Revert "crossgcc: Upgrade GCC to 9.2.0"

Revert the upgrade as it breaks at least the devicetree parser on
aarch64, tested on qemu aarch64 target.

This reverts commit dfd3f211740be4cf0d234bf4621ac384758a24ce.

Change-Id: I65607817188db21533014caa6d15be9a2004d498
Signed-off-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/39571
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>

9 files changed, 21174 insertions, 3 deletions

diff --git a/util/crossgcc/buildgcc b/util/crossgcc/buildgcc
index 992079979625..150e6166525e 100755
--- a/util/crossgcc/buildgcc
+++ b/util/crossgcc/buildgcc
@@ -50,7 +50,7 @@ THREADS=1
 GMP_VERSION=6.1.2
 MPFR_VERSION=4.0.2
 MPC_VERSION=1.1.0
-GCC_VERSION=9.2.0
+GCC_VERSION=8.3.0
 GCC_AUTOCONF_VERSION=2.69
 BINUTILS_VERSION=2.33.1
 GDB_VERSION=8.3.1
diff --git a/util/crossgcc/patches/gcc-9.2.0_ada-musl_workaround.patch b/util/crossgcc/patches/gcc-8.3.0_ada-musl_workaround.patch
index 1f504a4b5ebb..1f504a4b5ebb 100644
--- a/util/crossgcc/patches/gcc-9.2.0_ada-musl_workaround.patch
+++ b/util/crossgcc/patches/gcc-8.3.0_ada-musl_workaround.patch
diff --git a/util/crossgcc/patches/gcc-8.3.0_gnat-bad_constant.patch b/util/crossgcc/patches/gcc-8.3.0_gnat-bad_constant.patch
new file mode 100644
index 000000000000..e98f933a1338
--- /dev/null
+++ b/util/crossgcc/patches/gcc-8.3.0_gnat-bad_constant.patch
@@ -0,0 +1,150 @@
+commit b6f742f96c62bab0582021455328ae3be58e16d3
+Author: pmderodat <pmderodat@138bc75d-0d04-0410-961f-82ee72b054a4>
+Date:   Fri May 25 09:05:10 2018 +0000
+
+    [Ada] Remove "constant" attribute on Osint.Unknown_Attributes
+    
+    2018-05-25  Arnaud Charlet  <charlet@adacore.com>
+    
+    gcc/ada/
+    
+            * exp_aggr.adb (Convert_To_Positional): Bump default for
+            Max_Others_Replicate to 32. Update comments.
+            * osint.ads (Unknown_Attributes): No longer pretend this is a constant.
+            (No_File_Info_Cache): Initialize separately.
+            * osint.adb (No_File_Info_Cache): Update initializer.
+    
+    git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@260739 138bc75d-0d04-0410-961f-82ee72b054a4
+
+diff --git a/gcc/ada/ChangeLog b/gcc/ada/ChangeLog
+index e4127e472aa..d56240b7b82 100644
+--- a/gcc/ada/ChangeLog
++++ b/gcc/ada/ChangeLog
+@@ -188,6 +188,14 @@
+ 	an allocator if the type is an unconstrained record type with default
+ 	discriminant.
+ 
++2018-05-25  Arnaud Charlet  <charlet@adacore.com>
++
++	* exp_aggr.adb (Convert_To_Positional): Bump default for
++	Max_Others_Replicate to 32. Update comments.
++	* osint.ads (Unknown_Attributes): No longer pretend this is a constant.
++	(No_File_Info_Cache): Initialize separately.
++	* osint.adb (No_File_Info_Cache): Update initializer.
++
+ 2018-05-04  John Marino  <gnugcc@marino.st>
+ 
+ 	PR ada/85635
+diff --git a/gcc/ada/exp_aggr.adb b/gcc/ada/exp_aggr.adb
+index f723c1b4d99..ff5210eb4e4 100644
+--- a/gcc/ada/exp_aggr.adb
++++ b/gcc/ada/exp_aggr.adb
+@@ -284,14 +284,14 @@ package body Exp_Aggr is
+ 
+    procedure Convert_To_Positional
+      (N                    : Node_Id;
+-      Max_Others_Replicate : Nat     := 5;
++      Max_Others_Replicate : Nat     := 32;
+       Handle_Bit_Packed    : Boolean := False);
+    --  If possible, convert named notation to positional notation. This
+    --  conversion is possible only in some static cases. If the conversion is
+    --  possible, then N is rewritten with the analyzed converted aggregate.
+    --  The parameter Max_Others_Replicate controls the maximum number of
+    --  values corresponding to an others choice that will be converted to
+-   --  positional notation (the default of 5 is the normal limit, and reflects
++   --  positional notation (the default of 32 is the normal limit, and reflects
+    --  the fact that normally the loop is better than a lot of separate
+    --  assignments). Note that this limit gets overridden in any case if
+    --  either of the restrictions No_Elaboration_Code or No_Implicit_Loops is
+@@ -301,11 +301,6 @@ package body Exp_Aggr is
+    --  Packed_Array_Aggregate_Handled, we set this parameter to True, since
+    --  these are cases we handle in there.
+ 
+-   --  It would seem useful to have a higher default for Max_Others_Replicate,
+-   --  but aggregates in the compiler make this impossible: the compiler
+-   --  bootstrap fails if Max_Others_Replicate is greater than 25. This
+-   --  is unexpected ???
+-
+    procedure Expand_Array_Aggregate (N : Node_Id);
+    --  This is the top-level routine to perform array aggregate expansion.
+    --  N is the N_Aggregate node to be expanded.
+@@ -4292,7 +4287,7 @@ package body Exp_Aggr is
+ 
+    procedure Convert_To_Positional
+      (N                    : Node_Id;
+-      Max_Others_Replicate : Nat     := 5;
++      Max_Others_Replicate : Nat     := 32;
+       Handle_Bit_Packed    : Boolean := False)
+    is
+       Typ : constant Entity_Id := Etype (N);
+diff --git a/gcc/ada/osint.adb b/gcc/ada/osint.adb
+index 0c23761b6dc..896fbc7ee37 100644
+--- a/gcc/ada/osint.adb
++++ b/gcc/ada/osint.adb
+@@ -250,8 +250,7 @@ package body Osint is
+       Attr : aliased File_Attributes;
+    end record;
+ 
+-   No_File_Info_Cache : constant File_Info_Cache :=
+-                          (No_File, Unknown_Attributes);
++   No_File_Info_Cache : constant File_Info_Cache := (No_File, (others => 0));
+ 
+    package File_Name_Hash_Table is new GNAT.HTable.Simple_HTable (
+      Header_Num => File_Hash_Num,
+diff --git a/gcc/ada/osint.ads b/gcc/ada/osint.ads
+index 65a87fe4ce3..6c75b521456 100644
+--- a/gcc/ada/osint.ads
++++ b/gcc/ada/osint.ads
+@@ -255,10 +255,26 @@ package Osint is
+    --  from the disk and then cached in the File_Attributes parameter (possibly
+    --  along with other values).
+ 
+-   type File_Attributes is private;
+-   Unknown_Attributes : constant File_Attributes;
++   File_Attributes_Size : constant Natural := 32;
++   --  This should be big enough to fit a "struct file_attributes" on any
++   --  system. It doesn't cause any malfunction if it is too big (which avoids
++   --  the need for either mapping the struct exactly or importing the sizeof
++   --  from C, which would result in dynamic code). However, it does waste
++   --  space (e.g. when a component of this type appears in a record, if it is
++   --  unnecessarily large). Note: for runtime units, use System.OS_Constants.
++   --  SIZEOF_struct_file_attributes instead, which has the exact value.
++
++   type File_Attributes is
++     array (1 .. File_Attributes_Size)
++       of System.Storage_Elements.Storage_Element;
++   for File_Attributes'Alignment use Standard'Maximum_Alignment;
++
++   Unknown_Attributes : File_Attributes;
+    --  A cache for various attributes for a file (length, accessibility,...)
+-   --  This must be initialized to Unknown_Attributes prior to the first call.
++   --  Will be initialized properly at elaboration (for efficiency later on,
++   --  avoid function calls every time we want to reset the attributes) prior
++   --  to the first usage. We cannot make it constant since the compiler may
++   --  put it in a read-only section.
+ 
+    function Is_Directory
+      (Name : C_File_Name;
+@@ -754,22 +770,4 @@ private
+    --  detected, the file being written is deleted, and a fatal error is
+    --  signalled.
+ 
+-   File_Attributes_Size : constant Natural := 32;
+-   --  This should be big enough to fit a "struct file_attributes" on any
+-   --  system. It doesn't cause any malfunction if it is too big (which avoids
+-   --  the need for either mapping the struct exactly or importing the sizeof
+-   --  from C, which would result in dynamic code). However, it does waste
+-   --  space (e.g. when a component of this type appears in a record, if it is
+-   --  unnecessarily large). Note: for runtime units, use System.OS_Constants.
+-   --  SIZEOF_struct_file_attributes instead, which has the exact value.
+-
+-   type File_Attributes is
+-     array (1 .. File_Attributes_Size)
+-       of System.Storage_Elements.Storage_Element;
+-   for File_Attributes'Alignment use Standard'Maximum_Alignment;
+-
+-   Unknown_Attributes : constant File_Attributes := (others => 0);
+-   --  Will be initialized properly at elaboration (for efficiency later on,
+-   --  avoid function calls every time we want to reset the attributes).
+-
+ end Osint;
diff --git a/util/crossgcc/patches/gcc-9.2.0_gnat.patch b/util/crossgcc/patches/gcc-8.3.0_gnat.patch
index 2d7cecee245d..2d7cecee245d 100644
--- a/util/crossgcc/patches/gcc-9.2.0_gnat.patch
+++ b/util/crossgcc/patches/gcc-8.3.0_gnat.patch
diff --git a/util/crossgcc/patches/gcc-9.2.0_libgcc.patch b/util/crossgcc/patches/gcc-8.3.0_libgcc.patch
index 2f75c92b41fc..2f75c92b41fc 100644
--- a/util/crossgcc/patches/gcc-9.2.0_libgcc.patch
+++ b/util/crossgcc/patches/gcc-8.3.0_libgcc.patch
diff --git a/util/crossgcc/patches/gcc-8.3.0_nds32_ite.patch b/util/crossgcc/patches/gcc-8.3.0_nds32_ite.patch
new file mode 100644
index 000000000000..2f0780be7f8b
--- /dev/null
+++ b/util/crossgcc/patches/gcc-8.3.0_nds32_ite.patch
@@ -0,0 +1,21019 @@
+diff -urN gcc-8.2.0.orig/gcc/common/config/nds32/nds32-common.c gcc-8.2.0/gcc/common/config/nds32/nds32-common.c
+--- gcc-8.2.0.orig/gcc/common/config/nds32/nds32-common.c	2018-04-06 07:51:33.000000000 +0200
++++ gcc-8.2.0/gcc/common/config/nds32/nds32-common.c	2019-01-25 15:38:32.817242625 +0100
+@@ -53,6 +53,16 @@
+ 
+       return true;
+ 
++    case OPT_misr_secure_:
++      /* Check the valid security level: 0 1 2 3.  */
++      if (value < 0 || value > 3)
++	{
++	  error_at (loc, "for the option -misr-secure=X, the valid X "
++			 "must be: 0, 1, 2, or 3");
++	  return false;
++	}
++      return true;
++
+     case OPT_mcache_block_size_:
+       /* Check valid value: 4 8 16 32 64 128 256 512.  */
+       if (exact_log2 (value) < 2 || exact_log2 (value) > 9)
+@@ -74,12 +84,19 @@
+ /* Implement TARGET_OPTION_OPTIMIZATION_TABLE.  */
+ static const struct default_options nds32_option_optimization_table[] =
+ {
++#if TARGET_LINUX_ABI == 0
++  /* Disable -fdelete-null-pointer-checks by default in ELF toolchain.  */
++  { OPT_LEVELS_ALL,               OPT_fdelete_null_pointer_checks,
++							   NULL, 0 },
++#endif
+   /* Enable -fsched-pressure by default at -O1 and above.  */
+   { OPT_LEVELS_1_PLUS,            OPT_fsched_pressure,     NULL, 1 },
+   /* Enable -fomit-frame-pointer by default at all optimization levels.  */
+   { OPT_LEVELS_ALL,               OPT_fomit_frame_pointer, NULL, 1 },
+   /* Enable -mrelax-hint by default at all optimization levels.  */
+   { OPT_LEVELS_ALL,               OPT_mrelax_hint,         NULL, 1 },
++  /* Enalbe -malways-align by default at -O1 and above, but not -Os or -Og.  */
++  { OPT_LEVELS_1_PLUS_SPEED_ONLY, OPT_malways_align,       NULL, 1 },
+   /* Enable -mv3push by default at -Os, but it is useless under V2 ISA.  */
+   { OPT_LEVELS_SIZE,              OPT_mv3push,             NULL, 1 },
+ 
+@@ -87,6 +104,19 @@
+ };
+ 
+ /* ------------------------------------------------------------------------ */
++
++/* Implement TARGET_EXCEPT_UNWIND_INFO.  */
++static enum unwind_info_type
++nds32_except_unwind_info (struct gcc_options *opts ATTRIBUTE_UNUSED)
++{
++  if (TARGET_LINUX_ABI)
++    return UI_DWARF2;
++
++  return UI_SJLJ;
++}
++
++/* ------------------------------------------------------------------------ */
++
+ 
+ /* Run-time Target Specification.  */
+ 
+@@ -103,6 +133,7 @@
+      TARGET_EXT_PERF   : Generate performance extention instrcution.
+      TARGET_EXT_PERF2  : Generate performance extention version 2 instrcution.
+      TARGET_EXT_STRING : Generate string extention instrcution.
++     TARGET_HW_ABS     : Generate hardware abs instruction.
+      TARGET_CMOV       : Generate conditional move instruction.  */
+ #undef TARGET_DEFAULT_TARGET_FLAGS
+ #define TARGET_DEFAULT_TARGET_FLAGS		\
+@@ -113,6 +144,7 @@
+    | MASK_EXT_PERF				\
+    | MASK_EXT_PERF2				\
+    | MASK_EXT_STRING				\
++   | MASK_HW_ABS				\
+    | MASK_CMOV)
+ 
+ #undef TARGET_HANDLE_OPTION
+@@ -125,7 +157,7 @@
+ /* Defining the Output Assembler Language.  */
+ 
+ #undef TARGET_EXCEPT_UNWIND_INFO
+-#define TARGET_EXCEPT_UNWIND_INFO sjlj_except_unwind_info
++#define TARGET_EXCEPT_UNWIND_INFO nds32_except_unwind_info
+ 
+ /* ------------------------------------------------------------------------ */
+ 
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/constants.md gcc-8.2.0/gcc/config/nds32/constants.md
+--- gcc-8.2.0.orig/gcc/config/nds32/constants.md	2018-04-22 09:46:39.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/constants.md	2019-01-25 15:38:32.821242637 +0100
+@@ -23,6 +23,7 @@
+ (define_constants
+   [(R8_REGNUM  8)
+    (TA_REGNUM 15)
++   (TP_REGNUM 25)
+    (FP_REGNUM 28)
+    (GP_REGNUM 29)
+    (LP_REGNUM 30)
+@@ -49,6 +50,16 @@
+   UNSPEC_FFB
+   UNSPEC_FFMISM
+   UNSPEC_FLMISM
++  UNSPEC_KDMBB
++  UNSPEC_KDMBT
++  UNSPEC_KDMTB
++  UNSPEC_KDMTT
++  UNSPEC_KHMBB
++  UNSPEC_KHMBT
++  UNSPEC_KHMTB
++  UNSPEC_KHMTT
++  UNSPEC_KSLRAW
++  UNSPEC_KSLRAWU
+   UNSPEC_SVA
+   UNSPEC_SVS
+   UNSPEC_WSBH
+@@ -62,6 +73,29 @@
+   UNSPEC_UASTORE_HW
+   UNSPEC_UASTORE_W
+   UNSPEC_UASTORE_DW
++  UNSPEC_GOTINIT
++  UNSPEC_GOT
++  UNSPEC_GOTOFF
++  UNSPEC_PLT
++  UNSPEC_TLSGD
++  UNSPEC_TLSLD
++  UNSPEC_TLSIE
++  UNSPEC_TLSLE
++  UNSPEC_ROUND
++  UNSPEC_VEC_COMPARE
++  UNSPEC_KHM
++  UNSPEC_KHMX
++  UNSPEC_CLIP_OV
++  UNSPEC_CLIPS_OV
++  UNSPEC_BITREV
++  UNSPEC_KABS
++  UNSPEC_LOOP_END
++  UNSPEC_TLS_DESC
++  UNSPEC_TLS_IE
++  UNSPEC_ADD32
++  UNSPEC_ICT
++  UNSPEC_KADDH
++  UNSPEC_KSUBH
+ ])
+ 
+ ;; The unspec_volatile operation index.
+@@ -135,10 +169,14 @@
+   UNSPEC_VOLATILE_SET_TRIG_EDGE
+   UNSPEC_VOLATILE_GET_TRIG_TYPE
+   UNSPEC_VOLATILE_RELAX_GROUP
++  UNSPEC_VOLATILE_OMIT_FP_BEGIN
++  UNSPEC_VOLATILE_OMIT_FP_END
+   UNSPEC_VOLATILE_POP25_RETURN
+   UNSPEC_VOLATILE_UNALIGNED_FEATURE
+   UNSPEC_VOLATILE_ENABLE_UNALIGNED
+   UNSPEC_VOLATILE_DISABLE_UNALIGNED
++  UNSPEC_VOLATILE_RDOV
++  UNSPEC_VOLATILE_CLROV
+ ])
+ 
+ ;; ------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/constraints.md gcc-8.2.0/gcc/config/nds32/constraints.md
+--- gcc-8.2.0.orig/gcc/config/nds32/constraints.md	2018-04-06 07:51:33.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/constraints.md	2019-01-25 15:38:32.821242637 +0100
+@@ -127,6 +127,11 @@
+   (and (match_code "const_int")
+        (match_test "IN_RANGE (ival, -31, 0)")))
+ 
++(define_constraint "Iu06"
++  "Unsigned immediate 6-bit value"
++  (and (match_code "const_int")
++       (match_test "ival < (1 << 6) && ival >= 0")))
++
+ ;; Ip05 is special and dedicated for v3 movpi45 instruction.
+ ;; movpi45 has imm5u field but the range is 16 ~ 47.
+ (define_constraint "Ip05"
+@@ -136,10 +141,10 @@
+ 		    && ival >= (0 + 16)
+ 		    && (TARGET_ISA_V3 || TARGET_ISA_V3M)")))
+ 
+-(define_constraint "Iu06"
++(define_constraint "IU06"
+   "Unsigned immediate 6-bit value constraint for addri36.sp instruction"
+   (and (match_code "const_int")
+-       (match_test "ival < (1 << 6)
++       (match_test "ival < (1 << 8)
+ 		    && ival >= 0
+ 		    && (ival % 4 == 0)
+ 		    && (TARGET_ISA_V3 || TARGET_ISA_V3M)")))
+@@ -302,6 +307,25 @@
+        (match_test "(TARGET_ISA_V3 || TARGET_ISA_V3M)
+ 		    && (IN_RANGE (exact_log2 (ival + 1), 1, 8))")))
+ 
++(define_constraint "CVp5"
++  "Unsigned immediate 5-bit value for movpi45 instruction with range 16-47"
++  (and (match_code "const_vector")
++       (match_test "nds32_valid_CVp5_p (op)")))
++
++(define_constraint "CVs5"
++  "Signed immediate 5-bit value"
++  (and (match_code "const_vector")
++       (match_test "nds32_valid_CVs5_p (op)")))
++
++(define_constraint "CVs2"
++  "Signed immediate 20-bit value"
++  (and (match_code "const_vector")
++       (match_test "nds32_valid_CVs2_p (op)")))
++
++(define_constraint "CVhi"
++  "The immediate value that can be simply set high 20-bit"
++  (and (match_code "const_vector")
++       (match_test "nds32_valid_CVhi_p (op)")))
+ 
+ (define_memory_constraint "U33"
+   "Memory constraint for 333 format"
+@@ -349,4 +373,9 @@
+        (match_test "(TARGET_FPU_SINGLE || TARGET_FPU_DOUBLE)
+ 		     && nds32_float_mem_operand_p (op)")))
+ 
++(define_constraint "S"
++  "@internal
++   A constant call address."
++  (match_operand 0 "nds32_symbolic_operand"))
++
+ ;; ------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/elf.h gcc-8.2.0/gcc/config/nds32/elf.h
+--- gcc-8.2.0.orig/gcc/config/nds32/elf.h	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/elf.h	2019-01-25 15:38:32.821242637 +0100
+@@ -0,0 +1,81 @@
++/* Definitions of target machine of Andes NDS32 cpu for GNU compiler
++   Copyright (C) 2012-2014 Free Software Foundation, Inc.
++   Contributed by Andes Technology Corporation.
++
++   This file is part of GCC.
++
++   GCC is free software; you can redistribute it and/or modify it
++   under the terms of the GNU General Public License as published
++   by the Free Software Foundation; either version 3, or (at your
++   option) any later version.
++
++   GCC is distributed in the hope that it will be useful, but WITHOUT
++   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++   License for more details.
++
++   You should have received a copy of the GNU General Public License
++   along with GCC; see the file COPYING3.  If not see
++   <http://www.gnu.org/licenses/>.  */
++
++
++/* ------------------------------------------------------------------------ */
++
++#define TARGET_LINUX_ABI 0
++
++/* In the configure stage we may use options --enable-default-relax,
++   --enable-Os-default-ifc and --enable-Os-default-ex9.  They effect
++   the default spec of passing --relax, --mifc, and --mex9 to linker.
++   We use NDS32_RELAX_SPEC, NDS32_IFC_SPEC, and NDS32_EX9_SPEC
++   so that we can customize them conveniently.  */
++#define LINK_SPEC \
++  " %{G*}" \
++  " %{mbig-endian:-EB} %{mlittle-endian:-EL}" \
++  " %{shared:-shared}" \
++  NDS32_RELAX_SPEC
++
++#define LIB_SPEC \
++  " -lc -lgloss"
++
++#define LIBGCC_SPEC \
++  " -lgcc"
++
++/* The option -mno-ctor-dtor can disable constructor/destructor feature
++   by applying different crt stuff.  In the convention, crt0.o is the
++   startup file without constructor/destructor;
++   crt1.o, crti.o, crtbegin.o, crtend.o, and crtn.o are the
++   startup files with constructor/destructor.
++   Note that crt0.o, crt1.o, crti.o, and crtn.o are provided
++   by newlib/mculib/glibc/ublic, while crtbegin.o and crtend.o are
++   currently provided by GCC for nds32 target.
++
++   For nds32 target so far:
++   If -mno-ctor-dtor, we are going to link
++   "crt0.o [user objects]".
++   If -mctor-dtor, we are going to link
++   "crt1.o crtbegin1.o [user objects] crtend1.o".
++
++   Note that the TARGET_DEFAULT_CTOR_DTOR would effect the
++   default behavior.  Check gcc/config.gcc for more information.  */
++#ifdef TARGET_DEFAULT_CTOR_DTOR
++  #define STARTFILE_SPEC \
++    " %{!mno-ctor-dtor:crt1.o%s;:crt0.o%s}" \
++    " %{!mno-ctor-dtor:crtbegin1.o%s}" \
++    " %{mcrt-arg:crtarg.o%s}"
++  #define ENDFILE_SPEC \
++    " %{!mno-ctor-dtor:crtend1.o%s}"
++#else
++  #define STARTFILE_SPEC \
++    " %{mctor-dtor|coverage:crt1.o%s;:crt0.o%s}" \
++    " %{mctor-dtor|coverage:crtbegin1.o%s}" \
++    " %{mcrt-arg:crtarg.o%s}"
++  #define ENDFILE_SPEC \
++    " %{mctor-dtor|coverage:crtend1.o%s}"
++#endif
++
++#define STARTFILE_CXX_SPEC \
++  " %{!mno-ctor-dtor:crt1.o%s;:crt0.o%s}" \
++  " %{!mno-ctor-dtor:crtbegin1.o%s}" \
++  " %{mcrt-arg:crtarg.o%s}"
++#define ENDFILE_CXX_SPEC \
++  " %{!mno-ctor-dtor:crtend1.o%s}"
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/iterators.md gcc-8.2.0/gcc/config/nds32/iterators.md
+--- gcc-8.2.0.orig/gcc/config/nds32/iterators.md	2018-04-06 07:51:33.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/iterators.md	2019-01-25 15:38:32.821242637 +0100
+@@ -68,6 +68,28 @@
+ ;; shifts
+ (define_code_iterator shift_rotate [ashift ashiftrt lshiftrt rotatert])
+ 
++(define_code_iterator shifts [ashift ashiftrt lshiftrt])
++
++(define_code_iterator shiftrt [ashiftrt lshiftrt])
++
++(define_code_iterator sat_plus [ss_plus us_plus])
++
++(define_code_iterator all_plus [plus ss_plus us_plus])
++
++(define_code_iterator sat_minus [ss_minus us_minus])
++
++(define_code_iterator all_minus [minus ss_minus us_minus])
++
++(define_code_iterator plus_minus [plus minus])
++
++(define_code_iterator extend [sign_extend zero_extend])
++
++(define_code_iterator sumax [smax umax])
++
++(define_code_iterator sumin [smin umin])
++
++(define_code_iterator sumin_max [smax umax smin umin])
++
+ ;;----------------------------------------------------------------------------
+ ;; Code attributes.
+ ;;----------------------------------------------------------------------------
+@@ -76,5 +98,23 @@
+ (define_code_attr shift
+   [(ashift "ashl") (ashiftrt "ashr") (lshiftrt "lshr") (rotatert "rotr")])
+ 
++(define_code_attr su
++  [(ashiftrt "") (lshiftrt "u") (sign_extend "s") (zero_extend "u")])
++
++(define_code_attr zs
++  [(sign_extend "s") (zero_extend "z")])
++
++(define_code_attr uk
++  [(plus "") (ss_plus "k") (us_plus "uk")
++   (minus "") (ss_minus "k") (us_minus "uk")])
++
++(define_code_attr opcode
++  [(plus "add") (minus "sub") (smax "smax") (umax "umax") (smin "smin") (umin "umin")])
++
++(define_code_attr add_rsub
++  [(plus "a") (minus "rs")])
++
++(define_code_attr add_sub
++  [(plus "a") (minus "s")])
+ 
+ ;;----------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/linux.h gcc-8.2.0/gcc/config/nds32/linux.h
+--- gcc-8.2.0.orig/gcc/config/nds32/linux.h	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/linux.h	2019-01-25 15:38:32.821242637 +0100
+@@ -0,0 +1,86 @@
++/* Definitions of target machine of Andes NDS32 cpu for GNU compiler
++   Copyright (C) 2012-2014 Free Software Foundation, Inc.
++   Contributed by Andes Technology Corporation.
++
++   This file is part of GCC.
++
++   GCC is free software; you can redistribute it and/or modify it
++   under the terms of the GNU General Public License as published
++   by the Free Software Foundation; either version 3, or (at your
++   option) any later version.
++
++   GCC is distributed in the hope that it will be useful, but WITHOUT
++   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++   License for more details.
++
++   You should have received a copy of the GNU General Public License
++   along with GCC; see the file COPYING3.  If not see
++   <http://www.gnu.org/licenses/>.  */
++
++
++/* ------------------------------------------------------------------------ */
++
++#define TARGET_LINUX_ABI 1
++
++#undef  SIZE_TYPE
++#define SIZE_TYPE "unsigned int"
++
++#undef  PTRDIFF_TYPE
++#define PTRDIFF_TYPE "int"
++
++#define TARGET_OS_CPP_BUILTINS()                \
++  do                                            \
++    {                                           \
++      GNU_USER_TARGET_OS_CPP_BUILTINS();           \
++    }                                           \
++  while (0)
++
++#ifdef TARGET_BIG_ENDIAN_DEFAULT
++#define LD_SO_ENDIAN_SPEC "%{mlittle-endian:le}%{!mlittle-endian:be}"
++#else
++#define LD_SO_ENDIAN_SPEC "%{mbig-endian:be}%{!mbig-endian:le}"
++#endif
++
++/* Record arch version in TARGET_ARCH_DEFAULT. 0 means soft ABI,
++   1 means  hard ABI and using full floating-point instruction,
++   2 means hard ABI and only using single-precision floating-point
++   instruction  */
++#if TARGET_ARCH_DEFAULT
++#define LD_SO_ABI_SPEC "%{!mabi=2:f}"
++#else
++#define LD_SO_ABI_SPEC "%{mabi=2fp+:f}"
++#endif
++
++#define GLIBC_DYNAMIC_LINKER \
++  "/lib/ld-linux-nds32" LD_SO_ENDIAN_SPEC LD_SO_ABI_SPEC ".so.1"
++
++/* In the configure stage we may use options --enable-default-relax,
++   --enable-Os-default-ifc and --enable-Os-default-ex9.  They effect
++   the default spec of passing --relax, --mifc, and --mex9 to linker.
++   We use NDS32_RELAX_SPEC, NDS32_IFC_SPEC, and NDS32_EX9_SPEC
++   so that we can customize them conveniently.  */
++#define LINK_SPEC \
++ " %{G*}" \
++ " %{mbig-endian:-EB} %{mlittle-endian:-EL}" \
++ " %{shared:-shared} \
++  %{!shared: \
++    %{!static: \
++      %{rdynamic:-export-dynamic} \
++      -dynamic-linker " GNU_USER_DYNAMIC_LINKER "} \
++    %{static:-static}}" \
++  NDS32_RELAX_SPEC
++
++#define LINK_PIE_SPEC "%{pie:%{!fno-pie:%{!fno-PIE:%{!static:-pie}}}} "
++
++#define CPP_SPEC "%{pthread:-D_REENTRANT}"
++
++/* The SYNC operations are implemented as library functions, not
++   INSN patterns.  As a result, the HAVE defines for the patterns are
++   not defined.  We need to define them to generate the corresponding
++   __GCC_HAVE_SYNC_COMPARE_AND_SWAP_* and __GCC_ATOMIC_*_LOCK_FREE
++   defines.
++   Ref: https://sourceware.org/ml/libc-alpha/2014-09/msg00322.html  */
++#define HAVE_sync_compare_and_swapqi 1
++#define HAVE_sync_compare_and_swaphi 1
++#define HAVE_sync_compare_and_swapsi 1
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32.c gcc-8.2.0/gcc/config/nds32/nds32.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32.c	2018-05-07 03:38:02.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32.c	2019-01-25 15:38:32.833242671 +0100
+@@ -305,6 +305,7 @@
+   { "nested",       0,  0, false, false, false, false, NULL, NULL },
+   { "not_nested",   0,  0, false, false, false, false, NULL, NULL },
+   { "nested_ready", 0,  0, false, false, false, false, NULL, NULL },
++  { "critical",     0,  0, false, false, false, false, NULL, NULL },
+ 
+   /* The attributes describing isr register save scheme.  */
+   { "save_all",     0,  0, false, false, false, false, NULL, NULL },
+@@ -314,9 +315,19 @@
+   { "nmi",          1,  1, false, false, false, false, NULL, NULL },
+   { "warm",         1,  1, false, false, false, false, NULL, NULL },
+ 
++  /* The attributes describing isr security level. */
++  { "secure",       1,  1, false, false, false, false, NULL, NULL },
++
+   /* The attribute telling no prologue/epilogue.  */
+   { "naked",        0,  0, false, false, false, false, NULL, NULL },
+ 
++  /* The attribute is used to tell this function to be ROM patch.  */
++  { "indirect_call",0,  0, false, false, false, false, NULL, NULL },
++
++  /* FOR BACKWARD COMPATIBILITY,
++     this attribute also tells no prologue/epilogue.  */
++  { "no_prologue",  0,  0, false, false, false, false, NULL, NULL },
++
+   /* The last attribute spec is set to be NULL.  */
+   { NULL,           0,  0, false, false, false, false, NULL, NULL }
+ };
+@@ -345,6 +356,10 @@
+   /* Initially this function is not under strictly aligned situation.  */
+   machine->strict_aligned_p = 0;
+ 
++  /* Initially this function has no naked and no_prologue attributes.  */
++  machine->attr_naked_p = 0;
++  machine->attr_no_prologue_p = 0;
++
+   return machine;
+ }
+ 
+@@ -362,6 +377,15 @@
+      needs prologue/epilogue.  */
+   cfun->machine->naked_p = 0;
+ 
++  /* We need to mark whether this function has naked and no_prologue
++     attribute so that we can distinguish the difference if users applies
++     -mret-in-naked-func option.  */
++  cfun->machine->attr_naked_p
++    = lookup_attribute ("naked", DECL_ATTRIBUTES (current_function_decl))
++      ? 1 : 0;
++  cfun->machine->attr_no_prologue_p
++    = lookup_attribute ("no_prologue", DECL_ATTRIBUTES (current_function_decl))
++      ? 1 : 0;
+ 
+   /* If __builtin_eh_return is used, we better have frame pointer needed
+      so that we can easily locate the stack slot of return address.  */
+@@ -432,7 +456,8 @@
+ 
+   /* If $gp value is required to be saved on stack, it needs 4 bytes space.
+      Check whether we are using PIC code genration.  */
+-  cfun->machine->gp_size = (flag_pic) ? 4 : 0;
++  cfun->machine->gp_size =
++    (flag_pic && df_regs_ever_live_p (PIC_OFFSET_TABLE_REGNUM)) ? 4 : 0;
+ 
+   /* If $lp value is required to be saved on stack, it needs 4 bytes space.
+      Check whether $lp is ever live.  */
+@@ -497,7 +522,7 @@
+     }
+ 
+   /* Check if this function can omit prologue/epilogue code fragment.
+-     If there is 'naked' attribute in this function,
++     If there is 'no_prologue'/'naked'/'secure' attribute in this function,
+      we can set 'naked_p' flag to indicate that
+      we do not have to generate prologue/epilogue.
+      Or, if all the following conditions succeed,
+@@ -510,14 +535,17 @@
+ 		    is no outgoing size.
+        condition 3: There is no local_size, which means
+ 		    we do not need to adjust $sp.  */
+-  if (lookup_attribute ("naked", DECL_ATTRIBUTES (current_function_decl))
++  if (lookup_attribute ("no_prologue", DECL_ATTRIBUTES (current_function_decl))
++      || lookup_attribute ("naked", DECL_ATTRIBUTES (current_function_decl))
++      || lookup_attribute ("secure", DECL_ATTRIBUTES (current_function_decl))
+       || (cfun->machine->callee_saved_first_gpr_regno == SP_REGNUM
+ 	  && cfun->machine->callee_saved_last_gpr_regno == SP_REGNUM
+ 	  && cfun->machine->callee_saved_first_fpr_regno == SP_REGNUM
+ 	  && cfun->machine->callee_saved_last_fpr_regno == SP_REGNUM
+ 	  && !df_regs_ever_live_p (FP_REGNUM)
+ 	  && !df_regs_ever_live_p (LP_REGNUM)
+-	  && cfun->machine->local_size == 0))
++	  && cfun->machine->local_size == 0
++	  && !flag_pic))
+     {
+       /* Set this function 'naked_p' and other functions can check this flag.
+ 	 Note that in nds32 port, the 'naked_p = 1' JUST means there is no
+@@ -1259,6 +1287,32 @@
+   REG_NOTES (parallel_insn) = dwarf;
+ }
+ 
++static void
++nds32_emit_load_gp (void)
++{
++  rtx got_symbol, pat;
++
++  /* Initial GLOBAL OFFSET TABLE don't do the scheduling.  */
++  emit_insn (gen_blockage ());
++
++  got_symbol = gen_rtx_SYMBOL_REF (Pmode, "_GLOBAL_OFFSET_TABLE_");
++  /* sethi $gp, _GLOBAL_OFFSET_TABLE_ -8 */
++  pat = gen_rtx_UNSPEC (SImode, gen_rtvec (1, got_symbol), UNSPEC_GOTINIT);
++  pat = gen_rtx_CONST (SImode, gen_rtx_PLUS (Pmode, pat, GEN_INT (-8)));
++  emit_insn (gen_sethi (pic_offset_table_rtx,pat));
++
++  /* ori $gp, $gp, _GLOBAL_OFFSET_TABLE_ -4 */
++  pat = gen_rtx_UNSPEC (SImode, gen_rtvec (1, got_symbol), UNSPEC_GOTINIT);
++  pat = gen_rtx_CONST (SImode, gen_rtx_PLUS (Pmode, pat, GEN_INT (-4)));
++  emit_insn (gen_lo_sum (pic_offset_table_rtx, pic_offset_table_rtx, pat));
++
++  /* add5.pc $gp */
++  emit_insn (gen_add_pc (pic_offset_table_rtx, pic_offset_table_rtx));
++
++  /* Initial GLOBAL OFFSET TABLE don't do the scheduling.  */
++  emit_insn (gen_blockage ());
++}
++
+ /* Function that may creates more instructions
+    for large value on adjusting stack pointer.
+ 
+@@ -1342,17 +1396,25 @@
+ }
+ 
+ /* Return true if FUNC is a naked function.  */
+-static bool
++bool
+ nds32_naked_function_p (tree func)
+ {
+-  tree t;
++  /* FOR BACKWARD COMPATIBILITY,
++     we need to support 'no_prologue' attribute as well.  */
++  tree t_naked;
++  tree t_no_prologue;
+ 
+   if (TREE_CODE (func) != FUNCTION_DECL)
+     abort ();
+ 
+-  t = lookup_attribute ("naked", DECL_ATTRIBUTES (func));
++  /* We have to use lookup_attribute() to check attributes.
++     Because attr_naked_p and attr_no_prologue_p are set in
++     nds32_compute_stack_frame() and the function has not been
++     invoked yet.  */
++  t_naked       = lookup_attribute ("naked", DECL_ATTRIBUTES (func));
++  t_no_prologue = lookup_attribute ("no_prologue", DECL_ATTRIBUTES (func));
+ 
+-  return (t != NULL_TREE);
++  return ((t_naked != NULL_TREE) || (t_no_prologue != NULL_TREE));
+ }
+ 
+ /* Function that determine whether a load postincrement is a good thing to use
+@@ -1570,6 +1632,11 @@
+ nds32_register_passes (void)
+ {
+   nds32_register_pass (
++    make_pass_nds32_fp_as_gp,
++    PASS_POS_INSERT_BEFORE,
++    "ira");
++
++  nds32_register_pass (
+     make_pass_nds32_relax_opt,
+     PASS_POS_INSERT_AFTER,
+     "mach");
+@@ -1636,6 +1703,9 @@
+ {
+   int regno;
+ 
++  if (TARGET_LINUX_ABI)
++    fixed_regs[TP_REGNUM] = 1;
++
+   if (TARGET_HARD_FLOAT)
+     {
+       for (regno = NDS32_FIRST_FPR_REGNUM;
+@@ -1987,6 +2057,16 @@
+ 	  : PARM_BOUNDARY);
+ }
+ 
++bool
++nds32_vector_mode_supported_p (machine_mode mode)
++{
++  if (mode == V4QImode
++      || mode == V2HImode)
++    return NDS32_EXT_DSP_P ();
++
++  return false;
++}
++
+ /* -- How Scalar Function Values Are Returned.  */
+ 
+ static rtx
+@@ -2124,56 +2204,12 @@
+ nds32_asm_function_end_prologue (FILE *file)
+ {
+   fprintf (file, "\t! END PROLOGUE\n");
+-
+-  /* If frame pointer is NOT needed and -mfp-as-gp is issued,
+-     we can generate special directive: ".omit_fp_begin"
+-     to guide linker doing fp-as-gp optimization.
+-     However, for a naked function, which means
+-     it should not have prologue/epilogue,
+-     using fp-as-gp still requires saving $fp by push/pop behavior and
+-     there is no benefit to use fp-as-gp on such small function.
+-     So we need to make sure this function is NOT naked as well.  */
+-  if (!frame_pointer_needed
+-      && !cfun->machine->naked_p
+-      && cfun->machine->fp_as_gp_p)
+-    {
+-      fprintf (file, "\t! ----------------------------------------\n");
+-      fprintf (file, "\t! Guide linker to do "
+-		     "link time optimization: fp-as-gp\n");
+-      fprintf (file, "\t! We add one more instruction to "
+-		     "initialize $fp near to $gp location.\n");
+-      fprintf (file, "\t! If linker fails to use fp-as-gp transformation,\n");
+-      fprintf (file, "\t! this extra instruction should be "
+-		     "eliminated at link stage.\n");
+-      fprintf (file, "\t.omit_fp_begin\n");
+-      fprintf (file, "\tla\t$fp,_FP_BASE_\n");
+-      fprintf (file, "\t! ----------------------------------------\n");
+-    }
+ }
+ 
+ /* Before rtl epilogue has been expanded, this function is used.  */
+ static void
+ nds32_asm_function_begin_epilogue (FILE *file)
+ {
+-  /* If frame pointer is NOT needed and -mfp-as-gp is issued,
+-     we can generate special directive: ".omit_fp_end"
+-     to claim fp-as-gp optimization range.
+-     However, for a naked function,
+-     which means it should not have prologue/epilogue,
+-     using fp-as-gp still requires saving $fp by push/pop behavior and
+-     there is no benefit to use fp-as-gp on such small function.
+-     So we need to make sure this function is NOT naked as well.  */
+-  if (!frame_pointer_needed
+-      && !cfun->machine->naked_p
+-      && cfun->machine->fp_as_gp_p)
+-    {
+-      fprintf (file, "\t! ----------------------------------------\n");
+-      fprintf (file, "\t! Claim the range of fp-as-gp "
+-		     "link time optimization\n");
+-      fprintf (file, "\t.omit_fp_end\n");
+-      fprintf (file, "\t! ----------------------------------------\n");
+-    }
+-
+   fprintf (file, "\t! BEGIN EPILOGUE\n");
+ }
+ 
+@@ -2200,6 +2236,26 @@
+ 		? 1
+ 		: 0);
+ 
++  if (flag_pic)
++    {
++      fprintf (file, "\tsmw.adm\t$r31, [$r31], $r31, 4\n");
++      fprintf (file, "\tsethi\t%s, hi20(_GLOBAL_OFFSET_TABLE_-8)\n",
++		      reg_names [PIC_OFFSET_TABLE_REGNUM]);
++      fprintf (file, "\tori\t%s, %s, lo12(_GLOBAL_OFFSET_TABLE_-4)\n",
++		      reg_names [PIC_OFFSET_TABLE_REGNUM],
++		      reg_names [PIC_OFFSET_TABLE_REGNUM]);
++
++      if (TARGET_ISA_V3)
++	fprintf (file, "\tadd5.pc\t$gp\n");
++      else
++	{
++	  fprintf (file, "\tmfusr\t$ta, $pc\n");
++	  fprintf (file, "\tadd\t%s, $ta, %s\n",
++			  reg_names [PIC_OFFSET_TABLE_REGNUM],
++			  reg_names [PIC_OFFSET_TABLE_REGNUM]);
++	}
++    }
++
+   if (delta != 0)
+     {
+       if (satisfies_constraint_Is15 (GEN_INT (delta)))
+@@ -2224,9 +2280,23 @@
+ 	}
+     }
+ 
+-  fprintf (file, "\tb\t");
+-  assemble_name (file, XSTR (XEXP (DECL_RTL (function), 0), 0));
+-  fprintf (file, "\n");
++  if (flag_pic)
++    {
++      fprintf (file, "\tla\t$ta, ");
++      assemble_name (file, XSTR (XEXP (DECL_RTL (function), 0), 0));
++      fprintf (file, "@PLT\n");
++      fprintf (file, "\t! epilogue\n");
++      fprintf (file, "\tlwi.bi\t%s, [%s], 4\n",
++	       reg_names[PIC_OFFSET_TABLE_REGNUM],
++	       reg_names[STACK_POINTER_REGNUM]);
++      fprintf (file, "\tbr\t$ta\n");
++    }
++  else
++    {
++      fprintf (file, "\tb\t");
++      assemble_name (file, XSTR (XEXP (DECL_RTL (function), 0), 0));
++      fprintf (file, "\n");
++    }
+ 
+   final_end_function ();
+ }
+@@ -2242,15 +2312,20 @@
+ 
+   /* 1. Do not apply sibling call if -mv3push is enabled,
+ 	because pop25 instruction also represents return behavior.
+-     2. If this function is a variadic function, do not apply sibling call
++     2. If this function is a isr function, do not apply sibling call
++	because it may perform the behavior that user does not expect.
++     3. If this function is a variadic function, do not apply sibling call
+ 	because the stack layout may be a mess.
+-     3. We don't want to apply sibling call optimization for indirect
++     4. We don't want to apply sibling call optimization for indirect
+ 	sibcall because the pop behavior in epilogue may pollute the
+ 	content of caller-saved regsiter when the register is used for
+-	indirect sibcall.  */
++	indirect sibcall.
++     5. In pic mode, it may use some registers for PLT call.  */
+   return (!TARGET_V3PUSH
++	  && !nds32_isr_function_p (current_function_decl)
+ 	  && (cfun->machine->va_args_size == 0)
+-	  && decl);
++	  && decl
++	  && !flag_pic);
+ }
+ 
+ /* Determine whether we need to enable warning for function return check.  */
+@@ -2566,6 +2641,13 @@
+ 
+     case SYMBOL_REF:
+       /* (mem (symbol_ref A)) => [symbol_ref] */
++
++      if (flag_pic || SYMBOL_REF_TLS_MODEL (x))
++	return false;
++
++      if (TARGET_ICT_MODEL_LARGE && nds32_indirect_call_referenced_p (x))
++	return false;
++
+       /* If -mcmodel=large, the 'symbol_ref' is not a valid address
+ 	 during or after LRA/reload phase.  */
+       if (TARGET_CMODEL_LARGE
+@@ -2577,7 +2659,8 @@
+ 	 the 'symbol_ref' is not a valid address during or after
+ 	 LRA/reload phase.  */
+       if (TARGET_CMODEL_MEDIUM
+-	  && NDS32_SYMBOL_REF_RODATA_P (x)
++	  && (NDS32_SYMBOL_REF_RODATA_P (x)
++	      || CONSTANT_POOL_ADDRESS_P (x))
+ 	  && (reload_completed
+ 	      || reload_in_progress
+ 	      || lra_in_progress))
+@@ -2599,6 +2682,10 @@
+ 	    {
+ 	      /* Now we see the [ + const_addr ] pattern, but we need
+ 		 some further checking.  */
++
++	      if (flag_pic || SYMBOL_REF_TLS_MODEL (op0))
++		return false;
++
+ 	      /* If -mcmodel=large, the 'const_addr' is not a valid address
+ 		 during or after LRA/reload phase.  */
+ 	      if (TARGET_CMODEL_LARGE
+@@ -2675,17 +2762,202 @@
+ 
+     case LO_SUM:
+       /* (mem (lo_sum (reg) (symbol_ref))) */
+-      /* (mem (lo_sum (reg) (const))) */
+-      gcc_assert (REG_P (XEXP (x, 0)));
+-      if (GET_CODE (XEXP (x, 1)) == SYMBOL_REF
+-	  || GET_CODE (XEXP (x, 1)) == CONST)
+-	return nds32_legitimate_address_p (mode, XEXP (x, 1), strict);
+-      else
++      /* (mem (lo_sum (reg) (const (plus (symbol_ref) (reg)))) */
++      /* TLS case: (mem (lo_sum (reg) (const (unspec symbol_ref X)))) */
++      /* The LO_SUM is a valid address if and only if we would like to
++	 generate 32-bit full address memory access with any of following
++	 circumstance:
++	   1. -mcmodel=large.
++	   2. -mcmodel=medium and the symbol_ref references to rodata.  */
++      {
++	rtx sym = NULL_RTX;
++
++	if (flag_pic)
++	  return false;
++
++	if (!REG_P (XEXP (x, 0)))
++	  return false;
++
++	if (GET_CODE (XEXP (x, 1)) == SYMBOL_REF)
++	  sym = XEXP (x, 1);
++	else if (GET_CODE (XEXP (x, 1)) == CONST)
++	  {
++	    rtx plus = XEXP(XEXP (x, 1), 0);
++	    if (GET_CODE (plus) == PLUS)
++	      sym = XEXP (plus, 0);
++	    else if (GET_CODE (plus) == UNSPEC)
++	      sym = XVECEXP (plus, 0, 0);
++	  }
++	else
++	  return false;
++
++	gcc_assert (GET_CODE (sym) == SYMBOL_REF);
++
++	if (TARGET_ICT_MODEL_LARGE
++	    && nds32_indirect_call_referenced_p (sym))
++	  return true;
++
++	if (TARGET_CMODEL_LARGE)
++	  return true;
++	else if (TARGET_CMODEL_MEDIUM
++		 && NDS32_SYMBOL_REF_RODATA_P (sym))
++	  return true;
++	else
++	  return false;
++      }
++
++    default:
++      return false;
++    }
++}
++
++static rtx
++nds32_legitimize_address (rtx x,
++			  rtx oldx ATTRIBUTE_UNUSED,
++			  machine_mode mode ATTRIBUTE_UNUSED)
++{
++  if (nds32_tls_referenced_p (x))
++    x = nds32_legitimize_tls_address (x);
++  else if (flag_pic && SYMBOLIC_CONST_P (x))
++    x = nds32_legitimize_pic_address (x);
++  else if (TARGET_ICT_MODEL_LARGE && nds32_indirect_call_referenced_p (x))
++    x = nds32_legitimize_ict_address (x);
++
++  return x;
++}
++
++static bool
++nds32_legitimate_constant_p (machine_mode mode, rtx x)
++{
++  switch (GET_CODE (x))
++    {
++    case CONST_DOUBLE:
++      if ((TARGET_FPU_SINGLE || TARGET_FPU_DOUBLE)
++	  && (mode == DFmode || mode == SFmode))
+ 	return false;
++      break;
++    case CONST:
++      x = XEXP (x, 0);
++
++      if (GET_CODE (x) == PLUS)
++	{
++	  if (!CONST_INT_P (XEXP (x, 1)))
++	    return false;
++	  x = XEXP (x, 0);
++	}
++
++      if (GET_CODE (x) == UNSPEC)
++	{
++	  switch (XINT (x, 1))
++	    {
++	    case UNSPEC_GOT:
++	    case UNSPEC_GOTOFF:
++	    case UNSPEC_PLT:
++	    case UNSPEC_TLSGD:
++	    case UNSPEC_TLSLD:
++	    case UNSPEC_TLSIE:
++	    case UNSPEC_TLSLE:
++	    case UNSPEC_ICT:
++	      return false;
++	    default:
++	      return true;
++	    }
++	}
++      break;
++    case SYMBOL_REF:
++      /* TLS symbols need a call to resolve in
++	 precompute_register_parameters.  */
++      if (SYMBOL_REF_TLS_MODEL (x))
++	return false;
++      break;
++    default:
++      return true;
++    }
++
++  return true;
++}
++
++/* Reorgnize the UNSPEC CONST and return its direct symbol.  */
++static rtx
++nds32_delegitimize_address (rtx x)
++{
++  x = delegitimize_mem_from_attrs (x);
++
++  if (GET_CODE(x) == CONST)
++    {
++      rtx inner = XEXP (x, 0);
++
++      /* Handle for GOTOFF.  */
++      if (GET_CODE (inner) == PLUS)
++	inner = XEXP (inner, 0);
++
++      if (GET_CODE (inner) == UNSPEC)
++	{
++	  switch (XINT (inner, 1))
++	    {
++	    case UNSPEC_GOTINIT:
++	    case UNSPEC_GOT:
++	    case UNSPEC_GOTOFF:
++	    case UNSPEC_PLT:
++	    case UNSPEC_TLSGD:
++	    case UNSPEC_TLSLD:
++	    case UNSPEC_TLSIE:
++	    case UNSPEC_TLSLE:
++	    case UNSPEC_ICT:
++	      x = XVECEXP (inner, 0, 0);
++	      break;
++	    default:
++	      break;
++	    }
++	}
++    }
++  return x;
++}
+ 
++static machine_mode
++nds32_vectorize_preferred_simd_mode (scalar_mode mode)
++{
++  if (!NDS32_EXT_DSP_P ())
++    return word_mode;
++
++  switch (mode)
++    {
++    case E_QImode:
++      return V4QImode;
++    case E_HImode:
++      return V2HImode;
++    default:
++      return word_mode;
++    }
++}
++
++static bool
++nds32_cannot_force_const_mem (machine_mode mode ATTRIBUTE_UNUSED, rtx x)
++{
++  switch (GET_CODE (x))
++    {
++    case CONST:
++      return !nds32_legitimate_constant_p (mode, x);
++    case SYMBOL_REF:
++      /* All symbols have to be accessed through gp-relative in PIC mode.  */
++      /* We don't want to force symbol as constant pool in .text section,
++	 because we use the gp-relatived instruction to load in small
++	 or medium model.  */
++      if (flag_pic
++	  || SYMBOL_REF_TLS_MODEL (x)
++	  || TARGET_CMODEL_SMALL
++	  || TARGET_CMODEL_MEDIUM)
++	return true;
++      break;
++    case CONST_INT:
++    case CONST_DOUBLE:
++      if (flag_pic && (lra_in_progress || reload_completed))
++	return true;
++      break;
+     default:
+       return false;
+     }
++  return false;
+ }
+ 
+ 
+@@ -2731,13 +3003,33 @@
+ /* Describing Relative Costs of Operations.  */
+ 
+ static int
+-nds32_register_move_cost (machine_mode mode ATTRIBUTE_UNUSED,
++nds32_register_move_cost (machine_mode mode,
+ 			  reg_class_t from,
+ 			  reg_class_t to)
+ {
++  /* In garywolf cpu, FPR to GPR is chaper than other cpu.  */
++  if (TARGET_PIPELINE_GRAYWOLF)
++    {
++      if (GET_MODE_SIZE (mode) == 8)
++	{
++	  /* DPR to GPR.  */
++	  if (from == FP_REGS && to != FP_REGS)
++	    return 3;
++	  /* GPR to DPR.  */
++	  if (from != FP_REGS && to == FP_REGS)
++	    return 2;
++	}
++      else
++	{
++	  if ((from == FP_REGS && to != FP_REGS)
++	      || (from != FP_REGS && to == FP_REGS))
++	    return 2;
++	}
++    }
++
+   if ((from == FP_REGS && to != FP_REGS)
+       || (from != FP_REGS && to == FP_REGS))
+-    return 9;
++    return 3;
+   else if (from == HIGH_REGS || to == HIGH_REGS)
+     return optimize_size ? 6 : 2;
+   else
+@@ -2825,6 +3117,9 @@
+ {
+   default_file_start ();
+ 
++  if (flag_pic)
++    fprintf (asm_out_file, "\t.pic\n");
++
+   /* Tell assembler which ABI we are using.  */
+   fprintf (asm_out_file, "\t! ABI version\n");
+   if (TARGET_HARD_FLOAT)
+@@ -2835,10 +3130,36 @@
+   /* Tell assembler that this asm code is generated by compiler.  */
+   fprintf (asm_out_file, "\t! This asm file is generated by compiler\n");
+   fprintf (asm_out_file, "\t.flag\tverbatim\n");
+-  /* Give assembler the size of each vector for interrupt handler.  */
+-  fprintf (asm_out_file, "\t! This vector size directive is required "
+-			 "for checking inconsistency on interrupt handler\n");
+-  fprintf (asm_out_file, "\t.vec_size\t%d\n", nds32_isr_vector_size);
++
++  /* Insert directive for linker to distinguish object's ict flag.  */
++  if (!TARGET_LINUX_ABI)
++    {
++      if (TARGET_ICT_MODEL_LARGE)
++	fprintf (asm_out_file, "\t.ict_model\tlarge\n");
++      else
++	fprintf (asm_out_file, "\t.ict_model\tsmall\n");
++    }
++
++  /* We need to provide the size of each vector for interrupt handler
++     under elf toolchain.  */
++  if (!TARGET_LINUX_ABI)
++    {
++      fprintf (asm_out_file, "\t! This vector size directive is required "
++			     "for checking inconsistency on interrupt handler\n");
++      fprintf (asm_out_file, "\t.vec_size\t%d\n", nds32_isr_vector_size);
++    }
++
++  /* If user enables '-mforce-fp-as-gp' or compiles programs with -Os,
++     the compiler may produce 'la $fp,_FP_BASE_' instruction
++     at prologue for fp-as-gp optimization.
++     We should emit weak reference of _FP_BASE_ to avoid undefined reference
++     in case user does not pass '--relax' option to linker.  */
++  if (!TARGET_LINUX_ABI && (TARGET_FORCE_FP_AS_GP || optimize_size))
++    {
++      fprintf (asm_out_file, "\t! This weak reference is required to do "
++			     "fp-as-gp link time optimization\n");
++      fprintf (asm_out_file, "\t.weak\t_FP_BASE_\n");
++    }
+ 
+   fprintf (asm_out_file, "\t! ------------------------------------\n");
+ 
+@@ -2849,6 +3170,49 @@
+   if (TARGET_ISA_V3M)
+     fprintf (asm_out_file, "\t! ISA family\t\t: %s\n", "V3M");
+ 
++  switch (nds32_cpu_option)
++    {
++    case CPU_N6:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "N6");
++      break;
++
++    case CPU_N7:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "N7");
++      break;
++
++    case CPU_N8:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "N8");
++      break;
++
++    case CPU_E8:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "E8");
++      break;
++
++    case CPU_N9:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "N9");
++      break;
++
++    case CPU_N10:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "N10");
++      break;
++
++    case CPU_GRAYWOLF:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "Graywolf");
++      break;
++
++    case CPU_N12:
++    case CPU_N13:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "N13");
++      break;
++
++    case CPU_SIMPLE:
++      fprintf (asm_out_file, "\t! Pipeline model\t: %s\n", "SIMPLE");
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
+   if (TARGET_CMODEL_SMALL)
+     fprintf (asm_out_file, "\t! Code model\t\t: %s\n", "SMALL");
+   if (TARGET_CMODEL_MEDIUM)
+@@ -2926,9 +3290,65 @@
+ {
+   nds32_asm_file_end_for_isr ();
+ 
++  /* The NDS32 Linux stack is mapped non-executable by default, so add a
++     .note.GNU-stack section.  */
++  if (TARGET_LINUX_ABI)
++    file_end_indicate_exec_stack ();
++
+   fprintf (asm_out_file, "\t! ------------------------------------\n");
+ }
+ 
++static bool
++nds32_asm_output_addr_const_extra (FILE *file, rtx x)
++{
++  if (GET_CODE (x) == UNSPEC)
++    {
++      switch (XINT (x, 1))
++	{
++	case UNSPEC_GOTINIT:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  break;
++	case UNSPEC_GOTOFF:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@GOTOFF", file);
++	  break;
++	case UNSPEC_GOT:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@GOT", file);
++	  break;
++	case UNSPEC_PLT:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@PLT", file);
++	  break;
++	case UNSPEC_TLSGD:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@TLSDESC", file);
++	  break;
++	case UNSPEC_TLSLD:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@TLSDESC", file);
++	  break;
++	case UNSPEC_TLSIE:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@GOTTPOFF", file);
++	  break;
++	case UNSPEC_TLSLE:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@TPOFF", file);
++	  break;
++	case UNSPEC_ICT:
++	  output_addr_const (file, XVECEXP (x, 0, 0));
++	  fputs ("@ICT", file);
++	  break;
++	default:
++	  return false;
++	}
++      return true;
++    }
++  else
++    return false;
++}
++
+ /* -- Output and Generation of Labels.  */
+ 
+ static void
+@@ -2978,6 +3398,18 @@
+ 
+       /* No need to handle following process, so return immediately.  */
+       return;
++
++    case 'v':
++      gcc_assert (CONST_INT_P (x)
++		  && (INTVAL (x) == 0
++		      || INTVAL (x) == 8
++		      || INTVAL (x) == 16
++		      || INTVAL (x) == 24));
++      fprintf (stream, HOST_WIDE_INT_PRINT_DEC, INTVAL (x) / 8);
++
++      /* No need to handle following process, so return immediately.  */
++      return;
++
+     case 'B':
+       /* Use exact_log2() to search the 1-bit position.  */
+       gcc_assert (CONST_INT_P (x));
+@@ -3084,8 +3516,15 @@
+   switch (GET_CODE (x))
+     {
+     case LABEL_REF:
++      output_addr_const (stream, x);
++      break;
++
+     case SYMBOL_REF:
+       output_addr_const (stream, x);
++
++      if (!TARGET_LINUX_ABI && nds32_indirect_call_referenced_p (x))
++	fprintf (stream, "@ICT");
++
+       break;
+ 
+     case REG:
+@@ -3168,6 +3607,17 @@
+       output_addr_const (stream, x);
+       break;
+ 
++    case CONST_VECTOR:
++      fprintf (stream, HOST_WIDE_INT_PRINT_HEX, const_vector_to_hwint (x));
++      break;
++
++    case LO_SUM:
++      /* This is a special case for inline assembly using memory address 'p'.
++	 The inline assembly code is expected to use pesudo instruction
++	 for the operand.  EX: la  */
++      output_addr_const (stream, XEXP(x, 1));
++      break;
++
+     default:
+       /* Generally, output_addr_const () is able to handle most cases.
+ 	 We want to see what CODE could appear,
+@@ -3179,7 +3629,9 @@
+ }
+ 
+ static void
+-nds32_print_operand_address (FILE *stream, machine_mode /*mode*/, rtx x)
++nds32_print_operand_address (FILE *stream,
++			     machine_mode mode ATTRIBUTE_UNUSED,
++			     rtx x)
+ {
+   rtx op0, op1;
+ 
+@@ -3194,6 +3646,16 @@
+       fputs ("]", stream);
+       break;
+ 
++    case LO_SUM:
++      /* This is a special case for inline assembly using memory operand 'm'.
++	 The inline assembly code is expected to use pesudo instruction
++	 for the operand.  EX: [ls].[bhw]  */
++      fputs ("[ + ", stream);
++      op1 = XEXP (x, 1);
++      output_addr_const (stream, op1);
++      fputs ("]", stream);
++      break;
++
+     case REG:
+       /* Forbid using static chain register ($r16)
+ 	 on reduced-set registers configuration.  */
+@@ -3260,6 +3722,20 @@
+ 			   reg_names[REGNO (XEXP (op0, 0))],
+ 			   sv);
+ 	}
++      else if (GET_CODE (op0) == ASHIFT && REG_P (op1))
++	{
++	  /* [Ra + Rb << sv]
++	     In normal, ASHIFT can be converted to MULT like above case.
++	     But when the address rtx does not go through canonicalize_address
++	     defined in fwprop, we'll need this case.  */
++	  int sv = INTVAL (XEXP (op0, 1));
++	  gcc_assert (sv <= 3 && sv >=0);
++
++	  fprintf (stream, "[%s + %s << %d]",
++		   reg_names[REGNO (op1)],
++		   reg_names[REGNO (XEXP (op0, 0))],
++		   sv);
++	}
+       else
+ 	{
+ 	  /* The control flow is not supposed to be here.  */
+@@ -3454,6 +3930,27 @@
+ static void
+ nds32_insert_attributes (tree decl, tree *attributes)
+ {
++  /* A "indirect_call" function attribute implies "noinline" and "noclone"
++     for elf toolchain to support ROM patch mechanism.  */
++  if (TREE_CODE (decl) == FUNCTION_DECL
++      && lookup_attribute ("indirect_call", *attributes) != NULL)
++    {
++      tree new_attrs = *attributes;
++
++      if (TARGET_LINUX_ABI)
++	error("cannot use indirect_call attribute under linux toolchain");
++
++      if (lookup_attribute ("noinline", new_attrs) == NULL)
++	new_attrs = tree_cons (get_identifier ("noinline"), NULL, new_attrs);
++      if (lookup_attribute ("noclone", new_attrs) == NULL)
++	new_attrs = tree_cons (get_identifier ("noclone"), NULL, new_attrs);
++
++      if (!TREE_PUBLIC (decl))
++	error("indirect_call attribute can't apply for static function");
++
++      *attributes = new_attrs;
++    }
++
+   /* For function declaration, we need to check isr-specific attributes:
+        1. Call nds32_check_isr_attrs_conflict() to check any conflict.
+        2. Check valid integer value for interrupt/exception.
+@@ -3479,6 +3976,38 @@
+       excp  = lookup_attribute ("exception", func_attrs);
+       reset = lookup_attribute ("reset", func_attrs);
+ 
++      /* The following code may use attribute arguments.  If there is no
++	 argument from source code, it will cause segmentation fault.
++	 Therefore, return dircetly and report error message later.  */
++      if ((intr && TREE_VALUE (intr) == NULL)
++	  || (excp && TREE_VALUE (excp) == NULL)
++	  || (reset && TREE_VALUE (reset) == NULL))
++	return;
++
++      /* ------------------------------------------------------------- */
++      /* FIXME:
++	 FOR BACKWARD COMPATIBILITY, we need to support following patterns:
++
++	     __attribute__((interrupt("XXX;YYY;id=ZZZ")))
++	     __attribute__((exception("XXX;YYY;id=ZZZ")))
++	     __attribute__((reset("vectors=XXX;nmi_func=YYY;warm_func=ZZZ")))
++
++	 If interrupt/exception/reset appears and its argument is a
++	 STRING_CST, we will use other functions to parse string in the
++	 nds32_construct_isr_vectors_information() and then set necessary
++	 isr information in the nds32_isr_vectors[] array.  Here we can
++	 just return immediately to avoid new-syntax checking.  */
++      if (intr != NULL_TREE
++	  && TREE_CODE (TREE_VALUE (TREE_VALUE (intr))) == STRING_CST)
++	return;
++      if (excp != NULL_TREE
++	  && TREE_CODE (TREE_VALUE (TREE_VALUE (excp))) == STRING_CST)
++	return;
++      if (reset != NULL_TREE
++	  && TREE_CODE (TREE_VALUE (TREE_VALUE (reset))) == STRING_CST)
++	return;
++      /* ------------------------------------------------------------- */
++
+       if (intr || excp)
+ 	{
+ 	  /* Deal with interrupt/exception.  */
+@@ -3598,7 +4127,9 @@
+     }
+   if (TARGET_ISA_V3)
+     {
+-      /* Under V3 ISA, currently nothing should be strictly set.  */
++      /* If this is ARCH_V3J, we need to enable TARGET_REDUCED_REGS.  */
++      if (nds32_arch_option == ARCH_V3J)
++	target_flags |= MASK_REDUCED_REGS;
+     }
+   if (TARGET_ISA_V3M)
+     {
+@@ -3610,6 +4141,9 @@
+       target_flags &= ~MASK_EXT_PERF2;
+       /* Under V3M ISA, we need to strictly disable TARGET_EXT_STRING.  */
+       target_flags &= ~MASK_EXT_STRING;
++
++      if (flag_pic)
++	error ("not support -fpic option for v3m toolchain");
+     }
+ 
+   /* See if we are using reduced-set registers:
+@@ -3627,6 +4161,12 @@
+ 	fixed_regs[r] = call_used_regs[r] = 1;
+     }
+ 
++  /* See if user explicitly would like to use fp-as-gp optimization.
++     If so, we must prevent $fp from being allocated
++     during register allocation.  */
++  if (TARGET_FORCE_FP_AS_GP)
++    fixed_regs[FP_REGNUM] = call_used_regs[FP_REGNUM] = 1;
++
+   if (!TARGET_16_BIT)
+     {
+       /* Under no 16 bit ISA, we need to strictly disable TARGET_V3PUSH.  */
+@@ -3643,9 +4183,7 @@
+ 	       "must be enable '-mext-fpu-sp' or '-mext-fpu-dp'");
+     }
+ 
+-  /* Currently, we don't support PIC code generation yet.  */
+-  if (flag_pic)
+-    sorry ("position-independent code not supported");
++  nds32_init_rtx_costs ();
+ 
+   nds32_register_passes ();
+ }
+@@ -3659,8 +4197,11 @@
+ 		     vec<const char *> &constraints ATTRIBUTE_UNUSED,
+ 		     vec<rtx> &clobbers, HARD_REG_SET &clobbered_regs)
+ {
+-  clobbers.safe_push (gen_rtx_REG (SImode, TA_REGNUM));
+-  SET_HARD_REG_BIT (clobbered_regs, TA_REGNUM);
++  if (!flag_inline_asm_r15)
++    {
++      clobbers.safe_push (gen_rtx_REG (SImode, TA_REGNUM));
++      SET_HARD_REG_BIT (clobbered_regs, TA_REGNUM);
++    }
+   return NULL;
+ }
+ 
+@@ -3687,6 +4228,13 @@
+   return nds32_expand_builtin_impl (exp, target, subtarget, mode, ignore);
+ }
+ 
++/* Implement TARGET_INIT_LIBFUNCS.  */
++static void
++nds32_init_libfuncs (void)
++{
++  if (TARGET_LINUX_ABI)
++    init_sync_libfuncs (UNITS_PER_WORD);
++}
+ 
+ /* ------------------------------------------------------------------------ */
+ 
+@@ -3703,6 +4251,16 @@
+   builtin_define ("__nds32__");
+   builtin_define ("__NDS32__");
+ 
++  /* We need to provide builtin macro to describe the size of
++     each vector for interrupt handler under elf toolchain.  */
++  if (!TARGET_LINUX_ABI)
++    {
++      if (TARGET_ISR_VECTOR_SIZE_4_BYTE)
++	builtin_define ("__NDS32_ISR_VECTOR_SIZE_4__");
++      else
++	builtin_define ("__NDS32_ISR_VECTOR_SIZE_16__");
++    }
++
+   if (TARGET_HARD_FLOAT)
+     builtin_define ("__NDS32_ABI_2FP_PLUS__");
+   else
+@@ -3770,6 +4328,8 @@
+     builtin_define ("__NDS32_GP_DIRECT__");
+   if (TARGET_VH)
+     builtin_define ("__NDS32_VH__");
++  if (NDS32_EXT_DSP_P ())
++    builtin_define ("__NDS32_EXT_DSP__");
+ 
+   if (TARGET_BIG_ENDIAN)
+     builtin_define ("__big_endian__");
+@@ -4042,6 +4602,10 @@
+      The result will be in cfun->machine.  */
+   nds32_compute_stack_frame ();
+ 
++  /* Check frame_pointer_needed again to prevent fp is need after reload.  */
++  if (frame_pointer_needed)
++    cfun->machine->fp_as_gp_p = false;
++
+   /* If this is a variadic function, first we need to push argument
+      registers that hold the unnamed argument value.  */
+   if (cfun->machine->va_args_size != 0)
+@@ -4066,7 +4630,7 @@
+ 
+   /* If the function is 'naked',
+      we do not have to generate prologue code fragment.  */
+-  if (cfun->machine->naked_p)
++  if (cfun->machine->naked_p && !flag_pic)
+     return;
+ 
+   /* Get callee_first_regno and callee_last_regno.  */
+@@ -4195,9 +4759,15 @@
+ 			       -1 * sp_adjust);
+     }
+ 
+-  /* Prevent the instruction scheduler from
+-     moving instructions across the boundary.  */
+-  emit_insn (gen_blockage ());
++  /* Emit gp setup instructions for -fpic.  */
++  if (flag_pic && df_regs_ever_live_p (PIC_OFFSET_TABLE_REGNUM))
++    nds32_emit_load_gp ();
++
++  /* If user applies -mno-sched-prolog-epilog option,
++     we need to prevent instructions of function body from being
++     scheduled with stack adjustment in prologue.  */
++  if (!flag_sched_prolog_epilog)
++    emit_insn (gen_blockage ());
+ }
+ 
+ /* Function for normal multiple pop epilogue.  */
+@@ -4211,9 +4781,11 @@
+      The result will be in cfun->machine.  */
+   nds32_compute_stack_frame ();
+ 
+-  /* Prevent the instruction scheduler from
+-     moving instructions across the boundary.  */
+-  emit_insn (gen_blockage ());
++  /* If user applies -mno-sched-prolog-epilog option,
++     we need to prevent instructions of function body from being
++     scheduled with stack adjustment in epilogue.  */
++  if (!flag_sched_prolog_epilog)
++    emit_insn (gen_blockage ());
+ 
+   /* If the function is 'naked', we do not have to generate
+      epilogue code fragment BUT 'ret' instruction.
+@@ -4239,7 +4811,16 @@
+       /* Generate return instruction by using 'return_internal' pattern.
+ 	 Make sure this instruction is after gen_blockage().  */
+       if (!sibcall_p)
+-	emit_jump_insn (gen_return_internal ());
++	{
++	  /* We need to further check attributes to determine whether
++	     there should be return instruction at epilogue.
++	     If the attribute naked exists but -mno-ret-in-naked-func
++	     is issued, there is NO need to generate return instruction.  */
++	  if (cfun->machine->attr_naked_p && !flag_ret_in_naked_func)
++	    return;
++
++	  emit_jump_insn (gen_return_internal ());
++	}
+       return;
+     }
+ 
+@@ -4436,9 +5017,13 @@
+   if (cfun->machine->callee_saved_gpr_regs_size > 0)
+     df_set_regs_ever_live (FP_REGNUM, 1);
+ 
++  /* Check frame_pointer_needed again to prevent fp is need after reload.  */
++  if (frame_pointer_needed)
++    cfun->machine->fp_as_gp_p = false;
++
+   /* If the function is 'naked',
+      we do not have to generate prologue code fragment.  */
+-  if (cfun->machine->naked_p)
++  if (cfun->machine->naked_p && !flag_pic)
+     return;
+ 
+   /* Get callee_first_regno and callee_last_regno.  */
+@@ -4566,6 +5151,10 @@
+ 			       -1 * sp_adjust);
+     }
+ 
++  /* Emit gp setup instructions for -fpic.  */
++  if (flag_pic && df_regs_ever_live_p (PIC_OFFSET_TABLE_REGNUM))
++    nds32_emit_load_gp ();
++
+   /* Prevent the instruction scheduler from
+      moving instructions across the boundary.  */
+   emit_insn (gen_blockage ());
+@@ -4591,9 +5180,19 @@
+   if (cfun->machine->naked_p)
+     {
+       /* Generate return instruction by using 'return_internal' pattern.
+-	 Make sure this instruction is after gen_blockage().  */
++	 Make sure this instruction is after gen_blockage().
++	 First we need to check this is a function without sibling call.  */
+       if (!sibcall_p)
+-	emit_jump_insn (gen_return_internal ());
++	{
++	  /* We need to further check attributes to determine whether
++	     there should be return instruction at epilogue.
++	     If the attribute naked exists but -mno-ret-in-naked-func
++	     is issued, there is NO need to generate return instruction.  */
++	  if (cfun->machine->attr_naked_p && !flag_ret_in_naked_func)
++	    return;
++
++	  emit_jump_insn (gen_return_internal ());
++	}
+       return;
+     }
+ 
+@@ -4757,6 +5356,11 @@
+   if (!reload_completed)
+     return 0;
+ 
++  /* If attribute 'naked' appears but -mno-ret-in-naked-func is used,
++     we cannot use return instruction.  */
++  if (cfun->machine->attr_naked_p && !flag_ret_in_naked_func)
++    return 0;
++
+   sp_adjust = cfun->machine->local_size
+ 	      + cfun->machine->out_args_size
+ 	      + cfun->machine->callee_saved_area_gpr_padding_bytes
+@@ -5010,6 +5614,9 @@
+ #undef TARGET_FUNCTION_ARG_BOUNDARY
+ #define TARGET_FUNCTION_ARG_BOUNDARY nds32_function_arg_boundary
+ 
++#undef TARGET_VECTOR_MODE_SUPPORTED_P
++#define TARGET_VECTOR_MODE_SUPPORTED_P nds32_vector_mode_supported_p
++
+ /* -- How Scalar Function Values Are Returned.  */
+ 
+ #undef TARGET_FUNCTION_VALUE
+@@ -5087,6 +5694,21 @@
+ #undef TARGET_LEGITIMATE_ADDRESS_P
+ #define TARGET_LEGITIMATE_ADDRESS_P nds32_legitimate_address_p
+ 
++#undef TARGET_LEGITIMIZE_ADDRESS
++#define TARGET_LEGITIMIZE_ADDRESS nds32_legitimize_address
++
++#undef TARGET_LEGITIMATE_CONSTANT_P
++#define TARGET_LEGITIMATE_CONSTANT_P nds32_legitimate_constant_p
++
++#undef TARGET_VECTORIZE_PREFERRED_SIMD_MODE
++#define TARGET_VECTORIZE_PREFERRED_SIMD_MODE nds32_vectorize_preferred_simd_mode
++
++#undef TARGET_CANNOT_FORCE_CONST_MEM
++#define TARGET_CANNOT_FORCE_CONST_MEM nds32_cannot_force_const_mem
++
++#undef TARGET_DELEGITIMIZE_ADDRESS
++#define TARGET_DELEGITIMIZE_ADDRESS nds32_delegitimize_address
++
+ 
+ /* Anchored Addresses.  */
+ 
+@@ -5147,6 +5769,9 @@
+ #undef TARGET_ASM_ALIGNED_SI_OP
+ #define TARGET_ASM_ALIGNED_SI_OP "\t.word\t"
+ 
++#undef TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA
++#define TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA nds32_asm_output_addr_const_extra
++
+ /* -- Output of Uninitialized Variables.  */
+ 
+ /* -- Output and Generation of Labels.  */
+@@ -5216,6 +5841,9 @@
+ 
+ /* Emulating TLS.  */
+ 
++#undef TARGET_HAVE_TLS
++#define TARGET_HAVE_TLS TARGET_LINUX_ABI
++
+ 
+ /* Defining coprocessor specifics for MIPS targets.  */
+ 
+@@ -5243,6 +5871,8 @@
+ #undef TARGET_EXPAND_BUILTIN
+ #define TARGET_EXPAND_BUILTIN nds32_expand_builtin
+ 
++#undef TARGET_INIT_LIBFUNCS
++#define TARGET_INIT_LIBFUNCS nds32_init_libfuncs
+ 
+ #undef TARGET_USE_BLOCKS_FOR_CONSTANT_P
+ #define TARGET_USE_BLOCKS_FOR_CONSTANT_P nds32_use_blocks_for_constant_p
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-cost.c gcc-8.2.0/gcc/config/nds32/nds32-cost.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-cost.c	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-cost.c	2019-01-25 15:38:32.821242637 +0100
+@@ -34,66 +34,379 @@
+ #include "optabs.h"		/* For GEN_FCN.  */
+ #include "recog.h"
+ #include "tm-constrs.h"
++#include "tree-pass.h"
+ 
+ /* ------------------------------------------------------------------------ */
+ 
+-bool
+-nds32_rtx_costs_impl (rtx x,
+-		      machine_mode mode ATTRIBUTE_UNUSED,
+-		      int outer_code,
+-		      int opno ATTRIBUTE_UNUSED,
+-		      int *total,
+-		      bool speed)
+-{
+-  int code = GET_CODE (x);
++typedef bool (*rtx_cost_func) (rtx, int, int, int, int*);
+ 
+-  /* According to 'speed', goto suitable cost model section.  */
+-  if (speed)
+-    goto performance_cost;
+-  else
+-    goto size_cost;
+-
+-
+-performance_cost:
+-  /* This is section for performance cost model.  */
++struct rtx_cost_model_t {
++  rtx_cost_func speed_prefer;
++  rtx_cost_func size_prefer;
++};
++
++static rtx_cost_model_t rtx_cost_model;
++
++static int insn_size_16bit; /* Initial at nds32_init_rtx_costs.  */
++static const int insn_size_32bit = 4;
++
++static bool
++nds32_rtx_costs_speed_prefer (rtx x ATTRIBUTE_UNUSED,
++			      int code,
++			      int outer_code ATTRIBUTE_UNUSED,
++			      int opno ATTRIBUTE_UNUSED,
++			      int *total)
++{
++  rtx op0;
++  rtx op1;
++  machine_mode mode = GET_MODE (x);
++  /* Scale cost by mode size.  */
++  int cost = COSTS_N_INSNS (GET_MODE_SIZE (mode) / GET_MODE_SIZE (SImode));
+ 
+-  /* In gcc/rtl.h, the default value of COSTS_N_INSNS(N) is N*4.
+-     We treat it as 4-cycle cost for each instruction
+-     under performance consideration.  */
+   switch (code)
+     {
+-    case SET:
+-      /* For 'SET' rtx, we need to return false
+-         so that it can recursively calculate costs.  */
+-      return false;
+-
+     case USE:
+       /* Used in combine.c as a marker.  */
+       *total = 0;
+-      break;
++      return true;
++
++    case CONST_INT:
++      /* When not optimizing for size, we care more about the cost
++	 of hot code, and hot code is often in a loop.  If a constant
++	 operand needs to be forced into a register, we will often be
++	 able to hoist the constant load out of the loop, so the load
++	 should not contribute to the cost.  */
++      if (outer_code == SET || outer_code == PLUS)
++	*total = satisfies_constraint_Is20 (x) ? 0 : 4;
++      else if (outer_code == AND || outer_code == IOR || outer_code == XOR
++	       || outer_code == MINUS)
++	*total = satisfies_constraint_Iu15 (x) ? 0 : 4;
++      else if (outer_code == ASHIFT || outer_code == ASHIFTRT
++	       || outer_code == LSHIFTRT)
++	*total = satisfies_constraint_Iu05 (x) ? 0 : 4;
++      else if (GET_RTX_CLASS (outer_code) == RTX_COMPARE
++	       || GET_RTX_CLASS (outer_code) == RTX_COMM_COMPARE)
++	*total = satisfies_constraint_Is16 (x) ? 0 : 4;
++      else
++	*total = COSTS_N_INSNS (1);
++      return true;
++
++    case CONST:
++    case LO_SUM:
++    case HIGH:
++    case SYMBOL_REF:
++      *total = COSTS_N_INSNS (1);
++      return true;
++
++    case MEM:
++      *total = COSTS_N_INSNS (1);
++      return true;
++
++    case SET:
++      op0 = SET_DEST (x);
++      op1 = SET_SRC (x);
++      mode = GET_MODE (op0);
++      /* Scale cost by mode size.  */
++      cost = COSTS_N_INSNS (GET_MODE_SIZE (mode) / GET_MODE_SIZE (SImode));
++
++      switch (GET_CODE (op1))
++	{
++	case REG:
++	case SUBREG:
++	  /* Register move and Store instructions.  */
++	  if ((REG_P (op0) || MEM_P (op0))
++	      && GET_MODE_SIZE (mode) <= GET_MODE_SIZE (DImode))
++	    *total = COSTS_N_INSNS (1);
++	  else
++	    *total = cost;
++	  return true;
++
++	case MEM:
++	  /* Load instructions.  */
++	  if (REG_P (op0) && GET_MODE_SIZE (mode) <= GET_MODE_SIZE (DImode))
++	    *total = COSTS_N_INSNS (1);
++	  else
++	    *total = cost;
++	  return true;
++
++	case CONST_INT:
++	  /* movi instruction.  */
++	  if (REG_P (op0) && GET_MODE_SIZE (mode) < GET_MODE_SIZE (DImode))
++	    {
++	      if (satisfies_constraint_Is20 (op1))
++		*total = COSTS_N_INSNS (1) - 1;
++	      else
++		*total = COSTS_N_INSNS (2);
++	    }
++	  else
++	    *total = cost;
++	  return true;
++
++	case CONST:
++	case SYMBOL_REF:
++	case LABEL_REF:
++	  /* la instruction.  */
++	  if (REG_P (op0) && GET_MODE_SIZE (mode) < GET_MODE_SIZE (DImode))
++	    *total = COSTS_N_INSNS (1) - 1;
++	  else
++	    *total = cost;
++	  return true;
++	case VEC_SELECT:
++	  *total = cost;
++	  return true;
++
++	default:
++	  *total = cost;
++	  return true;
++	}
++
++    case PLUS:
++      op0 = XEXP (x, 0);
++      op1 = XEXP (x, 1);
++
++      if (GET_MODE_SIZE (mode) >= GET_MODE_SIZE (DImode))
++	*total = cost;
++      else if (GET_CODE (op0) == MULT || GET_CODE (op0) == LSHIFTRT
++	       || GET_CODE (op1) == MULT || GET_CODE (op1) == LSHIFTRT)
++	/* ALU_SHIFT */
++	*total = COSTS_N_INSNS (2);
++
++      else if ((GET_CODE (op1) == CONST_INT
++		&& satisfies_constraint_Is15 (op1))
++		|| REG_P (op1))
++	/* ADD instructions */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* ADD instructions: IMM out of range.  */
++	*total = COSTS_N_INSNS (2);
++      return true;
++
++    case MINUS:
++      op0 = XEXP (x, 0);
++      op1 = XEXP (x, 1);
++
++      if (GET_MODE_SIZE (mode) >= GET_MODE_SIZE (DImode))
++	*total = cost;
++      else if (GET_CODE (op0) == MULT || GET_CODE (op0) == LSHIFTRT
++	       || GET_CODE (op1) == MULT || GET_CODE (op1) == LSHIFTRT)
++	/* ALU_SHIFT */
++	*total = COSTS_N_INSNS (2);
++      else if ((GET_CODE (op0) == CONST_INT
++		&& satisfies_constraint_Is15 (op0))
++		|| REG_P (op0))
++	/* SUB instructions */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* SUB instructions: IMM out of range.  */
++	*total = COSTS_N_INSNS (2);
++      return true;
++
++    case TRUNCATE:
++      /* TRUNCATE and AND behavior is same. */
++      *total = COSTS_N_INSNS (1);
++      return true;
++
++    case AND:
++    case IOR:
++    case XOR:
++      op0 = XEXP (x, 0);
++      op1 = XEXP (x, 1);
++
++      if (NDS32_EXT_DSP_P ())
++	{
++	  /* We prefer (and (ior) (ior)) than (ior (and) (and)) for
++	     synthetize pk** and insb instruction.  */
++	  if (code == AND && GET_CODE (op0) == IOR && GET_CODE (op1) == IOR)
++	    return COSTS_N_INSNS (1);
++
++	  if (code == IOR && GET_CODE (op0) == AND && GET_CODE (op1) == AND)
++	    return COSTS_N_INSNS (10);
++	}
++
++      if (GET_MODE_SIZE (mode) >= GET_MODE_SIZE (DImode))
++	*total = cost;
++      else if (GET_CODE (op0) == ASHIFT || GET_CODE (op0) == LSHIFTRT)
++	*total = COSTS_N_INSNS (2);
++      else if ((GET_CODE (op1) == CONST_INT
++	       && satisfies_constraint_Iu15 (op1))
++	       || REG_P (op1))
++	/* AND, OR, XOR instructions */
++	*total = COSTS_N_INSNS (1);
++      else if (code == AND || GET_CODE (op0) == NOT)
++	/* BITC instruction */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* AND, OR, XOR instructions: IMM out of range.  */
++	*total = COSTS_N_INSNS (2);
++      return true;
+ 
+     case MULT:
++      if (GET_MODE (x) == DImode
++	  || GET_CODE (XEXP (x, 1)) == SIGN_EXTEND
++	  || GET_CODE (XEXP (x, 1)) == ZERO_EXTEND)
++	/* MUL instructions */
++	*total = COSTS_N_INSNS (1);
++      else if (GET_MODE_SIZE (mode) >= GET_MODE_SIZE (DImode))
++	*total = cost;
++      else if (outer_code == PLUS || outer_code == MINUS)
++	*total = COSTS_N_INSNS (2);
++      else if ((GET_CODE (XEXP (x, 1)) == CONST_INT
++	       && satisfies_constraint_Iu05 (XEXP (x, 1)))
++	       || REG_P (XEXP (x, 1)))
++	/* MUL instructions */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* MUL instructions: IMM out of range.  */
++	*total = COSTS_N_INSNS (2);
++
++      if (TARGET_MUL_SLOW)
++	*total += COSTS_N_INSNS (4);
++
++      return true;
++
++    case LSHIFTRT:
++      if (GET_MODE_SIZE (mode) >= GET_MODE_SIZE (DImode))
++	*total = cost;
++      else if (outer_code == PLUS || outer_code == MINUS
++	       || outer_code == AND || outer_code == IOR
++	       || outer_code == XOR)
++	*total = COSTS_N_INSNS (2);
++      else if ((GET_CODE (XEXP (x, 1)) == CONST_INT
++	       && satisfies_constraint_Iu05 (XEXP (x, 1)))
++	       || REG_P (XEXP (x, 1)))
++	/* SRL instructions */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* SRL instructions: IMM out of range.  */
++	*total = COSTS_N_INSNS (2);
++      return true;
++
++    case ASHIFT:
++      if (GET_MODE_SIZE (mode) >= GET_MODE_SIZE (DImode))
++	*total = cost;
++      else if (outer_code == AND || outer_code == IOR
++	       || outer_code == XOR)
++	*total = COSTS_N_INSNS (2);
++      else if ((GET_CODE (XEXP (x, 1)) == CONST_INT
++	       && satisfies_constraint_Iu05 (XEXP (x, 1)))
++	       || REG_P (XEXP (x, 1)))
++	/* SLL instructions */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* SLL instructions: IMM out of range.  */
++	*total = COSTS_N_INSNS (2);
++      return true;
++
++    case ASHIFTRT:
++    case ROTATERT:
++      if (GET_MODE_SIZE (mode) >= GET_MODE_SIZE (DImode))
++	*total = cost;
++      else if ((GET_CODE (XEXP (x, 1)) == CONST_INT
++	       && satisfies_constraint_Iu05 (XEXP (x, 1)))
++	       || REG_P (XEXP (x, 1)))
++	/* ROTR, SLL instructions */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* ROTR, SLL instructions: IMM out of range.  */
++	*total = COSTS_N_INSNS (2);
++      return true;
++
++    case LT:
++    case LTU:
++      if (outer_code == SET)
++	{
++	  if ((GET_CODE (XEXP (x, 1)) == CONST_INT
++	      && satisfies_constraint_Iu15 (XEXP (x, 1)))
++	      || REG_P (XEXP (x, 1)))
++	    /* SLT, SLTI instructions */
++	    *total = COSTS_N_INSNS (1);
++	  else
++	    /* SLT, SLT instructions: IMM out of range.  */
++	    *total = COSTS_N_INSNS (2);
++	}
++      else
++	/* branch */
++	*total = COSTS_N_INSNS (2);
++      return true;
++
++    case EQ:
++    case NE:
++    case GE:
++    case LE:
++    case GT:
++      /* branch */
++      *total = COSTS_N_INSNS (2);
++      return true;
++
++    case IF_THEN_ELSE:
++      if (GET_CODE (XEXP (x, 1)) == LABEL_REF)
++	/* branch */
++	*total = COSTS_N_INSNS (2);
++      else
++	/* cmovz, cmovn instructions */
++	*total = COSTS_N_INSNS (1);
++      return true;
++
++    case LABEL_REF:
++      if (outer_code == IF_THEN_ELSE)
++	/* branch */
++	*total = COSTS_N_INSNS (2);
++      else
++	*total = COSTS_N_INSNS (1);
++      return true;
++
++    case ZERO_EXTEND:
++    case SIGN_EXTEND:
++      if (MEM_P (XEXP (x, 0)))
++	/* Using memory access. */
++	*total = COSTS_N_INSNS (1);
++      else
++	/* Zero extend and sign extend instructions.  */
++	*total = COSTS_N_INSNS (1);
++      return true;
++
++    case NEG:
++    case NOT:
+       *total = COSTS_N_INSNS (1);
+-      break;
++      return true;
+ 
+     case DIV:
+     case UDIV:
+     case MOD:
+     case UMOD:
+-      *total = COSTS_N_INSNS (7);
+-      break;
+-
+-    default:
+-      *total = COSTS_N_INSNS (1);
+-      break;
+-    }
++      *total = COSTS_N_INSNS (20);
++      return true;
+ 
+-  return true;
++    case CALL:
++      *total = COSTS_N_INSNS (2);
++      return true;
+ 
++    case CLZ:
++    case SMIN:
++    case SMAX:
++    case ZERO_EXTRACT:
++      if (TARGET_EXT_PERF)
++	*total = COSTS_N_INSNS (1);
++      else
++	*total = COSTS_N_INSNS (3);
++      return true;
++    case VEC_SELECT:
++      *total = COSTS_N_INSNS (1);
++      return true;
+ 
+-size_cost:
+-  /* This is section for size cost model.  */
++    default:
++      *total = COSTS_N_INSNS (3);
++      return true;
++    }
++}
+ 
++static bool
++nds32_rtx_costs_size_prefer (rtx x,
++			     int code,
++			     int outer_code,
++			     int opno ATTRIBUTE_UNUSED,
++			     int *total)
++{
+   /* In gcc/rtl.h, the default value of COSTS_N_INSNS(N) is N*4.
+      We treat it as 4-byte cost for each instruction
+      under code size consideration.  */
+@@ -118,85 +431,162 @@
+ 	     (set X imm20s), use movi, 4-byte cost.
