// Name the struct return argument.
if (CGM.ReturnTypeUsesSRet(FI)) {
AI->setName("agg.result");
+ AI->addAttr(llvm::Attribute::NoAlias);
++AI;
}
AggValueSlot AggSlot =
AggValueSlot::forAddr(V, Qualifiers(),
AggValueSlot::IsDestructed,
- AggValueSlot::DoesNotNeedGCBarriers);
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased);
CGF.EmitAggExpr(BaseInit->getInit(), AggSlot);
AggValueSlot Slot =
AggValueSlot::forAddr(Dest, LHS.getQuals(),
AggValueSlot::IsDestructed,
- AggValueSlot::DoesNotNeedGCBarriers);
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased);
CGF.EmitAggExpr(MemberInit->getInit(), Slot);
}
AggValueSlot AggSlot =
AggValueSlot::forAddr(ThisPtr, Qualifiers(),
AggValueSlot::IsDestructed,
- AggValueSlot::DoesNotNeedGCBarriers);
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased);
EmitAggExpr(Ctor->init_begin()[0]->getInit(), AggSlot);
// TODO: how can we delay here if D is captured by its initializer?
EmitAggExpr(init, AggValueSlot::forLValue(lvalue,
AggValueSlot::IsDestructed,
- AggValueSlot::DoesNotNeedGCBarriers));
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased));
}
}
CGF.EmitComplexExprIntoAddr(Init, DeclPtr, lv.isVolatile());
} else {
CGF.EmitAggExpr(Init, AggValueSlot::forLValue(lv,AggValueSlot::IsDestructed,
- AggValueSlot::DoesNotNeedGCBarriers));
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased));
}
}
= AggValueSlot::IsDestructed_t(InitializedDecl != 0);
AggSlot = AggValueSlot::forAddr(ReferenceTemporary, Qualifiers(),
isDestructed,
- AggValueSlot::DoesNotNeedGCBarriers);
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased);
}
if (InitializedDecl) {
AggValueSlot Dest;
bool IgnoreResult;
+ /// We want to use 'dest' as the return slot except under two
+ /// conditions:
+ /// - The destination slot requires garbage collection, so we
+ /// need to use the GC API.
+ /// - The destination slot is potentially aliased.
+ bool shouldUseDestForReturnSlot() const {
+ return !(Dest.requiresGCollection() || Dest.isPotentiallyAliased());
+ }
+
ReturnValueSlot getReturnValueSlot() const {
- // If the destination slot requires garbage collection, we can't
- // use the real return value slot, because we have to use the GC
- // API.
- if (Dest.requiresGCollection()) return ReturnValueSlot();
+ if (!shouldUseDestForReturnSlot())
+ return ReturnValueSlot();
return ReturnValueSlot(Dest.getAddr(), Dest.isVolatile());
}
void EmitFinalDestCopy(const Expr *E, LValue Src, bool Ignore = false);
void EmitFinalDestCopy(const Expr *E, RValue Src, bool Ignore = false);
- void EmitGCMove(const Expr *E, RValue Src);
+ void EmitMoveFromReturnSlot(const Expr *E, RValue Src);
AggValueSlot::NeedsGCBarriers_t needsGC(QualType T) {
if (CGF.getLangOptions().getGCMode() && TypeRequiresGCollection(T))
return Record->hasObjectMember();
}
-/// \brief Perform the final move to DestPtr if RequiresGCollection is set.
+/// \brief Perform the final move to DestPtr if for some reason
+/// getReturnValueSlot() didn't use it directly.
///
/// The idea is that you do something like this:
/// RValue Result = EmitSomething(..., getReturnValueSlot());
-/// EmitGCMove(E, Result);
-/// If GC doesn't interfere, this will cause the result to be emitted
-/// directly into the return value slot. If GC does interfere, a final
-/// move will be performed.
-void AggExprEmitter::EmitGCMove(const Expr *E, RValue Src) {
- if (Dest.requiresGCollection()) {
- CharUnits size = CGF.getContext().getTypeSizeInChars(E->getType());
- llvm::Type *SizeTy = CGF.ConvertType(CGF.getContext().getSizeType());
- llvm::Value *SizeVal = llvm::ConstantInt::get(SizeTy, size.getQuantity());
- CGF.CGM.getObjCRuntime().EmitGCMemmoveCollectable(CGF, Dest.getAddr(),
- Src.getAggregateAddr(),
- SizeVal);
+/// EmitMoveFromReturnSlot(E, Result);
+///
+/// If nothing interferes, this will cause the result to be emitted
+/// directly into the return value slot. Otherwise, a final move
+/// will be performed.
