[X86][Clang] Add AVX512 Integer Comparison Intrinsics for constexpr Evaluation #164026
+3,314 −3,485
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
…add AVX512 integer comparison intrinsics to be used in constexpr
llvmbot added clang 
| @llvm/pr-subscribers-clang Author: None (sskzakaria) ChangesEnables constexpr evaluation for the following AVX512 Integer Comparison Intrinsics: Part 1 Full diff: https://github.com/llvm/llvm-project/pull/164026.diff 5 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td index 62c70fba946be..a2a1746fd1cae 100644 --- a/clang/include/clang/Basic/BuiltinsX86.td +++ b/clang/include/clang/Basic/BuiltinsX86.td @@ -1272,81 +1272,99 @@ let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in { def knotdi : X86Builtin<"unsigned long long int(unsigned long long int)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def cmpb128_mask : X86Builtin<"unsigned short(_Vector<16, char>, _Vector<16, char>, _Constant int, unsigned short)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def cmpd128_mask : X86Builtin<"unsigned char(_Vector<4, int>, _Vector<4, int>, _Constant int, unsigned char)">; def cmpq128_mask : X86Builtin<"unsigned char(_Vector<2, long long int>, _Vector<2, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def cmpw128_mask : X86Builtin<"unsigned char(_Vector<8, short>, _Vector<8, short>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def cmpb256_mask : X86Builtin<"unsigned int(_Vector<32, char>, _Vector<32, char>, _Constant int, unsigned int)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def cmpd256_mask : X86Builtin<"unsigned char(_Vector<8, int>, _Vector<8, int>, _Constant int, unsigned char)">; def cmpq256_mask : X86Builtin<"unsigned char(_Vector<4, long long int>, _Vector<4, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def cmpw256_mask : X86Builtin<"unsigned short(_Vector<16, short>, _Vector<16, short>, _Constant int, unsigned short)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def cmpb512_mask : X86Builtin<"unsigned long long int(_Vector<64, char>, _Vector<64, char>, _Constant int, unsigned long long int)">; } -let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512f", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def cmpd512_mask : X86Builtin<"unsigned short(_Vector<16, int>, _Vector<16, int>, _Constant int, unsigned short)">; def cmpq512_mask : X86Builtin<"unsigned char(_Vector<8, long long int>, _Vector<8, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def cmpw512_mask : X86Builtin<"unsigned int(_Vector<32, short>, _Vector<32, short>, _Constant int, unsigned int)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def ucmpb128_mask : X86Builtin<"unsigned short(_Vector<16, char>, _Vector<16, char>, _Constant int, unsigned short)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def ucmpd128_mask : X86Builtin<"unsigned char(_Vector<4, int>, _Vector<4, int>, _Constant int, unsigned char)">; def ucmpq128_mask : X86Builtin<"unsigned char(_Vector<2, long long int>, _Vector<2, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def ucmpw128_mask : X86Builtin<"unsigned char(_Vector<8, short>, _Vector<8, short>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def ucmpb256_mask : X86Builtin<"unsigned int(_Vector<32, char>, _Vector<32, char>, _Constant int, unsigned int)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def ucmpd256_mask : X86Builtin<"unsigned char(_Vector<8, int>, _Vector<8, int>, _Constant int, unsigned char)">; def ucmpq256_mask : X86Builtin<"unsigned char(_Vector<4, long long int>, _Vector<4, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def ucmpw256_mask : X86Builtin<"unsigned short(_Vector<16, short>, _Vector<16, short>, _Constant int, unsigned short)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def ucmpb512_mask : X86Builtin<"unsigned long long int(_Vector<64, char>, _Vector<64, char>, _Constant int, unsigned long long int)">; } -let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512f", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def ucmpd512_mask : X86Builtin<"unsigned