maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/CodeGen/blocks.c

125 lines
5.6 KiB
C
Raw Normal View History

Track whether an AggValueSlot is potentially aliased, and do not emit call results into potentially aliased slots. This allows us to properly mark indirect return slots as noalias, at the cost of requiring an extra memcpy when assigning an aggregate call result into a l-value. It also brings us into compliance with the x86-64 ABI. llvm-svn: 138599
2011-08-26 07:04:34 +08:00
// RUN: %clang_cc1 -triple i386-unknown-unknown %s -emit-llvm -o - -fblocks | FileCheck %s
[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: %[[STRUCT_BLOCK_DESCRIPTOR:.*]] = type { i32, i32 }
[CodeGen] Before returning a copy/dispose helper function, bitcast it to a void pointer type. This fixes a bug introduced in r339438. llvm-svn: 339633
2018-08-14 08:15:42 +08:00
// CHECK: @{{.*}} = internal constant { i32, i32, i8*, i8*, i8*, i8* } { i32 0, i32 24, i8* bitcast (void (i8*, i8*)* @__copy_helper_block_4_20r to i8*), i8* bitcast (void (i8*)* @__destroy_helper_block_4_20r to i8*), i8* getelementptr inbounds ([6 x i8], [6 x i8]* @{{.*}}, i32 0, i32 0), i8* null }, align 4
// CHECK: @[[BLOCK_DESCRIPTOR_TMP21:.*]] = internal constant { i32, i32, i8*, i8*, i8*, i8* } { i32 0, i32 24, i8* bitcast (void (i8*, i8*)* @__copy_helper_block_4_20r to i8*), i8* bitcast (void (i8*)* @__destroy_helper_block_4_20r to i8*), i8* getelementptr inbounds ([6 x i8], [6 x i8]* @{{.*}}, i32 0, i32 0), i8* null }, align 4
[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
Do some blocks cleanup and simplification. Fix a crash, and add a test case. llvm-svn: 65746
2009-03-01 09:09:12 +08:00
void (^f)(void) = ^{};
Don't assume that a block always has a FunctionProtoType. Fixes rdar://6768379. llvm-svn: 68583
2009-04-08 10:55:55 +08:00
// rdar://6768379
int f0(int (^a0)()) {
return a0(1, 2, 3);
}
Attributes on block functions were not being set. - <rdar://problem/6800351> clang not producing correct large struct return code for Blocks llvm-svn: 69337
2009-04-17 08:48:04 +08:00
// Verify that attributes on blocks are set correctly.
typedef struct s0 T;
struct s0 {
int a[64];
};
Update testcases due to Attribute sorting improvements. llvm-svn: 175253
2013-02-15 13:25:49 +08:00
// CHECK: define internal void @__f2_block_invoke(%struct.s0* noalias sret {{%.*}}, i8* {{%.*}}, %struct.s0* byval align 4 {{.*}})
Attributes on block functions were not being set. - <rdar://problem/6800351> clang not producing correct large struct return code for Blocks llvm-svn: 69337
2009-04-17 08:48:04 +08:00
struct s0 f2(struct s0 a0) {
return ^(struct s0 a1){ return a1; }(a0);
}
use of predefined identifiers like __func__ at global scope warn in sema, but crashed codegen. Fix this to report the name of the llvm function. This fixes rdar://6808051 llvm-svn: 69658
2009-04-21 12:41:23 +08:00
// This should not crash: rdar://6808051
void *P = ^{
void *Q = __func__;
};
Don't assert when we think we need copy/dispose, but don't need them. Radar 6838889 llvm-svn: 70525
2009-05-01 09:31:57 +08:00
void (^test1)(void) = ^(void) {
__block int i;
^ { i = 1; }();
};
use of predefined identifiers like __func__ at global scope warn in sema, but crashed codegen. Fix this to report the name of the llvm function. This fixes rdar://6808051 llvm-svn: 69658
2009-04-21 12:41:23 +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_4_20r(i8*, i8*) unnamed_addr
// CHECK: %[[_ADDR:.*]] = alloca i8*, align 4
// CHECK-NEXT: %[[_ADDR1:.*]] = alloca i8*, align 4
// CHECK-NEXT: store i8* %0, i8** %[[_ADDR]], align 4
// CHECK-NEXT: store i8* %1, i8** %[[_ADDR1]], align 4
// CHECK-NEXT: %[[V2:.*]] = load i8*, i8** %[[_ADDR1]], align 4
// CHECK-NEXT: %[[BLOCK_SOURCE:.*]] = bitcast i8* %[[V2]] to <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>*
// CHECK-NEXT: %[[V3:.*]] = load i8*, i8** %[[_ADDR]], align 4
// CHECK-NEXT: %[[BLOCK_DEST:.*]] = bitcast i8* %[[V3]] to <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>*
// CHECK-NEXT: %[[V4:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK_SOURCE]], i32 0, i32 5
// CHECK-NEXT: %[[V5:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK_DEST]], i32 0, i32 5
// CHECK-NEXT: %[[BLOCKCOPY_SRC:.*]] = load i8*, i8** %[[V4]], align 4
// CHECK-NEXT: %[[V6:.*]] = bitcast i8** %[[V5]] to i8*
