[clang] [Headers][X86] Allow basic AVX512 predicate ops to be used in constexpr (PR #159998)
via cfe-commits
cfe-commits at lists.llvm.org
Mon Sep 22 11:08:08 PDT 2025
https://github.com/fennecJ updated https://github.com/llvm/llvm-project/pull/159998
>From 7b91ea75b304a0b0fb3f14945b5c2f21f035bffe Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Mon, 22 Sep 2025 02:40:44 +0800
Subject: [PATCH 1/8] Allow kand* to be used in constexpr
---
clang/include/clang/Basic/BuiltinsX86.td | 6 +++---
clang/lib/AST/ByteCode/InterpBuiltin.cpp | 8 ++++++++
clang/lib/AST/ExprConstant.cpp | 22 ++++++++++++++++++++++
clang/lib/Headers/avx512bwintrin.h | 9 ++++-----
clang/lib/Headers/avx512dqintrin.h | 5 ++---
clang/lib/Headers/avx512fintrin.h | 5 ++---
clang/test/CodeGen/X86/avx512bw-builtins.c | 12 ++++++++++++
clang/test/CodeGen/X86/avx512dq-builtins.c | 6 ++++++
clang/test/CodeGen/X86/avx512f-builtins.c | 6 ++++++
9 files changed, 65 insertions(+), 14 deletions(-)
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index aac502091b57e..e5555236c7666 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -3083,15 +3083,15 @@ let Features = "avx512bw", Attributes = [NoThrow, Const] in {
def kadddi : X86Builtin<"unsigned long long int(unsigned long long int, unsigned long long int)">;
}
-let Features = "avx512dq", Attributes = [NoThrow, Const] in {
+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {
def kandqi : X86Builtin<"unsigned char(unsigned char, unsigned char)">;
}
-let Features = "avx512f", Attributes = [NoThrow, Const] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr] in {
def kandhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;
}
-let Features = "avx512bw", Attributes = [NoThrow, Const] in {
+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kandsi : X86Builtin<"unsigned int(unsigned int, unsigned int)">;
def kanddi : X86Builtin<"unsigned long long int(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 40b9e04aa335c..41d5d76c8c659 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -3588,6 +3588,14 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
case X86::BI__builtin_ia32_selectpd_512:
return interp__builtin_select(S, OpPC, Call);
+ case X86::BI__builtin_ia32_kandqi:
+ case X86::BI__builtin_ia32_kandhi:
+ case X86::BI__builtin_ia32_kandsi:
+ case X86::BI__builtin_ia32_kanddi:
+ return interp__builtin_elementwise_int_binop(
+ S, OpPC, Call,
+ [](const APSInt &LHS, const APSInt &RHS) { return LHS & RHS; });
+
case Builtin::BI__builtin_elementwise_fshl:
return interp__builtin_elementwise_triop(S, OpPC, Call,
llvm::APIntOps::fshl);
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 12e4e25bd29c4..cd41c54087d08 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -13542,6 +13542,20 @@ static bool getBuiltinAlignArguments(const CallExpr *E, EvalInfo &Info,
bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
unsigned BuiltinOp) {
+
+ auto HandleMaskBinOp =
+ [&](llvm::function_ref<APSInt(const APSInt &, const APSInt &)> Fn)
+ -> bool {
+ APValue LHS, RHS;
+ if (!Evaluate(LHS, Info, E->getArg(0)) ||
+ !Evaluate(RHS, Info, E->getArg(1)))
+ return false;
+
+ APSInt ResultInt = Fn(LHS.getInt(), RHS.getInt());
+
+ return Success(APValue(ResultInt), E);
+ };
+
switch (BuiltinOp) {
default:
return false;
@@ -14640,6 +14654,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
Result.setBitVal(P++, Val[I]);
return Success(Result, E);
}
+
+ case X86::BI__builtin_ia32_kandqi:
+ case X86::BI__builtin_ia32_kandhi:
+ case X86::BI__builtin_ia32_kandsi:
+ case X86::BI__builtin_ia32_kanddi: {
+ return HandleMaskBinOp(
+ [](const APSInt &LHS, const APSInt &RHS) { return LHS & RHS; });
+ }
}
}
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 599cfbe479676..cfe16062d8f65 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -42,14 +42,13 @@ static __inline __mmask64 __DEFAULT_FN_ATTRS _knot_mask64(__mmask64 __M) {
return __builtin_ia32_knotdi(__M);
}
-static __inline__ __mmask32 __DEFAULT_FN_ATTRS
-_kand_mask32(__mmask32 __A, __mmask32 __B)
-{
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR
+_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_CONSTEXPR
+_kand_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kanddi((__mmask64)__A, (__mmask64)__B);
}
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index 95fdc2851cb9b..52b3e9d148d6e 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -35,9 +35,8 @@ _knot_mask8(__mmask8 __M)
return __builtin_ia32_knotqi(__M);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS
-_kand_mask8(__mmask8 __A, __mmask8 __B)
-{
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR
+_kand_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_kandqi((__mmask8)__A, (__mmask8)__B);
}
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 8ebfb75170e17..9ffde30cba915 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8116,9 +8116,8 @@ _mm512_mask_permutexvar_epi32 (__m512i __W, __mmask16 __M, __m512i __X,
#define _mm512_mask_permutevar_epi32 _mm512_mask_permutexvar_epi32
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
-_mm512_kand (__mmask16 __A, __mmask16 __B)
-{
+static __inline__ __mmask16
+ __DEFAULT_FN_ATTRS_CONSTEXPR _mm512_kand(__mmask16 __A, __mmask16 __B) {
return (__mmask16) __builtin_ia32_kandhi ((__mmask16) __A, (__mmask16) __B);
}
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
index 3be708aea8a4d..28e9ff611472a 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -36,6 +36,12 @@ __mmask32 test_kand_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_kand_mask32(0xCCCCCCCC, 0xAAAAAAAA) == 0x88888888); // data correctness
+TEST_CONSTEXPR(_kand_mask32(0x123456789, 0xFFFFFFFF) == 0x23456789); // should be truncated
+TEST_CONSTEXPR(_kand_mask32(0xABCDEF01, 0x00000000) == 0x00000000); // all-zero
+TEST_CONSTEXPR(_kand_mask32(0x56789ABC, 0xFFFFFFFF) == 0x56789ABC); // all-one
+TEST_CONSTEXPR(_kand_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000); // disjoint
+
__mmask64 test_kand_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kand_mask64
// CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
@@ -46,6 +52,12 @@ __mmask64 test_kand_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_kand_mask64(0xCCCCCCCCCCCCCCCCull, 0xAAAAAAAAAAAAAAAAull) == 0x8888888888888888ull); // data correctness
+TEST_CONSTEXPR(_kand_mask64(0x123456789ABCDEF0ull, 0xFFFFFFFFFFFFFFFFull) == 0x123456789ABCDEF0ull); // full 64-bit value
+TEST_CONSTEXPR(_kand_mask64(0xABCDEF0123456789ull, 0x0000000000000000ull) == 0x0000000000000000ull); // all-zero
+TEST_CONSTEXPR(_kand_mask64(0x56789ABCDEF01234ull, 0xFFFFFFFFFFFFFFFFull) == 0x56789ABCDEF01234ull); // all-one
+TEST_CONSTEXPR(_kand_mask64(0xAAAAAAAAAAAAAAAAull, 0x5555555555555555ull) == 0x0000000000000000ull); // disjoint
+
__mmask32 test_kandn_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask32
// CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
index df096e3607f30..cc15517fe5bb5 100644
--- a/clang/test/CodeGen/X86/avx512dq-builtins.c
+++ b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -29,6 +29,12 @@ __mmask8 test_kand_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m
__E, __F);
}
+TEST_CONSTEXPR(_kand_mask8(0x0C, 0x0A) == 0x08); // data correctness
+TEST_CONSTEXPR(_kand_mask8(0x123, 0xFF) == 0x23); // should be truncated
+TEST_CONSTEXPR(_kand_mask8(0xAB, 0x00) == 0x00); // all-zero
+TEST_CONSTEXPR(_kand_mask8(0x56, 0xFF) == 0x56); // all-one
+TEST_CONSTEXPR(_kand_mask8(0xAA, 0x55) == 0x00); // disjoint
+
__mmask8 test_kandn_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask8
// CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index f93216e546a63..70a7ff80f9026 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -8502,6 +8502,12 @@ __mmask16 test_mm512_kand(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
+TEST_CONSTEXPR(_mm512_kand(0xCC, 0xAA) == 0x88); // data correctness
+TEST_CONSTEXPR(_mm512_kand(0x12345, 0xFFFF) == 0x2345); // should be truncated
+TEST_CONSTEXPR(_mm512_kand(0xABCD, 0x0000) == 0x0000); // all-zero
+TEST_CONSTEXPR(_mm512_kand(0x5678, 0xFFFF) == 0x5678); // all-one
+TEST_CONSTEXPR(_mm512_kand(0xAAAA, 0x5555) == 0x0000); // disjoint
+
__mmask16 test_mm512_kandn(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_mm512_kandn
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
>From a7f6dad463a3b1386412736a31f6632f2b92876a Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Mon, 22 Sep 2025 03:13:30 +0800
Subject: [PATCH 2/8] Allow int2mask, mask2int to be used in constexpr
---
clang/lib/Headers/avx512fintrin.h | 10 ++++------
clang/test/CodeGen/X86/avx512f-builtins.c | 4 ++++
2 files changed, 8 insertions(+), 6 deletions(-)
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 9ffde30cba915..5a45082f76e81 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -520,15 +520,13 @@ _mm512_castsi512_si256(__m512i __A) {
return (__m256i)__builtin_shufflevector(__A, __A , 0, 1, 2, 3);
}
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
-_mm512_int2mask(int __a)
-{
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_int2mask(int __a) {
return (__mmask16)__a;
}
-static __inline__ int __DEFAULT_FN_ATTRS
-_mm512_mask2int(__mmask16 __a)
-{
+static __inline__ int __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_mask2int(__mmask16 __a) {
return (int)__a;
}
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index 70a7ff80f9026..e7f6b5aac8279 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -11037,6 +11037,8 @@ __mmask16 test_mm512_int2mask(int __a)
return _mm512_int2mask(__a);
}
+TEST_CONSTEXPR(_mm512_int2mask((int)0xDEADBEEF) == 0xBEEF);
+
int test_mm512_mask2int(__mmask16 __a)
{
// CHECK-LABEL: test_mm512_mask2int
@@ -11044,6 +11046,8 @@ int test_mm512_mask2int(__mmask16 __a)
return _mm512_mask2int(__a);
}
+TEST_CONSTEXPR(_mm512_mask2int(0x8000) == 0x00008000); // Zero-extended
+
__m128 test_mm_mask_move_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
{
// CHECK-LABEL: test_mm_mask_move_ss
>From 666a158e34aa508d59c548b4ac1e8a763230eff1 Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Mon, 22 Sep 2025 20:58:35 +0800
Subject: [PATCH 3/8] Add constexpr testcase for _kand_mask16
_kand_mask16 is an alias of _mm512_kand.
