public inbox for glibc-cvs@sourceware.org
help / color / mirror / Atom feed
* [glibc/maskray/clang] x86-64: Remove Prefer_AVX2_STRCMP
@ 2021-11-02 19:59 Fangrui Song
0 siblings, 0 replies; only message in thread
From: Fangrui Song @ 2021-11-02 19:59 UTC (permalink / raw)
To: glibc-cvs
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=14dbbf46a007ae5df36646b51ad0c9e5f5259f30
commit 14dbbf46a007ae5df36646b51ad0c9e5f5259f30
Author: H.J. Lu <hjl.tools@gmail.com>
Date: Fri Oct 29 12:56:53 2021 -0700
x86-64: Remove Prefer_AVX2_STRCMP
Remove Prefer_AVX2_STRCMP to enable EVEX strcmp. When comparing 2 32-byte
strings, EVEX strcmp has been improved to require 1 load, 1 VPTESTM, 1
VPCMP, 1 KMOVD and 1 INCL instead of 2 loads, 3 VPCMPs, 2 KORDs, 1 KMOVD
and 1 TESTL while AVX2 strcmp requires 1 load, 2 VPCMPEQs, 1 VPMINU, 1
VPMOVMSKB and 1 TESTL. EVEX strcmp is now faster than AVX2 strcmp by up
to 40% on Tiger Lake and Ice Lake.
Diff:
---
sysdeps/x86/cpu-features.c | 8 --------
sysdeps/x86/cpu-tunables.c | 2 --
sysdeps/x86/include/cpu-features-preferred_feature_index_1.def | 1 -
sysdeps/x86_64/multiarch/strcmp.c | 3 +--
sysdeps/x86_64/multiarch/strncmp.c | 3 +--
5 files changed, 2 insertions(+), 15 deletions(-)
diff --git a/sysdeps/x86/cpu-features.c b/sysdeps/x86/cpu-features.c
index 645bba6314..be2498b2e7 100644
--- a/sysdeps/x86/cpu-features.c
+++ b/sysdeps/x86/cpu-features.c
@@ -546,14 +546,6 @@ init_cpu_features (struct cpu_features *cpu_features)
if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
cpu_features->preferred[index_arch_Prefer_No_VZEROUPPER]
|= bit_arch_Prefer_No_VZEROUPPER;
-
- /* Since to compare 2 32-byte strings, 256-bit EVEX strcmp
- requires 2 loads, 3 VPCMPs and 2 KORDs while AVX2 strcmp
- requires 1 load, 2 VPCMPEQs, 1 VPMINU and 1 VPMOVMSKB,
- AVX2 strcmp is faster than EVEX strcmp. */
- if (CPU_FEATURE_USABLE_P (cpu_features, AVX2))
- cpu_features->preferred[index_arch_Prefer_AVX2_STRCMP]
- |= bit_arch_Prefer_AVX2_STRCMP;
}
/* Avoid avoid short distance REP MOVSB on processor with FSRM. */
diff --git a/sysdeps/x86/cpu-tunables.c b/sysdeps/x86/cpu-tunables.c
index 00fe5045eb..61b05e5b1d 100644
--- a/sysdeps/x86/cpu-tunables.c
+++ b/sysdeps/x86/cpu-tunables.c
@@ -239,8 +239,6 @@ TUNABLE_CALLBACK (set_hwcaps) (tunable_val_t *valp)
CHECK_GLIBC_IFUNC_PREFERRED_BOTH (n, cpu_features,
Fast_Copy_Backward,
disable, 18);
- CHECK_GLIBC_IFUNC_PREFERRED_NEED_BOTH
- (n, cpu_features, Prefer_AVX2_STRCMP, AVX2, disable, 18);
}
break;
case 19:
diff --git a/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def b/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
index d7c93f00c5..1530d594b3 100644
--- a/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
+++ b/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
@@ -32,5 +32,4 @@ BIT (Prefer_ERMS)
BIT (Prefer_No_AVX512)
BIT (MathVec_Prefer_No_AVX512)
BIT (Prefer_FSRM)
-BIT (Prefer_AVX2_STRCMP)
BIT (Avoid_Short_Distance_REP_MOVSB)
diff --git a/sysdeps/x86_64/multiarch/strcmp.c b/sysdeps/x86_64/multiarch/strcmp.c
index 62b7abeeee..7c2901bf44 100644
--- a/sysdeps/x86_64/multiarch/strcmp.c
+++ b/sysdeps/x86_64/multiarch/strcmp.c
@@ -43,8 +43,7 @@ IFUNC_SELECTOR (void)
{
if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
&& CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
- && CPU_FEATURE_USABLE_P (cpu_features, BMI2)
- && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_AVX2_STRCMP))
+ && CPU_FEATURE_USABLE_P (cpu_features, BMI2))
return OPTIMIZE (evex);
if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
diff --git a/sysdeps/x86_64/multiarch/strncmp.c b/sysdeps/x86_64/multiarch/strncmp.c
index 60ba0fe356..f94a421784 100644
--- a/sysdeps/x86_64/multiarch/strncmp.c
+++ b/sysdeps/x86_64/multiarch/strncmp.c
@@ -43,8 +43,7 @@ IFUNC_SELECTOR (void)
{
if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
&& CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
- && CPU_FEATURE_USABLE_P (cpu_features, BMI2)
- && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_AVX2_STRCMP))
+ && CPU_FEATURE_USABLE_P (cpu_features, BMI2))
return OPTIMIZE (evex);
if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2021-11-02 19:59 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-11-02 19:59 [glibc/maskray/clang] x86-64: Remove Prefer_AVX2_STRCMP Fangrui Song
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).