maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/OpenMP/task_codegen.cpp

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

333 lines
18 KiB
C++
Raw Normal View History

[OPENMP] -fopenmp enables OpenMP support (fix for http://llvm.org/PR23492) -fopenmp turns on OpenMP support and links libiomp5 as OpenMP library. Also there is -fopenmp={libiomp5|libgomp} option that allows to override effect of -fopenmp and link libgomp library (if -fopenmp=libgomp is specified). Differential Revision: http://reviews.llvm.org/D9736 llvm-svn: 237769
2015-05-20 12:24:19 +08:00
// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -x c++ -emit-llvm %s -o - | FileCheck %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
[OpenMP] Lower taskwait using OpenMP IR Builder The code generation is exactly the same as it was. But not that the special handling of untied tasks is still handled by emitUntiedSwitch in clang. Differential Revision: https://reviews.llvm.org/D69828
2019-12-12 17:19:10 +08:00
//
// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fopenmp-enable-irbuilder -x c++ -emit-llvm %s -o - | FileCheck %s
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
[OPENMP] Support for -fopenmp-simd option with compilation of simd loops only. Added support for -fopenmp-simd option that allows compilation of simd-based constructs without emission of OpenMP runtime calls. llvm-svn: 321560
2017-12-30 02:07:07 +08:00
// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -x c++ -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-apple-darwin10 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
// CHECK-DAG: [[IDENT_T:%.+]] = type { i32, i32, i32, i32, i8* }
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK-DAG: [[STRUCT_SHAREDS:%.+]] = type { i8*, [2 x [[STRUCT_S:%.+]]]* }
// CHECK-DAG: [[STRUCT_SHAREDS1:%.+]] = type { [2 x [[STRUCT_S:%.+]]]* }
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK-DAG: [[KMP_TASK_T:%.+]] = type { i8*, i32 (i32, i8*)*, i32, %union{{.+}}, %union{{.+}} }
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK-DAG: [[KMP_DEPEND_INFO:%.+]] = type { i64, i64, i8 }
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
struct S {
int a;
S() : a(0) {}
S(const S &s) : a(s.a) {}
~S() {}
};
int a;
Fix CHECK directives that weren't checking. llvm-svn: 246492
2015-09-01 05:48:52 +08:00
// CHECK-LABEL: @main
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
int main() {
// CHECK: [[B:%.+]] = alloca i8
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: [[S:%.+]] = alloca [2 x [[STRUCT_S]]]
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
char b;
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
S s[2];
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
int arr[10][a];
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T]]* @{{.+}})
// CHECK: [[B_REF:%.+]] = getelementptr inbounds [[STRUCT_SHAREDS]], [[STRUCT_SHAREDS]]* [[CAPTURES:%.+]], i32 0, i32 0
// CHECK: store i8* [[B]], i8** [[B_REF]]
// CHECK: [[S_REF:%.+]] = getelementptr inbounds [[STRUCT_SHAREDS]], [[STRUCT_SHAREDS]]* [[CAPTURES]], i32 0, i32 1
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: store [2 x [[STRUCT_S]]]* [[S]], [2 x [[STRUCT_S]]]** [[S_REF]]
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 33, i64 40, i64 16, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY1:@.+]] to i32 (i32, i8*)*))
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: [[SHAREDS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR:%.+]], i32 0, i32 0
// CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_PTR]]
// CHECK: [[BITCAST:%.+]] = bitcast [[STRUCT_SHAREDS]]* [[CAPTURES]] to i8*
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.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[BITCAST]], i64 16, i1 false)
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[PRIORITY_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR]], i32 0, i32 4
// CHECK: [[PRIORITY:%.+]] = bitcast %union{{.+}}* [[PRIORITY_REF_PTR]] to i32*
// CHECK: store i32 {{.+}}, i32* [[PRIORITY]]
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
[OPENMP 4.0] Fixed codegen for destructors in task-based directives. If private variables require destructors call at the deletion of the task, additional flag in task flags must be set. Patch fixes this problem. llvm-svn: 269039
2016-05-10 18:36:51 +08:00
#pragma omp task shared(a, b, s) priority(b)
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
{
a = 15;
b = a;