Although they are semantically identical, the existing test suite
contains a dedicated set of tests for _kand_mask16.
This commit follows that convention by adding the same constexpr
test case for completeness.
---
clang/test/CodeGen/X86/avx512f-builtins.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index e7f6b5aac8279..5a131ab6af367 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -8656,6 +8656,12 @@ __mmask16 test_kand_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_kand_mask16(0xCC, 0xAA) == 0x88); // data correctness
+TEST_CONSTEXPR(_kand_mask16(0x12345, 0xFFFF) == 0x2345); // should be truncated
+TEST_CONSTEXPR(_kand_mask16(0xABCD, 0x0000) == 0x0000); // all-zero
+TEST_CONSTEXPR(_kand_mask16(0x5678, 0xFFFF) == 0x5678); // all-one
+TEST_CONSTEXPR(_kand_mask16(0xAAAA, 0x5555) == 0x0000); // intersect
+
__mmask16 test_kandn_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask16
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
>From 03225502b6e66dadf9703338f00fc809357515f9 Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Mon, 22 Sep 2025 21:16:43 +0800
Subject: [PATCH 4/8] Allow kandn* to be used in constexpr
---
clang/include/clang/Basic/BuiltinsX86.td | 6 +++---
clang/lib/AST/ByteCode/InterpBuiltin.cpp | 8 ++++++++
clang/lib/AST/ExprConstant.cpp | 8 ++++++++
clang/lib/Headers/avx512bwintrin.h | 9 ++++-----
clang/lib/Headers/avx512dqintrin.h | 5 ++---
clang/lib/Headers/avx512fintrin.h | 5 ++---
clang/test/CodeGen/X86/avx512bw-builtins.c | 23 +++++++++++++++++-----
clang/test/CodeGen/X86/avx512dq-builtins.c | 6 ++++++
clang/test/CodeGen/X86/avx512f-builtins.c | 12 +++++++++++
9 files changed, 63 insertions(+), 19 deletions(-)
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index e5555236c7666..dc74685299736 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -3096,15 +3096,15 @@ let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kanddi : X86Builtin<"unsigned long long int(unsigned long long int, unsigned long long int)">;
}
-let Features = "avx512dq", Attributes = [NoThrow, Const] in {
+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {
def kandnqi : X86Builtin<"unsigned char(unsigned char, unsigned char)">;
}
-let Features = "avx512f", Attributes = [NoThrow, Const] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr] in {
def kandnhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;
}
-let Features = "avx512bw", Attributes = [NoThrow, Const] in {
+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kandnsi : X86Builtin<"unsigned int(unsigned int, unsigned int)">;
def kandndi : X86Builtin<"unsigned long long int(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 41d5d76c8c659..28a7e838d8e98 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -3596,6 +3596,14 @@ 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_kandnqi:
+ case X86::BI__builtin_ia32_kandnhi:
+ case X86::BI__builtin_ia32_kandnsi:
+ case X86::BI__builtin_ia32_kandndi:
+ return interp__builtin_elementwise_int_binop(
+ S, OpPC, Call,
+ [](const APSInt &LHS, const APSInt &RHS) { return ~LHS & RHS; });
+
case Builtin::BI__builtin_elementwise_fshl:
return interp__builtin_elementwise_triop(S, OpPC, Call,
llvm::APIntOps::fshl);
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index cd41c54087d08..257ffbe8b74e4 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -14662,6 +14662,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
return HandleMaskBinOp(
[](const APSInt &LHS, const APSInt &RHS) { return LHS & RHS; });
}
+
+ case X86::BI__builtin_ia32_kandnqi:
+ case X86::BI__builtin_ia32_kandnhi:
+ case X86::BI__builtin_ia32_kandnsi:
+ case X86::BI__builtin_ia32_kandndi: {
+ return HandleMaskBinOp(
+ [](const APSInt &LHS, const APSInt &RHS) { return ~LHS & RHS; });
+ }
}
}
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index cfe16062d8f65..bdabc2382c5e0 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -52,14 +52,13 @@ _kand_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kanddi((__mmask64)__A, (__mmask64)__B);
}
-static __inline__ __mmask32 __DEFAULT_FN_ATTRS
-_kandn_mask32(__mmask32 __A, __mmask32 __B)
-{
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR
+_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_CONSTEXPR
+_kandn_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kandndi((__mmask64)__A, (__mmask64)__B);
}
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index 52b3e9d148d6e..dbc50b464396d 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -40,9 +40,8 @@ _kand_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_kandqi((__mmask8)__A, (__mmask8)__B);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS
-_kandn_mask8(__mmask8 __A, __mmask8 __B)
-{
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR
+_kandn_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_kandnqi((__mmask8)__A, (__mmask8)__B);
}
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 5a45082f76e81..9dee968fad0a8 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8119,9 +8119,8 @@ static __inline__ __mmask16
return (__mmask16) __builtin_ia32_kandhi ((__mmask16) __A, (__mmask16) __B);
}
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
-_mm512_kandn (__mmask16 __A, __mmask16 __B)
-{
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_kandn(__mmask16 __A, __mmask16 __B) {
return (__mmask16) __builtin_ia32_kandnhi ((__mmask16) __A, (__mmask16) __B);
}
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
index 28e9ff611472a..a55b913fc859b 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -52,11 +52,11 @@ __mmask64 test_kand_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
-TEST_CONSTEXPR(_kand_mask64(0xCCCCCCCCCCCCCCCCull, 0xAAAAAAAAAAAAAAAAull) == 0x8888888888888888ull); // data correctness
-TEST_CONSTEXPR(_kand_mask64(0x123456789ABCDEF0ull, 0xFFFFFFFFFFFFFFFFull) == 0x123456789ABCDEF0ull); // full 64-bit value
-TEST_CONSTEXPR(_kand_mask64(0xABCDEF0123456789ull, 0x0000000000000000ull) == 0x0000000000000000ull); // all-zero
-TEST_CONSTEXPR(_kand_mask64(0x56789ABCDEF01234ull, 0xFFFFFFFFFFFFFFFFull) == 0x56789ABCDEF01234ull); // all-one
-TEST_CONSTEXPR(_kand_mask64(0xAAAAAAAAAAAAAAAAull, 0x5555555555555555ull) == 0x0000000000000000ull); // disjoint
+TEST_CONSTEXPR(_kand_mask64(0xCCCCCCCCCCCCCCCC, 0xAAAAAAAAAAAAAAAA) == 0x8888888888888888); // data correctness
+TEST_CONSTEXPR(_kand_mask64(0x123456789ABCDEF0, 0xFFFFFFFFFFFFFFFF) == 0x123456789ABCDEF0); // full 64-bit value
+TEST_CONSTEXPR(_kand_mask64(0xABCDEF0123456789, 0x0000000000000000) == 0x0000000000000000); // all-zero
+TEST_CONSTEXPR(_kand_mask64(0x56789ABCDEF01234, 0xFFFFFFFFFFFFFFFF) == 0x56789ABCDEF01234); // all-one
+TEST_CONSTEXPR(_kand_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0x0000000000000000); // disjoint
__mmask32 test_kandn_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask32
@@ -68,6 +68,13 @@ __mmask32 test_kandn_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
_mm512_cmpneq_epu16_mask(__C, __D)),
__E, __F);
}
+// 1100
+//
+TEST_CONSTEXPR(_kandn_mask32(0xA0A0F0F0, 0xCCCCCCCC) == 0x4C4C0C0C); // data correctness
+TEST_CONSTEXPR(_kandn_mask32(0x123456789, 0xFFFFFFFF) == 0xDCBA9876); // truncated and inverted
+TEST_CONSTEXPR(_kandn_mask32(0x00000000, 0x1234ABCD) == 0x1234ABCD); // all-ones (~0)
+TEST_CONSTEXPR(_kandn_mask32(0xFFFFFFFF, 0x87654321) == 0x00000000); // all-zero (~0xFFFFFFFF)
+TEST_CONSTEXPR(_kandn_mask32(0xAAAAAAAA, 0xAAAAAAAA) == 0x00000000); // ~A & A is 0
__mmask64 test_kandn_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask64
@@ -80,6 +87,12 @@ __mmask64 test_kandn_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
+TEST_CONSTEXPR(_kandn_mask64(0xA0A0F0F0C3C33C3C, 0xCCCCCCCCFFFF0000) == 0x4C4C0C0C3C3C0000); // data correctness
+TEST_CONSTEXPR(_kandn_mask64(0x0123456789ABCDEF, 0xFFFFFFFFFFFFFFFF) == 0xFEDCBA9876543210); // inverted with all-ones mask
+TEST_CONSTEXPR(_kandn_mask64(0x0, 0x1122334455667788) == 0x1122334455667788); // all-ones (~0)
+TEST_CONSTEXPR(_kandn_mask64(0xFFFFFFFFFFFFFFFF, 0x8877665544332211) == 0x0); // all-zero (~0xFFFFFFFFFFFFFFFF)
+TEST_CONSTEXPR(_kandn_mask64(0xAAAAAAAAAAAAAAAA, 0xAAAAAAAAAAAAAAAA) == 0x0); // ~A & A is 0
+
__mmask32 test_kor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask32
// CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
index cc15517fe5bb5..6a6d6611dc18e 100644
--- a/clang/test/CodeGen/X86/avx512dq-builtins.c
+++ b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -46,6 +46,12 @@ __mmask8 test_kandn_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
+TEST_CONSTEXPR(_kandn_mask8(0xC3, 0xA5) == 0x24); // data correctness
+TEST_CONSTEXPR(_kandn_mask8(0x1F0, 0xFF) == 0x0F); // truncated (to 0xF0) and inverted
+TEST_CONSTEXPR(_kandn_mask8(0x00, 0xB7) == 0xB7); // all-ones (~0)
+TEST_CONSTEXPR(_kandn_mask8(0xFF, 0x7E) == 0x00); // all-zero (~0xFF)
+TEST_CONSTEXPR(_kandn_mask8(0x55, 0x55) == 0x00); // ~A & A is 0
+
__mmask8 test_kor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask8
// CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index 5a131ab6af367..b18ba76473360 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -8520,6 +8520,12 @@ __mmask16 test_mm512_kandn(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_mm512_kandn(0xCC, 0xAA) == 0x22); // data correctness
+TEST_CONSTEXPR(_mm512_kandn(0x12345, 0xFFFF) == 0xDCBA); // truncated and inverted
+TEST_CONSTEXPR(_mm512_kandn(0x0000, 0xABCD) == 0xABCD); // all-ones (~0)