+ 	     (set X BIG_INT), use sethi/ori, 8-byte cost.  */
+ 	  if (satisfies_constraint_Is05 (x))
+-	    *total = COSTS_N_INSNS (1) - 2;
++	    *total = insn_size_16bit;
+ 	  else if (satisfies_constraint_Is20 (x))
+-	    *total = COSTS_N_INSNS (1);
++	    *total = insn_size_32bit;
+ 	  else
+-	    *total = COSTS_N_INSNS (2);
++	    *total = insn_size_32bit * 2;
+ 	}
+       else if (outer_code == PLUS || outer_code == MINUS)
+ 	{
+ 	  /* Possible addi333/subi333 or subi45/addi45, 2-byte cost.
+ 	     General case, cost 1 instruction with 4-byte.  */
+ 	  if (satisfies_constraint_Iu05 (x))
+-	    *total = COSTS_N_INSNS (1) - 2;
++	    *total = insn_size_16bit;
+ 	  else
+-	    *total = COSTS_N_INSNS (1);
++	    *total = insn_size_32bit;
+ 	}
+       else if (outer_code == ASHIFT)
+ 	{
+ 	  /* Possible slli333, 2-byte cost.
+ 	     General case, cost 1 instruction with 4-byte.  */
+ 	  if (satisfies_constraint_Iu03 (x))
+-	    *total = COSTS_N_INSNS (1) - 2;
++	    *total = insn_size_16bit;
+ 	  else
+-	    *total = COSTS_N_INSNS (1);
++	    *total = insn_size_32bit;
+ 	}
+       else if (outer_code == ASHIFTRT || outer_code == LSHIFTRT)
+ 	{
+ 	  /* Possible srai45 or srli45, 2-byte cost.
+ 	     General case, cost 1 instruction with 4-byte.  */
+ 	  if (satisfies_constraint_Iu05 (x))
+-	    *total = COSTS_N_INSNS (1) - 2;
++	    *total = insn_size_16bit;
+ 	  else
+-	    *total = COSTS_N_INSNS (1);
++	    *total = insn_size_32bit;
+ 	}
+       else
+ 	{
+ 	  /* For other cases, simply set it 4-byte cost.  */
+-	  *total = COSTS_N_INSNS (1);
++	  *total = insn_size_32bit;
+ 	}
+       break;
+ 
+     case CONST_DOUBLE:
+       /* It requires high part and low part processing, set it 8-byte cost.  */
+-      *total = COSTS_N_INSNS (2);
++      *total = insn_size_32bit * 2;
++      break;
++
++    case CONST:
++    case SYMBOL_REF:
++      *total = insn_size_32bit * 2;
+       break;
+ 
+     default:
+       /* For other cases, generally we set it 4-byte cost
+-         and stop resurively traversing.  */
+-      *total = COSTS_N_INSNS (1);
++	 and stop resurively traversing.  */
++      *total = insn_size_32bit;
+       break;
+     }
+ 
+   return true;
+ }
+ 
+-int
+-nds32_address_cost_impl (rtx address,
+-			 machine_mode mode ATTRIBUTE_UNUSED,
+-			 addr_space_t as ATTRIBUTE_UNUSED,
+-			 bool speed)
++void
++nds32_init_rtx_costs (void)
++{
++  rtx_cost_model.speed_prefer = nds32_rtx_costs_speed_prefer;
++  rtx_cost_model.size_prefer  = nds32_rtx_costs_size_prefer;
++
++  if (TARGET_16_BIT)
++    insn_size_16bit = 2;
++  else
++    insn_size_16bit = 4;
++}
++
++/* This target hook describes the relative costs of RTL expressions.
++   Return 'true' when all subexpressions of x have been processed.
++   Return 'false' to sum the costs of sub-rtx, plus cost of this operation.
++   Refer to gcc/rtlanal.c for more information.  */
++bool
++nds32_rtx_costs_impl (rtx x,
++		      machine_mode mode ATTRIBUTE_UNUSED,
++		      int outer_code,
++		      int opno,
++		      int *total,
++		      bool speed)
++{
++  int code = GET_CODE (x);
++
++  /* According to 'speed', use suitable cost model section.  */
++  if (speed)
++    return rtx_cost_model.speed_prefer(x, code, outer_code, opno, total);
++  else
++    return rtx_cost_model.size_prefer(x, code, outer_code, opno, total);
++}
++
++
++int nds32_address_cost_speed_prefer (rtx address)
+ {
+   rtx plus0, plus1;
+   enum rtx_code code;
+ 
+   code = GET_CODE (address);
+ 
+-  /* According to 'speed', goto suitable cost model section.  */
+-  if (speed)
+-    goto performance_cost;
+-  else
+-    goto size_cost;
++  switch (code)
++    {
++    case POST_MODIFY:
++    case POST_INC:
++    case POST_DEC:
++      /* We encourage that rtx contains
++	 POST_MODIFY/POST_INC/POST_DEC behavior.  */
++      return COSTS_N_INSNS (1) - 2;
++
++    case SYMBOL_REF:
++      /* We can have gp-relative load/store for symbol_ref.
++	Have it 4-byte cost.  */
++      return COSTS_N_INSNS (2);
++
++    case CONST:
++      /* It is supposed to be the pattern (const (plus symbol_ref const_int)).
++	 Have it 4-byte cost.  */
++      return COSTS_N_INSNS (2);
++
++    case REG:
++      /* Simply return 4-byte costs.  */
++      return COSTS_N_INSNS (1) - 2;
++
++    case PLUS:
++      /* We do not need to check if the address is a legitimate address,
++	 because this hook is never called with an invalid address.
++	 But we better check the range of
++	 const_int value for cost, if it exists.  */
++      plus0 = XEXP (address, 0);
++      plus1 = XEXP (address, 1);
+ 
+-performance_cost:
+-  /* This is section for performance cost model.  */
++      if (REG_P (plus0) && CONST_INT_P (plus1))
++	return COSTS_N_INSNS (1) - 2;
++      else if (ARITHMETIC_P (plus0) || ARITHMETIC_P (plus1))
++	return COSTS_N_INSNS (1) - 1;
++      else if (REG_P (plus0) && REG_P (plus1))
++	return COSTS_N_INSNS (1);
+ 
+-  /* FALLTHRU, currently we use same cost model as size_cost.  */
++      /* For other 'plus' situation, make it cost 4-byte.  */
++      return COSTS_N_INSNS (1);
+ 
+-size_cost:
+-  /* This is section for size cost model.  */
++    default:
++      break;
++    }
++
++  return COSTS_N_INSNS (4);
++
++}
++
++int nds32_address_cost_speed_fwprop (rtx address)
++{
++  rtx plus0, plus1;
++  enum rtx_code code;
++
++  code = GET_CODE (address);
+ 
+   switch (code)
+     {
+@@ -210,12 +600,12 @@
+     case SYMBOL_REF:
+       /* We can have gp-relative load/store for symbol_ref.
+ 	 Have it 4-byte cost.  */
+-      return COSTS_N_INSNS (1);
++      return COSTS_N_INSNS (2);
+ 
+     case CONST:
+       /* It is supposed to be the pattern (const (plus symbol_ref const_int)).
+ 	 Have it 4-byte cost.  */
+-      return COSTS_N_INSNS (1);
++      return COSTS_N_INSNS (2);
+ 
+     case REG:
+       /* Simply return 4-byte costs.  */
+@@ -233,11 +623,78 @@
+ 	{
+ 	  /* If it is possible to be lwi333/swi333 form,
+ 	     make it 2-byte cost.  */
+-	  if (satisfies_constraint_Iu05 (plus1))
++	  if (satisfies_constraint_Iu03 (plus1))
+ 	    return (COSTS_N_INSNS (1) - 2);
+ 	  else
+ 	    return COSTS_N_INSNS (1);
+ 	}
++      if (ARITHMETIC_P (plus0) || ARITHMETIC_P (plus1))
++	return COSTS_N_INSNS (1) - 2;
++      else if (REG_P (plus0) && REG_P (plus1))
++	return COSTS_N_INSNS (1);
++
++      /* For other 'plus' situation, make it cost 4-byte.  */
++      return COSTS_N_INSNS (1);
++
++    default:
++      break;
++    }
++
++  return COSTS_N_INSNS (4);
++}
++
++
++int nds32_address_cost_size_prefer (rtx address)
++{
++  rtx plus0, plus1;
++  enum rtx_code code;
++
++  code = GET_CODE (address);
++
++  switch (code)
++    {
++    case POST_MODIFY:
++    case POST_INC:
++    case POST_DEC:
++      /* We encourage that rtx contains
++	 POST_MODIFY/POST_INC/POST_DEC behavior.  */
++      return 0;
++
++    case SYMBOL_REF:
++      /* We can have gp-relative load/store for symbol_ref.
++	 Have it 4-byte cost.  */
++      return COSTS_N_INSNS (2);
++
++    case CONST:
++      /* It is supposed to be the pattern (const (plus symbol_ref const_int)).
++	 Have it 4-byte cost.  */
++      return COSTS_N_INSNS (2);
++
++    case REG:
++      /* Simply return 4-byte costs.  */
++      return COSTS_N_INSNS (1) - 1;
++
++    case PLUS:
++      /* We do not need to check if the address is a legitimate address,
++	 because this hook is never called with an invalid address.
++	 But we better check the range of
++	 const_int value for cost, if it exists.  */
++      plus0 = XEXP (address, 0);
++      plus1 = XEXP (address, 1);
++
++      if (REG_P (plus0) && CONST_INT_P (plus1))
++	{
++	  /* If it is possible to be lwi333/swi333 form,
++	     make it 2-byte cost.  */
++	  if (satisfies_constraint_Iu03 (plus1))
++	    return (COSTS_N_INSNS (1) - 2);
++	  else
++	    return COSTS_N_INSNS (1) - 1;
++	}
++
++      /* (plus (reg) (mult (reg) (const))) */
++      if (ARITHMETIC_P (plus0) || ARITHMETIC_P (plus1))
++	return (COSTS_N_INSNS (1) - 1);
+ 
+       /* For other 'plus' situation, make it cost 4-byte.  */
+       return COSTS_N_INSNS (1);
+@@ -247,6 +704,23 @@
+     }
+ 
+   return COSTS_N_INSNS (4);
++
++}
++
++int nds32_address_cost_impl (rtx address,
++			     machine_mode mode ATTRIBUTE_UNUSED,
++			     addr_space_t as ATTRIBUTE_UNUSED,
++			     bool speed_p)
++{
++  if (speed_p)
++    {
++      if (current_pass->tv_id == TV_FWPROP)
++	return nds32_address_cost_speed_fwprop (address);
++      else
++	return nds32_address_cost_speed_prefer (address);
++    }
++  else
++    return nds32_address_cost_size_prefer (address);
+ }
+ 
+ /* ------------------------------------------------------------------------ */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-doubleword.md gcc-8.2.0/gcc/config/nds32/nds32-doubleword.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-doubleword.md	2018-05-07 04:09:58.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-doubleword.md	2019-01-25 15:38:32.821242637 +0100
+@@ -136,10 +136,28 @@
+   }
+ )
+ 
++;; Split move_di pattern when the hard register is odd.
++(define_split
++  [(set (match_operand:DIDF 0 "register_operand" "")
++	(match_operand:DIDF 1 "register_operand" ""))]
++  "(NDS32_IS_GPR_REGNUM (REGNO (operands[0]))
++    && ((REGNO (operands[0]) & 0x1) == 1))
++   || (NDS32_IS_GPR_REGNUM (REGNO (operands[1]))
++       && ((REGNO (operands[1]) & 0x1) == 1))"
++  [(set (match_dup 2) (match_dup 3))
++   (set (match_dup 4) (match_dup 5))]
++  {
++     operands[2] = gen_lowpart (SImode, operands[0]);
++     operands[4] = gen_highpart (SImode, operands[0]);
++     operands[3] = gen_lowpart (SImode, operands[1]);
++     operands[5] = gen_highpart (SImode, operands[1]);
++  }
++)
++
+ (define_split
+   [(set (match_operand:DIDF 0 "register_operand"     "")
+ 	(match_operand:DIDF 1 "const_double_operand" ""))]
+-  "reload_completed"
++  "flag_pic || reload_completed"
+   [(set (match_dup 2) (match_dup 3))
+    (set (match_dup 4) (match_dup 5))]
+ {
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-dspext.md gcc-8.2.0/gcc/config/nds32/nds32-dspext.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-dspext.md	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-dspext.md	2019-01-25 15:38:32.825242648 +0100
+@@ -0,0 +1,5278 @@
++;; Machine description of Andes NDS32 cpu for GNU compiler
++;; Copyright (C) 2012-2018 Free Software Foundation, Inc.
++;; Contributed by Andes Technology Corporation.
++;;
++;; This file is part of GCC.
++;;
++;; GCC is free software; you can redistribute it and/or modify it
++;; under the terms of the GNU General Public License as published
++;; by the Free Software Foundation; either version 3, or (at your
++;; option) any later version.
++;;
++;; GCC is distributed in the hope that it will be useful, but WITHOUT
++;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++;; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++;; License for more details.
++;;
++;; You should have received a copy of the GNU General Public License
++;; along with GCC; see the file COPYING3.  If not see
++;; <http://www.gnu.org/licenses/>.
++
++(define_expand "mov<mode>"
++  [(set (match_operand:VQIHI 0 "general_operand" "")
++	(match_operand:VQIHI 1 "general_operand" ""))]
++  "NDS32_EXT_DSP_P ()"
++{
++  /* Need to force register if mem <- !reg.  */
++  if (MEM_P (operands[0]) && !REG_P (operands[1]))
++    operands[1] = force_reg (<MODE>mode, operands[1]);
++
++  /* If operands[1] is a large constant and cannot be performed
++     by a single instruction, we need to split it.  */
++  if (GET_CODE (operands[1]) == CONST_VECTOR
++      && !satisfies_constraint_CVs2 (operands[1])
++      && !satisfies_constraint_CVhi (operands[1]))
++    {
++      HOST_WIDE_INT ival = const_vector_to_hwint (operands[1]);
++      rtx tmp_rtx;
++
++      tmp_rtx = can_create_pseudo_p ()
++		? gen_reg_rtx (SImode)
++		: simplify_gen_subreg (SImode, operands[0], <MODE>mode, 0);
++
++      emit_move_insn (tmp_rtx, gen_int_mode (ival, SImode));
++      convert_move (operands[0], tmp_rtx, false);
++      DONE;
++    }
++
++  if (REG_P (operands[0]) && SYMBOLIC_CONST_P (operands[1]))
++    {
++      if (nds32_tls_referenced_p (operands [1]))
++	{
++	  nds32_expand_tls_move (operands);
++	  DONE;
++	}
++      else if (flag_pic)
++	{
++	  nds32_expand_pic_move (operands);
++	  DONE;
++	}
++    }
++})
++
++(define_insn "*mov<mode>"
++  [(set (match_operand:VQIHI 0 "nonimmediate_operand" "=r, r,$U45,$U33,$U37,$U45, m,$  l,$  l,$  l,$  d,  d, r,$   d,    r,    r,    r, *f, *f,  r, *f,  Q")
++	(match_operand:VQIHI 1 "nds32_vmove_operand"  " r, r,   l,   l,   l,   d, r, U45, U33, U37, U45,Ufe, m, CVp5, CVs5, CVs2, CVhi, *f,  r, *f,  Q, *f"))]
++  "NDS32_EXT_DSP_P ()
++   && (register_operand(operands[0], <MODE>mode)
++       || register_operand(operands[1], <MODE>mode))"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "mov55\t%0, %1";
++    case 1:
++      return "ori\t%0, %1, 0";
++    case 2:
++    case 3:
++    case 4:
++    case 5:
++      return nds32_output_16bit_store (operands, <byte>);
++    case 6:
++      return nds32_output_32bit_store (operands, <byte>);
++    case 7:
++    case 8:
++    case 9:
++    case 10:
++    case 11:
++      return nds32_output_16bit_load (operands, <byte>);
++    case 12:
++      return nds32_output_32bit_load (operands, <byte>);
++    case 13:
++      return "movpi45\t%0, %1";
++    case 14:
++      return "movi55\t%0, %1";
++    case 15:
++      return "movi\t%0, %1";
++    case 16:
++      return "sethi\t%0, hi20(%1)";
++    case 17:
++      if (TARGET_FPU_SINGLE)
++	return "fcpyss\t%0, %1, %1";
++      else
++	return "#";
++    case 18:
++      return "fmtsr\t%1, %0";
++    case 19:
++      return "fmfsr\t%0, %1";
++    case 20:
++      return nds32_output_float_load (operands);
++    case 21:
++      return nds32_output_float_store (operands);
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"    "alu,alu,store,store,store,store,store,load,load,load,load,load,load,alu,alu,alu,alu,fcpy,fmtsr,fmfsr,fload,fstore")
++   (set_attr "length"  "  2,  4,    2,    2,    2,    2,    4,   2,   2,   2,   2,   2,   4,  2,  2,  4,  4,   4,    4,    4,    4,     4")
++   (set_attr "feature" " v1, v1,   v1,   v1,   v1,   v1,   v1,  v1,  v1,  v1,  v1, v3m,  v1, v1, v1, v1, v1, fpu,  fpu,  fpu,  fpu,   fpu")])
++
++(define_expand "movv2si"
++  [(set (match_operand:V2SI 0 "general_operand" "")
++	(match_operand:V2SI 1 "general_operand" ""))]
++  "NDS32_EXT_DSP_P ()"
++{
++  /* Need to force register if mem <- !reg.  */
++  if (MEM_P (operands[0]) && !REG_P (operands[1]))
++    operands[1] = force_reg (V2SImode, operands[1]);
++})
++
++(define_insn "*movv2si"
++  [(set (match_operand:V2SI 0 "nonimmediate_operand" "=r, r,  r, r, Da, m, f, Q, f, r, f")
++	(match_operand:V2SI 1 "general_operand"      " r, i, Da, m,  r, r, Q, f, f, f, r"))]
++  "NDS32_EXT_DSP_P ()
++   && (register_operand(operands[0], V2SImode)
++       || register_operand(operands[1], V2SImode))"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "movd44\t%0, %1";
++    case 1:
++      /* reg <- const_int, we ask gcc to split instruction.  */
++      return "#";
++    case 2:
++      /* The memory format is (mem (reg)),
++	 we can generate 'lmw.bi' instruction.  */
++      return nds32_output_double (operands, true);
++    case 3:
++      /* We haven't 64-bit load instruction,
++	 we split this pattern to two SImode pattern.  */
++      return "#";
++    case 4:
++      /* The memory format is (mem (reg)),
++	 we can generate 'smw.bi' instruction.  */
++      return nds32_output_double (operands, false);
++    case 5:
++      /* We haven't 64-bit store instruction,
++	 we split this pattern to two SImode pattern.  */
++      return "#";
++    case 6:
++      return nds32_output_float_load (operands);
++    case 7:
++      return nds32_output_float_store (operands);
++    case 8:
++      return "fcpysd\t%0, %1, %1";
++    case 9:
++      return "fmfdr\t%0, %1";
++    case 10:
++      return "fmtdr\t%1, %0";
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"    "alu,alu,load,load,store,store,unknown,unknown,unknown,unknown,unknown")
++   (set_attr_alternative "length"
++     [
++       ;; Alternative 0
++       (if_then_else (match_test "!TARGET_16_BIT")
++		     (const_int 4)
++		     (const_int 2))
++       ;; Alternative 1
++       (const_int 16)
++       ;; Alternative 2
++       (const_int 4)
++       ;; Alternative 3
++       (const_int 8)
++       ;; Alternative 4
++       (const_int 4)
++       ;; Alternative 5
++       (const_int 8)
++       ;; Alternative 6
++       (const_int 4)
++       ;; Alternative 7
++       (const_int 4)
++       ;; Alternative 8
++       (const_int 4)
++       ;; Alternative 9
++       (const_int 4)
++       ;; Alternative 10
++       (const_int 4)
++     ])
++   (set_attr "feature" " v1, v1,  v1,  v1,   v1,   v1,    fpu,    fpu,    fpu,    fpu,    fpu")])
++
++(define_expand "movmisalign<mode>"
++  [(set (match_operand:VQIHI 0 "general_operand" "")
++	(match_operand:VQIHI 1 "general_operand" ""))]
++  "NDS32_EXT_DSP_P ()"
++{
++  rtx addr;
++  if (MEM_P (operands[0]) && !REG_P (operands[1]))
++    operands[1] = force_reg (<MODE>mode, operands[1]);
++
++  if (MEM_P (operands[0]))
++    {
++      addr = force_reg (Pmode, XEXP (operands[0], 0));
++      emit_insn (gen_unaligned_store<mode> (addr, operands[1]));
++    }
++  else
++    {
++      addr = force_reg (Pmode, XEXP (operands[1], 0));
++      emit_insn (gen_unaligned_load<mode> (operands[0], addr));
++    }
++  DONE;
++})
++
++(define_expand "unaligned_load<mode>"
++  [(set (match_operand:VQIHI 0 "register_operand" "=r")
++	(unspec:VQIHI [(mem:VQIHI (match_operand:SI 1 "register_operand" "r"))] UNSPEC_UALOAD_W))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_ISA_V3M)
++    nds32_expand_unaligned_load (operands, <MODE>mode);
++  else
++    emit_insn (gen_unaligned_load_w<mode> (operands[0], gen_rtx_MEM (<MODE>mode, operands[1])));
++  DONE;
++})
++
++(define_insn "unaligned_load_w<mode>"
++  [(set (match_operand:VQIHI 0 "register_operand"                          "=  r")
++	(unspec:VQIHI [(match_operand:VQIHI 1 "nds32_lmw_smw_base_operand" " Umw")] UNSPEC_UALOAD_W))]
++  "NDS32_EXT_DSP_P ()"
++{
++  return nds32_output_lmw_single_word (operands);
++}
++  [(set_attr "type"   "load")
++   (set_attr "length"    "4")]
++)
++
++(define_expand "unaligned_store<mode>"
++  [(set (mem:VQIHI (match_operand:SI 0 "register_operand" "r"))
++	(unspec:VQIHI [(match_operand:VQIHI 1 "register_operand" "r")] UNSPEC_UASTORE_W))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_ISA_V3M)
++    nds32_expand_unaligned_store (operands, <MODE>mode);
++  else
++    emit_insn (gen_unaligned_store_w<mode> (gen_rtx_MEM (<MODE>mode, operands[0]), operands[1]));
++  DONE;
++})
++
++(define_insn "unaligned_store_w<mode>"
++  [(set (match_operand:VQIHI 0 "nds32_lmw_smw_base_operand"      "=Umw")
++	(unspec:VQIHI [(match_operand:VQIHI 1 "register_operand" "   r")] UNSPEC_UASTORE_W))]
++  "NDS32_EXT_DSP_P ()"
++{
++  return nds32_output_smw_single_word (operands);
++}
++  [(set_attr "type"   "store")
++   (set_attr "length"     "4")]
++)
++
++(define_insn "<uk>add<mode>3"
++  [(set (match_operand:VQIHI 0 "register_operand"                 "=r")
++	(all_plus:VQIHI (match_operand:VQIHI 1 "register_operand" " r")
++			(match_operand:VQIHI 2 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "<uk>add<bits> %0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "<uk>adddi3"
++  [(set (match_operand:DI 0 "register_operand"              "=r")
++	(all_plus:DI (match_operand:DI 1 "register_operand" " r")
++		     (match_operand:DI 2 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "<uk>add64 %0, %1, %2"
++  [(set_attr "type"    "dalu64")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "raddv4qi3"
++  [(set (match_operand:V4QI 0 "register_operand"                  "=r")
++	(truncate:V4QI
++	  (ashiftrt:V4HI
++	    (plus:V4HI (sign_extend:V4HI (match_operand:V4QI 1 "register_operand" " r"))
++		       (sign_extend:V4HI (match_operand:V4QI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "radd8\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++
++(define_insn "uraddv4qi3"
++  [(set (match_operand:V4QI 0 "register_operand"                  "=r")
++	(truncate:V4QI
++	  (lshiftrt:V4HI
++	    (plus:V4HI (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" " r"))
++		       (zero_extend:V4HI (match_operand:V4QI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "uradd8\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "raddv2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                                  "=r")
++	(truncate:V2HI
++	  (ashiftrt:V2SI
++	    (plus:V2SI (sign_extend:V2SI (match_operand:V2HI 1 "register_operand" " r"))
++		       (sign_extend:V2SI (match_operand:V2HI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "radd16\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "uraddv2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                                  "=r")
++	(truncate:V2HI
++	  (lshiftrt:V2SI
++	    (plus:V2SI (zero_extend:V2SI (match_operand:V2HI 1 "register_operand" " r"))
++		       (zero_extend:V2SI (match_operand:V2HI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "uradd16\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "radddi3"
++  [(set (match_operand:DI 0 "register_operand"            "=r")
++	(truncate:DI
++	  (ashiftrt:TI
++	    (plus:TI (sign_extend:TI (match_operand:DI 1 "register_operand" " r"))
++		     (sign_extend:TI (match_operand:DI 2 "register_operand" " r")))
++	  (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "radd64\t%0, %1, %2"
++  [(set_attr "type"    "dalu64")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++
++(define_insn "uradddi3"
++  [(set (match_operand:DI 0 "register_operand"            "=r")
++	(truncate:DI
++	  (lshiftrt:TI
++	    (plus:TI (zero_extend:TI (match_operand:DI 1 "register_operand" " r"))
++		     (zero_extend:TI (match_operand:DI 2 "register_operand" " r")))
++	  (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "uradd64\t%0, %1, %2"
++  [(set_attr "type"    "dalu64")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "<uk>sub<mode>3"
++  [(set (match_operand:VQIHI 0 "register_operand"                  "=r")
++	(all_minus:VQIHI (match_operand:VQIHI 1 "register_operand" " r")
++			 (match_operand:VQIHI 2 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "<uk>sub<bits> %0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "<uk>subdi3"
++  [(set (match_operand:DI 0 "register_operand"               "=r")
++	(all_minus:DI (match_operand:DI 1 "register_operand" " r")
++		      (match_operand:DI 2 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "<uk>sub64 %0, %1, %2"
++  [(set_attr "type"    "dalu64")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++(define_insn "rsubv4qi3"
++  [(set (match_operand:V4QI 0 "register_operand"                                   "=r")
++	(truncate:V4QI
++	  (ashiftrt:V4HI
++	    (minus:V4HI (sign_extend:V4HI (match_operand:V4QI 1 "register_operand" " r"))
++			(sign_extend:V4HI (match_operand:V4QI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "rsub8\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")])
++
++(define_insn "ursubv4qi3"
++  [(set (match_operand:V4QI 0 "register_operand"                                   "=r")
++	(truncate:V4QI
++	  (lshiftrt:V4HI
++	    (minus:V4HI (zero_extend:V4HI (match_operand:V4QI 1 "register_operand" " r"))
++			(zero_extend:V4HI (match_operand:V4QI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "ursub8\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")])
++
++(define_insn "rsubv2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                                   "=r")
++	(truncate:V2HI
++	  (ashiftrt:V2SI
++	    (minus:V2SI (sign_extend:V2SI (match_operand:V2HI 1 "register_operand" " r"))
++			(sign_extend:V2SI (match_operand:V2HI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "rsub16\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")])
++
++(define_insn "ursubv2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                                   "=r")
++	(truncate:V2HI
++	  (lshiftrt:V2SI
++	    (minus:V2SI (zero_extend:V2SI (match_operand:V2HI 1 "register_operand" " r"))
++			(zero_extend:V2SI (match_operand:V2HI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "ursub16\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")])
++
++(define_insn "rsubdi3"
++  [(set (match_operand:DI 0 "register_operand"                   "=r")
++	(truncate:DI
++	  (ashiftrt:TI
++	    (minus:TI (sign_extend:TI (match_operand:DI 1 "register_operand" " r"))
++		      (sign_extend:TI (match_operand:DI 2 "register_operand" " r")))
++	  (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "rsub64\t%0, %1, %2"
++  [(set_attr "type"    "dalu64")
++   (set_attr "length"  "4")])
++
++
++(define_insn "ursubdi3"
++  [(set (match_operand:DI 0 "register_operand"                   "=r")
++	(truncate:DI
++	  (lshiftrt:TI
++	    (minus:TI (zero_extend:TI (match_operand:DI 1 "register_operand" " r"))
++		      (zero_extend:TI (match_operand:DI 2 "register_operand" " r")))
++	  (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "ursub64\t%0, %1, %2"
++  [(set_attr "type"    "dalu64")
++   (set_attr "length"  "4")])
++
++(define_expand "cras16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_cras16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_cras16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "cras16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (plus:HI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 1)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "cras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "cras16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (plus:HI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "cras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "kcras16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kcras16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_kcras16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "kcras16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (ss_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (ss_plus:HI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 1)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "kcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "kcras16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (ss_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (ss_plus:HI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "kcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "ukcras16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_ukcras16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_ukcras16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "ukcras16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (us_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (us_plus:HI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 1)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "ukcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "ukcras16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (us_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (us_plus:HI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "ukcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "crsa16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_crsa16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_crsa16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "crsa16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (plus:HI
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "crsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "crsa16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (plus:HI
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "crsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "kcrsa16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kcrsa16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_kcrsa16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "kcrsa16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (ss_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (ss_plus:HI
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "kcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "kcrsa16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (ss_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (ss_plus:HI
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "kcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "ukcrsa16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_ukcrsa16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_ukcrsa16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "ukcrsa16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (us_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (us_plus:HI
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "ukcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "ukcrsa16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (us_minus:HI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand" " r")
++		(parallel [(const_int 0)]))
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (us_plus:HI
++	      (vec_select:HI
++		(match_dup 1)
++		(parallel [(const_int 1)]))
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "ukcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "rcras16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_rcras16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_rcras16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "rcras16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++		(minus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 0)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++		(plus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 0)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "rcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "rcras16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++		(minus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 1)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++		(plus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 1)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "rcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "urcras16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_urcras16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_urcras16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "urcras16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++		(minus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 0)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++		(plus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 0)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "urcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "urcras16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++		(minus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 1)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++		(plus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 1)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "urcras16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "rcrsa16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_rcrsa16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_rcrsa16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "rcrsa16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++	        (minus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 1)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++		(plus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 0)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "rcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "rcrsa16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++	        (minus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 0)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (ashiftrt:SI
++		(plus:SI
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 1)])))
++		  (sign_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "rcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "urcrsa16_1"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_urcrsa16_1_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_urcrsa16_1_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_insn "urcrsa16_1_le"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++	        (minus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 1)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++		(plus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 0)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "urcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_insn "urcrsa16_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"           "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++	        (minus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 1 "register_operand" " r")
++		      (parallel [(const_int 0)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand" " r")
++		      (parallel [(const_int 1)]))))
++		(const_int 1))))
++	  (vec_duplicate:V2HI
++	    (truncate:HI
++	      (lshiftrt:SI
++		(plus:SI
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 1)
++		      (parallel [(const_int 1)])))
++		  (zero_extend:SI
++		    (vec_select:HI
++		      (match_dup 2)
++		      (parallel [(const_int 0)]))))
++		(const_int 1))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "urcrsa16\t%0, %1, %2"
++  [(set_attr "type" "dalu")]
++)
++
++(define_expand "<shift>v2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                  "")
++	(shifts:V2HI (match_operand:V2HI 1 "register_operand"     "")
++		     (match_operand:SI   2 "nds32_rimm4u_operand" "")))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (operands[2] == const0_rtx)
++    {
++      emit_move_insn (operands[0], operands[1]);
++      DONE;
++    }
++})
++
++(define_insn "*ashlv2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                "=   r, r")
++	(ashift:V2HI (match_operand:V2HI 1 "register_operand"   "    r, r")
++		     (match_operand:SI 2 "nds32_rimm4u_operand" " Iu04, r")))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   slli16\t%0, %1, %2
++   sll16\t%0, %1, %2"
++  [(set_attr "type"   "dalu,dalu")
++   (set_attr "length" "   4,   4")])
++
++(define_insn "kslli16"
++  [(set (match_operand:V2HI 0 "register_operand"                   "=   r, r")
++	(ss_ashift:V2HI (match_operand:V2HI 1 "register_operand"   "    r, r")
++			(match_operand:SI 2 "nds32_rimm4u_operand" " Iu04, r")))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   kslli16\t%0, %1, %2
++   ksll16\t%0, %1, %2"
++  [(set_attr "type"   "dalu,dalu")
++   (set_attr "length" "   4,   4")])
++
++(define_insn "*ashrv2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                  "=   r, r")
++	(ashiftrt:V2HI (match_operand:V2HI 1 "register_operand"   "    r, r")
++		       (match_operand:SI 2 "nds32_rimm4u_operand" " Iu04, r")))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   srai16\t%0, %1, %2
++   sra16\t%0, %1, %2"
++  [(set_attr "type"   "dalu,dalu")
++   (set_attr "length" "   4,   4")])
++
++(define_insn "sra16_round"
++  [(set (match_operand:V2HI 0 "register_operand"                                "=   r, r")
++	(unspec:V2HI [(ashiftrt:V2HI (match_operand:V2HI 1 "register_operand"   "    r, r")
++				     (match_operand:SI 2 "nds32_rimm4u_operand" " Iu04, r"))]
++		     UNSPEC_ROUND))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   srai16.u\t%0, %1, %2
++   sra16.u\t%0, %1, %2"
++  [(set_attr "type"   "daluround,daluround")
++   (set_attr "length" "         4,       4")])
++
++(define_insn "*lshrv2hi3"
++  [(set (match_operand:V2HI 0 "register_operand"                  "=   r, r")
++	(lshiftrt:V2HI (match_operand:V2HI 1 "register_operand"   "    r, r")
++		       (match_operand:SI 2 "nds32_rimm4u_operand" " Iu04, r")))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   srli16\t%0, %1, %2
++   srl16\t%0, %1, %2"
++  [(set_attr "type"   "dalu,dalu")
++   (set_attr "length" "   4,   4")])
++
++(define_insn "srl16_round"
++  [(set (match_operand:V2HI 0 "register_operand"                                "=   r, r")
++	(unspec:V2HI [(lshiftrt:V2HI (match_operand:V2HI 1 "register_operand"   "    r, r")
++				     (match_operand:SI 2 "nds32_rimm4u_operand" " Iu04, r"))]
++		     UNSPEC_ROUND))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   srli16.u\t%0, %1, %2
++   srl16.u\t%0, %1, %2"
++  [(set_attr "type"   "daluround,daluround")
++   (set_attr "length" "        4,        4")])
++
++(define_insn "kslra16"
++  [(set (match_operand:V2HI 0 "register_operand"                  "=r")
++	(if_then_else:V2HI
++	  (lt:SI (match_operand:SI 2 "register_operand"           " r")
++		 (const_int 0))
++	  (ashiftrt:V2HI (match_operand:V2HI 1 "register_operand" " r")
++			 (neg:SI (match_dup 2)))
++	  (ashift:V2HI (match_dup 1)
++		       (match_dup 2))))]
++  "NDS32_EXT_DSP_P ()"
++  "kslra16\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")])
++
++(define_insn "kslra16_round"
++  [(set (match_operand:V2HI 0 "register_operand"                  "=r")
++	(if_then_else:V2HI
++	  (lt:SI (match_operand:SI 2 "register_operand"           " r")
++		 (const_int 0))
++	  (unspec:V2HI [(ashiftrt:V2HI (match_operand:V2HI 1 "register_operand" " r")
++				       (neg:SI (match_dup 2)))]
++		       UNSPEC_ROUND)
++	  (ashift:V2HI (match_dup 1)
++		       (match_dup 2))))]
++  "NDS32_EXT_DSP_P ()"
++  "kslra16.u\t%0, %1, %2"
++  [(set_attr "type"    "daluround")
++   (set_attr "length"  "4")])
++
++(define_insn "cmpeq<bits>"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(unspec:SI [(eq:SI (match_operand:VQIHI 1 "register_operand" " r")
++			   (match_operand:VQIHI 2 "register_operand" " r"))]
++		   UNSPEC_VEC_COMPARE))]
++  "NDS32_EXT_DSP_P ()"
++  "cmpeq<bits>\t%0, %1, %2"
++  [(set_attr "type"    "dcmp")
++   (set_attr "length"  "4")])
++
++(define_insn "scmplt<bits>"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(unspec:SI [(lt:SI (match_operand:VQIHI 1 "register_operand" " r")
++			   (match_operand:VQIHI 2 "register_operand" " r"))]
++		   UNSPEC_VEC_COMPARE))]
++  "NDS32_EXT_DSP_P ()"
++  "scmplt<bits>\t%0, %1, %2"
++  [(set_attr "type"    "dcmp")
++   (set_attr "length"  "4")])
++
++(define_insn "scmple<bits>"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(unspec:SI [(le:SI (match_operand:VQIHI 1 "register_operand" " r")
++			   (match_operand:VQIHI 2 "register_operand" " r"))]
++		   UNSPEC_VEC_COMPARE))]
++  "NDS32_EXT_DSP_P ()"
++  "scmple<bits>\t%0, %1, %2"
++  [(set_attr "type"    "dcmp")
++   (set_attr "length"  "4")])
++
++(define_insn "ucmplt<bits>"
++  [(set (match_operand:SI 0 "register_operand"                        "=r")
++	(unspec:SI [(ltu:SI (match_operand:VQIHI 1 "register_operand" " r")
++			    (match_operand:VQIHI 2 "register_operand" " r"))]
++		   UNSPEC_VEC_COMPARE))]
++  "NDS32_EXT_DSP_P ()"
++  "ucmplt<bits>\t%0, %1, %2"
++  [(set_attr "type"    "dcmp")
++   (set_attr "length"  "4")])
++
++(define_insn "ucmple<bits>"
++  [(set (match_operand:SI 0 "register_operand"                        "=r")
++	(unspec:SI [(leu:SI (match_operand:VQIHI 1 "register_operand" " r")
++			    (match_operand:VQIHI 2 "register_operand" " r"))]
++		   UNSPEC_VEC_COMPARE))]
++  "NDS32_EXT_DSP_P ()"
++  "ucmple<bits>\t%0, %1, %2"
++  [(set_attr "type"    "dcmp")
++   (set_attr "length"  "4")])
++
++(define_insn "sclip16"
++  [(set (match_operand:V2HI 0 "register_operand"                "=   r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand"  "    r")
++		      (match_operand:SI 2 "nds32_imm4u_operand" " Iu04")]
++		     UNSPEC_CLIPS))]
++  "NDS32_EXT_DSP_P ()"
++  "sclip16\t%0, %1, %2"
++  [(set_attr "type"    "dclip")
++   (set_attr "length"  "4")])
++
++(define_insn "uclip16"
++  [(set (match_operand:V2HI 0 "register_operand"                "=   r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand"  "    r")
++		      (match_operand:SI 2 "nds32_imm4u_operand" " Iu04")]
++		     UNSPEC_CLIP))]
++  "NDS32_EXT_DSP_P ()"
++  "uclip16\t%0, %1, %2"
++  [(set_attr "type"    "dclip")
++   (set_attr "length"  "4")])
++
++(define_insn "khm16"
++  [(set (match_operand:V2HI 0 "register_operand"                "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand"  " r")
++		      (match_operand:V2HI 2 "register_operand" "  r")]
++		     UNSPEC_KHM))]
++  "NDS32_EXT_DSP_P ()"
++  "khm16\t%0, %1, %2"
++  [(set_attr "type"    "dmul")
++   (set_attr "length"  "4")])
++
++(define_insn "khmx16"
++  [(set (match_operand:V2HI 0 "register_operand"                "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand"  " r")
++		      (match_operand:V2HI 2 "register_operand" "  r")]
++		     UNSPEC_KHMX))]
++  "NDS32_EXT_DSP_P ()"
++  "khmx16\t%0, %1, %2"
++  [(set_attr "type"    "dmul")
++   (set_attr "length"  "4")])
++
++(define_expand "vec_setv4qi"
++  [(match_operand:V4QI 0 "register_operand" "")
++   (match_operand:QI 1 "register_operand" "")
++   (match_operand:SI 2 "immediate_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  HOST_WIDE_INT pos = INTVAL (operands[2]);
++  if (pos > 4)
++    gcc_unreachable ();
++  HOST_WIDE_INT elem = (HOST_WIDE_INT) 1 << pos;
++  emit_insn (gen_vec_setv4qi_internal (operands[0], operands[1],
++				       operands[0], GEN_INT (elem)));
++  DONE;
++})
++
++(define_expand "insb"
++  [(match_operand:V4QI 0 "register_operand" "")
++   (match_operand:V4QI 1 "register_operand" "")
++   (match_operand:SI 2 "register_operand" "")
++   (match_operand:SI 3 "const_int_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (INTVAL (operands[3]) > 3 || INTVAL (operands[3]) < 0)
++    gcc_unreachable ();
++
++  rtx src = gen_reg_rtx (QImode);
++
++  convert_move (src, operands[2], false);
++
++  HOST_WIDE_INT selector_index;
++  /* Big endian need reverse index. */
++  if (TARGET_BIG_ENDIAN)
++    selector_index = 4 - INTVAL (operands[3]) - 1;
++  else
++    selector_index = INTVAL (operands[3]);
++  rtx selector = gen_int_mode (1 << selector_index, SImode);
++  emit_insn (gen_vec_setv4qi_internal (operands[0], src,
++				       operands[1], selector));
++  DONE;
++})
++
++(define_expand "insvsi"
++  [(set (zero_extract:SI (match_operand:SI 0 "register_operand" "")
++			 (match_operand:SI 1 "const_int_operand" "")
++			 (match_operand:SI 2 "nds32_insv_operand" ""))
++	(match_operand:SI 3 "register_operand" ""))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (INTVAL (operands[1]) != 8)
++    FAIL;
++}
++  [(set_attr "type"    "dinsb")
++   (set_attr "length"  "4")])
++
++
++(define_insn "insvsi_internal"
++  [(set (zero_extract:SI (match_operand:SI 0 "register_operand"   "+r")
++			 (const_int 8)
++			 (match_operand:SI 1 "nds32_insv_operand"  "i"))
++	(match_operand:SI 2                  "register_operand"    "r"))]
++  "NDS32_EXT_DSP_P ()"
++  "insb\t%0, %2, %v1"
++  [(set_attr "type"    "dinsb")
++   (set_attr "length"  "4")])
++
++(define_insn "insvsiqi_internal"
++  [(set (zero_extract:SI (match_operand:SI 0 "register_operand"   "+r")
++			 (const_int 8)
++			 (match_operand:SI 1 "nds32_insv_operand"  "i"))
++	(zero_extend:SI (match_operand:QI 2 "register_operand"    "r")))]
++  "NDS32_EXT_DSP_P ()"
++  "insb\t%0, %2, %v1"
++  [(set_attr "type"    "dinsb")
++   (set_attr "length"  "4")])
++
++;; Intermedium pattern for synthetize insvsiqi_internal
++;; v0 = ((v1 & 0xff) << 8)
++(define_insn_and_split "and0xff_s8"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(and:SI (ashift:SI (match_operand:SI 1 "register_operand" "r")
++			   (const_int 8))
++		(const_int 65280)))]
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (SImode);
++  emit_insn (gen_ashlsi3 (tmp, operands[1], gen_int_mode (8, SImode)));
++  emit_insn (gen_andsi3 (operands[0], tmp, gen_int_mode (0xffff, SImode)));
++  DONE;
++})
++
++;; v0 = (v1 & 0xff00ffff) | ((v2 << 16) | 0xff0000)
++(define_insn_and_split "insbsi2"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI (and:SI (match_operand:SI 1 "register_operand" "0")
++			(const_int -16711681))
++		(and:SI (ashift:SI (match_operand:SI 2 "register_operand" "r")
++				   (const_int 16))
++			(const_int 16711680))))]
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (SImode);
++  emit_move_insn (tmp, operands[1]);
++  emit_insn (gen_insvsi_internal (tmp, gen_int_mode(16, SImode), operands[2]));
++  emit_move_insn (operands[0], tmp);
++  DONE;
++})
++
++;; v0 = (v1 & 0xff00ffff) | v2
++(define_insn_and_split "ior_and0xff00ffff_reg"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI (and:SI (match_operand:SI 1 "register_operand" "r")
++			(const_int -16711681))
++		(match_operand:SI 2 "register_operand" "r")))]
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (SImode);
++  emit_insn (gen_andsi3 (tmp, operands[1], gen_int_mode (0xff00ffff, SImode)));
++  emit_insn (gen_iorsi3 (operands[0], tmp, operands[2]));
++  DONE;
++})
++
++(define_insn "vec_setv4qi_internal"
++  [(set (match_operand:V4QI 0 "register_operand"          "=   r,    r,    r,    r")
++	(vec_merge:V4QI
++	  (vec_duplicate:V4QI
++	    (match_operand:QI 1 "register_operand"        "    r,    r,    r,    r"))
++	  (match_operand:V4QI 2 "register_operand"        "    0,    0,    0,    0")
++	  (match_operand:SI 3 "nds32_imm_1_2_4_8_operand" " Iv01, Iv02, Iv04, Iv08")))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++       const char *pats[] = { "insb\t%0, %1, 3",
++			      "insb\t%0, %1, 2",
++			      "insb\t%0, %1, 1",
++			      "insb\t%0, %1, 0" };
++      return pats[which_alternative];
++    }
++  else
++    {
++       const char *pats[] = { "insb\t%0, %1, 0",
++			      "insb\t%0, %1, 1",
++			      "insb\t%0, %1, 2",
++			      "insb\t%0, %1, 3" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"    "dinsb")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_setv4qi_internal_vec"
++  [(set (match_operand:V4QI 0 "register_operand"          "=   r,    r,    r,    r")
++	(vec_merge:V4QI
++	  (vec_duplicate:V4QI
++	    (vec_select:QI
++	      (match_operand:V4QI 1 "register_operand"    "    r,    r,    r,    r")
++	      (parallel [(const_int 0)])))
++	  (match_operand:V4QI 2 "register_operand"        "    0,    0,    0,    0")
++	  (match_operand:SI 3 "nds32_imm_1_2_4_8_operand" " Iv01, Iv02, Iv04, Iv08")))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   insb\t%0, %1, 0
++   insb\t%0, %1, 1
++   insb\t%0, %1, 2
++   insb\t%0, %1, 3"
++  [(set_attr "type"    "dinsb")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_mergev4qi_and_cv0_1"
++  [(set (match_operand:V4QI 0 "register_operand"       "=$l,r")
++	(vec_merge:V4QI
++	  (vec_duplicate:V4QI
++	    (vec_select:QI
++	      (match_operand:V4QI 1 "register_operand" "  l,r")
++	      (parallel [(const_int 0)])))
++	  (const_vector:V4QI [
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)])
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeb33\t%0, %1
++   zeb\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_mergev4qi_and_cv0_2"
++  [(set (match_operand:V4QI 0 "register_operand"       "=$l,r")
++	(vec_merge:V4QI
++	  (const_vector:V4QI [
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)])
++	  (vec_duplicate:V4QI
++	    (vec_select:QI
++	      (match_operand:V4QI 1 "register_operand" "  l,r")
++	      (parallel [(const_int 0)])))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeb33\t%0, %1
++   zeb\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_mergeqi_and_cv0_1"
++  [(set (match_operand:V4QI 0 "register_operand"                     "=$l,r")
++	(vec_merge:V4QI
++	  (vec_duplicate:V4QI (match_operand:QI 1 "register_operand" "  l,r"))
++	  (const_vector:V4QI [
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)])
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeb33\t%0, %1
++   zeb\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_mergeqi_and_cv0_2"
++  [(set (match_operand:V4QI 0 "register_operand"                     "=$l,r")
++	(vec_merge:V4QI
++	  (const_vector:V4QI [
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)
++	    (const_int 0)])
++	  (vec_duplicate:V4QI (match_operand:QI 1 "register_operand" "  l,r"))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeb33\t%0, %1
++   zeb\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_expand "vec_setv2hi"
++  [(match_operand:V2HI 0 "register_operand" "")
++   (match_operand:HI 1 "register_operand" "")
++   (match_operand:SI 2 "immediate_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  HOST_WIDE_INT pos = INTVAL (operands[2]);
++  if (pos > 2)
++    gcc_unreachable ();
++  HOST_WIDE_INT elem = (HOST_WIDE_INT) 1 << pos;
++  emit_insn (gen_vec_setv2hi_internal (operands[0], operands[1],
++				       operands[0], GEN_INT (elem)));
++  DONE;
++})
++
++(define_insn "vec_setv2hi_internal"
++  [(set (match_operand:V2HI 0 "register_operand"      "=   r,    r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (match_operand:HI 1 "register_operand"    "    r,    r"))
++	  (match_operand:V2HI 2 "register_operand"    "    r,    r")
++	  (match_operand:SI 3 "nds32_imm_1_2_operand" " Iv01, Iv02")))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "pkbb16\t%0, %1, %2",
++			     "pktb16\t%0, %2, %1" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "pktb16\t%0, %2, %1",
++			     "pkbb16\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_mergev2hi_and_cv0_1"
++  [(set (match_operand:V2HI 0 "register_operand"       "=$l,r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (vec_select:HI
++	      (match_operand:V2HI 1 "register_operand" "  l,r")
++	      (parallel [(const_int 0)])))
++	  (const_vector:V2HI [
++	    (const_int 0)
++	    (const_int 0)])
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeh33\t%0, %1
++   zeh\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_mergev2hi_and_cv0_2"
++  [(set (match_operand:V2HI 0 "register_operand"       "=$l,r")
++	(vec_merge:V2HI
++	  (const_vector:V2HI [
++	    (const_int 0)
++	    (const_int 0)])
++	  (vec_duplicate:V2HI
++	    (vec_select:HI
++	      (match_operand:V2HI 1 "register_operand" "  l,r")
++	      (parallel [(const_int 0)])))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeh33\t%0, %1
++   zeh\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_mergehi_and_cv0_1"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=$l,r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI (match_operand:HI 1 "register_operand" "  l,r"))
++	  (const_vector:V2HI [
++	    (const_int 0)
++	    (const_int 0)])
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeh33\t%0, %1
++   zeh\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_mergehi_and_cv0_2"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=$l,r")
++	(vec_merge:V2HI
++	  (const_vector:V2HI [
++	    (const_int 0)
++	    (const_int 0)])
++	  (vec_duplicate:V2HI (match_operand:HI 1 "register_operand" "  l,r"))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   zeh33\t%0, %1
++   zeh\t%0, %1"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_expand "pkbb"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V2HI 1 "register_operand")
++   (match_operand:V2HI 2 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (1), GEN_INT (1), GEN_INT (1)));
++    }
++  else
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (2), GEN_INT (0), GEN_INT (0)));
++    }
++  DONE;
++})
++
++(define_insn "pkbbsi_1"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI (and:SI (match_operand:SI 1 "register_operand" "r")
++			(const_int 65535))
++		(ashift:SI (match_operand:SI 2 "register_operand" "r")
++			   (const_int 16))))]
++  "NDS32_EXT_DSP_P ()"
++  "pkbb16\t%0, %2, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "pkbbsi_2"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI	(ashift:SI (match_operand:SI 2 "register_operand" "r")
++			   (const_int 16))
++		(and:SI (match_operand:SI 1 "register_operand" "r")
++			(const_int 65535))))]
++  "NDS32_EXT_DSP_P ()"
++  "pkbb16\t%0, %2, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "pkbbsi_3"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI (zero_extend:SI	(match_operand:HI 1 "register_operand" "r"))
++		(ashift:SI (match_operand:SI 2 "register_operand" "r")
++			   (const_int 16))))]
++  "NDS32_EXT_DSP_P ()"
++  "pkbb16\t%0, %2, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "pkbbsi_4"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI	(ashift:SI (match_operand:SI 2 "register_operand" "r")
++			   (const_int 16))
++		(zero_extend:SI (match_operand:HI 1 "register_operand" "r"))))]
++  "NDS32_EXT_DSP_P ()"
++  "pkbb16\t%0, %2, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++;; v0 = (v1 & 0xffff0000) | (v2 & 0xffff)
++(define_insn "pktbsi_1"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI (and:SI (match_operand:SI 1 "register_operand" "r")
++			(const_int -65536))
++		(zero_extend:SI (match_operand:HI 2 "register_operand" "r"))))]
++  "NDS32_EXT_DSP_P ()"
++  "pktb16\t%0, %1, %2"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "pktbsi_2"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ior:SI (and:SI (match_operand:SI 1 "register_operand" "r")
++			(const_int -65536))
++		(and:SI (match_operand:SI 2 "register_operand" "r")
++			(const_int 65535))))]
++  "NDS32_EXT_DSP_P ()"
++  "pktb16\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "pktbsi_3"
++  [(set (zero_extract:SI (match_operand:SI 0 "register_operand" "+r")
++			 (const_int 16 )
++			 (const_int 0))
++	(match_operand:SI 1 "register_operand"                  " r"))]
++  "NDS32_EXT_DSP_P ()"
++  "pktb16\t%0, %0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "pktbsi_4"
++  [(set (zero_extract:SI (match_operand:SI 0 "register_operand" "+r")
++			 (const_int 16 )
++			 (const_int 0))
++	(zero_extend:SI (match_operand:HI 1 "register_operand"  " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "pktb16\t%0, %0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "pkttsi"
++  [(set (match_operand:SI 0 "register_operand"                      "=r")
++	(ior:SI (and:SI (match_operand:SI 1 "register_operand"      " r")
++			(const_int -65536))
++		(lshiftrt:SI (match_operand:SI 2 "register_operand" " r")
++			     (const_int 16))))]
++  "NDS32_EXT_DSP_P ()"
++  "pktt16\t%0, %1, %2"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_expand "pkbt"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V2HI 1 "register_operand")
++   (match_operand:V2HI 2 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (1), GEN_INT (1), GEN_INT (0)));
++    }
++  else
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (2), GEN_INT (0), GEN_INT (1)));
++    }
++  DONE;
++})
++
++(define_expand "pktt"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V2HI 1 "register_operand")
++   (match_operand:V2HI 2 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (1), GEN_INT (0), GEN_INT (0)));
++    }
++  else
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (2), GEN_INT (1), GEN_INT (1)));
++    }
++  DONE;
++})
++
++(define_expand "pktb"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V2HI 1 "register_operand")
++   (match_operand:V2HI 2 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (1), GEN_INT (0), GEN_INT (1)));
++    }
++  else
++    {
++      emit_insn (gen_vec_mergevv (operands[0], operands[1], operands[2],
++				  GEN_INT (2), GEN_INT (1), GEN_INT (0)));
++    }
++  DONE;
++})
++
++(define_insn "vec_mergerr"
++  [(set (match_operand:V2HI 0 "register_operand"      "=   r,    r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (match_operand:HI 1 "register_operand"    "    r,    r"))
++	  (vec_duplicate:V2HI
++	    (match_operand:HI 2 "register_operand"    "    r,    r"))
++	  (match_operand:SI 3 "nds32_imm_1_2_operand" " Iv01, Iv02")))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   pkbb16\t%0, %2, %1
++   pkbb16\t%0, %1, %2"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++
++(define_insn "vec_merge"
++  [(set (match_operand:V2HI 0 "register_operand"      "=   r,    r")
++	(vec_merge:V2HI
++	  (match_operand:V2HI 1 "register_operand"    "    r,    r")
++	  (match_operand:V2HI 2 "register_operand"    "    r,    r")
++	  (match_operand:SI 3 "nds32_imm_1_2_operand" " Iv01, Iv02")))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "pktb16\t%0, %1, %2",
++			     "pktb16\t%0, %2, %1" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "pktb16\t%0, %2, %1",
++			     "pktb16\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_mergerv"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=   r,    r,    r,    r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (match_operand:HI 1 "register_operand"                   "    r,    r,    r,    r"))
++	  (vec_duplicate:V2HI
++	    (vec_select:HI
++	      (match_operand:V2HI 2 "register_operand"               "    r,    r,    r,    r")
++	      (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iv00, Iv01, Iv00, Iv01")])))
++	  (match_operand:SI 3 "nds32_imm_1_2_operand"                " Iv01, Iv01, Iv02, Iv02")))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   pkbb16\t%0, %2, %1
++   pktb16\t%0, %2, %1
++   pkbb16\t%0, %1, %2
++   pkbt16\t%0, %1, %2"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_mergevr"
++  [(set (match_operand:V2HI 0 "register_operand"                      "=   r,    r,    r,    r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (vec_select:HI
++	      (match_operand:V2HI 1 "register_operand"                "    r,    r,    r,    r")
++	       (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iv00, Iv01, Iv00, Iv01")])))
++	  (vec_duplicate:V2HI
++	    (match_operand:HI 2 "register_operand"                    "    r,    r,    r,    r"))
++	  (match_operand:SI 3 "nds32_imm_1_2_operand"                 " Iv01, Iv01, Iv02, Iv02")))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   pkbb16\t%0, %2, %1
++   pkbt16\t%0, %2, %1
++   pkbb16\t%0, %1, %2
++   pktb16\t%0, %1, %2"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_mergevv"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=   r,    r,    r,    r,    r,    r,    r,    r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (vec_select:HI
++	      (match_operand:V2HI 1 "register_operand"               "    r,    r,    r,    r,    r,    r,    r,    r")
++	      (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iv00, Iv00, Iv01, Iv01, Iv00, Iv00, Iv01, Iv01")])))
++	  (vec_duplicate:V2HI
++	    (vec_select:HI
++	      (match_operand:V2HI 2 "register_operand"               "    r,    r,    r,    r,    r,    r,    r,    r")
++	      (parallel [(match_operand:SI 5 "nds32_imm_0_1_operand" " Iv00, Iv01, Iv01, Iv00, Iv00, Iv01, Iv01, Iv00")])))
++	  (match_operand:SI 3 "nds32_imm_1_2_operand"                " Iv01, Iv01, Iv01, Iv01, Iv02, Iv02, Iv02, Iv02")))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "pktt16\t%0, %1, %2",
++			     "pktb16\t%0, %1, %2",
++			     "pkbb16\t%0, %1, %2",
++			     "pkbt16\t%0, %1, %2",
++			     "pktt16\t%0, %2, %1",
++			     "pkbt16\t%0, %2, %1",
++			     "pkbb16\t%0, %2, %1",
++			     "pktb16\t%0, %2, %1" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "pkbb16\t%0, %2, %1",
++			     "pktb16\t%0, %2, %1",
++			     "pktt16\t%0, %2, %1",
++			     "pkbt16\t%0, %2, %1",
++			     "pkbb16\t%0, %1, %2",
++			     "pkbt16\t%0, %1, %2",
++			     "pktt16\t%0, %1, %2",
++			     "pktb16\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_expand "vec_extractv4qi"
++  [(set (match_operand:QI 0 "register_operand" "")
++	(vec_select:QI
++	  (match_operand:V4QI 1          "nonimmediate_operand" "")
++	  (parallel [(match_operand:SI 2 "const_int_operand" "")])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  if (INTVAL (operands[2]) != 0
++      && INTVAL (operands[2]) != 1
++      && INTVAL (operands[2]) != 2
++      && INTVAL (operands[2]) != 3)
++    gcc_unreachable ();
++
++  if (INTVAL (operands[2]) != 0 && MEM_P (operands[0]))
++    FAIL;
++})
++
++(define_insn "vec_extractv4qi0"
++  [(set (match_operand:QI 0 "register_operand"         "=l,r,r")
++	(vec_select:QI
++	  (match_operand:V4QI 1 "nonimmediate_operand" " l,r,m")
++	  (parallel [(const_int 0)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "zeb33\t%0, %1";
++    case 1:
++      return "zeb\t%0, %1";
++    case 2:
++      return nds32_output_32bit_load (operands, 1);
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_extractv4qi0_ze"
++  [(set (match_operand:SI 0 "register_operand"         "=l,r,r")
++	(zero_extend:SI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "nonimmediate_operand" " l,r,m")
++	    (parallel [(const_int 0)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "zeb33\t%0, %1";
++    case 1:
++      return "zeb\t%0, %1";
++    case 2:
++      return nds32_output_32bit_load (operands, 1);
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_extractv4qi0_se"
++  [(set (match_operand:SI 0 "register_operand"         "=l,r,r")
++	(sign_extend:SI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "nonimmediate_operand" " l,r,m")
++	    (parallel [(const_int 0)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "seb33\t%0, %1";
++    case 1:
++      return "seb\t%0, %1";
++    case 2:
++      return nds32_output_32bit_load_s (operands, 1);
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn_and_split "vec_extractv4qi1"
++  [(set (match_operand:QI 0 "register_operand" "=r")
++	(vec_select:QI
++	  (match_operand:V4QI 1 "register_operand" " r")
++	  (parallel [(const_int 1)])))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (V4QImode);
++  emit_insn (gen_rotrv4qi_1 (tmp, operands[1]));
++  emit_insn (gen_vec_extractv4qi0 (operands[0], tmp));
++  DONE;
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn_and_split "vec_extractv4qi2"
++  [(set (match_operand:QI 0 "register_operand" "=r")
++	(vec_select:QI
++	  (match_operand:V4QI 1 "register_operand" " r")
++	  (parallel [(const_int 2)])))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (V4QImode);
++  emit_insn (gen_rotrv4qi_2 (tmp, operands[1]));
++  emit_insn (gen_vec_extractv4qi0 (operands[0], tmp));
++  DONE;
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn_and_split "vec_extractv4qi3"
++  [(set (match_operand:QI 0 "register_operand" "=r")
++	(vec_select:QI
++	  (match_operand:V4QI 1 "register_operand" " r")
++	  (parallel [(const_int 3)])))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (V4QImode);
++  emit_insn (gen_rotrv4qi_3 (tmp, operands[1]));
++  emit_insn (gen_vec_extractv4qi0 (operands[0], tmp));
++  DONE;
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "vec_extractv4qi3_se"
++  [(set (match_operand:SI 0 "register_operand"       "=$d,r")
++	(sign_extend:SI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "register_operand" "  0,r")
++	    (parallel [(const_int 3)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   srai45\t%0, 24
++   srai\t%0, %1, 24"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_extractv4qi3_ze"
++  [(set (match_operand:SI 0 "register_operand"       "=$d,r")
++	(zero_extend:SI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "register_operand" "  0,r")
++	    (parallel [(const_int 3)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   srli45\t%0, 24
++   srli\t%0, %1, 24"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn_and_split "vec_extractv4qihi0"
++  [(set (match_operand:HI 0 "register_operand" "=r")
++	(sign_extend:HI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "register_operand" " r")
++	    (parallel [(const_int 0)]))))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (QImode);
++  emit_insn (gen_vec_extractv4qi0 (tmp, operands[1]));
++  emit_insn (gen_extendqihi2 (operands[0], tmp));
++  DONE;
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn_and_split "vec_extractv4qihi1"
++  [(set (match_operand:HI 0 "register_operand" "=r")
++	(sign_extend:HI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "register_operand" " r")
++	    (parallel [(const_int 1)]))))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (QImode);
++  emit_insn (gen_vec_extractv4qi1 (tmp, operands[1]));
++  emit_insn (gen_extendqihi2 (operands[0], tmp));
++  DONE;
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn_and_split "vec_extractv4qihi2"
++  [(set (match_operand:HI 0 "register_operand" "=r")
++	(sign_extend:HI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "register_operand" " r")
++	    (parallel [(const_int 2)]))))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (QImode);
++  emit_insn (gen_vec_extractv4qi2 (tmp, operands[1]));
++  emit_insn (gen_extendqihi2 (operands[0], tmp));
++  DONE;
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_insn_and_split "vec_extractv4qihi3"
++  [(set (match_operand:HI 0 "register_operand" "=r")
++	(sign_extend:HI
++	  (vec_select:QI
++	    (match_operand:V4QI 1 "register_operand" " r")
++	    (parallel [(const_int 3)]))))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx tmp = gen_reg_rtx (QImode);
++  emit_insn (gen_vec_extractv4qi3 (tmp, operands[1]));
++  emit_insn (gen_extendqihi2 (operands[0], tmp));
++  DONE;
++}
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")])
++
++(define_expand "vec_extractv2hi"
++  [(set (match_operand:HI 0 "register_operand" "")
++	(vec_select:HI
++	  (match_operand:V2HI 1          "nonimmediate_operand" "")
++	  (parallel [(match_operand:SI 2 "const_int_operand" "")])))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (INTVAL (operands[2]) != 0
++      && INTVAL (operands[2]) != 1)
++    gcc_unreachable ();
++
++  if (INTVAL (operands[2]) != 0 && MEM_P (operands[0]))
++    FAIL;
++})
++
++(define_insn "vec_extractv2hi0"
++  [(set (match_operand:HI 0 "register_operand"         "=$l,r,r")
++	(vec_select:HI
++	  (match_operand:V2HI 1 "nonimmediate_operand" "  l,r,m")
++	  (parallel [(const_int 0)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "seh33\t%0, %1";
++    case 1:
++      return "seh\t%0, %1";
++    case 2:
++      return nds32_output_32bit_load_s (operands, 2);
++
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"    "alu,alu,load")
++   (set_attr "length"  "  2,  4,   4")])
++
++(define_insn "vec_extractv2hi0_ze"
++  [(set (match_operand:SI 0 "register_operand"         "=$l, r,$  l, *r")
++        (zero_extend:SI
++	  (vec_select:HI
++	    (match_operand:V2HI 1 "nonimmediate_operand" "  l, r, U33,  m")
++	    (parallel [(const_int 0)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "zeh33\t%0, %1";
++    case 1:
++      return "zeh\t%0, %1";
++    case 2:
++      return nds32_output_16bit_load (operands, 2);
++    case 3:
++      return nds32_output_32bit_load (operands, 2);
++
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"   "alu,alu,load,load")
++   (set_attr "length" "  2,  4,   2,   4")])
++
++(define_insn "vec_extractv2hi0_se"
++  [(set (match_operand:SI 0 "register_operand"         "=$l, r, r")
++        (sign_extend:SI
++	  (vec_select:HI
++	    (match_operand:V2HI 1 "nonimmediate_operand" "  l,r,m")
++	    (parallel [(const_int 0)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "seh33\t%0, %1";
++    case 1:
++      return "seh\t%0, %1";
++    case 2:
++      return nds32_output_32bit_load_s (operands, 2);
++
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"   "alu,alu,load")
++   (set_attr "length" "  2,  4,   4")])
++
++(define_insn "vec_extractv2hi0_be"
++  [(set (match_operand:HI 0 "register_operand"     "=$d,r")
++	(vec_select:HI
++	  (match_operand:V2HI 1 "register_operand" "  0,r")
++	  (parallel [(const_int 0)])))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "@
++   srai45\t%0, 16
++   srai\t%0, %1, 16"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_extractv2hi1"
++  [(set (match_operand:HI 0 "register_operand"     "=$d,r")
++	(vec_select:HI
++	  (match_operand:V2HI 1 "register_operand" "  0,r")
++	  (parallel [(const_int 1)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   srai45\t%0, 16
++   srai\t%0, %1, 16"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_extractv2hi1_se"
++  [(set (match_operand:SI 0 "register_operand"     "=$d,r")
++	(sign_extend:SI
++	  (vec_select:HI
++	    (match_operand:V2HI 1 "register_operand" "  0,r")
++	    (parallel [(const_int 1)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   srai45\t%0, 16
++   srai\t%0, %1, 16"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_extractv2hi1_ze"
++  [(set (match_operand:SI 0 "register_operand"     "=$d,r")
++	(zero_extend:SI
++	  (vec_select:HI
++	    (match_operand:V2HI 1 "register_operand" "  0,r")
++	    (parallel [(const_int 1)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "@
++   srli45\t%0, 16
++   srli\t%0, %1, 16"
++  [(set_attr "type"    "alu,alu")
++   (set_attr "length"  "  2,  4")])
++
++(define_insn "vec_extractv2hi1_be"
++  [(set (match_operand:HI 0 "register_operand"         "=$l,r,r")
++	(vec_select:HI
++	  (match_operand:V2HI 1 "nonimmediate_operand" "  l,r,m")
++	  (parallel [(const_int 1)])))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++{
++  switch (which_alternative)
++    {
++    case 0:
++      return "seh33\t%0, %1";
++    case 1:
++      return "seh\t%0, %1";
++    case 2:
++      return nds32_output_32bit_load_s (operands, 2);
++
++    default:
++      gcc_unreachable ();
++    }
++}
++  [(set_attr "type"    "alu,alu,load")
++   (set_attr "length"  "  2,  4,   4")])
++
++(define_insn "<su>mul16"
++  [(set (match_operand:V2SI 0 "register_operand"                         "=r")
++	(mult:V2SI (extend:V2SI (match_operand:V2HI 1 "register_operand" "%r"))
++		   (extend:V2SI (match_operand:V2HI 2 "register_operand" " r"))))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>mul16\t%0, %1, %2"
++  [(set_attr "type"   "dmul")
++   (set_attr "length"   "4")])
++
++(define_insn "<su>mulx16"
++  [(set (match_operand:V2SI 0 "register_operand"         "=r")
++	(vec_merge:V2SI
++	  (vec_duplicate:V2SI
++	    (mult:SI
++	      (extend:SI
++		(vec_select:HI
++		  (match_operand:V2HI 1 "register_operand" " r")
++		  (parallel [(const_int 0)])))
++	      (extend:SI
++		(vec_select:HI
++		  (match_operand:V2HI 2 "register_operand" " r")
++		  (parallel [(const_int 1)])))))
++	  (vec_duplicate:V2SI
++	    (mult:SI
++	      (extend:SI
++		(vec_select:HI
++		  (match_dup 1)
++		  (parallel [(const_int 1)])))
++	      (extend:SI
++		(vec_select:HI
++		  (match_dup 2)
++		  (parallel [(const_int 0)])))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>mulx16\t%0, %1, %2"
++  [(set_attr "type"    "dmul")
++   (set_attr "length"   "4")])
++
++(define_insn "rotrv2hi_1"
++  [(set (match_operand:V2HI 0 "register_operand"    "=r")
++	(vec_select:V2HI
++	   (match_operand:V2HI 1 "register_operand" " r")
++	   (parallel [(const_int 1) (const_int 0)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 16"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "rotrv2hi_1_be"
++  [(set (match_operand:V2HI 0 "register_operand"    "=r")
++	(vec_select:V2HI
++	   (match_operand:V2HI 1 "register_operand" " r")
++	   (parallel [(const_int 0) (const_int 1)])))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 16"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "rotrv4qi_1"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 1) (const_int 2) (const_int 3) (const_int 0)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 8"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "rotrv4qi_1_be"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 2) (const_int 1) (const_int 0) (const_int 3)])))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 8"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "rotrv4qi_2"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 2) (const_int 3) (const_int 0) (const_int 1)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 16"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "rotrv4qi_2_be"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 1) (const_int 0) (const_int 3) (const_int 2)])))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 16"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "rotrv4qi_3"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 3) (const_int 0) (const_int 1) (const_int 2)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 24"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "rotrv4qi_3_be"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 0) (const_int 3) (const_int 2) (const_int 1)])))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "rotri\t%0, %1, 24"
++  [(set_attr "type"   "alu")
++   (set_attr "length"  "4")])
++
++(define_insn "v4qi_dup_10"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 0) (const_int 1) (const_int 0) (const_int 1)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "pkbb\t%0, %1, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "v4qi_dup_32"
++  [(set (match_operand:V4QI 0 "register_operand"    "=r")
++	(vec_select:V4QI
++	   (match_operand:V4QI 1 "register_operand" " r")
++	   (parallel [(const_int 2) (const_int 3) (const_int 2) (const_int 3)])))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "pktt\t%0, %1, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_expand "vec_unpacks_lo_v4qi"
++  [(match_operand:V2HI 0 "register_operand" "=r")
++   (match_operand:V4QI 1 "register_operand" " r")]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++{
++  emit_insn (gen_sunpkd810 (operands[0], operands[1]));
++  DONE;
++})
++
++(define_expand "sunpkd810"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_sunpkd810_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_sunpkd810_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_insn "<zs>unpkd810_imp"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd810\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd810_imp_inv"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 1)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd810\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd810_imp_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 2)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 3)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd810\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd810_imp_inv_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 3)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 2)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd810\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_expand "sunpkd820"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_sunpkd820_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_sunpkd820_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_insn "<zs>unpkd820_imp"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 2)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd820\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd820_imp_inv"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 2)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd820\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd820_imp_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 3)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd820\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd820_imp_inv_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 3)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 1)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd820\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_expand "sunpkd830"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_sunpkd830_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_sunpkd830_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_insn "<zs>unpkd830_imp"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 3)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd830\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd830_imp_inv"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 3)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd830\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd830_imp_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 3)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd830\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd830_imp_inv_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 3)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd830\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_expand "sunpkd831"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_sunpkd831_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_sunpkd831_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_insn "<zs>unpkd831_imp"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 3)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 1)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd831\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd831_imp_inv"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 3)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<zs>unpkd831\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd831_imp_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 0)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 2)]))))
++	  (const_int 1)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd831\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_insn "<zs>unpkd831_imp_inv_be"
++  [(set (match_operand:V2HI 0 "register_operand"                     "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_operand:V4QI 1 "register_operand"             " r")
++		(parallel [(const_int 2)]))))
++	  (vec_duplicate:V2HI
++	    (extend:HI
++	      (vec_select:QI
++		(match_dup 1)
++		(parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "<zs>unpkd831\t%0, %1"
++  [(set_attr "type"    "dpack")
++   (set_attr "length"  "4")])
++
++(define_expand "zunpkd810"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_zunpkd810_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_zunpkd810_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_expand "zunpkd820"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_zunpkd820_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_zunpkd820_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_expand "zunpkd830"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_zunpkd830_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_zunpkd830_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_expand "zunpkd831"
++  [(match_operand:V2HI 0 "register_operand")
++   (match_operand:V4QI 1 "register_operand")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_zunpkd831_imp_be (operands[0], operands[1]));
++  else
++    emit_insn (gen_zunpkd831_imp (operands[0], operands[1]));
++  DONE;
++})
++
++(define_expand "smbb"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_mulhisi3v (operands[0], operands[1], operands[2],
++			      GEN_INT (1), GEN_INT (1)));
++  else
++    emit_insn (gen_mulhisi3v (operands[0], operands[1], operands[2],
++			      GEN_INT (0), GEN_INT (0)));
++  DONE;
++})
++
++(define_expand "smbt"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_mulhisi3v (operands[0], operands[1], operands[2],
++			      GEN_INT (1), GEN_INT (0)));
++  else
++    emit_insn (gen_mulhisi3v (operands[0], operands[1], operands[2],
++			      GEN_INT (0), GEN_INT (1)));
++  DONE;
++})
++
++(define_expand "smtt"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_mulhisi3v (operands[0], operands[1], operands[2],
++			      GEN_INT (0), GEN_INT (0)));
++  else
++    emit_insn (gen_mulhisi3v (operands[0], operands[1], operands[2],
++			      GEN_INT (1), GEN_INT (1)));
++  DONE;
++})
++
++(define_insn "mulhisi3v"
++  [(set (match_operand:SI 0 "register_operand"                         "=   r,    r,    r,    r")
++	(mult:SI
++	  (sign_extend:SI
++	     (vec_select:HI
++	       (match_operand:V2HI 1 "register_operand"                "    r,    r,    r,    r")
++	       (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand"  " Iv00, Iv00, Iv01, Iv01")])))
++	  (sign_extend:SI (vec_select:HI
++	       (match_operand:V2HI 2 "register_operand"                "    r,    r,    r,    r")
++	       (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand"  " Iv00, Iv01, Iv01, Iv00")])))))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "smtt\t%0, %1, %2",
++			     "smbt\t%0, %2, %1",
++			     "smbb\t%0, %1, %2",
++			     "smbt\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "smbb\t%0, %1, %2",
++			     "smbt\t%0, %1, %2",
++			     "smtt\t%0, %1, %2",
++			     "smbt\t%0, %2, %1" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_expand "kmabb"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kma_internal (operands[0], operands[2], operands[3],
++				 GEN_INT (1), GEN_INT (1),
++				 operands[1]));
++  else
++    emit_insn (gen_kma_internal (operands[0], operands[2], operands[3],
++				 GEN_INT (0), GEN_INT (0),
++				 operands[1]));
++  DONE;
++})
++
++(define_expand "kmabt"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kma_internal (operands[0], operands[2], operands[3],
++				 GEN_INT (1), GEN_INT (0),
++				 operands[1]));
++  else
++    emit_insn (gen_kma_internal (operands[0], operands[2], operands[3],
++				 GEN_INT (0), GEN_INT (1),
++				 operands[1]));
++  DONE;
++})
++
++(define_expand "kmatt"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kma_internal (operands[0], operands[2], operands[3],
++				 GEN_INT (0), GEN_INT (0),
++				 operands[1]));
++  else
++    emit_insn (gen_kma_internal (operands[0], operands[2], operands[3],
++				 GEN_INT (1), GEN_INT (1),
++				 operands[1]));
++  DONE;
++})
++
++(define_insn "kma_internal"
++  [(set (match_operand:SI 0 "register_operand"                          "=    r,    r,    r,    r")
++	(ss_plus:SI
++	  (mult:SI
++	    (sign_extend:SI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand"                "    r,    r,    r,    r")
++	        (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand"  " Iv00, Iv00, Iv01, Iv01")])))
++	    (sign_extend:SI
++	      (vec_select:HI
++	        (match_operand:V2HI 2 "register_operand"                "    r,    r,    r,    r")
++	        (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand"  " Iv00, Iv01, Iv01, Iv00")]))))
++	  (match_operand:SI 5 "register_operand"                        "     0,    0,    0,    0")))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "kmatt\t%0, %1, %2",
++			     "kmabt\t%0, %2, %1",
++			     "kmabb\t%0, %1, %2",
++			     "kmabt\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "kmabb\t%0, %1, %2",
++			     "kmabt\t%0, %1, %2",
++			     "kmatt\t%0, %1, %2",
++			     "kmabt\t%0, %2, %1" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"    "dmac")
++   (set_attr "length"   "4")])
++
++(define_expand "smds"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smds_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_smds_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_expand "smds_le"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" " r")
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" " r")
++			      (parallel [(const_int 1)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 0)]))))))]
++  "NDS32_EXT_DSP_P ()"
++{
++})
++
++(define_expand "smds_be"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" " r")
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" " r")
++			      (parallel [(const_int 0)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 1)]))))))]
++  "NDS32_EXT_DSP_P ()"
++{
++})
++
++(define_expand "smdrs"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smdrs_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_smdrs_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++(define_expand "smdrs_le"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" " r")
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" " r")
++			      (parallel [(const_int 0)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 1)]))))))]
++  "NDS32_EXT_DSP_P ()"
++{
++})
++
++(define_expand "smdrs_be"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" " r")
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" " r")
++			      (parallel [(const_int 1)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 0)]))))))]
++  "NDS32_EXT_DSP_P ()"
++{
++})
++
++(define_expand "smxdsv"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:V2HI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smxdsv_be (operands[0], operands[1], operands[2]));
++  else
++    emit_insn (gen_smxdsv_le (operands[0], operands[1], operands[2]));
++  DONE;
++})
++
++
++(define_expand "smxdsv_le"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" " r")
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" " r")
++			      (parallel [(const_int 0)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 1)]))))))]
++  "NDS32_EXT_DSP_P ()"
++{
++})
++
++(define_expand "smxdsv_be"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" " r")
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" " r")
++			      (parallel [(const_int 1)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 0)]))))))]
++  "NDS32_EXT_DSP_P ()"
++{
++})
++
++(define_insn "smal1"
++  [(set (match_operand:DI 0 "register_operand"             "=r")
++	(plus:DI (match_operand:DI 1 "register_operand"    " r")
++	  (sign_extend:DI
++	    (mult:SI
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_operand:V2HI 2 "register_operand" " r")
++		  (parallel [(const_int 0)])))
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_dup 2)
++		  (parallel [(const_int 1)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smal2"
++  [(set (match_operand:DI 0 "register_operand"           "=r")
++	(plus:DI (match_operand:DI 1 "register_operand"  " r")
++	  (mult:DI
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)])))
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))))))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smal3"
++  [(set (match_operand:DI 0 "register_operand"             "=r")
++	(plus:DI (match_operand:DI 1 "register_operand"    " r")
++	  (sign_extend:DI
++	    (mult:SI
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_operand:V2HI 2 "register_operand" " r")
++		  (parallel [(const_int 1)])))
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_dup 2)
++		  (parallel [(const_int 0)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smal4"
++  [(set (match_operand:DI 0 "register_operand"           "=r")
++	(plus:DI (match_operand:DI 1 "register_operand"  " r")
++	  (mult:DI
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)])))
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))))))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smal5"
++  [(set (match_operand:DI 0 "register_operand"             "=r")
++	(plus:DI
++	  (sign_extend:DI
++	    (mult:SI
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_operand:V2HI 2 "register_operand" " r")
++		  (parallel [(const_int 0)])))
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_dup 2)
++		  (parallel [(const_int 1)])))))
++	  (match_operand:DI 1 "register_operand"           " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smal6"
++  [(set (match_operand:DI 0 "register_operand"           "=r")
++	(plus:DI
++	  (mult:DI
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 0)])))
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 1)]))))
++	  (match_operand:DI 1 "register_operand"         " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smal7"
++  [(set (match_operand:DI 0 "register_operand"             "=r")
++	(plus:DI
++	  (sign_extend:DI
++	    (mult:SI
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_operand:V2HI 2 "register_operand" " r")
++		  (parallel [(const_int 1)])))
++	      (sign_extend:SI
++		(vec_select:HI
++		  (match_dup 2)
++		  (parallel [(const_int 0)])))))
++	  (match_operand:DI 1 "register_operand"           " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"    "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smal8"
++  [(set (match_operand:DI 0 "register_operand"           "=r")
++	(plus:DI
++	  (mult:DI
