[clang] [llvm] [X86][AVX10] Allow 64-bit mask register used without EVEX512 (PR #75571)
Phoebe Wang via cfe-commits
cfe-commits at lists.llvm.org
Fri Dec 15 00:27:22 PST 2023
https://github.com/phoebewang updated https://github.com/llvm/llvm-project/pull/75571
>From 2e31cbd1a0b5b9c1689f664c6c261cabbc656f2a Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.wang at intel.com>
Date: Fri, 15 Dec 2023 16:11:02 +0800
Subject: [PATCH 1/2] [X86][AVX10] Allow 64-bit mask register used without
EVEX512
This is to reflect new document change that 64-bit mask is support by
AVX10 256-bit targets.
Latest documents can be found in:
https://cdrdv2.intel.com/v1/dl/getContent/784267
https://cdrdv2.intel.com/v1/dl/getContent/784343
---
clang/include/clang/Basic/BuiltinsX86.def | 30 +++++++--------
clang/lib/Headers/avx512bwintrin.h | 37 +++++++++---------
clang/test/CodeGen/X86/avx512-error.c | 46 +++++++----------------
llvm/lib/Target/X86/X86RegisterInfo.cpp | 6 ---
4 files changed, 47 insertions(+), 72 deletions(-)
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index e4802f8ab1c156..60b752ad48548f 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -979,7 +979,7 @@ TARGET_BUILTIN(__builtin_ia32_scatterpfqps, "vUcV8Oiv*IiIi", "nV:512:", "avx512p
TARGET_BUILTIN(__builtin_ia32_knotqi, "UcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_knothi, "UsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_knotsi, "UiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_knotdi, "UOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_knotdi, "UOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_cmpb128_mask, "UsV16cV16cIiUs", "ncV:128:", "avx512vl,avx512bw")
TARGET_BUILTIN(__builtin_ia32_cmpd128_mask, "UcV4iV4iIiUc", "ncV:128:", "avx512vl")
@@ -1349,7 +1349,7 @@ TARGET_BUILTIN(__builtin_ia32_vpmadd52luq128, "V2OiV2OiV2OiV2Oi", "ncV:128:", "a
TARGET_BUILTIN(__builtin_ia32_vpmadd52luq256, "V4OiV4OiV4OiV4Oi", "ncV:256:", "avx512ifma,avx512vl|avxifma")
TARGET_BUILTIN(__builtin_ia32_vcomisd, "iV2dV2dIiIi", "ncV:128:", "avx512f")
TARGET_BUILTIN(__builtin_ia32_vcomiss, "iV4fV4fIiIi", "ncV:128:", "avx512f")
-TARGET_BUILTIN(__builtin_ia32_kunpckdi, "UOiUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kunpckdi, "UOiUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kunpcksi, "UiUiUi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_loaddquhi512_mask, "V32sV32sC*V32sUi", "nV:512:", "avx512bw,evex512")
TARGET_BUILTIN(__builtin_ia32_loaddquqi512_mask, "V64cV64cC*V64cUOi", "nV:512:", "avx512bw,evex512")
@@ -1665,56 +1665,56 @@ TARGET_BUILTIN(__builtin_ia32_fpclassss_mask, "UcV4fIiUc", "ncV:128:", "avx512dq
TARGET_BUILTIN(__builtin_ia32_kaddqi, "UcUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kaddhi, "UsUsUs", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kaddsi, "UiUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kadddi, "UOiUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kadddi, "UOiUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kandqi, "UcUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kandhi, "UsUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kandsi, "UiUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kanddi, "UOiUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kanddi, "UOiUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kandnqi, "UcUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kandnhi, "UsUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kandnsi, "UiUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kandndi, "UOiUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kandndi, "UOiUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_korqi, "UcUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_korhi, "UsUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_korsi, "UiUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kordi, "UOiUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kordi, "UOiUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kortestcqi, "iUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kortestzqi, "iUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kortestchi, "iUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kortestzhi, "iUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kortestcsi, "iUiUi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kortestzsi, "iUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kortestcdi, "iUOiUOi", "nc", "avx512bw,evex512")
