Skip to content

Commit c5c8040

Browse files
shiltianshiltian
committed
[OpenMP] Introduce kernel environment
This patch introduces per kernel environment. Previously, flags such as execution mode are set through global variables with name like `__kernel_name_exec_mode`. They are accessible on the host by reading the corresponding global variable, but not from the device. Besides, some assumptions, such as no nested parallelism, are not per kernel basis, preventing us applying per kernel optimization in the device runtime. This is a combination and refinement of patch series D116908, D116909, and D116910. Depend on D155886. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D142569
1 parent 14b466b commit c5c8040

File tree

88 files changed

+8846
-3492
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

88 files changed

+8846
-3492
lines changed

‎clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp

Lines changed: 1 addition & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -795,7 +795,7 @@ void CGOpenMPRuntimeGPU::emitKernelDeinit(CodeGenFunction &CGF,
795795
emitGenericVarsEpilog(CGF);
796796

797797
CGBuilderTy &Bld = CGF.Builder;
798-
OMPBuilder.createTargetDeinit(Bld, IsSPMD);
798+
OMPBuilder.createTargetDeinit(Bld);
799799
}
800800

801801
void CGOpenMPRuntimeGPU::emitSPMDKernel(const OMPExecutableDirective &D,
@@ -833,24 +833,6 @@ void CGOpenMPRuntimeGPU::emitSPMDKernel(const OMPExecutableDirective &D,
833833
IsInTTDRegion = false;
834834
}
835835

836-
// Create a unique global variable to indicate the execution mode of this target
837-
// region. The execution mode is either 'generic', or 'spmd' depending on the
838-
// target directive. This variable is picked up by the offload library to setup
839-
// the device appropriately before kernel launch. If the execution mode is
840-
// 'generic', the runtime reserves one warp for the master, otherwise, all
841-
// warps participate in parallel work.
842-
static void setPropertyExecutionMode(CodeGenModule &CGM, StringRef Name,
843-
bool Mode) {
844-
auto *GVMode = new llvm::GlobalVariable(
845-
CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,
846-
llvm::GlobalValue::WeakAnyLinkage,
847-
llvm::ConstantInt::get(CGM.Int8Ty, Mode ? OMP_TGT_EXEC_MODE_SPMD
848-
: OMP_TGT_EXEC_MODE_GENERIC),
849-
Twine(Name, "_exec_mode"));
850-
GVMode->setVisibility(llvm::GlobalVariable::ProtectedVisibility);
851-
CGM.addCompilerUsedGlobal(GVMode);
852-
}
853-
854836
void CGOpenMPRuntimeGPU::emitTargetOutlinedFunction(
855837
const OMPExecutableDirective &D, StringRef ParentName,
856838
llvm::Function *&OutlinedFn, llvm::Constant *&OutlinedFnID,
@@ -867,8 +849,6 @@ void CGOpenMPRuntimeGPU::emitTargetOutlinedFunction(
867849
else
868850
emitNonSPMDKernel(D, ParentName, OutlinedFn, OutlinedFnID, IsOffloadEntry,
869851
CodeGen);
870-
871-
setPropertyExecutionMode(CGM, OutlinedFn->getName(), Mode);
872852
}
873853

