|
 |
e354a5 |
commit 875a50ff63b2c86af770949d563ee851d08eb46e
|
|
|
e354a5 |
Author: H.J. Lu <hjl.tools@gmail.com>
|
|
|
e354a5 |
Date: Thu Oct 8 08:33:45 2020 -0700
|
|
|
e354a5 |
|
|
|
e354a5 |
<sys/platform/x86.h>: Add AVX-VNNI support
|
|
|
e354a5 |
|
|
|
e354a5 |
Add AVX-VNNI support to <sys/platform/x86.h>.
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/manual/platform.texi b/manual/platform.texi
|
|
|
e354a5 |
index 4f5fdff9d9ef16fd..283f255679643d3e 100644
|
|
|
e354a5 |
--- a/manual/platform.texi
|
|
|
e354a5 |
+++ b/manual/platform.texi
|
|
|
e354a5 |
@@ -198,6 +198,9 @@ The supported processor features are:
|
|
|
e354a5 |
@item
|
|
|
e354a5 |
@code{AVX2} -- The AVX2 instruction extensions.
|
|
|
e354a5 |
|
|
|
e354a5 |
+@item
|
|
|
e354a5 |
+@code{AVX_VNNI} -- The AVX-VNNI instruction extensions.
|
|
|
e354a5 |
+
|
|
|
e354a5 |
@item
|
|
|
e354a5 |
@code{AVX512_4FMAPS} -- The AVX512_4FMAPS instruction extensions.
|
|
|
e354a5 |
|
|
|
e354a5 |
diff --git a/sysdeps/x86/cpu-features.c b/sysdeps/x86/cpu-features.c
|
|
|
e354a5 |
index 67f137259fccf4ad..3e5b9341c9756009 100644
|
|
|
e354a5 |
--- a/sysdeps/x86/cpu-features.c
|
|
|
e354a5 |
+++ b/sysdeps/x86/cpu-features.c
|
|
|
e354a5 |
@@ -119,6 +119,8 @@ update_usable (struct cpu_features *cpu_features)
|
|
|
e354a5 |
cpu_features->preferred[index_arch_AVX_Fast_Unaligned_Load]
|
|
|
e354a5 |
|= bit_arch_AVX_Fast_Unaligned_Load;
|
|
|
e354a5 |
}
|
|
|
e354a5 |
+ /* Determine if AVX-VNNI is usable. */
|
|
|
e354a5 |
+ CPU_FEATURE_SET_USABLE (cpu_features, AVX_VNNI);
|
|
|
e354a5 |
/* Determine if FMA is usable. */
|
|
|
e354a5 |
CPU_FEATURE_SET_USABLE (cpu_features, FMA);
|
|
|
e354a5 |
/* Determine if VAES is usable. */
|
|
|
e354a5 |
diff --git a/sysdeps/x86/sys/platform/x86.h b/sysdeps/x86/sys/platform/x86.h
|
|
|
e354a5 |
index 0b18257e20105ea4..0942ad7a7f7d4ce2 100644
|
|
|
e354a5 |
--- a/sysdeps/x86/sys/platform/x86.h
|
|
|
e354a5 |
+++ b/sysdeps/x86/sys/platform/x86.h
|
|
|
e354a5 |
@@ -311,6 +311,7 @@ extern const struct cpu_features *__x86_get_cpu_features (unsigned int)
|
|
|
e354a5 |
/* COMMON_CPUID_INDEX_7_ECX_1. */
|
|
|
e354a5 |
|
|
|
e354a5 |
/* EAX. */
|
|
|
e354a5 |
+#define bit_cpu_AVX_VNNI (1u << 4)
|
|
|
e354a5 |
#define bit_cpu_AVX512_BF16 (1u << 5)
|
|
|
e354a5 |
|
|
|
e354a5 |
/* COMMON_CPUID_INDEX_19. */
|
|
|
e354a5 |
@@ -530,6 +531,7 @@ extern const struct cpu_features *__x86_get_cpu_features (unsigned int)
|
|
|
e354a5 |
/* COMMON_CPUID_INDEX_7_ECX_1. */
|
|
|
e354a5 |
|
|
|
e354a5 |
/* EAX. */
|
|
|
e354a5 |
+#define index_cpu_AVX_VNNI COMMON_CPUID_INDEX_7_ECX_1
|
|
|
e354a5 |
#define index_cpu_AVX512_BF16 COMMON_CPUID_INDEX_7_ECX_1
|
|
|
e354a5 |
|
|
|
e354a5 |
/* COMMON_CPUID_INDEX_19. */
|
|
|
e354a5 |
@@ -749,6 +751,7 @@ extern const struct cpu_features *__x86_get_cpu_features (unsigned int)
|
|
|
e354a5 |
/* COMMON_CPUID_INDEX_7_ECX_1. */
|
|
|
e354a5 |
|
|
|
e354a5 |
/* EAX. */
|
|
|
e354a5 |
+#define reg_AVX_VNNI eax
|
|
|
e354a5 |
#define reg_AVX512_BF16 eax
|
|
|
e354a5 |
|
|
|
e354a5 |
/* COMMON_CPUID_INDEX_19. */
|
|
|
e354a5 |
diff --git a/sysdeps/x86/tst-get-cpu-features.c b/sysdeps/x86/tst-get-cpu-features.c
|
|
|
e354a5 |
index bcdeb243a82c4adc..8894d9f08ac36633 100644
|
|
|
e354a5 |
--- a/sysdeps/x86/tst-get-cpu-features.c
|
|
|
e354a5 |
+++ b/sysdeps/x86/tst-get-cpu-features.c
|
|
|
e354a5 |
@@ -219,6 +219,7 @@ do_test (void)
|
|
|
e354a5 |
CHECK_CPU_FEATURE (XFD);
|
|
|
e354a5 |
CHECK_CPU_FEATURE (INVARIANT_TSC);
|
|
|
e354a5 |
CHECK_CPU_FEATURE (WBNOINVD);
|
|
|
e354a5 |
+ CHECK_CPU_FEATURE (AVX_VNNI);
|
|
|
e354a5 |
CHECK_CPU_FEATURE (AVX512_BF16);
|
|
|
e354a5 |
CHECK_CPU_FEATURE (AESKLE);
|
|
|
e354a5 |
CHECK_CPU_FEATURE (WIDE_KL);
|
|
|
e354a5 |
@@ -374,6 +375,7 @@ do_test (void)
|
|
|
e354a5 |
CHECK_CPU_FEATURE_USABLE (XFD);
|
|
|
e354a5 |
CHECK_CPU_FEATURE_USABLE (INVARIANT_TSC);
|
|
|
e354a5 |
CHECK_CPU_FEATURE_USABLE (WBNOINVD);
|
|
|
e354a5 |
+ CHECK_CPU_FEATURE_USABLE (AVX_VNNI);
|
|
|
e354a5 |
CHECK_CPU_FEATURE_USABLE (AVX512_BF16);
|
|
|
e354a5 |
CHECK_CPU_FEATURE_USABLE (AESKLE);
|
|
|
e354a5 |
CHECK_CPU_FEATURE_USABLE (WIDE_KL);
|