x86: Optimize and shrink st{r|p}{n}{cat|cpy}-avx2 functions

Optimizations are:
    1. Use more overlapping stores to avoid branches.
    2. Reduce how unrolled the aligning copies are (this is more of a
       code-size save, its a negative for some sizes in terms of
       perf).
    3. For st{r|p}n{cat|cpy} re-order the branches to minimize the
       number that are taken.

Performance Changes:

    Times are from N = 10 runs of the benchmark suite and are
    reported as geometric mean of all ratios of
    New Implementation / Old Implementation.

    strcat-avx2      -> 0.998
    strcpy-avx2      -> 0.937
    stpcpy-avx2      -> 0.971

    strncpy-avx2     -> 0.793
    stpncpy-avx2     -> 0.775

    strncat-avx2     -> 0.962

Code Size Changes:
    function         -> Bytes New / Bytes Old -> Ratio

    strcat-avx2      ->  685 / 1639 -> 0.418
    strcpy-avx2      ->  560 /  903 -> 0.620
    stpcpy-avx2      ->  592 /  939 -> 0.630

    strncpy-avx2     -> 1176 / 2390 -> 0.492
    stpncpy-avx2     -> 1268 / 2438 -> 0.520

    strncat-avx2     -> 1042 / 2563 -> 0.407

Notes:
    1. Because of the significant difference between the
       implementations they are split into three files.

           strcpy-avx2.S    -> strcpy, stpcpy, strcat
           strncpy-avx2.S   -> strncpy
           strncat-avx2.S    > strncat

       I couldn't find a way to merge them without making the
       ifdefs incredibly difficult to follow.

Full check passes on x86-64 and build succeeds for all ISA levels w/
and w/o multiarch.

13 files changed, 1594 insertions, 1234 deletions

diff --git a/sysdeps/x86_64/multiarch/stpcpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/stpcpy-avx2-rtm.S
index 2b9c07a..90e532d 100644
--- a/sysdeps/x86_64/multiarch/stpcpy-avx2-rtm.S
+++ b/sysdeps/x86_64/multiarch/stpcpy-avx2-rtm.S
@@ -1,3 +1,3 @@
-#define USE_AS_STPCPY
-#define STRCPY __stpcpy_avx2_rtm
-#include "strcpy-avx2-rtm.S"
+#define STPCPY	__stpcpy_avx2_rtm
+#include "x86-avx-rtm-vecs.h"
+#include "stpcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/stpncpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/stpncpy-avx2-rtm.S
index 60a2ccf..46ee07b 100644
--- a/sysdeps/x86_64/multiarch/stpncpy-avx2-rtm.S
+++ b/sysdeps/x86_64/multiarch/stpncpy-avx2-rtm.S
@@ -1,4 +1,3 @@
-#define USE_AS_STPCPY
-#define USE_AS_STRNCPY
-#define STRCPY __stpncpy_avx2_rtm
-#include "strcpy-avx2-rtm.S"
+#define STPNCPY	__stpncpy_avx2_rtm
+#include "x86-avx-rtm-vecs.h"
+#include "stpncpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/stpncpy-avx2.S b/sysdeps/x86_64/multiarch/stpncpy-avx2.S
index b2f8c19..a46a8ed 100644
--- a/sysdeps/x86_64/multiarch/stpncpy-avx2.S
+++ b/sysdeps/x86_64/multiarch/stpncpy-avx2.S
@@ -3,6 +3,5 @@
 #endif
 
 #define USE_AS_STPCPY
-#define USE_AS_STRNCPY
-#define STRCPY	STPNCPY
-#include "strcpy-avx2.S"
+#define STRNCPY	STPNCPY
+#include "strncpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strcat-avx2-rtm.S b/sysdeps/x86_64/multiarch/strcat-avx2-rtm.S
index 637fb55..e84f4f1 100644
--- a/sysdeps/x86_64/multiarch/strcat-avx2-rtm.S
+++ b/sysdeps/x86_64/multiarch/strcat-avx2-rtm.S
@@ -1,12 +1,3 @@
-#ifndef STRCAT
-# define STRCAT __strcat_avx2_rtm
-#endif
-
-#define ZERO_UPPER_VEC_REGISTERS_RETURN \
-  ZERO_UPPER_VEC_REGISTERS_RETURN_XTEST
-
-#define VZEROUPPER_RETURN jmp	 L(return_vzeroupper)
-
-#define SECTION(p) p##.avx.rtm
-
+#define STRCAT	__strcat_avx2_rtm
+#include "x86-avx-rtm-vecs.h"
 #include "strcat-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S b/sysdeps/x86_64/multiarch/strcat-avx2.S
index d9b7fb2..3f914fa 100644
--- a/sysdeps/x86_64/multiarch/strcat-avx2.S
+++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
@@ -16,266 +16,10 @@
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
-#include <isa-level.h>
-
-#if ISA_SHOULD_BUILD (3)
-
-
-# include <sysdep.h>
-
-# ifndef STRCAT
-#  define STRCAT  __strcat_avx2
-# endif
-
-# define USE_AS_STRCAT
-
-/* Number of bytes in a vector register */
-# define VEC_SIZE	32
-
-# ifndef SECTION
-#  define SECTION(p)	p##.avx
-# endif
-
-	.section SECTION(.text),"ax",@progbits
-ENTRY (STRCAT)
-	mov	%rdi, %r9
-# ifdef USE_AS_STRNCAT
-	mov	%rdx, %r8
-# endif
-
-	xor	%eax, %eax
-	mov	%edi, %ecx
-	and	$((VEC_SIZE * 4) - 1), %ecx
-	vpxor	%xmm6, %xmm6, %xmm6
-	cmp	$(VEC_SIZE * 3), %ecx
-	ja	L(fourth_vector_boundary)
-	vpcmpeqb (%rdi), %ymm6, %ymm0
-	vpmovmskb %ymm0, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_first_vector)
-	mov	%rdi, %rax
-	and	$-VEC_SIZE, %rax
-	jmp	L(align_vec_size_start)
-L(fourth_vector_boundary):
-	mov	%rdi, %rax
-	and	$-VEC_SIZE, %rax
-	vpcmpeqb	(%rax), %ymm6, %ymm0
-	mov	$-1, %r10d
-	sub	%rax, %rcx
-	shl	%cl, %r10d
-	vpmovmskb %ymm0, %edx
-	and	%r10d, %edx
-	jnz	L(exit)
-
-L(align_vec_size_start):
-	vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm0
-	vpmovmskb %ymm0, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_second_vector)
-
-	vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
-	vpmovmskb %ymm1, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_third_vector)
-
-	vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
-	vpmovmskb %ymm2, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fourth_vector)
-
-	vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
-	vpmovmskb %ymm3, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fifth_vector)
-
-	vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
-	add	$(VEC_SIZE * 4), %rax
-	vpmovmskb %ymm0, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_second_vector)
-
-	vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
-	vpmovmskb %ymm1, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_third_vector)
-
-	vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
-	vpmovmskb %ymm2, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fourth_vector)
-
-	vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
-	vpmovmskb %ymm3, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fifth_vector)
-
-	vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
-	add	$(VEC_SIZE * 4), %rax
-	vpmovmskb %ymm0, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_second_vector)
-
-	vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
-	vpmovmskb %ymm1, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_third_vector)
-
-	vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
-	vpmovmskb %ymm2, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fourth_vector)
-
-	vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
-	vpmovmskb %ymm3, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fifth_vector)
-
-	vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
-	add	$(VEC_SIZE * 4), %rax
-	vpmovmskb %ymm0, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_second_vector)
-
-	vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
-	vpmovmskb %ymm1, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_third_vector)
-
-	vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
-	vpmovmskb %ymm2, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fourth_vector)
-
-	vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
-	vpmovmskb %ymm3, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fifth_vector)
-
-	test	$((VEC_SIZE * 4) - 1), %rax
-	jz	L(align_four_vec_loop)
-
-	vpcmpeqb (VEC_SIZE * 5)(%rax), %ymm6, %ymm0
-	add	$(VEC_SIZE * 5), %rax
-	vpmovmskb %ymm0, %edx
-	test	%edx, %edx
-	jnz	L(exit)
-
-	test	$((VEC_SIZE * 4) - 1), %rax
-	jz	L(align_four_vec_loop)
-
-	vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm1
-	add	$VEC_SIZE, %rax
-	vpmovmskb %ymm1, %edx
-	test	%edx, %edx
-	jnz	L(exit)
-
-	test	$((VEC_SIZE * 4) - 1), %rax
-	jz	L(align_four_vec_loop)
-
-	vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm2
-	add	$VEC_SIZE, %rax
-	vpmovmskb %ymm2, %edx
-	test	%edx, %edx
-	jnz	L(exit)
-
-	test	$((VEC_SIZE * 4) - 1), %rax
-	jz	L(align_four_vec_loop)
-
-	vpcmpeqb VEC_SIZE(%rax), %ymm6, %ymm3
-	add	$VEC_SIZE, %rax
-	vpmovmskb %ymm3, %edx
-	test	%edx, %edx
-	jnz	L(exit)
-
-	add	$VEC_SIZE, %rax
-
-	.p2align 4
-L(align_four_vec_loop):
-	vmovaps	(%rax),	%ymm4
-	vpminub	VEC_SIZE(%rax),	%ymm4, %ymm4
-	vmovaps	(VEC_SIZE * 2)(%rax),	%ymm5
-	vpminub	(VEC_SIZE * 3)(%rax),	%ymm5, %ymm5
-	add	$(VEC_SIZE * 4),	%rax
-	vpminub	%ymm4,	%ymm5, %ymm5
-	vpcmpeqb %ymm5,	%ymm6, %ymm5
-	vpmovmskb %ymm5,	%edx
-	test	%edx,	%edx
-	jz	L(align_four_vec_loop)
-
-	vpcmpeqb -(VEC_SIZE * 4)(%rax), %ymm6, %ymm0
-	sub	$(VEC_SIZE * 5),	%rax
-	vpmovmskb %ymm0, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_second_vector)
-
-	vpcmpeqb (VEC_SIZE * 2)(%rax), %ymm6, %ymm1
-	vpmovmskb %ymm1, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_third_vector)
-
-	vpcmpeqb (VEC_SIZE * 3)(%rax), %ymm6, %ymm2
-	vpmovmskb %ymm2, %edx
-	test	%edx, %edx
-	jnz	L(exit_null_on_fourth_vector)
-
-	vpcmpeqb (VEC_SIZE * 4)(%rax), %ymm6, %ymm3
-	vpmovmskb %ymm3, %edx
-	sub	%rdi, %rax
-	bsf	%rdx, %rdx
-	add	%rdx, %rax
-	add	$(VEC_SIZE * 4), %rax
-	jmp	L(StartStrcpyPart)
-
-	.p2align 4
-L(exit):
-	sub	%rdi, %rax
-L(exit_null_on_first_vector):
-	bsf	%rdx, %rdx
-	add	%rdx, %rax
-	jmp	L(StartStrcpyPart)
-
-	.p2align 4
-L(exit_null_on_second_vector):
-	sub	%rdi, %rax
-	bsf	%rdx, %rdx
-	add	%rdx, %rax
-	add	$VEC_SIZE, %rax
-	jmp	L(StartStrcpyPart)
-
-	.p2align 4
-L(exit_null_on_third_vector):
-	sub	%rdi, %rax
-	bsf	%rdx, %rdx
-	add	%rdx, %rax
-	add	$(VEC_SIZE * 2), %rax
-	jmp	L(StartStrcpyPart)
-
-	.p2align 4
-L(exit_null_on_fourth_vector):
-	sub	%rdi, %rax
-	bsf	%rdx, %rdx
-	add	%rdx, %rax
-	add	$(VEC_SIZE * 3), %rax
-	jmp	L(StartStrcpyPart)
-
-	.p2align 4
-L(exit_null_on_fifth_vector):
-	sub	%rdi, %rax
-	bsf	%rdx, %rdx
-	add	%rdx, %rax
-	add	$(VEC_SIZE * 4), %rax
-
-	.p2align 4
-L(StartStrcpyPart):
-	lea	(%r9, %rax), %rdi
-	mov	%rsi, %rcx
-	mov	%r9, %rax      /* save result */
-
-# ifdef USE_AS_STRNCAT
-	test	%r8, %r8
-	jz	L(ExitZero)
-#  define USE_AS_STRNCPY
-# endif
-
-# include "strcpy-avx2.S"
+#ifndef STRCAT
+# define STRCAT	__strcat_avx2
 #endif
+
+#define USE_AS_STRCAT
+#define STRCPY	STRCAT
+#include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strcat-strlen-avx2.h.S b/sysdeps/x86_64/multiarch/strcat-strlen-avx2.h.S
new file mode 100644
index 0000000..f50514e
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strcat-strlen-avx2.h.S
@@ -0,0 +1,101 @@
+/* strlen used for begining of str{n}cat using AVX2.
+   Copyright (C) 2011-2022 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+
+/* NOTE: This file is meant to be included by strcat-avx2 or
+   strncat-avx2 and does not standalone.  Before including %rdi
+   must be saved in %rax.  */
+
+
+/* Simple strlen implementation that ends at
+   L(strcat_strlen_done).  */
+	movq	%rdi, %r8
+	andq	$(VEC_SIZE * -1), %r8
+	VPCMPEQ	(%r8), %VZERO, %VMM(0)
+	vpmovmskb %VMM(0), %ecx
+	shrxl	%edi, %ecx, %ecx
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v0)
+
+	VPCMPEQ	VEC_SIZE(%r8), %VZERO, %VMM(0)
+	vpmovmskb %VMM(0), %ecx
+	leaq	(VEC_SIZE)(%r8), %rdi
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v0)
+
+	VPCMPEQ	(VEC_SIZE * 1)(%rdi), %VZERO, %VMM(0)
+	vpmovmskb %VMM(0), %ecx
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v1)
+
+	VPCMPEQ	(VEC_SIZE * 2)(%rdi), %VZERO, %VMM(0)
+	vpmovmskb %VMM(0), %ecx
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v2)
+
+	VPCMPEQ	(VEC_SIZE * 3)(%rdi), %VZERO, %VMM(0)
+	vpmovmskb %VMM(0), %ecx
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v3)
+
+	orq	$(VEC_SIZE * 4 - 1), %rdi
+	.p2align 4,, 8
+L(loop_2x_vec):
+	VMOVA	(VEC_SIZE * 0 + 1)(%rdi), %VMM(0)
+	VPMIN	(VEC_SIZE * 1 + 1)(%rdi), %VMM(0), %VMM(1)
+	VMOVA	(VEC_SIZE * 2 + 1)(%rdi), %VMM(2)
+	VPMIN	(VEC_SIZE * 3 + 1)(%rdi), %VMM(2), %VMM(3)
+	VPMIN	%VMM(1), %VMM(3), %VMM(3)
+	VPCMPEQ	%VMM(3), %VZERO, %VMM(3)
+	vpmovmskb %VMM(3), %r8d
+	subq	$(VEC_SIZE * -4), %rdi
+	testl	%r8d, %r8d
+	jz	L(loop_2x_vec)
+
+	addq	$(VEC_SIZE * -4 + 1), %rdi
+
+	VPCMPEQ	%VMM(0), %VZERO, %VMM(0)
+	vpmovmskb %VMM(0), %ecx
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v0)
+
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(1)
+	vpmovmskb %VMM(1), %ecx
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v1)
+
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(2)
+	vpmovmskb %VMM(2), %ecx
+	testl	%ecx, %ecx
+	jnz	L(bsf_and_done_v2)
+
+	movl	%r8d, %ecx
+L(bsf_and_done_v3):
+	addq	$VEC_SIZE, %rdi
+L(bsf_and_done_v2):
+	bsfl	%ecx, %ecx
+	leaq	(VEC_SIZE * 2)(%rdi, %rcx), %rdi
+	jmp	L(strcat_strlen_done)
+
+	.p2align 4,, 4
+L(bsf_and_done_v1):
+	addq	$VEC_SIZE, %rdi
+L(bsf_and_done_v0):
+	bsfl	%ecx, %ecx
+	addq	%rcx, %rdi
+L(strcat_strlen_done):
diff --git a/sysdeps/x86_64/multiarch/strcpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/strcpy-avx2-rtm.S
index c2c581e..3ae2de8 100644
--- a/sysdeps/x86_64/multiarch/strcpy-avx2-rtm.S
+++ b/sysdeps/x86_64/multiarch/strcpy-avx2-rtm.S
@@ -1,12 +1,3 @@
-#ifndef STRCPY
-# define STRCPY __strcpy_avx2_rtm
-#endif
-
-#define ZERO_UPPER_VEC_REGISTERS_RETURN \
-  ZERO_UPPER_VEC_REGISTERS_RETURN_XTEST
-
-#define VZEROUPPER_RETURN jmp	 L(return_vzeroupper)
-
-#define SECTION(p) p##.avx.rtm
-
+#define STRCPY	__strcpy_avx2_rtm
+#include "x86-avx-rtm-vecs.h"
 #include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strcpy-avx2.S b/sysdeps/x86_64/multiarch/strcpy-avx2.S
index c725834..32f86ba 100644
--- a/sysdeps/x86_64/multiarch/strcpy-avx2.S
+++ b/sysdeps/x86_64/multiarch/strcpy-avx2.S
@@ -20,984 +20,378 @@
 
