llvm-project/clang/lib/Basic/OpenMPKinds.cpp

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

698 lines
26 KiB
C++
Raw Normal View History

//===--- OpenMPKinds.cpp - Token Kinds Support ----------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
/// \file
/// This file implements the OpenMP enum and support functions.
///
//===----------------------------------------------------------------------===//
#include "clang/Basic/OpenMPKinds.h"
#include "clang/Basic/IdentifierTable.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/StringSwitch.h"
#include "llvm/Support/ErrorHandling.h"
#include <cassert>
using namespace clang;
using namespace llvm::omp;
unsigned clang::getOpenMPSimpleClauseType(OpenMPClauseKind Kind, StringRef Str,
unsigned OpenMPVersion) {
switch (Kind) {
case OMPC_default:
return llvm::StringSwitch<unsigned>(Str)
#define OMP_DEFAULT_KIND(Enum, Name) .Case(Name, unsigned(Enum))
#include "llvm/Frontend/OpenMP/OMPKinds.def"
.Default(unsigned(llvm::omp::OMP_DEFAULT_unknown));
case OMPC_proc_bind:
return llvm::StringSwitch<unsigned>(Str)
#define OMP_PROC_BIND_KIND(Enum, Name, Value) .Case(Name, Value)
#include "llvm/Frontend/OpenMP/OMPKinds.def"
.Default(unsigned(llvm::omp::OMP_PROC_BIND_unknown));
case OMPC_schedule:
return llvm::StringSwitch<unsigned>(Str)
#define OPENMP_SCHEDULE_KIND(Name) \
.Case(#Name, static_cast<unsigned>(OMPC_SCHEDULE_##Name))
#define OPENMP_SCHEDULE_MODIFIER(Name) \
.Case(#Name, static_cast<unsigned>(OMPC_SCHEDULE_MODIFIER_##Name))
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_SCHEDULE_unknown);
case OMPC_depend:
return llvm::StringSwitch<OpenMPDependClauseKind>(Str)
#define OPENMP_DEPEND_KIND(Name) .Case(#Name, OMPC_DEPEND_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_DEPEND_unknown);
case OMPC_linear:
return llvm::StringSwitch<OpenMPLinearClauseKind>(Str)
#define OPENMP_LINEAR_KIND(Name) .Case(#Name, OMPC_LINEAR_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_LINEAR_unknown);
case OMPC_map: {
unsigned Type = llvm::StringSwitch<unsigned>(Str)
#define OPENMP_MAP_KIND(Name) \
.Case(#Name, static_cast<unsigned>(OMPC_MAP_##Name))
#define OPENMP_MAP_MODIFIER_KIND(Name) \
.Case(#Name, static_cast<unsigned>(OMPC_MAP_MODIFIER_##Name))
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_MAP_unknown);
if (OpenMPVersion < 51 && Type == OMPC_MAP_MODIFIER_present)
return OMPC_MAP_MODIFIER_unknown;
return Type;
}
case OMPC_to:
case OMPC_from: {
unsigned Type = llvm::StringSwitch<unsigned>(Str)
#define OPENMP_MOTION_MODIFIER_KIND(Name) \
.Case(#Name, static_cast<unsigned>(OMPC_MOTION_MODIFIER_##Name))
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_MOTION_MODIFIER_unknown);
if (OpenMPVersion < 51 && Type == OMPC_MOTION_MODIFIER_present)
return OMPC_MOTION_MODIFIER_unknown;
return Type;
}
case OMPC_dist_schedule:
return llvm::StringSwitch<OpenMPDistScheduleClauseKind>(Str)
#define OPENMP_DIST_SCHEDULE_KIND(Name) .Case(#Name, OMPC_DIST_SCHEDULE_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_DIST_SCHEDULE_unknown);
case OMPC_defaultmap:
return llvm::StringSwitch<unsigned>(Str)
#define OPENMP_DEFAULTMAP_KIND(Name) \
.Case(#Name, static_cast<unsigned>(OMPC_DEFAULTMAP_##Name))
#define OPENMP_DEFAULTMAP_MODIFIER(Name) \
.Case(#Name, static_cast<unsigned>(OMPC_DEFAULTMAP_MODIFIER_##Name))
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_DEFAULTMAP_unknown);
case OMPC_atomic_default_mem_order:
return llvm::StringSwitch<OpenMPAtomicDefaultMemOrderClauseKind>(Str)
#define OPENMP_ATOMIC_DEFAULT_MEM_ORDER_KIND(Name) \
.Case(#Name, OMPC_ATOMIC_DEFAULT_MEM_ORDER_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_ATOMIC_DEFAULT_MEM_ORDER_unknown);
