From: Noah Goldstein <goldstein.w.n@gmail.com>
To: libc-alpha@sourceware.org
Cc: goldstein.w.n@gmail.com, hjl.tools@gmail.com, carlos@systemhalted.org
Subject: [PATCH v3 5/5] x86: Add avx2 optimized functions for the wchar_t strcpy family
Date: Fri, 4 Nov 2022 13:13:14 -0700 [thread overview]
Message-ID: <20221104201314.401813-5-goldstein.w.n@gmail.com> (raw)
In-Reply-To: <20221104201314.401813-1-goldstein.w.n@gmail.com>
Implemented:
wcscat-avx2{+rtm} (+ 744 * 2 bytes
wcscpy-avx2{+rtm} (+ 539 * 2 bytes)
wcpcpy-avx2{+rtm} (+ 577 * 2 bytes)
wcsncpy-avx2{+rtm} (+1108 * 2 bytes)
wcpncpy-avx2{+rtm} (+1214 * 2 bytes)
wcsncat-avx2{+rtm} (+1085 * 2 bytes)
Performance Changes:
Times are from N = 10 runs of the benchmark suite and are reported
as geometric mean of all ratios of New Implementation / Best Old
Implementation. Best Old Implementation was determined with the
highest ISA implementation.
wcscat-avx2 -> 0.975
wcscpy-avx2 -> 0.591
wcpcpy-avx2 -> 0.698
wcsncpy-avx2 -> 0.730
wcpncpy-avx2 -> 0.711
wcsncat-avx2 -> 0.954
Code Size Changes:
This change increase the size of libc.so by ~11kb bytes. For
reference the patch optimizing the normal strcpy family functions
decreases libc.so by ~5.2kb.
Full check passes on x86-64 and build succeeds for all ISA levels w/
and w/o multiarch.
---
sysdeps/x86_64/multiarch/Makefile | 12 ++++
sysdeps/x86_64/multiarch/ifunc-impl-list.c | 66 +++++++++++++++++++++
sysdeps/x86_64/multiarch/ifunc-wcs.h | 11 ++++
sysdeps/x86_64/multiarch/wcpcpy-avx2-rtm.S | 3 +
sysdeps/x86_64/multiarch/wcpcpy-avx2.S | 8 +++
sysdeps/x86_64/multiarch/wcpcpy-generic.c | 2 +-
sysdeps/x86_64/multiarch/wcpncpy-avx2-rtm.S | 3 +
sysdeps/x86_64/multiarch/wcpncpy-avx2.S | 8 +++
sysdeps/x86_64/multiarch/wcpncpy-generic.c | 2 +-
sysdeps/x86_64/multiarch/wcscat-avx2-rtm.S | 3 +
sysdeps/x86_64/multiarch/wcscat-avx2.S | 10 ++++
sysdeps/x86_64/multiarch/wcscat-generic.c | 2 +-
sysdeps/x86_64/multiarch/wcscpy-avx2-rtm.S | 3 +
sysdeps/x86_64/multiarch/wcscpy-avx2.S | 7 +++
sysdeps/x86_64/multiarch/wcscpy-generic.c | 2 +-
sysdeps/x86_64/multiarch/wcscpy.c | 9 +++
sysdeps/x86_64/multiarch/wcsncat-avx2-rtm.S | 3 +
sysdeps/x86_64/multiarch/wcsncat-avx2.S | 9 +++
sysdeps/x86_64/multiarch/wcsncat-generic.c | 2 +-
sysdeps/x86_64/multiarch/wcsncpy-avx2-rtm.S | 3 +
sysdeps/x86_64/multiarch/wcsncpy-avx2.S | 7 +++
sysdeps/x86_64/multiarch/wcsncpy-generic.c | 2 +-
sysdeps/x86_64/wcpcpy-generic.c | 2 +-
sysdeps/x86_64/wcpcpy.S | 3 +-
sysdeps/x86_64/wcpncpy-generic.c | 2 +-
sysdeps/x86_64/wcpncpy.S | 3 +-
sysdeps/x86_64/wcscat-generic.c | 2 +-
sysdeps/x86_64/wcscat.S | 3 +-
sysdeps/x86_64/wcscpy.S | 1 +
sysdeps/x86_64/wcsncat-generic.c | 2 +-
sysdeps/x86_64/wcsncat.S | 3 +-
sysdeps/x86_64/wcsncpy-generic.c | 2 +-
sysdeps/x86_64/wcsncpy.S | 3 +-
33 files changed, 187 insertions(+), 16 deletions(-)
create mode 100644 sysdeps/x86_64/multiarch/wcpcpy-avx2-rtm.S
create mode 100644 sysdeps/x86_64/multiarch/wcpcpy-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcpncpy-avx2-rtm.S
create mode 100644 sysdeps/x86_64/multiarch/wcpncpy-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcscat-avx2-rtm.S