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
s[0].a = 10;
}
// CHECK: [[S_REF:%.+]] = getelementptr inbounds [[STRUCT_SHAREDS1]], [[STRUCT_SHAREDS1]]* [[CAPTURES:%.+]], i32 0, i32 0
// CHECK: store [2 x [[STRUCT_S]]]* [[S]], [2 x [[STRUCT_S]]]** [[S_REF]]
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{[^,]+}}, i32 [[GTID]], i32 1, i64 40, i64 8,
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: [[SHAREDS_REF_PTR:%.+]] = getelementptr inbounds [[KMP_TASK_T]], [[KMP_TASK_T]]* [[TASK_PTR:%.+]], i32 0, i32 0
// CHECK: [[SHAREDS_REF:%.+]] = load i8*, i8** [[SHAREDS_REF_PTR]]
// CHECK: [[BITCAST:%.+]] = bitcast [[STRUCT_SHAREDS1]]* [[CAPTURES]] to i8*
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.memcpy.p0i8.p0i8.i64(i8* align 8 [[SHAREDS_REF]], i8* align 8 [[BITCAST]], i64 8, i1 false)
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: [[DEP_BASE:%.*]] = getelementptr inbounds [4 x [[KMP_DEPEND_INFO]]], [4 x [[KMP_DEPEND_INFO]]]* [[DEPENDENCIES:%.*]], i64 0, i64 0
// CHECK: [[DEP:%.+]] = getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP_BASE]], i64 0
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: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 0
// CHECK: store i64 ptrtoint (i32* @{{.+}} to i64), i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 1
// CHECK: store i64 4, i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 2
// CHECK: store i8 1, i8* [[T0]]
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: [[DEP:%.*]] = getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP_BASE]], i64 1
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: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 0
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: ptrtoint i8* [[B]] to i64
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: store i64 %{{[^,]+}}, i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 1
// CHECK: store i64 1, i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 2
// CHECK: store i8 1, i8* [[T0]]
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: [[DEP:%.*]] = getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP_BASE]], i64 2
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: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 0
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: ptrtoint [2 x [[STRUCT_S]]]* [[S]] to i64
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: store i64 %{{[^,]+}}, i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 1
// CHECK: store i64 8, i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 2
// CHECK: store i8 1, i8* [[T0]]
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: [[IDX1:%.+]] = mul nsw i64 0, [[A_VAL:%.+]]
// CHECK: [[START:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
// CHECK: [[IDX1:%.+]] = mul nsw i64 9, [[A_VAL]]
// CHECK: [[END:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
// CHECK: [[END1:%.+]] = getelementptr i32, i32* [[END]], i32 1
// CHECK: [[START_INT:%.+]] = ptrtoint i32* [[START]] to i64
// CHECK: [[END_INT:%.+]] = ptrtoint i32* [[END1]] to i64
// CHECK: [[SIZEOF:%.+]] = sub nuw i64 [[END_INT]], [[START_INT]]
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: [[DEP:%.*]] = getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP_BASE]], i64 3
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: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* [[DEP]], i32 0, i32 0
// CHECK: [[T1:%.*]] = ptrtoint i32* [[START]] to i64
// CHECK: store i64 [[T1]], i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 [[SIZEOF]], i64* [[T0]]
// CHECK: [[T0:%.*]] = getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
// CHECK: store i8 1, i8* [[T0]]
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: bitcast [[KMP_DEPEND_INFO]]* [[DEP_BASE]] to i8*
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: call i32 @__kmpc_omp_task_with_deps([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]], i32 4, i8* %{{[^,]+}}, i32 0, i8* null)
#pragma omp task shared(a, s) depend(in : a, b, s, arr[:])
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
{
a = 15;
s[1].a = 10;
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
}
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY2:@.+]] to i32 (i32, i8*)*))
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
#pragma omp task untied
{
