x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2

Message ID 20181008135950.9113-1-leonardo.sandoval.gonzalez@linux.intel.com
State New
Headers show
Series
  • x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
Related show

Commit Message

Leonardo Sandoval Oct. 8, 2018, 1:59 p.m.
From: Leonardo Sandoval <leonardo.sandoval.gonzalez@linux.intel.com>

Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
It uses vector comparison as much as possible. In general, the larger the
source string, the greater performance gain observed, reaching speedups of
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
preferred and AVX unaligned load is fast.

	* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
	strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
	stpcpy-avx2 and stpncpy-avx2.
	* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
	(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
	__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
	and __stpncpy_avx2.
	* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
	ifunc-strcpy.h}: rename header for a more generic name.
	* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
	(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
	AVX unaligned load is fast and vzeroupper is preferred.
	* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
	* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
	* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
	* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
	* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
	* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
---
 sysdeps/x86_64/multiarch/Makefile             |    3 +
 sysdeps/x86_64/multiarch/ifunc-impl-list.c    |   12 +
 ...ifunc-unaligned-ssse3.h => ifunc-strcpy.h} |    6 +
 sysdeps/x86_64/multiarch/stpcpy-avx2.S        |    3 +
 sysdeps/x86_64/multiarch/stpcpy.c             |    2 +-
 sysdeps/x86_64/multiarch/stpncpy-avx2.S       |    4 +
 sysdeps/x86_64/multiarch/stpncpy.c            |    2 +-
 sysdeps/x86_64/multiarch/strcat-avx2.S        |  275 +++++
 sysdeps/x86_64/multiarch/strcat.c             |    2 +-
 sysdeps/x86_64/multiarch/strcpy-avx2.S        | 1022 +++++++++++++++++
 sysdeps/x86_64/multiarch/strcpy.c             |    2 +-
 sysdeps/x86_64/multiarch/strncat-avx2.S       |    3 +
 sysdeps/x86_64/multiarch/strncat.c            |    2 +-
 sysdeps/x86_64/multiarch/strncpy-avx2.S       |    3 +
 sysdeps/x86_64/multiarch/strncpy.c            |    2 +-
 15 files changed, 1337 insertions(+), 6 deletions(-)
 rename sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h => ifunc-strcpy.h} (83%)
 create mode 100644 sysdeps/x86_64/multiarch/stpcpy-avx2.S
 create mode 100644 sysdeps/x86_64/multiarch/stpncpy-avx2.S
 create mode 100644 sysdeps/x86_64/multiarch/strcat-avx2.S
 create mode 100644 sysdeps/x86_64/multiarch/strcpy-avx2.S
 create mode 100644 sysdeps/x86_64/multiarch/strncat-avx2.S
 create mode 100644 sysdeps/x86_64/multiarch/strncpy-avx2.S

Comments

Leonardo Sandoval Oct. 31, 2018, 4:29 p.m. | #1
ping. OK for master?