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand" " r")
++		(parallel [(const_int 1)])))
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_dup 2)
++		(parallel [(const_int 0)]))))
++	  (match_operand:DI 1 "register_operand"         " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "smal\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++;; We need this dummy pattern for smal
++(define_insn_and_split "extendsidi2"
++  [(set (match_operand:DI 0 "register_operand" "")
++	(sign_extend:DI (match_operand:SI 1 "nds32_move_operand" "")))]
++  "NDS32_EXT_DSP_P ()"
++  "#"
++  "NDS32_EXT_DSP_P ()"
++  [(const_int 0)]
++{
++  rtx high_part_dst, low_part_dst;
++
++  low_part_dst = nds32_di_low_part_subreg (operands[0]);
++  high_part_dst = nds32_di_high_part_subreg (operands[0]);
++
++  emit_move_insn (low_part_dst, operands[1]);
++  emit_insn (gen_ashrsi3 (high_part_dst, low_part_dst, GEN_INT (31)));
++  DONE;
++}
++  [(set_attr "type"   "alu")
++   (set_attr "length"   "4")])
++
++;; We need this dummy pattern for usmar64/usmsr64
++(define_insn_and_split "zero_extendsidi2"
++  [(set (match_operand:DI 0 "register_operand" "")
++	(zero_extend:DI (match_operand:SI 1 "nds32_move_operand" "")))]
++  "NDS32_EXT_DSP_P ()"
++  "#"
++  "NDS32_EXT_DSP_P ()"
++  [(const_int 0)]
++{
++  rtx high_part_dst, low_part_dst;
++
++  low_part_dst = nds32_di_low_part_subreg (operands[0]);
++  high_part_dst = nds32_di_high_part_subreg (operands[0]);
++
++  emit_move_insn (low_part_dst, operands[1]);
++  emit_move_insn (high_part_dst, const0_rtx);
++  DONE;
++}
++  [(set_attr "type"   "alu")
++   (set_attr "length"   "4")])
++
++(define_insn_and_split "extendhidi2"
++  [(set (match_operand:DI 0 "register_operand" "")
++	(sign_extend:DI (match_operand:HI 1 "nonimmediate_operand" "")))]
++  "NDS32_EXT_DSP_P ()"
++  "#"
++  "NDS32_EXT_DSP_P ()"
++  [(const_int 0)]
++{
++  rtx high_part_dst, low_part_dst;
++
++  low_part_dst = nds32_di_low_part_subreg (operands[0]);
++  high_part_dst = nds32_di_high_part_subreg (operands[0]);
++
++
++  emit_insn (gen_extendhisi2 (low_part_dst, operands[1]));
++  emit_insn (gen_ashrsi3 (high_part_dst, low_part_dst, GEN_INT (31)));
++  DONE;
++}
++  [(set_attr "type"   "alu")
++   (set_attr "length"   "4")])
++
++(define_insn "extendqihi2"
++  [(set (match_operand:HI 0 "register_operand"                 "=r")
++	(sign_extend:HI (match_operand:QI 1 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "sunpkd820\t%0, %1"
++  [(set_attr "type"       "dpack")
++   (set_attr "length"     "4")])
++
++(define_insn "smulsi3_highpart"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (mult:DI
++	      (sign_extend:DI (match_operand:SI 1 "register_operand" " r"))
++	      (sign_extend:DI (match_operand:SI 2 "register_operand" " r")))
++	    (const_int 32))))]
++  "NDS32_EXT_DSP_P ()"
++  "smmul\t%0, %1, %2"
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_insn "smmul_round"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (unspec:DI [(mult:DI
++		  	  (sign_extend:DI (match_operand:SI 1 "register_operand" " r"))
++			  (sign_extend:DI (match_operand:SI 2 "register_operand" " r")))]
++		       UNSPEC_ROUND)
++	    (const_int 32))))]
++  "NDS32_EXT_DSP_P ()"
++  "smmul.u\t%0, %1, %2"
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_insn "kmmac"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(ss_plus:SI (match_operand:SI 1 "register_operand"             " 0")
++	  (truncate:SI
++	    (lshiftrt:DI
++	      (mult:DI
++		(sign_extend:DI (match_operand:SI 2 "register_operand" " r"))
++		(sign_extend:DI (match_operand:SI 3 "register_operand" " r")))
++	      (const_int 32)))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmmac\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmmac_round"
++  [(set (match_operand:SI 0 "register_operand"                                     "=r")
++	(ss_plus:SI (match_operand:SI 1 "register_operand"                         " 0")
++	  (truncate:SI
++	    (lshiftrt:DI
++	      (unspec:DI [(mult:DI
++			    (sign_extend:DI (match_operand:SI 2 "register_operand" " r"))
++			    (sign_extend:DI (match_operand:SI 3 "register_operand" " r")))]
++			 UNSPEC_ROUND)
++	      (const_int 32)))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmmac.u\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmmsb"
++  [(set (match_operand:SI 0 "register_operand"                         "=r")
++	(ss_minus:SI (match_operand:SI 1 "register_operand"            " 0")
++	  (truncate:SI
++	    (lshiftrt:DI
++	      (mult:DI
++		(sign_extend:DI (match_operand:SI 2 "register_operand" " r"))
++		(sign_extend:DI (match_operand:SI 3 "register_operand" " r")))
++	      (const_int 32)))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmmsb\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmmsb_round"
++  [(set (match_operand:SI 0 "register_operand"                                     "=r")
++	(ss_minus:SI (match_operand:SI 1 "register_operand"                        " 0")
++	  (truncate:SI
++	    (lshiftrt:DI
++	      (unspec:DI [(mult:DI
++			    (sign_extend:DI (match_operand:SI 2 "register_operand" " r"))
++			    (sign_extend:DI (match_operand:SI 3 "register_operand" " r")))]
++			 UNSPEC_ROUND)
++	      (const_int 32)))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmmsb.u\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kwmmul"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (ss_mult:DI
++	      (mult:DI (sign_extend:DI (match_operand:SI 1 "register_operand" " r")) (const_int 2))
++	      (mult:DI (sign_extend:DI (match_operand:SI 2 "register_operand" " r")) (const_int 2)))
++	    (const_int 32))))]
++  "NDS32_EXT_DSP_P ()"
++  "kwmmul\t%0, %1, %2"
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_insn "kwmmul_round"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (unspec:DI [
++	      (ss_mult:DI
++		(mult:DI (sign_extend:DI (match_operand:SI 1 "register_operand" " r")) (const_int 2))
++		(mult:DI (sign_extend:DI (match_operand:SI 2 "register_operand" " r")) (const_int 2)))]
++	      UNSPEC_ROUND)
++	    (const_int 32))))]
++  "NDS32_EXT_DSP_P ()"
++  "kwmmul.u\t%0, %1, %2"
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_expand "smmwb"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smulhisi3_highpart_1 (operands[0], operands[1], operands[2], GEN_INT (1)));
++  else
++    emit_insn (gen_smulhisi3_highpart_1 (operands[0], operands[1], operands[2], GEN_INT (0)));
++  DONE;
++})
++
++(define_expand "smmwt"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smulhisi3_highpart_1 (operands[0], operands[1], operands[2], GEN_INT (0)));
++  else
++    emit_insn (gen_smulhisi3_highpart_1 (operands[0], operands[1], operands[2], GEN_INT (1)));
++  DONE;
++})
++
++(define_insn "smulhisi3_highpart_1"
++  [(set (match_operand:SI 0 "register_operand"                           "=   r,    r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (mult:DI
++	      (sign_extend:DI (match_operand:SI 1 "register_operand"     "    r,    r"))
++	      (sign_extend:DI
++	        (vec_select:HI
++		  (match_operand:V2HI 2 "register_operand"               "    r,    r")
++		  (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iv00, Iv01")]))))
++	    (const_int 16))))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "smmwt\t%0, %1, %2",
++			     "smmwb\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "smmwb\t%0, %1, %2",
++			     "smmwt\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_insn "smulhisi3_highpart_2"
++  [(set (match_operand:SI 0 "register_operand"                           "=   r,    r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (mult:DI
++	      (sign_extend:DI
++	        (vec_select:HI
++		  (match_operand:V2HI 1 "register_operand"               "    r,    r")
++		  (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iv00, Iv01")])))
++	      (sign_extend:DI (match_operand:SI 2 "register_operand"     "    r,    r")))
++	    (const_int 16))))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "smmwt\t%0, %1, %2",
++			     "smmwb\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "smmwb\t%0, %1, %2",
++			     "smmwt\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_expand "smmwb_round"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smmw_round_internal (operands[0], operands[1], operands[2], GEN_INT (1)));
++  else
++    emit_insn (gen_smmw_round_internal (operands[0], operands[1], operands[2], GEN_INT (0)));
++  DONE;
++})
++
++(define_expand "smmwt_round"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smmw_round_internal (operands[0], operands[1], operands[2], GEN_INT (0)));
++  else
++    emit_insn (gen_smmw_round_internal (operands[0], operands[1], operands[2], GEN_INT (1)));
++  DONE;
++})
++
++(define_insn "smmw_round_internal"
++  [(set (match_operand:SI 0 "register_operand"                           "=   r,    r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (unspec:DI
++	      [(mult:DI
++		 (sign_extend:DI (match_operand:SI 1 "register_operand"     "    r,    r"))
++		 (sign_extend:DI
++		   (vec_select:HI
++		     (match_operand:V2HI 2 "register_operand"               "    r,    r")
++		     (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iv00, Iv01")]))))]
++	      UNSPEC_ROUND)
++	    (const_int 16))))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "smmwt.u\t%0, %1, %2",
++			     "smmwb.u\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "smmwb.u\t%0, %1, %2",
++			     "smmwt.u\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmul")
++   (set_attr "length"   "4")])
++
++(define_expand "kmmawb"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:SI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kmmaw_internal (operands[0], operands[2], operands[3], GEN_INT (1), operands[1]));
++  else
++    emit_insn (gen_kmmaw_internal (operands[0], operands[2], operands[3], GEN_INT (0), operands[1]));
++  DONE;
++})
++
++(define_expand "kmmawt"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:SI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kmmaw_internal (operands[0], operands[2], operands[3], GEN_INT (0), operands[1]));
++  else
++    emit_insn (gen_kmmaw_internal (operands[0], operands[2], operands[3], GEN_INT (1), operands[1]));
++  DONE;
++})
++
++(define_insn "kmmaw_internal"
++  [(set (match_operand:SI 0 "register_operand"                               "=   r,    r")
++	(ss_plus:SI
++	  (match_operand:SI 4 "register_operand"                             "    0,    0")
++	  (truncate:SI
++	    (lshiftrt:DI
++	      (mult:DI
++		(sign_extend:DI (match_operand:SI 1 "register_operand"       "    r,    r"))
++		  (sign_extend:DI
++		    (vec_select:HI
++		      (match_operand:V2HI 2 "register_operand"               "    r,    r")
++		      (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iv00, Iv01")]))))
++	      (const_int 16)))))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "kmmawt\t%0, %1, %2",
++			     "kmmawb\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "kmmawb\t%0, %1, %2",
++			     "kmmawt\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_expand "kmmawb_round"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:SI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kmmaw_round_internal (operands[0], operands[2], operands[3], GEN_INT (1), operands[1]));
++  else
++    emit_insn (gen_kmmaw_round_internal (operands[0], operands[2], operands[3], GEN_INT (0), operands[1]));
++  DONE;
++}
++  [(set_attr "type"   "alu")
++   (set_attr "length"   "4")])
++
++(define_expand "kmmawt_round"
++  [(match_operand:SI 0 "register_operand" "")
++   (match_operand:SI 1 "register_operand" "")
++   (match_operand:SI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_kmmaw_round_internal (operands[0], operands[2], operands[3], GEN_INT (0), operands[1]));
++  else
++    emit_insn (gen_kmmaw_round_internal (operands[0], operands[2], operands[3], GEN_INT (1), operands[1]));
++  DONE;
++}
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++
++(define_insn "kmmaw_round_internal"
++  [(set (match_operand:SI 0 "register_operand"                                "=   r,    r")
++	(ss_plus:SI
++	  (match_operand:SI 4 "register_operand"                              "    0,    0")
++	  (truncate:SI
++	    (lshiftrt:DI
++	      (unspec:DI
++		[(mult:DI
++		   (sign_extend:DI (match_operand:SI 1 "register_operand"     "    r,    r"))
++		   (sign_extend:DI
++		     (vec_select:HI
++		       (match_operand:V2HI 2 "register_operand"               "    r,    r")
++		       (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iv00, Iv01")]))))]
++		UNSPEC_ROUND)
++	      (const_int 16)))))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "kmmawt.u\t%0, %1, %2",
++			     "kmmawb.u\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "kmmawb.u\t%0, %1, %2",
++			     "kmmawt.u\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_expand "smalbb"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smaddhidi (operands[0], operands[2],
++			      operands[3], operands[1],
++			      GEN_INT (1), GEN_INT (1)));
++  else
++    emit_insn (gen_smaddhidi (operands[0], operands[2],
++			      operands[3], operands[1],
++			      GEN_INT (0), GEN_INT (0)));
++  DONE;
++})
++
++(define_expand "smalbt"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smaddhidi (operands[0], operands[2],
++			      operands[3], operands[1],
++			      GEN_INT (1), GEN_INT (0)));
++  else
++    emit_insn (gen_smaddhidi (operands[0], operands[2],
++			      operands[3], operands[1],
++			      GEN_INT (0), GEN_INT (1)));
++  DONE;
++})
++
++(define_expand "smaltt"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" "")
++   (match_operand:V2HI 3 "register_operand" "")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smaddhidi (operands[0], operands[2],
++			      operands[3], operands[1],
++			      GEN_INT (0), GEN_INT (0)));
++  else
++    emit_insn (gen_smaddhidi (operands[0], operands[2],
++			      operands[3], operands[1],
++			      GEN_INT (1), GEN_INT (1)));
++  DONE;
++})
++
++(define_insn "smaddhidi"
++  [(set (match_operand:DI 0 "register_operand"                         "=   r,    r,    r,    r")
++	(plus:DI
++	  (match_operand:DI 3 "register_operand"                       "    0,    0,    0,    0")
++	  (mult:DI
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand"               "    r,    r,    r,    r")
++		(parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iv00, Iv00, Iv01, Iv01")])))
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand"               "    r,    r,    r,    r")
++		(parallel [(match_operand:SI 5 "nds32_imm_0_1_operand" " Iv00, Iv01, Iv01, Iv00")]))))))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "smaltt\t%0, %1, %2",
++			     "smalbt\t%0, %2, %1",
++			     "smalbb\t%0, %1, %2",
++			     "smalbt\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "smalbb\t%0, %1, %2",
++			     "smalbt\t%0, %1, %2",
++			     "smaltt\t%0, %1, %2",
++			     "smalbt\t%0, %2, %1" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smaddhidi2"
++  [(set (match_operand:DI 0 "register_operand"                         "=   r,    r,    r,    r")
++	(plus:DI
++	  (mult:DI
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 1 "register_operand"               "    r,    r,    r,    r")
++		(parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iv00, Iv00, Iv01, Iv01")])))
++	    (sign_extend:DI
++	      (vec_select:HI
++		(match_operand:V2HI 2 "register_operand"               "    r,    r,    r,    r")
++		(parallel [(match_operand:SI 5 "nds32_imm_0_1_operand" " Iv00, Iv01, Iv01, Iv00")]))))
++	  (match_operand:DI 3 "register_operand"                       "    0,    0,    0,    0")))]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    {
++      const char *pats[] = { "smaltt\t%0, %1, %2",
++			     "smalbt\t%0, %2, %1",
++			     "smalbb\t%0, %1, %2",
++			     "smalbt\t%0, %1, %2" };
++      return pats[which_alternative];
++    }
++  else
++    {
++      const char *pats[] = { "smalbb\t%0, %1, %2",
++			     "smalbt\t%0, %1, %2",
++			     "smaltt\t%0, %1, %2",
++			     "smalbt\t%0, %2, %1" };
++      return pats[which_alternative];
++    }
++}
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_expand "smalda1"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" " r")
++   (match_operand:V2HI 3 "register_operand" " r")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smalda1_be (operands[0], operands[1], operands[2], operands[3]));
++  else
++    emit_insn (gen_smalda1_le (operands[0], operands[1], operands[2], operands[3]));
++  DONE;
++})
++
++(define_expand "smalds1"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" " r")
++   (match_operand:V2HI 3 "register_operand" " r")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smalds1_be (operands[0], operands[1], operands[2], operands[3]));
++  else
++    emit_insn (gen_smalds1_le (operands[0], operands[1], operands[2], operands[3]));
++  DONE;
++})
++
++(define_insn "smalda1_le"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (plus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 1)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 0)]))))))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "smalda\t%0, %2, %3"
++  [(set_attr "type"    "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smalds1_le"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (minus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 1)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 0)]))))))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "smalds\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smalda1_be"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (plus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 0)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 1)]))))))))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "smalda\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smalds1_be"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (minus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 0)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 1)]))))))))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "smalds\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_expand "smaldrs3"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" " r")
++   (match_operand:V2HI 3 "register_operand" " r")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smaldrs3_be (operands[0], operands[1], operands[2], operands[3]));
++  else
++    emit_insn (gen_smaldrs3_le (operands[0], operands[1], operands[2], operands[3]));
++  DONE;
++})
++
++(define_insn "smaldrs3_le"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (minus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 0)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 1)]))))))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "smaldrs\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smaldrs3_be"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (minus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 1)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 0)]))))))))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "smaldrs\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_expand "smalxda1"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" " r")
++   (match_operand:V2HI 3 "register_operand" " r")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smalxda1_be (operands[0], operands[1], operands[2], operands[3]));
++  else
++    emit_insn (gen_smalxda1_le (operands[0], operands[1], operands[2], operands[3]));
++  DONE;
++})
++
++(define_expand "smalxds1"
++  [(match_operand:DI 0 "register_operand" "")
++   (match_operand:DI 1 "register_operand" "")
++   (match_operand:V2HI 2 "register_operand" " r")
++   (match_operand:V2HI 3 "register_operand" " r")]
++  "NDS32_EXT_DSP_P ()"
++{
++  if (TARGET_BIG_ENDIAN)
++    emit_insn (gen_smalxds1_be (operands[0], operands[1], operands[2], operands[3]));
++  else
++    emit_insn (gen_smalxds1_le (operands[0], operands[1], operands[2], operands[3]));
++  DONE;
++})
++
++(define_insn "smalxd<add_sub>1_le"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (plus_minus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 0)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 1)]))))))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "smalxd<add_sub>\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++
++(define_insn "smalxd<add_sub>1_be"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"                           " 0")
++	  (sign_extend:DI
++	    (plus_minus:SI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 0)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 1)]))))
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 2)
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_dup 3)
++				  (parallel [(const_int 0)]))))))))]
++  "NDS32_EXT_DSP_P () && TARGET_BIG_ENDIAN"
++  "smalxd<add_sub>\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smslda1"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(minus:DI
++	  (minus:DI
++	    (match_operand:DI 1 "register_operand"                           " 0")
++	    (sign_extend:DI
++	      (mult:SI
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 2 "register_operand" " r")
++				  (parallel [(const_int 1)])))
++		(sign_extend:SI (vec_select:HI
++				  (match_operand:V2HI 3 "register_operand" " r")
++				  (parallel [(const_int 1)]))))))
++	  (sign_extend:DI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 0)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "smslda\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "smslxda1"
++  [(set (match_operand:DI 0 "register_operand"                             "=r")
++	(minus:DI
++	  (minus:DI
++	    (match_operand:DI 1 "register_operand"                           " 0")
++	      (sign_extend:DI
++		(mult:SI
++		  (sign_extend:SI (vec_select:HI
++				    (match_operand:V2HI 2 "register_operand" " r")
++				    (parallel [(const_int 1)])))
++		  (sign_extend:SI (vec_select:HI
++				    (match_operand:V2HI 3 "register_operand" " r")
++				    (parallel [(const_int 0)]))))))
++	  (sign_extend:DI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 1)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "smslxda\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++;; mada for synthetize smalda
++(define_insn_and_split "mada1"
++  [(set (match_operand:SI 0 "register_operand"                          "=r")
++	(plus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" "r")
++			      (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" "r")
++			      (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iu01")]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(match_operand:SI 5 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(match_operand:SI 6 "nds32_imm_0_1_operand" " Iu01")]))))))]
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx result0 = gen_reg_rtx (SImode);
++  rtx result1 = gen_reg_rtx (SImode);
++  emit_insn (gen_mulhisi3v (result0, operands[1], operands[2],
++			    operands[3], operands[4]));
++  emit_insn (gen_mulhisi3v (result1, operands[1], operands[2],
++			    operands[5], operands[6]));
++  emit_insn (gen_addsi3 (operands[0], result0, result1));
++  DONE;
++})
++
++(define_insn_and_split "mada2"
++  [(set (match_operand:SI 0 "register_operand"                          "=r")
++	(plus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" "r")
++			      (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" "r")
++			      (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iu01")]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(match_operand:SI 5 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(match_operand:SI 6 "nds32_imm_0_1_operand" " Iu01")]))))))]
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 1)]
++{
++  rtx result0 = gen_reg_rtx (SImode);
++  rtx result1 = gen_reg_rtx (SImode);
++  emit_insn (gen_mulhisi3v (result0, operands[1], operands[2],
++			    operands[3], operands[4]));
++  emit_insn (gen_mulhisi3v (result1, operands[1], operands[2],
++			    operands[6], operands[5]));
++  emit_insn (gen_addsi3 (operands[0], result0, result1));
++  DONE;
++})
++
++;; sms for synthetize smalds
++(define_insn_and_split "sms1"
++  [(set (match_operand:SI 0 "register_operand"                                       "=   r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand"               "    r")
++			      (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand"               "    r")
++			      (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iu01")]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(match_operand:SI 5 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(match_operand:SI 6 "nds32_imm_0_1_operand" " Iu01")]))))))]
++  "NDS32_EXT_DSP_P ()
++   && (!reload_completed
++       || !nds32_need_split_sms_p (operands[3], operands[4],
++				   operands[5], operands[6]))"
++
++{
++  return nds32_output_sms (operands[3], operands[4],
++			   operands[5], operands[6]);
++}
++  "NDS32_EXT_DSP_P ()
++   && !reload_completed
++   && nds32_need_split_sms_p (operands[3], operands[4],
++			      operands[5], operands[6])"
++  [(const_int 1)]
++{
++  nds32_split_sms (operands[0], operands[1], operands[2],
++		   operands[3], operands[4],
++		   operands[5], operands[6]);
++  DONE;
++}
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn_and_split "sms2"
++  [(set (match_operand:SI 0 "register_operand"                                       "=   r")
++	(minus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand"               "    r")
++			      (parallel [(match_operand:SI 3 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand"               "    r")
++			      (parallel [(match_operand:SI 4 "nds32_imm_0_1_operand" " Iu01")]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(match_operand:SI 5 "nds32_imm_0_1_operand" " Iu01")])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(match_operand:SI 6 "nds32_imm_0_1_operand" " Iu01")]))))))]
++  "NDS32_EXT_DSP_P ()
++   && (!reload_completed
++       || !nds32_need_split_sms_p (operands[3], operands[4],
++				   operands[6], operands[5]))"
++{
++  return nds32_output_sms (operands[3], operands[4],
++			   operands[6], operands[5]);
++}
++  "NDS32_EXT_DSP_P ()
++   && !reload_completed
++   && nds32_need_split_sms_p (operands[3], operands[4],
++			      operands[6], operands[5])"
++  [(const_int 1)]
++{
++  nds32_split_sms (operands[0], operands[1], operands[2],
++		   operands[3], operands[4],
++		   operands[6], operands[5]);
++  DONE;
++}
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmda"
++  [(set (match_operand:SI 0 "register_operand"                          "=r")
++	(ss_plus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" "r")
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" "r")
++			      (parallel [(const_int 1)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 0)]))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmda\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmxda"
++  [(set (match_operand:SI 0 "register_operand"                          "=r")
++	(ss_plus:SI
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 1 "register_operand" "r")
++			      (parallel [(const_int 1)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_operand:V2HI 2 "register_operand" "r")
++			      (parallel [(const_int 0)]))))
++	  (mult:SI
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 1)
++			      (parallel [(const_int 0)])))
++	    (sign_extend:SI (vec_select:HI
++			      (match_dup 2)
++			      (parallel [(const_int 1)]))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmxda\t%0, %1, %2"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmada"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_plus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_plus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 1)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 0)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmada\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmada2"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_plus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_plus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 0)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 1)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmada\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmaxda"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_plus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_plus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 0)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 1)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmaxda\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmads"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_plus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_minus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 1)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 0)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmads\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmadrs"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_plus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_minus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 0)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 1)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmadrs\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmaxds"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_plus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_minus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 0)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 1)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmaxds\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmsda"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_minus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_minus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 1)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 0)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmsda\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmsxda"
++  [(set (match_operand:SI 0 "register_operand"                           "=r")
++	(ss_minus:SI
++	  (match_operand:SI 1 "register_operand"                         " 0")
++	  (ss_minus:SI
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 2 "register_operand" " r")
++				(parallel [(const_int 1)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_operand:V2HI 3 "register_operand" " r")
++				(parallel [(const_int 0)]))))
++	    (mult:SI
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 2)
++				(parallel [(const_int 0)])))
++	      (sign_extend:SI (vec_select:HI
++				(match_dup 3)
++				(parallel [(const_int 1)])))))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmsxda\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++;; smax[8|16] and umax[8|16]
++(define_insn "<opcode><mode>3"
++  [(set (match_operand:VQIHI 0 "register_operand"               "=r")
++	(sumax:VQIHI (match_operand:VQIHI 1 "register_operand" " r")
++		     (match_operand:VQIHI 2 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "<opcode><bits>\t%0, %1, %2"
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++;; smin[8|16] and umin[8|16]
++(define_insn "<opcode><mode>3"
++  [(set (match_operand:VQIHI 0 "register_operand"              "=r")
++	(sumin:VQIHI (match_operand:VQIHI 1 "register_operand" " r")
++		     (match_operand:VQIHI 2 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "<opcode><bits>\t%0, %1, %2"
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_insn "<opcode><mode>3_bb"
++  [(set (match_operand:<VELT> 0 "register_operand"                    "=r")
++	(sumin_max:<VELT> (vec_select:<VELT>
++			    (match_operand:VQIHI 1 "register_operand" " r")
++			    (parallel [(const_int 0)]))
++			  (vec_select:<VELT>
++			    (match_operand:VQIHI 2 "register_operand" " r")
++			    (parallel [(const_int 0)]))))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "<opcode><bits>\t%0, %1, %2"
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_insn_and_split "<opcode><mode>3_tt"
++  [(set (match_operand:<VELT> 0 "register_operand"                    "=r")
++	(sumin_max:<VELT> (vec_select:<VELT>
++			    (match_operand:VQIHI 1 "register_operand" " r")
++			    (parallel [(const_int 1)]))
++			  (vec_select:<VELT>
++			    (match_operand:VQIHI 2 "register_operand" " r")
++			    (parallel [(const_int 1)]))))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 0)]
++{
++  rtx tmp = gen_reg_rtx (<MODE>mode);
++  emit_insn (gen_<opcode><mode>3 (tmp, operands[1], operands[2]));
++  emit_insn (gen_rotr<mode>_1 (tmp, tmp));
++  emit_move_insn (operands[0], simplify_gen_subreg (<VELT>mode, tmp, <MODE>mode, 0));
++  DONE;
++}
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_insn_and_split "<opcode>v4qi3_22"
++  [(set (match_operand:QI 0 "register_operand"                   "=r")
++	(sumin_max:QI (vec_select:QI
++			(match_operand:V4QI 1 "register_operand" " r")
++			(parallel [(const_int 2)]))
++		      (vec_select:QI
++			(match_operand:V4QI 2 "register_operand" " r")
++			(parallel [(const_int 2)]))))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 0)]
++{
++  rtx tmp = gen_reg_rtx (V4QImode);
++  emit_insn (gen_<opcode>v4qi3 (tmp, operands[1], operands[2]));
++  emit_insn (gen_rotrv4qi_2 (tmp, tmp));
++  emit_move_insn (operands[0], simplify_gen_subreg (QImode, tmp, V4QImode, 0));
++  DONE;
++}
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_insn_and_split "<opcode>v4qi3_33"
++  [(set (match_operand:QI 0 "register_operand"                   "=r")
++	(sumin_max:QI (vec_select:QI
++			(match_operand:V4QI 1 "register_operand" " r")
++			(parallel [(const_int 3)]))
++		      (vec_select:QI
++			(match_operand:V4QI 2 "register_operand" " r")
++			(parallel [(const_int 3)]))))]
++  "NDS32_EXT_DSP_P () && !reload_completed && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 0)]
++{
++  rtx tmp = gen_reg_rtx (V4QImode);
++  emit_insn (gen_<opcode>v4qi3 (tmp, operands[1], operands[2]));
++  emit_insn (gen_rotrv4qi_3 (tmp, tmp));
++  emit_move_insn (operands[0], simplify_gen_subreg (QImode, tmp, V4QImode, 0));
++  DONE;
++}
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_insn_and_split "<opcode>v2hi3_bbtt"
++  [(set (match_operand:V2HI 0 "register_operand"                         "=r")
++	(vec_merge:V2HI
++	  (vec_duplicate:V2HI
++	    (sumin_max:HI (vec_select:HI
++			    (match_operand:V2HI 1 "register_operand" " r")
++			    (parallel [(const_int 1)]))
++			  (vec_select:HI
++			    (match_operand:V2HI 2 "register_operand" " r")
++			    (parallel [(const_int 1)]))))
++	  (vec_duplicate:V2HI
++	    (sumin_max:HI (vec_select:HI
++			    (match_dup:V2HI 1)
++			    (parallel [(const_int 0)]))
++			  (vec_select:HI
++			    (match_dup:V2HI 2)
++			    (parallel [(const_int 0)]))))
++	  (const_int 2)))]
++  "NDS32_EXT_DSP_P () && !TARGET_BIG_ENDIAN"
++  "#"
++  "NDS32_EXT_DSP_P ()"
++  [(const_int 0)]
++{
++  emit_insn (gen_<opcode>v2hi3 (operands[0], operands[1], operands[2]));
++  DONE;
++}
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_expand "abs<mode>2"
++  [(set (match_operand:VQIHI 0 "register_operand"                "=r")
++	(ss_abs:VQIHI (match_operand:VQIHI 1 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P () && TARGET_HW_ABS && !flag_wrapv"
++{
++})
++
++(define_insn "kabs<mode>2"
++  [(set (match_operand:VQIHI 0 "register_operand"                "=r")
++	(ss_abs:VQIHI (match_operand:VQIHI 1 "register_operand" " r")))]
++  "NDS32_EXT_DSP_P ()"
++  "kabs<bits>\t%0, %1"
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_insn "<su>mar64_1"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (mult:DI
++	    (extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>mar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "<su>mar64_2"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(plus:DI
++	  (mult:DI
++	    (extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))
++	  (match_operand:DI 1 "register_operand"     " 0")))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>mar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "<su>mar64_3"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(plus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (extend:DI
++	    (mult:SI
++	      (match_operand:SI 2 "register_operand" " r")
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>mar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "<su>mar64_4"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(plus:DI
++	  (extend:DI
++	  (mult:SI
++	      (match_operand:SI 2 "register_operand" " r")
++	      (match_operand:SI 3 "register_operand" " r")))
++	  (match_operand:DI 1 "register_operand"     " 0")))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>mar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "<su>msr64"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(minus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (mult:DI
++	    (extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>msr64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "<su>msr64_2"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(minus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (extend:DI
++	    (mult:SI
++	      (match_operand:SI 2 "register_operand" " r")
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "<su>msr64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++;; kmar64, kmsr64, ukmar64 and ukmsr64
++(define_insn "kmar64_1"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(ss_plus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (mult:DI
++	    (sign_extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (sign_extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmar64_2"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(ss_plus:DI
++	  (mult:DI
++	    (sign_extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (sign_extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))
++	  (match_operand:DI 1 "register_operand"     " 0")))]
++  "NDS32_EXT_DSP_P ()"
++  "kmar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "kmsr64"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(ss_minus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (mult:DI
++	    (sign_extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (sign_extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "kmsr64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "ukmar64_1"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(us_plus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (mult:DI
++	    (zero_extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (zero_extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "ukmar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "ukmar64_2"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(us_plus:DI
++	  (mult:DI
++	    (zero_extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (zero_extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))
++	  (match_operand:DI 1 "register_operand"     " 0")))]
++  "NDS32_EXT_DSP_P ()"
++  "ukmar64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "ukmsr64"
++  [(set (match_operand:DI 0 "register_operand"       "=r")
++	(us_minus:DI
++	  (match_operand:DI 1 "register_operand"     " 0")
++	  (mult:DI
++	    (zero_extend:DI
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (zero_extend:DI
++	      (match_operand:SI 3 "register_operand" " r")))))]
++  "NDS32_EXT_DSP_P ()"
++  "ukmsr64\t%0, %2, %3"
++  [(set_attr "type"     "dmac")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick1"
++  [(set (match_operand:SI 0 "register_operand"       "=r")
++	  (ior:SI
++	    (and:SI
++	      (match_operand:SI 1 "register_operand" " r")
++	      (match_operand:SI 3 "register_operand" " r"))
++	    (and:SI
++	      (match_operand:SI 2 "register_operand" " r")
++	      (not:SI (match_dup 3)))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %1, %2, %3"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick2"
++  [(set (match_operand:SI 0 "register_operand"       "=r")
++	  (ior:SI
++	    (and:SI
++	      (match_operand:SI 1 "register_operand" " r")
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (and:SI
++	      (not:SI (match_dup 2))
++	      (match_operand:SI 3 "register_operand" " r"))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %1, %3, %2"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick3"
++  [(set (match_operand:SI 0 "register_operand"       "=r")
++	  (ior:SI
++	    (and:SI
++	      (match_operand:SI 1 "register_operand" " r")
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (and:SI
++	      (match_operand:SI 3 "register_operand" " r")
++	      (not:SI (match_dup 1)))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %2, %3, %1"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick4"
++  [(set (match_operand:SI 0 "register_operand"       "=r")
++	  (ior:SI
++	    (and:SI
++	      (match_operand:SI 1 "register_operand" " r")
++	      (match_operand:SI 2 "register_operand" " r"))
++	    (and:SI
++	      (not:SI (match_dup 1))
++	      (match_operand:SI 3 "register_operand" " r"))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %2, %3, %1"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick5"
++  [(set (match_operand:SI 0 "register_operand"               "=r")
++	  (ior:SI
++	    (and:SI
++	      (match_operand:SI 1 "register_operand"         " r")
++	      (not:SI (match_operand:SI 2 "register_operand" " r")))
++	    (and:SI
++	      (match_operand:SI 3 "register_operand"         " r")
++	      (match_dup 2))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %3, %1, %2"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick6"
++  [(set (match_operand:SI 0 "register_operand"               "=r")
++	  (ior:SI
++	    (and:SI
++	      (not:SI (match_operand:SI 1 "register_operand" " r"))
++	      (match_operand:SI 2 "register_operand"         " r"))
++	    (and:SI
++	      (match_operand:SI 3 "register_operand" " r")
++	      (match_dup 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %3, %2, %1"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick7"
++  [(set (match_operand:SI 0 "register_operand"               "=r")
++	  (ior:SI
++	    (and:SI
++	      (match_operand:SI 1 "register_operand"         " r")
++	      (not:SI (match_operand:SI 2 "register_operand" " r")))
++	    (and:SI
++	      (match_dup 2)
++	      (match_operand:SI 3 "register_operand"         " r"))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %3, %1, %2"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "bpick8"
++  [(set (match_operand:SI 0 "register_operand"               "=r")
++	  (ior:SI
++	    (and:SI
++	      (not:SI (match_operand:SI 1 "register_operand" " r"))
++	      (match_operand:SI 2 "register_operand"         " r"))
++	    (and:SI
++	      (match_dup 1)
++	      (match_operand:SI 3 "register_operand"         " r"))))]
++  "NDS32_EXT_DSP_P ()"
++  "bpick\t%0, %3, %2, %1"
++  [(set_attr "type"     "dbpick")
++   (set_attr "length"   "4")])
++
++(define_insn "sraiu"
++  [(set (match_operand:SI 0 "register_operand"                              "=   r, r")
++	(unspec:SI [(ashiftrt:SI (match_operand:SI 1 "register_operand"     "    r, r")
++				 (match_operand:SI 2 "nds32_rimm5u_operand" " Iu05, r"))]
++		    UNSPEC_ROUND))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   srai.u\t%0, %1, %2
++   sra.u\t%0, %1, %2"
++  [(set_attr "type"   "daluround")
++   (set_attr "length" "4")])
++
++(define_insn "kssl"
++  [(set (match_operand:SI 0 "register_operand"                   "=   r, r")
++	(ss_ashift:SI (match_operand:SI 1 "register_operand"     "    r, r")
++		      (match_operand:SI 2 "nds32_rimm5u_operand" " Iu05, r")))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   kslli\t%0, %1, %2
++   ksll\t%0, %1, %2"
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")])
++
++(define_insn "kslraw_round"
++  [(set (match_operand:SI 0 "register_operand"                  "=r")
++	(if_then_else:SI
++	  (lt:SI (match_operand:SI 2 "register_operand"        " r")
++		 (const_int 0))
++	  (unspec:SI [(ashiftrt:SI (match_operand:SI 1 "register_operand" " r")
++				   (neg:SI (match_dup 2)))]
++		     UNSPEC_ROUND)
++	  (ss_ashift:SI (match_dup 1)
++			(match_dup 2))))]
++  "NDS32_EXT_DSP_P ()"
++  "kslraw.u\t%0, %1, %2"
++  [(set_attr "type"    "daluround")
++   (set_attr "length"  "4")])
++
++(define_insn_and_split "<shift>di3"
++  [(set (match_operand:DI 0 "register_operand" "")
++	(shift_rotate:DI (match_operand:DI 1 "register_operand" "")
++			 (match_operand:SI 2 "nds32_rimm6u_operand" "")))]
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  "#"
++  "NDS32_EXT_DSP_P () && !reload_completed"
++  [(const_int 0)]
++{
++  if (REGNO (operands[0]) == REGNO (operands[1]))
++    {
++      rtx tmp = gen_reg_rtx (DImode);
++      nds32_split_<code>di3 (tmp, operands[1], operands[2]);
++      emit_move_insn (operands[0], tmp);
++    }
++  else
++    nds32_split_<code>di3 (operands[0], operands[1], operands[2]);
++  DONE;
++})
++
++(define_insn "sclip32"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "r")
++		    (match_operand:SI 2 "immediate_operand" "i")] UNSPEC_CLIPS_OV))]
++  "NDS32_EXT_DSP_P ()"
++  "sclip32\t%0, %1, %2"
++  [(set_attr "type"   "dclip")
++   (set_attr "length" "4")]
++)
++
++(define_insn "uclip32"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "r")
++		    (match_operand:SI 2 "immediate_operand" "i")] UNSPEC_CLIP_OV))]
++  "NDS32_EXT_DSP_P ()"
++  "uclip32\t%0, %1, %2"
++  [(set_attr "type"   "dclip")
++   (set_attr "length" "4")]
++)
++
++(define_insn "bitrev"
++  [(set (match_operand:SI 0 "register_operand"                 "=r,    r")
++	(unspec:SI [(match_operand:SI 1 "register_operand"     " r,    r")
++		    (match_operand:SI 2 "nds32_rimm5u_operand" " r, Iu05")]
++		   UNSPEC_BITREV))]
++  ""
++  "@
++   bitrev\t%0, %1, %2
++   bitrevi\t%0, %1, %2"
++  [(set_attr "type"   "dalu")
++   (set_attr "length" "4")]
++)
++
++;; wext, wexti
++(define_insn "<su>wext"
++  [(set (match_operand:SI 0 "register_operand"                "=r,   r")
++	(truncate:SI
++	  (shiftrt:DI
++	    (match_operand:DI 1 "register_operand"            " r,   r")
++	    (match_operand:SI 2 "nds32_rimm5u_operand"        " r,Iu05"))))]
++  "NDS32_EXT_DSP_P ()"
++  "@
++   wext\t%0, %1, %2
++   wexti\t%0, %1, %2"
++  [(set_attr "type"     "dwext")
++   (set_attr "length"   "4")])
++
++;; 32-bit add/sub instruction: raddw and rsubw.
++(define_insn "r<opcode>si3"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(truncate:SI
++	  (ashiftrt:DI
++	    (plus_minus:DI
++	      (sign_extend:DI (match_operand:SI 1 "register_operand" " r"))
++	      (sign_extend:DI (match_operand:SI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "r<opcode>w\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")])
++
++;; 32-bit add/sub instruction: uraddw and ursubw.
++(define_insn "ur<opcode>si3"
++  [(set (match_operand:SI 0 "register_operand"                       "=r")
++	(truncate:SI
++	  (lshiftrt:DI
++	    (plus_minus:DI
++	      (zero_extend:DI (match_operand:SI 1 "register_operand" " r"))
++	      (zero_extend:DI (match_operand:SI 2 "register_operand" " r")))
++	    (const_int 1))))]
++  "NDS32_EXT_DSP_P ()"
++  "ur<opcode>w\t%0, %1, %2"
++  [(set_attr "type"    "dalu")
++   (set_attr "length"  "4")])
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-elf.opt gcc-8.2.0/gcc/config/nds32/nds32-elf.opt
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-elf.opt	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-elf.opt	2019-01-25 15:38:32.825242648 +0100
+@@ -0,0 +1,16 @@
++mcmodel=
++Target RejectNegative Joined Enum(nds32_cmodel_type) Var(nds32_cmodel_option) Init(CMODEL_MEDIUM)
++Specify the address generation strategy for code model.
++
++Enum
++Name(nds32_cmodel_type) Type(enum nds32_cmodel_type)
++Known cmodel types (for use with the -mcmodel= option):
++
++EnumValue
++Enum(nds32_cmodel_type) String(small) Value(CMODEL_SMALL)
++
++EnumValue
++Enum(nds32_cmodel_type) String(medium) Value(CMODEL_MEDIUM)
++
++EnumValue
++Enum(nds32_cmodel_type) String(large) Value(CMODEL_LARGE)
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-fp-as-gp.c gcc-8.2.0/gcc/config/nds32/nds32-fp-as-gp.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-fp-as-gp.c	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-fp-as-gp.c	2019-01-25 15:38:32.825242648 +0100
+@@ -26,19 +26,256 @@
+ #include "system.h"
+ #include "coretypes.h"
+ #include "backend.h"
++#include "hard-reg-set.h"
++#include "tm_p.h"
++#include "rtl.h"
++#include "memmodel.h"
++#include "emit-rtl.h"
++#include "insn-config.h"
++#include "regs.h"
++#include "hard-reg-set.h"
++#include "ira.h"
++#include "ira-int.h"
++#include "df.h"
++#include "tree-core.h"
++#include "tree-pass.h"
++#include "nds32-protos.h"
+ 
+ /* ------------------------------------------------------------------------ */
+ 
++/* A helper function to check if this function should contain prologue.  */
++static bool
++nds32_have_prologue_p (void)
++{
++  int i;
++
++  for (i = 0; i < 28; i++)
++    if (NDS32_REQUIRED_CALLEE_SAVED_P (i))
++      return true;
++
++  return (flag_pic
++	  || NDS32_REQUIRED_CALLEE_SAVED_P (FP_REGNUM)
++	  || NDS32_REQUIRED_CALLEE_SAVED_P (LP_REGNUM));
++}
++
++static int
++nds32_get_symbol_count (void)
++{
++  int symbol_count = 0;
++  rtx_insn *insn;
++  basic_block bb;
++
++  FOR_EACH_BB_FN (bb, cfun)
++    {
++      FOR_BB_INSNS (bb, insn)
++	{
++	  /* Counting the insn number which the addressing mode is symbol.  */
++	  if (single_set (insn) && nds32_symbol_load_store_p (insn))
++	    {
++	      rtx pattern = PATTERN (insn);
++	      rtx mem;
++	      gcc_assert (GET_CODE (pattern) == SET);
++	      if (GET_CODE (SET_SRC (pattern)) == REG )
++		mem = SET_DEST (pattern);
++	      else
++		mem = SET_SRC (pattern);
++
++	      /* We have only lwi37 and swi37 for fp-as-gp optimization,
++		 so don't count any other than SImode.
++		 MEM for QImode and HImode will wrap by ZERO_EXTEND
++		 or SIGN_EXTEND */
++	      if (GET_CODE (mem) == MEM)
++		symbol_count++;
++	    }
++	}
++    }
++
++  return symbol_count;
++}
++
+ /* Function to determine whether it is worth to do fp_as_gp optimization.
+-   Return 0: It is NOT worth to do fp_as_gp optimization.
+-   Return 1: It is APPROXIMATELY worth to do fp_as_gp optimization.
++   Return false: It is NOT worth to do fp_as_gp optimization.
++   Return true: It is APPROXIMATELY worth to do fp_as_gp optimization.
+    Note that if it is worth to do fp_as_gp optimization,
+    we MUST set FP_REGNUM ever live in this function.  */
+-int
++static bool
+ nds32_fp_as_gp_check_available (void)
+ {
+-  /* By default we return 0.  */
+-  return 0;
++  basic_block bb;
++  basic_block exit_bb;
++  edge_iterator ei;
++  edge e;
++  bool first_exit_blocks_p;
++
++  /* If there exists ANY of following conditions,
++     we DO NOT perform fp_as_gp optimization:
++       1. TARGET_FORBID_FP_AS_GP is set
++	  regardless of the TARGET_FORCE_FP_AS_GP.
++       2. User explicitly uses 'naked'/'no_prologue' attribute.
++	  We use nds32_naked_function_p() to help such checking.
++       3. Not optimize for size.
++       4. Need frame pointer.
++       5. If $fp is already required to be saved,
++	  it means $fp is already choosen by register allocator.
++	  Thus we better not to use it for fp_as_gp optimization.
++       6. This function is a vararg function.
++	  DO NOT apply fp_as_gp optimization on this function
++	  because it may change and break stack frame.
++       7. The epilogue is empty.
++	  This happens when the function uses exit()
++	  or its attribute is no_return.
++	  In that case, compiler will not expand epilogue
++	  so that we have no chance to output .omit_fp_end directive.  */
++  if (TARGET_FORBID_FP_AS_GP
++      || nds32_naked_function_p (current_function_decl)
++      || !optimize_size
++      || frame_pointer_needed
++      || NDS32_REQUIRED_CALLEE_SAVED_P (FP_REGNUM)
++      || (cfun->stdarg == 1)
++      || (find_fallthru_edge (EXIT_BLOCK_PTR_FOR_FN (cfun)->preds) == NULL))
++    return false;
++
++  /* Disable fp_as_gp if there is any infinite loop since the fp may
++     reuse in infinite loops by register rename.
++     For check infinite loops we should make sure exit_bb is post dominate
++     all other basic blocks if there is no infinite loops.  */
++  first_exit_blocks_p = true;
++  exit_bb = NULL;
++
++  FOR_EACH_EDGE (e, ei, EXIT_BLOCK_PTR_FOR_FN (cfun)->preds)
++    {
++      /* More than one exit block also do not perform fp_as_gp optimization.  */
++      if (!first_exit_blocks_p)
++	return false;
++
++      exit_bb = e->src;
++      first_exit_blocks_p = false;
++    }
++
++  /* Not found exit_bb? just abort fp_as_gp!  */
++  if (!exit_bb)
++    return false;
++
++  /* Each bb should post dominate by exit_bb if there is no infinite loop! */
++  FOR_EACH_BB_FN (bb, cfun)
++    {
++      if (!dominated_by_p (CDI_POST_DOMINATORS,
++			   bb,
++			   exit_bb))
++	return false;
++    }
++
++  /* Now we can check the possibility of using fp_as_gp optimization.  */
++  if (TARGET_FORCE_FP_AS_GP)
++    {
++      /* User explicitly issues -mforce-fp-as-gp option.  */
++      return true;
++    }
++  else
++    {
++      /* In the following we are going to evaluate whether
++	 it is worth to do fp_as_gp optimization.  */
++      bool good_gain = false;
++      int symbol_count;
++
++      int threshold;
++
++      /* We check if there already requires prologue.
++	 Note that $gp will be saved in prologue for PIC code generation.
++	 After that, we can set threshold by the existence of prologue.
++	 Each fp-implied instruction will gain 2-byte code size
++	 from gp-aware instruction, so we have following heuristics.  */
++      if (flag_pic
++	  || nds32_have_prologue_p ())
++	{
++	  /* Have-prologue:
++	       Compiler already intends to generate prologue content,
++	       so the fp_as_gp optimization will only insert
++	       'la $fp,_FP_BASE_' instruction, which will be
++	       converted into 4-byte instruction at link time.
++	       The threshold is "3" symbol accesses, 2 + 2 + 2 > 4.  */
++	  threshold = 3;
++	}
++      else
++	{
++	  /* None-prologue:
++	       Compiler originally does not generate prologue content,
++	       so the fp_as_gp optimization will NOT ONLY insert
++	       'la $fp,_FP_BASE' instruction, but also causes
++	       push/pop instructions.
++	       If we are using v3push (push25/pop25),
++	       the threshold is "5" symbol accesses, 5*2 > 4 + 2 + 2;
++	       If we are using normal push (smw/lmw),
++	       the threshold is "5+2" symbol accesses 7*2 > 4 + 4 + 4.  */
++	  threshold = 5 + (TARGET_V3PUSH ? 0 : 2);
++	}
++
++      symbol_count = nds32_get_symbol_count ();
++
++      if (symbol_count >= threshold)
++	good_gain = true;
++
++      /* Enable fp_as_gp optimization when potential gain is good enough.  */
++      return good_gain;
++    }
++}
++
++static unsigned int
++nds32_fp_as_gp (void)
++{
++  bool fp_as_gp_p;
++  calculate_dominance_info (CDI_POST_DOMINATORS);
++  fp_as_gp_p = nds32_fp_as_gp_check_available ();
++
++  /* Here is a hack to IRA for enable/disable a hard register per function.
++     We *MUST* review this way after migrate gcc 4.9! */
++  if (fp_as_gp_p) {
++    SET_HARD_REG_BIT(this_target_ira_int->x_no_unit_alloc_regs, FP_REGNUM);
++    df_set_regs_ever_live (FP_REGNUM, 1);
++  } else {
++    CLEAR_HARD_REG_BIT(this_target_ira_int->x_no_unit_alloc_regs, FP_REGNUM);
++  }
++
++  cfun->machine->fp_as_gp_p = fp_as_gp_p;
++
++  free_dominance_info (CDI_POST_DOMINATORS);
++  return 1;
++}
++
++const pass_data pass_data_nds32_fp_as_gp =
++{
++  RTL_PASS,				/* type */
++  "fp_as_gp",				/* name */
++  OPTGROUP_NONE,			/* optinfo_flags */
++  TV_MACH_DEP,				/* tv_id */
++  0,					/* properties_required */
++  0,					/* properties_provided */
++  0,					/* properties_destroyed */
++  0,					/* todo_flags_start */
++  0					/* todo_flags_finish */
++};
++
++class pass_nds32_fp_as_gp : public rtl_opt_pass
++{
++public:
++  pass_nds32_fp_as_gp (gcc::context *ctxt)
++    : rtl_opt_pass (pass_data_nds32_fp_as_gp, ctxt)
++  {}
++
++  /* opt_pass methods: */
++  bool gate (function *)
++  {
++    return !TARGET_LINUX_ABI
++	   && TARGET_16_BIT
++	   && optimize_size;
++  }
++  unsigned int execute (function *) { return nds32_fp_as_gp (); }
++};
++
++rtl_opt_pass *
++make_pass_nds32_fp_as_gp (gcc::context *ctxt)
++{
++  return new pass_nds32_fp_as_gp (ctxt);
+ }
+ 
+ /* ------------------------------------------------------------------------ */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-fpu.md gcc-8.2.0/gcc/config/nds32/nds32-fpu.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-fpu.md	2018-04-06 07:51:33.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-fpu.md	2019-01-25 15:38:32.825242648 +0100
+@@ -1,5 +1,5 @@
+ ;; Machine description of Andes NDS32 cpu for GNU compiler
+-;; Copyright (C) 2012-2015 Free Software Foundation, Inc.
++;; Copyright (C) 2012-2018 Free Software Foundation, Inc.
+ ;; Contributed by Andes Technology Corporation.
+ ;;
+ ;; This file is part of GCC.
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-graywolf.md gcc-8.2.0/gcc/config/nds32/nds32-graywolf.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-graywolf.md	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-graywolf.md	2019-01-25 15:38:32.825242648 +0100
+@@ -0,0 +1,471 @@
++;; Pipeline descriptions of Andes NDS32 cpu for GNU compiler
++;; Copyright (C) 2012-2013 Free Software Foundation, Inc.
++;; Contributed by Andes Technology Corporation.
++;;
++;; This file is part of GCC.
++;;
++;; GCC is free software; you can redistribute it and/or modify it
++;; under the terms of the GNU General Public License as published
++;; by the Free Software Foundation; either version 3, or (at your
++;; option) any later version.
++;;
++;; GCC is distributed in the hope that it will be useful, but WITHOUT
++;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++;; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++;; License for more details.
++;;
++;; You should have received a copy of the GNU General Public License
++;; along with GCC; see the file COPYING3.  If not see
++;; <http://www.gnu.org/licenses/>.
++
++;; ------------------------------------------------------------------------
++;; Define Graywolf pipeline settings.
++;; ------------------------------------------------------------------------
++
++(define_automaton "nds32_graywolf_machine")
++
++(define_cpu_unit "gw_ii_0" "nds32_graywolf_machine")
++(define_cpu_unit "gw_ii_1" "nds32_graywolf_machine")
++(define_cpu_unit "gw_ex_p0" "nds32_graywolf_machine")
++(define_cpu_unit "gw_mm_p0" "nds32_graywolf_machine")
++(define_cpu_unit "gw_wb_p0" "nds32_graywolf_machine")
++(define_cpu_unit "gw_ex_p1" "nds32_graywolf_machine")
++(define_cpu_unit "gw_mm_p1" "nds32_graywolf_machine")
++(define_cpu_unit "gw_wb_p1" "nds32_graywolf_machine")
++(define_cpu_unit "gw_iq_p2" "nds32_graywolf_machine")
++(define_cpu_unit "gw_rf_p2" "nds32_graywolf_machine")
++(define_cpu_unit "gw_e1_p2" "nds32_graywolf_machine")
++(define_cpu_unit "gw_e2_p2" "nds32_graywolf_machine")
++(define_cpu_unit "gw_e3_p2" "nds32_graywolf_machine")
++(define_cpu_unit "gw_e4_p2" "nds32_graywolf_machine")
++
++(define_reservation "gw_ii" "gw_ii_0 | gw_ii_1")
++(define_reservation "gw_ex" "gw_ex_p0 | gw_ex_p1")
++(define_reservation "gw_mm" "gw_mm_p0 | gw_mm_p1")
++(define_reservation "gw_wb" "gw_wb_p0 | gw_wb_p1")
++
++(define_reservation "gw_ii_all" "gw_ii_0 + gw_ii_1")
++
++(define_insn_reservation "nds_gw_unknown" 1
++  (and (eq_attr "type" "unknown")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_misc" 1
++  (and (eq_attr "type" "misc")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_mmu" 1
++  (and (eq_attr "type" "mmu")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_alu" 1
++  (and (and (eq_attr "type" "alu")
++            (match_test "!nds32::movd44_insn_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_movd44" 1
++  (and (and (eq_attr "type" "alu")
++            (match_test "nds32::movd44_insn_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_alu_shift" 1
++  (and (eq_attr "type" "alu_shift")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex*2, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_pbsad" 1
++  (and (eq_attr "type" "pbsad")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex*3, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_pbsada" 1
++  (and (eq_attr "type" "pbsada")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex*3, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_load" 1
++  (and (and (eq_attr "type" "load")
++            (match_test "!nds32::post_update_insn_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_2w" 1
++  (and (and (eq_attr "type" "load")
++            (match_test "nds32::post_update_insn_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_all, gw_ex_p1, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store" 1
++  (and (and (eq_attr "type" "store")
++            (match_test "!nds32::store_offset_reg_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_3r" 1
++  (and (and (eq_attr "type" "store")
++            (match_test "nds32::store_offset_reg_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_all, gw_ex_p1, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_1" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "1"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_2" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "2"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*2, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_3" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "3"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*3, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_4" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "4"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_5" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "5"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_6" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "6"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_7" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "7"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_8" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "8"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_load_multiple_12" 1
++  (and (and (eq_attr "type" "load_multiple")
++            (eq_attr "combo" "12"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_1" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "1"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_2" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "2"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*2, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_3" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "3"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*3, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_4" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "4"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_5" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "5"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_6" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "6"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_7" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "7"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_8" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "8"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_store_multiple_12" 1
++  (and (and (eq_attr "type" "store_multiple")
++            (eq_attr "combo" "12"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_1, gw_ex_p1*4, gw_mm_p1, gw_wb_p1")
++
++(define_insn_reservation "nds_gw_mul_fast1" 1
++  (and (match_test "nds32_mul_config == MUL_TYPE_FAST_1")
++       (and (eq_attr "type" "mul")
++       (eq_attr "pipeline_model" "graywolf")))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_mul_fast2" 1
++  (and (match_test "nds32_mul_config == MUL_TYPE_FAST_2")
++       (and (eq_attr "type" "mul")
++       (eq_attr "pipeline_model" "graywolf")))
++  "gw_ii_0, gw_ex_p0*2, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_mul_slow" 1
++  (and (match_test "nds32_mul_config == MUL_TYPE_SLOW")
++       (and (eq_attr "type" "mul")
++       (eq_attr "pipeline_model" "graywolf")))
++  "gw_ii_0, gw_ex_p0*4, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_mac_fast1" 1
++  (and (match_test "nds32_mul_config == MUL_TYPE_FAST_1")
++       (and (eq_attr "type" "mac")
++       (eq_attr "pipeline_model" "graywolf")))
++  "gw_ii_all, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_mac_fast2" 1
++  (and (match_test "nds32_mul_config == MUL_TYPE_FAST_2")
++       (and (eq_attr "type" "mac")
++       (eq_attr "pipeline_model" "graywolf")))
++  "gw_ii_all, gw_ex_p0*2, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_mac_slow" 1
++  (and (match_test "nds32_mul_config == MUL_TYPE_SLOW")
++       (and (eq_attr "type" "mac")
++       (eq_attr "pipeline_model" "graywolf")))
++  "gw_ii_all, gw_ex_p0*4, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_div" 1
++  (and (and (eq_attr "type" "div")
++            (match_test "!nds32::divmod_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0*4, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_div_2w" 1
++  (and (and (eq_attr "type" "div")
++            (match_test "nds32::divmod_p (insn)"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_all, gw_ex_p0*4, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_branch" 1
++  (and (eq_attr "type" "branch")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_alu" 1
++  (and (eq_attr "type" "dalu")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ex, gw_mm, gw_wb")
++
++(define_insn_reservation "nds_gw_dsp_alu64" 1
++  (and (eq_attr "type" "dalu64")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_all, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_alu_round" 1
++  (and (eq_attr "type" "daluround")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_cmp" 1
++  (and (eq_attr "type" "dcmp")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_clip" 1
++  (and (eq_attr "type" "dclip")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_mul" 1
++  (and (eq_attr "type" "dmul")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_mac" 1
++  (and (eq_attr "type" "dmac")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_all, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_insb" 1
++  (and (eq_attr "type" "dinsb")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_pack" 1
++  (and (eq_attr "type" "dpack")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_bpick" 1
++  (and (eq_attr "type" "dbpick")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_0, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_dsp_wext" 1
++  (and (eq_attr "type" "dwext")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii_all, gw_ex_p0, gw_mm_p0, gw_wb_p0")
++
++(define_insn_reservation "nds_gw_fpu_alu" 4
++  (and (eq_attr "type" "falu")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_muls" 4
++  (and (eq_attr "type" "fmuls")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_muld" 4
++  (and (eq_attr "type" "fmuld")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2*2, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_macs" 4
++  (and (eq_attr "type" "fmacs")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2*3, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_macd" 4
++  (and (eq_attr "type" "fmacd")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2*4, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_divs" 4
++  (and (ior (eq_attr "type" "fdivs")
++	    (eq_attr "type" "fsqrts"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2*14, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_divd" 4
++  (and (ior (eq_attr "type" "fdivd")
++	    (eq_attr "type" "fsqrtd"))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2*28, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_fast_alu" 2
++  (and (ior (eq_attr "type" "fcmp")
++	    (ior (eq_attr "type" "fabs")
++		 (ior (eq_attr "type" "fcpy")
++		      (eq_attr "type" "fcmov"))))
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_fmtsr" 1
++  (and (eq_attr "type" "fmtsr")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_fmtdr" 1
++  (and (eq_attr "type" "fmtdr")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ii+gw_iq_p2, gw_iq_p2+gw_rf_p2, gw_rf_p2+gw_e1_p2, gw_e1_p2+gw_e2_p2, gw_e2_p2+gw_e3_p2, gw_e3_p2+gw_e4_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_fmfsr" 1
++  (and (eq_attr "type" "fmfsr")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_fmfdr" 1
++  (and (eq_attr "type" "fmfdr")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_ii+gw_iq_p2, gw_iq_p2+gw_rf_p2, gw_rf_p2+gw_e1_p2, gw_e1_p2+gw_e2_p2, gw_e2_p2+gw_e3_p2, gw_e3_p2+gw_e4_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_load" 3
++  (and (eq_attr "type" "fload")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2, gw_e3_p2, gw_e4_p2")
++
++(define_insn_reservation "nds_gw_fpu_store" 1
++  (and (eq_attr "type" "fstore")
++       (eq_attr "pipeline_model" "graywolf"))
++  "gw_ii, gw_iq_p2, gw_rf_p2, gw_e1_p2, gw_e2_p2, gw_e3_p2, gw_e4_p2")
++
++;; FPU_ADDR_OUT -> FPU_ADDR_IN
++;; Main pipeline rules don't need this because those default latency is 1.
++(define_bypass 1
++  "nds_gw_fpu_load, nds_gw_fpu_store"
++  "nds_gw_fpu_load, nds_gw_fpu_store"
++  "nds32_gw_ex_to_ex_p"
++)
++
++;; LD, MUL, MAC, DIV, DALU64, DMUL, DMAC, DALUROUND, DBPICK, DWEXT
++;;   -> ALU, ALU_SHIFT_Rb, PBSAD, PBSADA_RaRb, MOVD44, MUL, MAC_RaRb, DIV, ADDR_IN, BR, MMU,
++;;      DALU, DALUROUND, DMUL, DMAC_RaRb, DPACK, DINSB, DCMP, DCLIP, WEXT_O, BPICK_RaRb
++(define_bypass 2
++  "nds_gw_load, nds_gw_load_2w,\
++   nds_gw_mul_fast1, nds_gw_mul_fast2, nds_gw_mul_slow,\
++   nds_gw_mac_fast1, nds_gw_mac_fast2, nds_gw_mac_slow,\
++   nds_gw_div, nds_gw_div_2w,\
++   nds_gw_dsp_alu64, nds_gw_dsp_mul, nds_gw_dsp_mac,\
++   nds_gw_dsp_alu_round, nds_gw_dsp_bpick, nds_gw_dsp_wext"
++  "nds_gw_alu, nds_gw_movd44, nds_gw_alu_shift,\
++   nds_gw_pbsad, nds_gw_pbsada,\
++   nds_gw_mul_fast1, nds_gw_mul_fast2, nds_gw_mul_slow,\
++   nds_gw_mac_fast1, nds_gw_mac_fast2, nds_gw_mac_slow,\
++   nds_gw_branch,\
++   nds_gw_div, nds_gw_div_2w,\
++   nds_gw_load, nds_gw_load_2w, nds_gw_store, nds_gw_store_3r,\
++   nds_gw_load_multiple_1,nds_gw_load_multiple_2, nds_gw_load_multiple_3,\
++   nds_gw_load_multiple_4,nds_gw_load_multiple_5, nds_gw_load_multiple_6,\
++   nds_gw_load_multiple_7,nds_gw_load_multiple_8, nds_gw_load_multiple_12,\
++   nds_gw_store_multiple_1,nds_gw_store_multiple_2, nds_gw_store_multiple_3,\
++   nds_gw_store_multiple_4,nds_gw_store_multiple_5, nds_gw_store_multiple_6,\
++   nds_gw_store_multiple_7,nds_gw_store_multiple_8, nds_gw_store_multiple_12,\
++   nds_gw_mmu,\
++   nds_gw_dsp_alu, nds_gw_dsp_alu_round,\
++   nds_gw_dsp_mul, nds_gw_dsp_mac, nds_gw_dsp_pack,\
++   nds_gw_dsp_insb, nds_gw_dsp_cmp, nds_gw_dsp_clip,\
++   nds_gw_dsp_wext, nds_gw_dsp_bpick"
++  "nds32_gw_mm_to_ex_p"
++)
++
++;; LMW(N, N)
++;;   -> ALU, ALU_SHIFT_Rb, PBSAD, PBSADA_RaRb, MOVD44, MUL, MAC_RaRb, DIV, ADDR_IN, BR, MMU
++;;      DALU, DALUROUND, DMUL, DMAC_RaRb, DPACK, DINSB, DCMP, DCLIP, WEXT_O, BPICK_RaRb
++(define_bypass 2
++  "nds_gw_load_multiple_1,nds_gw_load_multiple_2, nds_gw_load_multiple_3,\
++   nds_gw_load_multiple_4,nds_gw_load_multiple_5, nds_gw_load_multiple_6,\
++   nds_gw_load_multiple_7,nds_gw_load_multiple_8, nds_gw_load_multiple_12"
++  "nds_gw_alu, nds_gw_movd44, nds_gw_alu_shift,\
++   nds_gw_pbsad, nds_gw_pbsada,\
++   nds_gw_mul_fast1, nds_gw_mul_fast2, nds_gw_mul_slow,\
++   nds_gw_mac_fast1, nds_gw_mac_fast2, nds_gw_mac_slow,\
++   nds_gw_branch,\
++   nds_gw_div, nds_gw_div_2w,\
++   nds_gw_load, nds_gw_load_2w, nds_gw_store, nds_gw_store_3r,\
++   nds_gw_load_multiple_1,nds_gw_load_multiple_2, nds_gw_load_multiple_3,\
++   nds_gw_load_multiple_4,nds_gw_load_multiple_5, nds_gw_load_multiple_6,\
++   nds_gw_load_multiple_7,nds_gw_load_multiple_8, nds_gw_load_multiple_12,\
++   nds_gw_store_multiple_1,nds_gw_store_multiple_2, nds_gw_store_multiple_3,\
++   nds_gw_store_multiple_4,nds_gw_store_multiple_5, nds_gw_store_multiple_6,\
++   nds_gw_store_multiple_7,nds_gw_store_multiple_8, nds_gw_store_multiple_12,\
++   nds_gw_mmu,\
++   nds_gw_dsp_alu, nds_gw_dsp_alu_round,\
++   nds_gw_dsp_mul, nds_gw_dsp_mac, nds_gw_dsp_pack,\
++   nds_gw_dsp_insb, nds_gw_dsp_cmp, nds_gw_dsp_clip,\
++   nds_gw_dsp_wext, nds_gw_dsp_bpick"
++  "nds32_gw_last_load_to_ex_p"
++)
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32.h gcc-8.2.0/gcc/config/nds32/nds32.h
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32.h	2018-05-07 03:27:52.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32.h	2019-01-25 15:44:03.534160189 +0100
+@@ -36,6 +36,16 @@
+ #define NDS32_SYMBOL_REF_RODATA_P(x) \
+   ((SYMBOL_REF_FLAGS (x) & NDS32_SYMBOL_FLAG_RODATA) != 0)
+ 
++enum nds32_relax_insn_type
++{
++  RELAX_ORI,
++  RELAX_PLT_ADD,
++  RELAX_TLS_ADD_or_LW,
++  RELAX_TLS_ADD_LW,
++  RELAX_TLS_LW_JRAL,
++  RELAX_DONE
++};
++
+ /* Classifies expand result for expand helper function.  */
+ enum nds32_expand_result_type
+ {
+@@ -140,6 +150,9 @@
+    Check gcc/common/config/nds32/nds32-common.c for the optimizations that
+    apply -malways-align.  */
+ #define NDS32_ALIGN_P() (TARGET_ALWAYS_ALIGN)
++
++#define NDS32_EXT_DSP_P() (TARGET_EXT_DSP && !TARGET_FORCE_NO_EXT_DSP)
++
+ /* Get alignment according to mode or type information.
+    When 'type' is nonnull, there is no need to look at 'mode'.  */
+ #define NDS32_MODE_TYPE_ALIGN(mode, type) \
+@@ -305,6 +318,10 @@
+        2. The rtl lowering and optimization are close to target code.
+ 	  For this case we need address to be strictly aligned.  */
+   int strict_aligned_p;
++
++  /* Record two similar attributes status.  */
++  int attr_naked_p;
++  int attr_no_prologue_p;
+ };
+ 
+ /* A C structure that contains the arguments information.  */
+@@ -350,7 +367,8 @@
+ {
+   NDS32_NESTED,
+   NDS32_NOT_NESTED,
+-  NDS32_NESTED_READY
++  NDS32_NESTED_READY,
++  NDS32_CRITICAL
+ };
+ 
+ /* Define structure to record isr information.
+@@ -378,6 +396,13 @@
+      unless user specifies attribute to change it.  */
+   enum nds32_isr_nested_type nested_type;
+ 
++  /* Secure isr level.
++     Currently we have 0-3 security level.
++     It should be set to 0 by default.
++     For security processors, this is determined by secure
++     attribute or compiler options.  */
++  unsigned int security_level;
++
+   /* Total vectors.
+      The total vectors = interrupt + exception numbers + reset.
+      It should be set to 0 by default.
+@@ -439,7 +464,30 @@
+   NDS32_BUILTIN_FFB,
+   NDS32_BUILTIN_FFMISM,
+   NDS32_BUILTIN_FLMISM,
+-
++  NDS32_BUILTIN_KADDW,
++  NDS32_BUILTIN_KSUBW,
++  NDS32_BUILTIN_KADDH,
++  NDS32_BUILTIN_KSUBH,
++  NDS32_BUILTIN_KDMBB,
++  NDS32_BUILTIN_V_KDMBB,
++  NDS32_BUILTIN_KDMBT,
++  NDS32_BUILTIN_V_KDMBT,
++  NDS32_BUILTIN_KDMTB,
++  NDS32_BUILTIN_V_KDMTB,
++  NDS32_BUILTIN_KDMTT,
++  NDS32_BUILTIN_V_KDMTT,
++  NDS32_BUILTIN_KHMBB,
++  NDS32_BUILTIN_V_KHMBB,
++  NDS32_BUILTIN_KHMBT,
++  NDS32_BUILTIN_V_KHMBT,
++  NDS32_BUILTIN_KHMTB,
++  NDS32_BUILTIN_V_KHMTB,
++  NDS32_BUILTIN_KHMTT,
++  NDS32_BUILTIN_V_KHMTT,
++  NDS32_BUILTIN_KSLRAW,
++  NDS32_BUILTIN_KSLRAW_U,
++  NDS32_BUILTIN_RDOV,
++  NDS32_BUILTIN_CLROV,
+   NDS32_BUILTIN_ROTR,
+   NDS32_BUILTIN_SVA,
+   NDS32_BUILTIN_SVS,
+@@ -512,7 +560,295 @@
+   NDS32_BUILTIN_SET_TRIG_LEVEL,
+   NDS32_BUILTIN_SET_TRIG_EDGE,
+   NDS32_BUILTIN_GET_TRIG_TYPE,
+-
++  NDS32_BUILTIN_DSP_BEGIN,
++  NDS32_BUILTIN_ADD16,
++  NDS32_BUILTIN_V_UADD16,
++  NDS32_BUILTIN_V_SADD16,
++  NDS32_BUILTIN_RADD16,
++  NDS32_BUILTIN_V_RADD16,
++  NDS32_BUILTIN_URADD16,
++  NDS32_BUILTIN_V_URADD16,
++  NDS32_BUILTIN_KADD16,
++  NDS32_BUILTIN_V_KADD16,
++  NDS32_BUILTIN_UKADD16,
++  NDS32_BUILTIN_V_UKADD16,
++  NDS32_BUILTIN_SUB16,
++  NDS32_BUILTIN_V_USUB16,
++  NDS32_BUILTIN_V_SSUB16,
++  NDS32_BUILTIN_RSUB16,
++  NDS32_BUILTIN_V_RSUB16,
++  NDS32_BUILTIN_URSUB16,
++  NDS32_BUILTIN_V_URSUB16,
++  NDS32_BUILTIN_KSUB16,
++  NDS32_BUILTIN_V_KSUB16,
++  NDS32_BUILTIN_UKSUB16,
++  NDS32_BUILTIN_V_UKSUB16,
++  NDS32_BUILTIN_CRAS16,
++  NDS32_BUILTIN_V_UCRAS16,
++  NDS32_BUILTIN_V_SCRAS16,
++  NDS32_BUILTIN_RCRAS16,
++  NDS32_BUILTIN_V_RCRAS16,
++  NDS32_BUILTIN_URCRAS16,
++  NDS32_BUILTIN_V_URCRAS16,
++  NDS32_BUILTIN_KCRAS16,
++  NDS32_BUILTIN_V_KCRAS16,
++  NDS32_BUILTIN_UKCRAS16,
++  NDS32_BUILTIN_V_UKCRAS16,
++  NDS32_BUILTIN_CRSA16,
++  NDS32_BUILTIN_V_UCRSA16,
++  NDS32_BUILTIN_V_SCRSA16,
++  NDS32_BUILTIN_RCRSA16,
++  NDS32_BUILTIN_V_RCRSA16,
++  NDS32_BUILTIN_URCRSA16,
++  NDS32_BUILTIN_V_URCRSA16,
++  NDS32_BUILTIN_KCRSA16,
++  NDS32_BUILTIN_V_KCRSA16,
++  NDS32_BUILTIN_UKCRSA16,
++  NDS32_BUILTIN_V_UKCRSA16,
++  NDS32_BUILTIN_ADD8,
++  NDS32_BUILTIN_V_UADD8,
++  NDS32_BUILTIN_V_SADD8,
++  NDS32_BUILTIN_RADD8,
++  NDS32_BUILTIN_V_RADD8,
++  NDS32_BUILTIN_URADD8,
++  NDS32_BUILTIN_V_URADD8,
++  NDS32_BUILTIN_KADD8,
++  NDS32_BUILTIN_V_KADD8,
++  NDS32_BUILTIN_UKADD8,
++  NDS32_BUILTIN_V_UKADD8,
++  NDS32_BUILTIN_SUB8,
++  NDS32_BUILTIN_V_USUB8,
++  NDS32_BUILTIN_V_SSUB8,
++  NDS32_BUILTIN_RSUB8,
++  NDS32_BUILTIN_V_RSUB8,
++  NDS32_BUILTIN_URSUB8,
++  NDS32_BUILTIN_V_URSUB8,
++  NDS32_BUILTIN_KSUB8,
++  NDS32_BUILTIN_V_KSUB8,
++  NDS32_BUILTIN_UKSUB8,
++  NDS32_BUILTIN_V_UKSUB8,
++  NDS32_BUILTIN_SRA16,
++  NDS32_BUILTIN_V_SRA16,
++  NDS32_BUILTIN_SRA16_U,
++  NDS32_BUILTIN_V_SRA16_U,
++  NDS32_BUILTIN_SRL16,
++  NDS32_BUILTIN_V_SRL16,
++  NDS32_BUILTIN_SRL16_U,
++  NDS32_BUILTIN_V_SRL16_U,
++  NDS32_BUILTIN_SLL16,
++  NDS32_BUILTIN_V_SLL16,
++  NDS32_BUILTIN_KSLL16,
++  NDS32_BUILTIN_V_KSLL16,
++  NDS32_BUILTIN_KSLRA16,
++  NDS32_BUILTIN_V_KSLRA16,
++  NDS32_BUILTIN_KSLRA16_U,
++  NDS32_BUILTIN_V_KSLRA16_U,
++  NDS32_BUILTIN_CMPEQ16,
++  NDS32_BUILTIN_V_SCMPEQ16,
++  NDS32_BUILTIN_V_UCMPEQ16,
++  NDS32_BUILTIN_SCMPLT16,
++  NDS32_BUILTIN_V_SCMPLT16,
++  NDS32_BUILTIN_SCMPLE16,
++  NDS32_BUILTIN_V_SCMPLE16,
++  NDS32_BUILTIN_UCMPLT16,
++  NDS32_BUILTIN_V_UCMPLT16,
++  NDS32_BUILTIN_UCMPLE16,
++  NDS32_BUILTIN_V_UCMPLE16,
++  NDS32_BUILTIN_CMPEQ8,
++  NDS32_BUILTIN_V_SCMPEQ8,
++  NDS32_BUILTIN_V_UCMPEQ8,
++  NDS32_BUILTIN_SCMPLT8,
++  NDS32_BUILTIN_V_SCMPLT8,
++  NDS32_BUILTIN_SCMPLE8,
++  NDS32_BUILTIN_V_SCMPLE8,
++  NDS32_BUILTIN_UCMPLT8,
++  NDS32_BUILTIN_V_UCMPLT8,
++  NDS32_BUILTIN_UCMPLE8,
++  NDS32_BUILTIN_V_UCMPLE8,
++  NDS32_BUILTIN_SMIN16,
++  NDS32_BUILTIN_V_SMIN16,
++  NDS32_BUILTIN_UMIN16,
++  NDS32_BUILTIN_V_UMIN16,
++  NDS32_BUILTIN_SMAX16,
++  NDS32_BUILTIN_V_SMAX16,
++  NDS32_BUILTIN_UMAX16,
++  NDS32_BUILTIN_V_UMAX16,
++  NDS32_BUILTIN_SCLIP16,
++  NDS32_BUILTIN_V_SCLIP16,
++  NDS32_BUILTIN_UCLIP16,
++  NDS32_BUILTIN_V_UCLIP16,
++  NDS32_BUILTIN_KHM16,
++  NDS32_BUILTIN_V_KHM16,
++  NDS32_BUILTIN_KHMX16,
++  NDS32_BUILTIN_V_KHMX16,
++  NDS32_BUILTIN_KABS16,
++  NDS32_BUILTIN_V_KABS16,
++  NDS32_BUILTIN_SMIN8,
++  NDS32_BUILTIN_V_SMIN8,
++  NDS32_BUILTIN_UMIN8,
++  NDS32_BUILTIN_V_UMIN8,
++  NDS32_BUILTIN_SMAX8,
++  NDS32_BUILTIN_V_SMAX8,
++  NDS32_BUILTIN_UMAX8,
++  NDS32_BUILTIN_V_UMAX8,
++  NDS32_BUILTIN_KABS8,
++  NDS32_BUILTIN_V_KABS8,
++  NDS32_BUILTIN_SUNPKD810,
++  NDS32_BUILTIN_V_SUNPKD810,
++  NDS32_BUILTIN_SUNPKD820,
++  NDS32_BUILTIN_V_SUNPKD820,
++  NDS32_BUILTIN_SUNPKD830,
++  NDS32_BUILTIN_V_SUNPKD830,
++  NDS32_BUILTIN_SUNPKD831,
++  NDS32_BUILTIN_V_SUNPKD831,
++  NDS32_BUILTIN_ZUNPKD810,
++  NDS32_BUILTIN_V_ZUNPKD810,
++  NDS32_BUILTIN_ZUNPKD820,
++  NDS32_BUILTIN_V_ZUNPKD820,
++  NDS32_BUILTIN_ZUNPKD830,
++  NDS32_BUILTIN_V_ZUNPKD830,
++  NDS32_BUILTIN_ZUNPKD831,
++  NDS32_BUILTIN_V_ZUNPKD831,
++  NDS32_BUILTIN_RADDW,
++  NDS32_BUILTIN_URADDW,
++  NDS32_BUILTIN_RSUBW,
++  NDS32_BUILTIN_URSUBW,
++  NDS32_BUILTIN_SRA_U,
++  NDS32_BUILTIN_KSLL,
++  NDS32_BUILTIN_PKBB16,
++  NDS32_BUILTIN_V_PKBB16,
++  NDS32_BUILTIN_PKBT16,
++  NDS32_BUILTIN_V_PKBT16,
++  NDS32_BUILTIN_PKTB16,
++  NDS32_BUILTIN_V_PKTB16,
++  NDS32_BUILTIN_PKTT16,
++  NDS32_BUILTIN_V_PKTT16,
++  NDS32_BUILTIN_SMMUL,
++  NDS32_BUILTIN_SMMUL_U,
++  NDS32_BUILTIN_KMMAC,
++  NDS32_BUILTIN_KMMAC_U,
++  NDS32_BUILTIN_KMMSB,
++  NDS32_BUILTIN_KMMSB_U,
++  NDS32_BUILTIN_KWMMUL,
++  NDS32_BUILTIN_KWMMUL_U,
++  NDS32_BUILTIN_SMMWB,
++  NDS32_BUILTIN_V_SMMWB,
++  NDS32_BUILTIN_SMMWB_U,
++  NDS32_BUILTIN_V_SMMWB_U,
++  NDS32_BUILTIN_SMMWT,
++  NDS32_BUILTIN_V_SMMWT,
++  NDS32_BUILTIN_SMMWT_U,
++  NDS32_BUILTIN_V_SMMWT_U,
++  NDS32_BUILTIN_KMMAWB,
++  NDS32_BUILTIN_V_KMMAWB,
++  NDS32_BUILTIN_KMMAWB_U,
++  NDS32_BUILTIN_V_KMMAWB_U,
++  NDS32_BUILTIN_KMMAWT,
++  NDS32_BUILTIN_V_KMMAWT,
++  NDS32_BUILTIN_KMMAWT_U,
++  NDS32_BUILTIN_V_KMMAWT_U,
++  NDS32_BUILTIN_SMBB,
++  NDS32_BUILTIN_V_SMBB,
++  NDS32_BUILTIN_SMBT,
++  NDS32_BUILTIN_V_SMBT,
++  NDS32_BUILTIN_SMTT,
++  NDS32_BUILTIN_V_SMTT,
++  NDS32_BUILTIN_KMDA,
++  NDS32_BUILTIN_V_KMDA,
++  NDS32_BUILTIN_KMXDA,
++  NDS32_BUILTIN_V_KMXDA,
++  NDS32_BUILTIN_SMDS,
++  NDS32_BUILTIN_V_SMDS,
++  NDS32_BUILTIN_SMDRS,
++  NDS32_BUILTIN_V_SMDRS,
++  NDS32_BUILTIN_SMXDS,
++  NDS32_BUILTIN_V_SMXDS,
++  NDS32_BUILTIN_KMABB,
++  NDS32_BUILTIN_V_KMABB,
++  NDS32_BUILTIN_KMABT,
++  NDS32_BUILTIN_V_KMABT,
++  NDS32_BUILTIN_KMATT,
++  NDS32_BUILTIN_V_KMATT,
++  NDS32_BUILTIN_KMADA,
++  NDS32_BUILTIN_V_KMADA,
++  NDS32_BUILTIN_KMAXDA,
++  NDS32_BUILTIN_V_KMAXDA,
++  NDS32_BUILTIN_KMADS,
++  NDS32_BUILTIN_V_KMADS,
++  NDS32_BUILTIN_KMADRS,
++  NDS32_BUILTIN_V_KMADRS,
++  NDS32_BUILTIN_KMAXDS,
++  NDS32_BUILTIN_V_KMAXDS,
++  NDS32_BUILTIN_KMSDA,
++  NDS32_BUILTIN_V_KMSDA,
++  NDS32_BUILTIN_KMSXDA,
++  NDS32_BUILTIN_V_KMSXDA,
++  NDS32_BUILTIN_SMAL,
++  NDS32_BUILTIN_V_SMAL,
++  NDS32_BUILTIN_BITREV,
++  NDS32_BUILTIN_WEXT,
++  NDS32_BUILTIN_BPICK,
++  NDS32_BUILTIN_INSB,
++  NDS32_BUILTIN_SADD64,
++  NDS32_BUILTIN_UADD64,
++  NDS32_BUILTIN_RADD64,
++  NDS32_BUILTIN_URADD64,
++  NDS32_BUILTIN_KADD64,
++  NDS32_BUILTIN_UKADD64,
++  NDS32_BUILTIN_SSUB64,
++  NDS32_BUILTIN_USUB64,
++  NDS32_BUILTIN_RSUB64,
++  NDS32_BUILTIN_URSUB64,
++  NDS32_BUILTIN_KSUB64,
++  NDS32_BUILTIN_UKSUB64,
++  NDS32_BUILTIN_SMAR64,
++  NDS32_BUILTIN_SMSR64,
++  NDS32_BUILTIN_UMAR64,
++  NDS32_BUILTIN_UMSR64,
++  NDS32_BUILTIN_KMAR64,
++  NDS32_BUILTIN_KMSR64,
++  NDS32_BUILTIN_UKMAR64,
++  NDS32_BUILTIN_UKMSR64,
++  NDS32_BUILTIN_SMALBB,
++  NDS32_BUILTIN_V_SMALBB,
++  NDS32_BUILTIN_SMALBT,
++  NDS32_BUILTIN_V_SMALBT,
++  NDS32_BUILTIN_SMALTT,
++  NDS32_BUILTIN_V_SMALTT,
++  NDS32_BUILTIN_SMALDA,
++  NDS32_BUILTIN_V_SMALDA,
++  NDS32_BUILTIN_SMALXDA,
++  NDS32_BUILTIN_V_SMALXDA,
++  NDS32_BUILTIN_SMALDS,
++  NDS32_BUILTIN_V_SMALDS,
++  NDS32_BUILTIN_SMALDRS,
++  NDS32_BUILTIN_V_SMALDRS,
++  NDS32_BUILTIN_SMALXDS,
++  NDS32_BUILTIN_V_SMALXDS,
++  NDS32_BUILTIN_SMUL16,
++  NDS32_BUILTIN_V_SMUL16,
++  NDS32_BUILTIN_SMULX16,
++  NDS32_BUILTIN_V_SMULX16,
++  NDS32_BUILTIN_UMUL16,
++  NDS32_BUILTIN_V_UMUL16,
++  NDS32_BUILTIN_UMULX16,
++  NDS32_BUILTIN_V_UMULX16,
++  NDS32_BUILTIN_SMSLDA,
++  NDS32_BUILTIN_V_SMSLDA,
++  NDS32_BUILTIN_SMSLXDA,
++  NDS32_BUILTIN_V_SMSLXDA,
++  NDS32_BUILTIN_UCLIP32,
++  NDS32_BUILTIN_SCLIP32,
++  NDS32_BUILTIN_KABS,
++  NDS32_BUILTIN_UALOAD_U16,
++  NDS32_BUILTIN_UALOAD_S16,
++  NDS32_BUILTIN_UALOAD_U8,
++  NDS32_BUILTIN_UALOAD_S8,
++  NDS32_BUILTIN_UASTORE_U16,
++  NDS32_BUILTIN_UASTORE_S16,
++  NDS32_BUILTIN_UASTORE_U8,
++  NDS32_BUILTIN_UASTORE_S8,
++  NDS32_BUILTIN_DSP_END,
+   NDS32_BUILTIN_UNALIGNED_FEATURE,
+   NDS32_BUILTIN_ENABLE_UNALIGNED,
+   NDS32_BUILTIN_DISABLE_UNALIGNED,
+@@ -521,16 +857,30 @@
+ 
+ /* ------------------------------------------------------------------------ */
+ 
+-#define TARGET_ISA_V2   (nds32_arch_option == ARCH_V2)
++#define TARGET_ISR_VECTOR_SIZE_4_BYTE \
++  (nds32_isr_vector_size == 4)
+ 
++#define TARGET_ISA_V2   (nds32_arch_option == ARCH_V2)
+ #define TARGET_ISA_V3 \
+   (nds32_arch_option == ARCH_V3 \
++   || nds32_arch_option == ARCH_V3J \
+    || nds32_arch_option == ARCH_V3F \
+    || nds32_arch_option == ARCH_V3S)
+ #define TARGET_ISA_V3M  (nds32_arch_option == ARCH_V3M)
+ 
++#define TARGET_PIPELINE_N7 \
++  (nds32_cpu_option == CPU_N7)
++#define TARGET_PIPELINE_N8 \
++  (nds32_cpu_option == CPU_N6 \
++   || nds32_cpu_option == CPU_N8)
+ #define TARGET_PIPELINE_N9 \
+   (nds32_cpu_option == CPU_N9)
++#define TARGET_PIPELINE_N10 \
++  (nds32_cpu_option == CPU_N10)
++#define TARGET_PIPELINE_N13 \
++  (nds32_cpu_option == CPU_N12 || nds32_cpu_option == CPU_N13)
++#define TARGET_PIPELINE_GRAYWOLF \
++  (nds32_cpu_option == CPU_GRAYWOLF)
+ #define TARGET_PIPELINE_SIMPLE \
+   (nds32_cpu_option == CPU_SIMPLE)
+ 
+@@ -541,6 +891,12 @@
+ #define TARGET_CMODEL_LARGE \
+    (nds32_cmodel_option == CMODEL_LARGE)
+ 
++#define TARGET_ICT_MODEL_SMALL \
++   (nds32_ict_model == ICT_MODEL_SMALL)
++
++#define TARGET_ICT_MODEL_LARGE \
++   (nds32_ict_model == ICT_MODEL_LARGE)
++
+ /* When -mcmodel=small or -mcmodel=medium,
+    compiler may generate gp-base instruction directly.  */
+ #define TARGET_GP_DIRECT \
+@@ -576,6 +932,21 @@
+ #endif
+ 
+ #define TARGET_CONFIG_FPU_DEFAULT NDS32_CONFIG_FPU_2
++
++/* ------------------------------------------------------------------------ */
++
++#ifdef TARGET_DEFAULT_RELAX
++#  define NDS32_RELAX_SPEC " %{!mno-relax:--relax}"
++#else
++#  define NDS32_RELAX_SPEC " %{mrelax:--relax}"
++#endif
++
++#ifdef TARGET_DEFAULT_EXT_DSP
++#  define NDS32_EXT_DSP_SPEC " %{!mno-ext-dsp:-mext-dsp}"
++#else
++#  define NDS32_EXT_DSP_SPEC ""
++#endif
++
+ /* ------------------------------------------------------------------------ */
+ 
+ /* Controlling the Compilation Driver.  */
+@@ -591,11 +962,15 @@
+   {"float", "%{!mfloat-abi=*:-mfloat-abi=%(VALUE)}" }
+ 
+ #define CC1_SPEC \
+-  ""
++  NDS32_EXT_DSP_SPEC
+ 
+ #define ASM_SPEC \
+   " %{mbig-endian:-EB} %{mlittle-endian:-EL}" \
+   " %{march=*:-march=%*}" \
++  " %{mno-16-bit|mno-16bit:-mno-16bit-ext}" \
++  " %{march=v3m:%{!mfull-regs:%{!mreduced-regs:-mreduced-regs}}}" \
++  " %{mfull-regs:-mno-reduced-regs}" \
++  " %{mreduced-regs:-mreduced-regs}" \
+   " %{mabi=*:-mabi=v%*}" \
+   " %{mconfig-fpu=*:-mfpu-freg=%*}" \
+   " %{mext-fpu-mac:-mmac}" \
+@@ -603,35 +978,9 @@
+   " %{mext-fpu-sp:-mfpu-sp-ext}" \
+   " %{mno-ext-fpu-sp:-mno-fpu-sp-ext}" \
+   " %{mext-fpu-dp:-mfpu-dp-ext}" \
+-  " %{mno-ext-fpu-sp:-mno-fpu-dp-ext}"
+-
+-/* If user issues -mrelax, we need to pass '--relax' to linker.  */
+-#define LINK_SPEC \
+-  " %{mbig-endian:-EB} %{mlittle-endian:-EL}" \
+-  " %{mrelax:--relax}"
+-
+-#define LIB_SPEC \
+-  " -lc -lgloss"
+-
+-/* The option -mno-ctor-dtor can disable constructor/destructor feature
+-   by applying different crt stuff.  In the convention, crt0.o is the
+-   startup file without constructor/destructor;
+-   crt1.o, crti.o, crtbegin.o, crtend.o, and crtn.o are the
+-   startup files with constructor/destructor.
+-   Note that crt0.o, crt1.o, crti.o, and crtn.o are provided
+-   by newlib/mculib/glibc/ublic, while crtbegin.o and crtend.o are
+-   currently provided by GCC for nds32 target.
+-
+-   For nds32 target so far:
+-   If -mno-ctor-dtor, we are going to link
+-   "crt0.o [user objects]".
+-   If general cases, we are going to link
+-   "crt1.o crtbegin1.o [user objects] crtend1.o".  */
+-#define STARTFILE_SPEC \
+-  " %{!mno-ctor-dtor:crt1.o%s;:crt0.o%s}" \
+-  " %{!mno-ctor-dtor:crtbegin1.o%s}"
+-#define ENDFILE_SPEC \
+-  " %{!mno-ctor-dtor:crtend1.o%s}"
++  " %{mno-ext-fpu-sp:-mno-fpu-dp-ext}" \
++  " %{mext-dsp:-mdsp-ext}" \
++  " %{O|O1|O2|O3|Ofast:-O1;:-Os}"
+ 
+ /* The TARGET_BIG_ENDIAN_DEFAULT is defined if we
+    configure gcc with --target=nds32be-* setting.
