aboutsummaryrefslogtreecommitdiff
path: root/clang/lib
diff options
context:
space:
mode:
Diffstat (limited to 'clang/lib')
-rw-r--r--clang/lib/AST/ByteCode/Interp.cpp43
-rw-r--r--clang/lib/AST/ByteCode/InterpBuiltin.cpp29
-rw-r--r--clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp60
-rw-r--r--clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h109
-rw-r--r--clang/lib/Headers/avx512vlintrin.h63
-rw-r--r--clang/lib/Sema/SemaConcept.cpp4
-rw-r--r--clang/lib/Sema/SemaDecl.cpp11
-rw-r--r--clang/lib/Sema/SemaOpenACC.cpp21
-rw-r--r--clang/lib/Sema/SemaOverload.cpp13
9 files changed, 143 insertions, 210 deletions
diff --git a/clang/lib/AST/ByteCode/Interp.cpp b/clang/lib/AST/ByteCode/Interp.cpp
index 21af3d6..8904396 100644
--- a/clang/lib/AST/ByteCode/Interp.cpp
+++ b/clang/lib/AST/ByteCode/Interp.cpp
@@ -1638,6 +1638,36 @@ bool Call(InterpState &S, CodePtr OpPC, const Function *Func,
return true;
}
+static bool GetDynamicDecl(InterpState &S, CodePtr OpPC, Pointer TypePtr,
+ const CXXRecordDecl *&DynamicDecl) {
+ while (TypePtr.isBaseClass())
+ TypePtr = TypePtr.getBase();
+
+ QualType DynamicType = TypePtr.getType();
+ if (TypePtr.isStatic() || TypePtr.isConst()) {
+ const VarDecl *VD = TypePtr.getDeclDesc()->asVarDecl();
+ if (!VD->isConstexpr()) {
+ const Expr *E = S.Current->getExpr(OpPC);
+ APValue V = TypePtr.toAPValue(S.getASTContext());
+ QualType TT = S.getASTContext().getLValueReferenceType(DynamicType);
+ S.FFDiag(E, diag::note_constexpr_polymorphic_unknown_dynamic_type)
+ << AccessKinds::AK_MemberCall << V.getAsString(S.getASTContext(), TT);
+ return false;
+ }
+ }
+
+ if (DynamicType->isPointerType() || DynamicType->isReferenceType()) {
+ DynamicDecl = DynamicType->getPointeeCXXRecordDecl();
+ } else if (DynamicType->isArrayType()) {
+ const Type *ElemType = DynamicType->getPointeeOrArrayElementType();
+ assert(ElemType);
+ DynamicDecl = ElemType->getAsCXXRecordDecl();
+ } else {
+ DynamicDecl = DynamicType->getAsCXXRecordDecl();
+ }
+ return true;
+}
+
bool CallVirt(InterpState &S, CodePtr OpPC, const Function *Func,
uint32_t VarArgSize) {
assert(Func->hasThisPointer());
@@ -1662,17 +1692,8 @@ bool CallVirt(InterpState &S, CodePtr OpPC, const Function *Func,
}
const CXXRecordDecl *DynamicDecl = nullptr;
- {
- Pointer TypePtr = ThisPtr;
- while (TypePtr.isBaseClass())
- TypePtr = TypePtr.getBase();
-
- QualType DynamicType = TypePtr.getType();
- if (DynamicType->isPointerType() || DynamicType->isReferenceType())
- DynamicDecl = DynamicType->getPointeeCXXRecordDecl();
- else
- DynamicDecl = DynamicType->getAsCXXRecordDecl();
- }
+ if (!GetDynamicDecl(S, OpPC, ThisPtr, DynamicDecl))
+ return false;
assert(DynamicDecl);
const auto *StaticDecl = cast<CXXRecordDecl>(Func->getParentDecl());
diff --git a/clang/lib/AST/ByteCode/InterpBuiltin.cpp b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
index 68ebfdf..a3c4ba5 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltin.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltin.cpp
@@ -736,25 +736,6 @@ static bool interp__builtin_expect(InterpState &S, CodePtr OpPC,
return true;
}
-/// rotateleft(value, amount)
-static bool interp__builtin_rotate(InterpState &S, CodePtr OpPC,
- const InterpFrame *Frame,
- const CallExpr *Call, bool Right) {
- APSInt Amount = popToAPSInt(S, Call->getArg(1));
- APSInt Value = popToAPSInt(S, Call->getArg(0));
-
- APSInt Result;
- if (Right)
- Result = APSInt(Value.rotr(Amount.urem(Value.getBitWidth())),
- /*IsUnsigned=*/true);
- else // Left.
