[PATCH v4] faster strlen on x64
Ondřej Bílka
neleai@seznam.cz
Wed Feb 13 11:43:00 GMT 2013
Hello,
I wrote at previous version that unaligned read of first 16 bytes is bad
tradeoff. When I made faster strcpy header I realized that it was because
I was doing separate check if it crosses page.
When I do only check if next 64 bytes do not cross page and first do
unaligned 16 byte load then it causes only small overhead for larger
strings. This makes my implementation faster for wider family of
workloads. It speed up gcc benchmark and most other programs.
On unit tests revised version is somewhat slower than previous version.
It is caused by choosing first 16 bytes only rarely which causes branch
misprediction.
I did two additional small improvements, first is squashing padding patch.
Second bit is test to cross page can be done as x%4096 < 4096-48 instead
x%4096 <= 4096-64 because I align x into 16 bytes.
I updated benchmarks, difference between new and revised version is at
http://kam.mff.cuni.cz/~ondra/benchmark_string/strlen_profile.html
http://kam.mff.cuni.cz/~ondra/strlen_profile.tar.bz2
Ondra
2013-01-31 Ondrej Bilka <neleai@seznam.cz>
* sysdeps/x86_64/strlen.S: Replace with new SSE2 based
implementation which is faster on all x86_64 architectures.
Tested on AMD, Intel Nehalem, SNB, IVB.
* sysdeps/x86_64/strnlen.S: Likewise.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines):
Remove all multiarch strlen and strnlen versions.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c: Update.
Remove strlen and strnlen related parts.
* sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S: Update.
Inline strlen part.
* sysdeps/x86_64/multiarch/strcat-ssse3.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.S: Remove.
* sysdeps/x86_64/multiarch/strlen-sse2-no-bsf.S: Remove.
* sysdeps/x86_64/multiarch/strlen-sse2-pminub.S: Remove.
* sysdeps/x86_64/multiarch/rtld-strlen.S: Remove.
* sysdeps/x86_64/multiarch/strlen-sse4.S: Remove.
* sysdeps/x86_64/multiarch/strnlen.S: Remove.
* sysdeps/x86_64/multiarch/strnlen-sse2-no-bsf.S: Remove.
-------------- next part --------------
>From e7b469b93cf4cfd6475b644b0f2d72b8ae47170f Mon Sep 17 00:00:00 2001
From: Ondrej Bilka <neleai@seznam.cz>
Date: Wed, 30 Jan 2013 18:13:22 +0100
Subject: [PATCH] Faster strlen on x86-64.
---
sysdeps/x86_64/multiarch/Makefile | 6 +-
sysdeps/x86_64/multiarch/ifunc-impl-list.c | 13 -
sysdeps/x86_64/multiarch/rtld-strlen.S | 1 -
sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S | 229 +++++++-
sysdeps/x86_64/multiarch/strcat-ssse3.S | 312 ++++++++++-
sysdeps/x86_64/multiarch/strlen-sse2-no-bsf.S | 685 ----------------------
sysdeps/x86_64/multiarch/strlen-sse2-pminub.S | 259 --------
sysdeps/x86_64/multiarch/strlen-sse4.S | 84 ---
sysdeps/x86_64/multiarch/strlen.S | 68 ---
sysdeps/x86_64/multiarch/strnlen-sse2-no-bsf.S | 3 -
sysdeps/x86_64/multiarch/strnlen.S | 57 --
sysdeps/x86_64/strlen.S | 265 +++++++--
sysdeps/x86_64/strnlen.S | 66 +--
13 files changed, 742 insertions(+), 1306 deletions(-)
delete mode 100644 sysdeps/x86_64/multiarch/rtld-strlen.S
delete mode 100644 sysdeps/x86_64/multiarch/strlen-sse2-no-bsf.S
delete mode 100644 sysdeps/x86_64/multiarch/strlen-sse2-pminub.S
delete mode 100644 sysdeps/x86_64/multiarch/strlen-sse4.S
delete mode 100644 sysdeps/x86_64/multiarch/strlen.S
delete mode 100644 sysdeps/x86_64/multiarch/strnlen-sse2-no-bsf.S
delete mode 100644 sysdeps/x86_64/multiarch/strnlen.S
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index dd6c27d..67686ad 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -10,14 +10,12 @@ sysdep_routines += strncat-c stpncpy-c strncpy-c strcmp-ssse3 strncmp-ssse3 \
strend-sse4 memcmp-sse4 memcpy-ssse3 mempcpy-ssse3 \
memmove-ssse3 memcpy-ssse3-back mempcpy-ssse3-back \
memmove-ssse3-back strcasestr-nonascii strcasecmp_l-ssse3 \
- strncase_l-ssse3 strlen-sse4 strlen-sse2-no-bsf memset-x86-64 \
+ strncase_l-ssse3 memset-x86-64 strcat-ssse3 strncat-ssse3\
strcpy-ssse3 strncpy-ssse3 stpcpy-ssse3 stpncpy-ssse3 \
strcpy-sse2-unaligned strncpy-sse2-unaligned \
stpcpy-sse2-unaligned stpncpy-sse2-unaligned \
strcat-sse2-unaligned strncat-sse2-unaligned \
- strcat-ssse3 strncat-ssse3 strlen-sse2-pminub \
- strnlen-sse2-no-bsf strrchr-sse2-no-bsf strchr-sse2-no-bsf \
- memcmp-ssse3
+ strrchr-sse2-no-bsf strchr-sse2-no-bsf memcmp-ssse3
ifeq (yes,$(config-cflags-sse4))
sysdep_routines += strcspn-c strpbrk-c strspn-c strstr-c strcasestr-c varshift
CFLAGS-varshift.c += -msse4
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index 643cb2d..848991e 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -187,11 +187,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__strncpy_sse2_unaligned)
IFUNC_IMPL_ADD (array, i, strncpy, 1, __strncpy_sse2))
- /* Support sysdeps/x86_64/multiarch/strnlen.S. */
- IFUNC_IMPL (i, name, strnlen,
- IFUNC_IMPL_ADD (array, i, strnlen, 1, __strnlen_sse2_no_bsf)
- IFUNC_IMPL_ADD (array, i, strnlen, 1, __strnlen_sse2))
-
/* Support sysdeps/x86_64/multiarch/strpbrk.S. */
IFUNC_IMPL (i, name, strpbrk,
IFUNC_IMPL_ADD (array, i, strpbrk, HAS_SSE4_2,
@@ -262,14 +257,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__mempcpy_ssse3)
