From 7956a3d27c6552f57c8b1c3893d55e501fe30e14 Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hongjiu.lu@intel.com>
Date: Sun, 26 Jul 2009 13:32:28 -0700
Subject: [PATCH] Add SSE2 support to str{,n}cmp for x86-64.

---
 sysdeps/x86_64/multiarch/Makefile       |    2 +-
 sysdeps/x86_64/multiarch/strcmp.S       |  361 +++----
 sysdeps/x86_64/multiarch/strncmp-c.c    |    8 -
 sysdeps/x86_64/{multiarch => }/strcmp.S | 1611 ++++++++++++++++++-------------
 sysdeps/x86_64/strncmp.S                |    3 +
 5 files changed, 1070 insertions(+), 915 deletions(-)
 delete mode 100644 sysdeps/x86_64/multiarch/strncmp-c.c
 copy sysdeps/x86_64/{multiarch => }/strcmp.S (57%)
 create mode 100644 sysdeps/x86_64/strncmp.S

diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index 5ce14aad8d..b066402204 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -4,7 +4,7 @@ gen-as-const-headers += ifunc-defines.sym
 endif
 
 ifeq ($(subdir),string)
-sysdep_routines += stpncpy-c strncpy-c strncmp-c
+sysdep_routines += stpncpy-c strncpy-c
 ifeq (yes,$(config-cflags-sse4))
 sysdep_routines += strcspn-c strpbrk-c strspn-c strstr-c strcasestr-c
 CFLAGS-strcspn-c.c += -msse4
diff --git a/sysdeps/x86_64/multiarch/strcmp.S b/sysdeps/x86_64/multiarch/strcmp.S
index 15148e4f7f..1a315737af 100644
--- a/sysdeps/x86_64/multiarch/strcmp.S
+++ b/sysdeps/x86_64/multiarch/strcmp.S
@@ -28,9 +28,9 @@
 	/* calculate left number to compare */		\
 	lea	-16(%rcx, %r11), %r9;			\
 	cmp	%r9, %r11;				\
-	jb	LABEL(strcmp_exitz);			\
+	jb	LABEL(strcmp_exitz_sse4_2);		\
 	test	%r9, %r9;				\
-	je	LABEL(strcmp_exitz);			\
+	je	LABEL(strcmp_exitz_sse4_2);		\
 	mov	%r9, %r11
 
 #define STRCMP_SSE42	__strncmp_sse42
@@ -106,9 +106,9 @@ STRCMP_SSE42:
  */
 #ifdef USE_AS_STRNCMP
 	test	%rdx, %rdx
-	je	LABEL(strcmp_exitz)
+	je	LABEL(strcmp_exitz_sse4_2)
 	cmp	$1, %rdx
-	je	LABEL(Byte0)
+	je	LABEL(Byte0_sse4_2)
 	mov	%rdx, %r11
 #endif
 	mov	%esi, %ecx
@@ -117,9 +117,9 @@ STRCMP_SSE42:
 	and	$0x3f, %rcx		/* rsi alignment in cache line */
 	and	$0x3f, %rax		/* rdi alignment in cache line */
 	cmp	$0x30, %ecx
-	ja	LABEL(crosscache)	/* rsi: 16-byte load will cross cache line */
+	ja	LABEL(crosscache_sse4_2)/* rsi: 16-byte load will cross cache line */
 	cmp	$0x30, %eax
-	ja	LABEL(crosscache)	/* rdi: 16-byte load will cross cache line */
+	ja	LABEL(crosscache_sse4_2)/* rdi: 16-byte load will cross cache line */
 	movdqu	(%rdi), %xmm1
 	movdqu	(%rsi), %xmm2
 	pxor	%xmm0, %xmm0		/* clear %xmm0 for null char checks */
@@ -128,10 +128,10 @@ STRCMP_SSE42:
 	psubb	%xmm0, %xmm1		/* packed sub of comparison results*/
 	pmovmskb %xmm1, %edx
 	sub	$0xffff, %edx		/* if first 16 bytes are same, edx == 0xffff */
-	jnz	LABEL(less16bytes)	/* If not, find different value or null char */
+	jnz	LABEL(less16bytes_sse4_2)/* If not, find different value or null char */
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)	/* finish comparision */
+	jbe	LABEL(strcmp_exitz_sse4_2)/* finish comparision */
 #endif
 	add	$16, %rsi		/* prepare to search next 16 bytes */
 	add	$16, %rdi		/* prepare to search next 16 bytes */
@@ -142,7 +142,7 @@ STRCMP_SSE42:
 	 * below to use.
 	 */
 	.p2align 4
-LABEL(crosscache):
+LABEL(crosscache_sse4_2):
 	and	$0xfffffffffffffff0, %rsi	/* force %rsi is 16 byte aligned */
 	and	$0xfffffffffffffff0, %rdi	/* force %rdi is 16 byte aligned */
 	mov	$0xffff, %edx			/* for equivalent offset */
@@ -150,15 +150,15 @@ LABEL(crosscache):
 	and	$0xf, %ecx			/* offset of rsi */
 	and	$0xf, %eax			/* offset of rdi */
 	cmp	%eax, %ecx
-	je	LABEL(ashr_0)			/* rsi and rdi relative offset same */
-	ja	LABEL(bigger)
+	je	LABEL(ashr_0_sse4_2)		/* rsi and rdi relative offset same */
+	ja	LABEL(bigger_sse4_2)
 	mov	%edx, %r8d			/* r8d is offset flag for exit tail */
 	xchg	%ecx, %eax
 	xchg	%rsi, %rdi
-LABEL(bigger):
+LABEL(bigger_sse4_2):
 	lea	15(%rax), %r9
 	sub	%rcx, %r9
-	lea	LABEL(unaligned_table)(%rip), %r10
+	lea	LABEL(unaligned_table_sse4_2)(%rip), %r10
 	movslq	(%r10, %r9,4), %r9
 	lea	(%r10, %r9), %r10
 	jmp	*%r10				/* jump to corresponding case */
@@ -169,7 +169,7 @@ LABEL(bigger):
  *        n(0~15)            n(0~15)           15(15+ n-n)         ashr_0
  */
 	.p2align 4
-LABEL(ashr_0):
+LABEL(ashr_0_sse4_2):
 
 	movdqa	(%rsi), %xmm1
 	pxor	%xmm0, %xmm0			/* clear %xmm0 for null char check */
@@ -184,7 +184,7 @@ LABEL(ashr_0):
 	 * edx must be the same with r9d if in left byte (16-rcx) is equal to
 	 * the start from (16-rax) and no null char was seen.
 	 */
-	jne	LABEL(less32bytes)		/* mismatch or null char */
+	jne	LABEL(less32bytes_sse4_2)	/* mismatch or null char */
 	UPDATE_STRNCMP_COUNTER
 	mov	$16, %rcx
 	mov	$16, %r9
@@ -203,7 +203,7 @@ LABEL(ashr_0_use_sse4_2):
 	jbe	LABEL(ashr_0_use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	movdqa	(%rdi,%rdx), %xmm0
@@ -212,17 +212,17 @@ LABEL(ashr_0_use_sse4_2):
 	jbe	LABEL(ashr_0_use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	jmp	LABEL(ashr_0_use_sse4_2)
 