case OMPC_device_type:
return llvm::StringSwitch<OpenMPDeviceType>(Str)
#define OPENMP_DEVICE_TYPE_KIND(Name) .Case(#Name, OMPC_DEVICE_TYPE_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_DEVICE_TYPE_unknown);
case OMPC_lastprivate:
return llvm::StringSwitch<OpenMPLastprivateModifier>(Str)
#define OPENMP_LASTPRIVATE_KIND(Name) .Case(#Name, OMPC_LASTPRIVATE_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_LASTPRIVATE_unknown);
case OMPC_order:
return llvm::StringSwitch<OpenMPOrderClauseKind>(Str)
#define OPENMP_ORDER_KIND(Name) .Case(#Name, OMPC_ORDER_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_ORDER_unknown);
case OMPC_update:
return llvm::StringSwitch<OpenMPDependClauseKind>(Str)
#define OPENMP_DEPEND_KIND(Name) .Case(#Name, OMPC_DEPEND_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_DEPEND_unknown);
case OMPC_device:
return llvm::StringSwitch<OpenMPDeviceClauseModifier>(Str)
#define OPENMP_DEVICE_MODIFIER(Name) .Case(#Name, OMPC_DEVICE_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_DEVICE_unknown);
case OMPC_reduction:
return llvm::StringSwitch<OpenMPReductionClauseModifier>(Str)
#define OPENMP_REDUCTION_MODIFIER(Name) .Case(#Name, OMPC_REDUCTION_##Name)
#include "clang/Basic/OpenMPKinds.def"
.Default(OMPC_REDUCTION_unknown);
case OMPC_unknown:
case OMPC_threadprivate:
case OMPC_if:
case OMPC_final:
case OMPC_num_threads:
case OMPC_safelen:
case OMPC_simdlen:
[OpenMP] Implement '#pragma omp tile', by Michael Kruse (@Meinersbur). The tile directive is in OpenMP's Technical Report 8 and foreseeably will be part of the upcoming OpenMP 5.1 standard. This implementation is based on an AST transformation providing a de-sugared loop nest. This makes it simple to forward the de-sugared transformation to loop associated directives taking the tiled loops. In contrast to other loop associated directives, the OMPTileDirective does not use CapturedStmts. Letting loop associated directives consume loops from different capture context would be difficult. A significant amount of code generation logic is taking place in the Sema class. Eventually, I would prefer if these would move into the CodeGen component such that we could make use of the OpenMPIRBuilder, together with flang. Only expressions converting between the language's iteration variable and the logical iteration space need to take place in the semantic analyzer: Getting the of iterations (e.g. the overload resolution of `std::distance`) and converting the logical iteration number to the iteration variable (e.g. overload resolution of `iteration + .omp.iv`). In clang, only CXXForRangeStmt is also represented by its de-sugared components. However, OpenMP loop are not defined as syntatic sugar. Starting with an AST-based approach allows us to gradually move generated AST statements into CodeGen, instead all at once. I would also like to refactor `checkOpenMPLoop` into its functionalities in a follow-up. In this patch it is used twice. Once for checking proper nesting and emitting diagnostics, and additionally for deriving the logical iteration space per-loop (instead of for the loop nest). Differential Revision: https://reviews.llvm.org/D76342
2021-02-13 03:26:59 +08:00
case OMPC_sizes:
case OMPC_allocator:
case OMPC_allocate:
case OMPC_collapse:
case OMPC_private:
case OMPC_firstprivate:
case OMPC_shared:
case OMPC_task_reduction:
case OMPC_in_reduction:
case OMPC_aligned:
case OMPC_copyin:
case OMPC_copyprivate:
case OMPC_ordered:
case OMPC_nowait:
case OMPC_untied:
case OMPC_mergeable:
case OMPC_flush:
case OMPC_depobj:
case OMPC_read:
case OMPC_write:
case OMPC_capture:
case OMPC_seq_cst:
case OMPC_acq_rel:
case OMPC_acquire:
case OMPC_release:
case OMPC_relaxed:
case OMPC_threads:
case OMPC_simd:
case OMPC_num_teams:
case OMPC_thread_limit:
case OMPC_priority:
case OMPC_grainsize:
case OMPC_nogroup:
case OMPC_num_tasks:
case OMPC_hint:
case OMPC_uniform:
case OMPC_use_device_ptr:
case OMPC_use_device_addr:
case OMPC_is_device_ptr:
case OMPC_unified_address:
case OMPC_unified_shared_memory:
case OMPC_reverse_offload:
case OMPC_dynamic_allocators:
case OMPC_match:
case OMPC_nontemporal:
case OMPC_destroy:
case OMPC_novariants:
case OMPC_nocontext:
case OMPC_detach:
case OMPC_inclusive:
case OMPC_exclusive:
case OMPC_uses_allocators:
case OMPC_affinity:
break;
default:
break;
}
llvm_unreachable("Invalid OpenMP simple clause kind");
}
const char *clang::getOpenMPSimpleClauseTypeName(OpenMPClauseKind Kind,
unsigned Type) {
switch (Kind) {
case OMPC_default:
switch (llvm::omp::DefaultKind(Type)) {
#define OMP_DEFAULT_KIND(Enum, Name) \
case Enum: \
return Name;
#include "llvm/Frontend/OpenMP/OMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'default' clause type");
case OMPC_proc_bind:
switch (Type) {
#define OMP_PROC_BIND_KIND(Enum, Name, Value) \
case Value: \
return Name;
#include "llvm/Frontend/OpenMP/OMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'proc_bind' clause type");
case OMPC_schedule:
switch (Type) {
case OMPC_SCHEDULE_unknown:
case OMPC_SCHEDULE_MODIFIER_last:
return "unknown";
#define OPENMP_SCHEDULE_KIND(Name) \
case OMPC_SCHEDULE_##Name: \
return #Name;
#define OPENMP_SCHEDULE_MODIFIER(Name) \
case OMPC_SCHEDULE_MODIFIER_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'schedule' clause type");
case OMPC_depend:
switch (Type) {
case OMPC_DEPEND_unknown:
return "unknown";
#define OPENMP_DEPEND_KIND(Name) \
case OMPC_DEPEND_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'depend' clause type");
case OMPC_linear:
switch (Type) {
case OMPC_LINEAR_unknown:
return "unknown";
#define OPENMP_LINEAR_KIND(Name) \
case OMPC_LINEAR_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'linear' clause type");
case OMPC_map:
switch (Type) {
case OMPC_MAP_unknown:
case OMPC_MAP_MODIFIER_last:
return "unknown";
#define OPENMP_MAP_KIND(Name) \
case OMPC_MAP_##Name: \
return #Name;
#define OPENMP_MAP_MODIFIER_KIND(Name) \
case OMPC_MAP_MODIFIER_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
default:
break;
}
llvm_unreachable("Invalid OpenMP 'map' clause type");
case OMPC_to:
case OMPC_from:
switch (Type) {
case OMPC_MOTION_MODIFIER_unknown:
return "unknown";
#define OPENMP_MOTION_MODIFIER_KIND(Name) \
case OMPC_MOTION_MODIFIER_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
default:
break;
}
llvm_unreachable("Invalid OpenMP 'to' or 'from' clause type");
case OMPC_dist_schedule:
switch (Type) {
case OMPC_DIST_SCHEDULE_unknown:
return "unknown";
#define OPENMP_DIST_SCHEDULE_KIND(Name) \
case OMPC_DIST_SCHEDULE_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'dist_schedule' clause type");
case OMPC_defaultmap:
switch (Type) {
case OMPC_DEFAULTMAP_unknown:
case OMPC_DEFAULTMAP_MODIFIER_last:
return "unknown";
#define OPENMP_DEFAULTMAP_KIND(Name) \
case OMPC_DEFAULTMAP_##Name: \
return #Name;
#define OPENMP_DEFAULTMAP_MODIFIER(Name) \
case OMPC_DEFAULTMAP_MODIFIER_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'schedule' clause type");
case OMPC_atomic_default_mem_order:
switch (Type) {
case OMPC_ATOMIC_DEFAULT_MEM_ORDER_unknown:
return "unknown";
#define OPENMP_ATOMIC_DEFAULT_MEM_ORDER_KIND(Name) \
case OMPC_ATOMIC_DEFAULT_MEM_ORDER_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'atomic_default_mem_order' clause type");
case OMPC_device_type:
switch (Type) {
case OMPC_DEVICE_TYPE_unknown:
return "unknown";
