maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/CodeGenCXX/microsoft-abi-virtual-inher...

558 lines
22 KiB
C++
Raw Normal View History

[MS] Apply adjustments after storing 'this' Summary: The MS ABI convention is that the 'this' pointer on entry is the address of the vfptr that was used to make the virtual method call. In other words, the pointer on entry always points to the base subobject that introduced the virtual method. Consider this hierarchy: struct A { virtual void f() = 0; }; struct B { virtual void g() = 0; }; struct C : A, B { void f() override; void g() override; }; On entry to C::g, [ER]CX will contain the address of C's B subobject, and C::g will have to subtract sizeof(A) to recover a pointer to C. Before this change, we applied this adjustment in the prologue and stored the new value into the "this" local variable alloca used for debug info. However, MSVC does not do this, presumably because it is often profitable to fold the adjustment into later field accesses. This creates a problem, because the debugger expects the variable to be unadjusted. Unfortunately, CodeView doesn't have anything like DWARF expressions for computing variables that aren't in the program anymore, so we have to declare 'this' to be the unadjusted value if we want the debugger to see the right value. This has the side benefit that, in optimized builds, the 'this' pointer will usually be available on function entry because it doesn't require any adjustment. Reviewers: hans Subscribers: aprantl, cfe-commits Differential Revision: https://reviews.llvm.org/D40109 llvm-svn: 318440
2017-11-17 03:09:36 +08:00
// RUN: %clang_cc1 %s -fno-rtti -std=c++11 -Wno-inaccessible-base -triple=i386-pc-win32 -emit-llvm -o %t
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// RUN: FileCheck %s < %t
// RUN: FileCheck --check-prefix=CHECK2 %s < %t
[ms-cxxabi] Move CodeGenVTables::needsVTTParameter to ItaniumCXXABI. This function only makes sense there. Eventually it should no longer be part of the CGCXXABI interface, as it is an Itanium-specific detail. Differential Revision: http://llvm-reviews.chandlerc.com/D821 llvm-svn: 185213
2013-06-29 04:45:28 +08:00
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// For now, just make sure x86_64 doesn't crash.
[MS] Apply adjustments after storing 'this' Summary: The MS ABI convention is that the 'this' pointer on entry is the address of the vfptr that was used to make the virtual method call. In other words, the pointer on entry always points to the base subobject that introduced the virtual method. Consider this hierarchy: struct A { virtual void f() = 0; }; struct B { virtual void g() = 0; }; struct C : A, B { void f() override; void g() override; }; On entry to C::g, [ER]CX will contain the address of C's B subobject, and C::g will have to subtract sizeof(A) to recover a pointer to C. Before this change, we applied this adjustment in the prologue and stored the new value into the "this" local variable alloca used for debug info. However, MSVC does not do this, presumably because it is often profitable to fold the adjustment into later field accesses. This creates a problem, because the debugger expects the variable to be unadjusted. Unfortunately, CodeView doesn't have anything like DWARF expressions for computing variables that aren't in the program anymore, so we have to declare 'this' to be the unadjusted value if we want the debugger to see the right value. This has the side benefit that, in optimized builds, the 'this' pointer will usually be available on function entry because it doesn't require any adjustment. Reviewers: hans Subscribers: aprantl, cfe-commits Differential Revision: https://reviews.llvm.org/D40109 llvm-svn: 318440
2017-11-17 03:09:36 +08:00
// RUN: %clang_cc1 %s -fno-rtti -std=c++11 -Wno-inaccessible-base -triple=x86_64-pc-win32 -emit-llvm -o %t
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
struct VBase {
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
virtual ~VBase();
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
virtual void foo();
virtual void bar();
int field;
[ms-cxxabi] Move CodeGenVTables::needsVTTParameter to ItaniumCXXABI. This function only makes sense there. Eventually it should no longer be part of the CGCXXABI interface, as it is an Itanium-specific detail. Differential Revision: http://llvm-reviews.chandlerc.com/D821 llvm-svn: 185213
2013-06-29 04:45:28 +08:00
};
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
struct B : virtual VBase {
Abstract out the emission of vtables, add basic support for vtable emission when using -cxx-abi microsoft Reviewed at http://llvm-reviews.chandlerc.com/D1532 llvm-svn: 191523
2013-09-27 22:48:01 +08:00
B();
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
virtual ~B();
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
virtual void foo();
virtual void bar();
[ms-cxxabi] Move CodeGenVTables::needsVTTParameter to ItaniumCXXABI. This function only makes sense there. Eventually it should no longer be part of the CGCXXABI interface, as it is an Itanium-specific detail. Differential Revision: http://llvm-reviews.chandlerc.com/D821 llvm-svn: 185213
2013-06-29 04:45:28 +08:00
};
Abstract out the emission of vtables, add basic support for vtable emission when using -cxx-abi microsoft Reviewed at http://llvm-reviews.chandlerc.com/D1532 llvm-svn: 191523
2013-09-27 22:48:01 +08:00
B::B() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc %struct.B* @"??0B@@QAE@XZ"
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[THIS:.*]] = load %struct.B*, %struct.B**
Abstract out the emission of vtables, add basic support for vtable emission when using -cxx-abi microsoft Reviewed at http://llvm-reviews.chandlerc.com/D1532 llvm-svn: 191523
2013-09-27 22:48:01 +08:00
// CHECK: br i1 %{{.*}}, label %[[INIT_VBASES:.*]], label %[[SKIP_VBASES:.*]]
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// Don't check the INIT_VBASES case as it's covered by the ctor tests.
Abstract out the emission of vtables, add basic support for vtable emission when using -cxx-abi microsoft Reviewed at http://llvm-reviews.chandlerc.com/D1532 llvm-svn: 191523
2013-09-27 22:48:01 +08:00
// CHECK: %[[SKIP_VBASES]]
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 0
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// ...
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VFPTR_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 %{{.*}}
Cast vtable address points to i32 (...)** to enable more globalopt We currently use i32 (...)** as the type of the vptr field in the LLVM struct type. LLVM's GlobalOpt prefers any bitcasts to be on the side of the data being stored rather than on the pointer being stored to. Reviewers: majnemer Differential Revision: http://reviews.llvm.org/D5916 llvm-svn: 223267
2014-12-04 05:00:21 +08:00
// CHECK: %[[VFPTR:.*]] = bitcast i8* %[[VFPTR_i8]] to i32 (...)***
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: store i32 (...)** bitcast ({ [3 x i8*] }* @"??_7B@@6B@" to i32 (...)**), i32 (...)*** %[[VFPTR]]
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// Initialize vtorDisp:
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 0
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// ...
