maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/CodeGenObjCXX/mrc-weak.mm

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

184 lines
6.2 KiB
Plaintext
Raw Normal View History

Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// RUN: %clang_cc1 -triple x86_64-apple-darwin10 -fobjc-runtime=macosx-10.10 -emit-llvm -fblocks -fobjc-weak -o - %s | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-MODERN
// RUN: %clang_cc1 -triple i386-apple-darwin10 -fobjc-runtime=macosx-fragile-10.10 -emit-llvm -fblocks -fobjc-weak -o - %s | FileCheck %s -check-prefix=CHECK -check-prefix=CHECK-FRAGILE
@interface Object
- (instancetype) retain;
- (void) run;
@end
// CHECK-MODERN: @OBJC_CLASS_NAME_{{.*}} = {{.*}} c"\01\00"
[CodeGen][ObjC] Remove the leading `l_` from ObjC symbols and make private symbols in the __DATA segment internal. This prevents the linker from removing the symbol names. Keeping the symbols visible enables tools to collect various information about the symbols, for example, tools that discover whether or not a symbol gets dirtied. rdar://problem/48887111 Differential Revision: https://reviews.llvm.org/D61454 llvm-svn: 360359
2019-05-10 01:43:52 +08:00
// CHECK-MODERN: @"_OBJC_CLASS_RO_$_Foo" = {{.*}} { i32 772
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// 772 == 0x304
// ^ HasMRCWeakIvars
// ^ HasCXXDestructorOnly
// ^ HasCXXStructors
// CHECK-FRAGILE: @OBJC_CLASS_NAME_{{.*}} = {{.*}} c"\01\00"
// CHECK-FRAGILE: @OBJC_CLASS_Foo = {{.*}} i32 134225921,
// 134225921 == 0x08002001
// ^ HasMRCWeakIvars
// ^ HasCXXStructors
// ^ Factory
@interface Foo : Object {
__weak id ivar;
}
@end
@implementation Foo
// CHECK-LABEL: define internal void @"\01-[Foo .cxx_destruct]"
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: call void @llvm.objc.destroyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
@end
void test1(__weak id x) {}
// CHECK-LABEL: define void @_Z5test1P11objc_object(
// CHECK: [[X:%.*]] = alloca i8*,
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK-NEXT: llvm.objc.initWeak
// CHECK-NEXT: llvm.objc.destroyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-NEXT: ret void
void test2(id y) {
__weak id z = y;
}
// CHECK-LABEL: define void @_Z5test2P11objc_object(
// CHECK: [[Y:%.*]] = alloca i8*,
// CHECK-NEXT: [[Z:%.*]] = alloca i8*,
// CHECK-NEXT: store
// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[Y]]
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK-NEXT: call i8* @llvm.objc.initWeak(i8** [[Z]], i8* [[T0]])
// CHECK-NEXT: call void @llvm.objc.destroyWeak(i8** [[Z]])
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-NEXT: ret void
void test3(id y) {
__weak id z;
z = y;
}
// CHECK-LABEL: define void @_Z5test3P11objc_object(
// CHECK: [[Y:%.*]] = alloca i8*,
// CHECK-NEXT: [[Z:%.*]] = alloca i8*,
// CHECK-NEXT: store
// CHECK-NEXT: store i8* null, i8** [[Z]]
// CHECK-NEXT: [[T0:%.*]] = load i8*, i8** [[Y]]
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK-NEXT: call i8* @llvm.objc.storeWeak(i8** [[Z]], i8* [[T0]])
// CHECK-NEXT: call void @llvm.objc.destroyWeak(i8** [[Z]])
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-NEXT: ret void
void test4(__weak id *p) {
id y = *p;
}
// CHECK-LABEL: define void @_Z5test4PU6__weakP11objc_object(
// CHECK: [[P:%.*]] = alloca i8**,
// CHECK-NEXT: [[Y:%.*]] = alloca i8*,
// CHECK-NEXT: store
// CHECK-NEXT: [[T0:%.*]] = load i8**, i8*** [[P]]
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK-NEXT: [[T1:%.*]] = call i8* @llvm.objc.loadWeak(i8** [[T0]])
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-NEXT: store i8* [[T1]], i8** [[Y]]
// CHECK-NEXT: ret void
void test5(__weak id *p) {
id y = [*p retain];
}
// CHECK-LABEL: define void @_Z5test5PU6__weakP11objc_object
// CHECK: [[P:%.*]] = alloca i8**,
// CHECK-NEXT: [[Y:%.*]] = alloca i8*,
// CHECK-NEXT: store
// CHECK-NEXT: [[T0:%.*]] = load i8**, i8*** [[P]]
