X86-64: Remove previous default/SSE2/AVX2 memcpy/memmove

Since the new SSE2/AVX2 memcpy/memmove are faster than the previous ones,
we can remove the previous SSE2/AVX2 memcpy/memmove and replace them with
the new ones.

No change in IFUNC selection if SSE2 and AVX2 memcpy/memmove weren't used
before.  If SSE2 or AVX2 memcpy/memmove were used, the new SSE2 or AVX2
memcpy/memmove optimized with Enhanced REP MOVSB will be used for
processors with ERMS.  The new AVX512 memcpy/memmove will be used for
processors with AVX512 which prefer vzeroupper.

Since the new SSE2 memcpy/memmove are faster than the previous default
memcpy/memmove used in libc.a and ld.so, we also remove the previous
default memcpy/memmove and make them the default memcpy/memmove, except
that non-temporal store isn't used in ld.so.

Together, it reduces the size of libc.so by about 6 KB and the size of
ld.so by about 2 KB.

	[BZ #19776]
	* sysdeps/x86_64/memcpy.S: Make it dummy.
	* sysdeps/x86_64/mempcpy.S: Likewise.
	* sysdeps/x86_64/memmove.S: New file.
	* sysdeps/x86_64/memmove_chk.S: Likewise.
	* sysdeps/x86_64/multiarch/memmove.S: Likewise.
	* sysdeps/x86_64/multiarch/memmove_chk.S: Likewise.
	* sysdeps/x86_64/memmove.c: Removed.
	* sysdeps/x86_64/multiarch/memcpy-avx-unaligned.S: Likewise.
	* sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S: Likewise.
	* sysdeps/x86_64/multiarch/memmove-avx-unaligned.S: Likewise.
	* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
	Likewise.
	* sysdeps/x86_64/multiarch/memmove.c: Likewise.
	* sysdeps/x86_64/multiarch/memmove_chk.c: Likewise.
	* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Remove
	memcpy-sse2-unaligned, memmove-avx-unaligned,
	memcpy-avx-unaligned and memmove-sse2-unaligned-erms.
	* sysdeps/x86_64/multiarch/ifunc-impl-list.c
	(__libc_ifunc_impl_list): Replace
	__memmove_chk_avx512_unaligned_2 with
	__memmove_chk_avx512_unaligned.  Remove
	__memmove_chk_avx_unaligned_2.  Replace
	__memmove_chk_sse2_unaligned_2 with
	__memmove_chk_sse2_unaligned.  Remove __memmove_chk_sse2 and
	__memmove_avx_unaligned_2.  Replace __memmove_avx512_unaligned_2
	with __memmove_avx512_unaligned.  Replace
	__memmove_sse2_unaligned_2 with __memmove_sse2_unaligned.
	Remove __memmove_sse2.  Replace __memcpy_chk_avx512_unaligned_2
	with __memcpy_chk_avx512_unaligned.  Remove
	__memcpy_chk_avx_unaligned_2.  Replace
	__memcpy_chk_sse2_unaligned_2 with __memcpy_chk_sse2_unaligned.
	Remove __memcpy_chk_sse2.  Remove __memcpy_avx_unaligned_2.
	Replace __memcpy_avx512_unaligned_2 with
	__memcpy_avx512_unaligned.  Remove __memcpy_sse2_unaligned_2
	and __memcpy_sse2.  Replace __mempcpy_chk_avx512_unaligned_2
	with __mempcpy_chk_avx512_unaligned.  Remove
	__mempcpy_chk_avx_unaligned_2.  Replace
	__mempcpy_chk_sse2_unaligned_2 with
	__mempcpy_chk_sse2_unaligned.  Remove __mempcpy_chk_sse2.
	Replace __mempcpy_avx512_unaligned_2 with
	__mempcpy_avx512_unaligned.  Remove __mempcpy_avx_unaligned_2.
	Replace __mempcpy_sse2_unaligned_2 with
	__mempcpy_sse2_unaligned.  Remove __mempcpy_sse2.
	* sysdeps/x86_64/multiarch/memcpy.S (__new_memcpy): Support
	__memcpy_avx512_unaligned_erms and __memcpy_avx512_unaligned.
	Use __memcpy_avx_unaligned_erms and __memcpy_sse2_unaligned_erms
	if processor has ERMS.  Default to __memcpy_sse2_unaligned.
	(ENTRY): Removed.
	(END): Likewise.
	(ENTRY_CHK): Likewise.
	(libc_hidden_builtin_def): Likewise.
	Don't include ../memcpy.S.
	* sysdeps/x86_64/multiarch/memcpy_chk.S (__memcpy_chk): Support
	__memcpy_chk_avx512_unaligned_erms and
	__memcpy_chk_avx512_unaligned.  Use
	__memcpy_chk_avx_unaligned_erms and
	__memcpy_chk_sse2_unaligned_erms if if processor has ERMS.
	Default to __memcpy_chk_sse2_unaligned.
	* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S
	Change function suffix from unaligned_2 to unaligned.
	* sysdeps/x86_64/multiarch/mempcpy.S (__mempcpy): Support
	__mempcpy_avx512_unaligned_erms and __mempcpy_avx512_unaligned.
	Use __mempcpy_avx_unaligned_erms and __mempcpy_sse2_unaligned_erms
	if processor has ERMS.  Default to __mempcpy_sse2_unaligned.
	(ENTRY): Removed.
	(END): Likewise.
	(ENTRY_CHK): Likewise.
	(libc_hidden_builtin_def): Likewise.
	Don't include ../mempcpy.S.
	(mempcpy): New.  Add a weak alias.
	* sysdeps/x86_64/multiarch/mempcpy_chk.S (__mempcpy_chk): Support
	__mempcpy_chk_avx512_unaligned_erms and
	__mempcpy_chk_avx512_unaligned.  Use
	__mempcpy_chk_avx_unaligned_erms and
	__mempcpy_chk_sse2_unaligned_erms if if processor has ERMS.
	Default to __mempcpy_chk_sse2_unaligned.

