diff mbox series

[v2,3/5] x86_64: Add sse4_1 optimized bcmp implementation in memcmp-sse4.S

Message ID 20210914063039.1126196-3-goldstein.w.n@gmail.com
State New
Headers show
Series [v2,1/5] x86_64: Add support for bcmp using sse2, sse4_1, avx2, and evex | expand

Commit Message

Noah Goldstein Sept. 14, 2021, 6:30 a.m. UTC 
No bug. This commit does not modify any of the memcmp
implementation. It just adds bcmp ifdefs to skip obvious cases
where computing the proper 1/-1 required by memcmp is not needed.

test-memcmp, test-bcmp, and test-wmemcmp are all passing.
---
 sysdeps/x86_64/multiarch/memcmp-sse4.S | 761 ++++++++++++++++++++++++-
 1 file changed, 746 insertions(+), 15 deletions(-)
diff mbox series

Patch

diff --git a/sysdeps/x86_64/multiarch/memcmp-sse4.S b/sysdeps/x86_64/multiarch/memcmp-sse4.S
index b82adcd5fa..b9528ed58e 100644
--- a/sysdeps/x86_64/multiarch/memcmp-sse4.S
+++ b/sysdeps/x86_64/multiarch/memcmp-sse4.S
@@ -72,7 +72,11 @@  L(79bytesormore):
 	movdqu	(%rdi), %xmm2
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 	mov	%rsi, %rcx
 	and	$-16, %rsi
 	add	$16, %rsi
@@ -91,34 +95,58 @@  L(less128bytes):
 	movdqu	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqu	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 
 	movdqu	32(%rdi), %xmm2
 	pxor	32(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(48bytesin256)
+# endif
 
 	movdqu	48(%rdi), %xmm2
 	pxor	48(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesin256)
+# endif
 	cmp	$32, %rdx
 	jb	L(less32bytesin64)
 
 	movdqu	64(%rdi), %xmm2
 	pxor	64(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(80bytesin256)
+# endif
 
 	movdqu	80(%rdi), %xmm2
 	pxor	80(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(96bytesin256)
+# endif
 	sub	$32, %rdx
 	add	$32, %rdi
 	add	$32, %rsi
@@ -140,42 +168,74 @@  L(less256bytes):
 	movdqu	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqu	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 
 	movdqu	32(%rdi), %xmm2
 	pxor	32(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(48bytesin256)
+# endif
 
 	movdqu	48(%rdi), %xmm2
 	pxor	48(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesin256)
+# endif
 
 	movdqu	64(%rdi), %xmm2
 	pxor	64(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(80bytesin256)
+# endif
 
 	movdqu	80(%rdi), %xmm2
 	pxor	80(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(96bytesin256)
+# endif
 
 	movdqu	96(%rdi), %xmm2
 	pxor	96(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(112bytesin256)
+# endif
 
 	movdqu	112(%rdi), %xmm2
 	pxor	112(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(128bytesin256)
+# endif
 
 	add	$128, %rsi
 	add	$128, %rdi
@@ -189,12 +249,20 @@  L(less256bytes):
 	movdqu	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqu	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 	sub	$32, %rdx
 	add	$32, %rdi
 	add	$32, %rsi
@@ -208,82 +276,146 @@  L(less512bytes):
 	movdqu	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqu	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 
 	movdqu	32(%rdi), %xmm2
 	pxor	32(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(48bytesin256)
+# endif
 
 	movdqu	48(%rdi), %xmm2
 	pxor	48(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesin256)
+# endif
 
 	movdqu	64(%rdi), %xmm2
 	pxor	64(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(80bytesin256)
+# endif
 
 	movdqu	80(%rdi), %xmm2
 	pxor	80(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(96bytesin256)
+# endif
 
 	movdqu	96(%rdi), %xmm2
 	pxor	96(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(112bytesin256)
+# endif
 
 	movdqu	112(%rdi), %xmm2
 	pxor	112(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(128bytesin256)
+# endif
 
 	movdqu	128(%rdi), %xmm2
 	pxor	128(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(144bytesin256)
+# endif
 
 	movdqu	144(%rdi), %xmm2
 	pxor	144(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(160bytesin256)
+# endif
 
 	movdqu	160(%rdi), %xmm2
 	pxor	160(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(176bytesin256)
+# endif
 
 	movdqu	176(%rdi), %xmm2
 	pxor	176(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(192bytesin256)
+# endif
 
