[Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - add AVX512 VPTERNLOGD/VPTERNLOGQ intrinsics to be used in constexpr #158703

kimsh02 · 2025-09-15T18:05:40Z

Add handling for __builtin_ia32_pternlog[d/q][128/256/512]_mask[z] intrinsics to VectorExprEvaluator::VisitCallExpr and InterpBuiltin.cpp with the corresponding test coverage:

_mm_mask_ternarylogic_epi32
_mm_maskz_ternarylogic_epi32
_mm_ternarylogic_epi32
_mm256_mask_ternarylogic_epi32
_mm256_maskz_ternarylogic_epi32
_mm256_ternarylogic_epi32
_mm512_mask_ternarylogic_epi32
_mm512_maskz_ternarylogic_epi32
_mm512_ternarylogic_epi32
_mm_mask_ternarylogic_epi64
_mm_maskz_ternarylogic_epi64
_mm_ternarylogic_epi64
_mm256_mask_ternarylogic_epi64
_mm256_maskz_ternarylogic_epi64
_mm256_ternarylogic_epi64
_mm512_mask_ternarylogic_epi64
_mm512_maskz_ternarylogic_epi64
_mm512_ternarylogic_epi64

llvmbot · 2025-09-15T18:06:18Z

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-x86

Author: Shawn (kimsh02)

Changes

Fix #157698

Add handling for __builtin_ia32_pternlog[d/q][128/256/512]_mask[z] intrinsics to VectorExprEvaluator::VisitCallExpr and InterpBuiltin.cpp with the corresponding test coverage:

_mm_mask_ternarylogic_epi32
_mm_maskz_ternarylogic_epi32
_mm_ternarylogic_epi32
_mm256_mask_ternarylogic_epi32
_mm256_maskz_ternarylogic_epi32
_mm256_ternarylogic_epi32
_mm512_mask_ternarylogic_epi32
_mm512_maskz_ternarylogic_epi32
_mm512_ternarylogic_epi32
_mm_mask_ternarylogic_epi64
_mm_maskz_ternarylogic_epi64
_mm_ternarylogic_epi64
_mm256_mask_ternarylogic_epi64
_mm256_maskz_ternarylogic_epi64
_mm256_ternarylogic_epi64
_mm512_mask_ternarylogic_epi64
_mm512_maskz_ternarylogic_epi64
_mm512_ternarylogic_epi64

Patch is 35.42 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/158703.diff

5 Files Affected:

(modified) clang/include/clang/Basic/BuiltinsX86.td (+12-4)
(modified) clang/lib/AST/ByteCode/InterpBuiltin.cpp (+63-1)
(modified) clang/lib/AST/ExprConstant.cpp (+90)
(modified) clang/test/CodeGen/X86/avx512f-builtins.c (+138)
(modified) clang/test/CodeGen/X86/avx512vl-builtins.c (+277)

diff --git a/clang/include/clang/Basic/BuiltinsX86.td b/clang/include/clang/Basic/BuiltinsX86.td
index 1a8645f99e281..a4335256f6deb 100644
--- a/clang/include/clang/Basic/BuiltinsX86.td
+++ b/clang/include/clang/Basic/BuiltinsX86.td
@@ -2398,28 +2398,36 @@ let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<512>
   def psraq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<2, long long int>)">;
   def psrld512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<4, int>)">;
   def psrlq512 : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<2, long long int>)">;
+}
+
+let Features = "avx512f",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<512>] in {
   def pternlogd512_mask : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>, _Constant int, unsigned short)">;
   def pternlogd512_maskz : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>, _Constant int, unsigned short)">;
   def pternlogq512_mask : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>, _Constant int, unsigned char)">;
   def pternlogq512_maskz : X86Builtin<"_Vector<8, long long int>(_Vector<8, long long int>, _Vector<8, long long int>, _Vector<8, long long int>, _Constant int, unsigned char)">;
 }
 
