Skip to content

Commit 55e27fd

Browse files
yuxuanchen1997yuxuanchen1997
committed
[Clang] Introduce [[clang::coro_await_elidable]]
1 parent 971a1ac commit 55e27fd

24 files changed

+335
-110
lines changed

‎clang/include/clang/AST/Expr.h

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2991,6 +2991,9 @@ class CallExpr : public Expr {
29912991

29922992
bool hasStoredFPFeatures() const { return CallExprBits.HasFPFeatures; }
29932993

2994+
bool isCoroMustElide() const { return CallExprBits.IsCoroMustElide; }
2995+
void setCoroMustElide(bool V = true) { CallExprBits.IsCoroMustElide = V; }
2996+
29942997
Decl *getCalleeDecl() { return getCallee()->getReferencedDeclOfCallee(); }
29952998
const Decl *getCalleeDecl() const {
29962999
return getCallee()->getReferencedDeclOfCallee();

‎clang/include/clang/AST/Stmt.h

Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -561,8 +561,11 @@ class alignas(void *) Stmt {
561561
LLVM_PREFERRED_TYPE(bool)
562562
unsigned HasFPFeatures : 1;
563563

564+
/// True if the call expression is a must-elide call to a coroutine.
565+
unsigned IsCoroMustElide : 1;
566+
564567
/// Padding used to align OffsetToTrailingObjects to a byte multiple.
565-
unsigned : 24 - 3 - NumExprBits;
568+
unsigned : 24 - 4 - NumExprBits;
566569

567570
/// The offset in bytes from the this pointer to the start of the
568571
/// trailing objects belonging to CallExpr. Intentionally byte sized

‎clang/include/clang/Basic/Attr.td

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1220,6 +1220,14 @@ def CoroDisableLifetimeBound : InheritableAttr {
12201220
let SimpleHandler = 1;
12211221
}
12221222

1223+
def CoroAwaitElidable : InheritableAttr {
1224+
let Spellings = [Clang<"coro_await_elidable">];
1225+
let Subjects = SubjectList<[CXXRecord]>;
1226+
let LangOpts = [CPlusPlus];
1227+
let Documentation = [CoroAwaitElidableDoc];
1228+
let SimpleHandler = 1;
1229+
}
1230+
12231231
// OSObject-based attributes.
12241232
def OSConsumed : InheritableParamAttr {
12251233
let Spellings = [Clang<"os_consumed">];

‎clang/include/clang/Basic/AttrDocs.td

Lines changed: 29 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8119,6 +8119,35 @@ but do not pass them to the underlying coroutine or pass them by value.
81198119
}];
81208120
}
81218121

8122+
def CoroAwaitElidableDoc : Documentation {
8123+
let Category = DocCatDecl;
8124+
let Content = [{
8125+
The ``[[clang::coro_await_elidable]]`` is a class attribute which can be applied
8126+
to a coroutine return type.
8127+
8128+
When a coroutine function that returns such a type calls another coroutine function,
8129+
the compiler performs heap allocation elision when the call to the coroutine function
8130+
is immediately co_awaited as a prvalue.
8131+
8132+
Example:
8133+
8134+
.. code-block:: c++
8135+
8136+
class [[clang::coro_await_elidable]] Task { ... };
8137+
8138+
Task foo();
8139+
Task bar() {
8140+
co_await foo(); // foo()'s coroutine frame on this line is elidable
8141+
auto t = foo(); // foo()'s coroutine frame on this line is NOT elidable
8142+
co_await t;
8143+
}
8144+
8145+
The behavior is undefined if the caller coroutine is destroyed earlier than the
8146+
callee coroutine.
8147+
8148+
}];
8149+
}
8150+
81228151
def CountedByDocs : Documentation {
81238152
let Category = DocCatField;
81248153
let Content = [{
@@ -8278,4 +8307,3 @@ Declares that a function potentially allocates heap memory, and prevents any pot
82788307
of ``nonallocating`` by the compiler.
82798308
}];
82808309
}
8281-