// CHECK-NEXT: call void @_Block_object_assign(i8* %[[V6]], i8* %[[BLOCKCOPY_SRC]], i32 8)
// CHECK-NEXT: ret void
// CHECK-LABEL: define linkonce_odr hidden void @__destroy_helper_block_4_20r(i8*) unnamed_addr
// CHECK: %[[_ADDR:.*]] = alloca i8*, align 4
// CHECK-NEXT: store i8* %0, i8** %[[_ADDR]], align 4
// CHECK-NEXT: %[[V1:.*]] = load i8*, i8** %[[_ADDR]], align 4
// CHECK-NEXT: %[[BLOCK:.*]] = bitcast i8* %[[V1]] to <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>*
// CHECK-NEXT: %[[V2:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %[[BLOCK]], i32 0, i32 5
// CHECK-NEXT: %[[V3:.*]] = load i8*, i8** %[[V2]], align 4
// CHECK-NEXT: call void @_Block_object_dispose(i8* %[[V3]], i32 8)
// CHECK-NEXT: ret void
Preserve more information from a block's original function declarator, if one was given. Remove some unnecessary accounting from BlockScopeInfo. Handle typedef'ed function types until such time as we decide not. llvm-svn: 105478
2010-06-05 03:02:56 +08:00
typedef double ftype(double);
// It's not clear that we *should* support this syntax, but until that decision
// is made, we should support it properly and not crash.
ftype ^test2 = ^ftype {
return 0;
};
When emitting l-values for bool non-__block decl references, make a pointer using the memory type; fixes an assert. Fixes rdar://problem/8605032 llvm-svn: 117610
2010-10-29 05:37:57 +08:00
// rdar://problem/8605032
void f3_helper(void (^)(void));
void f3() {
_Bool b = 0;
f3_helper(^{ if (b) {} });
}
Fix a bug with block layout when the block contains something more aligned than the block header but also contains something smaller than the block-header alignment but not exactly half the difference between the large alignment and the header alignment. Got that? I'm really not sure what I was thinking with the buggy computation here, but the fix is pretty obvious. llvm-svn: 155662
2012-04-27 05:14:42 +08:00
// rdar://problem/11322251
During block layout, after padding up to the max field alignment, the alignment might actually exceed the max field alignment; don't assert in this case. llvm-svn: 155937
2012-05-02 04:28:00 +08:00
// The bool can fill in between the header and the long long.
// Add the appropriate amount of padding between them.
Fix a bug with block layout when the block contains something more aligned than the block header but also contains something smaller than the block-header alignment but not exactly half the difference between the large alignment and the header alignment. Got that? I'm really not sure what I was thinking with the buggy computation here, but the fix is pretty obvious. llvm-svn: 155662
2012-04-27 05:14:42 +08:00
void f4_helper(long long (^)(void));
CHECK-LABEL-ify some code gen tests to improve diagnostic experience when tests fail. llvm-svn: 188447
2013-08-15 14:47:53 +08:00
// CHECK-LABEL: define void @f4()
Fix a bug with block layout when the block contains something more aligned than the block header but also contains something smaller than the block-header alignment but not exactly half the difference between the large alignment and the header alignment. Got that? I'm really not sure what I was thinking with the buggy computation here, but the fix is pretty obvious. llvm-svn: 155662
2012-04-27 05:14:42 +08:00
void f4(void) {
_Bool b = 0;
long long ll = 0;
During block layout, after padding up to the max field alignment, the alignment might actually exceed the max field alignment; don't assert in this case. llvm-svn: 155937
2012-05-02 04:28:00 +08:00
// CHECK: alloca <{ i8*, i32, i32, i8*, {{%.*}}*, i8, [3 x i8], i64 }>, align 8
Fix a bug with block layout when the block contains something more aligned than the block header but also contains something smaller than the block-header alignment but not exactly half the difference between the large alignment and the header alignment. Got that? I'm really not sure what I was thinking with the buggy computation here, but the fix is pretty obvious. llvm-svn: 155662
2012-04-27 05:14:42 +08:00
f4_helper(^{ if (b) return ll; return 0LL; });
}
During block layout, after padding up to the max field alignment, the alignment might actually exceed the max field alignment; don't assert in this case. llvm-svn: 155937
2012-05-02 04:28:00 +08:00
// rdar://problem/11354538
// The alignment after rounding up to the align of F5 is actually
// greater than the required alignment. Don't assert.