19 files changed, 394 insertions, 1490 deletions

diff --git a/sysdeps/x86_64/memcpy.S b/sysdeps/x86_64/memcpy.S
index f6e3d93..d98500a 100644
--- a/sysdeps/x86_64/memcpy.S
+++ b/sysdeps/x86_64/memcpy.S
@@ -1,584 +1 @@
-/*
-   Optimized memcpy for x86-64.
-
-   Copyright (C) 2007-2016 Free Software Foundation, Inc.
-   Contributed by Evandro Menezes <evandro.menezes@amd.com>, 2007.
-
-   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
-   <http://www.gnu.org/licenses/>.
-*/
-
-#include <sysdep.h>
-#include "asm-syntax.h"
-
-/* Stack slots in the red-zone. */
-
-#ifdef USE_AS_MEMPCPY
-#  define RETVAL	(0)
-#else
-#  define RETVAL	(-8)
-#  if defined SHARED && !defined USE_MULTIARCH && IS_IN (libc)
-#    define memcpy	__memcpy
-#    undef libc_hidden_builtin_def
-#    define libc_hidden_builtin_def(name) \
-	.globl __GI_memcpy; __GI_memcpy = __memcpy
-#  endif
-#endif
-#define SAVE0	(RETVAL - 8)
-#define SAVE1	(SAVE0	- 8)
-#define SAVE2	(SAVE1	- 8)
-#define SAVE3	(SAVE2	- 8)
-
-        .text
-
-#if defined PIC && IS_IN (libc)
-ENTRY_CHK (__memcpy_chk)
-
-	cmpq	%rdx, %rcx
-	jb	HIDDEN_JUMPTARGET (__chk_fail)
-
-END_CHK (__memcpy_chk)
-#endif
-
-ENTRY(memcpy)				/* (void *, const void*, size_t) */
-
-/* Handle tiny blocks. */
-
-L(1try):				/* up to 32B */
-	cmpq	$32, %rdx
-#ifndef USE_AS_MEMPCPY
-	movq	%rdi, %rax		/* save return value */
-#endif
-	jae	L(1after)
-
-L(1):					/* 1-byte once */
-	testb	$1, %dl
-	jz	L(1a)
-
-	movzbl	(%rsi),	%ecx
-	movb	%cl, (%rdi)
-
-	incq	%rsi
-	incq	%rdi
-
-	.p2align 4,, 4
-
-L(1a):					/* 2-byte once */
-	testb	$2, %dl
-	jz	L(1b)
-
-	movzwl	(%rsi),	%ecx
-	movw	%cx, (%rdi)
-
-	addq	$2, %rsi
-	addq	$2, %rdi
-
-	.p2align 4,, 4
-
-L(1b):					/* 4-byte once */
-	testb	$4, %dl
-	jz	L(1c)
-
-	movl	(%rsi),	%ecx
-	movl	%ecx, (%rdi)
-
-	addq	$4, %rsi
-	addq	$4, %rdi
-
-	.p2align 4,, 4
-
-L(1c):					/* 8-byte once */
-	testb	$8, %dl
-	jz	L(1d)
-
-	movq	(%rsi), %rcx
-	movq	%rcx, (%rdi)
-
-	addq	$8, %rsi
-	addq	$8, %rdi
-
-	.p2align 4,, 4
-
-L(1d):					/* 16-byte loop */
-	andl	$0xf0, %edx
-	jz	L(exit)
-
-	.p2align 4
-
-L(1loop):
-	movq	 (%rsi), %rcx
-	movq	8(%rsi), %r8
-	movq	%rcx,  (%rdi)
-	movq	 %r8, 8(%rdi)
-
-	subl	$16, %edx
-
-	leaq	16(%rsi), %rsi
-	leaq	16(%rdi), %rdi
-
-	jnz	L(1loop)
-
-	.p2align 4,, 4
-
-L(exit):				/* exit */
-#ifdef USE_AS_MEMPCPY
-	movq	%rdi, %rax		/* return value */
-#else
-	rep
-#endif
-	retq
-
-	.p2align 4
-
-L(1after):
-#ifndef USE_AS_MEMPCPY
-	movq	%rax, RETVAL(%rsp)	/* save return value */
-#endif
-
-/* Align to the natural word size. */
-
-L(aligntry):
-	movl	%esi, %ecx      	/* align by source */
-
-	andl	$7, %ecx
-	jz	L(alignafter)  		/* already aligned */
-
-L(align):		      		/* align */
-	leaq	-8(%rcx, %rdx), %rdx	/* calculate remaining bytes */
-	subl	$8, %ecx
-
-	.p2align 4
-
-L(alignloop):				/* 1-byte alignment loop */
-	movzbl	(%rsi), %eax
-	movb	%al, (%rdi)
-
-	incl	%ecx
-
-	leaq	1(%rsi), %rsi
-	leaq	1(%rdi), %rdi
-
-	jnz	L(alignloop)
-
-	.p2align 4
-
-L(alignafter):
-
-/* Handle mid-sized blocks. */
-
-L(32try):				/* up to 1KB */
-	cmpq	$1024, %rdx
-	ja	L(32after)
-
-L(32):					/* 32-byte loop */
-	movl	%edx, %ecx
-	shrl	$5, %ecx
-	jz	L(32skip)
-
-	.p2align 4
-
-L(32loop):
-	decl	%ecx
-
-	movq	  (%rsi), %rax
-	movq	 8(%rsi), %r8
-	movq	16(%rsi), %r9
-	movq	24(%rsi), %r10
-
-	movq	%rax,   (%rdi)
-	movq	 %r8,  8(%rdi)
-	movq	 %r9, 16(%rdi)
-	movq	%r10, 24(%rdi)
-
-	leaq	32(%rsi), %rsi
-	leaq	32(%rdi), %rdi
-
-	jz	L(32skip)		/* help out smaller blocks */
-
-	decl	%ecx
-
-	movq	  (%rsi), %rax
-	movq	 8(%rsi), %r8
-	movq	16(%rsi), %r9
-	movq	24(%rsi), %r10
-
-	movq	%rax,   (%rdi)
-	movq	 %r8,  8(%rdi)
-	movq	 %r9, 16(%rdi)
-	movq	%r10, 24(%rdi)
-
-	leaq	32(%rsi), %rsi
-	leaq	32(%rdi), %rdi
-
-	jnz	L(32loop)
-
-	.p2align 4
-
-L(32skip):
-	andl	$31, %edx		/* check for left overs */
-#ifdef USE_AS_MEMPCPY
-	jnz	L(1)
-
-	movq	%rdi, %rax
-#else
-	movq	RETVAL(%rsp), %rax
-	jnz	L(1)
-
-	rep
-#endif
-	retq				/* exit */
-
-	.p2align 4
-
-L(32after):
-
-/*
-	In order to minimize code-size in RTLD, algorithms specific for
-	larger blocks are excluded when building for RTLD.
-*/
-
-/* Handle blocks smaller than 1/2 L1. */
-
-L(fasttry):				/* first 1/2 L1 */
-#if IS_IN (libc)			/* only up to this algorithm outside of libc.so */
-	mov	__x86_data_cache_size_half(%rip), %R11_LP
-	cmpq	%rdx, %r11		/* calculate the smaller of */
-	cmovaq	%rdx, %r11		/* remaining bytes and 1/2 L1 */
-#endif
-
-L(fast):				/* good ol' MOVS */
-#if IS_IN (libc)
-	movq	%r11, %rcx
-	andq	$-8, %r11
-#else
-	movq	%rdx, %rcx
-#endif
-	shrq	$3, %rcx
-	jz	L(fastskip)
-
-	rep
-	movsq
-
-	.p2align 4,, 4
-
-L(fastskip):
-#if IS_IN (libc)
-	subq	%r11, %rdx		/* check for more */
-	testq	$-8, %rdx
-	jnz	L(fastafter)
-#endif
-
-	andl	$7, %edx		/* check for left overs */
-#ifdef USE_AS_MEMPCPY
-	jnz	L(1)
-
-	movq	%rdi, %rax
-#else
-	movq	RETVAL(%rsp), %rax
-	jnz	L(1)
-
-	rep
-#endif
-	retq				/* exit */
-
-#if IS_IN (libc)			/* none of the algorithms below for RTLD */
-
-	.p2align 4
-
-L(fastafter):
-
-/* Handle large blocks smaller than 1/2 L2. */
-
-L(pretry):				/* first 1/2 L2 */
-	mov	__x86_shared_cache_size_half (%rip), %R8_LP
-	cmpq	%rdx, %r8		/* calculate the lesser of */
-	cmovaq	%rdx, %r8		/* remaining bytes and 1/2 L2 */
-
-L(pre):					/* 64-byte with prefetching */
-	movq	%r8, %rcx
-	andq	$-64, %r8
-	shrq	$6, %rcx
-	jz	L(preskip)
-
-	movq	%r14, SAVE0(%rsp)
-	cfi_rel_offset (%r14, SAVE0)
-	movq	%r13, SAVE1(%rsp)
-	cfi_rel_offset (%r13, SAVE1)
-	movq	%r12, SAVE2(%rsp)
-	cfi_rel_offset (%r12, SAVE2)
-	movq	%rbx, SAVE3(%rsp)
-	cfi_rel_offset (%rbx, SAVE3)
-
-	cmpl	$0, __x86_prefetchw(%rip)
-	jz	L(preloop)		/* check if PREFETCHW OK */
-
-	.p2align 4
-
-/* ... when PREFETCHW is available (less cache-probe traffic in MP systems). */
-
-L(prewloop):				/* cache-line in state M */
-	decq	%rcx
-
-	movq	   (%rsi), %rax
-	movq	 8 (%rsi), %rbx
-	movq	16 (%rsi), %r9
-	movq	24 (%rsi), %r10
-	movq	32 (%rsi), %r11
-	movq	40 (%rsi), %r12
-	movq	48 (%rsi), %r13
-	movq	56 (%rsi), %r14
-
-	prefetcht0	 0 + 896 (%rsi)
-	prefetcht0	64 + 896 (%rsi)
-
-	movq	%rax,   (%rdi)
-	movq	%rbx,  8(%rdi)
-	movq	 %r9, 16(%rdi)
-	movq	%r10, 24(%rdi)
-	movq	%r11, 32(%rdi)
-	movq	%r12, 40(%rdi)
-	movq	%r13, 48(%rdi)
-	movq	%r14, 56(%rdi)
-
-	leaq	64(%rsi), %rsi
-	leaq	64(%rdi), %rdi
-
-	jz	L(prebail)
-
-	decq	%rcx
-
-	movq	  (%rsi), %rax
-	movq	 8(%rsi), %rbx
-	movq	16(%rsi), %r9
-	movq	24(%rsi), %r10
-	movq	32(%rsi), %r11
-	movq	40(%rsi), %r12
-	movq	48(%rsi), %r13
-	movq	56(%rsi), %r14
-
-	movq	%rax,   (%rdi)
-	movq	%rbx,  8(%rdi)
-	movq	 %r9, 16(%rdi)
-	movq	%r10, 24(%rdi)
-	movq	%r11, 32(%rdi)
-	movq	%r12, 40(%rdi)
-	movq	%r13, 48(%rdi)
-	movq	%r14, 56(%rdi)
-
-	prefetchw	896 - 64(%rdi)
-	prefetchw	896 -  0(%rdi)
-
-	leaq	64(%rsi), %rsi
-	leaq	64(%rdi), %rdi
-
-	jnz	L(prewloop)
-	jmp	L(prebail)
-
-	.p2align 4
-
-/* ... when PREFETCHW is not available. */
-
-L(preloop):				/* cache-line in state E */
-	decq	%rcx
-
-	movq	  (%rsi), %rax
-	movq	 8(%rsi), %rbx
-	movq	16(%rsi), %r9
-	movq	24(%rsi), %r10
-	movq	32(%rsi), %r11
-	movq	40(%rsi), %r12
-	movq	48(%rsi), %r13
-	movq	56(%rsi), %r14
-
-	prefetcht0	896 +  0(%rsi)
-	prefetcht0	896 + 64(%rsi)
-
-	movq	%rax,   (%rdi)
-	movq	%rbx,  8(%rdi)
-	movq	 %r9, 16(%rdi)
-	movq	%r10, 24(%rdi)
-	movq	%r11, 32(%rdi)
-	movq	%r12, 40(%rdi)
-	movq	%r13, 48(%rdi)
-	movq	%r14, 56(%rdi)
-
-	leaq	64 (%rsi), %rsi
-	leaq	64 (%rdi), %rdi
-
-	jz	L(prebail)
-
-	decq	%rcx
-
-	movq	  (%rsi), %rax
-	movq	 8(%rsi), %rbx
-	movq	16(%rsi), %r9
-	movq	24(%rsi), %r10
-	movq	32(%rsi), %r11
-	movq	40(%rsi), %r12
-	movq	48(%rsi), %r13
-	movq	56(%rsi), %r14
-
-	prefetcht0	896 - 64(%rdi)
-	prefetcht0	896 -  0(%rdi)
-
-	movq	%rax,   (%rdi)
-	movq	%rbx,  8(%rdi)
-	movq	 %r9, 16(%rdi)
-	movq	%r10, 24(%rdi)
-	movq	%r11, 32(%rdi)
-	movq	%r12, 40(%rdi)
-	movq	%r13, 48(%rdi)
-	movq	%r14, 56(%rdi)
-
-	leaq	64(%rsi), %rsi
-	leaq	64(%rdi), %rdi
-
-	jnz	L(preloop)
-
-L(prebail):
-	movq	SAVE3(%rsp), %rbx
-	cfi_restore (%rbx)
-	movq	SAVE2(%rsp), %r12
-	cfi_restore (%r12)
-	movq	SAVE1(%rsp), %r13
-	cfi_restore (%r13)
-	movq	SAVE0(%rsp), %r14
-	cfi_restore (%r14)
-
-/*       .p2align 4 */
-
-L(preskip):
-	subq	%r8, %rdx		/* check for more */
-	testq	$-64, %rdx
-	jnz	L(preafter)
-
-	andl	$63, %edx		/* check for left overs */
-#ifdef USE_AS_MEMPCPY
-	jnz	L(1)
-
-	movq	%rdi, %rax
-#else
-	movq	RETVAL(%rsp), %rax
-	jnz	L(1)
-
-	rep
-#endif
-	retq				/* exit */
-
-	.p2align 4
-
-L(preafter):
-
-/* Handle huge blocks. */
-
-L(NTtry):
-
-L(NT):					/* non-temporal 128-byte */
-	movq	%rdx, %rcx
-	shrq	$7, %rcx
-	jz	L(NTskip)
-
-	movq	%r14, SAVE0(%rsp)
-	cfi_rel_offset (%r14, SAVE0)
-	movq	%r13, SAVE1(%rsp)
-	cfi_rel_offset (%r13, SAVE1)
-	movq	%r12, SAVE2(%rsp)
-	cfi_rel_offset (%r12, SAVE2)
-
-       .p2align 4
-
-L(NTloop):
-	prefetchnta	768(%rsi)
-	prefetchnta	832(%rsi)
-
-	decq	%rcx
-
-	movq	  (%rsi), %rax
-	movq	 8(%rsi), %r8
-	movq	16(%rsi), %r9
-	movq	24(%rsi), %r10
-	movq	32(%rsi), %r11
-	movq	40(%rsi), %r12
-	movq	48(%rsi), %r13
-	movq	56(%rsi), %r14
-
-	movntiq	%rax,   (%rdi)
-	movntiq	 %r8,  8(%rdi)
-	movntiq	 %r9, 16(%rdi)
-	movntiq	%r10, 24(%rdi)
-	movntiq	%r11, 32(%rdi)
-	movntiq	%r12, 40(%rdi)
-	movntiq	%r13, 48(%rdi)
-	movntiq	%r14, 56(%rdi)
-
-	movq	 64(%rsi), %rax
-	movq	 72(%rsi), %r8
-	movq	 80(%rsi), %r9
-	movq	 88(%rsi), %r10
-	movq	 96(%rsi), %r11
-	movq	104(%rsi), %r12
-	movq	112(%rsi), %r13
-	movq	120(%rsi), %r14
-
-	movntiq	%rax,  64(%rdi)
-	movntiq	 %r8,  72(%rdi)
-	movntiq	 %r9,  80(%rdi)
-	movntiq	%r10,  88(%rdi)
-	movntiq	%r11,  96(%rdi)
-	movntiq	%r12, 104(%rdi)
-	movntiq	%r13, 112(%rdi)
-	movntiq	%r14, 120(%rdi)
-
-	leaq	128(%rsi), %rsi
-	leaq	128(%rdi), %rdi
-
-	jnz	L(NTloop)
-
-	sfence				/* serialize memory stores */
-
-	movq	SAVE2(%rsp), %r12
-	cfi_restore (%r12)
-	movq	SAVE1(%rsp), %r13
-	cfi_restore (%r13)
-	movq	SAVE0(%rsp), %r14
-	cfi_restore (%r14)
-
-L(NTskip):
-	andl	$127, %edx		/* check for left overs */
-#ifdef USE_AS_MEMPCPY
-	jnz	L(1)
-
-	movq	%rdi, %rax
-#else
-	movq	RETVAL(%rsp), %rax
-	jnz	L(1)
-
-	rep
-#endif
-	retq				/* exit */
-
-#endif /* IS_IN (libc) */
-
-END(memcpy)
-
-#ifndef USE_AS_MEMPCPY
-libc_hidden_builtin_def (memcpy)
-# if defined SHARED && !defined USE_MULTIARCH && IS_IN (libc)
-#  undef memcpy
-#  include <shlib-compat.h>
-versioned_symbol (libc, __memcpy, memcpy, GLIBC_2_14);
-# endif
-#endif
+/* Implemented in memcpy.S.  */
diff --git a/sysdeps/x86_64/memmove.S b/sysdeps/x86_64/memmove.S
new file mode 100644
index 0000000..a7ae453
--- /dev/null
+++ b/sysdeps/x86_64/memmove.S
@@ -0,0 +1,71 @@
+/* Optimized memmove for x86-64.
+   Copyright (C) 2016 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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+
+#define VEC_SIZE	16
+#define VEC(i)		xmm##i
+#define PREFETCHNT	prefetchnta
+#define VMOVNT		movntdq
+/* Use movups and movaps for smaller code sizes.  */
+#define VMOVU		movups
+#define VMOVA		movaps
+
+#define SECTION(p)		p
+
+#ifdef USE_MULTIARCH
+# if !defined SHARED || !IS_IN (libc)
+#  define MEMCPY_SYMBOL(p,s)		memcpy
+# endif
+#else
+# if defined SHARED && IS_IN (libc)
+#  define MEMCPY_SYMBOL(p,s)		__memcpy
+# else
+#  define MEMCPY_SYMBOL(p,s)		memcpy
+# endif
+#endif
+#if !defined SHARED || !defined USE_MULTIARCH || !IS_IN (libc)
+# define MEMPCPY_SYMBOL(p,s)		__mempcpy
+#endif
+#ifndef MEMMOVE_SYMBOL
+# define MEMMOVE_CHK_SYMBOL(p,s)	p
+# define MEMMOVE_SYMBOL(p,s)		memmove
+#endif
+
+#include "multiarch/memmove-vec-unaligned-erms.S"
+
+#ifndef USE_MULTIARCH
+libc_hidden_builtin_def (memmove)
+# if defined SHARED && IS_IN (libc)
+strong_alias (memmove, __memcpy)
+libc_hidden_ver (memmove, memcpy)
+# endif
+libc_hidden_def (__mempcpy)
+weak_alias (__mempcpy, mempcpy)
+libc_hidden_builtin_def (mempcpy)
+
+# if defined SHARED && IS_IN (libc)
+#  undef memcpy
+#  include <shlib-compat.h>
+versioned_symbol (libc, __memcpy, memcpy, GLIBC_2_14);
+
+#  if SHLIB_COMPAT (libc, GLIBC_2_2_5, GLIBC_2_14)
+compat_symbol (libc, memmove, memcpy, GLIBC_2_2_5);
+#  endif
+# endif
+#endif
diff --git a/sysdeps/x86_64/memmove.c b/sysdeps/x86_64/memmove_chk.S
index 07f8185..ee154f1 100644
--- a/sysdeps/x86_64/memmove.c
+++ b/sysdeps/x86_64/memmove_chk.S
@@ -1,4 +1,5 @@
-/* Copyright (C) 2011-2016 Free Software Foundation, Inc.
+/* Checking memmove for x86-64.
+   Copyright (C) 2016 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
@@ -15,12 +16,18 @@
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
-#include "string/memmove.c"
+#include <sysdep.h>
+#include "asm-syntax.h"
 