// CHECK: %[[VBASE_OFFSET:.*]] = add nsw i32 0, %{{.*}}
// CHECK: %[[VTORDISP_VAL:.*]] = sub i32 %[[VBASE_OFFSET]], 8
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBASE_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 %[[VBASE_OFFSET]]
// CHECK: %[[VTORDISP_i8:.*]] = getelementptr i8, i8* %[[VBASE_i8]], i32 -4
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// CHECK: %[[VTORDISP_PTR:.*]] = bitcast i8* %[[VTORDISP_i8]] to i32*
// CHECK: store i32 %[[VTORDISP_VAL]], i32* %[[VTORDISP_PTR]]
// CHECK: ret
}
B::~B() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc void @"??1B@@UAE@XZ"
[MS] Apply adjustments after storing 'this' Summary: The MS ABI convention is that the 'this' pointer on entry is the address of the vfptr that was used to make the virtual method call. In other words, the pointer on entry always points to the base subobject that introduced the virtual method. Consider this hierarchy: struct A { virtual void f() = 0; }; struct B { virtual void g() = 0; }; struct C : A, B { void f() override; void g() override; }; On entry to C::g, [ER]CX will contain the address of C's B subobject, and C::g will have to subtract sizeof(A) to recover a pointer to C. Before this change, we applied this adjustment in the prologue and stored the new value into the "this" local variable alloca used for debug info. However, MSVC does not do this, presumably because it is often profitable to fold the adjustment into later field accesses. This creates a problem, because the debugger expects the variable to be unadjusted. Unfortunately, CodeView doesn't have anything like DWARF expressions for computing variables that aren't in the program anymore, so we have to declare 'this' to be the unadjusted value if we want the debugger to see the right value. This has the side benefit that, in optimized builds, the 'this' pointer will usually be available on function entry because it doesn't require any adjustment. Reviewers: hans Subscribers: aprantl, cfe-commits Differential Revision: https://reviews.llvm.org/D40109 llvm-svn: 318440
2017-11-17 03:09:36 +08:00
// Store initial this:
// CHECK: %[[THIS_ADDR:.*]] = alloca %struct.B*
// CHECK: store %struct.B* %{{.*}}, %struct.B** %[[THIS_ADDR]], align 4
// Reload and adjust the this parameter:
// CHECK: %[[THIS_RELOAD:.*]] = load %struct.B*, %struct.B** %[[THIS_ADDR]]
// CHECK: %[[THIS_UNADJ_i8:.*]] = bitcast %struct.B* %[[THIS_RELOAD]] to i8*
// CHECK: %[[THIS_ADJ_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_UNADJ_i8]], i32 -8
// CHECK: %[[THIS:.*]] = bitcast i8* %[[THIS_ADJ_i8]] to %struct.B*
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// Restore the vfptr that could have been changed by a subclass.
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 0
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// ...
Abstract out the emission of vtables, add basic support for vtable emission when using -cxx-abi microsoft Reviewed at http://llvm-reviews.chandlerc.com/D1532 llvm-svn: 191523
2013-09-27 22:48:01 +08:00
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VFPTR_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 %{{.*}}
Cast vtable address points to i32 (...)** to enable more globalopt We currently use i32 (...)** as the type of the vptr field in the LLVM struct type. LLVM's GlobalOpt prefers any bitcasts to be on the side of the data being stored rather than on the pointer being stored to. Reviewers: majnemer Differential Revision: http://reviews.llvm.org/D5916 llvm-svn: 223267
2014-12-04 05:00:21 +08:00
// CHECK: %[[VFPTR:.*]] = bitcast i8* %[[VFPTR_i8]] to i32 (...)***
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: store i32 (...)** bitcast ({ [3 x i8*] }* @"??_7B@@6B@" to i32 (...)**), i32 (...)*** %[[VFPTR]]
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// Initialize vtorDisp:
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 0
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// ...
// CHECK: %[[VBASE_OFFSET:.*]] = add nsw i32 0, %{{.*}}
// CHECK: %[[VTORDISP_VAL:.*]] = sub i32 %[[VBASE_OFFSET]], 8
// CHECK: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBASE_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 %[[VBASE_OFFSET]]
// CHECK: %[[VTORDISP_i8:.*]] = getelementptr i8, i8* %[[VBASE_i8]], i32 -4
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// CHECK: %[[VTORDISP_PTR:.*]] = bitcast i8* %[[VTORDISP_i8]] to i32*
// CHECK: store i32 %[[VTORDISP_VAL]], i32* %[[VTORDISP_PTR]]
foo(); // Avoid the "trivial destructor" optimization.
Abstract out the emission of vtables, add basic support for vtable emission when using -cxx-abi microsoft Reviewed at http://llvm-reviews.chandlerc.com/D1532 llvm-svn: 191523
2013-09-27 22:48:01 +08:00
// CHECK: ret
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK2-LABEL: define linkonce_odr dso_local x86_thiscallcc void @"??_DB@@QAEXXZ"(%struct.B*
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK2: %[[THIS:.*]] = load %struct.B*, %struct.B** {{.*}}
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK2: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK2: %[[B_i8:.*]] = getelementptr i8, i8* %[[THIS_i8]], i32 8
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK2: %[[B:.*]] = bitcast i8* %[[B_i8]] to %struct.B*
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK2: call x86_thiscallcc void @"??1B@@UAE@XZ"(%struct.B* %[[B]])
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK2: %[[THIS_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
// CHECK2: %[[VBASE_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_i8]], i32 8
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK2: %[[VBASE:.*]] = bitcast i8* %[[VBASE_i8]] to %struct.VBase*
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK2: call x86_thiscallcc void @"??1VBase@@UAE@XZ"(%struct.VBase* %[[VBASE]])
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK2: ret
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK2-LABEL: define linkonce_odr dso_local x86_thiscallcc i8* @"??_GB@@UAEPAXI@Z"
[MS] Apply adjustments after storing 'this' Summary: The MS ABI convention is that the 'this' pointer on entry is the address of the vfptr that was used to make the virtual method call. In other words, the pointer on entry always points to the base subobject that introduced the virtual method. Consider this hierarchy: struct A { virtual void f() = 0; }; struct B { virtual void g() = 0; }; struct C : A, B { void f() override; void g() override; }; On entry to C::g, [ER]CX will contain the address of C's B subobject, and C::g will have to subtract sizeof(A) to recover a pointer to C. Before this change, we applied this adjustment in the prologue and stored the new value into the "this" local variable alloca used for debug info. However, MSVC does not do this, presumably because it is often profitable to fold the adjustment into later field accesses. This creates a problem, because the debugger expects the variable to be unadjusted. Unfortunately, CodeView doesn't have anything like DWARF expressions for computing variables that aren't in the program anymore, so we have to declare 'this' to be the unadjusted value if we want the debugger to see the right value. This has the side benefit that, in optimized builds, the 'this' pointer will usually be available on function entry because it doesn't require any adjustment. Reviewers: hans Subscribers: aprantl, cfe-commits Differential Revision: https://reviews.llvm.org/D40109 llvm-svn: 318440
2017-11-17 03:09:36 +08:00
// CHECK2: store %struct.B* %{{.*}}, %struct.B** %[[THIS_ADDR:.*]], align 4
// CHECK2: %[[THIS:.*]] = load %struct.B*, %struct.B** %[[THIS_ADDR]]
// CHECK2: %[[THIS_PARAM_i8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK2: %[[THIS_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_PARAM_i8:.*]], i32 -8
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK2: %[[THIS:.*]] = bitcast i8* %[[THIS_i8]] to %struct.B*
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK2: call x86_thiscallcc void @"??_DB@@QAEXXZ"(%struct.B* %[[THIS]])
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// ...
// CHECK2: ret
Abstract out the emission of vtables, add basic support for vtable emission when using -cxx-abi microsoft Reviewed at http://llvm-reviews.chandlerc.com/D1532 llvm-svn: 191523
2013-09-27 22:48:01 +08:00
}
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
void B::foo() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc void @"?foo@B@@UAEXXZ"(i8*
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
//
// B::foo gets 'this' cast to VBase* in ECX (i.e. this+8) so we
// need to adjust 'this' before use.