+@@ -642,9 +991,11 @@
+ #  define NDS32_ENDIAN_DEFAULT "mlittle-endian"
+ #endif
+ 
+-/* Currently we only have elf toolchain,
+-   where -mcmodel=medium is always the default.  */
+-#define NDS32_CMODEL_DEFAULT "mcmodel=medium"
++#if TARGET_ELF
++#  define NDS32_CMODEL_DEFAULT "mcmodel=medium"
++#else
++#  define NDS32_CMODEL_DEFAULT "mcmodel=large"
++#endif
+ 
+ #define MULTILIB_DEFAULTS \
+   { NDS32_ENDIAN_DEFAULT, NDS32_CMODEL_DEFAULT }
+@@ -1139,6 +1490,11 @@
+ 
+ #define PIC_OFFSET_TABLE_REGNUM GP_REGNUM
+ 
++#define SYMBOLIC_CONST_P(X)	\
++(GET_CODE (X) == SYMBOL_REF						\
++ || GET_CODE (X) == LABEL_REF						\
++ || (GET_CODE (X) == CONST && symbolic_reference_mentioned_p (X)))
++
+ 
+ /* Defining the Output Assembler Language.  */
+ 
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32_init.inc gcc-8.2.0/gcc/config/nds32/nds32_init.inc
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32_init.inc	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32_init.inc	2019-01-25 15:38:32.833242671 +0100
+@@ -0,0 +1,43 @@
++/*
++ * nds32_init.inc
++ *
++ * NDS32 architecture startup assembler header file
++ *
++ */
++
++.macro nds32_init
++
++	! Initialize GP for data access
++	la      $gp, _SDA_BASE_
++
++#if defined(__NDS32_EXT_EX9__)
++	! Check HW for EX9
++	mfsr    $r0, $MSC_CFG
++	li      $r1, (1 << 24)
++	and     $r2, $r0, $r1
++	beqz    $r2, 1f
++
++	! Initialize the table base of EX9 instruction
++	la      $r0, _ITB_BASE_
++	mtusr   $r0, $ITB
++1:
++#endif
++
++#if defined(__NDS32_EXT_FPU_DP__) || defined(__NDS32_EXT_FPU_SP__)
++	! Enable FPU
++	mfsr    $r0, $FUCOP_CTL
++	ori     $r0, $r0, #0x1
++	mtsr    $r0, $FUCOP_CTL
++	dsb
++
++	! Enable denormalized flush-to-Zero mode
++	fmfcsr  $r0
++	ori     $r0,$r0,#0x1000
++	fmtcsr  $r0
++	dsb
++#endif
++
++	! Initialize default stack pointer
++	la      $sp, _stack
++
++.endm
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-intrinsic.c gcc-8.2.0/gcc/config/nds32/nds32-intrinsic.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-intrinsic.c	2018-04-22 09:46:39.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-intrinsic.c	2019-01-25 15:38:32.825242648 +0100
+@@ -519,6 +519,7 @@
+ {
+   NDS32_BUILTIN(unspec_fmfcfg, "fmfcfg", FMFCFG)
+   NDS32_BUILTIN(unspec_fmfcsr, "fmfcsr", FMFCSR)
++  NDS32_BUILTIN(unspec_volatile_rdov, "rdov", RDOV)
+   NDS32_BUILTIN(unspec_get_current_sp, "get_current_sp", GET_CURRENT_SP)
+   NDS32_BUILTIN(unspec_return_address, "return_address", RETURN_ADDRESS)
+   NDS32_BUILTIN(unspec_get_all_pending_int, "get_all_pending_int",
+@@ -558,6 +559,31 @@
+   NDS32_NO_TARGET_BUILTIN(unspec_ret_itoff, "ret_itoff", RET_ITOFF)
+   NDS32_NO_TARGET_BUILTIN(unspec_set_current_sp,
+ 			  "set_current_sp", SET_CURRENT_SP)
++  NDS32_BUILTIN(kabsv2hi2, "kabs16", KABS16)
++  NDS32_BUILTIN(kabsv2hi2, "v_kabs16", V_KABS16)
++  NDS32_BUILTIN(kabsv4qi2, "kabs8", KABS8)
++  NDS32_BUILTIN(kabsv4qi2, "v_kabs8", V_KABS8)
++  NDS32_BUILTIN(sunpkd810, "sunpkd810", SUNPKD810)
++  NDS32_BUILTIN(sunpkd810, "v_sunpkd810", V_SUNPKD810)
++  NDS32_BUILTIN(sunpkd820, "sunpkd820", SUNPKD820)
++  NDS32_BUILTIN(sunpkd820, "v_sunpkd820", V_SUNPKD820)
++  NDS32_BUILTIN(sunpkd830, "sunpkd830", SUNPKD830)
++  NDS32_BUILTIN(sunpkd830, "v_sunpkd830", V_SUNPKD830)
++  NDS32_BUILTIN(sunpkd831, "sunpkd831", SUNPKD831)
++  NDS32_BUILTIN(sunpkd831, "v_sunpkd831", V_SUNPKD831)
++  NDS32_BUILTIN(zunpkd810, "zunpkd810", ZUNPKD810)
++  NDS32_BUILTIN(zunpkd810, "v_zunpkd810", V_ZUNPKD810)
++  NDS32_BUILTIN(zunpkd820, "zunpkd820", ZUNPKD820)
++  NDS32_BUILTIN(zunpkd820, "v_zunpkd820", V_ZUNPKD820)
++  NDS32_BUILTIN(zunpkd830, "zunpkd830", ZUNPKD830)
++  NDS32_BUILTIN(zunpkd830, "v_zunpkd830", V_ZUNPKD830)
++  NDS32_BUILTIN(zunpkd831, "zunpkd831", ZUNPKD831)
++  NDS32_BUILTIN(zunpkd831, "v_zunpkd831", V_ZUNPKD831)
++  NDS32_BUILTIN(unspec_kabs, "kabs", KABS)
++  NDS32_BUILTIN(unaligned_loadv2hi, "get_unaligned_u16x2", UALOAD_U16)
++  NDS32_BUILTIN(unaligned_loadv2hi, "get_unaligned_s16x2", UALOAD_S16)
++  NDS32_BUILTIN(unaligned_loadv4qi, "get_unaligned_u8x4", UALOAD_U8)
++  NDS32_BUILTIN(unaligned_loadv4qi, "get_unaligned_s8x4", UALOAD_S8)
+ };
+ 
+ /* Intrinsics that take just one argument. and the argument is immediate.  */
+@@ -593,6 +619,28 @@
+   NDS32_BUILTIN(unspec_ffb, "ffb", FFB)
+   NDS32_BUILTIN(unspec_ffmism, "ffmsim", FFMISM)
+   NDS32_BUILTIN(unspec_flmism, "flmism", FLMISM)
++  NDS32_BUILTIN(unspec_kaddw, "kaddw", KADDW)
++  NDS32_BUILTIN(unspec_kaddh, "kaddh", KADDH)
++  NDS32_BUILTIN(unspec_ksubw, "ksubw", KSUBW)
++  NDS32_BUILTIN(unspec_ksubh, "ksubh", KSUBH)
++  NDS32_BUILTIN(unspec_kdmbb, "kdmbb", KDMBB)
++  NDS32_BUILTIN(unspec_kdmbb, "v_kdmbb", V_KDMBB)
++  NDS32_BUILTIN(unspec_kdmbt, "kdmbt", KDMBT)
++  NDS32_BUILTIN(unspec_kdmbt, "v_kdmbt", V_KDMBT)
++  NDS32_BUILTIN(unspec_kdmtb, "kdmtb", KDMTB)
++  NDS32_BUILTIN(unspec_kdmtb, "v_kdmtb", V_KDMTB)
++  NDS32_BUILTIN(unspec_kdmtt, "kdmtt", KDMTT)
++  NDS32_BUILTIN(unspec_kdmtt, "v_kdmtt", V_KDMTT)
++  NDS32_BUILTIN(unspec_khmbb, "khmbb", KHMBB)
++  NDS32_BUILTIN(unspec_khmbb, "v_khmbb", V_KHMBB)
++  NDS32_BUILTIN(unspec_khmbt, "khmbt", KHMBT)
++  NDS32_BUILTIN(unspec_khmbt, "v_khmbt", V_KHMBT)
++  NDS32_BUILTIN(unspec_khmtb, "khmtb", KHMTB)
++  NDS32_BUILTIN(unspec_khmtb, "v_khmtb", V_KHMTB)
++  NDS32_BUILTIN(unspec_khmtt, "khmtt", KHMTT)
++  NDS32_BUILTIN(unspec_khmtt, "v_khmtt", V_KHMTT)
++  NDS32_BUILTIN(unspec_kslraw, "kslraw", KSLRAW)
++  NDS32_BUILTIN(unspec_kslrawu, "kslraw_u", KSLRAW_U)
+   NDS32_BUILTIN(rotrsi3, "rotr", ROTR)
+   NDS32_BUILTIN(unspec_sva, "sva", SVA)
+   NDS32_BUILTIN(unspec_svs, "svs", SVS)
+@@ -603,7 +651,202 @@
+   NDS32_NO_TARGET_BUILTIN(unaligned_store_hw, "unaligned_store_hw", UASTORE_HW)
+   NDS32_NO_TARGET_BUILTIN(unaligned_storesi, "unaligned_store_hw", UASTORE_W)
+   NDS32_NO_TARGET_BUILTIN(unaligned_storedi, "unaligned_store_hw", UASTORE_DW)
+-
++  NDS32_BUILTIN(addv2hi3, "add16", ADD16)
++  NDS32_BUILTIN(addv2hi3, "v_uadd16", V_UADD16)
++  NDS32_BUILTIN(addv2hi3, "v_sadd16", V_SADD16)
++  NDS32_BUILTIN(raddv2hi3, "radd16", RADD16)
++  NDS32_BUILTIN(raddv2hi3, "v_radd16", V_RADD16)
++  NDS32_BUILTIN(uraddv2hi3, "uradd16", URADD16)
++  NDS32_BUILTIN(uraddv2hi3, "v_uradd16", V_URADD16)
++  NDS32_BUILTIN(kaddv2hi3, "kadd16", KADD16)
++  NDS32_BUILTIN(kaddv2hi3, "v_kadd16", V_KADD16)
++  NDS32_BUILTIN(ukaddv2hi3, "ukadd16", UKADD16)
++  NDS32_BUILTIN(ukaddv2hi3, "v_ukadd16", V_UKADD16)
++  NDS32_BUILTIN(subv2hi3, "sub16", SUB16)
++  NDS32_BUILTIN(subv2hi3, "v_usub16", V_USUB16)
++  NDS32_BUILTIN(subv2hi3, "v_ssub16", V_SSUB16)
++  NDS32_BUILTIN(rsubv2hi3, "rsub16", RSUB16)
++  NDS32_BUILTIN(rsubv2hi3, "v_rsub16", V_RSUB16)
++  NDS32_BUILTIN(ursubv2hi3, "ursub16", URSUB16)
++  NDS32_BUILTIN(ursubv2hi3, "v_ursub16", V_URSUB16)
++  NDS32_BUILTIN(ksubv2hi3, "ksub16", KSUB16)
++  NDS32_BUILTIN(ksubv2hi3, "v_ksub16", V_KSUB16)
++  NDS32_BUILTIN(uksubv2hi3, "uksub16", UKSUB16)
++  NDS32_BUILTIN(uksubv2hi3, "v_uksub16", V_UKSUB16)
++  NDS32_BUILTIN(cras16_1, "cras16", CRAS16)
++  NDS32_BUILTIN(cras16_1, "v_ucras16", V_UCRAS16)
++  NDS32_BUILTIN(cras16_1, "v_scras16", V_SCRAS16)
++  NDS32_BUILTIN(rcras16_1, "rcras16", RCRAS16)
++  NDS32_BUILTIN(rcras16_1, "v_rcras16", V_RCRAS16)
++  NDS32_BUILTIN(urcras16_1, "urcras16", URCRAS16)
++  NDS32_BUILTIN(urcras16_1, "v_urcras16", V_URCRAS16)
++  NDS32_BUILTIN(kcras16_1, "kcras16", KCRAS16)
++  NDS32_BUILTIN(kcras16_1, "v_kcras16", V_KCRAS16)
++  NDS32_BUILTIN(ukcras16_1, "ukcras16", UKCRAS16)
++  NDS32_BUILTIN(ukcras16_1, "v_ukcras16", V_UKCRAS16)
++  NDS32_BUILTIN(crsa16_1, "crsa16", CRSA16)
++  NDS32_BUILTIN(crsa16_1, "v_ucrsa16", V_UCRSA16)
++  NDS32_BUILTIN(crsa16_1, "v_scrsa16", V_SCRSA16)
++  NDS32_BUILTIN(rcrsa16_1, "rcrsa16", RCRSA16)
++  NDS32_BUILTIN(rcrsa16_1, "v_rcrsa16", V_RCRSA16)
++  NDS32_BUILTIN(urcrsa16_1, "urcrsa16", URCRSA16)
++  NDS32_BUILTIN(urcrsa16_1, "v_urcrsa16", V_URCRSA16)
++  NDS32_BUILTIN(kcrsa16_1, "kcrsa16", KCRSA16)
++  NDS32_BUILTIN(kcrsa16_1, "v_kcrsa16", V_KCRSA16)
++  NDS32_BUILTIN(ukcrsa16_1, "ukcrsa16", UKCRSA16)
++  NDS32_BUILTIN(ukcrsa16_1, "v_ukcrsa16", V_UKCRSA16)
++  NDS32_BUILTIN(addv4qi3, "add8", ADD8)
++  NDS32_BUILTIN(addv4qi3, "v_uadd8", V_UADD8)
++  NDS32_BUILTIN(addv4qi3, "v_sadd8", V_SADD8)
++  NDS32_BUILTIN(raddv4qi3, "radd8", RADD8)
++  NDS32_BUILTIN(raddv4qi3, "v_radd8", V_RADD8)
++  NDS32_BUILTIN(uraddv4qi3, "uradd8", URADD8)
++  NDS32_BUILTIN(uraddv4qi3, "v_uradd8", V_URADD8)
++  NDS32_BUILTIN(kaddv4qi3, "kadd8", KADD8)
++  NDS32_BUILTIN(kaddv4qi3, "v_kadd8", V_KADD8)
++  NDS32_BUILTIN(ukaddv4qi3, "ukadd8", UKADD8)
++  NDS32_BUILTIN(ukaddv4qi3, "v_ukadd8", V_UKADD8)
++  NDS32_BUILTIN(subv4qi3, "sub8", SUB8)
++  NDS32_BUILTIN(subv4qi3, "v_usub8", V_USUB8)
++  NDS32_BUILTIN(subv4qi3, "v_ssub8", V_SSUB8)
++  NDS32_BUILTIN(rsubv4qi3, "rsub8", RSUB8)
++  NDS32_BUILTIN(rsubv4qi3, "v_rsub8", V_RSUB8)
++  NDS32_BUILTIN(ursubv4qi3, "ursub8", URSUB8)
++  NDS32_BUILTIN(ursubv4qi3, "v_ursub8", V_URSUB8)
++  NDS32_BUILTIN(ksubv4qi3, "ksub8", KSUB8)
++  NDS32_BUILTIN(ksubv4qi3, "v_ksub8", V_KSUB8)
++  NDS32_BUILTIN(uksubv4qi3, "uksub8", UKSUB8)
++  NDS32_BUILTIN(uksubv4qi3, "v_uksub8", V_UKSUB8)
++  NDS32_BUILTIN(ashrv2hi3, "sra16", SRA16)
++  NDS32_BUILTIN(ashrv2hi3, "v_sra16", V_SRA16)
++  NDS32_BUILTIN(sra16_round, "sra16_u", SRA16_U)
++  NDS32_BUILTIN(sra16_round, "v_sra16_u", V_SRA16_U)
++  NDS32_BUILTIN(lshrv2hi3, "srl16", SRL16)
++  NDS32_BUILTIN(lshrv2hi3, "v_srl16", V_SRL16)
++  NDS32_BUILTIN(srl16_round, "srl16_u", SRL16_U)
++  NDS32_BUILTIN(srl16_round, "v_srl16_u", V_SRL16_U)
++  NDS32_BUILTIN(ashlv2hi3, "sll16", SLL16)
++  NDS32_BUILTIN(ashlv2hi3, "v_sll16", V_SLL16)
++  NDS32_BUILTIN(kslli16, "ksll16", KSLL16)
++  NDS32_BUILTIN(kslli16, "v_ksll16", V_KSLL16)
++  NDS32_BUILTIN(kslra16, "kslra16", KSLRA16)
++  NDS32_BUILTIN(kslra16, "v_kslra16", V_KSLRA16)
++  NDS32_BUILTIN(kslra16_round, "kslra16_u", KSLRA16_U)
++  NDS32_BUILTIN(kslra16_round, "v_kslra16_u", V_KSLRA16_U)
++  NDS32_BUILTIN(cmpeq16, "cmpeq16", CMPEQ16)
++  NDS32_BUILTIN(cmpeq16, "v_scmpeq16", V_SCMPEQ16)
++  NDS32_BUILTIN(cmpeq16, "v_ucmpeq16", V_UCMPEQ16)
++  NDS32_BUILTIN(scmplt16, "scmplt16", SCMPLT16)
++  NDS32_BUILTIN(scmplt16, "v_scmplt16", V_SCMPLT16)
++  NDS32_BUILTIN(scmple16, "scmple16", SCMPLE16)
++  NDS32_BUILTIN(scmple16, "v_scmple16", V_SCMPLE16)
++  NDS32_BUILTIN(ucmplt16, "ucmplt16", UCMPLT16)
++  NDS32_BUILTIN(ucmplt16, "v_ucmplt16", V_UCMPLT16)
++  NDS32_BUILTIN(ucmplt16, "ucmple16", UCMPLE16)
++  NDS32_BUILTIN(ucmplt16, "v_ucmple16", V_UCMPLE16)
++  NDS32_BUILTIN(cmpeq8, "cmpeq8", CMPEQ8)
++  NDS32_BUILTIN(cmpeq8, "v_scmpeq8", V_SCMPEQ8)
++  NDS32_BUILTIN(cmpeq8, "v_ucmpeq8", V_UCMPEQ8)
++  NDS32_BUILTIN(scmplt8, "scmplt8", SCMPLT8)
++  NDS32_BUILTIN(scmplt8, "v_scmplt8", V_SCMPLT8)
++  NDS32_BUILTIN(scmple8, "scmple8", SCMPLE8)
++  NDS32_BUILTIN(scmple8, "v_scmple8", V_SCMPLE8)
++  NDS32_BUILTIN(ucmplt8, "ucmplt8", UCMPLT8)
++  NDS32_BUILTIN(ucmplt8, "v_ucmplt8", V_UCMPLT8)
++  NDS32_BUILTIN(ucmplt8, "ucmple8", UCMPLE8)
++  NDS32_BUILTIN(ucmplt8, "v_ucmple8", V_UCMPLE8)
++  NDS32_BUILTIN(sminv2hi3, "smin16", SMIN16)
++  NDS32_BUILTIN(sminv2hi3, "v_smin16", V_SMIN16)
++  NDS32_BUILTIN(uminv2hi3, "umin16", UMIN16)
++  NDS32_BUILTIN(uminv2hi3, "v_umin16", V_UMIN16)
++  NDS32_BUILTIN(smaxv2hi3, "smax16", SMAX16)
++  NDS32_BUILTIN(smaxv2hi3, "v_smax16", V_SMAX16)
++  NDS32_BUILTIN(umaxv2hi3, "umax16", UMAX16)
++  NDS32_BUILTIN(umaxv2hi3, "v_umax16", V_UMAX16)
++  NDS32_BUILTIN(khm16, "khm16", KHM16)
++  NDS32_BUILTIN(khm16, "v_khm16", V_KHM16)
++  NDS32_BUILTIN(khmx16, "khmx16", KHMX16)
++  NDS32_BUILTIN(khmx16, "v_khmx16", V_KHMX16)
++  NDS32_BUILTIN(sminv4qi3, "smin8", SMIN8)
++  NDS32_BUILTIN(sminv4qi3, "v_smin8", V_SMIN8)
++  NDS32_BUILTIN(uminv4qi3, "umin8", UMIN8)
++  NDS32_BUILTIN(uminv4qi3, "v_umin8", V_UMIN8)
++  NDS32_BUILTIN(smaxv4qi3, "smax8", SMAX8)
++  NDS32_BUILTIN(smaxv4qi3, "v_smax8", V_SMAX8)
++  NDS32_BUILTIN(umaxv4qi3, "umax8", UMAX8)
++  NDS32_BUILTIN(umaxv4qi3, "v_umax8", V_UMAX8)
++  NDS32_BUILTIN(raddsi3, "raddw", RADDW)
++  NDS32_BUILTIN(uraddsi3, "uraddw", URADDW)
++  NDS32_BUILTIN(rsubsi3, "rsubw", RSUBW)
++  NDS32_BUILTIN(ursubsi3, "ursubw", URSUBW)
++  NDS32_BUILTIN(sraiu, "sra_u", SRA_U)
++  NDS32_BUILTIN(kssl, "ksll", KSLL)
++  NDS32_BUILTIN(pkbb, "pkbb16", PKBB16)
++  NDS32_BUILTIN(pkbb, "v_pkbb16", V_PKBB16)
++  NDS32_BUILTIN(pkbt, "pkbt16", PKBT16)
++  NDS32_BUILTIN(pkbt, "v_pkbt16", V_PKBT16)
++  NDS32_BUILTIN(pktb, "pktb16", PKTB16)
++  NDS32_BUILTIN(pktb, "v_pktb16", V_PKTB16)
++  NDS32_BUILTIN(pktt, "pktt16", PKTT16)
++  NDS32_BUILTIN(pktt, "v_pktt16", V_PKTT16)
++  NDS32_BUILTIN(smulsi3_highpart, "smmul", SMMUL)
++  NDS32_BUILTIN(smmul_round, "smmul_u", SMMUL_U)
++  NDS32_BUILTIN(smmwb, "smmwb", SMMWB)
++  NDS32_BUILTIN(smmwb, "v_smmwb", V_SMMWB)
++  NDS32_BUILTIN(smmwb_round, "smmwb_u", SMMWB_U)
++  NDS32_BUILTIN(smmwb_round, "v_smmwb_u", V_SMMWB_U)
++  NDS32_BUILTIN(smmwt, "smmwt", SMMWT)
++  NDS32_BUILTIN(smmwt, "v_smmwt", V_SMMWT)
++  NDS32_BUILTIN(smmwt_round, "smmwt_u", SMMWT_U)
++  NDS32_BUILTIN(smmwt_round, "v_smmwt_u", V_SMMWT_U)
++  NDS32_BUILTIN(smbb, "smbb", SMBB)
++  NDS32_BUILTIN(smbb, "v_smbb", V_SMBB)
++  NDS32_BUILTIN(smbt, "smbt", SMBT)
++  NDS32_BUILTIN(smbt, "v_smbt", V_SMBT)
++  NDS32_BUILTIN(smtt, "smtt", SMTT)
++  NDS32_BUILTIN(smtt, "v_smtt", V_SMTT)
++  NDS32_BUILTIN(kmda, "kmda", KMDA)
++  NDS32_BUILTIN(kmda, "v_kmda", V_KMDA)
++  NDS32_BUILTIN(kmxda, "kmxda", KMXDA)
++  NDS32_BUILTIN(kmxda, "v_kmxda", V_KMXDA)
++  NDS32_BUILTIN(smds, "smds", SMDS)
++  NDS32_BUILTIN(smds, "v_smds", V_SMDS)
++  NDS32_BUILTIN(smdrs, "smdrs", SMDRS)
++  NDS32_BUILTIN(smdrs, "v_smdrs", V_SMDRS)
++  NDS32_BUILTIN(smxdsv, "smxds", SMXDS)
++  NDS32_BUILTIN(smxdsv, "v_smxds", V_SMXDS)
++  NDS32_BUILTIN(smal1, "smal", SMAL)
++  NDS32_BUILTIN(smal1, "v_smal", V_SMAL)
++  NDS32_BUILTIN(bitrev, "bitrev", BITREV)
++  NDS32_BUILTIN(wext, "wext", WEXT)
++  NDS32_BUILTIN(adddi3, "sadd64", SADD64)
++  NDS32_BUILTIN(adddi3, "uadd64", UADD64)
++  NDS32_BUILTIN(radddi3, "radd64", RADD64)
++  NDS32_BUILTIN(uradddi3, "uradd64", URADD64)
++  NDS32_BUILTIN(kadddi3, "kadd64", KADD64)
++  NDS32_BUILTIN(ukadddi3, "ukadd64", UKADD64)
++  NDS32_BUILTIN(subdi3, "ssub64", SSUB64)
++  NDS32_BUILTIN(subdi3, "usub64", USUB64)
++  NDS32_BUILTIN(rsubdi3, "rsub64", RSUB64)
++  NDS32_BUILTIN(ursubdi3, "ursub64", URSUB64)
++  NDS32_BUILTIN(ksubdi3, "ksub64", KSUB64)
++  NDS32_BUILTIN(uksubdi3, "uksub64", UKSUB64)
++  NDS32_BUILTIN(smul16, "smul16", SMUL16)
++  NDS32_BUILTIN(smul16, "v_smul16", V_SMUL16)
++  NDS32_BUILTIN(smulx16, "smulx16", SMULX16)
++  NDS32_BUILTIN(smulx16, "v_smulx16", V_SMULX16)
++  NDS32_BUILTIN(umul16, "umul16", UMUL16)
++  NDS32_BUILTIN(umul16, "v_umul16", V_UMUL16)
++  NDS32_BUILTIN(umulx16, "umulx16", UMULX16)
++  NDS32_BUILTIN(umulx16, "v_umulx16", V_UMULX16)
++  NDS32_BUILTIN(kwmmul, "kwmmul", KWMMUL)
++  NDS32_BUILTIN(kwmmul_round, "kwmmul_u", KWMMUL_U)
++  NDS32_NO_TARGET_BUILTIN(unaligned_storev2hi,
++			  "put_unaligned_u16x2", UASTORE_U16)
++  NDS32_NO_TARGET_BUILTIN(unaligned_storev2hi,
++			  "put_unaligned_s16x2", UASTORE_S16)
++  NDS32_NO_TARGET_BUILTIN(unaligned_storev4qi, "put_unaligned_u8x4", UASTORE_U8)
++  NDS32_NO_TARGET_BUILTIN(unaligned_storev4qi, "put_unaligned_s8x4", UASTORE_S8)
+ };
+ 
+ /* Two-argument intrinsics with an immediate second argument.  */
+@@ -617,6 +860,22 @@
+   NDS32_BUILTIN(unspec_clips, "clips", CLIPS)
+   NDS32_NO_TARGET_BUILTIN(unspec_teqz, "teqz", TEQZ)
+   NDS32_NO_TARGET_BUILTIN(unspec_tnez, "tnez", TNEZ)
++  NDS32_BUILTIN(ashrv2hi3, "srl16", SRL16)
++  NDS32_BUILTIN(ashrv2hi3, "v_srl16", V_SRL16)
++  NDS32_BUILTIN(srl16_round, "srl16_u", SRL16_U)
++  NDS32_BUILTIN(srl16_round, "v_srl16_u", V_SRL16_U)
++  NDS32_BUILTIN(kslli16, "ksll16", KSLL16)
++  NDS32_BUILTIN(kslli16, "v_ksll16", V_KSLL16)
++  NDS32_BUILTIN(sclip16, "sclip16", SCLIP16)
++  NDS32_BUILTIN(sclip16, "v_sclip16", V_SCLIP16)
++  NDS32_BUILTIN(uclip16, "uclip16", UCLIP16)
++  NDS32_BUILTIN(uclip16, "v_uclip16", V_UCLIP16)
++  NDS32_BUILTIN(sraiu, "sra_u", SRA_U)
++  NDS32_BUILTIN(kssl, "ksll", KSLL)
++  NDS32_BUILTIN(bitrev, "bitrev", BITREV)
++  NDS32_BUILTIN(wext, "wext", WEXT)
++  NDS32_BUILTIN(uclip32, "uclip32", UCLIP32)
++  NDS32_BUILTIN(sclip32, "sclip32", SCLIP32)
+ };
+ 
+ /* Intrinsics that take three arguments.  */
+@@ -625,6 +884,67 @@
+   NDS32_BUILTIN(unspec_pbsada, "pbsada", PBSADA)
+   NDS32_NO_TARGET_BUILTIN(bse, "bse", BSE)
+   NDS32_NO_TARGET_BUILTIN(bsp, "bsp", BSP)
++  NDS32_BUILTIN(kmabb, "kmabb", KMABB)
++  NDS32_BUILTIN(kmabb, "v_kmabb", V_KMABB)
++  NDS32_BUILTIN(kmabt, "kmabt", KMABT)
++  NDS32_BUILTIN(kmabt, "v_kmabt", V_KMABT)
++  NDS32_BUILTIN(kmatt, "kmatt", KMATT)
++  NDS32_BUILTIN(kmatt, "v_kmatt", V_KMATT)
++  NDS32_BUILTIN(kmada, "kmada", KMADA)
++  NDS32_BUILTIN(kmada, "v_kmada", V_KMADA)
++  NDS32_BUILTIN(kmaxda, "kmaxda", KMAXDA)
++  NDS32_BUILTIN(kmaxda, "v_kmaxda", V_KMAXDA)
++  NDS32_BUILTIN(kmads, "kmads", KMADS)
++  NDS32_BUILTIN(kmads, "v_kmads", V_KMADS)
++  NDS32_BUILTIN(kmadrs, "kmadrs", KMADRS)
++  NDS32_BUILTIN(kmadrs, "v_kmadrs", V_KMADRS)
++  NDS32_BUILTIN(kmaxds, "kmaxds", KMAXDS)
++  NDS32_BUILTIN(kmaxds, "v_kmaxds", V_KMAXDS)
++  NDS32_BUILTIN(kmsda, "kmsda", KMSDA)
++  NDS32_BUILTIN(kmsda, "v_kmsda", V_KMSDA)
++  NDS32_BUILTIN(kmsxda, "kmsxda", KMSXDA)
++  NDS32_BUILTIN(kmsxda, "v_kmsxda", V_KMSXDA)
++  NDS32_BUILTIN(bpick1, "bpick", BPICK)
++  NDS32_BUILTIN(smar64_1, "smar64", SMAR64)
++  NDS32_BUILTIN(smsr64, "smsr64", SMSR64)
++  NDS32_BUILTIN(umar64_1, "umar64", UMAR64)
++  NDS32_BUILTIN(umsr64, "umsr64", UMSR64)
++  NDS32_BUILTIN(kmar64_1, "kmar64", KMAR64)
++  NDS32_BUILTIN(kmsr64, "kmsr64", KMSR64)
++  NDS32_BUILTIN(ukmar64_1, "ukmar64", UKMAR64)
++  NDS32_BUILTIN(ukmsr64, "ukmsr64", UKMSR64)
++  NDS32_BUILTIN(smalbb, "smalbb", SMALBB)
++  NDS32_BUILTIN(smalbb, "v_smalbb", V_SMALBB)
++  NDS32_BUILTIN(smalbt, "smalbt", SMALBT)
++  NDS32_BUILTIN(smalbt, "v_smalbt", V_SMALBT)
++  NDS32_BUILTIN(smaltt, "smaltt", SMALTT)
++  NDS32_BUILTIN(smaltt, "v_smaltt", V_SMALTT)
++  NDS32_BUILTIN(smalda1, "smalda", SMALDA)
++  NDS32_BUILTIN(smalda1, "v_smalda", V_SMALDA)
++  NDS32_BUILTIN(smalxda1, "smalxda", SMALXDA)
++  NDS32_BUILTIN(smalxda1, "v_smalxda", V_SMALXDA)
++  NDS32_BUILTIN(smalds1, "smalds", SMALDS)
++  NDS32_BUILTIN(smalds1, "v_smalds", V_SMALDS)
++  NDS32_BUILTIN(smaldrs3, "smaldrs", SMALDRS)
++  NDS32_BUILTIN(smaldrs3, "v_smaldrs", V_SMALDRS)
++  NDS32_BUILTIN(smalxds1, "smalxds", SMALXDS)
++  NDS32_BUILTIN(smalxds1, "v_smalxds", V_SMALXDS)
++  NDS32_BUILTIN(smslda1, "smslda", SMSLDA)
++  NDS32_BUILTIN(smslda1, "v_smslda", V_SMSLDA)
++  NDS32_BUILTIN(smslxda1, "smslxda", SMSLXDA)
++  NDS32_BUILTIN(smslxda1, "v_smslxda", V_SMSLXDA)
++  NDS32_BUILTIN(kmmawb, "kmmawb", KMMAWB)
++  NDS32_BUILTIN(kmmawb, "v_kmmawb", V_KMMAWB)
++  NDS32_BUILTIN(kmmawb_round, "kmmawb_u", KMMAWB_U)
++  NDS32_BUILTIN(kmmawb_round, "v_kmmawb_u", V_KMMAWB_U)
++  NDS32_BUILTIN(kmmawt, "kmmawt", KMMAWT)
++  NDS32_BUILTIN(kmmawt, "v_kmmawt", V_KMMAWT)
++  NDS32_BUILTIN(kmmawt_round, "kmmawt_u", KMMAWT_U)
++  NDS32_BUILTIN(kmmawt_round, "v_kmmawt_u", V_KMMAWT_U)
++  NDS32_BUILTIN(kmmac, "kmmac", KMMAC)
++  NDS32_BUILTIN(kmmac_round, "kmmac_u", KMMAC_U)
++  NDS32_BUILTIN(kmmsb, "kmmsb", KMMSB)
++  NDS32_BUILTIN(kmmsb_round, "kmmsb_u", KMMSB_U)
+ };
+ 
+ /* Three-argument intrinsics with an immediate third argument.  */
+@@ -634,6 +954,7 @@
+   NDS32_NO_TARGET_BUILTIN(prefetch_hw, "prefetch_hw", DPREF_HW)
+   NDS32_NO_TARGET_BUILTIN(prefetch_w, "prefetch_w", DPREF_W)
+   NDS32_NO_TARGET_BUILTIN(prefetch_dw, "prefetch_dw", DPREF_DW)
++  NDS32_BUILTIN(insb, "insb", INSB)
+ };
+ 
+ /* Intrinsics that load a value.  */
+@@ -676,6 +997,11 @@
+   unsigned i;
+   struct builtin_description *d;
+ 
++  if (!NDS32_EXT_DSP_P ()
++      && fcode > NDS32_BUILTIN_DSP_BEGIN
++      && fcode < NDS32_BUILTIN_DSP_END)
++    error ("don't support DSP extension instructions");
++
+   switch (fcode)
+     {
+     /* FPU Register Transfer.  */
+@@ -812,6 +1138,9 @@
+     case NDS32_BUILTIN_CCTL_L1D_WBALL_ONE_LVL:
+       emit_insn (gen_cctl_l1d_wball_one_lvl());
+       return target;
++    case NDS32_BUILTIN_CLROV:
++      emit_insn (gen_unspec_volatile_clrov ());
++      return target;
+     case NDS32_BUILTIN_STANDBY_NO_WAKE_GRANT:
+       emit_insn (gen_unspec_standby_no_wake_grant ());
+       return target;
+@@ -947,10 +1276,18 @@
+ 			NDS32_BUILTIN_ ## CODE, BUILT_IN_MD, NULL, NULL_TREE)
+ 
+   /* Looking for return type and argument can be found in tree.h file.  */
++  tree ptr_char_type_node = build_pointer_type (char_type_node);
+   tree ptr_uchar_type_node = build_pointer_type (unsigned_char_type_node);
+   tree ptr_ushort_type_node = build_pointer_type (short_unsigned_type_node);
++  tree ptr_short_type_node = build_pointer_type (short_integer_type_node);
+   tree ptr_uint_type_node = build_pointer_type (unsigned_type_node);
+   tree ptr_ulong_type_node = build_pointer_type (long_long_unsigned_type_node);
++  tree v4qi_type_node = build_vector_type (intQI_type_node, 4);
++  tree u_v4qi_type_node = build_vector_type (unsigned_intQI_type_node, 4);
++  tree v2hi_type_node = build_vector_type (intHI_type_node, 2);
++  tree u_v2hi_type_node = build_vector_type (unsigned_intHI_type_node, 2);
++  tree v2si_type_node = build_vector_type (intSI_type_node, 2);
++  tree u_v2si_type_node = build_vector_type (unsigned_intSI_type_node, 2);
+ 
+   /* Cache.  */
+   ADD_NDS32_BUILTIN1 ("isync", void, ptr_uint, ISYNC);
+@@ -1050,6 +1387,31 @@
+   ADD_NDS32_BUILTIN2 ("se_ffmism", integer, unsigned, unsigned, FFMISM);
+   ADD_NDS32_BUILTIN2 ("se_flmism", integer, unsigned, unsigned, FLMISM);
+ 
++  /* SATURATION  */
++  ADD_NDS32_BUILTIN2 ("kaddw", integer, integer, integer, KADDW);
++  ADD_NDS32_BUILTIN2 ("ksubw", integer, integer, integer, KSUBW);
++  ADD_NDS32_BUILTIN2 ("kaddh", integer, integer, integer, KADDH);
++  ADD_NDS32_BUILTIN2 ("ksubh", integer, integer, integer, KSUBH);
++  ADD_NDS32_BUILTIN2 ("kdmbb", integer, unsigned, unsigned, KDMBB);
++  ADD_NDS32_BUILTIN2 ("v_kdmbb", integer, v2hi, v2hi, V_KDMBB);
++  ADD_NDS32_BUILTIN2 ("kdmbt", integer, unsigned, unsigned, KDMBT);
++  ADD_NDS32_BUILTIN2 ("v_kdmbt", integer, v2hi, v2hi, V_KDMBT);
++  ADD_NDS32_BUILTIN2 ("kdmtb", integer, unsigned, unsigned, KDMTB);
++  ADD_NDS32_BUILTIN2 ("v_kdmtb", integer, v2hi, v2hi, V_KDMTB);
++  ADD_NDS32_BUILTIN2 ("kdmtt", integer, unsigned, unsigned, KDMTT);
++  ADD_NDS32_BUILTIN2 ("v_kdmtt", integer, v2hi, v2hi, V_KDMTT);
++  ADD_NDS32_BUILTIN2 ("khmbb", integer, unsigned, unsigned, KHMBB);
++  ADD_NDS32_BUILTIN2 ("v_khmbb", integer, v2hi, v2hi, V_KHMBB);
++  ADD_NDS32_BUILTIN2 ("khmbt", integer, unsigned, unsigned, KHMBT);
++  ADD_NDS32_BUILTIN2 ("v_khmbt", integer, v2hi, v2hi, V_KHMBT);
++  ADD_NDS32_BUILTIN2 ("khmtb", integer, unsigned, unsigned, KHMTB);
++  ADD_NDS32_BUILTIN2 ("v_khmtb", integer, v2hi, v2hi, V_KHMTB);
++  ADD_NDS32_BUILTIN2 ("khmtt", integer, unsigned, unsigned, KHMTT);
++  ADD_NDS32_BUILTIN2 ("v_khmtt", integer, v2hi, v2hi, V_KHMTT);
++  ADD_NDS32_BUILTIN2 ("kslraw", integer, integer, integer, KSLRAW);
++  ADD_NDS32_BUILTIN2 ("kslraw_u", integer, integer, integer, KSLRAW_U);
++  ADD_NDS32_BUILTIN0 ("rdov", unsigned, RDOV);
++  ADD_NDS32_BUILTIN0 ("clrov", void, CLROV);
+ 
+   /* ROTR  */
+   ADD_NDS32_BUILTIN2 ("rotr", unsigned, unsigned, unsigned, ROTR);
+@@ -1109,4 +1471,384 @@
+   ADD_NDS32_BUILTIN0 ("enable_unaligned", void, ENABLE_UNALIGNED);
+   ADD_NDS32_BUILTIN0 ("disable_unaligned", void, DISABLE_UNALIGNED);
+ 
++  /* DSP Extension: SIMD 16bit Add and Subtract.  */
++  ADD_NDS32_BUILTIN2 ("add16", unsigned, unsigned, unsigned, ADD16);
++  ADD_NDS32_BUILTIN2 ("v_uadd16", u_v2hi, u_v2hi, u_v2hi, V_UADD16);
++  ADD_NDS32_BUILTIN2 ("v_sadd16", v2hi, v2hi, v2hi, V_SADD16);
++  ADD_NDS32_BUILTIN2 ("radd16", unsigned, unsigned, unsigned, RADD16);
++  ADD_NDS32_BUILTIN2 ("v_radd16", v2hi, v2hi, v2hi, V_RADD16);
++  ADD_NDS32_BUILTIN2 ("uradd16", unsigned, unsigned, unsigned, URADD16);
++  ADD_NDS32_BUILTIN2 ("v_uradd16", u_v2hi, u_v2hi, u_v2hi, V_URADD16);
++  ADD_NDS32_BUILTIN2 ("kadd16", unsigned, unsigned, unsigned, KADD16);
++  ADD_NDS32_BUILTIN2 ("v_kadd16", v2hi, v2hi, v2hi, V_KADD16);
++  ADD_NDS32_BUILTIN2 ("ukadd16", unsigned, unsigned, unsigned, UKADD16);
++  ADD_NDS32_BUILTIN2 ("v_ukadd16", u_v2hi, u_v2hi, u_v2hi, V_UKADD16);
++  ADD_NDS32_BUILTIN2 ("sub16", unsigned, unsigned, unsigned, SUB16);
++  ADD_NDS32_BUILTIN2 ("v_usub16", u_v2hi, u_v2hi, u_v2hi, V_USUB16);
++  ADD_NDS32_BUILTIN2 ("v_ssub16", v2hi, v2hi, v2hi, V_SSUB16);
++  ADD_NDS32_BUILTIN2 ("rsub16", unsigned, unsigned, unsigned, RSUB16);
++  ADD_NDS32_BUILTIN2 ("v_rsub16", v2hi, v2hi, v2hi, V_RSUB16);
++  ADD_NDS32_BUILTIN2 ("ursub16", unsigned, unsigned, unsigned, URSUB16);
++  ADD_NDS32_BUILTIN2 ("v_ursub16", u_v2hi, u_v2hi, u_v2hi, V_URSUB16);
++  ADD_NDS32_BUILTIN2 ("ksub16", unsigned, unsigned, unsigned, KSUB16);
++  ADD_NDS32_BUILTIN2 ("v_ksub16", v2hi, v2hi, v2hi, V_KSUB16);
++  ADD_NDS32_BUILTIN2 ("uksub16", unsigned, unsigned, unsigned, UKSUB16);
++  ADD_NDS32_BUILTIN2 ("v_uksub16", u_v2hi, u_v2hi, u_v2hi, V_UKSUB16);
++  ADD_NDS32_BUILTIN2 ("cras16", unsigned, unsigned, unsigned, CRAS16);
++  ADD_NDS32_BUILTIN2 ("v_ucras16", u_v2hi, u_v2hi, u_v2hi, V_UCRAS16);
++  ADD_NDS32_BUILTIN2 ("v_scras16", v2hi, v2hi, v2hi, V_SCRAS16);
++  ADD_NDS32_BUILTIN2 ("rcras16", unsigned, unsigned, unsigned, RCRAS16);
++  ADD_NDS32_BUILTIN2 ("v_rcras16", v2hi, v2hi, v2hi, V_RCRAS16);
++  ADD_NDS32_BUILTIN2 ("urcras16", unsigned, unsigned, unsigned, URCRAS16);
++  ADD_NDS32_BUILTIN2 ("v_urcras16", u_v2hi, u_v2hi, u_v2hi, V_URCRAS16);
++  ADD_NDS32_BUILTIN2 ("kcras16", unsigned, unsigned, unsigned, KCRAS16);
++  ADD_NDS32_BUILTIN2 ("v_kcras16", v2hi, v2hi, v2hi, V_KCRAS16);
++  ADD_NDS32_BUILTIN2 ("ukcras16", unsigned, unsigned, unsigned, UKCRAS16);
++  ADD_NDS32_BUILTIN2 ("v_ukcras16", u_v2hi, u_v2hi, u_v2hi, V_UKCRAS16);
++  ADD_NDS32_BUILTIN2 ("crsa16", unsigned, unsigned, unsigned, CRSA16);
++  ADD_NDS32_BUILTIN2 ("v_ucrsa16", u_v2hi, u_v2hi, u_v2hi, V_UCRSA16);
++  ADD_NDS32_BUILTIN2 ("v_scrsa16", v2hi, v2hi, v2hi, V_SCRSA16);
++  ADD_NDS32_BUILTIN2 ("rcrsa16", unsigned, unsigned, unsigned, RCRSA16);
++  ADD_NDS32_BUILTIN2 ("v_rcrsa16", v2hi, v2hi, v2hi, V_RCRSA16);
++  ADD_NDS32_BUILTIN2 ("urcrsa16", unsigned, unsigned, unsigned, URCRSA16);
++  ADD_NDS32_BUILTIN2 ("v_urcrsa16", u_v2hi, u_v2hi, u_v2hi, V_URCRSA16);
++  ADD_NDS32_BUILTIN2 ("kcrsa16", unsigned, unsigned, unsigned, KCRSA16);
++  ADD_NDS32_BUILTIN2 ("v_kcrsa16", v2hi, v2hi, v2hi, V_KCRSA16);
++  ADD_NDS32_BUILTIN2 ("ukcrsa16", unsigned, unsigned, unsigned, UKCRSA16);
++  ADD_NDS32_BUILTIN2 ("v_ukcrsa16", u_v2hi, u_v2hi, u_v2hi, V_UKCRSA16);
++
++  /* DSP Extension: SIMD 8bit Add and Subtract.  */
++  ADD_NDS32_BUILTIN2 ("add8", integer, integer, integer, ADD8);
++  ADD_NDS32_BUILTIN2 ("v_uadd8", u_v4qi, u_v4qi, u_v4qi, V_UADD8);
++  ADD_NDS32_BUILTIN2 ("v_sadd8", v4qi, v4qi, v4qi, V_SADD8);
++  ADD_NDS32_BUILTIN2 ("radd8", unsigned, unsigned, unsigned, RADD8);
++  ADD_NDS32_BUILTIN2 ("v_radd8", v4qi, v4qi, v4qi, V_RADD8);
++  ADD_NDS32_BUILTIN2 ("uradd8", unsigned, unsigned, unsigned, URADD8);
++  ADD_NDS32_BUILTIN2 ("v_uradd8", u_v4qi, u_v4qi, u_v4qi, V_URADD8);
++  ADD_NDS32_BUILTIN2 ("kadd8", unsigned, unsigned, unsigned, KADD8);
++  ADD_NDS32_BUILTIN2 ("v_kadd8", v4qi, v4qi, v4qi, V_KADD8);
++  ADD_NDS32_BUILTIN2 ("ukadd8", unsigned, unsigned, unsigned, UKADD8);
++  ADD_NDS32_BUILTIN2 ("v_ukadd8", u_v4qi, u_v4qi, u_v4qi, V_UKADD8);
++  ADD_NDS32_BUILTIN2 ("sub8", integer, integer, integer, SUB8);
++  ADD_NDS32_BUILTIN2 ("v_usub8", u_v4qi, u_v4qi, u_v4qi, V_USUB8);
++  ADD_NDS32_BUILTIN2 ("v_ssub8", v4qi, v4qi, v4qi, V_SSUB8);
++  ADD_NDS32_BUILTIN2 ("rsub8", unsigned, unsigned, unsigned, RSUB8);
++  ADD_NDS32_BUILTIN2 ("v_rsub8", v4qi, v4qi, v4qi, V_RSUB8);
++  ADD_NDS32_BUILTIN2 ("ursub8", unsigned, unsigned, unsigned, URSUB8);
++  ADD_NDS32_BUILTIN2 ("v_ursub8", u_v4qi, u_v4qi, u_v4qi, V_URSUB8);
++  ADD_NDS32_BUILTIN2 ("ksub8", unsigned, unsigned, unsigned, KSUB8);
++  ADD_NDS32_BUILTIN2 ("v_ksub8", v4qi, v4qi, v4qi, V_KSUB8);
++  ADD_NDS32_BUILTIN2 ("uksub8", unsigned, unsigned, unsigned, UKSUB8);
++  ADD_NDS32_BUILTIN2 ("v_uksub8", u_v4qi, u_v4qi, u_v4qi, V_UKSUB8);
++
++  /* DSP Extension: SIMD 16bit Shift.  */
++  ADD_NDS32_BUILTIN2 ("sra16", unsigned, unsigned, unsigned, SRA16);
++  ADD_NDS32_BUILTIN2 ("v_sra16", v2hi, v2hi, unsigned, V_SRA16);
++  ADD_NDS32_BUILTIN2 ("sra16_u", unsigned, unsigned, unsigned, SRA16_U);
++  ADD_NDS32_BUILTIN2 ("v_sra16_u", v2hi, v2hi, unsigned, V_SRA16_U);
++  ADD_NDS32_BUILTIN2 ("srl16", unsigned, unsigned, unsigned, SRL16);
++  ADD_NDS32_BUILTIN2 ("v_srl16", u_v2hi, u_v2hi, unsigned, V_SRL16);
++  ADD_NDS32_BUILTIN2 ("srl16_u", unsigned, unsigned, unsigned, SRL16_U);
++  ADD_NDS32_BUILTIN2 ("v_srl16_u", u_v2hi, u_v2hi, unsigned, V_SRL16_U);
++  ADD_NDS32_BUILTIN2 ("sll16", unsigned, unsigned, unsigned, SLL16);
++  ADD_NDS32_BUILTIN2 ("v_sll16", u_v2hi, u_v2hi, unsigned, V_SLL16);
++  ADD_NDS32_BUILTIN2 ("ksll16", unsigned, unsigned, unsigned, KSLL16);
++  ADD_NDS32_BUILTIN2 ("v_ksll16", v2hi, v2hi, unsigned, V_KSLL16);
++  ADD_NDS32_BUILTIN2 ("kslra16", unsigned, unsigned, unsigned, KSLRA16);
++  ADD_NDS32_BUILTIN2 ("v_kslra16", v2hi, v2hi, unsigned, V_KSLRA16);
++  ADD_NDS32_BUILTIN2 ("kslra16_u", unsigned, unsigned, unsigned, KSLRA16_U);
++  ADD_NDS32_BUILTIN2 ("v_kslra16_u", v2hi, v2hi, unsigned, V_KSLRA16_U);
++
++  /* DSP Extension: 16bit Compare.  */
++  ADD_NDS32_BUILTIN2 ("cmpeq16", unsigned, unsigned, unsigned, CMPEQ16);
++  ADD_NDS32_BUILTIN2 ("v_scmpeq16", u_v2hi, v2hi, v2hi, V_SCMPEQ16);
++  ADD_NDS32_BUILTIN2 ("v_ucmpeq16", u_v2hi, u_v2hi, u_v2hi, V_UCMPEQ16);
++  ADD_NDS32_BUILTIN2 ("scmplt16", unsigned, unsigned, unsigned, SCMPLT16);
++  ADD_NDS32_BUILTIN2 ("v_scmplt16", u_v2hi, v2hi, v2hi, V_SCMPLT16);
++  ADD_NDS32_BUILTIN2 ("scmple16", unsigned, unsigned, unsigned, SCMPLE16);
++  ADD_NDS32_BUILTIN2 ("v_scmple16", u_v2hi, v2hi, v2hi, V_SCMPLE16);
++  ADD_NDS32_BUILTIN2 ("ucmplt16", unsigned, unsigned, unsigned, UCMPLT16);
++  ADD_NDS32_BUILTIN2 ("v_ucmplt16", u_v2hi, u_v2hi, u_v2hi, V_UCMPLT16);
++  ADD_NDS32_BUILTIN2 ("ucmple16", unsigned, unsigned, unsigned, UCMPLE16);
++  ADD_NDS32_BUILTIN2 ("v_ucmple16", u_v2hi, u_v2hi, u_v2hi, V_UCMPLE16);
++
++  /* DSP Extension: 8bit Compare.  */
++  ADD_NDS32_BUILTIN2 ("cmpeq8", unsigned, unsigned, unsigned, CMPEQ8);
++  ADD_NDS32_BUILTIN2 ("v_scmpeq8", u_v4qi, v4qi, v4qi, V_SCMPEQ8);
++  ADD_NDS32_BUILTIN2 ("v_ucmpeq8", u_v4qi, u_v4qi, u_v4qi, V_UCMPEQ8);
++  ADD_NDS32_BUILTIN2 ("scmplt8", unsigned, unsigned, unsigned, SCMPLT8);
++  ADD_NDS32_BUILTIN2 ("v_scmplt8", u_v4qi, v4qi, v4qi, V_SCMPLT8);
++  ADD_NDS32_BUILTIN2 ("scmple8", unsigned, unsigned, unsigned, SCMPLE8);
++  ADD_NDS32_BUILTIN2 ("v_scmple8", u_v4qi, v4qi, v4qi, V_SCMPLE8);
++  ADD_NDS32_BUILTIN2 ("ucmplt8", unsigned, unsigned, unsigned, UCMPLT8);
++  ADD_NDS32_BUILTIN2 ("v_ucmplt8", u_v4qi, u_v4qi, u_v4qi, V_UCMPLT8);
++  ADD_NDS32_BUILTIN2 ("ucmple8", unsigned, unsigned, unsigned, UCMPLE8);
++  ADD_NDS32_BUILTIN2 ("v_ucmple8", u_v4qi, u_v4qi, u_v4qi, V_UCMPLE8);
++
++  /* DSP Extension: SIMD 16bit MISC.  */
++  ADD_NDS32_BUILTIN2 ("smin16", unsigned, unsigned, unsigned, SMIN16);
++  ADD_NDS32_BUILTIN2 ("v_smin16", v2hi, v2hi, v2hi, V_SMIN16);
++  ADD_NDS32_BUILTIN2 ("umin16", unsigned, unsigned, unsigned, UMIN16);
++  ADD_NDS32_BUILTIN2 ("v_umin16", u_v2hi, u_v2hi, u_v2hi, V_UMIN16);
++  ADD_NDS32_BUILTIN2 ("smax16", unsigned, unsigned, unsigned, SMAX16);
++  ADD_NDS32_BUILTIN2 ("v_smax16", v2hi, v2hi, v2hi, V_SMAX16);
++  ADD_NDS32_BUILTIN2 ("umax16", unsigned, unsigned, unsigned, UMAX16);
++  ADD_NDS32_BUILTIN2 ("v_umax16", u_v2hi, u_v2hi, u_v2hi, V_UMAX16);
++  ADD_NDS32_BUILTIN2 ("sclip16", unsigned, unsigned, unsigned, SCLIP16);
++  ADD_NDS32_BUILTIN2 ("v_sclip16", v2hi, v2hi, unsigned, V_SCLIP16);
++  ADD_NDS32_BUILTIN2 ("uclip16", unsigned, unsigned, unsigned, UCLIP16);
++  ADD_NDS32_BUILTIN2 ("v_uclip16", v2hi, v2hi, unsigned, V_UCLIP16);
++  ADD_NDS32_BUILTIN2 ("khm16", unsigned, unsigned, unsigned, KHM16);
++  ADD_NDS32_BUILTIN2 ("v_khm16", v2hi, v2hi, v2hi, V_KHM16);
++  ADD_NDS32_BUILTIN2 ("khmx16", unsigned, unsigned, unsigned, KHMX16);
++  ADD_NDS32_BUILTIN2 ("v_khmx16", v2hi, v2hi, v2hi, V_KHMX16);
++  ADD_NDS32_BUILTIN1 ("kabs16", unsigned, unsigned, KABS16);
++  ADD_NDS32_BUILTIN1 ("v_kabs16", v2hi, v2hi, V_KABS16);
++  ADD_NDS32_BUILTIN2 ("smul16", long_long_unsigned, unsigned, unsigned, SMUL16);
++  ADD_NDS32_BUILTIN2 ("v_smul16", v2si, v2hi, v2hi, V_SMUL16);
++  ADD_NDS32_BUILTIN2 ("smulx16",
++		      long_long_unsigned, unsigned, unsigned, SMULX16);
++  ADD_NDS32_BUILTIN2 ("v_smulx16", v2si, v2hi, v2hi, V_SMULX16);
++  ADD_NDS32_BUILTIN2 ("umul16", long_long_unsigned, unsigned, unsigned, UMUL16);
++  ADD_NDS32_BUILTIN2 ("v_umul16", u_v2si, u_v2hi, u_v2hi, V_UMUL16);
++  ADD_NDS32_BUILTIN2 ("umulx16",
++		      long_long_unsigned, unsigned, unsigned, UMULX16);
++  ADD_NDS32_BUILTIN2 ("v_umulx16", u_v2si, u_v2hi, u_v2hi, V_UMULX16);
++
++  /* DSP Extension: SIMD 8bit MISC.  */
++  ADD_NDS32_BUILTIN2 ("smin8", unsigned, unsigned, unsigned, SMIN8);
++  ADD_NDS32_BUILTIN2 ("v_smin8", v4qi, v4qi, v4qi, V_SMIN8);
++  ADD_NDS32_BUILTIN2 ("umin8", unsigned, unsigned, unsigned, UMIN8);
++  ADD_NDS32_BUILTIN2 ("v_umin8", u_v4qi, u_v4qi, u_v4qi, V_UMIN8);
++  ADD_NDS32_BUILTIN2 ("smax8", unsigned, unsigned, unsigned, SMAX8);
++  ADD_NDS32_BUILTIN2 ("v_smax8", v4qi, v4qi, v4qi, V_SMAX8);
++  ADD_NDS32_BUILTIN2 ("umax8", unsigned, unsigned, unsigned, UMAX8);
++  ADD_NDS32_BUILTIN2 ("v_umax8", u_v4qi, u_v4qi, u_v4qi, V_UMAX8);
++  ADD_NDS32_BUILTIN1 ("kabs8", unsigned, unsigned, KABS8);
++  ADD_NDS32_BUILTIN1 ("v_kabs8", v4qi, v4qi, V_KABS8);
++
++  /* DSP Extension: 8bit Unpacking.  */
++  ADD_NDS32_BUILTIN1 ("sunpkd810", unsigned, unsigned, SUNPKD810);
++  ADD_NDS32_BUILTIN1 ("v_sunpkd810", v2hi, v4qi, V_SUNPKD810);
++  ADD_NDS32_BUILTIN1 ("sunpkd820", unsigned, unsigned, SUNPKD820);
++  ADD_NDS32_BUILTIN1 ("v_sunpkd820", v2hi, v4qi, V_SUNPKD820);
++  ADD_NDS32_BUILTIN1 ("sunpkd830", unsigned, unsigned, SUNPKD830);
++  ADD_NDS32_BUILTIN1 ("v_sunpkd830", v2hi, v4qi, V_SUNPKD830);
++  ADD_NDS32_BUILTIN1 ("sunpkd831", unsigned, unsigned, SUNPKD831);
++  ADD_NDS32_BUILTIN1 ("v_sunpkd831", v2hi, v4qi, V_SUNPKD831);
++  ADD_NDS32_BUILTIN1 ("zunpkd810", unsigned, unsigned, ZUNPKD810);
++  ADD_NDS32_BUILTIN1 ("v_zunpkd810", u_v2hi, u_v4qi, V_ZUNPKD810);
++  ADD_NDS32_BUILTIN1 ("zunpkd820", unsigned, unsigned, ZUNPKD820);
++  ADD_NDS32_BUILTIN1 ("v_zunpkd820", u_v2hi, u_v4qi, V_ZUNPKD820);
++  ADD_NDS32_BUILTIN1 ("zunpkd830", unsigned, unsigned, ZUNPKD830);
++  ADD_NDS32_BUILTIN1 ("v_zunpkd830", u_v2hi, u_v4qi, V_ZUNPKD830);
++  ADD_NDS32_BUILTIN1 ("zunpkd831", unsigned, unsigned, ZUNPKD831);
++  ADD_NDS32_BUILTIN1 ("v_zunpkd831", u_v2hi, u_v4qi, V_ZUNPKD831);
++
++  /* DSP Extension: 32bit Add and Subtract.  */
++  ADD_NDS32_BUILTIN2 ("raddw", integer, integer, integer, RADDW);
++  ADD_NDS32_BUILTIN2 ("uraddw", unsigned, unsigned, unsigned, URADDW);
++  ADD_NDS32_BUILTIN2 ("rsubw", integer, integer, integer, RSUBW);
++  ADD_NDS32_BUILTIN2 ("ursubw", unsigned, unsigned, unsigned, URSUBW);
++
++  /* DSP Extension: 32bit Shift.  */
++  ADD_NDS32_BUILTIN2 ("sra_u", integer, integer, unsigned, SRA_U);
++  ADD_NDS32_BUILTIN2 ("ksll", integer, integer, unsigned, KSLL);
++
++  /* DSP Extension: 16bit Packing.  */
++  ADD_NDS32_BUILTIN2 ("pkbb16", unsigned, unsigned, unsigned, PKBB16);
++  ADD_NDS32_BUILTIN2 ("v_pkbb16", u_v2hi, u_v2hi, u_v2hi, V_PKBB16);
++  ADD_NDS32_BUILTIN2 ("pkbt16", unsigned, unsigned, unsigned, PKBT16);
++  ADD_NDS32_BUILTIN2 ("v_pkbt16", u_v2hi, u_v2hi, u_v2hi, V_PKBT16);
++  ADD_NDS32_BUILTIN2 ("pktb16", unsigned, unsigned, unsigned, PKTB16);
++  ADD_NDS32_BUILTIN2 ("v_pktb16", u_v2hi, u_v2hi, u_v2hi, V_PKTB16);
++  ADD_NDS32_BUILTIN2 ("pktt16", unsigned, unsigned, unsigned, PKTT16);
++  ADD_NDS32_BUILTIN2 ("v_pktt16", u_v2hi, u_v2hi, u_v2hi, V_PKTT16);
++
++  /* DSP Extension: Signed MSW 32x32 Multiply and ADD.  */
++  ADD_NDS32_BUILTIN2 ("smmul", integer, integer, integer, SMMUL);
++  ADD_NDS32_BUILTIN2 ("smmul_u", integer, integer, integer, SMMUL_U);
++  ADD_NDS32_BUILTIN3 ("kmmac", integer, integer, integer, integer, KMMAC);
++  ADD_NDS32_BUILTIN3 ("kmmac_u", integer, integer, integer, integer, KMMAC_U);
++  ADD_NDS32_BUILTIN3 ("kmmsb", integer, integer, integer, integer, KMMSB);
++  ADD_NDS32_BUILTIN3 ("kmmsb_u", integer, integer, integer, integer, KMMSB_U);
++  ADD_NDS32_BUILTIN2 ("kwmmul", integer, integer, integer, KWMMUL);
++  ADD_NDS32_BUILTIN2 ("kwmmul_u", integer, integer, integer, KWMMUL_U);
++
++  /* DSP Extension: Most Significant Word 32x16 Multiply and ADD.  */
++  ADD_NDS32_BUILTIN2 ("smmwb", integer, integer, unsigned, SMMWB);
++  ADD_NDS32_BUILTIN2 ("v_smmwb", integer, integer, v2hi, V_SMMWB);
++  ADD_NDS32_BUILTIN2 ("smmwb_u", integer, integer, unsigned, SMMWB_U);
++  ADD_NDS32_BUILTIN2 ("v_smmwb_u", integer, integer, v2hi, V_SMMWB_U);
++  ADD_NDS32_BUILTIN2 ("smmwt", integer, integer, unsigned, SMMWT);
++  ADD_NDS32_BUILTIN2 ("v_smmwt", integer, integer, v2hi, V_SMMWT);
++  ADD_NDS32_BUILTIN2 ("smmwt_u", integer, integer, unsigned, SMMWT_U);
++  ADD_NDS32_BUILTIN2 ("v_smmwt_u", integer, integer, v2hi, V_SMMWT_U);
++  ADD_NDS32_BUILTIN3 ("kmmawb", integer, integer, integer, unsigned, KMMAWB);
++  ADD_NDS32_BUILTIN3 ("v_kmmawb", integer, integer, integer, v2hi, V_KMMAWB);
++  ADD_NDS32_BUILTIN3 ("kmmawb_u",
++		      integer, integer, integer, unsigned, KMMAWB_U);
++  ADD_NDS32_BUILTIN3 ("v_kmmawb_u",
++		      integer, integer, integer, v2hi, V_KMMAWB_U);
++  ADD_NDS32_BUILTIN3 ("kmmawt", integer, integer, integer, unsigned, KMMAWT);
++  ADD_NDS32_BUILTIN3 ("v_kmmawt", integer, integer, integer, v2hi, V_KMMAWT);
++  ADD_NDS32_BUILTIN3 ("kmmawt_u",
++		      integer, integer, integer, unsigned, KMMAWT_U);
++  ADD_NDS32_BUILTIN3 ("v_kmmawt_u",
++		      integer, integer, integer, v2hi, V_KMMAWT_U);
++
++  /* DSP Extension: Signed 16bit Multiply with ADD/Subtract.  */
++  ADD_NDS32_BUILTIN2 ("smbb", integer, unsigned, unsigned, SMBB);
++  ADD_NDS32_BUILTIN2 ("v_smbb", integer, v2hi, v2hi, V_SMBB);
++  ADD_NDS32_BUILTIN2 ("smbt", integer, unsigned, unsigned, SMBT);
++  ADD_NDS32_BUILTIN2 ("v_smbt", integer, v2hi, v2hi, V_SMBT);
++  ADD_NDS32_BUILTIN2 ("smtt", integer, unsigned, unsigned, SMTT);
++  ADD_NDS32_BUILTIN2 ("v_smtt", integer, v2hi, v2hi, V_SMTT);
++  ADD_NDS32_BUILTIN2 ("kmda", integer, unsigned, unsigned, KMDA);
++  ADD_NDS32_BUILTIN2 ("v_kmda", integer, v2hi, v2hi, V_KMDA);
++  ADD_NDS32_BUILTIN2 ("kmxda", integer, unsigned, unsigned, KMXDA);
++  ADD_NDS32_BUILTIN2 ("v_kmxda", integer, v2hi, v2hi, V_KMXDA);
++  ADD_NDS32_BUILTIN2 ("smds", integer, unsigned, unsigned, SMDS);
++  ADD_NDS32_BUILTIN2 ("v_smds", integer, v2hi, v2hi, V_SMDS);
++  ADD_NDS32_BUILTIN2 ("smdrs", integer, unsigned, unsigned, SMDRS);
++  ADD_NDS32_BUILTIN2 ("v_smdrs", integer, v2hi, v2hi, V_SMDRS);
++  ADD_NDS32_BUILTIN2 ("smxds", integer, unsigned, unsigned, SMXDS);
++  ADD_NDS32_BUILTIN2 ("v_smxds", integer, v2hi, v2hi, V_SMXDS);
++  ADD_NDS32_BUILTIN3 ("kmabb", integer, integer, unsigned, unsigned, KMABB);
++  ADD_NDS32_BUILTIN3 ("v_kmabb", integer, integer, v2hi, v2hi, V_KMABB);
++  ADD_NDS32_BUILTIN3 ("kmabt", integer, integer, unsigned, unsigned, KMABT);
++  ADD_NDS32_BUILTIN3 ("v_kmabt", integer, integer, v2hi, v2hi, V_KMABT);
++  ADD_NDS32_BUILTIN3 ("kmatt", integer, integer, unsigned, unsigned, KMATT);
++  ADD_NDS32_BUILTIN3 ("v_kmatt", integer, integer, v2hi, v2hi, V_KMATT);
++  ADD_NDS32_BUILTIN3 ("kmada", integer, integer, unsigned, unsigned, KMADA);
++  ADD_NDS32_BUILTIN3 ("v_kmada", integer, integer, v2hi, v2hi, V_KMADA);
++  ADD_NDS32_BUILTIN3 ("kmaxda", integer, integer, unsigned, unsigned, KMAXDA);
++  ADD_NDS32_BUILTIN3 ("v_kmaxda", integer, integer, v2hi, v2hi, V_KMAXDA);
++  ADD_NDS32_BUILTIN3 ("kmads", integer, integer, unsigned, unsigned, KMADS);
++  ADD_NDS32_BUILTIN3 ("v_kmads", integer, integer, v2hi, v2hi, V_KMADS);
++  ADD_NDS32_BUILTIN3 ("kmadrs", integer, integer, unsigned, unsigned, KMADRS);
++  ADD_NDS32_BUILTIN3 ("v_kmadrs", integer, integer, v2hi, v2hi, V_KMADRS);
++  ADD_NDS32_BUILTIN3 ("kmaxds", integer, integer, unsigned, unsigned, KMAXDS);
++  ADD_NDS32_BUILTIN3 ("v_kmaxds", integer, integer, v2hi, v2hi, V_KMAXDS);
++  ADD_NDS32_BUILTIN3 ("kmsda", integer, integer, unsigned, unsigned, KMSDA);
++  ADD_NDS32_BUILTIN3 ("v_kmsda", integer, integer, v2hi, v2hi, V_KMSDA);
++  ADD_NDS32_BUILTIN3 ("kmsxda", integer, integer, unsigned, unsigned, KMSXDA);
++  ADD_NDS32_BUILTIN3 ("v_kmsxda", integer, integer, v2hi, v2hi, V_KMSXDA);
++
++  /* DSP Extension: Signed 16bit Multiply with 64bit ADD/Subtract.  */
++  ADD_NDS32_BUILTIN2 ("smal", long_long_integer,
++		      long_long_integer, unsigned, SMAL);
++  ADD_NDS32_BUILTIN2 ("v_smal", long_long_integer,
++		      long_long_integer, v2hi, V_SMAL);
++
++  /* DSP Extension: 32bit MISC.  */
++  ADD_NDS32_BUILTIN2 ("bitrev", unsigned, unsigned, unsigned, BITREV);
++  ADD_NDS32_BUILTIN2 ("wext", unsigned, long_long_integer, unsigned, WEXT);
++  ADD_NDS32_BUILTIN3 ("bpick", unsigned, unsigned, unsigned, unsigned, BPICK);
++  ADD_NDS32_BUILTIN3 ("insb", unsigned, unsigned, unsigned, unsigned, INSB);
++
++  /* DSP Extension: 64bit Add and Subtract.  */
++  ADD_NDS32_BUILTIN2 ("sadd64", long_long_integer,
++		      long_long_integer, long_long_integer, SADD64);
++  ADD_NDS32_BUILTIN2 ("uadd64", long_long_unsigned,
++		      long_long_unsigned, long_long_unsigned, UADD64);
++  ADD_NDS32_BUILTIN2 ("radd64", long_long_integer,
++		      long_long_integer, long_long_integer, RADD64);
++  ADD_NDS32_BUILTIN2 ("uradd64", long_long_unsigned,
++		      long_long_unsigned, long_long_unsigned, URADD64);
++  ADD_NDS32_BUILTIN2 ("kadd64", long_long_integer,
++		      long_long_integer, long_long_integer, KADD64);
++  ADD_NDS32_BUILTIN2 ("ukadd64", long_long_unsigned,
++		      long_long_unsigned, long_long_unsigned, UKADD64);
++  ADD_NDS32_BUILTIN2 ("ssub64", long_long_integer,
++		      long_long_integer, long_long_integer, SSUB64);
++  ADD_NDS32_BUILTIN2 ("usub64", long_long_unsigned,
++		      long_long_unsigned, long_long_unsigned, USUB64);
++  ADD_NDS32_BUILTIN2 ("rsub64", long_long_integer,
++		      long_long_integer, long_long_integer, RSUB64);
++  ADD_NDS32_BUILTIN2 ("ursub64", long_long_unsigned,
++		      long_long_unsigned, long_long_unsigned, URSUB64);
++  ADD_NDS32_BUILTIN2 ("ksub64", long_long_integer,
++		      long_long_integer, long_long_integer, KSUB64);
++  ADD_NDS32_BUILTIN2 ("uksub64", long_long_unsigned,
++		      long_long_unsigned, long_long_unsigned, UKSUB64);
++
++  /* DSP Extension: 32bit Multiply with 64bit Add/Subtract.  */
++  ADD_NDS32_BUILTIN3 ("smar64", long_long_integer,
++		      long_long_integer, integer, integer, SMAR64);
++  ADD_NDS32_BUILTIN3 ("smsr64", long_long_integer,
++		      long_long_integer, integer, integer, SMSR64);
++  ADD_NDS32_BUILTIN3 ("umar64", long_long_unsigned,
++		      long_long_unsigned, unsigned, unsigned, UMAR64);
++  ADD_NDS32_BUILTIN3 ("umsr64", long_long_unsigned,
++		      long_long_unsigned, unsigned, unsigned, UMSR64);
++  ADD_NDS32_BUILTIN3 ("kmar64", long_long_integer,
++		      long_long_integer, integer, integer, KMAR64);
++  ADD_NDS32_BUILTIN3 ("kmsr64", long_long_integer,
++		      long_long_integer, integer, integer, KMSR64);
++  ADD_NDS32_BUILTIN3 ("ukmar64", long_long_unsigned,
++		      long_long_unsigned, unsigned, unsigned, UKMAR64);
++  ADD_NDS32_BUILTIN3 ("ukmsr64", long_long_unsigned,
++		      long_long_unsigned, unsigned, unsigned, UKMSR64);
++
++  /* DSP Extension: Signed 16bit Multiply with 64bit Add/Subtract.  */
++  ADD_NDS32_BUILTIN3 ("smalbb", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALBB);
++  ADD_NDS32_BUILTIN3 ("v_smalbb", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALBB);
++  ADD_NDS32_BUILTIN3 ("smalbt", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALBT);
++  ADD_NDS32_BUILTIN3 ("v_smalbt", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALBT);
++  ADD_NDS32_BUILTIN3 ("smaltt", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALTT);
++  ADD_NDS32_BUILTIN3 ("v_smaltt", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALTT);
++  ADD_NDS32_BUILTIN3 ("smalda", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALDA);
++  ADD_NDS32_BUILTIN3 ("v_smalda", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALDA);
++  ADD_NDS32_BUILTIN3 ("smalxda", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALXDA);
++  ADD_NDS32_BUILTIN3 ("v_smalxda", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALXDA);
++  ADD_NDS32_BUILTIN3 ("smalds", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALDS);
++  ADD_NDS32_BUILTIN3 ("v_smalds", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALDS);
++  ADD_NDS32_BUILTIN3 ("smaldrs", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALDRS);
++  ADD_NDS32_BUILTIN3 ("v_smaldrs", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALDRS);
++  ADD_NDS32_BUILTIN3 ("smalxds", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMALXDS);
++  ADD_NDS32_BUILTIN3 ("v_smalxds", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMALXDS);
++  ADD_NDS32_BUILTIN3 ("smslda", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMSLDA);
++  ADD_NDS32_BUILTIN3 ("v_smslda", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMSLDA);
++  ADD_NDS32_BUILTIN3 ("smslxda", long_long_integer,
++		      long_long_integer, unsigned, unsigned, SMSLXDA);
++  ADD_NDS32_BUILTIN3 ("v_smslxda", long_long_integer,
++		      long_long_integer, v2hi, v2hi, V_SMSLXDA);
++
++  /* DSP Extension: augmented baseline.  */
++  ADD_NDS32_BUILTIN2 ("uclip32", unsigned, integer, unsigned, UCLIP32);
++  ADD_NDS32_BUILTIN2 ("sclip32", integer, integer, unsigned, SCLIP32);
++  ADD_NDS32_BUILTIN1 ("kabs", integer, integer, KABS);
++
++  /* DSP Extension: vector type unaligned Load/Store  */
++  ADD_NDS32_BUILTIN1 ("get_unaligned_u16x2", u_v2hi, ptr_ushort, UALOAD_U16);
++  ADD_NDS32_BUILTIN1 ("get_unaligned_s16x2", v2hi, ptr_short, UALOAD_S16);
++  ADD_NDS32_BUILTIN1 ("get_unaligned_u8x4", u_v4qi, ptr_uchar, UALOAD_U8);
++  ADD_NDS32_BUILTIN1 ("get_unaligned_s8x4", v4qi, ptr_char, UALOAD_S8);
++  ADD_NDS32_BUILTIN2 ("put_unaligned_u16x2", void, ptr_ushort,
++		      u_v2hi, UASTORE_U16);
++  ADD_NDS32_BUILTIN2 ("put_unaligned_s16x2", void, ptr_short,
++		      v2hi, UASTORE_S16);
++  ADD_NDS32_BUILTIN2 ("put_unaligned_u8x4", void, ptr_uchar,
++		      u_v4qi, UASTORE_U8);
++  ADD_NDS32_BUILTIN2 ("put_unaligned_s8x4", void, ptr_char,
++		      v4qi, UASTORE_S8);
+ }
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32_intrinsic.h gcc-8.2.0/gcc/config/nds32/nds32_intrinsic.h
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32_intrinsic.h	2018-04-22 09:46:39.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32_intrinsic.h	2019-01-25 15:38:32.833242671 +0100
+@@ -26,6 +26,13 @@
+ #ifndef _NDS32_INTRINSIC_H
+ #define _NDS32_INTRINSIC_H
+ 
++typedef signed char int8x4_t __attribute ((vector_size(4)));
++typedef short int16x2_t __attribute ((vector_size(4)));
++typedef int int32x2_t __attribute__((vector_size(8)));
++typedef unsigned char uint8x4_t __attribute__ ((vector_size (4)));
++typedef unsigned short uint16x2_t __attribute__ ((vector_size (4)));
++typedef unsigned int uint32x2_t __attribute__((vector_size(8)));
++
+ /* General instrinsic register names.  */
+ enum nds32_intrinsic_registers
+ {
+@@ -691,6 +698,55 @@
+ #define __nds32__tlbop_flua() \
+ (__builtin_nds32_tlbop_flua())
+ 
++#define __nds32__kaddw(a, b) \
++  (__builtin_nds32_kaddw ((a), (b)))
++#define __nds32__kaddh(a, b) \
++  (__builtin_nds32_kaddh ((a), (b)))
++#define __nds32__ksubw(a, b) \
++  (__builtin_nds32_ksubw ((a), (b)))
++#define __nds32__ksubh(a, b) \
++  (__builtin_nds32_ksubh ((a), (b)))
++#define __nds32__kdmbb(a, b) \
++  (__builtin_nds32_kdmbb ((a), (b)))
++#define __nds32__v_kdmbb(a, b) \
++  (__builtin_nds32_v_kdmbb ((a), (b)))
++#define __nds32__kdmbt(a, b) \
++  (__builtin_nds32_kdmbt ((a), (b)))
++#define __nds32__v_kdmbt(a, b) \
++  (__builtin_nds32_v_kdmbt ((a), (b)))
++#define __nds32__kdmtb(a, b) \
++  (__builtin_nds32_kdmtb ((a), (b)))
++#define __nds32__v_kdmtb(a, b) \
++  (__builtin_nds32_v_kdmtb ((a), (b)))
++#define __nds32__kdmtt(a, b) \
++  (__builtin_nds32_kdmtt ((a), (b)))
++#define __nds32__v_kdmtt(a, b) \
++  (__builtin_nds32_v_kdmtt ((a), (b)))
++#define __nds32__khmbb(a, b) \
++  (__builtin_nds32_khmbb ((a), (b)))
++#define __nds32__v_khmbb(a, b) \
++  (__builtin_nds32_v_khmbb ((a), (b)))
++#define __nds32__khmbt(a, b) \
++  (__builtin_nds32_khmbt ((a), (b)))
++#define __nds32__v_khmbt(a, b) \
++  (__builtin_nds32_v_khmbt ((a), (b)))
++#define __nds32__khmtb(a, b) \
++  (__builtin_nds32_khmtb ((a), (b)))
++#define __nds32__v_khmtb(a, b) \
++  (__builtin_nds32_v_khmtb ((a), (b)))
++#define __nds32__khmtt(a, b) \
++  (__builtin_nds32_khmtt ((a), (b)))
++#define __nds32__v_khmtt(a, b) \
++  (__builtin_nds32_v_khmtt ((a), (b)))
++#define __nds32__kslraw(a, b) \
++  (__builtin_nds32_kslraw ((a), (b)))
++#define __nds32__kslraw_u(a, b) \
++  (__builtin_nds32_kslraw_u ((a), (b)))
++
++#define __nds32__rdov() \
++  (__builtin_nds32_rdov())
++#define __nds32__clrov() \
++  (__builtin_nds32_clrov())
+ #define __nds32__gie_dis() \
+   (__builtin_nds32_gie_dis())
+ #define __nds32__gie_en() \
+@@ -720,10 +776,622 @@
+ #define __nds32__get_trig_type(a) \
+   (__builtin_nds32_get_trig_type ((a)))
+ 
++#define __nds32__get_unaligned_hw(a) \
++  (__builtin_nds32_unaligned_load_hw ((a)))
++#define __nds32__get_unaligned_w(a) \
++  (__builtin_nds32_unaligned_load_w ((a)))
++#define __nds32__get_unaligned_dw(a) \
++  (__builtin_nds32_unaligned_load_dw ((a)))
++#define __nds32__put_unaligned_hw(a, data) \
++  (__builtin_nds32_unaligned_store_hw ((a), (data)))
++#define __nds32__put_unaligned_w(a, data) \
++  (__builtin_nds32_unaligned_store_w ((a), (data)))
++#define __nds32__put_unaligned_dw(a, data) \
++  (__builtin_nds32_unaligned_store_dw ((a), (data)))
++
++#define __nds32__add16(a, b) \
++  (__builtin_nds32_add16 ((a), (b)))
++#define __nds32__v_uadd16(a, b) \
++  (__builtin_nds32_v_uadd16 ((a), (b)))
++#define __nds32__v_sadd16(a, b) \
++  (__builtin_nds32_v_sadd16 ((a), (b)))
++#define __nds32__radd16(a, b) \
++  (__builtin_nds32_radd16 ((a), (b)))
++#define __nds32__v_radd16(a, b) \
++  (__builtin_nds32_v_radd16 ((a), (b)))
++#define __nds32__uradd16(a, b) \
++  (__builtin_nds32_uradd16 ((a), (b)))
++#define __nds32__v_uradd16(a, b) \
++  (__builtin_nds32_v_uradd16 ((a), (b)))
++#define __nds32__kadd16(a, b) \
++  (__builtin_nds32_kadd16 ((a), (b)))
++#define __nds32__v_kadd16(a, b) \
++  (__builtin_nds32_v_kadd16 ((a), (b)))
++#define __nds32__ukadd16(a, b) \
++  (__builtin_nds32_ukadd16 ((a), (b)))
++#define __nds32__v_ukadd16(a, b) \
++  (__builtin_nds32_v_ukadd16 ((a), (b)))
++#define __nds32__sub16(a, b) \
++  (__builtin_nds32_sub16 ((a), (b)))
++#define __nds32__v_usub16(a, b) \
++  (__builtin_nds32_v_usub16 ((a), (b)))
++#define __nds32__v_ssub16(a, b) \
++  (__builtin_nds32_v_ssub16 ((a), (b)))
++#define __nds32__rsub16(a, b) \
++  (__builtin_nds32_rsub16 ((a), (b)))
++#define __nds32__v_rsub16(a, b) \
++  (__builtin_nds32_v_rsub16 ((a), (b)))
++#define __nds32__ursub16(a, b) \
++  (__builtin_nds32_ursub16 ((a), (b)))
++#define __nds32__v_ursub16(a, b) \
++  (__builtin_nds32_v_ursub16 ((a), (b)))
++#define __nds32__ksub16(a, b) \
++  (__builtin_nds32_ksub16 ((a), (b)))
++#define __nds32__v_ksub16(a, b) \
++  (__builtin_nds32_v_ksub16 ((a), (b)))
++#define __nds32__uksub16(a, b) \
++  (__builtin_nds32_uksub16 ((a), (b)))
++#define __nds32__v_uksub16(a, b) \
++  (__builtin_nds32_v_uksub16 ((a), (b)))
++#define __nds32__cras16(a, b) \
++  (__builtin_nds32_cras16 ((a), (b)))
++#define __nds32__v_ucras16(a, b) \
++  (__builtin_nds32_v_ucras16 ((a), (b)))
++#define __nds32__v_scras16(a, b) \
++  (__builtin_nds32_v_scras16 ((a), (b)))
++#define __nds32__rcras16(a, b) \
++  (__builtin_nds32_rcras16 ((a), (b)))
++#define __nds32__v_rcras16(a, b) \
++  (__builtin_nds32_v_rcras16 ((a), (b)))
++#define __nds32__urcras16(a, b) \
++  (__builtin_nds32_urcras16 ((a), (b)))
++#define __nds32__v_urcras16(a, b) \
++  (__builtin_nds32_v_urcras16 ((a), (b)))
++#define __nds32__kcras16(a, b) \
++  (__builtin_nds32_kcras16 ((a), (b)))
++#define __nds32__v_kcras16(a, b) \
++  (__builtin_nds32_v_kcras16 ((a), (b)))
++#define __nds32__ukcras16(a, b) \
++  (__builtin_nds32_ukcras16 ((a), (b)))
++#define __nds32__v_ukcras16(a, b) \
++  (__builtin_nds32_v_ukcras16 ((a), (b)))
++#define __nds32__crsa16(a, b) \
++  (__builtin_nds32_crsa16 ((a), (b)))
++#define __nds32__v_ucrsa16(a, b) \
++  (__builtin_nds32_v_ucrsa16 ((a), (b)))
++#define __nds32__v_scrsa16(a, b) \
++  (__builtin_nds32_v_scrsa16 ((a), (b)))
++#define __nds32__rcrsa16(a, b) \
++  (__builtin_nds32_rcrsa16 ((a), (b)))
++#define __nds32__v_rcrsa16(a, b) \
++  (__builtin_nds32_v_rcrsa16 ((a), (b)))
++#define __nds32__urcrsa16(a, b) \
++  (__builtin_nds32_urcrsa16 ((a), (b)))
++#define __nds32__v_urcrsa16(a, b) \
++  (__builtin_nds32_v_urcrsa16 ((a), (b)))
++#define __nds32__kcrsa16(a, b) \
++  (__builtin_nds32_kcrsa16 ((a), (b)))
++#define __nds32__v_kcrsa16(a, b) \
++  (__builtin_nds32_v_kcrsa16 ((a), (b)))
++#define __nds32__ukcrsa16(a, b) \
++  (__builtin_nds32_ukcrsa16 ((a), (b)))
++#define __nds32__v_ukcrsa16(a, b) \
++  (__builtin_nds32_v_ukcrsa16 ((a), (b)))
++
++#define __nds32__add8(a, b) \
++  (__builtin_nds32_add8 ((a), (b)))
++#define __nds32__v_uadd8(a, b) \
++  (__builtin_nds32_v_uadd8 ((a), (b)))
++#define __nds32__v_sadd8(a, b) \
++  (__builtin_nds32_v_sadd8 ((a), (b)))
++#define __nds32__radd8(a, b) \
++  (__builtin_nds32_radd8 ((a), (b)))
++#define __nds32__v_radd8(a, b) \
++  (__builtin_nds32_v_radd8 ((a), (b)))
++#define __nds32__uradd8(a, b) \
++  (__builtin_nds32_uradd8 ((a), (b)))
++#define __nds32__v_uradd8(a, b) \
++  (__builtin_nds32_v_uradd8 ((a), (b)))
++#define __nds32__kadd8(a, b) \
++  (__builtin_nds32_kadd8 ((a), (b)))
++#define __nds32__v_kadd8(a, b) \
++  (__builtin_nds32_v_kadd8 ((a), (b)))
++#define __nds32__ukadd8(a, b) \
++  (__builtin_nds32_ukadd8 ((a), (b)))
++#define __nds32__v_ukadd8(a, b) \
++  (__builtin_nds32_v_ukadd8 ((a), (b)))
++#define __nds32__sub8(a, b) \
++  (__builtin_nds32_sub8 ((a), (b)))
++#define __nds32__v_usub8(a, b) \
++  (__builtin_nds32_v_usub8 ((a), (b)))
++#define __nds32__v_ssub8(a, b) \
++  (__builtin_nds32_v_ssub8 ((a), (b)))
++#define __nds32__rsub8(a, b) \
++  (__builtin_nds32_rsub8 ((a), (b)))
++#define __nds32__v_rsub8(a, b) \
++  (__builtin_nds32_v_rsub8 ((a), (b)))
++#define __nds32__ursub8(a, b) \
++  (__builtin_nds32_ursub8 ((a), (b)))
++#define __nds32__v_ursub8(a, b) \
++  (__builtin_nds32_v_ursub8 ((a), (b)))
++#define __nds32__ksub8(a, b) \
++  (__builtin_nds32_ksub8 ((a), (b)))
++#define __nds32__v_ksub8(a, b) \
++  (__builtin_nds32_v_ksub8 ((a), (b)))
++#define __nds32__uksub8(a, b) \
++  (__builtin_nds32_uksub8 ((a), (b)))
++#define __nds32__v_uksub8(a, b) \
++  (__builtin_nds32_v_uksub8 ((a), (b)))
++
++#define __nds32__sra16(a, b) \
++  (__builtin_nds32_sra16 ((a), (b)))
++#define __nds32__v_sra16(a, b) \
++  (__builtin_nds32_v_sra16 ((a), (b)))
++#define __nds32__sra16_u(a, b) \
++  (__builtin_nds32_sra16_u ((a), (b)))
++#define __nds32__v_sra16_u(a, b) \
++  (__builtin_nds32_v_sra16_u ((a), (b)))
++#define __nds32__srl16(a, b) \
++  (__builtin_nds32_srl16 ((a), (b)))
++#define __nds32__v_srl16(a, b) \
++  (__builtin_nds32_v_srl16 ((a), (b)))
++#define __nds32__srl16_u(a, b) \
++  (__builtin_nds32_srl16_u ((a), (b)))
++#define __nds32__v_srl16_u(a, b) \
++  (__builtin_nds32_v_srl16_u ((a), (b)))
++#define __nds32__sll16(a, b) \
++  (__builtin_nds32_sll16 ((a), (b)))
++#define __nds32__v_sll16(a, b) \
++  (__builtin_nds32_v_sll16 ((a), (b)))
++#define __nds32__ksll16(a, b) \
++  (__builtin_nds32_ksll16 ((a), (b)))
++#define __nds32__v_ksll16(a, b) \
++  (__builtin_nds32_v_ksll16 ((a), (b)))
++#define __nds32__kslra16(a, b) \
++  (__builtin_nds32_kslra16 ((a), (b)))
++#define __nds32__v_kslra16(a, b) \
++  (__builtin_nds32_v_kslra16 ((a), (b)))
++#define __nds32__kslra16_u(a, b) \
++  (__builtin_nds32_kslra16_u ((a), (b)))
++#define __nds32__v_kslra16_u(a, b) \
++  (__builtin_nds32_v_kslra16_u ((a), (b)))
++
++#define __nds32__cmpeq16(a, b) \
++  (__builtin_nds32_cmpeq16 ((a), (b)))
++#define __nds32__v_scmpeq16(a, b) \
++  (__builtin_nds32_v_scmpeq16 ((a), (b)))
++#define __nds32__v_ucmpeq16(a, b) \
++  (__builtin_nds32_v_ucmpeq16 ((a), (b)))
++#define __nds32__scmplt16(a, b) \
++  (__builtin_nds32_scmplt16 ((a), (b)))
++#define __nds32__v_scmplt16(a, b) \
++  (__builtin_nds32_v_scmplt16 ((a), (b)))
++#define __nds32__scmple16(a, b) \
++  (__builtin_nds32_scmple16 ((a), (b)))
++#define __nds32__v_scmple16(a, b) \
++  (__builtin_nds32_v_scmple16 ((a), (b)))
++#define __nds32__ucmplt16(a, b) \
++  (__builtin_nds32_ucmplt16 ((a), (b)))
++#define __nds32__v_ucmplt16(a, b) \
++  (__builtin_nds32_v_ucmplt16 ((a), (b)))
++#define __nds32__ucmple16(a, b) \
++  (__builtin_nds32_ucmple16 ((a), (b)))
++#define __nds32__v_ucmple16(a, b) \
++  (__builtin_nds32_v_ucmple16 ((a), (b)))
++
++#define __nds32__cmpeq8(a, b) \
++  (__builtin_nds32_cmpeq8 ((a), (b)))
++#define __nds32__v_scmpeq8(a, b) \
++  (__builtin_nds32_v_scmpeq8 ((a), (b)))
++#define __nds32__v_ucmpeq8(a, b) \
++  (__builtin_nds32_v_ucmpeq8 ((a), (b)))
++#define __nds32__scmplt8(a, b) \
++  (__builtin_nds32_scmplt8 ((a), (b)))
++#define __nds32__v_scmplt8(a, b) \
++  (__builtin_nds32_v_scmplt8 ((a), (b)))
++#define __nds32__scmple8(a, b) \
++  (__builtin_nds32_scmple8 ((a), (b)))
++#define __nds32__v_scmple8(a, b) \
++  (__builtin_nds32_v_scmple8 ((a), (b)))
++#define __nds32__ucmplt8(a, b) \
++  (__builtin_nds32_ucmplt8 ((a), (b)))
++#define __nds32__v_ucmplt8(a, b) \
++  (__builtin_nds32_v_ucmplt8 ((a), (b)))
++#define __nds32__ucmple8(a, b) \
++  (__builtin_nds32_ucmple8 ((a), (b)))
++#define __nds32__v_ucmple8(a, b) \
++  (__builtin_nds32_v_ucmple8 ((a), (b)))
++
++#define __nds32__smin16(a, b) \
++  (__builtin_nds32_smin16 ((a), (b)))
++#define __nds32__v_smin16(a, b) \
++  (__builtin_nds32_v_smin16 ((a), (b)))
++#define __nds32__umin16(a, b) \
++  (__builtin_nds32_umin16 ((a), (b)))
++#define __nds32__v_umin16(a, b) \
++  (__builtin_nds32_v_umin16 ((a), (b)))
++#define __nds32__smax16(a, b) \
++  (__builtin_nds32_smax16 ((a), (b)))
++#define __nds32__v_smax16(a, b) \
++  (__builtin_nds32_v_smax16 ((a), (b)))
++#define __nds32__umax16(a, b) \
++  (__builtin_nds32_umax16 ((a), (b)))
++#define __nds32__v_umax16(a, b) \
++  (__builtin_nds32_v_umax16 ((a), (b)))
++#define __nds32__sclip16(a, b) \
++  (__builtin_nds32_sclip16 ((a), (b)))
++#define __nds32__v_sclip16(a, b) \
++  (__builtin_nds32_v_sclip16 ((a), (b)))
++#define __nds32__uclip16(a, b) \
++  (__builtin_nds32_uclip16 ((a), (b)))
++#define __nds32__v_uclip16(a, b) \
++  (__builtin_nds32_v_uclip16 ((a), (b)))
++#define __nds32__khm16(a, b) \
++  (__builtin_nds32_khm16 ((a), (b)))
++#define __nds32__v_khm16(a, b) \
++  (__builtin_nds32_v_khm16 ((a), (b)))
++#define __nds32__khmx16(a, b) \
++  (__builtin_nds32_khmx16 ((a), (b)))
++#define __nds32__v_khmx16(a, b) \
++  (__builtin_nds32_v_khmx16 ((a), (b)))
++#define __nds32__kabs16(a) \
++  (__builtin_nds32_kabs16 ((a)))
++#define __nds32__v_kabs16(a) \
++  (__builtin_nds32_v_kabs16 ((a)))
++
++#define __nds32__smin8(a, b) \
++  (__builtin_nds32_smin8 ((a), (b)))
++#define __nds32__v_smin8(a, b) \
++  (__builtin_nds32_v_smin8 ((a), (b)))
++#define __nds32__umin8(a, b) \
++  (__builtin_nds32_umin8 ((a), (b)))
++#define __nds32__v_umin8(a, b) \
++  (__builtin_nds32_v_umin8 ((a), (b)))
++#define __nds32__smax8(a, b) \
++  (__builtin_nds32_smax8 ((a), (b)))
++#define __nds32__v_smax8(a, b) \
++  (__builtin_nds32_v_smax8 ((a), (b)))
++#define __nds32__umax8(a, b) \
++  (__builtin_nds32_umax8 ((a), (b)))
++#define __nds32__v_umax8(a, b) \
++  (__builtin_nds32_v_umax8 ((a), (b)))
++#define __nds32__kabs8(a) \
++  (__builtin_nds32_kabs8 ((a)))
++#define __nds32__v_kabs8(a) \
++  (__builtin_nds32_v_kabs8 ((a)))
++
++#define __nds32__sunpkd810(a) \
++  (__builtin_nds32_sunpkd810 ((a)))
++#define __nds32__v_sunpkd810(a) \
++  (__builtin_nds32_v_sunpkd810 ((a)))
++#define __nds32__sunpkd820(a) \
++  (__builtin_nds32_sunpkd820 ((a)))
++#define __nds32__v_sunpkd820(a) \
++  (__builtin_nds32_v_sunpkd820 ((a)))
++#define __nds32__sunpkd830(a) \
++  (__builtin_nds32_sunpkd830 ((a)))
++#define __nds32__v_sunpkd830(a) \
++  (__builtin_nds32_v_sunpkd830 ((a)))
++#define __nds32__sunpkd831(a) \
++  (__builtin_nds32_sunpkd831 ((a)))
++#define __nds32__v_sunpkd831(a) \
++  (__builtin_nds32_v_sunpkd831 ((a)))
++#define __nds32__zunpkd810(a) \
++  (__builtin_nds32_zunpkd810 ((a)))
++#define __nds32__v_zunpkd810(a) \
++  (__builtin_nds32_v_zunpkd810 ((a)))
++#define __nds32__zunpkd820(a) \
++  (__builtin_nds32_zunpkd820 ((a)))
++#define __nds32__v_zunpkd820(a) \
++  (__builtin_nds32_v_zunpkd820 ((a)))
++#define __nds32__zunpkd830(a) \
++  (__builtin_nds32_zunpkd830 ((a)))
++#define __nds32__v_zunpkd830(a) \
++  (__builtin_nds32_v_zunpkd830 ((a)))
++#define __nds32__zunpkd831(a) \
++  (__builtin_nds32_zunpkd831 ((a)))
++#define __nds32__v_zunpkd831(a) \
++  (__builtin_nds32_v_zunpkd831 ((a)))
++
++#define __nds32__raddw(a, b) \
++  (__builtin_nds32_raddw ((a), (b)))
++#define __nds32__uraddw(a, b) \
++  (__builtin_nds32_uraddw ((a), (b)))
++#define __nds32__rsubw(a, b) \
++  (__builtin_nds32_rsubw ((a), (b)))
++#define __nds32__ursubw(a, b) \
++  (__builtin_nds32_ursubw ((a), (b)))
++
++#define __nds32__sra_u(a, b) \
++  (__builtin_nds32_sra_u ((a), (b)))
++#define __nds32__ksll(a, b) \
++  (__builtin_nds32_ksll ((a), (b)))
++#define __nds32__pkbb16(a, b) \
++  (__builtin_nds32_pkbb16 ((a), (b)))
++#define __nds32__v_pkbb16(a, b) \
++  (__builtin_nds32_v_pkbb16 ((a), (b)))
++#define __nds32__pkbt16(a, b) \
++  (__builtin_nds32_pkbt16 ((a), (b)))
++#define __nds32__v_pkbt16(a, b) \
++  (__builtin_nds32_v_pkbt16 ((a), (b)))
++#define __nds32__pktb16(a, b) \
++  (__builtin_nds32_pktb16 ((a), (b)))
++#define __nds32__v_pktb16(a, b) \
++  (__builtin_nds32_v_pktb16 ((a), (b)))
++#define __nds32__pktt16(a, b) \
++  (__builtin_nds32_pktt16 ((a), (b)))
++#define __nds32__v_pktt16(a, b) \
++  (__builtin_nds32_v_pktt16 ((a), (b)))
++
++#define __nds32__smmul(a, b) \
++  (__builtin_nds32_smmul ((a), (b)))
++#define __nds32__smmul_u(a, b) \
++  (__builtin_nds32_smmul_u ((a), (b)))
++#define __nds32__kmmac(r, a, b) \
++  (__builtin_nds32_kmmac ((r), (a), (b)))
++#define __nds32__kmmac_u(r, a, b) \
++  (__builtin_nds32_kmmac_u ((r), (a), (b)))
++#define __nds32__kmmsb(r, a, b) \
++  (__builtin_nds32_kmmsb ((r), (a), (b)))
++#define __nds32__kmmsb_u(r, a, b) \
++  (__builtin_nds32_kmmsb_u ((r), (a), (b)))
++#define __nds32__kwmmul(a, b) \
++  (__builtin_nds32_kwmmul ((a), (b)))
++#define __nds32__kwmmul_u(a, b) \
++  (__builtin_nds32_kwmmul_u ((a), (b)))
++
++#define __nds32__smmwb(a, b) \
++  (__builtin_nds32_smmwb ((a), (b)))
++#define __nds32__v_smmwb(a, b) \
++  (__builtin_nds32_v_smmwb ((a), (b)))
++#define __nds32__smmwb_u(a, b) \
++  (__builtin_nds32_smmwb_u ((a), (b)))
++#define __nds32__v_smmwb_u(a, b) \
++  (__builtin_nds32_v_smmwb_u ((a), (b)))
++#define __nds32__smmwt(a, b) \
++  (__builtin_nds32_smmwt ((a), (b)))
++#define __nds32__v_smmwt(a, b) \
++  (__builtin_nds32_v_smmwt ((a), (b)))
++#define __nds32__smmwt_u(a, b) \
++  (__builtin_nds32_smmwt_u ((a), (b)))
++#define __nds32__v_smmwt_u(a, b) \
++  (__builtin_nds32_v_smmwt_u ((a), (b)))
++#define __nds32__kmmawb(r, a, b) \
++  (__builtin_nds32_kmmawb ((r), (a), (b)))
++#define __nds32__v_kmmawb(r, a, b) \
++  (__builtin_nds32_v_kmmawb ((r), (a), (b)))
++#define __nds32__kmmawb_u(r, a, b) \
++  (__builtin_nds32_kmmawb_u ((r), (a), (b)))
++#define __nds32__v_kmmawb_u(r, a, b) \
++  (__builtin_nds32_v_kmmawb_u ((r), (a), (b)))
++#define __nds32__kmmawt(r, a, b) \
++  (__builtin_nds32_kmmawt ((r), (a), (b)))
++#define __nds32__v_kmmawt(r, a, b) \
++  (__builtin_nds32_v_kmmawt ((r), (a), (b)))
++#define __nds32__kmmawt_u(r, a, b) \
++  (__builtin_nds32_kmmawt_u ((r), (a), (b)))
++#define __nds32__v_kmmawt_u(r, a, b) \
++  (__builtin_nds32_v_kmmawt_u ((r), (a), (b)))
++
++#define __nds32__smbb(a, b) \
++  (__builtin_nds32_smbb ((a), (b)))
++#define __nds32__v_smbb(a, b) \
++  (__builtin_nds32_v_smbb ((a), (b)))
++#define __nds32__smbt(a, b) \
++  (__builtin_nds32_smbt ((a), (b)))
++#define __nds32__v_smbt(a, b) \
++  (__builtin_nds32_v_smbt ((a), (b)))
++#define __nds32__smtt(a, b) \
++  (__builtin_nds32_smtt ((a), (b)))
++#define __nds32__v_smtt(a, b) \
++  (__builtin_nds32_v_smtt ((a), (b)))
++#define __nds32__kmda(a, b) \
++  (__builtin_nds32_kmda ((a), (b)))
++#define __nds32__v_kmda(a, b) \
++  (__builtin_nds32_v_kmda ((a), (b)))
++#define __nds32__kmxda(a, b) \
++  (__builtin_nds32_kmxda ((a), (b)))
++#define __nds32__v_kmxda(a, b) \
++  (__builtin_nds32_v_kmxda ((a), (b)))
++#define __nds32__smds(a, b) \
++  (__builtin_nds32_smds ((a), (b)))
++#define __nds32__v_smds(a, b) \
++  (__builtin_nds32_v_smds ((a), (b)))
++#define __nds32__smdrs(a, b) \
++  (__builtin_nds32_smdrs ((a), (b)))
++#define __nds32__v_smdrs(a, b) \
++  (__builtin_nds32_v_smdrs ((a), (b)))
++#define __nds32__smxds(a, b) \
++  (__builtin_nds32_smxds ((a), (b)))
++#define __nds32__v_smxds(a, b) \
++  (__builtin_nds32_v_smxds ((a), (b)))
++#define __nds32__kmabb(r, a, b) \
++  (__builtin_nds32_kmabb ((r), (a), (b)))
++#define __nds32__v_kmabb(r, a, b) \
++  (__builtin_nds32_v_kmabb ((r), (a), (b)))
++#define __nds32__kmabt(r, a, b) \
++  (__builtin_nds32_kmabt ((r), (a), (b)))
++#define __nds32__v_kmabt(r, a, b) \
++  (__builtin_nds32_v_kmabt ((r), (a), (b)))
++#define __nds32__kmatt(r, a, b) \
++  (__builtin_nds32_kmatt ((r), (a), (b)))
++#define __nds32__v_kmatt(r, a, b) \
++  (__builtin_nds32_v_kmatt ((r), (a), (b)))
++#define __nds32__kmada(r, a, b) \
++  (__builtin_nds32_kmada ((r), (a), (b)))
++#define __nds32__v_kmada(r, a, b) \
++  (__builtin_nds32_v_kmada ((r), (a), (b)))
++#define __nds32__kmaxda(r, a, b) \
++  (__builtin_nds32_kmaxda ((r), (a), (b)))
++#define __nds32__v_kmaxda(r, a, b) \
++  (__builtin_nds32_v_kmaxda ((r), (a), (b)))
++#define __nds32__kmads(r, a, b) \
++  (__builtin_nds32_kmads ((r), (a), (b)))
++#define __nds32__v_kmads(r, a, b) \
++  (__builtin_nds32_v_kmads ((r), (a), (b)))
++#define __nds32__kmadrs(r, a, b) \
++  (__builtin_nds32_kmadrs ((r), (a), (b)))
++#define __nds32__v_kmadrs(r, a, b) \
++  (__builtin_nds32_v_kmadrs ((r), (a), (b)))
++#define __nds32__kmaxds(r, a, b) \
++  (__builtin_nds32_kmaxds ((r), (a), (b)))
++#define __nds32__v_kmaxds(r, a, b) \
++  (__builtin_nds32_v_kmaxds ((r), (a), (b)))
++#define __nds32__kmsda(r, a, b) \
++  (__builtin_nds32_kmsda ((r), (a), (b)))
++#define __nds32__v_kmsda(r, a, b) \
++  (__builtin_nds32_v_kmsda ((r), (a), (b)))
++#define __nds32__kmsxda(r, a, b) \
++  (__builtin_nds32_kmsxda ((r), (a), (b)))
++#define __nds32__v_kmsxda(r, a, b) \
++  (__builtin_nds32_v_kmsxda ((r), (a), (b)))
++
++#define __nds32__smal(a, b) \
++  (__builtin_nds32_smal ((a), (b)))
++#define __nds32__v_smal(a, b) \
++  (__builtin_nds32_v_smal ((a), (b)))
++
++#define __nds32__bitrev(a, b) \
++  (__builtin_nds32_bitrev ((a), (b)))
++#define __nds32__wext(a, b) \
++  (__builtin_nds32_wext ((a), (b)))
++#define __nds32__bpick(r, a, b) \
++  (__builtin_nds32_bpick ((r), (a), (b)))
++#define __nds32__insb(r, a, b) \
++  (__builtin_nds32_insb ((r), (a), (b)))
++
++#define __nds32__sadd64(a, b) \
++  (__builtin_nds32_sadd64 ((a), (b)))
++#define __nds32__uadd64(a, b) \
++  (__builtin_nds32_uadd64 ((a), (b)))
++#define __nds32__radd64(a, b) \
++  (__builtin_nds32_radd64 ((a), (b)))
++#define __nds32__uradd64(a, b) \
++  (__builtin_nds32_uradd64 ((a), (b)))
++#define __nds32__kadd64(a, b) \
++  (__builtin_nds32_kadd64 ((a), (b)))
++#define __nds32__ukadd64(a, b) \
++  (__builtin_nds32_ukadd64 ((a), (b)))
++#define __nds32__ssub64(a, b) \
++  (__builtin_nds32_ssub64 ((a), (b)))
++#define __nds32__usub64(a, b) \
++  (__builtin_nds32_usub64 ((a), (b)))
++#define __nds32__rsub64(a, b) \
++  (__builtin_nds32_rsub64 ((a), (b)))
++#define __nds32__ursub64(a, b) \
++  (__builtin_nds32_ursub64 ((a), (b)))
++#define __nds32__ksub64(a, b) \
++  (__builtin_nds32_ksub64 ((a), (b)))
++#define __nds32__uksub64(a, b) \
++  (__builtin_nds32_uksub64 ((a), (b)))
++
++#define __nds32__smar64(r, a, b) \
++  (__builtin_nds32_smar64 ((r), (a), (b)))
++#define __nds32__smsr64(r, a, b) \
++  (__builtin_nds32_smsr64 ((r), (a), (b)))
++#define __nds32__umar64(r, a, b) \
++  (__builtin_nds32_umar64 ((r), (a), (b)))
++#define __nds32__umsr64(r, a, b) \
++  (__builtin_nds32_umsr64 ((r), (a), (b)))
++#define __nds32__kmar64(r, a, b) \
++  (__builtin_nds32_kmar64 ((r), (a), (b)))
++#define __nds32__kmsr64(r, a, b) \
++  (__builtin_nds32_kmsr64 ((r), (a), (b)))
++#define __nds32__ukmar64(r, a, b) \
++  (__builtin_nds32_ukmar64 ((r), (a), (b)))
++#define __nds32__ukmsr64(r, a, b) \
++  (__builtin_nds32_ukmsr64 ((r), (a), (b)))
++
++#define __nds32__smalbb(r, a, b) \
++  (__builtin_nds32_smalbb ((r), (a), (b)))
++#define __nds32__v_smalbb(r, a, b) \
++  (__builtin_nds32_v_smalbb ((r), (a), (b)))
++#define __nds32__smalbt(r, a, b) \
++  (__builtin_nds32_smalbt ((r), (a), (b)))
++#define __nds32__v_smalbt(r, a, b) \
++  (__builtin_nds32_v_smalbt ((r), (a), (b)))
++#define __nds32__smaltt(r, a, b) \
++  (__builtin_nds32_smaltt ((r), (a), (b)))
++#define __nds32__v_smaltt(r, a, b) \
++  (__builtin_nds32_v_smaltt ((r), (a), (b)))
++#define __nds32__smalda(r, a, b) \
++  (__builtin_nds32_smalda ((r), (a), (b)))
++#define __nds32__v_smalda(r, a, b) \
++  (__builtin_nds32_v_smalda ((r), (a), (b)))
++#define __nds32__smalxda(r, a, b) \
++  (__builtin_nds32_smalxda ((r), (a), (b)))
++#define __nds32__v_smalxda(r, a, b) \
++  (__builtin_nds32_v_smalxda ((r), (a), (b)))
++#define __nds32__smalds(r, a, b) \
++  (__builtin_nds32_smalds ((r), (a), (b)))
++#define __nds32__v_smalds(r, a, b) \
++  (__builtin_nds32_v_smalds ((r), (a), (b)))
++#define __nds32__smaldrs(r, a, b) \
++  (__builtin_nds32_smaldrs ((r), (a), (b)))
++#define __nds32__v_smaldrs(r, a, b) \
++  (__builtin_nds32_v_smaldrs ((r), (a), (b)))
++#define __nds32__smalxds(r, a, b) \
++  (__builtin_nds32_smalxds ((r), (a), (b)))
++#define __nds32__v_smalxds(r, a, b) \
++  (__builtin_nds32_v_smalxds ((r), (a), (b)))
++#define __nds32__smslda(r, a, b) \
++  (__builtin_nds32_smslda ((r), (a), (b)))
++#define __nds32__v_smslda(r, a, b) \
++  (__builtin_nds32_v_smslda ((r), (a), (b)))
++#define __nds32__smslxda(r, a, b) \
++  (__builtin_nds32_smslxda ((r), (a), (b)))
++#define __nds32__v_smslxda(r, a, b) \
++  (__builtin_nds32_v_smslxda ((r), (a), (b)))
++
++#define __nds32__smul16(a, b) \
++  (__builtin_nds32_smul16 ((a), (b)))
++#define __nds32__v_smul16(a, b) \
++  (__builtin_nds32_v_smul16 ((a), (b)))
++#define __nds32__smulx16(a, b) \
++  (__builtin_nds32_smulx16 ((a), (b)))
++#define __nds32__v_smulx16(a, b) \
++  (__builtin_nds32_v_smulx16 ((a), (b)))
++#define __nds32__umul16(a, b) \
++  (__builtin_nds32_umul16 ((a), (b)))
++#define __nds32__v_umul16(a, b) \
++  (__builtin_nds32_v_umul16 ((a), (b)))
++#define __nds32__umulx16(a, b) \
++  (__builtin_nds32_umulx16 ((a), (b)))
++#define __nds32__v_umulx16(a, b) \
++  (__builtin_nds32_v_umulx16 ((a), (b)))
++
++#define __nds32__uclip32(a, imm) \
++  (__builtin_nds32_uclip32 ((a), (imm)))
++#define __nds32__sclip32(a, imm) \
++  (__builtin_nds32_sclip32 ((a), (imm)))
++#define __nds32__kabs(a) \
++  (__builtin_nds32_kabs ((a)))
++
+ #define __nds32__unaligned_feature() \
+   (__builtin_nds32_unaligned_feature())
+ #define __nds32__enable_unaligned() \
+   (__builtin_nds32_enable_unaligned())
+ #define __nds32__disable_unaligned() \
+   (__builtin_nds32_disable_unaligned())
++
++#define __nds32__get_unaligned_u16x2(a) \
++  (__builtin_nds32_get_unaligned_u16x2 ((a)))
++#define __nds32__get_unaligned_s16x2(a) \
++  (__builtin_nds32_get_unaligned_s16x2 ((a)))
++#define __nds32__get_unaligned_u8x4(a) \
++  (__builtin_nds32_get_unaligned_u8x4 ((a)))
++#define __nds32__get_unaligned_s8x4(a) \
++  (__builtin_nds32_get_unaligned_s8x4 ((a)))
++
++#define __nds32__put_unaligned_u16x2(a, data) \
++  (__builtin_nds32_put_unaligned_u16x2 ((a), (data)))
++#define __nds32__put_unaligned_s16x2(a, data) \
++  (__builtin_nds32_put_unaligned_s16x2 ((a), (data)))
++#define __nds32__put_unaligned_u8x4(a, data) \
++  (__builtin_nds32_put_unaligned_u8x4 ((a), (data)))
++#define __nds32__put_unaligned_s8x4(a, data) \
++  (__builtin_nds32_put_unaligned_s8x4 ((a), (data)))
++
++#define NDS32ATTR_SIGNATURE              __attribute__((signature))
++
+ #endif /* nds32_intrinsic.h */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-intrinsic.md gcc-8.2.0/gcc/config/nds32/nds32-intrinsic.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-intrinsic.md	2018-04-22 09:46:39.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-intrinsic.md	2019-01-25 15:38:32.825242648 +0100
+@@ -1037,6 +1037,187 @@
+    (set_attr "length" "4")]
+ )
+ 
++;; SATURATION
++
++(define_insn "unspec_kaddw"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ss_plus:SI (match_operand:SI 1 "register_operand" "r")
++		    (match_operand:SI 2 "register_operand" "r")))]
++  ""
++  "kaddw\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_ksubw"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(ss_minus:SI (match_operand:SI 1 "register_operand" "r")
++		     (match_operand:SI 2 "register_operand" "r")))]
++  ""
++  "ksubw\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kaddh"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "r")
++		    (match_operand:SI 2 "register_operand" "r")] UNSPEC_KADDH))]
++  ""
++  "kaddh\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_ksubh"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "r")
++		    (match_operand:SI 2 "register_operand" "r")] UNSPEC_KSUBH))]
++  ""
++  "ksubh\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kaddh_dsp"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(plus:SI (match_operand:SI 1 "register_operand" "r")
++			     (match_operand:SI 2 "register_operand" "r"))
++		    (const_int 15)] UNSPEC_CLIPS))]
++  "NDS32_EXT_DSP_P ()"
++  "kaddh\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_ksubh_dsp"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(minus:SI (match_operand:SI 1 "register_operand" "r")
++			      (match_operand:SI 2 "register_operand" "r"))
++		    (const_int 15)] UNSPEC_CLIPS))]
++  "NDS32_EXT_DSP_P ()"
++  "ksubh\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kdmbb"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KDMBB))]
++  ""
++  "kdmbb\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kdmbt"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KDMBT))]
++  ""
++  "kdmbt\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kdmtb"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KDMTB))]
++  ""
++  "kdmtb\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kdmtt"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KDMTT))]
++  ""
++  "kdmtt\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_khmbb"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KHMBB))]
++  ""
++  "khmbb\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_khmbt"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KHMBT))]
++  ""
++  "khmbt\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_khmtb"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KHMTB))]
++  ""
++  "khmtb\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_khmtt"
++  [(set (match_operand:V2HI 0 "register_operand" "=r")
++	(unspec:V2HI [(match_operand:V2HI 1 "register_operand" "r")
++		      (match_operand:V2HI 2 "register_operand" "r")] UNSPEC_KHMTT))]
++  ""
++  "khmtt\t%0, %1, %2"
++  [(set_attr "type"    "mul")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kslraw"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "r")
++		    (match_operand:SI 2 "register_operand" "r")] UNSPEC_KSLRAW))]
++  ""
++  "kslraw\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_kslrawu"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "r")
++		    (match_operand:SI 2 "register_operand" "r")] UNSPEC_KSLRAWU))]
++  ""
++  "kslraw.u\t%0, %1, %2"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_volatile_rdov"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec_volatile:SI [(const_int 0)] UNSPEC_VOLATILE_RDOV))]
++  ""
++  "rdov\t%0"
++  [(set_attr "type"   "misc")
++   (set_attr "length"    "4")]
++)
++
++(define_insn "unspec_volatile_clrov"
++  [(unspec_volatile:SI [(const_int 0)] UNSPEC_VOLATILE_CLROV)]
++  ""
++  "clrov"
++  [(set_attr "type"   "misc")
++   (set_attr "length"    "4")]
++)
++
+ ;; System
+ 
+ (define_insn "unspec_sva"
+@@ -1415,22 +1596,17 @@
+   if (TARGET_ISA_V3M)
+     nds32_expand_unaligned_store (operands, DImode);
+   else
+-    emit_insn (gen_unaligned_store_dw (operands[0], operands[1]));
++    emit_insn (gen_unaligned_store_dw (gen_rtx_MEM (DImode, operands[0]),
++				       operands[1]));
+   DONE;
+ })
+ 
+ (define_insn "unaligned_store_dw"
+-  [(set (mem:DI (match_operand:SI 0 "register_operand" "r"))
+-	(unspec:DI [(match_operand:DI 1 "register_operand" "r")] UNSPEC_UASTORE_DW))]
++  [(set (match_operand:DI 0 "nds32_lmw_smw_base_operand"   "=Umw")
++	(unspec:DI [(match_operand:DI 1 "register_operand" "   r")] UNSPEC_UASTORE_DW))]
+   ""
+ {
+-  rtx otherops[3];
+-  otherops[0] = gen_rtx_REG (SImode, REGNO (operands[1]));
+-  otherops[1] = gen_rtx_REG (SImode, REGNO (operands[1]) + 1);
+-  otherops[2] = operands[0];
+-
+-  output_asm_insn ("smw.bi\t%0, [%2], %1, 0", otherops);
+-  return "";
++  return nds32_output_smw_double_word (operands);
+ }
+   [(set_attr "type"   "store")
+    (set_attr "length"     "4")]
+@@ -1495,4 +1671,15 @@
+   DONE;
+ })
+ 
++;; abs alias kabs
++
++(define_insn "unspec_kabs"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "r")] UNSPEC_KABS))]
++  ""
++  "kabs\t%0, %1"
++  [(set_attr "type" "alu")
++   (set_attr "length" "4")]
++)
++
+ ;; ------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-isr.c gcc-8.2.0/gcc/config/nds32/nds32-isr.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-isr.c	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-isr.c	2019-01-25 15:38:32.829242659 +0100
+@@ -43,7 +43,260 @@
+    We use an array to record essential information for each vector.  */
+ static struct nds32_isr_info nds32_isr_vectors[NDS32_N_ISR_VECTORS];
+ 
+-/* ------------------------------------------------------------------------ */
++/* ------------------------------------------------------------- */
++/* FIXME:
++   FOR BACKWARD COMPATIBILITY, we need to support following patterns:
++
++       __attribute__((interrupt("XXX;YYY;id=ZZZ")))
++       __attribute__((exception("XXX;YYY;id=ZZZ")))
++       __attribute__((reset("vectors=XXX;nmi_func=YYY;warm_func=ZZZ")))
++
++   We provide several functions to parse the strings.  */
++
++static void
++nds32_interrupt_attribute_parse_string (const char *original_str,
++					const char *func_name,
++					unsigned int s_level)
++{
++  char target_str[100];
++  enum nds32_isr_save_reg save_reg;
++  enum nds32_isr_nested_type nested_type;
++
++  char *save_all_regs_str, *save_caller_regs_str;
++  char *nested_str, *not_nested_str, *ready_nested_str, *critical_str;
++  char *id_str, *value_str;
++
++  /* Copy original string into a character array so that
++     the string APIs can handle it.  */
++  strcpy (target_str, original_str);
++
++  /* 1. Detect 'save_all_regs'    : NDS32_SAVE_ALL
++	       'save_caller_regs' : NDS32_PARTIAL_SAVE */
++  save_all_regs_str    = strstr (target_str, "save_all_regs");
++  save_caller_regs_str = strstr (target_str, "save_caller_regs");
++
++  /* Note that if no argument is found,
++     use NDS32_PARTIAL_SAVE by default.  */
++  if (save_all_regs_str)
++    save_reg = NDS32_SAVE_ALL;
++  else if (save_caller_regs_str)
++    save_reg = NDS32_PARTIAL_SAVE;
++  else
++    save_reg = NDS32_PARTIAL_SAVE;
++
++  /* 2. Detect 'nested'       : NDS32_NESTED
++	       'not_nested'   : NDS32_NOT_NESTED
++	       'ready_nested' : NDS32_NESTED_READY
++	       'critical'     : NDS32_CRITICAL */
++  nested_str       = strstr (target_str, "nested");
++  not_nested_str   = strstr (target_str, "not_nested");
++  ready_nested_str = strstr (target_str, "ready_nested");
++  critical_str     = strstr (target_str, "critical");
++
++  /* Note that if no argument is found,
++     use NDS32_NOT_NESTED by default.
