From: "H.J. Lu" <hjl.tools@gmail.com>
To: libc-alpha@sourceware.org
Subject: [PATCH v2 10/10] x86-64: Use ZMM16-ZMM31 in AVX512 memmove family functions
Date: Mon, 15 Mar 2021 07:25:20 -0700 [thread overview]
Message-ID: <20210315142520.1661407-11-hjl.tools@gmail.com> (raw)
In-Reply-To: <20210315142520.1661407-1-hjl.tools@gmail.com>
Update ifunc-memmove.h to select the function optimized with AVX512
instructions using ZMM16-ZMM31 registers to avoid RTM abort with usable
AVX512VL since VZEROUPPER isn't needed at function exit.
---
sysdeps/x86_64/multiarch/ifunc-impl-list.c | 24 +++++++++---------
sysdeps/x86_64/multiarch/ifunc-memmove.h | 12 +++++----
.../multiarch/memmove-avx512-unaligned-erms.S | 25 +++++++++++++++++--
3 files changed, 42 insertions(+), 19 deletions(-)
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index 37f17075fa..770986e41a 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -83,10 +83,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
CPU_FEATURE_USABLE (AVX512F),
__memmove_chk_avx512_no_vzeroupper)
IFUNC_IMPL_ADD (array, i, __memmove_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memmove_chk_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, __memmove_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memmove_chk_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __memmove_chk,
CPU_FEATURE_USABLE (AVX),
@@ -148,10 +148,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
CPU_FEATURE_USABLE (AVX512F),
__memmove_avx512_no_vzeroupper)
IFUNC_IMPL_ADD (array, i, memmove,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memmove_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, memmove,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memmove_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, memmove, CPU_FEATURE_USABLE (SSSE3),
__memmove_ssse3_back)
@@ -733,10 +733,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
CPU_FEATURE_USABLE (AVX512F),
__memcpy_chk_avx512_no_vzeroupper)
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memcpy_chk_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memcpy_chk_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
CPU_FEATURE_USABLE (AVX),
@@ -802,10 +802,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
CPU_FEATURE_USABLE (AVX512F),
__memcpy_avx512_no_vzeroupper)
IFUNC_IMPL_ADD (array, i, memcpy,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memcpy_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, memcpy,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__memcpy_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_sse2_unaligned)
IFUNC_IMPL_ADD (array, i, memcpy, 1,
@@ -819,10 +819,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
CPU_FEATURE_USABLE (AVX512F),
__mempcpy_chk_avx512_no_vzeroupper)
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__mempcpy_chk_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__mempcpy_chk_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
CPU_FEATURE_USABLE (AVX),
@@ -864,10 +864,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
CPU_FEATURE_USABLE (AVX512F),
__mempcpy_avx512_no_vzeroupper)
IFUNC_IMPL_ADD (array, i, mempcpy,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__mempcpy_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, mempcpy,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__mempcpy_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, mempcpy,
CPU_FEATURE_USABLE (AVX),
diff --git a/sysdeps/x86_64/multiarch/ifunc-memmove.h b/sysdeps/x86_64/multiarch/ifunc-memmove.h
index 4eba926eca..a14718a970 100644
--- a/sysdeps/x86_64/multiarch/ifunc-memmove.h
+++ b/sysdeps/x86_64/multiarch/ifunc-memmove.h
@@ -56,13 +56,15 @@ IFUNC_SELECTOR (void)
if (CPU_FEATURE_USABLE_P (cpu_features, AVX512F)
&& !CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_AVX512))
{
- if (CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER))
- return OPTIMIZE (avx512_no_vzeroupper);
+ if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL))
+ {
+ if (CPU_FEATURE_USABLE_P (cpu_features, ERMS))
+ return OPTIMIZE (avx512_unaligned_erms);
- if (CPU_FEATURE_USABLE_P (cpu_features, ERMS))
- return OPTIMIZE (avx512_unaligned_erms);
+ return OPTIMIZE (avx512_unaligned);
+ }
- return OPTIMIZE (avx512_unaligned);
+ return OPTIMIZE (avx512_no_vzeroupper);
}
if (CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
diff --git a/sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S b/sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S
index aac1515cf6..848848ab39 100644
--- a/sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S
+++ b/sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S
@@ -1,11 +1,32 @@
#if IS_IN (libc)
# define VEC_SIZE 64
-# define VEC(i) zmm##i
+# define XMM0 xmm16
+# define XMM1 xmm17
+# define YMM0 ymm16
+# define YMM1 ymm17
+# define VEC0 zmm16
+# define VEC1 zmm17
+# define VEC2 zmm18
+# define VEC3 zmm19
+# define VEC4 zmm20
+# define VEC5 zmm21
+# define VEC6 zmm22
+# define VEC7 zmm23
+# define VEC8 zmm24
+# define VEC9 zmm25
+# define VEC10 zmm26
+# define VEC11 zmm27
+# define VEC12 zmm28
+# define VEC13 zmm29
+# define VEC14 zmm30
+# define VEC15 zmm31
+# define VEC(i) VEC##i
# define VMOVNT vmovntdq
# define VMOVU vmovdqu64
# define VMOVA vmovdqa64
+# define VZEROUPPER
-# define SECTION(p) p##.avx512
+# define SECTION(p) p##.evex512
# define MEMMOVE_SYMBOL(p,s) p##_avx512_##s
# include "memmove-vec-unaligned-erms.S"
--
2.30.2
next prev parent reply other threads:[~2021-03-15 14:25 UTC|newest]
Thread overview: 14+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-15 14:25 [PATCH v2 00/10] x86-64: Avoid RTM abort inside a RTM region H.J. Lu
2021-03-15 14:25 ` [PATCH v2 01/10] x86: Set Prefer_No_VZEROUPPER and add Prefer_AVX2_STRCMP H.J. Lu
2021-03-15 14:25 ` [PATCH v2 02/10] x86-64: Add ifunc-avx2.h functions with 256-bit EVEX H.J. Lu
2021-03-15 14:25 ` [PATCH v2 03/10] x86-64: Add strcpy family " H.J. Lu
2021-03-15 14:25 ` [PATCH v2 04/10] x86-64: Add memmove " H.J. Lu
2021-03-15 14:25 ` [PATCH v2 05/10] x86-64: Add memset " H.J. Lu
2021-03-15 14:25 ` [PATCH v2 06/10] x86-64: Add memcmp " H.J. Lu
2021-03-15 14:25 ` [PATCH v2 07/10] x86-64: Add AVX optimized string/memory functions for RTM H.J. Lu
2021-03-15 14:25 ` [PATCH v2 08/10] x86: Add string/memory function tests in RTM region H.J. Lu
2021-03-15 14:25 ` [PATCH v2 09/10] x86-64: Use ZMM16-ZMM31 in AVX512 memset family functions H.J. Lu
2021-03-15 14:25 ` H.J. Lu [this message]
2021-03-24 18:03 ` [PATCH v2 00/10] x86-64: Avoid RTM abort inside a RTM region H.J. Lu
2021-03-29 23:06 ` H.J. Lu
2022-01-27 17:13 ` H.J. Lu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20210315142520.1661407-11-hjl.tools@gmail.com \
--to=hjl.tools@gmail.com \
--cc=libc-alpha@sourceware.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).