+void AggExprEmitter::EmitMoveFromReturnSlot(const Expr *E, RValue Src) {
+ if (shouldUseDestForReturnSlot()) {
+ // Logically, Dest.getAddr() should equal Src.getAggregateAddr().
+ // The possibility of undef rvalues complicates that a lot,
+ // though, so we can't really assert.
+ return;
}
+
+ // Otherwise, do a final copy,
+ assert(Dest.getAddr() != Src.getAggregateAddr());
+ EmitFinalDestCopy(E, Src, /*Ignore*/ true);
}
/// EmitFinalDestCopy - Perform the final copy to DestPtr, if desired.
LValue LV = CGF.EmitLValue(E->getSubExpr());
assert(LV.isPropertyRef());
RValue RV = CGF.EmitLoadOfPropertyRefLValue(LV, getReturnValueSlot());
- EmitGCMove(E, RV);
+ EmitMoveFromReturnSlot(E, RV);
break;
}
}
RValue RV = CGF.EmitCallExpr(E, getReturnValueSlot());
- EmitGCMove(E, RV);
+ EmitMoveFromReturnSlot(E, RV);
}
void AggExprEmitter::VisitObjCMessageExpr(ObjCMessageExpr *E) {
RValue RV = CGF.EmitObjCMessageExpr(E, getReturnValueSlot());
- EmitGCMove(E, RV);
+ EmitMoveFromReturnSlot(E, RV);
}
void AggExprEmitter::VisitObjCPropertyRefExpr(ObjCPropertyRefExpr *E) {
CGF.EmitAggExpr(E, AggValueSlot::forLValue(LV,
AggValueSlot::IsDestructed,
AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased,
Dest.isZeroed()));
} else if (LV.isSimple()) {
CGF.EmitScalarInit(E, /*D=*/0, LV, /*Captured=*/false);
LoadObjCSelf(), Ivar, 0);
EmitAggExpr(IvarInit->getInit(),
AggValueSlot::forLValue(LV, AggValueSlot::IsDestructed,
- AggValueSlot::DoesNotNeedGCBarriers));
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased));
}
// constructor returns 'self'.
CodeGenTypes &Types = CGM.getTypes();
} else {
EmitAggExpr(RV, AggValueSlot::forAddr(ReturnValue, Qualifiers(),
AggValueSlot::IsDestructed,
- AggValueSlot::DoesNotNeedGCBarriers));
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased));
}
EmitBranchThroughCleanup(ReturnBlock);
/// be set.
bool ZeroedFlag : 1;
+ /// AliasedFlag - This generally defaults to false, but can be true
+ /// if the memory is known not to be aliased.
+ bool AliasedFlag : 1;
+
public:
- enum IsZeroed_t { IsNotZeroed, IsZeroed };
+ enum IsAliased_t { IsNotAliased, IsAliased };
enum IsDestructed_t { IsNotDestructed, IsDestructed };
+ enum IsZeroed_t { IsNotZeroed, IsZeroed };
enum NeedsGCBarriers_t { DoesNotNeedGCBarriers, NeedsGCBarriers };
/// ignored - Returns an aggregate value slot indicating that the
AggValueSlot AV;
AV.Addr = 0;
AV.Quals = Qualifiers();
- AV.LifetimeFlag = AV.RequiresGCollection = AV.ZeroedFlag = 0;
+ AV.LifetimeFlag = AV.RequiresGCollection = AV.ZeroedFlag = false;
+
+ // If there's ever an address here, it will be a temporary.