-let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
+let Features = "avx512vl",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
   def pternlogd128_mask : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>, _Constant int, unsigned char)">;
   def pternlogd128_maskz : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>, _Constant int, unsigned char)">;
 }
 
-let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
+let Features = "avx512vl",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
   def pternlogd256_mask : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>, _Constant int, unsigned char)">;
   def pternlogd256_maskz : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>, _Constant int, unsigned char)">;
 }
 
-let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
+let Features = "avx512vl",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<128>] in {
   def pternlogq128_mask : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>, _Constant int, unsigned char)">;
   def pternlogq128_maskz : X86Builtin<"_Vector<2, long long int>(_Vector<2, long long int>, _Vector<2, long long int>, _Vector<2, long long int>, _Constant int, unsigned char)">;
 }
 
-let Features = "avx512vl", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
+let Features = "avx512vl",
+    Attributes = [NoThrow, Const, Constexpr, RequiredVectorWidth<256>] in {
   def pternlogq256_mask : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>, _Constant int, unsigned char)">;
   def pternlogq256_maskz : X86Builtin<"_Vector<4, long long int>(_Vector<4, long long int>, _Vector<4, long long int>, _Vector<4, long long int>, _Constant int, unsigned char)">;
 }
diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
index 4461731c25648..4004e9fca86b8 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -2903,6 +2903,55 @@ static bool interp__builtin_elementwise_triop(
   return true;
 }
 
+static bool interp__builtin_pternlog(InterpState &S, CodePtr OpPC,
+                                     const CallExpr *Call, bool MaskZ) {
+  assert(Call->getNumArgs() == 5);
+
+  const VectorType *VecT = Call->getArg(0)->getType()->castAs<VectorType>();
+  unsigned DstLen = VecT->getNumElements();
+  PrimType DstElemT = *S.getContext().classify(VecT->getElementType());
+
+  APSInt U = popToAPSInt(S.Stk, *S.getContext().classify(Call->getArg(4)));
+  APSInt Imm = popToAPSInt(S.Stk, *S.getContext().classify(Call->getArg(3)));
+  const Pointer &C = S.Stk.pop<Pointer>();
+  const Pointer &B = S.Stk.pop<Pointer>();
+  const Pointer &A = S.Stk.pop<Pointer>();
+
+  const Pointer &Dst = S.Stk.peek<Pointer>();
+
+  for (unsigned I = 0; I < DstLen; ++I) {
+    APSInt a, b, c;
+    INT_TYPE_SWITCH(DstElemT, {
+      a = A.elem<T>(I).toAPSInt();
+      b = B.elem<T>(I).toAPSInt();
+      c = C.elem<T>(I).toAPSInt();
+    });
+
+    unsigned BitWidth = a.getBitWidth();
+    APInt R(BitWidth, 0);
+    bool DstUnsigned = a.isUnsigned();
+
+    if (U[I]) {
+      for (unsigned Bit = 0; Bit < BitWidth; ++Bit) {
+        unsigned Idx = (a[Bit] << 2) | (b[Bit] << 1) | (c[Bit]);
+        R.setBitVal(Bit, Imm[Idx]);
+      }
+      INT_TYPE_SWITCH_NO_BOOL(DstElemT, {
+        Dst.elem<T>(I) = static_cast<T>(APSInt(R, DstUnsigned));
+      });
+    } else if (MaskZ) {
+      INT_TYPE_SWITCH_NO_BOOL(DstElemT, {
+        Dst.elem<T>(I) = static_cast<T>(APSInt(R, DstUnsigned));
+      });
+    } else {
+      INT_TYPE_SWITCH_NO_BOOL(DstElemT,
+                              { Dst.elem<T>(I) = static_cast<T>(a); });
+    }
+  }
+  Dst.initializeAllElements();
+  return true;
+}
+
 bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
                       uint32_t BuiltinID) {
   if (!S.getASTContext().BuiltinInfo.isConstantEvaluated(BuiltinID))
@@ -3543,7 +3592,20 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
   case X86::BI__builtin_ia32_selectpd_256:
   case X86::BI__builtin_ia32_selectpd_512:
     return interp__builtin_select(S, OpPC, Call);
-
+  case X86::BI__builtin_ia32_pternlogd128_mask:
+  case X86::BI__builtin_ia32_pternlogd256_mask:
+  case X86::BI__builtin_ia32_pternlogd512_mask:
+  case X86::BI__builtin_ia32_pternlogq128_mask:
+  case X86::BI__builtin_ia32_pternlogq256_mask:
+  case X86::BI__builtin_ia32_pternlogq512_mask:
+    return interp__builtin_pternlog(S, OpPC, Call, false);
+  case X86::BI__builtin_ia32_pternlogd128_maskz:
+  case X86::BI__builtin_ia32_pternlogd256_maskz:
+  case X86::BI__builtin_ia32_pternlogd512_maskz:
+  case X86::BI__builtin_ia32_pternlogq128_maskz:
+  case X86::BI__builtin_ia32_pternlogq256_maskz:
+  case X86::BI__builtin_ia32_pternlogq512_maskz:
+    return interp__builtin_pternlog(S, OpPC, Call, true);
   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 5145896930153..1f063308aeca5 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -11997,6 +11997,96 @@ bool VectorExprEvaluator::VisitCallExpr(const CallExpr *E) {
 