‎clang/lib/AST/Expr.cpp

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1474,6 +1474,7 @@ CallExpr::CallExpr(StmtClass SC, Expr *Fn, ArrayRef<Expr *> PreArgs,
14741474
this->computeDependence();
14751475

14761476
CallExprBits.HasFPFeatures = FPFeatures.requiresTrailingStorage();
1477+
CallExprBits.IsCoroMustElide = false;
14771478
if (hasStoredFPFeatures())
14781479
setStoredFPFeatures(FPFeatures);
14791480
}
@@ -1489,6 +1490,7 @@ CallExpr::CallExpr(StmtClass SC, unsigned NumPreArgs, unsigned NumArgs,
14891490
assert((CallExprBits.OffsetToTrailingObjects == OffsetToTrailingObjects) &&
14901491
"OffsetToTrailingObjects overflow!");
14911492
CallExprBits.HasFPFeatures = HasFPFeatures;
1493+
CallExprBits.IsCoroMustElide = false;
14921494
}
14931495

14941496
CallExpr *CallExpr::Create(const ASTContext &Ctx, Expr *Fn,

‎clang/lib/CodeGen/CGBlocks.cpp

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1163,7 +1163,8 @@ llvm::Type *CodeGenModule::getGenericBlockLiteralType() {
11631163
}
11641164

11651165
RValue CodeGenFunction::EmitBlockCallExpr(const CallExpr *E,
1166-
ReturnValueSlot ReturnValue) {
1166+
ReturnValueSlot ReturnValue,
1167+
llvm::CallBase **CallOrInvoke) {
11671168
const auto *BPT = E->getCallee()->getType()->castAs<BlockPointerType>();
11681169
llvm::Value *BlockPtr = EmitScalarExpr(E->getCallee());
11691170
llvm::Type *GenBlockTy = CGM.getGenericBlockLiteralType();
@@ -1220,7 +1221,7 @@ RValue CodeGenFunction::EmitBlockCallExpr(const CallExpr *E,
12201221
CGCallee Callee(CGCalleeInfo(), Func);
12211222

12221223
// And call the block.
1223-
return EmitCall(FnInfo, Callee, ReturnValue, Args);
1224+
return EmitCall(FnInfo, Callee, ReturnValue, Args, CallOrInvoke);
12241225
}
12251226

12261227
Address CodeGenFunction::GetAddrOfBlockDecl(const VarDecl *variable) {

‎clang/lib/CodeGen/CGCUDARuntime.cpp

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -25,7 +25,8 @@ CGCUDARuntime::~CGCUDARuntime() {}
2525

2626
RValue CGCUDARuntime::EmitCUDAKernelCallExpr(CodeGenFunction &CGF,
2727
const CUDAKernelCallExpr *E,
28-
ReturnValueSlot ReturnValue) {
28+
ReturnValueSlot ReturnValue,
29+
llvm::CallBase **CallOrInvoke) {
2930
llvm::BasicBlock *ConfigOKBlock = CGF.createBasicBlock("kcall.configok");
3031
llvm::BasicBlock *ContBlock = CGF.createBasicBlock("kcall.end");
3132

@@ -35,7 +36,7 @@ RValue CGCUDARuntime::EmitCUDAKernelCallExpr(CodeGenFunction &CGF,
3536

3637
eval.begin(CGF);
3738
CGF.EmitBlock(ConfigOKBlock);
38-
CGF.EmitSimpleCallExpr(E, ReturnValue);
39+
CGF.EmitSimpleCallExpr(E, ReturnValue, CallOrInvoke);
3940
CGF.EmitBranch(ContBlock);
4041

4142
CGF.EmitBlock(ContBlock);

‎clang/lib/CodeGen/CGCUDARuntime.h

Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -21,6 +21,7 @@
2121
#include "llvm/IR/GlobalValue.h"
2222