 #if ISA_SHOULD_BUILD (3)
 
+# include <sysdep.h>
 
-# ifndef USE_AS_STRCAT
-#  include <sysdep.h>
-
-#  ifndef STRCPY
-#   define STRCPY  __strcpy_avx2
-#  endif
-
-# endif
-
-/* Number of bytes in a vector register */
 # ifndef VEC_SIZE
-#  define VEC_SIZE	32
-# endif
-
-# ifndef VZEROUPPER
-#  define VZEROUPPER	vzeroupper
-# endif
-
-# ifndef SECTION
-#  define SECTION(p)	p##.avx
-# endif
-
-/* zero register */
-#define xmmZ	xmm0
-#define ymmZ	ymm0
-
-/* mask register */
-#define ymmM	ymm1
-
-# ifndef USE_AS_STRCAT
-
-	.section SECTION(.text),"ax",@progbits
-ENTRY (STRCPY)
-#  ifdef USE_AS_STRNCPY
-	mov	%RDX_LP, %R8_LP
-	test	%R8_LP, %R8_LP
-	jz	L(ExitZero)
-#  endif
-	mov	%rsi, %rcx
-#  ifndef USE_AS_STPCPY
-	mov	%rdi, %rax      /* save result */
-#  endif
-
+#  include "x86-avx-vecs.h"
 # endif
 
-	vpxor	%xmmZ, %xmmZ, %xmmZ
-
-	and	$((VEC_SIZE * 4) - 1), %ecx
-	cmp	$(VEC_SIZE * 2), %ecx
-	jbe	L(SourceStringAlignmentLessTwoVecSize)
-
-	and	$-VEC_SIZE, %rsi
-	and	$(VEC_SIZE - 1), %ecx
-
-	vpcmpeqb (%rsi), %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	shr	%cl, %rdx
-
-# ifdef USE_AS_STRNCPY
-#  if defined USE_AS_STPCPY || defined USE_AS_STRCAT
-	mov	$VEC_SIZE, %r10
-	sub	%rcx, %r10
-	cmp	%r10, %r8
-#  else
-	mov	$(VEC_SIZE + 1), %r10
-	sub	%rcx, %r10
-	cmp	%r10, %r8
-#  endif
-	jbe	L(CopyVecSizeTailCase2OrCase3)
+# ifndef STRCPY
+#  define STRCPY	__strcpy_avx2
 # endif
-	test	%edx, %edx
-	jnz	L(CopyVecSizeTail)
 
-	vpcmpeqb VEC_SIZE(%rsi), %ymmZ, %ymm2
-	vpmovmskb %ymm2, %edx
+	/* Use movsb in page cross case to save code size.  */
+# define USE_MOVSB_IN_PAGE_CROSS	1
 
-# ifdef USE_AS_STRNCPY
-	add	$VEC_SIZE, %r10
-	cmp	%r10, %r8
-	jbe	L(CopyTwoVecSizeCase2OrCase3)
-# endif
-	test	%edx, %edx
-	jnz	L(CopyTwoVecSize)
-
-	vmovdqu (%rsi, %rcx), %ymm2   /* copy VEC_SIZE bytes */
-	vmovdqu %ymm2, (%rdi)
-
-/* If source address alignment != destination address alignment */
-	.p2align 4
-L(UnalignVecSizeBoth):
-	sub	%rcx, %rdi
-# ifdef USE_AS_STRNCPY
-	add	%rcx, %r8
-	sbb	%rcx, %rcx
-	or	%rcx, %r8
-# endif
-	mov	$VEC_SIZE, %rcx
-	vmovdqa (%rsi, %rcx), %ymm2
-	vmovdqu %ymm2, (%rdi, %rcx)
-	vmovdqa VEC_SIZE(%rsi, %rcx), %ymm2
-	vpcmpeqb %ymm2, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	add	$VEC_SIZE, %rcx
-# ifdef USE_AS_STRNCPY
-	sub	$(VEC_SIZE * 3), %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-# endif
-	test	%edx, %edx
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec2)
+# ifdef USE_AS_WCSCPY
+#  define VPCMPEQ	vpcmpeqd
+#  define VPMIN	vpminud
+#  define CHAR_SIZE	4
 # else
-	jnz	L(CopyVecSize)
+#  define VPCMPEQ	vpcmpeqb
+#  define VPMIN	vpminub
+#  define CHAR_SIZE	1
 # endif
 
-	vmovdqu %ymm2, (%rdi, %rcx)
-	vmovdqa VEC_SIZE(%rsi, %rcx), %ymm3
-	vpcmpeqb %ymm3, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	add	$VEC_SIZE, %rcx
-# ifdef USE_AS_STRNCPY
-	sub	$VEC_SIZE, %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-# endif
-	test	%edx, %edx
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec3)
-# else
-	jnz	L(CopyVecSize)
-# endif
+# define PAGE_SIZE	4096
 
-	vmovdqu %ymm3, (%rdi, %rcx)
-	vmovdqa VEC_SIZE(%rsi, %rcx), %ymm4
-	vpcmpeqb %ymm4, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	add	$VEC_SIZE, %rcx
-# ifdef USE_AS_STRNCPY
-	sub	$VEC_SIZE, %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-# endif
-	test	%edx, %edx
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec4)
+# ifdef USE_AS_STPCPY
+#  define END_REG	rax
 # else
-	jnz	L(CopyVecSize)
+#  define END_REG	rdi, %rdx
 # endif
 
-	vmovdqu %ymm4, (%rdi, %rcx)
-	vmovdqa VEC_SIZE(%rsi, %rcx), %ymm2
-	vpcmpeqb %ymm2, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	add	$VEC_SIZE, %rcx
-# ifdef USE_AS_STRNCPY
-	sub	$VEC_SIZE, %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-# endif
-	test	%edx, %edx
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec2)
+# ifdef USE_AS_STRCAT
+#  define PAGE_ALIGN_REG	ecx
 # else
-	jnz	L(CopyVecSize)
+#  define PAGE_ALIGN_REG	eax
 # endif
 
-	vmovdqu %ymm2, (%rdi, %rcx)
-	vmovdqa VEC_SIZE(%rsi, %rcx), %ymm2
-	vpcmpeqb %ymm2, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	add	$VEC_SIZE, %rcx
-# ifdef USE_AS_STRNCPY
-	sub	$VEC_SIZE, %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-# endif
-	test	%edx, %edx
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec2)
-# else
-	jnz	L(CopyVecSize)
-# endif
+# define VZERO	VMM(7)
+# define VZERO_128	VMM_128(7)
 
-	vmovdqa VEC_SIZE(%rsi, %rcx), %ymm3
-	vmovdqu %ymm2, (%rdi, %rcx)
-	vpcmpeqb %ymm3, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	add	$VEC_SIZE, %rcx
-# ifdef USE_AS_STRNCPY
-	sub	$VEC_SIZE, %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-# endif
-	test	%edx, %edx
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec3)
-# else
-	jnz	L(CopyVecSize)
-# endif
+	.section SECTION(.text), "ax", @progbits
+ENTRY(STRCPY)
+	vpxor	%VZERO_128, %VZERO_128, %VZERO_128
 
-	vmovdqu %ymm3, (%rdi, %rcx)
-	mov	%rsi, %rdx
-	lea	VEC_SIZE(%rsi, %rcx), %rsi
-	and	$-(VEC_SIZE * 4), %rsi
-	sub	%rsi, %rdx
-	sub	%rdx, %rdi
-# ifdef USE_AS_STRNCPY
-	lea	(VEC_SIZE * 8)(%r8, %rdx), %r8
-# endif
-L(UnalignedFourVecSizeLoop):
-	vmovdqa (%rsi), %ymm4
-	vmovdqa VEC_SIZE(%rsi), %ymm5
-	vmovdqa (VEC_SIZE * 2)(%rsi), %ymm6
-	vmovdqa (VEC_SIZE * 3)(%rsi), %ymm7
-	vpminub %ymm5, %ymm4, %ymm2
-	vpminub %ymm7, %ymm6, %ymm3
-	vpminub %ymm2, %ymm3, %ymm3
-	vpcmpeqb %ymmM, %ymm3, %ymm3
-	vpmovmskb %ymm3, %edx
-# ifdef USE_AS_STRNCPY
-	sub	$(VEC_SIZE * 4), %r8
-	jbe	L(UnalignedLeaveCase2OrCase3)
-# endif
-	test	%edx, %edx
-	jnz	L(UnalignedFourVecSizeLeave)
-
-L(UnalignedFourVecSizeLoop_start):
-	add	$(VEC_SIZE * 4), %rdi
-	add	$(VEC_SIZE * 4), %rsi
-	vmovdqu %ymm4, -(VEC_SIZE * 4)(%rdi)
-	vmovdqa (%rsi), %ymm4
-	vmovdqu %ymm5, -(VEC_SIZE * 3)(%rdi)
-	vmovdqa VEC_SIZE(%rsi), %ymm5
-	vpminub %ymm5, %ymm4, %ymm2
-	vmovdqu %ymm6, -(VEC_SIZE * 2)(%rdi)
-	vmovdqa (VEC_SIZE * 2)(%rsi), %ymm6
-	vmovdqu %ymm7, -VEC_SIZE(%rdi)
-	vmovdqa (VEC_SIZE * 3)(%rsi), %ymm7
-	vpminub %ymm7, %ymm6, %ymm3
-	vpminub %ymm2, %ymm3, %ymm3
-	vpcmpeqb %ymmM, %ymm3, %ymm3
-	vpmovmskb %ymm3, %edx
-# ifdef USE_AS_STRNCPY
-	sub	$(VEC_SIZE * 4), %r8
-	jbe	L(UnalignedLeaveCase2OrCase3)
-# endif
-	test	%edx, %edx
-	jz	L(UnalignedFourVecSizeLoop_start)
-
-L(UnalignedFourVecSizeLeave):
-	vpcmpeqb %ymm4, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	test	%edx, %edx
-	jnz	L(CopyVecSizeUnaligned_0)
-
-	vpcmpeqb %ymm5, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %ecx
-	test	%ecx, %ecx
-	jnz	L(CopyVecSizeUnaligned_16)
-
-	vpcmpeqb %ymm6, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	test	%edx, %edx
-	jnz	L(CopyVecSizeUnaligned_32)
-
-	vpcmpeqb %ymm7, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %ecx
-	bsf	%ecx, %edx
-	vmovdqu %ymm4, (%rdi)
-	vmovdqu %ymm5, VEC_SIZE(%rdi)
-	vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-# ifdef USE_AS_STPCPY
-	lea	(VEC_SIZE * 3)(%rdi, %rdx), %rax
-# endif
-	vmovdqu %ymm7, (VEC_SIZE * 3)(%rdi)
-	add	$(VEC_SIZE - 1), %r8
-	sub	%rdx, %r8
-	lea	((VEC_SIZE * 3) + 1)(%rdi, %rdx), %rdi
-	jmp	L(StrncpyFillTailWithZero)
-# else
-	add	$(VEC_SIZE * 3), %rsi
-	add	$(VEC_SIZE * 3), %rdi
-	jmp	L(CopyVecSizeExit)
+# ifdef USE_AS_STRCAT
+	movq	%rdi, %rax
+#  include "strcat-strlen-avx2.h.S"
 # endif
 