 
 	.p2align 4
 LABEL(ashr_0_use_sse4_2_exit):
-	jnc	LABEL(strcmp_exitz)
+	jnc	LABEL(strcmp_exitz_sse4_2)
 #ifdef USE_AS_STRNCMP
 	sub	%rcx, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	lea	-16(%rdx, %rcx), %rcx
 	movzbl	(%rdi, %rcx), %eax
@@ -239,7 +239,7 @@ LABEL(ashr_0_use_sse4_2_exit):
  *        n(15)            n -15            0(15 +(n-15) - n)         ashr_1
  */
 	.p2align 4
-LABEL(ashr_1):
+LABEL(ashr_1_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -251,7 +251,7 @@ LABEL(ashr_1):
 	shr	%cl, %edx		/* adjust 0xffff for offset */
 	shr	%cl, %r9d		/* adjust for 16-byte offset */
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)	/* mismatch or null char seen */
+	jnz	LABEL(less32bytes_sse4_2)/* mismatch or null char seen */
 	movdqa	(%rdi), %xmm3
 	UPDATE_STRNCMP_COUNTER
 
@@ -279,7 +279,7 @@ LABEL(loop_ashr_1_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -292,7 +292,7 @@ LABEL(loop_ashr_1_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_1_use_sse4_2)
@@ -318,7 +318,7 @@ LABEL(nibble_ashr_1_use_sse4_2):
  *        n(14~15)            n -14         1(15 +(n-14) - n)         ashr_2
  */
 	.p2align 4
-LABEL(ashr_2):
+LABEL(ashr_2_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -330,7 +330,7 @@ LABEL(ashr_2):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 	UPDATE_STRNCMP_COUNTER
 
@@ -358,7 +358,7 @@ LABEL(loop_ashr_2_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -371,7 +371,7 @@ LABEL(loop_ashr_2_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_2_use_sse4_2)
@@ -397,7 +397,7 @@ LABEL(nibble_ashr_2_use_sse4_2):
  *        n(13~15)            n -13         2(15 +(n-13) - n)         ashr_3
  */
 	.p2align 4
-LABEL(ashr_3):
+LABEL(ashr_3_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -409,7 +409,7 @@ LABEL(ashr_3):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -437,7 +437,7 @@ LABEL(loop_ashr_3_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -450,7 +450,7 @@ LABEL(loop_ashr_3_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_3_use_sse4_2)
@@ -476,7 +476,7 @@ LABEL(nibble_ashr_3_use_sse4_2):
  *        n(12~15)            n -12         3(15 +(n-12) - n)         ashr_4
  */
 	.p2align 4
-LABEL(ashr_4):
+LABEL(ashr_4_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -488,7 +488,7 @@ LABEL(ashr_4):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -517,7 +517,7 @@ LABEL(loop_ashr_4_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -530,7 +530,7 @@ LABEL(loop_ashr_4_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_4_use_sse4_2)
@@ -556,7 +556,7 @@ LABEL(nibble_ashr_4_use_sse4_2):
  *        n(11~15)          n - 11      	  4(15 +(n-11) - n)         ashr_5
  */
 	.p2align 4
-LABEL(ashr_5):
+LABEL(ashr_5_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -568,7 +568,7 @@ LABEL(ashr_5):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -597,7 +597,7 @@ LABEL(loop_ashr_5_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -611,7 +611,7 @@ LABEL(loop_ashr_5_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_5_use_sse4_2)
@@ -637,7 +637,7 @@ LABEL(nibble_ashr_5_use_sse4_2):
  *        n(10~15)          n - 10      	  5(15 +(n-10) - n)         ashr_6
  */
 	.p2align 4
-LABEL(ashr_6):
+LABEL(ashr_6_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -649,7 +649,7 @@ LABEL(ashr_6):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -678,7 +678,7 @@ LABEL(loop_ashr_6_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -691,7 +691,7 @@ LABEL(loop_ashr_6_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_6_use_sse4_2)
@@ -717,7 +717,7 @@ LABEL(nibble_ashr_6_use_sse4_2):
  *        n(9~15)          n - 9      	        6(15 +(n - 9) - n)         ashr_7
  */
 	.p2align 4
-LABEL(ashr_7):
+LABEL(ashr_7_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -729,7 +729,7 @@ LABEL(ashr_7):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -758,7 +758,7 @@ LABEL(loop_ashr_7_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -771,7 +771,7 @@ LABEL(loop_ashr_7_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_7_use_sse4_2)
@@ -797,7 +797,7 @@ LABEL(nibble_ashr_7_use_sse4_2):
  *        n(8~15)          n - 8      	        7(15 +(n - 8) - n)         ashr_8
  */
 	.p2align 4
-LABEL(ashr_8):
+LABEL(ashr_8_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -809,7 +809,7 @@ LABEL(ashr_8):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -838,7 +838,7 @@ LABEL(loop_ashr_8_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -851,7 +851,7 @@ LABEL(loop_ashr_8_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_8_use_sse4_2)
@@ -877,7 +877,7 @@ LABEL(nibble_ashr_8_use_sse4_2):
  *        n(7~15)          n - 7      	        8(15 +(n - 7) - n)         ashr_9
  */
 	.p2align 4
-LABEL(ashr_9):
+LABEL(ashr_9_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -889,7 +889,7 @@ LABEL(ashr_9):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -919,7 +919,7 @@ LABEL(loop_ashr_9_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -932,7 +932,7 @@ LABEL(loop_ashr_9_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_9_use_sse4_2)
@@ -958,7 +958,7 @@ LABEL(nibble_ashr_9_use_sse4_2):
  *        n(6~15)          n - 6      	        9(15 +(n - 6) - n)         ashr_10
  */
 	.p2align 4
-LABEL(ashr_10):
+LABEL(ashr_10_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -970,7 +970,7 @@ LABEL(ashr_10):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -999,7 +999,7 @@ LABEL(loop_ashr_10_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1012,7 +1012,7 @@ LABEL(loop_ashr_10_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_10_use_sse4_2)
@@ -1038,7 +1038,7 @@ LABEL(nibble_ashr_10_use_sse4_2):
  *        n(5~15)          n - 5      	        10(15 +(n - 5) - n)         ashr_11
  */
 	.p2align 4
-LABEL(ashr_11):
+LABEL(ashr_11_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1050,7 +1050,7 @@ LABEL(ashr_11):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1079,7 +1079,7 @@ LABEL(loop_ashr_11_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1092,7 +1092,7 @@ LABEL(loop_ashr_11_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_11_use_sse4_2)
@@ -1118,7 +1118,7 @@ LABEL(nibble_ashr_11_use_sse4_2):
  *        n(4~15)          n - 4      	        11(15 +(n - 4) - n)         ashr_12
  */
 	.p2align 4
-LABEL(ashr_12):
+LABEL(ashr_12_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1130,7 +1130,7 @@ LABEL(ashr_12):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1159,7 +1159,7 @@ LABEL(loop_ashr_12_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1172,7 +1172,7 @@ LABEL(loop_ashr_12_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_12_use_sse4_2)
@@ -1198,7 +1198,7 @@ LABEL(nibble_ashr_12_use_sse4_2):
  *        n(3~15)          n - 3      	        12(15 +(n - 3) - n)         ashr_13
  */
 	.p2align 4
-LABEL(ashr_13):
+LABEL(ashr_13_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1210,7 +1210,7 @@ LABEL(ashr_13):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1240,7 +1240,7 @@ LABEL(loop_ashr_13_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1253,7 +1253,7 @@ LABEL(loop_ashr_13_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_13_use_sse4_2)
@@ -1279,7 +1279,7 @@ LABEL(nibble_ashr_13_use_sse4_2):
  *        n(2~15)          n - 2      	        13(15 +(n - 2) - n)         ashr_14
  */
 	.p2align 4
-LABEL(ashr_14):
+LABEL(ashr_14_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1291,7 +1291,7 @@ LABEL(ashr_14):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1321,7 +1321,7 @@ LABEL(loop_ashr_14_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1334,7 +1334,7 @@ LABEL(loop_ashr_14_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_14_use_sse4_2)
@@ -1360,7 +1360,7 @@ LABEL(nibble_ashr_14_use_sse4_2):
  *        n(1~15)          n - 1      	        14(15 +(n - 1) - n)         ashr_15
  */
 	.p2align 4
-LABEL(ashr_15):
+LABEL(ashr_15_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1372,7 +1372,7 @@ LABEL(ashr_15):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 
 	movdqa	(%rdi), %xmm3
 
