summaryrefslogtreecommitdiffstats
path: root/arch/x86/lib/memcpy_32.c
diff options
context:
space:
mode:
authorMa Ling <ling.ma@intel.com>2010-09-17 03:12:40 +0800
committerH. Peter Anvin <hpa@zytor.com>2010-09-24 18:57:11 -0700
commit3b4b682becdfa9f42321aa024d5cc84f71f06d8c (patch)
tree807cac0bfbc45d45f156dae742b31f9a66aaa559 /arch/x86/lib/memcpy_32.c
parent59daa706fbec745684702741b9f5373142dd9fdc (diff)
downloadlinux-3b4b682becdfa9f42321aa024d5cc84f71f06d8c.tar.gz
linux-3b4b682becdfa9f42321aa024d5cc84f71f06d8c.tar.bz2
linux-3b4b682becdfa9f42321aa024d5cc84f71f06d8c.zip
x86, mem: Optimize memmove for small size and unaligned cases
movs instruction will combine data to accelerate moving data, however we need to concern two cases about it. 1. movs instruction need long lantency to startup, so here we use general mov instruction to copy data. 2. movs instruction is not good for unaligned case, even if src offset is 0x10, dest offset is 0x0, we avoid and handle the case by general mov instruction. Signed-off-by: Ma Ling <ling.ma@intel.com> LKML-Reference: <1284664360-6138-1-git-send-email-ling.ma@intel.com> Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Diffstat (limited to 'arch/x86/lib/memcpy_32.c')
-rw-r--r--arch/x86/lib/memcpy_32.c213
1 files changed, 182 insertions, 31 deletions
diff --git a/arch/x86/lib/memcpy_32.c b/arch/x86/lib/memcpy_32.c
index 81130d477ee2..b908a59eccf5 100644
--- a/arch/x86/lib/memcpy_32.c
+++ b/arch/x86/lib/memcpy_32.c
@@ -22,36 +22,187 @@ EXPORT_SYMBOL(memset);
void *memmove(void *dest, const void *src, size_t n)
{
- int d0, d1, d2;
-
- if (dest < src) {
- if ((dest + n) < src)
- return memcpy(dest, src, n);
- else
- __asm__ __volatile__(
- "rep\n\t"
- "movsb\n\t"
- : "=&c" (d0), "=&S" (d1), "=&D" (d2)
- :"0" (n),
- "1" (src),
- "2" (dest)
- :"memory");
- } else {
- if((src + n) < dest)
- return memcpy(dest, src, n);
- else
- __asm__ __volatile__(
- "std\n\t"
- "rep\n\t"
- "movsb\n\t"
- "cld"
- : "=&c" (d0), "=&S" (d1), "=&D" (d2)
- :"0" (n),
- "1" (n-1+src),
- "2" (n-1+dest)
- :"memory");
- }
-
- return dest;
+ int d0,d1,d2,d3,d4,d5;
+ char *ret = dest;
+
+ __asm__ __volatile__(
+ /* Handle more 16bytes in loop */
+ "cmp $0x10, %0\n\t"
+ "jb 1f\n\t"
+
+ /* Decide forward/backward copy mode */
+ "cmp %2, %1\n\t"
+ "jb 2f\n\t"
+
+ /*
+ * movs instruction have many startup latency
+ * so we handle small size by general register.
+ */
+ "cmp $680, %0\n\t"
+ "jb 3f\n\t"
+ /*
+ * movs instruction is only good for aligned case.
+ */
+ "mov %1, %3\n\t"
+ "xor %2, %3\n\t"
+ "and $0xff, %3\n\t"
+ "jz 4f\n\t"
+ "3:\n\t"
+ "sub $0x10, %0\n\t"
+
+ /*
+ * We gobble 16byts forward in each loop.
+ */
+ "3:\n\t"
+ "sub $0x10, %0\n\t"
+ "mov 0*4(%1), %3\n\t"
+ "mov 1*4(%1), %4\n\t"
+ "mov %3, 0*4(%2)\n\t"
+ "mov %4, 1*4(%2)\n\t"
+ "mov 2*4(%1), %3\n\t"
+ "mov 3*4(%1), %4\n\t"
+ "mov %3, 2*4(%2)\n\t"
+ "mov %4, 3*4(%2)\n\t"
+ "lea 0x10(%1), %1\n\t"
+ "lea 0x10(%2), %2\n\t"
+ "jae 3b\n\t"
+ "add $0x10, %0\n\t"
+ "jmp 1f\n\t"
+
+ /*
+ * Handle data forward by movs.