-/* If source address alignment == destination address alignment */
-
-L(SourceStringAlignmentLessTwoVecSize):
-	vmovdqu (%rsi), %ymm3
-	vmovdqu VEC_SIZE(%rsi), %ymm2
-	vpcmpeqb %ymm3, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-
-# ifdef USE_AS_STRNCPY
-#  if defined USE_AS_STPCPY || defined USE_AS_STRCAT
-	cmp	$VEC_SIZE, %r8
-#  else
-	cmp	$(VEC_SIZE + 1), %r8
-#  endif
-	jbe	L(CopyVecSizeTail1Case2OrCase3)
+	movl	%esi, %PAGE_ALIGN_REG
+	andl	$(PAGE_SIZE - 1), %PAGE_ALIGN_REG
+	cmpl	$(PAGE_SIZE - VEC_SIZE), %PAGE_ALIGN_REG
+	ja	L(page_cross)
+L(page_cross_continue):
+# if !defined USE_AS_STPCPY && !defined USE_AS_STRCAT
+	movq	%rdi, %rax
 # endif
-	test	%edx, %edx
-	jnz	L(CopyVecSizeTail1)
-
-	vmovdqu %ymm3, (%rdi)
-	vpcmpeqb %ymm2, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-
-# ifdef USE_AS_STRNCPY
-#  if defined USE_AS_STPCPY || defined USE_AS_STRCAT
-	cmp	$(VEC_SIZE * 2), %r8
-#  else
-	cmp	$((VEC_SIZE * 2) + 1), %r8
-#  endif
-	jbe	L(CopyTwoVecSize1Case2OrCase3)
-# endif
-	test	%edx, %edx
-	jnz	L(CopyTwoVecSize1)
-
-	and	$-VEC_SIZE, %rsi
-	and	$(VEC_SIZE - 1), %ecx
-	jmp	L(UnalignVecSizeBoth)
+	VMOVU	(%rsi), %VMM(0)
+	VPCMPEQ	%VMM(0), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
 
-/*------End of main part with loops---------------------*/
+	testl	%ecx, %ecx
+	jz	L(more_1x_vec)
 
-/* Case1 */
+	/* No longer need ymm registers so just vzeroupper so it doesn't
+	   need to be duplicated at each return statement.  */
+	COND_VZEROUPPER
 
-# if (!defined USE_AS_STRNCPY) || (defined USE_AS_STRCAT)
-	.p2align 4
-L(CopyVecSize):
-	add	%rcx, %rdi
-# endif
-L(CopyVecSizeTail):
-	add	%rcx, %rsi
-L(CopyVecSizeTail1):
-	bsf	%edx, %edx
-L(CopyVecSizeExit):
-	cmp	$32, %edx
-	jae	L(Exit32_63)
-	cmp	$16, %edx
-	jae	L(Exit16_31)
-	cmp	$8, %edx
-	jae	L(Exit8_15)
-	cmp	$4, %edx
-	jae	L(Exit4_7)
-	cmp	$3, %edx
-	je	L(Exit3)
-	cmp	$1, %edx
-	ja	L(Exit2)
-	je	L(Exit1)
-	movb	$0, (%rdi)
+	xorl	%edx, %edx
+	bsfl	%ecx, %edx
 # ifdef USE_AS_STPCPY
-	lea	(%rdi), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	$1, %r8
-	lea	1(%rdi), %rdi
-	jnz	L(StrncpyFillTailWithZero)
-# endif
-L(return_vzeroupper):
-	ZERO_UPPER_VEC_REGISTERS_RETURN
-
-	.p2align 4
-L(CopyTwoVecSize1):
-	add	$VEC_SIZE, %rsi
-	add	$VEC_SIZE, %rdi
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	$VEC_SIZE, %r8
-# endif
-	jmp	L(CopyVecSizeTail1)
-
-	.p2align 4
-L(CopyTwoVecSize):
-	bsf	%edx, %edx
-	add	%rcx, %rsi
-	add	$VEC_SIZE, %edx
-	sub	%ecx, %edx
-	jmp	L(CopyVecSizeExit)
-
-	.p2align 4
-L(CopyVecSizeUnaligned_0):
-	bsf	%edx, %edx
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-# ifdef USE_AS_STPCPY
-	lea	(%rdi, %rdx), %rax
-# endif
-	vmovdqu %ymm4, (%rdi)
-	add	$((VEC_SIZE * 4) - 1), %r8
-	sub	%rdx, %r8
-	lea	1(%rdi, %rdx), %rdi
-	jmp	L(StrncpyFillTailWithZero)
-# else
-	jmp	L(CopyVecSizeExit)
-# endif
-
-	.p2align 4
-L(CopyVecSizeUnaligned_16):
-	bsf	%ecx, %edx
-	vmovdqu %ymm4, (%rdi)
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-# ifdef USE_AS_STPCPY
-	lea	VEC_SIZE(%rdi, %rdx), %rax
-# endif
-	vmovdqu %ymm5, VEC_SIZE(%rdi)
-	add	$((VEC_SIZE * 3) - 1), %r8
-	sub	%rdx, %r8
-	lea	(VEC_SIZE + 1)(%rdi, %rdx), %rdi
-	jmp	L(StrncpyFillTailWithZero)
+	leaq	(%rdi, %rdx), %rax
+# endif
+
+	/* Use mask bits in rcx to detect which copy we need. If the low
+	   mask is zero then there must be a bit set in the upper half.
+	   I.e if ecx != 0 and cx == 0, then match must be upper 16
+	   bits so we use L(copy_16_31).  */
+	testw	%cx, %cx
+	jz	L(copy_16_31)
+
+	testb	%cl, %cl
+	jz	L(copy_8_15)
+# ifdef USE_AS_WCSCPY
+	vmovd	%xmm0, (%rdi)
+	movl	$0, (%END_REG)
+	ret
 # else
-	add	$VEC_SIZE, %rsi
-	add	$VEC_SIZE, %rdi
-	jmp	L(CopyVecSizeExit)
-# endif
-
-	.p2align 4
-L(CopyVecSizeUnaligned_32):
-	bsf	%edx, %edx
-	vmovdqu %ymm4, (%rdi)
-	vmovdqu %ymm5, VEC_SIZE(%rdi)
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-# ifdef USE_AS_STPCPY
-	lea	(VEC_SIZE * 2)(%rdi, %rdx), %rax
-# endif
-	vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
-	add	$((VEC_SIZE * 2) - 1), %r8
-	sub	%rdx, %r8
-	lea	((VEC_SIZE * 2) + 1)(%rdi, %rdx), %rdi
-	jmp	L(StrncpyFillTailWithZero)
+	testb	$0x7, %cl
+	jz	L(copy_4_7)
+
+	testl	%edx, %edx
+	jz	L(set_null_term)
+	vmovd	%xmm0, %ecx
+	movw	%cx, (%rdi)
+
+	.p2align 4,, 2
+L(set_null_term):
+	movb	$0, (%END_REG)
+	ret
+
+	.p2align 4,, 12
+L(copy_4_7):
+	movl	-3(%rsi, %rdx), %ecx
+	vmovd	%xmm0, (%rdi)
+	movl	%ecx, -3(%END_REG)
+	ret
+# endif
+
+	.p2align 4,, 10
+L(copy_16_31):
+	VMOVU	-(16 - CHAR_SIZE)(%rsi, %rdx), %xmm1
+	VMOVU	%xmm0, (%rdi)
+	VMOVU	%xmm1, -(16 - CHAR_SIZE)(%END_REG)
+	ret
+
+	.p2align 4,, 10
+L(copy_8_15):
+# ifdef USE_AS_WCSCPY
+	movl	-(8 - CHAR_SIZE)(%rsi, %rdx), %ecx
 # else
-	add	$(VEC_SIZE * 2), %rsi
-	add	$(VEC_SIZE * 2), %rdi
-	jmp	L(CopyVecSizeExit)
-# endif
-
-# ifdef USE_AS_STRNCPY
-#  ifndef USE_AS_STRCAT
-	.p2align 4
-L(CopyVecSizeUnalignedVec6):
-	vmovdqu %ymm6, (%rdi, %rcx)
-	jmp	L(CopyVecSizeVecExit)
-
-	.p2align 4
-L(CopyVecSizeUnalignedVec5):
-	vmovdqu %ymm5, (%rdi, %rcx)
-	jmp	L(CopyVecSizeVecExit)
-
-	.p2align 4
-L(CopyVecSizeUnalignedVec4):
-	vmovdqu %ymm4, (%rdi, %rcx)
-	jmp	L(CopyVecSizeVecExit)
-
-	.p2align 4
-L(CopyVecSizeUnalignedVec3):
-	vmovdqu %ymm3, (%rdi, %rcx)
-	jmp	L(CopyVecSizeVecExit)
-#  endif
-
-/* Case2 */
-
-	.p2align 4
-L(CopyVecSizeCase2):
-	add	$VEC_SIZE, %r8
-	add	%rcx, %rdi
-	add	%rcx, %rsi
-	bsf	%edx, %edx
-	cmp	%r8d, %edx
-	jb	L(CopyVecSizeExit)
-	jmp	L(StrncpyExit)
-
-	.p2align 4
-L(CopyTwoVecSizeCase2):
-	add	%rcx, %rsi
-	bsf	%edx, %edx
-	add	$VEC_SIZE, %edx
-	sub	%ecx, %edx
-	cmp	%r8d, %edx
-	jb	L(CopyVecSizeExit)
-	jmp	L(StrncpyExit)
-
-L(CopyVecSizeTailCase2):
-	add	%rcx, %rsi
-	bsf	%edx, %edx
-	cmp	%r8d, %edx
-	jb	L(CopyVecSizeExit)
-	jmp	L(StrncpyExit)
-
-L(CopyVecSizeTail1Case2):
-	bsf	%edx, %edx
-	cmp	%r8d, %edx
-	jb	L(CopyVecSizeExit)
-	jmp	L(StrncpyExit)
-
-/* Case2 or Case3,  Case3 */
-
-	.p2align 4
-L(CopyVecSizeCase2OrCase3):
-	test	%rdx, %rdx
-	jnz	L(CopyVecSizeCase2)
-L(CopyVecSizeCase3):
-	add	$VEC_SIZE, %r8
-	add	%rcx, %rdi
-	add	%rcx, %rsi
-	jmp	L(StrncpyExit)
-
-	.p2align 4
-L(CopyTwoVecSizeCase2OrCase3):
-	test	%rdx, %rdx
-	jnz	L(CopyTwoVecSizeCase2)
-	add	%rcx, %rsi
-	jmp	L(StrncpyExit)
-
-	.p2align 4
-L(CopyVecSizeTailCase2OrCase3):
-	test	%rdx, %rdx
-	jnz	L(CopyVecSizeTailCase2)
-	add	%rcx, %rsi
-	jmp	L(StrncpyExit)
-
-	.p2align 4
-L(CopyTwoVecSize1Case2OrCase3):
-	add	$VEC_SIZE, %rdi
-	add	$VEC_SIZE, %rsi
-	sub	$VEC_SIZE, %r8
-L(CopyVecSizeTail1Case2OrCase3):
-	test	%rdx, %rdx
-	jnz	L(CopyVecSizeTail1Case2)
-	jmp	L(StrncpyExit)
-# endif
-
-/*------------End labels regarding with copying 1-VEC_SIZE bytes--and 1-(VEC_SIZE*2) bytes----*/
-
-	.p2align 4
-L(Exit1):
-	movzwl	(%rsi), %edx
-	mov	%dx, (%rdi)
-# ifdef USE_AS_STPCPY
-	lea	1(%rdi), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	$2, %r8
-	lea	2(%rdi), %rdi
-	jnz	L(StrncpyFillTailWithZero)
-# endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Exit2):
-	movzwl	(%rsi), %ecx
-	mov	%cx, (%rdi)
-	movb	$0, 2(%rdi)
-# ifdef USE_AS_STPCPY
-	lea	2(%rdi), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	$3, %r8
-	lea	3(%rdi), %rdi
-	jnz	L(StrncpyFillTailWithZero)
-# endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Exit3):
-	mov	(%rsi), %edx
-	mov	%edx, (%rdi)
+	movq	-(8 - CHAR_SIZE)(%rsi, %rdx), %rcx
+# endif
+	vmovq	%xmm0, (%rdi)
+	movq	%rcx, -(8 - CHAR_SIZE)(%END_REG)
+	ret
+
+
+	.p2align 4,, 8
+L(more_1x_vec):
+# if defined USE_AS_STPCPY || defined USE_AS_STRCAT
+	VMOVU	%VMM(0), (%rdi)
+# endif
+	subq	%rsi, %rdi
+	orq	$(VEC_SIZE - 1), %rsi
+	addq	%rsi, %rdi
+	VMOVA	1(%rsi), %VMM(1)
+
+	/* Try and order stores after as many loads as is reasonable to
+	   avoid potential false dependencies.  */
+# if !defined USE_AS_STPCPY && !defined USE_AS_STRCAT
+	VMOVU	%VMM(0), (%rax)
+# endif
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x1)
+
+	VMOVA	(VEC_SIZE + 1)(%rsi), %VMM(2)
+	VMOVU	%VMM(1), 1(%rdi)
+
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x2)
+
+	VMOVA	(VEC_SIZE * 2 + 1)(%rsi), %VMM(3)
+	VMOVU	%VMM(2), (VEC_SIZE + 1)(%rdi)
+
+	VPCMPEQ	%VMM(3), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x3)
+
+	VMOVA	(VEC_SIZE * 3 + 1)(%rsi), %VMM(4)
+	VMOVU	%VMM(3), (VEC_SIZE * 2 + 1)(%rdi)
+	VPCMPEQ	%VMM(4), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %edx
+	testl	%edx, %edx
+	jnz	L(ret_vec_x4)
+
+	VMOVU	%VMM(4), (VEC_SIZE * 3 + 1)(%rdi)
+
+	/* Subtract rsi from rdi before aligning. Adding back rsi will
+	   get proper rdi (dst) for new src.  */
+	subq	%rsi, %rdi
+	incq	%rsi
+	orq	$(VEC_SIZE * 4 - 1), %rsi
+
+	/* Do first half of loop ahead of time so loop can just start by
+	   storing.  */
+	VMOVA	(VEC_SIZE * 0 + 1)(%rsi), %VMM(0)
+	VMOVA	(VEC_SIZE * 1 + 1)(%rsi), %VMM(1)
+	VMOVA	(VEC_SIZE * 2 + 1)(%rsi), %VMM(2)
+	VMOVA	(VEC_SIZE * 3 + 1)(%rsi), %VMM(3)
+
+	VPMIN	%VMM(0), %VMM(1), %VMM(4)
+	VPMIN	%VMM(2), %VMM(3), %VMM(6)
+	VPMIN	%VMM(4), %VMM(6), %VMM(6)
+	VPCMPEQ	%VMM(6), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %edx
+	addq	%rsi, %rdi
+
+	testl	%edx, %edx
+	jnz	L(loop_4x_done)
+
+	.p2align 4,, 11
+L(loop_4x_vec):
+
+	VMOVU	%VMM(0), (VEC_SIZE * 0 + 1)(%rdi)
+	VMOVU	%VMM(1), (VEC_SIZE * 1 + 1)(%rdi)
+	subq	$(VEC_SIZE * -4), %rsi
+	VMOVU	%VMM(2), (VEC_SIZE * 2 + 1)(%rdi)
+	VMOVU	%VMM(3), (VEC_SIZE * 3 + 1)(%rdi)
+
+
+	VMOVA	(VEC_SIZE * 0 + 1)(%rsi), %VMM(0)
+	VMOVA	(VEC_SIZE * 1 + 1)(%rsi), %VMM(1)
+	VMOVA	(VEC_SIZE * 2 + 1)(%rsi), %VMM(2)
+	VMOVA	(VEC_SIZE * 3 + 1)(%rsi), %VMM(3)
+
+	VPMIN	%VMM(0), %VMM(1), %VMM(4)
+	VPMIN	%VMM(2), %VMM(3), %VMM(6)
+	VPMIN	%VMM(4), %VMM(6), %VMM(6)
+	VPCMPEQ	%VMM(6), %VZERO, %VMM(6)
+
+	vpmovmskb %VMM(6), %edx
+	subq	$(VEC_SIZE * -4), %rdi
+	testl	%edx, %edx
+	jz	L(loop_4x_vec)
+
+L(loop_4x_done):
+	VPCMPEQ	%VMM(0), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x1)
+	VMOVU	%VMM(0), (VEC_SIZE * 0 + 1)(%rdi)
+
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x2)
+	VMOVU	%VMM(1), (VEC_SIZE * 1 + 1)(%rdi)
+
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x3)
+	VMOVU	%VMM(2), (VEC_SIZE * 2 + 1)(%rdi)
+L(ret_vec_x4):
+	bsfl	%edx, %edx
+	VMOVU	((VEC_SIZE * 3 + 1)-(VEC_SIZE - CHAR_SIZE))(%rsi, %rdx), %VMM(1)
+	VMOVU	%VMM(1), ((VEC_SIZE * 3 + 1)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
 # ifdef USE_AS_STPCPY
-	lea	3(%rdi), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	$4, %r8
-	lea	4(%rdi), %rdi
-	jnz	L(StrncpyFillTailWithZero)
+	leaq	(VEC_SIZE * 3 + 1)(%rdx, %rdi), %rax
 # endif
+L(return_end):
 	VZEROUPPER_RETURN
 