-TARGET_BUILTIN(__builtin_ia32_kortestzdi, "iUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kortestcdi, "iUOiUOi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_kortestzdi, "iUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_ktestcqi, "iUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_ktestzqi, "iUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_ktestchi, "iUsUs", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_ktestzhi, "iUsUs", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_ktestcsi, "iUiUi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_ktestzsi, "iUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_ktestcdi, "iUOiUOi", "nc", "avx512bw,evex512")
-TARGET_BUILTIN(__builtin_ia32_ktestzdi, "iUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_ktestcdi, "iUOiUOi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_ktestzdi, "iUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kunpckhi, "UsUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kxnorqi, "UcUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kxnorhi, "UsUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kxnorsi, "UiUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kxnordi, "UOiUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kxnordi, "UOiUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kxorqi, "UcUcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kxorhi, "UsUsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kxorsi, "UiUiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kxordi, "UOiUOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kxordi, "UOiUOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kshiftliqi, "UcUcIUi", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kshiftlihi, "UsUsIUi", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kshiftlisi, "UiUiIUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kshiftlidi, "UOiUOiIUi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kshiftlidi, "UOiUOiIUi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kshiftriqi, "UcUcIUi", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kshiftrihi, "UsUsIUi", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kshiftrisi, "UiUiIUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kshiftridi, "UOiUOiIUi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kshiftridi, "UOiUOiIUi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_kmovb, "UcUc", "nc", "avx512dq")
TARGET_BUILTIN(__builtin_ia32_kmovw, "UsUs", "nc", "avx512f")
TARGET_BUILTIN(__builtin_ia32_kmovd, "UiUi", "nc", "avx512bw")
-TARGET_BUILTIN(__builtin_ia32_kmovq, "UOiUOi", "nc", "avx512bw,evex512")
+TARGET_BUILTIN(__builtin_ia32_kmovq, "UOiUOi", "nc", "avx512bw")
TARGET_BUILTIN(__builtin_ia32_palignr512, "V64cV64cV64cIi", "ncV:512:", "avx512bw,evex512")
TARGET_BUILTIN(__builtin_ia32_dbpsadbw128, "V8sV16cV16cIi", "ncV:128:", "avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_dbpsadbw256, "V16sV32cV32cIi", "ncV:256:", "avx512bw,avx512vl")
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 51dba5427b0fc0..e90e3ce74477c5 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -19,7 +19,6 @@ typedef unsigned long long __mmask64;
/* Define the default attributes for the functions in this file. */