short(_Vector<16, int>, _Vector<16, int>, _Constant int, unsigned short)">; def ucmpq512_mask : X86Builtin<"unsigned char(_Vector<8, long long int>, _Vector<8, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def ucmpw512_mask : X86Builtin<"unsigned int(_Vector<32, short>, _Vector<32, short>, _Constant int, unsigned int)">; } diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp index 0ee18be166845..2526c13b529a7 100644 --- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp +++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp @@ -3101,6 +3101,62 @@ static bool interp__builtin_vec_set(InterpState &S, CodePtr OpPC, return true; } +static bool interp__builtin_cmp_mask(InterpState &S, CodePtr OpPC, + const CallExpr *Call, unsigned ID, + bool IsUnsigned) { + assert(Call->getNumArgs() == 4); + + APSInt Mask = popToAPSInt(S, Call->getArg(3)); + APSInt Opcode = popToAPSInt(S, Call->getArg(2)); + const Pointer &LHS = S.Stk.pop<Pointer>(); + const Pointer &RHS = S.Stk.pop<Pointer>(); + + assert(LHS.getNumElems() == RHS.getNumElems()); + + APInt RetMask = APInt::getZero(LHS.getNumElems()); + unsigned VectorLen = LHS.getNumElems(); + PrimType ElemT = LHS.getFieldDesc()->getPrimType(); + + INT_TYPE_SWITCH_NO_BOOL(ElemT, { + for (unsigned ElemNum = 0; ElemNum < VectorLen; ++ElemNum) { + APSInt A = LHS.elem<T>(ElemNum).toAPSInt(); + APSInt B = RHS.elem<T>(ElemNum).toAPSInt(); + bool result = false; + switch (Opcode.getExtValue() & 0x7) { + case 0x00: // _MM_CMPINT_EQ + result = (A == B); + break; + case 0x01: // _MM_CMPINT_LT + result = IsUnsigned ? A.ult(B) : A.slt(B); + break; + case 0x02: // _MM_CMPINT_LE + result = IsUnsigned ? A.ule(B) : A.sle(B); + break; + case 0x03: // _MM_CMPINT_FALSE + result = false; + break; + case 0x04: // _MM_CMPINT_NE + result = (A != B); + break; + case 0x05: // _MM_CMPINT_NLT (>=) + result = IsUnsigned ? A.uge(B) : A.sge(B); + break; + case 0x06: // _MM_CMPINT_NLE (>) + result = IsUnsigned ? A.ugt(B) : A.sgt(B); + break; + case 0x07: // _MM_CMPINT_TRUE + result = true; + break; + } + + RetMask.setBitVal(ElemNum, Mask[ElemNum] && result); + } + }); + + pushInteger(S, RetMask, Call->getType()); + return true; +} + static bool interp__builtin_ia32_vpconflict(InterpState &S, CodePtr OpPC, const CallExpr *Call) { assert(Call->getNumArgs() == 1); @@ -4141,6 +4197,36 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call, case X86::BI__builtin_ia32_vec_set_v4di: return interp__builtin_vec_set(S, OpPC, Call, BuiltinID); + case X86::BI__builtin_ia32_cmpb128_mask: + case X86::BI__builtin_ia32_cmpw128_mask: + case X86::BI__builtin_ia32_cmpd128_mask: + case X86::BI__builtin_ia32_cmpq128_mask: + case X86::BI__builtin_ia32_cmpb256_mask: + case X86::BI__builtin_ia32_cmpw256_mask: + case X86::BI__builtin_ia32_cmpd256_mask: + case X86::BI__builtin_ia32_cmpq256_mask: + case X86::BI__builtin_ia32_cmpb512_mask: + case X86::BI__builtin_ia32_cmpw512_mask: + case X86::BI__builtin_ia32_cmpd512_mask: + case X86::BI__builtin_ia32_cmpq512_mask: + return interp__builtin_cmp_mask(S, OpPC, Call, BuiltinID, + /*IsUnsigned*/ false); + + case X86::BI__builtin_ia32_ucmpb128_mask: + case X86::BI__builtin_ia32_ucmpw128_mask: + case X86::BI__builtin_ia32_ucmpd128_mask: + case X86::BI__builtin_ia32_ucmpq128_mask: + case X86::BI__builtin_ia32_ucmpb256_mask: + case X86::BI__builtin_ia32_ucmpw256_mask: + case X86::BI__builtin_ia32_ucmpd256_mask: + case X86::BI__builtin_ia32_ucmpq256_mask: + case X86::BI__builtin_ia32_ucmpb512_mask: + case X86::BI__builtin_ia32_ucmpw512_mask: + case X86::BI__builtin_ia32_ucmpd512_mask: + case X86::BI__builtin_ia32_ucmpq512_mask: + return interp__builtin_cmp_mask(S, OpPC, Call, BuiltinID, + /*IsUnsigned*/ true); + default: S.FFDiag(S.Current->getLocation(OpPC), diag::note_invalid_subexpr_in_const_expr) diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp index 16141b27f4ce8..e1e717ebe210b 100644 --- a/clang/lib/AST/ExprConstant.cpp +++ b/clang/lib/AST/ExprConstant.cpp @@ -15449,6 +15449,88 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E, unsigned Idx = static_cast<unsigned>(IdxAPS.getZExtValue() & (N - 1)); return