IFUNC_IMPL_ADD (array, i, mempcpy, 1, __mempcpy_sse2))
- /* Support sysdeps/x86_64/multiarch/strlen.S. */
- IFUNC_IMPL (i, name, strlen,
- IFUNC_IMPL_ADD (array, i, strlen, HAS_SSE4_2, __strlen_sse42)
- IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2_pminub)
- IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2_no_bsf)
- IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2)
- IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2))
-
/* Support sysdeps/x86_64/multiarch/strncmp.S. */
IFUNC_IMPL (i, name, strncmp,
IFUNC_IMPL_ADD (array, i, strncmp, HAS_SSE4_2,
diff --git a/sysdeps/x86_64/multiarch/rtld-strlen.S b/sysdeps/x86_64/multiarch/rtld-strlen.S
deleted file mode 100644
index 596e054..0000000
--- a/sysdeps/x86_64/multiarch/rtld-strlen.S
+++ /dev/null
@@ -1 +0,0 @@
-#include "../rtld-strlen.S"
diff --git a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
index 72bb609..6d9951e 100644
--- a/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
+++ b/sysdeps/x86_64/multiarch/strcat-sse2-unaligned.S
@@ -34,10 +34,233 @@ ENTRY (STRCAT)
mov %rdx, %r8
# endif
-# define RETURN jmp L(StartStrcpyPart)
-# include "strlen-sse2-pminub.S"
-# undef RETURN
+ xor %rax, %rax
+ mov %edi, %ecx
+ and $0x3f, %ecx
+ pxor %xmm0, %xmm0
+ cmp $0x30, %ecx
+ ja L(next)
+ movdqu (%rdi), %xmm1
+ pcmpeqb %xmm1, %xmm0
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ jnz L(exit_less16)
+ mov %rdi, %rax
+ and $-16, %rax
+ jmp L(align16_start)
+L(next):
+ mov %rdi, %rax
+ and $-16, %rax
+ pcmpeqb (%rax), %xmm0
+ mov $-1, %r10d
+ sub %rax, %rcx
+ shl %cl, %r10d
+ pmovmskb %xmm0, %edx
+ and %r10d, %edx
+ jnz L(exit)
+L(align16_start):
+ pxor %xmm0, %xmm0
+ pxor %xmm1, %xmm1
+ pxor %xmm2, %xmm2
+ pxor %xmm3, %xmm3
+ pcmpeqb 16(%rax), %xmm0
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ jnz L(exit16)
+
+ pcmpeqb 32(%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ jnz L(exit32)
+
+ pcmpeqb 48(%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ jnz L(exit48)
+
+ pcmpeqb 64(%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ jnz L(exit64)
+
+ pcmpeqb 80(%rax), %xmm0
+ add $64, %rax
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ jnz L(exit16)
+
+ pcmpeqb 32(%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ jnz L(exit32)
+
+ pcmpeqb 48(%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ jnz L(exit48)
+
+ pcmpeqb 64(%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ jnz L(exit64)
+
+ pcmpeqb 80(%rax), %xmm0
+ add $64, %rax
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ jnz L(exit16)
+
+ pcmpeqb 32(%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ jnz L(exit32)
+
+ pcmpeqb 48(%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ jnz L(exit48)
+
+ pcmpeqb 64(%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ jnz L(exit64)
+
+ pcmpeqb 80(%rax), %xmm0
+ add $64, %rax
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ jnz L(exit16)
+
+ pcmpeqb 32(%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ jnz L(exit32)
+
+ pcmpeqb 48(%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ jnz L(exit48)
+
+ pcmpeqb 64(%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ jnz L(exit64)
+
+ test $0x3f, %rax
+ jz L(align64_loop)
+
+ pcmpeqb 80(%rax), %xmm0
+ add $80, %rax
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ jnz L(exit)
+
+ test $0x3f, %rax
+ jz L(align64_loop)
+
+ pcmpeqb 16(%rax), %xmm1
+ add $16, %rax
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ jnz L(exit)
+
+ test $0x3f, %rax
+ jz L(align64_loop)
+
+ pcmpeqb 16(%rax), %xmm2
+ add $16, %rax
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ jnz L(exit)
+
+ test $0x3f, %rax
+ jz L(align64_loop)
+
+ pcmpeqb 16(%rax), %xmm3
+ add $16, %rax
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ jnz L(exit)
+
+ add $16, %rax
+ .p2align 4
+ L(align64_loop):
+ movaps (%rax), %xmm4
+ pminub 16(%rax), %xmm4
+ movaps 32(%rax), %xmm5
+ pminub 48(%rax), %xmm5
+ add $64, %rax
+ pminub %xmm4, %xmm5
+ pcmpeqb %xmm0, %xmm5
+ pmovmskb %xmm5, %edx
+ test %edx, %edx
+ jz L(align64_loop)
+
+ pcmpeqb -64(%rax), %xmm0
+ sub $80, %rax
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ jnz L(exit16)
+
+ pcmpeqb 32(%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ jnz L(exit32)
+
+ pcmpeqb 48(%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ jnz L(exit48)
+
+ pcmpeqb 64(%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ sub %rdi, %rax
+ bsf %rdx, %rdx
+ add %rdx, %rax
+ add $64, %rax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit):
+ sub %rdi, %rax
+L(exit_less16):
+ bsf %rdx, %rdx
+ add %rdx, %rax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit16):
+ sub %rdi, %rax
+ bsf %rdx, %rdx
+ add %rdx, %rax
+ add $16, %rax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit32):
+ sub %rdi, %rax
+ bsf %rdx, %rdx
+ add %rdx, %rax
+ add $32, %rax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit48):
+ sub %rdi, %rax
+ bsf %rdx, %rdx
+ add %rdx, %rax
+ add $48, %rax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit64):
+ sub %rdi, %rax
+ bsf %rdx, %rdx
+ add %rdx, %rax
+ add $64, %rax
+
+ .p2align 4
L(StartStrcpyPart):
lea (%r9, %rax), %rdi