     return Success(APValue(ResultElements.data(), ResultElements.size()), E);
   }
+  case X86::BI__builtin_ia32_pternlogd128_mask:
+  case X86::BI__builtin_ia32_pternlogd256_mask:
+  case X86::BI__builtin_ia32_pternlogd512_mask:
+  case X86::BI__builtin_ia32_pternlogq128_mask:
+  case X86::BI__builtin_ia32_pternlogq256_mask:
+  case X86::BI__builtin_ia32_pternlogq512_mask: {
+    APValue AValue, BValue, CValue, ImmValue, UValue;
+    if (!EvaluateAsRValue(Info, E->getArg(0), AValue) ||
+        !EvaluateAsRValue(Info, E->getArg(1), BValue) ||
+        !EvaluateAsRValue(Info, E->getArg(2), CValue) ||
+        !EvaluateAsRValue(Info, E->getArg(3), ImmValue) ||
+        !EvaluateAsRValue(Info, E->getArg(4), UValue))
+      return false;
+
+    QualType DestEltTy = E->getType()->castAs<VectorType>()->getElementType();
+    bool DestUnsigned = DestEltTy->isUnsignedIntegerOrEnumerationType();
+    APInt Imm = ImmValue.getInt();
+    APInt U = UValue.getInt();
+    unsigned ResultLen = AValue.getVectorLength();
+    SmallVector<APValue, 16> ResultElements;
+    ResultElements.reserve(ResultLen);
+
+    for (unsigned EltNum = 0; EltNum < ResultLen; ++EltNum) {
+      APInt ALane = AValue.getVectorElt(EltNum).getInt();
+      APInt BLane = BValue.getVectorElt(EltNum).getInt();
+      APInt CLane = CValue.getVectorElt(EltNum).getInt();
+
+      if (U[EltNum]) {
+        unsigned BitWidth = ALane.getBitWidth();
+        APInt ResLane(BitWidth, 0);
+
+        for (unsigned Bit = 0; Bit < BitWidth; ++Bit) {
+          unsigned ABit = ALane[Bit];
+          unsigned BBit = BLane[Bit];
+          unsigned CBit = CLane[Bit];
+
+          unsigned Idx = (ABit << 2) | (BBit << 1) | CBit;
+          ResLane.setBitVal(Bit, Imm[Idx]);
+        }
+        ResultElements.push_back(APValue(APSInt(ResLane, DestUnsigned)));
+      } else {
+        ResultElements.push_back(APValue(APSInt(ALane, DestUnsigned)));
+      }
+    }
+    return Success(APValue(ResultElements.data(), ResultElements.size()), E);
+  }
+  case X86::BI__builtin_ia32_pternlogd128_maskz:
+  case X86::BI__builtin_ia32_pternlogd256_maskz:
+  case X86::BI__builtin_ia32_pternlogd512_maskz:
+  case X86::BI__builtin_ia32_pternlogq128_maskz:
+  case X86::BI__builtin_ia32_pternlogq256_maskz:
+  case X86::BI__builtin_ia32_pternlogq512_maskz: {
+    APValue AValue, BValue, CValue, ImmValue, UValue;
+    if (!EvaluateAsRValue(Info, E->getArg(0), AValue) ||
+        !EvaluateAsRValue(Info, E->getArg(1), BValue) ||
+        !EvaluateAsRValue(Info, E->getArg(2), CValue) ||
+        !EvaluateAsRValue(Info, E->getArg(3), ImmValue) ||
+        !EvaluateAsRValue(Info, E->getArg(4), UValue))
+      return false;
+
+    QualType DestEltTy = E->getType()->castAs<VectorType>()->getElementType();
+    bool DestUnsigned = DestEltTy->isUnsignedIntegerOrEnumerationType();
+    APInt Imm = ImmValue.getInt();
+    APInt U = UValue.getInt();
+    unsigned ResultLen = AValue.getVectorLength();
+    SmallVector<APValue, 16> ResultElements;
+    ResultElements.reserve(ResultLen);
+
+    for (unsigned EltNum = 0; EltNum < ResultLen; ++EltNum) {
+      APInt ALane = AValue.getVectorElt(EltNum).getInt();
+      APInt BLane = BValue.getVectorElt(EltNum).getInt();
+      APInt CLane = CValue.getVectorElt(EltNum).getInt();
+
+      unsigned BitWidth = ALane.getBitWidth();
+      APInt ResLane(BitWidth, 0);
+
+      if (U[EltNum]) {
+        for (unsigned Bit = 0; Bit < BitWidth; ++Bit) {
+          unsigned ABit = ALane[Bit];
+          unsigned BBit = BLane[Bit];
+          unsigned CBit = CLane[Bit];
+
+          unsigned Idx = (ABit << 2) | (BBit << 1) | CBit;
+          ResLane.setBitVal(Bit, Imm[Idx]);
+        }
+      }
+      ResultElements.push_back(APValue(APSInt(ResLane, DestUnsigned)));
+    }
+    return Success(APValue(ResultElements.data(), ResultElements.size()), E);
+  }
   case Builtin::BI__builtin_elementwise_ctlz:
   case Builtin::BI__builtin_elementwise_cttz: {
     APValue SourceLHS;
diff --git a/clang/test/CodeGen/X86/avx512f-builtins.c b/clang/test/CodeGen/X86/avx512f-builtins.c
index f93216e546a63..9138bdef7f92b 100644
--- a/clang/test/CodeGen/X86/avx512f-builtins.c
+++ b/clang/test/CodeGen/X86/avx512f-builtins.c
@@ -6206,6 +6206,27 @@ __m512i test_mm512_ternarylogic_epi32(__m512i __A, __m512i __B, __m512i __C) {
   // CHECK: @llvm.x86.avx512.pternlog.d.512({{.*}}, i32 240)
   return _mm512_ternarylogic_epi32(__A, __B, __C, _MM_TERNLOG_A);
 }
+TEST_CONSTEXPR(match_v16si(
+  _mm512_ternarylogic_epi32(
+    ((__m512i)((__v16si){-0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0})),
+    ((__m512i)((__v16si){0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB})),
+    ((__m512i)((__v16si){0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC})),
+    (unsigned char)0xCA),
+  0xB, 0xC, 0xB, 0xC, 0xB, 0xC, 0xB, 0xC, 0xB, 0xC, 0xB, 0xC, 0xB, 0xC, 0xB, 0xC));
+TEST_CONSTEXPR(match_v16si(
+  _mm512_ternarylogic_epi32(
+    ((__m512i)((__v16si){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    ((__m512i)((__v16si){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v16si){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0xFE),
+  0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF));
+TEST_CONSTEXPR(match_v16si(
+  _mm512_ternarylogic_epi32(
+    ((__m512i)((__v16si){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    ((__m512i)((__v16si){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v16si){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0x80),
+  0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0));
 
