Allow kadd* to be used in constexpr

fennecJ · fennecJ · commit ff71edf11751 · 2025-09-23T15:43:04.000+08:00
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
@@ -3073,12 +3073,12 @@ let Features = "avx512dq", Attributes = [NoThrow, Const, RequiredVectorWidth<128
   def fpclassss_mask : X86Builtin<"unsigned char(_Vector<4, float>, _Constant int, unsigned char)">;
 }
 
-let Features = "avx512dq", Attributes = [NoThrow, Const] in {
+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {
   def kaddqi : X86Builtin<"unsigned char(unsigned char, unsigned char)">;
   def kaddhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;
 }
 
-let Features = "avx512bw", Attributes = [NoThrow, Const] in {
+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {
   def kaddsi : X86Builtin<"unsigned int(unsigned int, unsigned int)">;
   def kadddi : 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
@@ -3644,6 +3644,14 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
   case X86::BI__builtin_ia32_knotdi:
     return interp__builtin_knot(S, OpPC, Frame, Call);
 
+  case X86::BI__builtin_ia32_kaddqi:
+  case X86::BI__builtin_ia32_kaddhi:
+  case X86::BI__builtin_ia32_kaddsi:
+  case X86::BI__builtin_ia32_kadddi:
+    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
@@ -14705,6 +14705,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,
     APSInt Result = ~Val;
     return Success(APValue(Result), E);
   }
+
+  case X86::BI__builtin_ia32_kaddqi:
+  case X86::BI__builtin_ia32_kaddhi:
+  case X86::BI__builtin_ia32_kaddsi:
+  case X86::BI__builtin_ia32_kadddi: {
+    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
@@ -160,14 +160,13 @@ _ktest_mask64_u8(__mmask64 __A, __mmask64 __B, unsigned char *__C) {
   return (unsigned char)__builtin_ia32_ktestzdi(__A, __B);
 }
 
-static __inline__ __mmask32 __DEFAULT_FN_ATTRS
-_kadd_mask32(__mmask32 __A, __mmask32 __B)
-{
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR
+_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_CONSTEXPR
+_kadd_mask64(__mmask64 __A, __mmask64 __B) {
   return (__mmask64)__builtin_ia32_kadddi((__mmask64)__A, (__mmask64)__B);
 }
 
diff --git a/clang/lib/Headers/avx512dqintrin.h b/clang/lib/Headers/avx512dqintrin.h
@@ -113,15 +113,13 @@ _ktest_mask16_u8(__mmask16 __A, __mmask16 __B, unsigned char *__C) {
   return (unsigned char)__builtin_ia32_ktestzhi(__A, __B);
 }
 
-static __inline__ __mmask8 __DEFAULT_FN_ATTRS
-_kadd_mask8(__mmask8 __A, __mmask8 __B)
-{
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR
+_kadd_mask8(__mmask8 __A, __mmask8 __B) {
   return (__mmask8)__builtin_ia32_kaddqi((__mmask8)__A, (__mmask8)__B);
 }
 
-static __inline__ __mmask16 __DEFAULT_FN_ATTRS
-_kadd_mask16(__mmask16 __A, __mmask16 __B)
-{
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR
+_kadd_mask16(__mmask16 __A, __mmask16 __B) {
   return (__mmask16)__builtin_ia32_kaddhi((__mmask16)__A, (__mmask16)__B);
 }
 
diff --git a/clang/test/CodeGen/X86/avx512bw-builtins.c b/clang/test/CodeGen/X86/avx512bw-builtins.c
@@ -366,6 +366,12 @@ __mmask32 test_kadd_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
                                                     __E, __F);
 }
 
+TEST_CONSTEXPR(_kadd_mask32(100000, 200000) == 300000);
+TEST_CONSTEXPR(_kadd_mask32(2147483648, 0) == 2147483648);
+TEST_CONSTEXPR(_kadd_mask32(0xFFFFFFFF, 1) == 0);
+TEST_CONSTEXPR(_kadd_mask32(0xEE6B2800, 0x1DCD6500) == 0x0C388D00);
+TEST_CONSTEXPR(_kadd_mask32(0xFFFFFFFA, 10) == 4);
+
 __mmask64 test_kadd_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
   // CHECK-LABEL: test_kadd_mask64
   // CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
@@ -376,6 +382,12 @@ __mmask64 test_kadd_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
                                                    __E, __F);
 }
 
