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

[OPENMP, NVPTX] Do not use SPMD mode for target simd and target teams 

distribute simd directives.

Directives `target simd` and `target teams distribute simd` must be
executed in non-SPMD mode.

llvm-svn: 332129
This commit is contained in:
Alexey Bataev 2018-05-11 19:45:14 +00:00
parent 304d0d5560
commit df093e7b45
3 changed files with 43 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
View File

@ -628,9 +628,8 @@ static bool hasNestedSPMDDirective(ASTContext &Ctx,
OpenMPDirectiveKind DKind = NestedDir->getDirectiveKind();
switch (D.getDirectiveKind()) {
case OMPD_target:
if ((isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NestedDir)) ||
isOpenMPSimdDirective(DKind))
if (isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NestedDir))
return true;
if (DKind == OMPD_teams || DKind == OMPD_teams_distribute) {
Body = NestedDir->getInnermostCapturedStmt()->IgnoreContainers();
@ -639,9 +638,8 @@ static bool hasNestedSPMDDirective(ASTContext &Ctx,
ChildStmt = getSingleCompoundChild(Body);
if (const auto *NND = dyn_cast<OMPExecutableDirective>(ChildStmt)) {
DKind = NND->getDirectiveKind();
if ((isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NND)) ||
isOpenMPSimdDirective(DKind))
if (isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NND))
return true;
if (DKind == OMPD_distribute) {
Body = NestedDir->getInnermostCapturedStmt()->IgnoreContainers();
@ -652,18 +650,16 @@ static bool hasNestedSPMDDirective(ASTContext &Ctx,
return false;
if (const auto *NND = dyn_cast<OMPExecutableDirective>(ChildStmt)) {
DKind = NND->getDirectiveKind();
return (isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NND)) ||
isOpenMPSimdDirective(DKind);
return isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NND);
}
}
}
}
return false;
case OMPD_target_teams:
if ((isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NestedDir)) ||
isOpenMPSimdDirective(DKind))
if (isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NestedDir))
return true;
if (DKind == OMPD_distribute) {
Body = NestedDir->getInnermostCapturedStmt()->IgnoreContainers();
@ -672,16 +668,14 @@ static bool hasNestedSPMDDirective(ASTContext &Ctx,
ChildStmt = getSingleCompoundChild(Body);
if (const auto *NND = dyn_cast<OMPExecutableDirective>(ChildStmt)) {
DKind = NND->getDirectiveKind();
return (isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NND)) ||
isOpenMPSimdDirective(DKind);
return isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NND);
}
}
return false;
case OMPD_target_teams_distribute:
return (isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NestedDir)) ||
isOpenMPSimdDirective(DKind);
return isOpenMPParallelDirective(DKind) &&
!hasParallelIfClause(Ctx, *NestedDir);
case OMPD_target_simd:
case OMPD_target_parallel:
case OMPD_target_parallel_for:
@ -755,7 +749,7 @@ static bool supportsSPMDExecutionMode(ASTContext &Ctx,
return !hasParallelIfClause(Ctx, D);
case OMPD_target_simd:
case OMPD_target_teams_distribute_simd:
return true;
return false;
case OMPD_parallel:
case OMPD_for:
case OMPD_parallel_for:

30
clang/test/OpenMP/nvptx_target_simd_codegen.cpp
View File

@ -8,11 +8,11 @@
#ifndef HEADER
#define HEADER
// Check that the execution mode of all 2 target regions on the gpu is set to SPMD Mode.
// CHECK-DAG: {{@__omp_offloading_.+l25}}_exec_mode = weak constant i8 0
// CHECK-DAG: {{@__omp_offloading_.+l30}}_exec_mode = weak constant i8 0
// CHECK-DAG: {{@__omp_offloading_.+l35}}_exec_mode = weak constant i8 0
// CHECK-DAG: {{@__omp_offloading_.+l40}}_exec_mode = weak constant i8 0
// Check that the execution mode of all 2 target regions on the gpu is set to NonSPMD Mode.
// CHECK-DAG: {{@__omp_offloading_.+l25}}_exec_mode = weak constant i8 1
// CHECK-DAG: {{@__omp_offloading_.+l30}}_exec_mode = weak constant i8 1
// CHECK-DAG: {{@__omp_offloading_.+l35}}_exec_mode = weak constant i8 1
// CHECK-DAG: {{@__omp_offloading_.+l40}}_exec_mode = weak constant i8 1
#define N 1000
@ -54,38 +54,34 @@ int bar(int n){
}
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l25}}(
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK-NOT: call void @__kmpc_for_static_init
// CHECK-NOT: call void @__kmpc_for_static_fini
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l30}}(
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK-NOT: call void @__kmpc_for_static_init
// CHECK-NOT: call void @__kmpc_for_static_fini
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l35}}(
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK-NOT: call void @__kmpc_for_static_init
// CHECK-NOT: call void @__kmpc_for_static_fini
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+l40}}(
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK-NOT: call void @__kmpc_for_static_init
// CHECK-NOT: call void @__kmpc_for_static_fini
// CHECK: [[RES:%.+]] = call i32 @__kmpc_nvptx_simd_reduce_nowait(i32 %{{.+}}, i32 1, i{{64|32}} {{8|4}}, i8* %{{.+}}, void (i8*, i16, i16, i16)* @{{.+}}, void (i8*, i32)* @{{.+}})
// CHECK: switch i32 [[RES]]
// CHECK: call void @__kmpc_nvptx_end_reduce_nowait(i32 %{{.+}})
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void

38
clang/test/OpenMP/nvptx_target_teams_distribute_simd_codegen.cpp
View File

@ -8,11 +8,11 @@
#ifndef HEADER
#define HEADER
// Check that the execution mode of all 2 target regions on the gpu is set to SPMD Mode.
// CHECK-DAG: {{@__omp_offloading_.+l30}}_exec_mode = weak constant i8 0
// CHECK-DAG: {{@__omp_offloading_.+l36}}_exec_mode = weak constant i8 0
// CHECK-DAG: {{@__omp_offloading_.+l41}}_exec_mode = weak constant i8 0
// CHECK-DAG: {{@__omp_offloading_.+l46}}_exec_mode = weak constant i8 0
// Check that the execution mode of all 2 target regions on the gpu is set to NonSPMD Mode.
// CHECK-DAG: {{@__omp_offloading_.+l30}}_exec_mode = weak constant i8 1
// CHECK-DAG: {{@__omp_offloading_.+l36}}_exec_mode = weak constant i8 1
// CHECK-DAG: {{@__omp_offloading_.+l41}}_exec_mode = weak constant i8 1
// CHECK-DAG: {{@__omp_offloading_.+l46}}_exec_mode = weak constant i8 1
#define N 1000
#define M 10
@ -62,38 +62,34 @@ int bar(int n){
return a;
}
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l30(
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 91,
// CHECK: call void @__kmpc_for_static_fini(
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l36(
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92,
// CHECK: call void @__kmpc_for_static_fini(
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}(
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK-LABEL: define {{.*}}void {{@__omp_offloading_.+}}_l41(
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92,
// CHECK: call void @__kmpc_for_static_fini(
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void
// CHECK: define {{.*}}void {{@__omp_offloading_.+}}({{.+}}, i{{32|64}} [[F_IN:%.+]])
// CHECK: define {{.*}}void {{@__omp_offloading_.+}}_l46({{.+}}, i{{32|64}} [[F_IN:%.+]])
// CHECK: store {{.+}} [[F_IN]], {{.+}}* {{.+}},
// CHECK-DAG: [[THREAD_LIMIT:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.ntid.x()
// CHECK: call void @__kmpc_spmd_kernel_init(i32 [[THREAD_LIMIT]],
// CHECK: call void @__kmpc_kernel_init(i32 %{{.+}}, i16 1)
// CHECK: store {{.+}} 99, {{.+}}* [[COMB_UB:%.+]], align
// CHECK: call void @__kmpc_for_static_init_4({{.+}}, {{.+}}, {{.+}} 92, {{.+}}, {{.+}}, {{.+}}* [[COMB_UB]],
// CHECK: call void @__kmpc_for_static_fini(
// CHECK: call void @__kmpc_spmd_kernel_deinit()
// CHECK: call void @__kmpc_kernel_deinit(i16 1)
// CHECK: ret void
#endif