mov %rsi, %rcx
diff --git a/sysdeps/x86_64/multiarch/strcat-ssse3.S b/sysdeps/x86_64/multiarch/strcat-ssse3.S
index fea9d11..901e66f 100644
--- a/sysdeps/x86_64/multiarch/strcat-ssse3.S
+++ b/sysdeps/x86_64/multiarch/strcat-ssse3.S
@@ -33,11 +33,317 @@ ENTRY (STRCAT)
mov %rdx, %r8
# endif
-# define RETURN jmp L(StartStrcpyPart)
-# include "strlen-sse2-no-bsf.S"
+ xor %eax, %eax
+ cmpb $0, (%rdi)
+ jz L(exit_tail0)
+ cmpb $0, 1(%rdi)
+ jz L(exit_tail1)
+ cmpb $0, 2(%rdi)
+ jz L(exit_tail2)
+ cmpb $0, 3(%rdi)
+ jz L(exit_tail3)
+
+ cmpb $0, 4(%rdi)
+ jz L(exit_tail4)
+ cmpb $0, 5(%rdi)
+ jz L(exit_tail5)
+ cmpb $0, 6(%rdi)
+ jz L(exit_tail6)
+ cmpb $0, 7(%rdi)
+ jz L(exit_tail7)
+
+ cmpb $0, 8(%rdi)
+ jz L(exit_tail8)
+ cmpb $0, 9(%rdi)
+ jz L(exit_tail9)
+ cmpb $0, 10(%rdi)
+ jz L(exit_tail10)
+ cmpb $0, 11(%rdi)
+ jz L(exit_tail11)
+
+ cmpb $0, 12(%rdi)
+ jz L(exit_tail12)
+ cmpb $0, 13(%rdi)
+ jz L(exit_tail13)
+ cmpb $0, 14(%rdi)
+ jz L(exit_tail14)
+ cmpb $0, 15(%rdi)
+ jz L(exit_tail15)
+ pxor %xmm0, %xmm0
+ lea 16(%rdi), %rcx
+ lea 16(%rdi), %rax
+ and $-16, %rax
+
+ pcmpeqb (%rax), %xmm0
+ pmovmskb %xmm0, %edx
+ pxor %xmm1, %xmm1
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ pxor %xmm2, %xmm2
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ pxor %xmm3, %xmm3
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm0
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm0
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm0
+ pmovmskb %xmm0, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm1
+ pmovmskb %xmm1, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm2
+ pmovmskb %xmm2, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ pcmpeqb (%rax), %xmm3
+ pmovmskb %xmm3, %edx
+ test %edx, %edx
+ lea 16(%rax), %rax
+ jnz L(exit)
+
+ and $-0x40, %rax
-# undef RETURN
+ .p2align 4
+L(aligned_64):
+ pcmpeqb (%rax), %xmm0
+ pcmpeqb 16(%rax), %xmm1
+ pcmpeqb 32(%rax), %xmm2
+ pcmpeqb 48(%rax), %xmm3
+ pmovmskb %xmm0, %edx
+ pmovmskb %xmm1, %r11d
+ pmovmskb %xmm2, %r10d
+ pmovmskb %xmm3, %r9d
+ or %edx, %r9d
+ or %r11d, %r9d
+ or %r10d, %r9d
+ lea 64(%rax), %rax
+ jz L(aligned_64)
+
+ test %edx, %edx
+ jnz L(aligned_64_exit_16)
+ test %r11d, %r11d
+ jnz L(aligned_64_exit_32)
+ test %r10d, %r10d
+ jnz L(aligned_64_exit_48)
+
+L(aligned_64_exit_64):
+ pmovmskb %xmm3, %edx
+ jmp L(exit)
+
+L(aligned_64_exit_48):
+ lea -16(%rax), %rax
+ mov %r10d, %edx
+ jmp L(exit)
+
+L(aligned_64_exit_32):
+ lea -32(%rax), %rax
+ mov %r11d, %edx
+ jmp L(exit)
+
+L(aligned_64_exit_16):
+ lea -48(%rax), %rax
+
+L(exit):
+ sub %rcx, %rax
+ test %dl, %dl
+ jz L(exit_high)
+ test $0x01, %dl
+ jnz L(exit_tail0)
+
+ test $0x02, %dl
+ jnz L(exit_tail1)
+
+ test $0x04, %dl
+ jnz L(exit_tail2)
+
+ test $0x08, %dl
+ jnz L(exit_tail3)
+
+ test $0x10, %dl
+ jnz L(exit_tail4)
+
+ test $0x20, %dl
+ jnz L(exit_tail5)
+
+ test $0x40, %dl
+ jnz L(exit_tail6)
+ add $7, %eax
+L(exit_tail0):
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_high):
+ add $8, %eax
+ test $0x01, %dh
+ jnz L(exit_tail0)
+
+ test $0x02, %dh
+ jnz L(exit_tail1)
+
+ test $0x04, %dh
+ jnz L(exit_tail2)
+
+ test $0x08, %dh
+ jnz L(exit_tail3)
+
+ test $0x10, %dh
+ jnz L(exit_tail4)
+
+ test $0x20, %dh
+ jnz L(exit_tail5)
+
+ test $0x40, %dh
+ jnz L(exit_tail6)
+ add $7, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail1):
+ add $1, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail2):
+ add $2, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail3):
+ add $3, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail4):
+ add $4, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail5):
+ add $5, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail6):
+ add $6, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail7):
+ add $7, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail8):
+ add $8, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail9):
+ add $9, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail10):
+ add $10, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail11):
+ add $11, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail12):
+ add $12, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail13):
+ add $13, %eax
+ jmp L(StartStrcpyPart)
+ .p2align 4
+L(exit_tail14):
+ add $14, %eax
+ jmp L(StartStrcpyPart)
+
+ .p2align 4
+L(exit_tail15):
+ add $15, %eax
+
+ .p2align 4
L(StartStrcpyPart):
mov %rsi, %rcx
lea (%rdi, %rax), %rdx
diff --git a/sysdeps/x86_64/multiarch/strlen-sse2-no-bsf.S b/sysdeps/x86_64/multiarch/strlen-sse2-no-bsf.S
deleted file mode 100644
index ff2ab70..0000000
--- a/sysdeps/x86_64/multiarch/strlen-sse2-no-bsf.S
+++ /dev/null
@@ -1,685 +0,0 @@
-/* strlen SSE2 without bsf
- Copyright (C) 2010-2013 Free Software Foundation, Inc.
- Contributed by Intel Corporation.
- This file is part of the GNU C Library.
-
- The GNU C Library is free software; you can redistribute it and/or
- modify it under the terms of the GNU Lesser General Public
- License as published by the Free Software Foundation; either
- version 2.1 of the License, or (at your option) any later version.
-
- The GNU C Library 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
