maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[OPENMP] ThreadId in serialized parallel regions is 0. 

The first argument for the parallel outlined functions, called as
serialized parallel regions, should be a pointer to the global thread id
that always is 0.

llvm-svn: 337957
This commit is contained in:
Alexey Bataev 2018-07-25 20:03:01 +00:00
parent b6613ac665
commit 8521ff6ec4
4 changed files with 28 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
clang/lib/CodeGen/CGOpenMPRuntime.cpp
View File

@ -2839,12 +2839,12 @@ void CGOpenMPRuntime::emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
RT.createRuntimeFunction(OMPRTL__kmpc_serialized_parallel), Args); RT.createRuntimeFunction(OMPRTL__kmpc_serialized_parallel), Args);
// OutlinedFn(&GTid, &zero, CapturedStruct); // OutlinedFn(&GTid, &zero, CapturedStruct);
Address ThreadIDAddr = RT.emitThreadIDAddress(CGF, Loc);
Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty, Address ZeroAddr = CGF.CreateDefaultAlignTempAlloca(CGF.Int32Ty,
/*Name*/ ".zero.addr"); /*Name*/ ".zero.addr");
CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0)); CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs; llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
OutlinedFnArgs.push_back(ThreadIDAddr.getPointer()); // ThreadId for serialized parallels is 0.
OutlinedFnArgs.push_back(ZeroAddr.getPointer());
OutlinedFnArgs.push_back(ZeroAddr.getPointer()); OutlinedFnArgs.push_back(ZeroAddr.getPointer());
OutlinedFnArgs.append(CapturedVars.begin(), CapturedVars.end()); OutlinedFnArgs.append(CapturedVars.begin(), CapturedVars.end());
RT.emitOutlinedFunctionCall(CGF, Loc, OutlinedFn, OutlinedFnArgs); RT.emitOutlinedFunctionCall(CGF, Loc, OutlinedFn, OutlinedFnArgs);

21
clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
View File

@ -1784,8 +1784,9 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDParallelCall(
/*DestWidth=*/32, /*Signed=*/1), /*DestWidth=*/32, /*Signed=*/1),
".zero.addr"); ".zero.addr");
CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0)); CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
Address ThreadIDAddr = emitThreadIDAddress(CGF, Loc); // ThreadId for serialized parallels is 0.
auto &&CodeGen = [this, Fn, CapturedVars, Loc, ZeroAddr, ThreadIDAddr]( Address ThreadIDAddr = ZeroAddr;
auto &&CodeGen = [this, Fn, CapturedVars, Loc, ZeroAddr, &ThreadIDAddr](
CodeGenFunction &CGF, PrePostActionTy &Action) { CodeGenFunction &CGF, PrePostActionTy &Action) {
Action.Enter(CGF); Action.Enter(CGF);
@ -1883,8 +1884,9 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDParallelCall(
Work.emplace_back(WFn); Work.emplace_back(WFn);
}; };
auto &&LNParallelGen = [this, Loc, &SeqGen, &L0ParallelGen, &CodeGen]( auto &&LNParallelGen = [this, Loc, &SeqGen, &L0ParallelGen, &CodeGen,
CodeGenFunction &CGF, PrePostActionTy &Action) { &ThreadIDAddr](CodeGenFunction &CGF,
PrePostActionTy &Action) {
RegionCodeGenTy RCG(CodeGen); RegionCodeGenTy RCG(CodeGen);
if (IsInParallelRegion) { if (IsInParallelRegion) {
SeqGen(CGF, Action); SeqGen(CGF, Action);
@ -1936,6 +1938,8 @@ void CGOpenMPRuntimeNVPTX::emitNonSPMDParallelCall(
// There is no need to emit line number for unconditional branch. // There is no need to emit line number for unconditional branch.
(void)ApplyDebugLocation::CreateEmpty(CGF); (void)ApplyDebugLocation::CreateEmpty(CGF);
CGF.EmitBlock(ElseBlock); CGF.EmitBlock(ElseBlock);
// In the worker need to use the real thread id.
ThreadIDAddr = emitThreadIDAddress(CGF, Loc);
RCG(CGF); RCG(CGF);
// There is no need to emit line number for unconditional branch. // There is no need to emit line number for unconditional branch.
(void)ApplyDebugLocation::CreateEmpty(CGF); (void)ApplyDebugLocation::CreateEmpty(CGF);
@ -1965,10 +1969,11 @@ void CGOpenMPRuntimeNVPTX::emitSPMDParallelCall(
/*DestWidth=*/32, /*Signed=*/1), /*DestWidth=*/32, /*Signed=*/1),
".zero.addr"); ".zero.addr");
CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0)); CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
Address ThreadIDAddr = emitThreadIDAddress(CGF, Loc); // ThreadId for serialized parallels is 0.
Address ThreadIDAddr = ZeroAddr;
auto &&CodeGen = [this, OutlinedFn, CapturedVars, Loc, ZeroAddr, auto &&CodeGen = [this, OutlinedFn, CapturedVars, Loc, ZeroAddr,
ThreadIDAddr](CodeGenFunction &CGF, &ThreadIDAddr](CodeGenFunction &CGF,
PrePostActionTy &Action) { PrePostActionTy &Action) {
Action.Enter(CGF); Action.Enter(CGF);
llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs; llvm::SmallVector<llvm::Value *, 16> OutlinedFnArgs;
@ -1995,6 +2000,8 @@ void CGOpenMPRuntimeNVPTX::emitSPMDParallelCall(
}; };
if (IsInTargetMasterThreadRegion) { if (IsInTargetMasterThreadRegion) {
// In the worker need to use the real thread id.
ThreadIDAddr = emitThreadIDAddress(CGF, Loc);
RegionCodeGenTy RCG(CodeGen); RegionCodeGenTy RCG(CodeGen);
RCG(CGF); RCG(CGF);
} else { } else {