 __m512i test_mm512_mask_ternarylogic_epi32(__m512i __A, __mmask16 __U, __m512i __B, __m512i __C) {
   // CHECK-LABEL: test_mm512_mask_ternarylogic_epi32
@@ -6213,6 +6234,30 @@ __m512i test_mm512_mask_ternarylogic_epi32(__m512i __A, __mmask16 __U, __m512i _
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
   return _mm512_mask_ternarylogic_epi32(__A, __U, __B, __C, _MM_TERNLOG_B);
 }
+TEST_CONSTEXPR(match_v16si(
+  _mm512_mask_ternarylogic_epi32(
+    ((__m512i)((__v16si){-0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0})),
+    (__mmask16)0x3333,
+    ((__m512i)((__v16si){0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB})),
+    ((__m512i)((__v16si){0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC})),
+    (unsigned char)0xCA),
+  0xB, 0xC, -0x1, 0x0, 0xB, 0xC, -0x1, 0x0, 0xB, 0xC, -0x1, 0x0, 0xB, 0xC, -0x1, 0x0));
+TEST_CONSTEXPR(match_v16si(
+  _mm512_mask_ternarylogic_epi32(
+    ((__m512i)((__v16si){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    (__mmask16)0xCCCC,
+    ((__m512i)((__v16si){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v16si){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0xFE),
+  0x9, 0x9, 0xF, 0xF, 0x9, 0x9, 0xF, 0xF, 0x9, 0x9, 0xF, 0xF, 0x9, 0x9, 0xF, 0xF));
+TEST_CONSTEXPR(match_v16si(
+  _mm512_mask_ternarylogic_epi32(
+    ((__m512i)((__v16si){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    (__mmask16)0x5555,
+    ((__m512i)((__v16si){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v16si){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0x80),
+  0x0, 0x9, 0x0, 0x9, 0x0, 0x9, 0x0, 0x9, 0x0, 0x9, 0x0, 0x9, 0x0, 0x9, 0x0, 0x9));
 
