r356702 - [OPENMP]Codegen support for allocate directive on global variables.
Alexey Bataev via cfe-commits
cfe-commits at lists.llvm.org
Thu Mar 21 12:35:27 PDT 2019
Author: abataev
Date: Thu Mar 21 12:35:27 2019
New Revision: 356702
URL: http://llvm.org/viewvc/llvm-project?rev=356702&view=rev
Log:
[OPENMP]Codegen support for allocate directive on global variables.
For the global variables the allocate directive must specify only the
predefined allocator. This allocator must be translated into the correct
form of the address space for the targets that support different address
spaces.
Added:
cfe/trunk/test/OpenMP/nvptx_allocate_codegen.cpp
Modified:
cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp
cfe/trunk/lib/CodeGen/CGOpenMPRuntime.h
cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
cfe/trunk/lib/CodeGen/CodeGenModule.cpp
Modified: cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp?rev=356702&r1=356701&r2=356702&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp (original)
+++ cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp Thu Mar 21 12:35:27 2019
@@ -8928,6 +8928,30 @@ void CGOpenMPRuntime::adjustTargetSpecif
" Expected target-based directive.");
}
+bool CGOpenMPRuntime::hasAllocateAttributeForGlobalVar(const VarDecl *VD,
+ LangAS &AS) {
+ if (!VD || !VD->hasAttr<OMPAllocateDeclAttr>())
+ return false;
+ const auto *A = VD->getAttr<OMPAllocateDeclAttr>();
+ switch(A->getAllocatorType()) {
+ case OMPAllocateDeclAttr::OMPDefaultMemAlloc:
+ // Not supported, fallback to the default mem space.
+ case OMPAllocateDeclAttr::OMPLargeCapMemAlloc:
+ case OMPAllocateDeclAttr::OMPCGroupMemAlloc:
+ case OMPAllocateDeclAttr::OMPHighBWMemAlloc:
+ case OMPAllocateDeclAttr::OMPLowLatMemAlloc:
+ case OMPAllocateDeclAttr::OMPThreadMemAlloc:
+ case OMPAllocateDeclAttr::OMPConstMemAlloc:
+ case OMPAllocateDeclAttr::OMPPTeamMemAlloc:
+ AS = LangAS::Default;
+ return true;
+ case OMPAllocateDeclAttr::OMPUserDefinedMemAlloc:
+ llvm_unreachable("Expected predefined allocator for the variables with the "
+ "static storage.");
+ }
+ return false;
+}
+
CGOpenMPRuntime::DisableAutoDeclareTargetRAII::DisableAutoDeclareTargetRAII(
CodeGenModule &CGM)
: CGM(CGM) {
Modified: cfe/trunk/lib/CodeGen/CGOpenMPRuntime.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGOpenMPRuntime.h?rev=356702&r1=356701&r2=356702&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGOpenMPRuntime.h (original)
+++ cfe/trunk/lib/CodeGen/CGOpenMPRuntime.h Thu Mar 21 12:35:27 2019
@@ -1598,6 +1598,11 @@ public:
/// Perform check on requires decl to ensure that target architecture
/// supports unified addressing
virtual void checkArchForUnifiedAddressing(const OMPRequiresDecl *D) const {}
+
+ /// Checks if the variable has associated OMPAllocateDeclAttr attribute with
+ /// the predefined allocator and translates it into the corresponding address
+ /// space.
+ virtual bool hasAllocateAttributeForGlobalVar(const VarDecl *VD, LangAS &AS);
};
/// Class supports emissionof SIMD-only code.
Modified: cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp?rev=356702&r1=356701&r2=356702&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp (original)
+++ cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp Thu Mar 21 12:35:27 2019
@@ -4840,6 +4840,34 @@ unsigned CGOpenMPRuntimeNVPTX::getDefaul
return CGM.getContext().getTargetAddressSpace(LangAS::cuda_constant);
}
+bool CGOpenMPRuntimeNVPTX::hasAllocateAttributeForGlobalVar(const VarDecl *VD,
+ LangAS &AS) {
+ if (!VD || !VD->hasAttr<OMPAllocateDeclAttr>())
+ return false;
+ const auto *A = VD->getAttr<OMPAllocateDeclAttr>();
+ switch(A->getAllocatorType()) {
+ case OMPAllocateDeclAttr::OMPDefaultMemAlloc:
+ // Not supported, fallback to the default mem space.
