https://github.com/chaitanyav created https://github.com/llvm/llvm-project/pull/161056
Resolves #160498 >From 33d313bac475fc3aa558be8657b7373a2c724342 Mon Sep 17 00:00:00 2001 From: NagaChaitanya Vellanki <[email protected]> Date: Sun, 28 Sep 2025 00:29:57 -0700 Subject: [PATCH] [X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - Allow AVX/AVX512 IFMA madd52 intrinsics to be used in constexpr Resolves #160498 --- clang/include/clang/Basic/BuiltinsX86.td | 42 +++++++++---------- clang/lib/AST/ByteCode/InterpBuiltin.cpp | 19 +++++++++ clang/lib/AST/ExprConstant.cpp | 51 ++++++++++++++++++++++++ clang/lib/Headers/avx512ifmaintrin.h | 19 ++++++--- clang/lib/Headers/avxifmaintrin.h | 18 +++++++-- 5 files changed, 118 insertions(+), 31 deletions(-) diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td index 77e599587edc3..a5247629e255f 100644 --- a/clang/include/clang/Basic/BuiltinsX86.td +++ b/clang/include/clang/Basic/BuiltinsX86.td @@ -2101,27 +2101,6 @@ let Features = "avx512vl", Attributes = [NoThrow, RequiredVectorWidth<256>] in { def movdqa64store256_mask : X86Builtin<"void(_Vector<4, long long int *>, _Vector<4, long long int>, unsigned char)">; } -let Features = "avx512ifma", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in { - def vpmadd52huq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">; - def vpmadd52luq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">; -} - -let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { - def vpmadd52huq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">; -} - -let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { - def vpmadd52huq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">; -} - -let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { - def vpmadd52luq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">; -} - -let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in { - def vpmadd52luq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">; -} - let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in { def vcomisd : X86Builtin<"int(_Vector<2, double>, _Vector<2, double>, _Constant int, _Constant int)">; def vcomiss : X86Builtin<"int(_Vector<4, float>, _Vector<4, float>, _Constant int, _Constant int)">; @@ -3128,6 +3107,27 @@ let Features = "avx512bw", Attributes = [NoThrow, Const, Constexpr] in { def kordi : X86Builtin<"unsigned long long int(unsigned long long int, unsigned long long int)">; } +let Features = "avx512ifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in { + def vpmadd52huq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">; + def vpmadd52luq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>)">; +} + +let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { + def vpmadd52huq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">; +} + +let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { + def vpmadd52huq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">; +} + +let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { + def vpmadd52luq128 : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>)">; +} + +let Features = "avx512ifma,avx512vl|avxifma", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in { + def vpmadd52luq256 : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>)">; +} + let Features = "avx512dq", Attributes = [NoThrow, Const] in { def kortestcqi : X86Builtin<"int(unsigned char, unsigned char)">; def kortestzqi : X86Builtin<"int(unsigned char, unsigned char)">; diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp index 891344d4e6ed0..4b29eee057a51 100644 --- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp +++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp @@ -3564,6 +3564,25 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call, return F; }); + case X86::BI__builtin_ia32_vpmadd52luq128: + case X86::BI__builtin_ia32_vpmadd52luq256: + case X86::BI__builtin_ia32_vpmadd52luq512: + return interp__builtin_elementwise_triop( + S, OpPC, Call, [](const APSInt &A, const