On Mon, 2018-10-08 at 08:59 -0500,
leonardo.sandoval.gonzalez@linux.intel.com wrote:
> From: Leonardo Sandoval <leonardo.sandoval.gonzalez@linux.intel.com>
> 
> Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy
> with AVX2.
> It uses vector comparison as much as possible. In general, the larger
> the
> source string, the greater performance gain observed, reaching
> speedups of
> 1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
> strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper
> is
> preferred and AVX unaligned load is fast.
> 
> 	* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
> 	strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
> 	stpcpy-avx2 and stpncpy-avx2.
> 	* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
> 	(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
> 	__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
> 	and __stpncpy_avx2.
> 	* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
> 	ifunc-strcpy.h}: rename header for a more generic name.
> 	* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
> 	(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
> 	AVX unaligned load is fast and vzeroupper is preferred.
> 	* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
> 	* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
> 	* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
> 	* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
> 	* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
> 	* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
> ---
>  sysdeps/x86_64/multiarch/Makefile             |    3 +
>  sysdeps/x86_64/multiarch/ifunc-impl-list.c    |   12 +
>  ...ifunc-unaligned-ssse3.h => ifunc-strcpy.h} |    6 +
>  sysdeps/x86_64/multiarch/stpcpy-avx2.S        |    3 +
>  sysdeps/x86_64/multiarch/stpcpy.c             |    2 +-
>  sysdeps/x86_64/multiarch/stpncpy-avx2.S       |    4 +
>  sysdeps/x86_64/multiarch/stpncpy.c            |    2 +-
>  sysdeps/x86_64/multiarch/strcat-avx2.S        |  275 +++++
>  sysdeps/x86_64/multiarch/strcat.c             |    2 +-
>  sysdeps/x86_64/multiarch/strcpy-avx2.S        | 1022
> +++++++++++++++++
>  sysdeps/x86_64/multiarch/strcpy.c             |    2 +-
>  sysdeps/x86_64/multiarch/strncat-avx2.S       |    3 +
>  sysdeps/x86_64/multiarch/strncat.c            |    2 +-
>  sysdeps/x86_64/multiarch/strncpy-avx2.S       |    3 +
>  sysdeps/x86_64/multiarch/strncpy.c            |    2 +-
>  15 files changed, 1337 insertions(+), 6 deletions(-)
>  rename sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h => ifunc-
> strcpy.h} (83%)
>  create mode 100644 sysdeps/x86_64/multiarch/stpcpy-avx2.S
>  create mode 100644 sysdeps/x86_64/multiarch/stpncpy-avx2.S
>  create mode 100644 sysdeps/x86_64/multiarch/strcat-avx2.S
>  create mode 100644 sysdeps/x86_64/multiarch/strcpy-avx2.S
>  create mode 100644 sysdeps/x86_64/multiarch/strncat-avx2.S
>  create mode 100644 sysdeps/x86_64/multiarch/strncpy-avx2.S
>
H.J. Lu Oct. 31, 2018, 5:25 p.m. | #2
On Mon, Oct 8, 2018 at 6:59 AM
<leonardo.sandoval.gonzalez@linux.intel.com> wrote:
>
> From: Leonardo Sandoval <leonardo.sandoval.gonzalez@linux.intel.com>
>
> Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
> It uses vector comparison as much as possible. In general, the larger the
> source string, the greater performance gain observed, reaching speedups of
> 1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
> strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
> preferred and AVX unaligned load is fast.
>
>         * sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
>         strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
>         stpcpy-avx2 and stpncpy-avx2.
>         * sysdeps/x86_64/multiarch/ifunc-impl-list.c:
>         (__libc_ifunc_impl_list): Add tests for __strcat_avx2,
>         __strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
>         and __stpncpy_avx2.
>         * sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
>         ifunc-strcpy.h}: rename header for a more generic name.
>         * sysdeps/x86_64/multiarch/ifunc-strcpy.h:
>         (IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
>         AVX unaligned load is fast and vzeroupper is preferred.
>         * sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
>         * sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
>         * sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
>         * sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
>         * sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
>         * sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise

LGTM.

Thanks.

H.J.
Adhemerval Zanella Oct. 31, 2018, 6:36 p.m. | #3
> diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S b/sysdeps/x86_64/multiarch/strcat-avx2.S
> new file mode 100644
> index 00000000000..b0623564276
> --- /dev/null
> +++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
> @@ -0,0 +1,275 @@
> +/* strcat with AVX2