+ AV.AliasedFlag = false;
return AV;
}
static AggValueSlot forAddr(llvm::Value *addr, Qualifiers quals,
IsDestructed_t isDestructed,
NeedsGCBarriers_t needsGC,
+ IsAliased_t isAliased = IsAliased,
IsZeroed_t isZeroed = IsNotZeroed) {
AggValueSlot AV;
AV.Addr = addr;
AV.LifetimeFlag = isDestructed;
AV.RequiresGCollection = needsGC;
AV.ZeroedFlag = isZeroed;
+ AV.AliasedFlag = isAliased;
return AV;
}
static AggValueSlot forLValue(LValue LV, IsDestructed_t isDestructed,
NeedsGCBarriers_t needsGC,
+ IsAliased_t isAliased = IsAliased,
IsZeroed_t isZeroed = IsNotZeroed) {
return forAddr(LV.getAddress(), LV.getQuals(),
- isDestructed, needsGC, isZeroed);
+ isDestructed, needsGC, isAliased, isZeroed);
}
IsDestructed_t isLifetimeExternallyManaged() const {
return Addr == 0;
}
+ IsAliased_t isPotentiallyAliased() const {
+ return IsAliased_t(AliasedFlag);
+ }
+
RValue asRValue() const {
return RValue::getAggregate(getAddr(), isVolatile());
}
AggValueSlot CreateAggTemp(QualType T, const Twine &Name = "tmp") {
return AggValueSlot::forAddr(CreateMemTemp(T, Name), T.getQualifiers(),
AggValueSlot::IsNotDestructed,
- AggValueSlot::DoesNotNeedGCBarriers);
+ AggValueSlot::DoesNotNeedGCBarriers,
+ AggValueSlot::IsNotAliased);
}
/// Emit a cast to void* in the appropriate address space.
struct s4 f4(void) {}
// APCS-GNU: define void @f5(
-// APCS-GNU: struct.s5* sret
+// APCS-GNU: struct.s5* noalias sret
// AAPCS: define arm_aapcscc i32 @f5()
struct s5 { struct { } f0; int f1; };
struct s5 f5(void) {}
// APCS-GNU: define void @f6(
-// APCS-GNU: struct.s6* sret
+// APCS-GNU: struct.s6* noalias sret
// AAPCS: define arm_aapcscc i32 @f6()
struct s6 { int f0[1]; };
struct s6 f6(void) {}
struct s7 f7(void) {}
// APCS-GNU: define void @f8(
-// APCS-GNU: struct.s8* sret
+// APCS-GNU: struct.s8* noalias sret
// AAPCS: define arm_aapcscc void @f8()
struct s8 { struct { int : 0; } f0[1]; };
struct s8 f8(void) {}
struct s10 f10(void) {}
// APCS-GNU: define void @f11(
-// APCS-GNU: struct.s11* sret
+// APCS-GNU: struct.s11* noalias sret
// AAPCS: define arm_aapcscc i32 @f11()
struct s11 { int : 0; int f0; };
struct s11 f11(void) {}
union u12 f12(void) {}
// APCS-GNU: define void @f13(
-// APCS-GNU: struct.s13* sret
+// APCS-GNU: struct.s13* noalias sret
// FIXME: This should return a float.
// AAPCS-FIXME: darm_aapcscc efine float @f13()
struct s13 f13(void) {}
// APCS-GNU: define void @f14(
-// APCS-GNU: union.u14* sret
+// APCS-GNU: union.u14* noalias sret
// AAPCS: define arm_aapcscc i32 @f14()
union u14 { float f0; };
union u14 f14(void) {}
struct s18 f18(void) {}
// APCS-GNU: define void @f19(
-// APCS-GNU: struct.s19* sret
+// APCS-GNU: struct.s19* noalias sret
// AAPCS: define arm_aapcscc i32 @f19()
struct s19 { int f0; struct s8 f1; };
struct s19 f19(void) {}
// APCS-GNU: define void @f20(
-// APCS-GNU: struct.s20* sret
+// APCS-GNU: struct.s20* noalias sret
// AAPCS: define arm_aapcscc i32 @f20()
struct s20 { struct s8 f1; int f0; };
struct s20 f20(void) {}
// APCS-GNU: define i128 @f27()
// AAPCS: define arm_aapcscc i16 @f22()
// AAPCS: define arm_aapcscc i32 @f23()
-// AAPCS: define arm_aapcscc void @f24({{.*}} sret
-// AAPCS: define arm_aapcscc void @f25({{.*}} sret
-// AAPCS: define arm_aapcscc void @f26({{.*}} sret
-// AAPCS: define arm_aapcscc void @f27({{.*}} sret
+// AAPCS: define arm_aapcscc void @f24({{.*}} noalias sret
+// AAPCS: define arm_aapcscc void @f25({{.*}} noalias sret
+// AAPCS: define arm_aapcscc void @f26({{.*}} noalias sret
+// AAPCS: define arm_aapcscc void @f27({{.*}} noalias sret
_Complex char f22(void) {}
_Complex short f23(void) {}
_Complex int f24(void) {}