 	movdqu	192(%rdi), %xmm2
 	pxor	192(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(208bytesin256)
+# endif
 
 	movdqu	208(%rdi), %xmm2
 	pxor	208(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(224bytesin256)
+# endif
 
 	movdqu	224(%rdi), %xmm2
 	pxor	224(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(240bytesin256)
+# endif
 
 	movdqu	240(%rdi), %xmm2
 	pxor	240(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(256bytesin256)
+# endif
 
 	add	$256, %rsi
 	add	$256, %rdi
@@ -300,12 +432,20 @@  L(less512bytes):
 	movdqu	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqu	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 	sub	$32, %rdx
 	add	$32, %rdi
 	add	$32, %rsi
@@ -346,7 +486,11 @@  L(64bytesormore_loop):
 	por	%xmm5, %xmm1
 
 	ptest	%xmm1, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesormore_loop_end)
+# endif
 	add	$64, %rsi
 	add	$64, %rdi
 	sub	$64, %rdx
@@ -380,7 +524,11 @@  L(L2_L3_unaligned_128bytes_loop):
 	por	%xmm5, %xmm1
 
 	ptest	%xmm1, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesormore_loop_end)
+# endif
 	add	$64, %rsi
 	add	$64, %rdi
 	sub	$64, %rdx
@@ -404,34 +552,58 @@  L(less128bytesin2aligned):
 	movdqa	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqa	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 
 	movdqa	32(%rdi), %xmm2
 	pxor	32(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(48bytesin256)
+# endif
 
 	movdqa	48(%rdi), %xmm2
 	pxor	48(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesin256)
+# endif
 	cmp	$32, %rdx
 	jb	L(less32bytesin64in2alinged)
 
 	movdqa	64(%rdi), %xmm2
 	pxor	64(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(80bytesin256)
+# endif
 
 	movdqa	80(%rdi), %xmm2
 	pxor	80(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(96bytesin256)
+# endif
 	sub	$32, %rdx
 	add	$32, %rdi
 	add	$32, %rsi
@@ -454,42 +626,74 @@  L(less256bytesin2alinged):
 	movdqa	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqa	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 
 	movdqa	32(%rdi), %xmm2
 	pxor	32(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(48bytesin256)
+# endif
 
 	movdqa	48(%rdi), %xmm2
 	pxor	48(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesin256)
+# endif
 
 	movdqa	64(%rdi), %xmm2
 	pxor	64(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(80bytesin256)
+# endif
 
 	movdqa	80(%rdi), %xmm2
 	pxor	80(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(96bytesin256)
+# endif
 
 	movdqa	96(%rdi), %xmm2
 	pxor	96(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(112bytesin256)
+# endif
 
 	movdqa	112(%rdi), %xmm2
 	pxor	112(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(128bytesin256)
+# endif
 
 	add	$128, %rsi
 	add	$128, %rdi
@@ -503,12 +707,20 @@  L(less256bytesin2alinged):
 	movdqu	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqu	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 	sub	$32, %rdx
 	add	$32, %rdi
 	add	$32, %rsi
@@ -524,82 +736,146 @@  L(256bytesormorein2aligned):
 	movdqa	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqa	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 
 	movdqa	32(%rdi), %xmm2
 	pxor	32(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(48bytesin256)
+# endif
 
 	movdqa	48(%rdi), %xmm2
 	pxor	48(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesin256)
+# endif
 
 	movdqa	64(%rdi), %xmm2
 	pxor	64(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(80bytesin256)
+# endif
 
 	movdqa	80(%rdi), %xmm2
 	pxor	80(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(96bytesin256)
+# endif
 
 	movdqa	96(%rdi), %xmm2
 	pxor	96(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(112bytesin256)
+# endif
 
 	movdqa	112(%rdi), %xmm2
 	pxor	112(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(128bytesin256)
+# endif
 
 	movdqa	128(%rdi), %xmm2
 	pxor	128(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(144bytesin256)
+# endif
 
 	movdqa	144(%rdi), %xmm2
 	pxor	144(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(160bytesin256)
+# endif
 
 	movdqa	160(%rdi), %xmm2
 	pxor	160(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(176bytesin256)
+# endif
 
 	movdqa	176(%rdi), %xmm2
 	pxor	176(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(192bytesin256)
+# endif
 