-	.p2align 4
-L(Exit4_7):
-	mov	(%rsi), %ecx
-	mov	%ecx, (%rdi)
-	mov	-3(%rsi, %rdx), %ecx
-	mov	%ecx, -3(%rdi, %rdx)
+	.p2align 4,, 8
+L(ret_vec_x1):
+	bsfl	%ecx, %ecx
+	VMOVU	(1 -(VEC_SIZE - CHAR_SIZE))(%rsi, %rcx), %VMM(1)
+	VMOVU	%VMM(1), (1 -(VEC_SIZE - CHAR_SIZE))(%rdi, %rcx)
 # ifdef USE_AS_STPCPY
-	lea	(%rdi, %rdx), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	%rdx, %r8
-	sub	$1, %r8
-	lea	1(%rdi, %rdx), %rdi
-	jnz	L(StrncpyFillTailWithZero)
+	leaq	1(%rcx, %rdi), %rax
 # endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Exit8_15):
-	mov	(%rsi), %rcx
-	mov	-7(%rsi, %rdx), %r9
-	mov	%rcx, (%rdi)
-	mov	%r9, -7(%rdi, %rdx)
-# ifdef USE_AS_STPCPY
-	lea	(%rdi, %rdx), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	%rdx, %r8
-	sub	$1, %r8
-	lea	1(%rdi, %rdx), %rdi
-	jnz	L(StrncpyFillTailWithZero)
-# endif
-	VZEROUPPER_RETURN
+L(return_vzeroupper):
+	ZERO_UPPER_VEC_REGISTERS_RETURN
 
-	.p2align 4
-L(Exit16_31):
-	vmovdqu (%rsi), %xmm2
-	vmovdqu -15(%rsi, %rdx), %xmm3
-	vmovdqu %xmm2, (%rdi)
-	vmovdqu %xmm3, -15(%rdi, %rdx)
+	.p2align 4,, 8
+L(ret_vec_x2):
+	bsfl	%ecx, %ecx
+	VMOVU	((VEC_SIZE + 1)-(VEC_SIZE - CHAR_SIZE))(%rsi, %rcx), %VMM(1)
+	VMOVU	%VMM(1), ((VEC_SIZE + 1)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rcx)
 # ifdef USE_AS_STPCPY
-	lea	(%rdi, %rdx), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub %rdx, %r8
-	sub $1, %r8
-	lea 1(%rdi, %rdx), %rdi
-	jnz L(StrncpyFillTailWithZero)
+	leaq	(VEC_SIZE * 1 + 1)(%rcx, %rdi), %rax
 # endif
 	VZEROUPPER_RETURN
 
-	.p2align 4
-L(Exit32_63):
-	vmovdqu (%rsi), %ymm2
-	vmovdqu -31(%rsi, %rdx), %ymm3
-	vmovdqu %ymm2, (%rdi)
-	vmovdqu %ymm3, -31(%rdi, %rdx)
+	.p2align 4,, 8
+L(ret_vec_x3):
+	bsfl	%ecx, %ecx
+	VMOVU	((VEC_SIZE * 2 + 1)-(VEC_SIZE - CHAR_SIZE))(%rsi, %rcx), %VMM(1)
+	VMOVU	%VMM(1), ((VEC_SIZE * 2 + 1)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rcx)
 # ifdef USE_AS_STPCPY
-	lea	(%rdi, %rdx), %rax
-# endif
-# if defined USE_AS_STRNCPY && !defined USE_AS_STRCAT
-	sub	%rdx, %r8
-	sub	$1, %r8
-	lea	1(%rdi, %rdx), %rdi
-	jnz	L(StrncpyFillTailWithZero)
+	leaq	(VEC_SIZE * 2 + 1)(%rcx, %rdi), %rax
 # endif
 	VZEROUPPER_RETURN
 
-# ifdef USE_AS_STRNCPY
 
-	.p2align 4
-L(StrncpyExit1):
-	movzbl	(%rsi), %edx
-	mov	%dl, (%rdi)
+	.p2align 4,, 4
+L(page_cross):
+	movq	%rsi, %rcx
+	andq	$(VEC_SIZE * -1), %rcx
+
+	VPCMPEQ	(%rcx), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	shrxl	%esi, %ecx, %ecx
+# if USE_MOVSB_IN_PAGE_CROSS
+	/* Optimizing more aggressively for space as this is very cold
+	   code. This saves 2x cache lines.  */
+
+	/* This adds once to the later result which will get correct
+	   copy bounds. NB: this can never zero-out a non-zero RCX as
+	   to be in the page cross case rsi cannot be aligned and we
+	   already right-shift rcx by the misalignment.  */
+	shll	$CHAR_SIZE, %ecx
+	jz	L(page_cross_continue)
+	bsfl	%ecx, %ecx
+#  if !defined USE_AS_STPCPY && !defined USE_AS_STRCAT
+	movq	%rdi, %rax
+#  endif
+	rep	movsb
 #  ifdef USE_AS_STPCPY
-	lea	1(%rdi), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, 1(%rdi)
+	leaq	-CHAR_SIZE(%rdi), %rax
 #  endif
-	VZEROUPPER_RETURN
 
-	.p2align 4
-L(StrncpyExit2):
-	movzwl	(%rsi), %edx
-	mov	%dx, (%rdi)
-#  ifdef USE_AS_STPCPY
-	lea	2(%rdi), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, 2(%rdi)
-#  endif
 	VZEROUPPER_RETURN
 
-	.p2align 4
-L(StrncpyExit3_4):
-	movzwl	(%rsi), %ecx
-	movzwl	-2(%rsi, %r8), %edx
-	mov	%cx, (%rdi)
-	mov	%dx, -2(%rdi, %r8)
-#  ifdef USE_AS_STPCPY
-	lea	(%rdi, %r8), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, (%rdi, %r8)
-#  endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(StrncpyExit5_8):
-	mov	(%rsi), %ecx
-	mov	-4(%rsi, %r8), %edx
-	mov	%ecx, (%rdi)
-	mov	%edx, -4(%rdi, %r8)
-#  ifdef USE_AS_STPCPY
-	lea	(%rdi, %r8), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, (%rdi, %r8)
-#  endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(StrncpyExit9_16):
-	mov	(%rsi), %rcx
-	mov	-8(%rsi, %r8), %rdx
-	mov	%rcx, (%rdi)
-	mov	%rdx, -8(%rdi, %r8)
-#  ifdef USE_AS_STPCPY
-	lea	(%rdi, %r8), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, (%rdi, %r8)
-#  endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(StrncpyExit17_32):
-	vmovdqu (%rsi), %xmm2
-	vmovdqu -16(%rsi, %r8), %xmm3
-	vmovdqu %xmm2, (%rdi)
-	vmovdqu %xmm3, -16(%rdi, %r8)
-#  ifdef USE_AS_STPCPY
-	lea	(%rdi, %r8), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, (%rdi, %r8)
-#  endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(StrncpyExit33_64):
-	/*  0/32, 31/16 */
-	vmovdqu (%rsi), %ymm2
-	vmovdqu -VEC_SIZE(%rsi, %r8), %ymm3
-	vmovdqu %ymm2, (%rdi)
-	vmovdqu %ymm3, -VEC_SIZE(%rdi, %r8)
-#  ifdef USE_AS_STPCPY
-	lea	(%rdi, %r8), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, (%rdi, %r8)
-#  endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(StrncpyExit65):
-	/* 0/32, 32/32, 64/1 */
-	vmovdqu (%rsi), %ymm2
-	vmovdqu 32(%rsi), %ymm3
-	mov	64(%rsi), %cl
-	vmovdqu %ymm2, (%rdi)
-	vmovdqu %ymm3, 32(%rdi)
-	mov	%cl, 64(%rdi)
-#  ifdef USE_AS_STPCPY
-	lea	65(%rdi), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, 65(%rdi)
-#  endif
-	VZEROUPPER_RETURN
+# else
+	testl	%ecx, %ecx
+	jz	L(page_cross_continue)
 