//
Reland "[CodeGen] Fix crash when a function taking transparent union is redeclared." When a function taking transparent union is declared as taking one of union members earlier in the translation unit, clang would hit an "Invalid cast" assertion during EmitFunctionProlog. This case corresponds to function f1 in test/CodeGen/transparent-union-redecl.c. We decided to cast i32 to union because after merging function declarations function parameter type becomes int, CGFunctionInfo::ArgInfo type matches with ABIArgInfo type, so we decide it is a trivial case. But these types should also be castable to parameter declaration type which is not the case here. Now the fix is in converting from ABIArgInfo type to VarDecl type and using argument demotion when necessary. Additional tests in Sema/transparent-union.c capture current behavior and make sure there are no regressions. rdar://problem/34949329 Reviewers: rjmccall, rafael Reviewed By: rjmccall Subscribers: aemerson, cfe-commits, kristof.beyls, ahatanak Differential Revision: https://reviews.llvm.org/D41311 llvm-svn: 323156
2018-01-23 06:29:24 +08:00
// Coerce this to correct type:
// CHECK: %[[THIS_STORE:.*]] = alloca %struct.B*
Revert "[CodeGen] Fix crash when a function taking transparent union is redeclared." This reverts commit r321296. It caused performance regressions FAIL: imp.execution_time FAIL: 2007-01-04-KNR-Args.execution_time FAIL: sse_expandfft.execution_time FAIL: sse_stepfft.execution_time llvm-svn: 321306
2017-12-22 04:52:59 +08:00
// CHECK: %[[THIS_ADDR:.*]] = alloca %struct.B*
Reland "[CodeGen] Fix crash when a function taking transparent union is redeclared." When a function taking transparent union is declared as taking one of union members earlier in the translation unit, clang would hit an "Invalid cast" assertion during EmitFunctionProlog. This case corresponds to function f1 in test/CodeGen/transparent-union-redecl.c. We decided to cast i32 to union because after merging function declarations function parameter type becomes int, CGFunctionInfo::ArgInfo type matches with ABIArgInfo type, so we decide it is a trivial case. But these types should also be castable to parameter declaration type which is not the case here. Now the fix is in converting from ABIArgInfo type to VarDecl type and using argument demotion when necessary. Additional tests in Sema/transparent-union.c capture current behavior and make sure there are no regressions. rdar://problem/34949329 Reviewers: rjmccall, rafael Reviewed By: rjmccall Subscribers: aemerson, cfe-commits, kristof.beyls, ahatanak Differential Revision: https://reviews.llvm.org/D41311 llvm-svn: 323156
2018-01-23 06:29:24 +08:00
// CHECK: %[[COERCE_VAL:.*]] = bitcast i8* %{{.*}} to %struct.B*
// CHECK: store %struct.B* %[[COERCE_VAL]], %struct.B** %[[THIS_STORE]], align 4
//
// Store initial this:
// CHECK: %[[THIS_INIT:.*]] = load %struct.B*, %struct.B** %[[THIS_STORE]]
// CHECK: store %struct.B* %[[THIS_INIT]], %struct.B** %[[THIS_ADDR]], align 4
[MS] Apply adjustments after storing 'this' Summary: The MS ABI convention is that the 'this' pointer on entry is the address of the vfptr that was used to make the virtual method call. In other words, the pointer on entry always points to the base subobject that introduced the virtual method. Consider this hierarchy: struct A { virtual void f() = 0; }; struct B { virtual void g() = 0; }; struct C : A, B { void f() override; void g() override; }; On entry to C::g, [ER]CX will contain the address of C's B subobject, and C::g will have to subtract sizeof(A) to recover a pointer to C. Before this change, we applied this adjustment in the prologue and stored the new value into the "this" local variable alloca used for debug info. However, MSVC does not do this, presumably because it is often profitable to fold the adjustment into later field accesses. This creates a problem, because the debugger expects the variable to be unadjusted. Unfortunately, CodeView doesn't have anything like DWARF expressions for computing variables that aren't in the program anymore, so we have to declare 'this' to be the unadjusted value if we want the debugger to see the right value. This has the side benefit that, in optimized builds, the 'this' pointer will usually be available on function entry because it doesn't require any adjustment. Reviewers: hans Subscribers: aprantl, cfe-commits Differential Revision: https://reviews.llvm.org/D40109 llvm-svn: 318440
2017-11-17 03:09:36 +08:00
//
// Reload and adjust the this parameter:
// CHECK: %[[THIS_RELOAD:.*]] = load %struct.B*, %struct.B** %[[THIS_ADDR]]
// CHECK: %[[THIS_UNADJ_i8:.*]] = bitcast %struct.B* %[[THIS_RELOAD]] to i8*
// CHECK: %[[THIS_ADJ_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_UNADJ_i8]], i32 -8
// CHECK: %[[THIS:.*]] = bitcast i8* %[[THIS_ADJ_i8]] to %struct.B*
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
field = 42;
// CHECK: %[[THIS8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[THIS8]], i32 0
MS ABI: Emit more canonical vbptr stores and loads This eliminates some i8* GEPs and makes the IR that clang emits a bit more canonical. More work is needed for vftables, but that isn't a clear win so I plan to send it for review. llvm-svn: 220398
2014-10-23 01:26:00 +08:00
// CHECK: %[[VBPTR8:.*]] = bitcast i8* %[[VBPTR]] to i32**
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBTABLE:.*]] = load i32*, i32** %[[VBPTR8]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBENTRY:.*]] = getelementptr inbounds i32, i32* %[[VBTABLE]], i32 1
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBOFFSET32:.*]] = load i32, i32* %[[VBENTRY]]
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
// CHECK: %[[VBOFFSET:.*]] = add nsw i32 0, %[[VBOFFSET32]]
// CHECK: %[[THIS8:.*]] = bitcast %struct.B* %[[THIS]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBASE_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS8]], i32 %[[VBOFFSET]]
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
// CHECK: %[[VBASE:.*]] = bitcast i8* %[[VBASE_i8]] to %struct.VBase*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[FIELD:.*]] = getelementptr inbounds %struct.VBase, %struct.VBase* %[[VBASE]], i32 0, i32 1
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
// CHECK: store i32 42, i32* %[[FIELD]], align 4
//
// CHECK: ret void
}
void call_vbase_bar(B *obj) {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local void @"?call_vbase_bar@@YAXPAUB@@@Z"(%struct.B* %obj)
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
// CHECK: %[[OBJ:.*]] = load %struct.B
obj->bar();
// When calling a vbase's virtual method, one needs to adjust 'this'
// at the caller site.