#define __DEFAULT_FN_ATTRS512 __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,evex512"), __min_vector_width__(512)))
-#define __DEFAULT_FN_ATTRS64 __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,evex512")))
#define __DEFAULT_FN_ATTRS \
__attribute__((__always_inline__, __nodebug__, \
__target__("avx512bw,no-evex512")))
@@ -30,7 +29,7 @@ _knot_mask32(__mmask32 __M)
return __builtin_ia32_knotsi(__M);
}
-static __inline __mmask64 __DEFAULT_FN_ATTRS64
+static __inline __mmask64 __DEFAULT_FN_ATTRS
_knot_mask64(__mmask64 __M)
{
return __builtin_ia32_knotdi(__M);
@@ -42,7 +41,7 @@ _kand_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kandsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_kand_mask64(__mmask64 __A, __mmask64 __B)
{
return (__mmask64)__builtin_ia32_kanddi((__mmask64)__A, (__mmask64)__B);
@@ -54,7 +53,7 @@ _kandn_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kandnsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_kandn_mask64(__mmask64 __A, __mmask64 __B)
{
return (__mmask64)__builtin_ia32_kandndi((__mmask64)__A, (__mmask64)__B);
@@ -66,7 +65,7 @@ _kor_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_korsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_kor_mask64(__mmask64 __A, __mmask64 __B)
{
return (__mmask64)__builtin_ia32_kordi((__mmask64)__A, (__mmask64)__B);
@@ -78,7 +77,7 @@ _kxnor_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kxnorsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_kxnor_mask64(__mmask64 __A, __mmask64 __B)
{
return (__mmask64)__builtin_ia32_kxnordi((__mmask64)__A, (__mmask64)__B);
@@ -90,7 +89,7 @@ _kxor_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kxorsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_kxor_mask64(__mmask64 __A, __mmask64 __B)
{
return (__mmask64)__builtin_ia32_kxordi((__mmask64)__A, (__mmask64)__B);
@@ -114,19 +113,19 @@ _kortest_mask32_u8(__mmask32 __A, __mmask32 __B, unsigned char *__C) {
return (unsigned char)__builtin_ia32_kortestzsi(__A, __B);
}
-static __inline__ unsigned char __DEFAULT_FN_ATTRS64
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
_kortestc_mask64_u8(__mmask64 __A, __mmask64 __B)
{
return (unsigned char)__builtin_ia32_kortestcdi(__A, __B);
}
-static __inline__ unsigned char __DEFAULT_FN_ATTRS64
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
_kortestz_mask64_u8(__mmask64 __A, __mmask64 __B)
{
return (unsigned char)__builtin_ia32_kortestzdi(__A, __B);
}
-static __inline__ unsigned char __DEFAULT_FN_ATTRS64
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
_kortest_mask64_u8(__mmask64 __A, __mmask64 __B, unsigned char *__C) {
*__C = (unsigned char)__builtin_ia32_kortestcdi(__A, __B);
return (unsigned char)__builtin_ia32_kortestzdi(__A, __B);
@@ -150,19 +149,19 @@ _ktest_mask32_u8(__mmask32 __A, __mmask32 __B, unsigned char *__C) {
return (unsigned char)__builtin_ia32_ktestzsi(__A, __B);
}
-static __inline__ unsigned char __DEFAULT_FN_ATTRS64
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
_ktestc_mask64_u8(__mmask64 __A, __mmask64 __B)
{
return (unsigned char)__builtin_ia32_ktestcdi(__A, __B);
}
-static __inline__ unsigned char __DEFAULT_FN_ATTRS64
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
_ktestz_mask64_u8(__mmask64 __A, __mmask64 __B)
{
return (unsigned char)__builtin_ia32_ktestzdi(__A, __B);
}
-static __inline__ unsigned char __DEFAULT_FN_ATTRS64
+static __inline__ unsigned char __DEFAULT_FN_ATTRS
_ktest_mask64_u8(__mmask64 __A, __mmask64 __B, unsigned char *__C) {
*__C = (unsigned char)__builtin_ia32_ktestcdi(__A, __B);
return (unsigned char)__builtin_ia32_ktestzdi(__A, __B);
@@ -174,7 +173,7 @@ _kadd_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kaddsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_kadd_mask64(__mmask64 __A, __mmask64 __B)
{
return (__mmask64)__builtin_ia32_kadddi((__mmask64)__A, (__mmask64)__B);
@@ -197,7 +196,7 @@ _cvtmask32_u32(__mmask32 __A) {