- Result = APSInt(Value.rotl(Amount.urem(Value.getBitWidth())),
- /*IsUnsigned=*/true);
-
- pushInteger(S, Result, Call->getType());
- return true;
-}
-
static bool interp__builtin_ffs(InterpState &S, CodePtr OpPC,
const InterpFrame *Frame,
const CallExpr *Call) {
@@ -3160,7 +3141,10 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
case Builtin::BI_rotl:
case Builtin::BI_lrotl:
case Builtin::BI_rotl64:
- return interp__builtin_rotate(S, OpPC, Frame, Call, /*Right=*/false);
+ return interp__builtin_elementwise_int_binop(
+ S, OpPC, Call, [](const APSInt &Value, const APSInt &Amount) -> APInt {
+ return Value.rotl(Amount);
+ });
case Builtin::BI__builtin_rotateright8:
case Builtin::BI__builtin_rotateright16:
@@ -3171,7 +3155,10 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const CallExpr *Call,
case Builtin::BI_rotr:
case Builtin::BI_lrotr:
case Builtin::BI_rotr64:
- return interp__builtin_rotate(S, OpPC, Frame, Call, /*Right=*/true);
+ return interp__builtin_elementwise_int_binop(
+ S, OpPC, Call, [](const APSInt &Value, const APSInt &Amount) -> APInt {
+ return Value.rotr(Amount);
+ });
case Builtin::BI__builtin_ffs:
case Builtin::BI__builtin_ffsl:
diff --git a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp
index bbc45e5..24a5fc2 100644
--- a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.cpp
@@ -221,10 +221,9 @@ mlir::Value OpenACCRecipeBuilderBase::makeBoundsAlloca(
return initialAlloca;
}
-mlir::Value
-OpenACCRecipeBuilderBase::createBoundsLoop(mlir::Value subscriptedValue,
- mlir::Value bound,
- mlir::Location loc, bool inverse) {
+std::pair<mlir::Value, mlir::Value> OpenACCRecipeBuilderBase::createBoundsLoop(
+ mlir::Value subscriptedValue, mlir::Value subscriptedValue2,
+ mlir::Value bound, mlir::Location loc, bool inverse) {
mlir::Operation *bodyInsertLoc;
mlir::Type itrTy = cgf.cgm.convertType(cgf.getContext().UnsignedLongLongTy);
@@ -249,7 +248,6 @@ OpenACCRecipeBuilderBase::createBoundsLoop(mlir::Value subscriptedValue,
return cir::PtrStrideOp::create(builder, loc, eltLoad.getType(), eltLoad,
idxLoad);
-
};
auto forStmtBuilder = [&]() {
@@ -303,6 +301,8 @@ OpenACCRecipeBuilderBase::createBoundsLoop(mlir::Value subscriptedValue,
if (subscriptedValue)
subscriptedValue = doSubscriptOp(subscriptedValue, load);
+ if (subscriptedValue2)
+ subscriptedValue2 = doSubscriptOp(subscriptedValue2, load);
bodyInsertLoc = builder.createYield(loc);
},
/*stepBuilder=*/
@@ -325,7 +325,7 @@ OpenACCRecipeBuilderBase::createBoundsLoop(mlir::Value subscriptedValue,
// Leave the insertion point to be inside the body, so we can loop over
// these things.
builder.setInsertionPoint(bodyInsertLoc);
- return subscriptedValue;
+ return {subscriptedValue, subscriptedValue2};
}
mlir::acc::ReductionOperator
@@ -434,7 +434,7 @@ void OpenACCRecipeBuilderBase::createInitRecipe(
mlir::Location loc, mlir::Location locEnd, SourceRange exprRange,
mlir::Value mainOp, mlir::Region &recipeInitRegion, size_t numBounds,
llvm::ArrayRef<QualType> boundTypes, const VarDecl *allocaDecl,
- QualType origType) {
+ QualType origType, bool emitInitExpr) {
assert(allocaDecl && "Required recipe variable not set?");
CIRGenFunction::DeclMapRevertingRAII declMapRAII{cgf, allocaDecl};
@@ -464,14 +464,15 @@ void OpenACCRecipeBuilderBase::createInitRecipe(
// initialize this variable correctly.
