Skip extra copy from aggregate where it isn't necessary; rdar://problem/8139919 . This shouldn't make much of a difference at -O3, but should substantially reduce the number of generated memcpy's at -O0.

llvm-svn: 130717
This commit is contained in:
Eli Friedman
2011-05-02 20:24:29 +00:00
parent 39b56b4b9f
commit 30458b51e3
3 changed files with 38 additions and 4 deletions

View File

@@ -1166,6 +1166,15 @@ void CodeGenFunction::EmitCallArg(CallArgList &args, const Expr *E,
return args.add(EmitReferenceBindingToExpr(E, /*InitializedDecl=*/0),
type);
if (hasAggregateLLVMType(type) && isa<ImplicitCastExpr>(E) &&
cast<CastExpr>(E)->getCastKind() == CK_LValueToRValue) {
LValue L = EmitLValue(cast<CastExpr>(E)->getSubExpr());
assert(L.isSimple());
args.add(RValue::getAggregate(L.getAddress(), L.isVolatileQualified()),
type, /*NeedsCopy*/true);
return;
}
args.add(EmitAnyExprToTemp(E), type);
}
@@ -1231,6 +1240,10 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
Alignment, I->Ty);
else
StoreComplexToAddr(RV.getComplexVal(), Args.back(), false);
} else if (I->NeedsCopy && !ArgInfo.getIndirectByVal()) {
Args.push_back(CreateMemTemp(I->Ty));
EmitAggregateCopy(Args.back(), RV.getAggregateAddr(), I->Ty,
RV.isVolatileQualified());
} else {
Args.push_back(RV.getAggregateAddr());
}