[OPENMP]Fix PR39422: variables are not firstprivatized in task context. According to the OpenMP standard, In a task generating construct, if no default clause is present, a variable for which the data-sharing attribute is not determined by the rules above is firstprivatized. Compiler tries to implement this, but if the variable is not directly used in the task context, this variable may not be firstprivatized. Patch fixes this problem. llvm-svn: 345277
2018-10-25 23:35:27 +08:00
#pragma omp critical
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
a = 1;
}
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 40, i64 1,
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: getelementptr inbounds [2 x [[STRUCT_S]]], [2 x [[STRUCT_S]]]* [[S]], i64 0, i64 0
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
// CHECK: ptrtoint [[STRUCT_S]]* %{{.+}} to i64
// CHECK: store i64 %{{[^,]+}}, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 4, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
[OPENMP] 'out' dependency for 'task' directives must be the same as 'inout'. Runtime library requires, that codegen for 'depend' clause for 'out' dependency kind must be the same as codegen for 'depend' clause with 'inout' dependency. llvm-svn: 253866
2015-11-23 21:33:42 +08:00
// CHECK: store i8 3, i8*
[OPENMP 4.0] Fixed support of array sections/array subscripts. Codegen for array sections/array subscripts worked only for expressions with arrays as base. Patch fixes codegen for bases with pointer/reference types. llvm-svn: 259776
2016-02-04 19:27:03 +08:00
// CHECK: [[B_VAL:%.+]] = load i8, i8* [[B]]
// CHECK: [[IDX2:%.+]] = sext i8 [[B_VAL]] to i64
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: [[IDX1:%.+]] = mul nsw i64 4, [[A_VAL]]
// CHECK: [[START:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
[OPENMP 4.0] Fixed support of array sections/array subscripts. Codegen for array sections/array subscripts worked only for expressions with arrays as base. Patch fixes codegen for bases with pointer/reference types. llvm-svn: 259776
2016-02-04 19:27:03 +08:00
// CHECK: [[START1:%.+]] = getelementptr inbounds i32, i32* [[START]], i64 [[IDX2]]
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: [[B_VAL:%.+]] = load i8, i8* [[B]]
// CHECK: [[IDX2:%.+]] = sext i8 [[B_VAL]] to i64
// CHECK: [[IDX1:%.+]] = mul nsw i64 9, [[A_VAL]]
// CHECK: [[END:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
// CHECK: [[END1:%.+]] = getelementptr inbounds i32, i32* [[END]], i64 [[IDX2]]
// CHECK: [[END2:%.+]] = getelementptr i32, i32* [[END1]], i32 1
// CHECK: [[START_INT:%.+]] = ptrtoint i32* [[START1]] to i64
// CHECK: [[END_INT:%.+]] = ptrtoint i32* [[END2]] to i64
// CHECK: [[SIZEOF:%.+]] = sub nuw i64 [[END_INT]], [[START_INT]]
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i64 1
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
// CHECK: ptrtoint i32* [[START1]] to i64
// CHECK: store i64 %{{[^,]+}}, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 [[SIZEOF]], i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
[OPENMP] 'out' dependency for 'task' directives must be the same as 'inout'. Runtime library requires, that codegen for 'depend' clause for 'out' dependency kind must be the same as codegen for 'depend' clause with 'inout' dependency. llvm-svn: 253866
2015-11-23 21:33:42 +08:00
// CHECK: store i8 3, i8*
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: bitcast [[KMP_DEPEND_INFO]]* %{{.+}} to i8*
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: call i32 @__kmpc_omp_task_with_deps([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]], i32 2, i8* %{{[^,]+}}, i32 0, i8* null)
#pragma omp task untied depend(out : s[0], arr[4:][b])
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
{
a = 1;
}
[OpenMP] Adding support to the mutexinoutset dep-type Summary: this commit adds support to a new dependence type introduced in OpenMP 5.0. The LLVM OpenMP RTL already supports this feature, so we only need to modify CLANG to take advantage of them. Differential Revision: https://reviews.llvm.org/D57576 llvm-svn: 353018
2019-02-04 15:33:19 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 40, i64 1,
// CHECK: getelementptr inbounds [2 x [[STRUCT_S]]], [2 x [[STRUCT_S]]]* [[S]], i64 0, i64 0
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
// CHECK: ptrtoint [[STRUCT_S]]* %{{.+}} to i64
// CHECK: store i64 %{{[^,]+}}, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 4, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
// CHECK: store i8 4, i8*
// CHECK: [[B_VAL:%.+]] = load i8, i8* [[B]]
// CHECK: [[IDX2:%.+]] = sext i8 [[B_VAL]] to i64