CIRGenFunction::AutoVarEmission tempDeclEmission =
cgf.emitAutoVarAlloca(*allocaDecl, builder.saveInsertionPoint());
- cgf.emitAutoVarInit(tempDeclEmission);
+ if (emitInitExpr)
+ cgf.emitAutoVarInit(tempDeclEmission);
} else {
mlir::Value alloca = makeBoundsAlloca(
block, exprRange, loc, allocaDecl->getName(), numBounds, boundTypes);
// If the initializer is trivial, there is nothing to do here, so save
// ourselves some effort.
- if (allocaDecl->getInit() &&
+ if (emitInitExpr && allocaDecl->getInit() &&
(!cgf.isTrivialInitializer(allocaDecl->getInit()) ||
cgf.getContext().getLangOpts().getTrivialAutoVarInit() !=
LangOptions::TrivialAutoVarInitKind::Uninitialized))
@@ -484,35 +485,42 @@ void OpenACCRecipeBuilderBase::createInitRecipe(
void OpenACCRecipeBuilderBase::createFirstprivateRecipeCopy(
mlir::Location loc, mlir::Location locEnd, mlir::Value mainOp,
- CIRGenFunction::AutoVarEmission tempDeclEmission,
- mlir::acc::FirstprivateRecipeOp recipe, const VarDecl *varRecipe,
- const VarDecl *temporary) {
- mlir::Block *block =
- createRecipeBlock(recipe.getCopyRegion(), mainOp.getType(), loc,
- /*numBounds=*/0, /*isInit=*/false);
- builder.setInsertionPointToEnd(&recipe.getCopyRegion().back());
+ const VarDecl *allocaDecl, const VarDecl *temporary,
+ mlir::Region &copyRegion, size_t numBounds) {
+ mlir::Block *block = createRecipeBlock(copyRegion, mainOp.getType(), loc,
+ numBounds, /*isInit=*/false);
+ builder.setInsertionPointToEnd(&copyRegion.back());
CIRGenFunction::LexicalScope ls(cgf, loc, block);
- mlir::BlockArgument fromArg = block->getArgument(0);
- mlir::BlockArgument toArg = block->getArgument(1);
+ mlir::Value fromArg = block->getArgument(0);
+ mlir::Value toArg = block->getArgument(1);
- mlir::Type elementTy =
- mlir::cast<cir::PointerType>(mainOp.getType()).getPointee();
+ llvm::MutableArrayRef<mlir::BlockArgument> boundsRange =
+ block->getArguments().drop_front(2);
- // Set the address of the emission to be the argument, so that we initialize
- // that instead of the variable in the other block.
- tempDeclEmission.setAllocatedAddress(
- Address{toArg, elementTy, cgf.getContext().getDeclAlign(varRecipe)});
+ for (mlir::BlockArgument boundArg : llvm::reverse(boundsRange))
+ std::tie(fromArg, toArg) =
+ createBoundsLoop(fromArg, toArg, boundArg, loc, /*inverse=*/false);
+
+ // Set up the 'to' address.
+ mlir::Type elementTy =
+ mlir::cast<cir::PointerType>(toArg.getType()).getPointee();
+ CIRGenFunction::AutoVarEmission tempDeclEmission(*allocaDecl);
tempDeclEmission.emittedAsOffload = true;
+ tempDeclEmission.setAllocatedAddress(
+ Address{toArg, elementTy, cgf.getContext().getDeclAlign(allocaDecl)});
+ // Set up the 'from' address from the temporary.