+TEST_CONSTEXPR(_mm512_kandn(0xFFFF, 0x5678) == 0x0000); // all-zero (~0xFFFF)
+TEST_CONSTEXPR(_mm512_kandn(0xAAAA, 0xAAAA) == 0x0000); // ~A & A is 0
+
__mmask16 test_mm512_kor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_mm512_kor
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
@@ -8674,6 +8680,12 @@ __mmask16 test_kandn_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
+TEST_CONSTEXPR(_kandn_mask16(0xCC, 0xAA) == 0x22); // data correctness
+TEST_CONSTEXPR(_kandn_mask16(0x12345, 0xFFFF) == 0xDCBA); // truncated and inverted
+TEST_CONSTEXPR(_kandn_mask16(0x0000, 0xABCD) == 0xABCD); // all-ones (~0)
+TEST_CONSTEXPR(_kandn_mask16(0xFFFF, 0x5678) == 0x0000); // all-zero (~0xFFFF)
+TEST_CONSTEXPR(_kandn_mask16(0xAAAA, 0xAAAA) == 0x0000); // ~A & A is 0
+
__mmask16 test_kor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask16
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
>From 2312a1c1ca36b7f36908e642c903ae61001ec228 Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Mon, 22 Sep 2025 22:20:12 +0800
Subject: [PATCH 5/8] Allow kor* to be used in constexpr
---
clang/include/clang/Basic/BuiltinsX86.td | 6 +++---
clang/lib/AST/ByteCode/InterpBuiltin.cpp | 8 ++++++++
clang/lib/AST/ExprConstant.cpp | 8 ++++++++
clang/lib/Headers/avx512bwintrin.h | 9 ++++-----
clang/lib/Headers/avx512dqintrin.h | 5 ++---
clang/lib/Headers/avx512fintrin.h | 5 ++---
clang/test/CodeGen/X86/avx512bw-builtins.c | 20 ++++++++++++++++----
clang/test/CodeGen/X86/avx512dq-builtins.c | 6 ++++++
clang/test/CodeGen/X86/avx512f-builtins.c | 12 ++++++++++++
9 files changed, 61 insertions(+), 18 deletions(-)
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index dc74685299736..edae7d66f037f 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -3109,15 +3109,15 @@ let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kandndi : X86Builtin<"unsigned long long int(unsigned long long int, unsigned long long int)">;
}
-let Features = "avx512dq", Attributes = [NoThrow, Const] in {
+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {
def korqi : X86Builtin<"unsigned char(unsigned char, unsigned char)">;
}
-let Features = "avx512f", Attributes = [NoThrow, Const] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr] in {
def korhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;
}
-let Features = "avx512bw", Attributes = [NoThrow, Const] in {
+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def korsi : X86Builtin<"unsigned int(unsigned int, unsigned int)">;
def kordi : X86Builtin<"unsigned long long int(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 28a7e838d8e98..0701af2ac49f2 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -3604,6 +3604,14 @@ 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_korqi:
+ case X86::BI__builtin_ia32_korhi:
+ case X86::BI__builtin_ia32_korsi:
+ case X86::BI__builtin_ia32_kordi:
+ return interp__builtin_elementwise_int_binop(
+ S, OpPC, Call,
+ [](const APSInt &LHS, const APSInt &RHS) { return LHS | RHS; });
+
case Builtin::BI__builtin_elementwise_fshl:
return interp__builtin_elementwise_triop(S, OpPC, Call,
llvm::APIntOps::fshl);
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 257ffbe8b74e4..61682314dfea2 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -14670,6 +14670,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
return HandleMaskBinOp(
[](const APSInt &LHS, const APSInt &RHS) { return ~LHS & RHS; });
}
+
+ case X86::BI__builtin_ia32_korqi:
+ case X86::BI__builtin_ia32_korhi:
+ case X86::BI__builtin_ia32_korsi:
+ case X86::BI__builtin_ia32_kordi: {
+ return HandleMaskBinOp(
+ [](const APSInt &LHS, const APSInt &RHS) { return LHS | RHS; });
+ }
}
}
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index bdabc2382c5e0..30874f54815a7 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -62,14 +62,13 @@ _kandn_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kandndi((__mmask64)__A, (__mmask64)__B);
}
-static __inline__ __mmask32 __DEFAULT_FN_ATTRS
-_kor_mask32(__mmask32 __A, __mmask32 __B)
-{
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR
+_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_CONSTEXPR
+_kor_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kordi((__mmask64)__A, (__mmask64)__B);
}
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index dbc50b464396d..1d9b772da7840 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -45,9 +45,8 @@ _kandn_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_kandnqi((__mmask8)__A, (__mmask8)__B);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS
-_kor_mask8(__mmask8 __A, __mmask8 __B)
-{
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR
+_kor_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_korqi((__mmask8)__A, (__mmask8)__B);
}
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 9dee968fad0a8..28a66df1b7534 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8124,9 +8124,8 @@ _mm512_kandn(__mmask16 __A, __mmask16 __B) {
return (__mmask16) __builtin_ia32_kandnhi ((__mmask16) __A, (__mmask16) __B);
}
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
-_mm512_kor (__mmask16 __A, __mmask16 __B)
-{
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_kor(__mmask16 __A, __mmask16 __B) {
return (__mmask16) __builtin_ia32_korhi ((__mmask16) __A, (__mmask16) __B);
}
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
index a55b913fc859b..91916ab5a0081 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -36,11 +36,11 @@ __mmask32 test_kand_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
-TEST_CONSTEXPR(_kand_mask32(0xCCCCCCCC, 0xAAAAAAAA) == 0x88888888); // data correctness
+TEST_CONSTEXPR(_kand_mask32(0xCCCCCCCC, 0xAAAAAAAA) == 0x88888888); // data correctness
TEST_CONSTEXPR(_kand_mask32(0x123456789, 0xFFFFFFFF) == 0x23456789); // should be truncated
-TEST_CONSTEXPR(_kand_mask32(0xABCDEF01, 0x00000000) == 0x00000000); // all-zero
-TEST_CONSTEXPR(_kand_mask32(0x56789ABC, 0xFFFFFFFF) == 0x56789ABC); // all-one
-TEST_CONSTEXPR(_kand_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000); // disjoint
+TEST_CONSTEXPR(_kand_mask32(0xABCDEF01, 0x00000000) == 0x00000000); // all-zero
+TEST_CONSTEXPR(_kand_mask32(0x56789ABC, 0xFFFFFFFF) == 0x56789ABC); // all-one
+TEST_CONSTEXPR(_kand_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000); // disjoint
__mmask64 test_kand_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kand_mask64
@@ -103,6 +103,12 @@ __mmask32 test_kor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
+TEST_CONSTEXPR(_kor_mask32(0xF0F0A5A5, 0x0F0F5A5A) == 0xFFFFFFFF); // data correctness
+TEST_CONSTEXPR(_kor_mask32(0x12345ABCD, 0x12345ABCD) == 0x2345ABCD); // truncated
+TEST_CONSTEXPR(_kor_mask32(0x1A2B3C4D, 0x00000000) == 0x1A2B3C4D); // all-zero
+TEST_CONSTEXPR(_kor_mask32(0xDEADBEEF, 0xFFFFFFFF) == 0xFFFFFFFF); // all-ones
+TEST_CONSTEXPR(_kor_mask32(0xAAAAAAAA, 0x55555555) == 0xFFFFFFFF); // disjoint
+
__mmask64 test_kor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask64
// CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
@@ -113,6 +119,12 @@ __mmask64 test_kor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
+TEST_CONSTEXPR(_kor_mask64(0xF0A5C33C00FF11EE, 0x0F5AC33CFF00EE11) == 0xFFFFC33CFFFFFFFF); // data correctness
+TEST_CONSTEXPR(_kor_mask64(0x123456789ABCDEF0, 0x123456789ABCDEF0) == 0x123456789ABCDEF0); // full 64-bit value
+TEST_CONSTEXPR(_kor_mask64(0x1122334455667788, 0x0) == 0x1122334455667788); // all-zero
+TEST_CONSTEXPR(_kor_mask64(0x8877665544332211, 0xFFFFFFFFFFFFFFFF) == 0xFFFFFFFFFFFFFFFF); // all-ones
+TEST_CONSTEXPR(_kor_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0xFFFFFFFFFFFFFFFF); // disjoint
+
__mmask32 test_kxnor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask32
// CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
index 6a6d6611dc18e..49a47021d337c 100644
--- a/clang/test/CodeGen/X86/avx512dq-builtins.c
+++ b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -62,6 +62,12 @@ __mmask8 test_kor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m5
__E, __F);
}
+TEST_CONSTEXPR(_kor_mask8(0xB3, 0x6C) == 0xFF); // data correctness
+TEST_CONSTEXPR(_kor_mask8(0x1A5, 0x1A5) == 0xA5); // truncated
+TEST_CONSTEXPR(_kor_mask8(0xDE, 0x00) == 0xDE); // all-zero
+TEST_CONSTEXPR(_kor_mask8(0x42, 0xFF) == 0xFF); // all-ones
+TEST_CONSTEXPR(_kor_mask8(0xAA, 0x55) == 0xFF); // disjoint
+
__mmask8 test_kxnor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask8
// CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index b18ba76473360..90d09da8807fe 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -8537,6 +8537,12 @@ __mmask16 test_mm512_kor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m
__E, __F);
}
+TEST_CONSTEXPR(_mm512_kor(0xC1, 0xA8) == 0xE9); // data correctness
+TEST_CONSTEXPR(_mm512_kor(0x12345, 0x12345) == 0x2345); // truncated
+TEST_CONSTEXPR(_mm512_kor(0xABCD, 0x0000) == 0xABCD); // all-zero
+TEST_CONSTEXPR(_mm512_kor(0xABCD, 0xFFFF) == 0xFFFF); // all-ones
+TEST_CONSTEXPR(_mm512_kor(0xAAAA, 0x5555) == 0xFFFF); // disjoint
+
int test_mm512_kortestc(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
// CHECK-LABEL: test_mm512_kortestc
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
@@ -8697,6 +8703,12 @@ __mmask16 test_kor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
+TEST_CONSTEXPR(_kor_mask16(0xC1, 0xA8) == 0xE9); // data correctness