return (unsigned int)__builtin_ia32_kmovd((__mmask32)__A);
}
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS64
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS
_cvtmask64_u64(__mmask64 __A) {
return (unsigned long long)__builtin_ia32_kmovq((__mmask64)__A);
}
@@ -207,7 +206,7 @@ _cvtu32_mask32(unsigned int __A) {
return (__mmask32)__builtin_ia32_kmovd((__mmask32)__A);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_cvtu64_mask64(unsigned long long __A) {
return (__mmask64)__builtin_ia32_kmovq((__mmask64)__A);
}
@@ -217,7 +216,7 @@ _load_mask32(__mmask32 *__A) {
return (__mmask32)__builtin_ia32_kmovd(*(__mmask32 *)__A);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_load_mask64(__mmask64 *__A) {
return (__mmask64)__builtin_ia32_kmovq(*(__mmask64 *)__A);
}
@@ -227,7 +226,7 @@ _store_mask32(__mmask32 *__A, __mmask32 __B) {
*(__mmask32 *)__A = __builtin_ia32_kmovd((__mmask32)__B);
}
-static __inline__ void __DEFAULT_FN_ATTRS64
+static __inline__ void __DEFAULT_FN_ATTRS
_store_mask64(__mmask64 *__A, __mmask64 __B) {
*(__mmask64 *)__A = __builtin_ia32_kmovq((__mmask64)__B);
}
@@ -1717,7 +1716,7 @@ _mm512_maskz_set1_epi8 (__mmask64 __M, char __A)
(__v64qi) _mm512_setzero_si512());
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS64
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
_mm512_kunpackd (__mmask64 __A, __mmask64 __B)
{
return (__mmask64) __builtin_ia32_kunpckdi ((__mmask64) __A,
diff --git a/clang/test/CodeGen/X86/avx512-error.c b/clang/test/CodeGen/X86/avx512-error.c
index 99824a373e8ebf..422cc7a8679dcf 100644
--- a/clang/test/CodeGen/X86/avx512-error.c
+++ b/clang/test/CodeGen/X86/avx512-error.c
@@ -1,21 +1,24 @@
-// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512bw -target-feature -evex512 -emit-llvm -o /dev/null -verify=noevex -DFEATURE_TEST=1
-// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512bw -target-feature -evex512 -emit-llvm -o /dev/null -verify=noevex -DFEATURE_TEST=2
-// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512bw -emit-llvm -o /dev/null -verify -DFEATURE_TEST=3
-// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx10.1-256 -emit-llvm -o /dev/null -verify=noevex -DFEATURE_TEST=1
-// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx10.1-256 -emit-llvm -o /dev/null -verify=noevex -DFEATURE_TEST=2
-// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx10.1-512 -emit-llvm -o /dev/null -verify -DFEATURE_TEST=3
+// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512bw -target-feature -evex512 -emit-llvm -o /dev/null -verify=noevex
+// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512bw -emit-llvm -o /dev/null -verify
+// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx10.1-256 -emit-llvm -o /dev/null -verify=noevex
+// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx10.1-512 -emit-llvm -o /dev/null -verify
#include <immintrin.h>
-#if FEATURE_TEST & 3
-// expected-no-diagnostics
-#endif
+// No error emitted whether we have "evex512" feature or not.
+__attribute__((target("avx512bw,no-evex512")))
+__mmask64 k64_verify_1(__mmask64 a) {
+ return _knot_mask64(a); // expected-no-diagnostics
+}
+
+__mmask64 k64_verify_2(__mmask64 a) {
+ return _knot_mask64(a); // expected-no-diagnostic
+}
-#if FEATURE_TEST & 1
__attribute__((target("avx512bw,evex512")))
__m512d zmm_verify_ok(__m512d a) {
// No error emitted if we have "evex512" feature.
- return __builtin_ia32_sqrtpd512(a, _MM_FROUND_CUR_DIRECTION);
+ return __builtin_ia32_sqrtpd512(a, _MM_FROUND_CUR_DIRECTION); // expected-no-diagnostic
}
__m512d zmm_error(__m512d a) {
@@ -27,24 +30,3 @@ __m512d zmm_error(__m512d a) {
// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
#endif
-#endif
-
-#if FEATURE_TEST & 2
-__attribute__((target("avx512bw,evex512")))
-__mmask64 k64_verify_ok(__mmask64 a) {
- // No error emitted if we have "evex512" feature.