APSInt &B, const APSInt &C) { + APSInt result = A * B + C; + APSInt mask(APSInt::getAllOnes(52).zext(64)); + return result & mask; + }); + case X86::BI__builtin_ia32_vpmadd52huq128: + case X86::BI__builtin_ia32_vpmadd52huq256: + case X86::BI__builtin_ia32_vpmadd52huq512: + return interp__builtin_elementwise_triop( + S, OpPC, Call, [](const APSInt &A, const APSInt &B, const APSInt &C) { + APSInt result = A * B + C; + APSInt mask(APSInt::getAllOnes(52).zext(64)); + return (result >> 52) & mask; + }); + case X86::BI__builtin_ia32_vpshldd128: case X86::BI__builtin_ia32_vpshldd256: case X86::BI__builtin_ia32_vpshldd512: diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp index b706b14945b6d..092d91f908b43 100644 --- a/clang/lib/AST/ExprConstant.cpp +++ b/clang/lib/AST/ExprConstant.cpp @@ -60,10 +60,12 @@ #include "llvm/ADT/StringExtras.h" #include "llvm/Support/Casting.h" #include "llvm/Support/Debug.h" +#include "llvm/Support/LSP/Logging.h" #include "llvm/Support/SaveAndRestore.h" #include "llvm/Support/SipHash.h" #include "llvm/Support/TimeProfiler.h" #include "llvm/Support/raw_ostream.h" + #include <cstring> #include <functional> #include <limits> @@ -11869,6 +11871,55 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) { return Success(APValue(ResultElements.data(), ResultElements.size()), E); } + case X86::BI__builtin_ia32_vpmadd52luq128: + case X86::BI__builtin_ia32_vpmadd52luq256: + case X86::BI__builtin_ia32_vpmadd52luq512: { + APValue A, B, C; + if (!EvaluateAsRValue(Info, E->getArg(0), A) || + !EvaluateAsRValue(Info, E->getArg(1), B) || + !EvaluateAsRValue(Info, E->getArg(2), C)) + return false; + + unsigned ALen = A.getVectorLength(); + SmallVector<APValue, 4> ResultElements; + ResultElements.reserve(ALen); + + for (unsigned EltNum = 0; EltNum < ALen; EltNum += 1) { + APInt AElt = A.getVectorElt(EltNum).getInt(); + APInt BElt = B.getVectorElt(EltNum).getInt(); + APInt CElt = C.getVectorElt(EltNum).getInt(); + APInt ResElt(AElt * BElt + CElt); + APInt Mask(64, 0x000FFFFFFFFFFFFFULL); + ResultElements.push_back(APValue(APSInt(ResElt.trunc(64) & Mask))); + } + + return Success(APValue(ResultElements.data(), ResultElements.size()), E); + } + case X86::BI__builtin_ia32_vpmadd52huq128: + case X86::BI__builtin_ia32_vpmadd52huq256: + case X86::BI__builtin_ia32_vpmadd52huq512: { + APValue A, B, C; + if (!EvaluateAsRValue(Info, E->getArg(0), A) || + !EvaluateAsRValue(Info, E->getArg(1), B) || + !EvaluateAsRValue(Info, E->getArg(2), C)) + return false; + + unsigned ALen = A.getVectorLength(); + SmallVector<APValue, 4> ResultElements; + ResultElements.reserve(ALen); + + for (unsigned EltNum = 0; EltNum < ALen; EltNum += 1) { + APInt AElt = A.getVectorElt(EltNum).getInt(); + APInt BElt = B.getVectorElt(EltNum).getInt(); + APInt CElt = C.getVectorElt(EltNum).getInt(); + APInt ResElt(AElt * BElt + CElt); + APInt Mask(64, 0x000FFFFFFFFFFFFFULL); + ResultElements.push_back( + APValue(APSInt(ResElt.lshr(52).trunc(64) & Mask))); + } + + return Success(APValue(ResultElements.data(), ResultElements.size()), E); + } case clang::X86::BI__builtin_ia32_vprotbi: case clang::X86::BI__builtin_ia32_vprotdi: case clang::X86::BI__builtin_ia32_vprotqi: diff --git a/clang/lib/Headers/avx512ifmaintrin.h b/clang/lib/Headers/avx512ifmaintrin.h index f01b322ce7787..0c22637b63444 100644 --- a/clang/lib/Headers/avx512ifmaintrin.h +++ b/clang/lib/Headers/avx512ifmaintrin.h @@ -19,14 +19,20 @@ __attribute__((__always_inline__, __nodebug__, __target__("avx512ifma"), \ __min_vector_width__(512))) -static __inline__ __m512i __DEFAULT_FN_ATTRS +#if defined(__cplusplus) && (__cplusplus >= 201103L) +#define __DEFAULT_FN_ATTRS_CONSTEXPR __DEFAULT_FN_ATTRS constexpr +#else +#define __DEFAULT_FN_ATTRS_CONSTEXPR __DEFAULT_FN_ATTRS +#endif + +static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR _mm512_madd52hi_epu64 (__m512i __X, __m512i __Y, __m512i __Z) { return (__m512i)__builtin_ia32_vpmadd52huq512((__v8di) __X, (__v8di) __Y, (__v8di) __Z); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR _mm512_mask_madd52hi_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y) { return (__m512i)__builtin_ia32_selectq_512(__M, @@ -34,7 +40,7 @@ _mm512_mask_madd52hi_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y) (__v8di)__W); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR _mm512_maskz_madd52hi_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z) { return (__m512i)__builtin_ia32_selectq_512(__M, @@ -42,14 +48,14 @@ _mm512_maskz_madd52hi_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z (__v8di)_mm512_setzero_si512()); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR _mm512_madd52lo_epu64 (__m512i __X, __m512i __Y, __m512i __Z) { return (__m512i)__builtin_ia32_vpmadd52luq512((__v8di) __X, (__v8di) __Y, (__v8di) __Z); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR _mm512_mask_madd52lo_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y) { return (__m512i)__builtin_ia32_selectq_512(__M, @@ -57,7 +63,7 @@ _mm512_mask_madd52lo_epu64 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y) (__v8di)__W); } -static __inline__ __m512i __DEFAULT_FN_ATTRS +static __inline__ __m512i __DEFAULT_FN_ATTRS_CONSTEXPR _mm512_maskz_madd52lo_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z) { return (__m512i)__builtin_ia32_selectq_512(__M, @@ -66,5 +72,6 @@ _mm512_maskz_madd52lo_epu64 (__mmask8 __M, __m512i __X, __m512i __Y, __m512i __Z } #undef __DEFAULT_FN_ATTRS +#undef __DEFAULT_FN_ATTRS_CONSTEXPR #endif diff --git a/clang/lib/Headers/avxifmaintrin.h b/clang/lib/Headers/avxifmaintrin.h index 5c782d2a5b865..1a9aaaf53affa 100644 --- a/clang/lib/Headers/avxifmaintrin.h +++ b/clang/lib/Headers/avxifmaintrin.h @@ -22,6 +22,14 @@ __attribute__((__always_inline__, __nodebug__, __target__("avxifma"), \ __min_vector_width__(256))) +#if defined(__cplusplus) && (__cplusplus >= 201103L) +#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256 constexpr +#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128 constexpr +#else +#define __DEFAULT_FN_ATTRS256_CONSTEXPR __DEFAULT_FN_ATTRS256 +#define __DEFAULT_FN_ATTRS128_CONSTEXPR __DEFAULT_FN_ATTRS128 +#endif + // must vex-encoding /// Multiply packed unsigned 52-bit integers in each 64-bit element of \a __Y @@ -55,7 +63,7 @@ /// ENDFOR /// dst[MAX:128] := 0 /// \endcode -static __inline__ __m128i __DEFAULT_FN_ATTRS128 +static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR _mm_madd52hi_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) { return (__m128i)__builtin_ia32_vpmadd52huq128((__v2di)__X, (__v2di)__Y, (__v2di)__Z); @@ -92,7 +100,7 @@ _mm_madd52hi_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) { /// ENDFOR /// dst[MAX:256] := 0 /// \endcode -static __inline__ __m256i __DEFAULT_FN_ATTRS256 +static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR _mm256_madd52hi_avx_epu64(__m256i __X, __m256i __Y, __m256i __Z) { return (__m256i)__builtin_ia32_vpmadd52huq256((__v4di)__X, (__v4di)__Y, (__v4di)__Z); @@ -129,7 +137,7 @@ _mm256_madd52hi_avx_epu64(__m256i __X, __m256i __Y, __m256i __Z) { /// ENDFOR /// dst[MAX:128] := 0 /// \endcode -static __inline__ __m128i __DEFAULT_FN_ATTRS128 +static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR _mm_madd52lo_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) { return (__m128i)__builtin_ia32_vpmadd52luq128((__v2di)__X, (__v2di)__Y, (__v2di)__Z); @@ -166,12 +174,14 @@ _mm_madd52lo_avx_epu64(__m128i __X, __m128i __Y, __m128i __Z) { /// ENDFOR /// dst[MAX:256] := 0 /// \endcode -static __inline__ __m256i __DEFAULT_FN_ATTRS256 +static __inline__ __m256i __DEFAULT_FN_ATTRS256_CONSTEXPR _mm256_madd52lo_avx_epu64(__m256i __X, __m256i __Y, __m256i __Z) { return (__m256i)__builtin_ia32_vpmadd52luq256((__v4di)__X, (__v4di)__Y, (__v4di)__Z); } #undef __DEFAULT_FN_ATTRS128 #undef __DEFAULT_FN_ATTRS256 +#undef __DEFAULT_FN_ATTRS256_CONSTEXPR +#undef __DEFAULT_FN_ATTRS128_CONSTEXPR #endif // __AVXIFMAINTRIN_H _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