Success(Vec.getVectorElt(Idx).getInt(), E); } + + case clang::X86::BI__builtin_ia32_cmpb128_mask: + case clang::X86::BI__builtin_ia32_cmpw128_mask: + case clang::X86::BI__builtin_ia32_cmpd128_mask: + case clang::X86::BI__builtin_ia32_cmpq128_mask: + case clang::X86::BI__builtin_ia32_cmpb256_mask: + case clang::X86::BI__builtin_ia32_cmpw256_mask: + case clang::X86::BI__builtin_ia32_cmpd256_mask: + case clang::X86::BI__builtin_ia32_cmpq256_mask: + case clang::X86::BI__builtin_ia32_cmpb512_mask: + case clang::X86::BI__builtin_ia32_cmpw512_mask: + case clang::X86::BI__builtin_ia32_cmpd512_mask: + case clang::X86::BI__builtin_ia32_cmpq512_mask: + case clang::X86::BI__builtin_ia32_ucmpb128_mask: + case clang::X86::BI__builtin_ia32_ucmpw128_mask: + case clang::X86::BI__builtin_ia32_ucmpd128_mask: + case clang::X86::BI__builtin_ia32_ucmpq128_mask: + case clang::X86::BI__builtin_ia32_ucmpb256_mask: + case clang::X86::BI__builtin_ia32_ucmpw256_mask: + case clang::X86::BI__builtin_ia32_ucmpd256_mask: + case clang::X86::BI__builtin_ia32_ucmpq256_mask: + case clang::X86::BI__builtin_ia32_ucmpb512_mask: + case clang::X86::BI__builtin_ia32_ucmpw512_mask: + case clang::X86::BI__builtin_ia32_ucmpd512_mask: + case clang::X86::BI__builtin_ia32_ucmpq512_mask: { + assert(E->getNumArgs() == 4); + + bool IsUnsigned = + (BuiltinOp >= clang::X86::BI__builtin_ia32_ucmpb128_mask && + BuiltinOp <= clang::X86::BI__builtin_ia32_ucmpq512_mask); + + APValue LHS, RHS; + APSInt Mask, Opcode; + if (!EvaluateVector(E->getArg(0), LHS, Info) || + !EvaluateVector(E->getArg(1), RHS, Info) || + !EvaluateInteger(E->getArg(2), Opcode, Info) || + !EvaluateInteger(E->getArg(3), Mask, Info)) + return false; + + assert(LHS.getVectorLength() == RHS.getVectorLength()); + + APSInt RetMask = APSInt::getUnsigned(0); + unsigned VectorLen = LHS.getVectorLength(); + + for (unsigned ElemNum = 0; ElemNum < VectorLen; ++ElemNum) { + APSInt A = LHS.getVectorElt(ElemNum).getInt(); + APSInt B = RHS.getVectorElt(ElemNum).getInt(); + bool result = false; + + switch (Opcode.getExtValue() & 0x7) { + case 0: // _MM_CMPINT_EQ + result = (A == B); + break; + case 1: // _MM_CMPINT_LT + result = IsUnsigned ? A.ult(B) : A.slt(B); + break; + case 2: // _MM_CMPINT_LE + result = IsUnsigned ? A.ule(B) : A.sle(B); + break; + case 3: // _MM_CMPINT_FALSE + result = false; + break; + case 4: // _MM_CMPINT_NE + result = (A != B); + break; + case 5: // _MM_CMPINT_NLT (>=) + result = IsUnsigned ? A.uge(B) : A.sge(B); + break; + case 6: // _MM_CMPINT_NLE (>) + result = IsUnsigned ? A.ugt(B) : A.sgt(B); + break; + case 7: // _MM_CMPINT_TRUE + result = true; + break; + } + + RetMask.setBitVal(ElemNum, Mask[ElemNum] && result); + } + + RetMask.setIsUnsigned(true); + return Success(APValue(RetMask), E); + } } } diff --git a/clang/lib/Headers/avx512vlbwintrin.h b/clang/lib/Headers/avx512vlbwintrin.h index 639fb60f476c6..97e48357f3ccc 100644 --- a/clang/lib/Headers/avx512vlbwintrin.h +++ b/clang/lib/Headers/avx512vlbwintrin.h @@ -2385,22 +2385,19 @@ _mm256_mask_storeu_epi8 (void *__P, __mmask32 __U, __m256i __A) (__mmask32) __U); } -static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 -_mm_test_epi8_mask (__m128i __A, __m128i __B) -{ +static __inline__ __mmask16 __DEFAULT_FN_ATTRS128_CONSTEXPR +_mm_test_epi8_mask(__m128i __A, __m128i __B) { return _mm_cmpneq_epi8_mask (_mm_and_si128(__A, __B), _mm_setzero_si128()); } -static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 -_mm_mask_test_epi8_mask (__mmask16 __U, __m128i __A, __m128i __B) -{ +static __inline__ __mmask16 __DEFAULT_FN_ATTRS128_CONSTEXPR +_mm_mask_test_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { return _mm_mask_cmpneq_epi8_mask (__U, _mm_and_si128 (__A, __B), _mm_setzero_si128()); } -static __inline__ __mmask32 __DEFAULT_FN_ATTRS256 -_mm256_test_epi8_mask (__m256i __A, __m256i __B) -{ +static __inline__ __mmask32 __DEFAULT_FN_ATTRS256_CONSTEXPR +_mm256_test_epi8_mask(__m256i __A, __m256i __B) { return _mm256_cmpneq_epi8_mask (_mm256_and_si256(__A, __B), _mm256_setzero_si256()); } @@ -2439,9 +2436,8 @@ _mm256_mask_test_epi16_mask (__mmask16 __U, __m256i __A, __m256i __B) _mm256_setzero_si256()); } -static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 -_mm_testn_epi8_mask (__m128i __A, __m128i __B) -{ +static __inline__ __mmask16 __DEFAULT_FN_ATTRS128_CONSTEXPR +_mm_testn_epi8_mask(__m128i __A, __m128i __B) { return _mm_cmpeq_epi8_mask (_mm_and_si128 (__A, __B), _mm_setzero_si128()); } diff --git a/clang/test/CodeGen/X86/avx512vlbw-builtins.c b/clang/test/CodeGen/X86/avx512vlbw-builtins.c index d569283928a0a..1cc4518484c19 100644 --- a/clang/test/CodeGen/X86/avx512vlbw-builtins.c +++ b/clang/test/CodeGen/X86/avx512vlbw-builtins.c @@ -2890,6 +2890,12 @@ __mmask16 test_mm_test_epi8_mask(__m128i __A, __m128i __B) { return _mm_test_epi8_mask(__A, __B); } +TEST_CONSTEXPR(_mm_test_epi8_mask( + (__m128i)(__v16qi){1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}, + (__m128i)(__v16qi){1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} +) +== (__mmask16)0xfffb); + __mmask16 test_mm_mask_test_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { // CHECK-LABEL: test_mm_mask_test_epi8_mask // CHECK: and <2 x i64> %{{.*}}, %{{.*}} @@ -2897,6 +2903,12 @@ __mmask16 test_mm_mask_test_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { // CHECK: and <16 x i1> %{{.*}}, %{{.*}} return _mm_mask_test_epi8_mask(__U, __A, __B); } +TEST_CONSTEXPR(_mm_mask_test_epi8_mask( + 0xFFFF, + (__m128i)(__v16qi){1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}, + (__m128i)(__v16qi){1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} +) +== (__mmask16)0xfffb); __mmask32 test_mm256_test_epi8_mask(__m256i __A, __m256i __B) { // CHECK-LABEL: test_mm256_test_epi8_mask @@ -2904,6 +2916,11 @@ __mmask32 test_mm256_test_epi8_mask(__m256i __A, __m256i __B) { // CHECK: icmp ne <32 x i8> %{{.*}}, %{{.*}} return _mm256_test_epi8_mask(__A, __B); } +TEST_CONSTEXPR(_mm256_test_epi8_mask( + (__m256i)(__v32qi){1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}, + (__m256i)(__v32qi){1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} +) +== (__mmask32)0xfffbfffb); __mmask32 test_mm256_mask_test_epi8_mask(__mmask32 __U, __m256i __A, __m256i __B) { // CHECK-LABEL: test_mm256_mask_test_epi8_mask @@ -2950,6 +2967,12 @@ __mmask16 test_mm_testn_epi8_mask(__m128i __A, __m128i __B) { return _mm_testn_epi8_mask(__A, __B); } +TEST_CONSTEXPR(_mm_testn_epi8_mask( + (__m128i)(__v16qi){1, 2, 77, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 1, 16, 16}, + (__m128i)(__v16qi){2, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15} +) +== (__mmask16)0xe001); + __mmask16 test_mm_mask_testn_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { // CHECK-LABEL: test_mm_mask_testn_epi8_mask // CHECK: and <2 x i64> %{{.*}}, %{{.*}} |
| @llvm/pr-subscribers-backend-x86 Author: None (sskzakaria) ChangesEnables constexpr evaluation for the following AVX512 Integer Comparison Intrinsics: Part 1 Full diff: https://github.com/llvm/llvm-project/pull/164026.diff 5 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td index 62c70fba946be..a2a1746fd1cae 100644 --- a/clang/include/clang/Basic/BuiltinsX86.td +++ b/clang/include/clang/Basic/BuiltinsX86.td @@ -1272,81 +1272,99 @@ let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in { def knotdi : X86Builtin<"unsigned long long int(unsigned long long int)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def cmpb128_mask : X86Builtin<"unsigned short(_Vector<16, char>, _Vector<16, char>, _Constant int, unsigned short)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def cmpd128_mask : X86Builtin<"unsigned char(_Vector<4, int>, _Vector<4, int>, _Constant int, unsigned char)">; def cmpq128_mask : X86Builtin<"unsigned char(_Vector<2, long long int>, _Vector<2, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def cmpw128_mask : X86Builtin<"unsigned char(_Vector<8, short>, _Vector<8, short>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def cmpb256_mask : X86Builtin<"unsigned int(_Vector<32, char>, _Vector<32, char>, _Constant int, unsigned int)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def cmpd256_mask : X86Builtin<"unsigned