From ea2785e96fa503f3a2b5dd9f3a6ca65622b3c5f2 Mon Sep 17 00:00:00 2001 From: "H.J. Lu" Date: Sun, 3 Apr 2016 12:38:25 -0700 Subject: Fix memmove-vec-unaligned-erms.S __mempcpy_erms and __memmove_erms can't be placed between __memmove_chk and __memmove it breaks __memmove_chk. Don't check source == destination first since it is less common. * sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S: (__mempcpy_erms, __memmove_erms): Moved before __mempcpy_chk with unaligned_erms. (__memmove_erms): Skip if source == destination. (__memmove_unaligned_erms): Don't check source == destination first. --- .../x86_64/multiarch/memmove-vec-unaligned-erms.S | 54 ++++++++++++---------- 1 file changed, 30 insertions(+), 24 deletions(-) (limited to 'sysdeps/x86_64') diff --git a/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S b/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S index cf645dd7ff..66779a3bec 100644 --- a/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S +++ b/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S @@ -95,46 +95,30 @@ L(start): ret END (MEMMOVE_SYMBOL (__memmove, unaligned_2)) -# ifdef SHARED -ENTRY (MEMMOVE_SYMBOL (__mempcpy_chk, unaligned_erms)) - cmpq %rdx, %rcx - jb HIDDEN_JUMPTARGET (__chk_fail) -END (MEMMOVE_SYMBOL (__mempcpy_chk, unaligned_erms)) -# endif - -ENTRY (MEMMOVE_SYMBOL (__mempcpy, unaligned_erms)) - movq %rdi, %rax - addq %rdx, %rax - jmp L(start_erms) -END (MEMMOVE_SYMBOL (__mempcpy, unaligned_erms)) - -# ifdef SHARED -ENTRY (MEMMOVE_SYMBOL (__memmove_chk, unaligned_erms)) - cmpq %rdx, %rcx - jb HIDDEN_JUMPTARGET (__chk_fail) -END (MEMMOVE_SYMBOL (__memmove_chk, unaligned_erms)) -# endif - # if VEC_SIZE == 16 /* Only used to measure performance of REP MOVSB. */ # ifdef SHARED ENTRY (__mempcpy_erms) movq %rdi, %rax addq %rdx, %rax - jmp L(movsb) + jmp L(start_movsb) END (__mempcpy_erms) # endif ENTRY (__memmove_erms) movq %rdi, %rax +L(start_movsb): movq %rdx, %rcx cmpq %rsi, %rdi - jbe 1f + jb 1f + /* Source == destination is less common. */ + je 2f leaq (%rsi,%rcx), %rdx cmpq %rdx, %rdi jb L(movsb_backward) 1: rep movsb +2: ret L(movsb_backward): leaq -1(%rdi,%rcx), %rdi @@ -147,6 +131,26 @@ END (__memmove_erms) strong_alias (__memmove_erms, __memcpy_erms) # endif +# ifdef SHARED +ENTRY (MEMMOVE_SYMBOL (__mempcpy_chk, unaligned_erms)) + cmpq %rdx, %rcx + jb HIDDEN_JUMPTARGET (__chk_fail) +END (MEMMOVE_SYMBOL (__mempcpy_chk, unaligned_erms)) +# endif + +ENTRY (MEMMOVE_SYMBOL (__mempcpy, unaligned_erms)) + movq %rdi, %rax + addq %rdx, %rax + jmp L(start_erms) +END (MEMMOVE_SYMBOL (__mempcpy, unaligned_erms)) + +# ifdef SHARED +ENTRY (MEMMOVE_SYMBOL (__memmove_chk, unaligned_erms)) + cmpq %rdx, %rcx + jb HIDDEN_JUMPTARGET (__chk_fail) +END (MEMMOVE_SYMBOL (__memmove_chk, unaligned_erms)) +# endif + ENTRY (MEMMOVE_SYMBOL (__memmove, unaligned_erms)) movq %rdi, %rax L(start_erms): @@ -166,8 +170,9 @@ L(return): L(movsb): cmpq %rsi, %rdi - je L(nop) jb 1f + /* Source == destination is less common. */ + je L(nop) leaq (%rsi,%rdx), %r9 cmpq %r9, %rdi /* Avoid slow backward REP MOVSB. */ @@ -191,8 +196,9 @@ L(movsb_more_2x_vec): L(more_2x_vec): /* More than 2 * VEC. */ cmpq %rsi, %rdi - je L(nop) jb L(copy_forward) + /* Source == destination is less common. */ + je L(nop) leaq (%rsi,%rdx), %rcx cmpq %rcx, %rdi jb L(more_2x_vec_overlap) -- cgit 1.4.1