 __m512i test_mm512_maskz_ternarylogic_epi32(__mmask16 __U, __m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: test_mm512_maskz_ternarylogic_epi32
@@ -6220,12 +6265,57 @@ __m512i test_mm512_maskz_ternarylogic_epi32(__mmask16 __U, __m512i __A, __m512i
   // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> zeroinitializer
   return _mm512_maskz_ternarylogic_epi32(__U, __A, __B, __C, _MM_TERNLOG_C);
 }
+TEST_CONSTEXPR(match_v16si(
+  _mm512_maskz_ternarylogic_epi32(
+    (__mmask16)0x3333,
+    ((__m512i)((__v16si){-0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0})),
+    ((__m512i)((__v16si){0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB})),
+    ((__m512i)((__v16si){0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC})),
+    (unsigned char)0xCA),
+  0xB, 0xC, 0x0, 0x0, 0xB, 0xC, 0x0, 0x0, 0xB, 0xC, 0x0, 0x0, 0xB, 0xC, 0x0, 0x0));
+TEST_CONSTEXPR(match_v16si(
+  _mm512_maskz_ternarylogic_epi32(
+    (__mmask16)0xCCCC,
+    ((__m512i)((__v16si){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    ((__m512i)((__v16si){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v16si){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0xFE),
+  0x0, 0x0, 0xF, 0xF, 0x0, 0x0, 0xF, 0xF, 0x0, 0x0, 0xF, 0xF, 0x0, 0x0, 0xF, 0xF));
+TEST_CONSTEXPR(match_v16si(
+  _mm512_maskz_ternarylogic_epi32(
+    (__mmask16)0x5555,
+    ((__m512i)((__v16si){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    ((__m512i)((__v16si){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v16si){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0x80),
+  0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0));
 