char(_Vector<8, int>, _Vector<8, int>, _Constant int, unsigned char)">; def cmpq256_mask : X86Builtin<"unsigned char(_Vector<4, long long int>, _Vector<4, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def cmpw256_mask : X86Builtin<"unsigned short(_Vector<16, short>, _Vector<16, short>, _Constant int, unsigned short)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def cmpb512_mask : X86Builtin<"unsigned long long int(_Vector<64, char>, _Vector<64, char>, _Constant int, unsigned long long int)">; } -let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512f", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def cmpd512_mask : X86Builtin<"unsigned short(_Vector<16, int>, _Vector<16, int>, _Constant int, unsigned short)">; def cmpq512_mask : X86Builtin<"unsigned char(_Vector<8, long long int>, _Vector<8, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def cmpw512_mask : X86Builtin<"unsigned int(_Vector<32, short>, _Vector<32, short>, _Constant int, unsigned int)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def ucmpb128_mask : X86Builtin<"unsigned short(_Vector<16, char>, _Vector<16, char>, _Constant int, unsigned short)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def ucmpd128_mask : X86Builtin<"unsigned char(_Vector<4, int>, _Vector<4, int>, _Constant int, unsigned char)">; def ucmpq128_mask : X86Builtin<"unsigned char(_Vector<2, long long int>, _Vector<2, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { def ucmpw128_mask : X86Builtin<"unsigned char(_Vector<8, short>, _Vector<8, short>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def ucmpb256_mask : X86Builtin<"unsigned int(_Vector<32, char>, _Vector<32, char>, _Constant int, unsigned int)">; } -let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def ucmpd256_mask : X86Builtin<"unsigned char(_Vector<8, int>, _Vector<8, int>, _Constant int, unsigned char)">; def ucmpq256_mask : X86Builtin<"unsigned char(_Vector<4, long long int>, _Vector<4, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512vl,avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { +let Features = "avx512vl,avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { def ucmpw256_mask : X86Builtin<"unsigned short(_Vector<16, short>, _Vector<16, short>, _Constant int, unsigned short)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def ucmpb512_mask : X86Builtin<"unsigned long long int(_Vector<64, char>, _Vector<64, char>, _Constant int, unsigned long long int)">; } -let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512f", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def ucmpd512_mask : X86Builtin<"unsigned short(_Vector<16, int>, _Vector<16, int>, _Constant int, unsigned short)">; def ucmpq512_mask : X86Builtin<"unsigned char(_Vector<8, long long int>, _Vector<8, long long int>, _Constant int, unsigned char)">; } -let Features = "avx512bw", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { +let Features = "avx512bw", + Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { def ucmpw512_mask : X86Builtin<"unsigned int(_Vector<32, short>, _Vector<32, short>, _Constant int, unsigned int)">; } diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp index 0ee18be166845..2526c13b529a7 100644 --- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp +++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp @@ -3101,6 +3101,62 @@ static bool interp__builtin_vec_set(InterpState &S, CodePtr OpPC, return true; } +static bool interp__builtin_cmp_mask(InterpState &S, CodePtr OpPC, + const CallExpr *Call, unsigned ID, + bool IsUnsigned) { + assert(Call->getNumArgs() == 4); + + APSInt Mask = popToAPSInt(S, Call->getArg(3)); + APSInt Opcode = popToAPSInt(S, Call->getArg(2)); + const Pointer &LHS = S.Stk.pop<Pointer>(); + const Pointer &RHS = S.Stk.pop<Pointer>(); + + assert(LHS.getNumElems() == RHS.getNumElems()); + + APInt RetMask = APInt::getZero(LHS.getNumElems()); + unsigned VectorLen = LHS.getNumElems(); + PrimType ElemT = LHS.getFieldDesc()->getPrimType(); + + INT_TYPE_SWITCH_NO_BOOL(ElemT, { + for (unsigned ElemNum = 0; ElemNum < VectorLen; ++ElemNum) { + APSInt