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public
- License along with the GNU C Library; if not, see
- <http://www.gnu.org/licenses/>. */
-
-/* only for strlen case we don't use optimized version for STATIC build just for SHARED */
-
-#if (defined SHARED || defined USE_AS_STRCAT || defined USE_AS_STRNLEN) && !defined NOT_IN_libc
-
-# ifndef USE_AS_STRCAT
-
-# include <sysdep.h>
-
-# define RETURN ret
-
-# ifndef STRLEN
-# define STRLEN __strlen_sse2_no_bsf
-# endif
-
- atom_text_section
-ENTRY (STRLEN)
-# endif
- xor %eax, %eax
-# ifdef USE_AS_STRNLEN
- mov %rsi, %r8
- sub $4, %rsi
- jbe L(len_less4_prolog)
-# endif
- cmpb $0, (%rdi)
- jz L(exit_tail0)
- cmpb $0, 1(%rdi)
- jz L(exit_tail1)
- cmpb $0, 2(%rdi)
- jz L(exit_tail2)
- cmpb $0, 3(%rdi)
- jz L(exit_tail3)
-
-# ifdef USE_AS_STRNLEN
- sub $4, %rsi
- jbe L(len_less8_prolog)
-# endif
-
- cmpb $0, 4(%rdi)
- jz L(exit_tail4)
- cmpb $0, 5(%rdi)
- jz L(exit_tail5)
- cmpb $0, 6(%rdi)
- jz L(exit_tail6)
- cmpb $0, 7(%rdi)
- jz L(exit_tail7)
-
-# ifdef USE_AS_STRNLEN
- sub $4, %rsi
- jbe L(len_less12_prolog)
-# endif
-
- cmpb $0, 8(%rdi)
- jz L(exit_tail8)
- cmpb $0, 9(%rdi)
- jz L(exit_tail9)
- cmpb $0, 10(%rdi)
- jz L(exit_tail10)
- cmpb $0, 11(%rdi)
- jz L(exit_tail11)
-
-# ifdef USE_AS_STRNLEN
- sub $4, %rsi
- jbe L(len_less16_prolog)
-# endif
-
- cmpb $0, 12(%rdi)
- jz L(exit_tail12)
- cmpb $0, 13(%rdi)
- jz L(exit_tail13)
- cmpb $0, 14(%rdi)
- jz L(exit_tail14)
- cmpb $0, 15(%rdi)
- jz L(exit_tail15)
- pxor %xmm0, %xmm0
- lea 16(%rdi), %rcx
- lea 16(%rdi), %rax
- and $-16, %rax
-
-# ifdef USE_AS_STRNLEN
- and $15, %rdi
- add %rdi, %rsi
- sub $64, %rsi
- jbe L(len_less64)
-# endif
-
- pcmpeqb (%rax), %xmm0
- pmovmskb %xmm0, %edx
- pxor %xmm1, %xmm1
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm1
- pmovmskb %xmm1, %edx
- pxor %xmm2, %xmm2
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm2
- pmovmskb %xmm2, %edx
- pxor %xmm3, %xmm3
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
-# ifdef USE_AS_STRNLEN
- sub $64, %rsi
- jbe L(len_less64)
-# endif
-
- pcmpeqb (%rax), %xmm0
- pmovmskb %xmm0, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
-# ifdef USE_AS_STRNLEN
- sub $64, %rsi
- jbe L(len_less64)
-# endif
-
- pcmpeqb (%rax), %xmm0
- pmovmskb %xmm0, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
-# ifdef USE_AS_STRNLEN
- sub $64, %rsi
- jbe L(len_less64)
-# endif
-
- pcmpeqb (%rax), %xmm0
- pmovmskb %xmm0, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
- pcmpeqb (%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- lea 16(%rax), %rax
- jnz L(exit)
-
-# ifdef USE_AS_STRNLEN
- mov %rax, %rdx
- and $63, %rdx
- add %rdx, %rsi
-# endif
-
- and $-0x40, %rax
-
- .p2align 4
-L(aligned_64):
-# ifdef USE_AS_STRNLEN
- sub $64, %rsi
- jbe L(len_less64)
-# endif
- pcmpeqb (%rax), %xmm0
- pcmpeqb 16(%rax), %xmm1
- pcmpeqb 32(%rax), %xmm2
- pcmpeqb 48(%rax), %xmm3
- pmovmskb %xmm0, %edx
- pmovmskb %xmm1, %r11d
- pmovmskb %xmm2, %r10d
- pmovmskb %xmm3, %r9d
- or %edx, %r9d
- or %r11d, %r9d
- or %r10d, %r9d
- lea 64(%rax), %rax
- jz L(aligned_64)
-
- test %edx, %edx
- jnz L(aligned_64_exit_16)
- test %r11d, %r11d
- jnz L(aligned_64_exit_32)
- test %r10d, %r10d
- jnz L(aligned_64_exit_48)
-L(aligned_64_exit_64):
- pmovmskb %xmm3, %edx
- jmp L(aligned_64_exit)
-L(aligned_64_exit_48):
- lea -16(%rax), %rax
- mov %r10d, %edx
- jmp L(aligned_64_exit)
-L(aligned_64_exit_32):
- lea -32(%rax), %rax
- mov %r11d, %edx
- jmp L(aligned_64_exit)
-L(aligned_64_exit_16):
- lea -48(%rax), %rax
-L(aligned_64_exit):
-L(exit):
- sub %rcx, %rax
- test %dl, %dl
- jz L(exit_high)
- test $0x01, %dl
- jnz L(exit_tail0)
-
- test $0x02, %dl
- jnz L(exit_tail1)
-
- test $0x04, %dl
- jnz L(exit_tail2)
-
- test $0x08, %dl
- jnz L(exit_tail3)
-
- test $0x10, %dl
- jnz L(exit_tail4)
-
- test $0x20, %dl
- jnz L(exit_tail5)
-
- test $0x40, %dl
- jnz L(exit_tail6)
- add $7, %eax
-L(exit_tail0):
- RETURN
-
-L(exit_high):
- add $8, %eax
- test $0x01, %dh
- jnz L(exit_tail0)
-
- test $0x02, %dh
- jnz L(exit_tail1)
-
- test $0x04, %dh
- jnz L(exit_tail2)
-
- test $0x08, %dh
- jnz L(exit_tail3)
-
- test $0x10, %dh
- jnz L(exit_tail4)
-
- test $0x20, %dh
- jnz L(exit_tail5)
-
- test $0x40, %dh
- jnz L(exit_tail6)
- add $7, %eax
- RETURN
-
-# ifdef USE_AS_STRNLEN
-
- .p2align 4
-L(len_less64):
- pxor %xmm0, %xmm0
- add $64, %rsi
-
- pcmpeqb (%rax), %xmm0
- pmovmskb %xmm0, %edx
- pxor %xmm1, %xmm1
- lea 16(%rax), %rax
- test %edx, %edx
- jnz L(strnlen_exit)
-
- sub $16, %rsi
- jbe L(return_start_len)
-
- pcmpeqb (%rax), %xmm1
- pmovmskb %xmm1, %edx
- lea 16(%rax), %rax
- test %edx, %edx
- jnz L(strnlen_exit)
-
- sub $16, %rsi
- jbe L(return_start_len)
-
- pcmpeqb (%rax), %xmm0
- pmovmskb %xmm0, %edx
- lea 16(%rax), %rax
- test %edx, %edx
- jnz L(strnlen_exit)
-
- sub $16, %rsi
- jbe L(return_start_len)
-
- pcmpeqb (%rax), %xmm1
- pmovmskb %xmm1, %edx
- lea 16(%rax), %rax
- test %edx, %edx
- jnz L(strnlen_exit)
-
- mov %r8, %rax
- ret
-
- .p2align 4
-L(strnlen_exit):
- sub %rcx, %rax
-
- test %dl, %dl
- jz L(strnlen_exit_high)
- mov %dl, %cl
- and $15, %cl
- jz L(strnlen_exit_8)
- test $0x01, %dl
- jnz L(exit_tail0)
- test $0x02, %dl