 	movdqa	192(%rdi), %xmm2
 	pxor	192(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(208bytesin256)
+# endif
 
 	movdqa	208(%rdi), %xmm2
 	pxor	208(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(224bytesin256)
+# endif
 
 	movdqa	224(%rdi), %xmm2
 	pxor	224(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(240bytesin256)
+# endif
 
 	movdqa	240(%rdi), %xmm2
 	pxor	240(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(256bytesin256)
+# endif
 
 	add	$256, %rsi
 	add	$256, %rdi
@@ -616,12 +892,20 @@  L(256bytesormorein2aligned):
 	movdqa	(%rdi), %xmm2
 	pxor	(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(16bytesin256)
+# endif
 
 	movdqa	16(%rdi), %xmm2
 	pxor	16(%rsi), %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(32bytesin256)
+# endif
 	sub	$32, %rdx
 	add	$32, %rdi
 	add	$32, %rsi
@@ -663,7 +947,11 @@  L(64bytesormore_loopin2aligned):
 	por	%xmm5, %xmm1
 
 	ptest	%xmm1, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesormore_loop_end)
+# endif
 	add	$64, %rsi
 	add	$64, %rdi
 	sub	$64, %rdx
@@ -697,7 +985,11 @@  L(L2_L3_aligned_128bytes_loop):
 	por	%xmm5, %xmm1
 
 	ptest	%xmm1, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(64bytesormore_loop_end)
+# endif
 	add	$64, %rsi
 	add	$64, %rdi
 	sub	$64, %rdx
@@ -708,7 +1000,7 @@  L(L2_L3_aligned_128bytes_loop):
 	add	%rdx, %rdi
 	BRANCH_TO_JMPTBL_ENTRY(L(table_64bytes), %rdx, 4)
 
-
+# ifndef USE_AS_BCMP
 	.p2align 4
 L(64bytesormore_loop_end):
 	add	$16, %rdi
@@ -791,17 +1083,29 @@  L(32bytesin256):
 L(16bytesin256):
 	add	$16, %rdi
 	add	$16, %rsi
+# endif
 L(16bytes):
 	mov	-16(%rdi), %rax
 	mov	-16(%rsi), %rcx
 	cmp	%rax, %rcx
+# ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+# else
 	jne	L(diffin8bytes)
+# endif
 L(8bytes):
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
+# ifdef USE_AS_BCMP
+	sub	%rcx, %rax
+	mov	%rax, %rcx
+	shr	$32, %rcx
+	or	%ecx, %eax
+# else
 	cmp	%rax, %rcx
 	jne	L(diffin8bytes)
 	xor	%eax, %eax
+# endif
 	ret
 
 	.p2align 4
@@ -809,16 +1113,26 @@  L(12bytes):
 	mov	-12(%rdi), %rax
 	mov	-12(%rsi), %rcx
 	cmp	%rax, %rcx
+# ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+# else
 	jne	L(diffin8bytes)
+# endif
 L(4bytes):
 	mov	-4(%rsi), %ecx
-# ifndef USE_AS_WMEMCMP
+# ifdef USE_AS_BCMP
 	mov	-4(%rdi), %eax
-	cmp	%eax, %ecx
+	sub	%ecx, %eax
+	ret
 # else
+#  ifndef USE_AS_WMEMCMP
+	mov	-4(%rdi), %eax
+	cmp	%eax, %ecx
+#  else
 	cmp	-4(%rdi), %ecx
-# endif
+#  endif
 	jne	L(diffin4bytes)
+# endif
 L(0bytes):
 	xor	%eax, %eax
 	ret
@@ -832,31 +1146,51 @@  L(65bytes):
 	mov	$-65, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(49bytes):
 	movdqu	-49(%rdi), %xmm1
 	movdqu	-49(%rsi), %xmm2
 	mov	$-49, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(33bytes):
 	movdqu	-33(%rdi), %xmm1
 	movdqu	-33(%rsi), %xmm2
 	mov	$-33, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(17bytes):
 	mov	-17(%rdi), %rax
 	mov	-17(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 L(9bytes):
 	mov	-9(%rdi), %rax
 	mov	-9(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	movzbl	-1(%rdi), %eax
 	movzbl	-1(%rsi), %edx
 	sub	%edx, %eax
@@ -867,12 +1201,23 @@  L(13bytes):
 	mov	-13(%rdi), %rax
 	mov	-13(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
+#  ifdef USE_AS_BCMP
+	sub	%rcx, %rax
+	mov	%rax, %rcx
+	shr	$32, %rcx
+	or	%ecx, %eax
+#  else
 	cmp	%rax, %rcx
 	jne	L(diffin8bytes)
 	xor	%eax, %eax
+#  endif
 	ret
 