 __m512i test_mm512_ternarylogic_epi64(__m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: test_mm512_ternarylogic_epi64
   // CHECK: @llvm.x86.avx512.pternlog.q.512({{.*}}, i32 192)
   return _mm512_ternarylogic_epi64(__A, __B, __C, _MM_TERNLOG_A & _MM_TERNLOG_B);
 }
+TEST_CONSTEXPR(match_v8di(
+  _mm512_ternarylogic_epi64(
+    ((__m512i)((__v8di){-0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0})),
+    ((__m512i)((__v8di){0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB})),
+    ((__m512i)((__v8di){0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC})),
+    (unsigned char)0xCA),
+  0xB, 0xC, 0xB, 0xC, 0xB, 0xC, 0xB, 0xC));
+TEST_CONSTEXPR(match_v8di(
+  _mm512_ternarylogic_epi64(
+    ((__m512i)((__v8di){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    ((__m512i)((__v8di){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v8di){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0xFE),
+  0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF, 0xF));
+TEST_CONSTEXPR(match_v8di(
+  _mm512_ternarylogic_epi64(
+    ((__m512i)((__v8di){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    ((__m512i)((__v8di){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v8di){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0x80),
+  0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0));
 
 __m512i test_mm512_mask_ternarylogic_epi64(__m512i __A, __mmask8 __U, __m512i __B, __m512i __C) {
   // CHECK-LABEL: test_mm512_mask_ternarylogic_epi64
@@ -6233,6 +6323,30 @@ __m512i test_mm512_mask_ternarylogic_epi64(__m512i __A, __mmask8 __U, __m512i __
   // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
   return _mm512_mask_ternarylogic_epi64(__A, __U, __B, __C, _MM_TERNLOG_B | _MM_TERNLOG_C);
 }
+TEST_CONSTEXPR(match_v8di(
+  _mm512_mask_ternarylogic_epi64(
+    ((__m512i)((__v8di){-0x1, 0x0, -0x1, 0x0, -0x1, 0x0, -0x1, 0x0})),
+    (__mmask8)0x33,
+    ((__m512i)((__v8di){0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB, 0xB})),
+    ((__m512i)((__v8di){0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC, 0xC})),
+    (unsigned char)0xCA),
+  0xB, 0xC, -0x1, 0x0, 0xB, 0xC, -0x1, 0x0));
+TEST_CONSTEXPR(match_v8di(
+  _mm512_mask_ternarylogic_epi64(
+    ((__m512i)((__v8di){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    (__mmask8)0xCC,
+    ((__m512i)((__v8di){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v8di){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0xFE),
+  0x9, 0x9, 0xF, 0xF, 0x9, 0x9, 0xF, 0xF));
+TEST_CONSTEXPR(match_v8di(
+  _mm512_mask_ternarylogic_epi64(
+    ((__m512i)((__v8di){0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9, 0x9})),
+    (__mmask8)0x55,
+    ((__m512i)((__v8di){0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4, 0x4})),
+    ((__m512i)((__v8di){0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2, 0x2})),
+    (unsigned char)0x80),
+  0x0, 0x9, 0x0, 0x9, 0x0, 0x9, 0x0, 0x9));
 