+	/* Traditional copy case, essentially same as used in non-page-
+	   cross case but since we can't reuse VMM(0) we need twice as
+	   many loads from rsi.  */
 #  ifndef USE_AS_STRCAT
-
-	.p2align 4
-L(Fill1):
-	mov	%dl, (%rdi)
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Fill2):
-	mov	%dx, (%rdi)
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Fill3_4):
-	mov	%dx, (%rdi)
-	mov     %dx, -2(%rdi, %r8)
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Fill5_8):
-	mov	%edx, (%rdi)
-	mov     %edx, -4(%rdi, %r8)
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Fill9_16):
-	mov	%rdx, (%rdi)
-	mov	%rdx, -8(%rdi, %r8)
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(Fill17_32):
-	vmovdqu %xmmZ, (%rdi)
-	vmovdqu %xmmZ, -16(%rdi, %r8)
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(CopyVecSizeUnalignedVec2):
-	vmovdqu %ymm2, (%rdi, %rcx)
-
-	.p2align 4
-L(CopyVecSizeVecExit):
-	bsf	%edx, %edx
-	add	$(VEC_SIZE - 1), %r8
-	add	%rcx, %rdi
-#   ifdef USE_AS_STPCPY
-	lea	(%rdi, %rdx), %rax
-#   endif
-	sub	%rdx, %r8
-	lea	1(%rdi, %rdx), %rdi
-
-	.p2align 4
-L(StrncpyFillTailWithZero):
-	xor	%edx, %edx
-	sub	$VEC_SIZE, %r8
-	jbe	L(StrncpyFillExit)
-
-	vmovdqu %ymmZ, (%rdi)
-	add	$VEC_SIZE, %rdi
-
-	mov	%rdi, %rsi
-	and	$(VEC_SIZE - 1), %esi
-	sub	%rsi, %rdi
-	add	%rsi, %r8
-	sub	$(VEC_SIZE * 4), %r8
-	jb	L(StrncpyFillLessFourVecSize)
-
-L(StrncpyFillLoopVmovdqa):
-	vmovdqa %ymmZ, (%rdi)
-	vmovdqa %ymmZ, VEC_SIZE(%rdi)
-	vmovdqa %ymmZ, (VEC_SIZE * 2)(%rdi)
-	vmovdqa %ymmZ, (VEC_SIZE * 3)(%rdi)
-	add	$(VEC_SIZE * 4), %rdi
-	sub	$(VEC_SIZE * 4), %r8
-	jae	L(StrncpyFillLoopVmovdqa)
-
-L(StrncpyFillLessFourVecSize):
-	add	$(VEC_SIZE * 2), %r8
-	jl	L(StrncpyFillLessTwoVecSize)
-	vmovdqa %ymmZ, (%rdi)
-	vmovdqa %ymmZ, VEC_SIZE(%rdi)
-	add	$(VEC_SIZE * 2), %rdi
-	sub	$VEC_SIZE, %r8
-	jl	L(StrncpyFillExit)
-	vmovdqa %ymmZ, (%rdi)
-	add	$VEC_SIZE, %rdi
-	jmp	L(Fill)
-
-	.p2align 4
-L(StrncpyFillLessTwoVecSize):
-	add	$VEC_SIZE, %r8
-	jl	L(StrncpyFillExit)
-	vmovdqa %ymmZ, (%rdi)
-	add	$VEC_SIZE, %rdi
-	jmp	L(Fill)
-
-	.p2align 4
-L(StrncpyFillExit):
-	add	$VEC_SIZE, %r8
-L(Fill):
-	cmp	$17, %r8d
-	jae	L(Fill17_32)
-	cmp	$9, %r8d
-	jae	L(Fill9_16)
-	cmp	$5, %r8d
-	jae	L(Fill5_8)
-	cmp	$3, %r8d
-	jae	L(Fill3_4)
-	cmp	$1, %r8d
-	ja	L(Fill2)
-	je	L(Fill1)
-	VZEROUPPER_RETURN
-
-/* end of ifndef USE_AS_STRCAT */
+	xorl	%edx, %edx
 #  endif
-
-	.p2align 4
-L(UnalignedLeaveCase2OrCase3):
-	test	%rdx, %rdx
-	jnz	L(UnalignedFourVecSizeLeaveCase2)
-L(UnalignedFourVecSizeLeaveCase3):
-	lea	(VEC_SIZE * 4)(%r8), %rcx
-	and	$-VEC_SIZE, %rcx
-	add	$(VEC_SIZE * 3), %r8
-	jl	L(CopyVecSizeCase3)
-	vmovdqu %ymm4, (%rdi)
-	sub	$VEC_SIZE, %r8
-	jb	L(CopyVecSizeCase3)
-	vmovdqu %ymm5, VEC_SIZE(%rdi)
-	sub	$VEC_SIZE, %r8
-	jb	L(CopyVecSizeCase3)
-	vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
-	sub	$VEC_SIZE, %r8
-	jb	L(CopyVecSizeCase3)
-	vmovdqu %ymm7, (VEC_SIZE * 3)(%rdi)
+	bsfl	%ecx, %edx
 #  ifdef USE_AS_STPCPY
-	lea	(VEC_SIZE * 4)(%rdi), %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, (VEC_SIZE * 4)(%rdi)
+	leaq	(%rdi, %rdx), %rax
+#  elif !defined USE_AS_STRCAT
+	movq	%rdi, %rax
 #  endif
-	VZEROUPPER_RETURN
 
-	.p2align 4
-L(UnalignedFourVecSizeLeaveCase2):
-	xor	%ecx, %ecx
-	vpcmpeqb %ymm4, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	add	$(VEC_SIZE * 3), %r8
-	jle	L(CopyVecSizeCase2OrCase3)
-	test	%edx, %edx
-#  ifndef USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec4)
-#  else
-	jnz	L(CopyVecSize)
-#  endif
-	vpcmpeqb %ymm5, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	vmovdqu %ymm4, (%rdi)
-	add	$VEC_SIZE, %rcx
-	sub	$VEC_SIZE, %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-	test	%edx, %edx
-#  ifndef USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec5)
-#  else
-	jnz	L(CopyVecSize)
-#  endif
+	/* vzeroupper early to avoid duplicating at each return.  */
+	COND_VZEROUPPER
 
-	vpcmpeqb %ymm6, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	vmovdqu %ymm5, VEC_SIZE(%rdi)
-	add	$VEC_SIZE, %rcx
-	sub	$VEC_SIZE, %r8
-	jbe	L(CopyVecSizeCase2OrCase3)
-	test	%edx, %edx
-#  ifndef USE_AS_STRCAT
-	jnz	L(CopyVecSizeUnalignedVec6)
-#  else
-	jnz	L(CopyVecSize)
-#  endif
+	testw	%cx, %cx
+	jz	L(page_cross_copy_16_31)
 
-	vpcmpeqb %ymm7, %ymmZ, %ymmM
-	vpmovmskb %ymmM, %edx
-	vmovdqu %ymm6, (VEC_SIZE * 2)(%rdi)
-	lea	VEC_SIZE(%rdi, %rcx), %rdi
-	lea	VEC_SIZE(%rsi, %rcx), %rsi
-	bsf	%edx, %edx
-	cmp	%r8d, %edx
-	jb	L(CopyVecSizeExit)
-L(StrncpyExit):
-	cmp	$65, %r8d
-	je	L(StrncpyExit65)
-	cmp	$33, %r8d
-	jae	L(StrncpyExit33_64)
-	cmp	$17, %r8d
-	jae	L(StrncpyExit17_32)
-	cmp	$9, %r8d
-	jae	L(StrncpyExit9_16)
-	cmp	$5, %r8d
-	jae	L(StrncpyExit5_8)
-	cmp	$3, %r8d
-	jae	L(StrncpyExit3_4)
-	cmp	$1, %r8d
-	ja	L(StrncpyExit2)
-	je	L(StrncpyExit1)
-#  ifdef USE_AS_STPCPY
-	mov	%rdi, %rax
-#  endif
-#  ifdef USE_AS_STRCAT
-	movb	$0, (%rdi)
-#  endif
-	VZEROUPPER_RETURN
-
-	.p2align 4
-L(ExitZero):
-#  ifndef USE_AS_STRCAT
-	mov	%rdi, %rax
-#  endif
-	VZEROUPPER_RETURN
+	testb	%cl, %cl
+	jz	L(page_cross_copy_8_15)
 
-# endif
+	testl	$0x7, %cl
+	jz	L(page_cross_copy_4_7)
 
-# ifndef USE_AS_STRCAT
-END (STRCPY)
-# else
-END (STRCAT)
-# endif
+	testl	%edx, %edx
+	jz	L(page_cross_set_null_term)
+	movzwl	(%rsi), %ecx
+	movw	%cx, (%rdi)
+L(page_cross_set_null_term):
+	movb	$0, (%END_REG)
+	ret
+
+	.p2align 4,, 4
+L(page_cross_copy_4_7):
+	movl	(%rsi), %ecx
+	movl	-3(%rsi, %rdx), %esi
+	movl	%ecx, (%rdi)
+	movl	%esi, -3(%END_REG)
+	ret
+
+	.p2align 4,, 4
+L(page_cross_copy_8_15):
+	movq	(%rsi), %rcx
+	movq	-7(%rsi, %rdx), %rsi
+	movq	%rcx, (%rdi)
+	movq	%rsi, -7(%END_REG)
+	ret
+
+
+	.p2align 4,, 3
+L(page_cross_copy_16_31):
+	VMOVU	(%rsi), %xmm0
+	VMOVU	-15(%rsi, %rdx), %xmm1
+	VMOVU	%xmm0, (%rdi)
+	VMOVU	%xmm1, -15(%END_REG)
+	ret
+# endif
+
+END(STRCPY)
 #endif
diff --git a/sysdeps/x86_64/multiarch/strncat-avx2-rtm.S b/sysdeps/x86_64/multiarch/strncat-avx2-rtm.S
index 0dcea18..7272dee 100644
--- a/sysdeps/x86_64/multiarch/strncat-avx2-rtm.S
+++ b/sysdeps/x86_64/multiarch/strncat-avx2-rtm.S
@@ -1,3 +1,3 @@
-#define USE_AS_STRNCAT
-#define STRCAT __strncat_avx2_rtm
-#include "strcat-avx2-rtm.S"
+#define STRNCAT	__strncat_avx2_rtm
+#include "x86-avx-rtm-vecs.h"
+#include "strncat-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strncat-avx2.S b/sysdeps/x86_64/multiarch/strncat-avx2.S
index 52ecbca..ffa58bd 100644
--- a/sysdeps/x86_64/multiarch/strncat-avx2.S
+++ b/sysdeps/x86_64/multiarch/strncat-avx2.S
@@ -1,7 +1,419 @@
-#ifndef STRNCAT
-# define STRNCAT	__strncat_avx2
-#endif
+/* strncat with AVX2
+   Copyright (C) 2022 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <isa-level.h>
+
+#if ISA_SHOULD_BUILD (3)
+
+# include <sysdep.h>
+
+# ifndef VEC_SIZE
+#  include "x86-avx-vecs.h"
+# endif
+
+# ifndef STRNCAT
+#  define STRNCAT	__strncat_avx2
+# endif
+
+# ifdef USE_AS_WCSCPY
+#  define MOVCHAR	movl
+#  define VPCMPEQ	vpcmpeqd
+#  define VPMIN	vpminud
+#  define CHAR_SIZE	4
+# else
+#  define MOVCHAR	movb
+#  define VPCMPEQ	vpcmpeqb
+#  define VPMIN	vpminub
+#  define CHAR_SIZE	1
+# endif
+
+# include "strncpy-or-cat-overflow-def.h"
+
+# define PAGE_SIZE	4096
+
+# define VZERO	VMM(7)
+# define VZERO_128	VMM_128(7)
+
+	.section SECTION(.text), "ax", @progbits
+ENTRY(STRNCAT)
+	/* Filter zero length strings and very long strings.  Zero
+	   length strings just return, very long strings are handled by
+	   using the non-length variant {wcs|str}cat.  */
+	movq	%rdi, %rax
+# ifdef USE_AS_WCSCPY
+	leaq	-1(%rdx), %rcx
+	shr	$56, %rcx
+	jnz	L(zero_len)
+	salq	$2, %rdx
+# else
+	test	%rdx, %rdx
+	jl	L(zero_len)
+# endif
+	vpxor	%VZERO_128, %VZERO_128, %VZERO_128
+
+# include "strcat-strlen-avx2.h.S"
+
+	movl	%esi, %ecx
+	andl	$(PAGE_SIZE - 1), %ecx
+	cmpl	$(PAGE_SIZE - VEC_SIZE), %ecx
+	ja	L(page_cross)
+L(page_cross_continue):
+	VMOVU	(%rsi), %VMM(0)
+	VPCMPEQ	%VMM(0), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+
+	tzcnt	%ecx, %r8d
+	cmpq	%r8, %rdx
+	jbe	L(less_1x_vec)
+
+	testl	%ecx, %ecx
+	jz	L(more_1x_vec)
+
+	/* Hoist this to save code size.  */
+
+	movl	%r8d, %edx
+
+L(less_1x_vec):
+	COND_VZEROUPPER
+
+	cmpl	$16, %edx
+	jae	L(copy_16_31)
+	cmpl	$8, %edx
+	jae	L(copy_8_15)
+
+
+# ifdef USE_AS_WCSCPY
+	vmovd	%VMM_128(0), (%rdi)
+	MOVCHAR	$0, (%rdi, %rdx)
+	ret
+# else
+	cmpl	$4, %edx
+	jae	L(copy_4_7)
+
+	movzbl	(%rsi), %ecx
+	cmpl	$1, %edx
+	jbe	L(set_null_term)
+
+	/* NB: make this `vmovw` if support for AVX512-FP16 is added.
+	 */
+	movzwl	1(%rsi), %esi
+	movw	%si, 1(%rdi)
+
+	.p2align 4,, 1
+L(set_null_term):
+	movb	%cl, (%rdi)
+	MOVCHAR	$0, (%rdi, %rdx)
+	ret
+
+	.p2align 4,, 11
+L(copy_4_7):
+	movl	-(4)(%rsi, %rdx), %ecx
+	vmovd	%xmm0, (%rdi)
+	movl	%ecx, -(4)(%rdi, %rdx)
+	MOVCHAR	$0, (%rdi, %rdx)
+	ret
+# endif
+
+
+	.p2align 4,, 10
+L(copy_16_31):
+	VMOVU	-(16)(%rsi, %rdx), %xmm1
+	VMOVU	%xmm0, (%rdi)
+	VMOVU	%xmm1, -(16)(%rdi, %rdx)
+	MOVCHAR	$0, (%rdi, %rdx)
+	ret
+
+	.p2align 4,, 10
+L(copy_8_15):
+	movq	-(8)(%rsi, %rdx), %rcx
+	vmovq	%xmm0, (%rdi)
+	movq	%rcx, -(8)(%rdi, %rdx)
+	MOVCHAR	$0, (%rdi, %rdx)
+	ret
+
+	.p2align 4,, 8
+	.p2align 6,, 14
+L(more_1x_vec):
+	VMOVU	%VMM(0), (%rdi)
+
+	/* Align rsi (src) and just rdx/rdi (length/dst).  */
+	addq	%rsi, %rdx
+	subq	%rsi, %rdi
+	orq	$(VEC_SIZE - 1), %rsi
+	incq	%rsi
+	addq	%rsi, %rdi
+L(loop_last_4x_vec):
+	subq	%rsi, %rdx
+	VMOVA	0(%rsi), %VMM(1)
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	cmpq	$(VEC_SIZE * 2), %rdx
+	ja	L(more_2x_vec)
+L(last_2x_vec):
+	tzcnt	%ecx, %ecx
+	cmpl	%ecx, %edx
+	jbe	L(ret_vec_x1_len)
+
+	cmpl	$VEC_SIZE, %ecx
+	jnz	L(ret_vec_x1)
+
+	VMOVA	(VEC_SIZE * 1)(%rsi), %VMM(2)
+	VMOVU	%VMM(1), (%rdi)
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	addl	$-VEC_SIZE, %edx
+	bzhil	%edx, %ecx, %r8d
+	jz	L(ret_vec_x2_len)
+L(ret_vec_x2):
+	bsfl	%ecx, %edx
+L(ret_vec_x2_len):
+	VMOVU	(%rsi, %rdx), %VMM(0)
+	MOVCHAR	$0, (VEC_SIZE)(%rdi, %rdx)
+	VMOVU	%VMM(0), (%rdi, %rdx)
+L(return_vzeroupper):
+	ZERO_UPPER_VEC_REGISTERS_RETURN
+
+
+	.p2align 4,, 12
+L(ret_vec_x1_len):
+	movl	%edx, %ecx
+L(ret_vec_x1):
+	VMOVU	-(VEC_SIZE)(%rsi, %rcx), %VMM(1)
+	MOVCHAR	$0, (%rdi, %rcx)
+	VMOVU	%VMM(1), -VEC_SIZE(%rdi, %rcx)
+	VZEROUPPER_RETURN
+
+	.p2align 4,, 8
+L(last_4x_vec):
+	subq	$-(VEC_SIZE * 4), %rsi
+	VMOVA	0(%rsi), %VMM(1)
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	subq	$-(VEC_SIZE * 4), %rdi
+	addl	$-(VEC_SIZE * 4), %edx
+	cmpl	$(VEC_SIZE * 2), %edx
+	jbe	L(last_2x_vec)
+	.p2align 4,, 8
+L(more_2x_vec):
+	/* L(ret_vec_x1) expects ecx to have position of first match so
+	   test with bsf.  */
+	bsfl	%ecx, %ecx
+	jnz	L(ret_vec_x1)
+
+	VMOVA	(VEC_SIZE * 1)(%rsi), %VMM(2)
+	VMOVU	%VMM(1), (%rdi)
+
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x2)
+
 