@@ -1404,7 +1404,7 @@ LABEL(loop_ashr_15_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1417,7 +1417,7 @@ LABEL(loop_ashr_15_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_15_use_sse4_2)
@@ -1439,56 +1439,37 @@ LABEL(nibble_ashr_use_sse4_2_exit):
 	pcmpistri      $0x1a,(%rsi,%rdx), %xmm0
 	.p2align 4
 LABEL(use_sse4_2_exit):
-	jnc	LABEL(strcmp_exitz)
+	jnc	LABEL(strcmp_exitz_sse4_2)
 #ifdef USE_AS_STRNCMP
 	sub	%rcx, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	%rcx, %rdx
 	lea	-16(%rdi, %r9), %rdi
 	movzbl	(%rdi, %rdx), %eax
 	movzbl	(%rsi, %rdx), %edx
 	test	%r8d, %r8d
-	jz	LABEL(use_sse4_2_ret)
+	jz	LABEL(use_sse4_2_ret_sse4_2)
 	xchg	%eax, %edx
-LABEL(use_sse4_2_ret):
+LABEL(use_sse4_2_ret_sse4_2):
 	sub	%edx, %eax
 	ret
 
-#if 0
-	/* This code was in the origial submission but isn't used.
-	   --drepper */
-	.p2align 4
-LABEL(aftertail):
-	pcmpeqb	%xmm3, %xmm1
-	psubb	%xmm0, %xmm1
-	pmovmskb %xmm1, %edx
-	not	%edx
-
-	.p2align 4
-LABEL(exit):
-	lea	-16(%r9, %rcx), %rax	/* locate the exact offset for rdi */
-#endif
-
-LABEL(less32bytes):
+LABEL(less32bytes_sse4_2):
 	lea	(%rdi, %rax), %rdi	/* locate the exact address for first operand(rdi) */
 	lea	(%rsi, %rcx), %rsi	/* locate the exact address for second operand(rsi) */
 	test	%r8d, %r8d
-	jz	LABEL(ret)
+	jz	LABEL(ret_sse4_2)
 	xchg	%rsi, %rdi		/* recover original order according to flag(%r8d) */
 
 	.p2align 4
-LABEL(ret):
-LABEL(less16bytes):
-	/*
-	 * Check to see if BSF is fast on this processor. If not, use a different
-	 * exit tail.
-	 */
+LABEL(ret_sse4_2):
+LABEL(less16bytes_sse4_2):
 	bsf	%rdx, %rdx		/* find and store bit index in %rdx */
 
 #ifdef USE_AS_STRNCMP
 	sub	%rdx, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	movzbl	(%rsi, %rdx), %ecx
 	movzbl	(%rdi, %rdx), %eax
@@ -1496,164 +1477,40 @@ LABEL(less16bytes):
 	sub	%ecx, %eax
 	ret
 
-LABEL(strcmp_exitz):
+LABEL(strcmp_exitz_sse4_2):
 	xor	%eax, %eax
 	ret
 
 	.p2align 4
-LABEL(Byte0):
-	/*
-	 * never need to handle byte 0 for strncmpy
-#ifdef USE_AS_STRNCMP
-	sub	$0, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	*/
+LABEL(Byte0_sse4_2):
 	movzx	(%rsi), %ecx
 	movzx	(%rdi), %eax
 
 	sub	%ecx, %eax
 	ret
-
-	.p2align 4
-LABEL(Byte1):
-
-#ifdef USE_AS_STRNCMP
-	sub	$1, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	1(%rsi), %ecx
-	movzx	1(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte2):
-
-#ifdef USE_AS_STRNCMP
-	sub	$2, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	2(%rsi), %ecx
-	movzx	2(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte3):
-
-#ifdef USE_AS_STRNCMP
-	sub	$3, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	3(%rsi), %ecx
-	movzx	3(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte4):
-
-#ifdef USE_AS_STRNCMP
-	sub	$4, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	4(%rsi), %ecx
-	movzx	4(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte5):
-
-#ifdef USE_AS_STRNCMP
-	sub	$5, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	5(%rsi), %ecx
-	movzx	5(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte6):
-
-#ifdef USE_AS_STRNCMP
-	sub	$6, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	6(%rsi), %ecx
-	movzx	6(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(next_8_bytes):
-	add	$8, %rdi
-	add	$8, %rsi
-#ifdef USE_AS_STRNCMP
-	sub	$8, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	test	$0x01, %dh
-	jnz	LABEL(Byte0)
-
-	test	$0x02, %dh
-	jnz	LABEL(Byte1)
-
-	test	$0x04, %dh
-	jnz	LABEL(Byte2)
-
-	test	$0x08, %dh
-	jnz	LABEL(Byte3)
-
-	test	$0x10, %dh
-	jnz	LABEL(Byte4)
-
-	test	$0x20, %dh
-	jnz	LABEL(Byte5)
-
-	test	$0x40, %dh
-	jnz	LABEL(Byte6)
-
-#ifdef USE_AS_STRNCMP
-	sub	$7, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	7(%rsi), %ecx
-	movzx	7(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
 	cfi_endproc
 	.size	STRCMP_SSE42, .-STRCMP_SSE42
 