 	.p2align 4
@@ -880,7 +1225,11 @@  L(5bytes):
 	mov	-5(%rdi), %eax
 	mov	-5(%rsi), %ecx
 	cmp	%eax, %ecx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin4bytes)
+#  endif
 	movzbl	-1(%rdi), %eax
 	movzbl	-1(%rsi), %edx
 	sub	%edx, %eax
@@ -893,37 +1242,59 @@  L(66bytes):
 	mov	$-66, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(50bytes):
 	movdqu	-50(%rdi), %xmm1
 	movdqu	-50(%rsi), %xmm2
 	mov	$-50, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(34bytes):
 	movdqu	-34(%rdi), %xmm1
 	movdqu	-34(%rsi), %xmm2
 	mov	$-34, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(18bytes):
 	mov	-18(%rdi), %rax
 	mov	-18(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 L(10bytes):
 	mov	-10(%rdi), %rax
 	mov	-10(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	movzwl	-2(%rdi), %eax
 	movzwl	-2(%rsi), %ecx
+#  ifndef USE_AS_BCMP
 	cmp	%cl, %al
 	jne	L(end)
 	and	$0xffff, %eax
 	and	$0xffff, %ecx
+#  endif
 	sub	%ecx, %eax
 	ret
 
@@ -932,12 +1303,23 @@  L(14bytes):
 	mov	-14(%rdi), %rax
 	mov	-14(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
+#  ifdef USE_AS_BCMP
+	sub	%rcx, %rax
+	mov	%rax, %rcx
+	shr	$32, %rcx
+	or	%ecx, %eax
+#  else
 	cmp	%rax, %rcx
 	jne	L(diffin8bytes)
 	xor	%eax, %eax
+#  endif
 	ret
 
 	.p2align 4
@@ -945,14 +1327,20 @@  L(6bytes):
 	mov	-6(%rdi), %eax
 	mov	-6(%rsi), %ecx
 	cmp	%eax, %ecx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin4bytes)
+#  endif
 L(2bytes):
 	movzwl	-2(%rsi), %ecx
 	movzwl	-2(%rdi), %eax
+#  ifndef USE_AS_BCMP
 	cmp	%cl, %al
 	jne	L(end)
 	and	$0xffff, %eax
 	and	$0xffff, %ecx
+#  endif
 	sub	%ecx, %eax
 	ret
 
@@ -963,36 +1351,60 @@  L(67bytes):
 	mov	$-67, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(51bytes):
 	movdqu	-51(%rdi), %xmm2
 	movdqu	-51(%rsi), %xmm1
 	mov	$-51, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(35bytes):
 	movdqu	-35(%rsi), %xmm1
 	movdqu	-35(%rdi), %xmm2
 	mov	$-35, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(19bytes):
 	mov	-19(%rdi), %rax
 	mov	-19(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 L(11bytes):
 	mov	-11(%rdi), %rax
 	mov	-11(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	mov	-4(%rdi), %eax
 	mov	-4(%rsi), %ecx
+#  ifdef USE_AS_BCMP
+	sub	%ecx, %eax
+#  else
 	cmp	%eax, %ecx
 	jne	L(diffin4bytes)
 	xor	%eax, %eax
+#  endif
 	ret
 
 	.p2align 4
@@ -1000,12 +1412,23 @@  L(15bytes):
 	mov	-15(%rdi), %rax
 	mov	-15(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
+#  ifdef USE_AS_BCMP
+	sub	%rcx, %rax
+	mov	%rax, %rcx
+	shr	$32, %rcx
+	or	%ecx, %eax
+#  else
 	cmp	%rax, %rcx
 	jne	L(diffin8bytes)
 	xor	%eax, %eax
+#  endif
 	ret
 
 	.p2align 4
@@ -1013,12 +1436,20 @@  L(7bytes):
 	mov	-7(%rdi), %eax
 	mov	-7(%rsi), %ecx
 	cmp	%eax, %ecx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin4bytes)
+#  endif
 	mov	-4(%rdi), %eax
 	mov	-4(%rsi), %ecx
+#  ifdef USE_AS_BCMP
+	sub	%ecx, %eax
+#  else
 	cmp	%eax, %ecx
 	jne	L(diffin4bytes)
 	xor	%eax, %eax
+#  endif
 	ret
 