+TEST_CONSTEXPR(_kor_mask16(0x12345, 0x12345) == 0x2345); // truncated
+TEST_CONSTEXPR(_kor_mask16(0xABCD, 0x0000) == 0xABCD); // all-zero
+TEST_CONSTEXPR(_kor_mask16(0xABCD, 0xFFFF) == 0xFFFF); // all-ones
+TEST_CONSTEXPR(_kor_mask16(0xAAAA, 0x5555) == 0xFFFF); // disjoint
+
__mmask16 test_kxnor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask16
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
>From 1132cd33804fab47a9a85976238c558c9b602f76 Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Tue, 23 Sep 2025 00:42:42 +0800
Subject: [PATCH 6/8] Allow kxnor* to be used in constexpr
---
clang/include/clang/Basic/BuiltinsX86.td | 6 +++---
clang/lib/AST/ByteCode/InterpBuiltin.cpp | 8 ++++++++
clang/lib/AST/ExprConstant.cpp | 8 ++++++++
clang/lib/Headers/avx512bwintrin.h | 9 ++++-----
clang/lib/Headers/avx512dqintrin.h | 5 ++---
clang/lib/Headers/avx512fintrin.h | 5 ++---
clang/test/CodeGen/X86/avx512bw-builtins.c | 12 ++++++++++++
clang/test/CodeGen/X86/avx512dq-builtins.c | 6 ++++++
clang/test/CodeGen/X86/avx512f-builtins.c | 14 +++++++++++++-
9 files changed, 58 insertions(+), 15 deletions(-)
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index edae7d66f037f..0d4034922f3d5 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -3157,15 +3157,15 @@ let Features = "avx512f", Attributes = [NoThrow, Const] in {
def kunpckhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;
}
-let Features = "avx512dq", Attributes = [NoThrow, Const] in {
+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {
def kxnorqi : X86Builtin<"unsigned char(unsigned char, unsigned char)">;
}
-let Features = "avx512f", Attributes = [NoThrow, Const] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr] in {
def kxnorhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;
}
-let Features = "avx512bw", Attributes = [NoThrow, Const] in {
+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kxnorsi : X86Builtin<"unsigned int(unsigned int, unsigned int)">;
def kxnordi : X86Builtin<"unsigned long long int(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 0701af2ac49f2..98bb8932a76b5 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -3612,6 +3612,14 @@ 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_kxnorqi:
+ case X86::BI__builtin_ia32_kxnorhi:
+ case X86::BI__builtin_ia32_kxnorsi:
+ case X86::BI__builtin_ia32_kxnordi:
+ return interp__builtin_elementwise_int_binop(
+ S, OpPC, Call,
+ [](const APSInt &LHS, const APSInt &RHS) { return ~(LHS ^ RHS); });
+
case Builtin::BI__builtin_elementwise_fshl:
return interp__builtin_elementwise_triop(S, OpPC, Call,
llvm::APIntOps::fshl);
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 61682314dfea2..710b4c36499d1 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -14678,6 +14678,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
return HandleMaskBinOp(
[](const APSInt &LHS, const APSInt &RHS) { return LHS | RHS; });
}
+
+ case X86::BI__builtin_ia32_kxnorqi:
+ case X86::BI__builtin_ia32_kxnorhi:
+ case X86::BI__builtin_ia32_kxnorsi:
+ case X86::BI__builtin_ia32_kxnordi: {
+ return HandleMaskBinOp(
+ [](const APSInt &LHS, const APSInt &RHS) { return ~(LHS ^ RHS); });
+ }
}
}
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 30874f54815a7..0c470d0253c80 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -72,14 +72,13 @@ _kor_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kordi((__mmask64)__A, (__mmask64)__B);
}
-static __inline__ __mmask32 __DEFAULT_FN_ATTRS
-_kxnor_mask32(__mmask32 __A, __mmask32 __B)
-{
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR
+_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_CONSTEXPR
+_kxnor_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kxnordi((__mmask64)__A, (__mmask64)__B);
}
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index 1d9b772da7840..bf7f620de1629 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -50,9 +50,8 @@ _kor_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_korqi((__mmask8)__A, (__mmask8)__B);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS
-_kxnor_mask8(__mmask8 __A, __mmask8 __B)
-{
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR
+_kxnor_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_kxnorqi((__mmask8)__A, (__mmask8)__B);
}
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 28a66df1b7534..4b27f20192edb 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8165,9 +8165,8 @@ _mm512_kunpackb (__mmask16 __A, __mmask16 __B)
return (__mmask16) __builtin_ia32_kunpckhi ((__mmask16) __A, (__mmask16) __B);
}
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
-_mm512_kxnor (__mmask16 __A, __mmask16 __B)
-{
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_kxnor(__mmask16 __A, __mmask16 __B) {
return (__mmask16) __builtin_ia32_kxnorhi ((__mmask16) __A, (__mmask16) __B);
}
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
index 91916ab5a0081..0b596b6322d6a 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -136,6 +136,12 @@ __mmask32 test_kxnor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
+TEST_CONSTEXPR(_kxnor_mask32(0x1234ABCD, 0xFFFF0000) == 0x12345432); // data correctness
+TEST_CONSTEXPR(_kxnor_mask32(0x123456789ABCDEF0, 0xFFFFFFFF) == 0x9ABCDEF0); // should be truncated to 32 bits
+TEST_CONSTEXPR(_kxnor_mask32(0xAABBCCDD, 0x00000000) == 0x55443322); // all-zero mask, res = ~LHS
+TEST_CONSTEXPR(_kxnor_mask32(0x87654321, 0xFFFFFFFF) == 0x87654321); // all-one mask, res = LHS
+TEST_CONSTEXPR(_kxnor_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000); // ~A xnor A == 0
+
__mmask64 test_kxnor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask64
// CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
@@ -147,6 +153,12 @@ __mmask64 test_kxnor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
+TEST_CONSTEXPR(_kxnor_mask64(0x0123456789ABCDEF, 0xFFFFFFFF00000000) == 0x0123456776543210); // data correctness
+TEST_CONSTEXPR(_kxnor_mask64(0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F) == 0xFFFFFFFFFFFFFFFF); // full 64 bits
+TEST_CONSTEXPR(_kxnor_mask64(0xFEDCBA9876543210, 0xFFFFFFFFFFFFFFFF) == 0xFEDCBA9876543210); // all-one mask, res = LHS
+TEST_CONSTEXPR(_kxnor_mask64(0xAABBCCDD11223344, 0x0000000000000000) == 0x55443322EEDDCCBB); // all-zero mask, res = ~LHS
+TEST_CONSTEXPR(_kxnor_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0x0000000000000000); // ~A xnor A == 0
+
__mmask32 test_kxor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxor_mask32
// CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
index 49a47021d337c..6480e8c4237ae 100644
--- a/clang/test/CodeGen/X86/avx512dq-builtins.c
+++ b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -79,6 +79,12 @@ __mmask8 test_kxnor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
+TEST_CONSTEXPR(_kxnor_mask8(0xC5, 0xAF) == 0x95); // data correctness
+TEST_CONSTEXPR(_kxnor_mask8(0x1234, 0xFF) == 0x34); // should be truncated to 8 bits
+TEST_CONSTEXPR(_kxnor_mask8(0x3A, 0x00) == 0xC5); // all-zero mask, res = ~LHS
+TEST_CONSTEXPR(_kxnor_mask8(0xB4, 0xFF) == 0xB4); // all-one mask, res = LHS
+TEST_CONSTEXPR(_kxnor_mask8(0xAA, 0x55) == 0x00); // ~A xnor A == 0
+
__mmask8 test_kxor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxor_mask8
// CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index 90d09da8807fe..7180ff1b31f6a 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -8638,6 +8638,12 @@ __mmask16 test_mm512_kxnor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_mm512_kxnor(0x00C5, 0xFFAF) == 0x95); // data correctness
+TEST_CONSTEXPR(_mm512_kxnor(0x12345, 0xFFFF) == 0x2345); // should be truncated
+TEST_CONSTEXPR(_mm512_kxnor(0xABCD, 0x0000) == 0x5432); // all-zero, res = ~LHS
+TEST_CONSTEXPR(_mm512_kxnor(0x5678, 0xFFFF) == 0x5678); // all-one, res = LHS
+TEST_CONSTEXPR(_mm512_kxnor(0xAAAA, 0x5555) == 0x0000); // ~A xnor A == 0
+
__mmask16 test_mm512_kxor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_mm512_kxor
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
@@ -8672,7 +8678,7 @@ TEST_CONSTEXPR(_kand_mask16(0xCC, 0xAA) == 0x88); // data correctness
TEST_CONSTEXPR(_kand_mask16(0x12345, 0xFFFF) == 0x2345); // should be truncated
TEST_CONSTEXPR(_kand_mask16(0xABCD, 0x0000) == 0x0000); // all-zero
TEST_CONSTEXPR(_kand_mask16(0x5678, 0xFFFF) == 0x5678); // all-one
-TEST_CONSTEXPR(_kand_mask16(0xAAAA, 0x5555) == 0x0000); // intersect
+TEST_CONSTEXPR(_kand_mask16(0xAAAA, 0x5555) == 0x0000); // disjoint
__mmask16 test_kandn_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask16
@@ -8721,6 +8727,12 @@ __mmask16 test_kxnor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
+TEST_CONSTEXPR(_kxnor_mask16(0x00C5, 0xFFAF) == 0x95); // data correctness
+TEST_CONSTEXPR(_kxnor_mask16(0x12345, 0xFFFF) == 0x2345); // should be truncated
+TEST_CONSTEXPR(_kxnor_mask16(0xABCD, 0x0000) == 0x5432); // all-zero, res = ~LHS
+TEST_CONSTEXPR(_kxnor_mask16(0x5678, 0xFFFF) == 0x5678); // all-one, res = LHS
+TEST_CONSTEXPR(_kxnor_mask16(0xAAAA, 0x5555) == 0x0000); // ~A xnor A == 0
+
__mmask16 test_kxor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxor_mask16
// CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
>From 641a09069d69ef83d7e9def7727336faef1ac2db Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Tue, 23 Sep 2025 01:03:38 +0800
Subject: [PATCH 7/8] Remove too verbose comments
Some comments in the constexpr test suite were a bit too verbose. This
commit removes them to keep the code concise.