 	/* Put all SSE 4.2 functions together.  */
 	.section .rodata.sse4.2,"a",@progbits
-	.p2align 4
-LABEL(unaligned_table):
-	.int	LABEL(ashr_1) - LABEL(unaligned_table)
-	.int	LABEL(ashr_2) - LABEL(unaligned_table)
-	.int	LABEL(ashr_3) - LABEL(unaligned_table)
-	.int	LABEL(ashr_4) - LABEL(unaligned_table)
-	.int	LABEL(ashr_5) - LABEL(unaligned_table)
-	.int	LABEL(ashr_6) - LABEL(unaligned_table)
-	.int	LABEL(ashr_7) - LABEL(unaligned_table)
-	.int	LABEL(ashr_8) - LABEL(unaligned_table)
-	.int	LABEL(ashr_9) - LABEL(unaligned_table)
-	.int	LABEL(ashr_10) - LABEL(unaligned_table)
-	.int	LABEL(ashr_11) - LABEL(unaligned_table)
-	.int	LABEL(ashr_12) - LABEL(unaligned_table)
-	.int	LABEL(ashr_13) - LABEL(unaligned_table)
-	.int	LABEL(ashr_14) - LABEL(unaligned_table)
-	.int	LABEL(ashr_15) - LABEL(unaligned_table)
-	.int	LABEL(ashr_0) - LABEL(unaligned_table)
+	.p2align 3
+LABEL(unaligned_table_sse4_2):
+	.int	LABEL(ashr_1_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_2_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_3_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_4_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_5_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_6_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_7_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_8_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_9_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_10_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_11_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_12_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_13_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_14_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_15_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_0_sse4_2) - LABEL(unaligned_table_sse4_2)
 
 
 # undef ENTRY
@@ -1673,6 +1530,4 @@ LABEL(unaligned_table):
 	.globl __GI_STRCMP; __GI_STRCMP = STRCMP_SSE2
 #endif
 
-#ifndef USE_AS_STRNCMP
 #include "../strcmp.S"
-#endif
diff --git a/sysdeps/x86_64/multiarch/strncmp-c.c b/sysdeps/x86_64/multiarch/strncmp-c.c
deleted file mode 100644
index d4f74a418d..0000000000
--- a/sysdeps/x86_64/multiarch/strncmp-c.c
+++ /dev/null
@@ -1,8 +0,0 @@
-#ifdef SHARED
-#define STRNCMP __strncmp_sse2
-#undef libc_hidden_builtin_def
-#define libc_hidden_builtin_def(name) \
-  __hidden_ver1 (__strncmp_sse2, __GI_strncmp, __strncmp_sse2);
-#endif
-
-#include "strncmp.c"
diff --git a/sysdeps/x86_64/multiarch/strcmp.S b/sysdeps/x86_64/strcmp.S
similarity index 57%
copy from sysdeps/x86_64/multiarch/strcmp.S
copy to sysdeps/x86_64/strcmp.S
index 15148e4f7f..340a64ba35 100644
--- a/sysdeps/x86_64/multiarch/strcmp.S
+++ b/sysdeps/x86_64/strcmp.S
@@ -1,7 +1,10 @@
-/* strcmp with SSE4.2
-   Copyright (C) 2009 Free Software Foundation, Inc.
-   Contributed by Intel Corporation.
+/* Highly optimized version for x86-64.
+   Copyright (C) 1999, 2000, 2002, 2003, 2005, 2009
+   Free Software Foundation, Inc.
    This file is part of the GNU C Library.
+   Based on i686 version contributed by Ulrich Drepper
+   <drepper@cygnus.com>, 1999.
+   Updated with SSE2 support contributed by Intel Corporation.
 
    The GNU C Library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
@@ -19,12 +22,20 @@
    02111-1307 USA.  */
 
 #include <sysdep.h>
-#include <ifunc-defines.h>
+#include "asm-syntax.h"
+#include "bp-sym.h"
+#include "bp-asm.h"
+
+#undef UPDATE_STRNCMP_COUNTER
+
+#ifndef LABEL
+#define LABEL(l) L(l)
+#endif
 
 #ifdef USE_AS_STRNCMP
 /* Since the counter, %r11, is unsigned, we branch to strcmp_exitz
    if the new counter > the old one or is 0.  */
-#define UPDATE_STRNCMP_COUNTER				\
+# define UPDATE_STRNCMP_COUNTER				\
 	/* calculate left number to compare */		\
 	lea	-16(%rcx, %r11), %r9;			\
 	cmp	%r9, %r11;				\
@@ -33,74 +44,34 @@
 	je	LABEL(strcmp_exitz);			\
 	mov	%r9, %r11
 
-#define STRCMP_SSE42	__strncmp_sse42
-#define STRCMP_SSE2	__strncmp_sse2
-#define __GI_STRCMP	__GI_strncmp
 #else
-#define UPDATE_STRNCMP_COUNTER
-#ifndef STRCMP
-#define STRCMP		strcmp
-#define STRCMP_SSE42	__strcmp_sse42
-#define STRCMP_SSE2	__strcmp_sse2
-#define __GI_STRCMP	__GI_strcmp
-#endif
+# define UPDATE_STRNCMP_COUNTER
+# ifndef STRCMP
+#  define STRCMP strcmp
+# endif
 #endif
 
-#ifndef LABEL
-#define LABEL(l) L(l)
-#endif
-
-/* Define multiple versions only for the definition in libc.  Don't
-   define multiple versions for strncmp in static library since we
-   need strncmp before the initialization happened.  */
-#if (defined SHARED || !defined USE_AS_STRNCMP) && !defined NOT_IN_libc
 	.text
-ENTRY(STRCMP)
-	.type	STRCMP, @gnu_indirect_function
-	cmpl	$0, __cpu_features+KIND_OFFSET(%rip)
-	jne	1f
-	call	__init_cpu_features
-1:	leaq	STRCMP_SSE2(%rip), %rax
-	testl	$(1<<20), __cpu_features+CPUID_OFFSET+COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET(%rip)
-	jz	2f
-	leaq	STRCMP_SSE42(%rip), %rax
-2:	ret
-END(STRCMP)
-
-/* We use 0x1a:
-	_SIDD_SBYTE_OPS
-	| _SIDD_CMP_EQUAL_EACH
-	| _SIDD_NEGATIVE_POLARITY
-	| _SIDD_LEAST_SIGNIFICANT
-   on pcmpistri to find out if two 16byte data elements are the same
-   and the offset of the first different byte.  There are 4 cases:
-
-   1. Both 16byte data elements are valid and identical.
-   2. Both 16byte data elements have EOS and identical.
-   3. Both 16byte data elements are valid and they differ at offset X.
-   4. At least one 16byte data element has EOS at offset X.  Two 16byte
-      data elements must differ at or before offset X.
-
-   Here is the table of ECX, CFlag, ZFlag and SFlag for 4 cases:
-
-   case		ECX	CFlag	ZFlag	SFlag
-    1		16	  0	  0	  0
-    2		16	  0	  1	  1
-    3		 X	  1	  0	  0
-    4	       0 <= X	  1	 0/1	 0/1
-
-   We exit from the loop for cases 2, 3 and 4 with jbe which branches
-   when either CFlag or ZFlag is 1.  If CFlag == 0, we return 0 for
-   case 2.  */
-
-	/* Put all SSE 4.2 functions together.  */
-	.section .text.sse4.2,"ax",@progbits
-	.align 	16
-	.type	STRCMP_SSE42, @function
-STRCMP_SSE42:
-	cfi_startproc
-	CALL_MCOUNT
+ENTRY (BP_SYM (STRCMP))
+#ifdef NOT_IN_libc
+/* Simple version since we can't use SSE registers in ld.so.  */
+L(oop):	movb	(%rdi), %al
+	cmpb	(%rsi), %al
+	jne	L(neq)
+	incq	%rdi
+	incq	%rsi
+	testb	%al, %al
+	jnz	L(oop)
+
+	xorl	%eax, %eax
+	ret
 