CIRGenFunction::DeclMapRevertingRAII declMapRAII{cgf, temporary};
cgf.setAddrOfLocalVar(
temporary,
- Address{fromArg, elementTy, cgf.getContext().getDeclAlign(varRecipe)});
-
+ Address{fromArg, elementTy, cgf.getContext().getDeclAlign(allocaDecl)});
cgf.emitAutoVarInit(tempDeclEmission);
+
+ builder.setInsertionPointToEnd(&copyRegion.back());
mlir::acc::YieldOp::create(builder, locEnd);
}
+
// This function generates the 'combiner' section for a reduction recipe. Note
// that this function is not 'insertion point' clean, in that it alters the
// insertion point to be inside of the 'combiner' section of the recipe, but
diff --git a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h
index 21707ad..a5da744 100644
--- a/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h
+++ b/clang/lib/CIR/CodeGen/CIRGenOpenACCRecipe.h
@@ -49,14 +49,16 @@ protected:
// Creates a loop through an 'acc.bounds', leaving the 'insertion' point to be
// the inside of the loop body. Traverses LB->UB UNLESS `inverse` is set.
// Returns the 'subscriptedValue' changed with the new bounds subscript.
+ std::pair<mlir::Value, mlir::Value>
+ createBoundsLoop(mlir::Value subscriptedValue, mlir::Value subscriptedValue2,
+ mlir::Value bound, mlir::Location loc, bool inverse);
+
mlir::Value createBoundsLoop(mlir::Value subscriptedValue, mlir::Value bound,
- mlir::Location loc, bool inverse);
+ mlir::Location loc, bool inverse) {
+ return createBoundsLoop(subscriptedValue, {}, bound, loc, inverse).first;
+ }
+
mlir::acc::ReductionOperator convertReductionOp(OpenACCReductionOperator op);
- void createFirstprivateRecipeCopy(
- mlir::Location loc, mlir::Location locEnd, mlir::Value mainOp,
- CIRGenFunction::AutoVarEmission tempDeclEmission,
- mlir::acc::FirstprivateRecipeOp recipe, const VarDecl *varRecipe,
- const VarDecl *temporary);
// This function generates the 'combiner' section for a reduction recipe. Note
// that this function is not 'insertion point' clean, in that it alters the
@@ -66,11 +68,19 @@ protected:
mlir::Value mainOp,
mlir::acc::ReductionRecipeOp recipe,
size_t numBounds);
+
void createInitRecipe(mlir::Location loc, mlir::Location locEnd,
SourceRange exprRange, mlir::Value mainOp,
mlir::Region &recipeInitRegion, size_t numBounds,
llvm::ArrayRef<QualType> boundTypes,
- const VarDecl *allocaDecl, QualType origType);
+ const VarDecl *allocaDecl, QualType origType,
+ bool emitInitExpr);
+
+ void createFirstprivateRecipeCopy(mlir::Location loc, mlir::Location locEnd,
+ mlir::Value mainOp,
+ const VarDecl *allocaDecl,
+ const VarDecl *temporary,
+ mlir::Region &copyRegion, size_t numBounds);
void createRecipeDestroySection(mlir::Location loc, mlir::Location locEnd,
mlir::Value mainOp, CharUnits alignment,
@@ -150,63 +160,6 @@ class OpenACCRecipeBuilder : OpenACCRecipeBuilderBase {
return recipeName;
}
- // Create the 'init' section of the recipe, including the 'copy' section for
- // 'firstprivate'. Note that this function is not 'insertion point' clean, in
- // that it alters the insertion point to be inside of the 'destroy' section of
- // the recipe, but doesn't restore it aftewards.
- void createRecipeInitCopy(mlir::Location loc, mlir::Location locEnd,
- SourceRange exprRange, mlir::Value mainOp,
- RecipeTy recipe, const VarDecl *varRecipe,
- const VarDecl *temporary) {
- // TODO: OpenACC: when we get the 'pointer' variants for
- // firstprivate/reduction, this probably should be removed/split into
- // functions for the BuilderBase.
- assert(varRecipe && "Required recipe variable not set?");
-
- CIRGenFunction::AutoVarEmission tempDeclEmission{
- CIRGenFunction::AutoVarEmission::invalid()};
- CIRGenFunction::DeclMapRevertingRAII declMapRAII{cgf, varRecipe};
-
- // Do the 'init' section of the recipe IR, which does an alloca, then the
- // initialization (except for firstprivate).
- mlir::Block *block =
- createRecipeBlock(recipe.getInitRegion(), mainOp.getType(), loc,
- /*numBounds=*/0, /*isInit=*/true);
- builder.setInsertionPointToEnd(&recipe.getInitRegion().back());
- CIRGenFunction::LexicalScope ls(cgf, loc, block);
-
- tempDeclEmission =
- cgf.emitAutoVarAlloca(*varRecipe, builder.saveInsertionPoint());
-
- // 'firstprivate' doesn't do its initialization in the 'init' section,
- // instead it does it in the 'copy' section. SO, only do 'init' here for
- // reduction.