++     Also, since 'not_nested' and 'ready_nested' both contains
++     'nested' string, we check 'nested' with lowest priority.  */
++  if (not_nested_str)
++    nested_type = NDS32_NOT_NESTED;
++  else if (ready_nested_str)
++    nested_type = NDS32_NESTED_READY;
++  else if (nested_str)
++    nested_type = NDS32_NESTED;
++  else if (critical_str)
++    nested_type = NDS32_CRITICAL;
++  else
++    nested_type = NDS32_NOT_NESTED;
++
++  /* 3. Traverse each id value and set corresponding information.  */
++  id_str = strstr (target_str, "id=");
++
++  /* If user forgets to assign 'id', issue an error message.  */
++  if (id_str == NULL)
++    error ("require id argument in the string");
++  /* Extract the value_str first.  */
++  id_str    = strtok (id_str, "=");
++  value_str = strtok (NULL, ";");
++
++  /* Pick up the first id value token.  */
++  value_str = strtok (value_str, ",");
++  while (value_str != NULL)
++    {
++      int i;
++      i = atoi (value_str);
++
++      /* For interrupt(0..63), the actual vector number is (9..72).  */
++      i = i + 9;
++      if (i < 9 || i > 72)
++	error ("invalid id value for interrupt attribute");
++
++      /* Setup nds32_isr_vectors[] array.  */
++      nds32_isr_vectors[i].category = NDS32_ISR_INTERRUPT;
++      strcpy (nds32_isr_vectors[i].func_name, func_name);
++      nds32_isr_vectors[i].save_reg = save_reg;
++      nds32_isr_vectors[i].nested_type = nested_type;
++      nds32_isr_vectors[i].security_level = s_level;
++
++      /* Fetch next token.  */
++      value_str = strtok (NULL, ",");
++    }
++
++  return;
++}
++
++static void
++nds32_exception_attribute_parse_string (const char *original_str,
++					const char *func_name,
++					unsigned int s_level)
++{
++  char target_str[100];
++  enum nds32_isr_save_reg save_reg;
++  enum nds32_isr_nested_type nested_type;
++
++  char *save_all_regs_str, *save_caller_regs_str;
++  char *nested_str, *not_nested_str, *ready_nested_str, *critical_str;
++  char *id_str, *value_str;
++
++  /* Copy original string into a character array so that
++     the string APIs can handle it.  */
++  strcpy (target_str, original_str);
++
++  /* 1. Detect 'save_all_regs'    : NDS32_SAVE_ALL
++	       'save_caller_regs' : NDS32_PARTIAL_SAVE */
++  save_all_regs_str    = strstr (target_str, "save_all_regs");
++  save_caller_regs_str = strstr (target_str, "save_caller_regs");
++
++  /* Note that if no argument is found,
++     use NDS32_PARTIAL_SAVE by default.  */
++  if (save_all_regs_str)
++    save_reg = NDS32_SAVE_ALL;
++  else if (save_caller_regs_str)
++    save_reg = NDS32_PARTIAL_SAVE;
++  else
++    save_reg = NDS32_PARTIAL_SAVE;
++
++  /* 2. Detect 'nested'       : NDS32_NESTED
++	       'not_nested'   : NDS32_NOT_NESTED
++	       'ready_nested' : NDS32_NESTED_READY
++	       'critical'     : NDS32_CRITICAL */
++  nested_str       = strstr (target_str, "nested");
++  not_nested_str   = strstr (target_str, "not_nested");
++  ready_nested_str = strstr (target_str, "ready_nested");
++  critical_str     = strstr (target_str, "critical");
++
++  /* Note that if no argument is found,
++     use NDS32_NOT_NESTED by default.
++     Also, since 'not_nested' and 'ready_nested' both contains
++     'nested' string, we check 'nested' with lowest priority.  */
++  if (not_nested_str)
++    nested_type = NDS32_NOT_NESTED;
++  else if (ready_nested_str)
++    nested_type = NDS32_NESTED_READY;
++  else if (nested_str)
++    nested_type = NDS32_NESTED;
++  else if (critical_str)
++    nested_type = NDS32_CRITICAL;
++  else
++    nested_type = NDS32_NOT_NESTED;
++
++  /* 3. Traverse each id value and set corresponding information.  */
++  id_str = strstr (target_str, "id=");
++
++  /* If user forgets to assign 'id', issue an error message.  */
++  if (id_str == NULL)
++    error ("require id argument in the string");
++  /* Extract the value_str first.  */
++  id_str    = strtok (id_str, "=");
++  value_str = strtok (NULL, ";");
++
++  /* Pick up the first id value token.  */
++  value_str = strtok (value_str, ",");
++  while (value_str != NULL)
++    {
++      int i;
++      i = atoi (value_str);
++
++      /* For exception(1..8), the actual vector number is (1..8).  */
++      if (i < 1 || i > 8)
++	error ("invalid id value for exception attribute");
++
++      /* Setup nds32_isr_vectors[] array.  */
++      nds32_isr_vectors[i].category = NDS32_ISR_EXCEPTION;
++      strcpy (nds32_isr_vectors[i].func_name, func_name);
++      nds32_isr_vectors[i].save_reg = save_reg;
++      nds32_isr_vectors[i].nested_type = nested_type;
++      nds32_isr_vectors[i].security_level = s_level;
++
++      /* Fetch next token.  */
++      value_str = strtok (NULL, ",");
++    }
++
++  return;
++}
++
++static void
++nds32_reset_attribute_parse_string (const char *original_str,
++				    const char *func_name)
++{
++  char target_str[100];
++  char *vectors_str, *nmi_str, *warm_str, *value_str;
++
++  /* Deal with reset attribute.  Its vector number is always 0.  */
++  nds32_isr_vectors[0].category = NDS32_ISR_RESET;
++
++
++  /* 1. Parse 'vectors=XXXX'.  */
++
++  /* Copy original string into a character array so that
++     the string APIs can handle it.  */
++  strcpy (target_str, original_str);
++  vectors_str = strstr (target_str, "vectors=");
++  /* The total vectors = interrupt + exception numbers + reset.
++     There are 8 exception and 1 reset in nds32 architecture.
++     If user forgets to assign 'vectors', user default 16 interrupts.  */
++  if (vectors_str != NULL)
++    {
++      /* Extract the value_str.  */
++      vectors_str = strtok (vectors_str, "=");
++      value_str  = strtok (NULL, ";");
++      nds32_isr_vectors[0].total_n_vectors = atoi (value_str) + 8 + 1;
++    }
++  else
++    nds32_isr_vectors[0].total_n_vectors = 16 + 8 + 1;
++  strcpy (nds32_isr_vectors[0].func_name, func_name);
++
++
++  /* 2. Parse 'nmi_func=YYYY'.  */
++
++  /* Copy original string into a character array so that
++     the string APIs can handle it.  */
++  strcpy (target_str, original_str);
++  nmi_str = strstr (target_str, "nmi_func=");
++  if (nmi_str != NULL)
++    {
++      /* Extract the value_str.  */
++      nmi_str = strtok (nmi_str, "=");
++      value_str  = strtok (NULL, ";");
++      strcpy (nds32_isr_vectors[0].nmi_name, value_str);
++    }
++
++  /* 3. Parse 'warm_func=ZZZZ'.  */
++
++  /* Copy original string into a character array so that
++     the string APIs can handle it.  */
++  strcpy (target_str, original_str);
++  warm_str = strstr (target_str, "warm_func=");
++  if (warm_str != NULL)
++    {
++      /* Extract the value_str.  */
++      warm_str = strtok (warm_str, "=");
++      value_str  = strtok (NULL, ";");
++      strcpy (nds32_isr_vectors[0].warm_name, value_str);
++    }
++
++  return;
++}
++/* ------------------------------------------------------------- */
+ 
+ /* A helper function to emit section head template.  */
+ static void
+@@ -79,6 +332,15 @@
+   char section_name[100];
+   char symbol_name[100];
+ 
++  /* A critical isr does not need jump table section because
++     its behavior is not performed by two-level handler.  */
++  if (nds32_isr_vectors[vector_id].nested_type == NDS32_CRITICAL)
++    {
++      fprintf (asm_out_file, "\t! The vector %02d is a critical isr !\n",
++			     vector_id);
++      return;
++    }
++
+   /* Prepare jmptbl section and symbol name.  */
+   snprintf (section_name, sizeof (section_name),
+ 	    ".nds32_jmptbl.%02d", vector_id);
+@@ -99,7 +361,6 @@
+   const char *c_str = "CATEGORY";
+   const char *sr_str = "SR";
+   const char *nt_str = "NT";
+-  const char *vs_str = "VS";
+   char first_level_handler_name[100];
+   char section_name[100];
+   char symbol_name[100];
+@@ -147,30 +408,47 @@
+     case NDS32_NESTED_READY:
+       nt_str = "nr";
+       break;
++    case NDS32_CRITICAL:
++      /* The critical isr is not performed by two-level handler.  */
++      nt_str = "";
++      break;
+     }
+ 
+-  /* Currently we have 4-byte or 16-byte size for each vector.
+-     If it is 4-byte, the first level handler name has suffix string "_4b".  */
+-  vs_str = (nds32_isr_vector_size == 4) ? "_4b" : "";
+-
+   /* Now we can create first level handler name.  */
+-  snprintf (first_level_handler_name, sizeof (first_level_handler_name),
+-	    "_nds32_%s_%s_%s%s", c_str, sr_str, nt_str, vs_str);
++  if (nds32_isr_vectors[vector_id].security_level == 0)
++    {
++      /* For security level 0, use normal first level handler name.  */
++      snprintf (first_level_handler_name, sizeof (first_level_handler_name),
++		"_nds32_%s_%s_%s", c_str, sr_str, nt_str);
++    }
++  else
++    {
++      /* For security level 1-3, use corresponding spl_1, spl_2, or spl_3.  */
++      snprintf (first_level_handler_name, sizeof (first_level_handler_name),
++		"_nds32_spl_%d", nds32_isr_vectors[vector_id].security_level);
++    }
+ 
+   /* Prepare vector section and symbol name.  */
+   snprintf (section_name, sizeof (section_name),
+ 	    ".nds32_vector.%02d", vector_id);
+   snprintf (symbol_name, sizeof (symbol_name),
+-	    "_nds32_vector_%02d%s", vector_id, vs_str);
++	    "_nds32_vector_%02d", vector_id);
+ 
+ 
+   /* Everything is ready.  We can start emit vector section content.  */
+   nds32_emit_section_head_template (section_name, symbol_name,
+ 				    floor_log2 (nds32_isr_vector_size), false);
+ 
+-  /* According to the vector size, the instructions in the
+-     vector section may be different.  */
+-  if (nds32_isr_vector_size == 4)
++  /* First we check if it is a critical isr.
++     If so, jump to user handler directly; otherwise, the instructions
++     in the vector section may be different according to the vector size.  */
++  if (nds32_isr_vectors[vector_id].nested_type == NDS32_CRITICAL)
++    {
++      /* This block is for critical isr.  Jump to user handler directly.  */
++      fprintf (asm_out_file, "\tj\t%s ! jump to user handler directly\n",
++			     nds32_isr_vectors[vector_id].func_name);
++    }
++  else if (nds32_isr_vector_size == 4)
+     {
+       /* This block is for 4-byte vector size.
+ 	 Hardware $VID support is necessary and only one instruction
+@@ -239,13 +517,11 @@
+ {
+   unsigned int i;
+   unsigned int total_n_vectors;
+-  const char *vs_str;
+   char reset_handler_name[100];
+   char section_name[100];
+   char symbol_name[100];
+ 
+   total_n_vectors = nds32_isr_vectors[0].total_n_vectors;
+-  vs_str = (nds32_isr_vector_size == 4) ? "_4b" : "";
+ 
+   fprintf (asm_out_file, "\t! RESET HANDLER CONTENT - BEGIN !\n");
+ 
+@@ -261,7 +537,7 @@
+   /* Emit vector references.  */
+   fprintf (asm_out_file, "\t ! references to vector section entries\n");
+   for (i = 0; i < total_n_vectors; i++)
+-    fprintf (asm_out_file, "\t.word\t_nds32_vector_%02d%s\n", i, vs_str);
++    fprintf (asm_out_file, "\t.word\t_nds32_vector_%02d\n", i);
+ 
+   /* Emit jmptbl_00 section.  */
+   snprintf (section_name, sizeof (section_name), ".nds32_jmptbl.00");
+@@ -275,9 +551,9 @@
+ 
+   /* Emit vector_00 section.  */
+   snprintf (section_name, sizeof (section_name), ".nds32_vector.00");
+-  snprintf (symbol_name, sizeof (symbol_name), "_nds32_vector_00%s", vs_str);
++  snprintf (symbol_name, sizeof (symbol_name), "_nds32_vector_00");
+   snprintf (reset_handler_name, sizeof (reset_handler_name),
+-	    "_nds32_reset%s", vs_str);
++	    "_nds32_reset");
+ 
+   fprintf (asm_out_file, "\t! ....................................\n");
+   nds32_emit_section_head_template (section_name, symbol_name,
+@@ -323,12 +599,12 @@
+ nds32_check_isr_attrs_conflict (tree func_decl, tree func_attrs)
+ {
+   int save_all_p, partial_save_p;
+-  int nested_p, not_nested_p, nested_ready_p;
++  int nested_p, not_nested_p, nested_ready_p, critical_p;
+   int intr_p, excp_p, reset_p;
+ 
+   /* Initialize variables.  */
+   save_all_p = partial_save_p = 0;
+-  nested_p = not_nested_p = nested_ready_p = 0;
++  nested_p = not_nested_p = nested_ready_p = critical_p = 0;
+   intr_p = excp_p = reset_p = 0;
+ 
+   /* We must check at MOST one attribute to set save-reg.  */
+@@ -347,8 +623,10 @@
+     not_nested_p = 1;
+   if (lookup_attribute ("nested_ready", func_attrs))
+     nested_ready_p = 1;
++  if (lookup_attribute ("critical", func_attrs))
++    critical_p = 1;
+ 
+-  if ((nested_p + not_nested_p + nested_ready_p) > 1)
++  if ((nested_p + not_nested_p + nested_ready_p + critical_p) > 1)
+     error ("multiple nested types attributes to function %qD", func_decl);
+ 
+   /* We must check at MOST one attribute to
+@@ -362,6 +640,17 @@
+ 
+   if ((intr_p + excp_p + reset_p) > 1)
+     error ("multiple interrupt attributes to function %qD", func_decl);
++
++  /* Do not allow isr attributes under linux toolchain.  */
++  if (TARGET_LINUX_ABI && intr_p)
++      error ("cannot use interrupt attributes to function %qD "
++	     "under linux toolchain", func_decl);
++  if (TARGET_LINUX_ABI && excp_p)
++      error ("cannot use exception attributes to function %qD "
++	     "under linux toolchain", func_decl);
++  if (TARGET_LINUX_ABI && reset_p)
++      error ("cannot use reset attributes to function %qD "
++	     "under linux toolchain", func_decl);
+ }
+ 
+ /* Function to construct isr vectors information array.
+@@ -373,15 +662,21 @@
+ 					 const char *func_name)
+ {
+   tree save_all, partial_save;
+-  tree nested, not_nested, nested_ready;
++  tree nested, not_nested, nested_ready, critical;
+   tree intr, excp, reset;
+ 
++  tree secure;
++  tree security_level_list;
++  tree security_level;
++  unsigned int s_level;
++
+   save_all     = lookup_attribute ("save_all", func_attrs);
+   partial_save = lookup_attribute ("partial_save", func_attrs);
+ 
+   nested       = lookup_attribute ("nested", func_attrs);
+   not_nested   = lookup_attribute ("not_nested", func_attrs);
+   nested_ready = lookup_attribute ("nested_ready", func_attrs);
++  critical     = lookup_attribute ("critical", func_attrs);
+ 
+   intr  = lookup_attribute ("interrupt", func_attrs);
+   excp  = lookup_attribute ("exception", func_attrs);
+@@ -391,6 +686,63 @@
+   if (!intr && !excp && !reset)
+     return;
+ 
++  /* At first, we need to retrieve security level.  */
++  secure = lookup_attribute ("secure", func_attrs);
++  if (secure != NULL)
++    {
++      security_level_list = TREE_VALUE (secure);
++      security_level = TREE_VALUE (security_level_list);
++      s_level = TREE_INT_CST_LOW (security_level);
++    }
++  else
++    {
++      /* If there is no secure attribute, the security level is set by
++	 nds32_isr_secure_level, which is controlled by -misr-secure=X option.
++	 By default nds32_isr_secure_level should be 0.  */
++      s_level = nds32_isr_secure_level;
++    }
++
++  /* ------------------------------------------------------------- */
++  /* FIXME:
++     FOR BACKWARD COMPATIBILITY, we need to support following patterns:
++
++	 __attribute__((interrupt("XXX;YYY;id=ZZZ")))
++	 __attribute__((exception("XXX;YYY;id=ZZZ")))
++	 __attribute__((reset("vectors=XXX;nmi_func=YYY;warm_func=ZZZ")))
++
++     If interrupt/exception/reset appears and its argument is a
++     STRING_CST, we will parse string with some auxiliary functions
++     which set necessary isr information in the nds32_isr_vectors[] array.
++     After that, we can return immediately to avoid new-syntax isr
++     information construction.  */
++  if (intr != NULL_TREE
++      && TREE_CODE (TREE_VALUE (TREE_VALUE (intr))) == STRING_CST)
++    {
++      tree string_arg = TREE_VALUE (TREE_VALUE (intr));
++      nds32_interrupt_attribute_parse_string (TREE_STRING_POINTER (string_arg),
++					      func_name,
++					      s_level);
++      return;
++    }
++  if (excp != NULL_TREE
++      && TREE_CODE (TREE_VALUE (TREE_VALUE (excp))) == STRING_CST)
++    {
++      tree string_arg = TREE_VALUE (TREE_VALUE (excp));
++      nds32_exception_attribute_parse_string (TREE_STRING_POINTER (string_arg),
++					      func_name,
++					      s_level);
++      return;
++    }
++  if (reset != NULL_TREE
++      && TREE_CODE (TREE_VALUE (TREE_VALUE (reset))) == STRING_CST)
++    {
++      tree string_arg = TREE_VALUE (TREE_VALUE (reset));
++      nds32_reset_attribute_parse_string (TREE_STRING_POINTER (string_arg),
++					  func_name);
++      return;
++    }
++  /* ------------------------------------------------------------- */
++
+   /* If we are here, either we have interrupt/exception,
+      or reset attribute.  */
+   if (intr || excp)
+@@ -417,6 +769,9 @@
+ 	  /* Add vector_number_offset to get actual vector number.  */
+ 	  vector_id = TREE_INT_CST_LOW (id) + vector_number_offset;
+ 
++	  /* Set security level.  */
++	  nds32_isr_vectors[vector_id].security_level = s_level;
++
+ 	  /* Enable corresponding vector and set function name.  */
+ 	  nds32_isr_vectors[vector_id].category = (intr)
+ 						  ? (NDS32_ISR_INTERRUPT)
+@@ -436,6 +791,8 @@
+ 	    nds32_isr_vectors[vector_id].nested_type = NDS32_NOT_NESTED;
+ 	  else if (nested_ready)
+ 	    nds32_isr_vectors[vector_id].nested_type = NDS32_NESTED_READY;
++	  else if (critical)
++	    nds32_isr_vectors[vector_id].nested_type = NDS32_CRITICAL;
+ 
+ 	  /* Advance to next id.  */
+ 	  id_list = TREE_CHAIN (id_list);
+@@ -492,7 +849,6 @@
+     }
+ }
+ 
+-/* A helper function to handle isr stuff at the beginning of asm file.  */
+ void
+ nds32_asm_file_start_for_isr (void)
+ {
+@@ -505,15 +861,14 @@
+       strcpy (nds32_isr_vectors[i].func_name, "");
+       nds32_isr_vectors[i].save_reg = NDS32_PARTIAL_SAVE;
+       nds32_isr_vectors[i].nested_type = NDS32_NOT_NESTED;
++      nds32_isr_vectors[i].security_level = 0;
+       nds32_isr_vectors[i].total_n_vectors = 0;
+       strcpy (nds32_isr_vectors[i].nmi_name, "");
+       strcpy (nds32_isr_vectors[i].warm_name, "");
+     }
+ }
+ 
+-/* A helper function to handle isr stuff at the end of asm file.  */
+-void
+-nds32_asm_file_end_for_isr (void)
++void nds32_asm_file_end_for_isr (void)
+ {
+   int i;
+ 
+@@ -547,6 +902,8 @@
+ 	  /* Found one vector which is interupt or exception.
+ 	     Output its jmptbl and vector section content.  */
+ 	  fprintf (asm_out_file, "\t! interrupt/exception vector %02d\n", i);
++	  fprintf (asm_out_file, "\t! security level: %d\n",
++		   nds32_isr_vectors[i].security_level);
+ 	  fprintf (asm_out_file, "\t! ------------------------------------\n");
+ 	  nds32_emit_isr_jmptbl_section (i);
+ 	  fprintf (asm_out_file, "\t! ....................................\n");
+@@ -580,4 +937,65 @@
+ 	  || (t_reset != NULL_TREE));
+ }
+ 
+-/* ------------------------------------------------------------------------ */
++/* Return true if FUNC is a isr function with critical attribute.  */
++bool
++nds32_isr_function_critical_p (tree func)
++{
++  tree t_intr;
++  tree t_excp;
++  tree t_critical;
++
++  tree attrs;
++
++  if (TREE_CODE (func) != FUNCTION_DECL)
++    abort ();
++
++  attrs = DECL_ATTRIBUTES (func);
++
++  t_intr  = lookup_attribute ("interrupt", attrs);
++  t_excp  = lookup_attribute ("exception", attrs);
++
++  t_critical = lookup_attribute ("critical", attrs);
++
++  /* If both interrupt and exception attribute does not appear,
++     we can return false immediately.  */
++  if ((t_intr == NULL_TREE) && (t_excp == NULL_TREE))
++    return false;
++
++  /* Here we can guarantee either interrupt or ecxception attribute
++     does exist, so further check critical attribute.
++     If it also appears, we can return true.  */
++  if (t_critical != NULL_TREE)
++    return true;
++
++  /* ------------------------------------------------------------- */
++  /* FIXME:
++     FOR BACKWARD COMPATIBILITY, we need to handle string type.
++     If the string 'critical' appears in the interrupt/exception
++     string argument, we can return true.  */
++  if (t_intr != NULL_TREE || t_excp != NULL_TREE)
++    {
++      char target_str[100];
++      char *critical_str;
++      tree t_check;
++      tree string_arg;
++
++      t_check = t_intr ? t_intr : t_excp;
++      if (TREE_CODE (TREE_VALUE (TREE_VALUE (t_check))) == STRING_CST)
++	{
++	  string_arg = TREE_VALUE (TREE_VALUE (t_check));
++	  strcpy (target_str, TREE_STRING_POINTER (string_arg));
++	  critical_str = strstr (target_str, "critical");
++
++	  /* Found 'critical' string, so return true.  */
++	  if (critical_str)
++	    return true;
++	}
++    }
++  /* ------------------------------------------------------------- */
++
++  /* Other cases, this isr function is not critical type.  */
++  return false;
++}
++
++/* ------------------------------------------------------------- */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32_isr.h gcc-8.2.0/gcc/config/nds32/nds32_isr.h
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32_isr.h	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32_isr.h	2019-01-25 15:38:32.833242671 +0100
+@@ -0,0 +1,526 @@
++/* Intrinsic definitions of Andes NDS32 cpu for GNU compiler
++   Copyright (C) 2012-2018 Free Software Foundation, Inc.
++   Contributed by Andes Technology Corporation.
++
++   This file is part of GCC.
++
++   GCC is free software; you can redistribute it and/or modify it
++   under the terms of the GNU General Public License as published
++   by the Free Software Foundation; either version 3, or (at your
++   option) any later version.
++
++   GCC is distributed in the hope that it will be useful, but WITHOUT
++   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++   License for more details.
++
++   Under Section 7 of GPL version 3, you are granted additional
++   permissions described in the GCC Runtime Library Exception, version
++   3.1, as published by the Free Software Foundation.
++
++   You should have received a copy of the GNU General Public License and
++   a copy of the GCC Runtime Library Exception along with this program;
++   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
++   <http://www.gnu.org/licenses/>.  */
++
++#ifndef _NDS32_ISR_H
++#define _NDS32_ISR_H
++
++/* Attribute of a interrupt or exception handler:
++
++   NDS32_READY_NESTED: This handler is interruptible if user re-enable GIE bit.
++   NDS32_NESTED      : This handler is interruptible.  This is not suitable
++                       exception handler.
++   NDS32_NOT_NESTED  : This handler is NOT interruptible.  Users have to do
++                       some work if nested is wanted
++   NDS32_CRITICAL    : This handler is critical ISR, which means it is small
++                       and efficient.  */
++#define NDS32_READY_NESTED   0
++#define NDS32_NESTED         1
++#define NDS32_NOT_NESTED     2
++#define NDS32_CRITICAL       3
++
++/* Attribute of a interrupt or exception handler:
++
++   NDS32_SAVE_ALL_REGS    : Save all registers in a table.
++   NDS32_SAVE_PARTIAL_REGS: Save partial registers.  */
++#define NDS32_SAVE_CALLER_REGS   0
++#define NDS32_SAVE_ALL_REGS      1
++
++/* There are two version of Register table for interrupt and exception handler,
++   one for 16-register CPU the other for 32-register CPU.  These structures are
++   used for context switching or system call handling.  The address of this
++   data can be get from the input argument of the handler functions.
++
++   For system call handling, r0 to r5 are used to pass arguments.  If more
++   arguments are used they are put into the stack and its starting address is
++   in sp.  Return value of system call can be put into r0 and r1 upon exit from
++   system call handler.  System call ID is in a system register and it can be
++   fetched via intrinsic function.  For more information please read ABI and
++   other related documents.
++
++   For context switching, at least 2 values need to saved in kernel.  One is
++   IPC and the other is the stack address of current task.  Use intrinsic
++   function to get IPC and  the input argument of the handler functions + 8 to
++   get stack address of current task.  To do context switching, you replace
++   new_sp with the stack address of new task and replace IPC system register
++   with IPC of new task, then, just return from handler.  The context switching
++   will happen.  */
++
++/* Register table for exception handler; 32-register version.  */
++typedef struct
++{
++  int r0;
++  int r1;
++  int r2;
++  int r3;
++  int r4;
++  int r5;
++  int r6;
++  int r7;
++  int r8;
++  int r9;
++  int r10;
++  int r11;
++  int r12;
++  int r13;
++  int r14;
++  int r15;
++  int r16;
++  int r17;
++  int r18;
++  int r19;
++  int r20;
++  int r21;
++  int r22;
++  int r23;
++  int r24;
++  int r25;
++  int r26;
++  int r27;
++  int fp;
++  int gp;
++  int lp;
++  int sp;
++} NDS32_GPR32;
++
++/* Register table for exception handler; 16-register version.  */
++typedef struct
++{
++  int r0;
++  int r1;
++  int r2;
++  int r3;
++  int r4;
++  int r5;
++  int r6;
++  int r7;
++  int r8;
++  int r9;
++  int r10;
++  int r15;
++  int fp;
++  int gp;
++  int lp;
++  int sp;
++} NDS32_GPR16;
++
++
++/* Use NDS32_REG32_TAB or NDS32_REG16_TAB in your program to
++   access register table.  */
++typedef struct
++{
++  union
++    {
++      int          reg_a[32] ;
++      NDS32_GPR32  reg_s ;
++    } u ;
++} NDS32_REG32_TAB;
++
++typedef struct
++{
++  union
++    {
++      int          reg_a[16] ;
++      NDS32_GPR16  reg_s ;
++    } u ;
++} NDS32_REG16_TAB;
++
++typedef struct
++{
++  int    d0lo;
++  int    d0hi;
++  int    d1lo;
++  int    d1hi;
++} NDS32_DX_TAB;
++
++typedef struct
++{
++#ifdef __NDS32_EB__
++  float    fsr0;
++  float    fsr1;
++  float    fsr2;
++  float    fsr3;
++  float    fsr4;
++  float    fsr5;
++  float    fsr6;
++  float    fsr7;
++#else
++  float    fsr1;
++  float    fsr0;
++  float    fsr3;
++  float    fsr2;
++  float    fsr5;
++  float    fsr4;
++  float    fsr7;
++  float    fsr6;
++#endif
++} NDS32_FSR8;
++
++typedef struct
++{
++  double   dsr0;
++  double   dsr1;
++  double   dsr2;
++  double   dsr3;
++} NDS32_DSR4;
++
++typedef struct
++{
++#ifdef __NDS32_EB__
++  float    fsr0;
++  float    fsr1;
++  float    fsr2;
++  float    fsr3;
++  float    fsr4;
++  float    fsr5;
++  float    fsr6;
++  float    fsr7;
++  float    fsr8;
++  float    fsr9;
++  float    fsr10;
++  float    fsr11;
++  float    fsr12;
++  float    fsr13;
++  float    fsr14;
++  float    fsr15;
++#else
++  float    fsr1;
++  float    fsr0;
++  float    fsr3;
++  float    fsr2;
++  float    fsr5;
++  float    fsr4;
++  float    fsr7;
++  float    fsr6;
++  float    fsr9;
++  float    fsr8;
++  float    fsr11;
++  float    fsr10;
++  float    fsr13;
++  float    fsr12;
++  float    fsr15;
++  float    fsr14;
++#endif
++} NDS32_FSR16;
++
++typedef struct
++{
++  double   dsr0;
++  double   dsr1;
++  double   dsr2;
++  double   dsr3;
++  double   dsr4;
++  double   dsr5;
++  double   dsr6;
++  double   dsr7;
++} NDS32_DSR8;
++
++typedef struct
++{
++#ifdef __NDS32_EB__
++  float    fsr0;
++  float    fsr1;
++  float    fsr2;
++  float    fsr3;
++  float    fsr4;
++  float    fsr5;
++  float    fsr6;
++  float    fsr7;
++  float    fsr8;
++  float    fsr9;
++  float    fsr10;
++  float    fsr11;
++  float    fsr12;
++  float    fsr13;
++  float    fsr14;
++  float    fsr15;
++  float    fsr16;
++  float    fsr17;
++  float    fsr18;
++  float    fsr19;
++  float    fsr20;
++  float    fsr21;
++  float    fsr22;
++  float    fsr23;
++  float    fsr24;
++  float    fsr25;
++  float    fsr26;
++  float    fsr27;
++  float    fsr28;
++  float    fsr29;
++  float    fsr30;
++  float    fsr31;
++#else
++  float    fsr1;
++  float    fsr0;
++  float    fsr3;
++  float    fsr2;
++  float    fsr5;
++  float    fsr4;
++  float    fsr7;
++  float    fsr6;
++  float    fsr9;
++  float    fsr8;
++  float    fsr11;
++  float    fsr10;
++  float    fsr13;
++  float    fsr12;
++  float    fsr15;
++  float    fsr14;
++  float    fsr17;
++  float    fsr16;
++  float    fsr19;
++  float    fsr18;
++  float    fsr21;
++  float    fsr20;
++  float    fsr23;
++  float    fsr22;
++  float    fsr25;
++  float    fsr24;
++  float    fsr27;
++  float    fsr26;
++  float    fsr29;
++  float    fsr28;
++  float    fsr31;
++  float    fsr30;
++#endif
++} NDS32_FSR32;
++
++typedef struct
++{
++  double   dsr0;
++  double   dsr1;
++  double   dsr2;
++  double   dsr3;
++  double   dsr4;
++  double   dsr5;
++  double   dsr6;
++  double   dsr7;
++  double   dsr8;
++  double   dsr9;
++  double   dsr10;
++  double   dsr11;
++  double   dsr12;
++  double   dsr13;
++  double   dsr14;
++  double   dsr15;
++} NDS32_DSR16;
++
++typedef struct
++{
++  double   dsr0;
++  double   dsr1;
++  double   dsr2;
++  double   dsr3;
++  double   dsr4;
++  double   dsr5;
++  double   dsr6;
++  double   dsr7;
++  double   dsr8;
++  double   dsr9;
++  double   dsr10;
++  double   dsr11;
++  double   dsr12;
++  double   dsr13;
++  double   dsr14;
++  double   dsr15;
++  double   dsr16;
++  double   dsr17;
++  double   dsr18;
++  double   dsr19;
++  double   dsr20;
++  double   dsr21;
++  double   dsr22;
++  double   dsr23;
++  double   dsr24;
++  double   dsr25;
++  double   dsr26;
++  double   dsr27;
++  double   dsr28;
++  double   dsr29;
++  double   dsr30;
++  double   dsr31;
++} NDS32_DSR32;
++
++typedef struct
++{
++  union
++    {
++      NDS32_FSR8   fsr_s ;
++      NDS32_DSR4   dsr_s ;
++    } u ;
++} NDS32_FPU8_TAB;
++
++typedef struct
++{
++  union
++    {
++      NDS32_FSR16  fsr_s ;
++      NDS32_DSR8   dsr_s ;
++    } u ;
++} NDS32_FPU16_TAB;
++
++typedef struct
++{
++  union
++    {
++      NDS32_FSR32  fsr_s ;
++      NDS32_DSR16  dsr_s ;
++    } u ;
++} NDS32_FPU32_TAB;
++
++typedef struct
++{
++  union
++    {
++      NDS32_FSR32  fsr_s ;
++      NDS32_DSR32  dsr_s ;
++    } u ;
++} NDS32_FPU64_TAB;
++
++typedef struct
++{
++  int    ipc;
++  int    ipsw;
++#if defined(NDS32_EXT_FPU_CONFIG_0)
++  NDS32_FPU8_TAB fpr;
++#elif defined(NDS32_EXT_FPU_CONFIG_1)
++  NDS32_FPU16_TAB fpr;
++#elif defined(NDS32_EXT_FPU_CONFIG_2)
++  NDS32_FPU32_TAB fpr;
++#elif defined(NDS32_EXT_FPU_CONFIG_3)
++  NDS32_FPU64_TAB fpr;
++#endif
++#if __NDS32_DX_REGS__
++  NDS32_DX_TAB dxr;
++#endif
++#if __NDS32_EXT_IFC__
++  int    ifc_lp;
++  int    filler;
++#endif
++#if __NDS32_REDUCED_REGS__ || __NDS32_REDUCE_REGS
++  NDS32_REG16_TAB gpr;
++#else
++  NDS32_REG32_TAB gpr;
++#endif
++} NDS32_CONTEXT;
++
++/* Predefined Vector Definition.
++
++   For IVIC Mode: 9 to 14 are for hardware interrupt
++                  and 15 is for software interrupt.
++   For EVIC Mode: 9 to 72 are for hardware interrupt
++                  and software interrupt can be routed to any one of them.
++
++   You may want to define your hardware interrupts in the following way
++   for easy maintainance.
++
++     IVIC mode:
++       #define MY_HW_IVIC_TIMER NDS32_VECTOR_INTERRUPT_HW0 + 1
++       #define MY_HW_IVIC_USB   NDS32_VECTOR_INTERRUPT_HW0 + 3
++     EVIC mode:
++     #define MY_HW_EVIC_DMA   NDS32_VECTOR_INTERRUPT_HW0 + 2
++     #define MY_HW_EVIC_SWI   NDS32_VECTOR_INTERRUPT_HW0 + 10 */
++#define NDS32_VECTOR_RESET               0
++#define NDS32_VECTOR_TLB_FILL            1
++#define NDS32_VECTOR_PTE_NOT_PRESENT     2
++#define NDS32_VECTOR_TLB_MISC            3
++#define NDS32_VECTOR_TLB_VLPT_MISS       4
++#define NDS32_VECTOR_MACHINE_ERROR       5
++#define NDS32_VECTOR_DEBUG_RELATED       6
++#define NDS32_VECTOR_GENERAL_EXCEPTION   7
++#define NDS32_VECTOR_SYSCALL             8
++#define NDS32_VECTOR_INTERRUPT_HW0       9
++#define NDS32_VECTOR_INTERRUPT_HW1       10
++#define NDS32_VECTOR_INTERRUPT_HW2       11
++#define NDS32_VECTOR_INTERRUPT_HW3       12
++#define NDS32_VECTOR_INTERRUPT_HW4       13
++#define NDS32_VECTOR_INTERRUPT_HW5       14
++#define NDS32_VECTOR_INTERRUPT_HW6       15
++#define NDS32_VECTOR_SWI                 15  /* THIS IS FOR IVIC MODE ONLY */
++#define NDS32_VECTOR_INTERRUPT_HW7       16
++#define NDS32_VECTOR_INTERRUPT_HW8       17
++#define NDS32_VECTOR_INTERRUPT_HW9       18
++#define NDS32_VECTOR_INTERRUPT_HW10      19
++#define NDS32_VECTOR_INTERRUPT_HW11      20
++#define NDS32_VECTOR_INTERRUPT_HW12      21
++#define NDS32_VECTOR_INTERRUPT_HW13      22
++#define NDS32_VECTOR_INTERRUPT_HW14      23
++#define NDS32_VECTOR_INTERRUPT_HW15      24
++#define NDS32_VECTOR_INTERRUPT_HW16      25
++#define NDS32_VECTOR_INTERRUPT_HW17      26
++#define NDS32_VECTOR_INTERRUPT_HW18      27
++#define NDS32_VECTOR_INTERRUPT_HW19      28
++#define NDS32_VECTOR_INTERRUPT_HW20      29
++#define NDS32_VECTOR_INTERRUPT_HW21      30
++#define NDS32_VECTOR_INTERRUPT_HW22      31
++#define NDS32_VECTOR_INTERRUPT_HW23      32
++#define NDS32_VECTOR_INTERRUPT_HW24      33
++#define NDS32_VECTOR_INTERRUPT_HW25      34
++#define NDS32_VECTOR_INTERRUPT_HW26      35
++#define NDS32_VECTOR_INTERRUPT_HW27      36
++#define NDS32_VECTOR_INTERRUPT_HW28      37
++#define NDS32_VECTOR_INTERRUPT_HW29      38
++#define NDS32_VECTOR_INTERRUPT_HW30      39
++#define NDS32_VECTOR_INTERRUPT_HW31      40
++#define NDS32_VECTOR_INTERRUPT_HW32      41
++#define NDS32_VECTOR_INTERRUPT_HW33      42
++#define NDS32_VECTOR_INTERRUPT_HW34      43
++#define NDS32_VECTOR_INTERRUPT_HW35      44
++#define NDS32_VECTOR_INTERRUPT_HW36      45
++#define NDS32_VECTOR_INTERRUPT_HW37      46
++#define NDS32_VECTOR_INTERRUPT_HW38      47
++#define NDS32_VECTOR_INTERRUPT_HW39      48
++#define NDS32_VECTOR_INTERRUPT_HW40      49
++#define NDS32_VECTOR_INTERRUPT_HW41      50
++#define NDS32_VECTOR_INTERRUPT_HW42      51
++#define NDS32_VECTOR_INTERRUPT_HW43      52
++#define NDS32_VECTOR_INTERRUPT_HW44      53
++#define NDS32_VECTOR_INTERRUPT_HW45      54
++#define NDS32_VECTOR_INTERRUPT_HW46      55
++#define NDS32_VECTOR_INTERRUPT_HW47      56
++#define NDS32_VECTOR_INTERRUPT_HW48      57
++#define NDS32_VECTOR_INTERRUPT_HW49      58
++#define NDS32_VECTOR_INTERRUPT_HW50      59
++#define NDS32_VECTOR_INTERRUPT_HW51      60
++#define NDS32_VECTOR_INTERRUPT_HW52      61
++#define NDS32_VECTOR_INTERRUPT_HW53      62
++#define NDS32_VECTOR_INTERRUPT_HW54      63
++#define NDS32_VECTOR_INTERRUPT_HW55      64
++#define NDS32_VECTOR_INTERRUPT_HW56      65
++#define NDS32_VECTOR_INTERRUPT_HW57      66
++#define NDS32_VECTOR_INTERRUPT_HW58      67
++#define NDS32_VECTOR_INTERRUPT_HW59      68
++#define NDS32_VECTOR_INTERRUPT_HW60      69
++#define NDS32_VECTOR_INTERRUPT_HW61      70
++#define NDS32_VECTOR_INTERRUPT_HW62      71
++#define NDS32_VECTOR_INTERRUPT_HW63      72
++
++#define NDS32ATTR_RESET(option)          __attribute__((reset(option)))
++#define NDS32ATTR_EXCEPT(type)           __attribute__((exception(type)))
++#define NDS32ATTR_EXCEPTION(type)        __attribute__((exception(type)))
++#define NDS32ATTR_INTERRUPT(type)        __attribute__((interrupt(type)))
++#define NDS32ATTR_ISR(type)              __attribute__((interrupt(type)))
++
++#endif /* nds32_isr.h */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-linux.opt gcc-8.2.0/gcc/config/nds32/nds32-linux.opt
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-linux.opt	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-linux.opt	2019-01-25 15:38:32.829242659 +0100
+@@ -0,0 +1,16 @@
++mcmodel=
++Target RejectNegative Joined Enum(nds32_cmodel_type) Var(nds32_cmodel_option) Init(CMODEL_LARGE)
++Specify the address generation strategy for code model.
++
++Enum
++Name(nds32_cmodel_type) Type(enum nds32_cmodel_type)
++Known cmodel types (for use with the -mcmodel= option):
++
++EnumValue
++Enum(nds32_cmodel_type) String(small) Value(CMODEL_SMALL)
++
++EnumValue
++Enum(nds32_cmodel_type) String(medium) Value(CMODEL_MEDIUM)
++
++EnumValue
++Enum(nds32_cmodel_type) String(large) Value(CMODEL_LARGE)
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32.md gcc-8.2.0/gcc/config/nds32/nds32.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32.md	2018-04-08 11:21:30.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32.md	2019-01-25 15:38:32.833242671 +0100
+@@ -56,24 +56,29 @@
+ ;; ------------------------------------------------------------------------
+ 
+ ;; CPU pipeline model.
+-(define_attr "pipeline_model" "n7,n8,e8,n9,simple"
++(define_attr "pipeline_model" "n7,n8,e8,n9,n10,graywolf,n13,simple"
+   (const
+     (cond [(match_test "nds32_cpu_option == CPU_N7")  (const_string "n7")
+ 	   (match_test "nds32_cpu_option == CPU_E8")  (const_string "e8")
+ 	   (match_test "nds32_cpu_option == CPU_N6 || nds32_cpu_option == CPU_N8")  (const_string "n8")
+ 	   (match_test "nds32_cpu_option == CPU_N9")  (const_string "n9")
++	   (match_test "nds32_cpu_option == CPU_N10") (const_string "n10")
++	   (match_test "nds32_cpu_option == CPU_GRAYWOLF") (const_string "graywolf")
++	   (match_test "nds32_cpu_option == CPU_N12") (const_string "n13")
++	   (match_test "nds32_cpu_option == CPU_N13") (const_string "n13")
+ 	   (match_test "nds32_cpu_option == CPU_SIMPLE") (const_string "simple")]
+ 	  (const_string "n9"))))
+ 
+ ;; Insn type, it is used to default other attribute values.
+ (define_attr "type"
+   "unknown,load,store,load_multiple,store_multiple,alu,alu_shift,pbsad,pbsada,mul,mac,div,branch,mmu,misc,\
+-   falu,fmuls,fmuld,fmacs,fmacd,fdivs,fdivd,fsqrts,fsqrtd,fcmp,fabs,fcpy,fcmov,fmfsr,fmfdr,fmtsr,fmtdr,fload,fstore"
++   falu,fmuls,fmuld,fmacs,fmacd,fdivs,fdivd,fsqrts,fsqrtd,fcmp,fabs,fcpy,fcmov,fmfsr,fmfdr,fmtsr,fmtdr,fload,fstore,\
++   dalu,dalu64,daluround,dcmp,dclip,dmul,dmac,dinsb,dpack,dbpick,dwext"
+   (const_string "unknown"))
+ 
+ ;; Insn sub-type
+ (define_attr "subtype"
+-  "simple,shift"
++  "simple,shift,saturation"
+   (const_string "simple"))
+ 
+ ;; Length, in bytes, default is 4-bytes.
+@@ -133,6 +138,7 @@
+ 
+ ;; ----------------------------------------------------------------------------
+ 
++(include "nds32-dspext.md")
+ 
+ ;; Move instructions.
+ 
+@@ -209,6 +215,27 @@
+ 						  low12_int));
+       DONE;
+     }
++
++  if ((REG_P (operands[0]) || GET_CODE (operands[0]) == SUBREG)
++       && SYMBOLIC_CONST_P (operands[1]))
++    {
++      if (TARGET_ICT_MODEL_LARGE
++	  && nds32_indirect_call_referenced_p (operands[1]))
++	{
++	  nds32_expand_ict_move (operands);
++	  DONE;
++	}
++      else if (nds32_tls_referenced_p (operands [1]))
++	{
++	  nds32_expand_tls_move (operands);
++	  DONE;
++	}
++      else if (flag_pic)
++	{
++	  nds32_expand_pic_move (operands);
++	  DONE;
++	}
++    }
+ })
+ 
+ (define_insn "*mov<mode>"
+@@ -271,8 +298,8 @@
+ ;; We use nds32_symbolic_operand to limit that only CONST/SYMBOL_REF/LABEL_REF
+ ;; are able to match such instruction template.
+ (define_insn "move_addr"
+-  [(set (match_operand:SI 0 "register_operand"       "=l, r")
+-	(match_operand:SI 1 "nds32_symbolic_operand" " i, i"))]
++  [(set (match_operand:SI 0 "nds32_general_register_operand"   "=l, r")
++	(match_operand:SI 1 "nds32_nonunspec_symbolic_operand" " i, i"))]
+   ""
+   "la\t%0, %1"
+   [(set_attr "type"  "alu")
+@@ -351,13 +378,58 @@
+ 
+ 
+ ;; ----------------------------------------------------------------------------
++(define_expand "extv"
++  [(set (match_operand 0 "register_operand" "")
++        (sign_extract (match_operand 1 "nonimmediate_operand" "")
++                      (match_operand 2 "const_int_operand" "")
++                      (match_operand 3 "const_int_operand" "")))]
++  ""
++{
++  enum nds32_expand_result_type result = nds32_expand_extv (operands);
++  switch (result)
++    {
++    case EXPAND_DONE:
++      DONE;
++      break;
++    case EXPAND_FAIL:
++      FAIL;
++      break;
++    case EXPAND_CREATE_TEMPLATE:
++      break;
++    default:
++      gcc_unreachable ();
++    }
++})
++
++(define_expand "insv"
++  [(set (zero_extract (match_operand 0 "nonimmediate_operand" "")
++                      (match_operand 1 "const_int_operand" "")
++                      (match_operand 2 "const_int_operand" ""))
++        (match_operand 3 "register_operand" ""))]
++  ""
++{
++  enum nds32_expand_result_type result = nds32_expand_insv (operands);
++  switch (result)
++    {
++    case EXPAND_DONE:
++      DONE;
++      break;
++    case EXPAND_FAIL:
++      FAIL;
++      break;
++    case EXPAND_CREATE_TEMPLATE:
++      break;
++    default:
++      gcc_unreachable ();
++    }
++})
+ 
+ ;; Arithmetic instructions.
+ 
+ (define_insn "addsi3"
+   [(set (match_operand:SI 0 "register_operand"               "=   d,   l,   d,   l, d, l,   k,   l,    r, r")
+ 	(plus:SI (match_operand:SI 1 "register_operand"      "%   0,   l,   0,   l, 0, l,   0,   k,    r, r")
+-		 (match_operand:SI 2 "nds32_rimm15s_operand" " In05,In03,Iu05,Iu03, r, l,Is10,Iu06, Is15, r")))]
++		 (match_operand:SI 2 "nds32_rimm15s_operand" " In05,In03,Iu05,Iu03, r, l,Is10,IU06, Is15, r")))]
+   ""
+ {
+   switch (which_alternative)
+@@ -1428,11 +1500,30 @@
+ 	      (clobber (reg:SI LP_REGNUM))
+ 	      (clobber (reg:SI TA_REGNUM))])]
+   ""
+-  ""
++  {
++    rtx insn;
++    rtx sym = XEXP (operands[0], 0);
++
++    if (TARGET_ICT_MODEL_LARGE
++	&& nds32_indirect_call_referenced_p (sym))
++      {
++	rtx reg = gen_reg_rtx (Pmode);
++	emit_move_insn (reg, sym);
++	operands[0] = gen_const_mem (Pmode, reg);
++      }
++
++    if (flag_pic)
++      {
++	insn = emit_call_insn (gen_call_internal
++			       (XEXP (operands[0], 0), GEN_INT (0)));
++	use_reg (&CALL_INSN_FUNCTION_USAGE (insn), pic_offset_table_rtx);
++	DONE;
++      }
++  }
+ )
+ 
+ (define_insn "call_internal"
+-  [(parallel [(call (mem (match_operand:SI 0 "nds32_call_address_operand" "r, i"))
++  [(parallel [(call (mem (match_operand:SI 0 "nds32_call_address_operand" "r, S"))
+ 		    (match_operand 1))
+ 	      (clobber (reg:SI LP_REGNUM))
+ 	      (clobber (reg:SI TA_REGNUM))])]
+@@ -1474,9 +1565,11 @@
+ 		     (const_int 2)
+ 		     (const_int 4))
+        ;; Alternative 1
+-       (if_then_else (match_test "nds32_long_call_p (operands[0])")
+-		     (const_int 12)
+-		     (const_int 4))
++       (if_then_else (match_test "flag_pic")
++		     (const_int 16)
++		     (if_then_else (match_test "nds32_long_call_p (operands[0])")
++				   (const_int 12)
++				   (const_int 4)))
+      ])]
+ )
+ 
+@@ -1492,11 +1585,33 @@
+ 		         (match_operand 2)))
+ 	      (clobber (reg:SI LP_REGNUM))
+ 	      (clobber (reg:SI TA_REGNUM))])]
+-  "")
++  ""
++  {
++    rtx insn;
++    rtx sym = XEXP (operands[1], 0);
++
++    if (TARGET_ICT_MODEL_LARGE
++	&& nds32_indirect_call_referenced_p (sym))
++      {
++	rtx reg = gen_reg_rtx (Pmode);
++	emit_move_insn (reg, sym);
++	operands[1] = gen_const_mem (Pmode, reg);
++      }
++
++    if (flag_pic)
++      {
++	insn =
++	  emit_call_insn (gen_call_value_internal
++			  (operands[0], XEXP (operands[1], 0), GEN_INT (0)));
++	use_reg (&CALL_INSN_FUNCTION_USAGE (insn), pic_offset_table_rtx);
++	DONE;
++      }
++  }
++)
+ 
+ (define_insn "call_value_internal"
+   [(parallel [(set (match_operand 0)
+-		   (call (mem (match_operand:SI 1 "nds32_call_address_operand" "r, i"))
++		   (call (mem (match_operand:SI 1 "nds32_call_address_operand" "r, S"))
+ 		         (match_operand 2)))
+ 	      (clobber (reg:SI LP_REGNUM))
+ 	      (clobber (reg:SI TA_REGNUM))])]
+@@ -1538,9 +1653,11 @@
+ 		     (const_int 2)
+ 		     (const_int 4))
+        ;; Alternative 1
+-       (if_then_else (match_test "nds32_long_call_p (operands[1])")
+-		     (const_int 12)
+-		     (const_int 4))
++       (if_then_else (match_test "flag_pic")
++		     (const_int 16)
++		     (if_then_else (match_test "nds32_long_call_p (operands[1])")
++				   (const_int 12)
++				   (const_int 4)))
+      ])]
+ )
+ 
+@@ -1583,10 +1700,21 @@
+ 		    (const_int 0))
+ 	      (clobber (reg:SI TA_REGNUM))
+ 	      (return)])]
+-  "")
++  ""
++{
++    rtx sym = XEXP (operands[0], 0);
++
++    if (TARGET_ICT_MODEL_LARGE
++	&& nds32_indirect_call_referenced_p (sym))
++      {
++	rtx reg = gen_reg_rtx (Pmode);
++	emit_move_insn (reg, sym);
++	operands[0] = gen_const_mem (Pmode, reg);
++      }
++})
+ 
+ (define_insn "sibcall_internal"
+-  [(parallel [(call (mem (match_operand:SI 0 "nds32_call_address_operand" "r, i"))
++  [(parallel [(call (mem (match_operand:SI 0 "nds32_call_address_operand" "r, S"))
+ 		    (match_operand 1))
+ 	      (clobber (reg:SI TA_REGNUM))
+ 	      (return)])]
+@@ -1617,9 +1745,11 @@
+ 		     (const_int 2)
+ 		     (const_int 4))
+        ;; Alternative 1
+-       (if_then_else (match_test "nds32_long_call_p (operands[0])")
+-		     (const_int 12)
+-		     (const_int 4))
++       (if_then_else (match_test "flag_pic")
++		     (const_int 16)
++		     (if_then_else (match_test "nds32_long_call_p (operands[0])")
++				   (const_int 12)
++				   (const_int 4)))
+      ])]
+ )
+ 
+@@ -1633,11 +1763,22 @@
+ 			 (const_int 0)))
+ 	      (clobber (reg:SI TA_REGNUM))
+ 	      (return)])]
+-  "")
++  ""
++{
++    rtx sym = XEXP (operands[1], 0);
++
++    if (TARGET_ICT_MODEL_LARGE
++	&& nds32_indirect_call_referenced_p (sym))
++      {
++	rtx reg = gen_reg_rtx (Pmode);
++	emit_move_insn (reg, sym);
++	operands[1] = gen_const_mem (Pmode, reg);
++      }
++})
+ 
+ (define_insn "sibcall_value_internal"
+   [(parallel [(set (match_operand 0)
+-		   (call (mem (match_operand:SI 1 "nds32_call_address_operand" "r, i"))
++		   (call (mem (match_operand:SI 1 "nds32_call_address_operand" "r, S"))
+ 			 (match_operand 2)))
+ 	      (clobber (reg:SI TA_REGNUM))
+ 	      (return)])]
+@@ -1668,9 +1809,11 @@
+ 		     (const_int 2)
+ 		     (const_int 4))
+        ;; Alternative 1
+-       (if_then_else (match_test "nds32_long_call_p (operands[1])")
+-		     (const_int 12)
+-		     (const_int 4))
++       (if_then_else (match_test "flag_pic")
++		     (const_int 16)
++		     (if_then_else (match_test "nds32_long_call_p (operands[1])")
++				   (const_int 12)
++				   (const_int 4)))
+      ])]
+ )
+ 
+@@ -1687,12 +1830,33 @@
+     nds32_expand_prologue_v3push ();
+   else
+     nds32_expand_prologue ();
++
++  /* If cfun->machine->fp_as_gp_p is true, we can generate special
++     directive to guide linker doing fp-as-gp optimization.
++     However, for a naked function, which means
++     it should not have prologue/epilogue,
++     using fp-as-gp still requires saving $fp by push/pop behavior and
++     there is no benefit to use fp-as-gp on such small function.
++     So we need to make sure this function is NOT naked as well.  */
++  if (cfun->machine->fp_as_gp_p && !cfun->machine->naked_p)
++    emit_insn (gen_omit_fp_begin (gen_rtx_REG (SImode, FP_REGNUM)));
++
+   DONE;
+ })
+ 
+ (define_expand "epilogue" [(const_int 0)]
+   ""
+ {
++  /* If cfun->machine->fp_as_gp_p is true, we can generate special
++     directive to guide linker doing fp-as-gp optimization.
++     However, for a naked function, which means
++     it should not have prologue/epilogue,
++     using fp-as-gp still requires saving $fp by push/pop behavior and
++     there is no benefit to use fp-as-gp on such small function.
++     So we need to make sure this function is NOT naked as well.  */
++  if (cfun->machine->fp_as_gp_p && !cfun->machine->naked_p)
++    emit_insn (gen_omit_fp_end (gen_rtx_REG (SImode, FP_REGNUM)));
++
+   /* Note that only under V3/V3M ISA, we could use v3pop epilogue.
+      In addition, we need to check if v3push is indeed available.  */
+   if (NDS32_V3PUSH_AVAILABLE_P)
+@@ -1792,7 +1956,8 @@
+   "nds32_can_use_return_insn ()"
+ {
+   /* Emit as the simple return.  */
+-  if (cfun->machine->naked_p
++  if (!cfun->machine->fp_as_gp_p
++      && cfun->machine->naked_p
+       && (cfun->machine->va_args_size == 0))
+     {
+       emit_jump_insn (gen_return_internal ());
+@@ -1802,9 +1967,14 @@
+ 
+ ;; This pattern is expanded only by the shrink-wrapping optimization
+ ;; on paths where the function prologue has not been executed.
++;; However, such optimization may reorder the prologue/epilogue blocks
++;; together with basic blocks within function body.
++;; So we must disable this pattern if we have already decided
++;; to perform fp_as_gp optimization, which requires prologue to be
++;; first block and epilogue to be last block.
+ (define_expand "simple_return"
+   [(simple_return)]
+-  ""
++  "!cfun->machine->fp_as_gp_p"
+   ""
+ )
+ 
+@@ -1823,6 +1993,9 @@
+   [(simple_return)]
+   ""
+ {
++  if (nds32_isr_function_critical_p (current_function_decl))
++    return "iret";
++
+   if (TARGET_16_BIT)
+     return "ret5";
+   else
+@@ -1831,9 +2004,11 @@
+   [(set_attr "type" "branch")
+    (set_attr "enabled" "yes")
+    (set (attr "length")
+-	(if_then_else (match_test "TARGET_16_BIT")
+-		      (const_int 2)
+-		      (const_int 4)))])
++	(if_then_else (match_test "nds32_isr_function_critical_p (current_function_decl)")
++		      (const_int 4)
++		      (if_then_else (match_test "TARGET_16_BIT")
++				    (const_int 2)
++				    (const_int 4))))])
+ 
+ 
+ ;; ----------------------------------------------------------------------------
+@@ -1868,6 +2043,7 @@
+ {
+   rtx add_tmp;
+   rtx reg, test;
++  rtx tmp_reg;
+ 
+   /* Step A: "k <-- (plus (operands[0]) (-operands[1]))".  */
+   if (operands[1] != const0_rtx)
+@@ -1889,9 +2065,14 @@
+   emit_jump_insn (gen_cbranchsi4 (test, operands[0], operands[2],
+ 				  operands[4]));
+ 
+-  /* Step C, D, E, and F, using another temporary register.  */
+-  rtx tmp = gen_reg_rtx (SImode);
+-  emit_jump_insn (gen_casesi_internal (operands[0], operands[3], tmp));
++  tmp_reg = gen_reg_rtx (SImode);
++  /* Step C, D, E, and F, using another temporary register tmp_reg.  */
++  if (flag_pic)
++    emit_use (pic_offset_table_rtx);
++
++  emit_jump_insn (gen_casesi_internal (operands[0],
++				       operands[3],
++				       tmp_reg));
+   DONE;
+ })
+ 
+@@ -1927,13 +2108,30 @@
+   else
+     return nds32_output_casesi (operands);
+ }
+-  [(set_attr "length" "20")
+-   (set_attr "type" "branch")])
++  [(set_attr "type" "branch")
++   (set (attr "length")
++	(if_then_else (match_test "flag_pic")
++		      (const_int 28)
++		      (const_int 20)))])
+ 
+ ;; ----------------------------------------------------------------------------
+ 
+ ;; Performance Extension
+ 
++; If -fwrapv option is issued, GCC expects there will be
++; signed overflow situation.  So the ABS(INT_MIN) is still INT_MIN
++; (e.g. ABS(0x80000000)=0x80000000).
++; However, the hardware ABS instruction of nds32 target
++; always performs saturation: abs 0x80000000 -> 0x7fffffff.
++; So that we can only enable abssi2 pattern if flag_wrapv is NOT presented.
++(define_insn "abssi2"
++  [(set (match_operand:SI 0 "register_operand"         "=r")
++	(abs:SI (match_operand:SI 1 "register_operand" " r")))]
++  "TARGET_EXT_PERF && TARGET_HW_ABS && !flag_wrapv"
++  "abs\t%0, %1"
++  [(set_attr "type" "alu")
++   (set_attr "length" "4")])
++
+ (define_insn "clzsi2"
+   [(set (match_operand:SI 0 "register_operand"         "=r")
+ 	(clz:SI (match_operand:SI 1 "register_operand" " r")))]
+@@ -1996,6 +2194,25 @@
+   [(set_attr "length" "0")]
+ )
+ 
++;; Output .omit_fp_begin for fp-as-gp optimization.
++;; Also we have to set $fp register.
++(define_insn "omit_fp_begin"
++  [(set (match_operand:SI 0 "register_operand" "=x")
++	(unspec_volatile:SI [(const_int 0)] UNSPEC_VOLATILE_OMIT_FP_BEGIN))]
++  ""
++  "! -----\;.omit_fp_begin\;la\t$fp,_FP_BASE_\;! -----"
++  [(set_attr "length" "8")]
++)
++
++;; Output .omit_fp_end for fp-as-gp optimization.
++;; Claim that we have to use $fp register.
++(define_insn "omit_fp_end"
++  [(unspec_volatile:SI [(match_operand:SI 0 "register_operand" "x")] UNSPEC_VOLATILE_OMIT_FP_END)]
++  ""
++  "! -----\;.omit_fp_end\;! -----"
++  [(set_attr "length" "0")]
++)
++
+ (define_insn "pop25return"
+   [(return)
+    (unspec_volatile:SI [(reg:SI LP_REGNUM)] UNSPEC_VOLATILE_POP25_RETURN)]
+@@ -2004,6 +2221,36 @@
+   [(set_attr "length" "0")]
+ )
+ 
++;; Add pc
++(define_insn "add_pc"
++  [(set (match_operand:SI 0 "register_operand"          "=r")
++	(plus:SI (match_operand:SI 1 "register_operand"  "0")
++		 (pc)))]
++  "TARGET_LINUX_ABI || flag_pic"
++  "add5.pc\t%0"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
++(define_expand "bswapsi2"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(bswap:SI (match_operand:SI 1 "register_operand" "r")))]
++  ""
++{
++  emit_insn (gen_unspec_wsbh (operands[0], operands[1]));
++  emit_insn (gen_rotrsi3 (operands[0], operands[0], GEN_INT (16)));
++  DONE;
++})
++
++(define_insn "bswaphi2"
++  [(set (match_operand:HI 0 "register_operand" "=r")
++	(bswap:HI (match_operand:HI 1 "register_operand" "r")))]
++  ""
++  "wsbh\t%0, %1"
++  [(set_attr "type"    "alu")
++   (set_attr "length"    "4")]
++)
++
+ ;; ----------------------------------------------------------------------------
+ 
+ ;; Patterns for exception handling
+@@ -2068,3 +2315,57 @@
+ })
+ 
+ ;; ----------------------------------------------------------------------------
++
++;; Patterns for TLS.
++;; The following two tls patterns don't be expanded directly because the
++;; intermediate value may be spilled into the stack.  As a result, it is
++;; hard to analyze the define-use chain in the relax_opt pass.
++
++
++;; There is a unspec operand to record RELAX_GROUP number because each
++;; emitted instruction need a relax_hint above it.
++(define_insn "tls_desc"
++  [(set (reg:SI 0)
++	(call (unspec_volatile:SI [(match_operand:SI 0 "nds32_symbolic_operand" "i")] UNSPEC_TLS_DESC)
++	      (const_int 1)))
++   (use (unspec [(match_operand:SI 1 "immediate_operand" "i")] UNSPEC_VOLATILE_RELAX_GROUP))
++   (use (reg:SI GP_REGNUM))
++   (clobber (reg:SI LP_REGNUM))
++   (clobber (reg:SI TA_REGNUM))]
++  ""
++  {
++    return nds32_output_tls_desc (operands);
++  }
++  [(set_attr "length" "20")
++   (set_attr "type" "branch")]
++)
++
++;; There is a unspec operand to record RELAX_GROUP number because each
++;; emitted instruction need a relax_hint above it.
++(define_insn "tls_ie"
++  [(set (match_operand:SI 0 "register_operand" "=r")
++	(unspec:SI [(match_operand:SI 1 "nds32_symbolic_operand" "i")] UNSPEC_TLS_IE))
++   (use (unspec [(match_operand:SI 2 "immediate_operand" "i")] UNSPEC_VOLATILE_RELAX_GROUP))
++   (use (reg:SI GP_REGNUM))]
++  ""
++  {
++    return nds32_output_tls_ie (operands);
++  }
++  [(set (attr "length") (if_then_else (match_test "flag_pic")
++				      (const_int 12)
++				      (const_int 8)))
++   (set_attr "type" "misc")]
++)
++
++;; The pattern is for some relaxation groups that have to keep addsi3 in 32-bit mode.
++(define_insn "addsi3_32bit"
++  [(set (match_operand:SI 0 "register_operand"             "=r")
++	(unspec:SI [(match_operand:SI 1 "register_operand" "%r")
++		    (match_operand:SI 2 "register_operand" " r")] UNSPEC_ADD32))]
++  ""
++  "add\t%0, %1, %2";
++  [(set_attr "type"    "alu")
++   (set_attr "length"  "4")
++   (set_attr "feature" "v1")])
++
++;; ----------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-md-auxiliary.c gcc-8.2.0/gcc/config/nds32/nds32-md-auxiliary.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-md-auxiliary.c	2018-04-08 08:00:34.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-md-auxiliary.c	2019-01-25 15:38:32.829242659 +0100
+@@ -39,6 +39,9 @@
+ #include "expr.h"
+ #include "emit-rtl.h"
+ #include "explow.h"
++#include "stringpool.h"
++#include "attribs.h"
++
+ 
+ /* ------------------------------------------------------------------------ */
+ 
+@@ -261,6 +264,118 @@
+   output_asm_insn (pattern, operands);
+ }
+ 
++static void
++nds32_split_shiftrtdi3 (rtx dst, rtx src, rtx shiftamount, bool logic_shift_p)
++{
++  rtx src_high_part;
++  rtx dst_high_part, dst_low_part;
++
++  dst_high_part = nds32_di_high_part_subreg (dst);
++  src_high_part = nds32_di_high_part_subreg (src);
++  dst_low_part = nds32_di_low_part_subreg (dst);
++
++  if (CONST_INT_P (shiftamount))
++    {
++      if (INTVAL (shiftamount) < 32)
++	{
++	  if (logic_shift_p)
++	    {
++	      emit_insn (gen_uwext (dst_low_part, src,
++						  shiftamount));
++	      emit_insn (gen_lshrsi3 (dst_high_part, src_high_part,
++						     shiftamount));
++	    }
++	  else
++	    {
++	      emit_insn (gen_wext (dst_low_part, src,
++						 shiftamount));
++	      emit_insn (gen_ashrsi3 (dst_high_part, src_high_part,
++						     shiftamount));
++	    }
++	}
++      else
++	{
++	  rtx new_shift_amout = gen_int_mode(INTVAL (shiftamount) - 32, SImode);
++
++	  if (logic_shift_p)
++	    {
++	      emit_insn (gen_lshrsi3 (dst_low_part, src_high_part,
++						    new_shift_amout));
++	      emit_move_insn (dst_high_part, const0_rtx);
++	    }
++	  else
++	    {
++	      emit_insn (gen_ashrsi3 (dst_low_part, src_high_part,
++						    new_shift_amout));
++	      emit_insn (gen_ashrsi3 (dst_high_part, src_high_part,
++						     GEN_INT (31)));
++	    }
++	}
++    }
++  else
++    {
++      rtx dst_low_part_l32, dst_high_part_l32;
++      rtx dst_low_part_g32, dst_high_part_g32;
++      rtx new_shift_amout, select_reg;
++      dst_low_part_l32 = gen_reg_rtx (SImode);
++      dst_high_part_l32 = gen_reg_rtx (SImode);
++      dst_low_part_g32 = gen_reg_rtx (SImode);
++      dst_high_part_g32 = gen_reg_rtx (SImode);
++      new_shift_amout = gen_reg_rtx (SImode);
++      select_reg = gen_reg_rtx (SImode);
++
++      emit_insn (gen_andsi3 (shiftamount, shiftamount, GEN_INT (0x3f)));
++
++      if (logic_shift_p)
++	{
++	  /*
++	     if (shiftamount < 32)
++	       dst_low_part = wext (src, shiftamount)
++	       dst_high_part = src_high_part >> shiftamount
++	     else
++	       dst_low_part = src_high_part >> (shiftamount & 0x1f)
++	       dst_high_part = 0
++	  */
++	  emit_insn (gen_uwext (dst_low_part_l32, src, shiftamount));
++	  emit_insn (gen_lshrsi3 (dst_high_part_l32, src_high_part,
++						     shiftamount));
++
++	  emit_insn (gen_andsi3 (new_shift_amout, shiftamount, GEN_INT (0x1f)));
++	  emit_insn (gen_lshrsi3 (dst_low_part_g32, src_high_part,
++						    new_shift_amout));
++	  emit_move_insn (dst_high_part_g32, const0_rtx);
++	}
++      else
++	{
++	  /*
++	     if (shiftamount < 32)
++	       dst_low_part = wext (src, shiftamount)
++	       dst_high_part = src_high_part >> shiftamount
++	     else
++	       dst_low_part = src_high_part >> (shiftamount & 0x1f)
++	       # shift 31 for sign extend
++	       dst_high_part = src_high_part >> 31
++	  */
++	  emit_insn (gen_wext (dst_low_part_l32, src, shiftamount));
++	  emit_insn (gen_ashrsi3 (dst_high_part_l32, src_high_part,
++						     shiftamount));
++
++	  emit_insn (gen_andsi3 (new_shift_amout, shiftamount, GEN_INT (0x1f)));
++	  emit_insn (gen_ashrsi3 (dst_low_part_g32, src_high_part,
++						    new_shift_amout));
++	  emit_insn (gen_ashrsi3 (dst_high_part_g32, src_high_part,
++						     GEN_INT (31)));
++	}
++
++      emit_insn (gen_slt_compare (select_reg, shiftamount, GEN_INT (32)));
++
++      emit_insn (gen_cmovnsi (dst_low_part, select_reg,
++			      dst_low_part_l32, dst_low_part_g32));
++      emit_insn (gen_cmovnsi (dst_high_part, select_reg,
++			      dst_high_part_l32, dst_high_part_g32));
++  }
++}
++
+ /* ------------------------------------------------------------------------ */
+ 
+ /* Auxiliary function for expand RTL pattern.  */
+@@ -1195,8 +1310,166 @@
+     }
+ }
+ 
++enum nds32_expand_result_type
++nds32_expand_extv (rtx *operands)
++{
++  gcc_assert (CONST_INT_P (operands[2]) && CONST_INT_P (operands[3]));
++  HOST_WIDE_INT width = INTVAL (operands[2]);
++  HOST_WIDE_INT bitpos = INTVAL (operands[3]);
++  rtx dst = operands[0];
++  rtx src = operands[1];
++
++  if (MEM_P (src)
++      && width == 32
++      && (bitpos % BITS_PER_UNIT)  == 0
++      && GET_MODE_BITSIZE (GET_MODE (dst)) == width)
++    {
++      rtx newmem = adjust_address (src, GET_MODE (dst),
++				   bitpos / BITS_PER_UNIT);
++
++      rtx base_addr = force_reg (Pmode, XEXP (newmem, 0));
++
++      emit_insn (gen_unaligned_loadsi (dst, base_addr));
++
++      return EXPAND_DONE;
++    }
++  return EXPAND_FAIL;
++}
++
++enum nds32_expand_result_type
++nds32_expand_insv (rtx *operands)
++{
++  gcc_assert (CONST_INT_P (operands[1]) && CONST_INT_P (operands[2]));
++  HOST_WIDE_INT width = INTVAL (operands[1]);
++  HOST_WIDE_INT bitpos = INTVAL (operands[2]);
++  rtx dst = operands[0];
++  rtx src = operands[3];
++
++  if (MEM_P (dst)
++      && width == 32
++      && (bitpos % BITS_PER_UNIT)  == 0
++      && GET_MODE_BITSIZE (GET_MODE (src)) == width)
++    {
++      rtx newmem = adjust_address (dst, GET_MODE (src),
++				      bitpos / BITS_PER_UNIT);
++
++      rtx base_addr = force_reg (Pmode, XEXP (newmem, 0));
++
++      emit_insn (gen_unaligned_storesi (base_addr, src));
++
++      return EXPAND_DONE;
++    }
++  return EXPAND_FAIL;
++}
++
+ /* ------------------------------------------------------------------------ */
+ 
++/* Function to generate PC relative jump table.
++   Refer to nds32.md for more details.
++
++   The following is the sample for the case that diff value
++   can be presented in '.short' size.
++
++     addi    $r1, $r1, -(case_lower_bound)
++     slti    $ta, $r1, (case_number)
++     beqz    $ta, .L_skip_label
++
++     la      $ta, .L35             ! get jump table address
++     lh      $r1, [$ta + $r1 << 1] ! load symbol diff from jump table entry
++     addi    $ta, $r1, $ta
++     jr5     $ta
++
++     ! jump table entry
++   L35:
++     .short  .L25-.L35
++     .short  .L26-.L35
++     .short  .L27-.L35
++     .short  .L28-.L35
++     .short  .L29-.L35
++     .short  .L30-.L35
++     .short  .L31-.L35
++     .short  .L32-.L35
++     .short  .L33-.L35
++     .short  .L34-.L35 */
++const char *
++nds32_output_casesi_pc_relative (rtx *operands)
++{
++  machine_mode mode;
++  rtx diff_vec;
++
++  diff_vec = PATTERN (NEXT_INSN (as_a <rtx_insn *> (operands[1])));
++
++  gcc_assert (GET_CODE (diff_vec) == ADDR_DIFF_VEC);
++
++  /* Step C: "t <-- operands[1]".  */
++  if (flag_pic)
++    {
++      output_asm_insn ("sethi\t$ta, hi20(%l1@GOTOFF)", operands);
++      output_asm_insn ("ori\t$ta, $ta, lo12(%l1@GOTOFF)", operands);
++      output_asm_insn ("add\t$ta, $ta, $gp", operands);
++    }
++  else
++    output_asm_insn ("la\t$ta, %l1", operands);
++
++  /* Get the mode of each element in the difference vector.  */
++  mode = GET_MODE (diff_vec);
++
++  /* Step D: "z <-- (mem (plus (operands[0] << m) t))",
++     where m is 0, 1, or 2 to load address-diff value from table.  */
++  switch (mode)
++    {
++    case E_QImode:
++      output_asm_insn ("lb\t%2, [$ta + %0 << 0]", operands);
++      break;
++    case E_HImode:
++      output_asm_insn ("lh\t%2, [$ta + %0 << 1]", operands);
++      break;
++    case E_SImode:
++      output_asm_insn ("lw\t%2, [$ta + %0 << 2]", operands);
++      break;
++    default:
++      gcc_unreachable ();
++    }
++
++  /* Step E: "t <-- z + t".
