gtbercea updated this revision to Diff 195200.
gtbercea added a comment.
- Remove const.
Repository:
rC Clang
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D60568/new/
https://reviews.llvm.org/D60568
Files:
lib/CodeGen/CGOpenMPRuntime.cpp
lib/CodeGen/CGOpenMPRuntime.h
lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
lib/CodeGen/CGOpenMPRuntimeNVPTX.h
lib/CodeGen/CodeGenModule.cpp
test/OpenMP/openmp_offload_registration.cpp
Index: test/OpenMP/openmp_offload_registration.cpp
===================================================================
--- test/OpenMP/openmp_offload_registration.cpp
+++ test/OpenMP/openmp_offload_registration.cpp
@@ -26,7 +26,7 @@
// CHECK: [[DESC:@.+]] = internal constant [[DSCTY]] { i32 2, [[DEVTY]]* getelementptr inbounds ([2 x [[DEVTY]]], [2 x [[DEVTY]]]* [[IMAGES]], i32 0, i32 0), [[ENTTY]]* [[ENTBEGIN]], [[ENTTY]]* [[ENTEND]] }, comdat($[[REGFN]])
// Check target registration is registered as a Ctor.
-// CHECK: appending global [1 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }]
+// CHECK: appending global [2 x { i32, void ()*, i8* }] [{ i32, void ()*, i8* } { i32 0, void ()* @.omp_offloading.requires_reg, i8* null }, { i32, void ()*, i8* } { i32 0, void ()* @[[REGFN]], i8* bitcast (void ()* @[[REGFN]] to i8*) }]
// Check presence of foo() and the outlined target region
// CHECK: define void [[FOO:@.+]]()
@@ -34,6 +34,11 @@
// Check registration and unregistration code.
+// CHECK: define internal void @.omp_offloading.requires_reg()
+// CHECK: call void @__tgt_register_requires(i64 0)
+// CHECK: ret void
+// CHECK: declare void @__tgt_register_requires(i64)
+
// CHECK: define internal void @[[UNREGFN:.+]](i8*)
// CHECK-SAME: comdat($[[REGFN]]) {
// CHECK: call i32 @__tgt_unregister_lib([[DSCTY]]* [[DESC]])
Index: lib/CodeGen/CodeGenModule.cpp
===================================================================
--- lib/CodeGen/CodeGenModule.cpp
+++ lib/CodeGen/CodeGenModule.cpp
@@ -410,6 +410,10 @@
AddGlobalCtor(CudaCtorFunction);
}
if (OpenMPRuntime) {
+ if (llvm::Function *OpenMPRequiresDirectiveRegFun =
+ OpenMPRuntime->emitRequiresDirectiveRegFun()) {
+ AddGlobalCtor(OpenMPRequiresDirectiveRegFun, 0);
+ }
if (llvm::Function *OpenMPRegistrationFunction =
OpenMPRuntime->emitRegistrationFunction()) {
auto ComdatKey = OpenMPRegistrationFunction->hasComdat() ?
Index: lib/CodeGen/CGOpenMPRuntimeNVPTX.h
===================================================================
--- lib/CodeGen/CGOpenMPRuntimeNVPTX.h
+++ lib/CodeGen/CGOpenMPRuntimeNVPTX.h
@@ -383,7 +383,7 @@
/// Perform check on requires decl to ensure that target architecture
/// supports unified addressing
- void checkArchForUnifiedAddressing(const OMPRequiresDecl *D) const override;
+ void checkArchForUnifiedAddressing(const OMPRequiresDecl *D) override;
/// Returns default address space for the constant firstprivates, __constant__
/// address space by default.
Index: lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
===================================================================
--- lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@@ -4942,7 +4942,7 @@
/// Check to see if target architecture supports unified addressing which is
/// a restriction for OpenMP requires clause "unified_shared_memory".
void CGOpenMPRuntimeNVPTX::checkArchForUnifiedAddressing(
- const OMPRequiresDecl *D) const {
+ const OMPRequiresDecl *D) {
for (const OMPClause *Clause : D->clauselists()) {
if (Clause->getClauseKind() == OMPC_unified_shared_memory) {
switch (getCudaArch(CGM)) {
@@ -4987,6 +4987,7 @@
}
}
}
+ CGOpenMPRuntime::checkArchForUnifiedAddressing(D);
}
/// Get number of SMs and number of blocks per SM.
Index: lib/CodeGen/CGOpenMPRuntime.h
===================================================================
--- lib/CodeGen/CGOpenMPRuntime.h
+++ lib/CodeGen/CGOpenMPRuntime.h
@@ -636,6 +636,10 @@
/// must be emitted.
llvm::SmallDenseSet<const VarDecl *> DeferredGlobalVariables;
+ /// Flag for keeping track of weather a requires unified_shared_memory
+ /// directive is present.
+ bool HasRequiresUnifiedSharedMemory = false;
+
/// Creates and registers offloading binary descriptor for the current
/// compilation unit. The function that does the registration is returned.
llvm::Function *createOffloadingBinaryDescriptorRegistration();
@@ -1429,6 +1433,10 @@
/// \param GD Global to scan.
virtual bool emitTargetGlobal(GlobalDecl GD);
+ /// Creates and returns a registration function for when at least one
+ /// requires directives was used in the current module.
+ llvm::Function *emitRequiresDirectiveRegFun();
+
/// Creates the offloading descriptor in the event any target region
/// was emitted in the current module and return the function that registers
/// it.