- if constexpr (std::is_same_v<RecipeTy, mlir::acc::ReductionRecipeOp>) {
- // Unlike Private, the recipe here is always required as it has to do
- // init, not just 'default' init.
- if (!varRecipe->getInit())
- cgf.cgm.errorNYI(exprRange, "reduction init recipe");
- cgf.emitAutoVarInit(tempDeclEmission);
- }
-
- mlir::acc::YieldOp::create(builder, locEnd);
-
- if constexpr (std::is_same_v<RecipeTy, mlir::acc::FirstprivateRecipeOp>) {
- if (!varRecipe->getInit()) {
- // If we don't have any initialization recipe, we failed during Sema to
- // initialize this correctly. If we disable the
- // Sema::TentativeAnalysisScopes in SemaOpenACC::CreateInitRecipe, it'll
- // emit an error to tell us. However, emitting those errors during
- // production is a violation of the standard, so we cannot do them.
- cgf.cgm.errorNYI(
- exprRange, "firstprivate copy-init recipe not properly generated");
- }
-
- createFirstprivateRecipeCopy(loc, locEnd, mainOp, tempDeclEmission,
- recipe, varRecipe, temporary);
- }
- }
-
public:
OpenACCRecipeBuilder(CIRGen::CIRGenFunction &cgf,
CIRGen::CIRGenBuilderTy &builder)
@@ -221,19 +174,6 @@ public:
BuiltinType::ArraySection) &&
"array section shouldn't make it to recipe creation");
- // TODO: OpenACC: This is a bit of a hackery to get this to not change for
- // the non-private recipes. This will be removed soon, when we get this
- // 'right' for firstprivate and reduction.
- if constexpr (std::is_same_v<RecipeTy, mlir::acc::FirstprivateRecipeOp>) {
- if (numBounds) {
- cgf.cgm.errorNYI(varRef->getSourceRange(),
- "firstprivate-init with bounds");
- }
- boundTypes = {};
- numBounds = 0;
- origType = baseType;
- }
-
mlir::ModuleOp mod = builder.getBlock()
->getParent()
->template getParentOfType<mlir::ModuleOp>();
@@ -262,21 +202,20 @@ public:
if constexpr (std::is_same_v<RecipeTy, mlir::acc::PrivateRecipeOp>) {
createInitRecipe(loc, locEnd, varRef->getSourceRange(), mainOp,
recipe.getInitRegion(), numBounds, boundTypes, varRecipe,
- origType);
+ origType, /*emitInitExpr=*/true);
} else if constexpr (std::is_same_v<RecipeTy,
mlir::acc::ReductionRecipeOp>) {
createInitRecipe(loc, locEnd, varRef->getSourceRange(), mainOp,
recipe.getInitRegion(), numBounds, boundTypes, varRecipe,
- origType);
+ origType, /*emitInitExpr=*/true);
createReductionRecipeCombiner(loc, locEnd, mainOp, recipe, numBounds);
} else {
static_assert(std::is_same_v<RecipeTy, mlir::acc::FirstprivateRecipeOp>);
- // TODO: OpenACC: we probably want this to call createInitRecipe as well,
- // but do so in a way that omits the 'initialization', so that we can do
- // it separately, since it belongs in the 'copy' region. It also might
- // need a way of getting the tempDeclEmission out of it for that purpose.