+TEST_CONSTEXPR(_kadd_mask64(10000000000, 20000000000) == 30000000000);
+TEST_CONSTEXPR(_kadd_mask64(0x8000000000000000, 0) == 0x8000000000000000);
+TEST_CONSTEXPR(_kadd_mask64(0xFFFFFFFFFFFFFFFF, 1) == 0);
+TEST_CONSTEXPR(_kadd_mask64(0xFFFFFFFFFFFFFFFA, 10) == 4);
+TEST_CONSTEXPR(_kadd_mask64(0xFA0A1F2C6C729C00, 0x0DE0B6B3A7640000) == 0x07EAD5E013D69C00);
+
 __mmask32 test_kshiftli_mask32(__m512i A, __m512i B, __m512i C, __m512i D) {
   // CHECK-LABEL: test_kshiftli_mask32
   // CHECK: [[VAL:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
diff --git a/clang/test/CodeGen/X86/avx512dq-builtins.c b/clang/test/CodeGen/X86/avx512dq-builtins.c
@@ -228,6 +228,12 @@ __mmask8 test_kadd_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m
                                                    __E, __F);
 }
 
+TEST_CONSTEXPR(_kadd_mask8(20, 30) == 50);
+TEST_CONSTEXPR(_kadd_mask8(128, 0) == 128);
+TEST_CONSTEXPR(_kadd_mask8(0xFF, 1) == 0);
+TEST_CONSTEXPR(_kadd_mask8(0xC8, 0x64) == 0x2C);
+TEST_CONSTEXPR(_kadd_mask8(0xFA, 0x0F) == 0x09);
+
 __mmask16 test_kadd_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
   // CHECK-LABEL: test_kadd_mask16
   // CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
@@ -238,6 +244,12 @@ __mmask16 test_kadd_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, _
                                                     __E, __F);
 }
 
+TEST_CONSTEXPR(_kadd_mask16(1000, 2000) == 3000);
+TEST_CONSTEXPR(_kadd_mask16(32768, 0) == 32768);
+TEST_CONSTEXPR(_kadd_mask16(0xFFFF, 1) == 0);
+TEST_CONSTEXPR(_kadd_mask16(0xEA60, 0x2710) == 0x1170);
+TEST_CONSTEXPR(_kadd_mask16(0xFFFA, 0x14) == 0x0E);
+
 __mmask8 test_kshiftli_mask8(__m512i A, __m512i B, __m512i C, __m512i D) {
   // CHECK-LABEL: test_kshiftli_mask8
   // CHECK: [[VAL:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -11106,7 +11106,7 @@ int test_mm512_mask2int(__mmask16 __a)
   return _mm512_mask2int(__a);
 }
 
-TEST_CONSTEXPR(_mm512_mask2int(0x8000) == 0x00008000); // Zero-extended
+TEST_CONSTEXPR(_mm512_mask2int(0x8000) == 0x00008000);
 