Is this really a gain on real work usage comparing to generic strcat (
(strcpy (dest + strlen (dest), src)) assuming optimized strcpy / strlen?
Wouldn't be simple and more i-cache friendly to use a custom generic 
implementation that calls AVX2 strcpy/strlen (such as powerpc64 does)?
Rich Felker Oct. 31, 2018, 6:54 p.m. | #4
On Wed, Oct 31, 2018 at 03:36:10PM -0300, Adhemerval Zanella wrote:
> 
> 
> 
> > diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S b/sysdeps/x86_64/multiarch/strcat-avx2.S
> > new file mode 100644
> > index 00000000000..b0623564276
> > --- /dev/null
> > +++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
> > @@ -0,0 +1,275 @@
> > +/* strcat with AVX2
> 
> Is this really a gain on real work usage comparing to generic strcat (
> (strcpy (dest + strlen (dest), src)) assuming optimized strcpy / strlen?
> Wouldn't be simple and more i-cache friendly to use a custom generic 
> implementation that calls AVX2 strcpy/strlen (such as powerpc64 does)?

I second this, and fail to see the advantage of increasing the volume
of asm without a good reason. In this case specifically:

- Improvement over trivial strcpy(dest+strlen(dest),src), assuming
  those functions are optimized, is at best a constant difference in
  overhead, vs the O(m+n) runtime of the operation.

- Use of strcat at all is a major antipattern, typically leading to
  O(n²) time and buffer overflows. Thus optimizing it at all seems
  dubious (further encouraging its use "because it's fast").

Rich
Leonardo Sandoval Oct. 31, 2018, 8:23 p.m. | #5
On Wed, 2018-10-31 at 15:36 -0300, Adhemerval Zanella wrote:
> 
> 
> > diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S
> > b/sysdeps/x86_64/multiarch/strcat-avx2.S
> > new file mode 100644
> > index 00000000000..b0623564276
> > --- /dev/null
> > +++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
> > @@ -0,0 +1,275 @@
> > +/* strcat with AVX2
> 
> Is this really a gain on real work usage comparing to generic strcat
> (
> (strcpy (dest + strlen (dest), src)) assuming optimized strcpy /
> strlen?

The speedup is briefly summarized on the commit description but the
comparison was done between SSE2 Unaligned and AVX2 (current code).
Either SSE2 and AVX2 are much faster the generic strcat (I can provide
numbers).


> Wouldn't be simple and more i-cache friendly to use a custom generic 
> implementation that calls AVX2 strcpy/strlen (such as powerpc64
> does)?

I believe this is what we have right now through IFUNC_SELECTOR.
Adhemerval Zanella Oct. 31, 2018, 8:59 p.m. | #6
On 31/10/2018 17:23, Leonardo Sandoval wrote:
> On Wed, 2018-10-31 at 15:36 -0300, Adhemerval Zanella wrote:
>>
>>
>>> diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S
>>> b/sysdeps/x86_64/multiarch/strcat-avx2.S
>>> new file mode 100644
>>> index 00000000000..b0623564276
>>> --- /dev/null
>>> +++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
>>> @@ -0,0 +1,275 @@
>>> +/* strcat with AVX2
>>
>> Is this really a gain on real work usage comparing to generic strcat
>> (
>> (strcpy (dest + strlen (dest), src)) assuming optimized strcpy /
>> strlen?
> 
> The speedup is briefly summarized on the commit description but the
> comparison was done between SSE2 Unaligned and AVX2 (current code).
> Either SSE2 and AVX2 are much faster the generic strcat (I can provide
> numbers).
> 

As Rich has put, this strcat optimization seems to use the very
strategy of strcpy/strlen and basically brings a constant gain with
downside of i-cache overhead plus maintainability cost (since a possible 
bug in strcpy/strlen implementation will need to be checked against
strcat as well).

I also tend to agree with Rich that strcat is an antipattern, and I
would prefer we use simpler generic optimization for such cases.