874854
CGOpenMPRuntimeGPU::CGOpenMPRuntimeGPU(CodeGenModule &CGM)

‎clang/test/OpenMP/amdgcn_target_codegen.cpp

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -37,7 +37,7 @@ int test_amdgcn_target_tid_threads_simd() {
3737
// CHECK-NEXT: [[I_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[I]] to ptr
3838
// CHECK-NEXT: store ptr [[ARR]], ptr [[ARR_ADDR_ASCAST]], align 8
3939
// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[ARR_ADDR_ASCAST]], align 8
40-
// CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr), i8 1, i1 true)
40+
// CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z30test_amdgcn_target_tid_threadsv_l14_kernel_environment to ptr))
4141
// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
4242
// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
4343
// CHECK: user_code.entry:
@@ -61,7 +61,7 @@ int test_amdgcn_target_tid_threads_simd() {
6161
// CHECK: worker.exit:
6262
// CHECK-NEXT: ret void
6363
// CHECK: for.end:
64-
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
64+
// CHECK-NEXT: call void @__kmpc_target_deinit()
6565
// CHECK-NEXT: ret void
6666
//
6767
//
@@ -78,7 +78,7 @@ int test_amdgcn_target_tid_threads_simd() {
7878
// CHECK-NEXT: [[I_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[I]] to ptr
7979
// CHECK-NEXT: store ptr [[ARR]], ptr [[ARR_ADDR_ASCAST]], align 8
8080
// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[ARR_ADDR_ASCAST]], align 8
81-
// CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 2, i1 false)
81+
// CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z35test_amdgcn_target_tid_threads_simdv_l23_kernel_environment to ptr))
8282
// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
8383
// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
8484
// CHECK: user_code.entry:
@@ -109,6 +109,6 @@ int test_amdgcn_target_tid_threads_simd() {
109109
// CHECK-NEXT: ret void
110110
// CHECK: omp.inner.for.end:
111111
// CHECK-NEXT: store i32 1000, ptr [[I_ASCAST]], align 4
112-
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 2)
112+
// CHECK-NEXT: call void @__kmpc_target_deinit()
113113
// CHECK-NEXT: ret void
114114
//