struct s29 { _Complex short f0; };
struct s29 f29() {}
-// APCS-GNU: define void @f30({{.*}} sret
-// AAPCS: define arm_aapcscc void @f30({{.*}} sret
+// APCS-GNU: define void @f30({{.*}} noalias sret
+// AAPCS: define arm_aapcscc void @f30({{.*}} noalias sret
struct s30 { _Complex int f0; };
struct s30 f30() {}
#include <arm_neon.h>
-// CHECK: define void @f0(%struct.int8x16x2_t* sret %agg.result, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
+// CHECK: define void @f0(%struct.int8x16x2_t* noalias sret %agg.result, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
int8x16x2_t f0(int8x16_t a0, int8x16_t a1) {
return vzipq_s8(a0, a1);
}
T_float32x2 f1_0(T_float32x2 a0) { return a0; }
// CHECK: define <4 x float> @f1_1(<4 x float> %{{.*}})
T_float32x4 f1_1(T_float32x4 a0) { return a0; }
-// CHECK: define void @f1_2(<8 x float>* sret %{{.*}}, <8 x float> %{{.*}})
+// CHECK: define void @f1_2(<8 x float>* noalias sret %{{.*}}, <8 x float> %{{.*}})
T_float32x8 f1_2(T_float32x8 a0) { return a0; }
-// CHECK: define void @f1_3(<16 x float>* sret %{{.*}}, <16 x float> %{{.*}})
+// CHECK: define void @f1_3(<16 x float>* noalias sret %{{.*}}, <16 x float> %{{.*}})
T_float32x16 f1_3(T_float32x16 a0) { return a0; }
-// RUN: %clang_cc1 -triple i386-unknown-unknown %s -emit-llvm -o %t -fblocks
+// RUN: %clang_cc1 -triple i386-unknown-unknown %s -emit-llvm -o - -fblocks | FileCheck %s
void (^f)(void) = ^{};
// rdar://6768379
int a[64];
};
-// RUN: grep 'internal void @__f2_block_invoke_0(.struct.s0\* sret .*, .*, .* byval .*)' %t
+// CHECK: define internal void @__f2_block_invoke_0(%struct.s0* noalias sret {{%.*}}, i8* {{%.*}}, %struct.s0* byval align 4 {{.*}})
struct s0 f2(struct s0 a0) {
return ^(struct s0 a1){ return a1; }(a0);
}
// Small vectors and 1 x {i64,double} are returned in registers
// CHECK: i32 @f11()
-// CHECK: void @f12(<2 x i32>* sret %agg.result)
+// CHECK: void @f12(<2 x i32>* noalias sret %agg.result)
// CHECK: i64 @f13()
// CHECK: i64 @f14()
// CHECK: <2 x i64> @f15()
// 128-bits).
// CHECK: i32 @f17()
-// CHECK: void @f18(%{{.*}}* sret %agg.result)
-// CHECK: void @f19(%{{.*}}* sret %agg.result)
-// CHECK: void @f20(%{{.*}}* sret %agg.result)
-// CHECK: void @f21(%{{.*}}* sret %agg.result)
-// CHECK: void @f22(%{{.*}}* sret %agg.result)
+// CHECK: void @f18(%{{.*}}* noalias sret %agg.result)
+// CHECK: void @f19(%{{.*}}* noalias sret %agg.result)
+// CHECK: void @f20(%{{.*}}* noalias sret %agg.result)
+// CHECK: void @f21(%{{.*}}* noalias sret %agg.result)
+// CHECK: void @f22(%{{.*}}* noalias sret %agg.result)
struct { T11 a; } f17(void) { while (1) {} }
struct { T12 a; } f18(void) { while (1) {} }
struct { T13 a; } f19(void) { while (1) {} }
// Small structures are handled recursively
// CHECK: i32 @f26()
-// CHECK: void @f27(%struct.s27* sret %agg.result)
+// CHECK: void @f27(%struct.s27* noalias sret %agg.result)
struct s26 { struct { char a, b; } a; struct { char a, b; } b; } f26(void) { while (1) {} }
struct s27 { struct { char a, b, c; } a; struct { char a; } b; } f27(void) { while (1) {} }
-// CHECK: void @f28(%struct.s28* sret %agg.result)
+// CHECK: void @f28(%struct.s28* noalias sret %agg.result)
struct s28 { int a; int b[]; } f28(void) { while (1) {} }
// CHECK: define i16 @f29()
// CHECK: define float @f37()
struct s37 { float c[1][1]; } f37(void) { while (1) {} }
-// CHECK: define void @f38(%struct.s38* sret %agg.result)
+// CHECK: define void @f38(%struct.s38* noalias sret %agg.result)
struct s38 { char a[3]; short b; } f38(void) { while (1) {} }
// CHECK: define void @f39(%struct.s39* byval align 16 %x)
// Test merging/passing of upper eightbyte with X87 class.