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK-NEXT: [[T1:%.*]] = call i8* @llvm.objc.loadWeakRetained(i8** [[T0]])
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-NEXT: store i8* [[T1]], i8** [[Y]]
// CHECK-NEXT: ret void
void test6(__weak Foo **p) {
Foo *y = [*p retain];
}
// CHECK-LABEL: define void @_Z5test6PU6__weakP3Foo
// CHECK: [[P:%.*]] = alloca [[FOO:%.*]]**,
// CHECK-NEXT: [[Y:%.*]] = alloca [[FOO]]*,
// CHECK-NEXT: store
// CHECK-NEXT: [[T0:%.*]] = load [[FOO]]**, [[FOO]]*** [[P]]
// CHECK-NEXT: [[T1:%.*]] = bitcast [[FOO]]** [[T0]] to i8**
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK-NEXT: [[T2:%.*]] = call i8* @llvm.objc.loadWeakRetained(i8** [[T1]])
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-NEXT: [[T3:%.*]] = bitcast i8* [[T2]] to [[FOO]]*
// CHECK-NEXT: store [[FOO]]* [[T3]], [[FOO]]** [[Y]]
// CHECK-NEXT: ret void
extern "C" id get_object(void);
extern "C" void use_block(void (^)(void));
void test7(void) {
__weak Foo *p = get_object();
use_block(^{ [p run ]; });
}
// CHECK-LABEL: define void @_Z5test7v
// CHECK: [[P:%.*]] = alloca [[FOO]]*,
// CHECK: [[T0:%.*]] = call i8* @get_object()
// CHECK-NEXT: [[T1:%.*]] = bitcast i8* [[T0]] to [[FOO]]*
// CHECK-NEXT: [[T2:%.*]] = bitcast [[FOO]]** [[P]] to i8**
// CHECK-NEXT: [[T3:%.*]] = bitcast [[FOO]]* [[T1]] to i8*
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK-NEXT: call i8* @llvm.objc.initWeak(i8** [[T2]], i8* [[T3]])
// CHECK: call void @llvm.objc.copyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK: call void @use_block
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: call void @llvm.objc.destroyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
[CodeGen] Merge equivalent block copy/helper functions. Clang generates copy and dispose helper functions for each block literal on the stack. Often these functions are equivalent for different blocks. This commit makes changes to merge equivalent copy and dispose helper functions and reduce code size. To enable merging equivalent copy/dispose functions, the captured object infomation is encoded into the helper function name. This allows IRGen to check whether an equivalent helper function has already been emitted and reuse the function instead of generating a new helper function whenever a block is defined. In addition, the helper functions are marked as linkonce_odr to enable merging helper functions that have the same name across translation units and marked as unnamed_addr to enable the linker's deduplication pass to merge functions that have different names but the same content. rdar://problem/42640608 Differential Revision: https://reviews.llvm.org/D50152 llvm-svn: 339438
2018-08-10 23:09:24 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__copy_helper_block
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: @llvm.objc.copyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
[CodeGen] Merge equivalent block copy/helper functions. Clang generates copy and dispose helper functions for each block literal on the stack. Often these functions are equivalent for different blocks. This commit makes changes to merge equivalent copy and dispose helper functions and reduce code size. To enable merging equivalent copy/dispose functions, the captured object infomation is encoded into the helper function name. This allows IRGen to check whether an equivalent helper function has already been emitted and reuse the function instead of generating a new helper function whenever a block is defined. In addition, the helper functions are marked as linkonce_odr to enable merging helper functions that have the same name across translation units and marked as unnamed_addr to enable the linker's deduplication pass to merge functions that have different names but the same content. rdar://problem/42640608 Differential Revision: https://reviews.llvm.org/D50152 llvm-svn: 339438
2018-08-10 23:09:24 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__destroy_helper_block
