On 07.03.2013, at 22:36, John McCall <[email protected]> wrote: > Author: rjmccall > Date: Thu Mar 7 15:36:54 2013 > New Revision: 176654 > > URL: http://llvm.org/viewvc/llvm-project?rev=176654&view=rev > Log: > Evaluate compound literals directly into the result aggregate > when that aggregate isn't potentially aliased. > > Modified: > cfe/trunk/lib/CodeGen/CGExprAgg.cpp > cfe/trunk/test/CodeGen/compound-literal.c > > Modified: cfe/trunk/lib/CodeGen/CGExprAgg.cpp > URL: > http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGExprAgg.cpp?rev=176654&r1=176653&r2=176654&view=diff > ============================================================================== > --- cfe/trunk/lib/CodeGen/CGExprAgg.cpp (original) > +++ cfe/trunk/lib/CodeGen/CGExprAgg.cpp Thu Mar 7 15:36:54 2013 > @@ -531,12 +531,10 @@ void AggExprEmitter::VisitOpaqueValueExp > > void > AggExprEmitter::VisitCompoundLiteralExpr(CompoundLiteralExpr *E) { > - if (E->getType().isPODType(CGF.getContext())) { > + if (Dest.isPotentiallyAliased() && > + E->getType().isPODType(CGF.getContext())) {
Does this really depend on the C++ version in use or could it always use the more lenient C++11 definition of POD? - Ben > // For a POD type, just emit a load of the lvalue + a copy, because our > // compound literal might alias the destination. > - // FIXME: This is a band-aid; the real problem appears to be in our > handling > - // of assignments, where we store directly into the LHS without checking > - // whether anything in the RHS aliases. > EmitAggLoadOfLValue(E); > return; > } > > Modified: cfe/trunk/test/CodeGen/compound-literal.c > URL: > http://llvm.org/viewvc/llvm-project/cfe/trunk/test/CodeGen/compound-literal.c?rev=176654&r1=176653&r2=176654&view=diff > ============================================================================== > --- cfe/trunk/test/CodeGen/compound-literal.c (original) > +++ cfe/trunk/test/CodeGen/compound-literal.c Thu Mar 7 15:36:54 2013 > @@ -32,3 +32,37 @@ void f() { > s = (S){s.y,s.x}; > // CHECK-NEXT: ret void > } > + > +// CHECK: define i48 @g( > +struct G { short x, y, z; }; > +struct G g(int x, int y, int z) { > + // CHECK: [[RESULT:%.*]] = alloca [[G:%.*]], align 2 > + // CHECK-NEXT: [[X:%.*]] = alloca i32, align 4 > + // CHECK-NEXT: [[Y:%.*]] = alloca i32, align 4 > + // CHECK-NEXT: [[Z:%.*]] = alloca i32, align 4 > + // CHECK-NEXT: [[COERCE_TEMP:%.*]] = alloca i48 > + // CHECK-NEXT: store i32 > + // CHECK-NEXT: store i32 > + // CHECK-NEXT: store i32 > + > + // Evaluate the compound literal directly in the result value slot. > + // CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[G]]* [[RESULT]], i32 > 0, i32 0 > + // CHECK-NEXT: [[T1:%.*]] = load i32* [[X]], align 4 > + // CHECK-NEXT: [[T2:%.*]] = trunc i32 [[T1]] to i16 > + // CHECK-NEXT: store i16 [[T2]], i16* [[T0]], align 2 > + // CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[G]]* [[RESULT]], i32 > 0, i32 1 > + // CHECK-NEXT: [[T1:%.*]] = load i32* [[Y]], align 4 > + // CHECK-NEXT: [[T2:%.*]] = trunc i32 [[T1]] to i16 > + // CHECK-NEXT: store i16 [[T2]], i16* [[T0]], align 2 > + // CHECK-NEXT: [[T0:%.*]] = getelementptr inbounds [[G]]* [[RESULT]], i32 > 0, i32 2 > + // CHECK-NEXT: [[T1:%.*]] = load i32* [[Z]], align 4 > + // CHECK-NEXT: [[T2:%.*]] = trunc i32 [[T1]] to i16 > + // CHECK-NEXT: store i16 [[T2]], i16* [[T0]], align 2 > + return (struct G) { x, y, z }; > + > + // CHECK-NEXT: [[T0:%.*]] = bitcast i48* [[COERCE_TEMP]] to i8* > + // CHECK-NEXT: [[T1:%.*]] = bitcast [[G]]* [[RESULT]] to i8* > + // CHECK-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* [[T0]], i8* > [[T1]], i64 6 > + // CHECK-NEXT: [[T0:%.*]] = load i48* [[COERCE_TEMP]] > + // CHECK-NEXT: ret i48 [[T0]] > +} > > > _______________________________________________ > cfe-commits mailing list > [email protected] > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