-#if !defined memmove && IS_IN (libc)
-#include <shlib-compat.h>
-
-#if SHLIB_COMPAT (libc, GLIBC_2_2_5, GLIBC_2_14)
-compat_symbol (libc, memmove, memcpy, GLIBC_2_2_5);
-#endif
+#ifndef SHARED
+	/* For libc.so this is defined in memmove.S.
+	   For libc.a, this is a separate source to avoid
+	   memmove bringing in __chk_fail and all routines
+	   it calls.  */
+        .text
+ENTRY (__memmove_chk)
+	cmpq	%rdx, %rcx
+	jb	__chk_fail
+	jmp	memmove
+END (__memmove_chk)
 #endif
diff --git a/sysdeps/x86_64/mempcpy.S b/sysdeps/x86_64/mempcpy.S
index acee5e5..d98500a 100644
--- a/sysdeps/x86_64/mempcpy.S
+++ b/sysdeps/x86_64/mempcpy.S
@@ -1,8 +1 @@
-#define USE_AS_MEMPCPY
-#define memcpy __mempcpy
-#define __memcpy_chk __mempcpy_chk
-#include <sysdeps/x86_64/memcpy.S>
-
-libc_hidden_def (__mempcpy)
-weak_alias (__mempcpy, mempcpy)
-libc_hidden_builtin_def (mempcpy)
+/* Implemented in memcpy.S.  */
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index d78e667..3736f54 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -6,10 +6,9 @@ ifeq ($(subdir),string)
 
 sysdep_routines += strncat-c stpncpy-c strncpy-c strcmp-ssse3 \
 		   strcmp-sse2-unaligned strncmp-ssse3 \
-		   memcmp-sse4 memcpy-ssse3 memcpy-sse2-unaligned \
+		   memcmp-sse4 memcpy-ssse3 \
 		   memmove-ssse3 \
-		   memcpy-ssse3-back memmove-avx-unaligned \
-		   memcpy-avx-unaligned \
+		   memcpy-ssse3-back \
 		   memmove-ssse3-back \
 		   memmove-avx512-no-vzeroupper strcasecmp_l-ssse3 \
 		   strncase_l-ssse3 strcat-ssse3 strncat-ssse3\
@@ -20,7 +19,6 @@ sysdep_routines += strncat-c stpncpy-c strncpy-c strcmp-ssse3 \
 		   strchr-sse2-no-bsf memcmp-ssse3 strstr-sse2-unaligned \
 		   strcspn-c strpbrk-c strspn-c varshift \
 		   memset-avx512-no-vzeroupper \
-		   memmove-sse2-unaligned-erms \
 		   memmove-avx-unaligned-erms \
 		   memmove-avx512-unaligned-erms \
 		   memset-avx2-unaligned-erms \
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index ca05ff6..449b046 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -54,7 +54,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memmove_chk_avx512_no_vzeroupper)
 	      IFUNC_IMPL_ADD (array, i, __memmove_chk,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
-			      __memmove_chk_avx512_unaligned_2)
+			      __memmove_chk_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __memmove_chk,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
 			      __memmove_chk_avx512_unaligned_erms)
@@ -64,9 +64,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memmove_chk_avx_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __memmove_chk,
 			      HAS_ARCH_FEATURE (AVX_Usable),
-			      __memmove_chk_avx_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, __memmove_chk,
-			      HAS_ARCH_FEATURE (AVX_Usable),
 			      __memmove_chk_avx_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, __memmove_chk,
 			      HAS_CPU_FEATURE (SSSE3),
@@ -75,11 +72,9 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      HAS_CPU_FEATURE (SSSE3),
 			      __memmove_chk_ssse3)
 	      IFUNC_IMPL_ADD (array, i, __memmove_chk, 1,
-			      __memmove_chk_sse2_unaligned_2)
+			      __memmove_chk_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __memmove_chk, 1,
-			      __memmove_chk_sse2_unaligned_erms)
-	      IFUNC_IMPL_ADD (array, i, __memmove_chk, 1,
-			      __memmove_chk_sse2))
+			      __memmove_chk_sse2_unaligned_erms))
 
   /* Support sysdeps/x86_64/multiarch/memmove.S.  */
   IFUNC_IMPL (i, name, memmove,
@@ -88,9 +83,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memmove_avx_unaligned)
 	      IFUNC_IMPL_ADD (array, i, memmove,
 			      HAS_ARCH_FEATURE (AVX_Usable),
-			      __memmove_avx_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, memmove,
-			      HAS_ARCH_FEATURE (AVX_Usable),
 			      __memmove_avx_unaligned_erms)
 #ifdef HAVE_AVX512_ASM_SUPPORT
 	      IFUNC_IMPL_ADD (array, i, memmove,
@@ -98,7 +90,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memmove_avx512_no_vzeroupper)
 	      IFUNC_IMPL_ADD (array, i, memmove,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
-			      __memmove_avx512_unaligned_2)
+			      __memmove_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, memmove,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
 			      __memmove_avx512_unaligned_erms)
@@ -109,10 +101,9 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memmove_ssse3)
 	      IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_erms)
 	      IFUNC_IMPL_ADD (array, i, memmove, 1,
-			      __memmove_sse2_unaligned_2)
+			      __memmove_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, memmove, 1,
-			      __memmove_sse2_unaligned_erms)
-	      IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_sse2))
+			      __memmove_sse2_unaligned_erms))
 
   /* Support sysdeps/x86_64/multiarch/memset_chk.S.  */
   IFUNC_IMPL (i, name, __memset_chk,
@@ -326,7 +317,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memcpy_chk_avx512_no_vzeroupper)
 	      IFUNC_IMPL_ADD (array, i, __memcpy_chk,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
-			      __memcpy_chk_avx512_unaligned_2)
+			      __memcpy_chk_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __memcpy_chk,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
 			      __memcpy_chk_avx512_unaligned_erms)
@@ -336,9 +327,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memcpy_chk_avx_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __memcpy_chk,
 			      HAS_ARCH_FEATURE (AVX_Usable),
-			      __memcpy_chk_avx_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, __memcpy_chk,
-			      HAS_ARCH_FEATURE (AVX_Usable),
 			      __memcpy_chk_avx_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, __memcpy_chk,
 			      HAS_CPU_FEATURE (SSSE3),
@@ -347,11 +335,9 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      HAS_CPU_FEATURE (SSSE3),
 			      __memcpy_chk_ssse3)
 	      IFUNC_IMPL_ADD (array, i, __memcpy_chk, 1,
-			      __memcpy_chk_sse2_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, __memcpy_chk, 1,
-			      __memcpy_chk_sse2_unaligned_erms)
+			      __memcpy_chk_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __memcpy_chk, 1,
-			      __memcpy_chk_sse2))
+			      __memcpy_chk_sse2_unaligned_erms))
 