---
clang/test/CodeGen/X86/avx512bw-builtins.c | 80 +++++++++++-----------
clang/test/CodeGen/X86/avx512dq-builtins.c | 40 +++++------
clang/test/CodeGen/X86/avx512f-builtins.c | 80 +++++++++++-----------
3 files changed, 100 insertions(+), 100 deletions(-)
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
index 0b596b6322d6a..3128078bfef6d 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -36,11 +36,11 @@ __mmask32 test_kand_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
-TEST_CONSTEXPR(_kand_mask32(0xCCCCCCCC, 0xAAAAAAAA) == 0x88888888); // data correctness
-TEST_CONSTEXPR(_kand_mask32(0x123456789, 0xFFFFFFFF) == 0x23456789); // should be truncated
-TEST_CONSTEXPR(_kand_mask32(0xABCDEF01, 0x00000000) == 0x00000000); // all-zero
-TEST_CONSTEXPR(_kand_mask32(0x56789ABC, 0xFFFFFFFF) == 0x56789ABC); // all-one
-TEST_CONSTEXPR(_kand_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000); // disjoint
+TEST_CONSTEXPR(_kand_mask32(0xCCCCCCCC, 0xAAAAAAAA) == 0x88888888);
+TEST_CONSTEXPR(_kand_mask32(0x123456789, 0xFFFFFFFF) == 0x23456789);
+TEST_CONSTEXPR(_kand_mask32(0xABCDEF01, 0x00000000) == 0x00000000);
+TEST_CONSTEXPR(_kand_mask32(0x56789ABC, 0xFFFFFFFF) == 0x56789ABC);
+TEST_CONSTEXPR(_kand_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000);
__mmask64 test_kand_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kand_mask64
@@ -52,11 +52,11 @@ __mmask64 test_kand_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
-TEST_CONSTEXPR(_kand_mask64(0xCCCCCCCCCCCCCCCC, 0xAAAAAAAAAAAAAAAA) == 0x8888888888888888); // data correctness
-TEST_CONSTEXPR(_kand_mask64(0x123456789ABCDEF0, 0xFFFFFFFFFFFFFFFF) == 0x123456789ABCDEF0); // full 64-bit value
-TEST_CONSTEXPR(_kand_mask64(0xABCDEF0123456789, 0x0000000000000000) == 0x0000000000000000); // all-zero
-TEST_CONSTEXPR(_kand_mask64(0x56789ABCDEF01234, 0xFFFFFFFFFFFFFFFF) == 0x56789ABCDEF01234); // all-one
-TEST_CONSTEXPR(_kand_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0x0000000000000000); // disjoint
+TEST_CONSTEXPR(_kand_mask64(0xCCCCCCCCCCCCCCCC, 0xAAAAAAAAAAAAAAAA) == 0x8888888888888888);
+TEST_CONSTEXPR(_kand_mask64(0x123456789ABCDEF0, 0xFFFFFFFFFFFFFFFF) == 0x123456789ABCDEF0);
+TEST_CONSTEXPR(_kand_mask64(0xABCDEF0123456789, 0x0000000000000000) == 0x0000000000000000);
+TEST_CONSTEXPR(_kand_mask64(0x56789ABCDEF01234, 0xFFFFFFFFFFFFFFFF) == 0x56789ABCDEF01234);
+TEST_CONSTEXPR(_kand_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0x0000000000000000);
__mmask32 test_kandn_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask32
@@ -70,11 +70,11 @@ __mmask32 test_kandn_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
}
// 1100
//
-TEST_CONSTEXPR(_kandn_mask32(0xA0A0F0F0, 0xCCCCCCCC) == 0x4C4C0C0C); // data correctness
-TEST_CONSTEXPR(_kandn_mask32(0x123456789, 0xFFFFFFFF) == 0xDCBA9876); // truncated and inverted
-TEST_CONSTEXPR(_kandn_mask32(0x00000000, 0x1234ABCD) == 0x1234ABCD); // all-ones (~0)
-TEST_CONSTEXPR(_kandn_mask32(0xFFFFFFFF, 0x87654321) == 0x00000000); // all-zero (~0xFFFFFFFF)
-TEST_CONSTEXPR(_kandn_mask32(0xAAAAAAAA, 0xAAAAAAAA) == 0x00000000); // ~A & A is 0
+TEST_CONSTEXPR(_kandn_mask32(0xA0A0F0F0, 0xCCCCCCCC) == 0x4C4C0C0C);
+TEST_CONSTEXPR(_kandn_mask32(0x123456789, 0xFFFFFFFF) == 0xDCBA9876);
+TEST_CONSTEXPR(_kandn_mask32(0x00000000, 0x1234ABCD) == 0x1234ABCD);
+TEST_CONSTEXPR(_kandn_mask32(0xFFFFFFFF, 0x87654321) == 0x00000000);
+TEST_CONSTEXPR(_kandn_mask32(0xAAAAAAAA, 0xAAAAAAAA) == 0x00000000);
__mmask64 test_kandn_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask64
@@ -87,11 +87,11 @@ __mmask64 test_kandn_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
-TEST_CONSTEXPR(_kandn_mask64(0xA0A0F0F0C3C33C3C, 0xCCCCCCCCFFFF0000) == 0x4C4C0C0C3C3C0000); // data correctness
-TEST_CONSTEXPR(_kandn_mask64(0x0123456789ABCDEF, 0xFFFFFFFFFFFFFFFF) == 0xFEDCBA9876543210); // inverted with all-ones mask
-TEST_CONSTEXPR(_kandn_mask64(0x0, 0x1122334455667788) == 0x1122334455667788); // all-ones (~0)
-TEST_CONSTEXPR(_kandn_mask64(0xFFFFFFFFFFFFFFFF, 0x8877665544332211) == 0x0); // all-zero (~0xFFFFFFFFFFFFFFFF)
-TEST_CONSTEXPR(_kandn_mask64(0xAAAAAAAAAAAAAAAA, 0xAAAAAAAAAAAAAAAA) == 0x0); // ~A & A is 0
+TEST_CONSTEXPR(_kandn_mask64(0xA0A0F0F0C3C33C3C, 0xCCCCCCCCFFFF0000) == 0x4C4C0C0C3C3C0000);
+TEST_CONSTEXPR(_kandn_mask64(0x0123456789ABCDEF, 0xFFFFFFFFFFFFFFFF) == 0xFEDCBA9876543210);
+TEST_CONSTEXPR(_kandn_mask64(0x0, 0x1122334455667788) == 0x1122334455667788);
+TEST_CONSTEXPR(_kandn_mask64(0xFFFFFFFFFFFFFFFF, 0x8877665544332211) == 0x0);
+TEST_CONSTEXPR(_kandn_mask64(0xAAAAAAAAAAAAAAAA, 0xAAAAAAAAAAAAAAAA) == 0x0);
__mmask32 test_kor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask32
@@ -103,11 +103,11 @@ __mmask32 test_kor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
-TEST_CONSTEXPR(_kor_mask32(0xF0F0A5A5, 0x0F0F5A5A) == 0xFFFFFFFF); // data correctness
-TEST_CONSTEXPR(_kor_mask32(0x12345ABCD, 0x12345ABCD) == 0x2345ABCD); // truncated
-TEST_CONSTEXPR(_kor_mask32(0x1A2B3C4D, 0x00000000) == 0x1A2B3C4D); // all-zero
-TEST_CONSTEXPR(_kor_mask32(0xDEADBEEF, 0xFFFFFFFF) == 0xFFFFFFFF); // all-ones
-TEST_CONSTEXPR(_kor_mask32(0xAAAAAAAA, 0x55555555) == 0xFFFFFFFF); // disjoint
+TEST_CONSTEXPR(_kor_mask32(0xF0F0A5A5, 0x0F0F5A5A) == 0xFFFFFFFF);
+TEST_CONSTEXPR(_kor_mask32(0x12345ABCD, 0x12345ABCD) == 0x2345ABCD);
+TEST_CONSTEXPR(_kor_mask32(0x1A2B3C4D, 0x00000000) == 0x1A2B3C4D);
+TEST_CONSTEXPR(_kor_mask32(0xDEADBEEF, 0xFFFFFFFF) == 0xFFFFFFFF);
+TEST_CONSTEXPR(_kor_mask32(0xAAAAAAAA, 0x55555555) == 0xFFFFFFFF);
__mmask64 test_kor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask64
@@ -119,11 +119,11 @@ __mmask64 test_kor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
-TEST_CONSTEXPR(_kor_mask64(0xF0A5C33C00FF11EE, 0x0F5AC33CFF00EE11) == 0xFFFFC33CFFFFFFFF); // data correctness
-TEST_CONSTEXPR(_kor_mask64(0x123456789ABCDEF0, 0x123456789ABCDEF0) == 0x123456789ABCDEF0); // full 64-bit value
-TEST_CONSTEXPR(_kor_mask64(0x1122334455667788, 0x0) == 0x1122334455667788); // all-zero
-TEST_CONSTEXPR(_kor_mask64(0x8877665544332211, 0xFFFFFFFFFFFFFFFF) == 0xFFFFFFFFFFFFFFFF); // all-ones
-TEST_CONSTEXPR(_kor_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0xFFFFFFFFFFFFFFFF); // disjoint
+TEST_CONSTEXPR(_kor_mask64(0xF0A5C33C00FF11EE, 0x0F5AC33CFF00EE11) == 0xFFFFC33CFFFFFFFF);
+TEST_CONSTEXPR(_kor_mask64(0x123456789ABCDEF0, 0x123456789ABCDEF0) == 0x123456789ABCDEF0);
+TEST_CONSTEXPR(_kor_mask64(0x1122334455667788, 0x0) == 0x1122334455667788);
+TEST_CONSTEXPR(_kor_mask64(0x8877665544332211, 0xFFFFFFFFFFFFFFFF) == 0xFFFFFFFFFFFFFFFF);
+TEST_CONSTEXPR(_kor_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0xFFFFFFFFFFFFFFFF);
__mmask32 test_kxnor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask32
@@ -136,11 +136,11 @@ __mmask32 test_kxnor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
-TEST_CONSTEXPR(_kxnor_mask32(0x1234ABCD, 0xFFFF0000) == 0x12345432); // data correctness
