OpenE2K/glibc
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Unroll the loop x86-64 SSE4.2 strlen.

Browse Source
This commit is contained in:
H.J. Lu 2010-01-13 07:51:48 -08:00 committed by Ulrich Drepper
parent 52e96a8092
commit 5a7af22fbb
2 changed files with 48 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ChangeLog
View File

@ -1,3 +1,7 @@
2010-01-12 H.J. Lu <hongjiu.lu@intel.com>
* sysdeps/x86_64/multiarch/strlen.S: Unroll the loop.
2010-01-13 Ulrich Drepper <drepper@redhat.com>
* stdlib/stdlib.h: Be a bit more relaxed about obsoleted mktemp symbol.

58
sysdeps/x86_64/multiarch/strlen.S
View File

@ -46,28 +46,58 @@ END(strlen)
__strlen_sse42:
cfi_startproc
CALL_MCOUNT
pxor %xmm2, %xmm2
movq %rdi, %rcx
pxor %xmm1, %xmm1
movl %edi, %ecx
movq %rdi, %r8
andq $~15, %rdi
movdqa %xmm2, %xmm1
pcmpeqb (%rdi), %xmm2
orl $0xffffffff, %esi
subq %rdi, %rcx
shll %cl, %esi
pmovmskb %xmm2, %edx
andl %esi, %edx
jnz 1f
xor %edi, %ecx
pcmpeqb (%rdi), %xmm1
pmovmskb %xmm1, %edx
shrl %cl, %edx
shll %cl, %edx
andl %edx, %edx
jnz L(less16bytes)
pxor %xmm1, %xmm1
2: pcmpistri $0x08, 16(%rdi), %xmm1
leaq 16(%rdi), %rdi
jnz 2b
.p2align 4
L(more64bytes_loop):
pcmpistri $0x08, 16(%rdi), %xmm1
jz L(more32bytes)
pcmpistri $0x08, 32(%rdi), %xmm1
jz L(more48bytes)
pcmpistri $0x08, 48(%rdi), %xmm1
jz L(more64bytes)
add $64, %rdi
pcmpistri $0x08, (%rdi), %xmm1
jnz L(more64bytes_loop)
leaq (%rdi,%rcx), %rax
subq %r8, %rax
ret
1: subq %r8, %rdi
.p2align 4
L(more32bytes):
leaq 16(%rdi,%rcx, 1), %rax
subq %r8, %rax
ret
.p2align 4
L(more48bytes):
leaq 32(%rdi,%rcx, 1), %rax
subq %r8, %rax
ret
.p2align 4
L(more64bytes):
leaq 48(%rdi,%rcx, 1), %rax
subq %r8, %rax
ret
.p2align 4
L(less16bytes):
subq %r8, %rdi
bsfl %edx, %eax
addq %rdi, %rax
ret