   /* Support sysdeps/x86_64/multiarch/memcpy.S.  */
   IFUNC_IMPL (i, name, memcpy,
@@ -360,9 +346,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memcpy_avx_unaligned)
 	      IFUNC_IMPL_ADD (array, i, memcpy,
 			      HAS_ARCH_FEATURE (AVX_Usable),
-			      __memcpy_avx_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, memcpy,
-			      HAS_ARCH_FEATURE (AVX_Usable),
 			      __memcpy_avx_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, memcpy, HAS_CPU_FEATURE (SSSE3),
 			      __memcpy_ssse3_back)
@@ -374,18 +357,15 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __memcpy_avx512_no_vzeroupper)
 	      IFUNC_IMPL_ADD (array, i, memcpy,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
-			      __memcpy_avx512_unaligned_2)
+			      __memcpy_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, memcpy,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
 			      __memcpy_avx512_unaligned_erms)
 #endif
 	      IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, memcpy, 1,
-			      __memcpy_sse2_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, memcpy, 1,
 			      __memcpy_sse2_unaligned_erms)
-	      IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_erms)
-	      IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_sse2))
+	      IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_erms))
 
   /* Support sysdeps/x86_64/multiarch/mempcpy_chk.S.  */
   IFUNC_IMPL (i, name, __mempcpy_chk,
@@ -395,7 +375,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __mempcpy_chk_avx512_no_vzeroupper)
 	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
-			      __mempcpy_chk_avx512_unaligned_2)
+			      __mempcpy_chk_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
 			      __mempcpy_chk_avx512_unaligned_erms)
@@ -405,9 +385,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __mempcpy_chk_avx_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
 			      HAS_ARCH_FEATURE (AVX_Usable),
-			      __mempcpy_chk_avx_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
-			      HAS_ARCH_FEATURE (AVX_Usable),
 			      __mempcpy_chk_avx_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
 			      HAS_CPU_FEATURE (SSSE3),
@@ -416,11 +393,9 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      HAS_CPU_FEATURE (SSSE3),
 			      __mempcpy_chk_ssse3)
 	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk, 1,
-			      __mempcpy_chk_sse2_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk, 1,
-			      __mempcpy_chk_sse2_unaligned_erms)
+			      __mempcpy_chk_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __mempcpy_chk, 1,
-			      __mempcpy_chk_sse2))
+			      __mempcpy_chk_sse2_unaligned_erms))
 
   /* Support sysdeps/x86_64/multiarch/mempcpy.S.  */
   IFUNC_IMPL (i, name, mempcpy,
@@ -430,7 +405,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __mempcpy_avx512_no_vzeroupper)
 	      IFUNC_IMPL_ADD (array, i, mempcpy,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
-			      __mempcpy_avx512_unaligned_2)
+			      __mempcpy_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, mempcpy,
 			      HAS_ARCH_FEATURE (AVX512F_Usable),
 			      __mempcpy_avx512_unaligned_erms)
@@ -440,20 +415,16 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      __mempcpy_avx_unaligned)
 	      IFUNC_IMPL_ADD (array, i, mempcpy,
 			      HAS_ARCH_FEATURE (AVX_Usable),
-			      __mempcpy_avx_unaligned_2)
-	      IFUNC_IMPL_ADD (array, i, mempcpy,
-			      HAS_ARCH_FEATURE (AVX_Usable),
 			      __mempcpy_avx_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, mempcpy, HAS_CPU_FEATURE (SSSE3),
 			      __mempcpy_ssse3_back)
 	      IFUNC_IMPL_ADD (array, i, mempcpy, HAS_CPU_FEATURE (SSSE3),
 			      __mempcpy_ssse3)
 	      IFUNC_IMPL_ADD (array, i, mempcpy, 1,
-			      __mempcpy_sse2_unaligned_2)
+			      __mempcpy_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, mempcpy, 1,
 			      __mempcpy_sse2_unaligned_erms)
-	      IFUNC_IMPL_ADD (array, i, mempcpy, 1, __mempcpy_erms)
-	      IFUNC_IMPL_ADD (array, i, mempcpy, 1, __mempcpy_sse2))
+	      IFUNC_IMPL_ADD (array, i, mempcpy, 1, __mempcpy_erms))
 