> 
>> Wouldn't be simple and more i-cache friendly to use a custom generic 
>> implementation that calls AVX2 strcpy/strlen (such as powerpc64
>> does)?
> 
> I believe this is what we have right now through IFUNC_SELECTOR.
> 

What I meant was use something like:

char *strcat_avx2 (char *dest, const char *src)
{
  __strcpy_avx2 (dest + __strlen_avx2 (dest), src);
  return dest;
}
Leonardo Sandoval Nov. 1, 2018, 5:29 p.m. | #7
On Wed, 2018-10-31 at 17:59 -0300, Adhemerval Zanella wrote:
> 
> On 31/10/2018 17:23, Leonardo Sandoval wrote:
> > On Wed, 2018-10-31 at 15:36 -0300, Adhemerval Zanella wrote:
> > > 
> > > > diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S
> > > > b/sysdeps/x86_64/multiarch/strcat-avx2.S
> > > > new file mode 100644
> > > > index 00000000000..b0623564276
> > > > --- /dev/null
> > > > +++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
> > > > @@ -0,0 +1,275 @@
> > > > +/* strcat with AVX2
> > > 
> > > Is this really a gain on real work usage comparing to generic
> > > strcat
> > > (
> > > (strcpy (dest + strlen (dest), src)) assuming optimized strcpy /
> > > strlen?
> > 
> > The speedup is briefly summarized on the commit description but the
> > comparison was done between SSE2 Unaligned and AVX2 (current code).
> > Either SSE2 and AVX2 are much faster the generic strcat (I can
> > provide
> > numbers).
> > 
> 
> As Rich has put, this strcat optimization seems to use the very
> strategy of strcpy/strlen and basically brings a constant gain with
> downside of i-cache overhead plus maintainability cost (since a
> possible 
> bug in strcpy/strlen implementation will need to be checked against
> strcat as well).
> 
> I also tend to agree with Rich that strcat is an antipattern, and I
> would prefer we use simpler generic optimization for such cases.
> 
> > > Wouldn't be simple and more i-cache friendly to use a custom
> > > generic 
> > > implementation that calls AVX2 strcpy/strlen (such as powerpc64
> > > does)?
> > 
> > I believe this is what we have right now through IFUNC_SELECTOR.
> > 
> 
> What I meant was use something like:
> 
> char *strcat_avx2 (char *dest, const char *src)
> {
>   __strcpy_avx2 (dest + __strlen_avx2 (dest), src);
>   return dest;
> }

Thanks Rich and Adhemerval. Agree in all comments. V2 will be out soon.

Leo

Patch

diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index bb5e9707352..395e432c092 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -24,11 +24,14 @@  sysdep_routines += strncat-c stpncpy-c strncpy-c \
 		   strchr-sse2 strchrnul-sse2 strchr-avx2 strchrnul-avx2 \
 		   strrchr-sse2 strrchr-avx2 \
 		   strlen-sse2 strnlen-sse2 strlen-avx2 strnlen-avx2 \
+		   strcat-avx2 strncat-avx2 \
 		   strcat-ssse3 strncat-ssse3\
+		   strcpy-avx2 strncpy-avx2 \
 		   strcpy-sse2 stpcpy-sse2 \
 		   strcpy-ssse3 strncpy-ssse3 stpcpy-ssse3 stpncpy-ssse3 \
 		   strcpy-sse2-unaligned strncpy-sse2-unaligned \
 		   stpcpy-sse2-unaligned stpncpy-sse2-unaligned \
+		   stpcpy-avx2 stpncpy-avx2 \
 		   strcat-sse2 \
 		   strcat-sse2-unaligned strncat-sse2-unaligned \
 		   strchr-sse2-no-bsf memcmp-ssse3 strstr-sse2-unaligned \
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index 9aaaef7251b..950bd9eccb7 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -199,6 +199,8 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
   IFUNC_IMPL (i, name, stpncpy,
 	      IFUNC_IMPL_ADD (array, i, stpncpy, HAS_CPU_FEATURE (SSSE3),
 			      __stpncpy_ssse3)
+	      IFUNC_IMPL_ADD (array, i, stpncpy, HAS_ARCH_FEATURE (AVX2_Usable),
+			      __stpncpy_avx2)
 	      IFUNC_IMPL_ADD (array, i, stpncpy, 1,
 			      __stpncpy_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, stpncpy, 1, __stpncpy_sse2))
@@ -207,6 +209,8 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
   IFUNC_IMPL (i, name, stpcpy,
 	      IFUNC_IMPL_ADD (array, i, stpcpy, HAS_CPU_FEATURE (SSSE3),
 			      __stpcpy_ssse3)
+	      IFUNC_IMPL_ADD (array, i, stpcpy, HAS_ARCH_FEATURE (AVX2_Usable),
+			      __stpcpy_avx2)
 	      IFUNC_IMPL_ADD (array, i, stpcpy, 1, __stpcpy_sse2_unaligned)
 	      IFUNC_IMPL_ADD (array, i, stpcpy, 1, __stpcpy_sse2))
 