create mode 100644 sysdeps/x86_64/multiarch/wcscat-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcscpy-avx2-rtm.S
create mode 100644 sysdeps/x86_64/multiarch/wcscpy-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcsncat-avx2-rtm.S
create mode 100644 sysdeps/x86_64/multiarch/wcsncat-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcsncpy-avx2-rtm.S
create mode 100644 sysdeps/x86_64/multiarch/wcsncpy-avx2.S
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index d6e01940c3..f848fc0e28 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -131,10 +131,16 @@ endif
ifeq ($(subdir),wcsmbs)
sysdep_routines += \
+ wcpcpy-avx2 \
+ wcpcpy-avx2-rtm \
wcpcpy-evex \
wcpcpy-generic \
+ wcpncpy-avx2 \
+ wcpncpy-avx2-rtm \
wcpncpy-evex \
wcpncpy-generic \
+ wcscat-avx2 \
+ wcscat-avx2-rtm \
wcscat-evex \
wcscat-generic \
wcschr-avx2 \
@@ -146,6 +152,8 @@ sysdep_routines += \
wcscmp-avx2-rtm \
wcscmp-evex \
wcscmp-sse2 \
+ wcscpy-avx2 \
+ wcscpy-avx2-rtm \
wcscpy-evex \
wcscpy-generic \
wcscpy-ssse3 \
@@ -155,11 +163,15 @@ sysdep_routines += \
wcslen-evex512 \
wcslen-sse2 \
wcslen-sse4_1 \
+ wcsncat-avx2 \
+ wcsncat-avx2-rtm \
wcsncat-evex \
wcsncat-generic \
wcsncmp-avx2 \
wcsncmp-avx2-rtm \
wcsncmp-evex \
+ wcsncpy-avx2 \
+ wcsncpy-avx2-rtm \
wcsncpy-evex \
wcsncpy-generic \
wcsnlen-avx2 \
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index 959cb0b420..71e8953e91 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -908,6 +908,17 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (BMI2)),
__wcscpy_evex)
X86_IFUNC_IMPL_ADD_V3 (array, i, wcscpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)),
+ __wcscpy_avx2)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcscpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)
+ && CPU_FEATURE_USABLE (RTM)),
+ __wcscpy_avx2_rtm)
+ X86_IFUNC_IMPL_ADD_V2 (array, i, wcscpy,
CPU_FEATURE_USABLE (SSSE3),
__wcscpy_ssse3)
X86_IFUNC_IMPL_ADD_V1 (array, i, wcscpy,
@@ -922,6 +933,17 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (BMI1)
&& CPU_FEATURE_USABLE (BMI2)),
__wcsncpy_evex)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcsncpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)),
+ __wcsncpy_avx2)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcsncpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)
+ && CPU_FEATURE_USABLE (RTM)),
+ __wcsncpy_avx2_rtm)
X86_IFUNC_IMPL_ADD_V1 (array, i, wcpncpy,
1,
__wcsncpy_generic))
@@ -934,6 +956,17 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (BMI1)
&& CPU_FEATURE_USABLE (BMI2)),
__wcpcpy_evex)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcpcpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)),
+ __wcpcpy_avx2)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcpcpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)
+ && CPU_FEATURE_USABLE (RTM)),
+ __wcpcpy_avx2_rtm)
X86_IFUNC_IMPL_ADD_V1 (array, i, wcpcpy,
1,
__wcpcpy_generic))
@@ -946,6 +979,17 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (BMI1)
&& CPU_FEATURE_USABLE (BMI2)),