- jnz L(strnlen_exit_tail1)
- test $0x04, %dl
- jnz L(strnlen_exit_tail2)
- sub $4, %rsi
- jb L(return_start_len)
- lea 3(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_8):
- test $0x10, %dl
- jnz L(strnlen_exit_tail4)
- test $0x20, %dl
- jnz L(strnlen_exit_tail5)
- test $0x40, %dl
- jnz L(strnlen_exit_tail6)
- sub $8, %rsi
- jb L(return_start_len)
- lea 7(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_high):
- mov %dh, %ch
- and $15, %ch
- jz L(strnlen_exit_high_8)
- test $0x01, %dh
- jnz L(strnlen_exit_tail8)
- test $0x02, %dh
- jnz L(strnlen_exit_tail9)
- test $0x04, %dh
- jnz L(strnlen_exit_tail10)
- sub $12, %rsi
- jb L(return_start_len)
- lea 11(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_high_8):
- test $0x10, %dh
- jnz L(strnlen_exit_tail12)
- test $0x20, %dh
- jnz L(strnlen_exit_tail13)
- test $0x40, %dh
- jnz L(strnlen_exit_tail14)
- sub $16, %rsi
- jb L(return_start_len)
- lea 15(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail1):
- sub $2, %rsi
- jb L(return_start_len)
- lea 1(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail2):
- sub $3, %rsi
- jb L(return_start_len)
- lea 2(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail4):
- sub $5, %rsi
- jb L(return_start_len)
- lea 4(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail5):
- sub $6, %rsi
- jb L(return_start_len)
- lea 5(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail6):
- sub $7, %rsi
- jb L(return_start_len)
- lea 6(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail8):
- sub $9, %rsi
- jb L(return_start_len)
- lea 8(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail9):
- sub $10, %rsi
- jb L(return_start_len)
- lea 9(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail10):
- sub $11, %rsi
- jb L(return_start_len)
- lea 10(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail12):
- sub $13, %rsi
- jb L(return_start_len)
- lea 12(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail13):
- sub $14, %rsi
- jb L(return_start_len)
- lea 13(%eax), %eax
- ret
-
- .p2align 4
-L(strnlen_exit_tail14):
- sub $15, %rsi
- jb L(return_start_len)
- lea 14(%eax), %eax
- ret
-
- .p2align 4
-L(return_start_len):
- mov %r8, %rax
- ret
-
-/* for prolog only */
-
- .p2align 4
-L(len_less4_prolog):
- add $4, %rsi
- jz L(exit_tail0)
-
- cmpb $0, (%rdi)
- jz L(exit_tail0)
- cmp $1, %esi
- je L(exit_tail1)
-
- cmpb $0, 1(%rdi)
- jz L(exit_tail1)
- cmp $2, %esi
- je L(exit_tail2)
-
- cmpb $0, 2(%rdi)
- jz L(exit_tail2)
- cmp $3, %esi
- je L(exit_tail3)
-
- cmpb $0, 3(%rdi)
- jz L(exit_tail3)
- mov $4, %eax
- ret
-
- .p2align 4
-L(len_less8_prolog):
- add $4, %rsi
-
- cmpb $0, 4(%rdi)
- jz L(exit_tail4)
- cmp $1, %esi
- je L(exit_tail5)
-
- cmpb $0, 5(%rdi)
- jz L(exit_tail5)
- cmp $2, %esi
- je L(exit_tail6)
-
- cmpb $0, 6(%rdi)
- jz L(exit_tail6)
- cmp $3, %esi
- je L(exit_tail7)
-
- cmpb $0, 7(%rdi)
- jz L(exit_tail7)
- mov $8, %eax
- ret
-
- .p2align 4
-L(len_less12_prolog):
- add $4, %rsi
-
- cmpb $0, 8(%rdi)
- jz L(exit_tail8)
- cmp $1, %esi
- je L(exit_tail9)
-
- cmpb $0, 9(%rdi)
- jz L(exit_tail9)
- cmp $2, %esi
- je L(exit_tail10)
-
- cmpb $0, 10(%rdi)
- jz L(exit_tail10)
- cmp $3, %esi
- je L(exit_tail11)
-
- cmpb $0, 11(%rdi)
- jz L(exit_tail11)
- mov $12, %eax
- ret
-
- .p2align 4
-L(len_less16_prolog):
- add $4, %rsi
-
- cmpb $0, 12(%rdi)
- jz L(exit_tail12)
- cmp $1, %esi
- je L(exit_tail13)
-
- cmpb $0, 13(%rdi)
- jz L(exit_tail13)
- cmp $2, %esi
- je L(exit_tail14)
-
- cmpb $0, 14(%rdi)
- jz L(exit_tail14)
- cmp $3, %esi
- je L(exit_tail15)
-
- cmpb $0, 15(%rdi)
- jz L(exit_tail15)
- mov $16, %eax
- ret
-# endif
-
- .p2align 4
-L(exit_tail1):
- add $1, %eax
- RETURN
-
- .p2align 4
-L(exit_tail2):
- add $2, %eax
- RETURN
-
- .p2align 4
-L(exit_tail3):
- add $3, %eax
- RETURN
-
- .p2align 4
-L(exit_tail4):
- add $4, %eax
- RETURN
-
- .p2align 4
-L(exit_tail5):
- add $5, %eax
- RETURN
-
- .p2align 4
-L(exit_tail6):
- add $6, %eax
- RETURN
-
- .p2align 4
-L(exit_tail7):
- add $7, %eax
- RETURN
-
- .p2align 4
-L(exit_tail8):
- add $8, %eax
- RETURN
-
- .p2align 4
-L(exit_tail9):
- add $9, %eax
- RETURN
-
- .p2align 4
-L(exit_tail10):
- add $10, %eax
- RETURN
-
- .p2align 4
-L(exit_tail11):
- add $11, %eax
- RETURN
-
- .p2align 4
-L(exit_tail12):
- add $12, %eax
- RETURN
-
- .p2align 4
-L(exit_tail13):
- add $13, %eax
- RETURN
-
- .p2align 4
-L(exit_tail14):
- add $14, %eax
- RETURN
-
- .p2align 4
-L(exit_tail15):
- add $15, %eax
-# ifndef USE_AS_STRCAT
- RETURN
-END (STRLEN)
-# endif
-#endif
diff --git a/sysdeps/x86_64/multiarch/strlen-sse2-pminub.S b/sysdeps/x86_64/multiarch/strlen-sse2-pminub.S
deleted file mode 100644
index cc4bb57..0000000
--- a/sysdeps/x86_64/multiarch/strlen-sse2-pminub.S
+++ /dev/null
@@ -1,259 +0,0 @@
-/* strlen SSE2
- Copyright (C) 2011-2013 Free Software Foundation, Inc.
- Contributed by Intel Corporation.
- This file is part of the GNU C Library.
-
- The GNU C Library is free software; you can redistribute it and/or
- modify it under the terms of the GNU Lesser General Public
- License as published by the Free Software Foundation; either
- version 2.1 of the License, or (at your option) any later version.