- createRecipeInitCopy(loc, locEnd, varRef->getSourceRange(), mainOp,
- recipe, varRecipe, temporary);
+ createInitRecipe(loc, locEnd, varRef->getSourceRange(), mainOp,
+ recipe.getInitRegion(), numBounds, boundTypes, varRecipe,
+ origType, /*emitInitExpr=*/false);
+ createFirstprivateRecipeCopy(loc, locEnd, mainOp, varRecipe, temporary,
+ recipe.getCopyRegion(), numBounds);
}
if (origType.isDestructedType())
diff --git a/clang/lib/Headers/avx512vlintrin.h b/clang/lib/Headers/avx512vlintrin.h
index 754f43a..965741f 100644
--- a/clang/lib/Headers/avx512vlintrin.h
+++ b/clang/lib/Headers/avx512vlintrin.h
@@ -7330,9 +7330,8 @@ _mm256_mask_cvtusepi64_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
__builtin_ia32_pmovusqw256mem_mask ((__v8hi *) __P, (__v4di) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_cvtepi32_epi8 (__m128i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_cvtepi32_epi8(__m128i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v4si)__A, __v4qi), (__v4qi){0, 0, 0, 0}, 0, 1,
2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7);
@@ -7360,9 +7359,8 @@ _mm_mask_cvtepi32_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
__builtin_ia32_pmovdb128mem_mask ((__v16qi *) __P, (__v4si) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_cvtepi32_epi8 (__m256i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_cvtepi32_epi8(__m256i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v8si)__A, __v8qi),
(__v8qi){0, 0, 0, 0, 0, 0, 0, 0}, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11,
@@ -7370,8 +7368,7 @@ _mm256_cvtepi32_epi8 (__m256i __A)
}
static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_mask_cvtepi32_epi8 (__m128i __O, __mmask8 __M, __m256i __A)
-{
+_mm256_mask_cvtepi32_epi8(__m128i __O, __mmask8 __M, __m256i __A) {
return (__m128i) __builtin_ia32_pmovdb256_mask ((__v8si) __A,
(__v16qi) __O, __M);
}
@@ -7390,9 +7387,8 @@ _mm256_mask_cvtepi32_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
__builtin_ia32_pmovdb256mem_mask ((__v16qi *) __P, (__v8si) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_cvtepi32_epi16 (__m128i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_cvtepi32_epi16(__m128i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v4si)__A, __v4hi), (__v4hi){0, 0, 0, 0}, 0, 1,
2, 3, 4, 5, 6, 7);
@@ -7419,9 +7415,8 @@ _mm_mask_cvtepi32_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
__builtin_ia32_pmovdw128mem_mask ((__v8hi *) __P, (__v4si) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_cvtepi32_epi16 (__m256i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_cvtepi32_epi16(__m256i __A) {
return (__m128i)__builtin_convertvector((__v8si)__A, __v8hi);
}
@@ -7446,9 +7441,8 @@ _mm256_mask_cvtepi32_storeu_epi16 (void * __P, __mmask8 __M, __m256i __A)
__builtin_ia32_pmovdw256mem_mask ((__v8hi *) __P, (__v8si) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_cvtepi64_epi8 (__m128i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_cvtepi64_epi8(__m128i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v2di)__A, __v2qi), (__v2qi){0, 0}, 0, 1, 2, 3,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3);
@@ -7475,9 +7469,8 @@ _mm_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m128i __A)
__builtin_ia32_pmovqb128mem_mask ((__v16qi *) __P, (__v2di) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_cvtepi64_epi8 (__m256i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_cvtepi64_epi8(__m256i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v4di)__A, __v4qi), (__v4qi){0, 0, 0, 0}, 0, 1,
2, 3, 4, 5, 6, 7, 7, 7, 7, 7, 7, 7, 7, 7);
@@ -7504,9 +7497,8 @@ _mm256_mask_cvtepi64_storeu_epi8 (void * __P, __mmask8 __M, __m256i __A)
