[OpenACC] Implement 'firstprivate' clause copy lowering #154150
Merged
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
This patch is the last of the 'firstprivate' clause lowering patches. It takes the already generated 'copy' init from Sema and uses it to generate the IR for the copy section of the recipe. However, one thing that this patch had to do, was come up with a way to hijack the decl registration in CIRGenFunction. Because these decls are being created in a 'different' place, we need to remove the things we've added. We could alternatively generate these 'differently', but it seems worth a little extra effort here to avoid having to re-implement variable initialization.
erichkeane
requested review from
andykaylor,
bcardosolopes,
lanza and
xlauko
as code owners
llvmbot
added
clang
Member
|
@llvm/pr-subscribers-clang Author: Erich Keane (erichkeane) ChangesThis patch is the last of the 'firstprivate' clause lowering patches. It takes the already generated 'copy' init from Sema and uses it to generate the IR for the copy section of the recipe. However, one thing that this patch had to do, was come up with a way to hijack the decl registration in CIRGenFunction. Because these decls are being created in a 'different' place, we need to remove the things we've added. We could alternatively generate these 'differently', but it seems worth a little extra effort here to avoid having to re-implement variable initialization. Patch is 59.78 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/154150.diff 6 Files Affected:
diff --git a/clang/lib/CIR/CodeGen/CIRGenDecl.cpp b/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
index 715d1017e6bb9..baa37dbd84d76 100644
--- a/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
@@ -77,7 +77,7 @@ bool CIRGenFunction::isTrivialInitializer(const Expr *init) {
}
void CIRGenFunction::emitAutoVarInit(
- const CIRGenFunction::AutoVarEmission &emission) {
+ const CIRGenFunction::AutoVarEmission &emission, bool allocatedSeparately) {
assert(emission.Variable && "emission was not valid!");
// If this was emitted as a global constant, we're done.
@@ -159,9 +159,10 @@ void CIRGenFunction::emitAutoVarInit(
mlir::Value val = lv.getAddress().getPointer();
assert(val && "Should have an address");
auto allocaOp = val.getDefiningOp<cir::AllocaOp>();
- assert(allocaOp && "Address should come straight out of the alloca");
+ assert((allocatedSeparately || allocaOp) &&
+ "Address should come straight out of the alloca");
- if (!allocaOp.use_empty())
+ if (allocaOp && !allocaOp.use_empty())
allocaOp.setInitAttr(mlir::UnitAttr::get(&getMLIRContext()));
return;
}
diff --git a/clang/lib/CIR/CodeGen/CIRGenFunction.h b/clang/lib/CIR/CodeGen/CIRGenFunction.h
index 9a887ec047f86..4a9b259c5c009 100644
--- a/clang/lib/CIR/CodeGen/CIRGenFunction.h
+++ b/clang/lib/CIR/CodeGen/CIRGenFunction.h
@@ -487,6 +487,11 @@ class CIRGenFunction : public CIRGenTypeCache {
/// address space for address space agnostic languages.
Address getAllocatedAddress() const { return Addr; }
+ // Changes the stored address for the emission. This function should only
+ // be used in extreme cases, and isn't required to model normal AST
+ // initialization/variables.
+ void setAllocatedAddress(Address A) { Addr = A; }
+
/// Returns the address of the object within this declaration.
/// Note that this does not chase the forwarding pointer for
/// __block decls.
@@ -518,12 +523,19 @@ class CIRGenFunction : public CIRGenTypeCache {
symbolTable.insert(vd, addr.getPointer());
}
- /// Removes a declaration from the address-relationship. This is a function
- /// that shouldn't need to be used except in cases where we're adding/removing
- /// things that aren't part of the language-semantics AST.
- void removeAddrOfLocalVar(const clang::VarDecl *vd) {
- localDeclMap.erase(vd);
- }
+ // A class to allow inserting things into the declaration map during some sort
+ // of alternative generation (used currently for the OpenACC recipe
+ // generation), then reverting changes after the fact.
