Skip to content

Commit 92966fb

Browse files
yuxuanchen1997yuxuanchen1997
committed
[Clang] Introduce [[clang::coro_await_elidable]]
1 parent cbd3024 commit 92966fb

26 files changed

+351
-110
lines changed

‎clang/docs/ReleaseNotes.rst

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -209,6 +209,9 @@ Attribute Changes in Clang
209209
- ``[[clang::lifetimebound]]`` is now explicitly disallowed on explicit object member functions
210210
where they were previously silently ignored.
211211

212+
- Introduced a new attribute ``[[clang::coro_await_elidable]]`` on coroutine return types
213+
to express elideability at call sites where the coroutine is co_awaited as a prvalue.
214+
212215
Improvements to Clang's diagnostics
213216
-----------------------------------
214217

‎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: 32 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -8173,6 +8173,38 @@ but do not pass them to the underlying coroutine or pass them by value.
81738173
}];
81748174
}
81758175

8176+
def CoroAwaitElidableDoc : Documentation {
8177+
let Category = DocCatDecl;
8178+
let Content = [{
8179+
The ``[[clang::coro_await_elidable]]`` is a class attribute which can be applied
8180+
to a coroutine return type.
8181+
8182+
When a coroutine function that returns such a type calls another coroutine function,
8183+
the compiler performs heap allocation elision when the call to the coroutine function
8184+
is immediately co_awaited as a prvalue. In this case, the coroutine frame for the
8185+
callee will be a local variable within the enclosing braces in the caller's stack
8186+
frame. And the local variable, like other variables in coroutines, may be collected
8187+
into the coroutine frame, which may be allocated on the heap.
8188+
8189+
Example:
8190+
8191+
.. code-block:: c++
8192+
8193+
class [[clang::coro_await_elidable]] Task { ... };
8194+
8195+
Task foo();
8196+
Task bar() {
8197+
co_await foo(); // foo()'s coroutine frame on this line is elidable
8198+
auto t = foo(); // foo()'s coroutine frame on this line is NOT elidable
8199+
co_await t;
8200+
}
8201+
8202+
The behavior is undefined if the caller coroutine is destroyed earlier than the
8203+
callee coroutine.
8204+
8205+
}];
8206+
}
8207+
81768208
def CountedByDocs : Documentation {
81778209
let Category = DocCatField;
81788210
let Content = [{
@@ -8332,4 +8364,3 @@ Declares that a function potentially allocates heap memory, and prevents any pot
83328364
of ``nonallocating`` by the compiler.
83338365
}];
83348366
}
8335-

‎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:

‎clang/lib/CodeGen/CGExpr.cpp

Lines changed: 39 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -33,6 +33,7 @@
3333
#include "clang/Basic/SourceManager.h"
3434
#include "llvm/ADT/Hashing.h"
3535
#include "llvm/ADT/STLExtras.h"
36+
#include "llvm/ADT/ScopeExit.h"
3637
#include "llvm/ADT/StringExtras.h"
3738
#include "llvm/IR/DataLayout.h"
3839
#include "llvm/IR/Intrinsics.h"
@@ -5478,24 +5479,38 @@ RValue CodeGenFunction::EmitRValueForField(LValue LV,
54785479
//===--------------------------------------------------------------------===//
54795480

54805481
RValue CodeGenFunction::EmitCallExpr(const CallExpr *E,
5481-
ReturnValueSlot ReturnValue) {
5482+
ReturnValueSlot ReturnValue,
5483+
llvm::CallBase **CallOrInvoke) {
5484+
llvm::CallBase *CallOrInvokeStorage;
5485+
if (!CallOrInvoke) {
5486+
CallOrInvoke = &CallOrInvokeStorage;
5487+
}
5488+
5489+
auto AddCoroMustElideOnExit = llvm::make_scope_exit([&] {
5490+
if (E->isCoroMustElide()) {
5491+
auto *I = *CallOrInvoke;
5492+
if (I)
5493+
I->addFnAttr(llvm::Attribute::CoroMustElide);
5494+
}
5495+
});
5496+
54825497
// Builtins never have block type.
54835498
if (E->getCallee()->getType()->isBlockPointerType())
5484-
return EmitBlockCallExpr(E, ReturnValue);
5499+
return EmitBlockCallExpr(E, ReturnValue, CallOrInvoke);
54855500

54865501
if (const auto *CE = dyn_cast<CXXMemberCallExpr>(E))
5487-
return EmitCXXMemberCallExpr(CE, ReturnValue);
5502+
return EmitCXXMemberCallExpr(CE, ReturnValue, CallOrInvoke);
54885503

54895504
if (const auto *CE = dyn_cast<CUDAKernelCallExpr>(E))
5490-
return EmitCUDAKernelCallExpr(CE, ReturnValue);
5505+
return EmitCUDAKernelCallExpr(CE, ReturnValue, CallOrInvoke);
54915506