‎clang/test/OpenMP/amdgcn_target_device_vla.cpp

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -111,7 +111,7 @@ int main() {
111111
// CHECK-NEXT: [[I1_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[I1]] to ptr
112112
// CHECK-NEXT: store ptr [[SUM]], ptr [[SUM_ADDR_ASCAST]], align 8
113113
// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[SUM_ADDR_ASCAST]], align 8
114-
// CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr), i8 1, i1 true)
114+
// CHECK-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4foo1v_l12_kernel_environment to ptr))
115115
// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
116116
// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
117117
// CHECK: user_code.entry:
@@ -169,7 +169,7 @@ int main() {
169169
// CHECK-NEXT: br label [[FOR_COND2]], !llvm.loop [[LOOP15:![0-9]+]]
170170
// CHECK: for.end9:
171171
// CHECK-NEXT: call void @__kmpc_free_shared(ptr [[A]], i64 [[TMP7]])
172-
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
172+
// CHECK-NEXT: call void @__kmpc_target_deinit()
173173
// CHECK-NEXT: ret void
174174
//
175175
//
@@ -193,18 +193,18 @@ int main() {
193193
// CHECK-NEXT: store ptr [[RESULT]], ptr [[RESULT_ADDR_ASCAST]], align 8
194194
// CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
195195
// CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr [[RESULT_ADDR_ASCAST]], align 8
196-
// CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 2, i1 false)
196+
// CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4foo2v_l30_kernel_environment to ptr))
197197
// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
198198
// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
199199
// CHECK: user_code.entry:
200-
// CHECK-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr))
200+
// CHECK-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_global_thread_num(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr))
201201
// CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[M_ADDR_ASCAST]], align 4
202202
// CHECK-NEXT: store i32 [[TMP4]], ptr [[M_CASTED_ASCAST]], align 4
203203
// CHECK-NEXT: [[TMP5:%.*]] = load i64, ptr [[M_CASTED_ASCAST]], align 8
204204
// CHECK-NEXT: store i32 0, ptr [[DOTZERO_ADDR_ASCAST]], align 4
205205
// CHECK-NEXT: store i32 [[TMP3]], ptr [[DOTTHREADID_TEMP__ASCAST]], align 4
206206
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4foo2v_l30_omp_outlined(ptr [[DOTTHREADID_TEMP__ASCAST]], ptr [[DOTZERO_ADDR_ASCAST]], i64 [[TMP5]], i64 [[TMP0]], ptr [[TMP1]]) #[[ATTR5:[0-9]+]]
207-
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 2)
207+
// CHECK-NEXT: call void @__kmpc_target_deinit()
208208
// CHECK-NEXT: ret void
209209
// CHECK: worker.exit:
210210
// CHECK-NEXT: ret void
@@ -559,7 +559,7 @@ int main() {
559559
// CHECK-NEXT: store ptr [[RESULT]], ptr [[RESULT_ADDR_ASCAST]], align 8
560560
// CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
561561
// CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr [[RESULT_ADDR_ASCAST]], align 8
562-
// CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1, i1 true)
562+
// CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4foo3v_l52_kernel_environment to ptr))
563563
// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
564564
// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
565565
// CHECK: user_code.entry:
@@ -570,7 +570,7 @@ int main() {
570570
// CHECK-NEXT: store i32 0, ptr [[DOTZERO_ADDR_ASCAST]], align 4
571571
// CHECK-NEXT: store i32 [[TMP3]], ptr [[DOTTHREADID_TEMP__ASCAST]], align 4
572572
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4foo3v_l52_omp_outlined(ptr [[DOTTHREADID_TEMP__ASCAST]], ptr [[DOTZERO_ADDR_ASCAST]], i64 [[TMP5]], i64 [[TMP0]], ptr [[TMP1]]) #[[ATTR5]]
573-
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
573+
// CHECK-NEXT: call void @__kmpc_target_deinit()
574574
// CHECK-NEXT: ret void
575575
// CHECK: worker.exit:
576576
// CHECK-NEXT: ret void
@@ -918,7 +918,7 @@ int main() {
918918
// CHECK-NEXT: store ptr [[RESULT]], ptr [[RESULT_ADDR_ASCAST]], align 8
919919
// CHECK-NEXT: [[TMP0:%.*]] = load i64, ptr [[VLA_ADDR_ASCAST]], align 8
920920
// CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr [[RESULT_ADDR_ASCAST]], align 8
921-
// CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1, i1 true)
921+
// CHECK-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4foo4v_l76_kernel_environment to ptr))
922922
// CHECK-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP2]], -1
923923
// CHECK-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
924924
// CHECK: user_code.entry:
@@ -932,7 +932,7 @@ int main() {
932932
// CHECK-NEXT: store i32 0, ptr [[DOTZERO_ADDR_ASCAST]], align 4
933933
// CHECK-NEXT: store i32 [[TMP3]], ptr [[DOTTHREADID_TEMP__ASCAST]], align 4
934934
// CHECK-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z4foo4v_l76_omp_outlined(ptr [[DOTTHREADID_TEMP__ASCAST]], ptr [[DOTZERO_ADDR_ASCAST]], i64 [[TMP5]], i64 [[TMP7]], i64 [[TMP0]], ptr [[TMP1]]) #[[ATTR5]]
935-
// CHECK-NEXT: call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 1)
935+
// CHECK-NEXT: call void @__kmpc_target_deinit()
936936
// CHECK-NEXT: ret void
937937
// CHECK: worker.exit:
938938
// CHECK-NEXT: ret void