+ class DeclMapRevertingRAII {
+ CIRGenFunction &cgf;
+ CIRGenFunction::DeclMapTy originalMap;
+
+ public:
+ DeclMapRevertingRAII(CIRGenFunction &cgf)
+ : cgf(cgf), originalMap(cgf.localDeclMap) {}
+
+ ~DeclMapRevertingRAII() { cgf.localDeclMap = std::move(originalMap); }
+ };
bool shouldNullCheckClassCastValue(const CastExpr *ce);
@@ -961,7 +973,13 @@ class CIRGenFunction : public CIRGenTypeCache {
void emitAutoVarDecl(const clang::VarDecl &d);
void emitAutoVarCleanups(const AutoVarEmission &emission);
- void emitAutoVarInit(const AutoVarEmission &emission);
+ /// Emit the initializer for an allocated variable. If this call is not
+ /// associated with the call to emitAutoVarAlloca (as the address of the
+ /// emission is not directly an alloca), the allocatedSeparately parameter can
+ /// be used to suppress the assertions. However, this should only be used in
+ /// extreme cases, as it doesn't properly reflect the language/AST.
+ void emitAutoVarInit(const AutoVarEmission &emission,
+ bool allocatedSeparately = false);
void emitAutoVarTypeCleanup(const AutoVarEmission &emission,
clang::QualType::DestructionKind dtorKind);
diff --git a/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp b/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp
index 72e2c533254c9..958a974b05d69 100644
--- a/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp
@@ -392,12 +392,27 @@ class OpenACCClauseCIREmitter final
CIRGenFunction::AutoVarEmission tempDeclEmission,
mlir::acc::FirstprivateRecipeOp recipe, const VarDecl *varRecipe,
const VarDecl *temporary) {
- builder.createBlock(&recipe.getCopyRegion(), recipe.getCopyRegion().end(),
- {mainOp.getType(), mainOp.getType()}, {loc, loc});
+ mlir::Block *block = builder.createBlock(
+ &recipe.getCopyRegion(), recipe.getCopyRegion().end(),
+ {mainOp.getType(), mainOp.getType()}, {loc, loc});
builder.setInsertionPointToEnd(&recipe.getCopyRegion().back());
- // TODO: OpenACC: Implement this copy to actually do something.
+ mlir::BlockArgument fromArg = block->getArgument(0);
+ mlir::BlockArgument toArg = block->getArgument(1);
+
+ mlir::Type elementTy =
+ mlir::cast<cir::PointerType>(mainOp.getType()).getPointee();
+
+ // 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)});
+
+ cgf.setAddrOfLocalVar(
+ temporary,
+ Address{fromArg, elementTy, cgf.getContext().getDeclAlign(varRecipe)});
+ cgf.emitAutoVarInit(tempDeclEmission, /*allocatedSeparately=*/true);
mlir::acc::YieldOp::create(builder, locEnd);
}
@@ -417,6 +432,7 @@ class OpenACCClauseCIREmitter final
CIRGenFunction::AutoVarEmission tempDeclEmission{
CIRGenFunction::AutoVarEmission::invalid()};
+ CIRGenFunction::DeclMapRevertingRAII declMapRAII{cgf};
// Do the 'init' section of the recipe IR, which does an alloca, then the
// initialization (except for firstprivate).
@@ -425,6 +441,7 @@ class OpenACCClauseCIREmitter final
builder.setInsertionPointToEnd(&recipe.getInitRegion().back());
tempDeclEmission =
cgf.emitAutoVarAlloca(*varRecipe, builder.saveInsertionPoint());
+
// 'firstprivate' doesn't do its initialization in the 'init' section,
// instead does it in the 'copy' section. SO only do init here.
// 'reduction' appears to use it too (rather than a 'copy' section), so
@@ -450,16 +467,19 @@ class OpenACCClauseCIREmitter final
mlir::acc::YieldOp::create(builder, locEnd);
if constexpr (std::is_same_v<RecipeTy, mlir::acc::FirstprivateRecipeOp>) {
- // TODO: OpenACC: we should have a errorNYI call here if
- // !varRecipe->getInit(), but as that generation isn't currently
- // implemented, it ends up being too noisy. So when we implement copy-init
- // generation both in Sema and here, we should have a diagnostic here.
+ 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);
}
-
- // Make sure we cleanup after ourselves here.