__wcpncpy_evex)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcpncpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)),
+ __wcpncpy_avx2)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcpncpy,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)
+ && CPU_FEATURE_USABLE (RTM)),
+ __wcpncpy_avx2_rtm)
X86_IFUNC_IMPL_ADD_V1 (array, i, wcsncpy,
1,
__wcpncpy_generic))
@@ -958,6 +1002,17 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (BMI1)
&& CPU_FEATURE_USABLE (BMI2)),
__wcscat_evex)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcscat,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)),
+ __wcscat_avx2)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcscat,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)
+ && CPU_FEATURE_USABLE (RTM)),
+ __wcscat_avx2_rtm)
X86_IFUNC_IMPL_ADD_V1 (array, i, wcscat,
1,
__wcscat_generic))
@@ -970,6 +1025,17 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
&& CPU_FEATURE_USABLE (BMI1)
&& CPU_FEATURE_USABLE (BMI2)),
__wcsncat_evex)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcsncat,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)),
+ __wcsncat_avx2)
+ X86_IFUNC_IMPL_ADD_V3 (array, i, wcsncat,
+ (CPU_FEATURE_USABLE (AVX2)
+ && CPU_FEATURE_USABLE (BMI1)
+ && CPU_FEATURE_USABLE (BMI2)
+ && CPU_FEATURE_USABLE (RTM)),
+ __wcsncat_avx2_rtm)
X86_IFUNC_IMPL_ADD_V1 (array, i, wcsncat,
1,
__wcsncat_generic))
diff --git a/sysdeps/x86_64/multiarch/ifunc-wcs.h b/sysdeps/x86_64/multiarch/ifunc-wcs.h
index da6e1b03d0..cda633d8fb 100644
--- a/sysdeps/x86_64/multiarch/ifunc-wcs.h
+++ b/sysdeps/x86_64/multiarch/ifunc-wcs.h
@@ -27,6 +27,9 @@
extern __typeof (REDIRECT_NAME) OPTIMIZE (evex) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2_rtm) attribute_hidden;
+
extern __typeof (REDIRECT_NAME) OPTIMIZE (GENERIC) attribute_hidden;
static inline void *
@@ -43,6 +46,14 @@ IFUNC_SELECTOR (void)
if (X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
&& X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX512BW))
return OPTIMIZE (evex);
+
+ if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
+ return OPTIMIZE (avx2_rtm);
+
+ if (X86_ISA_CPU_FEATURES_ARCH_P (cpu_features,
+ Prefer_No_VZEROUPPER, !))
+ return OPTIMIZE (avx2);
+
}
return OPTIMIZE (GENERIC);
diff --git a/sysdeps/x86_64/multiarch/wcpcpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/wcpcpy-avx2-rtm.S
new file mode 100644
index 0000000000..756280a3ab
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcpcpy-avx2-rtm.S
@@ -0,0 +1,3 @@
+#define WCPCPY __wcpcpy_avx2_rtm
+#include "x86-avx2-rtm-vecs.h"
+#include "wcpcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcpcpy-avx2.S b/sysdeps/x86_64/multiarch/wcpcpy-avx2.S
new file mode 100644
index 0000000000..0fffd912d3
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcpcpy-avx2.S
@@ -0,0 +1,8 @@
+#ifndef WCPCPY
+# define WCPCPY __wcpcpy_avx2
+#endif
+
+#define USE_AS_STPCPY
+#define USE_AS_WCSCPY
+#define STRCPY WCPCPY
+#include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcpcpy-generic.c b/sysdeps/x86_64/multiarch/wcpcpy-generic.c
index 6039196a3e..0ba29b081f 100644