 __m128 test_mm_mask_move_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)
 {

Original file line number	Diff line number	Diff line change
`@@ -3073,12 +3073,12 @@ let Features = "avx512dq", Attributes = [NoThrow, Const, RequiredVectorWidth<128`
`3073`	`3073`	`def fpclassss_mask : X86Builtin<"unsigned char(_Vector<4, float>, _Constant int, unsigned char)">;`
`3074`	`3074`	`}`
`3075`	`3075`
`3076`		`-let Features = "avx512dq", Attributes = [NoThrow, Const] in {`
	`3076`	`+let Features = "avx512dq", Attributes = [NoThrow, Const, Constexpr] in {`
`3077`	`3077`	`def kaddqi : X86Builtin<"unsigned char(unsigned char, unsigned char)">;`
`3078`	`3078`	`def kaddhi : X86Builtin<"unsigned short(unsigned short, unsigned short)">;`
`3079`	`3079`	`}`
`3080`	`3080`
`3081`		`-let Features = "avx512bw", Attributes = [NoThrow, Const] in {`
	`3081`	`+let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in {`
`3082`	`3082`	`def kaddsi : X86Builtin<"unsigned int(unsigned int, unsigned int)">;`
`3083`	`3083`	`def kadddi : X86Builtin<"unsigned long long int(unsigned long long int, unsigned long long int)">;`
`3084`	`3084`	`}`
Original file line number	Diff line number	Diff line change
`@@ -14705,6 +14705,14 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E,`
`14705`	`14705`	`APSInt Result = ~Val;`
`14706`	`14706`	`return Success(APValue(Result), E);`
`14707`	`14707`	`}`
	`14708`	`+`
	`14709`	`+ case X86::BI__builtin_ia32_kaddqi:`
	`14710`	`+ case X86::BI__builtin_ia32_kaddhi:`
	`14711`	`+ case X86::BI__builtin_ia32_kaddsi:`
	`14712`	`+ case X86::BI__builtin_ia32_kadddi: {`
	`14713`	`+ return HandleMaskBinOp(`
	`14714`	`+ [](const APSInt &LHS, const APSInt &RHS) { return LHS + RHS; });`
	`14715`	`+ }`
`14708`	`14716`	`}`
`14709`	`14717`	`}`
`14710`	`14718`
Original file line number	Diff line number	Diff line change
`@@ -160,14 +160,13 @@ _ktest_mask64_u8(__mmask64 __A, __mmask64 __B, unsigned char *__C) {`
`160`	`160`	`return (unsigned char)__builtin_ia32_ktestzdi(__A, __B);`
`161`	`161`	`}`
`162`	`162`
`163`		`-static __inline__ __mmask32 __DEFAULT_FN_ATTRS`
`164`		`-_kadd_mask32(__mmask32 __A, __mmask32 __B)`
`165`		`-{`
	`163`	`+static __inline__ __mmask32 __DEFAULT_FN_ATTRS_CONSTEXPR`
	`164`	`+_kadd_mask32(__mmask32 __A, __mmask32 __B) {`
`166`	`165`	`return (__mmask32)__builtin_ia32_kaddsi((__mmask32)__A, (__mmask32)__B);`
`167`	`166`	`}`
`168`	`167`
`169`		`-static __inline__ __mmask64 __DEFAULT_FN_ATTRS _kadd_mask64(__mmask64 __A,`
`170`		`- __mmask64 __B) {`
	`168`	`+static __inline__ __mmask64 __DEFAULT_FN_ATTRS_CONSTEXPR`
	`169`	`+_kadd_mask64(__mmask64 __A, __mmask64 __B) {`
`171`	`170`	`return (__mmask64)__builtin_ia32_kadddi((__mmask64)__A, (__mmask64)__B);`
`172`	`171`	`}`
`173`	`172`
Original file line number	Diff line number	Diff line change
`@@ -113,15 +113,13 @@ _ktest_mask16_u8(__mmask16 __A, __mmask16 __B, unsigned char *__C) {`
`113`	`113`	`return (unsigned char)__builtin_ia32_ktestzhi(__A, __B);`
`114`	`114`	`}`
`115`	`115`
`116`		`-static __inline__ __mmask8 __DEFAULT_FN_ATTRS`
`117`		`-_kadd_mask8(__mmask8 __A, __mmask8 __B)`
`118`		`-{`
	`116`	`+static __inline__ __mmask8 __DEFAULT_FN_ATTRS_CONSTEXPR`
	`117`	`+_kadd_mask8(__mmask8 __A, __mmask8 __B) {`
`119`	`118`	`return (__mmask8)__builtin_ia32_kaddqi((__mmask8)__A, (__mmask8)__B);`
`120`	`119`	`}`
`121`	`120`
`122`		`-static __inline__ __mmask16 __DEFAULT_FN_ATTRS`
`123`		`-_kadd_mask16(__mmask16 __A, __mmask16 __B)`
`124`		`-{`
	`121`	`+static __inline__ __mmask16 __DEFAULT_FN_ATTRS_CONSTEXPR`
	`122`	`+_kadd_mask16(__mmask16 __A, __mmask16 __B) {`
`125`	`123`	`return (__mmask16)__builtin_ia32_kaddhi((__mmask16)__A, (__mmask16)__B);`
`126`	`124`	`}`
`127`	`125`
Original file line number	Diff line number	Diff line change
`@@ -11106,7 +11106,7 @@ int test_mm512_mask2int(__mmask16 __a)`
`11106`	`11106`	`return _mm512_mask2int(__a);`
`11107`	`11107`	`}`
`11108`	`11108`
`11109`		`-TEST_CONSTEXPR(_mm512_mask2int(0x8000) == 0x00008000); // Zero-extended`
	`11109`	`+TEST_CONSTEXPR(_mm512_mask2int(0x8000) == 0x00008000);`
`11110`	`11110`
`11111`	`11111`	`__m128 test_mm_mask_move_ss (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B)`
`11112`	`11112`	`{`