   /* Support sysdeps/x86_64/multiarch/strncmp.S.  */
   IFUNC_IMPL (i, name, strncmp,
diff --git a/sysdeps/x86_64/multiarch/memcpy-avx-unaligned.S b/sysdeps/x86_64/multiarch/memcpy-avx-unaligned.S
deleted file mode 100644
index dd4187f..0000000
--- a/sysdeps/x86_64/multiarch/memcpy-avx-unaligned.S
+++ /dev/null
@@ -1,391 +0,0 @@
-/* memcpy with AVX
-   Copyright (C) 2014-2016 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
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-
-#if IS_IN (libc) \
-    && (defined SHARED \
-        || defined USE_AS_MEMMOVE \
-	|| !defined USE_MULTIARCH)
-
-#include "asm-syntax.h"
-#ifndef MEMCPY
-# define MEMCPY		__memcpy_avx_unaligned
-# define MEMCPY_CHK	__memcpy_chk_avx_unaligned
-# define MEMPCPY	__mempcpy_avx_unaligned
-# define MEMPCPY_CHK	__mempcpy_chk_avx_unaligned
-#endif
-
-	.section .text.avx,"ax",@progbits
-#if !defined USE_AS_MEMPCPY && !defined USE_AS_MEMMOVE
-ENTRY (MEMPCPY_CHK)
-	cmpq	%rdx, %rcx
-	jb	HIDDEN_JUMPTARGET (__chk_fail)
-END (MEMPCPY_CHK)
-
-ENTRY (MEMPCPY)
-	movq	%rdi, %rax
-	addq	%rdx, %rax
-	jmp	L(start)
-END (MEMPCPY)
-#endif
-
-#if !defined USE_AS_BCOPY
-ENTRY (MEMCPY_CHK)
-	cmpq	%rdx, %rcx
-	jb	HIDDEN_JUMPTARGET (__chk_fail)
-END (MEMCPY_CHK)
-#endif
-
-ENTRY (MEMCPY)
-	mov	%rdi, %rax
-#ifdef USE_AS_MEMPCPY
-	add	%rdx, %rax
-#endif
-L(start):
-	cmp	$256, %rdx
-	jae	L(256bytesormore)
-	cmp	$16, %dl
-	jb	L(less_16bytes)
-	cmp	$128, %dl
-	jb	L(less_128bytes)
-	vmovdqu (%rsi), %xmm0
-	lea	(%rsi, %rdx), %rcx
-	vmovdqu 0x10(%rsi), %xmm1
-	vmovdqu 0x20(%rsi), %xmm2
-	vmovdqu 0x30(%rsi), %xmm3
-	vmovdqu 0x40(%rsi), %xmm4
-	vmovdqu 0x50(%rsi), %xmm5
-	vmovdqu 0x60(%rsi), %xmm6
-	vmovdqu 0x70(%rsi), %xmm7
-	vmovdqu -0x80(%rcx), %xmm8
-	vmovdqu -0x70(%rcx), %xmm9
-	vmovdqu -0x60(%rcx), %xmm10
-	vmovdqu -0x50(%rcx), %xmm11
-	vmovdqu -0x40(%rcx), %xmm12
-	vmovdqu -0x30(%rcx), %xmm13
-	vmovdqu -0x20(%rcx), %xmm14
-	vmovdqu -0x10(%rcx), %xmm15
-	lea	(%rdi, %rdx), %rdx
-	vmovdqu %xmm0, (%rdi)
-	vmovdqu %xmm1, 0x10(%rdi)
-	vmovdqu %xmm2, 0x20(%rdi)
-	vmovdqu %xmm3, 0x30(%rdi)
-	vmovdqu %xmm4, 0x40(%rdi)
-	vmovdqu %xmm5, 0x50(%rdi)
-	vmovdqu %xmm6, 0x60(%rdi)
-	vmovdqu %xmm7, 0x70(%rdi)
-	vmovdqu %xmm8, -0x80(%rdx)
-	vmovdqu %xmm9, -0x70(%rdx)
-	vmovdqu %xmm10, -0x60(%rdx)
-	vmovdqu %xmm11, -0x50(%rdx)
-	vmovdqu %xmm12, -0x40(%rdx)
-	vmovdqu %xmm13, -0x30(%rdx)
-	vmovdqu %xmm14, -0x20(%rdx)
-	vmovdqu %xmm15, -0x10(%rdx)
-	ret
-	.p2align 4
-L(less_128bytes):
-	cmp	$64, %dl
-	jb	L(less_64bytes)
-	vmovdqu (%rsi), %xmm0
-	lea	(%rsi, %rdx), %rcx
-	vmovdqu 0x10(%rsi), %xmm1
-	vmovdqu 0x20(%rsi), %xmm2
-	lea	(%rdi, %rdx), %rdx
-	vmovdqu 0x30(%rsi), %xmm3
-	vmovdqu -0x40(%rcx), %xmm4
-	vmovdqu -0x30(%rcx), %xmm5
-	vmovdqu -0x20(%rcx), %xmm6
-	vmovdqu -0x10(%rcx), %xmm7
-	vmovdqu %xmm0, (%rdi)
-	vmovdqu %xmm1, 0x10(%rdi)
-	vmovdqu %xmm2, 0x20(%rdi)
-	vmovdqu %xmm3, 0x30(%rdi)
-	vmovdqu %xmm4, -0x40(%rdx)
-	vmovdqu %xmm5, -0x30(%rdx)
-	vmovdqu %xmm6, -0x20(%rdx)
-	vmovdqu %xmm7, -0x10(%rdx)
-	ret
-
-	.p2align 4
-L(less_64bytes):
-	cmp	$32, %dl
-	jb	L(less_32bytes)
-	vmovdqu (%rsi), %xmm0
-	vmovdqu 0x10(%rsi), %xmm1
-	vmovdqu -0x20(%rsi, %rdx), %xmm6
-	vmovdqu -0x10(%rsi, %rdx), %xmm7
-	vmovdqu %xmm0, (%rdi)
-	vmovdqu %xmm1, 0x10(%rdi)
-	vmovdqu %xmm6, -0x20(%rdi, %rdx)
-	vmovdqu %xmm7, -0x10(%rdi, %rdx)
-	ret
-
-	.p2align 4
-L(less_32bytes):
-	vmovdqu (%rsi), %xmm0
-	vmovdqu -0x10(%rsi, %rdx), %xmm7
-	vmovdqu %xmm0, (%rdi)
-	vmovdqu %xmm7, -0x10(%rdi, %rdx)
-	ret
-
-	.p2align 4
-L(less_16bytes):
-	cmp	$8, %dl
-	jb	L(less_8bytes)
-	movq -0x08(%rsi, %rdx),	%rcx
-	movq (%rsi),	%rsi
-	movq %rsi, (%rdi)
-	movq %rcx, -0x08(%rdi, %rdx)
-	ret
-
-	.p2align 4
-L(less_8bytes):
-	cmp	$4, %dl
-	jb	L(less_4bytes)
-	mov -0x04(%rsi, %rdx), %ecx
-	mov (%rsi),	%esi
-	mov %esi, (%rdi)
-	mov %ecx, -0x04(%rdi, %rdx)
-	ret
-
-L(less_4bytes):
-	cmp	$1, %dl
-	jbe	L(less_2bytes)
-	mov -0x02(%rsi, %rdx),	%cx
-	mov (%rsi),	%si
-	mov %si, (%rdi)
-	mov %cx, -0x02(%rdi, %rdx)
-	ret
-
-L(less_2bytes):
-	jb	L(less_0bytes)
-	mov	(%rsi), %cl
-	mov	%cl,	(%rdi)
-L(less_0bytes):
-	ret
-
-	.p2align 4
-L(256bytesormore):
-#ifdef USE_AS_MEMMOVE
-	mov	%rdi, %rcx
-	sub	%rsi, %rcx
-	cmp	%rdx, %rcx
-	jc	L(copy_backward)
-#endif
-	cmp	$2048, %rdx
-	jae	L(gobble_data_movsb)
-	mov	%rax, %r8
-	lea	(%rsi, %rdx), %rcx
-	mov	%rdi, %r10
-	vmovdqu -0x80(%rcx), %xmm5
-	vmovdqu -0x70(%rcx), %xmm6
-	mov	$0x80, %rax
-	and	$-32, %rdi
-	add	$32, %rdi
-	vmovdqu -0x60(%rcx), %xmm7
-	vmovdqu -0x50(%rcx), %xmm8
-	mov	%rdi, %r11
-	sub	%r10, %r11
-	vmovdqu -0x40(%rcx), %xmm9
-	vmovdqu -0x30(%rcx), %xmm10
-	sub	%r11, %rdx
-	vmovdqu -0x20(%rcx), %xmm11
-	vmovdqu -0x10(%rcx), %xmm12
-	vmovdqu	(%rsi), %ymm4
-	add	%r11, %rsi
-	sub	%eax, %edx
-L(goble_128_loop):
-	vmovdqu (%rsi), %ymm0
-	vmovdqu 0x20(%rsi), %ymm1
-	vmovdqu 0x40(%rsi), %ymm2
-	vmovdqu 0x60(%rsi), %ymm3
-	add	%rax, %rsi
-	vmovdqa %ymm0, (%rdi)
-	vmovdqa %ymm1, 0x20(%rdi)
-	vmovdqa %ymm2, 0x40(%rdi)
-	vmovdqa %ymm3, 0x60(%rdi)
-	add	%rax, %rdi
-	sub	%eax, %edx
-	jae	L(goble_128_loop)
-	add	%eax, %edx
-	add	%rdi, %rdx
-	vmovdqu	%ymm4, (%r10)
-	vzeroupper
-	vmovdqu %xmm5, -0x80(%rdx)
-	vmovdqu %xmm6, -0x70(%rdx)
-	vmovdqu %xmm7, -0x60(%rdx)
-	vmovdqu %xmm8, -0x50(%rdx)
-	vmovdqu %xmm9, -0x40(%rdx)
-	vmovdqu %xmm10, -0x30(%rdx)
-	vmovdqu %xmm11, -0x20(%rdx)
-	vmovdqu %xmm12, -0x10(%rdx)
-	mov	%r8, %rax
-	ret
-
-	.p2align 4
-L(gobble_data_movsb):
-#ifdef SHARED_CACHE_SIZE_HALF
-	mov	$SHARED_CACHE_SIZE_HALF, %rcx
-#else
-	mov	__x86_shared_cache_size_half(%rip), %rcx
-#endif
-	shl	$3, %rcx
-	cmp	%rcx, %rdx
-	jae	L(gobble_big_data_fwd)
-	mov	%rdx, %rcx
-	rep	movsb
-	ret
-
-	.p2align 4
-L(gobble_big_data_fwd):
-	lea	(%rsi, %rdx), %rcx
-	vmovdqu	(%rsi), %ymm4
-	vmovdqu -0x80(%rsi,%rdx), %xmm5
-	vmovdqu -0x70(%rcx), %xmm6
-	vmovdqu -0x60(%rcx), %xmm7
-	vmovdqu -0x50(%rcx), %xmm8
-	vmovdqu -0x40(%rcx), %xmm9
-	vmovdqu -0x30(%rcx), %xmm10
-	vmovdqu -0x20(%rcx), %xmm11
-	vmovdqu -0x10(%rcx), %xmm12
-	mov	%rdi, %r8
-	and	$-32, %rdi
-	add	$32, %rdi
-	mov	%rdi, %r10
-	sub	%r8, %r10
-	sub	%r10, %rdx
-	add	%r10, %rsi
-	lea	(%rdi, %rdx), %rcx
-	add	$-0x80, %rdx
-L(gobble_mem_fwd_loop):
-	prefetchnta 0x1c0(%rsi)
-	prefetchnta 0x280(%rsi)
-	vmovdqu	(%rsi), %ymm0
-	vmovdqu	0x20(%rsi), %ymm1
-	vmovdqu	0x40(%rsi), %ymm2
-	vmovdqu	0x60(%rsi), %ymm3
-	sub	$-0x80, %rsi
-	vmovntdq	%ymm0, (%rdi)
-	vmovntdq	%ymm1, 0x20(%rdi)
-	vmovntdq	%ymm2, 0x40(%rdi)
-	vmovntdq	%ymm3, 0x60(%rdi)
-	sub	$-0x80, %rdi
-	add	$-0x80, %rdx
-	jb	L(gobble_mem_fwd_loop)
-	sfence
-	vmovdqu	%ymm4, (%r8)
-	vzeroupper
-	vmovdqu %xmm5, -0x80(%rcx)
-	vmovdqu %xmm6, -0x70(%rcx)
-	vmovdqu %xmm7, -0x60(%rcx)
-	vmovdqu %xmm8, -0x50(%rcx)
-	vmovdqu %xmm9, -0x40(%rcx)
-	vmovdqu %xmm10, -0x30(%rcx)
-	vmovdqu %xmm11, -0x20(%rcx)
-	vmovdqu %xmm12, -0x10(%rcx)
-	ret
-
-#ifdef USE_AS_MEMMOVE
-	.p2align 4
-L(copy_backward):
-#ifdef SHARED_CACHE_SIZE_HALF
-	mov	$SHARED_CACHE_SIZE_HALF, %rcx
-#else
-	mov	__x86_shared_cache_size_half(%rip), %rcx
-#endif
-	shl	$3, %rcx
-	vmovdqu (%rsi), %xmm5
-	vmovdqu 0x10(%rsi), %xmm6
-	add	%rdx, %rdi
-	vmovdqu 0x20(%rsi), %xmm7
-	vmovdqu 0x30(%rsi), %xmm8
-	lea	-0x20(%rdi), %r10
-	mov %rdi, %r11
-	vmovdqu 0x40(%rsi), %xmm9
-	vmovdqu 0x50(%rsi), %xmm10
-	and	$0x1f, %r11
-	vmovdqu 0x60(%rsi), %xmm11
-	vmovdqu 0x70(%rsi), %xmm12
-	xor	%r11, %rdi
-	add	%rdx, %rsi
-	vmovdqu	-0x20(%rsi), %ymm4
-	sub	%r11, %rsi
-	sub	%r11, %rdx
-	cmp	%rcx, %rdx
-	ja	L(gobble_big_data_bwd)
-	add	$-0x80, %rdx
-L(gobble_mem_bwd_llc):
-	vmovdqu	-0x20(%rsi), %ymm0
-	vmovdqu	-0x40(%rsi), %ymm1
-	vmovdqu	-0x60(%rsi), %ymm2
-	vmovdqu	-0x80(%rsi), %ymm3
-	lea	-0x80(%rsi), %rsi
-	vmovdqa	%ymm0, -0x20(%rdi)
-	vmovdqa	%ymm1, -0x40(%rdi)
-	vmovdqa	%ymm2, -0x60(%rdi)
-	vmovdqa	%ymm3, -0x80(%rdi)
-	lea	-0x80(%rdi), %rdi
-	add	$-0x80, %rdx
-	jb	L(gobble_mem_bwd_llc)
-	vmovdqu	%ymm4, (%r10)
-	vzeroupper
-	vmovdqu %xmm5, (%rax)
-	vmovdqu %xmm6, 0x10(%rax)
-	vmovdqu %xmm7, 0x20(%rax)
-	vmovdqu %xmm8, 0x30(%rax)
-	vmovdqu %xmm9, 0x40(%rax)
-	vmovdqu %xmm10, 0x50(%rax)
-	vmovdqu %xmm11, 0x60(%rax)
-	vmovdqu %xmm12, 0x70(%rax)
-	ret
-
-	.p2align 4
-L(gobble_big_data_bwd):
-	add	$-0x80, %rdx
-L(gobble_mem_bwd_loop):
-	prefetchnta -0x1c0(%rsi)
-	prefetchnta -0x280(%rsi)
-	vmovdqu	-0x20(%rsi), %ymm0
-	vmovdqu	-0x40(%rsi), %ymm1
-	vmovdqu	-0x60(%rsi), %ymm2
-	vmovdqu	-0x80(%rsi), %ymm3
-	lea	-0x80(%rsi), %rsi
-	vmovntdq	%ymm0, -0x20(%rdi)
-	vmovntdq	%ymm1, -0x40(%rdi)
-	vmovntdq	%ymm2, -0x60(%rdi)
-	vmovntdq	%ymm3, -0x80(%rdi)
-	lea	-0x80(%rdi), %rdi
-	add	$-0x80, %rdx
-	jb	L(gobble_mem_bwd_loop)
-	sfence
-	vmovdqu	%ymm4, (%r10)
-	vzeroupper
-	vmovdqu %xmm5, (%rax)
-	vmovdqu %xmm6, 0x10(%rax)
-	vmovdqu %xmm7, 0x20(%rax)
-	vmovdqu %xmm8, 0x30(%rax)
-	vmovdqu %xmm9, 0x40(%rax)
-	vmovdqu %xmm10, 0x50(%rax)
-	vmovdqu %xmm11, 0x60(%rax)
-	vmovdqu %xmm12, 0x70(%rax)
-	ret
-#endif
-END (MEMCPY)
-#endif
diff --git a/sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S b/sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S
deleted file mode 100644
index c450983..0000000
--- a/sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S
+++ /dev/null
@@ -1,175 +0,0 @@
-/* memcpy with unaliged loads
-   Copyright (C) 2013-2016 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
-   <http://www.gnu.org/licenses/>.  */
-
-#if IS_IN (libc)
-
-#include <sysdep.h>
-
-#include "asm-syntax.h"
-
-
-ENTRY(__memcpy_sse2_unaligned)
-	movq	%rsi, %rax
-	leaq	(%rdx,%rdx), %rcx
-	subq	%rdi, %rax
-	subq	%rdx, %rax
-	cmpq	%rcx, %rax
-	jb	L(overlapping)
-	cmpq	$16, %rdx
-	jbe	L(less_16)
-	movdqu	(%rsi), %xmm8
-	cmpq	$32, %rdx
-	movdqu	%xmm8, (%rdi)
-	movdqu	-16(%rsi,%rdx), %xmm8
-	movdqu	%xmm8, -16(%rdi,%rdx)
-	ja	.L31
-L(return):
-	movq	%rdi, %rax
-	ret
-	.p2align 4,,10
-	.p2align 4
-.L31:
-	movdqu	16(%rsi), %xmm8
-	cmpq	$64, %rdx
-	movdqu	%xmm8, 16(%rdi)
-	movdqu	-32(%rsi,%rdx), %xmm8
-	movdqu	%xmm8, -32(%rdi,%rdx)
-	jbe	L(return)
-	movdqu	32(%rsi), %xmm8
-	cmpq	$128, %rdx
-	movdqu	%xmm8, 32(%rdi)
-	movdqu	-48(%rsi,%rdx), %xmm8
-	movdqu	%xmm8, -48(%rdi,%rdx)
-	movdqu	48(%rsi), %xmm8
-	movdqu	%xmm8, 48(%rdi)
-	movdqu	-64(%rsi,%rdx), %xmm8
-	movdqu	%xmm8, -64(%rdi,%rdx)
-	jbe	L(return)
-	leaq	64(%rdi), %rcx
-	addq	%rdi, %rdx
-	andq	$-64, %rdx
-	andq	$-64, %rcx
-	movq	%rcx, %rax
-	subq	%rdi, %rax
-	addq	%rax, %rsi
-	cmpq	%rdx, %rcx
-	je	L(return)
-	movq	%rsi, %r10
-	subq	%rcx, %r10
-	leaq	16(%r10), %r9
-	leaq	32(%r10), %r8
-	leaq	48(%r10), %rax
-	.p2align 4,,10
-	.p2align 4
-L(loop):
-	movdqu	(%rcx,%r10), %xmm8
-	movdqa	%xmm8, (%rcx)
-	movdqu	(%rcx,%r9), %xmm8
-	movdqa	%xmm8, 16(%rcx)
-	movdqu	(%rcx,%r8), %xmm8
-	movdqa	%xmm8, 32(%rcx)
-	movdqu	(%rcx,%rax), %xmm8
-	movdqa	%xmm8, 48(%rcx)
-	addq	$64, %rcx
-	cmpq	%rcx, %rdx
-	jne	L(loop)
-	jmp	L(return)
-L(overlapping):
-	cmpq	%rsi, %rdi
-	jae	.L3
-	testq	%rdx, %rdx
-	.p2align 4,,5
-	je	L(return)
-	movq	%rdx, %r9
-	leaq	16(%rsi), %rcx
-	leaq	16(%rdi), %r8
-	shrq	$4, %r9
-	movq	%r9, %rax
-	salq	$4, %rax
-	cmpq	%rcx, %rdi
-	setae	%cl
-	cmpq	%r8, %rsi
-	setae	%r8b
-	orl	%r8d, %ecx
-	cmpq	$15, %rdx
-	seta	%r8b
-	testb	%r8b, %cl
-	je	.L16
-	testq	%rax, %rax
-	je	.L16
-	xorl	%ecx, %ecx
-	xorl	%r8d, %r8d
-.L7:
-	movdqu	(%rsi,%rcx), %xmm8
-	addq	$1, %r8
-	movdqu	%xmm8, (%rdi,%rcx)
-	addq	$16, %rcx
-	cmpq	%r8, %r9
-	ja	.L7
-	cmpq	%rax, %rdx
-	je	L(return)
-.L21:
-	movzbl	(%rsi,%rax), %ecx
-	movb	%cl, (%rdi,%rax)
-	addq	$1, %rax
-	cmpq	%rax, %rdx
-	ja	.L21
-	jmp	L(return)
-L(less_16):
-	testb	$24, %dl
-	jne	L(between_9_16)
-	testb	$4, %dl
-	.p2align 4,,5
-	jne	L(between_5_8)
-	testq	%rdx, %rdx
-	.p2align 4,,2
-	je	L(return)
-	movzbl	(%rsi), %eax
-	testb	$2, %dl
-	movb	%al, (%rdi)
-	je	L(return)
-	movzwl	-2(%rsi,%rdx), %eax
-	movw	%ax, -2(%rdi,%rdx)
-	jmp	L(return)
-.L3:
-	leaq	-1(%rdx), %rax
-	.p2align 4,,10
-	.p2align 4
-.L11:
-	movzbl	(%rsi,%rax), %edx
-	movb	%dl, (%rdi,%rax)
-	subq	$1, %rax
-	jmp	.L11
-L(between_9_16):
-	movq	(%rsi), %rax
-	movq	%rax, (%rdi)
-	movq	-8(%rsi,%rdx), %rax
-	movq	%rax, -8(%rdi,%rdx)
-	jmp	L(return)
-.L16:
-	xorl	%eax, %eax
-	jmp	.L21
-L(between_5_8):
-	movl	(%rsi), %eax
-	movl	%eax, (%rdi)
-	movl	-4(%rsi,%rdx), %eax
-	movl	%eax, -4(%rdi,%rdx)
-	jmp	L(return)
-END(__memcpy_sse2_unaligned)
-
-#endif
diff --git a/sysdeps/x86_64/multiarch/memcpy.S b/sysdeps/x86_64/multiarch/memcpy.S
index 5b045d7..f6771a4 100644
--- a/sysdeps/x86_64/multiarch/memcpy.S
+++ b/sysdeps/x86_64/multiarch/memcpy.S
@@ -19,7 +19,6 @@
    <http://www.gnu.org/licenses/>.  */
 