--- a/sysdeps/x86_64/multiarch/wcpcpy-generic.c
+++ b/sysdeps/x86_64/multiarch/wcpcpy-generic.c
@@ -19,7 +19,7 @@
/* We always need to build this implementation as strspn-sse4 needs to
be able to fallback to it. */
#include <isa-level.h>
-#if ISA_SHOULD_BUILD (3)
+#if ISA_SHOULD_BUILD (2)
# define WCPCPY __wcpcpy_generic
# include <wcsmbs/wcpcpy.c>
diff --git a/sysdeps/x86_64/multiarch/wcpncpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/wcpncpy-avx2-rtm.S
new file mode 100644
index 0000000000..80600d6b01
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcpncpy-avx2-rtm.S
@@ -0,0 +1,3 @@
+#define WCPNCPY __wcpncpy_avx2_rtm
+#include "x86-avx2-rtm-vecs.h"
+#include "wcpncpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcpncpy-avx2.S b/sysdeps/x86_64/multiarch/wcpncpy-avx2.S
new file mode 100644
index 0000000000..b7e594f7b7
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcpncpy-avx2.S
@@ -0,0 +1,8 @@
+#ifndef WCPNCPY
+# define WCPNCPY __wcpncpy_avx2
+#endif
+
+#define USE_AS_WCSCPY
+#define USE_AS_STPCPY
+#define STRNCPY WCPNCPY
+#include "strncpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcpncpy-generic.c b/sysdeps/x86_64/multiarch/wcpncpy-generic.c
index de8d34320e..4aab4ecdd2 100644
--- a/sysdeps/x86_64/multiarch/wcpncpy-generic.c
+++ b/sysdeps/x86_64/multiarch/wcpncpy-generic.c
@@ -19,7 +19,7 @@
/* We always need to build this implementation as strspn-sse4 needs to
be able to fallback to it. */
#include <isa-level.h>
-#if ISA_SHOULD_BUILD (3)
+#if ISA_SHOULD_BUILD (2)
# define WCPNCPY __wcpncpy_generic
# include <wcsmbs/wcpncpy.c>
diff --git a/sysdeps/x86_64/multiarch/wcscat-avx2-rtm.S b/sysdeps/x86_64/multiarch/wcscat-avx2-rtm.S
new file mode 100644
index 0000000000..e99449a2dc
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcscat-avx2-rtm.S
@@ -0,0 +1,3 @@
+#define WCSCAT __wcscat_avx2_rtm
+#include "x86-avx2-rtm-vecs.h"
+#include "wcscat-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcscat-avx2.S b/sysdeps/x86_64/multiarch/wcscat-avx2.S
new file mode 100644
index 0000000000..a20f23c09d
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcscat-avx2.S
@@ -0,0 +1,10 @@
+#ifndef WCSCAT
+# define WCSCAT __wcscat_avx2
+#endif
+
+#define USE_AS_WCSCPY
+#define USE_AS_STRCAT
+
+#define STRCPY WCSCAT
+
+#include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcscat-generic.c b/sysdeps/x86_64/multiarch/wcscat-generic.c
index d86b4d5c00..6476f85bbb 100644
--- a/sysdeps/x86_64/multiarch/wcscat-generic.c
+++ b/sysdeps/x86_64/multiarch/wcscat-generic.c
@@ -19,7 +19,7 @@
/* We always need to build this implementation as strspn-sse4 needs to
be able to fallback to it. */
#include <isa-level.h>
-#if ISA_SHOULD_BUILD (3)
+#if ISA_SHOULD_BUILD (2)
# define WCSCAT __wcscat_generic
# include <wcsmbs/wcscat.c>
diff --git a/sysdeps/x86_64/multiarch/wcscpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/wcscpy-avx2-rtm.S
new file mode 100644
index 0000000000..2f800c8d3e
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcscpy-avx2-rtm.S
@@ -0,0 +1,3 @@
+#define WCSCPY __wcscpy_avx2_rtm
+#include "x86-avx2-rtm-vecs.h"