- cgf.removeAddrOfLocalVar(varRecipe);
}
void createRecipeDestroySection(mlir::Location loc, mlir::Location locEnd,
diff --git a/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp b/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
index 7571e5e3306f7..64e5a73667fff 100644
--- a/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
@@ -21,6 +21,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!cir.array<!rec_HasDtor x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_HasDtor x 5>> {{.*}}):
+// CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[ZERO]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[DECAY_TO]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[ONE]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[ONE_2:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[ONE_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[TWO:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[TWO]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[TWO_2:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[TWO_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[THREE:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[THREE]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[THREE_2:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[THREE_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[FOUR:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[FOUR]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FOUR_2:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[FOUR_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
//
// CHECK-NEXT: acc.yield
//
@@ -52,6 +85,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FORM:.*]]: !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>> {{.*}}):
+// CHECK-NEXT: %[[TO_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[FROM_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[FROM_DECAY]] : !cir.ptr<!rec_NonDefaultCtor>, %[[ZERO]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_DECAY]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[ONE]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[ONE_2:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[ONE_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[TWO:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[TWO]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[TWO_2:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[TWO_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[THREE:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[THREE]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[THREE_2:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[THREE_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[FOUR:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[FOUR]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FOUR_2:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[FOUR_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
//
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
@@ -62,6 +128,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FORM:.*]]: !cir.ptr<!cir.array<!rec_CopyConstruct x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_CopyConstruct x 5>> {{.*}}):
+// CHECK-NEXT: %[[TO_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[FROM_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[FROM_DECAY]] : !cir.ptr<!rec_CopyConstruct>, %[[ZERO]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_DECAY]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[ONE]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[ONE_2:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[ONE_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[TWO:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[TWO]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[TWO_2:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[TWO_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[THREE:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[THREE]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[THREE_2:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[THREE_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[FOUR:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[FOUR]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FOUR_2:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[FOUR_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
//
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
@@ -72,6 +171,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FORM:.*]]: !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>> {{.*}}):
+// CHECK-NEXT: %[[TO_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>>), !cir.ptr<!rec_NoCopyConstruct>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[FROM_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>>), !cir.ptr<!rec_NoCopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[FROM_DECAY]] : !cir.ptr<!rec_NoCopyConstruct>, %[[ZERO]] : !u64i), !cir.ptr<!rec_NoCopyConstruct>
+// CHECK-NEXT: cir.call @_ZN15NoCopyConstructC1ERKS_(%[[TO_DECAY]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NoCopyConstruct>, !cir.ptr<!rec_NoCopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[D...
[truncated]
|
Member
|
@llvm/pr-subscribers-clangir Author: Erich Keane (erichkeane) ChangesThis patch is the last of the 'firstprivate' clause lowering patches. It takes the already generated 'copy' init from Sema and uses it to generate the IR for the copy section of the recipe. However, one thing that this patch had to do, was come up with a way to hijack the decl registration in CIRGenFunction. Because these decls are being created in a 'different' place, we need to remove the things we've added. We could alternatively generate these 'differently', but it seems worth a little extra effort here to avoid having to re-implement variable initialization. Patch is 59.78 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/154150.diff 6 Files Affected:
diff --git a/clang/lib/CIR/CodeGen/CIRGenDecl.cpp b/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
index 715d1017e6bb9..baa37dbd84d76 100644
--- a/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
@@ -77,7 +77,7 @@ bool CIRGenFunction::isTrivialInitializer(const Expr *init) {
}
void CIRGenFunction::emitAutoVarInit(
- const CIRGenFunction::AutoVarEmission &emission) {
+ const CIRGenFunction::AutoVarEmission &emission, bool allocatedSeparately) {
assert(emission.Variable && "emission was not valid!");
// If this was emitted as a global constant, we're done.
@@ -159,9 +159,10 @@ void CIRGenFunction::emitAutoVarInit(
mlir::Value val = lv.getAddress().getPointer();
assert(val && "Should have an address");
auto allocaOp = val.getDefiningOp<cir::AllocaOp>();
- assert(allocaOp && "Address should come straight out of the alloca");
+ assert((allocatedSeparately || allocaOp) &&
+ "Address should come straight out of the alloca");
- if (!allocaOp.use_empty())
+ if (allocaOp && !allocaOp.use_empty())
allocaOp.setInitAttr(mlir::UnitAttr::get(&getMLIRContext()));
return;
}
diff --git a/clang/lib/CIR/CodeGen/CIRGenFunction.h b/clang/lib/CIR/CodeGen/CIRGenFunction.h
index 9a887ec047f86..4a9b259c5c009 100644
--- a/clang/lib/CIR/CodeGen/CIRGenFunction.h
+++ b/clang/lib/CIR/CodeGen/CIRGenFunction.h
@@ -487,6 +487,11 @@ class CIRGenFunction : public CIRGenTypeCache {
/// address space for address space agnostic languages.
Address getAllocatedAddress() const { return Addr; }
+ // Changes the stored address for the emission. This function should only
+ // be used in extreme cases, and isn't required to model normal AST
+ // initialization/variables.