+ */
+ ".p2align 4\n\t"
+ "4:\n\t"
+ "mov -4(%1, %0), %3\n\t"
+ "lea -4(%2, %0), %4\n\t"
+ "shr $2, %0\n\t"
+ "rep movsl\n\t"
+ "mov %3, (%4)\n\t"
+ "jmp 11f\n\t"
+ /*
+ * Handle data backward by movs.
+ */
+ ".p2align 4\n\t"
+ "6:\n\t"
+ "mov (%1), %3\n\t"
+ "mov %2, %4\n\t"
+ "lea -4(%1, %0), %1\n\t"
+ "lea -4(%2, %0), %2\n\t"
+ "shr $2, %0\n\t"
+ "std\n\t"
+ "rep movsl\n\t"
+ "mov %3,(%4)\n\t"
+ "cld\n\t"
+ "jmp 11f\n\t"
+
+ /*
+ * Start to prepare for backward copy.
+ */
+ ".p2align 4\n\t"
+ "2:\n\t"
+ "cmp $680, %0\n\t"
+ "jb 5f\n\t"
+ "mov %1, %3\n\t"
+ "xor %2, %3\n\t"
+ "and $0xff, %3\n\t"
+ "jz 6b\n\t"
+
+ /*
+ * Calculate copy position to tail.
+ */
+ "5:\n\t"
+ "add %0, %1\n\t"
+ "add %0, %2\n\t"
+ "sub $0x10, %0\n\t"
+
+ /*
+ * We gobble 16byts backward in each loop.
+ */
+ "7:\n\t"
+ "sub $0x10, %0\n\t"
+
+ "mov -1*4(%1), %3\n\t"
+ "mov -2*4(%1), %4\n\t"
+ "mov %3, -1*4(%2)\n\t"
+ "mov %4, -2*4(%2)\n\t"
+ "mov -3*4(%1), %3\n\t"
+ "mov -4*4(%1), %4\n\t"
+ "mov %3, -3*4(%2)\n\t"
+ "mov %4, -4*4(%2)\n\t"
+ "lea -0x10(%1), %1\n\t"
+ "lea -0x10(%2), %2\n\t"
+ "jae 7b\n\t"
+ /*
+ * Calculate copy position to head.
+ */
+ "add $0x10, %0\n\t"
+ "sub %0, %1\n\t"
+ "sub %0, %2\n\t"
+
+ /*
+ * Move data from 8 bytes to 15 bytes.
+ */
+ ".p2align 4\n\t"
+ "1:\n\t"
+ "cmp $8, %0\n\t"
+ "jb 8f\n\t"
+ "mov 0*4(%1), %3\n\t"
+ "mov 1*4(%1), %4\n\t"
+ "mov -2*4(%1, %0), %5\n\t"
+ "mov -1*4(%1, %0), %1\n\t"
+
+ "mov %3, 0*4(%2)\n\t"
+ "mov %4, 1*4(%2)\n\t"
+ "mov %5, -2*4(%2, %0)\n\t"
+ "mov %1, -1*4(%2, %0)\n\t"
+ "jmp 11f\n\t"
+
+ /*
+ * Move data from 4 bytes to 7 bytes.
+ */
+ ".p2align 4\n\t"
+ "8:\n\t"
+ "cmp $4, %0\n\t"
+ "jb 9f\n\t"
+ "mov 0*4(%1), %3\n\t"
+ "mov -1*4(%1, %0), %4\n\t"
+ "mov %3, 0*4(%2)\n\t"
+ "mov %4, -1*4(%2, %0)\n\t"
+ "jmp 11f\n\t"
+
+ /*
+ * Move data from 2 bytes to 3 bytes.
+ */
+ ".p2align 4\n\t"
+ "9:\n\t"
+ "cmp $2, %0\n\t"
+ "jb 10f\n\t"
+ "movw 0*2(%1), %%dx\n\t"
+ "movw -1*2(%1, %0), %%bx\n\t"
+ "movw %%dx, 0*2(%2)\n\t"
+ "movw %%bx, -1*2(%2, %0)\n\t"
+ "jmp 11f\n\t"
+
+ /*
+ * Move data for 1 byte.
+ */
+ ".p2align 4\n\t"
+ "10:\n\t"
+ "cmp $1, %0\n\t"
+ "jb 11f\n\t"
+ "movb (%1), %%cl\n\t"
+ "movb %%cl, (%2)\n\t"
+ ".p2align 4\n\t"
+ "11:"
+ : "=&c" (d0), "=&S" (d1), "=&D" (d2),
+ "=r" (d3),"=r" (d4), "=r"(d5)
+ :"0" (n),
+ "1" (src),
+ "2" (dest)
+ :"memory");
+
+ return ret;
+
}
EXPORT_SYMBOL(memmove);