- return _knot_mask64(a);
-}
-#if defined(__AVX10_1__) && !defined(__AVX10_1_512__)
-// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
-// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
-// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
-// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
-// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
-// noevex-warning@*:* {{invalid feature combination: +avx512bw +avx10.1-256; will be promoted to avx10.1-512}}
-#endif
-
-__mmask64 test_knot_mask64(__mmask64 a) {
- return _knot_mask64(a); // noevex-error {{always_inline function '_knot_mask64' requires target feature 'evex512', but would be inlined into function 'test_knot_mask64' that is compiled without support for 'evex512'}}
-}
-#endif
diff --git a/llvm/lib/Target/X86/X86RegisterInfo.cpp b/llvm/lib/Target/X86/X86RegisterInfo.cpp
index 5c32519dab3714..e76d0d7bf50e19 100644
--- a/llvm/lib/Target/X86/X86RegisterInfo.cpp
+++ b/llvm/lib/Target/X86/X86RegisterInfo.cpp
@@ -1077,12 +1077,6 @@ bool X86RegisterInfo::getRegAllocationHints(Register VirtReg,
VirtReg, Order, Hints, MF, VRM, Matrix);
unsigned ID = RC.getID();
- const X86Subtarget &Subtarget = MF.getSubtarget<X86Subtarget>();
- if ((ID == X86::VK64RegClassID || ID == X86::VK64WMRegClassID) &&
- Subtarget.hasAVX512() && !Subtarget.hasEVEX512())
- report_fatal_error(
- "64-bit mask registers are not supported without EVEX512");
-
if (ID != X86::TILERegClassID)
return BaseImplRetVal;
>From 6a266eb3e829d4582a2a5b15b3b14a8e96fd016c Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.wang at intel.com>
Date: Fri, 15 Dec 2023 16:27:08 +0800
Subject: [PATCH 2/2] clang-format
---
clang/lib/Headers/avx512bwintrin.h | 62 ++++++++++++------------------
1 file changed, 25 insertions(+), 37 deletions(-)
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index e90e3ce74477c5..c854720de6a650 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -18,7 +18,9 @@ typedef unsigned int __mmask32;
typedef unsigned long long __mmask64;
/* Define the default attributes for the functions in this file. */
-#define __DEFAULT_FN_ATTRS512 __attribute__((__always_inline__, __nodebug__, __target__("avx512bw,evex512"), __min_vector_width__(512)))
+#define __DEFAULT_FN_ATTRS512 \
+ __attribute__((__always_inline__, __nodebug__, \
+ __target__("avx512bw,evex512"), __min_vector_width__(512)))
#define __DEFAULT_FN_ATTRS \
__attribute__((__always_inline__, __nodebug__, \
__target__("avx512bw,no-evex512")))
@@ -29,9 +31,7 @@ _knot_mask32(__mmask32 __M)
return __builtin_ia32_knotsi(__M);
}
-static __inline __mmask64 __DEFAULT_FN_ATTRS
-_knot_mask64(__mmask64 __M)
-{
+static __inline __mmask64 __DEFAULT_FN_ATTRS _knot_mask64(__mmask64 __M) {
return __builtin_ia32_knotdi(__M);
}
@@ -41,9 +41,8 @@ _kand_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kandsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_kand_mask64(__mmask64 __A, __mmask64 __B)
-{
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _kand_mask64(__mmask64 __A,
+ __mmask64 __B) {
return (__mmask64)__builtin_ia32_kanddi((__mmask64)__A, (__mmask64)__B);
}
@@ -53,9 +52,8 @@ _kandn_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kandnsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_kandn_mask64(__mmask64 __A, __mmask64 __B)
-{
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _kandn_mask64(__mmask64 __A,
+ __mmask64 __B) {
return (__mmask64)__builtin_ia32_kandndi((__mmask64)__A, (__mmask64)__B);
}
@@ -65,9 +63,8 @@ _kor_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_korsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_kor_mask64(__mmask64 __A, __mmask64 __B)