4
clang/test/OpenMP/nvptx_target_codegen.cpp
View File

@ -562,7 +562,6 @@ int baz(int f, double &a) {
// CHECK: [[REC_ADDR:%.+]] = bitcast i8* [[PTR]] to %struct._globalized_locals_ty* // CHECK: [[REC_ADDR:%.+]] = bitcast i8* [[PTR]] to %struct._globalized_locals_ty*
// CHECK: [[F_PTR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[REC_ADDR]], i32 0, i32 0 // CHECK: [[F_PTR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[REC_ADDR]], i32 0, i32 0
// CHECK: store i32 %{{.+}}, i32* [[F_PTR]], // CHECK: store i32 %{{.+}}, i32* [[F_PTR]],
// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR]],
// CHECK: [[RES:%.+]] = call i8 @__kmpc_is_spmd_exec_mode() // CHECK: [[RES:%.+]] = call i8 @__kmpc_is_spmd_exec_mode()
// CHECK: icmp ne i8 [[RES]], 0 // CHECK: icmp ne i8 [[RES]], 0
@ -573,7 +572,7 @@ int baz(int f, double &a) {
// CHECK: br i1 // CHECK: br i1
// CHECK: call void @__kmpc_serialized_parallel(%struct.ident_t* @{{.+}}, i32 [[GTID]]) // CHECK: call void @__kmpc_serialized_parallel(%struct.ident_t* @{{.+}}, i32 [[GTID]])
// CHECK: call void [[OUTLINED:@.+]](i32* [[GTID_ADDR]], i32* [[ZERO_ADDR]], i32* [[F_PTR]], double* %{{.+}}) // CHECK: call void [[OUTLINED:@.+]](i32* [[ZERO_ADDR]], i32* [[ZERO_ADDR]], i32* [[F_PTR]], double* %{{.+}})
// CHECK: call void @__kmpc_end_serialized_parallel(%struct.ident_t* @{{.+}}, i32 [[GTID]]) // CHECK: call void @__kmpc_end_serialized_parallel(%struct.ident_t* @{{.+}}, i32 [[GTID]])
// CHECK: br label // CHECK: br label
@ -591,6 +590,7 @@ int baz(int f, double &a) {
// CHECK: call void @__kmpc_end_sharing_variables() // CHECK: call void @__kmpc_end_sharing_variables()
// CHECK: br label // CHECK: br label
// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR]],
// CHECK: call void [[OUTLINED]](i32* [[GTID_ADDR]], i32* [[ZERO_ADDR]], i32* [[F_PTR]], double* %{{.+}}) // CHECK: call void [[OUTLINED]](i32* [[GTID_ADDR]], i32* [[ZERO_ADDR]], i32* [[F_PTR]], double* %{{.+}})
// CHECK: br label // CHECK: br label

