[PATCH] x86_64: memcpy/memmove family optimized with AVX512
Paul Pluzhnikov
ppluzhnikov@google.com
Mon Jan 18 22:57:00 GMT 2016
On Mon, Jan 18, 2016 at 2:29 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Mon, Jan 18, 2016 at 10:50 AM, Paul Pluzhnikov
> <ppluzhnikov@google.com> wrote:
>> This patch appears to break my build:
>>
> Please try this.
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -49,8 +49,10 @@ __libc_ifunc_impl_list (const char *name, struct
libc_ifunc_impl *array,
/* Support sysdeps/x86_64/multiarch/memmove_chk.c. */
IFUNC_IMPL (i, name, __memmove_chk,
IFUNC_IMPL_ADD (array, i, __memmove_chk,
+#ifdef HAVE_AVX512_ASM_SUPPORT
HAS_ARCH_FEATURE (AVX512F_Usable),
__memmove_chk_avx512_no_vzeroupper)
+#endif
The #ifdef should be moved one line up, I think.
@@ -309,8 +317,10 @@ __libc_ifunc_impl_list (const char *name, struct
libc_ifunc_impl *array,
/* Support sysdeps/x86_64/multiarch/mempcpy.S. */
IFUNC_IMPL (i, name, mempcpy,
IFUNC_IMPL_ADD (array, i, mempcpy,
+# ifdef HAVE_AVX512_ASM_SUPPORT
HAS_ARCH_FEATURE (AVX512F_Usable),
__mempcpy_avx512_no_vzeroupper)
+# endif
IFUNC_IMPL_ADD (array, i, mempcpy,
HAS_ARCH_FEATURE (AVX_Usable),
__mempcpy_avx_unaligned)
Likewise.
You also missed __memmove_avx512_no_vzeroupper.
Attached patch makes it build and pass tests for me.
Thanks,
--
Paul Pluzhnikov
-------------- next part --------------
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index d9ff016..188b6d3 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -48,9 +48,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
/* Support sysdeps/x86_64/multiarch/memmove_chk.c. */
IFUNC_IMPL (i, name, __memmove_chk,
+#ifdef HAVE_AVX512_ASM_SUPPORT
IFUNC_IMPL_ADD (array, i, __memmove_chk,
HAS_ARCH_FEATURE (AVX512F_Usable),
__memmove_chk_avx512_no_vzeroupper)
+#endif
IFUNC_IMPL_ADD (array, i, __memmove_chk,
HAS_ARCH_FEATURE (AVX_Usable),
__memmove_chk_avx_unaligned)
@@ -68,9 +70,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, memmove,
HAS_ARCH_FEATURE (AVX_Usable),
__memmove_avx_unaligned)
+#ifdef HAVE_AVX512_ASM_SUPPORT
IFUNC_IMPL_ADD (array, i, memmove,
HAS_ARCH_FEATURE (AVX512F_Usable),
__memmove_avx512_no_vzeroupper)
+#endif
IFUNC_IMPL_ADD (array, i, memmove, HAS_CPU_FEATURE (SSSE3),
__memmove_ssse3_back)
IFUNC_IMPL_ADD (array, i, memmove, HAS_CPU_FEATURE (SSSE3),
@@ -259,9 +263,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
#ifdef SHARED
/* Support sysdeps/x86_64/multiarch/memcpy_chk.S. */
IFUNC_IMPL (i, name, __memcpy_chk,
+#ifdef HAVE_AVX512_ASM_SUPPORT
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
HAS_ARCH_FEATURE (AVX512F_Usable),
__memcpy_chk_avx512_no_vzeroupper)
+#endif
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
HAS_ARCH_FEATURE (AVX_Usable),
__memcpy_chk_avx_unaligned)
@@ -283,17 +289,21 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__memcpy_ssse3_back)
IFUNC_IMPL_ADD (array, i, memcpy, HAS_CPU_FEATURE (SSSE3),
__memcpy_ssse3)
+#ifdef HAVE_AVX512_ASM_SUPPORT
IFUNC_IMPL_ADD (array, i, memcpy,
HAS_ARCH_FEATURE (AVX512F_Usable),
__memcpy_avx512_no_vzeroupper)
+#endif
IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_sse2_unaligned)
IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_sse2))
/* Support sysdeps/x86_64/multiarch/mempcpy_chk.S. */
IFUNC_IMPL (i, name, __mempcpy_chk,
+#ifdef HAVE_AVX512_ASM_SUPPORT
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
HAS_ARCH_FEATURE (AVX512F_Usable),
__mempcpy_chk_avx512_no_vzeroupper)
+#endif
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
HAS_ARCH_FEATURE (AVX_Usable),
__mempcpy_chk_avx_unaligned)
@@ -308,9 +318,11 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
/* Support sysdeps/x86_64/multiarch/mempcpy.S. */
IFUNC_IMPL (i, name, mempcpy,
+#ifdef HAVE_AVX512_ASM_SUPPORT
IFUNC_IMPL_ADD (array, i, mempcpy,
HAS_ARCH_FEATURE (AVX512F_Usable),
__mempcpy_avx512_no_vzeroupper)
+#endif
IFUNC_IMPL_ADD (array, i, mempcpy,
HAS_ARCH_FEATURE (AVX_Usable),
__mempcpy_avx_unaligned)
More information about the Libc-alpha
mailing list