+L(neq):	movl	$1, %eax
+	movl	$-1, %ecx
+	cmovbl	%ecx, %eax
+	ret
+END (BP_SYM (STRCMP))
+#else	/* NOT_IN_libc */
 /*
  * This implementation uses SSE to compare up to 16 bytes at a time.
  */
@@ -120,8 +91,10 @@ STRCMP_SSE42:
 	ja	LABEL(crosscache)	/* rsi: 16-byte load will cross cache line */
 	cmp	$0x30, %eax
 	ja	LABEL(crosscache)	/* rdi: 16-byte load will cross cache line */
-	movdqu	(%rdi), %xmm1
-	movdqu	(%rsi), %xmm2
+	movlpd	(%rdi), %xmm1
+	movlpd	(%rsi), %xmm2
+	movhpd	8(%rdi), %xmm1
+	movhpd	8(%rsi), %xmm2
 	pxor	%xmm0, %xmm0		/* clear %xmm0 for null char checks */
 	pcmpeqb	%xmm1, %xmm0		/* Any null chars? */
 	pcmpeqb	%xmm2, %xmm1		/* compare first 16 bytes for equality */
@@ -194,44 +167,38 @@ LABEL(ashr_0):
 	 * Now both strings are aligned at 16-byte boundary. Loop over strings
 	 * checking 32-bytes per iteration.
 	 */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 	.p2align 4
-LABEL(ashr_0_use_sse4_2):
-	movdqa	(%rdi,%rdx), %xmm0
-	pcmpistri      $0x1a,(%rsi,%rdx), %xmm0
-	lea	16(%rdx), %rdx
-	jbe	LABEL(ashr_0_use_sse4_2_exit)
-#ifdef USE_AS_STRNCMP
-	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
+LABEL(loop_ashr_0):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)		/* mismatch or null char seen */
 
-	movdqa	(%rdi,%rdx), %xmm0
-	pcmpistri      $0x1a,(%rsi,%rdx), %xmm0
-	lea	16(%rdx), %rdx
-	jbe	LABEL(ashr_0_use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	jmp	LABEL(ashr_0_use_sse4_2)
-
+	add	$16, %rcx
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
 
-	.p2align 4
-LABEL(ashr_0_use_sse4_2_exit):
-	jnc	LABEL(strcmp_exitz)
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 #ifdef USE_AS_STRNCMP
-	sub	%rcx, %r11
+	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	lea	-16(%rdx, %rcx), %rcx
-	movzbl	(%rdi, %rcx), %eax
-	movzbl	(%rsi, %rcx), %edx
-	sub	%edx, %eax
-	ret
-
-
-
+	add	$16, %rcx
+	jmp	LABEL(loop_ashr_0)
 
 /*
  * The following cases will be handled by ashr_1
@@ -266,51 +233,91 @@ LABEL(ashr_1):
 	lea	1(%rdi), %r10
 	and	$0xfff, %r10		/* offset into 4K page */
 	sub	$0x1000, %r10		/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_1_use_sse4_2):
+LABEL(loop_ashr_1):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_1_use_sse4_2)
+	jg	LABEL(nibble_ashr_1)	/* cross page boundary */
+
+LABEL(gobble_ashr_1):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4		 /* store for next cycle */
+
+	psrldq	$1, %xmm3
+	pslldq	$15, %xmm2
+	por	%xmm3, %xmm2		/* merge into one 16byte value */
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $1, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
 
-	add	$16, %rdx
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_1_use_sse4_2)
+	jg	LABEL(nibble_ashr_1)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4		/* store for next cycle */
+
+	psrldq	$1, %xmm3
+	pslldq 	$15, %xmm2
+	por	%xmm3, %xmm2		/* merge into one 16byte value */
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $1, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_1_use_sse4_2)
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_1)
 
+	/*
+	 * Nibble avoids loads across page boundary. This is to avoid a potential
+	 * access into unmapped memory.
+	 */
 	.p2align 4
-LABEL(nibble_ashr_1_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$1, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_1):
+	pcmpeqb	%xmm3, %xmm0		 /* check nibble for null char*/
+	pmovmskb %xmm0, %edx
+	test	$0xfffe, %edx
+	jnz	LABEL(ashr_1_exittail)	/* find null char*/
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$14, %r11
+	jbe	LABEL(ashr_1_exittail)
 #endif
-	cmp	$14, %ecx
-	ja	LABEL(loop_ashr_1_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10		/* substract 4K from %r10 */
+	jmp	LABEL(gobble_ashr_1)
+
+	/*
+	 * Once find null char, determine if there is a string mismatch
+	 * before the null char.
+	 */
+	.p2align 4
+LABEL(ashr_1_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$1, %xmm0
+	psrldq	$1, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  * The following cases will be handled by ashr_2
@@ -345,51 +352,85 @@ LABEL(ashr_2):
 	lea	2(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_2_use_sse4_2):
+LABEL(loop_ashr_2):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_2_use_sse4_2)
+	jg	LABEL(nibble_ashr_2)
+
+LABEL(gobble_ashr_2):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$2, %xmm3
+	pslldq	$14, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $2, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_2_use_sse4_2)
+	jg	LABEL(nibble_ashr_2)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$2, %xmm3
+	pslldq 	$14, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $2, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_2_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_2)
 