54925507
// A CXXOperatorCallExpr is created even for explicit object methods, but
54935508
// these should be treated like static function call.
54945509
if (const auto *CE = dyn_cast<CXXOperatorCallExpr>(E))
54955510
if (const auto *MD =
54965511
dyn_cast_if_present<CXXMethodDecl>(CE->getCalleeDecl());
54975512
MD && MD->isImplicitObjectMemberFunction())
5498-
return EmitCXXOperatorMemberCallExpr(CE, MD, ReturnValue);
5513+
return EmitCXXOperatorMemberCallExpr(CE, MD, ReturnValue, CallOrInvoke);
54995514

55005515
CGCallee callee = EmitCallee(E->getCallee());
55015516

@@ -5508,14 +5523,17 @@ RValue CodeGenFunction::EmitCallExpr(const CallExpr *E,
55085523
return EmitCXXPseudoDestructorExpr(callee.getPseudoDestructorExpr());
55095524
}
55105525

5511-
return EmitCall(E->getCallee()->getType(), callee, E, ReturnValue);
5526+
return EmitCall(E->getCallee()->getType(), callee, E, ReturnValue,
5527+
/*Chain=*/nullptr, CallOrInvoke);
55125528
}
55135529

55145530
/// Emit a CallExpr without considering whether it might be a subclass.
55155531
RValue CodeGenFunction::EmitSimpleCallExpr(const CallExpr *E,
5516-
ReturnValueSlot ReturnValue) {
5532+
ReturnValueSlot ReturnValue,
5533+
llvm::CallBase **CallOrInvoke) {
55175534
CGCallee Callee = EmitCallee(E->getCallee());
5518-
return EmitCall(E->getCallee()->getType(), Callee, E, ReturnValue);
5535+
return EmitCall(E->getCallee()->getType(), Callee, E, ReturnValue,
5536+
/*Chain=*/nullptr, CallOrInvoke);
55195537
}
55205538

55215539
// Detect the unusual situation where an inline version is shadowed by a
@@ -5719,8 +5737,9 @@ LValue CodeGenFunction::EmitBinaryOperatorLValue(const BinaryOperator *E) {
57195737
llvm_unreachable("bad evaluation kind");
57205738
}
57215739

5722-
LValue CodeGenFunction::EmitCallExprLValue(const CallExpr *E) {
5723-
RValue RV = EmitCallExpr(E);
5740+
LValue CodeGenFunction::EmitCallExprLValue(const CallExpr *E,
5741+
llvm::CallBase **CallOrInvoke) {
5742+
RValue RV = EmitCallExpr(E, ReturnValueSlot(), CallOrInvoke);
57245743

57255744
if (!RV.isScalar())
57265745
return MakeAddrLValue(RV.getAggregateAddress(), E->getType(),
@@ -5843,9 +5862,11 @@ LValue CodeGenFunction::EmitStmtExprLValue(const StmtExpr *E) {
58435862
AlignmentSource::Decl);
58445863
}
58455864

5846-
RValue CodeGenFunction::EmitCall(QualType CalleeType, const CGCallee &OrigCallee,
5847-
const CallExpr *E, ReturnValueSlot ReturnValue,
5848-
llvm::Value *Chain) {
5865+
RValue CodeGenFunction::EmitCall(QualType CalleeType,
5866+
const CGCallee &OrigCallee, const CallExpr *E,
5867+
ReturnValueSlot ReturnValue,
5868+
llvm::Value *Chain,
5869+
llvm::CallBase **CallOrInvoke) {
58495870
// Get the actual function type. The callee type will always be a pointer to
58505871
// function type or a block pointer type.
58515872
assert(CalleeType->isFunctionPointerType() &&
@@ -6065,8 +6086,8 @@ RValue CodeGenFunction::EmitCall(QualType CalleeType, const CGCallee &OrigCallee
60656086
Address(Handle, Handle->getType(), CGM.getPointerAlign()));
60666087
Callee.setFunctionPointer(Stub);
60676088
}
6068-
llvm::CallBase *CallOrInvoke = nullptr;
6069-
RValue Call = EmitCall(FnInfo, Callee, ReturnValue, Args, &CallOrInvoke,
6089+
llvm::CallBase *LocalCallOrInvoke = nullptr;
6090+
RValue Call = EmitCall(FnInfo, Callee, ReturnValue, Args, &LocalCallOrInvoke,
60706091
E == MustTailCall, E->getExprLoc());
60716092

60726093
// Generate function declaration DISuprogram in order to be used
@@ -6075,11 +6096,13 @@ RValue CodeGenFunction::EmitCall(QualType CalleeType, const CGCallee &OrigCallee
60756096
if (auto *CalleeDecl = dyn_cast_or_null<FunctionDecl>(TargetDecl)) {
60766097
FunctionArgList Args;
60776098
QualType ResTy = BuildFunctionArgList(CalleeDecl, Args);
6078-
DI->EmitFuncDeclForCallSite(CallOrInvoke,
6099+
DI->EmitFuncDeclForCallSite(LocalCallOrInvoke,
60796100
DI->getFunctionType(CalleeDecl, ResTy, Args),
60806101
CalleeDecl);
60816102
}
60826103
}
6104+
if (CallOrInvoke)
6105+
*CallOrInvoke = LocalCallOrInvoke;
60836106

60846107
return Call;
60856108
}

0 commit comments

Comments
 (0)