+ void setAllocatedAddress(Address A) { Addr = A; }
+
/// Returns the address of the object within this declaration.
/// Note that this does not chase the forwarding pointer for
/// __block decls.
@@ -518,12 +523,19 @@ class CIRGenFunction : public CIRGenTypeCache {
symbolTable.insert(vd, addr.getPointer());
}
- /// Removes a declaration from the address-relationship. This is a function
- /// that shouldn't need to be used except in cases where we're adding/removing
- /// things that aren't part of the language-semantics AST.
- void removeAddrOfLocalVar(const clang::VarDecl *vd) {
- localDeclMap.erase(vd);
- }
+ // A class to allow inserting things into the declaration map during some sort
+ // of alternative generation (used currently for the OpenACC recipe
+ // generation), then reverting changes after the fact.
+ class DeclMapRevertingRAII {
+ CIRGenFunction &cgf;
+ CIRGenFunction::DeclMapTy originalMap;
+
+ public:
+ DeclMapRevertingRAII(CIRGenFunction &cgf)
+ : cgf(cgf), originalMap(cgf.localDeclMap) {}
+
+ ~DeclMapRevertingRAII() { cgf.localDeclMap = std::move(originalMap); }
+ };
bool shouldNullCheckClassCastValue(const CastExpr *ce);
@@ -961,7 +973,13 @@ class CIRGenFunction : public CIRGenTypeCache {
void emitAutoVarDecl(const clang::VarDecl &d);
void emitAutoVarCleanups(const AutoVarEmission &emission);
- void emitAutoVarInit(const AutoVarEmission &emission);
+ /// Emit the initializer for an allocated variable. If this call is not
+ /// associated with the call to emitAutoVarAlloca (as the address of the
+ /// emission is not directly an alloca), the allocatedSeparately parameter can
+ /// be used to suppress the assertions. However, this should only be used in
+ /// extreme cases, as it doesn't properly reflect the language/AST.
+ void emitAutoVarInit(const AutoVarEmission &emission,
+ bool allocatedSeparately = false);
void emitAutoVarTypeCleanup(const AutoVarEmission &emission,
clang::QualType::DestructionKind dtorKind);
diff --git a/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp b/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp
index 72e2c533254c9..958a974b05d69 100644
--- a/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenOpenACCClause.cpp
@@ -392,12 +392,27 @@ class OpenACCClauseCIREmitter final
CIRGenFunction::AutoVarEmission tempDeclEmission,
mlir::acc::FirstprivateRecipeOp recipe, const VarDecl *varRecipe,
const VarDecl *temporary) {
- builder.createBlock(&recipe.getCopyRegion(), recipe.getCopyRegion().end(),
- {mainOp.getType(), mainOp.getType()}, {loc, loc});
+ mlir::Block *block = builder.createBlock(
+ &recipe.getCopyRegion(), recipe.getCopyRegion().end(),
+ {mainOp.getType(), mainOp.getType()}, {loc, loc});
builder.setInsertionPointToEnd(&recipe.getCopyRegion().back());
- // TODO: OpenACC: Implement this copy to actually do something.
+ mlir::BlockArgument fromArg = block->getArgument(0);
+ mlir::BlockArgument toArg = block->getArgument(1);
+
+ mlir::Type elementTy =
+ mlir::cast<cir::PointerType>(mainOp.getType()).getPointee();
+
+ // 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)});
+
+ cgf.setAddrOfLocalVar(
+ temporary,
+ Address{fromArg, elementTy, cgf.getContext().getDeclAlign(varRecipe)});
+ cgf.emitAutoVarInit(tempDeclEmission, /*allocatedSeparately=*/true);
mlir::acc::YieldOp::create(builder, locEnd);
}
@@ -417,6 +432,7 @@ class OpenACCClauseCIREmitter final
CIRGenFunction::AutoVarEmission tempDeclEmission{
CIRGenFunction::AutoVarEmission::invalid()};
+ CIRGenFunction::DeclMapRevertingRAII declMapRAII{cgf};
// Do the 'init' section of the recipe IR, which does an alloca, then the
// initialization (except for firstprivate).
@@ -425,6 +441,7 @@ class OpenACCClauseCIREmitter final
builder.setInsertionPointToEnd(&recipe.getInitRegion().back());
tempDeclEmission =
cgf.emitAutoVarAlloca(*varRecipe, builder.saveInsertionPoint());
+
// 'firstprivate' doesn't do its initialization in the 'init' section,
// instead does it in the 'copy' section. SO only do init here.
