[clang] dae9139 - [X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow AVX512 kmov intrinsics to be used in constexp (#169895)
via cfe-commits
cfe-commits at lists.llvm.org
Mon Dec 1 09:24:21 PST 2025
Author: Muhammad Abdul
Date: 2025-12-01T17:24:16Z
New Revision: dae9139d8fecf09d975f59b012646bc04f694c35
URL: https://github.com/llvm/llvm-project/commit/dae9139d8fecf09d975f59b012646bc04f694c35
DIFF: https://github.com/llvm/llvm-project/commit/dae9139d8fecf09d975f59b012646bc04f694c35.diff
LOG: [X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow AVX512 kmov intrinsics to be used in constexp (#169895)
Resolves #166975
Added:
Modified:
clang/include/clang/Basic/BuiltinsX86.td
clang/lib/AST/ByteCode/InterpBuiltin.cpp
clang/lib/AST/ExprConstant.cpp
clang/lib/Headers/avx512bwintrin.h
clang/lib/Headers/avx512dqintrin.h
clang/lib/Headers/avx512fintrin.h
clang/test/CodeGen/X86/avx512bw-builtins.c
clang/test/CodeGen/X86/avx512dq-builtins.c
clang/test/CodeGen/X86/avx512f-builtins.c
Removed:
################################################################################
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index 98cea35beb0ea..36670a346a4ad 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -3160,15 +3160,15 @@ let Features = "avx512bw", Attributes = [NoThrow, Const] in {
def kshiftridi : X86Builtin<"unsigned long long int(unsigned long long int, _Constant unsigned int)">;
}
-let Features = "avx512dq", Attributes = [NoThrow, Const] in {
+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {
def kmovb : X86Builtin<"unsigned char(unsigned char)">;
}
-let Features = "avx512f", Attributes = [NoThrow, Const] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr] in {
def kmovw : X86Builtin<"unsigned short(unsigned short)">;
}
-let Features = "avx512bw", Attributes = [NoThrow, Const] in {
+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kmovd : X86Builtin<"unsigned int(unsigned int)">;
def kmovq : X86Builtin<"unsigned long long int(unsigned long long int)">;
}
diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
index 971fce541bb88..61043f29ac2af 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -5003,6 +5003,13 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
S, OpPC, Call,
[](const APSInt &LHS, const APSInt &RHS) { return LHS + RHS; });
+ case X86::BI__builtin_ia32_kmovb:
+ case X86::BI__builtin_ia32_kmovw:
+ case X86::BI__builtin_ia32_kmovd:
+ case X86::BI__builtin_ia32_kmovq:
+ return interp__builtin_elementwise_int_unaryop(
+ S, OpPC, Call, [](const APSInt &Src) { return Src; });
+
case X86::BI__builtin_ia32_kunpckhi:
case X86::BI__builtin_ia32_kunpckdi:
case X86::BI__builtin_ia32_kunpcksi:
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index e5af4cb049ba9..3eedf0d6795d8 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -16897,6 +16897,16 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
[](const APSInt &LHS, const APSInt &RHS) { return LHS + RHS; });
}
+ case X86::BI__builtin_ia32_kmovb:
+ case X86::BI__builtin_ia32_kmovw:
+ case X86::BI__builtin_ia32_kmovd:
+ case X86::BI__builtin_ia32_kmovq: {
+ APSInt Val;
+ if (!EvaluateInteger(E->getArg(0), Val, Info))
+ return false;
+ return Success(Val, E);
+ }
+
case clang::X86::BI__builtin_ia32_vec_ext_v4hi:
case clang::X86::BI__builtin_ia32_vec_ext_v16qi:
case clang::X86::BI__builtin_ia32_vec_ext_v8hi:
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 67e8461560b04..48b7c98df7b68 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -178,22 +178,22 @@ _kadd_mask64(__mmask64 __A, __mmask64 __B) {
#define _kshiftri_mask64(A, I) \
((__mmask64)__builtin_ia32_kshiftridi((__mmask64)(A), (unsigned int)(I)))
-static __inline__ unsigned int __DEFAULT_FN_ATTRS
-_cvtmask32_u32(__mmask32 __A) {
+static __inline__ unsigned int
+ __DEFAULT_FN_ATTRS_CONSTEXPR _cvtmask32_u32(__mmask32 __A) {
return (unsigned int)__builtin_ia32_kmovd((__mmask32)__A);
}
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS_CONSTEXPR
_cvtmask64_u64(__mmask64 __A) {
return (unsigned long long)__builtin_ia32_kmovq((__mmask64)__A);
}
-static __inline__ __mmask32 __DEFAULT_FN_ATTRS
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR
_cvtu32_mask32(unsigned int __A) {
return (__mmask32)__builtin_ia32_kmovd((__mmask32)__A);
}
-static __inline__ __mmask64 __DEFAULT_FN_ATTRS
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS_CONSTEXPR
_cvtu64_mask64(unsigned long long __A) {
return (__mmask64)__builtin_ia32_kmovq((__mmask64)__A);
}
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index f200b22f27ee1..ae02cdd47af2e 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -123,12 +123,12 @@ _kadd_mask16(__mmask16 __A, __mmask16 __B) {
#define _kshiftri_mask8(A, I) \
((__mmask8)__builtin_ia32_kshiftriqi((__mmask8)(A), (unsigned int)(I)))
-static __inline__ unsigned int __DEFAULT_FN_ATTRS
-_cvtmask8_u32(__mmask8 __A) {
+static __inline__ unsigned int
+ __DEFAULT_FN_ATTRS_CONSTEXPR _cvtmask8_u32(__mmask8 __A) {
return (unsigned int)__builtin_ia32_kmovb((__mmask8)__A);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR
_cvtu32_mask8(unsigned int __A) {
return (__mmask8)__builtin_ia32_kmovb((__mmask8)__A);
}
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 806a13c414c10..ca5d7e85061e6 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8069,12 +8069,12 @@ _mm512_kxor(__mmask16 __A, __mmask16 __B) {
#define _kshiftri_mask16(A, I) \
((__mmask16)__builtin_ia32_kshiftrihi((__mmask16)(A), (unsigned int)(I)))
-static __inline__ unsigned int __DEFAULT_FN_ATTRS
-_cvtmask16_u32(__mmask16 __A) {
+static __inline__ unsigned int
+ __DEFAULT_FN_ATTRS_CONSTEXPR _cvtmask16_u32(__mmask16 __A) {
return (unsigned int)__builtin_ia32_kmovw((__mmask16)__A);
}
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
_cvtu32_mask16(unsigned int __A) {
return (__mmask16)__builtin_ia32_kmovw((__mmask16)__A);
}
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
index c9c30dab389db..fd50ab97d2c42 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -561,21 +561,31 @@ unsigned int test_cvtmask32_u32(__m512i A, __m512i B) {
return _cvtmask32_u32(_mm512_cmpneq_epu16_mask(A, B));