+#include "wcscpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcscpy-avx2.S b/sysdeps/x86_64/multiarch/wcscpy-avx2.S
new file mode 100644
index 0000000000..6bc509da07
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcscpy-avx2.S
@@ -0,0 +1,7 @@
+#ifndef WCSCPY
+# define WCSCPY __wcscpy_avx2
+#endif
+
+#define USE_AS_WCSCPY
+#define STRCPY WCSCPY
+#include "strcpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcscpy-generic.c b/sysdeps/x86_64/multiarch/wcscpy-generic.c
index 4a1fffae4b..600d606c45 100644
--- a/sysdeps/x86_64/multiarch/wcscpy-generic.c
+++ b/sysdeps/x86_64/multiarch/wcscpy-generic.c
@@ -18,7 +18,7 @@
#include <isa-level.h>
-#if ISA_SHOULD_BUILD (3)
+#if ISA_SHOULD_BUILD (2)
# define WCSCPY __wcscpy_generic
# include <wcsmbs/wcscpy.c>
diff --git a/sysdeps/x86_64/multiarch/wcscpy.c b/sysdeps/x86_64/multiarch/wcscpy.c
index efe32e505f..7f6387817b 100644
--- a/sysdeps/x86_64/multiarch/wcscpy.c
+++ b/sysdeps/x86_64/multiarch/wcscpy.c
@@ -28,6 +28,9 @@
extern __typeof (REDIRECT_NAME) OPTIMIZE (evex) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2) attribute_hidden;
+extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2_rtm) attribute_hidden;
+
extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden;
extern __typeof (REDIRECT_NAME) OPTIMIZE (generic) attribute_hidden;
@@ -45,6 +48,12 @@ IFUNC_SELECTOR (void)
if (X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
&& X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX512BW))
return OPTIMIZE (evex);
+
+ if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
+ return OPTIMIZE (avx2_rtm);
+
+ if (X86_ISA_CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER, !))
+ return OPTIMIZE (avx2);
}
if (X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, SSSE3))
diff --git a/sysdeps/x86_64/multiarch/wcsncat-avx2-rtm.S b/sysdeps/x86_64/multiarch/wcsncat-avx2-rtm.S
new file mode 100644
index 0000000000..609d6e69c0
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsncat-avx2-rtm.S
@@ -0,0 +1,3 @@
+#define WCSNCAT __wcsncat_avx2_rtm
+#include "x86-avx2-rtm-vecs.h"
+#include "wcsncat-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcsncat-avx2.S b/sysdeps/x86_64/multiarch/wcsncat-avx2.S
new file mode 100644
index 0000000000..a72105b7e9
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsncat-avx2.S
@@ -0,0 +1,9 @@
+#ifndef WCSNCAT
+# define WCSNCAT __wcsncat_avx2
+#endif
+
+#define USE_AS_WCSCPY
+#define USE_AS_STRCAT
+
+#define STRNCAT WCSNCAT
+#include "strncat-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcsncat-generic.c b/sysdeps/x86_64/multiarch/wcsncat-generic.c
index 4b55cb40bc..9ced02b35e 100644
--- a/sysdeps/x86_64/multiarch/wcsncat-generic.c
+++ b/sysdeps/x86_64/multiarch/wcsncat-generic.c
@@ -19,7 +19,7 @@
/* We always need to build this implementation as strspn-sse4 needs to
be able to fallback to it. */
#include <isa-level.h>
-#if ISA_SHOULD_BUILD (3)
+#if ISA_SHOULD_BUILD (2)
# define WCSNCAT __wcsncat_generic
# include <wcsmbs/wcsncat.c>
diff --git a/sysdeps/x86_64/multiarch/wcsncpy-avx2-rtm.S b/sysdeps/x86_64/multiarch/wcsncpy-avx2-rtm.S