-TEST_CONSTEXPR(_kxnor_mask32(0x123456789ABCDEF0, 0xFFFFFFFF) == 0x9ABCDEF0); // should be truncated to 32 bits
-TEST_CONSTEXPR(_kxnor_mask32(0xAABBCCDD, 0x00000000) == 0x55443322); // all-zero mask, res = ~LHS
-TEST_CONSTEXPR(_kxnor_mask32(0x87654321, 0xFFFFFFFF) == 0x87654321); // all-one mask, res = LHS
-TEST_CONSTEXPR(_kxnor_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000); // ~A xnor A == 0
+TEST_CONSTEXPR(_kxnor_mask32(0x1234ABCD, 0xFFFF0000) == 0x12345432);
+TEST_CONSTEXPR(_kxnor_mask32(0x123456789ABCDEF0, 0xFFFFFFFF) == 0x9ABCDEF0);
+TEST_CONSTEXPR(_kxnor_mask32(0xAABBCCDD, 0x00000000) == 0x55443322);
+TEST_CONSTEXPR(_kxnor_mask32(0x87654321, 0xFFFFFFFF) == 0x87654321);
+TEST_CONSTEXPR(_kxnor_mask32(0xAAAAAAAA, 0x55555555) == 0x00000000);
__mmask64 test_kxnor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask64
@@ -153,11 +153,11 @@ __mmask64 test_kxnor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
-TEST_CONSTEXPR(_kxnor_mask64(0x0123456789ABCDEF, 0xFFFFFFFF00000000) == 0x0123456776543210); // data correctness
-TEST_CONSTEXPR(_kxnor_mask64(0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F) == 0xFFFFFFFFFFFFFFFF); // full 64 bits
-TEST_CONSTEXPR(_kxnor_mask64(0xFEDCBA9876543210, 0xFFFFFFFFFFFFFFFF) == 0xFEDCBA9876543210); // all-one mask, res = LHS
-TEST_CONSTEXPR(_kxnor_mask64(0xAABBCCDD11223344, 0x0000000000000000) == 0x55443322EEDDCCBB); // all-zero mask, res = ~LHS
-TEST_CONSTEXPR(_kxnor_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0x0000000000000000); // ~A xnor A == 0
+TEST_CONSTEXPR(_kxnor_mask64(0x0123456789ABCDEF, 0xFFFFFFFF00000000) == 0x0123456776543210);
+TEST_CONSTEXPR(_kxnor_mask64(0x0F0F0F0F0F0F0F0F, 0x0F0F0F0F0F0F0F0F) == 0xFFFFFFFFFFFFFFFF);
+TEST_CONSTEXPR(_kxnor_mask64(0xFEDCBA9876543210, 0xFFFFFFFFFFFFFFFF) == 0xFEDCBA9876543210);
+TEST_CONSTEXPR(_kxnor_mask64(0xAABBCCDD11223344, 0x0000000000000000) == 0x55443322EEDDCCBB);
+TEST_CONSTEXPR(_kxnor_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0x0000000000000000);
__mmask32 test_kxor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxor_mask32
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
index 6480e8c4237ae..fa4fd9ddee42e 100644
--- a/clang/test/CodeGen/X86/avx512dq-builtins.c
+++ b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -29,11 +29,11 @@ __mmask8 test_kand_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m
__E, __F);
}
-TEST_CONSTEXPR(_kand_mask8(0x0C, 0x0A) == 0x08); // data correctness
-TEST_CONSTEXPR(_kand_mask8(0x123, 0xFF) == 0x23); // should be truncated
-TEST_CONSTEXPR(_kand_mask8(0xAB, 0x00) == 0x00); // all-zero
-TEST_CONSTEXPR(_kand_mask8(0x56, 0xFF) == 0x56); // all-one
-TEST_CONSTEXPR(_kand_mask8(0xAA, 0x55) == 0x00); // disjoint
+TEST_CONSTEXPR(_kand_mask8(0x0C, 0x0A) == 0x08);
+TEST_CONSTEXPR(_kand_mask8(0x123, 0xFF) == 0x23);
+TEST_CONSTEXPR(_kand_mask8(0xAB, 0x00) == 0x00);
+TEST_CONSTEXPR(_kand_mask8(0x56, 0xFF) == 0x56);
+TEST_CONSTEXPR(_kand_mask8(0xAA, 0x55) == 0x00);
__mmask8 test_kandn_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask8
@@ -46,11 +46,11 @@ __mmask8 test_kandn_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
-TEST_CONSTEXPR(_kandn_mask8(0xC3, 0xA5) == 0x24); // data correctness
-TEST_CONSTEXPR(_kandn_mask8(0x1F0, 0xFF) == 0x0F); // truncated (to 0xF0) and inverted
-TEST_CONSTEXPR(_kandn_mask8(0x00, 0xB7) == 0xB7); // all-ones (~0)
-TEST_CONSTEXPR(_kandn_mask8(0xFF, 0x7E) == 0x00); // all-zero (~0xFF)
-TEST_CONSTEXPR(_kandn_mask8(0x55, 0x55) == 0x00); // ~A & A is 0
+TEST_CONSTEXPR(_kandn_mask8(0xC3, 0xA5) == 0x24);
+TEST_CONSTEXPR(_kandn_mask8(0x1F0, 0xFF) == 0x0F);
+TEST_CONSTEXPR(_kandn_mask8(0x00, 0xB7) == 0xB7);
+TEST_CONSTEXPR(_kandn_mask8(0xFF, 0x7E) == 0x00);
+TEST_CONSTEXPR(_kandn_mask8(0x55, 0x55) == 0x00);
__mmask8 test_kor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask8
@@ -62,11 +62,11 @@ __mmask8 test_kor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m5
__E, __F);
}
-TEST_CONSTEXPR(_kor_mask8(0xB3, 0x6C) == 0xFF); // data correctness
-TEST_CONSTEXPR(_kor_mask8(0x1A5, 0x1A5) == 0xA5); // truncated
-TEST_CONSTEXPR(_kor_mask8(0xDE, 0x00) == 0xDE); // all-zero
-TEST_CONSTEXPR(_kor_mask8(0x42, 0xFF) == 0xFF); // all-ones
-TEST_CONSTEXPR(_kor_mask8(0xAA, 0x55) == 0xFF); // disjoint
+TEST_CONSTEXPR(_kor_mask8(0xB3, 0x6C) == 0xFF);
+TEST_CONSTEXPR(_kor_mask8(0x1A5, 0x1A5) == 0xA5);
+TEST_CONSTEXPR(_kor_mask8(0xDE, 0x00) == 0xDE);
+TEST_CONSTEXPR(_kor_mask8(0x42, 0xFF) == 0xFF);
+TEST_CONSTEXPR(_kor_mask8(0xAA, 0x55) == 0xFF);
__mmask8 test_kxnor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask8
@@ -79,11 +79,11 @@ __mmask8 test_kxnor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
-TEST_CONSTEXPR(_kxnor_mask8(0xC5, 0xAF) == 0x95); // data correctness
-TEST_CONSTEXPR(_kxnor_mask8(0x1234, 0xFF) == 0x34); // should be truncated to 8 bits
-TEST_CONSTEXPR(_kxnor_mask8(0x3A, 0x00) == 0xC5); // all-zero mask, res = ~LHS
-TEST_CONSTEXPR(_kxnor_mask8(0xB4, 0xFF) == 0xB4); // all-one mask, res = LHS
-TEST_CONSTEXPR(_kxnor_mask8(0xAA, 0x55) == 0x00); // ~A xnor A == 0
+TEST_CONSTEXPR(_kxnor_mask8(0xC5, 0xAF) == 0x95);
+TEST_CONSTEXPR(_kxnor_mask8(0x1234, 0xFF) == 0x34);
+TEST_CONSTEXPR(_kxnor_mask8(0x3A, 0x00) == 0xC5);
+TEST_CONSTEXPR(_kxnor_mask8(0xB4, 0xFF) == 0xB4);
+TEST_CONSTEXPR(_kxnor_mask8(0xAA, 0x55) == 0x00);
__mmask8 test_kxor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxor_mask8
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index 7180ff1b31f6a..5e5b335e755a5 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -8502,11 +8502,11 @@ __mmask16 test_mm512_kand(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
-TEST_CONSTEXPR(_mm512_kand(0xCC, 0xAA) == 0x88); // data correctness
-TEST_CONSTEXPR(_mm512_kand(0x12345, 0xFFFF) == 0x2345); // should be truncated
-TEST_CONSTEXPR(_mm512_kand(0xABCD, 0x0000) == 0x0000); // all-zero
-TEST_CONSTEXPR(_mm512_kand(0x5678, 0xFFFF) == 0x5678); // all-one
-TEST_CONSTEXPR(_mm512_kand(0xAAAA, 0x5555) == 0x0000); // disjoint
+TEST_CONSTEXPR(_mm512_kand(0xCC, 0xAA) == 0x88);
+TEST_CONSTEXPR(_mm512_kand(0x12345, 0xFFFF) == 0x2345);
+TEST_CONSTEXPR(_mm512_kand(0xABCD, 0x0000) == 0x0000);
+TEST_CONSTEXPR(_mm512_kand(0x5678, 0xFFFF) == 0x5678);
+TEST_CONSTEXPR(_mm512_kand(0xAAAA, 0x5555) == 0x0000);
__mmask16 test_mm512_kandn(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_mm512_kandn
@@ -8520,11 +8520,11 @@ __mmask16 test_mm512_kandn(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
-TEST_CONSTEXPR(_mm512_kandn(0xCC, 0xAA) == 0x22); // data correctness
-TEST_CONSTEXPR(_mm512_kandn(0x12345, 0xFFFF) == 0xDCBA); // truncated and inverted
-TEST_CONSTEXPR(_mm512_kandn(0x0000, 0xABCD) == 0xABCD); // all-ones (~0)
-TEST_CONSTEXPR(_mm512_kandn(0xFFFF, 0x5678) == 0x0000); // all-zero (~0xFFFF)
-TEST_CONSTEXPR(_mm512_kandn(0xAAAA, 0xAAAA) == 0x0000); // ~A & A is 0
