llvm-project/clang/test/CodeGenCXX/thunks.cpp

406 lines
8.0 KiB
C++
Raw Normal View History

Cleanup the handling of noinline function attributes, -fno-inline, -fno-inline-functions, -O0, and optnone. These were really, really tangled together: - We used the noinline LLVM attribute for -fno-inline - But not for -fno-inline-functions (breaking LTO) - But we did use it for -finline-hint-functions (yay, LTO is happy!) - But we didn't for -O0 (LTO is sad yet again...) - We had weird structuring of CodeGenOpts with both an inlining enumeration and a boolean. They interacted in weird ways and needlessly. - A *lot* of set smashing went on with setting these, and then got worse when we considered optnone and other inlining-effecting attributes. - A bunch of inline affecting attributes were managed in a completely different place from -fno-inline. - Even with -fno-inline we failed to put the LLVM noinline attribute onto many generated function definitions because they didn't show up as AST-level functions. - If you passed -O0 but -finline-functions we would run the normal inliner pass in LLVM despite it being in the O0 pipeline, which really doesn't make much sense. - Lastly, we used things like '-fno-inline' to manipulate the pass pipeline which forced the pass pipeline to be much more parameterizable than it really needs to be. Instead we can *just* use the optimization level to select a pipeline and control the rest via attributes. Sadly, this causes a bunch of churn in tests because we don't run the optimizer in the tests and check the contents of attribute sets. It would be awesome if attribute sets were a bit more FileCheck friendly, but oh well. I think this is a significant improvement and should remove the semantic need to change what inliner pass we run in order to comply with the requested inlining semantics by relying completely on attributes. It also cleans up tho optnone and related handling a bit. One unfortunate aspect of this is that for generating alwaysinline routines like those in OpenMP we end up removing noinline and then adding alwaysinline. I tried a bunch of other approaches, but because we recompute function attributes from scratch and don't have a declaration here I couldn't find anything substantially cleaner than this. Differential Revision: https://reviews.llvm.org/D28053 llvm-svn: 290398
2016-12-23 09:24:49 +08:00
// RUN: %clang_cc1 %s -triple=x86_64-pc-linux-gnu -munwind-tables -emit-llvm -o - | FileCheck --check-prefix=CHECK --check-prefix=CHECK-NONOPT %s
// RUN: %clang_cc1 %s -triple=x86_64-pc-linux-gnu -munwind-tables -emit-llvm -o - -O1 -disable-llvm-passes | FileCheck --check-prefix=CHECK --check-prefix=CHECK-OPT %s
namespace Test1 {
// Check that we emit a non-virtual thunk for C::f.
struct A {
virtual void f();
};
struct B {
virtual void f();
};
struct C : A, B {
virtual void c();
virtual void f();
};
// CHECK-LABEL: define void @_ZThn8_N5Test11C1fEv(
void C::f() { }
}
namespace Test2 {
// Check that we emit a thunk for B::f since it's overriding a virtual base.
struct A {
virtual void f();
};
struct B : virtual A {
virtual void b();
virtual void f();
};
// CHECK-LABEL: define void @_ZTv0_n24_N5Test21B1fEv(
void B::f() { }
}
namespace Test3 {
// Check that we emit a covariant thunk for B::f.
struct V1 { };
struct V2 : virtual V1 { };
struct A {
virtual V1 *f();
};
struct B : A {
virtual void b();
virtual V2 *f();
};
// CHECK: define %{{.*}}* @_ZTch0_v0_n24_N5Test31B1fEv(
V2 *B::f() { return 0; }
}
namespace Test4 {
// Check that the thunk for 'C::f' has the same visibility as the function itself.
struct A {
virtual void f();
};
struct B {
virtual void f();
};
struct __attribute__((visibility("protected"))) C : A, B {
virtual void c();
virtual void f();
};
// CHECK-LABEL: define protected void @_ZThn8_N5Test41C1fEv(
void C::f() { }
}
// Check that the thunk gets internal linkage.
namespace Test4B {
struct A {
virtual void f();
};
struct B {
virtual void f();
};
namespace {
struct C : A, B {
virtual void c();
virtual void f();
};
}
void C::c() {}
void C::f() {}
// Force C::f to be used.
void f() {
C c;
c.f();
}
}
namespace Test5 {
// Check that the thunk for 'B::f' gets the same linkage as the function itself.