// CHECK: [[IDX1:%.+]] = mul nsw i64 4, [[A_VAL]]
// CHECK: [[START:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
// CHECK: [[START1:%.+]] = getelementptr inbounds i32, i32* [[START]], i64 [[IDX2]]
// CHECK: [[B_VAL:%.+]] = load i8, i8* [[B]]
// CHECK: [[IDX2:%.+]] = sext i8 [[B_VAL]] to i64
// CHECK: [[IDX1:%.+]] = mul nsw i64 9, [[A_VAL]]
// CHECK: [[END:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
// CHECK: [[END1:%.+]] = getelementptr inbounds i32, i32* [[END]], i64 [[IDX2]]
// CHECK: [[END2:%.+]] = getelementptr i32, i32* [[END1]], i32 1
// CHECK: [[START_INT:%.+]] = ptrtoint i32* [[START1]] to i64
// CHECK: [[END_INT:%.+]] = ptrtoint i32* [[END2]] to i64
// CHECK: [[SIZEOF:%.+]] = sub nuw i64 [[END_INT]], [[START_INT]]
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i64 1
[OpenMP] Adding support to the mutexinoutset dep-type Summary: this commit adds support to a new dependence type introduced in OpenMP 5.0. The LLVM OpenMP RTL already supports this feature, so we only need to modify CLANG to take advantage of them. Differential Revision: https://reviews.llvm.org/D57576 llvm-svn: 353018
2019-02-04 15:33:19 +08:00
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
// CHECK: ptrtoint i32* [[START1]] to i64
// CHECK: store i64 %{{[^,]+}}, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 [[SIZEOF]], i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
// CHECK: store i8 4, i8*
// CHECK: bitcast [[KMP_DEPEND_INFO]]* %{{.+}} to i8*
// CHECK: call i32 @__kmpc_omp_task_with_deps([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]], i32 2, i8* %{{[^,]+}}, i32 0, i8* null)
#pragma omp task untied depend(mutexinoutset: s[0], arr[4:][b])
{
a = 1;
}
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 3, i64 40, i64 1,
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
// CHECK: store i64 ptrtoint (i32* @{{.+}} to i64), i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 4, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
// CHECK: store i8 3, 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: getelementptr inbounds [2 x [[STRUCT_S]]], [2 x [[STRUCT_S]]]* [[S]], i64 0, i64 1
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i64 1
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
// CHECK: ptrtoint [[STRUCT_S]]* %{{.+}} to i64
// CHECK: store i64 %{{[^,]+}}, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 4, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
// CHECK: store i8 3, i8*
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: [[IDX1:%.+]] = mul nsw i64 0, [[A_VAL]]
// CHECK: [[START:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
// CHECK: [[START1:%.+]] = getelementptr inbounds i32, i32* [[START]], i64 3
// CHECK: [[NEW_A_VAL:%.+]] = load i32, i32* @{{.+}},
// CHECK: [[NEW_A_VAL_I64:%.+]] = sext i32 [[NEW_A_VAL]] to i64
[OPENMP 4.0] Fixed support of array sections/array subscripts. Codegen for array sections/array subscripts worked only for expressions with arrays as base. Patch fixes codegen for bases with pointer/reference types. llvm-svn: 259776
2016-02-04 19:27:03 +08:00
// CHECK: [[IDX2:%.+]] = sub nsw i64 [[NEW_A_VAL_I64]], 1
// CHECK: [[NEW_A_VAL:%.+]] = load i32, i32* @{{.+}},
// CHECK: [[NEW_A_VAL_I64:%.+]] = sext i32 [[NEW_A_VAL]] to i64
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: [[SUB:%.+]] = add nsw i64 -1, [[NEW_A_VAL_I64]]
// CHECK: [[IDX1:%.+]] = mul nsw i64 [[SUB]], [[A_VAL]]
// CHECK: [[END:%.+]] = getelementptr inbounds i32, i32* %{{.+}}, i64 [[IDX1]]
// CHECK: [[END1:%.+]] = getelementptr inbounds i32, i32* [[END]], i64 [[IDX2]]
// CHECK: [[END2:%.+]] = getelementptr i32, i32* [[END1]], i32 1
// CHECK: [[START_INT:%.+]] = ptrtoint i32* [[START1]] to i64
// CHECK: [[END_INT:%.+]] = ptrtoint i32* [[END2]] to i64
// CHECK: [[SIZEOF:%.+]] = sub nuw i64 [[END_INT]], [[START_INT]]
[OPENMP50]Codegen for iterator construct. Implemented codegen for the iterator expression in the depend clauses. Iterator construct is emitted the following way: iterator(cnt1, cnt2, ...), in : <dep> <TotalNumDeps> = <cnt1_size> * <cnt2_size> * ...; kmp_depend_t deps[<TotalNumDeps>]; deps_counter = 0; for (cnt1) { for (cnt2) { ... deps[deps_counter].base_addr = &<dep>; deps[deps_counter].size = sizeof(<dep>); deps[deps_counter].flags = in; deps_counter += 1; ... } } For depobj construct the codegen is very similar, but the memory is allocated dynamically and added extra first item reserved for internal use.