// 'reduction' appears to use it too (rather than a 'copy' section), so
@@ -450,16 +467,19 @@ class OpenACCClauseCIREmitter final
mlir::acc::YieldOp::create(builder, locEnd);
if constexpr (std::is_same_v<RecipeTy, mlir::acc::FirstprivateRecipeOp>) {
- // TODO: OpenACC: we should have a errorNYI call here if
- // !varRecipe->getInit(), but as that generation isn't currently
- // implemented, it ends up being too noisy. So when we implement copy-init
- // generation both in Sema and here, we should have a diagnostic here.
+ 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);
}
-
- // Make sure we cleanup after ourselves here.
- cgf.removeAddrOfLocalVar(varRecipe);
}
void createRecipeDestroySection(mlir::Location loc, mlir::Location locEnd,
diff --git a/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp b/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
index 7571e5e3306f7..64e5a73667fff 100644
--- a/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
+++ b/clang/test/CIR/CodeGenOpenACC/combined-firstprivate-clause.cpp
@@ -21,6 +21,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FROM:.*]]: !cir.ptr<!cir.array<!rec_HasDtor x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_HasDtor x 5>> {{.*}}):
+// CHECK-NEXT: %[[DECAY_TO:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[ZERO]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[DECAY_TO]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[ONE]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[ONE_2:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[ONE_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[TWO:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[TWO]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[TWO_2:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[TWO_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[THREE:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[THREE]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[THREE_2:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[THREE_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
+//
+// CHECK-NEXT: %[[FOUR:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_HasDtor>, %[[FOUR]] : !s64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FOUR_2:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_HasDtor x 5>>), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_HasDtor>, %[[FOUR_2]] : !u64i), !cir.ptr<!rec_HasDtor>
+// CHECK-NEXT: cir.call @_ZN7HasDtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_HasDtor>, !cir.ptr<!rec_HasDtor>) -> ()
//
// CHECK-NEXT: acc.yield
//
@@ -52,6 +85,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FORM:.*]]: !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>> {{.*}}):
+// CHECK-NEXT: %[[TO_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[FROM_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[FROM_DECAY]] : !cir.ptr<!rec_NonDefaultCtor>, %[[ZERO]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_DECAY]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[ONE]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[ONE_2:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[ONE_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[TWO:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[TWO]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[TWO_2:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[TWO_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[THREE:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[THREE]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[THREE_2:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[THREE_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
+//
+// CHECK-NEXT: %[[FOUR:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_NonDefaultCtor>, %[[FOUR]] : !s64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FOUR_2:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NonDefaultCtor x 5>>), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_NonDefaultCtor>, %[[FOUR_2]] : !u64i), !cir.ptr<!rec_NonDefaultCtor>
+// CHECK-NEXT: cir.call @_ZN14NonDefaultCtorC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NonDefaultCtor>, !cir.ptr<!rec_NonDefaultCtor>) -> ()
//
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
@@ -62,6 +128,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FORM:.*]]: !cir.ptr<!cir.array<!rec_CopyConstruct x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_CopyConstruct x 5>> {{.*}}):
+// CHECK-NEXT: %[[TO_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[FROM_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[FROM_DECAY]] : !cir.ptr<!rec_CopyConstruct>, %[[ZERO]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_DECAY]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[ONE]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[ONE_2:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[ONE_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[TWO:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[TWO]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[TWO_2:.*]] = cir.const #cir.int<2>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[TWO_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[THREE:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[THREE]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[THREE_2:.*]] = cir.const #cir.int<3>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[THREE_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[FOUR:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_TO]] : !cir.ptr<!rec_CopyConstruct>, %[[FOUR]] : !s64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FOUR_2:.*]] = cir.const #cir.int<4>