struct A {
virtual void f();
};
struct B : virtual A {
virtual void f() { }
};
void f(B b) {
b.f();
}
}
namespace Test6 {
struct X {
X();
X(const X&);
X &operator=(const X&);
~X();
};
struct P {
P();
P(const P&);
~P();
X first;
X second;
};
P getP();
struct Base1 {
int i;
virtual X f() { return X(); }
};
struct Base2 {
float real;
virtual X f() { return X(); }
};
struct Thunks : Base1, Base2 {
long l;
virtual X f();
};
// CHECK-LABEL: define void @_ZThn16_N5Test66Thunks1fEv
// CHECK-NOT: memcpy
// CHECK: {{call void @_ZN5Test66Thunks1fEv.*sret}}
// CHECK: ret void
X Thunks::f() { return X(); }
}
namespace Test7 {
// PR7188
struct X {
X();
X(const X&);
X &operator=(const X&);
~X();
};
struct Small { short s; };
struct Large {
char array[1024];
};
class A {
protected:
virtual void foo() = 0;
};
class B : public A {
protected:
virtual void bar() = 0;
};
class C : public A {
protected:
virtual void baz(X, X&, _Complex float, Small, Small&, Large) = 0;
};
class D : public B,
public C {
void foo() {}
void bar() {}
void baz(X, X&, _Complex float, Small, Small&, Large);
};
void D::baz(X, X&, _Complex float, Small, Small&, Large) { }
// CHECK-LABEL: define void @_ZThn8_N5Test71D3bazENS_1XERS1_CfNS_5SmallERS4_NS_5LargeE(
// CHECK-NOT: memcpy
// CHECK: ret void
void testD() { D d; }
}
namespace Test8 {
struct NonPOD { ~NonPOD(); int x, y, z; };
struct A { virtual void foo(); };
struct B { virtual void bar(NonPOD); };
struct C : A, B { virtual void bar(NonPOD); static void helper(NonPOD); };
// CHECK: define void @_ZN5Test81C6helperENS_6NonPODE([[NONPODTYPE:%.*]]*
void C::helper(NonPOD var) {}
// CHECK-LABEL: define void @_ZThn8_N5Test81C3barENS_6NonPODE(
// CHECK-NOT: load [[NONPODTYPE]], [[NONPODTYPE]]*
// CHECK-NOT: memcpy
// CHECK: ret void
void C::bar(NonPOD var) {}
}
// PR7241: Emitting thunks for a method shouldn't require the vtable for
// that class to be emitted.
namespace Test9 {
struct A { virtual ~A() { } };
struct B : A { virtual void test() const {} };
struct C : B { C(); ~C(); };
struct D : C { D() {} };
void test() {
D d;
}
}
namespace Test10 {
struct A { virtual void foo(); };
struct B { virtual void foo(); };
struct C : A, B { void foo() {} };
// Test later.
void test() {
C c;
}
}
// PR7611
namespace Test11 {
struct A { virtual A* f(); };
struct B : virtual A { virtual A* f(); };
struct C : B { virtual C* f(); };
C* C::f() { return 0; }
// C::f itself.
// CHECK: define {{.*}} @_ZN6Test111C1fEv(
// The this-adjustment and return-adjustment thunk required when
// C::f appears in a vtable where A is at a nonzero offset from C.
// CHECK: define {{.*}} @_ZTcv0_n24_v0_n32_N6Test111C1fEv(
// The return-adjustment thunk required when C::f appears in a vtable
// where A is at a zero offset from C.
// CHECK: define {{.*}} @_ZTch0_v0_n32_N6Test111C1fEv(
}
// Varargs thunk test.
namespace Test12 {
struct A {
virtual A* f(int x, ...);
};
struct B {
virtual B* f(int x, ...);
};
struct C : A, B {
virtual void c();
virtual C* f(int x, ...);
};
C* C::f(int x, ...) { return this; }
// C::f
// CHECK: define {{.*}} @_ZN6Test121C1fEiz
// Varargs thunk; check that both the this and covariant adjustments
// are generated.
// CHECK: define {{.*}} @_ZTchn8_h8_N6Test121C1fEiz
// CHECK: getelementptr inbounds i8, i8* {{.*}}, i64 -8
// CHECK: getelementptr inbounds i8, i8* {{.*}}, i64 8
}
// PR13832
namespace Test13 {
struct B1 {
virtual B1 &foo1();
};
struct Pad1 {
virtual ~Pad1();
};
struct Proxy1 : Pad1, B1 {
virtual ~Proxy1();
};
struct D : virtual Proxy1 {
virtual ~D();
virtual D &foo1();
};
D& D::foo1() {
return *this;
}
// CHECK: define {{.*}} @_ZTcvn8_n32_v8_n24_N6Test131D4foo1Ev
// CHECK: getelementptr inbounds i8, i8* {{.*}}, i64 -8
// CHECK: getelementptr inbounds i8, i8* {{.*}}, i64 -32
// CHECK: getelementptr inbounds i8, i8* {{.*}}, i64 -24
// CHECK: getelementptr inbounds i8, i8* {{.*}}, i64 8
// CHECK: ret %"struct.Test13::D"*
}
namespace Test14 {
class A {
virtual void f();
};
class B {
virtual void f();
};
class C : public A, public B {
virtual void f();
};
void C::f() {
}
// CHECK: define void @_ZThn8_N6Test141C1fEv({{.*}}) unnamed_addr [[NUW:#[0-9]+]]
}
// Varargs non-covariant thunk test.