A = LHS.elem<T>(ElemNum).toAPSInt(); + APSInt B = RHS.elem<T>(ElemNum).toAPSInt(); + bool result = false; + switch (Opcode.getExtValue() & 0x7) { + case 0x00: // _MM_CMPINT_EQ + result = (A == B); + break; + case 0x01: // _MM_CMPINT_LT + result = IsUnsigned ? A.ult(B) : A.slt(B); + break; + case 0x02: // _MM_CMPINT_LE + result = IsUnsigned ? A.ule(B) : A.sle(B); + break; + case 0x03: // _MM_CMPINT_FALSE + result = false; + break; + case 0x04: // _MM_CMPINT_NE + result = (A != B); + break; + case 0x05: // _MM_CMPINT_NLT (>=) + result = IsUnsigned ? A.uge(B) : A.sge(B); + break; + case 0x06: // _MM_CMPINT_NLE (>) + result = IsUnsigned ? A.ugt(B) : A.sgt(B); + break; + case 0x07: // _MM_CMPINT_TRUE + result = true; + break; + } + + RetMask.setBitVal(ElemNum, Mask[ElemNum] && result); + } + }); + + pushInteger(S, RetMask, Call->getType()); + return true; +} + static bool interp__builtin_ia32_vpconflict(InterpState &S, CodePtr OpPC, const CallExpr *Call) { assert(Call->getNumArgs() == 1); @@ -4141,6 +4197,36 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call, case X86::BI__builtin_ia32_vec_set_v4di: return interp__builtin_vec_set(S, OpPC, Call, BuiltinID); + case X86::BI__builtin_ia32_cmpb128_mask: + case X86::BI__builtin_ia32_cmpw128_mask: + case X86::BI__builtin_ia32_cmpd128_mask: + case X86::BI__builtin_ia32_cmpq128_mask: + case X86::BI__builtin_ia32_cmpb256_mask: + case X86::BI__builtin_ia32_cmpw256_mask: + case X86::BI__builtin_ia32_cmpd256_mask: + case X86::BI__builtin_ia32_cmpq256_mask: + case X86::BI__builtin_ia32_cmpb512_mask: + case X86::BI__builtin_ia32_cmpw512_mask: + case X86::BI__builtin_ia32_cmpd512_mask: + case X86::BI__builtin_ia32_cmpq512_mask: + return interp__builtin_cmp_mask(S, OpPC, Call, BuiltinID, + /*IsUnsigned*/ false); + + case X86::BI__builtin_ia32_ucmpb128_mask: + case X86::BI__builtin_ia32_ucmpw128_mask: + case X86::BI__builtin_ia32_ucmpd128_mask: + case X86::BI__builtin_ia32_ucmpq128_mask: + case X86::BI__builtin_ia32_ucmpb256_mask: + case X86::BI__builtin_ia32_ucmpw256_mask: + case X86::BI__builtin_ia32_ucmpd256_mask: + case X86::BI__builtin_ia32_ucmpq256_mask: + case X86::BI__builtin_ia32_ucmpb512_mask: + case X86::BI__builtin_ia32_ucmpw512_mask: + case X86::BI__builtin_ia32_ucmpd512_mask: + case X86::BI__builtin_ia32_ucmpq512_mask: + return interp__builtin_cmp_mask(S, OpPC, Call, BuiltinID, + /*IsUnsigned*/ true); + default: S.FFDiag(S.Current->getLocation(OpPC), diag::note_invalid_subexpr_in_const_expr) diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp index 16141b27f4ce8..e1e717ebe210b 100644 --- a/clang/lib/AST/ExprConstant.cpp +++ b/clang/lib/AST/ExprConstant.cpp @@ -15449,6 +15449,88 @@ bool IntExprEvaluator::VisitBuiltinCallExpr(const CallExpr *E, unsigned Idx = static_cast<unsigned>(IdxAPS.getZExtValue() & (N - 1)); return Success(Vec.getVectorElt(Idx).getInt(), E); } + + case clang::X86::BI__builtin_ia32_cmpb128_mask: + case clang::X86::BI__builtin_ia32_cmpw128_mask: + case clang::X86::BI__builtin_ia32_cmpd128_mask: + case clang::X86::BI__builtin_ia32_cmpq128_mask: + case clang::X86::BI__builtin_ia32_cmpb256_mask: + case clang::X86::BI__builtin_ia32_cmpw256_mask: + case clang::X86::BI__builtin_ia32_cmpd256_mask: + case clang::X86::BI__builtin_ia32_cmpq256_mask: + case clang::X86::BI__builtin_ia32_cmpb512_mask: + case clang::X86::BI__builtin_ia32_cmpw512_mask: + case clang::X86::BI__builtin_ia32_cmpd512_mask: + case clang::X86::BI__builtin_ia32_cmpq512_mask: + case clang::X86::BI__builtin_ia32_ucmpb128_mask: + case clang::X86::BI__builtin_ia32_ucmpw128_mask: + case clang::X86::BI__builtin_ia32_ucmpd128_mask: + case clang::X86::BI__builtin_ia32_ucmpq128_mask: + case clang::X86::BI__builtin_ia32_ucmpb256_mask: + case clang::X86::BI__builtin_ia32_ucmpw256_mask: + case clang::X86::BI__builtin_ia32_ucmpd256_mask: + case clang::X86::BI__builtin_ia32_ucmpq256_mask: + case clang::X86::BI__builtin_ia32_ucmpb512_mask: + case clang::X86::BI__builtin_ia32_ucmpw512_mask: + case clang::X86::BI__builtin_ia32_ucmpd512_mask: + case clang::X86::BI__builtin_ia32_ucmpq512_mask: { + assert(E->getNumArgs() == 4); + + bool IsUnsigned = + (BuiltinOp >= clang::X86::BI__builtin_ia32_ucmpb128_mask && + BuiltinOp <= clang::X86::BI__builtin_ia32_ucmpq512_mask); + + APValue LHS, RHS; + APSInt Mask, Opcode; + if (!EvaluateVector(E->getArg(0), LHS, Info) || + !EvaluateVector(E->getArg(1), RHS, Info) || + !EvaluateInteger(E->getArg(2), Opcode, Info) || + !EvaluateInteger(E->getArg(3), Mask, Info)) + return false; + + assert(LHS.getVectorLength() == RHS.getVectorLength()); + + APSInt RetMask = APSInt::getUnsigned(0); + unsigned VectorLen = LHS.getVectorLength(); + + for (unsigned ElemNum = 0; ElemNum < VectorLen; ++ElemNum) { + APSInt A = LHS.getVectorElt(ElemNum).getInt(); + APSInt B = RHS.getVectorElt(ElemNum).getInt(); + bool result = false; + + switch (Opcode.getExtValue() & 0x7) { + case 0: // _MM_CMPINT_EQ + result = (A == B); + break; + case 1: // _MM_CMPINT_LT + result = IsUnsigned ? A.ult(B) : A.slt(B); + break; + case 2: // _MM_CMPINT_LE + result = IsUnsigned ? A.ule(B) : A.sle(B); + break; + case 3: // _MM_CMPINT_FALSE + result = false; + break; + case 4: // _MM_CMPINT_NE + result = (A != B); + break; + case 5: // _MM_CMPINT_NLT (>=) + result = IsUnsigned ? A.uge(B) : A.sge(B); + break; + case 6: // _MM_CMPINT_NLE (>) + result = IsUnsigned ? A.ugt(B) : A.sgt(B); + break; + case 7: // _MM_CMPINT_TRUE + result = true; + break; + } + + RetMask.setBitVal(ElemNum, Mask[ElemNum] && result); + } + + RetMask.setIsUnsigned(true); + return Success(APValue(RetMask), E); + } } } diff --git a/clang/lib/Headers/avx512vlbwintrin.h b/clang/lib/Headers/avx512vlbwintrin.h index 639fb60f476c6..97e48357f3ccc 100644 --- a/clang/lib/Headers/avx512vlbwintrin.h +++ b/clang/lib/Headers/avx512vlbwintrin.h @@ -2385,22 +2385,19 @@ _mm256_mask_storeu_epi8 (void *__P, __mmask32 __U, __m256i __A) (__mmask32) __U); } -static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 -_mm_test_epi8_mask (__m128i __A, __m128i __B) -{ +static __inline__ __mmask16 __DEFAULT_FN_ATTRS128_CONSTEXPR +_mm_test_epi8_mask(__m128i __A, __m128i __B) { return _mm_cmpneq_epi8_mask (_mm_and_si128(__A, __B), _mm_setzero_si128()); } -static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 -_mm_mask_test_epi8_mask (__mmask16 __U, __m128i __A, __m128i __B) -{ +static __inline__ __mmask16 __DEFAULT_FN_ATTRS128_CONSTEXPR +_mm_mask_test_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { return _mm_mask_cmpneq_epi8_mask (__U, _mm_and_si128 (__A, __B), _mm_setzero_si128()); } -static __inline__ __mmask32 __DEFAULT_FN_ATTRS256 -_mm256_test_epi8_mask (__m256i __A, __m256i __B) -{ +static __inline__ __mmask32 __DEFAULT_FN_ATTRS256_CONSTEXPR +_mm256_test_epi8_mask(__m256i __A, __m256i __B) { return _mm256_cmpneq_epi8_mask (_mm256_and_si256(__A, __B), _mm256_setzero_si256()); } @@ -2439,9 +2436,8 @@ _mm256_mask_test_epi16_mask (__mmask16 __U, __m256i __A, __m256i __B) _mm256_setzero_si256()); } -static __inline__ __mmask16 __DEFAULT_FN_ATTRS128 -_mm_testn_epi8_mask (__m128i __A, __m128i __B) -{ +static __inline__ __mmask16 __DEFAULT_FN_ATTRS128_CONSTEXPR +_mm_testn_epi8_mask(__m128i __A, __m128i __B) { return _mm_cmpeq_epi8_mask (_mm_and_si128 (__A, __B), _mm_setzero_si128()); } diff --git a/clang/test/CodeGen/X86/avx512vlbw-builtins.c b/clang/test/CodeGen/X86/avx512vlbw-builtins.c index d569283928a0a..1cc4518484c19 100644 --- a/clang/test/CodeGen/X86/avx512vlbw-builtins.c +++ b/clang/test/CodeGen/X86/avx512vlbw-builtins.c @@ -2890,6 +2890,12 @@ __mmask16 test_mm_test_epi8_mask(__m128i __A, __m128i __B) { return _mm_test_epi8_mask(__A, __B); } +TEST_CONSTEXPR(_mm_test_epi8_mask( + (__m128i)(__v16qi){1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}, + (__m128i)(__v16qi){1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} +) +== (__mmask16)0xfffb); + __mmask16 test_mm_mask_test_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { // CHECK-LABEL: test_mm_mask_test_epi8_mask // CHECK: and <2 x i64> %{{.*}}, %{{.*}} @@ -2897,6 +2903,12 @@ __mmask16 test_mm_mask_test_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { // CHECK: and <16 x i1> %{{.*}}, %{{.