//
// CHECK: %[[OBJ_i8:.*]] = bitcast %struct.B* %[[OBJ]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 0
MS ABI: Emit more canonical vbptr stores and loads This eliminates some i8* GEPs and makes the IR that clang emits a bit more canonical. More work is needed for vftables, but that isn't a clear win so I plan to send it for review. llvm-svn: 220398
2014-10-23 01:26:00 +08:00
// CHECK: %[[VBPTR8:.*]] = bitcast i8* %[[VBPTR]] to i32**
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBTABLE:.*]] = load i32*, i32** %[[VBPTR8]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBENTRY:.*]] = getelementptr inbounds i32, i32* %[[VBTABLE]], i32 1
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBOFFSET32:.*]] = load i32, i32* %[[VBENTRY]]
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
// CHECK: %[[VBOFFSET:.*]] = add nsw i32 0, %[[VBOFFSET32]]
IRGen: Move vtable load after argument evaluation. This change reduces the live range of the loaded function pointer, resulting in a slight code size decrease (~10KB in clang), and also improves the security of CFI for virtual calls by making it less likely that the function pointer will be spilled, and ensuring that it is not spilled across a function call boundary. Fixes PR35353. Differential Revision: https://reviews.llvm.org/D42725 llvm-svn: 324286
2018-02-06 07:09:13 +08:00
// CHECK: %[[VBASE:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 %[[VBOFFSET]]
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
//
// CHECK: %[[OBJ_i8:.*]] = bitcast %struct.B* %[[OBJ]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 0
MS ABI: Emit more canonical vbptr stores and loads This eliminates some i8* GEPs and makes the IR that clang emits a bit more canonical. More work is needed for vftables, but that isn't a clear win so I plan to send it for review. llvm-svn: 220398
2014-10-23 01:26:00 +08:00
// CHECK: %[[VBPTR8:.*]] = bitcast i8* %[[VBPTR]] to i32**
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBTABLE:.*]] = load i32*, i32** %[[VBPTR8]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBENTRY:.*]] = getelementptr inbounds i32, i32* %[[VBTABLE]], i32 1
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBOFFSET32:.*]] = load i32, i32* %[[VBENTRY]]
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
// CHECK: %[[VBOFFSET:.*]] = add nsw i32 0, %[[VBOFFSET32]]
IRGen: Move vtable load after argument evaluation. This change reduces the live range of the loaded function pointer, resulting in a slight code size decrease (~10KB in clang), and also improves the security of CFI for virtual calls by making it less likely that the function pointer will be spilled, and ensuring that it is not spilled across a function call boundary. Fixes PR35353. Differential Revision: https://reviews.llvm.org/D42725 llvm-svn: 324286
2018-02-06 07:09:13 +08:00
// CHECK: %[[VBASE_i8:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 %[[VBOFFSET]]
// CHECK: %[[VFPTR:.*]] = bitcast i8* %[[VBASE_i8]] to void (i8*)***
// CHECK: %[[VFTABLE:.*]] = load void (i8*)**, void (i8*)*** %[[VFPTR]]
// CHECK: %[[VFUN:.*]] = getelementptr inbounds void (i8*)*, void (i8*)** %[[VFTABLE]], i64 2
// CHECK: %[[VFUN_VALUE:.*]] = load void (i8*)*, void (i8*)** %[[VFUN]]
Abstract out virtual calls and virtual function prologue code generation; implement them for -cxx-abi microsoft llvm-svn: 188870
2013-08-21 14:25:03 +08:00
//
// CHECK: call x86_thiscallcc void %[[VFUN_VALUE]](i8* %[[VBASE]])
//
// CHECK: ret void
}
Reland 192220 "Abstract out parts of thunk emission code, add support for simple thunks when using -cxx-abi microsoft" with relaxed assertions llvm-svn: 192285
2013-10-09 17:23:58 +08:00
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
void delete_B(B *obj) {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local void @"?delete_B@@YAXPAUB@@@Z"(%struct.B* %obj)
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: %[[OBJ:.*]] = load %struct.B
delete obj;
// CHECK: %[[OBJ_i8:.*]] = bitcast %struct.B* %[[OBJ]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 0
MS ABI: Emit more canonical vbptr stores and loads This eliminates some i8* GEPs and makes the IR that clang emits a bit more canonical. More work is needed for vftables, but that isn't a clear win so I plan to send it for review. llvm-svn: 220398
2014-10-23 01:26:00 +08:00
// CHECK: %[[VBPTR8:.*]] = bitcast i8* %[[VBPTR]] to i32**
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBTABLE:.*]] = load i32*, i32** %[[VBPTR8]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBENTRY:.*]] = getelementptr inbounds i32, i32* %[[VBTABLE]], i32 1
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBOFFSET32:.*]] = load i32, i32* %[[VBENTRY]]
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: %[[VBOFFSET:.*]] = add nsw i32 0, %[[VBOFFSET32]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBASE_i8:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 %[[VBOFFSET]]
IRGen: Move vtable load after argument evaluation. This change reduces the live range of the loaded function pointer, resulting in a slight code size decrease (~10KB in clang), and also improves the security of CFI for virtual calls by making it less likely that the function pointer will be spilled, and ensuring that it is not spilled across a function call boundary. Fixes PR35353. Differential Revision: https://reviews.llvm.org/D42725 llvm-svn: 324286
2018-02-06 07:09:13 +08:00
// CHECK: %[[VBASE:.*]] = bitcast i8* %[[VBASE_i8]] to %struct.B*
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
//
// CHECK: %[[OBJ_i8:.*]] = bitcast %struct.B* %[[OBJ]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBPTR:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 0
MS ABI: Emit more canonical vbptr stores and loads This eliminates some i8* GEPs and makes the IR that clang emits a bit more canonical. More work is needed for vftables, but that isn't a clear win so I plan to send it for review. llvm-svn: 220398
2014-10-23 01:26:00 +08:00
// CHECK: %[[VBPTR8:.*]] = bitcast i8* %[[VBPTR]] to i32**
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBTABLE:.*]] = load i32*, i32** %[[VBPTR8]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBENTRY:.*]] = getelementptr inbounds i32, i32* %[[VBTABLE]], i32 1
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VBOFFSET32:.*]] = load i32, i32* %[[VBENTRY]]
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: %[[VBOFFSET:.*]] = add nsw i32 0, %[[VBOFFSET32]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VBASE_i8:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 %[[VBOFFSET]]
IRGen: Move vtable load after argument evaluation. This change reduces the live range of the loaded function pointer, resulting in a slight code size decrease (~10KB in clang), and also improves the security of CFI for virtual calls by making it less likely that the function pointer will be spilled, and ensuring that it is not spilled across a function call boundary. Fixes PR35353. Differential Revision: https://reviews.llvm.org/D42725 llvm-svn: 324286
2018-02-06 07:09:13 +08:00
// CHECK: %[[VFPTR:.*]] = bitcast i8* %[[VBASE_i8]] to i8* (%struct.B*, i32)***
// CHECK: %[[VFTABLE:.*]] = load i8* (%struct.B*, i32)**, i8* (%struct.B*, i32)*** %[[VFPTR]]
// CHECK: %[[VFUN:.*]] = getelementptr inbounds i8* (%struct.B*, i32)*, i8* (%struct.B*, i32)** %[[VFTABLE]], i64 0
// CHECK: %[[VFUN_VALUE:.*]] = load i8* (%struct.B*, i32)*, i8* (%struct.B*, i32)** %[[VFUN]]
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
//
MS ABI: Properly call global delete when invoking virtual destructors Summary: The Itanium ABI approach of using offset-to-top isn't possible with the MS ABI, it doesn't have that kind of information lying around. Instead, we do the following: - Call the virtual deleting destructor with the "don't delete the object flag" set. The virtual deleting destructor will return a pointer to 'this' adjusted to the most derived class. - Call the global delete using the adjusted 'this' pointer. Reviewers: rnk Subscribers: cfe-commits Differential Revision: http://reviews.llvm.org/D5996 llvm-svn: 220993
2014-11-01 04:09:12 +08:00
// CHECK: call x86_thiscallcc i8* %[[VFUN_VALUE]](%struct.B* %[[VBASE]], i32 1)
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: ret void
}
void call_complete_dtor() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local void @"?call_complete_dtor@@YAXXZ"
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
B b;
Revert "[CodeGenCXX] Treat 'this' as noalias in constructors" This reverts commit https://reviews.llvm.org/rL344150 which causes MachineOutliner related failures on the ppc64le multistage buildbot. llvm-svn: 344526
2018-10-15 23:43:00 +08:00
// CHECK: call x86_thiscallcc %struct.B* @"??0B@@QAE@XZ"(%struct.B* %[[B:.*]], i32 1)
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK-NOT: getelementptr
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: call x86_thiscallcc void @"??_DB@@QAEXXZ"(%struct.B* %[[B]])
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: ret
}
Fix PR18967 -- Bad this adjustment for virtual methods in a diamond virtual inheritance hierarchy llvm-svn: 203222
2014-03-07 17:34:59 +08:00
struct C : B {
Reland 192220 "Abstract out parts of thunk emission code, add support for simple thunks when using -cxx-abi microsoft" with relaxed assertions llvm-svn: 192285
2013-10-09 17:23:58 +08:00
C();
// has an implicit vdtor.
};
// Used to crash on an assertion.