2020-04-06 23:21:43 +08:00
// CHECK: getelementptr [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i64 2
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 0
// CHECK: ptrtoint i32* [[START1]] to i64
// CHECK: store i64 %{{[^,]+}}, i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 1
// CHECK: store i64 [[SIZEOF]], i64*
// CHECK: getelementptr inbounds [[KMP_DEPEND_INFO]], [[KMP_DEPEND_INFO]]* %{{[^,]+}}, i32 0, i32 2
// CHECK: store i8 3, i8*
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
// CHECK: bitcast [[KMP_DEPEND_INFO]]* %{{.+}} to i8*
[OPENMP 4.0] Codegen for array sections. Added codegen for array section in 'depend' clause of 'task' directive. It emits to pointers, one for the begin of array section and another for the end of array section. Size of the section is calculated as (end + 1 - start) * sizeof(basic_element_type). llvm-svn: 246422
2015-08-31 15:32:19 +08:00
// CHECK: call i32 @__kmpc_omp_task_with_deps([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]], i32 3, i8* %{{[^,]+}}, i32 0, i8* null)
#pragma omp task final(true) depend(inout: a, s[1], arr[:a][3:])
[OPENMP] Codegen for 'depend' clause (OpenMP 4.0). If task directive has associated 'depend' clause then function kmp_int32 __kmpc_omp_task_with_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list,kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called instead of __kmpc_omp_task(). If this directive has associated 'if' clause then also before a call of kmpc_omp_task_begin_if0() a function void __kmpc_omp_wait_deps ( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list) must be called. Array sections are not supported yet. llvm-svn: 240532
2015-06-24 19:01:36 +08:00
{
a = 2;
}
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 3, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY3:@.+]] to i32 (i32, i8*)*))
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
#pragma omp task final(true)
{
a = 2;
}
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 1, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY4:@.+]] to i32 (i32, i8*)*))
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
const bool flag = false;
#pragma omp task final(flag)
{
a = 3;
}
// CHECK: [[B_VAL:%.+]] = load i8, i8* [[B]]
// CHECK: [[CMP:%.+]] = icmp ne i8 [[B_VAL]], 0
// CHECK: [[FINAL:%.+]] = select i1 [[CMP]], i32 2, i32 0
// CHECK: [[FLAGS:%.+]] = or i32 [[FINAL]], 1
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 [[FLAGS]], i64 40, i64 8, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY5:@.+]] to i32 (i32, i8*)*))
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
[OPENMP] Preserve alignment of the original variables for the captured references. Patch makes codegen to preserve alignment of the shared variables captured and used in OpenMP regions. llvm-svn: 247401
2015-09-11 18:29:41 +08:00
int c __attribute__((aligned(128)));
#pragma omp task final(b) shared(c)
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
{
a = 4;
[OPENMP] Preserve alignment of the original variables for the captured references. Patch makes codegen to preserve alignment of the shared variables captured and used in OpenMP regions. llvm-svn: 247401
2015-09-11 18:29:41 +08:00
c = 5;
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
}
[OPENMP 4.5] Fixed codegen for 'priority' and destructors in task-based directives. 'kmp_task_t' record type added a new field for 'priority' clause and changed the representation of pointer to destructors for privates used within loop-based directives. Old representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */ /* private vars */ } kmp_task_t; New representation: typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */ void *shareds; /**< pointer to block of pointers to shared vars */ kmp_routine_entry_t routine; /**< pointer to routine to call for executing task */ kmp_int32 part_id; /**< part id for the task */ kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */ kmp_cmplrdata_t data2; /* Process destructors first, priority second */ /* future data */ /* private vars */ } kmp_task_t; Also excessive initialization of 'destructors' fields to 'null' was removed from codegen if it is known that no destructors shal be used. Currently a special bit is used in 'kmp_tasking_flags_t' bitfields ('destructors_thunk' bitfield). llvm-svn: 271201