 #include <sysdep.h>
-#include <shlib-compat.h>
 #include <init-arch.h>
 
 /* Define multiple versions only for the definition in lib and for
@@ -30,21 +29,34 @@
 ENTRY(__new_memcpy)
 	.type	__new_memcpy, @gnu_indirect_function
 	LOAD_RTLD_GLOBAL_RO_RDX
-#ifdef HAVE_AVX512_ASM_SUPPORT
+# ifdef HAVE_AVX512_ASM_SUPPORT
 	HAS_ARCH_FEATURE (AVX512F_Usable)
 	jz	1f
+	lea	__memcpy_avx512_no_vzeroupper(%rip), %RAX_LP
 	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
-	jz	1f
-	lea    __memcpy_avx512_no_vzeroupper(%rip), %RAX_LP
+	jnz	2f
+	lea	__memcpy_avx512_unaligned_erms(%rip), %RAX_LP
+	HAS_CPU_FEATURE (ERMS)
+	jnz	2f
+	lea	__memcpy_avx512_unaligned(%rip), %RAX_LP
 	ret
-#endif
+# endif
 1:	lea	__memcpy_avx_unaligned(%rip), %RAX_LP
 	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
-	jnz	2f
+	jz	L(Fast_Unaligned_Load)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memcpy_avx_unaligned_erms(%rip), %RAX_LP
+	ret
+L(Fast_Unaligned_Load):
 	lea	__memcpy_sse2_unaligned(%rip), %RAX_LP
 	HAS_ARCH_FEATURE (Fast_Unaligned_Copy)
-	jnz	2f
-	lea	__memcpy_sse2(%rip), %RAX_LP
+	jz	L(SSSE3)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memcpy_sse2_unaligned_erms(%rip), %RAX_LP
+	ret
+L(SSSE3):
 	HAS_CPU_FEATURE (SSSE3)
 	jz	2f
 	lea    __memcpy_ssse3_back(%rip), %RAX_LP
@@ -54,37 +66,7 @@ ENTRY(__new_memcpy)
 2:	ret
 END(__new_memcpy)
 
-# undef ENTRY
-# define ENTRY(name) \
-	.type __memcpy_sse2, @function; \
-	.globl __memcpy_sse2; \
-	.hidden __memcpy_sse2; \
-	.p2align 4; \
-	__memcpy_sse2: cfi_startproc; \
-	CALL_MCOUNT
-# undef END
-# define END(name) \
-	cfi_endproc; .size __memcpy_sse2, .-__memcpy_sse2
-
-# undef ENTRY_CHK
-# define ENTRY_CHK(name) \
-	.type __memcpy_chk_sse2, @function; \
-	.globl __memcpy_chk_sse2; \
-	.p2align 4; \
-	__memcpy_chk_sse2: cfi_startproc; \
-	CALL_MCOUNT
-# undef END_CHK
-# define END_CHK(name) \
-	cfi_endproc; .size __memcpy_chk_sse2, .-__memcpy_chk_sse2
-
-# undef libc_hidden_builtin_def
-/* It doesn't make sense to send libc-internal memcpy calls through a PLT.
-   The speedup we get from using SSSE3 instruction is likely eaten away
-   by the indirect call in the PLT.  */
-# define libc_hidden_builtin_def(name) \
-	.globl __GI_memcpy; __GI_memcpy = __memcpy_sse2
-
+# undef memcpy
+# include <shlib-compat.h>
 versioned_symbol (libc, __new_memcpy, memcpy, GLIBC_2_14);
 #endif
-
-#include "../memcpy.S"
diff --git a/sysdeps/x86_64/multiarch/memcpy_chk.S b/sysdeps/x86_64/multiarch/memcpy_chk.S
index 648217e..11f1310 100644
--- a/sysdeps/x86_64/multiarch/memcpy_chk.S
+++ b/sysdeps/x86_64/multiarch/memcpy_chk.S
@@ -30,24 +30,40 @@
 ENTRY(__memcpy_chk)
 	.type	__memcpy_chk, @gnu_indirect_function
 	LOAD_RTLD_GLOBAL_RO_RDX
-#ifdef HAVE_AVX512_ASM_SUPPORT
+# ifdef HAVE_AVX512_ASM_SUPPORT
 	HAS_ARCH_FEATURE (AVX512F_Usable)
-	jz      1f
+	jz	1f
+	lea	__memcpy_chk_avx512_no_vzeroupper(%rip), %RAX_LP
 	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
-	jz      1f
-	leaq    __memcpy_chk_avx512_no_vzeroupper(%rip), %rax
+	jnz	2f
+	lea	__memcpy_chk_avx512_unaligned_erms(%rip), %RAX_LP
+	HAS_CPU_FEATURE (ERMS)
+	jnz	2f
+	lea	__memcpy_chk_avx512_unaligned(%rip), %RAX_LP
 	ret
-#endif
-1:	leaq	__memcpy_chk_sse2(%rip), %rax
+# endif
+1:	lea	__memcpy_chk_avx_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	jz	L(Fast_Unaligned_Load)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memcpy_chk_avx_unaligned_erms(%rip), %RAX_LP
+	ret
+L(Fast_Unaligned_Load):
+	lea	__memcpy_chk_sse2_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Unaligned_Copy)
+	jz	L(SSSE3)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memcpy_chk_sse2_unaligned_erms(%rip), %RAX_LP
+	ret
+L(SSSE3):
 	HAS_CPU_FEATURE (SSSE3)
 	jz	2f
-	leaq	__memcpy_chk_ssse3(%rip), %rax
+	lea    __memcpy_chk_ssse3_back(%rip), %RAX_LP
 	HAS_ARCH_FEATURE (Fast_Copy_Backward)
-	jz	2f
-	leaq	__memcpy_chk_ssse3_back(%rip), %rax
-	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
-	jz  2f
-	leaq    __memcpy_chk_avx_unaligned(%rip), %rax
+	jnz	2f
+	lea	__memcpy_chk_ssse3(%rip), %RAX_LP
 2:	ret
 END(__memcpy_chk)
 # else
diff --git a/sysdeps/x86_64/multiarch/memmove-avx-unaligned.S b/sysdeps/x86_64/multiarch/memmove-avx-unaligned.S
deleted file mode 100644
index 75e35f2..0000000
--- a/sysdeps/x86_64/multiarch/memmove-avx-unaligned.S
+++ /dev/null
@@ -1,22 +0,0 @@
-/* memmove with AVX
-   Copyright (C) 2014-2016 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
-   <http://www.gnu.org/licenses/>.  */
-
-#define USE_AS_MEMMOVE
-#define MEMCPY		__memmove_avx_unaligned
-#define MEMCPY_CHK	__memmove_chk_avx_unaligned
-#include "memcpy-avx-unaligned.S"
diff --git a/sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S b/sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S
deleted file mode 100644
index d7edb18..0000000
--- a/sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S
+++ /dev/null
@@ -1,13 +0,0 @@
-#if IS_IN (libc)
-# define VEC_SIZE	16
-# define VEC(i)		xmm##i
-# define VMOVNT		movntdq
-/* Use movups and movaps for smaller code sizes.  */
-# define VMOVU		movups
-# define VMOVA		movaps
-
-# define SECTION(p)		p
-# define MEMMOVE_SYMBOL(p,s)	p##_sse2_##s
-
-# include "memmove-vec-unaligned-erms.S"
-#endif
diff --git a/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S b/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S
index 3742c10..a2cce39 100644
--- a/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S
+++ b/sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S
@@ -106,28 +106,28 @@
 