struct F5 {
char buffer[32] __attribute((aligned));
};
void f5_helper(void (^)(struct F5 *));
CHECK-LABEL-ify some code gen tests to improve diagnostic experience when tests fail. llvm-svn: 188447
2013-08-15 14:47:53 +08:00
// CHECK-LABEL: define void @f5()
During block layout, after padding up to the max field alignment, the alignment might actually exceed the max field alignment; don't assert in this case. llvm-svn: 155937
2012-05-02 04:28:00 +08:00
void f5(void) {
struct F5 value;
// CHECK: alloca <{ i8*, i32, i32, i8*, {{%.*}}*, [12 x i8], [[F5:%.*]] }>, align 16
f5_helper(^(struct F5 *slot) { *slot = value; });
}
blocks: fixes an ast bug when block pointer variable is evaluated in a condition expression and then dereferenced to envoke the block. This is pr15663 and I applied a slight variation of the patch with a test case. (patch is from Arthur O'Dwyer). Also // rdar://14085217 llvm-svn: 183471
2013-06-07 08:48:14 +08:00
// rdar://14085217
void (^b)() = ^{};
int main() {
(b?: ^{})();
}
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: [[ZERO:%.*]] = load void (...)*, void (...)** @b
blocks: fixes an ast bug when block pointer variable is evaluated in a condition expression and then dereferenced to envoke the block. This is pr15663 and I applied a slight variation of the patch with a test case. (patch is from Arthur O'Dwyer). Also // rdar://14085217 llvm-svn: 183471
2013-06-07 08:48:14 +08:00
// CHECK-NEXT: [[TB:%.*]] = icmp ne void (...)* [[ZERO]], null
// CHECK-NEXT: br i1 [[TB]], label [[CT:%.*]], label [[CF:%.*]]
// CHECK: [[ONE:%.*]] = bitcast void (...)* [[ZERO]] to void ()*
// CHECK-NEXT: br label [[CE:%.*]]
NFC, Add a test that ensure that we don't emit helper code in copy/dispose routines for variables that are const-captured This is a preparation commit that improves code-coverage in code that emits block copy/dispose routines. llvm-svn: 296040
2017-02-24 07:41:50 +08:00
// Ensure that we don't emit helper code in copy/dispose routines for variables
// that are const-captured.
void testConstCaptureInCopyAndDestroyHelpers() {
const int x = 0;
__block int i;
(^ { i = x; })();
}
[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 void @testConstCaptureInCopyAndDestroyHelpers(
// CHECK: %[[BLOCK_DESCRIPTOR:.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>, <{ i8*, i32, i32, i8*, %[[STRUCT_BLOCK_DESCRIPTOR]]*, i8* }>* %{{.*}}, i32 0, i32 4
[CodeGen] Before returning a copy/dispose helper function, bitcast it to a void pointer type. This fixes a bug introduced in r339438. llvm-svn: 339633
2018-08-14 08:15:42 +08:00
// CHECK: store %[[STRUCT_BLOCK_DESCRIPTOR]]* bitcast ({ i32, i32, i8*, i8*, i8*, i8* }* @[[BLOCK_DESCRIPTOR_TMP21]] to %[[STRUCT_BLOCK_DESCRIPTOR]]*), %[[STRUCT_BLOCK_DESCRIPTOR]]** %[[BLOCK_DESCRIPTOR]], align 4
[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 internal void @__testConstCaptureInCopyAndDestroyHelpers_block_invoke