2016-05-30 17:06:50 +08:00
// CHECK: [[ORIG_TASK_PTR:%.+]] = call i8* @__kmpc_omp_task_alloc([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i32 0, i64 40, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, [[KMP_TASK_T]]{{.*}}*)* [[TASK_ENTRY6:@.+]] to i32 (i32, i8*)*))
[OPENMP] Codegen for untied tasks. If the untied clause is present on a task construct, any thread in the team can resume the task region after a suspension. Patch adds proper codegen for untied tasks. llvm-svn: 266853
2016-04-20 12:01:36 +08:00
// CHECK: call i32 @__kmpc_omp_task([[IDENT_T]]* @{{.+}}, i32 [[GTID]], i8* [[ORIG_TASK_PTR]])
#pragma omp task untied
{
S s1;
#pragma omp task
a = 4;
#pragma omp taskyield
s1 = S();
#pragma omp taskwait
}
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
return a;
}
IR: print value numbers for unnamed function arguments For consistency with normal instructions and clarity when reading IR, it's best to print the %0, %1, ... names of function arguments in definitions. Also modifies the parser to accept IR in that form for obvious reasons. llvm-svn: 367755
2019-08-03 22:28:34 +08:00
// CHECK: define internal i32 [[TASK_ENTRY1]](i32 %0, [[KMP_TASK_T]]{{.*}}* noalias %1)
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
// CHECK: store i32 15, i32* [[A_PTR:@.+]]
// CHECK: [[A_VAL:%.+]] = load i32, i32* [[A_PTR]]
// CHECK: [[A_VAL_I8:%.+]] = trunc i32 [[A_VAL]] to i8
// CHECK: store i8 [[A_VAL_I8]], i8* %{{.+}}
// CHECK: store i32 10, i32* %{{.+}}
IR: print value numbers for unnamed function arguments For consistency with normal instructions and clarity when reading IR, it's best to print the %0, %1, ... names of function arguments in definitions. Also modifies the parser to accept IR in that form for obvious reasons. llvm-svn: 367755
2019-08-03 22:28:34 +08:00
// CHECK: define internal i32 [[TASK_ENTRY2]](i32 %0, [[KMP_TASK_T]]{{.*}}* noalias %1)
[OPENMP] Codegen for untied tasks. If the untied clause is present on a task construct, any thread in the team can resume the task region after a suspension. Patch adds proper codegen for untied tasks. llvm-svn: 266853
2016-04-20 12:01:36 +08:00
// CHECK: store i32 1, i32* [[A_PTR]]
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
IR: print value numbers for unnamed function arguments For consistency with normal instructions and clarity when reading IR, it's best to print the %0, %1, ... names of function arguments in definitions. Also modifies the parser to accept IR in that form for obvious reasons. llvm-svn: 367755
2019-08-03 22:28:34 +08:00
// CHECK: define internal i32 [[TASK_ENTRY3]](i32 %0, [[KMP_TASK_T]]{{.*}}* noalias %1)
[OPENMP] Codegen for untied tasks. If the untied clause is present on a task construct, any thread in the team can resume the task region after a suspension. Patch adds proper codegen for untied tasks. llvm-svn: 266853
2016-04-20 12:01:36 +08:00
// CHECK: store i32 2, i32* [[A_PTR]]
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
IR: print value numbers for unnamed function arguments For consistency with normal instructions and clarity when reading IR, it's best to print the %0, %1, ... names of function arguments in definitions. Also modifies the parser to accept IR in that form for obvious reasons. llvm-svn: 367755
2019-08-03 22:28:34 +08:00
// CHECK: define internal i32 [[TASK_ENTRY4]](i32 %0, [[KMP_TASK_T]]{{.*}}* noalias %1)
[OPENMP] Codegen for untied tasks. If the untied clause is present on a task construct, any thread in the team can resume the task region after a suspension. Patch adds proper codegen for untied tasks. llvm-svn: 266853