@@ -239,6 +243,8 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 
   /* Support sysdeps/x86_64/multiarch/strcat.c.  */
   IFUNC_IMPL (i, name, strcat,
+	      IFUNC_IMPL_ADD (array, i, strcat, HAS_ARCH_FEATURE (AVX2_Usable),
+			      __strcat_avx2)
 	      IFUNC_IMPL_ADD (array, i, strcat, HAS_CPU_FEATURE (SSSE3),
 			      __strcat_ssse3)
 	      IFUNC_IMPL_ADD (array, i, strcat, 1, __strcat_sse2_unaligned)
@@ -280,6 +286,8 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 
   /* Support sysdeps/x86_64/multiarch/strcpy.c.  */
   IFUNC_IMPL (i, name, strcpy,
+	      IFUNC_IMPL_ADD (array, i, strcpy, HAS_ARCH_FEATURE (AVX2_Usable),
+			      __strcpy_avx2)
 	      IFUNC_IMPL_ADD (array, i, strcpy, HAS_CPU_FEATURE (SSSE3),
 			      __strcpy_ssse3)
 	      IFUNC_IMPL_ADD (array, i, strcpy, 1, __strcpy_sse2_unaligned)
@@ -321,6 +329,8 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 
   /* Support sysdeps/x86_64/multiarch/strncat.c.  */
   IFUNC_IMPL (i, name, strncat,
+	      IFUNC_IMPL_ADD (array, i, strncat, HAS_ARCH_FEATURE (AVX2_Usable),
+			      __strncat_avx2)
 	      IFUNC_IMPL_ADD (array, i, strncat, HAS_CPU_FEATURE (SSSE3),
 			      __strncat_ssse3)
 	      IFUNC_IMPL_ADD (array, i, strncat, 1,
@@ -329,6 +339,8 @@  __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 
   /* Support sysdeps/x86_64/multiarch/strncpy.c.  */
   IFUNC_IMPL (i, name, strncpy,
+	      IFUNC_IMPL_ADD (array, i, strncpy, HAS_ARCH_FEATURE (AVX2_Usable),
+ 			      __strncpy_avx2)
 	      IFUNC_IMPL_ADD (array, i, strncpy, HAS_CPU_FEATURE (SSSE3),
 			      __strncpy_ssse3)
 	      IFUNC_IMPL_ADD (array, i, strncpy, 1,
diff --git a/sysdeps/x86_64/multiarch/ifunc-unaligned-ssse3.h b/sysdeps/x86_64/multiarch/ifunc-strcpy.h
similarity index 83%
rename from sysdeps/x86_64/multiarch/ifunc-unaligned-ssse3.h
rename to sysdeps/x86_64/multiarch/ifunc-strcpy.h
index 81805f98323..4f2286fefcc 100644
--- a/sysdeps/x86_64/multiarch/ifunc-unaligned-ssse3.h
+++ b/sysdeps/x86_64/multiarch/ifunc-strcpy.h
@@ -24,12 +24,18 @@  extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden;
 extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2_unaligned)
   attribute_hidden;
 extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden;
 
 static inline void *
 IFUNC_SELECTOR (void)
 {
   const struct cpu_features* cpu_features = __get_cpu_features ();
 