#define OPENMP_DEVICE_TYPE_KIND(Name) \
case OMPC_DEVICE_TYPE_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'device_type' clause type");
case OMPC_lastprivate:
switch (Type) {
case OMPC_LASTPRIVATE_unknown:
return "unknown";
#define OPENMP_LASTPRIVATE_KIND(Name) \
case OMPC_LASTPRIVATE_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'lastprivate' clause type");
case OMPC_order:
switch (Type) {
case OMPC_ORDER_unknown:
return "unknown";
#define OPENMP_ORDER_KIND(Name) \
case OMPC_ORDER_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'order' clause type");
case OMPC_update:
switch (Type) {
case OMPC_DEPEND_unknown:
return "unknown";
#define OPENMP_DEPEND_KIND(Name) \
case OMPC_DEPEND_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'depend' clause type");
case OMPC_device:
switch (Type) {
case OMPC_DEVICE_unknown:
return "unknown";
#define OPENMP_DEVICE_MODIFIER(Name) \
case OMPC_DEVICE_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'device' clause modifier");
case OMPC_reduction:
switch (Type) {
case OMPC_REDUCTION_unknown:
return "unknown";
#define OPENMP_REDUCTION_MODIFIER(Name) \
case OMPC_REDUCTION_##Name: \
return #Name;
#include "clang/Basic/OpenMPKinds.def"
}
llvm_unreachable("Invalid OpenMP 'reduction' clause modifier");
case OMPC_unknown:
case OMPC_threadprivate:
case OMPC_if:
case OMPC_final:
case OMPC_num_threads:
case OMPC_safelen:
case OMPC_simdlen:
[OpenMP] Implement '#pragma omp tile', by Michael Kruse (@Meinersbur). The tile directive is in OpenMP's Technical Report 8 and foreseeably will be part of the upcoming OpenMP 5.1 standard. This implementation is based on an AST transformation providing a de-sugared loop nest. This makes it simple to forward the de-sugared transformation to loop associated directives taking the tiled loops. In contrast to other loop associated directives, the OMPTileDirective does not use CapturedStmts. Letting loop associated directives consume loops from different capture context would be difficult. A significant amount of code generation logic is taking place in the Sema class. Eventually, I would prefer if these would move into the CodeGen component such that we could make use of the OpenMPIRBuilder, together with flang. Only expressions converting between the language's iteration variable and the logical iteration space need to take place in the semantic analyzer: Getting the of iterations (e.g. the overload resolution of `std::distance`) and converting the logical iteration number to the iteration variable (e.g. overload resolution of `iteration + .omp.iv`). In clang, only CXXForRangeStmt is also represented by its de-sugared components. However, OpenMP loop are not defined as syntatic sugar. Starting with an AST-based approach allows us to gradually move generated AST statements into CodeGen, instead all at once. I would also like to refactor `checkOpenMPLoop` into its functionalities in a follow-up. In this patch it is used twice. Once for checking proper nesting and emitting diagnostics, and additionally for deriving the logical iteration space per-loop (instead of for the loop nest). Differential Revision: https://reviews.llvm.org/D76342
2021-02-13 03:26:59 +08:00
case OMPC_sizes:
case OMPC_allocator:
case OMPC_allocate:
case OMPC_collapse:
case OMPC_private:
case OMPC_firstprivate:
case OMPC_shared:
case OMPC_task_reduction:
case OMPC_in_reduction:
case OMPC_aligned:
case OMPC_copyin:
case OMPC_copyprivate:
case OMPC_ordered:
case OMPC_nowait:
case OMPC_untied:
case OMPC_mergeable:
case OMPC_flush:
case OMPC_depobj:
case OMPC_read:
case OMPC_write:
case OMPC_capture:
case OMPC_seq_cst:
case OMPC_acq_rel:
case OMPC_acquire:
case OMPC_release:
case OMPC_relaxed:
case OMPC_threads:
case OMPC_simd:
case OMPC_num_teams:
case OMPC_thread_limit:
case OMPC_priority:
case OMPC_grainsize:
case OMPC_nogroup:
case OMPC_num_tasks:
case OMPC_hint:
case OMPC_uniform:
case OMPC_use_device_ptr:
case OMPC_use_device_addr:
case OMPC_is_device_ptr:
case OMPC_unified_address:
case OMPC_unified_shared_memory:
case OMPC_reverse_offload:
case OMPC_dynamic_allocators:
case OMPC_match:
case OMPC_nontemporal:
case OMPC_destroy:
case OMPC_detach:
case OMPC_novariants:
case OMPC_nocontext:
case OMPC_inclusive:
case OMPC_exclusive:
case OMPC_uses_allocators:
case OMPC_affinity:
break;
default:
break;
}
llvm_unreachable("Invalid OpenMP simple clause kind");
}
bool clang::isOpenMPLoopDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_simd || DKind == OMPD_for || DKind == OMPD_for_simd ||
DKind == OMPD_parallel_for || DKind == OMPD_parallel_for_simd ||
DKind == OMPD_taskloop || DKind == OMPD_taskloop_simd ||
DKind == OMPD_master_taskloop || DKind == OMPD_master_taskloop_simd ||
DKind == OMPD_parallel_master_taskloop ||
DKind == OMPD_parallel_master_taskloop_simd ||
DKind == OMPD_distribute || DKind == OMPD_target_parallel_for ||
DKind == OMPD_distribute_parallel_for ||
DKind == OMPD_distribute_parallel_for_simd ||
DKind == OMPD_distribute_simd ||
DKind == OMPD_target_parallel_for_simd || DKind == OMPD_target_simd ||
DKind == OMPD_teams_distribute ||
DKind == OMPD_teams_distribute_simd ||
DKind == OMPD_teams_distribute_parallel_for_simd ||
DKind == OMPD_teams_distribute_parallel_for ||
DKind == OMPD_target_teams_distribute ||
DKind == OMPD_target_teams_distribute_parallel_for ||
DKind == OMPD_target_teams_distribute_parallel_for_simd ||
[OpenMP] Implement '#pragma omp tile', by Michael Kruse (@Meinersbur). The tile directive is in OpenMP's Technical Report 8 and foreseeably will be part of the upcoming OpenMP 5.1 standard. This implementation is based on an AST transformation providing a de-sugared loop nest. This makes it simple to forward the de-sugared transformation to loop associated directives taking the tiled loops. In contrast to other loop associated directives, the OMPTileDirective does not use CapturedStmts. Letting loop associated directives consume loops from different capture context would be difficult. A significant amount of code generation logic is taking place in the Sema class. Eventually, I would prefer if these would move into the CodeGen component such that we could make use of the OpenMPIRBuilder, together with flang. Only expressions converting between the language's iteration variable and the logical iteration space need to take place in the semantic analyzer: Getting the of iterations (e.g. the overload resolution of `std::distance`) and converting the logical iteration number to the iteration variable (e.g. overload resolution of `iteration + .omp.iv`). In clang, only CXXForRangeStmt is also represented by its de-sugared components. However, OpenMP loop are not defined as syntatic sugar. Starting with an AST-based approach allows us to gradually move generated AST statements into CodeGen, instead all at once. I would also like to refactor `checkOpenMPLoop` into its functionalities in a follow-up. In this patch it is used twice. Once for checking proper nesting and emitting diagnostics, and additionally for deriving the logical iteration space per-loop (instead of for the loop nest). Differential Revision: https://reviews.llvm.org/D76342
2021-02-13 03:26:59 +08:00
DKind == OMPD_target_teams_distribute_simd || DKind == OMPD_tile;
}
bool clang::isOpenMPWorksharingDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_for || DKind == OMPD_for_simd ||
DKind == OMPD_sections || DKind == OMPD_section ||