 	.p2align 4
@@ -1026,7 +1457,11 @@  L(3bytes):
 	movzwl	-3(%rdi), %eax
 	movzwl	-3(%rsi), %ecx
 	cmp	%eax, %ecx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin2bytes)
+#  endif
 L(1bytes):
 	movzbl	-1(%rdi), %eax
 	movzbl	-1(%rsi), %ecx
@@ -1041,38 +1476,58 @@  L(68bytes):
 	mov	$-68, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(52bytes):
 	movdqu	-52(%rdi), %xmm2
 	movdqu	-52(%rsi), %xmm1
 	mov	$-52, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(36bytes):
 	movdqu	-36(%rdi), %xmm2
 	movdqu	-36(%rsi), %xmm1
 	mov	$-36, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(20bytes):
 	movdqu	-20(%rdi), %xmm2
 	movdqu	-20(%rsi), %xmm1
 	mov	$-20, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 	mov	-4(%rsi), %ecx
-
-# ifndef USE_AS_WMEMCMP
+# ifdef USE_AS_BCMP
 	mov	-4(%rdi), %eax
-	cmp	%eax, %ecx
+	sub	%ecx, %eax
 # else
+#  ifndef USE_AS_WMEMCMP
+	mov	-4(%rdi), %eax
+	cmp	%eax, %ecx
+#  else
 	cmp	-4(%rdi), %ecx
-# endif
+#  endif
 	jne	L(diffin4bytes)
 	xor	%eax, %eax
+# endif
 	ret
 
 # ifndef USE_AS_WMEMCMP
@@ -1084,32 +1539,52 @@  L(69bytes):
 	mov	$-69, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(53bytes):
 	movdqu	-53(%rsi), %xmm1
 	movdqu	-53(%rdi), %xmm2
 	mov	$-53, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(37bytes):
 	movdqu	-37(%rsi), %xmm1
 	movdqu	-37(%rdi), %xmm2
 	mov	$-37, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(21bytes):
 	movdqu	-21(%rsi), %xmm1
 	movdqu	-21(%rdi), %xmm2
 	mov	$-21, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	xor	%eax, %eax
 	ret
 
@@ -1120,32 +1595,52 @@  L(70bytes):
 	mov	$-70, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(54bytes):
 	movdqu	-54(%rsi), %xmm1
 	movdqu	-54(%rdi), %xmm2
 	mov	$-54, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(38bytes):
 	movdqu	-38(%rsi), %xmm1
 	movdqu	-38(%rdi), %xmm2
 	mov	$-38, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(22bytes):
 	movdqu	-22(%rsi), %xmm1
 	movdqu	-22(%rdi), %xmm2
 	mov	$-22, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	xor	%eax, %eax
 	ret
 
@@ -1156,32 +1651,52 @@  L(71bytes):
 	mov	$-71, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(55bytes):
 	movdqu	-55(%rdi), %xmm2
 	movdqu	-55(%rsi), %xmm1
 	mov	$-55, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(39bytes):
 	movdqu	-39(%rdi), %xmm2
 	movdqu	-39(%rsi), %xmm1
 	mov	$-39, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(23bytes):
 	movdqu	-23(%rdi), %xmm2
 	movdqu	-23(%rsi), %xmm1
 	mov	$-23, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	xor	%eax, %eax
 	ret
 # endif
@@ -1193,33 +1708,53 @@  L(72bytes):
 	mov	$-72, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(56bytes):
 	movdqu	-56(%rdi), %xmm2
 	movdqu	-56(%rsi), %xmm1
 	mov	$-56, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(40bytes):
 	movdqu	-40(%rdi), %xmm2
 	movdqu	-40(%rsi), %xmm1
 	mov	$-40, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(24bytes):
 	movdqu	-24(%rdi), %xmm2
 	movdqu	-24(%rsi), %xmm1
 	mov	$-24, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 
 	mov	-8(%rsi), %rcx
 	mov	-8(%rdi), %rax
 	cmp	%rax, %rcx
+# ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+# else
 	jne	L(diffin8bytes)
+# endif
 	xor	%eax, %eax
 	ret
 