 	.p2align 4
-LABEL(nibble_ashr_2_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$2, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_2):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfffc, %edx
+	jnz	LABEL(ashr_2_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$13, %r11
+	jbe	LABEL(ashr_2_exittail)
 #endif
-	cmp	$13, %ecx
-	ja	LABEL(loop_ashr_2_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_2)
+
+	.p2align 4
+LABEL(ashr_2_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$2, %xmm0
+	psrldq	$2, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  * The following cases will be handled by ashr_3
@@ -425,50 +466,85 @@ LABEL(ashr_3):
 	lea	3(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
-LABEL(loop_ashr_3_use_sse4_2):
+	.p2align 4
+LABEL(loop_ashr_3):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_3_use_sse4_2)
+	jg	LABEL(nibble_ashr_3)
+
+LABEL(gobble_ashr_3):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$3, %xmm3
+	pslldq	$13, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $3, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_3_use_sse4_2)
+	jg	LABEL(nibble_ashr_3)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$3, %xmm3
+	pslldq 	$13, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $3, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_3_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_3)
 
 	.p2align 4
-LABEL(nibble_ashr_3_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$3, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_3):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfff8, %edx
+	jnz	LABEL(ashr_3_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$12, %r11
+	jbe	LABEL(ashr_3_exittail)
 #endif
-	cmp	$12, %ecx
-	ja	LABEL(loop_ashr_3_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_3)
+
+	.p2align 4
+LABEL(ashr_3_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$3, %xmm0
+	psrldq	$3, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  * The following cases will be handled by ashr_4
@@ -504,51 +580,85 @@ LABEL(ashr_4):
 	lea	4(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_4_use_sse4_2):
+LABEL(loop_ashr_4):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_4_use_sse4_2)
+	jg	LABEL(nibble_ashr_4)
+
+LABEL(gobble_ashr_4):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$4, %xmm3
+	pslldq	$12, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $4, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_4_use_sse4_2)
+	jg	LABEL(nibble_ashr_4)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$4, %xmm3
+	pslldq 	$12, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $4, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_4_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_4)
 
 	.p2align 4
-LABEL(nibble_ashr_4_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$4, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_4):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfff0, %edx
+	jnz	LABEL(ashr_4_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$11, %r11
+	jbe	LABEL(ashr_4_exittail)
 #endif
-	cmp	$11, %ecx
-	ja	LABEL(loop_ashr_4_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_4)
+
+	.p2align 4
+LABEL(ashr_4_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$4, %xmm0
+	psrldq	$4, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  * The following cases will be handled by ashr_5
@@ -584,52 +694,85 @@ LABEL(ashr_5):
 	lea	5(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_5_use_sse4_2):
+LABEL(loop_ashr_5):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_5_use_sse4_2)
+	jg	LABEL(nibble_ashr_5)
+
+LABEL(gobble_ashr_5):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$5, %xmm3
+	pslldq	$11, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $5, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_5_use_sse4_2)
+	jg	LABEL(nibble_ashr_5)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
 
-	movdqa	(%rdi, %rdx), %xmm0
+	psrldq	$5, %xmm3
+	pslldq 	$11, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	palignr $5, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_5_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_5)
 
 	.p2align 4
-LABEL(nibble_ashr_5_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$5, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_5):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xffe0, %edx
+	jnz	LABEL(ashr_5_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$10, %r11
+	jbe	LABEL(ashr_5_exittail)
 #endif
-	cmp	$10, %ecx
-	ja	LABEL(loop_ashr_5_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_5)
+
+	.p2align 4
+LABEL(ashr_5_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$5, %xmm0
+	psrldq	$5, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  * The following cases will be handled by ashr_6
@@ -665,51 +808,85 @@ LABEL(ashr_6):
 	lea	6(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_6_use_sse4_2):
+LABEL(loop_ashr_6):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_6_use_sse4_2)
+	jg	LABEL(nibble_ashr_6)
+
+LABEL(gobble_ashr_6):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$6, %xmm3
+	pslldq	$10, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $6, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_6_use_sse4_2)
+	jg	LABEL(nibble_ashr_6)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$6, %xmm3
+	pslldq 	$10, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $6, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_6_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_6)
 
 	.p2align 4
-LABEL(nibble_ashr_6_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$6, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_6):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xffc0, %edx
+	jnz	LABEL(ashr_6_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$9, %r11
+	jbe	LABEL(ashr_6_exittail)
 #endif
-	cmp	$9, %ecx
-	ja	LABEL(loop_ashr_6_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_6)
+
+	.p2align 4
+LABEL(ashr_6_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$6, %xmm0
+	psrldq	$6, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  * The following cases will be handled by ashr_7
@@ -745,51 +922,85 @@ LABEL(ashr_7):
 	lea	7(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_7_use_sse4_2):
+LABEL(loop_ashr_7):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_7_use_sse4_2)
+	jg	LABEL(nibble_ashr_7)
+
+LABEL(gobble_ashr_7):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$7, %xmm3
+	pslldq	$9, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $7, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_7_use_sse4_2)
+	jg	LABEL(nibble_ashr_7)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$7, %xmm3
+	pslldq 	$9, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $7, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_7_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_7)
 
 	.p2align 4
-LABEL(nibble_ashr_7_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$7, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_7):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xff80, %edx
+	jnz	LABEL(ashr_7_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$8, %r11
+	jbe	LABEL(ashr_7_exittail)
 #endif
-	cmp	$8, %ecx
-	ja	LABEL(loop_ashr_7_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_7)
+
+	.p2align 4
+LABEL(ashr_7_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$7, %xmm0
+	psrldq	$7, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_8
@@ -825,51 +1036,85 @@ LABEL(ashr_8):
 	lea	8(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_8_use_sse4_2):
+LABEL(loop_ashr_8):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_8_use_sse4_2)
+	jg	LABEL(nibble_ashr_8)
+
+LABEL(gobble_ashr_8):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$8, %xmm3
+	pslldq	$8, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $8, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_8_use_sse4_2)
+	jg	LABEL(nibble_ashr_8)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$8, %xmm3
+	pslldq 	$8, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $8, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_8_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_8)
 
 	.p2align 4
-LABEL(nibble_ashr_8_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$8, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_8):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xff00, %edx
+	jnz	LABEL(ashr_8_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$7, %r11
+	jbe	LABEL(ashr_8_exittail)
 #endif
-	cmp	$7, %ecx
-	ja	LABEL(loop_ashr_8_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_8)
+
+	.p2align 4
+LABEL(ashr_8_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$8, %xmm0
+	psrldq	$8, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_9
@@ -905,52 +1150,85 @@ LABEL(ashr_9):
 	lea	9(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_9_use_sse4_2):
+LABEL(loop_ashr_9):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_9_use_sse4_2)
+	jg	LABEL(nibble_ashr_9)
 
-	movdqa	(%rdi, %rdx), %xmm0
+LABEL(gobble_ashr_9):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$9, %xmm3
+	pslldq	$7, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	palignr $9, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_9_use_sse4_2)
+	jg	LABEL(nibble_ashr_9)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$9, %xmm3
+	pslldq 	$7, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $9, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_9_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3		/* store for next cycle */
+	jmp	LABEL(loop_ashr_9)
 