DKind == OMPD_single || DKind == OMPD_parallel_for ||
DKind == OMPD_parallel_for_simd || DKind == OMPD_parallel_sections ||
DKind == OMPD_target_parallel_for ||
DKind == OMPD_distribute_parallel_for ||
DKind == OMPD_distribute_parallel_for_simd ||
DKind == OMPD_target_parallel_for_simd ||
DKind == OMPD_teams_distribute_parallel_for_simd ||
DKind == OMPD_teams_distribute_parallel_for ||
DKind == OMPD_target_teams_distribute_parallel_for ||
DKind == OMPD_target_teams_distribute_parallel_for_simd;
}
bool clang::isOpenMPTaskLoopDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_taskloop || DKind == OMPD_taskloop_simd ||
DKind == OMPD_master_taskloop || DKind == OMPD_master_taskloop_simd ||
DKind == OMPD_parallel_master_taskloop ||
DKind == OMPD_parallel_master_taskloop_simd;
}
bool clang::isOpenMPParallelDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_parallel || DKind == OMPD_parallel_for ||
DKind == OMPD_parallel_for_simd || DKind == OMPD_parallel_sections ||
DKind == OMPD_target_parallel || DKind == OMPD_target_parallel_for ||
DKind == OMPD_distribute_parallel_for ||
DKind == OMPD_distribute_parallel_for_simd ||
DKind == OMPD_target_parallel_for_simd ||
DKind == OMPD_teams_distribute_parallel_for ||
DKind == OMPD_teams_distribute_parallel_for_simd ||
DKind == OMPD_target_teams_distribute_parallel_for ||
DKind == OMPD_target_teams_distribute_parallel_for_simd ||
DKind == OMPD_parallel_master ||
DKind == OMPD_parallel_master_taskloop ||
DKind == OMPD_parallel_master_taskloop_simd;
}
bool clang::isOpenMPTargetExecutionDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_target || DKind == OMPD_target_parallel ||
DKind == OMPD_target_parallel_for ||
DKind == OMPD_target_parallel_for_simd || DKind == OMPD_target_simd ||
DKind == OMPD_target_teams || DKind == OMPD_target_teams_distribute ||
DKind == OMPD_target_teams_distribute_parallel_for ||
DKind == OMPD_target_teams_distribute_parallel_for_simd ||
DKind == OMPD_target_teams_distribute_simd;
}
bool clang::isOpenMPTargetDataManagementDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_target_data || DKind == OMPD_target_enter_data ||
DKind == OMPD_target_exit_data || DKind == OMPD_target_update;
}
bool clang::isOpenMPNestingTeamsDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_teams || DKind == OMPD_teams_distribute ||
DKind == OMPD_teams_distribute_simd ||
DKind == OMPD_teams_distribute_parallel_for_simd ||
DKind == OMPD_teams_distribute_parallel_for;
}
bool clang::isOpenMPTeamsDirective(OpenMPDirectiveKind DKind) {
return isOpenMPNestingTeamsDirective(DKind) ||
DKind == OMPD_target_teams || DKind == OMPD_target_teams_distribute ||
DKind == OMPD_target_teams_distribute_parallel_for ||
DKind == OMPD_target_teams_distribute_parallel_for_simd ||
DKind == OMPD_target_teams_distribute_simd;
}
bool clang::isOpenMPSimdDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_simd || DKind == OMPD_for_simd ||
DKind == OMPD_parallel_for_simd || DKind == OMPD_taskloop_simd ||
DKind == OMPD_master_taskloop_simd ||
DKind == OMPD_parallel_master_taskloop_simd ||
DKind == OMPD_distribute_parallel_for_simd ||
DKind == OMPD_distribute_simd || DKind == OMPD_target_simd ||
DKind == OMPD_teams_distribute_simd ||
DKind == OMPD_teams_distribute_parallel_for_simd ||
DKind == OMPD_target_teams_distribute_parallel_for_simd ||
DKind == OMPD_target_teams_distribute_simd ||
DKind == OMPD_target_parallel_for_simd;
}
bool clang::isOpenMPNestingDistributeDirective(OpenMPDirectiveKind Kind) {
return Kind == OMPD_distribute || Kind == OMPD_distribute_parallel_for ||
Kind == OMPD_distribute_parallel_for_simd ||
Kind == OMPD_distribute_simd;
// TODO add next directives.