‎clang/test/OpenMP/amdgpu_target_with_aligned_attribute.c

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -33,19 +33,19 @@ void write_to_aligned_array(int *a, int N) {
3333
// CHECK-AMD-NEXT: [[DOTTHREADID_TEMP__ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTTHREADID_TEMP_]] to ptr
3434
// CHECK-AMD-NEXT: store i64 [[N]], ptr [[N_ADDR_ASCAST]], align 8
3535
// CHECK-AMD-NEXT: store ptr [[APTR]], ptr [[APTR_ADDR_ASCAST]], align 8
36-
// CHECK-AMD-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr), i8 2, i1 false)
36+
// CHECK-AMD-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_target_init(ptr addrspacecast (ptr addrspace(1) @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_write_to_aligned_array_l14_kernel_environment to ptr))
3737
// CHECK-AMD-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP0]], -1
3838
// CHECK-AMD-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
3939
// CHECK-AMD: user_code.entry:
40-
// CHECK-AMD-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr))
40+
// CHECK-AMD-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_global_thread_num(ptr addrspacecast (ptr addrspace(1) @[[GLOB1:[0-9]+]] to ptr))
4141
// CHECK-AMD-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR_ASCAST]], align 4
4242
// CHECK-AMD-NEXT: store i32 [[TMP2]], ptr [[N_CASTED_ASCAST]], align 4
4343
// CHECK-AMD-NEXT: [[TMP3:%.*]] = load i64, ptr [[N_CASTED_ASCAST]], align 8
4444
// CHECK-AMD-NEXT: [[TMP4:%.*]] = load ptr, ptr [[APTR_ADDR_ASCAST]], align 8
4545
// CHECK-AMD-NEXT: store i32 0, ptr [[DOTZERO_ADDR_ASCAST]], align 4
4646
// CHECK-AMD-NEXT: store i32 [[TMP1]], ptr [[DOTTHREADID_TEMP__ASCAST]], align 4
4747
// CHECK-AMD-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_write_to_aligned_array_l14_omp_outlined(ptr [[DOTTHREADID_TEMP__ASCAST]], ptr [[DOTZERO_ADDR_ASCAST]], i64 [[TMP3]], ptr [[TMP4]]) #[[ATTR2:[0-9]+]]
48-
// CHECK-AMD-NEXT: call void @__kmpc_target_deinit(ptr addrspacecast (ptr addrspace(1) @[[GLOB1]] to ptr), i8 2)
48+
// CHECK-AMD-NEXT: call void @__kmpc_target_deinit()
4949
// CHECK-AMD-NEXT: ret void
5050
// CHECK-AMD: worker.exit:
5151
// CHECK-AMD-NEXT: ret void

‎clang/test/OpenMP/declare_target_codegen_globalization.cpp

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -31,15 +31,15 @@ int maini1() {
3131
// CHECK1-NEXT: [[CAPTURED_VARS_ADDRS:%.*]] = alloca [1 x ptr], align 8
3232
// CHECK1-NEXT: store ptr [[A]], ptr [[A_ADDR]], align 8
3333
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
34-
// CHECK1-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @[[GLOB1:[0-9]+]], i8 2, i1 false)
34+
// CHECK1-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_target_init(ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6maini1v_l16_kernel_environment)
3535
// CHECK1-NEXT: [[EXEC_USER_CODE:%.*]] = icmp eq i32 [[TMP1]], -1
3636
// CHECK1-NEXT: br i1 [[EXEC_USER_CODE]], label [[USER_CODE_ENTRY:%.*]], label [[WORKER_EXIT:%.*]]
3737
// CHECK1: user_code.entry:
38-
// CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1]])
38+
// CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
3939
// CHECK1-NEXT: [[TMP3:%.*]] = getelementptr inbounds [1 x ptr], ptr [[CAPTURED_VARS_ADDRS]], i64 0, i64 0
4040
// CHECK1-NEXT: store ptr [[TMP0]], ptr [[TMP3]], align 8
4141
// CHECK1-NEXT: call void @__kmpc_parallel_51(ptr @[[GLOB1]], i32 [[TMP2]], i32 1, i32 -1, i32 -1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z6maini1v_l16_omp_outlined, ptr null, ptr [[CAPTURED_VARS_ADDRS]], i64 1)
42-
// CHECK1-NEXT: call void @__kmpc_target_deinit(ptr @[[GLOB1]], i8 2)
42+
// CHECK1-NEXT: call void @__kmpc_target_deinit()
4343
// CHECK1-NEXT: ret void
4444
// CHECK1: worker.exit:
4545
// CHECK1-NEXT: ret void

0 commit comments

Comments
 (0)