-#define USE_AS_STRNCAT
-#define STRCAT	STRNCAT
-#include "strcat-avx2.S"
+	VMOVA	(VEC_SIZE * 2)(%rsi), %VMM(3)
+	VMOVU	%VMM(2), (VEC_SIZE * 1)(%rdi)
+
+	VPCMPEQ	%VMM(3), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+
+	/* Check if length is greater than 4x VEC.  */
+	cmpq	$(VEC_SIZE * 4), %rdx
+	ja	L(more_4x_vec)
+
+	addl	$(VEC_SIZE * -2), %edx
+
+	tzcnt	%ecx, %ecx
+	cmpl	%ecx, %edx
+	jbe	L(ret_vec_x3_len)
+
+	cmpl	$VEC_SIZE, %ecx
+	jnz	L(ret_vec_x3)
+
+	VMOVA	(VEC_SIZE * 3 + 0)(%rsi), %VMM(4)
+	VMOVU	%VMM(3), (VEC_SIZE * 2 + 0)(%rdi)
+	VPCMPEQ	%VMM(4), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	addl	$-VEC_SIZE, %edx
+	bzhil	%edx, %ecx, %r8d
+	jz	L(ret_vec_x4_len)
+L(ret_vec_x4):
+	bsfl	%ecx, %edx
+L(ret_vec_x4_len):
+	VMOVU	(VEC_SIZE * 2)(%rsi, %rdx), %VMM(0)
+	MOVCHAR	$0, (VEC_SIZE * 3)(%rdi, %rdx)
+	VMOVU	%VMM(0), (VEC_SIZE * 2)(%rdi, %rdx)
+	VZEROUPPER_RETURN
+
+	.p2align 4,, 4
+L(ret_vec_x3_len):
+	movl	%edx, %ecx
+L(ret_vec_x3):
+	VMOVU	(VEC_SIZE)(%rsi, %rcx), %VMM(0)
+	MOVCHAR	$0, (VEC_SIZE * 2)(%rdi, %rcx)
+	VMOVU	%VMM(0), (VEC_SIZE)(%rdi, %rcx)
+	VZEROUPPER_RETURN
+
+
+	.p2align 4,, 8
+L(more_4x_vec):
+	bsfl	%ecx, %ecx
+	jnz	L(ret_vec_x3)
+
+	VMOVA	(VEC_SIZE * 3)(%rsi), %VMM(4)
+	VMOVU	%VMM(3), (VEC_SIZE * 2)(%rdi)
+	VPCMPEQ	%VMM(4), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x4)
+
+	VMOVU	%VMM(4), (VEC_SIZE * 3)(%rdi)
+
+
+	/* Recheck length before aligning.  */
+	cmpq	$(VEC_SIZE * 8), %rdx
+	jbe	L(last_4x_vec)
+
+	/* Align rsi (src) and just rdx/rdi (length/dst).  */
+	addq	%rsi, %rdx
+	subq	%rsi, %rdi
+	subq	$-(VEC_SIZE * 4), %rsi
+	andq	$(VEC_SIZE * -4), %rsi
+
+	/* Do first half of loop ahead of time so loop can just start by
+	   storing.  */
+	VMOVA	(VEC_SIZE * 0 + 0)(%rsi), %VMM(0)
+	VMOVA	(VEC_SIZE * 1 + 0)(%rsi), %VMM(1)
+	VMOVA	(VEC_SIZE * 2 + 0)(%rsi), %VMM(2)
+	VMOVA	(VEC_SIZE * 3 + 0)(%rsi), %VMM(3)
+
+	VPMIN	%VMM(0), %VMM(1), %VMM(4)
+	VPMIN	%VMM(2), %VMM(3), %VMM(6)
+	VPMIN	%VMM(4), %VMM(6), %VMM(6)
+	VPCMPEQ	%VMM(6), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %r8d
+	addq	%rsi, %rdi
+	testl	%r8d, %r8d
+	jnz	L(loop_4x_done)
+
+	/* Use r9 for end of region before handling last 4x VEC
+	   specially.  */
+	leaq	-(VEC_SIZE * 4)(%rdx), %r9
+
+	.p2align 4,, 11
+L(loop_4x_vec):
+
+	VMOVU	%VMM(0), (VEC_SIZE * 0 + 0)(%rdi)
+	VMOVU	%VMM(1), (VEC_SIZE * 1 + 0)(%rdi)
+	subq	$(VEC_SIZE * -4), %rsi
+	VMOVU	%VMM(2), (VEC_SIZE * 2 + 0)(%rdi)
+	VMOVU	%VMM(3), (VEC_SIZE * 3 + 0)(%rdi)
+
+	subq	$(VEC_SIZE * -4), %rdi
+	cmpq	%rsi, %r9
+	jbe	L(loop_last_4x_vec)
+
+	VMOVA	(VEC_SIZE * 0 + 0)(%rsi), %VMM(0)
+	VMOVA	(VEC_SIZE * 1 + 0)(%rsi), %VMM(1)
+	VMOVA	(VEC_SIZE * 2 + 0)(%rsi), %VMM(2)
+	VMOVA	(VEC_SIZE * 3 + 0)(%rsi), %VMM(3)
+
+	VPMIN	%VMM(0), %VMM(1), %VMM(4)
+	VPMIN	%VMM(2), %VMM(3), %VMM(6)
+	VPMIN	%VMM(4), %VMM(6), %VMM(6)
+	VPCMPEQ	%VMM(6), %VZERO, %VMM(6)
+
+	vpmovmskb %VMM(6), %r8d
+
+	testl	%r8d, %r8d
+	jz	L(loop_4x_vec)
+
+L(loop_4x_done):
+	VPCMPEQ	%VMM(0), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	/* L(ret_vec_x1) expects ecx to have position of first match so
+	   test with bsf.  */
+	bsfl	%ecx, %ecx
+	jnz	L(ret_vec_x1)
+	VMOVU	%VMM(0), (VEC_SIZE * 0 + 0)(%rdi)
+
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x2)
+	VMOVU	%VMM(1), (VEC_SIZE * 1 + 0)(%rdi)
+
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	bsfl	%ecx, %ecx
+	jnz	L(ret_vec_x3)
+
+	VMOVU	%VMM(2), (VEC_SIZE * 2 + 0)(%rdi)
+	bsfl	%r8d, %r8d
+	VMOVU	(VEC_SIZE * 2 + CHAR_SIZE)(%rsi, %r8), %VMM(1)
+	VMOVU	%VMM(1), (VEC_SIZE * 2 + CHAR_SIZE)(%rdi, %r8)
+	VZEROUPPER_RETURN
+
+
+
+	.p2align 4,, 4
+L(page_cross):
+	movq	%rsi, %r8
+	andq	$(VEC_SIZE * -1), %r8
+
+	VPCMPEQ	(%r8), %VZERO, %VMM(6)
+
+	vpmovmskb %VMM(6), %ecx
+	shrxl	%esi, %ecx, %ecx
+
+	subl	%esi, %r8d
+	andl	$(VEC_SIZE - 1), %r8d
+	cmpq	%r8, %rdx
+	jb	L(page_cross_small)
+
+	/* Optimizing more aggressively for space as this is very cold
+	   code. This saves 2x cache lines.  */
+
+	/* This adds once to the later result which will get correct
+	   copy bounds. NB: this can never zero-out a non-zero RCX as
+	   to be in the page cross case rsi cannot be aligned and we
+	   already right-shift rcx by the misalignment.  */
+	shll	$CHAR_SIZE, %ecx
+	jz	L(page_cross_continue)
+	bsfl	%ecx, %ecx
+	rep	movsb
+	VZEROUPPER_RETURN
+
+L(page_cross_small):
+	tzcntl	%ecx, %ecx
+	jz	L(page_cross_setz)
+	cmpl	%edx, %ecx
+	cmova	%edx, %ecx
+	rep	movsb
+L(page_cross_setz):
+	MOVCHAR	$0, (%rdi)
+	VZEROUPPER_RETURN
+L(zero_len):
+# ifdef USE_AS_WCSCPY
+	test	%rdx, %rdx
+# endif
+	jnz	OVERFLOW_STRCAT
+	ret
+
+
+END(STRNCAT)
+#endif
diff --git a/sysdeps/x86_64/multiarch/strncpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/strncpy-avx2-rtm.S
index 79e7083..d42ad88 100644
--- a/sysdeps/x86_64/multiarch/strncpy-avx2-rtm.S
+++ b/sysdeps/x86_64/multiarch/strncpy-avx2-rtm.S
@@ -1,3 +1,3 @@
-#define USE_AS_STRNCPY
-#define STRCPY __strncpy_avx2_rtm
-#include "strcpy-avx2-rtm.S"
+#define STRNCPY	__strncpy_avx2_rtm
+#include "x86-avx-rtm-vecs.h"
+#include "strncpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strncpy-avx2.S b/sysdeps/x86_64/multiarch/strncpy-avx2.S
index ce634e9..e9afd8f 100644
--- a/sysdeps/x86_64/multiarch/strncpy-avx2.S
+++ b/sysdeps/x86_64/multiarch/strncpy-avx2.S
@@ -1,7 +1,735 @@
-#ifndef STRNCPY
-# define STRNCPY	__strncpy_avx2
-#endif
+/* strncpy with AVX2
+   Copyright (C) 2022 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <https://www.gnu.org/licenses/>.  */
+
+#include <isa-level.h>
+
+#if ISA_SHOULD_BUILD (3)
+
+# include <sysdep.h>
+
+
+# ifndef VEC_SIZE
+#  include "x86-avx-vecs.h"
+# endif
+
+# ifndef STRNCPY
+#  define STRNCPY	__strncpy_avx2
+# endif
+
+
+# ifdef USE_AS_WCSCPY
+#  define VPCMPEQ	vpcmpeqd
+#  define VPMIN	vpminud
+#  define CHAR_SIZE	4
+# else
+#  define VPCMPEQ	vpcmpeqb
+#  define VPMIN	vpminub
+#  define CHAR_SIZE	1
+# endif
+
+# include "strncpy-or-cat-overflow-def.h"
+
+# define PAGE_SIZE	4096
+
+# define VZERO	VMM(7)
+# define VZERO_128	VMM_128(7)
+
+
+	.section SECTION(.text), "ax", @progbits
+ENTRY(STRNCPY)
+	/* Filter zero length strings and very long strings.  Zero
+	   length strings just return, very long strings are handled by
+	   just running rep stos{b|l} to zero set (which will almost
+	   certainly segfault), if that succeeds then just calling
+	   OVERFLOW_STRCPY (strcpy, stpcpy, wcscpy, wcpcpy).  */
+# ifdef USE_AS_WCSCPY
+	decq	%rdx