}
bool clang::isOpenMPDistributeDirective(OpenMPDirectiveKind Kind) {
return isOpenMPNestingDistributeDirective(Kind) ||
Kind == OMPD_teams_distribute || Kind == OMPD_teams_distribute_simd ||
Kind == OMPD_teams_distribute_parallel_for_simd ||
Kind == OMPD_teams_distribute_parallel_for ||
Kind == OMPD_target_teams_distribute ||
Kind == OMPD_target_teams_distribute_parallel_for ||
Kind == OMPD_target_teams_distribute_parallel_for_simd ||
Kind == OMPD_target_teams_distribute_simd;
}
bool clang::isOpenMPPrivate(OpenMPClauseKind Kind) {
return Kind == OMPC_private || Kind == OMPC_firstprivate ||
Kind == OMPC_lastprivate || Kind == OMPC_linear ||
Kind == OMPC_reduction || Kind == OMPC_task_reduction ||
Kind == OMPC_in_reduction; // TODO add next clauses like 'reduction'.
}
bool clang::isOpenMPThreadPrivate(OpenMPClauseKind Kind) {
return Kind == OMPC_threadprivate || Kind == OMPC_copyin;
}
bool clang::isOpenMPTaskingDirective(OpenMPDirectiveKind Kind) {
return Kind == OMPD_task || isOpenMPTaskLoopDirective(Kind);
}
bool clang::isOpenMPLoopBoundSharingDirective(OpenMPDirectiveKind Kind) {
return Kind == OMPD_distribute_parallel_for ||
Kind == OMPD_distribute_parallel_for_simd ||
Kind == OMPD_teams_distribute_parallel_for_simd ||
Kind == OMPD_teams_distribute_parallel_for ||
Kind == OMPD_target_teams_distribute_parallel_for ||
Kind == OMPD_target_teams_distribute_parallel_for_simd;
}
[OpenMP] Implement '#pragma omp tile', by Michael Kruse (@Meinersbur). The tile directive is in OpenMP's Technical Report 8 and foreseeably will be part of the upcoming OpenMP 5.1 standard. This implementation is based on an AST transformation providing a de-sugared loop nest. This makes it simple to forward the de-sugared transformation to loop associated directives taking the tiled loops. In contrast to other loop associated directives, the OMPTileDirective does not use CapturedStmts. Letting loop associated directives consume loops from different capture context would be difficult. A significant amount of code generation logic is taking place in the Sema class. Eventually, I would prefer if these would move into the CodeGen component such that we could make use of the OpenMPIRBuilder, together with flang. Only expressions converting between the language's iteration variable and the logical iteration space need to take place in the semantic analyzer: Getting the of iterations (e.g. the overload resolution of `std::distance`) and converting the logical iteration number to the iteration variable (e.g. overload resolution of `iteration + .omp.iv`). In clang, only CXXForRangeStmt is also represented by its de-sugared components. However, OpenMP loop are not defined as syntatic sugar. Starting with an AST-based approach allows us to gradually move generated AST statements into CodeGen, instead all at once. I would also like to refactor `checkOpenMPLoop` into its functionalities in a follow-up. In this patch it is used twice. Once for checking proper nesting and emitting diagnostics, and additionally for deriving the logical iteration space per-loop (instead of for the loop nest). Differential Revision: https://reviews.llvm.org/D76342
2021-02-13 03:26:59 +08:00
bool clang::isOpenMPLoopTransformationDirective(OpenMPDirectiveKind DKind) {
return DKind == OMPD_tile;
}
void clang::getOpenMPCaptureRegions(
SmallVectorImpl<OpenMPDirectiveKind> &CaptureRegions,
OpenMPDirectiveKind DKind) {
assert(unsigned(DKind) < llvm::omp::Directive_enumSize);
switch (DKind) {
case OMPD_parallel:
case OMPD_parallel_for:
case OMPD_parallel_for_simd:
case OMPD_parallel_master:
case OMPD_parallel_sections:
case OMPD_distribute_parallel_for:
case OMPD_distribute_parallel_for_simd:
CaptureRegions.push_back(OMPD_parallel);
break;
case OMPD_target_teams:
case OMPD_target_teams_distribute:
case OMPD_target_teams_distribute_simd:
CaptureRegions.push_back(OMPD_task);
CaptureRegions.push_back(OMPD_target);
CaptureRegions.push_back(OMPD_teams);
break;
case OMPD_teams:
case OMPD_teams_distribute:
case OMPD_teams_distribute_simd:
CaptureRegions.push_back(OMPD_teams);
break;
case OMPD_target:
case OMPD_target_simd:
CaptureRegions.push_back(OMPD_task);
CaptureRegions.push_back(OMPD_target);
break;
case OMPD_teams_distribute_parallel_for:
case OMPD_teams_distribute_parallel_for_simd:
CaptureRegions.push_back(OMPD_teams);
CaptureRegions.push_back(OMPD_parallel);
break;
case OMPD_target_parallel:
case OMPD_target_parallel_for:
case OMPD_target_parallel_for_simd:
CaptureRegions.push_back(OMPD_task);
CaptureRegions.push_back(OMPD_target);
CaptureRegions.push_back(OMPD_parallel);
break;
case OMPD_task:
case OMPD_target_enter_data:
case OMPD_target_exit_data:
case OMPD_target_update:
CaptureRegions.push_back(OMPD_task);
break;
case OMPD_taskloop:
case OMPD_taskloop_simd:
case OMPD_master_taskloop:
case OMPD_master_taskloop_simd:
CaptureRegions.push_back(OMPD_taskloop);
break;
case OMPD_parallel_master_taskloop:
case OMPD_parallel_master_taskloop_simd:
CaptureRegions.push_back(OMPD_parallel);
CaptureRegions.push_back(OMPD_taskloop);
break;
case OMPD_target_teams_distribute_parallel_for:
case OMPD_target_teams_distribute_parallel_for_simd:
CaptureRegions.push_back(OMPD_task);
CaptureRegions.push_back(OMPD_target);
CaptureRegions.push_back(OMPD_teams);
CaptureRegions.push_back(OMPD_parallel);
break;
case OMPD_simd:
case OMPD_for:
case OMPD_for_simd:
case OMPD_sections:
case OMPD_section:
case OMPD_single:
case OMPD_master:
case OMPD_critical:
case OMPD_taskgroup:
case OMPD_distribute:
case OMPD_ordered:
case OMPD_atomic:
case OMPD_target_data:
case OMPD_distribute_simd:
case OMPD_dispatch:
CaptureRegions.push_back(OMPD_unknown);
break;
[OpenMP] Implement '#pragma omp tile', by Michael Kruse (@Meinersbur). The tile directive is in OpenMP's Technical Report 8 and foreseeably will be part of the upcoming OpenMP 5.1 standard. This implementation is based on an AST transformation providing a de-sugared loop nest. This makes it simple to forward the de-sugared transformation to loop associated directives taking the tiled loops. In contrast to other loop associated directives, the OMPTileDirective does not use CapturedStmts. Letting loop associated directives consume loops from different capture context would be difficult. A significant amount of code generation logic is taking place in the Sema class. Eventually, I would prefer if these would move into the CodeGen component such that we could make use of the OpenMPIRBuilder, together with flang. Only expressions converting between the language's iteration variable and the logical iteration space need to take place in the semantic analyzer: Getting the of iterations (e.g. the overload resolution of `std::distance`) and converting the logical iteration number to the iteration variable (e.g. overload resolution of `iteration + .omp.iv`). In clang, only CXXForRangeStmt is also represented by its de-sugared components. However, OpenMP loop are not defined as syntatic sugar. Starting with an AST-based approach allows us to gradually move generated AST statements into CodeGen, instead all at once. I would also like to refactor `checkOpenMPLoop` into its functionalities in a follow-up. In this patch it is used twice. Once for checking proper nesting and emitting diagnostics, and additionally for deriving the logical iteration space per-loop (instead of for the loop nest). Differential Revision: https://reviews.llvm.org/D76342
2021-02-13 03:26:59 +08:00
case OMPD_tile:
// loop transformations do not introduce captures.
break;
case OMPD_threadprivate:
case OMPD_allocate:
case OMPD_taskyield:
case OMPD_barrier:
case OMPD_taskwait:
case OMPD_cancellation_point:
case OMPD_cancel:
case OMPD_flush:
case OMPD_depobj:
case OMPD_scan:
case OMPD_declare_reduction:
case OMPD_declare_mapper:
case OMPD_declare_simd:
case OMPD_declare_target:
case OMPD_end_declare_target:
case OMPD_requires:
case OMPD_declare_variant:
case OMPD_begin_declare_variant:
case OMPD_end_declare_variant:
llvm_unreachable("OpenMP Directive is not allowed");
case OMPD_unknown:
default:
llvm_unreachable("Unknown OpenMP directive");
}
}