+TEST_CONSTEXPR(_mm512_kandn(0xCC, 0xAA) == 0x22);
+TEST_CONSTEXPR(_mm512_kandn(0x12345, 0xFFFF) == 0xDCBA);
+TEST_CONSTEXPR(_mm512_kandn(0x0000, 0xABCD) == 0xABCD);
+TEST_CONSTEXPR(_mm512_kandn(0xFFFF, 0x5678) == 0x0000);
+TEST_CONSTEXPR(_mm512_kandn(0xAAAA, 0xAAAA) == 0x0000);
__mmask16 test_mm512_kor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_mm512_kor
@@ -8537,11 +8537,11 @@ __mmask16 test_mm512_kor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m
__E, __F);
}
-TEST_CONSTEXPR(_mm512_kor(0xC1, 0xA8) == 0xE9); // data correctness
-TEST_CONSTEXPR(_mm512_kor(0x12345, 0x12345) == 0x2345); // truncated
-TEST_CONSTEXPR(_mm512_kor(0xABCD, 0x0000) == 0xABCD); // all-zero
-TEST_CONSTEXPR(_mm512_kor(0xABCD, 0xFFFF) == 0xFFFF); // all-ones
-TEST_CONSTEXPR(_mm512_kor(0xAAAA, 0x5555) == 0xFFFF); // disjoint
+TEST_CONSTEXPR(_mm512_kor(0xC1, 0xA8) == 0xE9);
+TEST_CONSTEXPR(_mm512_kor(0x12345, 0x12345) == 0x2345);
+TEST_CONSTEXPR(_mm512_kor(0xABCD, 0x0000) == 0xABCD);
+TEST_CONSTEXPR(_mm512_kor(0xABCD, 0xFFFF) == 0xFFFF);
+TEST_CONSTEXPR(_mm512_kor(0xAAAA, 0x5555) == 0xFFFF);
int test_mm512_kortestc(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
// CHECK-LABEL: test_mm512_kortestc
@@ -8638,11 +8638,11 @@ __mmask16 test_mm512_kxnor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
-TEST_CONSTEXPR(_mm512_kxnor(0x00C5, 0xFFAF) == 0x95); // data correctness
-TEST_CONSTEXPR(_mm512_kxnor(0x12345, 0xFFFF) == 0x2345); // should be truncated
-TEST_CONSTEXPR(_mm512_kxnor(0xABCD, 0x0000) == 0x5432); // all-zero, res = ~LHS
-TEST_CONSTEXPR(_mm512_kxnor(0x5678, 0xFFFF) == 0x5678); // all-one, res = LHS
-TEST_CONSTEXPR(_mm512_kxnor(0xAAAA, 0x5555) == 0x0000); // ~A xnor A == 0
+TEST_CONSTEXPR(_mm512_kxnor(0x00C5, 0xFFAF) == 0x95);
+TEST_CONSTEXPR(_mm512_kxnor(0x12345, 0xFFFF) == 0x2345);
+TEST_CONSTEXPR(_mm512_kxnor(0xABCD, 0x0000) == 0x5432);
+TEST_CONSTEXPR(_mm512_kxnor(0x5678, 0xFFFF) == 0x5678);
+TEST_CONSTEXPR(_mm512_kxnor(0xAAAA, 0x5555) == 0x0000);
__mmask16 test_mm512_kxor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_mm512_kxor
@@ -8674,11 +8674,11 @@ __mmask16 test_kand_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
-TEST_CONSTEXPR(_kand_mask16(0xCC, 0xAA) == 0x88); // data correctness
-TEST_CONSTEXPR(_kand_mask16(0x12345, 0xFFFF) == 0x2345); // should be truncated
-TEST_CONSTEXPR(_kand_mask16(0xABCD, 0x0000) == 0x0000); // all-zero
-TEST_CONSTEXPR(_kand_mask16(0x5678, 0xFFFF) == 0x5678); // all-one
-TEST_CONSTEXPR(_kand_mask16(0xAAAA, 0x5555) == 0x0000); // disjoint
+TEST_CONSTEXPR(_kand_mask16(0xCC, 0xAA) == 0x88);
+TEST_CONSTEXPR(_kand_mask16(0x12345, 0xFFFF) == 0x2345);
+TEST_CONSTEXPR(_kand_mask16(0xABCD, 0x0000) == 0x0000);
+TEST_CONSTEXPR(_kand_mask16(0x5678, 0xFFFF) == 0x5678);
+TEST_CONSTEXPR(_kand_mask16(0xAAAA, 0x5555) == 0x0000);
__mmask16 test_kandn_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kandn_mask16
@@ -8692,11 +8692,11 @@ __mmask16 test_kandn_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
-TEST_CONSTEXPR(_kandn_mask16(0xCC, 0xAA) == 0x22); // data correctness
-TEST_CONSTEXPR(_kandn_mask16(0x12345, 0xFFFF) == 0xDCBA); // truncated and inverted
-TEST_CONSTEXPR(_kandn_mask16(0x0000, 0xABCD) == 0xABCD); // all-ones (~0)
-TEST_CONSTEXPR(_kandn_mask16(0xFFFF, 0x5678) == 0x0000); // all-zero (~0xFFFF)
-TEST_CONSTEXPR(_kandn_mask16(0xAAAA, 0xAAAA) == 0x0000); // ~A & A is 0
+TEST_CONSTEXPR(_kandn_mask16(0xCC, 0xAA) == 0x22);
+TEST_CONSTEXPR(_kandn_mask16(0x12345, 0xFFFF) == 0xDCBA);
+TEST_CONSTEXPR(_kandn_mask16(0x0000, 0xABCD) == 0xABCD);
+TEST_CONSTEXPR(_kandn_mask16(0xFFFF, 0x5678) == 0x0000);
+TEST_CONSTEXPR(_kandn_mask16(0xAAAA, 0xAAAA) == 0x0000);
__mmask16 test_kor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kor_mask16
@@ -8709,11 +8709,11 @@ __mmask16 test_kor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
-TEST_CONSTEXPR(_kor_mask16(0xC1, 0xA8) == 0xE9); // data correctness
-TEST_CONSTEXPR(_kor_mask16(0x12345, 0x12345) == 0x2345); // truncated
-TEST_CONSTEXPR(_kor_mask16(0xABCD, 0x0000) == 0xABCD); // all-zero
-TEST_CONSTEXPR(_kor_mask16(0xABCD, 0xFFFF) == 0xFFFF); // all-ones
-TEST_CONSTEXPR(_kor_mask16(0xAAAA, 0x5555) == 0xFFFF); // disjoint
+TEST_CONSTEXPR(_kor_mask16(0xC1, 0xA8) == 0xE9);
+TEST_CONSTEXPR(_kor_mask16(0x12345, 0x12345) == 0x2345);
+TEST_CONSTEXPR(_kor_mask16(0xABCD, 0x0000) == 0xABCD);
+TEST_CONSTEXPR(_kor_mask16(0xABCD, 0xFFFF) == 0xFFFF);
+TEST_CONSTEXPR(_kor_mask16(0xAAAA, 0x5555) == 0xFFFF);
__mmask16 test_kxnor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxnor_mask16
@@ -8727,11 +8727,11 @@ __mmask16 test_kxnor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
__E, __F);
}
-TEST_CONSTEXPR(_kxnor_mask16(0x00C5, 0xFFAF) == 0x95); // data correctness
-TEST_CONSTEXPR(_kxnor_mask16(0x12345, 0xFFFF) == 0x2345); // should be truncated
-TEST_CONSTEXPR(_kxnor_mask16(0xABCD, 0x0000) == 0x5432); // all-zero, res = ~LHS
-TEST_CONSTEXPR(_kxnor_mask16(0x5678, 0xFFFF) == 0x5678); // all-one, res = LHS
-TEST_CONSTEXPR(_kxnor_mask16(0xAAAA, 0x5555) == 0x0000); // ~A xnor A == 0
+TEST_CONSTEXPR(_kxnor_mask16(0x00C5, 0xFFAF) == 0x95);
+TEST_CONSTEXPR(_kxnor_mask16(0x12345, 0xFFFF) == 0x2345);
+TEST_CONSTEXPR(_kxnor_mask16(0xABCD, 0x0000) == 0x5432);
+TEST_CONSTEXPR(_kxnor_mask16(0x5678, 0xFFFF) == 0x5678);
+TEST_CONSTEXPR(_kxnor_mask16(0xAAAA, 0x5555) == 0x0000);
__mmask16 test_kxor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxor_mask16
>From a64ac24281e484e984a8eb7b42cdfeaf3ad058c5 Mon Sep 17 00:00:00 2001
From: fennecJ <hwahwa649 at gmail.com>
Date: Tue, 23 Sep 2025 01:17:25 +0800
Subject: [PATCH 8/8] Allow kxor* to be used in constexpr
---
clang/include/clang/Basic/BuiltinsX86.td | 6 +++---
clang/lib/AST/ByteCode/InterpBuiltin.cpp | 8 ++++++++
clang/lib/AST/ExprConstant.cpp | 8 ++++++++
clang/lib/Headers/avx512bwintrin.h | 9 ++++-----
clang/lib/Headers/avx512dqintrin.h | 5 ++---
clang/lib/Headers/avx512fintrin.h | 5 ++---
clang/test/CodeGen/X86/avx512bw-builtins.c | 15 +++++++++++++--
clang/test/CodeGen/X86/avx512dq-builtins.c | 6 ++++++
clang/test/CodeGen/X86/avx512f-builtins.c | 12 ++++++++++++
9 files changed, 58 insertions(+), 16 deletions(-)
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index 0d4034922f3d5..ef676ca24549d 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -3170,15 +3170,15 @@ let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kxnordi : X86Builtin<"unsigned long long int(unsigned long long int, unsigned long long int)">;
}
-let Features = "avx512dq", Attributes = [NoThrow, Const] in {
+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {
def kxorqi : X86Builtin<"unsigned char(unsigned char, unsigned char)">;
}
-let Features = "avx512f", Attributes = [NoThrow, Const] in {
+let Features = "avx512f", Attributes = [NoThrow, Const, Constexpr] in {
def kxorhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;
}
-let Features = "avx512bw", Attributes = [NoThrow, Const] in {