*}} return _mm_mask_test_epi8_mask(__U, __A, __B); } +TEST_CONSTEXPR(_mm_mask_test_epi8_mask( + 0xFFFF, + (__m128i)(__v16qi){1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}, + (__m128i)(__v16qi){1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} +) +== (__mmask16)0xfffb); __mmask32 test_mm256_test_epi8_mask(__m256i __A, __m256i __B) { // CHECK-LABEL: test_mm256_test_epi8_mask @@ -2904,6 +2916,11 @@ __mmask32 test_mm256_test_epi8_mask(__m256i __A, __m256i __B) { // CHECK: icmp ne <32 x i8> %{{.*}}, %{{.*}} return _mm256_test_epi8_mask(__A, __B); } +TEST_CONSTEXPR(_mm256_test_epi8_mask( + (__m256i)(__v32qi){1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 1, 2, 2, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16}, + (__m256i)(__v32qi){1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 1, 2, 4, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16} +) +== (__mmask32)0xfffbfffb); __mmask32 test_mm256_mask_test_epi8_mask(__mmask32 __U, __m256i __A, __m256i __B) { // CHECK-LABEL: test_mm256_mask_test_epi8_mask @@ -2950,6 +2967,12 @@ __mmask16 test_mm_testn_epi8_mask(__m128i __A, __m128i __B) { return _mm_testn_epi8_mask(__A, __B); } +TEST_CONSTEXPR(_mm_testn_epi8_mask( + (__m128i)(__v16qi){1, 2, 77, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 1, 16, 16}, + (__m128i)(__v16qi){2, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 15} +) +== (__mmask16)0xe001); + __mmask16 test_mm_mask_testn_epi8_mask(__mmask16 __U, __m128i __A, __m128i __B) { // CHECK-LABEL: test_mm_mask_testn_epi8_mask // CHECK: and <2 x i64> %{{.*}}, %{{.*}} |
sskzakaria force-pushed the constexpr_cmp branch from 375bc86 to dbcb924 Compare 
| ✅ With the latest revision this PR passed the C/C++ code formatter. |
tbaederr reviewed Oct 18, 2025
| case X86::BI__builtin_ia32_cmpd512_mask: | ||
| case X86::BI__builtin_ia32_cmpq512_mask: | ||
| return interp__builtin_cmp_mask(S, OpPC, Call, BuiltinID, | ||
| /*IsUnsigned*/ false); |
There was a problem hiding this comment.
Suggested change
| /*IsUnsigned*/ false); | |
| /*IsUnsigned=*/ false); |
| case X86::BI__builtin_ia32_ucmpd512_mask: | ||
| case X86::BI__builtin_ia32_ucmpq512_mask: | ||
| return interp__builtin_cmp_mask(S, OpPC, Call, BuiltinID, | ||
| /*IsUnsigned*/ true); |
There was a problem hiding this comment.
Suggested change
| /*IsUnsigned*/ true); | |
| /*IsUnsigned=*/ true); |
| for (unsigned ElemNum = 0; ElemNum < VectorLen; ++ElemNum) { | ||
| APSInt A = LHS.elem<T>(ElemNum).toAPSInt(); | ||
| APSInt B = RHS.elem<T>(ElemNum).toAPSInt(); | ||
| bool result = false; |
| unsigned VectorLen = LHS.getNumElems(); | ||
| PrimType ElemT = LHS.getFieldDesc()->getPrimType(); | ||
| | ||
| INT_TYPE_SWITCH_NO_BOOL(ElemT, { |
There was a problem hiding this comment.
Move the type switch into the loop body and only do the elem<>() calls in the type switch body. The macro expands to a lot of code otherwise.
RKSimon requested changes Oct 18, 2025
| bool Result = false; | ||
| switch (Opcode.getExtValue() & 0x7) { | ||
| case 0x00: // _MM_CMPINT_EQ | ||
| Result = (LHS.elem<T>(ElemNum).toAPSInt() == |
There was a problem hiding this comment.
Pull out the repeated elem(ElemNum).toAPSInt() calls to top of the loop - that should reduce the size of this code considerably.
clang/lib/AST/ExprConstant.cpp Outdated
| APSInt RetMask(llvm::APInt(RetWidth, 0), /*isUnsigned=*/true); | ||
| for (unsigned ElemNum = 0; ElemNum < VectorLen; ++ElemNum) { | ||
| APSInt A = LHS.getVectorElt(ElemNum).getInt(); | ||
| APSInt B = RHS.getVectorElt(ElemNum).getInt(); |
There was a problem hiding this comment.
const APSInt &A = LHS.getVectorElt(ElemNum).getInt(); const APSInt &B = RHS.getVectorElt(ElemNum).getInt(); | @RKSimon PTLA when you can, github-actions bot said there was a problem with the formatting that I weren't detected git clang-format HEAD and suggested I run clang-format -i on the files but that modified a lot of lines of code if that's alright. |
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
Enables constexpr evaluation for the following AVX512 Integer Comparison Intrinsics:
Part 1
FIXES : #162054