+  if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)
+      && CPU_FEATURES_ARCH_P (cpu_features, AVX2_Usable)
+      && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
+    return OPTIMIZE (avx2);
+
   if (CPU_FEATURES_ARCH_P (cpu_features, Fast_Unaligned_Load))
     return OPTIMIZE (sse2_unaligned);
 
diff --git a/sysdeps/x86_64/multiarch/stpcpy-avx2.S b/sysdeps/x86_64/multiarch/stpcpy-avx2.S
new file mode 100644
index 00000000000..f0bd3029fe3
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/stpcpy-avx2.S
@@ -0,0 +1,3 @@ 
+#define USE_AS_STPCPY
+#define STRCPY __stpcpy_avx2
+#include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/stpcpy.c b/sysdeps/x86_64/multiarch/stpcpy.c
index 1e340fca991..8ffd13b48c8 100644
--- a/sysdeps/x86_64/multiarch/stpcpy.c
+++ b/sysdeps/x86_64/multiarch/stpcpy.c
@@ -28,7 +28,7 @@ 
 # undef __stpcpy
 
 # define SYMBOL_NAME stpcpy
-# include "ifunc-unaligned-ssse3.h"
+# include "ifunc-strcpy.h"
 
 libc_ifunc_redirected (__redirect_stpcpy, __stpcpy, IFUNC_SELECTOR ());
 
diff --git a/sysdeps/x86_64/multiarch/stpncpy-avx2.S b/sysdeps/x86_64/multiarch/stpncpy-avx2.S
new file mode 100644
index 00000000000..032b0407d08
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/stpncpy-avx2.S
@@ -0,0 +1,4 @@ 
+#define USE_AS_STPCPY
+#define USE_AS_STRNCPY
+#define STRCPY __stpncpy_avx2
+#include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/stpncpy.c b/sysdeps/x86_64/multiarch/stpncpy.c
index 28842ece2b0..f3e203f78cc 100644
--- a/sysdeps/x86_64/multiarch/stpncpy.c
+++ b/sysdeps/x86_64/multiarch/stpncpy.c
@@ -26,7 +26,7 @@ 
 # undef __stpncpy
 
 # define SYMBOL_NAME stpncpy
-# include "ifunc-unaligned-ssse3.h"
+# include "ifunc-strcpy.h"
 
 libc_ifunc_redirected (__redirect_stpncpy, __stpncpy, IFUNC_SELECTOR ());
 
diff --git a/sysdeps/x86_64/multiarch/strcat-avx2.S b/sysdeps/x86_64/multiarch/strcat-avx2.S
new file mode 100644
index 00000000000..b0623564276
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strcat-avx2.S
@@ -0,0 +1,275 @@ 
+/* strcat with AVX2
+   Copyright (C) 2011-2018 Free Software Foundation, Inc.
+   Contributed by Intel Corporation.
+   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>
+
+# ifndef STRCAT
+#  define STRCAT  __strcat_avx2
+# endif
+
+# define USE_AS_STRCAT
+
+/* Number of bytes in a vector register */
+# define VEC_SIZE	32
+
+	.section .text.avx,"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"
+#endif
diff --git a/sysdeps/x86_64/multiarch/strcat.c b/sysdeps/x86_64/multiarch/strcat.c
index 1f7f6263f35..694b9b24058 100644
--- a/sysdeps/x86_64/multiarch/strcat.c
+++ b/sysdeps/x86_64/multiarch/strcat.c
@@ -24,7 +24,7 @@ 
 # undef strcat
 