+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
def kxorsi : X86Builtin<"unsigned int(unsigned int, unsigned int)">;
def kxordi : X86Builtin<"unsigned long long int(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 98bb8932a76b5..bd13ba75d051e 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -3620,6 +3620,14 @@ 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_kxorqi:
+ case X86::BI__builtin_ia32_kxorhi:
+ case X86::BI__builtin_ia32_kxorsi:
+ case X86::BI__builtin_ia32_kxordi:
+ return interp__builtin_elementwise_int_binop(
+ S, OpPC, Call,
+ [](const APSInt &LHS, const APSInt &RHS) { return LHS ^ RHS; });
+
case Builtin::BI__builtin_elementwise_fshl:
return interp__builtin_elementwise_triop(S, OpPC, Call,
llvm::APIntOps::fshl);
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 710b4c36499d1..58f992aec109c 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -14686,6 +14686,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
return HandleMaskBinOp(
[](const APSInt &LHS, const APSInt &RHS) { return ~(LHS ^ RHS); });
}
+
+ case X86::BI__builtin_ia32_kxorqi:
+ case X86::BI__builtin_ia32_kxorhi:
+ case X86::BI__builtin_ia32_kxorsi:
+ case X86::BI__builtin_ia32_kxordi: {
+ return HandleMaskBinOp(
+ [](const APSInt &LHS, const APSInt &RHS) { return LHS ^ RHS; });
+ }
}
}
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 0c470d0253c80..53d8593301140 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -82,14 +82,13 @@ _kxnor_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kxnordi((__mmask64)__A, (__mmask64)__B);
}
-static __inline__ __mmask32 __DEFAULT_FN_ATTRS
-_kxor_mask32(__mmask32 __A, __mmask32 __B)
-{
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR
+_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_CONSTEXPR
+_kxor_mask64(__mmask64 __A, __mmask64 __B) {
return (__mmask64)__builtin_ia32_kxordi((__mmask64)__A, (__mmask64)__B);
}
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
index bf7f620de1629..a2324a69007a7 100644
--- a/clang/lib/Headers/avx512dqintrin.h
+++ b/clang/lib/Headers/avx512dqintrin.h
@@ -55,9 +55,8 @@ _kxnor_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_kxnorqi((__mmask8)__A, (__mmask8)__B);
}
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS
-_kxor_mask8(__mmask8 __A, __mmask8 __B)
-{
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR
+_kxor_mask8(__mmask8 __A, __mmask8 __B) {
return (__mmask8)__builtin_ia32_kxorqi((__mmask8)__A, (__mmask8)__B);
}
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 4b27f20192edb..858d22e707214 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -8170,9 +8170,8 @@ _mm512_kxnor(__mmask16 __A, __mmask16 __B) {
return (__mmask16) __builtin_ia32_kxnorhi ((__mmask16) __A, (__mmask16) __B);
}
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
-_mm512_kxor (__mmask16 __A, __mmask16 __B)
-{
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
+_mm512_kxor(__mmask16 __A, __mmask16 __B) {
return (__mmask16) __builtin_ia32_kxorhi ((__mmask16) __A, (__mmask16) __B);
}
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
index 3128078bfef6d..fbfd170fe360c 100644
--- a/clang/test/CodeGen/X86/avx512bw-builtins.c
+++ b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -68,8 +68,7 @@ __mmask32 test_kandn_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D,
_mm512_cmpneq_epu16_mask(__C, __D)),
__E, __F);
}
-// 1100
-//
+
TEST_CONSTEXPR(_kandn_mask32(0xA0A0F0F0, 0xCCCCCCCC) == 0x4C4C0C0C);
TEST_CONSTEXPR(_kandn_mask32(0x123456789, 0xFFFFFFFF) == 0xDCBA9876);
TEST_CONSTEXPR(_kandn_mask32(0x00000000, 0x1234ABCD) == 0x1234ABCD);
@@ -169,6 +168,12 @@ __mmask32 test_kxor_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_kxor_mask32(0x1234ABCD, 0xFFFF0000) == 0xEDCBABCD);
+TEST_CONSTEXPR(_kxor_mask32(0x123456789ABCDEF0, 0x00000000) == 0x9ABCDEF0);
+TEST_CONSTEXPR(_kxor_mask32(0xAABBCCDD, 0x00000000) == 0xAABBCCDD);
+TEST_CONSTEXPR(_kxor_mask32(0x87654321, 0xFFFFFFFF) == 0x789ABCDE);
+TEST_CONSTEXPR(_kxor_mask32(0xAAAAAAAA, 0x55555555) == 0xFFFFFFFF);
+
__mmask64 test_kxor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
// CHECK-LABEL: test_kxor_mask64
// CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
@@ -179,6 +184,12 @@ __mmask64 test_kxor_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_kxor_mask64(0x0123456789ABCDEF, 0xFFFFFFFF00000000) == 0xFEDCBA9889ABCDEF);
+TEST_CONSTEXPR(_kxor_mask64(0xF0F0F0F0F0F0F0F0, 0x0F0F0F0F0F0F0F0F) == 0xFFFFFFFFFFFFFFFF);
+TEST_CONSTEXPR(_kxor_mask64(0xFEDCBA9876543210, 0xFFFFFFFFFFFFFFFF) == 0x0123456789ABCDEF);
+TEST_CONSTEXPR(_kxor_mask64(0xAABBCCDD11223344, 0x0000000000000000) == 0xAABBCCDD11223344);
+TEST_CONSTEXPR(_kxor_mask64(0xAAAAAAAAAAAAAAAA, 0x5555555555555555) == 0xFFFFFFFFFFFFFFFF);
+
unsigned char test_kortestz_mask32_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
// CHECK-LABEL: test_kortestz_mask32_u8
// CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
index fa4fd9ddee42e..ca60ee09245b7 100644
--- a/clang/test/CodeGen/X86/avx512dq-builtins.c
+++ b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -95,6 +95,12 @@ __mmask8 test_kxor_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m
__E, __F);
}
+TEST_CONSTEXPR(_kxor_mask8(0xC5, 0xAF) == 0x6A);
+TEST_CONSTEXPR(_kxor_mask8(0x1234, 0xFFFF) == 0xCB);
+TEST_CONSTEXPR(_kxor_mask8(0xCD, 0x00) == 0xCD);
+TEST_CONSTEXPR(_kxor_mask8(0x78, 0xFF) == 0x87);
+TEST_CONSTEXPR(_kxor_mask8(0xAA, 0x55) == 0xFF);
+
unsigned char test_kortestz_mask8_u8(__m512i __A, __m512i __B, __m512i __C, __m512i __D) {
// CHECK-LABEL: test_kortestz_mask8_u8
// CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index 5e5b335e755a5..d1dcf7ceee587 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -8655,6 +8655,12 @@ __mmask16 test_mm512_kxor(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __
__E, __F);
}
+TEST_CONSTEXPR(_mm512_kxor(0xC5, 0xAF) == 0x6A);
+TEST_CONSTEXPR(_mm512_kxor(0x12345, 0xFFFF) == 0xDCBA);
+TEST_CONSTEXPR(_mm512_kxor(0xABCD, 0x0000) == 0xABCD);
+TEST_CONSTEXPR(_mm512_kxor(0x5678, 0xFFFF) == 0xA987);
+TEST_CONSTEXPR(_mm512_kxor(0xAAAA, 0x5555) == 0xFFFF);
+
__mmask16 test_knot_mask16(__mmask16 a) {
// CHECK-LABEL: test_knot_mask16
// CHECK: [[IN:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
@@ -8744,6 +8750,12 @@ __mmask16 test_kxor_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
__E, __F);
}
+TEST_CONSTEXPR(_kxor_mask16(0xC5, 0xAF) == 0x6A);
+TEST_CONSTEXPR(_kxor_mask16(0x12345, 0xFFFF) == 0xDCBA);
+TEST_CONSTEXPR(_kxor_mask16(0xABCD, 0x0000) == 0xABCD);
+TEST_CONSTEXPR(_kxor_mask16(0x5678, 0xFFFF) == 0xA987);
+TEST_CONSTEXPR(_kxor_mask16(0xAAAA, 0x5555) == 0xFFFF);
+
__mmask16 test_kshiftli_mask16(__m512i A, __m512i B, __m512i C, __m512i D) {
// CHECK-LABEL: test_kshiftli_mask16
// CHECK: [[VAL:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
More information about the cfe-commits
mailing list