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.

2 files changed, 39 insertions, 24 deletions

diff --git a/ChangeLog b/ChangeLog
index d22231b95d..e93b7bf3b1 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+2016-04-03   H.J. Lu  <hongjiu.lu@intel.com>
+
+	* 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.
+
 2016-04-01   H.J. Lu  <hongjiu.lu@intel.com>
 
 	* sysdeps/x86/cpu-features.c (init_cpu_features): Don't set
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)