-
- The GNU C Library 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
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public
- License along with the GNU C Library; if not, see
- <http://www.gnu.org/licenses/>. */
-
-#if !defined NOT_IN_libc && (defined SHARED || defined USE_AS_STRCAT)
-
-# ifndef USE_AS_STRCAT
-
-# include <sysdep.h>
-
-# define RETURN ret
-
- .section .text.sse2,"ax",@progbits
-ENTRY (__strlen_sse2_pminub)
-
-# endif
- xor %rax, %rax
- mov %edi, %ecx
- and $0x3f, %ecx
- pxor %xmm0, %xmm0
- cmp $0x30, %ecx
- ja L(next)
- movdqu (%rdi), %xmm1
- pcmpeqb %xmm1, %xmm0
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit_less16)
- mov %rdi, %rax
- and $-16, %rax
- jmp L(align16_start)
-L(next):
- mov %rdi, %rax
- and $-16, %rax
- pcmpeqb (%rax), %xmm0
- mov $-1, %r10d
- sub %rax, %rcx
- shl %cl, %r10d
- pmovmskb %xmm0, %edx
- and %r10d, %edx
- jnz L(exit)
-L(align16_start):
- pxor %xmm0, %xmm0
- pxor %xmm1, %xmm1
- pxor %xmm2, %xmm2
- pxor %xmm3, %xmm3
- pcmpeqb 16(%rax), %xmm0
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit16)
-
- pcmpeqb 32(%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- jnz L(exit32)
-
- pcmpeqb 48(%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- jnz L(exit48)
-
- pcmpeqb 64(%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- jnz L(exit64)
-
- pcmpeqb 80(%rax), %xmm0
- add $64, %rax
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit16)
-
- pcmpeqb 32(%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- jnz L(exit32)
-
- pcmpeqb 48(%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- jnz L(exit48)
-
- pcmpeqb 64(%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- jnz L(exit64)
-
- pcmpeqb 80(%rax), %xmm0
- add $64, %rax
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit16)
-
- pcmpeqb 32(%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- jnz L(exit32)
-
- pcmpeqb 48(%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- jnz L(exit48)
-
- pcmpeqb 64(%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- jnz L(exit64)
-
- pcmpeqb 80(%rax), %xmm0
- add $64, %rax
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit16)
-
- pcmpeqb 32(%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- jnz L(exit32)
-
- pcmpeqb 48(%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- jnz L(exit48)
-
- pcmpeqb 64(%rax), %xmm3
- pmovmskb %xmm3, %edx
- test %edx, %edx
- jnz L(exit64)
-
-
- test $0x3f, %rax
- jz L(align64_loop)
-
- pcmpeqb 80(%rax), %xmm0
- add $80, %rax
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit)
-
- test $0x3f, %rax
- jz L(align64_loop)
-
- pcmpeqb 16(%rax), %xmm1
- add $16, %rax
- pmovmskb %xmm1, %edx
- test %edx, %edx
- jnz L(exit)
-
- test $0x3f, %rax
- jz L(align64_loop)
-
- pcmpeqb 16(%rax), %xmm2
- add $16, %rax
- pmovmskb %xmm2, %edx
- test %edx, %edx
- jnz L(exit)
-
- test $0x3f, %rax
- jz L(align64_loop)
-
- pcmpeqb 16(%rax), %xmm3
- add $16, %rax
- pmovmskb %xmm3, %edx
- test %edx, %edx
- jnz L(exit)
-
- add $16, %rax
- .p2align 4
- L(align64_loop):
- movaps (%rax), %xmm4
- pminub 16(%rax), %xmm4
- movaps 32(%rax), %xmm5
- pminub 48(%rax), %xmm5
- add $64, %rax
- pminub %xmm4, %xmm5
- pcmpeqb %xmm0, %xmm5
- pmovmskb %xmm5, %edx
- test %edx, %edx
- jz L(align64_loop)
-
-
- pcmpeqb -64(%rax), %xmm0
- sub $80, %rax
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit16)
-
- pcmpeqb 32(%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- jnz L(exit32)
-
- pcmpeqb 48(%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- jnz L(exit48)
-
- pcmpeqb 64(%rax), %xmm3
- pmovmskb %xmm3, %edx
- sub %rdi, %rax
- bsf %rdx, %rdx
- add %rdx, %rax
- add $64, %rax
- RETURN
-
- .p2align 4
-L(exit):
- sub %rdi, %rax
-L(exit_less16):
- bsf %rdx, %rdx
- add %rdx, %rax
- RETURN
- .p2align 4
-L(exit16):
- sub %rdi, %rax
- bsf %rdx, %rdx
- add %rdx, %rax
- add $16, %rax
- RETURN
- .p2align 4
-L(exit32):
- sub %rdi, %rax
- bsf %rdx, %rdx
- add %rdx, %rax
- add $32, %rax
- RETURN
- .p2align 4
-L(exit48):
- sub %rdi, %rax
- bsf %rdx, %rdx
- add %rdx, %rax
- add $48, %rax
- RETURN
- .p2align 4
-L(exit64):
- sub %rdi, %rax
- bsf %rdx, %rdx
- add %rdx, %rax
- add $64, %rax
-# ifndef USE_AS_STRCAT
- RETURN
-
-END (__strlen_sse2_pminub)
-# endif
-#endif
diff --git a/sysdeps/x86_64/multiarch/strlen-sse4.S b/sysdeps/x86_64/multiarch/strlen-sse4.S
deleted file mode 100644
index 8d685df..0000000
--- a/sysdeps/x86_64/multiarch/strlen-sse4.S
+++ /dev/null
@@ -1,84 +0,0 @@
-/* strlen with SSE4
- Copyright (C) 2009-2013 Free Software Foundation, Inc.
- Contributed by Ulrich Drepper <drepper@redhat.com>.
- This file is part of the GNU C Library.
-
- The GNU C Library is free software; you can redistribute it and/or
- modify it under the terms of the GNU Lesser General Public
- License as published by the Free Software Foundation; either
- version 2.1 of the License, or (at your option) any later version.
-
- The GNU C Library 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
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public
- License along with the GNU C Library; if not, see
- <http://www.gnu.org/licenses/>. */
-
-#if defined SHARED && !defined NOT_IN_libc
-
-#include <sysdep.h>
-
- .section .text.sse4.2,"ax",@progbits
-ENTRY (__strlen_sse42)
- pxor %xmm1, %xmm1
- movl %edi, %ecx
- movq %rdi, %r8
- andq $~15, %rdi
- xor %edi, %ecx
- pcmpeqb (%rdi), %xmm1
- pmovmskb %xmm1, %edx
- shrl %cl, %edx
- shll %cl, %edx
- andl %edx, %edx
- jnz L(less16bytes)
- pxor %xmm1, %xmm1
-
- .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
-
- .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
-
-END (__strlen_sse42)
-
-#endif
diff --git a/sysdeps/x86_64/multiarch/strlen.S b/sysdeps/x86_64/multiarch/strlen.S
deleted file mode 100644
index ab29cef..0000000
--- a/sysdeps/x86_64/multiarch/strlen.S
+++ /dev/null
@@ -1,68 +0,0 @@
-/* Multiple versions of strlen(str) -- determine the length of the string STR.
- All versions must be listed in ifunc-impl-list.c.
- Copyright (C) 2009-2013 Free Software Foundation, Inc.
- Contributed by Ulrich Drepper <drepper@redhat.com>.
- This file is part of the GNU C Library.
-
- The GNU C Library is free software; you can redistribute it and/or
- modify it under the terms of the GNU Lesser General Public
- License as published by the Free Software Foundation; either
- version 2.1 of the License, or (at your option) any later version.
-
- The GNU C Library 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
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public
- License along with the GNU C Library; if not, see
- <http://www.gnu.org/licenses/>. */
-
-#include <sysdep.h>
-#include <init-arch.h>
-
-
-/* Define multiple versions only for the definition in libc and for
- the DSO. In static binaries we need strlen before the initialization
- happened. */
-#if defined SHARED && !defined NOT_IN_libc
- .text
-ENTRY(strlen)
- .type strlen, @gnu_indirect_function
- cmpl $0, __cpu_features+KIND_OFFSET(%rip)
- jne 1f
- call __init_cpu_features
-1: leaq __strlen_sse2_pminub(%rip), %rax
- testl $bit_Prefer_PMINUB_for_stringop, __cpu_features+FEATURE_OFFSET+index_Prefer_PMINUB_for_stringop(%rip)
- jnz 2f
- leaq __strlen_sse2(%rip), %rax
- testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
- jz 2f
- leaq __strlen_sse42(%rip), %rax
- ret
-2: testl $bit_Slow_BSF, __cpu_features+FEATURE_OFFSET+index_Slow_BSF(%rip)
- jz 3f
- leaq __strlen_sse2_no_bsf(%rip), %rax
-3: ret
-END(strlen)
-
-# undef ENTRY
-# define ENTRY(name) \
- .type __strlen_sse2, @function; \
- .align 16; \
- .globl __strlen_sse2; \
- .hidden __strlen_sse2; \
- __strlen_sse2: cfi_startproc; \
- CALL_MCOUNT
-# undef END
-# define END(name) \
- cfi_endproc; .size __strlen_sse2, .-__strlen_sse2
-# undef libc_hidden_builtin_def
-/* It doesn't make sense to send libc-internal strlen calls through a PLT.