 	.p2align 4
-LABEL(nibble_ashr_9_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$9, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_9):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfe00, %edx
+	jnz	LABEL(ashr_9_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$6, %r11
+	jbe	LABEL(ashr_9_exittail)
 #endif
-	cmp	$6, %ecx
-	ja	LABEL(loop_ashr_9_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_9)
+
+	.p2align 4
+LABEL(ashr_9_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$9, %xmm0
+	psrldq	$9, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_10
@@ -986,51 +1264,85 @@ LABEL(ashr_10):
 	lea	10(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_10_use_sse4_2):
+LABEL(loop_ashr_10):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_10_use_sse4_2)
+	jg	LABEL(nibble_ashr_10)
+
+LABEL(gobble_ashr_10):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$10, %xmm3
+	pslldq	$6, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $10, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_10_use_sse4_2)
+	jg	LABEL(nibble_ashr_10)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$10, %xmm3
+	pslldq 	$6, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $10, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_10_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_10)
 
 	.p2align 4
-LABEL(nibble_ashr_10_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$10, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_10):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfc00, %edx
+	jnz	LABEL(ashr_10_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$5, %r11
+	jbe	LABEL(ashr_10_exittail)
 #endif
-	cmp	$5, %ecx
-	ja	LABEL(loop_ashr_10_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_10)
+
+	.p2align 4
+LABEL(ashr_10_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$10, %xmm0
+	psrldq	$10, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_11
@@ -1066,51 +1378,85 @@ LABEL(ashr_11):
 	lea	11(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_11_use_sse4_2):
+LABEL(loop_ashr_11):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_11_use_sse4_2)
+	jg	LABEL(nibble_ashr_11)
+
+LABEL(gobble_ashr_11):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$11, %xmm3
+	pslldq	$5, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $11, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_11_use_sse4_2)
+	jg	LABEL(nibble_ashr_11)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$11, %xmm3
+	pslldq 	$5, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $11, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_11_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_11)
 
 	.p2align 4
-LABEL(nibble_ashr_11_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$11, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_11):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xf800, %edx
+	jnz	LABEL(ashr_11_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$4, %r11
+	jbe	LABEL(ashr_11_exittail)
 #endif
-	cmp	$4, %ecx
-	ja	LABEL(loop_ashr_11_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_11)
+
+	.p2align 4
+LABEL(ashr_11_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$11, %xmm0
+	psrldq	$11, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_12
@@ -1146,51 +1492,85 @@ LABEL(ashr_12):
 	lea	12(%rdi), %r10
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
 
 	.p2align 4
-LABEL(loop_ashr_12_use_sse4_2):
+LABEL(loop_ashr_12):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_12_use_sse4_2)
+	jg	LABEL(nibble_ashr_12)
+
+LABEL(gobble_ashr_12):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$12, %xmm3
+	pslldq	$4, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $12, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_12_use_sse4_2)
+	jg	LABEL(nibble_ashr_12)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$12, %xmm3
+	pslldq 	$4, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $12, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_12_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_12)
 
 	.p2align 4
-LABEL(nibble_ashr_12_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$12, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_12):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xf000, %edx
+	jnz	LABEL(ashr_12_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$3, %r11
+	jbe	LABEL(ashr_12_exittail)
 #endif
-	cmp	$3, %ecx
-	ja	LABEL(loop_ashr_12_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_12)
+
+	.p2align 4
+LABEL(ashr_12_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$12, %xmm0
+	psrldq	$12, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_13
@@ -1227,51 +1607,84 @@ LABEL(ashr_13):
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
 
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
-
 	.p2align 4
-LABEL(loop_ashr_13_use_sse4_2):
+LABEL(loop_ashr_13):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_13_use_sse4_2)
+	jg	LABEL(nibble_ashr_13)
+
+LABEL(gobble_ashr_13):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$13, %xmm3
+	pslldq	$3, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $13, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_13_use_sse4_2)
+	jg	LABEL(nibble_ashr_13)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$13, %xmm3
+	pslldq 	$3, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $13, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_13_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_13)
 
 	.p2align 4
-LABEL(nibble_ashr_13_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$13, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_13):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xe000, %edx
+	jnz	LABEL(ashr_13_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$2, %r11
+	jbe	LABEL(ashr_13_exittail)
 #endif
-	cmp	$2, %ecx
-	ja	LABEL(loop_ashr_13_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_13)
+
+	.p2align 4
+LABEL(ashr_13_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq  $13, %xmm0
+	psrldq  $13, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_14
@@ -1308,51 +1721,84 @@ LABEL(ashr_14):
 	and	$0xfff, %r10	/* offset into 4K page */
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
 
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
-
 	.p2align 4
-LABEL(loop_ashr_14_use_sse4_2):
+LABEL(loop_ashr_14):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_14_use_sse4_2)
+	jg	LABEL(nibble_ashr_14)
+
+LABEL(gobble_ashr_14):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$14, %xmm3
+	pslldq	$2, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $14, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_14_use_sse4_2)
+	jg	LABEL(nibble_ashr_14)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$14, %xmm3
+	pslldq 	$2, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $14, -16(%rdi, %rdx), %xmm0
-	pcmpistri	$0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_14_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_14)
 
 	.p2align 4
-LABEL(nibble_ashr_14_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$14, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_14):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xc000, %edx
+	jnz	LABEL(ashr_14_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	cmp	$1, %r11
+	jbe	LABEL(ashr_14_exittail)
 #endif
-	cmp	$1, %ecx
-	ja	LABEL(loop_ashr_14_use_sse4_2)
 
-	jmp	LABEL(nibble_ashr_use_sse4_2_exit)
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_14)
+
+	.p2align 4
+LABEL(ashr_14_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$14, %xmm0
+	psrldq	$14, %xmm3
+	jmp	LABEL(aftertail)
 
 /*
  *  The following cases will be handled by ashr_15
@@ -1391,73 +1837,84 @@ LABEL(ashr_15):
 
 	sub	$0x1000, %r10	/* subtract 4K pagesize */
 
-	mov	%rcx, %rdx	/* only for offset of sse4 instruction loop*/
-
 	.p2align 4
-LABEL(loop_ashr_15_use_sse4_2):
+LABEL(loop_ashr_15):
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_15_use_sse4_2)
+	jg	LABEL(nibble_ashr_15)
+
+LABEL(gobble_ashr_15):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$15, %xmm3
+	pslldq	$1, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $15, -16(%rdi, %rdx), %xmm0
-	pcmpistri      $0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
 
-	add	$16, %rdx
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
 	add	$16, %r10
-	jg	LABEL(nibble_ashr_15_use_sse4_2)
+	jg	LABEL(nibble_ashr_15)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$15, %xmm3
+	pslldq 	$1, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
 
-	movdqa	(%rdi, %rdx), %xmm0
-	palignr $15, -16(%rdi, %rdx), %xmm0
-	pcmpistri      $0x1a,(%rsi,%rdx), %xmm0
-	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
 	jbe	LABEL(strcmp_exitz)
 #endif
-	add	$16, %rdx
-	jmp	LABEL(loop_ashr_15_use_sse4_2)
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_15)
 