+// CHECK-NEXT: %[[DECAY_FROM:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_CopyConstruct x 5>>), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[DECAY_FROM]] : !cir.ptr<!rec_CopyConstruct>, %[[FOUR_2]] : !u64i), !cir.ptr<!rec_CopyConstruct>
+// CHECK-NEXT: cir.call @_ZN13CopyConstructC1ERKS_(%[[TO_OFFSET]], %[[FROM_OFFSET]]) : (!cir.ptr<!rec_CopyConstruct>, !cir.ptr<!rec_CopyConstruct>) -> ()
//
// CHECK-NEXT: acc.yield
// CHECK-NEXT: }
@@ -72,6 +171,39 @@ struct HasDtor {
// CHECK-NEXT: acc.yield
// CHECK-NEXT: } copy {
// CHECK-NEXT: ^bb0(%[[ARG_FORM:.*]]: !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>> {{.*}}, %[[ARG_TO:.*]]: !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>> {{.*}}):
+// CHECK-NEXT: %[[TO_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_TO]] : !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>>), !cir.ptr<!rec_NoCopyConstruct>
+// CHECK-NEXT: %[[ZERO:.*]] = cir.const #cir.int<0>
+// CHECK-NEXT: %[[FROM_DECAY:.*]] = cir.cast(array_to_ptrdecay, %[[ARG_FROM]] : !cir.ptr<!cir.array<!rec_NoCopyConstruct x 5>>), !cir.ptr<!rec_NoCopyConstruct>
+// CHECK-NEXT: %[[FROM_OFFSET:.*]] = cir.ptr_stride(%[[FROM_DECAY]] : !cir.ptr<!rec_NoCopyConstruct>, %[[ZERO]] : !u64i), !cir.ptr<!rec_NoCopyConstruct>
+// CHECK-NEXT: cir.call @_ZN15NoCopyConstructC1ERKS_(%[[TO_DECAY]], %[[FROM_OFFSET]]) nothrow : (!cir.ptr<!rec_NoCopyConstruct>, !cir.ptr<!rec_NoCopyConstruct>) -> ()
+//
+// CHECK-NEXT: %[[ONE:.*]] = cir.const #cir.int<1>
+// CHECK-NEXT: %[[TO_OFFSET:.*]] = cir.ptr_stride(%[[D...
[truncated]
|
andykaylor
approved these changes
Aug 18, 2025
Contributor
andykaylor
left a comment
andykaylor
left a comment
There was a problem hiding this comment.
This looks good. I just have one suggestion about how to represent the different different emission behavior.
clang/lib/CIR/CodeGen/CIRGenDecl.cpp
Outdated
|
|
||
| void CIRGenFunction::emitAutoVarInit( | ||
| const CIRGenFunction::AutoVarEmission &emission) { | ||
| const CIRGenFunction::AutoVarEmission &emission, bool allocatedSeparately) { |
Contributor
There was a problem hiding this comment.
Rather than passing this in as an argument here, could you make it a member of AutoVarEmission?
clang/lib/CIR/CodeGen/CIRGenDecl.cpp
Outdated
| "Address should come straight out of the alloca"); | ||
|
|
||
| if (!allocaOp.use_empty()) | ||
| if (allocaOp && !allocaOp.use_empty()) |
Contributor
There was a problem hiding this comment.
Maybe put everything from line 161 through line 166 under if (emission.wasEmittedAsOffloadClause()) or something similar?
| // A class to allow inserting things into the declaration map during some sort | ||
| // of alternative generation (used currently for the OpenACC recipe | ||
| // generation), then reverting changes after the fact. | ||
| class DeclMapRevertingRAII { |
Contributor
There was a problem hiding this comment.
This seems to have a similar purpose to ParamReferenceReplacerRAII in the incubator. That one requires explicitly adding replacement variables and only removes the explicit mappings. The advantage is that it doesn't require copying the entire map.
In the uses here, do you expect any addresses to be added to the map other than clause parameters?
Collaborator
Author
There was a problem hiding this comment.
I don't really expect ANYTHING other than the clause parameters, but it seemed less error prone than making the add/remove the same operation. That said, I can probably make this 'less intrusive' than a copy + move by making me just record the decl that I'd like to replace the value of.
Contributor
There was a problem hiding this comment.
I could live with it either way, but it would be nice if we could somehow combine this with the coroutine parameter handling when that gets upstreamed.
Collaborator
Author
There was a problem hiding this comment.
Ping me when that gets to review, and I'll work with the person to combine them. At the moment, I'm a bit too 'simple' for that, but when that goes under review I'll work with the uathor to combine the uses.
inside AutoVarEmission instead.
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
bcardosolopes
approved these changes
Aug 19, 2025
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.
This patch is the last of the 'firstprivate' clause lowering patches. It takes the already generated 'copy' init from Sema and uses it to generate the IR for the copy section of the recipe.
However, one thing that this patch had to do, was come up with a way to hijack the decl registration in CIRGenFunction. Because these decls are being created in a 'different' place, we need to remove the things we've added. We could alternatively generate these 'differently', but it seems worth a little extra effort here to avoid having to re-implement variable initialization.