new file mode 100644
index 0000000000..cab5a6b820
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsncpy-avx2-rtm.S
@@ -0,0 +1,3 @@
+#define WCSNCPY __wcsncpy_avx2_rtm
+#include "x86-avx2-rtm-vecs.h"
+#include "wcsncpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcsncpy-avx2.S b/sysdeps/x86_64/multiarch/wcsncpy-avx2.S
new file mode 100644
index 0000000000..3a1a8a372c
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsncpy-avx2.S
@@ -0,0 +1,7 @@
+#ifndef WCSNCPY
+# define WCSNCPY __wcsncpy_avx2
+#endif
+
+#define USE_AS_WCSCPY
+#define STRNCPY WCSNCPY
+#include "strncpy-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcsncpy-generic.c b/sysdeps/x86_64/multiarch/wcsncpy-generic.c
index d0e8a86605..693521713b 100644
--- a/sysdeps/x86_64/multiarch/wcsncpy-generic.c
+++ b/sysdeps/x86_64/multiarch/wcsncpy-generic.c
@@ -19,7 +19,7 @@
/* We always need to build this implementation as strspn-sse4 needs to
be able to fallback to it. */
#include <isa-level.h>
-#if ISA_SHOULD_BUILD (3)
+#if ISA_SHOULD_BUILD (2)
# define WCSNCPY __wcsncpy_generic
# include <wcsmbs/wcsncpy.c>
diff --git a/sysdeps/x86_64/wcpcpy-generic.c b/sysdeps/x86_64/wcpcpy-generic.c
index d52525f288..2e4d69a500 100644
--- a/sysdeps/x86_64/wcpcpy-generic.c
+++ b/sysdeps/x86_64/wcpcpy-generic.c
@@ -24,7 +24,7 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL <= 3
+#if MINIMUM_X86_ISA_LEVEL <= 2
# include <wcsmbs/wcpcpy.c>
diff --git a/sysdeps/x86_64/wcpcpy.S b/sysdeps/x86_64/wcpcpy.S
index 97e9207c16..cfde4309fe 100644
--- a/sysdeps/x86_64/wcpcpy.S
+++ b/sysdeps/x86_64/wcpcpy.S
@@ -24,11 +24,12 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL >= 4
+#if MINIMUM_X86_ISA_LEVEL >= 3
# define WCPCPY __wcpcpy
# define DEFAULT_IMPL_V4 "multiarch/wcpcpy-evex.S"
+# define DEFAULT_IMPL_V3 "multiarch/wcpcpy-avx2.S"
/* isa-default-impl.h expects DEFAULT_IMPL_V1 to be defined but it
should never be used from here. */
# define DEFAULT_IMPL_V1 "ERROR -- Invalid ISA IMPL"
diff --git a/sysdeps/x86_64/wcpncpy-generic.c b/sysdeps/x86_64/wcpncpy-generic.c
index 871219a445..1f12a0e4c6 100644
--- a/sysdeps/x86_64/wcpncpy-generic.c
+++ b/sysdeps/x86_64/wcpncpy-generic.c
@@ -24,7 +24,7 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL <= 3
+#if MINIMUM_X86_ISA_LEVEL <= 2
# include <wcsmbs/wcpncpy.c>
diff --git a/sysdeps/x86_64/wcpncpy.S b/sysdeps/x86_64/wcpncpy.S
index 2169ed5545..2f89482d30 100644
--- a/sysdeps/x86_64/wcpncpy.S
+++ b/sysdeps/x86_64/wcpncpy.S
@@ -24,11 +24,12 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL >= 4
+#if MINIMUM_X86_ISA_LEVEL >= 3
# define WCPNCPY __wcpncpy
# define DEFAULT_IMPL_V4 "multiarch/wcpncpy-evex.S"
+# define DEFAULT_IMPL_V3 "multiarch/wcpncpy-avx2.S"
/* isa-default-impl.h expects DEFAULT_IMPL_V1 to be defined but it
should never be used from here. */
# define DEFAULT_IMPL_V1 "ERROR -- Invalid ISA IMPL"
diff --git a/sysdeps/x86_64/wcscat-generic.c b/sysdeps/x86_64/wcscat-generic.c
index 85f981a81f..3552167ebe 100644
--- a/sysdeps/x86_64/wcscat-generic.c
+++ b/sysdeps/x86_64/wcscat-generic.c
@@ -24,7 +24,7 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL <= 3
