diff options
Diffstat (limited to 'sysdeps/x86_64/multiarch/mempcpy.S')
| -rw-r--r-- | sysdeps/x86_64/multiarch/mempcpy.S | 39 |
1 files changed, 21 insertions, 18 deletions
diff --git a/sysdeps/x86_64/multiarch/mempcpy.S b/sysdeps/x86_64/multiarch/mempcpy.S index ad36840d54..450915f60f 100644 --- a/sysdeps/x86_64/multiarch/mempcpy.S +++ b/sysdeps/x86_64/multiarch/mempcpy.S @@ -28,41 +28,44 @@ ENTRY(__mempcpy) .type __mempcpy, @gnu_indirect_function LOAD_RTLD_GLOBAL_RO_RDX + leaq __mempcpy_avx_unaligned(%rip), %rax + HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load) + jnz 2f + leaq __mempcpy_sse2_unaligned(%rip), %rax + HAS_ARCH_FEATURE (Fast_Unaligned_Load) + jnz 2f leaq __mempcpy_sse2(%rip), %rax HAS_CPU_FEATURE (SSSE3) jz 2f - leaq __mempcpy_ssse3(%rip), %rax - HAS_ARCH_FEATURE (Fast_Copy_Backward) - jz 2f - leaq __mempcpy_ssse3_back(%rip), %rax - HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load) - jz 2f - leaq __mempcpy_avx_unaligned(%rip), %rax + leaq __mempcpy_ssse3_back(%rip), %rax + HAS_CPU_FEATURE (Fast_Copy_Backward) + jnz 2f + leaq __mempcpy_ssse3(%rip), %rax 2: ret END(__mempcpy) # undef ENTRY # define ENTRY(name) \ - .type __mempcpy_sse2, @function; \ + .type __mempcpy_sse2_unaligned, @function; \ .p2align 4; \ - .globl __mempcpy_sse2; \ - .hidden __mempcpy_sse2; \ - __mempcpy_sse2: cfi_startproc; \ + .globl __mempcpy_sse2_unaligned; \ + .hidden __mempcpy_sse2_unaligned; \ + __mempcpy_sse2_unaligned: cfi_startproc; \ CALL_MCOUNT # undef END # define END(name) \ - cfi_endproc; .size __mempcpy_sse2, .-__mempcpy_sse2 + cfi_endproc; .size __mempcpy_sse2_unaligned, .-__mempcpy_sse2_unaligned # undef ENTRY_CHK # define ENTRY_CHK(name) \ - .type __mempcpy_chk_sse2, @function; \ - .globl __mempcpy_chk_sse2; \ + .type __mempcpy_chk_sse2_unaligned, @function; \ + .globl __mempcpy_chk_sse2_unaligned; \ .p2align 4; \ - __mempcpy_chk_sse2: cfi_startproc; \ + __mempcpy_chk_sse2_unaligned: cfi_startproc; \ CALL_MCOUNT # undef END_CHK # define END_CHK(name) \ - cfi_endproc; .size __mempcpy_chk_sse2, .-__mempcpy_chk_sse2 + cfi_endproc; .size __mempcpy_chk_sse2_unaligned, .-__mempcpy_chk_sse2_unaligned # undef libc_hidden_def # undef libc_hidden_builtin_def @@ -70,9 +73,9 @@ END(__mempcpy) The speedup we get from using SSSE3 instruction is likely eaten away by the indirect call in the PLT. */ # define libc_hidden_def(name) \ - .globl __GI_mempcpy; __GI_mempcpy = __mempcpy_sse2 + .globl __GI_mempcpy; __GI_mempcpy = __mempcpy_sse2_unaligned # define libc_hidden_builtin_def(name) \ - .globl __GI___mempcpy; __GI___mempcpy = __mempcpy_sse2 + .globl __GI___mempcpy; __GI___mempcpy = __mempcpy_sse2_unaligned #endif #include "../mempcpy.S" |