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
C::C() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc %struct.C* @"??0C@@QAE@XZ"
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
}
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
namespace multiple_vbases {
struct A {
virtual void a();
};
struct B {
virtual void b();
};
struct C {
virtual void c();
};
struct D : virtual A, virtual B, virtual C {
virtual void a();
virtual void b();
virtual void c();
D();
};
D::D() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc %"struct.multiple_vbases::D"* @"??0D@multiple_vbases@@QAE@XZ"
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// Just make sure we emit 3 vtordisps after initializing vfptrs.
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: store i32 (...)** bitcast ({ [1 x i8*] }* @"??_7D@multiple_vbases@@6BA@1@@" to i32 (...)**), i32 (...)*** %{{.*}}
// CHECK: store i32 (...)** bitcast ({ [1 x i8*] }* @"??_7D@multiple_vbases@@6BB@1@@" to i32 (...)**), i32 (...)*** %{{.*}}
// CHECK: store i32 (...)** bitcast ({ [1 x i8*] }* @"??_7D@multiple_vbases@@6BC@1@@" to i32 (...)**), i32 (...)*** %{{.*}}
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
// ...
// CHECK: store i32 %{{.*}}, i32* %{{.*}}
// CHECK: store i32 %{{.*}}, i32* %{{.*}}
// CHECK: store i32 %{{.*}}, i32* %{{.*}}
// CHECK: ret
}
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
}
namespace diamond {
struct A {
A();
virtual ~A();
};
struct B : virtual A {
B();
~B();
};
struct C : virtual A {
C();
~C();
int c1, c2, c3;
};
struct Z {
int z;
};
struct D : virtual Z, B, C {
D();
~D();
} d;
D::~D() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc void @"??1D@diamond@@UAE@XZ"(%"struct.diamond::D"*{{.*}})
[MS] Apply adjustments after storing 'this' Summary: The MS ABI convention is that the 'this' pointer on entry is the address of the vfptr that was used to make the virtual method call. In other words, the pointer on entry always points to the base subobject that introduced the virtual method. Consider this hierarchy: struct A { virtual void f() = 0; }; struct B { virtual void g() = 0; }; struct C : A, B { void f() override; void g() override; }; On entry to C::g, [ER]CX will contain the address of C's B subobject, and C::g will have to subtract sizeof(A) to recover a pointer to C. Before this change, we applied this adjustment in the prologue and stored the new value into the "this" local variable alloca used for debug info. However, MSVC does not do this, presumably because it is often profitable to fold the adjustment into later field accesses. This creates a problem, because the debugger expects the variable to be unadjusted. Unfortunately, CodeView doesn't have anything like DWARF expressions for computing variables that aren't in the program anymore, so we have to declare 'this' to be the unadjusted value if we want the debugger to see the right value. This has the side benefit that, in optimized builds, the 'this' pointer will usually be available on function entry because it doesn't require any adjustment. Reviewers: hans Subscribers: aprantl, cfe-commits Differential Revision: https://reviews.llvm.org/D40109 llvm-svn: 318440
2017-11-17 03:09:36 +08:00
// Store initial this:
// CHECK: %[[THIS_ADDR:.*]] = alloca %"struct.diamond::D"*
// CHECK: store %"struct.diamond::D"* %{{.*}}, %"struct.diamond::D"** %[[THIS_ADDR]], align 4
//
// Reload and adjust the this parameter:
// CHECK: %[[THIS_RELOAD:.*]] = load %"struct.diamond::D"*, %"struct.diamond::D"** %[[THIS_ADDR]]
// CHECK: %[[THIS_UNADJ_i8:.*]] = bitcast %"struct.diamond::D"* %[[THIS_RELOAD]] to i8*
// CHECK: %[[THIS_ADJ_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_UNADJ_i8]], i32 -24
// CHECK: %[[THIS:.*]] = bitcast i8* %[[THIS_ADJ_i8]] to %"struct.diamond::D"*
//
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: %[[D_i8:.*]] = bitcast %"struct.diamond::D"* %[[THIS]] to i8*
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
// CHECK: %[[C_i8:.*]] = getelementptr inbounds i8, i8* %[[D_i8]], i32 4
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: %[[C:.*]] = bitcast i8* %[[C_i8]] to %"struct.diamond::C"*
// CHECK: %[[C_i8:.*]] = bitcast %"struct.diamond::C"* %[[C]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[ARG_i8:.*]] = getelementptr i8, i8* %{{.*}}, i32 16
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// FIXME: We might consider changing the dtor this parameter type to i8*.
// CHECK: %[[ARG:.*]] = bitcast i8* %[[ARG_i8]] to %"struct.diamond::C"*
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: call x86_thiscallcc void @"??1C@diamond@@UAE@XZ"(%"struct.diamond::C"* %[[ARG]])
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: %[[B:.*]] = bitcast %"struct.diamond::D"* %[[THIS]] to %"struct.diamond::B"*
// CHECK: %[[B_i8:.*]] = bitcast %"struct.diamond::B"* %[[B]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[ARG_i8:.*]] = getelementptr i8, i8* %[[B_i8]], i32 4
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: %[[ARG:.*]] = bitcast i8* %[[ARG_i8]] to %"struct.diamond::B"*
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: call x86_thiscallcc void @"??1B@diamond@@UAE@XZ"(%"struct.diamond::B"* %[[ARG]])
[-cxx-abi microsoft] Fix this argument/parameter offsets for virtual destructors in the presence of virtual bases Reviewed at http://llvm-reviews.chandlerc.com/D1939 llvm-svn: 192822
2013-10-17 02:24:06 +08:00
// CHECK: ret void
}
Initialize vtorDisp in class constructors and destructors Reviewed at http://llvm-reviews.chandlerc.com/D1867 llvm-svn: 192312
2013-10-10 02:16:58 +08:00
}
[ms-abi] Always generate complete constructors in the Microsoft C++ ABI Fixes PR18435, where we generated a base ctor instead of a complete ctor, and so failed to construct virtual bases when constructing the complete object. llvm-svn: 199160
2014-01-14 06:57:31 +08:00
namespace test2 {
struct A { A(); };
struct B : virtual A { B() {} };
struct C : B, A { C() {} };
// PR18435: Order mattered here. We were generating code for the delegating
// call to B() from C().