2323
namespace llvm {
24+
class CallBase;
2425
class Function;
2526
class GlobalVariable;
2627
}
@@ -82,9 +83,10 @@ class CGCUDARuntime {
8283
CGCUDARuntime(CodeGenModule &CGM) : CGM(CGM) {}
8384
virtual ~CGCUDARuntime();
8485

85-
virtual RValue EmitCUDAKernelCallExpr(CodeGenFunction &CGF,
86-
const CUDAKernelCallExpr *E,
87-
ReturnValueSlot ReturnValue);
86+
virtual RValue
87+
EmitCUDAKernelCallExpr(CodeGenFunction &CGF, const CUDAKernelCallExpr *E,
88+
ReturnValueSlot ReturnValue,
89+
llvm::CallBase **CallOrInvoke = nullptr);
8890

8991
/// Emits a kernel launch stub.
9092
virtual void emitDeviceStub(CodeGenFunction &CGF, FunctionArgList &Args) = 0;

‎clang/lib/CodeGen/CGCXXABI.h

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -485,11 +485,11 @@ class CGCXXABI {
485485
llvm::PointerUnion<const CXXDeleteExpr *, const CXXMemberCallExpr *>;
486486

487487
/// Emit the ABI-specific virtual destructor call.
488-
virtual llvm::Value *EmitVirtualDestructorCall(CodeGenFunction &CGF,
489-
const CXXDestructorDecl *Dtor,
490-
CXXDtorType DtorType,
491-
Address This,
492-
DeleteOrMemberCallExpr E) = 0;
488+
virtual llvm::Value *
489+
EmitVirtualDestructorCall(CodeGenFunction &CGF, const CXXDestructorDecl *Dtor,
490+
CXXDtorType DtorType, Address This,
491+
DeleteOrMemberCallExpr E,
492+
llvm::CallBase **CallOrInvoke) = 0;
493493

494494
virtual void adjustCallArgsForDestructorThunk(CodeGenFunction &CGF,
495495
GlobalDecl GD,

‎clang/lib/CodeGen/CGClass.cpp

Lines changed: 6 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -2192,15 +2192,11 @@ static bool canEmitDelegateCallArgs(CodeGenFunction &CGF,
21922192
return true;
21932193
}
21942194

2195-
void CodeGenFunction::EmitCXXConstructorCall(const CXXConstructorDecl *D,
2196-
CXXCtorType Type,
2197-
bool ForVirtualBase,
2198-
bool Delegating,
2199-
Address This,
2200-
CallArgList &Args,
2201-
AggValueSlot::Overlap_t Overlap,
2202-
SourceLocation Loc,
2203-
bool NewPointerIsChecked) {
2195+
void CodeGenFunction::EmitCXXConstructorCall(
2196+
const CXXConstructorDecl *D, CXXCtorType Type, bool ForVirtualBase,
2197+
bool Delegating, Address This, CallArgList &Args,
2198+
AggValueSlot::Overlap_t Overlap, SourceLocation Loc,
2199+
bool NewPointerIsChecked, llvm::CallBase **CallOrInvoke) {
22042200
const CXXRecordDecl *ClassDecl = D->getParent();
22052201

22062202
if (!NewPointerIsChecked)
@@ -2248,7 +2244,7 @@ void CodeGenFunction::EmitCXXConstructorCall(const CXXConstructorDecl *D,
22482244
const CGFunctionInfo &Info = CGM.getTypes().arrangeCXXConstructorCall(
22492245
Args, D, Type, ExtraArgs.Prefix, ExtraArgs.Suffix, PassPrototypeArgs);
22502246
CGCallee Callee = CGCallee::forDirect(CalleePtr, GlobalDecl(D, Type));
2251-
EmitCall(Info, Callee, ReturnValueSlot(), Args, nullptr, false, Loc);
2247+
EmitCall(Info, Callee, ReturnValueSlot(), Args, CallOrInvoke, false, Loc);
22522248

22532249
// Generate vtable assumptions if we're constructing a complete object
22542250
// with a vtable. We don't do this for base subobjects for two reasons:

0 commit comments

Comments
 (0)