+#if MINIMUM_X86_ISA_LEVEL <= 2
# include <wcsmbs/wcscat.c>
diff --git a/sysdeps/x86_64/wcscat.S b/sysdeps/x86_64/wcscat.S
index 8432087c7c..2e59987e76 100644
--- a/sysdeps/x86_64/wcscat.S
+++ b/sysdeps/x86_64/wcscat.S
@@ -24,11 +24,12 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL >= 4
+#if MINIMUM_X86_ISA_LEVEL >= 3
# define WCSCAT __wcscat
# define DEFAULT_IMPL_V4 "multiarch/wcscat-evex.S"
+# define DEFAULT_IMPL_V3 "multiarch/wcscat-avx2.S"
/* isa-default-impl.h expects DEFAULT_IMPL_V1 to be defined but it
should never be used from here. */
# define DEFAULT_IMPL_V1 "ERROR -- Invalid ISA IMPL"
diff --git a/sysdeps/x86_64/wcscpy.S b/sysdeps/x86_64/wcscpy.S
index ff8bdd3aea..ab9288ed74 100644
--- a/sysdeps/x86_64/wcscpy.S
+++ b/sysdeps/x86_64/wcscpy.S
@@ -29,6 +29,7 @@
# define WCSCPY __wcscpy
# define DEFAULT_IMPL_V4 "multiarch/wcscpy-evex.S"
+# define DEFAULT_IMPL_V3 "multiarch/wcscpy-avx2.S"
# define DEFAULT_IMPL_V2 "multiarch/wcscpy-ssse3.S"
/* isa-default-impl.h expects DEFAULT_IMPL_V1 to be defined but it
should never be used from here. */
diff --git a/sysdeps/x86_64/wcsncat-generic.c b/sysdeps/x86_64/wcsncat-generic.c
index 2cc0f7b11a..47f6a8ad56 100644
--- a/sysdeps/x86_64/wcsncat-generic.c
+++ b/sysdeps/x86_64/wcsncat-generic.c
@@ -24,7 +24,7 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL <= 3
+#if MINIMUM_X86_ISA_LEVEL <= 2
# include <wcsmbs/wcsncat.c>
diff --git a/sysdeps/x86_64/wcsncat.S b/sysdeps/x86_64/wcsncat.S
index 64e144a9c7..9a55499131 100644
--- a/sysdeps/x86_64/wcsncat.S
+++ b/sysdeps/x86_64/wcsncat.S
@@ -24,11 +24,12 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL >= 4
+#if MINIMUM_X86_ISA_LEVEL >= 3
# define WCSNCAT wcsncat
# define DEFAULT_IMPL_V4 "multiarch/wcsncat-evex.S"
+# define DEFAULT_IMPL_V3 "multiarch/wcsncat-avx2.S"
/* isa-default-impl.h expects DEFAULT_IMPL_V1 to be defined but it
should never be used from here. */
# define DEFAULT_IMPL_V1 "ERROR -- Invalid ISA IMPL"
diff --git a/sysdeps/x86_64/wcsncpy-generic.c b/sysdeps/x86_64/wcsncpy-generic.c
index 49d06b8ae8..7f19fcaddc 100644
--- a/sysdeps/x86_64/wcsncpy-generic.c
+++ b/sysdeps/x86_64/wcsncpy-generic.c
@@ -24,7 +24,7 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL <= 3
+#if MINIMUM_X86_ISA_LEVEL <= 2
# include <wcsmbs/wcsncpy.c>
diff --git a/sysdeps/x86_64/wcsncpy.S b/sysdeps/x86_64/wcsncpy.S
index 1450c1aa28..dc44b32395 100644
--- a/sysdeps/x86_64/wcsncpy.S
+++ b/sysdeps/x86_64/wcsncpy.S
@@ -24,11 +24,12 @@
#include <isa-level.h>
-#if MINIMUM_X86_ISA_LEVEL >= 4
+#if MINIMUM_X86_ISA_LEVEL >= 3
# define WCSNCPY __wcsncpy
# define DEFAULT_IMPL_V4 "multiarch/wcsncpy-evex.S"
+# define DEFAULT_IMPL_V3 "multiarch/wcsncpy-avx2.S"
/* isa-default-impl.h expects DEFAULT_IMPL_V1 to be defined but it
should never be used from here. */
# define DEFAULT_IMPL_V1 "ERROR -- Invalid ISA IMPL"
--
2.34.1
next prev parent reply other threads:[~2022-11-04 20:13 UTC|newest]
Thread overview: 42+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-11-03 8:53 [PATCH v1 1/4] benchtests: Make str{n}{cat|cpy} benchmarks output json Noah Goldstein