 	.p2align 4
-LABEL(nibble_ashr_15_use_sse4_2):
-	sub	$0x1000, %r10
-	movdqa	-16(%rdi, %rdx), %xmm0
-	psrldq	$15, %xmm0
-	pcmpistri      $0x3a,%xmm0, %xmm0
+LABEL(nibble_ashr_15):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0x8000, %edx
+	jnz	LABEL(ashr_15_exittail)
+
 #ifdef USE_AS_STRNCMP
-	cmp	%r11, %rcx
-	jae	LABEL(nibble_ashr_use_sse4_2_exit)
+	test	%r11, %r11
+	je	LABEL(ashr_15_exittail)
 #endif
-	cmp	$0, %ecx
-	ja	LABEL(loop_ashr_15_use_sse4_2)
 
-LABEL(nibble_ashr_use_sse4_2_exit):
-	pcmpistri      $0x1a,(%rsi,%rdx), %xmm0
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_15)
+
 	.p2align 4
-LABEL(use_sse4_2_exit):
-	jnc	LABEL(strcmp_exitz)
-#ifdef USE_AS_STRNCMP
-	sub	%rcx, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	add	%rcx, %rdx
-	lea	-16(%rdi, %r9), %rdi
-	movzbl	(%rdi, %rdx), %eax
-	movzbl	(%rsi, %rdx), %edx
-	test	%r8d, %r8d
-	jz	LABEL(use_sse4_2_ret)
-	xchg	%eax, %edx
-LABEL(use_sse4_2_ret):
-	sub	%edx, %eax
-	ret
+LABEL(ashr_15_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$15, %xmm3
+	psrldq	$15, %xmm0
 
-#if 0
-	/* This code was in the origial submission but isn't used.
-	   --drepper */
 	.p2align 4
 LABEL(aftertail):
 	pcmpeqb	%xmm3, %xmm1
@@ -1468,8 +1925,6 @@ LABEL(aftertail):
 	.p2align 4
 LABEL(exit):
 	lea	-16(%r9, %rcx), %rax	/* locate the exact offset for rdi */
-#endif
-
 LABEL(less32bytes):
 	lea	(%rdi, %rax), %rdi	/* locate the exact address for first operand(rdi) */
 	lea	(%rsi, %rcx), %rsi	/* locate the exact address for second operand(rsi) */
@@ -1480,10 +1935,6 @@ LABEL(less32bytes):
 	.p2align 4
 LABEL(ret):
 LABEL(less16bytes):
-	/*
-	 * Check to see if BSF is fast on this processor. If not, use a different
-	 * exit tail.
-	 */
 	bsf	%rdx, %rdx		/* find and store bit index in %rdx */
 
 #ifdef USE_AS_STRNCMP
@@ -1502,141 +1953,15 @@ LABEL(strcmp_exitz):
 
 	.p2align 4
 LABEL(Byte0):
-	/*
-	 * never need to handle byte 0 for strncmpy
-#ifdef USE_AS_STRNCMP
-	sub	$0, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	*/
 	movzx	(%rsi), %ecx
 	movzx	(%rdi), %eax
 
 	sub	%ecx, %eax
 	ret
+END (BP_SYM (STRCMP))
 
-	.p2align 4
-LABEL(Byte1):
-
-#ifdef USE_AS_STRNCMP
-	sub	$1, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	1(%rsi), %ecx
-	movzx	1(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte2):
-
-#ifdef USE_AS_STRNCMP
-	sub	$2, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	2(%rsi), %ecx
-	movzx	2(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte3):
-
-#ifdef USE_AS_STRNCMP
-	sub	$3, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	3(%rsi), %ecx
-	movzx	3(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte4):
-
-#ifdef USE_AS_STRNCMP
-	sub	$4, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	4(%rsi), %ecx
-	movzx	4(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte5):
-
-#ifdef USE_AS_STRNCMP
-	sub	$5, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	5(%rsi), %ecx
-	movzx	5(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte6):
-
-#ifdef USE_AS_STRNCMP
-	sub	$6, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	6(%rsi), %ecx
-	movzx	6(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(next_8_bytes):
-	add	$8, %rdi
-	add	$8, %rsi
-#ifdef USE_AS_STRNCMP
-	sub	$8, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	test	$0x01, %dh
-	jnz	LABEL(Byte0)
-
-	test	$0x02, %dh
-	jnz	LABEL(Byte1)
-
-	test	$0x04, %dh
-	jnz	LABEL(Byte2)
-
-	test	$0x08, %dh
-	jnz	LABEL(Byte3)
-
-	test	$0x10, %dh
-	jnz	LABEL(Byte4)
-
-	test	$0x20, %dh
-	jnz	LABEL(Byte5)
-
-	test	$0x40, %dh
-	jnz	LABEL(Byte6)
-
-#ifdef USE_AS_STRNCMP
-	sub	$7, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	7(%rsi), %ecx
-	movzx	7(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-	cfi_endproc
-	.size	STRCMP_SSE42, .-STRCMP_SSE42
-
-	/* Put all SSE 4.2 functions together.  */
-	.section .rodata.sse4.2,"a",@progbits
-	.p2align 4
+	.section .rodata,"a",@progbits
+	.p2align 3
 LABEL(unaligned_table):
 	.int	LABEL(ashr_1) - LABEL(unaligned_table)
 	.int	LABEL(ashr_2) - LABEL(unaligned_table)
@@ -1654,25 +1979,5 @@ LABEL(unaligned_table):
 	.int	LABEL(ashr_14) - LABEL(unaligned_table)
 	.int	LABEL(ashr_15) - LABEL(unaligned_table)
 	.int	LABEL(ashr_0) - LABEL(unaligned_table)
-
-
-# undef ENTRY
-# define ENTRY(name) \
-	.type STRCMP_SSE2, @function; \
-	.align 16; \
-	STRCMP_SSE2: cfi_startproc; \
-	CALL_MCOUNT
-# undef END
-# define END(name) \
-	cfi_endproc; .size STRCMP_SSE2, .-STRCMP_SSE2
-# undef libc_hidden_builtin_def
-/* It doesn't make sense to send libc-internal strcmp 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_STRCMP; __GI_STRCMP = STRCMP_SSE2
-#endif
-
-#ifndef USE_AS_STRNCMP
-#include "../strcmp.S"
-#endif
+#endif /* NOT_IN_libc */
+libc_hidden_builtin_def (STRCMP)
diff --git a/sysdeps/x86_64/strncmp.S b/sysdeps/x86_64/strncmp.S
new file mode 100644
index 0000000000..0af34e7f15
--- /dev/null
+++ b/sysdeps/x86_64/strncmp.S
@@ -0,0 +1,3 @@
+#define STRCMP strncmp
+#define USE_AS_STRNCMP
+#include "strcmp.S"
-- 
2.11.4.GIT