-{
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _kor_mask64(__mmask64 __A,
+ __mmask64 __B) {
return (__mmask64)__builtin_ia32_kordi((__mmask64)__A, (__mmask64)__B);
}
@@ -77,9 +74,8 @@ _kxnor_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kxnorsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_kxnor_mask64(__mmask64 __A, __mmask64 __B)
-{
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _kxnor_mask64(__mmask64 __A,
+ __mmask64 __B) {
return (__mmask64)__builtin_ia32_kxnordi((__mmask64)__A, (__mmask64)__B);
}
@@ -89,9 +85,8 @@ _kxor_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kxorsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_kxor_mask64(__mmask64 __A, __mmask64 __B)
-{
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _kxor_mask64(__mmask64 __A,
+ __mmask64 __B) {
return (__mmask64)__builtin_ia32_kxordi((__mmask64)__A, (__mmask64)__B);
}
@@ -114,14 +109,12 @@ _kortest_mask32_u8(__mmask32 __A, __mmask32 __B, unsigned char *__C) {
}
static __inline__ unsigned char __DEFAULT_FN_ATTRS
-_kortestc_mask64_u8(__mmask64 __A, __mmask64 __B)
-{
+_kortestc_mask64_u8(__mmask64 __A, __mmask64 __B) {
return (unsigned char)__builtin_ia32_kortestcdi(__A, __B);
}
static __inline__ unsigned char __DEFAULT_FN_ATTRS
-_kortestz_mask64_u8(__mmask64 __A, __mmask64 __B)
-{
+_kortestz_mask64_u8(__mmask64 __A, __mmask64 __B) {
return (unsigned char)__builtin_ia32_kortestzdi(__A, __B);
}
@@ -150,14 +143,12 @@ _ktest_mask32_u8(__mmask32 __A, __mmask32 __B, unsigned char *__C) {
}
static __inline__ unsigned char __DEFAULT_FN_ATTRS
-_ktestc_mask64_u8(__mmask64 __A, __mmask64 __B)
-{
+_ktestc_mask64_u8(__mmask64 __A, __mmask64 __B) {
return (unsigned char)__builtin_ia32_ktestcdi(__A, __B);
}
static __inline__ unsigned char __DEFAULT_FN_ATTRS
-_ktestz_mask64_u8(__mmask64 __A, __mmask64 __B)
-{
+_ktestz_mask64_u8(__mmask64 __A, __mmask64 __B) {
return (unsigned char)__builtin_ia32_ktestzdi(__A, __B);
}
@@ -173,9 +164,8 @@ _kadd_mask32(__mmask32 __A, __mmask32 __B)
return (__mmask32)__builtin_ia32_kaddsi((__mmask32)__A, (__mmask32)__B);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_kadd_mask64(__mmask64 __A, __mmask64 __B)
-{
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _kadd_mask64(__mmask64 __A,
+ __mmask64 __B) {
return (__mmask64)__builtin_ia32_kadddi((__mmask64)__A, (__mmask64)__B);
}
@@ -216,8 +206,7 @@ _load_mask32(__mmask32 *__A) {
return (__mmask32)__builtin_ia32_kmovd(*(__mmask32 *)__A);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_load_mask64(__mmask64 *__A) {
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _load_mask64(__mmask64 *__A) {
return (__mmask64)__builtin_ia32_kmovq(*(__mmask64 *)__A);
}
@@ -226,8 +215,8 @@ _store_mask32(__mmask32 *__A, __mmask32 __B) {
*(__mmask32 *)__A = __builtin_ia32_kmovd((__mmask32)__B);
}
-static __inline__ void __DEFAULT_FN_ATTRS
-_store_mask64(__mmask64 *__A, __mmask64 __B) {
+static __inline__ void __DEFAULT_FN_ATTRS _store_mask64(__mmask64 *__A,
+ __mmask64 __B) {
*(__mmask64 *)__A = __builtin_ia32_kmovq((__mmask64)__B);
}
@@ -1716,9 +1705,8 @@ _mm512_maskz_set1_epi8 (__mmask64 __M, char __A)
(__v64qi) _mm512_setzero_si512());
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
-_mm512_kunpackd (__mmask64 __A, __mmask64 __B)
-{
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS _mm512_kunpackd(__mmask64 __A,
+ __mmask64 __B) {
return (__mmask64) __builtin_ia32_kunpckdi ((__mmask64) __A,
(__mmask64) __B);
}
More information about the cfe-commits
mailing list