//
-// CHECK: define void @f8_1(%union.u8* sret %agg.result)
+// CHECK: define void @f8_1(%union.u8* noalias sret %agg.result)
// CHECK: define void @f8_2(%union.u8* byval align 16 %a0)
union u8 {
long double a;
struct s10 { int a; int b; int : 0; };
void f10(struct s10 a0) {}
-// CHECK: define void @f11(%union.anon* sret %agg.result)
+// CHECK: define void @f11(%union.anon* noalias sret %agg.result)
union { long double a; float b; } f11() { while (1) {} }
// CHECK: define i32 @f12_0()
// Check that sret parameter is accounted for when checking available integer
// registers.
-// CHECK: define void @f13(%struct.s13_0* sret %agg.result, i32 %a, i32 %b, i32 %c, i32 %d, {{.*}}* byval align 8 %e, i32 %f)
+// CHECK: define void @f13(%struct.s13_0* noalias sret %agg.result, i32 %a, i32 %b, i32 %c, i32 %d, {{.*}}* byval align 8 %e, i32 %f)
struct s13_0 { long long f0[3]; };
struct s13_1 { long long f0[2]; };
// CHECK-NEXT: call void @_ZN7Elision1AD1Ev([[A]]* [[I]])
}
- // CHECK: define void @_ZN7Elision5test2Ev([[A]]* sret
+ // CHECK: define void @_ZN7Elision5test2Ev([[A]]* noalias sret
A test2() {
// CHECK: call void @_ZN7Elision3fooEv()
// CHECK-NEXT: call void @_ZN7Elision1AC1Ev([[A]]* [[RET:%.*]])
return (foo(), A());
}
- // CHECK: define void @_ZN7Elision5test3EiNS_1AE([[A]]* sret
+ // CHECK: define void @_ZN7Elision5test3EiNS_1AE([[A]]* noalias sret
A test3(int v, A x) {
if (v < 5)
// CHECK: call void @_ZN7Elision1AC1Ev([[A]]* [[RET:%.*]])
}
// rdar://problem/8433352
- // CHECK: define void @_ZN7Elision5test5Ev([[A]]* sret
+ // CHECK: define void @_ZN7Elision5test5Ev([[A]]* noalias sret
struct B { A a; B(); };
A test5() {
// CHECK: [[AT0:%.*]] = alloca [[A]], align 8
short s;
};
-// CHECK: define void @_Z1fv(%struct.S* sret %
+// CHECK: define void @_Z1fv(%struct.S* noalias sret %
S f() { return S(); }
// CHECK: define void @_Z1f1S(%struct.S*)
void f(S) { }
double c;
};
-// CHECK: define void @_Z1gv(%class.C* sret %
+// CHECK: define void @_Z1gv(%class.C* noalias sret %
C g() { return C(); }
// CHECK: define void @_Z1f1C(%class.C*)
struct s7 : s7_0, s7_1 { };
s7 f7() { return s7(); }
-// CHECK: define void @_Z2f8v(%struct.s8* sret %agg.result)
+// CHECK: define void @_Z2f8v(%struct.s8* noalias sret %agg.result)
struct s8_0 { };
struct s8_1 { double x; };
struct s8 { s8_0 a; s8_1 b; };
s8 f8() { return s8(); }
-// CHECK: define void @_Z2f9v(%struct.s9* sret %agg.result)
+// CHECK: define void @_Z2f9v(%struct.s9* noalias sret %agg.result)
struct s9_0 { unsigned : 0; };
struct s9_1 { double x; };
struct s9 { s9_0 a; s9_1 b; };