diff options
Diffstat (limited to 'libc/string/x86_64/memset.S')
| -rw-r--r-- | libc/string/x86_64/memset.S | 20 |
1 files changed, 13 insertions, 7 deletions
diff --git a/libc/string/x86_64/memset.S b/libc/string/x86_64/memset.S index 3092e81eb..46751006b 100644 --- a/libc/string/x86_64/memset.S +++ b/libc/string/x86_64/memset.S @@ -55,8 +55,10 @@ ENTRY (memset) test $0x7,%edi /* Check for alignment. */ jz 2f - .p2align 4 -1: /* Align ptr to 8 byte. */ + /* Next 3 insns are 9 bytes total, make sure we decode them in one go */ + .p2align 4,,9 +1: + /* Align ptr to 8 byte. */ mov %sil,(%rcx) dec %rdx inc %rcx @@ -70,8 +72,10 @@ ENTRY (memset) cmp LARGE, %rdx jae 11f - .p2align 4 -3: /* Fill 64 bytes. */ + /* Next 3 insns are 11 bytes total, make sure we decode them in one go */ + .p2align 4,,11 +3: + /* Fill 64 bytes. */ mov %r8,(%rcx) mov %r8,0x8(%rcx) mov %r8,0x10(%rcx) @@ -114,9 +118,11 @@ ENTRY (memset) #endif retq - .p2align 4 -11: /* Fill 64 bytes without polluting the cache. */ - /* We could use movntdq %xmm0,(%rcx) here to further + /* Next 3 insns are 14 bytes total, make sure we decode them in one go */ + .p2align 4,,14 +11: + /* Fill 64 bytes without polluting the cache. */ + /* We could use movntdq %xmm0,(%rcx) here to further speed up for large cases but let's not use XMM registers. */ movnti %r8,(%rcx) movnti %r8,0x8(%rcx) |