++     Add table label_ref with address-diff value to
++     obtain target case address.  */
++  output_asm_insn ("add\t$ta, %2, $ta", operands);
++
++  /* Step F: jump to target with register t.  */
++  if (TARGET_16_BIT)
++    return "jr5\t$ta";
++  else
++    return "jr\t$ta";
++}
++
++/* Function to generate normal jump table.  */
++const char *
++nds32_output_casesi (rtx *operands)
++{
++  /* Step C: "t <-- operands[1]".  */
++  if (flag_pic)
++    {
++      output_asm_insn ("sethi\t$ta, hi20(%l1@GOTOFF)", operands);
++      output_asm_insn ("ori\t$ta, $ta, lo12(%l1@GOTOFF)", operands);
++      output_asm_insn ("add\t$ta, $ta, $gp", operands);
++    }
++  else
++    output_asm_insn ("la\t$ta, %l1", operands);
++
++  /* Step D: "z <-- (mem (plus (operands[0] << 2) t))".  */
++  output_asm_insn ("lw\t%2, [$ta + %0 << 2]", operands);
++
++  /* No need to perform Step E, which is only used for
++     pc relative jump table.  */
++
++  /* Step F: jump to target with register z.  */
++  if (TARGET_16_BIT)
++    return "jr5\t%2";
++  else
++    return "jr\t%2";
++}
++
+ /* Function to return memory format.  */
+ enum nds32_16bit_address_type
+ nds32_mem_format (rtx op)
+@@ -1757,11 +2030,8 @@
+ 
+   /* If we step here, we are going to do v3push or multiple push operation.  */
+ 
+-  /* The v3push/v3pop instruction should only be applied on
+-     none-isr and none-variadic function.  */
+-  if (TARGET_V3PUSH
+-      && !nds32_isr_function_p (current_function_decl)
+-      && (cfun->machine->va_args_size == 0))
++  /* Refer to nds32.h, where we comment when push25/pop25 are available.  */
++  if (NDS32_V3PUSH_AVAILABLE_P)
+     {
+       /* For stack v3push:
+ 	   operands[0]: Re
+@@ -1881,11 +2151,8 @@
+ 
+   /* If we step here, we are going to do v3pop or multiple pop operation.  */
+ 
+-  /* The v3push/v3pop instruction should only be applied on
+-     none-isr and none-variadic function.  */
+-  if (TARGET_V3PUSH
+-      && !nds32_isr_function_p (current_function_decl)
+-      && (cfun->machine->va_args_size == 0))
++  /* Refer to nds32.h, where we comment when push25/pop25 are available.  */
++  if (NDS32_V3PUSH_AVAILABLE_P)
+     {
+       /* For stack v3pop:
+ 	   operands[0]: Re
+@@ -2022,77 +2289,6 @@
+   return "";
+ }
+ 
+-/* Function to generate PC relative jump table.
+-   Refer to nds32.md for more details.
+-
+-   The following is the sample for the case that diff value
+-   can be presented in '.short' size.
+-
+-     addi    $r1, $r1, -(case_lower_bound)
+-     slti    $ta, $r1, (case_number)
+-     beqz    $ta, .L_skip_label
+-
+-     la      $ta, .L35             ! get jump table address
+-     lh      $r1, [$ta + $r1 << 1] ! load symbol diff from jump table entry
+-     addi    $ta, $r1, $ta
+-     jr5     $ta
+-
+-     ! jump table entry
+-   L35:
+-     .short  .L25-.L35
+-     .short  .L26-.L35
+-     .short  .L27-.L35
+-     .short  .L28-.L35
+-     .short  .L29-.L35
+-     .short  .L30-.L35
+-     .short  .L31-.L35
+-     .short  .L32-.L35
+-     .short  .L33-.L35
+-     .short  .L34-.L35 */
+-const char *
+-nds32_output_casesi_pc_relative (rtx *operands)
+-{
+-  machine_mode mode;
+-  rtx diff_vec;
+-
+-  diff_vec = PATTERN (NEXT_INSN (as_a <rtx_insn *> (operands[1])));
+-
+-  gcc_assert (GET_CODE (diff_vec) == ADDR_DIFF_VEC);
+-
+-  /* Step C: "t <-- operands[1]".  */
+-  output_asm_insn ("la\t$ta, %l1", operands);
+-
+-  /* Get the mode of each element in the difference vector.  */
+-  mode = GET_MODE (diff_vec);
+-
+-  /* Step D: "z <-- (mem (plus (operands[0] << m) t))",
+-     where m is 0, 1, or 2 to load address-diff value from table.  */
+-  switch (mode)
+-    {
+-    case E_QImode:
+-      output_asm_insn ("lb\t%2, [$ta + %0 << 0]", operands);
+-      break;
+-    case E_HImode:
+-      output_asm_insn ("lh\t%2, [$ta + %0 << 1]", operands);
+-      break;
+-    case E_SImode:
+-      output_asm_insn ("lw\t%2, [$ta + %0 << 2]", operands);
+-      break;
+-    default:
+-      gcc_unreachable ();
+-    }
+-
+-  /* Step E: "t <-- z + t".
+-     Add table label_ref with address-diff value to
+-     obtain target case address.  */
+-  output_asm_insn ("add\t$ta, %2, $ta", operands);
+-
+-  /* Step F: jump to target with register t.  */
+-  if (TARGET_16_BIT)
+-    return "jr5\t$ta";
+-  else
+-    return "jr\t$ta";
+-}
+ 
+ /* output a float load instruction */
+ const char *
+@@ -2250,52 +2446,51 @@
+   return "";
+ }
+ 
+-/* Function to generate normal jump table.  */
+ const char *
+-nds32_output_casesi (rtx *operands)
++nds32_output_smw_single_word (rtx *operands)
+ {
+-  /* Step C: "t <-- operands[1]".  */
+-  output_asm_insn ("la\t$ta, %l1", operands);
+-
+-  /* Step D: "z <-- (mem (plus (operands[0] << 2) t))".  */
+-  output_asm_insn ("lw\t%2, [$ta + %0 << 2]", operands);
+-
+-  /* No need to perform Step E, which is only used for
+-     pc relative jump table.  */
++  char buff[100];
++  unsigned regno;
++  int enable4;
++  bool update_base_p;
++  rtx base_addr = operands[0];
++  rtx base_reg;
++  rtx otherops[2];
+ 
+-  /* Step F: jump to target with register z.  */
+-  if (TARGET_16_BIT)
+-    return "jr5\t%2";
++  if (REG_P (XEXP (base_addr, 0)))
++    {
++      update_base_p = false;
++      base_reg = XEXP (base_addr, 0);
++    }
+   else
+-    return "jr\t%2";
+-}
++    {
++      update_base_p = true;
++      base_reg = XEXP (XEXP (base_addr, 0), 0);
++    }
+ 
+-/* Auxiliary functions for lwm/smw.  */
+-bool
+-nds32_valid_smw_lwm_base_p (rtx op)
+-{
+-  rtx base_addr;
++  const char *update_base = update_base_p ? "m" : "";
+ 
+-  if (!MEM_P (op))
+-    return false;
++  regno = REGNO (operands[1]);
+ 
+-  base_addr = XEXP (op, 0);
++  otherops[0] = base_reg;
++  otherops[1] = operands[1];
+ 
+-  if (REG_P (base_addr))
+-    return true;
++  if (regno >= 28)
++    {
++      enable4 = nds32_regno_to_enable4 (regno);
++      sprintf (buff, "smw.bi%s\t$sp, [%%0], $sp, %x", update_base, enable4);
++    }
+   else
+     {
+-      if (GET_CODE (base_addr) == POST_INC
+-	  && REG_P (XEXP (base_addr, 0)))
+-        return true;
++      sprintf (buff, "smw.bi%s\t%%1, [%%0], %%1", update_base);
+     }
+-
+-  return false;
++  output_asm_insn (buff, otherops);
++  return "";
+ }
+ 
+ /* ------------------------------------------------------------------------ */
+ const char *
+-nds32_output_smw_single_word (rtx *operands)
++nds32_output_smw_double_word (rtx *operands)
+ {
+   char buff[100];
+   unsigned regno;
+@@ -2303,7 +2498,7 @@
+   bool update_base_p;
+   rtx base_addr = operands[0];
+   rtx base_reg;
+-  rtx otherops[2];
++  rtx otherops[3];
+ 
+   if (REG_P (XEXP (base_addr, 0)))
+     {
+@@ -2322,15 +2517,22 @@
+ 
+   otherops[0] = base_reg;
+   otherops[1] = operands[1];
++  otherops[2] = gen_rtx_REG (SImode, REGNO (operands[1]) + 1);;
+ 
+   if (regno >= 28)
+     {
+-      enable4 = nds32_regno_to_enable4 (regno);
++      enable4 = nds32_regno_to_enable4 (regno)
++		| nds32_regno_to_enable4 (regno + 1);
+       sprintf (buff, "smw.bi%s\t$sp, [%%0], $sp, %x", update_base, enable4);
+     }
++  else if (regno == 27)
++    {
++      enable4 = nds32_regno_to_enable4 (regno + 1);
++      sprintf (buff, "smw.bi%s\t%%1, [%%0], %%1, %x", update_base, enable4);
++    }
+   else
+     {
+-      sprintf (buff, "smw.bi%s\t%%1, [%%0], %%1", update_base);
++      sprintf (buff, "smw.bi%s\t%%1, [%%0], %%2", update_base);
+     }
+   output_asm_insn (buff, otherops);
+   return "";
+@@ -2415,16 +2617,17 @@
+   if (mode == DImode)
+     {
+       /* Load doubleword, we need two registers to access.  */
+-      reg[0] = simplify_gen_subreg (SImode, operands[0],
+-				    GET_MODE (operands[0]), 0);
+-      reg[1] = simplify_gen_subreg (SImode, operands[0],
+-				    GET_MODE (operands[0]), 4);
++      reg[0] = nds32_di_low_part_subreg (operands[0]);
++      reg[1] = nds32_di_high_part_subreg (operands[0]);
+       /* A register only store 4 byte.  */
+       width = GET_MODE_SIZE (SImode) - 1;
+     }
+   else
+     {
+-      reg[0] = operands[0];
++      if (VECTOR_MODE_P (mode))
++	reg[0] = gen_reg_rtx (SImode);
++      else
++	reg[0] = operands[0];
+     }
+ 
+   for (num_reg = (mode == DImode) ? 2 : 1; num_reg > 0; num_reg--)
+@@ -2466,6 +2669,8 @@
+ 	  offset = offset + offset_adj;
+ 	}
+     }
++    if (VECTOR_MODE_P (mode))
++      convert_move (operands[0], reg[0], false);
+ }
+ 
+ void
+@@ -2499,16 +2704,20 @@
+   if (mode == DImode)
+     {
+       /* Load doubleword, we need two registers to access.  */
+-      reg[0] = simplify_gen_subreg (SImode, operands[1],
+-				    GET_MODE (operands[1]), 0);
+-      reg[1] = simplify_gen_subreg (SImode, operands[1],
+-				    GET_MODE (operands[1]), 4);
++      reg[0] = nds32_di_low_part_subreg (operands[1]);
++      reg[1] = nds32_di_high_part_subreg (operands[1]);
+       /* A register only store 4 byte.  */
+       width = GET_MODE_SIZE (SImode) - 1;
+     }
+   else
+     {
+-      reg[0] = operands[1];
++      if (VECTOR_MODE_P (mode))
++	{
++	  reg[0] = gen_reg_rtx (SImode);
++	  convert_move (reg[0], operands[1], false);
++	}
++      else
++	reg[0] = operands[1];
+     }
+ 
+   for (num_reg = (mode == DImode) ? 2 : 1; num_reg > 0; num_reg--)
+@@ -2765,6 +2974,36 @@
+   return "";
+ }
+ 
++const char *
++nds32_output_unpkd8 (rtx output, rtx input,
++		     rtx high_idx_rtx, rtx low_idx_rtx,
++		     bool signed_p)
++{
++  char pattern[100];
++  rtx output_operands[2];
++  HOST_WIDE_INT high_idx, low_idx;
++  high_idx = INTVAL (high_idx_rtx);
++  low_idx = INTVAL (low_idx_rtx);
++
++  gcc_assert (high_idx >= 0 && high_idx <= 3);
++  gcc_assert (low_idx >= 0 && low_idx <= 3);
++
++  /* We only have 10, 20, 30 and 31.  */
++  if ((low_idx != 0 || high_idx == 0) &&
++      !(low_idx == 1 && high_idx == 3))
++    return "#";
++
++  char sign_char = signed_p ? 's' : 'z';
++
++  sprintf (pattern,
++	   "%cunpkd8" HOST_WIDE_INT_PRINT_DEC HOST_WIDE_INT_PRINT_DEC "\t%%0, %%1",
++	   sign_char, high_idx, low_idx);
++  output_operands[0] = output;
++  output_operands[1] = input;
++  output_asm_insn (pattern, output_operands);
++  return "";
++}
++
+ /* Return true if SYMBOL_REF X binds locally.  */
+ 
+ static bool
+@@ -2782,22 +3021,15 @@
+   char pattern[100];
+   bool noreturn_p;
+ 
+-  if (GET_CODE (symbol) == CONST)
+-    {
+-      symbol= XEXP (symbol, 0);
+-
+-      if (GET_CODE (symbol) == PLUS)
+-        symbol = XEXP (symbol, 0);
+-    }
+-
+-  gcc_assert (GET_CODE (symbol) == SYMBOL_REF
+-	      || REG_P (symbol));
+-
+   if (nds32_long_call_p (symbol))
+     strcpy (pattern, long_call);
+   else
+     strcpy (pattern, call);
+ 
++  if (flag_pic && CONSTANT_P (symbol)
++      && !nds32_symbol_binds_local_p (symbol))
++    strcat (pattern, "@PLT");
++
+   if (align_p)
+     strcat (pattern, "\n\t.align 2");
+ 
+@@ -2815,6 +3047,91 @@
+   return "";
+ }
+ 
++bool
++nds32_need_split_sms_p (rtx in0_idx0, rtx in1_idx0,
++			rtx in0_idx1, rtx in1_idx1)
++{
++  /* smds or smdrs.  */
++  if (INTVAL (in0_idx0) == INTVAL (in1_idx0)
++      && INTVAL (in0_idx1) == INTVAL (in1_idx1)
++      && INTVAL (in0_idx0) != INTVAL (in0_idx1))
++    return false;
++
++  /* smxds.  */
++  if (INTVAL (in0_idx0) != INTVAL (in0_idx1)
++      && INTVAL (in1_idx0) != INTVAL (in1_idx1))
++    return false;
++
++  return true;
++}
++
++const char *
++nds32_output_sms (rtx in0_idx0, rtx in1_idx0,
++		  rtx in0_idx1, rtx in1_idx1)
++{
++  if (nds32_need_split_sms_p (in0_idx0, in1_idx0,
++			      in0_idx1, in1_idx1))
++    return "#";
++  /* out = in0[in0_idx0] * in1[in1_idx0] - in0[in0_idx1] * in1[in1_idx1] */
++
++  /* smds or smdrs.  */
++  if (INTVAL (in0_idx0) == INTVAL (in1_idx0)
++      && INTVAL (in0_idx1) == INTVAL (in1_idx1)
++      && INTVAL (in0_idx0) != INTVAL (in0_idx1))
++    {
++      if (INTVAL (in0_idx0) == 0)
++	{
++	  if (TARGET_BIG_ENDIAN)
++	    return "smds\t%0, %1, %2";
++	  else
++	    return "smdrs\t%0, %1, %2";
++	}
++      else
++	{
++	  if (TARGET_BIG_ENDIAN)
++	    return "smdrs\t%0, %1, %2";
++	  else
++	    return "smds\t%0, %1, %2";
++	}
++    }
++
++  if (INTVAL (in0_idx0) != INTVAL (in0_idx1)
++      && INTVAL (in1_idx0) != INTVAL (in1_idx1))
++    {
++      if (INTVAL (in0_idx0) == 1)
++	{
++	  if (TARGET_BIG_ENDIAN)
++	    return "smxds\t%0, %2, %1";
++	  else
++	    return "smxds\t%0, %1, %2";
++	}
++      else
++	{
++	  if (TARGET_BIG_ENDIAN)
++	    return "smxds\t%0, %1, %2";
++	  else
++	    return "smxds\t%0, %2, %1";
++	}
++    }
++
++  gcc_unreachable ();
++  return "";
++}
++
++void
++nds32_split_sms (rtx out, rtx in0, rtx in1,
++		 rtx in0_idx0, rtx in1_idx0,
++		 rtx in0_idx1, rtx in1_idx1)
++{
++  rtx result0 = gen_reg_rtx (SImode);
++  rtx result1 = gen_reg_rtx (SImode);
++  emit_insn (gen_mulhisi3v (result0, in0, in1,
++			    in0_idx0, in1_idx0));
++  emit_insn (gen_mulhisi3v (result1, in0, in1,
++			    in0_idx1, in1_idx1));
++  emit_insn (gen_subsi3 (out, result0, result1));
++}
++
+ /* Spilt a doubleword instrucion to two single word instructions.  */
+ void
+ nds32_spilt_doubleword (rtx *operands, bool load_p)
+@@ -2846,16 +3163,30 @@
+       /* generate low_part and high_part memory format:
+ 	   low_part:  (post_modify ((reg) (plus (reg) (const 4)))
+ 	   high_part: (post_modify ((reg) (plus (reg) (const -12))) */
+-      low_part[mem] = gen_frame_mem (SImode,
+-				     gen_rtx_POST_MODIFY (Pmode, sub_mem,
+-							  gen_rtx_PLUS (Pmode,
+-							  sub_mem,
+-							  GEN_INT (4))));
+-      high_part[mem] = gen_frame_mem (SImode,
+-				      gen_rtx_POST_MODIFY (Pmode, sub_mem,
+-							   gen_rtx_PLUS (Pmode,
+-							   sub_mem,
+-							   GEN_INT (-12))));
++      low_part[mem] = gen_rtx_MEM (SImode,
++				   gen_rtx_POST_MODIFY (Pmode, sub_mem,
++							gen_rtx_PLUS (Pmode,
++							sub_mem,
++							GEN_INT (4))));
++      high_part[mem] = gen_rtx_MEM (SImode,
++				    gen_rtx_POST_MODIFY (Pmode, sub_mem,
++							 gen_rtx_PLUS (Pmode,
++							 sub_mem,
++							 GEN_INT (-12))));
++    }
++  else if (GET_CODE (sub_mem) == POST_INC)
++    {
++      /* memory format is (post_inc (reg)),
++	 so that extract (reg) from the (post_inc (reg)) pattern.  */
++      sub_mem = XEXP (sub_mem, 0);
++
++      /* generate low_part and high_part memory format:
++	   low_part:  (post_inc (reg))
++	   high_part: (post_inc (reg)) */
++      low_part[mem] = gen_rtx_MEM (SImode,
++				   gen_rtx_POST_INC (Pmode, sub_mem));
++      high_part[mem] = gen_rtx_MEM (SImode,
++				    gen_rtx_POST_INC (Pmode, sub_mem));
+     }
+   else if (GET_CODE (sub_mem) == POST_MODIFY)
+     {
+@@ -2872,14 +3203,14 @@
+       /* Generate low_part and high_part memory format:
+ 	   low_part:  (post_modify ((reg) (plus (reg) (const)))
+ 	   high_part: ((plus (reg) (const 4))) */
+-      low_part[mem] = gen_frame_mem (SImode,
+-				     gen_rtx_POST_MODIFY (Pmode, post_mem,
+-							  gen_rtx_PLUS (Pmode,
+-							  post_mem,
+-							  post_val)));
+-      high_part[mem] = gen_frame_mem (SImode, plus_constant (Pmode,
+-							     post_mem,
+-							     4));
++      low_part[mem] = gen_rtx_MEM (SImode,
++				   gen_rtx_POST_MODIFY (Pmode, post_mem,
++							gen_rtx_PLUS (Pmode,
++							post_mem,
++							post_val)));
++      high_part[mem] = gen_rtx_MEM (SImode, plus_constant (Pmode,
++							   post_mem,
++							   4));
+     }
+   else
+     {
+@@ -2924,11 +3255,516 @@
+     }
+ }
+ 
++void
++nds32_split_ashiftdi3 (rtx dst, rtx src, rtx shiftamount)
++{
++  rtx src_high_part, src_low_part;
++  rtx dst_high_part, dst_low_part;
++
++  dst_high_part = nds32_di_high_part_subreg (dst);
++  dst_low_part = nds32_di_low_part_subreg (dst);
++
++  src_high_part = nds32_di_high_part_subreg (src);
++  src_low_part = nds32_di_low_part_subreg (src);
++
++  /* We need to handle shift more than 32 bit!!!! */
++  if (CONST_INT_P (shiftamount))
++    {
++      if (INTVAL (shiftamount) < 32)
++	{
++	  rtx ext_start;
++	  ext_start = gen_int_mode(32 - INTVAL (shiftamount), SImode);
++
++	  emit_insn (gen_wext (dst_high_part, src, ext_start));
++	  emit_insn (gen_ashlsi3 (dst_low_part, src_low_part, shiftamount));
++	}
++      else
++	{
++	  rtx new_shift_amout = gen_int_mode(INTVAL (shiftamount) - 32, SImode);
++
++	  emit_insn (gen_ashlsi3 (dst_high_part, src_low_part,
++						 new_shift_amout));
++
++	  emit_move_insn (dst_low_part, GEN_INT (0));
++	}
++    }
++  else
++    {
++      rtx dst_low_part_l32, dst_high_part_l32;
++      rtx dst_low_part_g32, dst_high_part_g32;
++      rtx new_shift_amout, select_reg;
++      dst_low_part_l32 = gen_reg_rtx (SImode);
++      dst_high_part_l32 = gen_reg_rtx (SImode);
++      dst_low_part_g32 = gen_reg_rtx (SImode);
++      dst_high_part_g32 = gen_reg_rtx (SImode);
++      new_shift_amout = gen_reg_rtx (SImode);
++      select_reg = gen_reg_rtx (SImode);
++
++      rtx ext_start;
++      ext_start = gen_reg_rtx (SImode);
++
++      /*
++	 if (shiftamount < 32)
++	   dst_low_part = src_low_part << shiftamout
++	   dst_high_part = wext (src, 32 - shiftamount)
++	   # wext can't handle wext (src, 32) since it's only take rb[0:4]
++	   # for extract.
++	   dst_high_part = shiftamount == 0 ? src_high_part : dst_high_part
++	 else
++	   dst_low_part = 0
++	   dst_high_part = src_low_part << shiftamount & 0x1f
++      */
++
++      emit_insn (gen_subsi3 (ext_start,
++			     gen_int_mode (32, SImode),
++			     shiftamount));
++      emit_insn (gen_wext (dst_high_part_l32, src, ext_start));
++
++      /* Handle for shiftamout == 0.  */
++      emit_insn (gen_cmovzsi (dst_high_part_l32, shiftamount,
++			      src_high_part, dst_high_part_l32));
++
++      emit_insn (gen_ashlsi3 (dst_low_part_l32, src_low_part, shiftamount));
++
++      emit_move_insn (dst_low_part_g32, const0_rtx);
++      emit_insn (gen_andsi3 (new_shift_amout, shiftamount, GEN_INT (0x1f)));
++      emit_insn (gen_ashlsi3 (dst_high_part_g32, src_low_part,
++						 new_shift_amout));
++
++      emit_insn (gen_slt_compare (select_reg, shiftamount, GEN_INT (32)));
++
++      emit_insn (gen_cmovnsi (dst_low_part, select_reg,
++			      dst_low_part_l32, dst_low_part_g32));
++      emit_insn (gen_cmovnsi (dst_high_part, select_reg,
++			      dst_high_part_l32, dst_high_part_g32));
++    }
++}
++
++void
++nds32_split_ashiftrtdi3 (rtx dst, rtx src, rtx shiftamount)
++{
++  nds32_split_shiftrtdi3 (dst, src, shiftamount, false);
++}
++
++void
++nds32_split_lshiftrtdi3 (rtx dst, rtx src, rtx shiftamount)
++{
++  nds32_split_shiftrtdi3 (dst, src, shiftamount, true);
++}
++
++void
++nds32_split_rotatertdi3 (rtx dst, rtx src, rtx shiftamount)
++{
++  rtx dst_low_part_l32, dst_high_part_l32;
++  rtx dst_low_part_g32, dst_high_part_g32;
++  rtx select_reg, low5bit, low5bit_inv, minus32sa;
++  rtx dst_low_part_g32_tmph;
++  rtx dst_low_part_g32_tmpl;
++  rtx dst_high_part_l32_tmph;
++  rtx dst_high_part_l32_tmpl;
++
++  rtx src_low_part, src_high_part;
++  rtx dst_high_part, dst_low_part;
++
++  shiftamount = force_reg (SImode, shiftamount);
++
++  emit_insn (gen_andsi3 (shiftamount,
++			 shiftamount,
++			 gen_int_mode (0x3f, SImode)));
++
++  dst_high_part = nds32_di_high_part_subreg (dst);
++  dst_low_part = nds32_di_low_part_subreg (dst);
++
++  src_high_part = nds32_di_high_part_subreg (src);
++  src_low_part = nds32_di_low_part_subreg (src);
++
++  dst_low_part_l32 = gen_reg_rtx (SImode);
++  dst_high_part_l32 = gen_reg_rtx (SImode);
++  dst_low_part_g32 = gen_reg_rtx (SImode);
++  dst_high_part_g32 = gen_reg_rtx (SImode);
++  low5bit = gen_reg_rtx (SImode);
++  low5bit_inv = gen_reg_rtx (SImode);
++  minus32sa = gen_reg_rtx (SImode);
++  select_reg = gen_reg_rtx (SImode);
++
++  dst_low_part_g32_tmph = gen_reg_rtx (SImode);
++  dst_low_part_g32_tmpl = gen_reg_rtx (SImode);
++
++  dst_high_part_l32_tmph = gen_reg_rtx (SImode);
++  dst_high_part_l32_tmpl = gen_reg_rtx (SImode);
++
++  emit_insn (gen_slt_compare (select_reg, shiftamount, GEN_INT (32)));
++
++  /* if shiftamount < 32
++       dst_low_part = wext(src, shiftamount)
++     else
++       dst_low_part = ((src_high_part >> (shiftamount & 0x1f))
++		       | (src_low_part << (32 - (shiftamount & 0x1f))))
++  */
++  emit_insn (gen_andsi3 (low5bit, shiftamount, gen_int_mode (0x1f, SImode)));
++  emit_insn (gen_subsi3 (low5bit_inv, gen_int_mode (32, SImode), low5bit));
++
++  emit_insn (gen_wext (dst_low_part_l32, src, shiftamount));
++
++  emit_insn (gen_lshrsi3 (dst_low_part_g32_tmpl, src_high_part, low5bit));
++  emit_insn (gen_ashlsi3 (dst_low_part_g32_tmph, src_low_part, low5bit_inv));
++
++  emit_insn (gen_iorsi3 (dst_low_part_g32,
++			 dst_low_part_g32_tmpl,
++			 dst_low_part_g32_tmph));
++
++  emit_insn (gen_cmovnsi (dst_low_part, select_reg,
++			  dst_low_part_l32, dst_low_part_g32));
++
++  /* if shiftamount < 32
++       dst_high_part = ((src_high_part >> shiftamount)
++			| (src_low_part << (32 - shiftamount)))
++       dst_high_part = shiftamount == 0 ? src_high_part : dst_high_part
++     else
++       dst_high_part = wext(src, shiftamount & 0x1f)
++  */
++
++  emit_insn (gen_subsi3 (minus32sa, gen_int_mode (32, SImode), shiftamount));
++
++  emit_insn (gen_lshrsi3 (dst_high_part_l32_tmpl, src_high_part, shiftamount));
++  emit_insn (gen_ashlsi3 (dst_high_part_l32_tmph, src_low_part, minus32sa));
++
++  emit_insn (gen_iorsi3 (dst_high_part_l32,
++			 dst_high_part_l32_tmpl,
++			 dst_high_part_l32_tmph));
++
++  emit_insn (gen_cmovzsi (dst_high_part_l32, shiftamount,
++			  src_high_part, dst_high_part_l32));
++
++  emit_insn (gen_wext (dst_high_part_g32, src, low5bit));
++
++  emit_insn (gen_cmovnsi (dst_high_part, select_reg,
++			  dst_high_part_l32, dst_high_part_g32));
++}
++
++/* Return true if OP contains a symbol reference.  */
++bool
++symbolic_reference_mentioned_p (rtx op)
++{
++  const char *fmt;
++  int i;
++
++  if (GET_CODE (op) == SYMBOL_REF || GET_CODE (op) == LABEL_REF)
++    return true;
++
++  fmt = GET_RTX_FORMAT (GET_CODE (op));
++  for (i = GET_RTX_LENGTH (GET_CODE (op)) - 1; i >= 0; i--)
++    {
++      if (fmt[i] == 'E')
++	{
++	  int j;
++
++	  for (j = XVECLEN (op, i) - 1; j >= 0; j--)
++	    if (symbolic_reference_mentioned_p (XVECEXP (op, i, j)))
++	      return true;
++	}
++
++      else if (fmt[i] == 'e' && symbolic_reference_mentioned_p (XEXP (op, i)))
++	return true;
++    }
++
++  return false;
++}
++
++/* Expand PIC code for @GOTOFF and @GOT.
++
++  Example for @GOTOFF:
++
++    la $r0, symbol@GOTOFF
++      -> sethi $ta, hi20(symbol@GOTOFF)
++	 ori $ta, $ta, lo12(symbol@GOTOFF)
++	 add $r0, $ta, $gp
++
++  Example for @GOT:
++
++    la $r0, symbol@GOT
++      -> sethi $ta, hi20(symbol@GOT)
++	 ori $ta, $ta, lo12(symbol@GOT)
++	 lw  $r0, [$ta + $gp]
++*/
++rtx
++nds32_legitimize_pic_address (rtx x)
++{
++  rtx addr = x;
++  rtx reg = gen_reg_rtx (Pmode);
++  rtx pat;
++
++  if (GET_CODE (x) == LABEL_REF
++      || (GET_CODE (x) == SYMBOL_REF
++	  && (CONSTANT_POOL_ADDRESS_P (x)
++	      || SYMBOL_REF_LOCAL_P (x))))
++    {
++      addr = gen_rtx_UNSPEC (SImode, gen_rtvec (1, x), UNSPEC_GOTOFF);
++      addr = gen_rtx_CONST (SImode, addr);
++      emit_insn (gen_sethi (reg, addr));
++      emit_insn (gen_lo_sum (reg, reg, addr));
++      x = gen_rtx_PLUS (Pmode, reg, pic_offset_table_rtx);
++    }
++  else if (GET_CODE (x) == SYMBOL_REF)
++    {
++      addr = gen_rtx_UNSPEC (SImode, gen_rtvec (1, x), UNSPEC_GOT);
++      addr = gen_rtx_CONST (SImode, addr);
++      emit_insn (gen_sethi (reg, addr));
++      emit_insn (gen_lo_sum (reg, reg, addr));
++
++      x = gen_const_mem (SImode, gen_rtx_PLUS (Pmode, pic_offset_table_rtx,
++					       reg));
++    }
++  else if (GET_CODE (x) == CONST)
++    {
++      /* We don't split constant in expand_pic_move because GOTOFF can combine
++	 the addend with the symbol.  */
++      addr = XEXP (x, 0);
++      gcc_assert (GET_CODE (addr) == PLUS);
++
++      rtx op0 = XEXP (addr, 0);
++      rtx op1 = XEXP (addr, 1);
++
++      if ((GET_CODE (op0) == LABEL_REF
++	   || (GET_CODE (op0) == SYMBOL_REF
++	       && (CONSTANT_POOL_ADDRESS_P (op0)
++		   || SYMBOL_REF_LOCAL_P (op0))))
++	  && GET_CODE (op1) == CONST_INT)
++	{
++	  pat = gen_rtx_UNSPEC (Pmode, gen_rtvec (1, op0), UNSPEC_GOTOFF);
++	  pat = gen_rtx_PLUS (Pmode, pat, op1);
++	  pat = gen_rtx_CONST (Pmode, pat);
++	  emit_insn (gen_sethi (reg, pat));
++	  emit_insn (gen_lo_sum (reg, reg, pat));
++	  x = gen_rtx_PLUS (Pmode, reg, pic_offset_table_rtx);
++	}
++      else if (GET_CODE (op0) == SYMBOL_REF
++	       && GET_CODE (op1) == CONST_INT)
++	{
++	  /* This is a constant offset from a @GOT symbol reference.  */
++	  addr = gen_rtx_UNSPEC (SImode, gen_rtvec (1, op0), UNSPEC_GOT);
++	  addr = gen_rtx_CONST (SImode, addr);
++	  emit_insn (gen_sethi (reg, addr));
++	  emit_insn (gen_lo_sum (reg, reg, addr));
++	  addr = gen_const_mem (SImode, gen_rtx_PLUS (Pmode,
++						      pic_offset_table_rtx,
++						      reg));
++	  emit_move_insn (reg, addr);
++	  if (satisfies_constraint_Is15 (op1))
++	    x = gen_rtx_PLUS (Pmode, reg, op1);
++	  else
++	    {
++	      rtx tmp_reg = gen_reg_rtx (SImode);
++	      emit_insn (gen_movsi (tmp_reg, op1));
++	      x = gen_rtx_PLUS (Pmode, reg, tmp_reg);
++	    }
++	}
++      else
++	{
++	  /* Don't handle this pattern.  */
++	  debug_rtx (x);
++	  gcc_unreachable ();
++	}
++    }
++  return x;
++}
++
++void
++nds32_expand_pic_move (rtx *operands)
++{
++  rtx src;
++
++  src = nds32_legitimize_pic_address (operands[1]);
++  emit_move_insn (operands[0], src);
++}
++
++/* Expand ICT symbol.
++    Example for @ICT and ICT model=large:
++
++    la $r0, symbol@ICT
++      -> sethi $rt, hi20(symbol@ICT)
++	 lwi $r0, [$rt + lo12(symbol@ICT)]
++
++*/
++rtx
++nds32_legitimize_ict_address (rtx x)
++{
++  rtx symbol = x;
++  rtx addr = x;
++  rtx reg = gen_reg_rtx (Pmode);
++  gcc_assert (GET_CODE (x) == SYMBOL_REF
++	      && nds32_indirect_call_referenced_p (x));
++
++  addr = gen_rtx_UNSPEC (SImode, gen_rtvec (1, symbol), UNSPEC_ICT);
++  addr = gen_rtx_CONST (SImode, addr);
++  emit_insn (gen_sethi (reg, addr));
++
++  x = gen_const_mem (SImode, gen_rtx_LO_SUM (Pmode, reg, addr));
++
++  return x;
++}
++
++void
++nds32_expand_ict_move (rtx *operands)
++{
++  rtx src = operands[1];
++
++  src = nds32_legitimize_ict_address (src);
++
++  emit_move_insn (operands[0], src);
++}
++
++/* Return true X is a indirect call symbol.  */
++bool
++nds32_indirect_call_referenced_p (rtx x)
++{
++  if (GET_CODE (x) == UNSPEC && XINT (x, 1) == UNSPEC_ICT)
++    x = XVECEXP (x, 0, 0);
++
++  if (GET_CODE (x) == SYMBOL_REF)
++    {
++      tree decl = SYMBOL_REF_DECL (x);
++
++      return decl
++	     && (lookup_attribute("indirect_call",
++				  DECL_ATTRIBUTES(decl))
++		 != NULL);
++    }
++
++  return false;
++}
++
+ /* Return true X is need use long call.  */
+ bool
+ nds32_long_call_p (rtx symbol)
+ {
+-  return TARGET_CMODEL_LARGE;
++  if (nds32_indirect_call_referenced_p (symbol))
++    return TARGET_ICT_MODEL_LARGE;
++  else
++    return TARGET_CMODEL_LARGE;
++}
++
++/* Return true if X contains a thread-local symbol.  */
++bool
++nds32_tls_referenced_p (rtx x)
++{
++  if (!targetm.have_tls)
++   return false;
++
++  if (GET_CODE (x) == CONST && GET_CODE (XEXP (x, 0)) == PLUS)
++    x = XEXP (XEXP (x, 0), 0);
++
++  if (GET_CODE (x) == SYMBOL_REF && SYMBOL_REF_TLS_MODEL (x))
++    return true;
++
++  return false;
++}
++
++/* ADDR contains a thread-local SYMBOL_REF.  Generate code to compute
++   this (thread-local) address.  */
++rtx
++nds32_legitimize_tls_address (rtx x)
++{
++  rtx tmp_reg;
++  rtx tp_reg = gen_rtx_REG (Pmode, TP_REGNUM);
++  rtx pat, insns, reg0;
++
++  if (GET_CODE (x) == SYMBOL_REF)
++    switch (SYMBOL_REF_TLS_MODEL (x))
++      {
++      case TLS_MODEL_GLOBAL_DYNAMIC:
++      case TLS_MODEL_LOCAL_DYNAMIC:
++	/* Emit UNSPEC_TLS_DESC rather than expand rtl directly because spill
++	   may destroy the define-use chain anylysis to insert relax_hint.  */
++	if (SYMBOL_REF_TLS_MODEL (x) == TLS_MODEL_GLOBAL_DYNAMIC)
++	  pat = gen_rtx_UNSPEC (SImode, gen_rtvec (1, x), UNSPEC_TLSGD);
++	else
++	  pat = gen_rtx_UNSPEC (SImode, gen_rtvec (1, x), UNSPEC_TLSLD);
++
++	pat = gen_rtx_CONST (SImode, pat);
++	reg0 = gen_rtx_REG (Pmode, 0);
++	/* If we can confirm all clobber reigsters, it doesn't have to use call
++	   instruction.  */
++	insns = emit_call_insn (gen_tls_desc (pat, GEN_INT (0)));
++	use_reg (&CALL_INSN_FUNCTION_USAGE (insns), pic_offset_table_rtx);
++	RTL_CONST_CALL_P (insns) = 1;
++	tmp_reg = gen_reg_rtx (SImode);
++	emit_move_insn (tmp_reg, reg0);
++	x = tmp_reg;
++	break;
++
++      case TLS_MODEL_INITIAL_EXEC:
++	pat = gen_rtx_UNSPEC (SImode, gen_rtvec (1, x), UNSPEC_TLSIE);
++	tmp_reg  = gen_reg_rtx (SImode);
++	pat = gen_rtx_CONST (SImode, pat);
++	emit_insn (gen_tls_ie (tmp_reg, pat, GEN_INT (0)));
++	if (flag_pic)
++	  emit_use (pic_offset_table_rtx);
++	x = gen_rtx_PLUS (Pmode, tmp_reg, tp_reg);
++	break;
++
++      case TLS_MODEL_LOCAL_EXEC:
++	/* Expand symbol_ref@TPOFF':
++	     sethi $ta, hi20(symbol_ref@TPOFF)
++	     ori   $ta, $ta, lo12(symbol_ref@TPOFF)
++	     add   $r0, $ta, $tp */
++	tmp_reg  = gen_reg_rtx (SImode);
++	pat = gen_rtx_UNSPEC (SImode, gen_rtvec (1, x), UNSPEC_TLSLE);
++	pat = gen_rtx_CONST (SImode, pat);
++	emit_insn (gen_sethi (tmp_reg, pat));
++	emit_insn (gen_lo_sum (tmp_reg, tmp_reg, pat));
++	x = gen_rtx_PLUS (Pmode, tmp_reg, tp_reg);
++	break;
++
++      default:
++	gcc_unreachable ();
++      }
++  else if (GET_CODE (x) == CONST)
++    {
++      rtx base, addend;
++      split_const (x, &base, &addend);
++
++      if (SYMBOL_REF_TLS_MODEL (base) == TLS_MODEL_LOCAL_EXEC)
++	{
++	  /* Expand symbol_ref@TPOFF':
++	     sethi $ta, hi20(symbol_ref@TPOFF + addend)
++	     ori   $ta, $ta, lo12(symbol_ref@TPOFF + addend)
++	     add   $r0, $ta, $tp */
++	  tmp_reg  = gen_reg_rtx (SImode);
++	  pat = gen_rtx_UNSPEC (SImode, gen_rtvec (1, base), UNSPEC_TLSLE);
++	  pat = gen_rtx_PLUS (SImode, pat, addend);
++	  pat = gen_rtx_CONST (SImode, pat);
++	  emit_insn (gen_sethi (tmp_reg, pat));
++	  emit_insn (gen_lo_sum (tmp_reg, tmp_reg, pat));
++	  x = gen_rtx_PLUS (Pmode, tmp_reg, tp_reg);
++	}
++    }
++
++  return x;
++}
++
++void
++nds32_expand_tls_move (rtx *operands)
++{
++  rtx src = operands[1];
++  rtx base, addend;
++
++  if (CONSTANT_P (src))
++    split_const (src, &base, &addend);
++
++  if (SYMBOL_REF_TLS_MODEL (base) == TLS_MODEL_LOCAL_EXEC)
++    src = nds32_legitimize_tls_address (src);
++  else
++    {
++      src = nds32_legitimize_tls_address (base);
++      if (addend != const0_rtx)
++	{
++	  src = gen_rtx_PLUS (SImode, src, addend);
++	  src = force_operand (src, operands[0]);
++	}
++    }
++
++  emit_move_insn (operands[0], src);
+ }
+ 
+ void
+@@ -2976,3 +3812,105 @@
+       emit_move_insn (target, gen_rtx_fmt_ee (AND, mode, source, temp));
+     }
+ }
++
++/* Auxiliary functions for lwm/smw.  */
++bool
++nds32_valid_smw_lwm_base_p (rtx op)
++{
++  rtx base_addr;
++
++  if (!MEM_P (op))
++    return false;
++
++  base_addr = XEXP (op, 0);
++
++  if (REG_P (base_addr))
++    return true;
++  else
++    {
++      if (GET_CODE (base_addr) == POST_INC
++	  && REG_P (XEXP (base_addr, 0)))
++	return true;
++    }
++
++  return false;
++}
++
++/* Auxiliary functions for manipulation DI mode.  */
++rtx nds32_di_high_part_subreg(rtx reg)
++{
++  unsigned high_part_offset = subreg_highpart_offset (SImode, DImode);
++
++  return simplify_gen_subreg (
++	   SImode, reg,
++	   DImode, high_part_offset);
++}
++
++rtx nds32_di_low_part_subreg(rtx reg)
++{
++  unsigned low_part_offset = subreg_lowpart_offset (SImode, DImode);
++
++  return simplify_gen_subreg (
++	   SImode, reg,
++	   DImode, low_part_offset);
++}
++
++/* ------------------------------------------------------------------------ */
++
++/* Auxiliary function for output TLS patterns.  */
++
++const char *
++nds32_output_tls_desc (rtx *operands)
++{
++  char pattern[1000];
++
++  if (TARGET_RELAX_HINT)
++    snprintf (pattern, sizeof (pattern),
++	      ".relax_hint %%1\n\tsethi $r0, hi20(%%0)\n\t"
++	      ".relax_hint %%1\n\tori $r0, $r0, lo12(%%0)\n\t"
++	      ".relax_hint %%1\n\tlw $r15, [$r0 + $gp]\n\t"
++	      ".relax_hint %%1\n\tadd $r0, $r0, $gp\n\t"
++	      ".relax_hint %%1\n\tjral $r15");
++  else
++    snprintf (pattern, sizeof (pattern),
++	      "sethi $r0, hi20(%%0)\n\t"
++	      "ori $r0, $r0, lo12(%%0)\n\t"
++	      "lw $r15, [$r0 + $gp]\n\t"
++	      "add $r0, $r0, $gp\n\t"
++	      "jral $r15");
++  output_asm_insn (pattern, operands);
++  return "";
++}
++
++const char *
++nds32_output_tls_ie (rtx *operands)
++{
++  char pattern[1000];
++
++  if (flag_pic)
++  {
++      if (TARGET_RELAX_HINT)
++	snprintf (pattern, sizeof (pattern),
++		  ".relax_hint %%2\n\tsethi %%0, hi20(%%1)\n\t"
++		  ".relax_hint %%2\n\tori %%0, %%0, lo12(%%1)\n\t"
++		  ".relax_hint %%2\n\tlw %%0, [%%0 + $gp]");
++      else
++	snprintf (pattern, sizeof (pattern),
++		  "sethi %%0, hi20(%%1)\n\t"
++		  "ori %%0, %%0, lo12(%%1)\n\t"
++		  "lw %%0, [%%0 + $gp]");
++  }
++  else
++    {
++      if (TARGET_RELAX_HINT)
++	snprintf (pattern, sizeof (pattern),
++		  ".relax_hint %%2\n\tsethi %%0, hi20(%%1)\n\t"
++		  ".relax_hint %%2\n\tlwi %%0, [%%0 + lo12(%%1)]");
++      else
++	snprintf (pattern, sizeof (pattern),
++		  "sethi %%0, hi20(%%1)\n\t"
++		  "lwi %%0, [%%0 + lo12(%%1)]");
++    }
++  output_asm_insn (pattern, operands);
++  return "";
++}
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-memory-manipulation.c gcc-8.2.0/gcc/config/nds32/nds32-memory-manipulation.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-memory-manipulation.c	2018-03-11 09:24:33.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-memory-manipulation.c	2019-01-25 15:38:32.829242659 +0100
+@@ -257,8 +257,124 @@
+ nds32_expand_movmemsi_loop_known_size (rtx dstmem, rtx srcmem,
+ 				       rtx size, rtx alignment)
+ {
+-  return nds32_expand_movmemsi_loop_unknown_size (dstmem, srcmem,
+-						  size, alignment);
++  rtx dst_base_reg, src_base_reg;
++  rtx dst_itr, src_itr;
++  rtx dstmem_m, srcmem_m, dst_itr_m, src_itr_m;
++  rtx dst_end;
++  rtx double_word_mode_loop, byte_mode_loop;
++  rtx tmp;
++  int start_regno;
++  bool align_to_4_bytes = (INTVAL (alignment) & 3) == 0;
++  unsigned HOST_WIDE_INT total_bytes = UINTVAL (size);
++
++  if (TARGET_ISA_V3M && !align_to_4_bytes)
++    return 0;
++
++  if (TARGET_REDUCED_REGS)
++    start_regno = 2;
++  else
++    start_regno = 16;
++
++  dst_itr = gen_reg_rtx (Pmode);
++  src_itr = gen_reg_rtx (Pmode);
++  dst_end = gen_reg_rtx (Pmode);
++  tmp = gen_reg_rtx (QImode);
++
++  double_word_mode_loop = gen_label_rtx ();
++  byte_mode_loop = gen_label_rtx ();
++
++  dst_base_reg = copy_to_mode_reg (Pmode, XEXP (dstmem, 0));
++  src_base_reg = copy_to_mode_reg (Pmode, XEXP (srcmem, 0));
++
++  if (total_bytes < 8)
++    {
++      /* Emit total_bytes less than 8 loop version of movmem.
++	add     $dst_end, $dst, $size
++	move    $dst_itr, $dst
++	.Lbyte_mode_loop:
++	lbi.bi  $tmp, [$src_itr], #1
++	sbi.bi  $tmp, [$dst_itr], #1
++	! Not readch upper bound. Loop.
++	bne     $dst_itr, $dst_end, .Lbyte_mode_loop */
++
++      /* add     $dst_end, $dst, $size */
++      dst_end = expand_binop (Pmode, add_optab, dst_base_reg, size,
++			      NULL_RTX, 0, OPTAB_WIDEN);
++      /* move    $dst_itr, $dst
++	 move    $src_itr, $src */
++      emit_move_insn (dst_itr, dst_base_reg);
++      emit_move_insn (src_itr, src_base_reg);
++
++      /* .Lbyte_mode_loop: */
++      emit_label (byte_mode_loop);
++
++      /* lbi.bi  $tmp, [$src_itr], #1 */
++      nds32_emit_post_inc_load_store (tmp, src_itr, QImode, true);
++
++      /* sbi.bi  $tmp, [$dst_itr], #1 */
++      nds32_emit_post_inc_load_store (tmp, dst_itr, QImode, false);
++      /* ! Not readch upper bound. Loop.
++	 bne     $dst_itr, $dst_end, .Lbyte_mode_loop */
++      emit_cmp_and_jump_insns (dst_itr, dst_end, NE, NULL,
++			       SImode, 1, byte_mode_loop);
++      return true;
++    }
++  else if (total_bytes % 8 == 0)
++    {
++      /* Emit multiple of 8 loop version of movmem.
++
++	 add     $dst_end, $dst, $size
++	 move    $dst_itr, $dst
++	 move    $src_itr, $src
++
++	.Ldouble_word_mode_loop:
++	lmw.bim $tmp-begin, [$src_itr], $tmp-end, #0 ! $src_itr' = $src_itr
++	smw.bim $tmp-begin, [$dst_itr], $tmp-end, #0 ! $dst_itr' = $dst_itr
++	! move will delete after register allocation
++	move    $src_itr, $src_itr'
++	move    $dst_itr, $dst_itr'
++	! Not readch upper bound. Loop.
++	bne     $double_word_end, $dst_itr, .Ldouble_word_mode_loop */
++
++      /* add     $dst_end, $dst, $size */
++      dst_end = expand_binop (Pmode, add_optab, dst_base_reg, size,
++			      NULL_RTX, 0, OPTAB_WIDEN);
++
++      /* move    $dst_itr, $dst
++	 move    $src_itr, $src */
++      emit_move_insn (dst_itr, dst_base_reg);
++      emit_move_insn (src_itr, src_base_reg);
++
++      /* .Ldouble_word_mode_loop: */
++      emit_label (double_word_mode_loop);
++      /* lmw.bim $tmp-begin, [$src_itr], $tmp-end, #0 ! $src_itr' = $src_itr
++	 smw.bim $tmp-begin, [$dst_itr], $tmp-end, #0 ! $dst_itr' = $dst_itr */
++      src_itr_m = src_itr;
++      dst_itr_m = dst_itr;
++      srcmem_m = srcmem;
++      dstmem_m = dstmem;
++      nds32_emit_mem_move_block (start_regno, 2,
++				 &dst_itr_m, &dstmem_m,
++				 &src_itr_m, &srcmem_m,
++				 true);
++      /* move    $src_itr, $src_itr'
++	 move    $dst_itr, $dst_itr' */
++      emit_move_insn (dst_itr, dst_itr_m);
++      emit_move_insn (src_itr, src_itr_m);
++
++      /* ! Not readch upper bound. Loop.
++	 bne     $double_word_end, $dst_itr, .Ldouble_word_mode_loop */
++      emit_cmp_and_jump_insns (dst_end, dst_itr, NE, NULL,
++			       Pmode, 1, double_word_mode_loop);
++    }
++  else
++    {
++      /* Handle size greater than 8, and not a multiple of 8.  */
++      return nds32_expand_movmemsi_loop_unknown_size (dstmem, srcmem,
++						      size, alignment);
++    }
++
++  return true;
+ }
+ 
+ static bool
+@@ -433,10 +549,8 @@
+ /* Auxiliary function for expand setmem pattern.  */
+ 
+ static rtx
+-nds32_gen_dup_4_byte_to_word_value (rtx value)
++nds32_gen_dup_4_byte_to_word_value_aux (rtx value, rtx value4word)
+ {
+-  rtx value4word = gen_reg_rtx (SImode);
+-
+   gcc_assert (GET_MODE (value) == QImode || CONST_INT_P (value));
+ 
+   if (CONST_INT_P (value))
+@@ -449,36 +563,74 @@
+     }
+   else
+     {
+-      /* ! prepare word
+-	 andi    $tmp1, $value, 0xff       ! $tmp1  <- 0x000000ab
+-	 slli    $tmp2, $tmp1, 8           ! $tmp2  <- 0x0000ab00
+-	 or      $tmp3, $tmp1, $tmp2       ! $tmp3  <- 0x0000abab
+-	 slli    $tmp4, $tmp3, 16          ! $tmp4  <- 0xabab0000
+-	 or      $val4word, $tmp3, $tmp4   ! $value4word  <- 0xabababab  */
+-
+-      rtx tmp1, tmp2, tmp3, tmp4, final_value;
+-      tmp1 = expand_binop (SImode, and_optab, value,
+-			   gen_int_mode (0xff, SImode),
+-			   NULL_RTX, 0, OPTAB_WIDEN);
+-      tmp2 = expand_binop (SImode, ashl_optab, tmp1,
+-			   gen_int_mode (8, SImode),
+-			   NULL_RTX, 0, OPTAB_WIDEN);
+-      tmp3 = expand_binop (SImode, ior_optab, tmp1, tmp2,
+-			   NULL_RTX, 0, OPTAB_WIDEN);
+-      tmp4 = expand_binop (SImode, ashl_optab, tmp3,
+-			   gen_int_mode (16, SImode),
+-			   NULL_RTX, 0, OPTAB_WIDEN);
+-
+-      final_value = expand_binop (SImode, ior_optab, tmp3, tmp4,
+-				  NULL_RTX, 0, OPTAB_WIDEN);
+-      emit_move_insn (value4word, final_value);
++      if (NDS32_EXT_DSP_P ())
++	{
++	  /* ! prepare word
++	     insb    $tmp, $value, 1         ! $tmp  <- 0x0000abab
++	     pkbb16  $tmp6, $tmp2, $tmp2   ! $value4word  <- 0xabababab */
++	  rtx tmp = gen_reg_rtx (SImode);
++
++	  convert_move (tmp, value, true);
++
++	  emit_insn (
++	    gen_insvsi_internal (tmp, gen_int_mode (0x8, SImode), tmp));
++
++	  emit_insn (gen_pkbbsi_1 (value4word, tmp, tmp));
++	}
++      else
++	{
++	  /* ! prepare word
++	     andi    $tmp1, $value, 0xff       ! $tmp1  <- 0x000000ab
++	     slli    $tmp2, $tmp1, 8           ! $tmp2  <- 0x0000ab00
++	     or      $tmp3, $tmp1, $tmp2       ! $tmp3  <- 0x0000abab
++	     slli    $tmp4, $tmp3, 16          ! $tmp4  <- 0xabab0000
++	     or      $val4word, $tmp3, $tmp4   ! $value4word  <- 0xabababab  */
++
++	  rtx tmp1, tmp2, tmp3, tmp4;
++	  tmp1 = expand_binop (SImode, and_optab, value,
++			       gen_int_mode (0xff, SImode),
++			       NULL_RTX, 0, OPTAB_WIDEN);
++	  tmp2 = expand_binop (SImode, ashl_optab, tmp1,
++			       gen_int_mode (8, SImode),
++			       NULL_RTX, 0, OPTAB_WIDEN);
++	  tmp3 = expand_binop (SImode, ior_optab, tmp1, tmp2,
++			       NULL_RTX, 0, OPTAB_WIDEN);
++	  tmp4 = expand_binop (SImode, ashl_optab, tmp3,
++			       gen_int_mode (16, SImode),
++			       NULL_RTX, 0, OPTAB_WIDEN);
++
++	  emit_insn (gen_iorsi3 (value4word, tmp3, tmp4));
++	}
+     }
+ 
+   return value4word;
+ }
+ 
+ static rtx
+-emit_setmem_word_loop (rtx itr, rtx size, rtx value)
++nds32_gen_dup_4_byte_to_word_value (rtx value)
++{
++  rtx value4word = gen_reg_rtx (SImode);
++  nds32_gen_dup_4_byte_to_word_value_aux (value, value4word);
++
++  return value4word;
++}
++
++static rtx
++nds32_gen_dup_8_byte_to_double_word_value (rtx value)
++{
++  rtx value4doubleword = gen_reg_rtx (DImode);
++
++  nds32_gen_dup_4_byte_to_word_value_aux (
++    value, nds32_di_low_part_subreg(value4doubleword));
++
++  emit_move_insn (nds32_di_high_part_subreg(value4doubleword),
++		  nds32_di_low_part_subreg(value4doubleword));
++  return value4doubleword;
++}
++
++
++static rtx
++emit_setmem_doubleword_loop (rtx itr, rtx size, rtx value)
+ {
+   rtx word_mode_label = gen_label_rtx ();
+   rtx word_mode_end_label = gen_label_rtx ();
+@@ -487,9 +639,9 @@
+   rtx word_mode_end = gen_reg_rtx (SImode);
+   rtx size_for_word = gen_reg_rtx (SImode);
+ 
+-  /* and     $size_for_word, $size, #~3  */
++  /* and     $size_for_word, $size, #~0x7  */
+   size_for_word = expand_binop (SImode, and_optab, size,
+-				gen_int_mode (~3, SImode),
++				gen_int_mode (~0x7, SImode),
+ 				NULL_RTX, 0, OPTAB_WIDEN);
+ 
+   emit_move_insn (byte_mode_size, size);
+@@ -501,8 +653,8 @@
+   word_mode_end = expand_binop (Pmode, add_optab, itr, size_for_word,
+ 				NULL_RTX, 0, OPTAB_WIDEN);
+ 
+-  /* andi    $byte_mode_size, $size, 3  */
+-  byte_mode_size_tmp = expand_binop (SImode, and_optab, size, GEN_INT (3),
++  /* andi    $byte_mode_size, $size, 0x7  */
++  byte_mode_size_tmp = expand_binop (SImode, and_optab, size, GEN_INT (0x7),
+ 				     NULL_RTX, 0, OPTAB_WIDEN);
+ 
+   emit_move_insn (byte_mode_size, byte_mode_size_tmp);
+@@ -512,9 +664,9 @@
+   /*   ! word-mode set loop
+        smw.bim $value4word, [$dst_itr], $value4word, 0
+        bne     $word_mode_end, $dst_itr, .Lword_mode  */
+-  emit_insn (gen_unaligned_store_update_base_w (itr,
+-						itr,
+-						value));
++  emit_insn (gen_unaligned_store_update_base_dw (itr,
++						 itr,
++						 value));
+   emit_cmp_and_jump_insns (word_mode_end, itr, NE, NULL,
+ 			   Pmode, 1, word_mode_label);
+ 
+@@ -566,7 +718,7 @@
+ static bool
+ nds32_expand_setmem_loop (rtx dstmem, rtx size, rtx value)
+ {
+-  rtx value4word;
++  rtx value4doubleword;
+   rtx value4byte;
+   rtx dst;
+   rtx byte_mode_size;
+@@ -609,7 +761,7 @@
+      or      $tmp3, $tmp1, $tmp2             ! $tmp3  <- 0x0000abab
+      slli    $tmp4, $tmp3, 16                ! $tmp4  <- 0xabab0000
+      or      $val4word, $tmp3, $tmp4         ! $value4word  <- 0xabababab  */
+-  value4word = nds32_gen_dup_4_byte_to_word_value (value);
++  value4doubleword = nds32_gen_dup_8_byte_to_double_word_value (value);
+ 
+   /*   and     $size_for_word, $size, #-4
+        beqz    $size_for_word, .Lword_mode_end
+@@ -622,7 +774,7 @@
+        smw.bim $value4word, [$dst], $value4word, 0
+        bne     $word_mode_end, $dst, .Lword_mode
+      .Lword_mode_end:  */
+-  byte_mode_size = emit_setmem_word_loop (dst, size, value4word);
++  byte_mode_size = emit_setmem_doubleword_loop (dst, size, value4doubleword);
+ 
+   /*   beqz    $byte_mode_size, .Lend
+        add     $byte_mode_end, $dst, $byte_mode_size
+@@ -633,8 +785,8 @@
+        bne     $byte_mode_end, $dst, .Lbyte_mode
+      .Lend: */
+ 
+-  value4byte = simplify_gen_subreg (QImode, value4word, SImode,
+-				    subreg_lowpart_offset (QImode, SImode));
++  value4byte = simplify_gen_subreg (QImode, value4doubleword, DImode,
++				    subreg_lowpart_offset (QImode, DImode));
+ 
+   emit_setmem_byte_loop (dst, byte_mode_size, value4byte, false);
+ 
+@@ -651,14 +803,15 @@
+   rtx byte_loop_size = gen_reg_rtx (SImode);
+   rtx remain_size = gen_reg_rtx (SImode);
+   rtx new_base_reg;
+-  rtx value4byte, value4word;
++  rtx value4byte, value4doubleword;
+   rtx byte_mode_size;
+   rtx last_byte_loop_label = gen_label_rtx ();
+ 
+   size = force_reg (SImode, size);
+ 
+-  value4word = nds32_gen_dup_4_byte_to_word_value (value);
+-  value4byte = simplify_gen_subreg (QImode, value4word, SImode, 0);
++  value4doubleword = nds32_gen_dup_8_byte_to_double_word_value (value);
++  value4byte = simplify_gen_subreg (QImode, value4doubleword, DImode,
++				    subreg_lowpart_offset (QImode, DImode));
+ 
+   emit_move_insn (byte_loop_size, size);
+   emit_move_insn (byte_loop_base, base_reg);
+@@ -686,9 +839,9 @@
+   emit_insn (gen_subsi3 (remain_size, size, need_align_bytes));
+ 
+   /* Set memory word by word. */
+-  byte_mode_size = emit_setmem_word_loop (new_base_reg,
+-					  remain_size,
+-					  value4word);
++  byte_mode_size = emit_setmem_doubleword_loop (new_base_reg,
++						remain_size,
++						value4doubleword);
+ 
+   emit_move_insn (byte_loop_base, new_base_reg);
+   emit_move_insn (byte_loop_size, byte_mode_size);
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-multiple.md gcc-8.2.0/gcc/config/nds32/nds32-multiple.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-multiple.md	2018-03-11 09:24:33.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-multiple.md	2019-01-25 15:38:32.829242659 +0100
+@@ -2854,6 +2854,25 @@
+    (set_attr "length"              "4")]
+ )
+ 
++(define_expand "unaligned_store_update_base_dw"
++  [(parallel [(set (match_operand:SI 0 "register_operand" "=r")
++		   (plus:SI (match_operand:SI 1 "register_operand" "0") (const_int 8)))
++	      (set (mem:DI (match_dup 1))
++		   (unspec:DI [(match_operand:DI 2 "register_operand" "r")] UNSPEC_UASTORE_DW))])]
++  ""
++{
++  /* DO NOT emit unaligned_store_w_m immediately since web pass don't
++     recognize post_inc, try it again after GCC 5.0.
++     REF: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63156  */
++  emit_insn (gen_unaligned_store_dw (gen_rtx_MEM (DImode, operands[1]), operands[2]));
++  emit_insn (gen_addsi3 (operands[0], operands[1], gen_int_mode (8, Pmode)));
++  DONE;
++}
++  [(set_attr "type"   "store_multiple")
++   (set_attr "combo"               "2")
++   (set_attr "length"              "4")]
++)
++
+ (define_insn "*stmsi25"
+   [(match_parallel 0 "nds32_store_multiple_operation"
+     [(set (mem:SI (match_operand:SI 1 "register_operand" "r"))
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-n10.md gcc-8.2.0/gcc/config/nds32/nds32-n10.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-n10.md	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-n10.md	2019-01-25 15:38:32.829242659 +0100
+@@ -0,0 +1,439 @@
++;; Pipeline descriptions of Andes NDS32 cpu for GNU compiler
++;; Copyright (C) 2012-2018 Free Software Foundation, Inc.
++;; Contributed by Andes Technology Corporation.
++;;
++;; This file is part of GCC.
++;;
++;; GCC is free software; you can redistribute it and/or modify it
++;; under the terms of the GNU General Public License as published
++;; by the Free Software Foundation; either version 3, or (at your
++;; option) any later version.
++;;
++;; GCC is distributed in the hope that it will be useful, but WITHOUT
++;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++;; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++;; License for more details.
++;;
++;; You should have received a copy of the GNU General Public License
++;; along with GCC; see the file COPYING3.  If not see
++;; <http://www.gnu.org/licenses/>.
++
++
++;; ------------------------------------------------------------------------
++;; Define N10 pipeline settings.
++;; ------------------------------------------------------------------------
++
++(define_automaton "nds32_n10_machine")
++
++;; ------------------------------------------------------------------------
++;; Pipeline Stages
++;; ------------------------------------------------------------------------
++;; IF - Instruction Fetch
++;; II - Instruction Issue / Instruction Decode
++;; EX - Instruction Execution
++;; MM - Memory Execution
++;; WB - Instruction Retire / Result Write-Back
++
++(define_cpu_unit "n10_ii" "nds32_n10_machine")
++(define_cpu_unit "n10_ex" "nds32_n10_machine")
++(define_cpu_unit "n10_mm" "nds32_n10_machine")
++(define_cpu_unit "n10_wb" "nds32_n10_machine")
++(define_cpu_unit "n10f_iq" "nds32_n10_machine")
++(define_cpu_unit "n10f_rf" "nds32_n10_machine")
++(define_cpu_unit "n10f_e1" "nds32_n10_machine")
++(define_cpu_unit "n10f_e2" "nds32_n10_machine")
++(define_cpu_unit "n10f_e3" "nds32_n10_machine")
++(define_cpu_unit "n10f_e4" "nds32_n10_machine")
++
++(define_insn_reservation "nds_n10_unknown" 1
++  (and (eq_attr "type" "unknown")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_misc" 1
++  (and (eq_attr "type" "misc")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_mmu" 1
++  (and (eq_attr "type" "mmu")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_alu" 1
++  (and (eq_attr "type" "alu")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_alu_shift" 1
++  (and (eq_attr "type" "alu_shift")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ii+n10_ex, n10_ex+n10_mm, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_pbsad" 1
++  (and (eq_attr "type" "pbsad")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex*3, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_pbsada" 1
++  (and (eq_attr "type" "pbsada")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex*3, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_load" 1
++  (and (match_test "nds32::load_single_p (insn)")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_store" 1
++  (and (match_test "nds32::store_single_p (insn)")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_1" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "1")))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_2" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (ior (and (eq_attr "type" "load_multiple")
++		 (eq_attr "combo" "2"))
++	    (match_test "nds32::load_double_p (insn)")))
++  "n10_ii, n10_ii+n10_ex, n10_ex+n10_mm, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_3" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "3")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_4" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "4")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, n10_ii+n10_ex+n10_mm+n10_wb, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_5" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "5")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*2, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_6" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "6")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*3, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_7" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "7")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*4, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_load_multiple_N" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "load_multiple")
++	    (match_test "get_attr_combo (insn) >= 8")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*5, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_1" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "1")))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_2" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (ior (and (eq_attr "type" "store_multiple")
++		 (eq_attr "combo" "2"))
++       (match_test "nds32::store_double_p (insn)")))
++  "n10_ii, n10_ii+n10_ex, n10_ex+n10_mm, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_3" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "3")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_4" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "4")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, n10_ii+n10_ex+n10_mm+n10_wb, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_5" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "5")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*2, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_6" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "6")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*3, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_7" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "7")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*4, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_store_multiple_N" 1
++  (and (eq_attr "pipeline_model" "n10")
++       (and (eq_attr "type" "store_multiple")
++	    (match_test "get_attr_combo (insn) >= 8")))
++  "n10_ii, n10_ii+n10_ex, n10_ii+n10_ex+n10_mm, (n10_ii+n10_ex+n10_mm+n10_wb)*5, n10_ex+n10_mm+n10_wb, n10_mm+n10_wb, n10_wb")
++
++(define_insn_reservation "nds_n10_mul" 1
++  (and (eq_attr "type" "mul")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_mac" 1
++       (and (eq_attr "type" "mac")
++	    (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_div" 1
++  (and (eq_attr "type" "div")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex*34, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_branch" 1
++  (and (eq_attr "type" "branch")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_alu" 1
++  (and (eq_attr "type" "dalu")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_alu64" 1
++  (and (eq_attr "type" "dalu64")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_alu_round" 1
++  (and (eq_attr "type" "daluround")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_cmp" 1
++  (and (eq_attr "type" "dcmp")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_clip" 1
++  (and (eq_attr "type" "dclip")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_mul" 1
++  (and (eq_attr "type" "dmul")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_mac" 1
++  (and (eq_attr "type" "dmac")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_insb" 1
++  (and (eq_attr "type" "dinsb")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_pack" 1
++  (and (eq_attr "type" "dpack")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_bpick" 1
++  (and (eq_attr "type" "dbpick")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_dsp_wext" 1
++  (and (eq_attr "type" "dwext")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ex, n10_mm, n10_wb")
++
++(define_insn_reservation "nds_n10_fpu_alu" 4
++  (and (eq_attr "type" "falu")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_muls" 4
++  (and (eq_attr "type" "fmuls")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_muld" 4
++  (and (eq_attr "type" "fmuld")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2*2, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_macs" 4
++  (and (eq_attr "type" "fmacs")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2*3, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_macd" 4
++  (and (eq_attr "type" "fmacd")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2*4, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_divs" 4
++  (and (ior (eq_attr "type" "fdivs")
++	    (eq_attr "type" "fsqrts"))
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2*14, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_divd" 4
++  (and (ior (eq_attr "type" "fdivd")
++	    (eq_attr "type" "fsqrtd"))
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2*28, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_fast_alu" 2
++  (and (ior (eq_attr "type" "fcmp")
++	    (ior (eq_attr "type" "fabs")
++		 (ior (eq_attr "type" "fcpy")
++		      (eq_attr "type" "fcmov"))))
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_fmtsr" 4
++  (and (eq_attr "type" "fmtsr")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_fmtdr" 4
++  (and (eq_attr "type" "fmtdr")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ii+n10f_iq, n10f_iq+n10f_rf, n10f_rf+n10f_e1, n10f_e1+n10f_e2, n10f_e2+n10f_e3, n10f_e3+n10f_e4, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_fmfsr" 2
++  (and (eq_attr "type" "fmfsr")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_fmfdr" 2
++  (and (eq_attr "type" "fmfdr")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10_ii+n10f_iq, n10f_iq+n10f_rf, n10f_rf+n10f_e1, n10f_e1+n10f_e2, n10f_e2+n10f_e3, n10f_e3+n10f_e4, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_load" 3
++  (and (eq_attr "type" "fload")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2, n10f_e3, n10f_e4")
++
++(define_insn_reservation "nds_n10_fpu_store" 1
++  (and (eq_attr "type" "fstore")
++       (eq_attr "pipeline_model" "n10"))
++  "n10_ii, n10f_iq, n10f_rf, n10f_e1, n10f_e2, n10f_e3, n10f_e4")
++
++;; ------------------------------------------------------------------------
++;; Comment Notations and Bypass Rules
++;; ------------------------------------------------------------------------
++;; Producers (LHS)
++;;   LD
++;;     Load data from the memory and produce the loaded data. The result is
++;;     ready at MM.
++;;   LMW(N, M)
++;;     There are N micro-operations within an instruction that loads multiple
++;;     words. The result produced by the M-th micro-operation is sent to
++;;     consumers. The result is ready at MM.
++;;   MUL, MAC
++;;     Compute data in the multiply-adder and produce the data. The result
++;;     is ready at MM.
++;;   DIV
++;;     Compute data in the divider and produce the data. The result is ready
++;;     at MM.
++;;
++;; Consumers (RHS)
++;;   ALU, MOVD44, PBSAD, PBSADA_RaRb, MUL, MAC, DIV, MMU
++;;     Require operands at EX.
++;;   ALU_SHIFT_Rb
++;;     An ALU-SHIFT instruction consists of a shift micro-operation followed
++;;     by an arithmetic micro-operation. The operand Rb is used by the first
++;;     micro-operation, and there are some latencies if data dependency occurs.
++;;   MAC_RaRb
++;;     A MAC instruction does multiplication at EX and does accumulation at MM,
++;;     so the operand Rt is required at MM, and operands Ra and Rb are required
++;;     at EX.
++;;   ADDR_IN
++;;     If an instruction requires an address as its input operand, the address
++;;     is required at EX.
++;;   ST
++;;     A store instruction requires its data at MM.
++;;   SMW(N, M)
++;;     There are N micro-operations within an instruction that stores multiple
++;;     words. Each M-th micro-operation requires its data at MM.
++;;   BR
++;;     If a branch instruction is conditional, its input data is required at EX.
++
++;; FPU_ADDR_OUT -> FPU_ADDR_IN
++;; Main pipeline rules don't need this because those default latency is 1.
++(define_bypass 1
++  "nds_n10_fpu_load, nds_n10_fpu_store"
++  "nds_n10_fpu_load, nds_n10_fpu_store"
++  "nds32_n10_ex_to_ex_p"
++)
++
++;; LD, MUL, MAC, DIV, DALU64, DMUL, DMAC, DALUROUND, DBPICK, DWEXT
++;;   -> ALU, ALU_SHIFT_Rb, PBSAD, PBSADA_RaRb, MOVD44, MUL, MAC_RaRb, DIV, ADDR_IN, BR, MMU,
++;;      DALU, DALUROUND, DMUL, DMAC_RaRb, DPACK, DINSB, DCMP, DCLIP, WEXT_O, BPICK_RaRb
++(define_bypass 2
++  "nds_n10_load, nds_n10_mul, nds_n10_mac, nds_n10_div,\
++   nds_n10_dsp_alu64, nds_n10_dsp_mul, nds_n10_dsp_mac,\
++   nds_n10_dsp_alu_round, nds_n10_dsp_bpick, nds_n10_dsp_wext"
++  "nds_n10_alu, nds_n10_alu_shift,\
++   nds_n10_pbsad, nds_n10_pbsada,\
++   nds_n10_mul, nds_n10_mac, nds_n10_div,\
++   nds_n10_branch,\
++   nds_n10_load, nds_n10_store,\
++   nds_n10_load_multiple_1, nds_n10_load_multiple_2, nds_n10_load_multiple_3,\
++   nds_n10_load_multiple_4, nds_n10_load_multiple_5, nds_n10_load_multiple_6,\
++   nds_n10_load_multiple_7, nds_n10_load_multiple_N,\
++   nds_n10_store_multiple_1, nds_n10_store_multiple_2, nds_n10_store_multiple_3,\
++   nds_n10_store_multiple_4, nds_n10_store_multiple_5, nds_n10_store_multiple_6,\
++   nds_n10_store_multiple_7, nds_n10_store_multiple_N,\
++   nds_n10_mmu,\
++   nds_n10_dsp_alu, nds_n10_dsp_alu_round,\
++   nds_n10_dsp_mul, nds_n10_dsp_mac, nds_n10_dsp_pack,\
++   nds_n10_dsp_insb, nds_n10_dsp_cmp, nds_n10_dsp_clip,\
++   nds_n10_dsp_wext, nds_n10_dsp_bpick"
++  "nds32_n10_mm_to_ex_p"
++)
++
++;; LMW(N, N)
++;;   -> ALU, ALU_SHIFT_Rb, PBSAD, PBSADA_RaRb, MOVD44, MUL, MAC_RaRb, DIV, ADDR_IN, BR, MMU
++;;      DALU, DALUROUND, DMUL, DMAC_RaRb, DPACK, DINSB, DCMP, DCLIP, WEXT_O, BPICK_RaRb
++(define_bypass 2
++  "nds_n10_load_multiple_1, nds_n10_load_multiple_2, nds_n10_load_multiple_3,\
++   nds_n10_load_multiple_4, nds_n10_load_multiple_5, nds_n10_load_multiple_6,\
++   nds_n10_load_multiple_7, nds_n10_load_multiple_N"
++  "nds_n10_alu, nds_n10_alu_shift,\
++   nds_n10_pbsad, nds_n10_pbsada,\
++   nds_n10_mul, nds_n10_mac, nds_n10_div,\
++   nds_n10_branch,\
++   nds_n10_load, nds_n10_store,\
++   nds_n10_load_multiple_1, nds_n10_load_multiple_2, nds_n10_load_multiple_3,\
++   nds_n10_load_multiple_4, nds_n10_load_multiple_5, nds_n10_load_multiple_6,\
++   nds_n10_load_multiple_7, nds_n10_load_multiple_N,\
++   nds_n10_store_multiple_1, nds_n10_store_multiple_2, nds_n10_store_multiple_3,\
++   nds_n10_store_multiple_4, nds_n10_store_multiple_5, nds_n10_store_multiple_6,\
++   nds_n10_store_multiple_7, nds_n10_store_multiple_N,\
++   nds_n10_mmu,\
++   nds_n10_dsp_alu, nds_n10_dsp_alu_round,\
++   nds_n10_dsp_mul, nds_n10_dsp_mac, nds_n10_dsp_pack,\
++   nds_n10_dsp_insb, nds_n10_dsp_cmp, nds_n10_dsp_clip,\
++   nds_n10_dsp_wext, nds_n10_dsp_bpick"
++  "nds32_n10_last_load_to_ex_p"
++)
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-n13.md gcc-8.2.0/gcc/config/nds32/nds32-n13.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-n13.md	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-n13.md	2019-01-25 15:38:32.829242659 +0100
+@@ -0,0 +1,401 @@
++;; Pipeline descriptions of Andes NDS32 cpu for GNU compiler
++;; Copyright (C) 2012-2018 Free Software Foundation, Inc.
++;; Contributed by Andes Technology Corporation.
++;;
++;; This file is part of GCC.
++;;
++;; GCC is free software; you can redistribute it and/or modify it
++;; under the terms of the GNU General Public License as published
++;; by the Free Software Foundation; either version 3, or (at your
++;; option) any later version.
++;;
++;; GCC is distributed in the hope that it will be useful, but WITHOUT
++;; ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++;; or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++;; License for more details.
++;;
++;; You should have received a copy of the GNU General Public License
++;; along with GCC; see the file COPYING3.  If not see
++;; <http://www.gnu.org/licenses/>.
++
++
++;; ------------------------------------------------------------------------
++;; Define N13 pipeline settings.
++;; ------------------------------------------------------------------------
++
++(define_automaton "nds32_n13_machine")
++
++;; ------------------------------------------------------------------------
++;; Pipeline Stages
++;; ------------------------------------------------------------------------
++;; F1 - Instruction Fetch First
++;;   Instruction Tag/Data Arrays
++;;   ITLB Address Translation
++;;   Branch Target Buffer Prediction
++;; F2 - Instruction Fetch Second
++;;   Instruction Cache Hit Detection
++;;   Cache Way Selection
++;;   Inustruction Alignment
++;; I1 - Instruction Issue First / Instruction Decode
++;;   Instruction Cache Replay Triggering
++;;   32/16-Bit Instruction Decode
++;;   Return Address Stack Prediction
++;; I2 - Instruction Issue Second / Register File Access
++;;   Instruction Issue Logic
++;;   Register File Access
++;; E1 - Instruction Execute First / Address Generation / MAC First
++;;   Data Access Address generation
++;;   Multiply Operation
++;; E2 - Instruction Execute Second / Data Access First / MAC Second /
++;;      ALU Execute
++;;   Skewed ALU
++;;   Branch/Jump/Return Resolution
++;;   Data Tag/Data arrays
++;;   DTLB address translation
++;;   Accumulation Operation
++;; E3 - Instruction Execute Third / Data Access Second
++;;   Data Cache Hit Detection
++;;   Cache Way Selection
++;;   Data Alignment
++;; E4 - Instruction Execute Fourth / Write Back
++;;   Interruption Resolution
++;;   Instruction Retire
++;;   Register File Write Back
++
++(define_cpu_unit "n13_i1" "nds32_n13_machine")
++(define_cpu_unit "n13_i2" "nds32_n13_machine")
++(define_cpu_unit "n13_e1" "nds32_n13_machine")
++(define_cpu_unit "n13_e2" "nds32_n13_machine")
++(define_cpu_unit "n13_e3" "nds32_n13_machine")
++(define_cpu_unit "n13_e4" "nds32_n13_machine")
++
++(define_insn_reservation "nds_n13_unknown" 1
++  (and (eq_attr "type" "unknown")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_misc" 1
++  (and (eq_attr "type" "misc")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_mmu" 1
++  (and (eq_attr "type" "mmu")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_alu" 1
++  (and (eq_attr "type" "alu")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_alu_shift" 1
++  (and (eq_attr "type" "alu_shift")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i2+n13_e1, n13_e1+n13_e2, n13_e2+n13_e3, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_pbsad" 1
++  (and (eq_attr "type" "pbsad")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2*2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_pbsada" 1
++  (and (eq_attr "type" "pbsada")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2*3, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_load" 1
++  (and (match_test "nds32::load_single_p (insn)")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_store" 1
++  (and (match_test "nds32::store_single_p (insn)")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_1" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "1"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_2" 1
++  (and (ior (and (eq_attr "type" "load_multiple")
++		 (eq_attr "combo" "2"))
++	    (match_test "nds32::load_double_p (insn)"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i2+n13_e1, n13_e1+n13_e2, n13_e2+n13_e3, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_3" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "3"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2+n13_i2, n13_i1+n13_i2+n13_e1, n13_i2+n13_e1+n13_e2, n13_e1+n13_e2+n13_e3, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_4" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "4"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i2+n13_e1+n13_e2+n13_e3, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_5" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "5"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_6" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "6"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_7" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "7"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, (n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4)*2, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_8" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "8"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, (n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4)*3, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_load_multiple_12" 1
++  (and (and (eq_attr "type" "load_multiple")
++	    (eq_attr "combo" "12"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, (n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4)*7, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_1" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "1"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_2" 1
++  (and (ior (and (eq_attr "type" "store_multiple")
++		 (eq_attr "combo" "2"))
++	    (match_test "nds32::store_double_p (insn)"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i2+n13_e1, n13_e1+n13_e2, n13_e2+n13_e3, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_3" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "3"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2+n13_i2, n13_i1+n13_i2+n13_e1, n13_i2+n13_e1+n13_e2, n13_e1+n13_e2+n13_e3, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_4" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "4"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i2+n13_e1+n13_e2+n13_e3, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_5" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "5"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_6" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "6"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_7" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "7"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, (n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4)*2, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_8" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "8"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, (n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4)*3, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++(define_insn_reservation "nds_n13_store_multiple_12" 1
++  (and (and (eq_attr "type" "store_multiple")
++	    (eq_attr "combo" "12"))
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i1+n13_i2, n13_i1+n13_i2+n13_e1, n13_i1+n13_i2+n13_e1+n13_e2, n13_i1+n13_i2+n13_e1+n13_e2+n13_e3, (n13_i1+n13_i2+n13_e1+n13_e2+n13_e3+n13_e4)*7, n13_i2+n13_e1+n13_e2+n13_e3+n13_e4, n13_e1+n13_e2+n13_e3+n13_e4, n13_e2+n13_e3+n13_e4, n13_e3+n13_e4, n13_e4")
++
++;; The multiplier at E1 takes two cycles.
++(define_insn_reservation "nds_n13_mul" 1
++  (and (eq_attr "type" "mul")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1*2, n13_e2, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_mac" 1
++  (and (eq_attr "type" "mac")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1*2, n13_e2, n13_e3, n13_e4")
++
++;; The cycles consumed at E2 are 32 - CLZ(abs(Ra)) + 2,
++;; so the worst case is 34.
++(define_insn_reservation "nds_n13_div" 1
++  (and (eq_attr "type" "div")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2*34, n13_e3, n13_e4")
++
++(define_insn_reservation "nds_n13_branch" 1
++  (and (eq_attr "type" "branch")
++       (eq_attr "pipeline_model" "n13"))
++  "n13_i1, n13_i2, n13_e1, n13_e2, n13_e3, n13_e4")
++
++;; ------------------------------------------------------------------------
++;; Comment Notations and Bypass Rules
++;; ------------------------------------------------------------------------
++;; Producers (LHS)
++;;   LD
++;;     Load data from the memory and produce the loaded data. The result is
++;;     ready at E3.
++;;   LMW(N, M)
++;;     There are N micro-operations within an instruction that loads multiple
++;;     words. The result produced by the M-th micro-operation is sent to
++;;     consumers. The result is ready at E3.
++;;   ADDR_OUT
++;;     Most load/store instructions can produce an address output if updating
++;;     the base register is required. The result is ready at E2, which is
++;;     produced by ALU.
++;;   ALU, ALU_SHIFT, SIMD
++;;     Compute data in ALU and produce the data. The result is ready at E2.
++;;   MUL, MAC
++;;     Compute data in the multiply-adder and produce the data. The result
++;;     is ready at E2.
++;;   DIV
++;;     Compute data in the divider and produce the data. The result is ready
++;;     at E2.
++;;   BR
++;;     Branch-with-link instructions produces a result containing the return
++;;     address. The result is ready at E2.
++;;
++;; Consumers (RHS)
++;;   ALU
++;;     General ALU instructions require operands at E2.
++;;   ALU_E1
++;;     Some special ALU instructions, such as BSE, BSP and MOVD44, require
++;;     operand at E1.
++;;   MUL, DIV, PBSAD, MMU
++;;     Operands are required at E1.
++;;   PBSADA_Rt, PBSADA_RaRb
++;;     Operands Ra and Rb are required at E1, and the operand Rt is required
++;;     at E2.
++;;   ALU_SHIFT_Rb
++;;     An ALU-SHIFT instruction consists of a shift micro-operation followed
++;;     by an arithmetic micro-operation. The operand Rb is used by the first
++;;     micro-operation, and there are some latencies if data dependency occurs.
++;;   MAC_RaRb
++;;     A MAC instruction does multiplication at E1 and does accumulation at E2,
++;;     so the operand Rt is required at E2, and operands Ra and Rb are required
++;;     at E1.
++;;   ADDR_IN
++;;     If an instruction requires an address as its input operand, the address
++;;     is required at E1.
++;;   ST
++;;     A store instruction requires its data at E2.
++;;   SMW(N, M)
++;;     There are N micro-operations within an instruction that stores multiple
++;;     words. Each M-th micro-operation requires its data at E2.
++;;   BR
++;;     If a branch instruction is conditional, its input data is required at E2.
++
++;; LD -> ALU_E1, PBSAD, PBSADA_RaRb, MUL, MAC_RaRb, DIV, MMU, ADDR_IN
++(define_bypass 3
++  "nds_n13_load"
++  "nds_n13_alu, nds_n13_pbsad, nds_n13_pbsada,\
++   nds_n13_mul, nds_n13_mac, nds_n13_div,\
++   nds_n13_mmu,\
++   nds_n13_load, nds_n13_store,\
++   nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12,\
++   nds_n13_store_multiple_1,nds_n13_store_multiple_2, nds_n13_store_multiple_3,\
++   nds_n13_store_multiple_4,nds_n13_store_multiple_5, nds_n13_store_multiple_6,\
++   nds_n13_store_multiple_7,nds_n13_store_multiple_8, nds_n13_store_multiple_12"
++  "nds32_n13_load_to_e1_p"
++)
++
++;; LD -> ALU, ALU_SHIFT_Rb, PBSADA_Rt, BR, ST, SMW(N, 1)
++(define_bypass 2
++  "nds_n13_load"
++  "nds_n13_alu, nds_n13_alu_shift, nds_n13_pbsada, nds_n13_branch, nds_n13_store,\
++   nds_n13_store_multiple_1,nds_n13_store_multiple_2, nds_n13_store_multiple_3,\
++   nds_n13_store_multiple_4,nds_n13_store_multiple_5, nds_n13_store_multiple_6,\
++   nds_n13_store_multiple_7,nds_n13_store_multiple_8, nds_n13_store_multiple_12"
++  "nds32_n13_load_to_e2_p"
++)
++
++;; LMW(N, N) -> ALU_E1, PBSAD, PBSADA_RaRb, MUL, MAC_RaRb, DIV, MMU, ADDR_IN
++(define_bypass 3
++  "nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12"
++  "nds_n13_alu, nds_n13_pbsad, nds_n13_pbsada,\
++   nds_n13_mul, nds_n13_mac, nds_n13_div,\
++   nds_n13_mmu,\
++   nds_n13_load, nds_n13_store,\
++   nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12,\
++   nds_n13_store_multiple_1,nds_n13_store_multiple_2, nds_n13_store_multiple_3,\
++   nds_n13_store_multiple_4,nds_n13_store_multiple_5, nds_n13_store_multiple_6,\
++   nds_n13_store_multiple_7,nds_n13_store_multiple_8, nds_n13_store_multiple_12"
++  "nds32_n13_last_load_to_e1_p")
++
++;; LMW(N, N) -> ALU, ALU_SHIFT_Rb, PBSADA_Rt, BR, ST, SMW(N, 1)
++(define_bypass 2
++  "nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12"
++  "nds_n13_alu, nds_n13_alu_shift, nds_n13_pbsada, nds_n13_branch, nds_n13_store,\
++   nds_n13_store_multiple_1,nds_n13_store_multiple_2, nds_n13_store_multiple_3,\
++   nds_n13_store_multiple_4,nds_n13_store_multiple_5, nds_n13_store_multiple_6,\
++   nds_n13_store_multiple_7,nds_n13_store_multiple_8, nds_n13_store_multiple_12"
++  "nds32_n13_last_load_to_e2_p"
++)
++
++;; LMW(N, N - 1) -> ALU_E1, PBSAD, PBSADA_RaRb, MUL, MAC_RaRb, DIV, MMU, ADDR_IN
++(define_bypass 2
++  "nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12"
++  "nds_n13_alu, nds_n13_pbsad, nds_n13_pbsada,\
++   nds_n13_mul, nds_n13_mac, nds_n13_div,\
++   nds_n13_mmu,\
++   nds_n13_load, nds_n13_store,\
++   nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12,\
++   nds_n13_store_multiple_1,nds_n13_store_multiple_2, nds_n13_store_multiple_3,\
++   nds_n13_store_multiple_4,nds_n13_store_multiple_5, nds_n13_store_multiple_6,\
++   nds_n13_store_multiple_7,nds_n13_store_multiple_8, nds_n13_store_multiple_12"
++  "nds32_n13_last_two_load_to_e1_p")
++
++;; ALU, ALU_SHIFT, SIMD, BR, MUL, MAC, DIV, ADDR_OUT
++;;   ->  ALU_E1, PBSAD, PBSADA_RaRb, MUL, MAC_RaRb, DIV, MMU, ADDR_IN
++(define_bypass 2
++  "nds_n13_alu, nds_n13_alu_shift, nds_n13_pbsad, nds_n13_pbsada, nds_n13_branch,\
++   nds_n13_mul, nds_n13_mac, nds_n13_div,\
++   nds_n13_load, nds_n13_store,\
++   nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12,\
++   nds_n13_store_multiple_1,nds_n13_store_multiple_2, nds_n13_store_multiple_3,\
++   nds_n13_store_multiple_4,nds_n13_store_multiple_5, nds_n13_store_multiple_6,\
++   nds_n13_store_multiple_7,nds_n13_store_multiple_8, nds_n13_store_multiple_12"
++  "nds_n13_alu, nds_n13_pbsad, nds_n13_pbsada,\
++   nds_n13_mul, nds_n13_mac, nds_n13_div,\
++   nds_n13_mmu,\
++   nds_n13_load, nds_n13_store,\
++   nds_n13_load_multiple_1,nds_n13_load_multiple_2, nds_n13_load_multiple_3,\
++   nds_n13_load_multiple_4,nds_n13_load_multiple_5, nds_n13_load_multiple_6,\
++   nds_n13_load_multiple_7,nds_n13_load_multiple_8, nds_n13_load_multiple_12,\
++   nds_n13_store_multiple_1,nds_n13_store_multiple_2, nds_n13_store_multiple_3,\
++   nds_n13_store_multiple_4,nds_n13_store_multiple_5, nds_n13_store_multiple_6,\
++   nds_n13_store_multiple_7,nds_n13_store_multiple_8, nds_n13_store_multiple_12"
++  "nds32_n13_e2_to_e1_p")
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32.opt gcc-8.2.0/gcc/config/nds32/nds32.opt
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32.opt	2018-04-22 12:10:00.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32.opt	2019-01-25 15:38:32.833242671 +0100
+@@ -32,6 +32,13 @@
+ Target RejectNegative Alias(mlittle-endian)
+ Generate code in little-endian mode.
+ 
++mfp-as-gp
++Target RejectNegative Alias(mforce-fp-as-gp)
++Force performing fp-as-gp optimization.
++
++mno-fp-as-gp
++Target RejectNegative Alias(mforbid-fp-as-gp)
++Forbid performing fp-as-gp optimization.
+ 
+ ; ---------------------------------------------------------------
+ 
+@@ -85,11 +92,36 @@
+ Target Undocumented RejectNegative Negative(mbig-endian) InverseMask(BIG_ENDIAN)
+ Generate code in little-endian mode.
+ 
++mforce-fp-as-gp
++Target Undocumented Mask(FORCE_FP_AS_GP)
++Prevent $fp being allocated during register allocation so that compiler is able to force performing fp-as-gp optimization.
++
++mforbid-fp-as-gp
++Target Undocumented Mask(FORBID_FP_AS_GP)
++Forbid using $fp to access static and global variables.  This option strictly forbids fp-as-gp optimization regardless of '-mforce-fp-as-gp'.
++
++mict-model=
++Target Undocumented RejectNegative Joined Enum(nds32_ict_model_type) Var(nds32_ict_model) Init(ICT_MODEL_SMALL)
++Specify the address generation strategy for ICT call's code model.
++
++Enum
++Name(nds32_ict_model_type) Type(enum nds32_ict_model_type)
++Known cmodel types (for use with the -mict-model= option):
++
++EnumValue
++Enum(nds32_ict_model_type) String(small) Value(ICT_MODEL_SMALL)
++
++EnumValue
++Enum(nds32_ict_model_type) String(large) Value(ICT_MODEL_LARGE)
+ 
+ mcmov
+ Target Report Mask(CMOV)
+ Generate conditional move instructions.
