From: "H.J. Lu" <hjl.tools@gmail.com>
To: libc-alpha@sourceware.org
Subject: [PATCH v2 09/10] x86-64: Use ZMM16-ZMM31 in AVX512 memset family functions
Date: Mon, 15 Mar 2021 07:25:19 -0700 [thread overview]
Message-ID: <20210315142520.1661407-10-hjl.tools@gmail.com> (raw)
In-Reply-To: <20210315142520.1661407-1-hjl.tools@gmail.com>
Update ifunc-memset.h/ifunc-wmemset.h to select the function optimized
with AVX512 instructions using ZMM16-ZMM31 registers to avoid RTM abort
with usable AVX512VL and AVX512BW since VZEROUPPER isn't needed at
function exit.
---
sysdeps/x86_64/multiarch/ifunc-impl-list.c | 14 +++++++++-----
sysdeps/x86_64/multiarch/ifunc-memset.h | 13 ++++++++-----
sysdeps/x86_64/multiarch/ifunc-wmemset.h | 12 ++++++------
.../multiarch/memset-avx512-unaligned-erms.S | 16 ++++++++--------
4 files changed, 31 insertions(+), 24 deletions(-)
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index 024913065b..37f17075fa 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -211,10 +211,12 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (AVX512BW)),
__memset_chk_evex_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __memset_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ (CPU_FEATURE_USABLE (AVX512VL)
+ && CPU_FEATURE_USABLE (AVX512BW)),
__memset_chk_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __memset_chk,
- CPU_FEATURE_USABLE (AVX512F),
+ (CPU_FEATURE_USABLE (AVX512VL)
+ && CPU_FEATURE_USABLE (AVX512BW)),
__memset_chk_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, __memset_chk,
CPU_FEATURE_USABLE (AVX512F),
@@ -252,10 +254,12 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (AVX512BW)),
__memset_evex_unaligned_erms)
IFUNC_IMPL_ADD (array, i, memset,
- CPU_FEATURE_USABLE (AVX512F),
+ (CPU_FEATURE_USABLE (AVX512VL)
+ && CPU_FEATURE_USABLE (AVX512BW)),
__memset_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, memset,
- CPU_FEATURE_USABLE (AVX512F),
+ (CPU_FEATURE_USABLE (AVX512VL)
+ && CPU_FEATURE_USABLE (AVX512BW)),
__memset_avx512_unaligned)
IFUNC_IMPL_ADD (array, i, memset,
CPU_FEATURE_USABLE (AVX512F),
@@ -719,7 +723,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
CPU_FEATURE_USABLE (AVX512VL),
__wmemset_evex_unaligned)
IFUNC_IMPL_ADD (array, i, wmemset,
- CPU_FEATURE_USABLE (AVX512F),
+ CPU_FEATURE_USABLE (AVX512VL),
__wmemset_avx512_unaligned))
#ifdef SHARED
diff --git a/sysdeps/x86_64/multiarch/ifunc-memset.h b/sysdeps/x86_64/multiarch/ifunc-memset.h
index 43655fb684..502f946a84 100644
--- a/sysdeps/x86_64/multiarch/ifunc-memset.h
+++ b/sysdeps/x86_64/multiarch/ifunc-memset.h
@@ -53,13 +53,16 @@ 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)
+ && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW))
+ {
+ 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_FEATURE_USABLE_P (cpu_features, AVX2))
diff --git a/sysdeps/x86_64/multiarch/ifunc-wmemset.h b/sysdeps/x86_64/multiarch/ifunc-wmemset.h
index 8d952eff99..756f0ccdbf 100644
--- a/sysdeps/x86_64/multiarch/ifunc-wmemset.h
+++ b/sysdeps/x86_64/multiarch/ifunc-wmemset.h
@@ -33,13 +33,13 @@ IFUNC_SELECTOR (void)
if (CPU_FEATURE_USABLE_P (cpu_features, AVX2)
&& CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
{
- if (CPU_FEATURE_USABLE_P (cpu_features, AVX512F)
- && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_AVX512)
- && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER))
- return OPTIMIZE (avx512_unaligned);
-
if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL))
- return OPTIMIZE (evex_unaligned);
+ {
+ if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_AVX512))
+ return OPTIMIZE (avx512_unaligned);
+
+ return OPTIMIZE (evex_unaligned);
+ }
if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
return OPTIMIZE (avx2_unaligned_rtm);
diff --git a/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S b/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
index 0783979ca5..22e7b187c8 100644
--- a/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
+++ b/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
@@ -1,22 +1,22 @@
#if IS_IN (libc)
# define VEC_SIZE 64
-# define VEC(i) zmm##i
+# define XMM0 xmm16
+# define YMM0 ymm16
+# define VEC0 zmm16
+# define VEC(i) VEC##i
# define VMOVU vmovdqu64
# define VMOVA vmovdqa64
+# define VZEROUPPER
# define MEMSET_VDUP_TO_VEC0_AND_SET_RETURN(d, r) \
- vmovd d, %xmm0; \
movq r, %rax; \
- vpbroadcastb %xmm0, %xmm0; \
- vpbroadcastq %xmm0, %zmm0
+ vpbroadcastb d, %VEC0
# define WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN(d, r) \
- vmovd d, %xmm0; \
movq r, %rax; \
- vpbroadcastd %xmm0, %xmm0; \
- vpbroadcastq %xmm0, %zmm0
+ vpbroadcastd d, %VEC0
-# define SECTION(p) p##.avx512
+# define SECTION(p) p##.evex512
# define MEMSET_SYMBOL(p,s) p##_avx512_##s
# define WMEMSET_SYMBOL(p,s) p##_avx512_##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 ` H.J. Lu [this message]
2021-03-15 14:25 ` [PATCH v2 10/10] x86-64: Use ZMM16-ZMM31 in AVX512 memmove family functions H.J. Lu
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=20210315142520.1661407-10-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).