diff options
author | Ondřej Bílka <neleai@seznam.cz> | 2013-09-11 17:07:38 +0200 |
---|---|---|
committer | Ondřej Bílka <neleai@seznam.cz> | 2013-09-11 17:07:38 +0200 |
commit | 5905e7b3e29139dbef84c065ca39315485f497e1 (patch) | |
tree | c0333d621bbb599eb32a3c773ebc21ce53d6c438 /sysdeps/x86_64/multiarch | |
parent | b987c776723eea2f223885289c466386f12e823a (diff) | |
download | glibc-5905e7b3e29139dbef84c065ca39315485f497e1.tar.gz glibc-5905e7b3e29139dbef84c065ca39315485f497e1.tar.xz glibc-5905e7b3e29139dbef84c065ca39315485f497e1.zip |
Faster strchr implementation.
Diffstat (limited to 'sysdeps/x86_64/multiarch')
-rw-r--r-- | sysdeps/x86_64/multiarch/ifunc-impl-list.c | 1 | ||||
-rw-r--r-- | sysdeps/x86_64/multiarch/strchr.S | 127 |
2 files changed, 0 insertions, 128 deletions
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c index f8756d7af2..1a65ac04ff 100644 --- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c +++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c @@ -110,7 +110,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, /* Support sysdeps/x86_64/multiarch/strchr.S. */ IFUNC_IMPL (i, name, strchr, - IFUNC_IMPL_ADD (array, i, strchr, HAS_SSE4_2, __strchr_sse42) IFUNC_IMPL_ADD (array, i, strchr, 1, __strchr_sse2_no_bsf) IFUNC_IMPL_ADD (array, i, strchr, 1, __strchr_sse2)) diff --git a/sysdeps/x86_64/multiarch/strchr.S b/sysdeps/x86_64/multiarch/strchr.S index f170238b55..3f0b2c5f5a 100644 --- a/sysdeps/x86_64/multiarch/strchr.S +++ b/sysdeps/x86_64/multiarch/strchr.S @@ -29,12 +29,6 @@ ENTRY(strchr) jne 1f call __init_cpu_features 1: leaq __strchr_sse2(%rip), %rax - testl $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip) - jnz 2f - testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip) - jz 2f - leaq __strchr_sse42(%rip), %rax - ret 2: testl $bit_Slow_BSF, __cpu_features+FEATURE_OFFSET+index_Slow_BSF(%rip) jz 3f leaq __strchr_sse2_no_bsf(%rip), %rax @@ -42,127 +36,6 @@ ENTRY(strchr) END(strchr) -/* - This implementation uses SSE4 instructions to compare up to 16 bytes - at a time looking for the first occurrence of the character c in the - string s: - - char *strchr (const char *s, int c); - - We use 0xa: - _SIDD_SBYTE_OPS - | _SIDD_CMP_EQUAL_EACH - | _SIDD_LEAST_SIGNIFICANT - on pcmpistri to compare xmm/mem128 - - 0 1 2 3 4 5 6 7 8 9 A B C D E F - X X X X X X X X X X X X X X X X - - against xmm - - 0 1 2 3 4 5 6 7 8 9 A B C D E F - C C C C C C C C C C C C C C C C - - to find out if the first 16byte data element has a byte C and the - offset of the first byte. There are 3 cases: - - 1. The first 16byte data element has the byte C at the offset X. - 2. The first 16byte data element has EOS and doesn't have the byte C. - 3. The first 16byte data element is valid and doesn't have the byte C. - - Here is the table of ECX, CFlag, ZFlag and SFlag for 3 cases: - - case ECX CFlag ZFlag SFlag - 1 X 1 0/1 0 - 2 16 0 1 0 - 3 16 0 0 0 - - We exit from the loop for cases 1 and 2 with jbe which branches - when either CFlag or ZFlag is 1. If CFlag == 1, ECX has the offset - X for case 1. */ - - .section .text.sse4.2,"ax",@progbits - .align 16 - .type __strchr_sse42, @function - .globl __strchr_sse42 - .hidden __strchr_sse42 -__strchr_sse42: - cfi_startproc - CALL_MCOUNT - testb %sil, %sil - je __strend_sse4 - pxor %xmm2, %xmm2 - movd %esi, %xmm1 - movl %edi, %ecx - pshufb %xmm2, %xmm1 - andl $15, %ecx - movq %rdi, %r8 - je L(aligned_start) - -/* Handle unaligned string. */ - andq $-16, %r8 - movdqa (%r8), %xmm0 - pcmpeqb %xmm0, %xmm2 - pcmpeqb %xmm1, %xmm0 - /* Find where NULL is. */ - pmovmskb %xmm2, %edx - /* Check if there is a match. */ - pmovmskb %xmm0, %esi - /* Remove the leading bytes. */ - sarl %cl, %edx - sarl %cl, %esi - testl %esi, %esi - je L(unaligned_no_match) - /* Check which byte is a match. */ - bsfl %esi, %eax - /* Is there a NULL? */ - testl %edx, %edx - je L(unaligned_match) - bsfl %edx, %esi - cmpl %esi, %eax - /* Return NULL if NULL comes first. */ - ja L(return_null) -L(unaligned_match): - addq %rdi, %rax - ret - - .p2align 4 -L(unaligned_no_match): - testl %edx, %edx - jne L(return_null) - -/* Loop start on aligned string. */ -L(loop): - addq $16, %r8 -L(aligned_start): - pcmpistri $0x2, (%r8), %xmm1 - jbe L(wrap) - addq $16, %r8 - pcmpistri $0x2, (%r8), %xmm1 - jbe L(wrap) - addq $16, %r8 - pcmpistri $0x2, (%r8), %xmm1 - jbe L(wrap) - addq $16, %r8 - pcmpistri $0x2, (%r8), %xmm1 - jbe L(wrap) - jmp L(loop) -L(wrap): - jc L(loop_exit) - -/* Return NULL. */ -L(return_null): - xorl %eax, %eax - ret - -/* Loop exit. */ - .p2align 4 -L(loop_exit): - leaq (%r8,%rcx), %rax - ret - cfi_endproc - .size __strchr_sse42, .-__strchr_sse42 - # undef ENTRY # define ENTRY(name) \ |