void callC() { C x; }
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define linkonce_odr dso_local x86_thiscallcc %"struct.test2::C"* @"??0C@test2@@QAE@XZ"
Revert "[CodeGenCXX] Treat 'this' as noalias in constructors" This reverts commit https://reviews.llvm.org/rL344150 which causes MachineOutliner related failures on the ppc64le multistage buildbot. llvm-svn: 344526
2018-10-15 23:43:00 +08:00
// CHECK: (%"struct.test2::C"* returned %this, i32 %is_most_derived)
[ms-abi] Always generate complete constructors in the Microsoft C++ ABI Fixes PR18435, where we generated a base ctor instead of a complete ctor, and so failed to construct virtual bases when constructing the complete object. llvm-svn: 199160
2014-01-14 06:57:31 +08:00
// CHECK: br i1
// Virtual bases
Revert "[CodeGenCXX] Treat 'this' as noalias in constructors" This reverts commit https://reviews.llvm.org/rL344150 which causes MachineOutliner related failures on the ppc64le multistage buildbot. llvm-svn: 344526
2018-10-15 23:43:00 +08:00
// CHECK: call x86_thiscallcc %"struct.test2::A"* @"??0A@test2@@QAE@XZ"(%"struct.test2::A"* %{{.*}})
[ms-abi] Always generate complete constructors in the Microsoft C++ ABI Fixes PR18435, where we generated a base ctor instead of a complete ctor, and so failed to construct virtual bases when constructing the complete object. llvm-svn: 199160
2014-01-14 06:57:31 +08:00
// CHECK: br label
// Non-virtual bases
Revert "[CodeGenCXX] Treat 'this' as noalias in constructors" This reverts commit https://reviews.llvm.org/rL344150 which causes MachineOutliner related failures on the ppc64le multistage buildbot. llvm-svn: 344526
2018-10-15 23:43:00 +08:00
// CHECK: call x86_thiscallcc %"struct.test2::B"* @"??0B@test2@@QAE@XZ"(%"struct.test2::B"* %{{.*}}, i32 0)
// CHECK: call x86_thiscallcc %"struct.test2::A"* @"??0A@test2@@QAE@XZ"(%"struct.test2::A"* %{{.*}})
[ms-abi] Always generate complete constructors in the Microsoft C++ ABI Fixes PR18435, where we generated a base ctor instead of a complete ctor, and so failed to construct virtual bases when constructing the complete object. llvm-svn: 199160
2014-01-14 06:57:31 +08:00
// CHECK: ret
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK2-LABEL: define linkonce_odr dso_local x86_thiscallcc %"struct.test2::B"* @"??0B@test2@@QAE@XZ"
Revert "[CodeGenCXX] Treat 'this' as noalias in constructors" This reverts commit https://reviews.llvm.org/rL344150 which causes MachineOutliner related failures on the ppc64le multistage buildbot. llvm-svn: 344526
2018-10-15 23:43:00 +08:00
// CHECK2: (%"struct.test2::B"* returned %this, i32 %is_most_derived)
// CHECK2: call x86_thiscallcc %"struct.test2::A"* @"??0A@test2@@QAE@XZ"(%"struct.test2::A"* %{{.*}})
[ms-abi] Always generate complete constructors in the Microsoft C++ ABI Fixes PR18435, where we generated a base ctor instead of a complete ctor, and so failed to construct virtual bases when constructing the complete object. llvm-svn: 199160
2014-01-14 06:57:31 +08:00
// CHECK2: ret
}
Fix PR19104: Incorrect handling of non-virtual calls of virtual methods Reviewed at http://llvm-reviews.chandlerc.com/D3054 llvm-svn: 203949
2014-03-15 01:43:37 +08:00
namespace test3 {
// PR19104: A non-virtual call of a virtual method doesn't use vftable thunks,
// so requires only static adjustment which is different to the one used
// for virtual calls.
struct A {
virtual void foo();
};
struct B : virtual A {
virtual void bar();
};
struct C : virtual A {
virtual void foo();
};
struct D : B, C {
virtual void bar();
int field; // Laid out between C and A subobjects in D.
};
void D::bar() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc void @"?bar@D@test3@@UAEXXZ"(%"struct.test3::D"* %this)
Fix PR19104: Incorrect handling of non-virtual calls of virtual methods Reviewed at http://llvm-reviews.chandlerc.com/D3054 llvm-svn: 203949
2014-03-15 01:43:37 +08:00
C::foo();
// Shouldn't need any vbtable lookups. All we have to do is adjust to C*,
// then compensate for the adjustment performed in the C::foo() prologue.
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK-NOT: load i8*, i8**
Fix PR19104: Incorrect handling of non-virtual calls of virtual methods Reviewed at http://llvm-reviews.chandlerc.com/D3054 llvm-svn: 203949
2014-03-15 01:43:37 +08:00
// CHECK: %[[OBJ_i8:.*]] = bitcast %"struct.test3::D"* %{{.*}} to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[C_i8:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 8
Fix PR19104: Incorrect handling of non-virtual calls of virtual methods Reviewed at http://llvm-reviews.chandlerc.com/D3054 llvm-svn: 203949
2014-03-15 01:43:37 +08:00
// CHECK: %[[C:.*]] = bitcast i8* %[[C_i8]] to %"struct.test3::C"*
// CHECK: %[[C_i8:.*]] = bitcast %"struct.test3::C"* %[[C]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[ARG:.*]] = getelementptr i8, i8* %[[C_i8]], i32 4
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: call x86_thiscallcc void @"?foo@C@test3@@UAEXXZ"(i8* %[[ARG]])
Fix PR19104: Incorrect handling of non-virtual calls of virtual methods Reviewed at http://llvm-reviews.chandlerc.com/D3054 llvm-svn: 203949
2014-03-15 01:43:37 +08:00
// CHECK: ret
}
}
Fix PR19172 - wrong this adjustment calculated for virtual destructor in a class with complex inheritance Reviewed at http://llvm-reviews.chandlerc.com/D3128 llvm-svn: 204394
2014-03-21 04:38:34 +08:00
namespace test4{
// PR19172: We used to merge method vftable locations wrong.
struct A {
virtual ~A() {}
};
struct B {
virtual ~B() {}
};
struct C : virtual A, B {
virtual ~C();
};
void foo(void*);
C::~C() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc void @"??1C@test4@@UAE@XZ"(%"struct.test4::C"* %this)
Fix PR19172 - wrong this adjustment calculated for virtual destructor in a class with complex inheritance Reviewed at http://llvm-reviews.chandlerc.com/D3128 llvm-svn: 204394
2014-03-21 04:38:34 +08:00
// In this case "this" points to the most derived class, so no GEPs needed.
// CHECK-NOT: getelementptr
// CHECK-NOT: bitcast
Cast vtable address points to i32 (...)** to enable more globalopt We currently use i32 (...)** as the type of the vptr field in the LLVM struct type. LLVM's GlobalOpt prefers any bitcasts to be on the side of the data being stored rather than on the pointer being stored to. Reviewers: majnemer Differential Revision: http://reviews.llvm.org/D5916 llvm-svn: 223267
2014-12-04 05:00:21 +08:00
// CHECK: %[[VFPTR_i8:.*]] = bitcast %"struct.test4::C"* %{{.*}} to i32 (...)***
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: store i32 (...)** bitcast ({ [1 x i8*] }* @"??_7C@test4@@6BB@1@@" to i32 (...)**), i32 (...)*** %[[VFPTR_i8]]
Fix PR19172 - wrong this adjustment calculated for virtual destructor in a class with complex inheritance Reviewed at http://llvm-reviews.chandlerc.com/D3128 llvm-svn: 204394
2014-03-21 04:38:34 +08:00
foo(this);
Add one more test for PR19172 llvm-svn: 204453
2014-03-21 19:07:24 +08:00
// CHECK: ret
Fix PR19172 - wrong this adjustment calculated for virtual destructor in a class with complex inheritance Reviewed at http://llvm-reviews.chandlerc.com/D3128 llvm-svn: 204394
2014-03-21 04:38:34 +08:00
}
void destroy(C *obj) {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local void @"?destroy@test4@@YAXPAUC@1@@Z"(%"struct.test4::C"* %obj)
Fix PR19172 - wrong this adjustment calculated for virtual destructor in a class with complex inheritance Reviewed at http://llvm-reviews.chandlerc.com/D3128 llvm-svn: 204394
2014-03-21 04:38:34 +08:00
delete obj;
MS ABI: Properly call global delete when invoking virtual destructors Summary: The Itanium ABI approach of using offset-to-top isn't possible with the MS ABI, it doesn't have that kind of information lying around. Instead, we do the following: - Call the virtual deleting destructor with the "don't delete the object flag" set. The virtual deleting destructor will return a pointer to 'this' adjusted to the most derived class. - Call the global delete using the adjusted 'this' pointer. Reviewers: rnk Subscribers: cfe-commits Differential Revision: http://reviews.llvm.org/D5996 llvm-svn: 220993
2014-11-01 04:09:12 +08:00
// CHECK: %[[VPTR:.*]] = bitcast %"struct.test4::C"* %[[OBJ:.*]] to i8* (%"struct.test4::C"*, i32)***
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VFTABLE:.*]] = load i8* (%"struct.test4::C"*, i32)**, i8* (%"struct.test4::C"*, i32)*** %[[VPTR]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VFTENTRY:.*]] = getelementptr inbounds i8* (%"struct.test4::C"*, i32)*, i8* (%"struct.test4::C"*, i32)** %[[VFTABLE]], i64 0
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VFUN:.*]] = load i8* (%"struct.test4::C"*, i32)*, i8* (%"struct.test4::C"*, i32)** %[[VFTENTRY]]
MS ABI: Properly call global delete when invoking virtual destructors Summary: The Itanium ABI approach of using offset-to-top isn't possible with the MS ABI, it doesn't have that kind of information lying around. Instead, we do the following: - Call the virtual deleting destructor with the "don't delete the object flag" set. The virtual deleting destructor will return a pointer to 'this' adjusted to the most derived class. - Call the global delete using the adjusted 'this' pointer. Reviewers: rnk Subscribers: cfe-commits Differential Revision: http://reviews.llvm.org/D5996 llvm-svn: 220993
2014-11-01 04:09:12 +08:00
// CHECK: call x86_thiscallcc i8* %[[VFUN]](%"struct.test4::C"* %[[OBJ]], i32 1)
Add one more test for PR19172 llvm-svn: 204453
2014-03-21 19:07:24 +08:00
// CHECK: ret
}
struct D {
virtual void d();
};
// The first non-virtual base doesn't have a vdtor,
// but "this adjustment" is not needed.