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: @llvm.objc.destroyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
void test8(void) {
__block __weak Foo *p = get_object();
use_block(^{ [p run ]; });
}
// CHECK-LABEL: define void @_Z5test8v
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: call i8* @llvm.objc.initWeak
// CHECK-NOT: call void @llvm.objc.copyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK: call void @use_block
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: call void @llvm.objc.destroyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-LABEL: define internal void @__Block_byref_object_copy
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: call void @llvm.objc.moveWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-LABEL: define internal void @__Block_byref_object_dispose
Generate objc intrinsics instead of runtime calls as the ARC optimizer now works only on intrinsics Differential Revision: https://reviews.llvm.org/D55802 Reviewers: rjmccall llvm-svn: 349535
2018-12-19 04:33:00 +08:00
// CHECK: call void @llvm.objc.destroyWeak
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
// CHECK-LABEL: define void @_Z14test9_baselinev()
[CodeGen] Merge equivalent block copy/helper functions. Clang generates copy and dispose helper functions for each block literal on the stack. Often these functions are equivalent for different blocks. This commit makes changes to merge equivalent copy and dispose helper functions and reduce code size. To enable merging equivalent copy/dispose functions, the captured object infomation is encoded into the helper function name. This allows IRGen to check whether an equivalent helper function has already been emitted and reuse the function instead of generating a new helper function whenever a block is defined. In addition, the helper functions are marked as linkonce_odr to enable merging helper functions that have the same name across translation units and marked as unnamed_addr to enable the linker's deduplication pass to merge functions that have different names but the same content. rdar://problem/42640608 Differential Revision: https://reviews.llvm.org/D50152 llvm-svn: 339438
2018-08-10 23:09:24 +08:00
// CHECK: define linkonce_odr hidden void @__copy_helper
// CHECK: define linkonce_odr hidden void @__destroy_helper
Don't actually add the __unsafe_unretained qualifier in MRC; driving a canonical difference between that and an unqualified type is a really bad idea when both are valid. Instead, remember that it was there in a non-canonical way, then look for that in the one place we really care about it: block captures. The net effect closely resembles the behavior of a decl attribute, except still closely following ARC's standard qualifier parsing rules. llvm-svn: 253534
2015-11-19 10:28:03 +08:00
void test9_baseline(void) {
Foo *p = get_object();
use_block(^{ [p run]; });
}
// CHECK-LABEL: define void @_Z5test9v()
// CHECK-NOT: define internal void @__copy_helper
// CHECK-NOT: define internal void @__destroy_helper
// CHECK: define void @_Z9test9_finv()
void test9(void) {
__unsafe_unretained Foo *p = get_object();
use_block(^{ [p run]; });
}
void test9_fin() {}
// CHECK-LABEL: define void @_Z6test10v()
// CHECK-NOT: define internal void @__copy_helper
// CHECK-NOT: define internal void @__destroy_helper
// CHECK: define void @_Z10test10_finv()
void test10(void) {
typedef __unsafe_unretained Foo *UnsafeFooPtr;
UnsafeFooPtr p = get_object();
use_block(^{ [p run]; });
}
void test10_fin() {}
// CHECK-LABEL: define weak_odr void @_Z6test11ILj0EEvv()
// CHECK-NOT: define internal void @__copy_helper
// CHECK-NOT: define internal void @__destroy_helper
// CHECK: define void @_Z10test11_finv()
template <unsigned i> void test11(void) {
typedef __unsafe_unretained Foo *UnsafeFooPtr;
UnsafeFooPtr p = get_object();
use_block(^{ [p run]; });
}
template void test11<0>();
void test11_fin() {}