 	.section SECTION(.text),"ax",@progbits
 #if defined SHARED && IS_IN (libc)
-ENTRY (MEMMOVE_CHK_SYMBOL (__mempcpy_chk, unaligned_2))
+ENTRY (MEMMOVE_CHK_SYMBOL (__mempcpy_chk, unaligned))
 	cmpq	%rdx, %rcx
 	jb	HIDDEN_JUMPTARGET (__chk_fail)
-END (MEMMOVE_CHK_SYMBOL (__mempcpy_chk, unaligned_2))
+END (MEMMOVE_CHK_SYMBOL (__mempcpy_chk, unaligned))
 #endif
 
 #if VEC_SIZE == 16 || defined SHARED
-ENTRY (MEMPCPY_SYMBOL (__mempcpy, unaligned_2))
+ENTRY (MEMPCPY_SYMBOL (__mempcpy, unaligned))
 	movq	%rdi, %rax
 	addq	%rdx, %rax
 	jmp	L(start)
-END (MEMPCPY_SYMBOL (__mempcpy, unaligned_2))
+END (MEMPCPY_SYMBOL (__mempcpy, unaligned))
 #endif
 
 #if defined SHARED && IS_IN (libc)
-ENTRY (MEMMOVE_CHK_SYMBOL (__memmove_chk, unaligned_2))
+ENTRY (MEMMOVE_CHK_SYMBOL (__memmove_chk, unaligned))
 	cmpq	%rdx, %rcx
 	jb	HIDDEN_JUMPTARGET (__chk_fail)
-END (MEMMOVE_CHK_SYMBOL (__memmove_chk, unaligned_2))
+END (MEMMOVE_CHK_SYMBOL (__memmove_chk, unaligned))
 #endif
 
-ENTRY (MEMMOVE_SYMBOL (__memmove, unaligned_2))
+ENTRY (MEMMOVE_SYMBOL (__memmove, unaligned))
 	movq	%rdi, %rax
 L(start):
 	cmpq	$VEC_SIZE, %rdx
@@ -148,7 +148,7 @@ L(nop):
 #endif
 	ret
 #if defined USE_MULTIARCH && IS_IN (libc)
-END (MEMMOVE_SYMBOL (__memmove, unaligned_2))
+END (MEMMOVE_SYMBOL (__memmove, unaligned))
 
 # if VEC_SIZE == 16 && defined SHARED
 /* Only used to measure performance of REP MOVSB.  */
@@ -539,11 +539,11 @@ strong_alias (MEMMOVE_SYMBOL (__memmove, unaligned_erms),
 strong_alias (MEMMOVE_SYMBOL (__memmove_chk, unaligned_erms),
 	      MEMMOVE_SYMBOL (__memcpy_chk, unaligned_erms))
 #  endif
-strong_alias (MEMMOVE_CHK_SYMBOL (__memmove_chk, unaligned_2),
-	      MEMMOVE_CHK_SYMBOL (__memcpy_chk, unaligned_2))
+strong_alias (MEMMOVE_CHK_SYMBOL (__memmove_chk, unaligned),
+	      MEMMOVE_CHK_SYMBOL (__memcpy_chk, unaligned))
 # endif
 #endif
 #if VEC_SIZE == 16 || defined SHARED
-strong_alias (MEMMOVE_SYMBOL (__memmove, unaligned_2),
-	      MEMCPY_SYMBOL (__memcpy, unaligned_2))
+strong_alias (MEMMOVE_SYMBOL (__memmove, unaligned),
+	      MEMCPY_SYMBOL (__memcpy, unaligned))
 #endif
diff --git a/sysdeps/x86_64/multiarch/memmove.S b/sysdeps/x86_64/multiarch/memmove.S
new file mode 100644
index 0000000..25c3586
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/memmove.S
@@ -0,0 +1,98 @@
+/* Multiple versions of memmove
+   All versions must be listed in ifunc-impl-list.c.
+   Copyright (C) 2016 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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+/* Define multiple versions only for the definition in lib and for
+   DSO.  */
+#if IS_IN (libc)
+	.text
+ENTRY(__libc_memmove)
+	.type	__libc_memmove, @gnu_indirect_function
+	LOAD_RTLD_GLOBAL_RO_RDX
+# ifdef HAVE_AVX512_ASM_SUPPORT
+	HAS_ARCH_FEATURE (AVX512F_Usable)
+	jz	1f
+	lea	__memmove_avx512_no_vzeroupper(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+	jnz	2f
+	lea	__memmove_avx512_unaligned_erms(%rip), %RAX_LP
+	HAS_CPU_FEATURE (ERMS)
+	jnz	2f
+	lea	__memmove_avx512_unaligned(%rip), %RAX_LP
+	ret
+# endif
+1:	lea	__memmove_avx_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	jz	L(Fast_Unaligned_Load)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memmove_avx_unaligned_erms(%rip), %RAX_LP
+	ret
+L(Fast_Unaligned_Load):
+	lea	__memmove_sse2_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Unaligned_Copy)
+	jz	L(SSSE3)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memmove_sse2_unaligned_erms(%rip), %RAX_LP
+	ret
+L(SSSE3):
+	HAS_CPU_FEATURE (SSSE3)
+	jz	2f
+	lea    __memmove_ssse3_back(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Copy_Backward)
+	jnz	2f
+	lea	__memmove_ssse3(%rip), %RAX_LP
+2:	ret
+END(__libc_memmove)
+#endif
+
+#if IS_IN (libc)
+# define MEMMOVE_SYMBOL(p,s)	p##_sse2_##s
+
+# ifdef SHARED
+libc_hidden_ver (__memmove_sse2_unaligned, memmove)
+libc_hidden_ver (__memcpy_sse2_unaligned, memcpy)
+libc_hidden_ver (__mempcpy_sse2_unaligned, mempcpy)
+libc_hidden_ver (__mempcpy_sse2_unaligned, __mempcpy)
+
+#  undef libc_hidden_builtin_def
+/* It doesn't make sense to send libc-internal memmove calls through a PLT.
+   The speedup we get from using SSE2 instructions is likely eaten away
+   by the indirect call in the PLT.  */
+#  define libc_hidden_builtin_def
+# endif
+strong_alias (__libc_memmove, memmove)
+#endif
+
+#if !defined SHARED || !IS_IN (libc)
+weak_alias (__mempcpy, mempcpy)
+#endif
+
+#include "../memmove.S"
+
+#if defined SHARED && IS_IN (libc)
+# include <shlib-compat.h>
+# if SHLIB_COMPAT (libc, GLIBC_2_2_5, GLIBC_2_14)
+/* Use __memmove_sse2_unaligned to support overlapping addresses.  */
+compat_symbol (libc, __memmove_sse2_unaligned, memcpy, GLIBC_2_2_5);
+# endif
+#endif
diff --git a/sysdeps/x86_64/multiarch/memmove.c b/sysdeps/x86_64/multiarch/memmove.c
deleted file mode 100644
index 8da5640..0000000
--- a/sysdeps/x86_64/multiarch/memmove.c
+++ /dev/null
@@ -1,73 +0,0 @@
-/* Multiple versions of memmove.
-   All versions must be listed in ifunc-impl-list.c.
-   Copyright (C) 2010-2016 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
-   <http://www.gnu.org/licenses/>.  */
-
-#if IS_IN (libc)
-# define MEMMOVE __memmove_sse2
-# ifdef SHARED
-#  undef libc_hidden_builtin_def
-#  define libc_hidden_builtin_def(name) \
-  __hidden_ver1 (__memmove_sse2, __GI_memmove, __memmove_sse2);
-# endif
-
-/* Redefine memmove so that the compiler won't complain about the type
-   mismatch with the IFUNC selector in strong_alias, below.  */
-# undef memmove
-# define memmove __redirect_memmove
-# include <string.h>
-# undef memmove
-
-extern __typeof (__redirect_memmove) __memmove_sse2 attribute_hidden;
-extern __typeof (__redirect_memmove) __memmove_ssse3 attribute_hidden;
-extern __typeof (__redirect_memmove) __memmove_ssse3_back attribute_hidden;
-extern __typeof (__redirect_memmove) __memmove_avx_unaligned attribute_hidden;
-# ifdef HAVE_AVX512_ASM_SUPPORT
-  extern __typeof (__redirect_memmove) __memmove_avx512_no_vzeroupper attribute_hidden;
-# endif
-
-#endif
-
-#include "string/memmove.c"
-
-#if IS_IN (libc)
-# include <shlib-compat.h>
-# include "init-arch.h"
-
-/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
-   ifunc symbol properly.  */
-extern __typeof (__redirect_memmove) __libc_memmove;
-libc_ifunc (__libc_memmove,
-#ifdef HAVE_AVX512_ASM_SUPPORT
-	    HAS_ARCH_FEATURE (AVX512F_Usable)
-	      && HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
-	    ? __memmove_avx512_no_vzeroupper
-	    :
-#endif
-	    (HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
-	    ? __memmove_avx_unaligned
-	    : (HAS_CPU_FEATURE (SSSE3)
-	       ? (HAS_ARCH_FEATURE (Fast_Copy_Backward)
-	          ? __memmove_ssse3_back : __memmove_ssse3)
-	       : __memmove_sse2)));
-
-strong_alias (__libc_memmove, memmove)
-
-# if SHLIB_COMPAT (libc, GLIBC_2_2_5, GLIBC_2_14)
-compat_symbol (libc, memmove, memcpy, GLIBC_2_2_5);
-# endif
-#endif
diff --git a/sysdeps/x86_64/multiarch/memmove_chk.S b/sysdeps/x86_64/multiarch/memmove_chk.S
new file mode 100644
index 0000000..cd639b8
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/memmove_chk.S
@@ -0,0 +1,71 @@
+/* Multiple versions of __memmove_chk
+   All versions must be listed in ifunc-impl-list.c.
+   Copyright (C) 2016 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
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+/* Define multiple versions only for the definition in lib and for
+   DSO.  There are no multiarch memmove functions for static binaries.
+ */
+#if IS_IN (libc)
+# ifdef SHARED
+	.text
+ENTRY(__memmove_chk)
+	.type	__memmove_chk, @gnu_indirect_function
+	LOAD_RTLD_GLOBAL_RO_RDX
+# ifdef HAVE_AVX512_ASM_SUPPORT
+	HAS_ARCH_FEATURE (AVX512F_Usable)
+	jz	1f
+	lea	__memmove_chk_avx512_no_vzeroupper(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+	jnz	2f
+	lea	__memmove_chk_avx512_unaligned_erms(%rip), %RAX_LP
+	HAS_CPU_FEATURE (ERMS)
+	jnz	2f
+	lea	__memmove_chk_avx512_unaligned(%rip), %RAX_LP
+	ret
+# endif
+1:	lea	__memmove_chk_avx_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	jz	L(Fast_Unaligned_Load)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memmove_chk_avx_unaligned_erms(%rip), %RAX_LP
+	ret
+L(Fast_Unaligned_Load):
+	lea	__memmove_chk_sse2_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Unaligned_Copy)
+	jz	L(SSSE3)
+	HAS_CPU_FEATURE (ERMS)
+	jz	2f
+	lea	__memmove_chk_sse2_unaligned_erms(%rip), %RAX_LP
+	ret
+L(SSSE3):
+	HAS_CPU_FEATURE (SSSE3)
+	jz	2f
+	lea    __memmove_chk_ssse3_back(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Copy_Backward)
+	jnz	2f
+	lea	__memmove_chk_ssse3(%rip), %RAX_LP
+2:	ret
+END(__memmove_chk)
+# else
+#  include "../memmove_chk.S"
+# endif
+#endif
diff --git a/sysdeps/x86_64/multiarch/memmove_chk.c b/sysdeps/x86_64/multiarch/memmove_chk.c
deleted file mode 100644
index f64da63..0000000
--- a/sysdeps/x86_64/multiarch/memmove_chk.c
+++ /dev/null
@@ -1,46 +0,0 @@
-/* Multiple versions of __memmove_chk.
-   All versions must be listed in ifunc-impl-list.c.
-   Copyright (C) 2010-2016 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
-   <http://www.gnu.org/licenses/>.  */
-
-#include <string.h>
-#include "init-arch.h"
-
-#define MEMMOVE_CHK __memmove_chk_sse2
-
-extern __typeof (__memmove_chk) __memmove_chk_sse2 attribute_hidden;
-extern __typeof (__memmove_chk) __memmove_chk_ssse3 attribute_hidden;
-extern __typeof (__memmove_chk) __memmove_chk_ssse3_back attribute_hidden;
-extern __typeof (__memmove_chk) __memmove_chk_avx_unaligned attribute_hidden;
-# ifdef HAVE_AVX512_ASM_SUPPORT
-  extern __typeof (__memmove_chk) __memmove_chk_avx512_no_vzeroupper attribute_hidden;
-# endif
-
-#include "debug/memmove_chk.c"
-
-libc_ifunc (__memmove_chk,
-#ifdef HAVE_AVX512_ASM_SUPPORT
-	    HAS_ARCH_FEATURE (AVX512F_Usable)
-	      && HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
-	    ? __memmove_chk_avx512_no_vzeroupper
-	    :
-#endif
-	    HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load) ? __memmove_chk_avx_unaligned :
-	    (HAS_CPU_FEATURE (SSSE3)
-	    ? (HAS_ARCH_FEATURE (Fast_Copy_Backward)
-	       ? __memmove_chk_ssse3_back : __memmove_chk_ssse3)
-	    : __memmove_chk_sse2));
diff --git a/sysdeps/x86_64/multiarch/mempcpy.S b/sysdeps/x86_64/multiarch/mempcpy.S
index ed78623..f9c6df3 100644
--- a/sysdeps/x86_64/multiarch/mempcpy.S
+++ b/sysdeps/x86_64/multiarch/mempcpy.S
@@ -25,62 +25,46 @@
    DSO.  In static binaries we need mempcpy before the initialization
    happened.  */
 #if defined SHARED && IS_IN (libc)
+	.text
 ENTRY(__mempcpy)
 	.type	__mempcpy, @gnu_indirect_function
 	LOAD_RTLD_GLOBAL_RO_RDX
-#ifdef HAVE_AVX512_ASM_SUPPORT
+# ifdef HAVE_AVX512_ASM_SUPPORT
 	HAS_ARCH_FEATURE (AVX512F_Usable)
 	jz	1f
+	lea	__mempcpy_avx512_no_vzeroupper(%rip), %RAX_LP
 	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
-	jz	1f
-	leaq    __mempcpy_avx512_no_vzeroupper(%rip), %rax
+	jnz	2f
+	lea	__mempcpy_avx512_unaligned_erms(%rip), %RAX_LP
+	HAS_CPU_FEATURE (ERMS)
+	jnz	2f
+	lea	__mempcpy_avx512_unaligned(%rip), %RAX_LP
 	ret
-#endif
-1:	leaq	__mempcpy_sse2(%rip), %rax
-	HAS_CPU_FEATURE (SSSE3)
+# endif
+1:	lea	__mempcpy_avx_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	jz	L(Fast_Unaligned_Load)
+	HAS_CPU_FEATURE (ERMS)
 	jz	2f
-	leaq	__mempcpy_ssse3(%rip), %rax
-	HAS_ARCH_FEATURE (Fast_Copy_Backward)
+	lea	__mempcpy_avx_unaligned_erms(%rip), %RAX_LP
+	ret
+L(Fast_Unaligned_Load):
+	lea	__mempcpy_sse2_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Unaligned_Copy)
+	jz	L(SSSE3)
+	HAS_CPU_FEATURE (ERMS)
 	jz	2f
-	leaq	__mempcpy_ssse3_back(%rip), %rax
-	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	lea	__mempcpy_sse2_unaligned_erms(%rip), %RAX_LP
+	ret
+L(SSSE3):
+	HAS_CPU_FEATURE (SSSE3)
 	jz	2f
-	leaq	__mempcpy_avx_unaligned(%rip), %rax
+	lea    __mempcpy_ssse3_back(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Copy_Backward)
+	jnz	2f
+	lea	__mempcpy_ssse3(%rip), %RAX_LP
 2:	ret
 END(__mempcpy)
 