- The speedup we get from using SSE4.2 instruction is likely eaten away
- by the indirect call in the PLT. */
-# define libc_hidden_builtin_def(name) \
- .globl __GI_strlen; __GI_strlen = __strlen_sse2
-#endif
-
-#include "../strlen.S"
diff --git a/sysdeps/x86_64/multiarch/strnlen-sse2-no-bsf.S b/sysdeps/x86_64/multiarch/strnlen-sse2-no-bsf.S
deleted file mode 100644
index 248328d..0000000
--- a/sysdeps/x86_64/multiarch/strnlen-sse2-no-bsf.S
+++ /dev/null
@@ -1,3 +0,0 @@
-#define USE_AS_STRNLEN
-#define STRLEN __strnlen_sse2_no_bsf
-#include "strlen-sse2-no-bsf.S"
diff --git a/sysdeps/x86_64/multiarch/strnlen.S b/sysdeps/x86_64/multiarch/strnlen.S
deleted file mode 100644
index 124f845..0000000
--- a/sysdeps/x86_64/multiarch/strnlen.S
+++ /dev/null
@@ -1,57 +0,0 @@
-/* multiple version of strnlen
- All versions must be listed in ifunc-impl-list.c.
- Copyright (C) 2011-2013 Free Software Foundation, Inc.
- This file is part of the GNU C Library.
-
- The GNU C Library is free software; you can redistribute it and/or
- modify it under the terms of the GNU Lesser General Public
- License as published by the Free Software Foundation; either
- version 2.1 of the License, or (at your option) any later version.
-
- The GNU C Library 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
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public
- License along with the GNU C Library; if not, see
- <http://www.gnu.org/licenses/>. */
-
-#include <sysdep.h>
-#include <init-arch.h>
-
-
-/* Define multiple versions only for the definition in libc. */
-#ifndef NOT_IN_libc
-
- .text
-ENTRY(__strnlen)
- .type __strnlen, @gnu_indirect_function
- cmpl $0, __cpu_features+KIND_OFFSET(%rip)
- jne 1f
- call __init_cpu_features
-1: leaq __strnlen_sse2(%rip), %rax
- testl $bit_Slow_BSF, __cpu_features+FEATURE_OFFSET+index_Slow_BSF(%rip)
- jz 2f
- leaq __strnlen_sse2_no_bsf(%rip), %rax
-2: ret
-END(__strnlen)
-
-# undef ENTRY
-# define ENTRY(name) \
- .type __strnlen_sse2, @function; \
- .align 16; \
- .globl __strnlen_sse2; \
- .hidden __strnlen_sse2; \
- __strnlen_sse2: cfi_startproc; \
- CALL_MCOUNT
-# undef END
-# define END(name) \
- cfi_endproc; .size __strnlen_sse2, .-__strnlen_sse2
-
-# undef libc_hidden_def
-# define libc_hidden_def(name) \
- .globl __GI_strnlen; __GI_strnlen = __strnlen_sse2
-#endif
-
-#include "../strnlen.S"
diff --git a/sysdeps/x86_64/strlen.S b/sysdeps/x86_64/strlen.S
index 4bdca0a..c8ced10 100644
--- a/sysdeps/x86_64/strlen.S
+++ b/sysdeps/x86_64/strlen.S
@@ -1,6 +1,5 @@
-/* strlen(str) -- determine the length of the string STR.
- Copyright (C) 2009-2013 Free Software Foundation, Inc.
- Contributed by Ulrich Drepper <drepper@redhat.com>.
+/* SSE2 version of strlen.
+ Copyright (C) 2012, 2013 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
@@ -19,83 +18,219 @@
#include <sysdep.h>
+/* Used in linker - use only %xmm8-%xmm15. */
- .text
+/* Long lived register are
+ strlen(s), strnlen(s, n):
+
+ %xmm11 - zero
+ %rdi - s
+ %r10 (s+n) & (~(64-1))
+ %r11 s+n
+*/
+
+
+.text
ENTRY(strlen)
+
+#define FIND_ZERO \
+ pcmpeqb (%rax), %xmm8; \
+ pcmpeqb 16(%rax), %xmm9; \
+ pcmpeqb 32(%rax), %xmm10; \
+ pcmpeqb 48(%rax), %xmm11; \
+ pmovmskb %xmm8, %esi; \
+ pmovmskb %xmm9, %edx; \
+ pmovmskb %xmm10, %r8d; \
+ pmovmskb %xmm11, %ecx; \
+ salq $16, %rdx; \
+ salq $16, %rcx; \
+ orq %rsi, %rdx; \
+ orq %r8, %rcx; \
+ salq $32, %rcx; \
+ orq %rcx, %rdx;
+
+#ifdef AS_STRNLEN
+/* Do not read anything when n==0. */
+ test %rsi, %rsi
+ jne L(n_nonzero)
xor %rax, %rax
- mov %edi, %ecx
- and $0x3f, %ecx
- pxor %xmm0, %xmm0
- cmp $0x30, %ecx
+ ret
+L(n_nonzero):
+
+/* Initialize long lived registers. */
+
+ add %rdi, %rsi
+ mov %rsi, %r10
+ and $-64, %r10
+ mov %rsi, %r11
+#endif
+
+ pxor %xmm8, %xmm8
+ pxor %xmm9, %xmm9
+ pxor %xmm10, %xmm10
+ pxor %xmm11, %xmm11
+ movq %rdi, %rax
+ movq %rdi, %rcx
+ andq $4095, %rcx
+/* Offsets 4032-4047 will be aligned into 4032 thus fit into page. */
+ cmpq $4047, %rcx
+/* We cannot unify this branching as it would be ~6 cycles slower. */
ja L(next)
- movdqu (%rdi), %xmm1
- pcmpeqb %xmm1, %xmm0
- pmovmskb %xmm0, %edx
+
+#ifdef AS_STRNLEN
+# define STRNLEN_PROLOG \
+ mov %r11, %rsi; \
+ subq %rax, %rsi; \
+ andq $-64, %rax; \
+ testq $-64, %rsi; \
+ je L(strnlen_ret)
+#else
+# define STRNLEN_PROLOG andq $-64, %rax;
+#endif
+
+#define PROLOG(lab) \
+ movq %rdi, %rcx; \
+ xorq %rax, %rcx; \
+ STRNLEN_PROLOG; \
+ sarq %cl, %rdx; \
+ test %rdx, %rdx; \
+ je L(lab); \
+ bsfq %rdx, %rax; \
+ ret
+
+#ifdef AS_STRNLEN
+ andq $-16, %rax
+ FIND_ZERO
+#else
+ movdqu (%rax), %xmm12
+ pcmpeqb %xmm8, %xmm12
+ pmovmskb %xmm12, %edx
test %edx, %edx
- jnz L(exit_less16)
- mov %rdi, %rax
- and $-16, %rax
- jmp L(align16_start)
+ je L(next48_bytes)