 __m512i test_mm512_maskz_ternarylogic_epi64(__mmask8 __U, __m512i __A, __m512i __B, __m512i __C) {
   // CHECK-LABEL: test_mm512_maskz_ternarylogic_epi64
@@ -6240,6 +6354,30 @@ __m512i test_mm512_maskz_ternarylogic_epi64(__mmask8 __U, __m512i __A, __m512i _
   // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> zeroinitializer
   return _mm512_maskz_ternarylogic_epi64(__U, __A, __B, __C, ~_MM_TERNLOG_A | (_MM_TERNLOG_B ^ _MM_TERNLOG_C));
 }
+TEST_CONSTEXPR(match_v8di(
+  _mm512_maskz_ternarylogic_epi64(
+    (__mmask8)0x33,
+    ((__m512i)((__v8di){-0x1, 0x0, -0x1, 0x0, -0x...
[truncated]

clang/lib/AST/ByteCode/InterpBuiltin.cpp

github-actions · 2025-09-16T07:51:28Z

✅ With the latest revision this PR passed the C/C++ code formatter.

clang/test/CodeGen/X86/avx512vl-builtins.c

clang/lib/AST/ByteCode/InterpBuiltin.cpp

kimsh02 · 2025-09-18T06:32:13Z

@phoebewang Sorry, accidentally requested your review, didn't mean to.

RKSimon

please can you merge against trunk

kimsh02 · 2025-09-19T22:14:03Z

I couldn't get clang-format to change any files, my recent push was just to rebase.

clang/test/CodeGen/X86/avx512f-builtins.c

tbaederr · 2025-09-28T04:48:29Z

Are you still stuck with a problem? I can have a look.

kimsh02 · 2025-09-28T05:50:49Z

@tbaederr I'm still working on it. Thanks for checking in!

kimsh02 · 2025-10-01T06:40:34Z

As an update, I made better test cases with varying ternary operations and values, but I was not able to debug the handling in the new evaluator. There are three tests which pass from the old evaluator but does not pass from the new evaluator, and I have commented these failing tests.

What the failing tests have in common is that they hit the pternlogq512 path (512-bit, 64-bit lanes), although the new evaluator passes other tests that hit this path as well. What I have tried was reviewing my logic, and trying to replicate the code exactly as I had written it in the old evaluator (since the old evaluator is passing everything).

@tbaederr If anyone could have a look, that would be great 😅

clang/test/CodeGen/X86/avx512f-builtins.c

tbaederr · 2025-10-01T06:46:24Z

Can you rebase this first? The patch doesn't apply on main right now.

clang/test/CodeGen/X86/avx512f-builtins.c

kimsh02 · 2025-10-01T06:50:11Z

Can you rebase this first? The patch doesn't apply on main right now.

Okay, will do.

kimsh02 · 2025-10-01T07:57:31Z

@tbaederr Is the branch fine now?

tbaederr · 2025-10-01T08:45:04Z

Yes, thanks.

tbaederr · 2025-10-01T11:08:46Z

Should work once #161506 is merged.

kimsh02 · 2025-10-01T12:41:39Z

  if (Value.isSigned())
    return this->emitConst(Value.getSExtValue(), Ty, E);
  return this->emitConst(Value.getZExtValue(), Ty, E);

❤️

kimsh02 · 2025-10-02T18:35:48Z

Better test coverage with help of LLM.

clang/lib/AST/ByteCode/InterpBuiltin.cpp

kimsh02 · 2025-10-04T17:07:10Z

@tbaederr 👍

RKSimon

LGTM - cheers

llvmbot added clang Clang issues not falling into any other category backend:X86 clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:bytecode Issues for the clang bytecode constexpr interpreter labels Sep 15, 2025