__builtin_ia32_pmovqb256mem_mask ((__v16qi *) __P, (__v4di) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_cvtepi64_epi32 (__m128i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_cvtepi64_epi32(__m128i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v2di)__A, __v2si), (__v2si){0, 0}, 0, 1, 2, 3);
}
@@ -7532,23 +7524,20 @@ _mm_mask_cvtepi64_storeu_epi32 (void * __P, __mmask8 __M, __m128i __A)
__builtin_ia32_pmovqd128mem_mask ((__v4si *) __P, (__v2di) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_cvtepi64_epi32 (__m256i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_cvtepi64_epi32(__m256i __A) {
return (__m128i)__builtin_convertvector((__v4di)__A, __v4si);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_mask_cvtepi64_epi32 (__m128i __O, __mmask8 __M, __m256i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_mask_cvtepi64_epi32(__m128i __O, __mmask8 __M, __m256i __A) {
return (__m128i)__builtin_ia32_selectd_128((__mmask8)__M,
(__v4si)_mm256_cvtepi64_epi32(__A),
(__v4si)__O);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_maskz_cvtepi64_epi32 (__mmask8 __M, __m256i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_maskz_cvtepi64_epi32(__mmask8 __M, __m256i __A) {
return (__m128i)__builtin_ia32_selectd_128((__mmask8)__M,
(__v4si)_mm256_cvtepi64_epi32(__A),
(__v4si)_mm_setzero_si128());
@@ -7560,9 +7549,8 @@ _mm256_mask_cvtepi64_storeu_epi32 (void * __P, __mmask8 __M, __m256i __A)
__builtin_ia32_pmovqd256mem_mask ((__v4si *) __P, (__v4di) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_cvtepi64_epi16 (__m128i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS128_CONSTEXPR
+_mm_cvtepi64_epi16(__m128i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v2di)__A, __v2hi), (__v2hi){0, 0}, 0, 1, 2, 3,
3, 3, 3, 3);
@@ -7590,9 +7578,8 @@ _mm_mask_cvtepi64_storeu_epi16 (void * __P, __mmask8 __M, __m128i __A)
__builtin_ia32_pmovqw128mem_mask ((__v8hi *) __P, (__v2di) __A, __M);
}
-static __inline__ __m128i __DEFAULT_FN_ATTRS256
-_mm256_cvtepi64_epi16 (__m256i __A)
-{
+static __inline__ __m128i __DEFAULT_FN_ATTRS256_CONSTEXPR
+_mm256_cvtepi64_epi16(__m256i __A) {
return (__m128i)__builtin_shufflevector(
__builtin_convertvector((__v4di)__A, __v4hi), (__v4hi){0, 0, 0, 0}, 0, 1,
2, 3, 4, 5, 6, 7);
diff --git a/clang/lib/Sema/SemaConcept.cpp b/clang/lib/Sema/SemaConcept.cpp
index 11d2d5c..999e302c 100644
--- a/clang/lib/Sema/SemaConcept.cpp
+++ b/clang/lib/Sema/SemaConcept.cpp
@@ -1049,6 +1049,7 @@ ExprResult ConstraintSatisfactionChecker::Evaluate(
case NormalizedConstraint::ConstraintKind::Compound:
return Evaluate(static_cast<const CompoundConstraint &>(Constraint), MLTAL);
}
+ llvm_unreachable("Unknown ConstraintKind enum");
}
static bool CheckConstraintSatisfaction(
@@ -2141,6 +2142,7 @@ bool SubstituteParameterMappings::substitute(NormalizedConstraint &N) {
return substitute(Compound.getRHS());
}
}
+ llvm_unreachable("Unknown ConstraintKind enum");
}
} // namespace
@@ -2561,7 +2563,6 @@ FormulaType SubsumptionChecker::Normalize(const NormalizedConstraint &NC) {
};
switch (NC.getKind()) {
-
case NormalizedConstraint::ConstraintKind::Atomic:
return {{find(&static_cast<const AtomicConstraint &>(NC))}};
@@ -2601,6 +2602,7 @@ FormulaType SubsumptionChecker::Normalize(const NormalizedConstraint &NC) {
return Res;
}
}
+ llvm_unreachable("Unknown ConstraintKind enum");
}
void SubsumptionChecker::AddUniqueClauseToFormula(Formula &F, Clause C) {
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 0069b08..6eaf7b9 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -11041,17 +11041,6 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
<< CUDA().getConfigureFuncName();
Context.setcudaConfigureCallDecl(NewFD);
}
-
- // Variadic functions, other than a *declaration* of printf, are not allowed
- // in device-side CUDA code, unless someone passed
- // -fcuda-allow-variadic-functions.