+ 
++mhw-abs
++Target Report Mask(HW_ABS)
++Generate hardware abs instructions.
++
+ mext-perf
+ Target Report Mask(EXT_PERF)
+ Generate performance extension instructions.
+@@ -102,6 +134,10 @@
+ Target Report Mask(EXT_STRING)
+ Generate string extension instructions.
+ 
++mext-dsp
++Target Report Mask(EXT_DSP)
++Generate DSP extension instructions.
++
+ mv3push
+ Target Report Mask(V3PUSH)
+ Generate v3 push25/pop25 instructions.
+@@ -115,13 +151,17 @@
+ Insert relax hint for linker to do relaxation.
+ 
+ mvh
+-Target Report Mask(VH)
++Target Report Mask(VH) Condition(!TARGET_LINUX_ABI)
+ Enable Virtual Hosting support.
+ 
+ misr-vector-size=
+ Target RejectNegative Joined UInteger Var(nds32_isr_vector_size) Init(NDS32_DEFAULT_ISR_VECTOR_SIZE)
+ Specify the size of each interrupt vector, which must be 4 or 16.
+ 
++misr-secure=
++Target RejectNegative Joined UInteger Var(nds32_isr_secure_level) Init(0)
++Specify the security level of c-isr for the whole file.
++
+ mcache-block-size=
+ Target RejectNegative Joined UInteger Var(nds32_cache_block_size) Init(NDS32_DEFAULT_CACHE_BLOCK_SIZE)
+ Specify the size of each cache block, which must be a power of 2 between 4 and 512.
+@@ -141,6 +181,9 @@
+ Enum(nds32_arch_type) String(v3) Value(ARCH_V3)
+ 
+ EnumValue
++Enum(nds32_arch_type) String(v3j) Value(ARCH_V3J)
++
++EnumValue
+ Enum(nds32_arch_type) String(v3m) Value(ARCH_V3M)
+ 
+ EnumValue
+@@ -149,23 +192,6 @@
+ EnumValue
+ Enum(nds32_arch_type) String(v3s) Value(ARCH_V3S)
+ 
+-mcmodel=
+-Target RejectNegative Joined Enum(nds32_cmodel_type) Var(nds32_cmodel_option) Init(CMODEL_LARGE)
+-Specify the address generation strategy for code model.
+-
+-Enum
+-Name(nds32_cmodel_type) Type(enum nds32_cmodel_type)
+-Known cmodel types (for use with the -mcmodel= option):
+-
+-EnumValue
+-Enum(nds32_cmodel_type) String(small) Value(CMODEL_SMALL)
+-
+-EnumValue
+-Enum(nds32_cmodel_type) String(medium) Value(CMODEL_MEDIUM)
+-
+-EnumValue
+-Enum(nds32_cmodel_type) String(large) Value(CMODEL_LARGE)
+-
+ mcpu=
+ Target RejectNegative Joined Enum(nds32_cpu_type) Var(nds32_cpu_option) Init(CPU_N9)
+ Specify the cpu for pipeline model.
+@@ -235,6 +261,99 @@
+ Enum(nds32_cpu_type) String(n968a) Value(CPU_N9)
+ 
+ EnumValue
++Enum(nds32_cpu_type) String(n10) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1033) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1033a) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1033-fpu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1033-spu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1068) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1068a) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1068-fpu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1068a-fpu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1068-spu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1068a-spu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(d10) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(d1088) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(d1088-fpu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) String(d1088-spu) Value(CPU_N10)
++
++EnumValue
++Enum(nds32_cpu_type) Undocumented String(graywolf) Value(CPU_GRAYWOLF)
++
++EnumValue
++Enum(nds32_cpu_type) String(n15) Value(CPU_GRAYWOLF)
++
++EnumValue
++Enum(nds32_cpu_type) String(d15) Value(CPU_GRAYWOLF)
++
++EnumValue
++Enum(nds32_cpu_type) String(n15s) Value(CPU_GRAYWOLF)
++
++EnumValue
++Enum(nds32_cpu_type) String(d15s) Value(CPU_GRAYWOLF)
++
++EnumValue
++Enum(nds32_cpu_type) String(n15f) Value(CPU_GRAYWOLF)
++
++EnumValue
++Enum(nds32_cpu_type) String(d15f) Value(CPU_GRAYWOLF)
++
++EnumValue
++Enum(nds32_cpu_type) String(n12) Value(CPU_N12)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1213) Value(CPU_N12)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1233) Value(CPU_N12)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1233-fpu) Value(CPU_N12)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1233-spu) Value(CPU_N12)
++
++EnumValue
++Enum(nds32_cpu_type) String(n13) Value(CPU_N13)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1337) Value(CPU_N13)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1337-fpu) Value(CPU_N13)
++
++EnumValue
++Enum(nds32_cpu_type) String(n1337-spu) Value(CPU_N13)
++
++EnumValue
+ Enum(nds32_cpu_type) String(simple) Value(CPU_SIMPLE)
+ 
+ mconfig-fpu=
+@@ -321,6 +440,18 @@
+ Target Report Mask(FPU_DOUBLE)
+ Generate double-precision floating-point instructions.
+ 
++mforce-no-ext-dsp
++Target Undocumented Report Mask(FORCE_NO_EXT_DSP)
++Force disable hardware loop, even use -mext-dsp.
++
++msched-prolog-epilog
++Target Var(flag_sched_prolog_epilog) Init(0)
++Permit scheduling of a function's prologue and epilogue sequence.
++
++mret-in-naked-func
++Target Var(flag_ret_in_naked_func) Init(1)
++Generate return instruction in naked function.
++
+ malways-save-lp
+ Target Var(flag_always_save_lp) Init(0)
+ Always save $lp in the stack.
+@@ -328,3 +459,7 @@
+ munaligned-access
+ Target Report Var(flag_unaligned_access) Init(0)
+ Enable unaligned word and halfword accesses to packed data.
++
++minline-asm-r15
++Target Report Var(flag_inline_asm_r15) Init(0)
++Allow use r15 for inline ASM.
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-opts.h gcc-8.2.0/gcc/config/nds32/nds32-opts.h
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-opts.h	2018-04-08 11:21:30.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-opts.h	2019-01-25 15:38:32.833242671 +0100
+@@ -29,6 +29,7 @@
+ {
+   ARCH_V2,
+   ARCH_V3,
++  ARCH_V3J,
+   ARCH_V3M,
+   ARCH_V3F,
+   ARCH_V3S
+@@ -42,6 +43,10 @@
+   CPU_N8,
+   CPU_E8,
+   CPU_N9,
++  CPU_N10,
++  CPU_GRAYWOLF,
++  CPU_N12,
++  CPU_N13,
+   CPU_SIMPLE
+ };
+ 
+@@ -53,6 +58,13 @@
+   CMODEL_LARGE
+ };
+ 
++/* The code model defines the address generation strategy.  */
++enum nds32_ict_model_type
++{
++  ICT_MODEL_SMALL,
++  ICT_MODEL_LARGE
++};
++
+ /* Multiply instruction configuration.  */
+ enum nds32_mul_type
+ {
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-peephole2.md gcc-8.2.0/gcc/config/nds32/nds32-peephole2.md
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-peephole2.md	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/nds32-peephole2.md	2019-01-25 15:38:32.833242671 +0100
+@@ -22,3 +22,139 @@
+ ;; Use define_peephole2 to handle possible target-specific optimization.
+ 
+ ;; ------------------------------------------------------------------------
++;; Try to utilize 16-bit instruction by swap operand if possible.
++;; ------------------------------------------------------------------------
++
++;; Try to make add as add45.
++(define_peephole2
++  [(set (match_operand:QIHISI 0 "register_operand"              "")
++	(plus:QIHISI (match_operand:QIHISI 1 "register_operand" "")
++		     (match_operand:QIHISI 2 "register_operand" "")))]
++  "reload_completed
++   && TARGET_16_BIT
++   && REGNO (operands[0]) == REGNO (operands[2])
++   && REGNO (operands[0]) != REGNO (operands[1])
++   && TEST_HARD_REG_BIT (reg_class_contents[MIDDLE_REGS], REGNO (operands[0]))"
++  [(set (match_dup 0) (plus:QIHISI (match_dup 2) (match_dup 1)))])
++
++;; Try to make xor/ior/and/mult as xor33/ior33/and33/mult33.
++(define_peephole2
++  [(set (match_operand:SI 0 "register_operand"    "")
++	(match_operator:SI 1 "nds32_have_33_inst_operator"
++	  [(match_operand:SI 2 "register_operand" "")
++	   (match_operand:SI 3 "register_operand" "")]))]
++  "reload_completed
++   && TARGET_16_BIT
++   && REGNO (operands[0]) == REGNO (operands[3])
++   && REGNO (operands[0]) != REGNO (operands[2])
++   && TEST_HARD_REG_BIT (reg_class_contents[LOW_REGS], REGNO (operands[0]))
++   && TEST_HARD_REG_BIT (reg_class_contents[LOW_REGS], REGNO (operands[2]))"
++  [(set (match_dup 0) (match_op_dup 1 [(match_dup 3) (match_dup 2)]))])
++
++(define_peephole
++  [(set (match_operand:SI 0 "register_operand" "")
++	(match_operand:SI 1 "register_operand" ""))
++   (set (match_operand:SI 2 "register_operand" "")
++	(match_operand:SI 3 "register_operand" ""))]
++  "TARGET_16_BIT
++   && !TARGET_ISA_V2
++   && NDS32_IS_GPR_REGNUM (REGNO (operands[0]))
++   && NDS32_IS_GPR_REGNUM (REGNO (operands[1]))
++   && ((REGNO (operands[0]) & 0x1) == 0)
++   && ((REGNO (operands[1]) & 0x1) == 0)
++   && (REGNO (operands[0]) + 1) == REGNO (operands[2])
++   && (REGNO (operands[1]) + 1) == REGNO (operands[3])"
++  "movd44\t%0, %1"
++  [(set_attr "type"   "alu")
++   (set_attr "length" "2")])
++
++;; Merge two fcpyss to fcpysd.
++(define_peephole2
++  [(set (match_operand:SF 0 "float_even_register_operand" "")
++	(match_operand:SF 1 "float_even_register_operand" ""))
++   (set (match_operand:SF 2 "float_odd_register_operand"  "")
++	(match_operand:SF 3 "float_odd_register_operand"  ""))]
++  "(TARGET_FPU_SINGLE || TARGET_FPU_DOUBLE)
++   && REGNO (operands[0]) == REGNO (operands[2]) - 1
++   && REGNO (operands[1]) == REGNO (operands[3]) - 1"
++  [(set (match_dup 4) (match_dup 5))]
++  {
++    operands[4] = gen_rtx_REG (DFmode, REGNO (operands[0]));
++    operands[5] = gen_rtx_REG (DFmode, REGNO (operands[1]));
++  })
++
++(define_peephole2
++  [(set (match_operand:SF 0 "float_odd_register_operand"  "")
++	(match_operand:SF 1 "float_odd_register_operand"  ""))
++   (set (match_operand:SF 2 "float_even_register_operand" "")
++	(match_operand:SF 3 "float_even_register_operand" ""))]
++  "(TARGET_FPU_SINGLE || TARGET_FPU_DOUBLE)
++   && REGNO (operands[2]) == REGNO (operands[0]) - 1
++   && REGNO (operands[3]) == REGNO (operands[1]) - 1"
++  [(set (match_dup 4) (match_dup 5))]
++  {
++    operands[4] = gen_rtx_REG (DFmode, REGNO (operands[2]));
++    operands[5] = gen_rtx_REG (DFmode, REGNO (operands[3]));
++  })
++
++;; ------------------------------------------------------------------------
++;; GCC will prefer [u]divmodsi3 rather than [u]divsi3 even remainder is
++;; unused, so we use split to drop mod operation for lower register pressure.
++
++(define_split
++  [(set (match_operand:SI 0 "register_operand")
++	(div:SI (match_operand:SI 1 "register_operand")
++		(match_operand:SI 2 "register_operand")))
++   (set (match_operand:SI 3 "register_operand")
++	(mod:SI (match_dup 1) (match_dup 2)))]
++  "find_regno_note (insn, REG_UNUSED, REGNO (operands[3])) != NULL
++   && can_create_pseudo_p ()"
++  [(set (match_dup 0)
++	(div:SI (match_dup 1)
++		(match_dup 2)))])
++
++(define_split
++  [(set (match_operand:SI 0 "register_operand")
++	(udiv:SI (match_operand:SI 1 "register_operand")
++		 (match_operand:SI 2 "register_operand")))
++   (set (match_operand:SI 3 "register_operand")
++	(umod:SI (match_dup 1) (match_dup 2)))]
++  "find_regno_note (insn, REG_UNUSED, REGNO (operands[3])) != NULL
++   && can_create_pseudo_p ()"
++  [(set (match_dup 0)
++	(udiv:SI (match_dup 1)
++		 (match_dup 2)))])
++
++(define_peephole2
++  [(set (match_operand:DI 0 "register_operand")
++	(mult:DI (sign_extend:DI (match_operand:SI 1 "register_operand"))
++		 (sign_extend:DI (match_operand:SI 2 "register_operand"))))]
++  "NDS32_EXT_DSP_P ()
++   && peep2_regno_dead_p (1, WORDS_BIG_ENDIAN ? REGNO (operands[0]) + 1 : REGNO (operands[0]))"
++  [(const_int 1)]
++{
++  rtx highpart = nds32_di_high_part_subreg (operands[0]);
++  emit_insn (gen_smulsi3_highpart (highpart, operands[1], operands[2]));
++  DONE;
++})
++
++(define_split
++  [(set (match_operand:DI 0 "nds32_general_register_operand" "")
++	(match_operand:DI 1 "nds32_general_register_operand" ""))]
++  "find_regno_note (insn, REG_UNUSED, REGNO (operands[0])) != NULL
++   || find_regno_note (insn, REG_UNUSED, REGNO (operands[0]) + 1) != NULL"
++  [(set (match_dup 0) (match_dup 1))]
++{
++  rtx dead_note = find_regno_note (curr_insn, REG_UNUSED, REGNO (operands[0]));
++  HOST_WIDE_INT offset;
++  if (dead_note == NULL_RTX)
++    offset = 0;
++  else
++    offset = 4;
++  operands[0] = simplify_gen_subreg (
++		  SImode, operands[0],
++		  DImode, offset);
++  operands[1] = simplify_gen_subreg (
++		  SImode, operands[1],
++		  DImode, offset);
++})
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-pipelines-auxiliary.c gcc-8.2.0/gcc/config/nds32/nds32-pipelines-auxiliary.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-pipelines-auxiliary.c	2018-04-08 11:21:30.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-pipelines-auxiliary.c	2019-01-25 15:38:32.833242671 +0100
+@@ -306,6 +306,19 @@
+   return false;
+ }
+ 
++/* Determine if the latency is occured when the consumer PBSADA_INSN uses the
++   value of DEF_REG in its Rt field.  */
++bool
++pbsada_insn_rt_dep_reg_p (rtx pbsada_insn, rtx def_reg)
++{
++  rtx pbsada_rt = SET_DEST (PATTERN (pbsada_insn));
++
++  if (rtx_equal_p (def_reg, pbsada_rt))
++    return true;
++
++  return false;
++}
++
+ /* Check if INSN is a movd44 insn consuming DEF_REG.  */
+ bool
+ movd44_even_dep_p (rtx_insn *insn, rtx def_reg)
+@@ -335,6 +348,103 @@
+   return false;
+ }
+ 
++/* Check if INSN is a wext insn consuming DEF_REG.  */
++bool
++wext_odd_dep_p (rtx insn, rtx def_reg)
++{
++  rtx shift_rtx = XEXP (SET_SRC (PATTERN (insn)), 0);
++  rtx use_reg = XEXP (shift_rtx, 0);
++  rtx pos_rtx = XEXP (shift_rtx, 1);
++
++  if (REG_P (pos_rtx) && reg_overlap_p (def_reg, pos_rtx))
++    return true;
++
++  if (GET_MODE (def_reg) == DImode)
++    return reg_overlap_p (def_reg, use_reg);
++
++  gcc_assert (REG_P (def_reg) || GET_CODE (def_reg) == SUBREG);
++  gcc_assert (REG_P (use_reg));
++
++  if (REG_P (def_reg))
++    {
++      if (!TARGET_BIG_ENDIAN)
++	return REGNO (def_reg) == REGNO (use_reg) + 1;
++      else
++	return  REGNO (def_reg) == REGNO (use_reg);
++    }
++
++  if (GET_CODE (def_reg) == SUBREG)
++    {
++      if (!reg_overlap_p (def_reg, use_reg))
++	return false;
++
++      if (!TARGET_BIG_ENDIAN)
++	return SUBREG_BYTE (def_reg) == 4;
++      else
++	return SUBREG_BYTE (def_reg) == 0;
++    }
++
++  return false;
++}
++
++/* Check if INSN is a bpick insn consuming DEF_REG.  */
++bool
++bpick_ra_rb_dep_p (rtx insn, rtx def_reg)
++{
++  rtx ior_rtx = SET_SRC (PATTERN (insn));
++  rtx and1_rtx = XEXP (ior_rtx, 0);
++  rtx and2_rtx = XEXP (ior_rtx, 1);
++  rtx reg1_0 = XEXP (and1_rtx, 0);
++  rtx reg1_1 = XEXP (and1_rtx, 1);
++  rtx reg2_0 = XEXP (and2_rtx, 0);
++  rtx reg2_1 = XEXP (and2_rtx, 1);
++
++  if (GET_CODE (reg1_0) == NOT)
++    {
++      if (rtx_equal_p (reg1_0, reg2_0))
++	return reg_overlap_p (def_reg, reg1_1)
++	       || reg_overlap_p (def_reg, reg2_1);
++
++      if (rtx_equal_p (reg1_0, reg2_1))
++	return reg_overlap_p (def_reg, reg1_1)
++	       || reg_overlap_p (def_reg, reg2_0);
++    }
++
++  if (GET_CODE (reg1_1) == NOT)
++    {
++      if (rtx_equal_p (reg1_1, reg2_0))
++	return reg_overlap_p (def_reg, reg1_0)
++	       || reg_overlap_p (def_reg, reg2_1);
++
++      if (rtx_equal_p (reg1_1, reg2_1))
++	return reg_overlap_p (def_reg, reg1_0)
++	       || reg_overlap_p (def_reg, reg2_0);
++    }
++
++  if (GET_CODE (reg2_0) == NOT)
++    {
++      if (rtx_equal_p (reg2_0, reg1_0))
++	return reg_overlap_p (def_reg, reg2_1)
++	       || reg_overlap_p (def_reg, reg1_1);
++
++      if (rtx_equal_p (reg2_0, reg1_1))
++	return reg_overlap_p (def_reg, reg2_1)
++	       || reg_overlap_p (def_reg, reg1_0);
++    }
++
++  if (GET_CODE (reg2_1) == NOT)
++    {
++      if (rtx_equal_p (reg2_1, reg1_0))
++	return reg_overlap_p (def_reg, reg2_0)
++	       || reg_overlap_p (def_reg, reg1_1);
++
++      if (rtx_equal_p (reg2_1, reg1_1))
++	return reg_overlap_p (def_reg, reg2_0)
++	       || reg_overlap_p (def_reg, reg1_0);
++    }
++
++  gcc_unreachable ();
++}
+ } // namespace scheduling
+ } // namespace nds32
+ 
+@@ -375,8 +485,7 @@
+       operations in order to write two registers. We have to check the
+       dependency from the producer to the first micro-operation.  */
+     case TYPE_DIV:
+-      if (INSN_CODE (consumer) == CODE_FOR_divmodsi4
+-	  || INSN_CODE (consumer) == CODE_FOR_udivmodsi4)
++      if (divmod_p (consumer))
+ 	use_rtx = SET_SRC (parallel_element (consumer, 0));
+       else
+ 	use_rtx = SET_SRC (PATTERN (consumer));
+@@ -506,8 +615,7 @@
+       operations in order to write two registers. We have to check the
+       dependency from the producer to the first micro-operation.  */
+     case TYPE_DIV:
+-      if (INSN_CODE (consumer) == CODE_FOR_divmodsi4
+-	  || INSN_CODE (consumer) == CODE_FOR_udivmodsi4)
++      if (divmod_p (consumer))
+ 	use_rtx = SET_SRC (parallel_element (consumer, 0));
+       else
+ 	use_rtx = SET_SRC (PATTERN (consumer));
+@@ -606,8 +714,7 @@
+       break;
+ 
+     case TYPE_DIV:
+-      if (INSN_CODE (consumer) == CODE_FOR_divmodsi4
+-	  || INSN_CODE (consumer) == CODE_FOR_udivmodsi4)
++      if (divmod_p (consumer))
+ 	use_rtx = SET_SRC (parallel_element (consumer, 0));
+       else
+ 	use_rtx = SET_SRC (PATTERN (consumer));
+@@ -706,13 +813,175 @@
+       We have to check the dependency from the producer to the first
+       micro-operation.  */
+     case TYPE_DIV:
+-      if (INSN_CODE (consumer) == CODE_FOR_divmodsi4
+-	  || INSN_CODE (consumer) == CODE_FOR_udivmodsi4)
++      if (divmod_p (consumer))
++	use_rtx = SET_SRC (parallel_element (consumer, 0));
++      else
++	use_rtx = SET_SRC (PATTERN (consumer));
++      break;
++
++    case TYPE_MMU:
++      if (GET_CODE (PATTERN (consumer)) == SET)
++	use_rtx = SET_SRC (PATTERN (consumer));
++      else
++	return true;
++      break;
++
++    case TYPE_LOAD:
++    case TYPE_STORE:
++      use_rtx = extract_mem_rtx (consumer);
++      break;
++
++    case TYPE_LOAD_MULTIPLE:
++    case TYPE_STORE_MULTIPLE:
++      use_rtx = extract_base_reg (consumer);
++      break;
++
++    case TYPE_BRANCH:
++      use_rtx = PATTERN (consumer);
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
++  if (reg_overlap_p (def_reg, use_rtx))
++    return true;
++
++  return false;
++}
++
++/* Check the dependency between the producer defining DEF_REG and CONSUMER
++   requiring input operand at EX.  */
++bool
++n10_consumed_by_ex_dep_p (rtx_insn *consumer, rtx def_reg)
++{
++  rtx use_rtx;
++
++  switch (get_attr_type (consumer))
++    {
++    case TYPE_ALU:
++    case TYPE_PBSAD:
++    case TYPE_MUL:
++    case TYPE_DALU:
++    case TYPE_DALU64:
++    case TYPE_DMUL:
++    case TYPE_DPACK:
++    case TYPE_DINSB:
++    case TYPE_DCMP:
++    case TYPE_DCLIP:
++    case TYPE_DALUROUND:
++      use_rtx = SET_SRC (PATTERN (consumer));
++      break;
++
++    case TYPE_ALU_SHIFT:
++      use_rtx = extract_shift_reg (consumer);
++      break;
++
++    case TYPE_PBSADA:
++      return pbsada_insn_ra_rb_dep_reg_p (consumer, def_reg);
++
++    case TYPE_MAC:
++    case TYPE_DMAC:
++      use_rtx = extract_mac_non_acc_rtx (consumer);
++      break;
++
++   /* Some special instructions, divmodsi4 and udivmodsi4, produce two
++      results, the quotient and the remainder.  */
++    case TYPE_DIV:
++      if (divmod_p (consumer))
++	use_rtx = SET_SRC (parallel_element (consumer, 0));
++      else
++	use_rtx = SET_SRC (PATTERN (consumer));
++      break;
++
++    case TYPE_DWEXT:
++      return wext_odd_dep_p (consumer, def_reg);
++
++    case TYPE_DBPICK:
++      return bpick_ra_rb_dep_p (consumer, def_reg);
++
++    case TYPE_MMU:
++      if (GET_CODE (PATTERN (consumer)) == SET)
++	use_rtx = SET_SRC (PATTERN (consumer));
++      else
++	return true;
++      break;
++
++    case TYPE_LOAD:
++    case TYPE_STORE:
++      use_rtx = extract_mem_rtx (consumer);
++      break;
++
++    case TYPE_LOAD_MULTIPLE:
++    case TYPE_STORE_MULTIPLE:
++      use_rtx = extract_base_reg (consumer);
++      break;
++
++    case TYPE_BRANCH:
++      use_rtx = PATTERN (consumer);
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
++  if (reg_overlap_p (def_reg, use_rtx))
++    return true;
++
++  return false;
++}
++
++/* Check the dependency between the producer defining DEF_REG and CONSUMER
++   requiring input operand at EX.  */
++bool
++gw_consumed_by_ex_dep_p (rtx_insn *consumer, rtx def_reg)
++{
++  rtx use_rtx;
++
++  switch (get_attr_type (consumer))
++    {
++    case TYPE_ALU:
++    case TYPE_PBSAD:
++    case TYPE_MUL:
++    case TYPE_DALU:
++    case TYPE_DALU64:
++    case TYPE_DMUL:
++    case TYPE_DPACK:
++    case TYPE_DINSB:
++    case TYPE_DCMP:
++    case TYPE_DCLIP:
++    case TYPE_DALUROUND:
++      use_rtx = SET_SRC (PATTERN (consumer));
++      break;
++
++    case TYPE_ALU_SHIFT:
++      use_rtx = extract_shift_reg (consumer);
++      break;
++
++    case TYPE_PBSADA:
++      return pbsada_insn_ra_rb_dep_reg_p (consumer, def_reg);
++
++    case TYPE_MAC:
++    case TYPE_DMAC:
++      use_rtx = extract_mac_non_acc_rtx (consumer);
++      break;
++
++   /* Some special instructions, divmodsi4 and udivmodsi4, produce two
++      results, the quotient and the remainder.  We have to check the
++      dependency from the producer to the first micro-operation.  */
++    case TYPE_DIV:
++      if (divmod_p (consumer))
+ 	use_rtx = SET_SRC (parallel_element (consumer, 0));
+       else
+ 	use_rtx = SET_SRC (PATTERN (consumer));
+       break;
+ 
++    case TYPE_DWEXT:
++      return wext_odd_dep_p (consumer, def_reg);
++
++    case TYPE_DBPICK:
++      return bpick_ra_rb_dep_p (consumer, def_reg);
++
+     case TYPE_MMU:
+       if (GET_CODE (PATTERN (consumer)) == SET)
+ 	use_rtx = SET_SRC (PATTERN (consumer));
+@@ -744,7 +1013,153 @@
+   return false;
+ }
+ 
++/* Check dependencies from any stages to ALU_E1 (E1).  This is a helper
++   function of n13_consumed_by_e1_dep_p ().  */
++bool
++n13_alu_e1_insn_dep_reg_p (rtx_insn *alu_e1_insn, rtx def_reg)
++{
++  rtx unspec_rtx, operand_ra, operand_rb;
++  rtx src_rtx, dst_rtx;
++
++  switch (INSN_CODE (alu_e1_insn))
++    {
++    /* BSP and BSE are supported by built-in functions, the corresponding
++       patterns are formed by UNSPEC RTXs.  We have to handle them
++       individually.  */
++    case CODE_FOR_unspec_bsp:
++    case CODE_FOR_unspec_bse:
++      unspec_rtx = SET_SRC (parallel_element (alu_e1_insn, 0));
++      gcc_assert (GET_CODE (unspec_rtx) == UNSPEC);
++
++      operand_ra = XVECEXP (unspec_rtx, 0, 0);
++      operand_rb = XVECEXP (unspec_rtx, 0, 1);
++
++      if (rtx_equal_p (def_reg, operand_ra)
++	  || rtx_equal_p (def_reg, operand_rb))
++	return true;
++
++      return false;
++
++    /* Unlink general ALU instructions, MOVD44 requires operands at E1.  */
++    case CODE_FOR_move_di:
++    case CODE_FOR_move_df:
++      src_rtx = SET_SRC (PATTERN (alu_e1_insn));
++      dst_rtx = SET_DEST (PATTERN (alu_e1_insn));
++
++      if (REG_P (dst_rtx) && REG_P (src_rtx)
++	  && rtx_equal_p (src_rtx, def_reg))
++	return true;
++
++      return false;
++
++    default:
++      return false;
++    }
++}
++
++/* Check the dependency between the producer defining DEF_REG and CONSUMER
++   requiring input operand at E1.  Because the address generation unti is
++   at E1, the address input should be ready at E1.  Note that the branch
++   target is also a kind of addresses, so we have to check it.  */
++bool
++n13_consumed_by_e1_dep_p (rtx_insn *consumer, rtx def_reg)
++{
++  rtx use_rtx;
++
++  switch (get_attr_type (consumer))
++    {
++    /* ALU_E1 */
++    case TYPE_ALU:
++      return n13_alu_e1_insn_dep_reg_p (consumer, def_reg);
++
++    case TYPE_PBSADA:
++      return pbsada_insn_ra_rb_dep_reg_p (consumer, def_reg);
++
++    case TYPE_PBSAD:
++    case TYPE_MUL:
++      use_rtx = SET_SRC (PATTERN (consumer));
++      break;
++
++    case TYPE_MAC:
++      use_rtx = extract_mac_non_acc_rtx (consumer);
++      break;
++
++    case TYPE_DIV:
++      if (divmod_p (consumer))
++	use_rtx = SET_SRC (parallel_element (consumer, 0));
++      else
++	use_rtx = SET_SRC (PATTERN (consumer));
++      break;
++
++    case TYPE_MMU:
++      if (GET_CODE (PATTERN (consumer)) == SET)
++	use_rtx = SET_SRC (PATTERN (consumer));
++      else
++	return true;
++      break;
++
++    case TYPE_BRANCH:
++      use_rtx = extract_branch_target_rtx (consumer);
++      break;
++
++    case TYPE_LOAD:
++    case TYPE_STORE:
++      use_rtx = extract_mem_rtx (consumer);
++      break;
++
++    case TYPE_LOAD_MULTIPLE:
++    case TYPE_STORE_MULTIPLE:
++      use_rtx = extract_base_reg (consumer);
++      break;
++
++    default:
++      return false;
++    }
++
++  if (reg_overlap_p (def_reg, use_rtx))
++    return true;
++
++  return false;
++}
++
++/* Check the dependency between the producer defining DEF_REG and CONSUMER
++   requiring input operand at E2.  */
++bool
++n13_consumed_by_e2_dep_p (rtx_insn *consumer, rtx def_reg)
++{
++  rtx use_rtx;
++
++  switch (get_attr_type (consumer))
++    {
++    case TYPE_ALU:
++    case TYPE_STORE:
++      use_rtx = SET_SRC (PATTERN (consumer));
++      break;
++
++    case TYPE_ALU_SHIFT:
++      use_rtx = extract_shift_reg (consumer);
++      break;
++
++    case TYPE_PBSADA:
++      return pbsada_insn_rt_dep_reg_p (consumer, def_reg);
++
++    case TYPE_STORE_MULTIPLE:
++      use_rtx = extract_nth_access_rtx (consumer, 0);
++      break;
++
++    case TYPE_BRANCH:
++      use_rtx = extract_branch_condition_rtx (consumer);
++      break;
++
++    default:
++      gcc_unreachable();
++    }
++
++  if (reg_overlap_p (def_reg, use_rtx))
++    return true;
+ 
++  return false;
++}
+ } // anonymous namespace
+ 
+ /* ------------------------------------------------------------------------ */
+@@ -837,8 +1252,7 @@
+       break;
+ 
+     case TYPE_DIV:
+-      if (INSN_CODE (producer) == CODE_FOR_divmodsi4
+-	  || INSN_CODE (producer) == CODE_FOR_udivmodsi4)
++      if (divmod_p (producer))
+ 	def_reg = SET_DEST (parallel_element (producer, 1));
+       else
+ 	def_reg = SET_DEST (PATTERN (producer));
+@@ -969,8 +1383,7 @@
+       break;
+ 
+     case TYPE_DIV:
+-      if (INSN_CODE (producer) == CODE_FOR_divmodsi4
+-	  || INSN_CODE (producer) == CODE_FOR_udivmodsi4)
++      if (divmod_p (producer))
+ 	{
+ 	  rtx def_reg1 = SET_DEST (parallel_element (producer, 0));
+ 	  rtx def_reg2 = SET_DEST (parallel_element (producer, 1));
+@@ -1073,8 +1486,7 @@
+       results, the quotient and the remainder.  We have to handle them
+       individually.  */
+     case TYPE_DIV:
+-      if (INSN_CODE (producer) == CODE_FOR_divmodsi4
+-	  || INSN_CODE (producer) == CODE_FOR_udivmodsi4)
++      if (divmod_p (producer))
+ 	{
+ 	  rtx def_reg1 = SET_DEST (parallel_element (producer, 0));
+ 	  rtx def_reg2 = SET_DEST (parallel_element (producer, 1));
+@@ -1132,4 +1544,245 @@
+     return n9_3r2w_consumed_by_ex_dep_p (consumer, last_def_reg);
+ }
+ 
++/* Guard functions for N10 cores.  */
++
++/* Check dependencies from EX to EX (ADDR_OUT -> ADDR_IN).  */
++bool
++nds32_n10_ex_to_ex_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  gcc_assert (get_attr_type (producer) == TYPE_FLOAD
++	      || get_attr_type (producer) == TYPE_FSTORE);
++  gcc_assert (get_attr_type (consumer) == TYPE_FLOAD
++	      || get_attr_type (consumer) == TYPE_FSTORE);
++
++  if (!post_update_insn_p (producer))
++    return false;
++
++  return reg_overlap_p (extract_base_reg (producer),
++			extract_mem_rtx (consumer));
++}
++
++/* Check dependencies from MM to EX.  */
++bool
++nds32_n10_mm_to_ex_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx def_reg;
++
++  switch (get_attr_type (producer))
++    {
++    case TYPE_LOAD:
++    case TYPE_MUL:
++    case TYPE_MAC:
++    case TYPE_DALU64:
++    case TYPE_DMUL:
++    case TYPE_DMAC:
++    case TYPE_DALUROUND:
++    case TYPE_DBPICK:
++    case TYPE_DWEXT:
++      def_reg = SET_DEST (PATTERN (producer));
++      break;
++
++   /* Some special instructions, divmodsi4 and udivmodsi4, produce two
++      results, the quotient and the remainder.  We have to handle them
++      individually.  */
++    case TYPE_DIV:
++      if (divmod_p (producer))
++	{
++	  rtx def_reg1 = SET_DEST (parallel_element (producer, 0));
++	  rtx def_reg2 = SET_DEST (parallel_element (producer, 1));
++
++	  return (n10_consumed_by_ex_dep_p (consumer, def_reg1)
++		  || n10_consumed_by_ex_dep_p (consumer, def_reg2));
++	}
++
++      def_reg = SET_DEST (PATTERN (producer));
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
++    return n10_consumed_by_ex_dep_p (consumer, def_reg);
++}
++
++/* Check dependencies from LMW(N, N) to EX.  */
++bool
++nds32_n10_last_load_to_ex_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx last_def_reg = extract_nth_access_reg (producer, -1);
++
++  return n10_consumed_by_ex_dep_p (consumer, last_def_reg);
++}
++
++/* Guard functions for Graywolf cores.  */
++
++/* Check dependencies from EX to EX (ADDR_OUT -> ADDR_IN).  */
++bool
++nds32_gw_ex_to_ex_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  return nds32_n10_ex_to_ex_p (producer, consumer);
++}
++
++/* Check dependencies from MM to EX.  */
++bool
++nds32_gw_mm_to_ex_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx def_reg;
++
++  switch (get_attr_type (producer))
++    {
++    case TYPE_LOAD:
++    case TYPE_MUL:
++    case TYPE_MAC:
++    case TYPE_DALU64:
++    case TYPE_DMUL:
++    case TYPE_DMAC:
++    case TYPE_DALUROUND:
++    case TYPE_DBPICK:
++    case TYPE_DWEXT:
++      def_reg = SET_DEST (PATTERN (producer));
++      break;
++
++   /* Some special instructions, divmodsi4 and udivmodsi4, produce two
++      results, the quotient and the remainder.  We have to handle them
++      individually.  */
++    case TYPE_DIV:
++      if (divmod_p (producer))
++	{
++	  rtx def_reg1 = SET_DEST (parallel_element (producer, 0));
++	  rtx def_reg2 = SET_DEST (parallel_element (producer, 1));
++
++	  return (gw_consumed_by_ex_dep_p (consumer, def_reg1)
++		  || gw_consumed_by_ex_dep_p (consumer, def_reg2));
++	}
++
++      def_reg = SET_DEST (PATTERN (producer));
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
++    return gw_consumed_by_ex_dep_p (consumer, def_reg);
++}
++
++/* Check dependencies from LMW(N, N) to EX.  */
++bool
++nds32_gw_last_load_to_ex_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx last_def_reg = extract_nth_access_reg (producer, -1);
++
++  return gw_consumed_by_ex_dep_p (consumer, last_def_reg);
++}
++
++/* Guard functions for N12/N13 cores.  */
++
++/* Check dependencies from E2 to E1.  */
++bool
++nds32_n13_e2_to_e1_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx def_reg;
++
++  switch (get_attr_type (producer))
++    {
++    /* Only post-update load/store instructions are considered.  These
++       instructions produces address output at E2.  */
++    case TYPE_LOAD:
++    case TYPE_STORE:
++    case TYPE_LOAD_MULTIPLE:
++    case TYPE_STORE_MULTIPLE:
++      if (!post_update_insn_p (producer))
++	return false;
++
++      def_reg = extract_base_reg (producer);
++      break;
++
++    case TYPE_ALU:
++    case TYPE_ALU_SHIFT:
++    case TYPE_PBSAD:
++    case TYPE_PBSADA:
++    case TYPE_MUL:
++    case TYPE_MAC:
++      def_reg = SET_DEST (PATTERN (producer));
++      break;
++
++    case TYPE_BRANCH:
++      return true;
++
++    case TYPE_DIV:
++      /* Some special instructions, divmodsi4 and udivmodsi4, produce two
++	 results, the quotient and the remainder.  We have to handle them
++	 individually.  */
++      if (divmod_p (producer))
++	{
++	  rtx def_reg1 = SET_DEST (parallel_element (producer, 0));
++	  rtx def_reg2 = SET_DEST (parallel_element (producer, 1));
++
++	  return (n13_consumed_by_e1_dep_p (consumer, def_reg1)
++		  || n13_consumed_by_e1_dep_p (consumer, def_reg2));
++	}
++
++      def_reg = SET_DEST (PATTERN (producer));
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
++  return n13_consumed_by_e1_dep_p (consumer, def_reg);
++}
++
++/* Check dependencies from Load-Store Unit (E3) to E1.  */
++bool
++nds32_n13_load_to_e1_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx def_reg = SET_DEST (PATTERN (producer));
++
++  gcc_assert (get_attr_type (producer) == TYPE_LOAD);
++  gcc_assert (REG_P (def_reg) || GET_CODE (def_reg) == SUBREG);
++
++  return n13_consumed_by_e1_dep_p (consumer, def_reg);
++}
++
++/* Check dependencies from Load-Store Unit (E3) to E2.  */
++bool
++nds32_n13_load_to_e2_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx def_reg = SET_DEST (PATTERN (producer));
++
++  gcc_assert (get_attr_type (producer) == TYPE_LOAD);
++  gcc_assert (REG_P (def_reg) || GET_CODE (def_reg) == SUBREG);
++
++  return n13_consumed_by_e2_dep_p (consumer, def_reg);
++}
++
++/* Check dependencies from LMW(N, N) to E1.  */
++bool
++nds32_n13_last_load_to_e1_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx last_def_reg = extract_nth_access_reg (producer, -1);
++
++  return n13_consumed_by_e1_dep_p (consumer, last_def_reg);
++}
++
++/* Check dependencies from LMW(N, N) to E2.  */
++bool
++nds32_n13_last_load_to_e2_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx last_def_reg = extract_nth_access_reg (producer, -1);
++
++  return n13_consumed_by_e2_dep_p (consumer, last_def_reg);
++}
++
++/* Check dependencies from LMW(N, N-1) to E2.  */
++bool
++nds32_n13_last_two_load_to_e1_p (rtx_insn *producer, rtx_insn *consumer)
++{
++  rtx last_two_def_reg = extract_nth_access_reg (producer, -2);
++
++  if (last_two_def_reg == NULL_RTX)
++    return false;
++
++  return n13_consumed_by_e1_dep_p (consumer, last_two_def_reg);
++}
+ /* ------------------------------------------------------------------------ */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-predicates.c gcc-8.2.0/gcc/config/nds32/nds32-predicates.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-predicates.c	2018-05-07 04:22:07.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-predicates.c	2019-01-25 15:41:54.217806220 +0100
+@@ -356,8 +356,8 @@
+ }
+ 
+ /* Function to check if 'bclr' instruction can be used with IVAL.  */
+-int
+-nds32_can_use_bclr_p (int ival)
++bool
++nds32_can_use_bclr_p (HOST_WIDE_INT ival)
+ {
+   int one_bit_count;
+   unsigned HOST_WIDE_INT mask = GET_MODE_MASK (SImode);
+@@ -373,8 +373,8 @@
+ }
+ 
+ /* Function to check if 'bset' instruction can be used with IVAL.  */
+-int
+-nds32_can_use_bset_p (int ival)
++bool
++nds32_can_use_bset_p (HOST_WIDE_INT ival)
+ {
+   int one_bit_count;
+   unsigned HOST_WIDE_INT mask = GET_MODE_MASK (SImode);
+@@ -389,8 +389,8 @@
+ }
+ 
+ /* Function to check if 'btgl' instruction can be used with IVAL.  */
+-int
+-nds32_can_use_btgl_p (int ival)
++bool
++nds32_can_use_btgl_p (HOST_WIDE_INT ival)
+ {
+   int one_bit_count;
+   unsigned HOST_WIDE_INT mask = GET_MODE_MASK (SImode);
+@@ -405,8 +405,8 @@
+ }
+ 
+ /* Function to check if 'bitci' instruction can be used with IVAL.  */
+-int
+-nds32_can_use_bitci_p (int ival)
++bool
++nds32_can_use_bitci_p (HOST_WIDE_INT ival)
+ {
+   /* If we are using V3 ISA, we have 'bitci' instruction.
+      Try to see if we can present 'andi' semantic with
+@@ -518,4 +518,117 @@
+ 
+   return val >= lower && val < upper;
+ }
++
++bool
++nds32_const_unspec_p (rtx x)
++{
++  if (GET_CODE (x) == CONST)
++    {
++      x = XEXP (x, 0);
++
++      if (GET_CODE (x) == PLUS)
++	x = XEXP (x, 0);
++
++      if (GET_CODE (x) == UNSPEC)
++	{
++	  switch (XINT (x, 1))
++	    {
++	    case UNSPEC_GOTINIT:
++	    case UNSPEC_GOT:
++	    case UNSPEC_GOTOFF:
++	    case UNSPEC_PLT:
++	    case UNSPEC_TLSGD:
++	    case UNSPEC_TLSLD:
++	    case UNSPEC_TLSIE:
++	    case UNSPEC_TLSLE:
++	      return false;
++	    default:
++	      return true;
++	    }
++	}
++    }
++
++  if (GET_CODE (x) == SYMBOL_REF
++      && SYMBOL_REF_TLS_MODEL (x))
++    return false;
++
++  return true;
++}
++
++HOST_WIDE_INT
++const_vector_to_hwint (rtx op)
++{
++  HOST_WIDE_INT hwint = 0;
++  HOST_WIDE_INT mask;
++  int i;
++  int shift_adv;
++  int shift = 0;
++  int nelem;
++
++  switch (GET_MODE (op))
++    {
++      case E_V2HImode:
++	mask = 0xffff;
++	shift_adv = 16;
++	nelem = 2;
++	break;
++      case E_V4QImode:
++	mask = 0xff;
++	shift_adv = 8;
++	nelem = 4;
++	break;
++      default:
++	gcc_unreachable ();
++    }
++
++  if (TARGET_BIG_ENDIAN)
++    {
++      for (i = 0; i < nelem; ++i)
++	{
++	  HOST_WIDE_INT val = XINT (XVECEXP (op, 0, nelem - i - 1), 0);
++	  hwint |= (val & mask) << shift;
++	  shift = shift + shift_adv;
++	}
++    }
++  else
++    {
++      for (i = 0; i < nelem; ++i)
++	{
++	  HOST_WIDE_INT val = XINT (XVECEXP (op, 0, i), 0);
++	  hwint |= (val & mask) << shift;
++	  shift = shift + shift_adv;
++	}
++    }
++
++  return hwint;
++}
++
++bool
++nds32_valid_CVp5_p (rtx op)
++{
++  HOST_WIDE_INT ival = const_vector_to_hwint (op);
++  return (ival < ((1 << 5) + 16)) && (ival >= (0 + 16));
++}
++
++bool
++nds32_valid_CVs5_p (rtx op)
++{
++  HOST_WIDE_INT ival = const_vector_to_hwint (op);
++  return (ival < (1 << 4)) && (ival >= -(1 << 4));
++}
++
++bool
++nds32_valid_CVs2_p (rtx op)
++{
++  HOST_WIDE_INT ival = const_vector_to_hwint (op);
++  return (ival < (1 << 19)) && (ival >= -(1 << 19));
++}
++
++bool
++nds32_valid_CVhi_p (rtx op)
++{
++  HOST_WIDE_INT ival = const_vector_to_hwint (op);
++  return (ival != 0) && ((ival & 0xfff) == 0);
++}
++
+ /* ------------------------------------------------------------------------ */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-protos.h gcc-8.2.0/gcc/config/nds32/nds32-protos.h
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-protos.h	2018-04-22 11:05:10.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-protos.h	2019-01-25 15:38:32.833242671 +0100
+@@ -69,9 +69,10 @@
+ 
+ /* ------------------------------------------------------------------------ */
+ 
+-/* Auxiliary functions for lwm/smw.  */
++/* Auxiliary functions for manipulation DI mode.  */
+ 
+-extern bool nds32_valid_smw_lwm_base_p (rtx);
++extern rtx nds32_di_high_part_subreg(rtx);
++extern rtx nds32_di_low_part_subreg(rtx);
+ 
+ /* Auxiliary functions for expanding rtl used in nds32-multiple.md.  */
+ 
+@@ -116,6 +117,20 @@
+ extern bool nds32_n9_3r2w_mm_to_ex_p (rtx_insn *, rtx_insn *);
+ extern bool nds32_n9_last_load_to_ex_p (rtx_insn *, rtx_insn *);
+ 
++extern bool nds32_n10_ex_to_ex_p (rtx_insn *, rtx_insn *);
++extern bool nds32_n10_mm_to_ex_p (rtx_insn *, rtx_insn *);
++extern bool nds32_n10_last_load_to_ex_p (rtx_insn *, rtx_insn *);
++
++extern bool nds32_gw_ex_to_ex_p (rtx_insn *, rtx_insn *);
++extern bool nds32_gw_mm_to_ex_p (rtx_insn *, rtx_insn *);
++extern bool nds32_gw_last_load_to_ex_p (rtx_insn *, rtx_insn *);
++
++extern bool nds32_n13_e2_to_e1_p (rtx_insn *, rtx_insn *);
++extern bool nds32_n13_load_to_e1_p (rtx_insn *, rtx_insn *);
++extern bool nds32_n13_load_to_e2_p (rtx_insn *, rtx_insn *);
++extern bool nds32_n13_last_load_to_e1_p (rtx_insn *, rtx_insn *);
++extern bool nds32_n13_last_load_to_e2_p (rtx_insn *, rtx_insn *);
++extern bool nds32_n13_last_two_load_to_e1_p (rtx_insn *, rtx_insn *);
+ 
+ /* Auxiliary functions for stack operation predicate checking.  */
+ 
+@@ -123,24 +138,25 @@
+ 
+ /* Auxiliary functions for bit operation detection.  */
+ 
+-extern int nds32_can_use_bclr_p (int);
+-extern int nds32_can_use_bset_p (int);
+-extern int nds32_can_use_btgl_p (int);
++extern bool nds32_can_use_bclr_p (HOST_WIDE_INT);
++extern bool nds32_can_use_bset_p (HOST_WIDE_INT);
++extern bool nds32_can_use_btgl_p (HOST_WIDE_INT);
+ 
+-extern int nds32_can_use_bitci_p (int);
++extern bool nds32_can_use_bitci_p (HOST_WIDE_INT);
+ 
+ extern bool nds32_const_double_range_ok_p (rtx, machine_mode,
+ 					   HOST_WIDE_INT, HOST_WIDE_INT);
+ 
++extern bool nds32_const_unspec_p (rtx x);
++
+ /* Auxiliary function for 'Computing the Length of an Insn'.  */
+ 
+ extern int nds32_adjust_insn_length (rtx_insn *, int);
+ 
+ /* Auxiliary functions for FP_AS_GP detection.  */
+ 
+-extern int nds32_fp_as_gp_check_available (void);
+-
+ extern bool nds32_symbol_load_store_p (rtx_insn *);
++extern bool nds32_naked_function_p (tree);
+ 
+ /* Auxiliary functions for jump table generation.  */
+ 
+@@ -159,10 +175,50 @@
+ extern enum nds32_expand_result_type nds32_expand_movcc (rtx *);
+ extern void nds32_expand_float_movcc (rtx *);
+ 
++/* Auxiliary functions for expand extv/insv instruction.  */
++
++extern enum nds32_expand_result_type nds32_expand_extv (rtx *);
++extern enum nds32_expand_result_type nds32_expand_insv (rtx *);
++
++/* Auxiliary functions for expand PIC instruction.  */
++
++extern void nds32_expand_pic_move (rtx *);
++
++/* Auxiliary functions to legitimize PIC address.  */
++
++extern rtx nds32_legitimize_pic_address (rtx);
++
++/* Auxiliary functions for expand TLS instruction.  */
++
++extern void nds32_expand_tls_move (rtx *);
++
++/* Auxiliary functions to legitimize TLS address.  */
++
++extern rtx nds32_legitimize_tls_address (rtx);
++
++/* Auxiliary functions to identify thread-local symbol.  */
++
++extern bool nds32_tls_referenced_p (rtx);
++
++/* Auxiliary functions for expand ICT instruction.  */
++
++extern void nds32_expand_ict_move (rtx *);
++
++/* Auxiliary functions to legitimize address for indirect-call symbol.  */
++
++extern rtx nds32_legitimize_ict_address (rtx);
++
++/* Auxiliary functions to identify indirect-call symbol.  */
++
++extern bool nds32_indirect_call_referenced_p (rtx);
+ 
+ /* Auxiliary functions to identify long-call symbol.  */
+ extern bool nds32_long_call_p (rtx);
+ 
++/* Auxiliary functions to identify SYMBOL_REF and LABEL_REF pattern.  */
++
++extern bool symbolic_reference_mentioned_p (rtx);
++
+ /* Auxiliary functions to identify conditional move comparison operand.  */
+ 
+ extern int nds32_cond_move_p (rtx);
+@@ -185,6 +241,7 @@
+ extern const char *nds32_output_float_load(rtx *);
+ extern const char *nds32_output_float_store(rtx *);
+ extern const char *nds32_output_smw_single_word (rtx *);
++extern const char *nds32_output_smw_double_word (rtx *);
+ extern const char *nds32_output_lmw_single_word (rtx *);
+ extern const char *nds32_output_double (rtx *, bool);
+ extern const char *nds32_output_cbranchsi4_equality_zero (rtx_insn *, rtx *);
+@@ -193,9 +250,12 @@
+ 								      rtx *);
+ extern const char *nds32_output_cbranchsi4_greater_less_zero (rtx_insn *, rtx *);
+ 
++extern const char *nds32_output_unpkd8 (rtx, rtx, rtx, rtx, bool);
++
+ extern const char *nds32_output_call (rtx, rtx *, rtx,
+ 				      const char *, const char *, bool);
+-
++extern const char *nds32_output_tls_desc (rtx *);
++extern const char *nds32_output_tls_ie (rtx *);
+ 
+ /* Auxiliary functions to output stack push/pop instruction.  */
+ 
+@@ -203,9 +263,19 @@
+ extern const char *nds32_output_stack_pop (rtx);
+ extern const char *nds32_output_return (void);
+ 
++
++/* Auxiliary functions to split/output sms pattern.  */
++extern bool nds32_need_split_sms_p (rtx, rtx, rtx, rtx);
++extern const char *nds32_output_sms (rtx, rtx, rtx, rtx);
++extern void nds32_split_sms (rtx, rtx, rtx, rtx, rtx, rtx, rtx);
++
+ /* Auxiliary functions to split double word RTX pattern.  */
+ 
+ extern void nds32_spilt_doubleword (rtx *, bool);
++extern void nds32_split_ashiftdi3 (rtx, rtx, rtx);
++extern void nds32_split_ashiftrtdi3 (rtx, rtx, rtx);
++extern void nds32_split_lshiftrtdi3 (rtx, rtx, rtx);
++extern void nds32_split_rotatertdi3 (rtx, rtx, rtx);
+ 
+ /* Auxiliary functions to split large constant RTX pattern.  */
+ 
+@@ -237,15 +307,29 @@
+ extern void nds32_asm_file_start_for_isr (void);
+ extern void nds32_asm_file_end_for_isr (void);
+ extern bool nds32_isr_function_p (tree);
++extern bool nds32_isr_function_critical_p (tree);
+ 
+ /* Auxiliary functions for cost calculation.  */
+ 
++extern void nds32_init_rtx_costs (void);
+ extern bool nds32_rtx_costs_impl (rtx, machine_mode, int, int, int *, bool);
+ extern int nds32_address_cost_impl (rtx, machine_mode, addr_space_t, bool);
+ 
+ /* Auxiliary functions for pre-define marco.  */
+ extern void nds32_cpu_cpp_builtins(struct cpp_reader *);
+ 
++/* Auxiliary functions for const_vector's constraints.  */
++
++extern HOST_WIDE_INT const_vector_to_hwint (rtx);
++extern bool nds32_valid_CVp5_p (rtx);
++extern bool nds32_valid_CVs5_p (rtx);
++extern bool nds32_valid_CVs2_p (rtx);
++extern bool nds32_valid_CVhi_p (rtx);
++
++/* Auxiliary functions for lwm/smw.  */
++
++extern bool nds32_valid_smw_lwm_base_p (rtx);
++
+ extern bool nds32_split_double_word_load_store_p (rtx *,bool);
+ 
+ namespace nds32 {
+@@ -258,11 +342,13 @@
+ bool store_single_p (rtx_insn *);
+ bool load_double_p (rtx_insn *);
+ bool store_double_p (rtx_insn *);
++bool store_offset_reg_p (rtx_insn *);
+ bool post_update_insn_p (rtx_insn *);
+ bool immed_offset_p (rtx);
+ int find_post_update_rtx (rtx_insn *);
+ rtx extract_mem_rtx (rtx_insn *);
+ rtx extract_base_reg (rtx_insn *);
++rtx extract_offset_rtx (rtx_insn *);
+ 
+ rtx extract_shift_reg (rtx);
+ 
+@@ -271,6 +357,8 @@
+ 
+ rtx extract_mac_non_acc_rtx (rtx_insn *);
+ 
++bool divmod_p (rtx_insn *);
++
+ rtx extract_branch_target_rtx (rtx_insn *);
+ rtx extract_branch_condition_rtx (rtx_insn *);
+ } // namespace nds32
+@@ -279,5 +367,6 @@
+ 
+ /* Functions for create nds32 specific optimization pass.  */
+ extern rtl_opt_pass *make_pass_nds32_relax_opt (gcc::context *);
++extern rtl_opt_pass *make_pass_nds32_fp_as_gp (gcc::context *);
+ 
+ /* ------------------------------------------------------------------------ */
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-relax-opt.c gcc-8.2.0/gcc/config/nds32/nds32-relax-opt.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-relax-opt.c	2018-04-01 12:07:40.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-relax-opt.c	2019-01-25 15:38:32.833242671 +0100
+@@ -52,6 +52,8 @@
+ #include "cfgrtl.h"
+ #include "tree-pass.h"
+ 
++using namespace nds32;
++
+ /* This is used to create unique relax hint id value.
+    The initial value is 0.  */
+ static int relax_group_id = 0;
+@@ -185,6 +187,121 @@
+   return false;
+ }
+ 
++/* Return true if x is const and the referance is ict symbol.  */
++static bool
++nds32_ict_const_p (rtx x)
++{
++  if (GET_CODE (x) == CONST)
++    {
++      x = XEXP (x, 0);
++      return nds32_indirect_call_referenced_p (x);
++    }
++  return FALSE;
++}
++
++/* Group the following pattern as relax candidates:
++
++   GOT:
++      sethi	$ra, hi20(sym)
++      ori	$ra, $ra, lo12(sym)
++      lw	$rb, [$ra + $gp]
++
++   GOTOFF, TLSLE:
++      sethi	$ra, hi20(sym)
++      ori	$ra, $ra, lo12(sym)
++      LS	$rb, [$ra + $gp]
++
++   GOTOFF, TLSLE:
++      sethi	$ra, hi20(sym)
++      ori	$ra, $ra, lo12(sym)
++      add	$rb, $ra, $gp($tp)
++
++   Initial GOT table:
++      sethi	$gp,hi20(sym)
++      ori	$gp, $gp, lo12(sym)
++      add5.pc	$gp  */
++
++static auto_vec<rtx_insn *, 32> nds32_group_infos;
++/* Group the PIC and TLS relax candidate instructions for linker.  */
++static bool
++nds32_pic_tls_group (rtx_insn *def_insn,
++		     enum nds32_relax_insn_type relax_type,
++		     int sym_type)
++{
++  df_ref def_record;
++  df_link *link;
++  rtx_insn *use_insn = NULL;
++  rtx pat, new_pat;
++  def_record = DF_INSN_DEFS (def_insn);
++  for (link = DF_REF_CHAIN (def_record); link; link = link->next)
++    {
++      if (!DF_REF_INSN_INFO (link->ref))
++	continue;
++
++      use_insn = DF_REF_INSN (link->ref);
++
++      /* Skip if define insn and use insn not in the same basic block.  */
++      if (!dominated_by_p (CDI_DOMINATORS,
++			   BLOCK_FOR_INSN (use_insn),
++			   BLOCK_FOR_INSN (def_insn)))
++	return FALSE;
++
++      /* Skip if use_insn not active insn.  */
++      if (!active_insn_p (use_insn))
++	return FALSE;
++
++      switch (relax_type)
++	{
++	case RELAX_ORI:
++
++	  /* GOTOFF, TLSLE:
++	     sethi	$ra, hi20(sym)
++	     ori	$ra, $ra, lo12(sym)
++	     add	$rb, $ra, $gp($tp)  */
++	  if ((sym_type == UNSPEC_TLSLE
++	       || sym_type == UNSPEC_GOTOFF)
++	      && (recog_memoized (use_insn) == CODE_FOR_addsi3))
++	    {
++	      pat = XEXP (PATTERN (use_insn), 1);
++	      new_pat =
++		gen_rtx_UNSPEC (SImode,
++				gen_rtvec (2, XEXP (pat, 0), XEXP (pat, 1)),
++				UNSPEC_ADD32);
++	      validate_replace_rtx (pat, new_pat, use_insn);
++	      nds32_group_infos.safe_push (use_insn);
++	    }
++	  else if (nds32_plus_reg_load_store_p (use_insn)
++		   && !nds32_sp_base_or_plus_load_store_p (use_insn))
++	    nds32_group_infos.safe_push (use_insn);
++	  else
++	    return FALSE;
++	  break;
++
++	default:
++	  return FALSE;
++	}
++    }
++  return TRUE;
++}
++
++static int
++nds32_pic_tls_symbol_type (rtx x)
++{
++  x = XEXP (SET_SRC (PATTERN (x)), 1);
++
++  if (GET_CODE (x) == CONST)
++    {
++      x = XEXP (x, 0);
++
++      if (GET_CODE (x) == PLUS)
++	x = XEXP (x, 0);
++
++      return XINT (x, 1);
++    }
++
++  return XINT (x, 1);
++}
++
+ /* Group the relax candidates with group id.  */
+ static void
+ nds32_group_insns (rtx sethi)
+@@ -193,6 +310,7 @@
+   df_link *link;
+   rtx_insn *use_insn = NULL;
+   rtx group_id;
++  bool valid;
+ 
+   def_record = DF_INSN_DEFS (sethi);
+ 
+@@ -242,6 +360,132 @@
+       /* Insert .relax_* directive.  */
+       if (active_insn_p (use_insn))
+ 	emit_insn_before (gen_relax_group (group_id), use_insn);
++
++      /* Find ori ra, ra, unspec(symbol) instruction.  */
++      if (use_insn != NULL
++	  && recog_memoized (use_insn) == CODE_FOR_lo_sum
++	  && !nds32_const_unspec_p (XEXP (SET_SRC (PATTERN (use_insn)), 1)))
++	{
++	  int sym_type = nds32_pic_tls_symbol_type (use_insn);
++	  valid = nds32_pic_tls_group (use_insn, RELAX_ORI, sym_type);
++
++	  /* Insert .relax_* directive.  */
++	  while (!nds32_group_infos.is_empty ())
++	    {
++	      use_insn = nds32_group_infos.pop ();
++	      if (valid)
++		emit_insn_before (gen_relax_group (group_id), use_insn);
++	    }
++	}
++    }
++
++  relax_group_id++;
++}
++
++/* Convert relax group id in rtl.  */
++
++static void
++nds32_group_tls_insn (rtx insn)
++{
++  rtx pat = PATTERN (insn);
++  rtx unspec_relax_group = XEXP (XVECEXP (pat, 0, 1), 0);
++
++  while (GET_CODE (pat) != SET && GET_CODE (pat) == PARALLEL)
++    {
++      pat = XVECEXP (pat, 0, 0);
++    }
++
++  if (GET_CODE (unspec_relax_group) == UNSPEC
++      && XINT (unspec_relax_group, 1) == UNSPEC_VOLATILE_RELAX_GROUP)
++    {
++      XVECEXP (unspec_relax_group, 0, 0) = GEN_INT (relax_group_id);
++    }
++
++  relax_group_id++;
++}
++
++static bool
++nds32_float_reg_load_store_p (rtx_insn *insn)
++{
++  rtx pat = PATTERN (insn);
++
++  if (get_attr_type (insn) == TYPE_FLOAD
++      && GET_CODE (pat) == SET
++      && (GET_MODE (XEXP (pat, 0)) == SFmode
++	  || GET_MODE (XEXP (pat, 0)) == DFmode)
++      && MEM_P (XEXP (pat, 1)))
++    {
++      rtx addr = XEXP (XEXP (pat, 1), 0);
++
++      /* [$ra] */
++      if (REG_P (addr))
++	return true;
++      /* [$ra + offset] */
++      if (GET_CODE (addr) == PLUS
++	  && REG_P (XEXP (addr, 0))
++	  && CONST_INT_P (XEXP (addr, 1)))
++	return true;
++    }
++  return false;
++}
++
++
++/* Group float load-store instructions:
++   la $ra, symbol
++   flsi $rt, [$ra + offset] */
++
++static void
++nds32_group_float_insns (rtx insn)
++{
++  df_ref def_record, use_record;
++  df_link *link;
++  rtx_insn *use_insn = NULL;
++  rtx group_id;
++
++  def_record = DF_INSN_DEFS (insn);
++
++  for (link = DF_REF_CHAIN (def_record); link; link = link->next)
++    {
++      if (!DF_REF_INSN_INFO (link->ref))
++	continue;
++
++      use_insn = DF_REF_INSN (link->ref);
++
++      /* Skip if define insn and use insn not in the same basic block.  */
++      if (!dominated_by_p (CDI_DOMINATORS,
++			   BLOCK_FOR_INSN (use_insn),
++			   BLOCK_FOR_INSN (insn)))
++	return;
++
++      /* Skip if the low-part used register is from different high-part
++	 instructions.  */
++      use_record = DF_INSN_USES (use_insn);
++      if (DF_REF_CHAIN (use_record) && DF_REF_CHAIN (use_record)->next)
++	return;
++
++      /* Skip if use_insn not active insn.  */
++      if (!active_insn_p (use_insn))
++	return;
++
++      if (!nds32_float_reg_load_store_p (use_insn)
++	  || find_post_update_rtx (use_insn) != -1)
++	return;
++    }
++
++  group_id = GEN_INT (relax_group_id);
++  /* Insert .relax_* directive for insn.  */
++  emit_insn_before (gen_relax_group (group_id), insn);
++
++  /* Scan the use insns and insert the directive.  */
++  for (link = DF_REF_CHAIN (def_record); link; link = link->next)
++    {
++      if (!DF_REF_INSN_INFO (link->ref))
++	continue;
++
++      use_insn = DF_REF_INSN (link->ref);
++
++      /* Insert .relax_* directive.  */
++	emit_insn_before (gen_relax_group (group_id), use_insn);
+     }
+ 
+   relax_group_id++;
+@@ -271,8 +515,21 @@
+ 	  /* Find sethi ra, symbol  instruction.  */
+ 	  if (recog_memoized (insn) == CODE_FOR_sethi
+ 	      && nds32_symbolic_operand (XEXP (SET_SRC (PATTERN (insn)), 0),
+-					 SImode))
++					 SImode)
++	      && !nds32_ict_const_p (XEXP (SET_SRC (PATTERN (insn)), 0)))
+ 	    nds32_group_insns (insn);
++	  else if (recog_memoized (insn) == CODE_FOR_tls_ie)
++	    nds32_group_tls_insn (insn);
++	  else if (TARGET_FPU_SINGLE
++		   && recog_memoized (insn) == CODE_FOR_move_addr
++		   && !nds32_ict_const_p (XEXP (SET_SRC (PATTERN (insn)), 0)))
++	    {
++	      nds32_group_float_insns (insn);
++	    }
++	}
++      else if (CALL_P (insn) && recog_memoized (insn) == CODE_FOR_tls_desc)
++	{
++	  nds32_group_tls_insn (insn);
+ 	}
+     }
+ 
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/nds32-utils.c gcc-8.2.0/gcc/config/nds32/nds32-utils.c
+--- gcc-8.2.0.orig/gcc/config/nds32/nds32-utils.c	2018-04-08 10:31:52.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/nds32-utils.c	2019-01-25 15:38:32.833242671 +0100
+@@ -142,6 +142,23 @@
+   return true;
+ }
+ 
++bool
++store_offset_reg_p (rtx_insn *insn)
++{
++  if (get_attr_type (insn) != TYPE_STORE)
++    return false;
++
++  rtx offset_rtx = extract_offset_rtx (insn);
++
++  if (offset_rtx == NULL_RTX)
++    return false;
++
++  if (REG_P (offset_rtx))
++    return true;
++
++  return false;
++}
++
+ /* Determine if INSN is a post update insn.  */
+ bool
+ post_update_insn_p (rtx_insn *insn)
+@@ -316,22 +333,114 @@
+   if (REG_P (XEXP (mem_rtx, 0)))
+     return XEXP (mem_rtx, 0);
+ 
++  /* (mem (lo_sum (reg) (symbol_ref)) */
++  if (GET_CODE (XEXP (mem_rtx, 0)) == LO_SUM)
++    return XEXP (XEXP (mem_rtx, 0), 0);
++
+   plus_rtx = XEXP (mem_rtx, 0);
+ 
+   if (GET_CODE (plus_rtx) == SYMBOL_REF
+       || GET_CODE (plus_rtx) == CONST)
+     return NULL_RTX;
+ 
+-  gcc_assert (GET_CODE (plus_rtx) == PLUS
+-	      || GET_CODE (plus_rtx) == POST_INC
+-	      || GET_CODE (plus_rtx) == POST_DEC
+-	      || GET_CODE (plus_rtx) == POST_MODIFY);
+-  gcc_assert (REG_P (XEXP (plus_rtx, 0)));
+   /* (mem (plus (reg) (const_int))) or
++     (mem (plus (mult (reg) (const_int 4)) (reg))) or
+      (mem (post_inc (reg))) or
+      (mem (post_dec (reg))) or
+      (mem (post_modify (reg) (plus (reg) (reg))))  */
+-  return XEXP (plus_rtx, 0);
++  gcc_assert (GET_CODE (plus_rtx) == PLUS
++	      || GET_CODE (plus_rtx) == POST_INC
++	      || GET_CODE (plus_rtx) == POST_DEC
++	      || GET_CODE (plus_rtx) == POST_MODIFY);
++
++  if (REG_P (XEXP (plus_rtx, 0)))
++    return XEXP (plus_rtx, 0);
++
++  gcc_assert (REG_P (XEXP (plus_rtx, 1)));
++  return XEXP (plus_rtx, 1);
++}
++
++/* Extract the offset rtx from load/store insns.  The function returns
++   NULL_RTX if offset is absent.  */
++rtx
++extract_offset_rtx (rtx_insn *insn)
++{
++  rtx mem_rtx;
++  rtx plus_rtx;
++  rtx offset_rtx;
++
++  /* Find the MEM rtx.  The multiple load/store insns doens't have
++     the offset field so we can return NULL_RTX here.  */
++  switch (get_attr_type (insn))
++    {
++    case TYPE_LOAD_MULTIPLE:
++    case TYPE_STORE_MULTIPLE:
++      return NULL_RTX;
++
++    case TYPE_LOAD:
++    case TYPE_FLOAD:
++    case TYPE_STORE:
++    case TYPE_FSTORE:
++      mem_rtx = extract_mem_rtx (insn);
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
++  gcc_assert (MEM_P (mem_rtx));
++
++  /* (mem (reg))  */
++  if (REG_P (XEXP (mem_rtx, 0)))
++    return NULL_RTX;
++
++  plus_rtx = XEXP (mem_rtx, 0);
++
++  switch (GET_CODE (plus_rtx))
++    {
++    case SYMBOL_REF:
++    case CONST:
++    case POST_INC:
++    case POST_DEC:
++      return NULL_RTX;
++
++    case PLUS:
++      /* (mem (plus (reg) (const_int))) or
++         (mem (plus (mult (reg) (const_int 4)) (reg))) */
++      if (REG_P (XEXP (plus_rtx, 0)))
++        offset_rtx = XEXP (plus_rtx, 1);
++      else
++	{
++	  gcc_assert (REG_P (XEXP (plus_rtx, 1)));
++	  offset_rtx = XEXP (plus_rtx, 0);
++	}
++
++      if (ARITHMETIC_P (offset_rtx))
++	{
++	  gcc_assert (GET_CODE (offset_rtx) == MULT);
++	  gcc_assert (REG_P (XEXP (offset_rtx, 0)));
++	  offset_rtx = XEXP (offset_rtx, 0);
++	}
++      break;
++
++    case LO_SUM:
++      /* (mem (lo_sum (reg) (symbol_ref)) */
++      offset_rtx = XEXP (plus_rtx, 1);
++      break;
++
++    case POST_MODIFY:
++      /* (mem (post_modify (reg) (plus (reg) (reg / const_int)))) */
++      gcc_assert (REG_P (XEXP (plus_rtx, 0)));
++      plus_rtx = XEXP (plus_rtx, 1);
++      gcc_assert (GET_CODE (plus_rtx) == PLUS);
++      offset_rtx = XEXP (plus_rtx, 0);
++      break;
++
++    default:
++      gcc_unreachable ();
++    }
++
++  return offset_rtx;
+ }
+ 
+ /* Extract the register of the shift operand from an ALU_SHIFT rtx.  */
+@@ -413,6 +522,7 @@
+   switch (get_attr_type (insn))
+     {
+     case TYPE_MAC:
++    case TYPE_DMAC:
+       if (REG_P (XEXP (exp, 0)))
+ 	return XEXP (exp, 1);
+       else
+@@ -423,6 +533,19 @@
+     }
+ }
+ 
++/* Check if the DIV insn needs two write ports.  */
++bool
++divmod_p (rtx_insn *insn)
++{
++  gcc_assert (get_attr_type (insn) == TYPE_DIV);
++
++  if (INSN_CODE (insn) == CODE_FOR_divmodsi4
++      || INSN_CODE (insn) == CODE_FOR_udivmodsi4)
++    return true;
++
++  return false;
++}
++
+ /* Extract the rtx representing the branch target to help recognize
+    data hazards.  */
+ rtx
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/pipelines.md gcc-8.2.0/gcc/config/nds32/pipelines.md
+--- gcc-8.2.0.orig/gcc/config/nds32/pipelines.md	2018-04-08 11:21:30.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/pipelines.md	2019-01-25 15:38:32.833242671 +0100
+@@ -44,6 +44,24 @@
+ 
+ 
+ ;; ------------------------------------------------------------------------
++;; Include N10 pipeline settings.
++;; ------------------------------------------------------------------------
++(include "nds32-n10.md")
++
++
++;; ------------------------------------------------------------------------
++;; Include Graywolf pipeline settings.
++;; ------------------------------------------------------------------------
++(include "nds32-graywolf.md")
++
++
++;; ------------------------------------------------------------------------
++;; Include N12/N13 pipeline settings.
++;; ------------------------------------------------------------------------
++(include "nds32-n13.md")
++
++
++;; ------------------------------------------------------------------------
+ ;; Define simple pipeline settings.
+ ;; ------------------------------------------------------------------------
+ 
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/predicates.md gcc-8.2.0/gcc/config/nds32/predicates.md
+--- gcc-8.2.0.orig/gcc/config/nds32/predicates.md	2018-04-06 07:51:33.000000000 +0200
++++ gcc-8.2.0/gcc/config/nds32/predicates.md	2019-01-25 15:38:32.833242671 +0100
+@@ -40,7 +40,15 @@
+   (match_code "mult,and,ior,xor"))
+ 
+ (define_predicate "nds32_symbolic_operand"
+-  (match_code "const,symbol_ref,label_ref"))
++  (and (match_code "const,symbol_ref,label_ref")
++       (match_test "!(TARGET_ICT_MODEL_LARGE
++		      && nds32_indirect_call_referenced_p (op))")))
++
++(define_predicate "nds32_nonunspec_symbolic_operand"
++  (and (match_code "const,symbol_ref,label_ref")
++       (match_test "!flag_pic && nds32_const_unspec_p (op)
++		    && !(TARGET_ICT_MODEL_LARGE
++			 && nds32_indirect_call_referenced_p (op))")))
+ 
+ (define_predicate "nds32_reg_constant_operand"
+   (ior (match_operand 0 "register_operand")
+@@ -56,14 +64,51 @@
+        (and (match_operand 0 "const_int_operand")
+ 	    (match_test "satisfies_constraint_Is11 (op)"))))
+ 
++(define_predicate "nds32_imm_0_1_operand"
++  (and (match_operand 0 "const_int_operand")
++       (ior (match_test "satisfies_constraint_Iv00 (op)")
++	    (match_test "satisfies_constraint_Iv01 (op)"))))
++
++(define_predicate "nds32_imm_1_2_operand"
++  (and (match_operand 0 "const_int_operand")
++       (ior (match_test "satisfies_constraint_Iv01 (op)")
++	    (match_test "satisfies_constraint_Iv02 (op)"))))
++
++(define_predicate "nds32_imm_1_2_4_8_operand"
++  (and (match_operand 0 "const_int_operand")
++       (ior (ior (match_test "satisfies_constraint_Iv01 (op)")
++		 (match_test "satisfies_constraint_Iv02 (op)"))
++	    (ior (match_test "satisfies_constraint_Iv04 (op)")
++		 (match_test "satisfies_constraint_Iv08 (op)")))))
++
++(define_predicate "nds32_imm2u_operand"
++  (and (match_operand 0 "const_int_operand")
++       (match_test "satisfies_constraint_Iu02 (op)")))
++
++(define_predicate "nds32_imm4u_operand"
++  (and (match_operand 0 "const_int_operand")
++       (match_test "satisfies_constraint_Iu04 (op)")))
++
+ (define_predicate "nds32_imm5u_operand"
+   (and (match_operand 0 "const_int_operand")
+        (match_test "satisfies_constraint_Iu05 (op)")))
+ 
++(define_predicate "nds32_imm6u_operand"
++  (and (match_operand 0 "const_int_operand")
++       (match_test "satisfies_constraint_Iu06 (op)")))
++
++(define_predicate "nds32_rimm4u_operand"
++  (ior (match_operand 0 "register_operand")
++       (match_operand 0 "nds32_imm4u_operand")))
++
+ (define_predicate "nds32_rimm5u_operand"
+   (ior (match_operand 0 "register_operand")
+        (match_operand 0 "nds32_imm5u_operand")))
+ 
++(define_predicate "nds32_rimm6u_operand"
++  (ior (match_operand 0 "register_operand")
++       (match_operand 0 "nds32_imm6u_operand")))
++
+ (define_predicate "nds32_move_operand"
+   (and (match_operand 0 "general_operand")
+        (not (match_code "high,const,symbol_ref,label_ref")))
+@@ -78,6 +123,20 @@
+   return true;
+ })
+ 
++(define_predicate "nds32_vmove_operand"
++  (and (match_operand 0 "general_operand")
++       (not (match_code "high,const,symbol_ref,label_ref")))
++{
++  /* If the constant op does NOT satisfy Is20 nor Ihig,
++     we can not perform move behavior by a single instruction.  */
++  if (GET_CODE (op) == CONST_VECTOR
++      && !satisfies_constraint_CVs2 (op)
++      && !satisfies_constraint_CVhi (op))
++    return false;
++
++  return true;
++})
++
+ (define_predicate "nds32_and_operand"
+   (match_operand 0 "nds32_reg_constant_operand")
+ {
+@@ -127,6 +186,15 @@
+   (ior (match_operand 0 "nds32_symbolic_operand")
+        (match_operand 0 "nds32_general_register_operand")))
+ 
++(define_predicate "nds32_insv_operand"
++  (match_code "const_int")
++{
++  return INTVAL (op) == 0
++	 || INTVAL (op) == 8
++	 || INTVAL (op) == 16
++	 || INTVAL (op) == 24;
++})
++
+ (define_predicate "nds32_lmw_smw_base_operand"
+   (and (match_code "mem")
+        (match_test "nds32_valid_smw_lwm_base_p (op)")))
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/t-elf gcc-8.2.0/gcc/config/nds32/t-elf
+--- gcc-8.2.0.orig/gcc/config/nds32/t-elf	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/t-elf	2019-01-25 15:38:32.833242671 +0100
+@@ -0,0 +1,42 @@
++# The multilib settings of Andes NDS32 cpu for GNU compiler
++# Copyright (C) 2012-2018 Free Software Foundation, Inc.
++# Contributed by Andes Technology Corporation.
++#
++# This file is part of GCC.
++#
++# GCC is free software; you can redistribute it and/or modify it
++# under the terms of the GNU General Public License as published
++# by the Free Software Foundation; either version 3, or (at your
++# option) any later version.
++#
++# GCC is distributed in the hope that it will be useful, but WITHOUT
++# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++# or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++# License for more details.
++#
++# You should have received a copy of the GNU General Public License
++# along with GCC; see the file COPYING3.  If not see
++# <http://www.gnu.org/licenses/>.
++
++# We also define a macro MULTILIB_DEFAULTS in nds32.h that tells the
++# driver program which options are defaults for this target and thus
++# do not need to be handled specially.
++MULTILIB_OPTIONS += mcmodel=small/mcmodel=medium/mcmodel=large mvh
++
++ifneq ($(filter graywolf,$(TM_MULTILIB_CONFIG)),)
++MULTILIB_OPTIONS += mcpu=graywolf
++endif
++
++ifneq ($(filter dsp,$(TM_MULTILIB_CONFIG)),)
++MULTILIB_OPTIONS += mext-dsp
++endif
++
++ifneq ($(filter zol,$(TM_MULTILIB_CONFIG)),)
++MULTILIB_OPTIONS += mext-zol
++endif
++
++ifneq ($(filter v3m+,$(TM_MULTILIB_CONFIG)),)
++MULTILIB_OPTIONS += march=v3m+
++endif
++
++# ------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/gcc/config/nds32/t-linux gcc-8.2.0/gcc/config/nds32/t-linux
+--- gcc-8.2.0.orig/gcc/config/nds32/t-linux	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/config/nds32/t-linux	2019-01-25 15:38:32.833242671 +0100
+@@ -0,0 +1,26 @@
++# The multilib settings of Andes NDS32 cpu for GNU compiler
++# Copyright (C) 2012-2018 Free Software Foundation, Inc.
++# Contributed by Andes Technology Corporation.
++#
++# This file is part of GCC.
++#
++# GCC is free software; you can redistribute it and/or modify it
++# under the terms of the GNU General Public License as published
++# by the Free Software Foundation; either version 3, or (at your
++# option) any later version.
++#
++# GCC is distributed in the hope that it will be useful, but WITHOUT
++# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++# or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++# License for more details.
++#
++# You should have received a copy of the GNU General Public License
++# along with GCC; see the file COPYING3.  If not see
++# <http://www.gnu.org/licenses/>.
++
++# We also define a macro MULTILIB_DEFAULTS in nds32.h that tells the
++# driver program which options are defaults for this target and thus
++# do not need to be handled specially.
++MULTILIB_OPTIONS +=
++
++# ------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/gcc/config.gcc gcc-8.2.0/gcc/config.gcc
+--- gcc-8.2.0.orig/gcc/config.gcc	2018-06-25 21:34:01.000000000 +0200
++++ gcc-8.2.0/gcc/config.gcc	2019-01-25 15:38:32.821242637 +0100
+@@ -445,7 +445,17 @@
+ 	;;
+ nds32*)
+ 	cpu_type=nds32
+-	extra_headers="nds32_intrinsic.h"
++	extra_headers="nds32_intrinsic.h nds32_isr.h nds32_init.inc"
++	case ${target} in
++	  nds32*-*-linux*)
++	    extra_options="${extra_options} nds32/nds32-linux.opt"
++	    ;;
++	  nds32*-*-elf*)
++	    extra_options="${extra_options} nds32/nds32-elf.opt"
++	    ;;
++	  *)
++	    ;;
++	esac
+ 	extra_objs="nds32-cost.o nds32-intrinsic.o nds32-isr.o nds32-md-auxiliary.o nds32-pipelines-auxiliary.o nds32-predicates.o nds32-memory-manipulation.o nds32-fp-as-gp.o nds32-relax-opt.o nds32-utils.o"
+ 	;;
+ nios2-*-*)
+@@ -2335,17 +2345,36 @@
+ 	tmake_file="${tmake_file} msp430/t-msp430"
+ 	extra_gcc_objs="driver-msp430.o"
+ 	;;
+-nds32le-*-*)
++nds32*-*-*)
+ 	target_cpu_default="0"
+ 	tm_defines="${tm_defines}"
+-	tm_file="dbxelf.h elfos.h newlib-stdint.h ${tm_file} nds32/nds32_intrinsic.h"
+-	tmake_file="nds32/t-nds32 nds32/t-mlibs"
+-	;;
+-nds32be-*-*)
+-	target_cpu_default="0|MASK_BIG_ENDIAN"
+-	tm_defines="${tm_defines} TARGET_BIG_ENDIAN_DEFAULT=1"
+-	tm_file="dbxelf.h elfos.h newlib-stdint.h ${tm_file} nds32/nds32_intrinsic.h"
+-	tmake_file="nds32/t-nds32 nds32/t-mlibs"
++	case ${target} in
++	  nds32le*-*-*)
++	    ;;
++	  nds32be-*-*)
++	    target_cpu_default="${target_cpu_default}|MASK_BIG_ENDIAN"
++	    tm_defines="${tm_defines} TARGET_BIG_ENDIAN_DEFAULT=1"
++	    ;;
++	esac
++	case ${target} in
++	  nds32*-*-elf*)
++	    tm_file="dbxelf.h elfos.h newlib-stdint.h ${tm_file} nds32/elf.h nds32/nds32_intrinsic.h"
++	    tmake_file="nds32/t-nds32 nds32/t-elf"
++	    ;;
++	  nds32*-*-linux*)
++	    tm_file="dbxelf.h elfos.h ${tm_file} gnu-user.h linux.h glibc-stdint.h nds32/linux.h nds32/nds32_intrinsic.h"
++	    tmake_file="${tmake_file} nds32/t-nds32 nds32/t-linux"
++	    ;;
++	esac
++
++	# Handle --enable-default-relax setting.
++	if test x${enable_default_relax} = xyes; then
++		tm_defines="${tm_defines} TARGET_DEFAULT_RELAX=1"
++	fi
++	# Handle --with-ext-dsp
++	if test x${with_ext_dsp} = xyes; then
++		tm_defines="${tm_defines} TARGET_DEFAULT_EXT_DSP=1"
++	fi
+ 	;;
+ nios2-*-*)
+ 	tm_file="elfos.h ${tm_file}"
+@@ -4318,11 +4347,11 @@
+ 		"")
+ 			with_cpu=n9
+ 			;;
+-		n6 | n7 | n8 | e8 | s8 | n9)
++		n6 | n7 |n8 | e8 | s8 | n9 | n10 | d10 | n12 | n13 | n15)
+ 			# OK
+ 			;;
+ 		*)
+-			echo "Cannot accept --with-cpu=$with_cpu, available values are: n6 n7 n8 e8 s8 n9" 1>&2
++			echo "Cannot accept --with-cpu=$with_cpu, available values are: n6 n7 n8 e8 s8 n9 n10 d10 n12 n13 n15" 1>&2
+ 			exit 1
+ 			;;
+ 		esac
+@@ -4332,15 +4361,30 @@
+ 		"")
+ 			# the default library is newlib
+ 			with_nds32_lib=newlib
++			tm_defines="${tm_defines} TARGET_DEFAULT_CTOR_DTOR=1"
+ 			;;
+ 		newlib)
+ 			# OK
++			tm_defines="${tm_defines} TARGET_DEFAULT_CTOR_DTOR=1"
+ 			;;
+ 		mculib)
+ 			# OK
++			# for the arch=v3f or arch=v3s under mculib toolchain,
++			# we would like to set -fno-math-errno as default
++			case "${with_arch}" in
++			v3f | v3s)
++				tm_defines="${tm_defines} TARGET_DEFAULT_NO_MATH_ERRNO=1"
++				;;
++			esac
++			;;
++		glibc)
++			# OK
++			tm_defines="${tm_defines}"
++			;;
++		uclibc)
+ 			;;
+ 		*)
+-			echo "Cannot accept --with-nds32-lib=$with_nds32_lib, available values are: newlib mculib" 1>&2
++			echo "Cannot accept --with-nds32-lib=$with_nds32_lib, available values are: newlib mculib glibc uclibc" 1>&2
+ 			exit 1
+ 			;;
+ 		esac
+diff -urN gcc-8.2.0.orig/gcc/configure gcc-8.2.0/gcc/configure
+--- gcc-8.2.0.orig/gcc/configure	2018-04-18 11:46:58.000000000 +0200
++++ gcc-8.2.0/gcc/configure	2019-01-25 15:38:32.837242683 +0100
+@@ -27784,7 +27784,7 @@
+ # version to the per-target configury.
+ case "$cpu_type" in
+   aarch64 | alpha | arc | arm | avr | bfin | cris | i386 | m32c | m68k \
+-  | microblaze | mips | nios2 | pa | riscv | rs6000 | score | sparc | spu \
++  | microblaze | mips | nds32 | nios2 | pa | riscv | rs6000 | score | sparc | spu \
+   | tilegx | tilepro | visium | xstormy16 | xtensa)
+     insn="nop"
+     ;;
+diff -urN gcc-8.2.0.orig/gcc/configure.ac gcc-8.2.0/gcc/configure.ac
+--- gcc-8.2.0.orig/gcc/configure.ac	2018-04-18 11:46:58.000000000 +0200
++++ gcc-8.2.0/gcc/configure.ac	2019-01-25 15:38:32.837242683 +0100
+@@ -4910,7 +4910,7 @@
+ # version to the per-target configury.
+ case "$cpu_type" in
+   aarch64 | alpha | arc | arm | avr | bfin | cris | i386 | m32c | m68k \
+-  | microblaze | mips | nios2 | pa | riscv | rs6000 | score | sparc | spu \
++  | microblaze | mips | nds32 | nios2 | pa | riscv | rs6000 | score | sparc | spu \
+   | tilegx | tilepro | visium | xstormy16 | xtensa)
+     insn="nop"
+     ;;
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.c-torture/execute/20010122-1.c gcc-8.2.0/gcc/testsuite/gcc.c-torture/execute/20010122-1.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.c-torture/execute/20010122-1.c	2015-12-10 20:20:14.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.c-torture/execute/20010122-1.c	2019-01-25 15:38:32.837242683 +0100
+@@ -1,4 +1,5 @@
+ /* { dg-skip-if "requires frame pointers" { *-*-* } "-fomit-frame-pointer" "" } */
++/* { dg-additional-options "-malways-save-lp" { target nds32*-*-* } } */
+ /* { dg-require-effective-target return_address } */
+ 
+ extern void exit (int);
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.dg/lower-subreg-1.c gcc-8.2.0/gcc/testsuite/gcc.dg/lower-subreg-1.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.dg/lower-subreg-1.c	2017-06-17 17:32:28.000000000 +0200
++++ gcc-8.2.0/gcc/testsuite/gcc.dg/lower-subreg-1.c	2019-01-25 15:38:32.837242683 +0100
+@@ -1,4 +1,4 @@
+-/* { dg-do compile { target { ! { mips64 || { aarch64*-*-* arm*-*-* ia64-*-* sparc*-*-* spu-*-* tilegx-*-* } } } } } */
++/* { dg-do compile { target { ! { mips64 || { aarch64*-*-* arm*-*-* ia64-*-* nds32*-*-* sparc*-*-* spu-*-* tilegx-*-* } } } } } */
+ /* { dg-options "-O -fdump-rtl-subreg1" } */
+ /* { dg-additional-options "-mno-stv" { target ia32 } } */
+ /* { dg-skip-if "" { { i?86-*-* x86_64-*-* } && x32 } } */
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.dg/stack-usage-1.c gcc-8.2.0/gcc/testsuite/gcc.dg/stack-usage-1.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.dg/stack-usage-1.c	2017-06-17 17:32:28.000000000 +0200
++++ gcc-8.2.0/gcc/testsuite/gcc.dg/stack-usage-1.c	2019-01-25 15:38:32.837242683 +0100
+@@ -2,6 +2,7 @@
+ /* { dg-options "-fstack-usage" } */
+ /* nvptx doesn't have a reg allocator, and hence no stack usage data.  */
+ /* { dg-skip-if "" { nvptx-*-* } } */
++/* { dg-options "-fstack-usage -fno-omit-frame-pointer" { target { nds32*-*-* } } } */
+ 
+ /* This is aimed at testing basic support for -fstack-usage in the back-ends.
+    See the SPARC back-end for example (grep flag_stack_usage_info in sparc.c).
