Skip to content

Conversation

kimsh02
Copy link
Contributor

@kimsh02 kimsh02 commented Oct 12, 2025

Fix #161336

@llvmbot llvmbot added clang Clang issues not falling into any other category backend:X86 clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:headers Headers provided by Clang, e.g. for intrinsics clang:bytecode Issues for the clang bytecode constexpr interpreter labels Oct 12, 2025
@llvmbot
Copy link
Member

llvmbot commented Oct 12, 2025

@llvm/pr-subscribers-backend-x86

@llvm/pr-subscribers-clang

Author: Shawn K (kimsh02)

Changes

Fix #161336


Full diff: https://github.com/llvm/llvm-project/pull/163041.diff

5 Files Affected:

  • (modified) clang/include/clang/Basic/BuiltinsX86.td (+3-2)
  • (modified) clang/lib/AST/ByteCode/InterpBuiltin.cpp (+44)
  • (modified) clang/lib/AST/ExprConstant.cpp (+34)
  • (modified) clang/lib/Headers/smmintrin.h (+2-1)
  • (modified) clang/test/CodeGen/X86/sse41-builtins.c (+10)
diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td index 217589d7add1d..1cd72707bf93f 100644 --- a/clang/include/clang/Basic/BuiltinsX86.td +++ b/clang/include/clang/Basic/BuiltinsX86.td @@ -327,8 +327,8 @@ let Features = "sse4.1", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] def ptestz128 : X86Builtin<"int(_Vector<2, long long int>, _Vector<2, long long int>)">; def ptestc128 : X86Builtin<"int(_Vector<2, long long int>, _Vector<2, long long int>)">; def ptestnzc128 : X86Builtin<"int(_Vector<2, long long int>, _Vector<2, long long int>)">; - def mpsadbw128 : X86Builtin<"_Vector<16, char>(_Vector<16, char>, _Vector<16, char>, _Constant char)">; - def phminposuw128 : X86Builtin<"_Vector<8, short>(_Vector<8, short>)">; + def mpsadbw128 : X86Builtin<"_Vector<16, char>(_Vector<16, char>, " + "_Vector<16, char>, _Constant char)">; } let Features = "sse4.1", Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in { @@ -341,6 +341,7 @@ let Features = "sse4.1", Attributes = [NoThrow, Const, Constexpr, RequiredVector def pmuldq128 : X86Builtin<"_Vector<2, long long int>(_Vector<4, int>, _Vector<4, int>)">; def packusdw128 : X86Builtin<"_Vector<8, short>(_Vector<4, int>, _Vector<4, int>)">; + def phminposuw128 : X86Builtin<"_Vector<8, short>(_Vector<8, short>)">; def vec_ext_v16qi : X86Builtin<"char(_Vector<16, char>, _Constant int)">; def vec_set_v16qi : X86Builtin<"_Vector<16, char>(_Vector<16, char>, char, _Constant int)">; diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp index 922d67940e22f..1b3eb4da69ec7 100644 --- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp +++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp @@ -2860,6 +2860,46 @@ static bool interp__builtin_x86_insert_subvector(InterpState &S, CodePtr OpPC, return true; } +static bool interp__builtin_ia32_phminposuw(InterpState &S, CodePtr OpPC, + const CallExpr *Call) { + assert(Call->getNumArgs() == 1); + + const Pointer &Source = S.Stk.pop<Pointer>(); + const Pointer &Dest = S.Stk.peek<Pointer>(); + + unsigned SourceLen = Source.getNumElems(); + QualType ElemQT = getElemType(Source); + OptPrimType ElemPT = S.getContext().classify(ElemQT); + unsigned LaneBitWidth = S.getASTContext().getTypeSize(ElemQT); + + INT_TYPE_SWITCH_NO_BOOL(*ElemPT, { + APInt MinIndex(LaneBitWidth, 0); + APInt MinVal = Source.elem<T>(0).toAPSInt(); + + for (unsigned I = 0; I != SourceLen; ++I) { + APInt Val = Source.elem<T>(I).toAPSInt(); + if (MinVal.ugt(Val)) { + MinVal = Val; + MinIndex = I; + } + } + + bool DestUnsigned = Call->getCallReturnType(S.getASTContext()) + ->castAs<VectorType>() + ->getElementType() + ->isUnsignedIntegerOrEnumerationType(); + + Dest.elem<T>(0) = static_cast<T>(APSInt(MinVal, DestUnsigned)); + Dest.elem<T>(1) = static_cast<T>(APSInt(MinIndex, DestUnsigned)); + for (unsigned I = 2; I != SourceLen; ++I) { + Dest.elem<T>(I) = + static_cast<T>(APSInt(APInt(LaneBitWidth, 0), DestUnsigned)); + } + }); + Dest.initializeAllElements(); + return true; +} + static bool interp__builtin_ia32_pternlog(InterpState &S, CodePtr OpPC, const CallExpr *Call, bool MaskZ) { assert(Call->getNumArgs() == 5); @@ -3809,6 +3849,10 @@ 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_phminposuw128: { + return interp__builtin_ia32_phminposuw(S, OpPC, Call); + } + case X86::BI__builtin_ia32_pternlogd128_mask: case X86::BI__builtin_ia32_pternlogd256_mask: case X86::BI__builtin_ia32_pternlogd512_mask: diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp index 35a866ea5010f..e686d17522260 100644 --- a/clang/lib/AST/ExprConstant.cpp +++ b/clang/lib/AST/ExprConstant.cpp @@ -12216,6 +12216,40 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) { return Success(R, E); } + case X86::BI__builtin_ia32_phminposuw128: { + APValue Source; + if (!Evaluate(Source, Info, E->getArg(0))) + return false; + unsigned SourceLen = Source.getVectorLength(); + const VectorType *VT = E->getArg(0)->getType()->castAs<VectorType>(); + QualType ElemQT = VT->getElementType(); + unsigned LaneBitWidth = Info.Ctx.getTypeSize(ElemQT); + + APInt MinIndex(LaneBitWidth, 0); + APInt MinVal = Source.getVectorElt(0).getInt(); + for (unsigned I = 0; I != SourceLen; ++I) { + APInt Val = Source.getVectorElt(I).getInt(); + if (MinVal.ugt(Val)) { + MinVal = Val; + MinIndex = I; + } + } + + bool ResultUnsigned = E->getCallReturnType(Info.Ctx) + ->castAs<VectorType>() + ->getElementType() + ->isUnsignedIntegerOrEnumerationType(); + + SmallVector<APValue, 8> Result; + Result.reserve(SourceLen); + Result.emplace_back(APSInt(MinVal, ResultUnsigned)); + Result.emplace_back(APSInt(MinIndex, ResultUnsigned)); + for (unsigned I = 0; I != SourceLen - 2; ++I) { + Result.emplace_back(APSInt(APInt(LaneBitWidth, 0), ResultUnsigned)); + } + return Success(APValue(Result.data(), Result.size()), E); + } + case X86::BI__builtin_ia32_pternlogd128_mask: case X86::BI__builtin_ia32_pternlogd256_mask: case X86::BI__builtin_ia32_pternlogd512_mask: diff --git a/clang/lib/Headers/smmintrin.h b/clang/lib/Headers/smmintrin.h index 5e63a1ae321bc..1cc2c7b9371a2 100644 --- a/clang/lib/Headers/smmintrin.h +++ b/clang/lib/Headers/smmintrin.h @@ -1524,7 +1524,8 @@ _mm_packus_epi32(__m128i __V1, __m128i __V2) { /// \returns A 128-bit value where bits [15:0] contain the minimum value found /// in parameter \a __V, bits [18:16] contain the index of the minimum value /// and the remaining bits are set to 0. -static __inline__ __m128i __DEFAULT_FN_ATTRS _mm_minpos_epu16(__m128i __V) { +static __inline__ __m128i __DEFAULT_FN_ATTRS_CONSTEXPR +_mm_minpos_epu16(__m128i __V) { return (__m128i)__builtin_ia32_phminposuw128((__v8hi)__V); } diff --git a/clang/test/CodeGen/X86/sse41-builtins.c b/clang/test/CodeGen/X86/sse41-builtins.c index eee479a755ab4..28b9b83237c88 100644 --- a/clang/test/CodeGen/X86/sse41-builtins.c +++ b/clang/test/CodeGen/X86/sse41-builtins.c @@ -376,6 +376,16 @@ __m128i test_mm_minpos_epu16(__m128i x) { // CHECK: call <8 x i16> @llvm.x86.sse41.phminposuw(<8 x i16> %{{.*}}) return _mm_minpos_epu16(x); } +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){0,0,0,0, 0,0,0,0}), 0,0,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){1,0,0,0, 0,0,0,0}), 0,1,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){65535,65535,65535,65535,65535,65535,65535,65535}), 65535,0,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){9,8,7,6,5,4,3,2}), 2,7,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){5,5,5,5,5,5,5,5}), 5,0,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){5,7,9,4,10,4,11,12}), 4,3,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){6,0,0,0,0,0,0,0}), 0,1,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){1000,2000,3000,4000,5000,6000,7000,1}), 1,7,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){1234,5678,42,9999,65535,0,4242,42}), 0,5,0,0, 0,0,0,0)); +TEST_CONSTEXPR(match_v8hu(_mm_minpos_epu16((__m128i)(__v8hu){400,500,12,600,12,700,800,900}), 12,2,0,0, 0,0,0,0)); __m128i test_mm_mpsadbw_epu8(__m128i x, __m128i y) { // CHECK-LABEL: test_mm_mpsadbw_epu8 
@kimsh02 kimsh02 changed the title X86 phminposuw [Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow SSE41 phminposuw intrinsic to be used in constexp Oct 12, 2025
@tbaederr tbaederr requested a review from RKSimon October 12, 2025 06:16
Copy link
Collaborator

@RKSimon RKSimon left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a couple of minors

@kimsh02
Copy link
Contributor Author

kimsh02 commented Oct 15, 2025

@RKSimon This PR can take a look

->isUnsignedIntegerOrEnumerationType();

INT_TYPE_SWITCH_NO_BOOL(*ElemT, {
APSInt MinIndex(ElemBitWidth, DestUnsigned);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Its probably better to just use:

Suggested change
APSInt MinIndex(ElemBitWidth, DestUnsigned);
unsigned MinIndex = 0;

We never make use of APSInt/APInt features for the index.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

backend:X86 clang:bytecode Issues for the clang bytecode constexpr interpreter clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:headers Headers provided by Clang, e.g. for intrinsics clang Clang issues not falling into any other category

4 participants