forked from rpms/glibc
75 lines
3.1 KiB
Diff
75 lines
3.1 KiB
Diff
|
From 1da50d4bda07f04135dca39f40e79fc9eabed1f8 Mon Sep 17 00:00:00 2001
|
||
|
|
From: "H.J. Lu" <hjl.tools@gmail.com>
|
||
|
|
Date: Fri, 26 Feb 2021 05:36:59 -0800
|
||
|
|
Subject: [PATCH] x86: Set Prefer_No_VZEROUPPER and add Prefer_AVX2_STRCMP
|
||
|
|
Content-type: text/plain; charset=UTF-8
|
||
|
|
|
||
|
|
1. Set Prefer_No_VZEROUPPER if RTM is usable to avoid RTM abort triggered
|
||
|
|
by VZEROUPPER inside a transactionally executing RTM region.
|
||
|
|
2. 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. Add
|
||
|
|
Prefer_AVX2_STRCMP to prefer AVX2 strcmp family functions.
|
||
|
|
---
|
||
|
|
sysdeps/x86/cpu-features.c | 20 +++++++++++++++++--
|
||
|
|
sysdeps/x86/cpu-tunables.c | 2 ++
|
||
|
|
...cpu-features-preferred_feature_index_1.def | 1 +
|
||
|
|
3 files changed, 21 insertions(+), 2 deletions(-)
|
||
|
|
|
||
|
|
diff --git a/sysdeps/x86/cpu-features.c b/sysdeps/x86/cpu-features.c
|
||
|
|
index 91042505..3610ee5c 100644
|
||
|
|
--- a/sysdeps/x86/cpu-features.c
|
||
|
|
+++ b/sysdeps/x86/cpu-features.c
|
||
|
|
@@ -524,8 +524,24 @@ init_cpu_features (struct cpu_features *cpu_features)
|
||
|
|
cpu_features->preferred[index_arch_Prefer_No_VZEROUPPER]
|
||
|
|
|= bit_arch_Prefer_No_VZEROUPPER;
|
||
|
|
else
|
||
|
|
- cpu_features->preferred[index_arch_Prefer_No_AVX512]
|
||
|
|
- |= bit_arch_Prefer_No_AVX512;
|
||
|
|
+ {
|
||
|
|
+ cpu_features->preferred[index_arch_Prefer_No_AVX512]
|
||
|
|
+ |= bit_arch_Prefer_No_AVX512;
|
||
|
|
+
|
||
|
|
+ /* Avoid RTM abort triggered by VZEROUPPER inside a
|
||
|
|
+ transactionally executing RTM region. */
|
||
|
|
+ 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;
|
||
|
|
+ }
|
||
|
|
}
|
||
|
|
/* This spells out "AuthenticAMD". */
|
||
|
|
else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
|
||
|
|
diff --git a/sysdeps/x86/cpu-tunables.c b/sysdeps/x86/cpu-tunables.c
|
||
|
|
index 3173b2b9..73adbaba 100644
|
||
|
|
--- a/sysdeps/x86/cpu-tunables.c
|
||
|
|
+++ b/sysdeps/x86/cpu-tunables.c
|
||
|
|
@@ -239,6 +239,8 @@ 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 17a5cc42..4ca70b40 100644
|
||
|
|
--- a/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
|
||
|
|
+++ b/sysdeps/x86/include/cpu-features-preferred_feature_index_1.def
|
||
|
|
@@ -32,3 +32,4 @@ BIT (Prefer_ERMS)
|
||
|
|
BIT (Prefer_FSRM)
|
||
|
|
BIT (Prefer_No_AVX512)
|
||
|
|
BIT (MathVec_Prefer_No_AVX512)
|
||
|
|
+BIT (Prefer_AVX2_STRCMP)
|
||
|
|
--
|
||
|
|
GitLab
|
||
|
|
|