@@ -1232,32 +1767,52 @@  L(73bytes):
 	mov	$-73, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(57bytes):
 	movdqu	-57(%rdi), %xmm2
 	movdqu	-57(%rsi), %xmm1
 	mov	$-57, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(41bytes):
 	movdqu	-41(%rdi), %xmm2
 	movdqu	-41(%rsi), %xmm1
 	mov	$-41, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(25bytes):
 	movdqu	-25(%rdi), %xmm2
 	movdqu	-25(%rsi), %xmm1
 	mov	$-25, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-9(%rdi), %rax
 	mov	-9(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	movzbl	-1(%rdi), %eax
 	movzbl	-1(%rsi), %ecx
 	sub	%ecx, %eax
@@ -1270,35 +1825,60 @@  L(74bytes):
 	mov	$-74, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(58bytes):
 	movdqu	-58(%rdi), %xmm2
 	movdqu	-58(%rsi), %xmm1
 	mov	$-58, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(42bytes):
 	movdqu	-42(%rdi), %xmm2
 	movdqu	-42(%rsi), %xmm1
 	mov	$-42, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(26bytes):
 	movdqu	-26(%rdi), %xmm2
 	movdqu	-26(%rsi), %xmm1
 	mov	$-26, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-10(%rdi), %rax
 	mov	-10(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	movzwl	-2(%rdi), %eax
 	movzwl	-2(%rsi), %ecx
+#  ifdef USE_AS_BCMP
+	sub	%ecx, %eax
+	ret
+#  else
 	jmp	L(diffin2bytes)
+#  endif
 
 	.p2align 4
 L(75bytes):
@@ -1307,37 +1887,61 @@  L(75bytes):
 	mov	$-75, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(59bytes):
 	movdqu	-59(%rdi), %xmm2
 	movdqu	-59(%rsi), %xmm1
 	mov	$-59, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(43bytes):
 	movdqu	-43(%rdi), %xmm2
 	movdqu	-43(%rsi), %xmm1
 	mov	$-43, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(27bytes):
 	movdqu	-27(%rdi), %xmm2
 	movdqu	-27(%rsi), %xmm1
 	mov	$-27, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-11(%rdi), %rax
 	mov	-11(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	mov	-4(%rdi), %eax
 	mov	-4(%rsi), %ecx
+#  ifdef USE_AS_BCMP
+	sub	%ecx, %eax
+#  else
 	cmp	%eax, %ecx
 	jne	L(diffin4bytes)
 	xor	%eax, %eax
+#  endif
 	ret
 # endif
 	.p2align 4
@@ -1347,41 +1951,66 @@  L(76bytes):
 	mov	$-76, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(60bytes):
 	movdqu	-60(%rdi), %xmm2
 	movdqu	-60(%rsi), %xmm1
 	mov	$-60, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(44bytes):
 	movdqu	-44(%rdi), %xmm2
 	movdqu	-44(%rsi), %xmm1
 	mov	$-44, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(28bytes):
 	movdqu	-28(%rdi), %xmm2
 	movdqu	-28(%rsi), %xmm1
 	mov	$-28, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 	mov	-12(%rdi), %rax
 	mov	-12(%rsi), %rcx
 	cmp	%rax, %rcx
+# ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+# else
 	jne	L(diffin8bytes)
+# endif
 	mov	-4(%rsi), %ecx
-# ifndef USE_AS_WMEMCMP
+# ifdef USE_AS_BCMP
 	mov	-4(%rdi), %eax
-	cmp	%eax, %ecx
+	sub	%ecx, %eax
 # else
+#  ifndef USE_AS_WMEMCMP
+	mov	-4(%rdi), %eax
+	cmp	%eax, %ecx
+#  else
 	cmp	-4(%rdi), %ecx
-# endif
+#  endif
 	jne	L(diffin4bytes)
 	xor	%eax, %eax
+# endif
 	ret
 
 # ifndef USE_AS_WMEMCMP
@@ -1393,38 +2022,62 @@  L(77bytes):
 	mov	$-77, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(61bytes):
 	movdqu	-61(%rdi), %xmm2
 	movdqu	-61(%rsi), %xmm1
 	mov	$-61, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(45bytes):
 	movdqu	-45(%rdi), %xmm2
 	movdqu	-45(%rsi), %xmm1
 	mov	$-45, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(29bytes):
 	movdqu	-29(%rdi), %xmm2
 	movdqu	-29(%rsi), %xmm1
 	mov	$-29, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 
 	mov	-13(%rdi), %rax
 	mov	-13(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	xor	%eax, %eax
 	ret
 