+	movq	%rdx, %rax
+	/* 56 is end of max supported address space.  */
+	shr	$56, %rax
+	jnz	L(zero_len)
+	salq	$2, %rdx
+# else
+	decq	%rdx
+	/* `dec` can macrofuse with `jl`. If the flag needs to become
+	   `jb` replace `dec` with `sub`.  */
+	jl	L(zero_len)
+# endif
+
+	vpxor	%VZERO_128, %VZERO_128, %VZERO_128
+	movl	%esi, %eax
+	andl	$(PAGE_SIZE - 1), %eax
+	cmpl	$(PAGE_SIZE - VEC_SIZE), %eax
+	ja	L(page_cross)
+
+L(page_cross_continue):
+	VMOVU	(%rsi), %VMM(0)
+	VPCMPEQ	%VMM(0), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+
+	/* If no STPCPY just save end ahead of time.  */
+# ifndef USE_AS_STPCPY
+	movq	%rdi, %rax
+# elif defined USE_AS_WCSCPY
+	/* Clear dependency as nearly all return code for wcpncpy uses
+	   `setc %al`.  */
+	xorl	%eax, %eax
+# endif
+
+	cmpq	$(VEC_SIZE - CHAR_SIZE), %rdx
+	/* `jb` because length rdx is now length - CHAR_SIZE.  */
+	jbe	L(less_1x_vec)
+
+	/* This may overset but thats fine because we still need to zero
+	   fill.  */
+	VMOVU	%VMM(0), (%rdi)
+
+	testl	%ecx, %ecx
+	jnz	L(zfill)
+
+	/* Align.  */
+	addq	%rsi, %rdx
+	subq	%rsi, %rdi
+	orq	$(VEC_SIZE - 1), %rsi
+	incq	%rsi
+L(last_4x_vec):
+	addq	%rsi, %rdi
+L(loop_last_4x_vec):
+	subq	%rsi, %rdx
+
+
+	VMOVA	0(%rsi), %VMM(1)
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+
+	cmpq	$(VEC_SIZE * 2), %rdx
+	jae	L(more_2x_vec)
+
+	cmpl	$(VEC_SIZE), %edx
+	jb	L(ret_vec_x1_len)
+
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x1)
+
+	VPCMPEQ	VEC_SIZE(%rsi), %VZERO, %VMM(6)
+	VMOVU	%VMM(1), (%rdi)
+	vpmovmskb %VMM(6), %ecx
+	shlq	$VEC_SIZE, %rcx
+L(ret_vec_x1_len):
+	tzcntq	%rcx, %rcx
+	cmpl	%ecx, %edx
+	jbe	L(ret_vec_x1_len_no_zfill)
+	/* Fall through (expectation) is copy len < buffer len.  */
+	VMOVU	%VZERO, ((0)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
+L(ret_vec_x1_len_no_zfill_mov):
+	movl	%ecx, %edx
+# ifdef USE_AS_STPCPY
+	/* clear flags.  */
+	xorl	%ecx, %ecx
+# endif
+L(ret_vec_x1_len_no_zfill):
+	VMOVU	((0)-(VEC_SIZE - CHAR_SIZE))(%rsi, %rdx), %VMM(1)
+	VMOVU	%VMM(1), ((0)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
+# ifdef USE_AS_STPCPY
+#  ifdef USE_AS_WCSCPY
+	setc	%al
+	addq	%rdx, %rdi
+	leaq	(%rdi, %rax, CHAR_SIZE), %rax
+#  else
+	movl	%edx, %eax
+	adcq	%rdi, %rax
+#  endif
+# endif
+L(return_vzeroupper):
+	ZERO_UPPER_VEC_REGISTERS_RETURN
+
+	.p2align 4,, 6
+L(ret_vec_x1):
+	bsfl	%ecx, %ecx
+	VMOVU	%VZERO, ((0)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
+	subl	%ecx, %edx
+	/* Check if we need to reload/store.  */
+	cmpl	$VEC_SIZE, %edx
+	jb	L(ret_vec_x1_len_no_zfill_mov)
+	/* Otherwise safe to just store directly.  */
+	VMOVU	%VMM(1), (%rdi)
+	VMOVU	%VZERO, (%rdi, %rcx)
+# ifdef USE_AS_STPCPY
+	leaq	(%rdi, %rcx), %rax
+# endif
+	VZEROUPPER_RETURN
+
+	.p2align 4,, 12
+L(more_2x_vec):
+	VMOVU	%VMM(1), (%rdi)
+	testl	%ecx, %ecx
+	/* Must fill at least 2x VEC.  */
+	jnz	L(zfill_vec1)
+
+	VMOVA	VEC_SIZE(%rsi), %VMM(2)
+	VMOVU	%VMM(2), VEC_SIZE(%rdi)
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	/* Must fill at least 1x VEC.  */
+	jnz	L(zfill_vec2)
+
+	VMOVA	(VEC_SIZE * 2)(%rsi), %VMM(3)
+	VPCMPEQ	%VMM(3), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+
+	/* Check if len is more 4x VEC. -CHAR_SIZE because rdx is len -
+	   CHAR_SIZE.  */
+	cmpq	$(VEC_SIZE * 4 - CHAR_SIZE), %rdx
+	ja	L(more_4x_vec)
+
+	subl	$(VEC_SIZE * 3), %edx
+	jb	L(ret_vec_x3_len)
+
+	testl	%ecx, %ecx
+	jnz	L(ret_vec_x3)
+
+	VPCMPEQ	(VEC_SIZE * 3)(%rsi), %VZERO, %VMM(6)
+	VMOVU	%VMM(3), (VEC_SIZE * 2)(%rdi)
+	vpmovmskb %VMM(6), %ecx
+	tzcntl	%ecx, %ecx
+	cmpl	%ecx, %edx
+	jbe	L(ret_vec_x4_len_no_zfill)
+	/* Fall through (expectation) is copy len < buffer len.  */
+	VMOVU	%VZERO, ((VEC_SIZE * 3)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
+	movl	%ecx, %edx
+L(ret_vec_x4_len_no_zfill):
+	VMOVU	((VEC_SIZE * 3)-(VEC_SIZE - CHAR_SIZE))(%rsi, %rdx), %VMM(1)
+	VMOVU	%VMM(1), ((VEC_SIZE * 3)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
+# ifdef USE_AS_STPCPY
+#  ifdef USE_AS_WCSCPY
+	setc	%al
+	addq	%rdx, %rdi
+	leaq	(VEC_SIZE * 3)(%rdi, %rax, CHAR_SIZE), %rax
+#  else
+	leal	(VEC_SIZE * 3 + 0)(%edx), %eax
+	adcq	%rdi, %rax
+#  endif
+# endif
+	VZEROUPPER_RETURN
+
+
+L(ret_vec_x3_len):
+	addl	$(VEC_SIZE * 1), %edx
+	tzcntl	%ecx, %ecx
+	cmpl	%ecx, %edx
+	jbe	L(ret_vec_x3_len_no_zfill)
+	/* Fall through (expectation) is copy len < buffer len.  */
+	VMOVU	%VZERO, ((VEC_SIZE * 2)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
+L(ret_vec_x3_len_no_zfill_mov):
+	movl	%ecx, %edx
+# ifdef USE_AS_STPCPY
+	/* clear flags.  */
+	xorl	%ecx, %ecx
+# endif
+	.p2align 4,, 4
+L(ret_vec_x3_len_no_zfill):
+	VMOVU	((VEC_SIZE * 2)-(VEC_SIZE - CHAR_SIZE))(%rsi, %rdx), %VMM(1)
+	VMOVU	%VMM(1), ((VEC_SIZE * 2)-(VEC_SIZE - CHAR_SIZE))(%rdi, %rdx)
+# ifdef USE_AS_STPCPY
+#  ifdef USE_AS_WCSCPY
+	setc	%al
+	addq	%rdx, %rdi
+	leaq	(VEC_SIZE * 2)(%rdi, %rax, CHAR_SIZE), %rax
+#  else
+	leal	(VEC_SIZE * 2 + 0)(%rdx), %eax
+	adcq	%rdi, %rax
+#  endif
+# endif
+	VZEROUPPER_RETURN
+
+
+	.p2align 4,, 8
+L(ret_vec_x3):
+	bsfl	%ecx, %ecx
+	VMOVU	%VZERO, (VEC_SIZE * 3 +(-(VEC_SIZE - CHAR_SIZE)))(%rdi, %rdx)
+	subl	%ecx, %edx
+	jl	L(ret_vec_x3_len_no_zfill_mov)
+	VMOVU	%VMM(3), (VEC_SIZE * 2)(%rdi)
+	VMOVU	%VZERO, (VEC_SIZE * 2)(%rdi, %rcx)
+# ifdef USE_AS_STPCPY
+	leaq	(VEC_SIZE * 2)(%rdi, %rcx), %rax
+# endif
+	VZEROUPPER_RETURN
+
+	.p2align 4,, 8
+L(more_4x_vec):
+
+	VMOVU	%VMM(3), (VEC_SIZE * 2)(%rdi)
+	testl	%ecx, %ecx
+	jnz	L(zfill_vec3)
+
+	VMOVA	(VEC_SIZE * 3)(%rsi), %VMM(4)
+	VMOVU	%VMM(4), (VEC_SIZE * 3)(%rdi)
+	VPCMPEQ	%VMM(4), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(zfill_vec4)
+
+	movq	%rdx, %rcx
+	addq	%rsi, %rdx
+	subq	%rsi, %rdi
+	subq	$-(VEC_SIZE * 4), %rsi
+	/* Recheck length before aligning.  */
+	cmpq	$(VEC_SIZE * 8 - CHAR_SIZE), %rcx
+	jbe	L(last_4x_vec)
+
+	andq	$(VEC_SIZE * -4), %rsi
+
+	/* Do first half of loop ahead of time so loop can just start by
+	   storing.  */
+	VMOVA	(VEC_SIZE * 0 + 0)(%rsi), %VMM(0)
+	VMOVA	(VEC_SIZE * 1 + 0)(%rsi), %VMM(1)
+	VMOVA	(VEC_SIZE * 2 + 0)(%rsi), %VMM(2)
+	VMOVA	(VEC_SIZE * 3 + 0)(%rsi), %VMM(3)
+
+	VPMIN	%VMM(0), %VMM(1), %VMM(4)
+	VPMIN	%VMM(2), %VMM(3), %VMM(6)
+	VPMIN	%VMM(4), %VMM(6), %VMM(6)
+	VPCMPEQ	%VMM(6), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %r8d
+	addq	%rsi, %rdi
+	testl	%r8d, %r8d
+	jnz	L(loop_4x_done)
+
+	/* Use r9 as end register.  */
+	leaq	-(VEC_SIZE * 4 - CHAR_SIZE)(%rdx), %r9
 