2022-11-03 8:53 ` [PATCH v1 2/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-evex functions Noah Goldstein
2022-11-03 8:55 ` Noah Goldstein
2022-11-04 23:04 ` [PATCH v4 1/4] " Noah Goldstein
2022-11-04 23:04 ` [PATCH v4 2/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-avx2 functions Noah Goldstein
2022-11-04 23:04 ` [PATCH v4 3/4] x86: Add evex optimized functions for the wchar_t strcpy family Noah Goldstein
2022-11-04 23:04 ` [PATCH v4 4/4] x86: Add avx2 " Noah Goldstein
2022-11-04 23:34 ` [PATCH v4 1/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-evex functions H.J. Lu
2022-11-09 1:38 ` [PATCH v5 " Noah Goldstein
2022-11-09 1:38 ` [PATCH v5 2/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-avx2 functions Noah Goldstein
2022-11-09 3:00 ` H.J. Lu
2022-11-09 1:38 ` [PATCH v5 3/4] x86: Add evex optimized functions for the wchar_t strcpy family Noah Goldstein
2022-11-09 3:01 ` H.J. Lu
2022-11-09 1:38 ` [PATCH v5 4/4] x86: Add avx2 " Noah Goldstein
2022-11-09 3:01 ` H.J. Lu
2022-11-09 3:00 ` [PATCH v5 1/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-evex functions H.J. Lu
2022-11-03 8:53 ` [PATCH v1 3/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-avx2 functions Noah Goldstein
2022-11-03 8:55 ` Noah Goldstein
2022-11-03 8:53 ` [PATCH v1 4/4] x86: Add optimized functions for the wide-character strcpy family Noah Goldstein
2022-11-03 9:06 ` [PATCH v1 1/4] benchtests: Make str{n}{cat|cpy} benchmarks output json Noah Goldstein
2022-11-04 8:20 ` [PATCH v2 " Noah Goldstein
2022-11-04 8:20 ` [PATCH v2 2/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-evex functions Noah Goldstein
2022-11-04 16:33 ` H.J. Lu
2022-11-04 20:20 ` Noah Goldstein
2022-11-04 8:20 ` [PATCH v2 3/4] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-avx2 functions Noah Goldstein
2022-11-04 16:45 ` H.J. Lu
2022-11-04 20:21 ` Noah Goldstein
2022-11-04 8:20 ` [PATCH v2 4/4] x86: Add optimized functions for the wide-character strcpy family Noah Goldstein
2022-11-04 16:47 ` H.J. Lu
2022-11-04 20:22 ` Noah Goldstein
2022-11-04 16:26 ` [PATCH v2 1/4] benchtests: Make str{n}{cat|cpy} benchmarks output json H.J. Lu
2022-11-04 20:13 ` [PATCH v3 1/5] " Noah Goldstein
2022-11-04 20:13 ` [PATCH v3 2/5] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-evex functions Noah Goldstein
2022-11-04 21:46 ` H.J. Lu
2022-11-04 22:27 ` Noah Goldstein
2022-11-04 22:47 ` H.J. Lu
2022-11-04 23:06 ` Noah Goldstein
2022-11-04 20:13 ` [PATCH v3 3/5] x86: Optimize and shrink st{r|p}{n}{cat|cpy}-avx2 functions Noah Goldstein
2022-11-04 20:13 ` [PATCH v3 4/5] x86: Add evex optimized functions for the wchar_t strcpy family Noah Goldstein
2022-11-04 20:13 ` Noah Goldstein [this message]
2022-11-04 21:01 ` [PATCH v3 1/5] benchtests: Make str{n}{cat|cpy} benchmarks output json H.J. Lu
2022-11-04 21:24 ` Noah Goldstein
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=20221104201314.401813-5-goldstein.w.n@gmail.com \
--to=goldstein.w.n@gmail.com \
--cc=carlos@systemhalted.org \
--cc=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).