// PR18098
namespace Test15 {
struct A {
virtual ~A();
};
struct B {
virtual void f(int x, ...);
};
struct C : A, B {
virtual void c();
virtual void f(int x, ...);
};
void C::c() {}
// C::c
// CHECK: declare void @_ZN6Test151C1fEiz
// non-virtual thunk to C::f
// CHECK: declare void @_ZThn8_N6Test151C1fEiz
}
namespace Test16 {
struct A {
virtual ~A();
};
struct B {
virtual void foo();
};
struct C : public A, public B {
void foo() {}
};
struct D : public C {
~D();
};
D::~D() {}
// CHECK: define linkonce_odr void @_ZThn8_N6Test161C3fooEv({{.*}}) {{.*}} comdat
}
/**** The following has to go at the end of the file ****/
// checking without opt
// CHECK-NONOPT-LABEL: define internal void @_ZThn8_N6Test4B12_GLOBAL__N_11C1fEv(
// CHECK-NONOPT-NOT: comdat
// This is from Test5:
// CHECK-NONOPT-LABEL: define linkonce_odr void @_ZTv0_n24_N5Test51B1fEv
// This is from Test10:
// CHECK-NONOPT-LABEL: define linkonce_odr void @_ZN6Test101C3fooEv
// CHECK-NONOPT-LABEL: define linkonce_odr void @_ZThn8_N6Test101C3fooEv
// Checking with opt
// CHECK-OPT-LABEL: define internal void @_ZThn8_N6Test4B12_GLOBAL__N_11C1fEv(%"struct.Test4B::(anonymous namespace)::C"* %this) unnamed_addr #0 align 2
// This is from Test5:
// CHECK-OPT-LABEL: define linkonce_odr void @_ZTv0_n24_N5Test51B1fEv
// This is from Test10:
// CHECK-OPT-LABEL: define linkonce_odr void @_ZN6Test101C3fooEv
// CHECK-OPT-LABEL: define linkonce_odr void @_ZThn8_N6Test101C3fooEv
Cleanup the handling of noinline function attributes, -fno-inline, -fno-inline-functions, -O0, and optnone. These were really, really tangled together: - We used the noinline LLVM attribute for -fno-inline - But not for -fno-inline-functions (breaking LTO) - But we did use it for -finline-hint-functions (yay, LTO is happy!) - But we didn't for -O0 (LTO is sad yet again...) - We had weird structuring of CodeGenOpts with both an inlining enumeration and a boolean. They interacted in weird ways and needlessly. - A *lot* of set smashing went on with setting these, and then got worse when we considered optnone and other inlining-effecting attributes. - A bunch of inline affecting attributes were managed in a completely different place from -fno-inline. - Even with -fno-inline we failed to put the LLVM noinline attribute onto many generated function definitions because they didn't show up as AST-level functions. - If you passed -O0 but -finline-functions we would run the normal inliner pass in LLVM despite it being in the O0 pipeline, which really doesn't make much sense. - Lastly, we used things like '-fno-inline' to manipulate the pass pipeline which forced the pass pipeline to be much more parameterizable than it really needs to be. Instead we can *just* use the optimization level to select a pipeline and control the rest via attributes. Sadly, this causes a bunch of churn in tests because we don't run the optimizer in the tests and check the contents of attribute sets. It would be awesome if attribute sets were a bit more FileCheck friendly, but oh well. I think this is a significant improvement and should remove the semantic need to change what inliner pass we run in order to comply with the requested inlining semantics by relying completely on attributes. It also cleans up tho optnone and related handling a bit. One unfortunate aspect of this is that for generating alwaysinline routines like those in OpenMP we end up removing noinline and then adding alwaysinline. I tried a bunch of other approaches, but because we recompute function attributes from scratch and don't have a declaration here I couldn't find anything substantially cleaner than this. Differential Revision: https://reviews.llvm.org/D28053 llvm-svn: 290398
2016-12-23 09:24:49 +08:00
// CHECK-NONOPT: attributes [[NUW]] = { noinline nounwind uwtable{{.*}} }
// CHECK-OPT: attributes [[NUW]] = { nounwind uwtable{{.*}} }