-#define USE_AS_STRNCPY
-#define STRCPY	STRNCPY
-#include "strcpy-avx2.S"
+	.p2align 4,, 11
+L(loop_4x_vec):
+
+	VMOVU	%VMM(0), (VEC_SIZE * 0 + 0)(%rdi)
+	VMOVU	%VMM(1), (VEC_SIZE * 1 + 0)(%rdi)
+	subq	$(VEC_SIZE * -4), %rsi
+	VMOVU	%VMM(2), (VEC_SIZE * 2 + 0)(%rdi)
+	VMOVU	%VMM(3), (VEC_SIZE * 3 + 0)(%rdi)
+
+	subq	$(VEC_SIZE * -4), %rdi
+	cmpq	%rsi, %r9
+	jbe	L(loop_last_4x_vec)
+
+	VMOVA	(VEC_SIZE * 0 + 0)(%rsi), %VMM(0)
+	VMOVA	(VEC_SIZE * 1 + 0)(%rsi), %VMM(1)
+	VMOVA	(VEC_SIZE * 2 + 0)(%rsi), %VMM(2)
+	VMOVA	(VEC_SIZE * 3 + 0)(%rsi), %VMM(3)
+
+	VPMIN	%VMM(0), %VMM(1), %VMM(4)
+	VPMIN	%VMM(2), %VMM(3), %VMM(6)
+	VPMIN	%VMM(4), %VMM(6), %VMM(6)
+	VPCMPEQ	%VMM(6), %VZERO, %VMM(6)
+
+	vpmovmskb %VMM(6), %r8d
+
+	testl	%r8d, %r8d
+	jz	L(loop_4x_vec)
+
+L(loop_4x_done):
+	subq	%rsi, %rdx
+	VMOVU	%VMM(0), (VEC_SIZE * 0 + 0)(%rdi)
+	VPCMPEQ	%VMM(0), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(zfill_vec1)
+
+	VMOVU	%VMM(1), (VEC_SIZE * 1 + 0)(%rdi)
+	VPCMPEQ	%VMM(1), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(zfill_vec2)
+
+	VMOVU	%VMM(2), (VEC_SIZE * 2 + 0)(%rdi)
+	VPCMPEQ	%VMM(2), %VZERO, %VMM(6)
+	vpmovmskb %VMM(6), %ecx
+	testl	%ecx, %ecx
+	jnz	L(zfill_vec3)
+
+	VMOVU	%VMM(3), (VEC_SIZE * 3 + 0)(%rdi)
+	movl	%r8d, %ecx
+
+	// Zfill more....
+
+	.p2align 4,, 4
+L(zfill_vec4):
+	addq	$(VEC_SIZE * 2), %rdi
+	subq	$(VEC_SIZE * 2), %rdx
+L(zfill_vec2):
+	shlq	$VEC_SIZE, %rcx
+L(zfill):
+	bsfq	%rcx, %rcx
+	subq	%rcx, %rdx
+	addq	%rcx, %rdi
+# ifdef USE_AS_STPCPY
+	movq	%rdi, %rax
+# endif
+L(zfill_from_page_cross):
+	cmpq	$VEC_SIZE, %rdx
+	jb	L(zfill_less_vec_vzeroupper)
+
+L(zfill_more_1x_vec):
+	VMOVU	%VZERO, CHAR_SIZE(%rdi)
+	VMOVU	%VZERO, (CHAR_SIZE - VEC_SIZE)(%rdi, %rdx)
+	cmpq	$(VEC_SIZE * 2), %rdx
+	jae	L(zfill_more_2x_vec)
+L(zfill_done0):
+	VZEROUPPER_RETURN
+
+	.p2align 4,, 8
+L(zfill_vec3):
+	addq	$(VEC_SIZE * 2), %rdi
+	subq	$(VEC_SIZE * 2), %rdx
+	.p2align 4,, 2
+L(zfill_vec1):
+	bsfl	%ecx, %ecx
+	addq	%rcx, %rdi
+	subq	%rcx, %rdx
+# ifdef USE_AS_STPCPY
+	movq	%rdi, %rax
+# endif
+	/* zfill from vec1/vec3 must have to set at least 2x VECS.  */
+
+	VMOVU	%VZERO, CHAR_SIZE(%rdi)
+	VMOVU	%VZERO, (CHAR_SIZE - VEC_SIZE)(%rdi, %rdx)
+	cmpq	$(VEC_SIZE * 2), %rdx
+	jb	L(zfill_done0)
+L(zfill_more_2x_vec):
+	VMOVU	%VZERO, (CHAR_SIZE - VEC_SIZE * 2)(%rdi, %rdx)
+	VMOVU	%VZERO, (VEC_SIZE + CHAR_SIZE)(%rdi)
+	subq	$(VEC_SIZE * 4 - CHAR_SIZE), %rdx
+	jbe	L(zfill_done)
+
+	addq	%rdi, %rdx
+	VMOVU	%VZERO, (VEC_SIZE * 2 + CHAR_SIZE)(%rdi)
+	VMOVU	%VZERO, (VEC_SIZE * 3 + CHAR_SIZE)(%rdi)
+
+
+	VMOVU	%VZERO, (VEC_SIZE * 0 + 0)(%rdx)
+	VMOVU	%VZERO, (VEC_SIZE * 1 + 0)(%rdx)
+
+	subq	$-(VEC_SIZE * 4 + CHAR_SIZE), %rdi
+	cmpq	%rdi, %rdx
+	jbe	L(zfill_done)
+
+	andq	$-(VEC_SIZE), %rdi
+	.p2align 4,, 12
+L(zfill_loop_4x_vec):
+	VMOVA	%VZERO, (VEC_SIZE * 0)(%rdi)
+	VMOVA	%VZERO, (VEC_SIZE * 1)(%rdi)
+	VMOVA	%VZERO, (VEC_SIZE * 2)(%rdi)
+	VMOVA	%VZERO, (VEC_SIZE * 3)(%rdi)
+	subq	$-(VEC_SIZE * 4), %rdi
+	cmpq	%rdi, %rdx
+	ja	L(zfill_loop_4x_vec)
+L(zfill_done):
+	VZEROUPPER_RETURN
+
+
+	.p2align 4,, 8
+L(copy_1x):
+	VMOVU	%VMM(0), (%rdi)
+	testl	%ecx, %ecx
+	jz	L(ret_32_32)
+L(zfill_less_vec):
+	bsfl	%ecx, %ecx
+L(zfill_less_vec_no_bsf):
+	subq	%rcx, %rdx
+	addq	%rcx, %rdi
+# ifdef USE_AS_STPCPY
+	movq	%rdi, %rax
+# endif
+L(zfill_less_vec_vzeroupper):
+	COND_VZEROUPPER
+	/* We are taking advantage of the fact that to be here we must
+	   be writing null-term as (%rdi, %rcx) we have a byte of lee-
+	   way for overwriting.  */
+	cmpl	$16, %edx
+	jb	L(zfill_less_16)
+	VMOVU	%VZERO_128, (%rdi)
+	VMOVU	%VZERO_128, -(16 - CHAR_SIZE)(%rdi, %rdx)
+	ret
+# ifdef USE_AS_STPCPY
+L(ret_32_32):
+	leaq	CHAR_SIZE(%rdi, %rdx), %rax
+	VZEROUPPER_RETURN
+# endif
+
+	.p2align 4,, 4
+L(copy_16_31):
+	/* Overfill to avoid branches.  */
+	vmovdqu	-(16 - CHAR_SIZE)(%rsi, %rdx), %xmm1
+	vmovdqu	%xmm0, (%rdi)
+	vmovdqu	%xmm1, -(16 - CHAR_SIZE)(%rdi, %rdx)
+	cmpl	%ecx, %edx
+	ja	L(zfill_less_vec_no_bsf)
+# ifndef USE_AS_STPCPY
+L(ret_32_32):
+# else
+#  ifdef USE_AS_WCSCPY
+	setc	%al
+	addq	%rdx, %rdi
+	leaq	(%rdi, %rax, CHAR_SIZE), %rax
+#  else
+	movl	%edx, %eax
+	adcq	%rdi, %rax
+#  endif
+# endif
+	VZEROUPPER_RETURN
+
+	.p2align 4,, 4
+L(copy_8_15):
+	/* Overfill to avoid branches.  */
+	movq	-(8 - CHAR_SIZE)(%rsi, %rdx), %rsi
+	vmovq	%xmm0, (%rdi)
+	movq	%rsi, -(8 - CHAR_SIZE)(%rdi, %rdx)
+	cmpl	%ecx, %edx
+	jbe	L(ret_8_15)
+	subq	%rcx, %rdx
+	addq	%rcx, %rdi
+# ifdef USE_AS_STPCPY
+	movq	%rdi, %rax
+# endif
+	.p2align 4,, 8
+L(zfill_less_16):
+	xorl	%ecx, %ecx
+	cmpl	$8, %edx
+	jb	L(zfill_less_8)
+	movq	%rcx, (%rdi)
+	movq	%rcx, -(8 - CHAR_SIZE)(%rdi, %rdx)
+# ifndef USE_AS_STPCPY
+L(ret_8_15):
+# endif
+	ret
+
+
+	.p2align 4,, 8
+L(less_1x_vec):
+	/* Reuse flag from `cmp $VEC_SIZE, %rdx`. The idea is many
+	   buffer sizes are aligned conventially.  */
+	je	L(copy_1x)
+
+	tzcntl	%ecx, %ecx
+	cmpl	$16, %edx
+	jae	L(copy_16_31)
+
+	COND_VZEROUPPER
+	cmpl	$8, %edx
+	jae	L(copy_8_15)
+# ifdef USE_AS_WCSCPY
+	testl	%ecx, %ecx
+	jz	L(zfill_less_8_set_ret)
+
+	movl	(%rsi, %rdx), %esi
+	vmovd	%xmm0, (%rdi)
+	movl	%esi, (%rdi, %rdx)
+
+#  ifdef USE_AS_STPCPY
+	cmpl	%ecx, %edx
+L(ret_8_15):
+	setc	%al
+	addq	%rdx, %rdi
+	leaq	(%rdi, %rax, CHAR_SIZE), %rax
+#  endif
+	ret
+L(zfill_less_8_set_ret):
+	xorl	%ecx, %ecx
+#  ifdef USE_AS_STPCPY
+	movq	%rdi, %rax
+#  endif
+L(zfill_less_8):
+	movl	%ecx, (%rdi)
+	movl	%ecx, (%rdi, %rdx)
+	ret
+
+# else
+	cmpl	$3, %edx
+	jb	L(copy_0_3)
+	/* Overfill to avoid branches.  */
+	movl	-3(%rsi, %rdx), %esi
+	vmovd	%xmm0, (%rdi)
+	movl	%esi, -3(%rdi, %rdx)
+	cmpl	%ecx, %edx
+	jbe	L(ret_4_7)
+	subq	%rcx, %rdx
+	addq	%rcx, %rdi
+#  ifdef USE_AS_STPCPY
+	movq	%rdi, %rax
+#  endif
+	xorl	%ecx, %ecx
+	.p2align 4,, 8
+L(zfill_less_8):
+	cmpl	$3, %edx
+	jb	L(zfill_less_3)
+	movl	%ecx, (%rdi)
+	movl	%ecx, -3(%rdi, %rdx)
+#  ifdef USE_AS_STPCPY
+	ret
+#  endif
+
+L(ret_4_7):
+#  ifdef USE_AS_STPCPY
+L(ret_8_15):
+	movl	%edx, %eax
+	adcq	%rdi, %rax
+#  endif
+	ret
+
+	.p2align 4,, 4
+L(zfill_less_3):
+	testl	%edx, %edx
+	jz	L(zfill_1)
+	movw	%cx, (%rdi)
+L(zfill_1):
+	movb	%cl, (%rdi, %rdx)
+	ret
+
+	.p2align 4,, 8
+L(copy_0_3):
+	vmovd	%xmm0, %r8d
+	testl	%edx, %edx
+	jz	L(copy_1)
+	movw	%r8w, (%rdi)
+	cmpl	%ecx, %edx
+	ja	L(zfill_from_1)
+	movzbl	(%rsi, %rdx), %r8d
+#  ifdef USE_AS_STPCPY
+	movl	%edx, %eax
+	adcq	%rdi, %rax
+	movb	%r8b, (%rdi, %rdx)
+	ret
+#  endif
+
+L(copy_1):
+#  ifdef USE_AS_STPCPY
+	movl	%edx, %eax
+	cmpl	%ecx, %edx
+	adcq	%rdi, %rax
+#  endif
+#  ifdef USE_AS_WCSCPY
+	vmovd	%xmm0, (%rdi)
+#  else
+	movb	%r8b, (%rdi, %rdx)
+#  endif
+	ret
+# endif
+
+	.p2align 4,, 2
+L(zero_len):
+	movq	%rdi, %rax
+	ret
+# ifndef USE_AS_WCSCPY
+	.p2align 4,, 8
+L(zfill_from_1):
+#  ifdef USE_AS_STPCPY
+	leaq	(%rdi, %rcx), %rax
+#  endif
+	movw	$0, -1(%rdi, %rdx)
+	ret
+# endif
+
+	.p2align 4,, 4
+	.p2align 6,, 8
+L(page_cross):
+	movq	%rsi, %rax
+	andq	$(VEC_SIZE * -1), %rax
+
+	VPCMPEQ	(%rax), %VZERO, %VMM(6)
+
+	vpmovmskb %VMM(6), %ecx
+	shrxl	%esi, %ecx, %ecx
+
+	subl	%esi, %eax
+	andl	$(VEC_SIZE - 1), %eax
+	cmpq	%rax, %rdx
+	jb	L(page_cross_small)
+	/* Optimizing more aggressively for space as this is very cold
+	   code. This saves 2x cache lines.  */
+
+	/* If rcx is non-zero then continue.  */
+	shl	$CHAR_SIZE, %ecx
+	jz	L(page_cross_continue)
+	bsf	%ecx, %ecx
+
+	subq	%rcx, %rdx
+# ifdef USE_AS_STPCPY
+	leaq	-CHAR_SIZE(%rdi, %rcx), %rax
+# else
+	movq	%rdi, %rax
+# endif
+
+	rep	movsb
+# ifdef USE_AS_WCSCPY
+	movl	$0, (%rdi)
+# else
+	movb	$0, (%rdi)
+# endif
+	jmp	L(zfill_from_page_cross)
+
+L(page_cross_small):
+	tzcntl	%ecx, %ecx
+	xorl	%eax, %eax
+	cmpl	%ecx, %edx
+	jbe	L(page_cross_copy_only)
+
+	/* Do a zfill of the tail before copying.  */
+	movq	%rdi, %r9
+	movl	%ecx, %r8d
+
+	subl	%ecx, %edx
+	leaq	CHAR_SIZE(%rdi, %rcx), %rdi
+	movl	%edx, %ecx
+	rep	stosb
+	movq	%r9, %rdi
+	movl	%r8d, %edx
+L(page_cross_copy_only):
+	leal	CHAR_SIZE(%rdx), %ecx
+# ifdef USE_AS_STPCPY
+#  ifdef USE_AS_WCSCPY
+	setc	%al
+	addq	%rdi, %rdx
+	leaq	(%rdx, %rax, CHAR_SIZE), %rax
+#  else
+	movl	%edx, %eax
+	adcq	%rdi, %rax
+#  endif
+# else
+	movq	%rdi, %rax
+# endif
+	rep	movsb
+	ret
+
+
+L(best_effort_strncpy):
+	movq	%rdx, %rcx
+	xorl	%eax, %eax
+	movq	%rdi, %r8
+	/* The length is >= 2^63. We very much so expect to segfault at
+	   rep stos. If that doesn't happen then just strcpy to finish.
+	 */
+# ifdef USE_AS_WCSCPY
+	rep	stosl
+# else
+	rep	stosb
+# endif
+	movq	%r8, %rdi
+	jmp	OVERFLOW_STRCPY
+END(STRNCPY)
+#endif
diff --git a/sysdeps/x86_64/multiarch/x86-avx-vecs.h b/sysdeps/x86_64/multiarch/x86-avx-vecs.h
index dca1089..275af75 100644
--- a/sysdeps/x86_64/multiarch/x86-avx-vecs.h
+++ b/sysdeps/x86_64/multiarch/x86-avx-vecs.h
@@ -27,7 +27,8 @@
 #define VEC_SIZE			32
 #include "x86-vec-macros.h"
 
-#define USE_WITH_AVX		1
+#define USE_WITH_AVX2		1
+
 #define SECTION(p)			p##.avx
 
 /* 4-byte mov instructions with AVX2.  */