- if (!getLangOpts().CUDAAllowVariadicFunctions && NewFD->isVariadic() &&
- (NewFD->hasAttr<CUDADeviceAttr>() ||
- NewFD->hasAttr<CUDAGlobalAttr>()) &&
- !(II && II->isStr("printf") && NewFD->isExternC() &&
- !D.isFunctionDefinition())) {
- Diag(NewFD->getLocation(), diag::err_variadic_device_fn);
- }
}
MarkUnusedFileScopedDecl(NewFD);
diff --git a/clang/lib/Sema/SemaOpenACC.cpp b/clang/lib/Sema/SemaOpenACC.cpp
index 7ad7049..8471f02 100644
--- a/clang/lib/Sema/SemaOpenACC.cpp
+++ b/clang/lib/Sema/SemaOpenACC.cpp
@@ -2724,16 +2724,6 @@ Expr *GenerateReductionInitRecipeExpr(ASTContext &Context,
return InitExpr;
}
-const Expr *StripOffBounds(const Expr *VarExpr) {
- while (isa_and_present<ArraySectionExpr, ArraySubscriptExpr>(VarExpr)) {
- if (const auto *AS = dyn_cast<ArraySectionExpr>(VarExpr))
- VarExpr = AS->getBase()->IgnoreParenImpCasts();
- else if (const auto *Sub = dyn_cast<ArraySubscriptExpr>(VarExpr))
- VarExpr = Sub->getBase()->IgnoreParenImpCasts();
- }
- return VarExpr;
-}
-
VarDecl *CreateAllocaDecl(ASTContext &Ctx, DeclContext *DC,
SourceLocation BeginLoc, IdentifierInfo *VarName,
QualType VarTy) {
@@ -2794,17 +2784,18 @@ OpenACCPrivateRecipe SemaOpenACC::CreatePrivateInitRecipe(const Expr *VarExpr) {
OpenACCFirstPrivateRecipe
SemaOpenACC::CreateFirstPrivateInitRecipe(const Expr *VarExpr) {
- // TODO: OpenACC: This shouldn't be necessary, see PrivateInitRecipe
- VarExpr = StripOffBounds(VarExpr);
-
+ // We don't strip bounds here, so that we are doing our recipe init at the
+ // 'lowest' possible level. Codegen is going to have to do its own 'looping'.
if (!VarExpr || VarExpr->getType()->isDependentType())
return OpenACCFirstPrivateRecipe::Empty();
QualType VarTy =
VarExpr->getType().getNonReferenceType().getUnqualifiedType();
- // TODO: OpenACC: for arrays/bounds versions, we're going to have to do a
- // different initializer, but for now we can go ahead with this.
+ // Array sections are special, and we have to treat them that way.
+ if (const auto *ASE =
+ dyn_cast<ArraySectionExpr>(VarExpr->IgnoreParenImpCasts()))
+ VarTy = ArraySectionExpr::getBaseOriginalType(ASE);
VarDecl *AllocaDecl = CreateAllocaDecl(
getASTContext(), SemaRef.getCurContext(), VarExpr->getBeginLoc(),
diff --git a/clang/lib/Sema/SemaOverload.cpp b/clang/lib/Sema/SemaOverload.cpp
index b870114..5657dfe 100644
--- a/clang/lib/Sema/SemaOverload.cpp
+++ b/clang/lib/Sema/SemaOverload.cpp
@@ -4413,14 +4413,23 @@ CompareImplicitConversionSequences(Sema &S, SourceLocation Loc,
Result = CompareStandardConversionSequences(S, Loc,
ICS1.Standard, ICS2.Standard);
else if (ICS1.isUserDefined()) {
+ // With lazy template loading, it is possible to find non-canonical
+ // FunctionDecls, depending on when redecl chains are completed. Make sure
+ // to compare the canonical decls of conversion functions. This avoids
+ // ambiguity problems for templated conversion operators.
+ const FunctionDecl *ConvFunc1 = ICS1.UserDefined.ConversionFunction;
+ if (ConvFunc1)
+ ConvFunc1 = ConvFunc1->getCanonicalDecl();
+ const FunctionDecl *ConvFunc2 = ICS2.UserDefined.ConversionFunction;
+ if (ConvFunc2)
+ ConvFunc2 = ConvFunc2->getCanonicalDecl();
// User-defined conversion sequence U1 is a better conversion
// sequence than another user-defined conversion sequence U2 if
// they contain the same user-defined conversion function or
// constructor and if the second standard conversion sequence of
// U1 is better than the second standard conversion sequence of
// U2 (C++ 13.3.3.2p3).
- if (ICS1.UserDefined.ConversionFunction ==
- ICS2.UserDefined.ConversionFunction)
+ if (ConvFunc1 == ConvFunc2)
Result = CompareStandardConversionSequences(S, Loc,
ICS1.UserDefined.After,
ICS2.UserDefined.After);
generated by cgit v1.1 at 2025-10-06 17:23:30 +0000