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-isb.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-isb.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-isb.c	2013-12-03 11:58:05.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-isb.c	1970-01-01 01:00:00.000000000 +0100
+@@ -1,11 +0,0 @@
+-/* Verify that we generate isb instruction with builtin function.  */
+-
+-/* { dg-do compile }  */
+-/* { dg-options "-O0" }  */
+-/* { dg-final { scan-assembler "\\tisb" } }  */
+-
+-void
+-test (void)
+-{
+-  __builtin_nds32_isb ();
+-}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-isync.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-isync.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-isync.c	2013-12-03 11:58:05.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-isync.c	1970-01-01 01:00:00.000000000 +0100
+@@ -1,12 +0,0 @@
+-/* Verify that we generate isync instruction with builtin function.  */
+-
+-/* { dg-do compile }  */
+-/* { dg-options "-O0" }  */
+-/* { dg-final { scan-assembler "\\tisync" } }  */
+-
+-void
+-test (void)
+-{
+-  int *addr = (int *) 0x53000000;
+-  __builtin_nds32_isync (addr);
+-}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-mfsr-mtsr.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-mfsr-mtsr.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-mfsr-mtsr.c	2013-12-03 11:58:05.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-mfsr-mtsr.c	1970-01-01 01:00:00.000000000 +0100
+@@ -1,17 +0,0 @@
+-/* Verify that we generate mfsr/mtsr instruction with builtin function.  */
+-
+-/* { dg-do compile }  */
+-/* { dg-options "-O0" }  */
+-/* { dg-final { scan-assembler "\\tmfsr" } }  */
+-/* { dg-final { scan-assembler "\\tmtsr" } }  */
+-
+-#include <nds32_intrinsic.h>
+-
+-void
+-test (void)
+-{
+-  int ipsw_value;
+-
+-  ipsw_value = __builtin_nds32_mfsr (__NDS32_REG_IPSW__);
+-  __builtin_nds32_mtsr (ipsw_value, __NDS32_REG_IPSW__);
+-}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-mfusr-mtusr.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-mfusr-mtusr.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-mfusr-mtusr.c	2013-12-03 11:58:05.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-mfusr-mtusr.c	1970-01-01 01:00:00.000000000 +0100
+@@ -1,17 +0,0 @@
+-/* Verify that we generate mfusr/mtusr instruction with builtin function.  */
+-
+-/* { dg-do compile }  */
+-/* { dg-options "-O0" }  */
+-/* { dg-final { scan-assembler "\\tmfusr" } }  */
+-/* { dg-final { scan-assembler "\\tmtusr" } }  */
+-
+-#include <nds32_intrinsic.h>
+-
+-void
+-test (void)
+-{
+-  int itype_value;
+-
+-  itype_value = __builtin_nds32_mfusr (__NDS32_REG_ITYPE__);
+-  __builtin_nds32_mtusr (itype_value, __NDS32_REG_ITYPE__);
+-}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-setgie-dis.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-setgie-dis.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-setgie-dis.c	2013-12-03 11:58:05.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-setgie-dis.c	1970-01-01 01:00:00.000000000 +0100
+@@ -1,11 +0,0 @@
+-/* Verify that we generate setgie.d instruction with builtin function.  */
+-
+-/* { dg-do compile }  */
+-/* { dg-options "-O0" }  */
+-/* { dg-final { scan-assembler "\\tsetgie.d" } }  */
+-
+-void
+-test (void)
+-{
+-  __builtin_nds32_setgie_dis ();
+-}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-setgie-en.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-setgie-en.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-setgie-en.c	2013-12-03 11:58:05.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-setgie-en.c	1970-01-01 01:00:00.000000000 +0100
+@@ -1,11 +0,0 @@
+-/* Verify that we generate setgie.e instruction with builtin function.  */
+-
+-/* { dg-do compile }  */
+-/* { dg-options "-O0" }  */
+-/* { dg-final { scan-assembler "\\tsetgie.e" } }  */
+-
+-void
+-test (void)
+-{
+-  __builtin_nds32_setgie_en ();
+-}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-setgie_mtsr_mfsr.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-setgie_mtsr_mfsr.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/builtin-setgie_mtsr_mfsr.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/builtin-setgie_mtsr_mfsr.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,36 @@
++/* This is a test program for checking gie with
++   mtsr/mfsr instruction.  */
++
++/* { dg-do run } */
++/* { dg-options "-O0" } */
++
++#include <nds32_intrinsic.h>
++#include <stdlib.h>
++
++int
++main ()
++{
++  unsigned int psw;
++  unsigned int gie;
++  unsigned int pfm_ctl;
++
++  __nds32__setgie_en ();
++  __nds32__dsb(); /* This is needed for waiting pipeline.  */
++  psw = __nds32__mfsr (NDS32_SR_PSW);
++
++  gie = psw & 0x00000001;
++
++  if (gie != 1)
++    abort ();
++
++  psw = psw & 0xFFFFFFFE;
++  __nds32__mtsr (psw,NDS32_SR_PSW);
++  __nds32__dsb(); /* This is needed for waiting pipeline.  */
++  psw = __nds32__mfsr (NDS32_SR_PSW);
++  gie = psw & 0x00000001;
++
++  if (gie != 0)
++    abort ();
++  else
++   exit (0);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,10 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++void
++main (void)
++{
++  __nds32__clr_pending_swint ();
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending-hw.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending-hw.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending-hw.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-clr-pending-hw.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,16 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++void
++main (void)
++{
++ __nds32__clr_pending_hwint (NDS32_INT_H0);
++ __nds32__clr_pending_hwint (NDS32_INT_H1);
++ __nds32__clr_pending_hwint (NDS32_INT_H2);
++
++ __nds32__clr_pending_hwint (NDS32_INT_H15);
++ __nds32__clr_pending_hwint (NDS32_INT_H16);
++ __nds32__clr_pending_hwint (NDS32_INT_H31);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-disable.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-disable.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-disable.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-disable.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,13 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++void
++main (void)
++{
++  __nds32__disable_int (NDS32_INT_H15);
++  __nds32__disable_int (NDS32_INT_H16);
++  __nds32__disable_int (NDS32_INT_H31);
++  __nds32__disable_int (NDS32_INT_SWI);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-enable.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-enable.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-enable.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-enable.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,13 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++void
++main (void)
++{
++  __nds32__enable_int (NDS32_INT_H15);
++  __nds32__enable_int (NDS32_INT_H16);
++  __nds32__enable_int (NDS32_INT_H31);
++  __nds32__enable_int (NDS32_INT_SWI);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-get-pending-int.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-get-pending-int.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-get-pending-int.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-get-pending-int.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,14 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++int
++main (void)
++{
++  int a = __nds32__get_pending_int (NDS32_INT_H15);
++  int b = __nds32__get_pending_int (NDS32_INT_SWI);
++  int c = __nds32__get_pending_int (NDS32_INT_H16);
++
++  return a + b + c;
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-get-trig.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-get-trig.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-get-trig.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-get-trig.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,14 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++int
++main (void)
++{
++  int a = __nds32__get_trig_type (NDS32_INT_H0);
++  int b = __nds32__get_trig_type (NDS32_INT_H15);
++  int c = __nds32__get_trig_type (NDS32_INT_H16);
++  int d = __nds32__get_trig_type (NDS32_INT_H31);
++  return a + b + c + d;
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-isb.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-isb.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-isb.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-isb.c	2019-01-25 15:38:32.837242683 +0100
+@@ -0,0 +1,11 @@
++/* Verify that we generate isb instruction with builtin function.  */
++
++/* { dg-do compile }  */
++/* { dg-options "-O0" }  */
++/* { dg-final { scan-assembler "\\tisb" } }  */
++
++void
++test (void)
++{
++  __builtin_nds32_isb ();
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-isync.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-isync.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-isync.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-isync.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,12 @@
++/* Verify that we generate isync instruction with builtin function.  */
++
++/* { dg-do compile }  */
++/* { dg-options "-O0" }  */
++/* { dg-final { scan-assembler "\\tisync" } }  */
++
++void
++test (void)
++{
++  int *addr = (int *) 0x53000000;
++  __builtin_nds32_isync (addr);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-mfsr-mtsr.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-mfsr-mtsr.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-mfsr-mtsr.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-mfsr-mtsr.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,17 @@
++/* Verify that we generate mfsr/mtsr instruction with builtin function.  */
++
++/* { dg-do compile }  */
++/* { dg-options "-O0" }  */
++/* { dg-final { scan-assembler "\\tmfsr" } }  */
++/* { dg-final { scan-assembler "\\tmtsr" } }  */
++
++#include <nds32_intrinsic.h>
++
++void
++test (void)
++{
++  int ipsw_value;
++
++  ipsw_value = __builtin_nds32_mfsr (__NDS32_REG_IPSW__);
++  __builtin_nds32_mtsr (ipsw_value, __NDS32_REG_IPSW__);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-mfusr-mtusr.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-mfusr-mtusr.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-mfusr-mtusr.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-mfusr-mtusr.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,17 @@
++/* Verify that we generate mfusr/mtusr instruction with builtin function.  */
++
++/* { dg-do compile }  */
++/* { dg-options "-O0" }  */
++/* { dg-final { scan-assembler "\\tmfusr" } }  */
++/* { dg-final { scan-assembler "\\tmtusr" } }  */
++
++#include <nds32_intrinsic.h>
++
++void
++test (void)
++{
++  int itype_value;
++
++  itype_value = __builtin_nds32_mfusr (__NDS32_REG_ITYPE__);
++  __builtin_nds32_mtusr (itype_value, __NDS32_REG_ITYPE__);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-dis.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-dis.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-dis.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-dis.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,13 @@
++/* Verify that we generate setgie.d instruction with builtin function.  */
++
++/* { dg-do compile } */
++/* { dg-options "-O0" } */
++/* { dg-final { scan-assembler "\\tsetgie.d" } } */
++
++#include <nds32_intrinsic.h>
++
++void
++test (void)
++{
++  __nds32__setgie_dis ();
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-en.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-en.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-en.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-setgie-en.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,13 @@
++/* Verify that we generate setgie.e instruction with builtin function.  */
++
++/* { dg-do compile */
++/* { dg-options "-O0" } */
++/* { dg-final { scan-assembler "\\tsetgie.e" } } */
++
++#include <nds32_intrinsic.h>
++
++void
++test (void)
++{
++  __nds32__setgie_en ();
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-set-pending.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-set-pending.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-set-pending.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-set-pending.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,10 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++int
++main (void)
++{
++  __nds32__set_pending_swint ();
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-edge.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-edge.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-edge.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-edge.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,13 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++void
++main (void)
++{
++  __nds32__set_trig_type_edge (NDS32_INT_H0);
++  __nds32__set_trig_type_edge (NDS32_INT_H15);
++  __nds32__set_trig_type_edge (NDS32_INT_H16);
++  __nds32__set_trig_type_edge (NDS32_INT_H31);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-level.c gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-level.c
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-level.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/compile/builtin-set-trig-level.c	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,13 @@
++/* { dg-do compile } */
++/* { dg-options "-O1" } */
++
++#include <nds32_intrinsic.h>
++
++void
++main (void)
++{
++  __nds32__set_trig_type_level (NDS32_INT_H0);
++  __nds32__set_trig_type_level (NDS32_INT_H15);
++  __nds32__set_trig_type_level (NDS32_INT_H16);
++  __nds32__set_trig_type_level (NDS32_INT_H31);
++}
+diff -urN gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/nds32.exp gcc-8.2.0/gcc/testsuite/gcc.target/nds32/nds32.exp
+--- gcc-8.2.0.orig/gcc/testsuite/gcc.target/nds32/nds32.exp	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/gcc/testsuite/gcc.target/nds32/nds32.exp	2019-01-25 15:38:32.841242694 +0100
+@@ -38,8 +38,10 @@
+ dg-init
+ 
+ # Main loop.
+-dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/*.\[cS\]]] \
++dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/compile/*.\[cS\]]] \
+ 	"" $DEFAULT_CFLAGS
++gcc-dg-runtest [lsort [glob -nocomplain $srcdir/$subdir/*.\[cS\]]] \
++	"" ""
+ 
+ # All done.
+ dg-finish
+diff -urN gcc-8.2.0.orig/gcc/testsuite/lib/target-supports.exp gcc-8.2.0/gcc/testsuite/lib/target-supports.exp
+--- gcc-8.2.0.orig/gcc/testsuite/lib/target-supports.exp	2018-06-29 00:23:51.000000000 +0200
++++ gcc-8.2.0/gcc/testsuite/lib/target-supports.exp	2019-01-25 15:38:32.841242694 +0100
+@@ -8783,6 +8783,7 @@
+ 	 || [istarget avr*-*-*]
+ 	 || [istarget crisv32-*-*] || [istarget cris-*-*]
+ 	 || [istarget mmix-*-*]
++	 || [istarget nds32*-*-*]
+ 	 || [istarget s390*-*-*]
+ 	 || [istarget powerpc*-*-*]
+ 	 || [istarget nios2*-*-*]
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/initfini.c gcc-8.2.0/libgcc/config/nds32/initfini.c
+--- gcc-8.2.0.orig/libgcc/config/nds32/initfini.c	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/initfini.c	2019-01-25 15:38:32.841242694 +0100
+@@ -25,6 +25,10 @@
+    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+    <http://www.gnu.org/licenses/>.  */
+ 
++#include <stddef.h>
++/* Need header file for `struct object' type.  */
++#include "../libgcc/unwind-dw2-fde.h"
++
+ /*  Declare a pointer to void function type.  */
+ typedef void (*func_ptr) (void);
+ 
+@@ -42,11 +46,59 @@
+    refer to only the __CTOR_END__ symbol in crtfini.o and the __DTOR_LIST__
+    symbol in crtinit.o, where they are defined.  */
+ 
+-static func_ptr __CTOR_LIST__[1] __attribute__ ((section (".ctors")))
+-     = { (func_ptr) (-1) };
++static func_ptr __CTOR_LIST__[1] __attribute__ ((section (".ctors"), used))
++     = { (func_ptr) 0 };
++
++static func_ptr __DTOR_LIST__[1] __attribute__ ((section (".dtors"), used))
++     = { (func_ptr) 0 };
++
++
++#ifdef SUPPORT_UNWINDING_DWARF2
++/* Preparation of exception handling with dwar2 mechanism registration.  */
++
++asm ("\n\
++	.section .eh_frame,\"aw\",@progbits\n\
++	.global __EH_FRAME_BEGIN__\n\
++	.type	__EH_FRAME_BEGIN__, @object\n\
++	.align 2\n\
++__EH_FRAME_BEGIN__:\n\
++	! Beginning location of eh_frame section\n\
++	.previous\n\
++");
++
++extern func_ptr __EH_FRAME_BEGIN__[];
++
+ 
+-static func_ptr __DTOR_LIST__[1] __attribute__ ((section (".dtors")))
+-     = { (func_ptr) (-1) };
++/* Note that the following two functions are going to be chained into
++   constructor and destructor list, repectively.  So these two declarations
++   must be placed after __CTOR_LIST__ and __DTOR_LIST.  */
++extern void __nds32_register_eh(void) __attribute__((constructor, used));
++extern void __nds32_deregister_eh(void) __attribute__((destructor, used));
++
++/* Register the exception handling table as the first constructor.  */
++void
++__nds32_register_eh (void)
++{
++  static struct object object;
++  if (__register_frame_info)
++    __register_frame_info (__EH_FRAME_BEGIN__, &object);
++}
++
++/* Unregister the exception handling table as a deconstructor.  */
++void
++__nds32_deregister_eh (void)
++{
++  static int completed = 0;
++
++  if (completed)
++    return;
++
++  if (__deregister_frame_info)
++    __deregister_frame_info (__EH_FRAME_BEGIN__);
++
++  completed = 1;
++}
++#endif
+ 
+ /* Run all the global destructors on exit from the program.  */
+ 
+@@ -63,7 +115,7 @@
+    same particular root executable or shared library file.  */
+ 
+ static void __do_global_dtors (void)
+-asm ("__do_global_dtors") __attribute__ ((section (".text")));
++asm ("__do_global_dtors") __attribute__ ((section (".text"), used));
+ 
+ static void
+ __do_global_dtors (void)
+@@ -116,23 +168,37 @@
+    last, these words naturally end up at the very ends of the two lists
+    contained in these two sections.  */
+ 
+-static func_ptr __CTOR_END__[1] __attribute__ ((section (".ctors")))
++static func_ptr __CTOR_END__[1] __attribute__ ((section (".ctors"), used))
+      = { (func_ptr) 0 };
+ 
+-static func_ptr __DTOR_END__[1] __attribute__ ((section (".dtors")))
++static func_ptr __DTOR_END__[1] __attribute__ ((section (".dtors"), used))
+      = { (func_ptr) 0 };
+ 
++#ifdef SUPPORT_UNWINDING_DWARF2
++/* ZERO terminator in .eh_frame section.  */
++asm ("\n\
++	.section .eh_frame,\"aw\",@progbits\n\
++	.global __EH_FRAME_END__\n\
++	.type	__EH_FRAME_END__, @object\n\
++	.align 2\n\
++__EH_FRAME_END__:\n\
++	! End location of eh_frame section with ZERO terminator\n\
++	.word 0\n\
++	.previous\n\
++");
++#endif
++
+ /* Run all global constructors for the program.
+    Note that they are run in reverse order.  */
+ 
+ static void __do_global_ctors (void)
+-asm ("__do_global_ctors") __attribute__ ((section (".text")));
++asm ("__do_global_ctors") __attribute__ ((section (".text"), used));
+ 
+ static void
+ __do_global_ctors (void)
+ {
+   func_ptr *p;
+-  for (p = __CTOR_END__ - 1; *p != (func_ptr) -1; p--)
++  for (p = __CTOR_END__ - 1; *p; p--)
+     (*p) ();
+ }
+ 
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/adj_intr_lvl.inc gcc-8.2.0/libgcc/config/nds32/isr-library/adj_intr_lvl.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/adj_intr_lvl.inc	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/adj_intr_lvl.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -26,13 +26,26 @@
+ .macro ADJ_INTR_LVL
+ #if defined(NDS32_NESTED) /* Nested handler.  */
+ 	mfsr	$r3, $PSW
++	/* By substracting 1 from $PSW, we can lower PSW.INTL
++	   and enable GIE simultaneously.  */
+ 	addi	$r3, $r3, #-0x1
++  #if __NDS32_EXT_ZOL__ || __NDS32_EXT_DSP__
++    ori   $r3, $r3, 0x2000  /* Set PSW.AEN(b'13) */
++  #endif
+ 	mtsr	$r3, $PSW
+ #elif defined(NDS32_NESTED_READY) /* Nested ready handler.  */
+ 	/* Save ipc and ipsw and lower INT level.  */
+ 	mfsr	$r3, $PSW
+ 	addi	$r3, $r3, #-0x2
++  #if __NDS32_EXT_ZOL__ || __NDS32_EXT_DSP__
++    ori   $r3, $r3, 0x2000  /* Set PSW.AEN(b'13) */
++  #endif
+ 	mtsr	$r3, $PSW
+ #else /* Not nested handler.  */
++  #if __NDS32_EXT_ZOL__ || __NDS32_EXT_DSP__
++    mfsr	$r3, $PSW
++    ori   $r3, $r3, 0x2000  /* Set PSW.AEN(b'13) */
++    mtsr	$r3, $PSW
++  #endif
+ #endif
+ .endm
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/excp_isr.S gcc-8.2.0/libgcc/config/nds32/isr-library/excp_isr.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/excp_isr.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/excp_isr.S	2019-01-25 15:38:32.841242694 +0100
+@@ -23,6 +23,7 @@
+    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+    <http://www.gnu.org/licenses/>.  */
+ 
++#include "save_usr_regs.inc"
+ #include "save_mac_regs.inc"
+ #include "save_fpu_regs.inc"
+ #include "save_fpu_regs_00.inc"
+@@ -32,35 +33,33 @@
+ #include "save_all.inc"
+ #include "save_partial.inc"
+ #include "adj_intr_lvl.inc"
+-#include "restore_mac_regs.inc"
+ #include "restore_fpu_regs_00.inc"
+ #include "restore_fpu_regs_01.inc"
+ #include "restore_fpu_regs_02.inc"
+ #include "restore_fpu_regs_03.inc"
+ #include "restore_fpu_regs.inc"
++#include "restore_mac_regs.inc"
++#include "restore_usr_regs.inc"
+ #include "restore_all.inc"
+ #include "restore_partial.inc"
++
+ 	.section .nds32_isr, "ax"       /* Put it in the section of 1st level handler. */
+ 	.align	1
+-/*
+-  First Level Handlers
+-  1. First Level Handlers are invokded in vector section via jump instruction
+-  with specific names for different configurations.
+-  2. Naming Format: _nds32_e_SR_NT for exception handlers.
+-		    _nds32_i_SR_NT for interrupt handlers.
+-  2.1 All upper case letters are replaced with specific lower case letters encodings.
+-  2.2 SR: Saved Registers
+-      sa: Save All regs (context)
+-      ps: Partial Save (all caller-saved regs)
+-  2.3 NT: Nested Type
+-      ns: nested
+-      nn: not nested
+-      nr: nested ready
+-*/
+-
+-/*
+-  This is original 16-byte vector size version.
+-*/
++
++/* First Level Handlers
++   1. First Level Handlers are invokded in vector section via jump instruction
++      with specific names for different configurations.
++   2. Naming Format: _nds32_e_SR_NT for exception handlers.
++                     _nds32_i_SR_NT for interrupt handlers.
++     2.1 All upper case letters are replaced with specific lower case letters encodings.
++     2.2 SR -- Saved Registers
++         sa: Save All regs (context)
++         ps: Partial Save (all caller-saved regs)
++     2.3 NT -- Nested Type
++         ns: nested
++         nn: not nested
++         nr: nested ready */
++
+ #ifdef NDS32_SAVE_ALL_REGS
+ #if defined(NDS32_NESTED)
+ 	.globl	_nds32_e_sa_ns
+@@ -91,21 +90,26 @@
+ #endif /* endif for Nest Type */
+ #endif /* not NDS32_SAVE_ALL_REGS */
+ 
+-/*
+-  This is 16-byte vector size version.
+-  The vector id was restored into $r0 in vector by compiler.
+-*/
++
++/* For 4-byte vector size version, the vector id is
++   extracted from $ITYPE and is set into $r0 by library.
++   For 16-byte vector size version, the vector id
++   is set into $r0 in vector section by compiler.  */
++
++/* Save used registers.  */
+ #ifdef NDS32_SAVE_ALL_REGS
+         SAVE_ALL
+ #else
+         SAVE_PARTIAL
+ #endif
++
+ 	/* Prepare to call 2nd level handler. */
+ 	la	$r2, _nds32_jmptbl_00
+ 	lw	$r2, [$r2 + $r0 << #2]
+ 	ADJ_INTR_LVL	/* Adjust INTR level. $r3 is clobbered.  */
+ 	jral    $r2
+-	/* Restore used registers. */
++
++/* Restore used registers.  */
+ #ifdef NDS32_SAVE_ALL_REGS
+ 	RESTORE_ALL
+ #else
+@@ -113,6 +117,7 @@
+ #endif
+ 	iret
+ 
++
+ #ifdef NDS32_SAVE_ALL_REGS
+ #if defined(NDS32_NESTED)
+ 	.size	_nds32_e_sa_ns, .-_nds32_e_sa_ns
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/intr_isr.S gcc-8.2.0/libgcc/config/nds32/isr-library/intr_isr.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/intr_isr.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/intr_isr.S	2019-01-25 15:38:32.841242694 +0100
+@@ -23,6 +23,7 @@
+    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+    <http://www.gnu.org/licenses/>.  */
+ 
++#include "save_usr_regs.inc"
+ #include "save_mac_regs.inc"
+ #include "save_fpu_regs.inc"
+ #include "save_fpu_regs_00.inc"
+@@ -32,35 +33,33 @@
+ #include "save_all.inc"
+ #include "save_partial.inc"
+ #include "adj_intr_lvl.inc"
+-#include "restore_mac_regs.inc"
+ #include "restore_fpu_regs_00.inc"
+ #include "restore_fpu_regs_01.inc"
+ #include "restore_fpu_regs_02.inc"
+ #include "restore_fpu_regs_03.inc"
+ #include "restore_fpu_regs.inc"
++#include "restore_mac_regs.inc"
++#include "restore_usr_regs.inc"
+ #include "restore_all.inc"
+ #include "restore_partial.inc"
++
+ 	.section .nds32_isr, "ax"       /* Put it in the section of 1st level handler. */
+ 	.align	1
+-/*
+-  First Level Handlers
+-  1. First Level Handlers are invokded in vector section via jump instruction
+-  with specific names for different configurations.
+-  2. Naming Format: _nds32_e_SR_NT for exception handlers.
+-		    _nds32_i_SR_NT for interrupt handlers.
+-  2.1 All upper case letters are replaced with specific lower case letters encodings.
+-  2.2 SR: Saved Registers
+-      sa: Save All regs (context)
+-      ps: Partial Save (all caller-saved regs)
+-  2.3 NT: Nested Type
+-      ns: nested
+-      nn: not nested
+-      nr: nested ready
+-*/
+-
+-/*
+-  This is original 16-byte vector size version.
+-*/
++
++/* First Level Handlers
++   1. First Level Handlers are invokded in vector section via jump instruction
++      with specific names for different configurations.
++   2. Naming Format: _nds32_e_SR_NT for exception handlers.
++                     _nds32_i_SR_NT for interrupt handlers.
++     2.1 All upper case letters are replaced with specific lower case letters encodings.
++     2.2 SR -- Saved Registers
++         sa: Save All regs (context)
++         ps: Partial Save (all caller-saved regs)
++     2.3 NT -- Nested Type
++         ns: nested
++         nn: not nested
++         nr: nested ready */
++
+ #ifdef NDS32_SAVE_ALL_REGS
+ #if defined(NDS32_NESTED)
+ 	.globl	_nds32_i_sa_ns
+@@ -91,21 +90,36 @@
+ #endif /* endif for Nest Type */
+ #endif /* not NDS32_SAVE_ALL_REGS */
+ 
+-/*
+-  This is 16-byte vector size version.
+-  The vector id was restored into $r0 in vector by compiler.
+-*/
++
++/* For 4-byte vector size version, the vector id is
++   extracted from $ITYPE and is set into $r0 by library.
++   For 16-byte vector size version, the vector id
++   is set into $r0 in vector section by compiler.  */
++
++/* Save used registers first.  */
+ #ifdef NDS32_SAVE_ALL_REGS
+         SAVE_ALL
+ #else
+         SAVE_PARTIAL
+ #endif
+-	/* Prepare to call 2nd level handler. */
++
++/* According to vector size, we need to have different implementation.  */
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* Prepare to call 2nd level handler.  */
++	la	$r2, _nds32_jmptbl_00
++	lw	$r2, [$r2 + $r0 << #2]
++	addi    $r0, $r0, #-9	/* Make interrput vector id zero-based.  */
++	ADJ_INTR_LVL	/* Adjust INTR level.  $r3 is clobbered.  */
++	jral    $r2
++#else /* not __NDS32_ISR_VECTOR_SIZE_4__ */
++	/* Prepare to call 2nd level handler.  */
+ 	la	$r2, _nds32_jmptbl_09	/* For zero-based vcetor id.  */
+ 	lw	$r2, [$r2 + $r0 << #2]
+ 	ADJ_INTR_LVL	/* Adjust INTR level. $r3 is clobbered.  */
+ 	jral    $r2
+-	/* Restore used registers. */
++#endif /* not __NDS32_ISR_VECTOR_SIZE_4__ */
++
++/* Restore used registers.  */
+ #ifdef NDS32_SAVE_ALL_REGS
+ 	RESTORE_ALL
+ #else
+@@ -113,6 +127,7 @@
+ #endif
+ 	iret
+ 
++
+ #ifdef NDS32_SAVE_ALL_REGS
+ #if defined(NDS32_NESTED)
+ 	.size	_nds32_i_sa_ns, .-_nds32_i_sa_ns
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/reset.S gcc-8.2.0/libgcc/config/nds32/isr-library/reset.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/reset.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/reset.S	2019-01-25 15:38:32.841242694 +0100
+@@ -26,22 +26,18 @@
+ 	.section .nds32_isr, "ax"	/* Put it in the section of 1st level handler.  */
+ 	.align	1
+ 	.weak	_SDA_BASE_	/* For reset handler only.  */
+-	.weak	_FP_BASE_	/* For reset handler only.  */
+ 	.weak	_nds32_init_mem	/* User defined memory initialization function.  */
+ 	.globl	_start
+ 	.globl	_nds32_reset
+ 	.type	_nds32_reset, @function
+ _nds32_reset:
+ _start:
+-#ifdef  NDS32_EXT_EX9
+-	.no_ex9_begin
+-#endif
+ 	/* Handle NMI and warm boot if any of them exists.  */
+ 	beqz	$sp, 1f		/* Reset, NMI or warm boot?  */
+ 	/* Either NMI or warm boot; save all regs.  */
+ 
+ 	/* Preserve registers for context-switching.  */
+-#ifdef __NDS32_REDUCED_REGS__
++#if __NDS32_REDUCED_REGS__ || __NDS32_REDUCE_REGS
+ 	/* For 16-reg mode.  */
+ 	smw.adm $r0, [$sp], $r10, #0x0
+ 	smw.adm $r15, [$sp], $r15, #0xf
+@@ -49,10 +45,9 @@
+ 	/* For 32-reg mode.  */
+ 	smw.adm $r0, [$sp], $r27, #0xf
+ #endif
+-#ifdef NDS32_EXT_IFC
++#if __NDS32_EXT_IFC__
+ 	mfusr   $r1, $IFC_LP
+-	smw.adm $r1, [$sp], $r2, #0x0	/* Save extra $r2 to keep
+-					   stack 8-byte alignment.  */
++	smw.adm $r1, [$sp], $r2, #0x0	/* Save extra $r2 to keep stack 8-byte alignment.  */
+ #endif
+ 
+ 	la	$gp, _SDA_BASE_	/* Init GP for small data access.  */
+@@ -71,12 +66,11 @@
+ 	bnez    $r0, 1f		/* If fail to resume, do cold boot.  */
+ 
+ 	/* Restore registers for context-switching.  */
+-#ifdef NDS32_EXT_IFC
+-	lmw.bim	$r1, [$sp], $r2, #0x0	/* Restore extra $r2 to keep
+-					   stack 8-byte alignment.  */
++#if __NDS32_EXT_IFC__
++	lmw.bim	$r1, [$sp], $r2, #0x0	/* Restore extra $r2 to keep stack 8-byte alignment.  */
+ 	mtusr   $r1, $IFC_LP
+ #endif
+-#ifdef __NDS32_REDUCED_REGS__
++#if __NDS32_REDUCED_REGS__ || __NDS32_REDUCE_REGS
+ 	/* For 16-reg mode.  */
+ 	lmw.bim	$r15, [$sp], $r15, #0xf
+ 	lmw.bim	$r0, [$sp], $r10, #0x0
+@@ -88,6 +82,17 @@
+ 
+ 
+ 1:	/* Cold boot.  */
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* With vector ID feature for v3 architecture, default vector size is 4-byte.  */
++	/* Set IVB.ESZ = 0 (vector table entry size = 4 bytes)  */
++	mfsr    $r0, $IVB
++	li      $r1, #0xc000
++	or      $r0, $r0, $r1
++	xor     $r0, $r0, $r1
++	mtsr    $r0, $IVB
++	dsb
++#else
++	/* There is no vector ID feature, so the vector size must be 16-byte.  */
+ 	/* Set IVB.ESZ = 1 (vector table entry size = 16 bytes)  */
+ 	mfsr    $r0, $IVB
+ 	li	$r1, #0xffff3fff
+@@ -95,36 +100,54 @@
+ 	ori	$r0, $r0, #0x4000
+ 	mtsr    $r0, $IVB
+ 	dsb
++#endif
+ 
+ 	la	$gp, _SDA_BASE_		/* Init $gp.  */
+-	la	$fp, _FP_BASE_		/* Init $fp.  */
+ 	la	$sp, _stack		/* Init $sp.  */
+-#ifdef  NDS32_EXT_EX9
+-/*
+- *	Initialize the table base of EX9 instruction
+- *	ex9 generation needs to disable before the ITB is set
+- */
+-	mfsr    $r0, $MSC_CFG	/* Check if HW support of EX9.  */
++
++#if __NDS32_EXT_EX9__
++.L_init_itb:
++	/* Initialization for Instruction Table Base (ITB).
++	   The symbol _ITB_BASE_ is determined by Linker.
++	   Set $ITB only if MSC_CFG.EIT (cr4.b'24) is set.  */
++	mfsr    $r0, $MSC_CFG
+ 	srli	$r0, $r0, 24
+ 	andi	$r0, $r0, 0x1
+-	beqz	$r0, 4f		/* Zero means HW does not support EX9.  */
+-	la      $r0, _ITB_BASE_	/* Init $ITB.  */
++	beqz	$r0, 4f		/* Fall through ?  */
++	la      $r0, _ITB_BASE_
+ 	mtusr   $r0, $ITB
+-	.no_ex9_end
+ 4:
+ #endif
+-	la	$r15, _nds32_init_mem	/* Call DRAM init. _nds32_init_mem
+-					  may written by C language.  */
++
++#if __NDS32_EXT_FPU_SP__ || __NDS32_EXT_FPU_DP__
++.L_init_fpu:
++	/* Initialize FPU
++	   Set FUCOP_CTL.CP0EN (fucpr.b'0).  */
++	mfsr    $r0, $FUCOP_CTL
++	ori     $r0, $r0, 0x1
++	mtsr    $r0, $FUCOP_CTL
++	dsb
++	/* According to [bugzilla #9425], set flush-to-zero mode.
++	   That is, set $FPCSR.DNZ(b'12) = 1.  */
++	FMFCSR	$r0
++	ori	$r0, $r0, 0x1000
++	FMTCSR	$r0
++	dsb
++#endif
++
++	/* Call DRAM init. _nds32_init_mem may written by C language.  */
++	la	$r15, _nds32_init_mem
+ 	beqz	$r15, 6f
+ 	jral	$r15
+ 6:
+ 	l.w	$r15, _nds32_jmptbl_00	/* Load reset handler.  */
+ 	jral	$r15
+-/* Reset handler() should never return in a RTOS or non-OS system.
+-   In case it does return, an exception will be generated.
+-   This exception will be caught either by default break handler or by EDM.
+-   Default break handle may just do an infinite loop.
+-   EDM will notify GDB and GDB will regain control when the ID is 0x7fff. */
++
++	/* Reset handler() should never return in a RTOS or non-OS system.
++	   In case it does return, an exception will be generated.
++	   This exception will be caught either by default break handler or by EDM.
++	   Default break handle may just do an infinite loop.
++	   EDM will notify GDB and GDB will regain control when the ID is 0x7fff.  */
+ 5:
+ 	break    #0x7fff
+ 	.size	_nds32_reset, .-_nds32_reset
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_all.inc gcc-8.2.0/libgcc/config/nds32/isr-library/restore_all.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_all.inc	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/restore_all.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -31,15 +31,11 @@
+ 	mtsr	$r2, $IPSW
+ 	RESTORE_FPU_REGS
+ 	RESTORE_MAC_REGS
+-#ifdef NDS32_EXT_IFC
+-	lmw.bim	$r1, [$sp], $r2, #0x0	/* Restore extra $r2 to keep
+-					   stack 8-byte alignment.  */
+-	mtusr   $r1, $IFC_LP
+-#endif
+-#ifdef __NDS32_REDUCED_REGS__
++  RESTORE_USR_REGS
++#if __NDS32_REDUCED_REGS__ || __NDS32_REDUCE_REGS
+ 	lmw.bim $r0, [$sp], $r10, #0x0	/* Restore all regs.  */
+ 	lmw.bim $r15, [$sp], $r15, #0xf
+-#else /* not __NDS32_REDUCED_REGS__ */
++#else
+ 	lmw.bim $r0, [$sp], $r27, #0xf	/* Restore all regs.  */
+ #endif
+ .endm
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_mac_regs.inc gcc-8.2.0/libgcc/config/nds32/isr-library/restore_mac_regs.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_mac_regs.inc	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/restore_mac_regs.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -24,7 +24,7 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ .macro RESTORE_MAC_REGS
+-#ifdef NDS32_DX_REGS
++#if __NDS32_DX_REGS__
+ 	lmw.bim	$r1, [$sp], $r4, #0x0
+ 	mtusr	$r1, $d0.lo
+ 	mtusr	$r2, $d0.hi
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_partial.inc gcc-8.2.0/libgcc/config/nds32/isr-library/restore_partial.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_partial.inc	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/restore_partial.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -31,15 +31,11 @@
+ 	mtsr $r1, $IPC	/* Set IPC.  */
+ 	mtsr $r2, $IPSW	/* Set IPSW.  */
+ #endif
+-	RESTORE_FPU_REGS
+-	RESTORE_MAC_REGS
+-#ifdef NDS32_EXT_IFC
+-	lmw.bim	$r1, [$sp], $r2, #0x0	/* Restore extra $r2 to keep
+-					   stack 8-byte alignment.  */
+-	mtusr   $r1, $IFC_LP
+-#endif
++  RESTORE_FPU_REGS
++  RESTORE_MAC_REGS
++  RESTORE_USR_REGS
+ 	lmw.bim $r0, [$sp], $r5, #0x0	/* Restore all regs.  */
+-#ifdef __NDS32_REDUCED_REGS__
++#if __NDS32_REDUCED_REGS__ || __NDS32_REDUCE_REGS
+ 	lmw.bim $r15, [$sp], $r15, #0x2
+ #else
+ 	lmw.bim $r15, [$sp], $r27, #0x2	/* Restore all regs.  */
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_usr_regs.inc gcc-8.2.0/libgcc/config/nds32/isr-library/restore_usr_regs.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/restore_usr_regs.inc	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/restore_usr_regs.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,42 @@
++/* c-isr library stuff of Andes NDS32 cpu for GNU compiler
++   Copyright (C) 2012-2018 Free Software Foundation, Inc.
++   Contributed by Andes Technology Corporation.
++
++   This file is part of GCC.
++
++   GCC is free software; you can redistribute it and/or modify it
++   under the terms of the GNU General Public License as published
++   by the Free Software Foundation; either version 3, or (at your
++   option) any later version.
++
++   GCC is distributed in the hope that it will be useful, but WITHOUT
++   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++   License for more details.
++
++   Under Section 7 of GPL version 3, you are granted additional
++   permissions described in the GCC Runtime Library Exception, version
++   3.1, as published by the Free Software Foundation.
++
++   You should have received a copy of the GNU General Public License and
++   a copy of the GCC Runtime Library Exception along with this program;
++   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
++   <http://www.gnu.org/licenses/>.  */
++
++.macro RESTORE_USR_REGS
++#if __NDS32_EXT_IFC__ && (__NDS32_EXT_ZOL__ || __NDS32_EXT_DSP__)
++  lmw.bim $r1, [$sp], $r4, #0x0
++  mtusr   $r1, $IFC_LP
++  mtusr   $r2, $LB
++  mtusr   $r3, $LE
++  mtusr   $r4, $LC
++#elif __NDS32_EXT_IFC__
++  lmw.bim	$r1, [$sp], $r2, #0x0
++  mtusr   $r1, $IFC_LP
++#elif __NDS32_EXT_ZOL__ || __NDS32_EXT_DSP__
++  lmw.bim $r1, [$sp], $r4, #0x0
++  mtusr   $r1, $LB
++  mtusr   $r2, $LE
++  mtusr   $r3, $LC
++#endif
++.endm
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_all.inc gcc-8.2.0/libgcc/config/nds32/isr-library/save_all.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_all.inc	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/save_all.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -23,45 +23,42 @@
+    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+    <http://www.gnu.org/licenses/>.  */
+ 
+-.macro SAVE_ALL_4B
+-#ifdef __NDS32_REDUCED_REGS__
++#if __NDS32_ISR_VECTOR_SIZE_4__
++
++/* If vector size is 4-byte, we have to save registers
++   in the macro implementation.  */
++.macro SAVE_ALL
++#if __NDS32_REDUCED_REGS__ || __NDS32_REDUCE_REGS
+ 	smw.adm $r15, [$sp], $r15, #0xf
+ 	smw.adm $r0, [$sp], $r10, #0x0
+-#else /* not __NDS32_REDUCED_REGS__ */
++#else
+ 	smw.adm $r0, [$sp], $r27, #0xf
+-#endif /* not __NDS32_REDUCED_REGS__ */
+-#ifdef NDS32_EXT_IFC
+-	mfusr   $r1, $IFC_LP
+-	smw.adm $r1, [$sp], $r2, #0x0	/* Save extra $r2 to keep
+-					   stack 8-byte alignment.  */
+ #endif
+-	SAVE_MAC_REGS
+-	SAVE_FPU_REGS
++  SAVE_USR_REGS
++  SAVE_MAC_REGS
++  SAVE_FPU_REGS
+ 	mfsr	$r1, $IPC	/* Get IPC.  */
+ 	mfsr	$r2, $IPSW	/* Get IPSW.  */
+ 	smw.adm	$r1, [$sp], $r2, #0x0	/* Push IPC, IPSW.  */
+ 	move	$r1, $sp	/* $r1 is ptr to NDS32_CONTEXT.  */
+ 	mfsr	$r0, $ITYPE	/* Get VID to $r0.  */
+ 	srli	$r0, $r0, #5
+-#ifdef __NDS32_ISA_V2__
+ 	andi	$r0, $r0, #127
+-#else
+-	fexti33	$r0, #6
+-#endif
+ .endm
+ 
++#else /* not __NDS32_ISR_VECTOR_SIZE_4__ */
++
++/* If vector size is 16-byte, some works can be done in
++   the vector section generated by compiler, so that we
++   can implement less in the macro.  */
+ .macro SAVE_ALL
+-/* SAVE_REG_TBL code has been moved to
+-   vector table generated by compiler.  */
+-#ifdef NDS32_EXT_IFC
+-	mfusr   $r1, $IFC_LP
+-	smw.adm $r1, [$sp], $r2, #0x0	/* Save extra $r2 to keep
+-					   stack 8-byte alignment.  */
+-#endif
+-	SAVE_MAC_REGS
+-	SAVE_FPU_REGS
++  SAVE_USR_REGS
++  SAVE_MAC_REGS
++  SAVE_FPU_REGS
+ 	mfsr	$r1, $IPC	/* Get IPC.  */
+ 	mfsr	$r2, $IPSW	/* Get IPSW.  */
+ 	smw.adm	$r1, [$sp], $r2, #0x0	/* Push IPC, IPSW.  */
+ 	move	$r1, $sp	/* $r1 is ptr to NDS32_CONTEXT.  */
+ .endm
++
++#endif /* not __NDS32_ISR_VECTOR_SIZE_4__ */
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_mac_regs.inc gcc-8.2.0/libgcc/config/nds32/isr-library/save_mac_regs.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_mac_regs.inc	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/save_mac_regs.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -24,7 +24,7 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ .macro SAVE_MAC_REGS
+-#ifdef NDS32_DX_REGS
++#if __NDS32_DX_REGS__
+ 	mfusr	$r1, $d0.lo
+ 	mfusr	$r2, $d0.hi
+ 	mfusr	$r3, $d1.lo
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_partial.inc gcc-8.2.0/libgcc/config/nds32/isr-library/save_partial.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_partial.inc	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/save_partial.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -23,20 +23,20 @@
+    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
+    <http://www.gnu.org/licenses/>.  */
+ 
+-.macro SAVE_PARTIAL_4B
+-#ifdef __NDS32_REDUCED_REGS__
++#if __NDS32_ISR_VECTOR_SIZE_4__
++
++/* If vector size is 4-byte, we have to save registers
++   in the macro implementation.  */
++.macro SAVE_PARTIAL
++#if __NDS32_REDUCED_REGS__ || __NDS32_REDUCE_REGS
+ 	smw.adm $r15, [$sp], $r15, #0x2
+-#else /* not __NDS32_REDUCED_REGS__ */
++#else
+ 	smw.adm $r15, [$sp], $r27, #0x2
+-#endif /* not __NDS32_REDUCED_REGS__ */
+-	smw.adm $r0, [$sp], $r5, #0x0
+-#ifdef NDS32_EXT_IFC
+-	mfusr   $r1, $IFC_LP
+-	smw.adm $r1, [$sp], $r2, #0x0	/* Save extra $r2 to keep
+-					   stack 8-byte alignment.  */
+ #endif
+-	SAVE_MAC_REGS
+-	SAVE_FPU_REGS
++	smw.adm $r0, [$sp], $r5, #0x0
++  SAVE_USR_REGS
++  SAVE_MAC_REGS
++  SAVE_FPU_REGS
+ #if defined(NDS32_NESTED) || defined(NDS32_NESTED_READY)
+        mfsr    $r1, $IPC       /* Get IPC.  */
+        mfsr    $r2, $IPSW      /* Get IPSW.  */
+@@ -44,26 +44,24 @@
+ #endif
+ 	mfsr	$r0, $ITYPE	/* Get VID to $r0.  */
+ 	srli	$r0, $r0, #5
+-#ifdef __NDS32_ISA_V2__
+ 	andi	$r0, $r0, #127
+-#else
+-	fexti33	$r0, #6
+-#endif
+ .endm
+ 
++#else /* not __NDS32_ISR_VECTOR_SIZE_4__ */
++
++/* If vector size is 16-byte, some works can be done in
++   the vector section generated by compiler, so that we
++   can implement less in the macro.  */
++
+ .macro SAVE_PARTIAL
+-/* SAVE_CALLER_REGS code has been moved to
+-   vector table generated by compiler.  */
+-#ifdef NDS32_EXT_IFC
+-	mfusr   $r1, $IFC_LP
+-	smw.adm $r1, [$sp], $r2, #0x0	/* Save extra $r2 to keep
+-					   stack 8-byte alignment.  */
+-#endif
+-	SAVE_MAC_REGS
+-	SAVE_FPU_REGS
++  SAVE_USR_REGS
++  SAVE_MAC_REGS
++  SAVE_FPU_REGS
+ #if defined(NDS32_NESTED) || defined(NDS32_NESTED_READY)
+        mfsr    $r1, $IPC       /* Get IPC.  */
+        mfsr    $r2, $IPSW      /* Get IPSW.  */
+        smw.adm $r1, [$sp], $r2, #0x0   /* Push IPC, IPSW.  */
+ #endif
+ .endm
++
++#endif /* not __NDS32_ISR_VECTOR_SIZE_4__ */
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_usr_regs.inc gcc-8.2.0/libgcc/config/nds32/isr-library/save_usr_regs.inc
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/save_usr_regs.inc	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/save_usr_regs.inc	2019-01-25 15:38:32.841242694 +0100
+@@ -0,0 +1,44 @@
++/* c-isr library stuff of Andes NDS32 cpu for GNU compiler
++   Copyright (C) 2012-2018 Free Software Foundation, Inc.
++   Contributed by Andes Technology Corporation.
++
++   This file is part of GCC.
++
++   GCC is free software; you can redistribute it and/or modify it
++   under the terms of the GNU General Public License as published
++   by the Free Software Foundation; either version 3, or (at your
++   option) any later version.
++
++   GCC is distributed in the hope that it will be useful, but WITHOUT
++   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++   License for more details.
++
++   Under Section 7 of GPL version 3, you are granted additional
++   permissions described in the GCC Runtime Library Exception, version
++   3.1, as published by the Free Software Foundation.
++
++   You should have received a copy of the GNU General Public License and
++   a copy of the GCC Runtime Library Exception along with this program;
++   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
++   <http://www.gnu.org/licenses/>.  */
++
++.macro SAVE_USR_REGS
++/* Store User Special Registers according to supported ISA extension
++   !!! WATCH OUT !!! Take care of 8-byte alignment issue.  */
++#if __NDS32_EXT_IFC__ && (__NDS32_EXT_ZOL__ || __NDS32_EXT_DSP__)
++  mfusr   $r1, $IFC_LP
++  mfusr   $r2, $LB
++  mfusr   $r3, $LE
++  mfusr   $r4, $LC
++  smw.adm $r1, [$sp], $r4, #0x0 /* Save even. Ok!  */
++#elif __NDS32_EXT_IFC__
++  mfusr   $r1, $IFC_LP
++  smw.adm $r1, [$sp], $r2, #0x0	/* Save extra $r2 to keep stack 8-byte aligned.  */
++#elif (__NDS32_EXT_ZOL__ || __NDS32_EXT_DSP__)
++  mfusr   $r1, $LB
++  mfusr   $r2, $LE
++  mfusr   $r3, $LC
++  smw.adm $r1, [$sp], $r4, #0x0	/* Save extra $r4 to keep stack 8-byte aligned.  */
++#endif
++.endm
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid00.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid00.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid00.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid00.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.00, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_00
+ 	.type	_nds32_vector_00, @function
+ _nds32_vector_00:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid01.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid01.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid01.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid01.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.01, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_01
+ 	.type	_nds32_vector_01, @function
+ _nds32_vector_01:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid02.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid02.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid02.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid02.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.02, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_02
+ 	.type	_nds32_vector_02, @function
+ _nds32_vector_02:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid03.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid03.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid03.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid03.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.03, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_03
+ 	.type	_nds32_vector_03, @function
+ _nds32_vector_03:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid04.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid04.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid04.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid04.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.04, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_04
+ 	.type	_nds32_vector_04, @function
+ _nds32_vector_04:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid05.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid05.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid05.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid05.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.05, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_05
+ 	.type	_nds32_vector_05, @function
+ _nds32_vector_05:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid06.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid06.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid06.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid06.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.06, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_06
+ 	.type	_nds32_vector_06, @function
+ _nds32_vector_06:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid07.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid07.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid07.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid07.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.07, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_07
+ 	.type	_nds32_vector_07, @function
+ _nds32_vector_07:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid08.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid08.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid08.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid08.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.08, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_08
+ 	.type	_nds32_vector_08, @function
+ _nds32_vector_08:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid09.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid09.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid09.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid09.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.09, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_09
+ 	.type	_nds32_vector_09, @function
+ _nds32_vector_09:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid10.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid10.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid10.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid10.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.10, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_10
+ 	.type	_nds32_vector_10, @function
+ _nds32_vector_10:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid11.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid11.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid11.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid11.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.11, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_11
+ 	.type	_nds32_vector_11, @function
+ _nds32_vector_11:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid12.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid12.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid12.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid12.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.12, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_12
+ 	.type	_nds32_vector_12, @function
+ _nds32_vector_12:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid13.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid13.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid13.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid13.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.13, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_13
+ 	.type	_nds32_vector_13, @function
+ _nds32_vector_13:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid14.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid14.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid14.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid14.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.14, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_14
+ 	.type	_nds32_vector_14, @function
+ _nds32_vector_14:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid15.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid15.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid15.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid15.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.15, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_15
+ 	.type	_nds32_vector_15, @function
+ _nds32_vector_15:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid16.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid16.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid16.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid16.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.16, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_16
+ 	.type	_nds32_vector_16, @function
+ _nds32_vector_16:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid17.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid17.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid17.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid17.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.17, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_17
+ 	.type	_nds32_vector_17, @function
+ _nds32_vector_17:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid18.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid18.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid18.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid18.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.18, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_18
+ 	.type	_nds32_vector_18, @function
+ _nds32_vector_18:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid19.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid19.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid19.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid19.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.19, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_19
+ 	.type	_nds32_vector_19, @function
+ _nds32_vector_19:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid20.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid20.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid20.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid20.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.20, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_20
+ 	.type	_nds32_vector_20, @function
+ _nds32_vector_20:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid21.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid21.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid21.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid21.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.21, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_21
+ 	.type	_nds32_vector_21, @function
+ _nds32_vector_21:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid22.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid22.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid22.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid22.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.22, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_22
+ 	.type	_nds32_vector_22, @function
+ _nds32_vector_22:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid23.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid23.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid23.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid23.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.23, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_23
+ 	.type	_nds32_vector_23, @function
+ _nds32_vector_23:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid24.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid24.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid24.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid24.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.24, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_24
+ 	.type	_nds32_vector_24, @function
+ _nds32_vector_24:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid25.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid25.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid25.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid25.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.25, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_25
+ 	.type	_nds32_vector_25, @function
+ _nds32_vector_25:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid26.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid26.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid26.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid26.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.26, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_26
+ 	.type	_nds32_vector_26, @function
+ _nds32_vector_26:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid27.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid27.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid27.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid27.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.27, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_27
+ 	.type	_nds32_vector_27, @function
+ _nds32_vector_27:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid28.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid28.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid28.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid28.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.28, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_28
+ 	.type	_nds32_vector_28, @function
+ _nds32_vector_28:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid29.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid29.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid29.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid29.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.29, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_29
+ 	.type	_nds32_vector_29, @function
+ _nds32_vector_29:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid30.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid30.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid30.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid30.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.30, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_30
+ 	.type	_nds32_vector_30, @function
+ _nds32_vector_30:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid31.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid31.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid31.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid31.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.31, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_31
+ 	.type	_nds32_vector_31, @function
+ _nds32_vector_31:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid32.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid32.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid32.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid32.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.32, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_32
+ 	.type	_nds32_vector_32, @function
+ _nds32_vector_32:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid33.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid33.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid33.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid33.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.33, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_33
+ 	.type	_nds32_vector_33, @function
+ _nds32_vector_33:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid34.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid34.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid34.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid34.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.34, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_34
+ 	.type	_nds32_vector_34, @function
+ _nds32_vector_34:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid35.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid35.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid35.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid35.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.35, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_35
+ 	.type	_nds32_vector_35, @function
+ _nds32_vector_35:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid36.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid36.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid36.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid36.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.36, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_36
+ 	.type	_nds32_vector_36, @function
+ _nds32_vector_36:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid37.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid37.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid37.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid37.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.37, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_37
+ 	.type	_nds32_vector_37, @function
+ _nds32_vector_37:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid38.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid38.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid38.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid38.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.38, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_38
+ 	.type	_nds32_vector_38, @function
+ _nds32_vector_38:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid39.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid39.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid39.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid39.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.39, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_39
+ 	.type	_nds32_vector_39, @function
+ _nds32_vector_39:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid40.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid40.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid40.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid40.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.40, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_40
+ 	.type	_nds32_vector_40, @function
+ _nds32_vector_40:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid41.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid41.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid41.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid41.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.41, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_41
+ 	.type	_nds32_vector_41, @function
+ _nds32_vector_41:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid42.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid42.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid42.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid42.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.42, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_42
+ 	.type	_nds32_vector_42, @function
+ _nds32_vector_42:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid43.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid43.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid43.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid43.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.43, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_43
+ 	.type	_nds32_vector_43, @function
+ _nds32_vector_43:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid44.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid44.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid44.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid44.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.44, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_44
+ 	.type	_nds32_vector_44, @function
+ _nds32_vector_44:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid45.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid45.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid45.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid45.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.45, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_45
+ 	.type	_nds32_vector_45, @function
+ _nds32_vector_45:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid46.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid46.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid46.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid46.S	2019-01-25 15:38:32.841242694 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.46, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_46
+ 	.type	_nds32_vector_46, @function
+ _nds32_vector_46:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid47.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid47.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid47.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid47.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.47, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_47
+ 	.type	_nds32_vector_47, @function
+ _nds32_vector_47:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid48.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid48.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid48.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid48.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.48, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_48
+ 	.type	_nds32_vector_48, @function
+ _nds32_vector_48:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid49.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid49.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid49.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid49.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.49, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_49
+ 	.type	_nds32_vector_49, @function
+ _nds32_vector_49:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid50.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid50.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid50.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid50.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.50, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_50
+ 	.type	_nds32_vector_50, @function
+ _nds32_vector_50:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid51.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid51.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid51.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid51.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.51, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_51
+ 	.type	_nds32_vector_51, @function
+ _nds32_vector_51:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid52.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid52.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid52.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid52.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.52, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_52
+ 	.type	_nds32_vector_52, @function
+ _nds32_vector_52:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid53.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid53.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid53.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid53.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.53, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_53
+ 	.type	_nds32_vector_53, @function
+ _nds32_vector_53:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid54.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid54.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid54.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid54.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.54, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_54
+ 	.type	_nds32_vector_54, @function
+ _nds32_vector_54:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid55.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid55.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid55.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid55.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.55, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_55
+ 	.type	_nds32_vector_55, @function
+ _nds32_vector_55:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid56.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid56.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid56.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid56.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.56, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_56
+ 	.type	_nds32_vector_56, @function
+ _nds32_vector_56:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid57.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid57.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid57.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid57.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.57, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_57
+ 	.type	_nds32_vector_57, @function
+ _nds32_vector_57:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid58.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid58.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid58.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid58.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.58, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_58
+ 	.type	_nds32_vector_58, @function
+ _nds32_vector_58:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid59.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid59.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid59.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid59.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.59, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_59
+ 	.type	_nds32_vector_59, @function
+ _nds32_vector_59:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid60.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid60.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid60.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid60.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.60, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_60
+ 	.type	_nds32_vector_60, @function
+ _nds32_vector_60:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid61.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid61.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid61.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid61.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.61, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_61
+ 	.type	_nds32_vector_61, @function
+ _nds32_vector_61:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid62.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid62.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid62.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid62.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.62, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_62
+ 	.type	_nds32_vector_62, @function
+ _nds32_vector_62:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid63.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid63.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid63.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid63.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.63, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_63
+ 	.type	_nds32_vector_63, @function
+ _nds32_vector_63:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid64.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid64.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid64.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid64.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.64, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_64
+ 	.type	_nds32_vector_64, @function
+ _nds32_vector_64:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid65.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid65.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid65.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid65.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.65, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_65
+ 	.type	_nds32_vector_65, @function
+ _nds32_vector_65:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid66.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid66.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid66.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid66.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.66, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_66
+ 	.type	_nds32_vector_66, @function
+ _nds32_vector_66:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid67.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid67.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid67.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid67.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.67, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_67
+ 	.type	_nds32_vector_67, @function
+ _nds32_vector_67:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid68.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid68.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid68.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid68.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.68, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_68
+ 	.type	_nds32_vector_68, @function
+ _nds32_vector_68:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid69.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid69.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid69.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid69.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.69, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_69
+ 	.type	_nds32_vector_69, @function
+ _nds32_vector_69:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid70.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid70.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid70.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid70.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.70, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_70
+ 	.type	_nds32_vector_70, @function
+ _nds32_vector_70:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid71.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid71.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid71.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid71.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.71, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_71
+ 	.type	_nds32_vector_71, @function
+ _nds32_vector_71:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid72.S gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid72.S
+--- gcc-8.2.0.orig/libgcc/config/nds32/isr-library/vec_vid72.S	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/isr-library/vec_vid72.S	2019-01-25 15:38:32.845242705 +0100
+@@ -24,8 +24,15 @@
+    <http://www.gnu.org/licenses/>.  */
+ 
+ 	.section	.nds32_vector.72, "ax"
++#if __NDS32_ISR_VECTOR_SIZE_4__
++	/* The vector size is default 4-byte for v3 architecture.  */
++	.vec_size	4
++	.align	2
++#else
++	/* The vector size is default 16-byte for other architectures.  */
+ 	.vec_size	16
+ 	.align	4
++#endif
+ 	.weak	_nds32_vector_72
+ 	.type	_nds32_vector_72, @function
+ _nds32_vector_72:
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/linux-atomic.c gcc-8.2.0/libgcc/config/nds32/linux-atomic.c
+--- gcc-8.2.0.orig/libgcc/config/nds32/linux-atomic.c	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/linux-atomic.c	2019-01-25 15:38:32.845242705 +0100
+@@ -0,0 +1,282 @@
++/* Linux-specific atomic operations for NDS32 Linux.
++   Copyright (C) 2012-2018 Free Software Foundation, Inc.
++
++This file is free software; you can redistribute it and/or modify it
++under the terms of the GNU General Public License as published by the
++Free Software Foundation; either version 3, or (at your option) any
++later version.
++
++This file is distributed in the hope that it will be useful, but
++WITHOUT ANY WARRANTY; without even the implied warranty of
++MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
++General Public License for more details.
++
++Under Section 7 of GPL version 3, you are granted additional
++permissions described in the GCC Runtime Library Exception, version
++3.1, as published by the Free Software Foundation.
++
++You should have received a copy of the GNU General Public License and
++a copy of the GCC Runtime Library Exception along with this program;
++see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
++<http://www.gnu.org/licenses/>.  */
++
++/* We implement byte, short and int versions of each atomic operation
++   using the kernel helper defined below.  There is no support for
++   64-bit operations yet.  */
++
++/* This function copy form NDS32 Linux-kernal. */
++static inline int
++__kernel_cmpxchg (int oldval, int newval, int *mem)
++{
++  int temp1, temp2, temp3, offset;
++
++  asm volatile ("msync\tall\n"
++		"movi\t%0, #0\n"
++		"1:\n"
++		"\tllw\t%1, [%4+%0]\n"
++		"\tsub\t%3, %1, %6\n"
++		"\tcmovz\t%2, %5, %3\n"
++		"\tcmovn\t%2, %1, %3\n"
++		"\tscw\t%2, [%4+%0]\n"
++		"\tbeqz\t%2, 1b\n"
++		: "=&r" (offset), "=&r" (temp3), "=&r" (temp2), "=&r" (temp1)
++		: "r" (mem), "r" (newval), "r" (oldval) : "memory");
++
++  return temp1;
++}
++
++#define HIDDEN __attribute__ ((visibility ("hidden")))
++
++#ifdef __NDS32_EL__
++#define INVERT_MASK_1 0
++#define INVERT_MASK_2 0
++#else
++#define INVERT_MASK_1 24
++#define INVERT_MASK_2 16
++#endif
++
++#define MASK_1 0xffu
++#define MASK_2 0xffffu
++
++#define FETCH_AND_OP_WORD(OP, PFX_OP, INF_OP)				\
++  int HIDDEN								\
++  __sync_fetch_and_##OP##_4 (int *ptr, int val)				\
++  {									\
++    int failure, tmp;							\
++									\
++    do {								\
++      tmp = __atomic_load_n (ptr, __ATOMIC_SEQ_CST);			\
++      failure = __kernel_cmpxchg (tmp, PFX_OP (tmp INF_OP val), ptr);	\
++    } while (failure != 0);						\
++									\
++    return tmp;								\
++  }
++
++FETCH_AND_OP_WORD (add,   , +)
++FETCH_AND_OP_WORD (sub,   , -)
++FETCH_AND_OP_WORD (or,    , |)
++FETCH_AND_OP_WORD (and,   , &)
++FETCH_AND_OP_WORD (xor,   , ^)
++FETCH_AND_OP_WORD (nand, ~, &)
++
++#define NAME_oldval(OP, WIDTH) __sync_fetch_and_##OP##_##WIDTH
++#define NAME_newval(OP, WIDTH) __sync_##OP##_and_fetch_##WIDTH
++
++/* Implement both __sync_<op>_and_fetch and __sync_fetch_and_<op> for
++   subword-sized quantities.  */
++
++#define SUBWORD_SYNC_OP(OP, PFX_OP, INF_OP, TYPE, WIDTH, RETURN)	\
++  TYPE HIDDEN								\
++  NAME##_##RETURN (OP, WIDTH) (TYPE *ptr, TYPE val)			\
++  {									\
++    int *wordptr = (int *) ((unsigned long) ptr & ~3);			\
++    unsigned int mask, shift, oldval, newval;				\
++    int failure;							\
++									\
++    shift = (((unsigned long) ptr & 3) << 3) ^ INVERT_MASK_##WIDTH;	\
++    mask = MASK_##WIDTH << shift;					\
++									\
++    do {								\
++      oldval = __atomic_load_n (wordptr, __ATOMIC_SEQ_CST);		\
++      newval = ((PFX_OP (((oldval & mask) >> shift)			\
++			 INF_OP (unsigned int) val)) << shift) & mask;	\
++      newval |= oldval & ~mask;						\
++      failure = __kernel_cmpxchg (oldval, newval, wordptr);		\
++    } while (failure != 0);						\
++									\
++    return (RETURN & mask) >> shift;					\
++  }
++
++
++SUBWORD_SYNC_OP (add,   , +, unsigned short, 2, oldval)
++SUBWORD_SYNC_OP (sub,   , -, unsigned short, 2, oldval)
++SUBWORD_SYNC_OP (or,    , |, unsigned short, 2, oldval)
++SUBWORD_SYNC_OP (and,   , &, unsigned short, 2, oldval)
++SUBWORD_SYNC_OP (xor,   , ^, unsigned short, 2, oldval)
++SUBWORD_SYNC_OP (nand, ~, &, unsigned short, 2, oldval)
++
++SUBWORD_SYNC_OP (add,   , +, unsigned char, 1, oldval)
++SUBWORD_SYNC_OP (sub,   , -, unsigned char, 1, oldval)
++SUBWORD_SYNC_OP (or,    , |, unsigned char, 1, oldval)
++SUBWORD_SYNC_OP (and,   , &, unsigned char, 1, oldval)
++SUBWORD_SYNC_OP (xor,   , ^, unsigned char, 1, oldval)
++SUBWORD_SYNC_OP (nand, ~, &, unsigned char, 1, oldval)
++
++#define OP_AND_FETCH_WORD(OP, PFX_OP, INF_OP)				\
++  int HIDDEN								\
++  __sync_##OP##_and_fetch_4 (int *ptr, int val)				\
++  {									\
++    int tmp, failure;							\
++									\
++    do {								\
++      tmp = __atomic_load_n (ptr, __ATOMIC_SEQ_CST);			\
++      failure = __kernel_cmpxchg (tmp, PFX_OP (tmp INF_OP val), ptr);	\
++    } while (failure != 0);						\
++									\
++    return PFX_OP (tmp INF_OP val);					\
++  }
++
++OP_AND_FETCH_WORD (add,   , +)
++OP_AND_FETCH_WORD (sub,   , -)
++OP_AND_FETCH_WORD (or,    , |)
++OP_AND_FETCH_WORD (and,   , &)
++OP_AND_FETCH_WORD (xor,   , ^)
++OP_AND_FETCH_WORD (nand, ~, &)
++
++SUBWORD_SYNC_OP (add,   , +, unsigned short, 2, newval)
++SUBWORD_SYNC_OP (sub,   , -, unsigned short, 2, newval)
++SUBWORD_SYNC_OP (or,    , |, unsigned short, 2, newval)
++SUBWORD_SYNC_OP (and,   , &, unsigned short, 2, newval)
++SUBWORD_SYNC_OP (xor,   , ^, unsigned short, 2, newval)
++SUBWORD_SYNC_OP (nand, ~, &, unsigned short, 2, newval)
++
++SUBWORD_SYNC_OP (add,   , +, unsigned char, 1, newval)
++SUBWORD_SYNC_OP (sub,   , -, unsigned char, 1, newval)
++SUBWORD_SYNC_OP (or,    , |, unsigned char, 1, newval)
++SUBWORD_SYNC_OP (and,   , &, unsigned char, 1, newval)
++SUBWORD_SYNC_OP (xor,   , ^, unsigned char, 1, newval)
++SUBWORD_SYNC_OP (nand, ~, &, unsigned char, 1, newval)
++
++int HIDDEN
++__sync_val_compare_and_swap_4 (int *ptr, int oldval, int newval)
++{
++  int actual_oldval, fail;
++
++  while (1)
++    {
++      actual_oldval = __atomic_load_n (ptr, __ATOMIC_SEQ_CST);
++
++      if (oldval != actual_oldval)
++	return actual_oldval;
++
++      fail = __kernel_cmpxchg (actual_oldval, newval, ptr);
++
++      if (!fail)
++	return oldval;
++    }
++}
++
++#define SUBWORD_VAL_CAS(TYPE, WIDTH)					\
++  TYPE HIDDEN								\
++  __sync_val_compare_and_swap_##WIDTH (TYPE *ptr, TYPE oldval,		\
++				       TYPE newval)			\
++  {									\
++    int *wordptr = (int *)((unsigned long) ptr & ~3), fail;		\
++    unsigned int mask, shift, actual_oldval, actual_newval;		\
++									\
++    shift = (((unsigned long) ptr & 3) << 3) ^ INVERT_MASK_##WIDTH;	\
++    mask = MASK_##WIDTH << shift;					\
++									\
++    while (1)								\
++      {									\
++	actual_oldval = __atomic_load_n (wordptr, __ATOMIC_SEQ_CST); 	\
++									\
++	if (((actual_oldval & mask) >> shift) != (unsigned int) oldval)	\
++	  return (actual_oldval & mask) >> shift;			\
++									\
++	actual_newval = (actual_oldval & ~mask)				\
++			| (((unsigned int) newval << shift) & mask);	\
++									\
++	fail = __kernel_cmpxchg (actual_oldval, actual_newval,		\
++				 wordptr);				\
++									\
++	if (!fail)							\
++	  return oldval;						\
++      }									\
++  }
++
++SUBWORD_VAL_CAS (unsigned short, 2)
++SUBWORD_VAL_CAS (unsigned char,  1)
++
++typedef unsigned char bool;
++
++bool HIDDEN
++__sync_bool_compare_and_swap_4 (int *ptr, int oldval, int newval)
++{
++  int failure = __kernel_cmpxchg (oldval, newval, ptr);
++  return (failure == 0);
++}
++
++#define SUBWORD_BOOL_CAS(TYPE, WIDTH)					\
++  bool HIDDEN								\
++  __sync_bool_compare_and_swap_##WIDTH (TYPE *ptr, TYPE oldval,		\
++					TYPE newval)			\
++  {									\
++    TYPE actual_oldval							\
++      = __sync_val_compare_and_swap_##WIDTH (ptr, oldval, newval);	\
++    return (oldval == actual_oldval);					\
++  }
++
++SUBWORD_BOOL_CAS (unsigned short, 2)
++SUBWORD_BOOL_CAS (unsigned char,  1)
++
++int HIDDEN
++__sync_lock_test_and_set_4 (int *ptr, int val)
++{
++  int failure, oldval;
++
++  do {
++    oldval = __atomic_load_n (ptr, __ATOMIC_SEQ_CST);
++    failure = __kernel_cmpxchg (oldval, val, ptr);
++  } while (failure != 0);
++
++  return oldval;
++}
++
++#define SUBWORD_TEST_AND_SET(TYPE, WIDTH)				\
++  TYPE HIDDEN								\
++  __sync_lock_test_and_set_##WIDTH (TYPE *ptr, TYPE val)		\
++  {									\
++    int failure;							\
++    unsigned int oldval, newval, shift, mask;				\
++    int *wordptr = (int *) ((unsigned long) ptr & ~3);			\
++									\
++    shift = (((unsigned long) ptr & 3) << 3) ^ INVERT_MASK_##WIDTH;	\
++    mask = MASK_##WIDTH << shift;					\
++									\
++    do {								\
++      oldval = __atomic_load_n (wordptr, __ATOMIC_SEQ_CST);		\
++      newval = (oldval & ~mask)						\
++	       | (((unsigned int) val << shift) & mask);		\
++      failure = __kernel_cmpxchg (oldval, newval, wordptr);		\
++    } while (failure != 0);						\
++									\
++    return (oldval & mask) >> shift;					\
++  }
++
++SUBWORD_TEST_AND_SET (unsigned short, 2)
++SUBWORD_TEST_AND_SET (unsigned char,  1)
++
++#define SYNC_LOCK_RELEASE(TYPE, WIDTH)					\
++  void HIDDEN								\
++  __sync_lock_release_##WIDTH (TYPE *ptr)				\
++  {									\
++    /* All writes before this point must be seen before we release	\
++       the lock itself.  */						\
++    __builtin_nds32_msync_all ();					\
++    *ptr = 0;								\
++  }
++
++SYNC_LOCK_RELEASE (int,   4)
++SYNC_LOCK_RELEASE (short, 2)
++SYNC_LOCK_RELEASE (char,  1)
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/linux-unwind.h gcc-8.2.0/libgcc/config/nds32/linux-unwind.h
+--- gcc-8.2.0.orig/libgcc/config/nds32/linux-unwind.h	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/linux-unwind.h	2019-01-25 15:38:32.845242705 +0100
+@@ -0,0 +1,143 @@
++/* DWARF2 EH unwinding support for NDS32 Linux signal frame.
++   Copyright (C) 2014-2015 Free Software Foundation, Inc.
++   Contributed by Andes Technology Corporation.
++
++   This file is part of GCC.
++
++   GCC is free software; you can redistribute it and/or modify it
++   under the terms of the GNU General Public License as published
++   by the Free Software Foundation; either version 3, or (at your
++   option) any later version.
++
++   GCC is distributed in the hope that it will be useful, but WITHOUT
++   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++   License for more details.
++
++   Under Section 7 of GPL version 3, you are granted additional
++   permissions described in the GCC Runtime Library Exception, version
++   3.1, as published by the Free Software Foundation.
++
++   You should have received a copy of the GNU General Public License and
++   a copy of the GCC Runtime Library Exception along with this program;
++   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
++   <http://www.gnu.org/licenses/>.  */
++
++#ifndef inhibit_libc
++
++/* Do code reading to identify a signal frame, and set the frame
++   state data appropriately.  See unwind-dw2.c for the structs.
++   The corresponding bits in the Linux kernel are in
++   arch/nds32/kernel/signal.c.  */
++
++#include <signal.h>
++#include <asm/unistd.h>
++#include <sys/ucontext.h>
++
++/* Exactly the same layout as the kernel structures, unique names.  */
++
++/* arch/nds32/kernel/signal.c */
++struct _rt_sigframe {
++  siginfo_t info;
++  struct ucontext_t uc;
++};
++
++#define RT_SIGRETURN 0x8b00f044
++
++#define MD_FALLBACK_FRAME_STATE_FOR nds32_fallback_frame_state
++
++/* This function is supposed to be invoked by uw_frame_state_for()
++   when there is no unwind data available.
++
++   Generally, given the _Unwind_Context CONTEXT for a stack frame,
++   we need to look up its caller and decode information into FS.
++   However, if the exception handling happens within a signal handler,
++   the return address of signal handler is a special module, which
++   contains signal return syscall and has no FDE in the .eh_frame section.
++   We need to implement MD_FALLBACK_FRAME_STATE_FOR so that we can
++   unwind through signal frames.  */
++static _Unwind_Reason_Code
++nds32_fallback_frame_state (struct _Unwind_Context *context,
++			    _Unwind_FrameState *fs)
++{
++  u_int32_t *pc = (u_int32_t *) context->ra;
++  struct sigcontext *sc_;
++  _Unwind_Ptr new_cfa;
++
++#ifdef __NDS32_EB__
++#error "Signal handler is not supported for force unwind."
++#endif
++
++  if ((_Unwind_Ptr) pc & 3)
++    return _URC_END_OF_STACK;
++
++  /* Check if we are going through a signal handler.
++     See arch/nds32/kernel/signal.c implementation.
++     FIXME: Currently we only handle little endian (EL) case.  */
++  if (pc[0] == RT_SIGRETURN)
++    {
++      /* Using '_sigfame' memory address to locate kernal's sigcontext.
++	 The sigcontext structures in arch/nds32/include/asm/sigcontext.h.  */
++      struct _rt_sigframe *rt_;
++      rt_ = context->cfa;
++      sc_ = &rt_->uc.uc_mcontext;
++    }
++  else
++    return _URC_END_OF_STACK;
++
++  /* Update cfa from sigcontext.  */
++  new_cfa = (_Unwind_Ptr) sc_;
++  fs->regs.cfa_how = CFA_REG_OFFSET;
++  fs->regs.cfa_reg = STACK_POINTER_REGNUM;
++  fs->regs.cfa_offset = new_cfa - (_Unwind_Ptr) context->cfa;
++
++#define NDS32_PUT_FS_REG(NUM, NAME) \
++  (fs->regs.reg[NUM].how = REG_SAVED_OFFSET, \
++   fs->regs.reg[NUM].loc.offset = (_Unwind_Ptr) &(sc_->NAME) - new_cfa)
++
++  /* Restore all registers value.  */
++  NDS32_PUT_FS_REG (0, nds32_r0);
++  NDS32_PUT_FS_REG (1, nds32_r1);
++  NDS32_PUT_FS_REG (2, nds32_r2);
++  NDS32_PUT_FS_REG (3, nds32_r3);
++  NDS32_PUT_FS_REG (4, nds32_r4);
++  NDS32_PUT_FS_REG (5, nds32_r5);
++  NDS32_PUT_FS_REG (6, nds32_r6);
++  NDS32_PUT_FS_REG (7, nds32_r7);
++  NDS32_PUT_FS_REG (8, nds32_r8);
++  NDS32_PUT_FS_REG (9, nds32_r9);
++  NDS32_PUT_FS_REG (10, nds32_r10);
++  NDS32_PUT_FS_REG (11, nds32_r11);
++  NDS32_PUT_FS_REG (12, nds32_r12);
++  NDS32_PUT_FS_REG (13, nds32_r13);
++  NDS32_PUT_FS_REG (14, nds32_r14);
++  NDS32_PUT_FS_REG (15, nds32_r15);
++  NDS32_PUT_FS_REG (16, nds32_r16);
++  NDS32_PUT_FS_REG (17, nds32_r17);
++  NDS32_PUT_FS_REG (18, nds32_r18);
++  NDS32_PUT_FS_REG (19, nds32_r19);
++  NDS32_PUT_FS_REG (20, nds32_r20);
++  NDS32_PUT_FS_REG (21, nds32_r21);
++  NDS32_PUT_FS_REG (22, nds32_r22);
++  NDS32_PUT_FS_REG (23, nds32_r23);
++  NDS32_PUT_FS_REG (24, nds32_r24);
++  NDS32_PUT_FS_REG (25, nds32_r25);
++
++  NDS32_PUT_FS_REG (28, nds32_fp);
++  NDS32_PUT_FS_REG (29, nds32_gp);
++  NDS32_PUT_FS_REG (30, nds32_lp);
++  NDS32_PUT_FS_REG (31, nds32_sp);
++
++  /* Restore PC, point to trigger signal instruction.  */
++  NDS32_PUT_FS_REG (32, nds32_ipc);
++
++#undef NDS32_PUT_FS_REG
++
++  /* The retaddr is PC, use PC to find FDE.  */
++  fs->retaddr_column = 32;
++  fs->signal_frame = 1;
++
++  return _URC_NO_REASON;
++}
++
++#endif
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/t-nds32-glibc gcc-8.2.0/libgcc/config/nds32/t-nds32-glibc
+--- gcc-8.2.0.orig/libgcc/config/nds32/t-nds32-glibc	1970-01-01 01:00:00.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/t-nds32-glibc	2019-01-25 15:38:37.357255536 +0100
+@@ -0,0 +1,34 @@
++# Rules of glibc library makefile of Andes NDS32 cpu for GNU compiler
++# Copyright (C) 2012-2015 Free Software Foundation, Inc.
++# Contributed by Andes Technology Corporation.
++#
++# This file is part of GCC.
++#
++# GCC is free software; you can redistribute it and/or modify it
++# under the terms of the GNU General Public License as published
++# by the Free Software Foundation; either version 3, or (at your
++# option) any later version.
++#
++# GCC is distributed in the hope that it will be useful, but WITHOUT
++# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
++# or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
++# License for more details.
++#
++# You should have received a copy of the GNU General Public License
++# along with GCC; see the file COPYING3.  If not see
++# <http://www.gnu.org/licenses/>.
++
++# Compiler flags to use when compiling 'libgcc2.c'
++HOST_LIBGCC2_CFLAGS = -O2 -fPIC -fwrapv
++LIB2ADD += $(srcdir)/config/nds32/linux-atomic.c
++
++#LIB1ASMSRC   = nds32/lib1asmsrc-newlib.S
++#LIB1ASMFUNCS = _divsi3 _modsi3 _udivsi3 _umodsi3
++
++# List of functions not to build from libgcc2.c.
++#LIB2FUNCS_EXCLUDE = _clzsi2
++
++# List of extra C and assembler files(*.S) to add to static libgcc2.
++#LIB2ADD_ST += $(srcdir)/config/nds32/lib2csrc-newlib/_clzsi2.c
++
++# ------------------------------------------------------------------------
+diff -urN gcc-8.2.0.orig/libgcc/config/nds32/t-nds32-isr gcc-8.2.0/libgcc/config/nds32/t-nds32-isr
+--- gcc-8.2.0.orig/libgcc/config/nds32/t-nds32-isr	2018-01-03 11:03:58.000000000 +0100
++++ gcc-8.2.0/libgcc/config/nds32/t-nds32-isr	2019-01-25 15:38:37.357255536 +0100
+@@ -23,11 +23,11 @@
+ # Makfile fragment rules for libnds32_isr.a to support ISR attribute extension
+ ###############################################################################
+ 
+-# basic flags setting
++# Basic flags setting.
+ ISR_CFLAGS = $(CFLAGS) -c
+ 
+-# the object files we would like to create
+-LIBNDS32_ISR_16B_OBJS = \
++# The object files we would like to create.
++LIBNDS32_ISR_VEC_OBJS = \
+ 		vec_vid00.o vec_vid01.o vec_vid02.o vec_vid03.o \
+ 		vec_vid04.o vec_vid05.o vec_vid06.o vec_vid07.o \
+ 		vec_vid08.o vec_vid09.o vec_vid10.o vec_vid11.o \
+@@ -46,40 +46,9 @@
+ 		vec_vid60.o vec_vid61.o vec_vid62.o vec_vid63.o \
+ 		vec_vid64.o vec_vid65.o vec_vid66.o vec_vid67.o \
+ 		vec_vid68.o vec_vid69.o vec_vid70.o vec_vid71.o \
+-		vec_vid72.o \
+-		excp_isr_ps_nn.o excp_isr_ps_ns.o excp_isr_ps_nr.o \
+-		excp_isr_sa_nn.o excp_isr_sa_ns.o excp_isr_sa_nr.o \
+-		intr_isr_ps_nn.o intr_isr_ps_ns.o intr_isr_ps_nr.o \
+-		intr_isr_sa_nn.o intr_isr_sa_ns.o intr_isr_sa_nr.o \
+-		reset.o
+-
+-LIBNDS32_ISR_4B_OBJS = \
+-		vec_vid00_4b.o vec_vid01_4b.o vec_vid02_4b.o vec_vid03_4b.o \
+-		vec_vid04_4b.o vec_vid05_4b.o vec_vid06_4b.o vec_vid07_4b.o \
+-		vec_vid08_4b.o vec_vid09_4b.o vec_vid10_4b.o vec_vid11_4b.o \
+-		vec_vid12_4b.o vec_vid13_4b.o vec_vid14_4b.o vec_vid15_4b.o \
+-		vec_vid16_4b.o vec_vid17_4b.o vec_vid18_4b.o vec_vid19_4b.o \
+-		vec_vid20_4b.o vec_vid21_4b.o vec_vid22_4b.o vec_vid23_4b.o \
+-		vec_vid24_4b.o vec_vid25_4b.o vec_vid26_4b.o vec_vid27_4b.o \
+-		vec_vid28_4b.o vec_vid29_4b.o vec_vid30_4b.o vec_vid31_4b.o \
+-		vec_vid32_4b.o vec_vid33_4b.o vec_vid34_4b.o vec_vid35_4b.o \
+-		vec_vid36_4b.o vec_vid37_4b.o vec_vid38_4b.o vec_vid39_4b.o \
+-		vec_vid40_4b.o vec_vid41_4b.o vec_vid42_4b.o vec_vid43_4b.o \
+-		vec_vid44_4b.o vec_vid45_4b.o vec_vid46_4b.o vec_vid47_4b.o \
+-		vec_vid48_4b.o vec_vid49_4b.o vec_vid50_4b.o vec_vid51_4b.o \
+-		vec_vid52_4b.o vec_vid53_4b.o vec_vid54_4b.o vec_vid55_4b.o \
+-		vec_vid56_4b.o vec_vid57_4b.o vec_vid58_4b.o vec_vid59_4b.o \
+-		vec_vid60_4b.o vec_vid61_4b.o vec_vid62_4b.o vec_vid63_4b.o \
+-		vec_vid64_4b.o vec_vid65_4b.o vec_vid66_4b.o vec_vid67_4b.o \
+-		vec_vid68_4b.o vec_vid69_4b.o vec_vid70_4b.o vec_vid71_4b.o \
+-		vec_vid72_4b.o \
+-		excp_isr_ps_nn_4b.o excp_isr_ps_ns_4b.o excp_isr_ps_nr_4b.o \
+-		excp_isr_sa_nn_4b.o excp_isr_sa_ns_4b.o excp_isr_sa_nr_4b.o \
+-		intr_isr_ps_nn_4b.o intr_isr_ps_ns_4b.o intr_isr_ps_nr_4b.o \
+-		intr_isr_sa_nn_4b.o intr_isr_sa_ns_4b.o intr_isr_sa_nr_4b.o \
+-		reset_4b.o
++		vec_vid72.o
+ 
+-LIBNDS32_ISR_COMMON_OBJS = \
++LIBNDS32_ISR_JMP_OBJS = \
+ 		jmptbl_vid00.o jmptbl_vid01.o jmptbl_vid02.o jmptbl_vid03.o \
+ 		jmptbl_vid04.o jmptbl_vid05.o jmptbl_vid06.o jmptbl_vid07.o \
+ 		jmptbl_vid08.o jmptbl_vid09.o jmptbl_vid10.o jmptbl_vid11.o \
+@@ -98,29 +67,32 @@
+ 		jmptbl_vid60.o jmptbl_vid61.o jmptbl_vid62.o jmptbl_vid63.o \
+ 		jmptbl_vid64.o jmptbl_vid65.o jmptbl_vid66.o jmptbl_vid67.o \
+ 		jmptbl_vid68.o jmptbl_vid69.o jmptbl_vid70.o jmptbl_vid71.o \
+-		jmptbl_vid72.o \
++		jmptbl_vid72.o
++
++LIBNDS32_ISR_COMMON_OBJS = \
++		excp_isr_ps_nn.o excp_isr_ps_ns.o excp_isr_ps_nr.o \
++		excp_isr_sa_nn.o excp_isr_sa_ns.o excp_isr_sa_nr.o \
++		intr_isr_ps_nn.o intr_isr_ps_ns.o intr_isr_ps_nr.o \
++		intr_isr_sa_nn.o intr_isr_sa_ns.o intr_isr_sa_nr.o \
++		reset.o \
+ 		nmih.o \
+ 		wrh.o
+ 
+-LIBNDS32_ISR_COMPLETE_OBJS = $(LIBNDS32_ISR_16B_OBJS) $(LIBNDS32_ISR_4B_OBJS) $(LIBNDS32_ISR_COMMON_OBJS)
+-
++LIBNDS32_ISR_COMPLETE_OBJS = $(LIBNDS32_ISR_VEC_OBJS) $(LIBNDS32_ISR_JMP_OBJS) $(LIBNDS32_ISR_COMMON_OBJS)
+ 
+-# Build common objects for ISR library
+-nmih.o: $(srcdir)/config/nds32/isr-library/nmih.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/nmih.S -o nmih.o
+ 
+-wrh.o: $(srcdir)/config/nds32/isr-library/wrh.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/wrh.S -o wrh.o
+ 
+-jmptbl_vid%.o: $(srcdir)/config/nds32/isr-library/jmptbl_vid%.S
++# Build vector vid objects for ISR library.
++vec_vid%.o: $(srcdir)/config/nds32/isr-library/vec_vid%.S
+ 	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $< -o $@
+ 
+ 
+-
+-# Build 16b version objects for ISR library. (no "_4b" postfix string)
+-vec_vid%.o: $(srcdir)/config/nds32/isr-library/vec_vid%.S
++# Build jump table objects for ISR library.
++jmptbl_vid%.o: $(srcdir)/config/nds32/isr-library/jmptbl_vid%.S
+ 	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $< -o $@
+ 
++
++# Build commen objects for ISR library.
+ excp_isr_ps_nn.o: $(srcdir)/config/nds32/isr-library/excp_isr.S
+ 	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/excp_isr.S -o excp_isr_ps_nn.o
+ 
+@@ -160,48 +132,12 @@
+ reset.o: $(srcdir)/config/nds32/isr-library/reset.S
+ 	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/reset.S -o reset.o
+ 
+-# Build 4b version objects for ISR library.
+-vec_vid%_4b.o: $(srcdir)/config/nds32/isr-library/vec_vid%_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $< -o $@
+-
+-excp_isr_ps_nn_4b.o: $(srcdir)/config/nds32/isr-library/excp_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/excp_isr_4b.S -o excp_isr_ps_nn_4b.o
+-
+-excp_isr_ps_ns_4b.o: $(srcdir)/config/nds32/isr-library/excp_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_NESTED $(srcdir)/config/nds32/isr-library/excp_isr_4b.S -o excp_isr_ps_ns_4b.o
+-
+-excp_isr_ps_nr_4b.o: $(srcdir)/config/nds32/isr-library/excp_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_NESTED_READY $(srcdir)/config/nds32/isr-library/excp_isr_4b.S -o excp_isr_ps_nr_4b.o
+-
+-excp_isr_sa_nn_4b.o: $(srcdir)/config/nds32/isr-library/excp_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_SAVE_ALL_REGS $(srcdir)/config/nds32/isr-library/excp_isr_4b.S -o excp_isr_sa_nn_4b.o
+-
+-excp_isr_sa_ns_4b.o: $(srcdir)/config/nds32/isr-library/excp_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_SAVE_ALL_REGS -DNDS32_NESTED $(srcdir)/config/nds32/isr-library/excp_isr_4b.S -o excp_isr_sa_ns_4b.o
+-
+-excp_isr_sa_nr_4b.o: $(srcdir)/config/nds32/isr-library/excp_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_SAVE_ALL_REGS -DNDS32_NESTED_READY $(srcdir)/config/nds32/isr-library/excp_isr_4b.S -o excp_isr_sa_nr_4b.o
+-
+-intr_isr_ps_nn_4b.o: $(srcdir)/config/nds32/isr-library/intr_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/intr_isr_4b.S -o intr_isr_ps_nn_4b.o
+-
+-intr_isr_ps_ns_4b.o: $(srcdir)/config/nds32/isr-library/intr_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_NESTED $(srcdir)/config/nds32/isr-library/intr_isr_4b.S -o intr_isr_ps_ns_4b.o
+-
+-intr_isr_ps_nr_4b.o: $(srcdir)/config/nds32/isr-library/intr_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_NESTED_READY $(srcdir)/config/nds32/isr-library/intr_isr_4b.S -o intr_isr_ps_nr_4b.o
+-
+-intr_isr_sa_nn_4b.o: $(srcdir)/config/nds32/isr-library/intr_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_SAVE_ALL_REGS $(srcdir)/config/nds32/isr-library/intr_isr_4b.S -o intr_isr_sa_nn_4b.o
+-
+-intr_isr_sa_ns_4b.o: $(srcdir)/config/nds32/isr-library/intr_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_SAVE_ALL_REGS -DNDS32_NESTED $(srcdir)/config/nds32/isr-library/intr_isr_4b.S -o intr_isr_sa_ns_4b.o
++nmih.o: $(srcdir)/config/nds32/isr-library/nmih.S
++	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/nmih.S -o nmih.o
+ 
+-intr_isr_sa_nr_4b.o: $(srcdir)/config/nds32/isr-library/intr_isr_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) -DNDS32_SAVE_ALL_REGS -DNDS32_NESTED_READY $(srcdir)/config/nds32/isr-library/intr_isr_4b.S -o intr_isr_sa_nr_4b.o
++wrh.o: $(srcdir)/config/nds32/isr-library/wrh.S
++	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/wrh.S -o wrh.o
+ 
+-reset_4b.o: $(srcdir)/config/nds32/isr-library/reset_4b.S
+-	$(GCC_FOR_TARGET) $(ISR_CFLAGS) $(srcdir)/config/nds32/isr-library/reset_4b.S -o reset_4b.o
+ 
+ 
+ # The rule to create libnds32_isr.a file
+diff -urN gcc-8.2.0.orig/libgcc/config.host gcc-8.2.0/libgcc/config.host
+--- gcc-8.2.0.orig/libgcc/config.host	2018-04-06 22:04:17.000000000 +0200
++++ gcc-8.2.0/libgcc/config.host	2019-01-25 15:38:32.841242694 +0100
+@@ -974,6 +974,23 @@
+ 	tmake_file="$tm_file t-crtstuff t-fdpbit msp430/t-msp430"
+         extra_parts="$extra_parts libmul_none.a libmul_16.a libmul_32.a libmul_f5.a"
+ 	;;
++nds32*-linux*)
++	# Basic makefile fragment and extra_parts for crt stuff.
++	# We also append c-isr library implementation.
++	tmake_file="${tmake_file} t-slibgcc-libgcc"
++	tmake_file="${tmake_file} nds32/t-nds32-glibc nds32/t-crtstuff t-softfp-sfdf t-softfp"
++	# The header file of defining MD_FALLBACK_FRAME_STATE_FOR.
++	md_unwind_header=nds32/linux-unwind.h
++	# Append library definition makefile fragment according to --with-nds32-lib=X setting.
++	case "${with_nds32_lib}" in
++	"" | glibc | uclibc )
++		;;
++	*)
++		echo "Cannot accept --with-nds32-lib=$with_nds32_lib, available values are: glibc uclibc" 1>&2
++		exit 1
++		;;
++	esac
++	;;
+ nds32*-elf*)
+ 	# Basic makefile fragment and extra_parts for crt stuff.
+ 	# We also append c-isr library implementation.
diff --git a/util/crossgcc/sum/gcc-8.3.0.tar.xz.cksum b/util/crossgcc/sum/gcc-8.3.0.tar.xz.cksum
new file mode 100644
index 000000000000..b46ef8c46f6a
--- /dev/null
+++ b/util/crossgcc/sum/gcc-8.3.0.tar.xz.cksum
@@ -0,0 +1 @@
+c27f4499dd263fe4fb01bcc5565917f3698583b2  tarballs/gcc-8.3.0.tar.xz
diff --git a/util/crossgcc/sum/gcc-9.2.0.tar.xz.cksum b/util/crossgcc/sum/gcc-9.2.0.tar.xz.cksum
deleted file mode 100644
index 767c6b0a981c..000000000000
--- a/util/crossgcc/sum/gcc-9.2.0.tar.xz.cksum
+++ /dev/null
@@ -1 +0,0 @@
-306d27c3465fa36862c206738d06d65fff5c3645  tarballs/gcc-9.2.0.tar.xz
diff --git a/util/xcompile/xcompile b/util/xcompile/xcompile
index 59908c5b08c3..3203d71899b9 100755
--- a/util/xcompile/xcompile
+++ b/util/xcompile/xcompile
@@ -185,6 +185,8 @@ detect_special_flags() {
 	testcc "$GCC"      "$CFLAGS_GCC -Wl,--build-id=none" &&
 		CFLAGS_GCC="$CFLAGS_GCC -Wl,--build-id=none"
 
+	testcc "$GCC"      "$CFLAGS_GCC -Wno-address-of-packed-member" &&
+		CFLAGS_GCC="$CFLAGS_GCC -Wno-address-of-packed-member"
 	case "$architecture" in
 	x86)
 		;;
@@ -219,7 +221,7 @@ SUBARCH_SUPPORTED+=${TSUPP-${TARCH}}
 GCC_CC_${TARCH}:=${GCC}
 GCC_CFLAGS_${TARCH}:=${CFLAGS_GCC}
 # Generally available for GCC's cc1:
-GCC_CFLAGS_${TARCH}+=-fno-delete-null-pointer-checks -Wlogical-op -Wno-address-of-packed-member
+GCC_CFLAGS_${TARCH}+=-fno-delete-null-pointer-checks -Wlogical-op
 GCC_ADAFLAGS_${TARCH}:=${CFLAGS_GCC}
 GCC_COMPILER_RT_${TARCH}:=${CC_RT_GCC}
 GCC_COMPILER_RT_FLAGS_${TARCH}:=${CC_RT_EXTRA_GCC}