-# undef ENTRY
-# define ENTRY(name) \
-	.type __mempcpy_sse2, @function; \
-	.p2align 4; \
-	.globl __mempcpy_sse2; \
-	.hidden __mempcpy_sse2; \
-	__mempcpy_sse2: cfi_startproc; \
-	CALL_MCOUNT
-# undef END
-# define END(name) \
-	cfi_endproc; .size __mempcpy_sse2, .-__mempcpy_sse2
-
-# undef ENTRY_CHK
-# define ENTRY_CHK(name) \
-	.type __mempcpy_chk_sse2, @function; \
-	.globl __mempcpy_chk_sse2; \
-	.p2align 4; \
-	__mempcpy_chk_sse2: cfi_startproc; \
-	CALL_MCOUNT
-# undef END_CHK
-# define END_CHK(name) \
-	cfi_endproc; .size __mempcpy_chk_sse2, .-__mempcpy_chk_sse2
-
-# undef libc_hidden_def
-# undef libc_hidden_builtin_def
-/* It doesn't make sense to send libc-internal mempcpy calls through a PLT.
-   The speedup we get from using SSSE3 instruction is likely eaten away
-   by the indirect call in the PLT.  */
-# define libc_hidden_def(name) \
-	.globl __GI_mempcpy; __GI_mempcpy = __mempcpy_sse2
-# define libc_hidden_builtin_def(name) \
-	.globl __GI___mempcpy; __GI___mempcpy = __mempcpy_sse2
+weak_alias (__mempcpy, mempcpy)
 #endif
-
-#include "../mempcpy.S"
diff --git a/sysdeps/x86_64/multiarch/mempcpy_chk.S b/sysdeps/x86_64/multiarch/mempcpy_chk.S
index 6e8a89d..80f460f 100644
--- a/sysdeps/x86_64/multiarch/mempcpy_chk.S
+++ b/sysdeps/x86_64/multiarch/mempcpy_chk.S
@@ -30,24 +30,40 @@
 ENTRY(__mempcpy_chk)
 	.type	__mempcpy_chk, @gnu_indirect_function
 	LOAD_RTLD_GLOBAL_RO_RDX
-#ifdef HAVE_AVX512_ASM_SUPPORT
+# ifdef HAVE_AVX512_ASM_SUPPORT
 	HAS_ARCH_FEATURE (AVX512F_Usable)
 	jz	1f
+	lea	__mempcpy_chk_avx512_no_vzeroupper(%rip), %RAX_LP
 	HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
-	jz	1f
-	leaq    __mempcpy_chk_avx512_no_vzeroupper(%rip), %rax
+	jnz	2f
+	lea	__mempcpy_chk_avx512_unaligned_erms(%rip), %RAX_LP
+	HAS_CPU_FEATURE (ERMS)
+	jnz	2f
+	lea	__mempcpy_chk_avx512_unaligned(%rip), %RAX_LP
 	ret
-#endif
-1:	leaq	__mempcpy_chk_sse2(%rip), %rax
-	HAS_CPU_FEATURE (SSSE3)
+# endif
+1:	lea	__mempcpy_chk_avx_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	jz	L(Fast_Unaligned_Load)
+	HAS_CPU_FEATURE (ERMS)
 	jz	2f
-	leaq	__mempcpy_chk_ssse3(%rip), %rax
-	HAS_ARCH_FEATURE (Fast_Copy_Backward)
+	lea	__mempcpy_chk_avx_unaligned_erms(%rip), %RAX_LP
+	ret
+L(Fast_Unaligned_Load):
+	lea	__mempcpy_chk_sse2_unaligned(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Unaligned_Copy)
+	jz	L(SSSE3)
+	HAS_CPU_FEATURE (ERMS)
 	jz	2f
-	leaq	__mempcpy_chk_ssse3_back(%rip), %rax
-	HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+	lea	__mempcpy_chk_sse2_unaligned_erms(%rip), %RAX_LP
+	ret
+L(SSSE3):
+	HAS_CPU_FEATURE (SSSE3)
 	jz	2f
-	leaq	__mempcpy_chk_avx_unaligned(%rip), %rax
+	lea    __mempcpy_chk_ssse3_back(%rip), %RAX_LP
+	HAS_ARCH_FEATURE (Fast_Copy_Backward)
+	jnz	2f
+	lea	__mempcpy_chk_ssse3(%rip), %RAX_LP
 2:	ret
 END(__mempcpy_chk)
 # else