RKSimon requested review from RKSimon and tbaederr September 15, 2025 18:13

RKSimon reviewed Sep 15, 2025

View reviewed changes

clang/lib/AST/ByteCode/InterpBuiltin.cpp Show resolved Hide resolved

RKSimon requested a review from phoebewang September 15, 2025 18:16

tbaederr reviewed Sep 16, 2025

View reviewed changes

kimsh02 commented Sep 16, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512vl-builtins.c Outdated Show resolved Hide resolved

kimsh02 commented Sep 16, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512vl-builtins.c Outdated Show resolved Hide resolved

kimsh02 commented Sep 16, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512vl-builtins.c Outdated Show resolved Hide resolved

kimsh02 requested a review from tbaederr September 17, 2025 19:41

tbaederr reviewed Sep 18, 2025

View reviewed changes

clang/lib/AST/ByteCode/InterpBuiltin.cpp Outdated Show resolved Hide resolved

kimsh02 requested a review from tbaederr September 18, 2025 06:29

RKSimon requested changes Sep 19, 2025

View reviewed changes

kimsh02 force-pushed the 157698 branch from a6ab43b to 9cf5aef Compare September 19, 2025 21:42

kimsh02 requested a review from RKSimon September 19, 2025 22:14

RKSimon reviewed Sep 22, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512f-builtins.c Outdated Show resolved Hide resolved

kimsh02 marked this pull request as draft September 23, 2025 04:02

kimsh02 commented Sep 23, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512f-builtins.c Show resolved Hide resolved

kimsh02 commented Oct 1, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512f-builtins.c Show resolved Hide resolved

kimsh02 commented Oct 1, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512f-builtins.c Show resolved Hide resolved

kimsh02 commented Oct 1, 2025

View reviewed changes

clang/test/CodeGen/X86/avx512f-builtins.c Show resolved Hide resolved

kimsh02 force-pushed the 157698 branch from 8953f45 to e55bd08 Compare October 1, 2025 07:55

kimsh02 force-pushed the 157698 branch 3 times, most recently from 51d6412 to ae6265f Compare October 2, 2025 18:08

kimsh02 marked this pull request as ready for review October 2, 2025 18:35

kimsh02 requested a review from RKSimon October 2, 2025 18:36

tbaederr reviewed Oct 4, 2025

View reviewed changes

kimsh02 added 6 commits October 4, 2025 10:04

Squash commits for rebase

fc9a28f

Clang-format

8e25dd8

Revert back to single handler for new evaluator

3c54f85

Rebase

4cd5bcb

Remove duplicate handler in new evaulator

28e413e

Apply feedback

fc62c1c

kimsh02 force-pushed the 157698 branch from c6a589a to fc62c1c Compare October 4, 2025 17:05

Merge branch 'main' into 157698

40a43a1

RKSimon approved these changes Oct 6, 2025

View reviewed changes

RKSimon enabled auto-merge (squash) October 6, 2025 16:21

RKSimon merged commit d7feeda into llvm:main Oct 6, 2025
9 checks passed

[Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - add AVX512 VPTERNLOGD/VPTERNLOGQ intrinsics to be used in constexpr #158703

[Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - add AVX512 VPTERNLOGD/VPTERNLOGQ intrinsics to be used in constexpr #158703

Uh oh!

Conversation

kimsh02 commented Sep 15, 2025

Uh oh!

llvmbot commented Sep 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Sep 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

kimsh02 commented Sep 18, 2025

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

kimsh02 commented Sep 19, 2025

Uh oh!

Uh oh!

Uh oh!

tbaederr commented Sep 28, 2025

Uh oh!

kimsh02 commented Sep 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kimsh02 commented Oct 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tbaederr commented Oct 1, 2025

Uh oh!

Uh oh!

kimsh02 commented Oct 1, 2025

Uh oh!

kimsh02 commented Oct 1, 2025

Uh oh!

tbaederr commented Oct 1, 2025

Uh oh!

tbaederr commented Oct 1, 2025

Uh oh!

kimsh02 commented Oct 1, 2025

Uh oh!

kimsh02 commented Oct 2, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

kimsh02 commented Oct 4, 2025

Uh oh!

RKSimon left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

llvmbot commented Sep 15, 2025 •

edited

Loading

github-actions bot commented Sep 16, 2025 •

edited

Loading

kimsh02 commented Sep 28, 2025 •

edited

Loading

kimsh02 commented Oct 1, 2025 •

edited

Loading