 # define SYMBOL_NAME strcat
-# include "ifunc-unaligned-ssse3.h"
+# include "ifunc-strcpy.h"
 
 libc_ifunc_redirected (__redirect_strcat, strcat, IFUNC_SELECTOR ());
 
diff --git a/sysdeps/x86_64/multiarch/strcpy-avx2.S b/sysdeps/x86_64/multiarch/strcpy-avx2.S
new file mode 100644
index 00000000000..81677f90607
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strcpy-avx2.S
@@ -0,0 +1,1022 @@ 
+/* strcpy with AVX2
+   Copyright (C) 2011-2018 Free Software Foundation, Inc.
+   Contributed by Intel Corporation.
+   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)
+
+# 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
+
+/* zero register */
+#define xmmZ	xmm0
+#define ymmZ	ymm0
+
+/* mask register */
+#define ymmM	ymm1
+
+# ifndef USE_AS_STRCAT
+
+	.section .text.avx,"ax",@progbits
+ENTRY (STRCPY)
+#  ifdef USE_AS_STRNCPY
+	mov	%rdx, %r8
+	test	%r8, %r8
+	jz	L(ExitZero)
+#  endif
+	mov	%rsi, %rcx
+#  ifndef USE_AS_STPCPY
+	mov	%rdi, %rax      /* save result */
+#  endif
+
+# 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)
+# endif
+	test	%edx, %edx
+	jnz	L(CopyVecSizeTail)
+
+	vpcmpeqb VEC_SIZE(%rsi), %ymmZ, %ymm2
+	vpmovmskb %ymm2, %edx
+
+# 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)
+# else
+	jnz	L(CopyVecSize)
+# 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
+
+	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)
+# else
+	jnz	L(CopyVecSize)
+# 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)
+# else
+	jnz	L(CopyVecSize)
+# 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
+
+	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
+
+	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)
+# 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)
+# 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)
+
+/*------End of main part with loops---------------------*/
+
+/* Case1 */
+
+# 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)
+# 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
+	VZEROUPPER
+	ret
+
+	.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)
+# 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)
+# 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
+	ret
+
+	.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
+	ret
+
+	.p2align 4
+L(Exit3):
+	mov	(%rsi), %edx
+	mov	%edx, (%rdi)
+# 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)
+# endif
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(Exit4_7):
+	mov	(%rsi), %ecx
+	mov	%ecx, (%rdi)
+	mov	-3(%rsi, %rdx), %ecx
+	mov	%ecx, -3(%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
+	ret
+
+	.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
+	ret
+
+	.p2align 4
+L(Exit16_31):
+	vmovdqu (%rsi), %xmm2
+	vmovdqu -15(%rsi, %rdx), %xmm3
+	vmovdqu %xmm2, (%rdi)
+	vmovdqu %xmm3, -15(%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
+	ret
+
+	.p2align 4
+L(Exit32_63):
+	vmovdqu (%rsi), %ymm2
+	vmovdqu -31(%rsi, %rdx), %ymm3
+	vmovdqu %ymm2, (%rdi)
+	vmovdqu %ymm3, -31(%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
+	ret
+
+# ifdef USE_AS_STRNCPY
+
+	.p2align 4
+L(StrncpyExit1):
+	movzbl	(%rsi), %edx
+	mov	%dl, (%rdi)
+#  ifdef USE_AS_STPCPY
+	lea	1(%rdi), %rax
+#  endif
+#  ifdef USE_AS_STRCAT
+	movb	$0, 1(%rdi)
+#  endif
+	VZEROUPPER
+	ret
+
+	.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
+	ret
+
+	.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
+	ret
+
+	.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
+	ret
+
+	.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
+	ret
+
+	.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
+	ret
+
+	.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
+	ret
+
+	.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
+	ret
+
+#  ifndef USE_AS_STRCAT
+
+	.p2align 4
+L(Fill1):
+	mov	%dl, (%rdi)
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(Fill2):
+	mov	%dx, (%rdi)
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(Fill3_4):
+	mov	%dx, (%rdi)
+	mov     %dx, -2(%rdi, %r8)
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(Fill5_8):
+	mov	%edx, (%rdi)
+	mov     %edx, -4(%rdi, %r8)
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(Fill9_16):
+	mov	%rdx, (%rdi)
+	mov	%rdx, -8(%rdi, %r8)
+	VZEROUPPER
+	ret
+
+	.p2align 4
+L(Fill17_32):
+	vmovdqu %xmmZ, (%rdi)
+	vmovdqu %xmmZ, -16(%rdi, %r8)
+	VZEROUPPER
+	ret
+
+	.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
+	ret
+
+/* end of ifndef USE_AS_STRCAT */
+#  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)
+#  ifdef USE_AS_STPCPY
+	lea	(VEC_SIZE * 4)(%rdi), %rax
+#  endif
+#  ifdef USE_AS_STRCAT
+	movb	$0, (VEC_SIZE * 4)(%rdi)
+#  endif
+	VZEROUPPER
+	ret
+
+	.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
+
+	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
+
+	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
+	ret
+
+	.p2align 4
+L(ExitZero):
+#  ifndef USE_AS_STRCAT
+	mov	%rdi, %rax
+#  endif
+	VZEROUPPER
+	ret
+
+# endif
+
+# ifndef USE_AS_STRCAT
+END (STRCPY)
+# else
+END (STRCAT)
+# endif
+#endif
diff --git a/sysdeps/x86_64/multiarch/strcpy.c b/sysdeps/x86_64/multiarch/strcpy.c
index 12e0e3ffe20..ecf90d4b044 100644
--- a/sysdeps/x86_64/multiarch/strcpy.c
+++ b/sysdeps/x86_64/multiarch/strcpy.c
@@ -24,7 +24,7 @@ 
 # undef strcpy
 