@@ -1597,7 +1605,7 @@
/// Perform check on requires decl to ensure that target architecture
/// supports unified addressing
- virtual void checkArchForUnifiedAddressing(const OMPRequiresDecl *D) const {}
+ virtual void checkArchForUnifiedAddressing(const OMPRequiresDecl *D);
/// Checks if the variable has associated OMPAllocateDeclAttr attribute with
/// the predefined allocator and translates it into the corresponding address
Index: lib/CodeGen/CGOpenMPRuntime.cpp
===================================================================
--- lib/CodeGen/CGOpenMPRuntime.cpp
+++ lib/CodeGen/CGOpenMPRuntime.cpp
@@ -457,6 +457,24 @@
LLVM_MARK_AS_BITMASK_ENUM(/*LargestValue=*/OMP_IDENT_WORK_DISTRIBUTE)
};
+namespace {
+LLVM_ENABLE_BITMASK_ENUMS_IN_NAMESPACE();
+/// Values for bit flags for marking which requires clauses have been used.
+enum OpenMPOffloadingRequiresDirFlags : int64_t {
+ /// no requires directive present.
+ OMP_REQ_NONE = 0x000,
+ /// reverse_offload clause.
+ OMP_REQ_REVERSE_OFFLOAD = 0x001,
+ /// unified_address clause.
+ OMP_REQ_UNIFIED_ADDRESS = 0x002,
+ /// unified_shared_memory clause.
+ OMP_REQ_UNIFIED_SHARED_MEMORY = 0x004,
+ /// dynamic_allocators clause.
+ OMP_REQ_DYNAMIC_ALLOCATORS = 0x008,
+ LLVM_MARK_AS_BITMASK_ENUM(/*LargestValue=*/OMP_REQ_DYNAMIC_ALLOCATORS)
+};
+} // anonymous namespace
+
/// Describes ident structure that describes a source location.
/// All descriptions are taken from
/// https://github.com/llvm/llvm-project/blob/master/openmp/runtime/src/kmp.h
@@ -694,6 +712,8 @@
// *host_ptr, int32_t arg_num, void** args_base, void **args, size_t
// *arg_sizes, int64_t *arg_types, int32_t num_teams, int32_t thread_limit);
OMPRTL__tgt_target_teams_nowait,
+ // Call to void __tgt_register_requires(int64_t flags);
+ OMPRTL__tgt_register_requires,
// Call to void __tgt_register_lib(__tgt_bin_desc *desc);
OMPRTL__tgt_register_lib,
// Call to void __tgt_unregister_lib(__tgt_bin_desc *desc);
@@ -2294,6 +2314,14 @@
RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_target_teams_nowait");
break;
}
+ case OMPRTL__tgt_register_requires: {
+ // Build void __tgt_register_requires(int64_t flags);
+ llvm::Type *TypeParams[] = {CGM.Int64Ty};
+ auto *FnTy =
+ llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
+ RTLFn = CGM.CreateRuntimeFunction(FnTy, "__tgt_register_requires");
+ break;
+ }
case OMPRTL__tgt_register_lib: {
// Build void __tgt_register_lib(__tgt_bin_desc *desc);
QualType ParamTy =
@@ -8935,6 +8963,16 @@
" Expected target-based directive.");
}
+void CGOpenMPRuntime::checkArchForUnifiedAddressing(
+ const OMPRequiresDecl *D) {
+ for (const OMPClause *Clause : D->clauselists()) {
+ if (Clause->getClauseKind() == OMPC_unified_shared_memory) {
+ HasRequiresUnifiedSharedMemory = true;
+ break;
+ }
+ }
+}
+
bool CGOpenMPRuntime::hasAllocateAttributeForGlobalVar(const VarDecl *VD,
LangAS &AS) {
if (!VD || !VD->hasAttr<OMPAllocateDeclAttr>())
@@ -8993,6 +9031,35 @@
return !AlreadyEmittedTargetFunctions.insert(Name).second;
}
+llvm::Function *CGOpenMPRuntime::emitRequiresDirectiveRegFun() {
+ // If we don't have entries or if we are emitting code for the device, we
+ // don't need to do anything.
+ if (CGM.getLangOpts().OpenMPSimd || CGM.getLangOpts().OpenMPIsDevice ||
+ OffloadEntriesInfoManager.empty())
+ return nullptr;
+
+ // Create and register the function that handles the requires directives.
+ ASTContext &C = CGM.getContext();
+
+ llvm::Function *RequiresRegFn;
+ {
+ CodeGenFunction CGF(CGM);
+ const auto &FI = CGM.getTypes().arrangeNullaryFunction();
+ llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
+ std::string ReqName = getName({"omp_offloading", "requires_reg"});
+ RequiresRegFn = CGM.CreateGlobalInitOrDestructFunction(FTy, ReqName, FI);
+ CGF.StartFunction(GlobalDecl(), C.VoidTy, RequiresRegFn, FI, {});
+ OpenMPOffloadingRequiresDirFlags Flags = OMP_REQ_NONE;
+ //TODO: check for other requires clauses.
+ if (HasRequiresUnifiedSharedMemory)
+ Flags |= OMP_REQ_UNIFIED_SHARED_MEMORY;
+ CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__tgt_register_requires),
+ llvm::ConstantInt::get(CGM.Int64Ty, Flags));
+ CGF.FinishFunction();
+ }
+ return RequiresRegFn;
+}
+
llvm::Function *CGOpenMPRuntime::emitRegistrationFunction() {
// If we have offloading in the current module, we need to emit the entries
// now and register the offloading descriptor.
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