2016-04-20 12:01:36 +08:00
// CHECK: store i32 3, i32* [[A_PTR]]
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
IR: print value numbers for unnamed function arguments For consistency with normal instructions and clarity when reading IR, it's best to print the %0, %1, ... names of function arguments in definitions. Also modifies the parser to accept IR in that form for obvious reasons. llvm-svn: 367755
2019-08-03 22:28:34 +08:00
// CHECK: define internal i32 [[TASK_ENTRY5]](i32 %0, [[KMP_TASK_T]]{{.*}}* noalias %1)
[OPENMP] Codegen for untied tasks. If the untied clause is present on a task construct, any thread in the team can resume the task region after a suspension. Patch adds proper codegen for untied tasks. llvm-svn: 266853
2016-04-20 12:01:36 +08:00
// CHECK: store i32 4, i32* [[A_PTR]]
[OPENMP] Preserve alignment of the original variables for the captured references. Patch makes codegen to preserve alignment of the shared variables captured and used in OpenMP regions. llvm-svn: 247401
2015-09-11 18:29:41 +08:00
// CHECK: store i32 5, i32* [[C_PTR:%.+]], align 128
[OPENMP] Codegen for untied tasks. If the untied clause is present on a task construct, any thread in the team can resume the task region after a suspension. Patch adds proper codegen for untied tasks. llvm-svn: 266853
2016-04-20 12:01:36 +08:00
// CHECK: define internal i32
// CHECK: store i32 4, i32* [[A_PTR]]
IR: print value numbers for unnamed function arguments For consistency with normal instructions and clarity when reading IR, it's best to print the %0, %1, ... names of function arguments in definitions. Also modifies the parser to accept IR in that form for obvious reasons. llvm-svn: 367755
2019-08-03 22:28:34 +08:00
// CHECK: define internal i32 [[TASK_ENTRY6]](i32 %0, [[KMP_TASK_T]]{{.*}}* noalias %1)
[OPENMP] Codegen for untied tasks. If the untied clause is present on a task construct, any thread in the team can resume the task region after a suspension. Patch adds proper codegen for untied tasks. llvm-svn: 266853
2016-04-20 12:01:36 +08:00
// CHECK: switch i32 %{{.+}}, label
// CHECK: load i32*, i32** %
// CHECK: store i32 1, i32* %
// CHECK: call i32 @__kmpc_omp_task(%
// CHECK: call i8* @__kmpc_omp_task_alloc(
// CHECK: call i32 @__kmpc_omp_task(%
// CHECK: load i32*, i32** %
// CHECK: store i32 2, i32* %
// CHECK: call i32 @__kmpc_omp_task(%
// CHECK: call i32 @__kmpc_omp_taskyield(%
// CHECK: load i32*, i32** %
// CHECK: store i32 3, i32* %
// CHECK: call i32 @__kmpc_omp_task(%
// CHECK: call i32 @__kmpc_omp_taskwait(%
// CHECK: load i32*, i32** %
// CHECK: store i32 4, i32* %
// CHECK: call i32 @__kmpc_omp_task(%
[OPENMP]Fix PR39366: do not try to private field if it is not captured. The compiler is crashing if we trying to post-capture the fields implicitly captured inside of the task constructs. Seems, this kind of processing is not supported and such fields should not be firstprivatized. llvm-svn: 345177
2018-10-25 02:53:12 +08:00
struct S1 {
int a;
S1() { taskinit(); }
void taskinit() {
#pragma omp task
a = 0;
}
} s1;
// CHECK-LABEL: taskinit
// CHECK: call i8* @__kmpc_omp_task_alloc(
[OPENMP] Initial codegen for 'omp task' directive. The task region is emmitted in several steps: Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds, kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the function: kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) { TaskFunction(gtid, tt->part_id, tt->shareds); return 0; } Copy a list of shared variables to field shareds of the resulting structure kmp_task_t returned by the previous call (if any). Copy a pointer to destructions function to field destructions of the resulting structure kmp_task_t. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *new_task), where new_task is a resulting structure from previous items. Differential Revision: http://reviews.llvm.org/D7560 llvm-svn: 231762
2015-03-10 15:28:44 +08:00
#endif