From 474fd6e80fe529e9adeeb7ea9d4e5d6c4da0b7fe Mon Sep 17 00:00:00 2001 From: Martin Schwidefsky Date: Tue, 23 Aug 2016 13:30:24 +0200 Subject: RAID/s390: add SIMD implementation for raid6 gen/xor Using vector registers is slightly faster: raid6: vx128x8 gen() 19705 MB/s raid6: vx128x8 xor() 11886 MB/s raid6: using algorithm vx128x8 gen() 19705 MB/s raid6: .... xor() 11886 MB/s, rmw enabled vs the software algorithms: raid6: int64x1 gen() 3018 MB/s raid6: int64x1 xor() 1429 MB/s raid6: int64x2 gen() 4661 MB/s raid6: int64x2 xor() 3143 MB/s raid6: int64x4 gen() 5392 MB/s raid6: int64x4 xor() 3509 MB/s raid6: int64x8 gen() 4441 MB/s raid6: int64x8 xor() 3207 MB/s raid6: using algorithm int64x4 gen() 5392 MB/s raid6: .... xor() 3509 MB/s, rmw enabled Signed-off-by: Martin Schwidefsky --- lib/raid6/s390vx.uc | 168 ++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 168 insertions(+) create mode 100644 lib/raid6/s390vx.uc (limited to 'lib/raid6/s390vx.uc') diff --git a/lib/raid6/s390vx.uc b/lib/raid6/s390vx.uc new file mode 100644 index 000000000000..7b45191a655f --- /dev/null +++ b/lib/raid6/s390vx.uc @@ -0,0 +1,168 @@ +/* + * raid6_vx$#.c + * + * $#-way unrolled RAID6 gen/xor functions for s390 + * based on the vector facility + * + * Copyright IBM Corp. 2016 + * Author(s): Martin Schwidefsky + * + * This file is postprocessed using unroll.awk. + */ + +#include +#include + +asm(".include \"asm/vx-insn.h\"\n"); + +#define NSIZE 16 + +static inline void LOAD_CONST(void) +{ + asm volatile("VREPIB %v24,7"); + asm volatile("VREPIB %v25,0x1d"); +} + +/* + * The SHLBYTE() operation shifts each of the 16 bytes in + * vector register y left by 1 bit and stores the result in + * vector register x. + */ +static inline void SHLBYTE(int x, int y) +{ + asm volatile ("VAB %0,%1,%1" : : "i" (x), "i" (y)); +} + +/* + * For each of the 16 bytes in the vector register y the MASK() + * operation returns 0xFF if the high bit of the byte is 1, + * or 0x00 if the high bit is 0. The result is stored in vector + * register x. + */ +static inline void MASK(int x, int y) +{ + asm volatile ("VESRAVB %0,%1,24" : : "i" (x), "i" (y)); +} + +static inline void AND(int x, int y, int z) +{ + asm volatile ("VN %0,%1,%2" : : "i" (x), "i" (y), "i" (z)); +} + +static inline void XOR(int x, int y, int z) +{ + asm volatile ("VX %0,%1,%2" : : "i" (x), "i" (y), "i" (z)); +} + +static inline void LOAD_DATA(int x, int n, u8 *ptr) +{ + typedef struct { u8 _[16*n]; } addrtype; + register addrtype *__ptr asm("1") = (addrtype *) ptr; + + asm volatile ("VLM %2,%3,0,%r1" + : : "m" (*__ptr), "a" (__ptr), "i" (x), "i" (x + n - 1)); +} + +static inline void STORE_DATA(int x, int n, u8 *ptr) +{ + typedef struct { u8 _[16*n]; } addrtype; + register addrtype *__ptr asm("1") = (addrtype *) ptr; + + asm volatile ("VSTM %2,%3,0,1" + : "=m" (*__ptr) : "a" (__ptr), "i" (x), "i" (x + n - 1)); +} + +static inline void COPY_VEC(int x, int y) +{ + asm volatile ("VLR %0,%1" : : "i" (x), "i" (y)); +} + +static void raid6_s390vx$#_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + struct kernel_fpu vxstate; + u8 **dptr, *p, *q; + int d, z, z0; + + kernel_fpu_begin(&vxstate, KERNEL_VXR); + LOAD_CONST(); + + dptr = (u8 **) ptrs; + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0 + 1]; /* XOR parity */ + q = dptr[z0 + 2]; /* RS syndrome */ + + for (d = 0; d < bytes; d += $#*NSIZE) { + LOAD_DATA(0,$#,&dptr[z0][d]); + COPY_VEC(8+$$,0+$$); + for (z = z0 - 1; z >= 0; z--) { + MASK(16+$$,8+$$); + AND(16+$$,16+$$,25); + SHLBYTE(8+$$,8+$$); + XOR(8+$$,8+$$,16+$$); + LOAD_DATA(16,$#,&dptr[z][d]); + XOR(0+$$,0+$$,16+$$); + XOR(8+$$,8+$$,16+$$); + } + STORE_DATA(0,$#,&p[d]); + STORE_DATA(8,$#,&q[d]); + } + kernel_fpu_end(&vxstate, KERNEL_VXR); +} + +static void raid6_s390vx$#_xor_syndrome(int disks, int start, int stop, + size_t bytes, void **ptrs) +{ + struct kernel_fpu vxstate; + u8 **dptr, *p, *q; + int d, z, z0; + + dptr = (u8 **) ptrs; + z0 = stop; /* P/Q right side optimization */ + p = dptr[disks - 2]; /* XOR parity */ + q = dptr[disks - 1]; /* RS syndrome */ + + kernel_fpu_begin(&vxstate, KERNEL_VXR); + LOAD_CONST(); + + for (d = 0; d < bytes; d += $#*NSIZE) { + /* P/Q data pages */ + LOAD_DATA(0,$#,&dptr[z0][d]); + COPY_VEC(8+$$,0+$$); + for (z = z0 - 1; z >= start; z--) { + MASK(16+$$,8+$$); + AND(16+$$,16+$$,25); + SHLBYTE(8+$$,8+$$); + XOR(8+$$,8+$$,16+$$); + LOAD_DATA(16,$#,&dptr[z][d]); + XOR(0+$$,0+$$,16+$$); + XOR(8+$$,8+$$,16+$$); + } + /* P/Q left side optimization */ + for (z = start - 1; z >= 0; z--) { + MASK(16+$$,8+$$); + AND(16+$$,16+$$,25); + SHLBYTE(8+$$,8+$$); + XOR(8+$$,8+$$,16+$$); + } + LOAD_DATA(16,$#,&p[d]); + XOR(16+$$,16+$$,0+$$); + STORE_DATA(16,$#,&p[d]); + LOAD_DATA(16,$#,&q[d]); + XOR(16+$$,16+$$,8+$$); + STORE_DATA(16,$#,&q[d]); + } + kernel_fpu_end(&vxstate, KERNEL_VXR); +} + +static int raid6_s390vx$#_valid(void) +{ + return MACHINE_HAS_VX; +} + +const struct raid6_calls raid6_s390vx$# = { + raid6_s390vx$#_gen_syndrome, + raid6_s390vx$#_xor_syndrome, + raid6_s390vx$#_valid, + "vx128x$#", + 1 +}; -- cgit v1.2.3