}
+TEST_CONSTEXPR(_cvtmask32_u32((__mmask32)0xDEADBEEF) == 0xDEADBEEF);
+
unsigned long long test_cvtmask64_u64(__m512i A, __m512i B) {
// CHECK-LABEL: test_cvtmask64_u64
return _cvtmask64_u64(_mm512_cmpneq_epu8_mask(A, B));
}
+TEST_CONSTEXPR(_cvtmask64_u64((__mmask64)0x123456789ABCDEF0ULL) == 0x123456789ABCDEF0ULL);
+
__mmask32 test_cvtu32_mask32(__m512i A, __m512i B, unsigned int C) {
// CHECK-LABEL: test_cvtu32_mask32
return _mm512_mask_cmpneq_epu16_mask(_cvtu32_mask32(C), A, B);
}
+TEST_CONSTEXPR(_cvtu32_mask32(0x13579BDF) == (__mmask32)0x13579BDF);
+TEST_CONSTEXPR(_cvtu32_mask32(_cvtmask32_u32((__mmask32)0x2468ACE0)) == (__mmask32)0x2468ACE0);
+
__mmask64 test_cvtu64_mask64(__m512i A, __m512i B, unsigned long long C) {
// CHECK-LABEL: test_cvtu64_mask64
return _mm512_mask_cmpneq_epu8_mask(_cvtu64_mask64(C), A, B);
}
+TEST_CONSTEXPR(_cvtu64_mask64(0x0F0F0F0F0F0F0F0FULL) == (__mmask64)0x0F0F0F0F0F0F0F0FULL);
+TEST_CONSTEXPR(_cvtu64_mask64(_cvtmask64_u64((__mmask64)0xF0F0F0F0F0F0F0F0ULL)) == (__mmask64)0xF0F0F0F0F0F0F0F0ULL);
+
__mmask32 test_load_mask32(__mmask32 *A, __m512i B, __m512i C) {
// CHECK-LABEL: test_load_mask32
// CHECK: [[LOAD:%.*]] = load i32, ptr %{{.*}}
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
index 542d4446a3690..b8d9587af0394 100644
--- a/clang/test/CodeGen/X86/avx512dq-builtins.c
+++ b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -378,12 +378,17 @@ unsigned int test_cvtmask8_u32(__m512i A, __m512i B) {
return _cvtmask8_u32(_mm512_cmpneq_epu64_mask(A, B));
}
+TEST_CONSTEXPR(_cvtmask8_u32((__mmask8)0x5A) == 0x5A);
+
__mmask8 test_cvtu32_mask8(__m512i A, __m512i B, unsigned int C) {
// CHECK-LABEL: test_cvtu32_mask8
// CHECK: trunc i32 %{{.*}} to i8
return _mm512_mask_cmpneq_epu64_mask(_cvtu32_mask8(C), A, B);
}
+TEST_CONSTEXPR(_cvtu32_mask8(0xB7) == (__mmask8)0xB7);
+TEST_CONSTEXPR(_cvtu32_mask8(_cvtmask8_u32((__mmask8)0xDE)) == (__mmask8)0xDE);
+
__mmask8 test_load_mask8(__mmask8 *A, __m512i B, __m512i C) {
// CHECK-LABEL: test_load_mask8
// CHECK: [[LOAD:%.*]] = load i8, ptr %{{.*}}
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index 6401a0e55a83b..257559261905e 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -9589,6 +9589,8 @@ unsigned int test_cvtmask16_u32(__m512i A, __m512i B) {
return _cvtmask16_u32(_mm512_cmpneq_epu32_mask(A, B));
}
+TEST_CONSTEXPR(_cvtmask16_u32((__mmask16)0xBEEF) == 0xBEEF);
+
__mmask16 test_cvtu32_mask16(__m512i A, __m512i B, unsigned int C) {
// CHECK-LABEL: test_cvtu32_mask16
// CHECK: trunc i32 %{{.*}} to i16
@@ -9596,6 +9598,9 @@ __mmask16 test_cvtu32_mask16(__m512i A, __m512i B, unsigned int C) {
return _mm512_mask_cmpneq_epu32_mask(_cvtu32_mask16(C), A, B);
}
+TEST_CONSTEXPR(_cvtu32_mask16(0xCAFE) == (__mmask16)0xCAFE);
+TEST_CONSTEXPR(_cvtu32_mask16(_cvtmask16_u32((__mmask16)0x1357)) == (__mmask16)0x1357);
+
__mmask16 test_load_mask16(__mmask16 *A, __m512i B, __m512i C) {
// CHECK-LABEL: test_load_mask16
// CHECK: [[LOAD:%.*]] = load i16, ptr %{{.*}}{{$}}
More information about the cfe-commits
mailing list