@@ -1435,36 +2088,60 @@  L(78bytes):
 	mov	$-78, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(62bytes):
 	movdqu	-62(%rdi), %xmm2
 	movdqu	-62(%rsi), %xmm1
 	mov	$-62, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(46bytes):
 	movdqu	-46(%rdi), %xmm2
 	movdqu	-46(%rsi), %xmm1
 	mov	$-46, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(30bytes):
 	movdqu	-30(%rdi), %xmm2
 	movdqu	-30(%rsi), %xmm1
 	mov	$-30, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-14(%rdi), %rax
 	mov	-14(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	xor	%eax, %eax
 	ret
 
@@ -1475,36 +2152,60 @@  L(79bytes):
 	mov	$-79, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(63bytes):
 	movdqu	-63(%rdi), %xmm2
 	movdqu	-63(%rsi), %xmm1
 	mov	$-63, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(47bytes):
 	movdqu	-47(%rdi), %xmm2
 	movdqu	-47(%rsi), %xmm1
 	mov	$-47, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 L(31bytes):
 	movdqu	-31(%rdi), %xmm2
 	movdqu	-31(%rsi), %xmm1
 	mov	$-31, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+#  ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+#  else
 	jnc	L(less16bytes)
+#  endif
 	mov	-15(%rdi), %rax
 	mov	-15(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
 	cmp	%rax, %rcx
+#  ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+#  else
 	jne	L(diffin8bytes)
+#  endif
 	xor	%eax, %eax
 	ret
 # endif
@@ -1515,37 +2216,58 @@  L(64bytes):
 	mov	$-64, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(48bytes):
 	movdqu	-48(%rdi), %xmm2
 	movdqu	-48(%rsi), %xmm1
 	mov	$-48, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 L(32bytes):
 	movdqu	-32(%rdi), %xmm2
 	movdqu	-32(%rsi), %xmm1
 	mov	$-32, %dl
 	pxor	%xmm1, %xmm2
 	ptest	%xmm2, %xmm0
+# ifdef USE_AS_BCMP
+	jnc	L(return_not_equals)
+# else
 	jnc	L(less16bytes)
+# endif
 
 	mov	-16(%rdi), %rax
 	mov	-16(%rsi), %rcx
 	cmp	%rax, %rcx
+# ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+# else
 	jne	L(diffin8bytes)
+# endif
 
 	mov	-8(%rdi), %rax
 	mov	-8(%rsi), %rcx
 	cmp	%rax, %rcx
+# ifdef USE_AS_BCMP
+	jne	L(return_not_equals)
+# else
 	jne	L(diffin8bytes)
+# endif
 	xor	%eax, %eax
 	ret
 
 /*
  * Aligned 8 bytes to avoid 2 branch "taken" in one 16 alinged code block.
  */
+# ifndef USE_AS_BCMP
 	.p2align 3
 L(less16bytes):
 	movsbq	%dl, %rdx
@@ -1561,16 +2283,16 @@  L(diffin8bytes):
 	shr	$32, %rcx
 	shr	$32, %rax
 
-# ifdef USE_AS_WMEMCMP
+#  ifdef USE_AS_WMEMCMP
 /* for wmemcmp */
 	cmp	%eax, %ecx
 	jne	L(diffin4bytes)
 	xor	%eax, %eax
 	ret
-# endif
+#  endif
 
 L(diffin4bytes):
-# ifndef USE_AS_WMEMCMP
+#  ifndef USE_AS_WMEMCMP
 	cmp	%cx, %ax
 	jne	L(diffin2bytes)
 	shr	$16, %ecx
@@ -1589,7 +2311,7 @@  L(end):
 	and	$0xff, %ecx
 	sub	%ecx, %eax
 	ret
-# else
+#  else
 
 /* for wmemcmp */
 	mov	$1, %eax
@@ -1601,6 +2323,15 @@  L(end):
 L(nequal_bigger):
 	ret
 
+L(unreal_case):
+	xor	%eax, %eax
+	ret
+#  endif
+# else
+	.p2align 4
+L(return_not_equals):
+	mov	$1, %eax
+	ret
 L(unreal_case):
 	xor	%eax, %eax
 	ret