struct E : D, B, virtual A {
virtual ~E();
};
E::~E() {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc void @"??1E@test4@@UAE@XZ"(%"struct.test4::E"* %this)
Add one more test for PR19172 llvm-svn: 204453
2014-03-21 19:07:24 +08:00
// In this case "this" points to the most derived class, so no GEPs needed.
// CHECK-NOT: getelementptr
// CHECK-NOT: bitcast
Cast vtable address points to i32 (...)** to enable more globalopt We currently use i32 (...)** as the type of the vptr field in the LLVM struct type. LLVM's GlobalOpt prefers any bitcasts to be on the side of the data being stored rather than on the pointer being stored to. Reviewers: majnemer Differential Revision: http://reviews.llvm.org/D5916 llvm-svn: 223267
2014-12-04 05:00:21 +08:00
// CHECK: %[[VFPTR_i8:.*]] = bitcast %"struct.test4::E"* %{{.*}} to i32 (...)***
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: store i32 (...)** bitcast ({ [1 x i8*] }* @"??_7E@test4@@6BD@1@@" to i32 (...)**), i32 (...)*** %[[VFPTR_i8]]
Add one more test for PR19172 llvm-svn: 204453
2014-03-21 19:07:24 +08:00
foo(this);
}
void destroy(E *obj) {
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local void @"?destroy@test4@@YAXPAUE@1@@Z"(%"struct.test4::E"* %obj)
Add one more test for PR19172 llvm-svn: 204453
2014-03-21 19:07:24 +08:00
// CHECK-NOT: getelementptr
IRGen: Move vtable load after argument evaluation. This change reduces the live range of the loaded function pointer, resulting in a slight code size decrease (~10KB in clang), and also improves the security of CFI for virtual calls by making it less likely that the function pointer will be spilled, and ensuring that it is not spilled across a function call boundary. Fixes PR35353. Differential Revision: https://reviews.llvm.org/D42725 llvm-svn: 324286
2018-02-06 07:09:13 +08:00
// CHECK: %[[OBJ_i8:.*]] = bitcast %"struct.test4::E"* %[[OBJ]] to i8*
// CHECK: %[[B_i8:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 4
// FIXME: in fact, the call should take i8* and the bitcast is redundant.
// CHECK: %[[B_as_E:.*]] = bitcast i8* %[[B_i8]] to %"struct.test4::E"*
Add one more test for PR19172 llvm-svn: 204453
2014-03-21 19:07:24 +08:00
// CHECK: %[[OBJ_i8:.*]] = bitcast %"struct.test4::E"* %[[OBJ:.*]] to i8*
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[B_i8:.*]] = getelementptr inbounds i8, i8* %[[OBJ_i8]], i32 4
MS ABI: Properly call global delete when invoking virtual destructors Summary: The Itanium ABI approach of using offset-to-top isn't possible with the MS ABI, it doesn't have that kind of information lying around. Instead, we do the following: - Call the virtual deleting destructor with the "don't delete the object flag" set. The virtual deleting destructor will return a pointer to 'this' adjusted to the most derived class. - Call the global delete using the adjusted 'this' pointer. Reviewers: rnk Subscribers: cfe-commits Differential Revision: http://reviews.llvm.org/D5996 llvm-svn: 220993
2014-11-01 04:09:12 +08:00
// CHECK: %[[VPTR:.*]] = bitcast i8* %[[B_i8]] to i8* (%"struct.test4::E"*, i32)***
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VFTABLE:.*]] = load i8* (%"struct.test4::E"*, i32)**, i8* (%"struct.test4::E"*, i32)*** %[[VPTR]]
Update Clang tests to handle explicitly typed gep changes in LLVM. llvm-svn: 230783
2015-02-28 03:18:17 +08:00
// CHECK: %[[VFTENTRY:.*]] = getelementptr inbounds i8* (%"struct.test4::E"*, i32)*, i8* (%"struct.test4::E"*, i32)** %[[VFTABLE]], i64 0
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: %[[VFUN:.*]] = load i8* (%"struct.test4::E"*, i32)*, i8* (%"struct.test4::E"*, i32)** %[[VFTENTRY]]
MS ABI: Properly call global delete when invoking virtual destructors Summary: The Itanium ABI approach of using offset-to-top isn't possible with the MS ABI, it doesn't have that kind of information lying around. Instead, we do the following: - Call the virtual deleting destructor with the "don't delete the object flag" set. The virtual deleting destructor will return a pointer to 'this' adjusted to the most derived class. - Call the global delete using the adjusted 'this' pointer. Reviewers: rnk Subscribers: cfe-commits Differential Revision: http://reviews.llvm.org/D5996 llvm-svn: 220993
2014-11-01 04:09:12 +08:00
// CHECK: call x86_thiscallcc i8* %[[VFUN]](%"struct.test4::E"* %[[B_as_E]], i32 1)
Add one more test for PR19172 llvm-svn: 204453
2014-03-21 19:07:24 +08:00
delete obj;
Fix PR19172 - wrong this adjustment calculated for virtual destructor in a class with complex inheritance Reviewed at http://llvm-reviews.chandlerc.com/D3128 llvm-svn: 204394
2014-03-21 04:38:34 +08:00
}
}
[MS ABI] Don't zero-initialize vbptrs in bases Certain CXXConstructExpr nodes require zero-initialization before a constructor is called. We had a bug in the case where the constructor is called on a virtual base: we zero-initialized the base's vbptr field. A complementary bug is present in MSVC where no zero-initialization occurs for the subobject at all. This fixes PR25370. llvm-svn: 251783
2015-11-02 17:01:44 +08:00
namespace test5 {
// PR25370: Don't zero-initialize vbptrs in virtual bases.