20
clang/test/OpenMP/parallel_if_codegen.cpp
View File

@ -29,12 +29,12 @@ void gtid_test() {
} }
// CHECK: define internal {{.*}}void [[GTID_TEST_REGION1]](i{{.+}}* noalias [[GTID_PARAM:%.+]], i32* noalias // CHECK: define internal {{.*}}void [[GTID_TEST_REGION1]](i{{.+}}* noalias [[GTID_PARAM:%.+]], i32* noalias
// CHECK: store i32 0, i32* [[ZERO_ADDR:%.+]],
// CHECK: store i{{[0-9]+}}* [[GTID_PARAM]], i{{[0-9]+}}** [[GTID_ADDR_REF:%.+]], // CHECK: store i{{[0-9]+}}* [[GTID_PARAM]], i{{[0-9]+}}** [[GTID_ADDR_REF:%.+]],
// CHECK: [[GTID_ADDR:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_REF]] // CHECK: [[GTID_ADDR:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_REF]]
// CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_ADDR]] // CHECK: [[GTID:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[GTID_ADDR]]
// CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]]) // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]])
// CHECK: [[GTID_ADDR:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[GTID_ADDR_REF]] // CHECK: call void [[GTID_TEST_REGION2:@.+]](i{{[0-9]+}}* [[ZERO_ADDR]]
// CHECK: call void [[GTID_TEST_REGION2:@.+]](i{{[0-9]+}}* [[GTID_ADDR]]
// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i{{.+}} [[GTID]])
// CHECK: ret void // CHECK: ret void
@ -56,12 +56,13 @@ int tmain(T Arg) {
// CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main() // CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main()
int main() { int main() {
// CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num( // CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num(
// CHECK: store i32 0, i32* [[ZERO_ADDR2:%.+]],
// CHECK: store i32 0, i32* [[ZERO_ADDR1:%.+]],
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN4:@.+]] to void // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN4:@.+]] to void
#pragma omp parallel if (true) #pragma omp parallel if (true)
fn4(); fn4();
// CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], // CHECK: call void [[CAP_FN5:@.+]](i32* [[ZERO_ADDR1]], i32* [[ZERO_ADDR1]])
// CHECK: call void [[CAP_FN5:@.+]](i32* [[GTID_ADDR]],
// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
#pragma omp parallel if (false) #pragma omp parallel if (false)
fn5(); fn5();
@ -72,8 +73,7 @@ int main() {
// CHECK: br label %[[OMP_END:.+]] // CHECK: br label %[[OMP_END:.+]]
// CHECK: [[OMP_ELSE]] // CHECK: [[OMP_ELSE]]
// CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], // CHECK: call void [[CAP_FN6]](i32* [[ZERO_ADDR2]], i32* [[ZERO_ADDR2]])
// CHECK: call void [[CAP_FN6]](i32* [[GTID_ADDR]],
// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
// CHECK: br label %[[OMP_END]] // CHECK: br label %[[OMP_END]]
// CHECK: [[OMP_END]] // CHECK: [[OMP_END]]
@ -97,10 +97,11 @@ int main() {
// CHECK-LABEL: define {{.+}} @{{.+}}tmain // CHECK-LABEL: define {{.+}} @{{.+}}tmain
// CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num( // CHECK: [[GTID:%.+]] = call {{.*}}i32 @__kmpc_global_thread_num(
// CHECK: store i32 0, i32* [[ZERO_ADDR2:%.+]],
// CHECK: store i32 0, i32* [[ZERO_ADDR1:%.+]],
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN1:@.+]] to void // CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 0, void {{.+}}* [[CAP_FN1:@.+]] to void
// CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], // CHECK: call void [[CAP_FN2:@.+]](i32* [[ZERO_ADDR1]], i32* [[ZERO_ADDR1]])
// CHECK: call void [[CAP_FN2:@.+]](i32* [[GTID_ADDR]],
// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
// CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]] // CHECK: br i1 %{{.+}}, label %[[OMP_THEN:.+]], label %[[OMP_ELSE:.+]]
// CHECK: [[OMP_THEN]] // CHECK: [[OMP_THEN]]
@ -108,8 +109,7 @@ int main() {
// CHECK: br label %[[OMP_END:.+]] // CHECK: br label %[[OMP_END:.+]]
// CHECK: [[OMP_ELSE]] // CHECK: [[OMP_ELSE]]
// CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
// CHECK: store i32 [[GTID]], i32* [[GTID_ADDR:%.+]], // CHECK: call void [[CAP_FN3]](i32* [[ZERO_ADDR2]], i32* [[ZERO_ADDR2]])
// CHECK: call void [[CAP_FN3]](i32* [[GTID_ADDR]],
// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]]) // CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(%{{.+}}* @{{.+}}, i32 [[GTID]])
// CHECK: br label %[[OMP_END]] // CHECK: br label %[[OMP_END]]
// CHECK: [[OMP_END]] // CHECK: [[OMP_END]]