+ bsfq %rdx, %rax
+ ret
+
+L(next48_bytes):
+/* Same as FIND_ZERO except we do not check first 16 bytes. */
+ andq $-16, %rax
+ pcmpeqb 16(%rax), %xmm9;
+ pcmpeqb 32(%rax), %xmm10;
+ pcmpeqb 48(%rax), %xmm11;
+ pmovmskb %xmm9, %edx;
+ pmovmskb %xmm10, %r8d;
+ pmovmskb %xmm11, %ecx;
+ salq $16, %rdx;
+ salq $16, %rcx;
+ orq %r8, %rcx;
+ salq $32, %rcx;
+ orq %rcx, %rdx;
+#endif
+
+ PROLOG(loop)
+
+ .p2align 4
L(next):
- mov %rdi, %rax
- and $-16, %rax
- pcmpeqb (%rax), %xmm0
- mov $-1, %esi
- sub %rax, %rcx
- shl %cl, %esi
- pmovmskb %xmm0, %edx
- and %esi, %edx
- jnz L(exit)
-L(align16_start):
- pxor %xmm0, %xmm0
- pxor %xmm1, %xmm1
- pxor %xmm2, %xmm2
- pxor %xmm3, %xmm3
+ andq $-64, %rax
+ FIND_ZERO
+ PROLOG(loop_init)
+
+#ifdef AS_STRNLEN
+/* We must do this check to correctly handle strnlen (s, -1). */
+L(strnlen_ret):
+ bts %rsi, %rdx
+ sarq %cl, %rdx
+ test %rdx, %rdx
+ je L(loop_init)
+ bsfq %rdx, %rax
+ ret
+#endif
.p2align 4
-L(align16_loop):
- pcmpeqb 16(%rax), %xmm0
- pmovmskb %xmm0, %edx
- test %edx, %edx
- jnz L(exit16)
+L(loop_init):
+ pxor %xmm9, %xmm9
+ pxor %xmm10, %xmm10
+ pxor %xmm11, %xmm11
+#ifdef AS_STRNLEN
+ .p2align 4
+L(loop):
- pcmpeqb 32(%rax), %xmm1
- pmovmskb %xmm1, %edx
- test %edx, %edx
- jnz L(exit32)
+ addq $64, %rax
+ cmpq %rax, %r10
+ je L(exit_end)
- pcmpeqb 48(%rax), %xmm2
- pmovmskb %xmm2, %edx
- test %edx, %edx
- jnz L(exit48)
+ movdqa (%rax), %xmm8
+ pminub 16(%rax), %xmm8
+ pminub 32(%rax), %xmm8
+ pminub 48(%rax), %xmm8
+ pcmpeqb %xmm11, %xmm8
+ pmovmskb %xmm8, %edx
+ testl %edx, %edx
+ jne L(exit)
+ jmp L(loop)
- pcmpeqb 64(%rax), %xmm3
- pmovmskb %xmm3, %edx
- lea 64(%rax), %rax
- test %edx, %edx
- jz L(align16_loop)
-L(exit):
- sub %rdi, %rax
-L(exit_less16):
- bsf %rdx, %rdx
- add %rdx, %rax
- ret
.p2align 4
-L(exit16):
- sub %rdi, %rax
- bsf %rdx, %rdx
- lea 16(%rdx,%rax), %rax
+L(exit_end):
+ cmp %rax, %r11
+ je L(first)
+ pxor %xmm8, %xmm8
+ FIND_ZERO
+
+L(first):
+ bts %r11, %rdx
+ bsfq %rdx, %rdx
+ addq %rdx, %rax
+ subq %rdi, %rax
ret
+
.p2align 4
-L(exit32):
- sub %rdi, %rax
- bsf %rdx, %rdx
- lea 32(%rdx,%rax), %rax
+L(exit):
+ pxor %xmm8, %xmm8
+ FIND_ZERO
+
+ bsfq %rdx, %rdx
+ addq %rdx, %rax
+ subq %rdi, %rax
ret
+
+#else
+ .p2align 4
+L(loop):
+
+ movdqa 64(%rax), %xmm8
+ pminub 80(%rax), %xmm8
+ pminub 96(%rax), %xmm8
+ pminub 112(%rax), %xmm8
+ pcmpeqb %xmm11, %xmm8
+ pmovmskb %xmm8, %edx
+ testl %edx, %edx
+ jne L(exit64)
+
+ subq $-128, %rax
+
+ movdqa (%rax), %xmm8
+ pminub 16(%rax), %xmm8
+ pminub 32(%rax), %xmm8
+ pminub 48(%rax), %xmm8
+ pcmpeqb %xmm11, %xmm8
+ pmovmskb %xmm8, %edx
+ testl %edx, %edx
+ jne L(exit0)
+ jmp L(loop)
+
.p2align 4
-L(exit48):
- sub %rdi, %rax
- bsf %rdx, %rdx
- lea 48(%rdx,%rax), %rax
+L(exit64):
+ addq $64, %rax
+L(exit0):
+ pxor %xmm8, %xmm8
+ FIND_ZERO
+
+ bsfq %rdx, %rdx
+ addq %rdx, %rax
+ subq %rdi, %rax
ret
+
+#endif
+
END(strlen)
+#ifndef AS_STRLEN
libc_hidden_builtin_def (strlen)
+#endif
diff --git a/sysdeps/x86_64/strnlen.S b/sysdeps/x86_64/strnlen.S
index 6e53503..d0694a5 100644
--- a/sysdeps/x86_64/strnlen.S
+++ b/sysdeps/x86_64/strnlen.S
@@ -1,63 +1,7 @@
-/* strnlen(str,maxlen) -- determine the length of the string STR up to MAXLEN.
- Copyright (C) 2010-2013 Free Software Foundation, Inc.
- Contributed by Ulrich Drepper <drepper@redhat.com>.
- This file is part of the GNU C Library.
+#define AS_STRNLEN
+#define strlen __strnlen
+#include "strlen.S"
- The GNU C Library is free software; you can redistribute it and/or
- modify it under the terms of the GNU Lesser General Public
- License as published by the Free Software Foundation; either
- version 2.1 of the License, or (at your option) any later version.
+weak_alias (__strnlen, strnlen);
+libc_hidden_builtin_def (strnlen)
- The GNU C Library 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
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public
- License along with the GNU C Library; if not, see
- <http://www.gnu.org/licenses/>. */
-
-#include <sysdep.h>
-
-
- .text
-ENTRY(__strnlen)
- movq %rsi, %rax
- testq %rsi, %rsi
- jz 3f
- pxor %xmm2, %xmm2
- movq %rdi, %rcx
- movq %rdi, %r8
- movq $16, %r9
- andq $~15, %rdi
- movdqa %xmm2, %xmm1
- pcmpeqb (%rdi), %xmm2
- orl $0xffffffff, %r10d
- subq %rdi, %rcx
- shll %cl, %r10d
- subq %rcx, %r9
- pmovmskb %xmm2, %edx
- andl %r10d, %edx
- jnz 1f
- subq %r9, %rsi
- jbe 3f
-
-2: movdqa 16(%rdi), %xmm0
- leaq 16(%rdi), %rdi
- pcmpeqb %xmm1, %xmm0
- pmovmskb %xmm0, %edx
- testl %edx, %edx
- jnz 1f
- subq $16, %rsi
- jnbe 2b
-3: ret
-
-1: subq %r8, %rdi
- bsfl %edx, %edx
- addq %rdi, %rdx
- cmpq %rdx, %rax
- cmovnbq %rdx, %rax
- ret
-END(__strnlen)
-weak_alias (__strnlen, strnlen)
-libc_hidden_def (strnlen)
--
1.7.4.4
More information about the Libc-alpha
mailing list