 # define SYMBOL_NAME strcpy
-# include "ifunc-unaligned-ssse3.h"
+# include "ifunc-strcpy.h"
 
 libc_ifunc_redirected (__redirect_strcpy, strcpy, IFUNC_SELECTOR ());
 
diff --git a/sysdeps/x86_64/multiarch/strncat-avx2.S b/sysdeps/x86_64/multiarch/strncat-avx2.S
new file mode 100644
index 00000000000..bfefa659bb6
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strncat-avx2.S
@@ -0,0 +1,3 @@ 
+#define USE_AS_STRNCAT
+#define STRCAT __strncat_avx2
+#include "strcat-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strncat.c b/sysdeps/x86_64/multiarch/strncat.c
index 841c165565a..74f7d028ae2 100644
--- a/sysdeps/x86_64/multiarch/strncat.c
+++ b/sysdeps/x86_64/multiarch/strncat.c
@@ -24,7 +24,7 @@ 
 # undef strncat
 
 # define SYMBOL_NAME strncat
-# include "ifunc-unaligned-ssse3.h"
+# include "ifunc-strcpy.h"
 
 libc_ifunc_redirected (__redirect_strncat, strncat, IFUNC_SELECTOR ());
 strong_alias (strncat, __strncat);
diff --git a/sysdeps/x86_64/multiarch/strncpy-avx2.S b/sysdeps/x86_64/multiarch/strncpy-avx2.S
new file mode 100644
index 00000000000..9ef8c87627d
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strncpy-avx2.S
@@ -0,0 +1,3 @@ 
+#define USE_AS_STRNCPY
+#define STRCPY __strncpy_avx2
+#include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strncpy.c b/sysdeps/x86_64/multiarch/strncpy.c
index 3c3de8b18eb..93dfb4cfde7 100644
--- a/sysdeps/x86_64/multiarch/strncpy.c
+++ b/sysdeps/x86_64/multiarch/strncpy.c
@@ -24,7 +24,7 @@ 
 # undef strncpy
 
 # define SYMBOL_NAME strncpy
-# include "ifunc-unaligned-ssse3.h"
+# include "ifunc-strcpy.h"
 
 libc_ifunc_redirected (__redirect_strncpy, strncpy, IFUNC_SELECTOR ());