struct A {
virtual void f();
};
struct B : virtual A {
int Field;
};
struct C : B {
C();
};
C::C() : B() {}
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc %"struct.test5::C"* @"??0C@test5@@QAE@XZ"(
[MS ABI] Don't zero-initialize vbptrs in bases Certain CXXConstructExpr nodes require zero-initialization before a constructor is called. We had a bug in the case where the constructor is called on a virtual base: we zero-initialized the base's vbptr field. A complementary bug is present in MSVC where no zero-initialization occurs for the subobject at all. This fixes PR25370. llvm-svn: 251783
2015-11-02 17:01:44 +08:00
// CHECK: %[[THIS:.*]] = load %"struct.test5::C"*, %"struct.test5::C"**
// CHECK: br i1 %{{.*}}, label %[[INIT_VBASES:.*]], label %[[SKIP_VBASES:.*]]
// CHECK: %[[SKIP_VBASES]]
// CHECK: %[[B:.*]] = bitcast %"struct.test5::C"* %[[THIS]] to %"struct.test5::B"*
// CHECK: %[[B_i8:.*]] = bitcast %"struct.test5::B"* %[[B]] to i8*
// CHECK: %[[FIELD:.*]] = getelementptr inbounds i8, i8* %[[B_i8]], i32 4
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// CHECK: call void @llvm.memset.p0i8.i32(i8* align 4 %[[FIELD]], i8 0, i32 4, i1 false)
[MS ABI] Don't zero-initialize vbptrs in bases Certain CXXConstructExpr nodes require zero-initialization before a constructor is called. We had a bug in the case where the constructor is called on a virtual base: we zero-initialized the base's vbptr field. A complementary bug is present in MSVC where no zero-initialization occurs for the subobject at all. This fixes PR25370. llvm-svn: 251783
2015-11-02 17:01:44 +08:00
}
[MS] Pass CalleeDecl to adjustThisArgumentForVirtualFunctionCall If we are devirtualizing, then we want to compute the 'this' adjustment of the devirtualized target, not the adjustment of the base's method definition, which is usually zero. Fixes PR27621 llvm-svn: 268418
2016-05-04 02:44:29 +08:00
namespace pr27621 {
// Devirtualization through a static_cast used to make us compute the 'this'
// adjustment for B::g instead of C::g. When we directly call C::g, 'this' is a
// B*, and the prologue of C::g will adjust it to a C*.
struct A { virtual void f(); };
struct B { virtual void g(); };
struct C final : A, B {
virtual void h();
void g() override;
};
void callit(C *p) {
static_cast<B*>(p)->g();
}
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local void @"?callit@pr27621@@YAXPAUC@1@@Z"(%"struct.pr27621::C"* %{{.*}})
Fix use of LLVM IR names in lit test llvm-svn: 268419
2016-05-04 02:48:50 +08:00
// CHECK: %[[B_i8:.*]] = getelementptr i8, i8* %{{.*}}, i32 4
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK: call x86_thiscallcc void @"?g@C@pr27621@@UAEXXZ"(i8* %[[B_i8]])
[MS] Pass CalleeDecl to adjustThisArgumentForVirtualFunctionCall If we are devirtualizing, then we want to compute the 'this' adjustment of the devirtualized target, not the adjustment of the base's method definition, which is usually zero. Fixes PR27621 llvm-svn: 268418
2016-05-04 02:44:29 +08:00
}
[MS ABI] Don't crash when zero-initializing a vbase which contains a vbase Bases can be zero-initialized: the storage is zero-initialized before the base constructor is run. The MS ABI has a quirk where base VBPtrs are not installed by the base constructor but by the most derived class. In particular, they are installed before the base constructor is run. The derived constructor must be careful to zero-initialize only the bits of the class which haven't already been populated by virtual base pointers. While we correctly avoided this scenario, we didn't handle the case where the base class has virtual bases which have virtual bases. llvm-svn: 269271
2016-05-12 11:51:52 +08:00
namespace test6 {
class A {};
class B : virtual A {};
class C : virtual B {
virtual void m_fn1();
float field;
};
class D : C {
D();
};
D::D() : C() {}
[MS] Don't escape MS C++ names with \01 It is not needed after LLVM r327734. Now it will be easier to copy-paste IR symbol names from Clang. llvm-svn: 327738
2018-03-17 04:36:49 +08:00
// CHECK-LABEL: define dso_local x86_thiscallcc %"class.test6::D"* @"??0D@test6@@AAE@XZ"(
[MS ABI] Don't crash when zero-initializing a vbase which contains a vbase Bases can be zero-initialized: the storage is zero-initialized before the base constructor is run. The MS ABI has a quirk where base VBPtrs are not installed by the base constructor but by the most derived class. In particular, they are installed before the base constructor is run. The derived constructor must be careful to zero-initialize only the bits of the class which haven't already been populated by virtual base pointers. While we correctly avoided this scenario, we didn't handle the case where the base class has virtual bases which have virtual bases. llvm-svn: 269271
2016-05-12 11:51:52 +08:00
// CHECK: %[[THIS:.*]] = load %"class.test6::D"*, %"class.test6::D"**
// CHECK: br i1 %{{.*}}, label %[[INIT_VBASES:.*]], label %[[SKIP_VBASES:.*]]
// CHECK: %[[SKIP_VBASES]]
// CHECK: %[[C:.*]] = bitcast %"class.test6::D"* %[[THIS]] to %"class.test6::C"*
// CHECK: %[[C_i8:.*]] = bitcast %"class.test6::C"* %[[C]] to i8*
// CHECK: %[[FIELD:.*]] = getelementptr inbounds i8, i8* %[[C_i8]], i32 8
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// CHECK: call void @llvm.memset.p0i8.i32(i8* align 4 %[[FIELD]], i8 0, i32 4, i1 false)
[MS ABI] Don't crash when zero-initializing a vbase which contains a vbase Bases can be zero-initialized: the storage is zero-initialized before the base constructor is run. The MS ABI has a quirk where base VBPtrs are not installed by the base constructor but by the most derived class. In particular, they are installed before the base constructor is run. The derived constructor must be careful to zero-initialize only the bits of the class which haven't already been populated by virtual base pointers. While we correctly avoided this scenario, we didn't handle the case where the base class has virtual bases which have virtual bases. llvm-svn: 269271
2016-05-12 11:51:52 +08:00
}
[MS] Fix bug in method vfptr location code We were assuming that vbtable indices were assigned in layout order in our comparison, which is not the case. When a virtual method, such as the destructor, appears in multiple vftables, the vftable that appears first in object layout order is the one that points to the main implementation of that method. The later vftables use thunks. In this layout, we adjusted "this" in the main implementation by the amount that is appropriate for 'B' instead of 'A', even though the main implementation is found in D's vftable for A: struct A { virtual ~A() {} }; struct B { virtual ~B() {} }; struct C : virtual B {}; struct D : virtual A, C {}; D's layout looks like: 0 D subobject (empty) 0 C base suboject 8 A base subobject 16 B base subobject With this fix, we correctly adjust by -8 in D's deleting destructor instead of -16. Fixes PR36921. llvm-svn: 328723
2018-03-29 02:23:35 +08:00
namespace pr36921 {
struct A {
virtual ~A() {}
};
struct B {
virtual ~B() {}
};
struct C : virtual B {};
struct D : virtual A, C {};
D d;
// CHECK-LABEL: define linkonce_odr dso_local x86_thiscallcc i8* @"??_GD@pr36921@@UAEPAXI@Z"(
// CHECK: %[[THIS:.*]] = load %"struct.pr36921::D"*, %"struct.pr36921::D"**
// CHECK: %[[THIS_UNADJ_i8:.*]] = bitcast %"struct.pr36921::D"* %[[THIS_RELOAD]] to i8*
// CHECK: %[[THIS_ADJ_i8:.*]] = getelementptr inbounds i8, i8* %[[THIS_UNADJ_i8]], i32 -4
// CHECK: %[[THIS:.*]] = bitcast i8* %[[THIS_ADJ_i8]] to %"struct.pr36921::D"*
}