+ case OMPAllocateDeclAttr::OMPThreadMemAlloc:
+ case OMPAllocateDeclAttr::OMPLargeCapMemAlloc:
+ case OMPAllocateDeclAttr::OMPCGroupMemAlloc:
+ case OMPAllocateDeclAttr::OMPHighBWMemAlloc:
+ case OMPAllocateDeclAttr::OMPLowLatMemAlloc:
+ AS = LangAS::Default;
+ return true;
+ case OMPAllocateDeclAttr::OMPConstMemAlloc:
+ AS = LangAS::cuda_constant;
+ return true;
+ case OMPAllocateDeclAttr::OMPPTeamMemAlloc:
+ AS = LangAS::cuda_shared;
+ return true;
+ case OMPAllocateDeclAttr::OMPUserDefinedMemAlloc:
+ llvm_unreachable("Expected predefined allocator for the variables with the "
+ "static storage.");
+ }
+ return false;
+}
+
// Get current CudaArch and ignore any unknown values
static CudaArch getCudaArch(CodeGenModule &CGM) {
if (!CGM.getTarget().hasFeature("ptx"))
Modified: cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.h?rev=356702&r1=356701&r2=356702&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.h (original)
+++ cfe/trunk/lib/CodeGen/CGOpenMPRuntimeNVPTX.h Thu Mar 21 12:35:27 2019
@@ -389,6 +389,11 @@ public:
/// address space by default.
unsigned getDefaultFirstprivateAddressSpace() const override;
+ /// Checks if the variable has associated OMPAllocateDeclAttr attribute with
+ /// the predefined allocator and translates it into the corresponding address
+ /// space.
+ bool hasAllocateAttributeForGlobalVar(const VarDecl *VD, LangAS &AS) override;
+
private:
/// Track the execution mode when codegening directives within a target
/// region. The appropriate mode (SPMD/NON-SPMD) is set on entry to the
Modified: cfe/trunk/lib/CodeGen/CodeGenModule.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CodeGenModule.cpp?rev=356702&r1=356701&r2=356702&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CodeGenModule.cpp (original)
+++ cfe/trunk/lib/CodeGen/CodeGenModule.cpp Thu Mar 21 12:35:27 2019
@@ -3387,6 +3387,11 @@ LangAS CodeGenModule::GetGlobalVarAddres
return LangAS::cuda_device;
}
+ if (LangOpts.OpenMP) {
+ LangAS AS;
+ if (OpenMPRuntime->hasAllocateAttributeForGlobalVar(D, AS))
+ return AS;
+ }
return getTargetCodeGenInfo().getGlobalVarAddressSpace(*this, D);
}
Added: cfe/trunk/test/OpenMP/nvptx_allocate_codegen.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/OpenMP/nvptx_allocate_codegen.cpp?rev=356702&view=auto
==============================================================================
--- cfe/trunk/test/OpenMP/nvptx_allocate_codegen.cpp (added)
+++ cfe/trunk/test/OpenMP/nvptx_allocate_codegen.cpp Thu Mar 21 12:35:27 2019
@@ -0,0 +1,71 @@
+// RUN: %clang_cc1 -verify -fopenmp -triple x86_64-apple-darwin10.6.0 -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc -o %t-host.bc %s
+// RUN: %clang_cc1 -verify -fopenmp -triple nvptx64-nvidia-cuda -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-host.bc -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+#pragma omp declare target
+typedef void **omp_allocator_handle_t;
+extern const omp_allocator_handle_t omp_default_mem_alloc;
+extern const omp_allocator_handle_t omp_large_cap_mem_alloc;
+extern const omp_allocator_handle_t omp_const_mem_alloc;
+extern const omp_allocator_handle_t omp_high_bw_mem_alloc;
+extern const omp_allocator_handle_t omp_low_lat_mem_alloc;
+extern const omp_allocator_handle_t omp_cgroup_mem_alloc;
+extern const omp_allocator_handle_t omp_pteam_mem_alloc;
+extern const omp_allocator_handle_t omp_thread_mem_alloc;
+
+// CHECK-DAG: @{{.+}}St1{{.+}}b{{.+}} = external global i32,
+// CHECK-DAG: @a = global i32 0,
+// CHECK-DAG: @b = addrspace(4) global i32 0,
+// CHECK-DAG: @c = global i32 0,
+// CHECK-DAG: @d = global %struct.St1 zeroinitializer,
+// CHECK-DAG: @{{.+}}ns{{.+}}a{{.+}} = addrspace(3) global i32 0,
+// CHECK-DAG: @{{.+}}main{{.+}}a{{.*}} = internal global i32 0,
+// CHECK-DAG: @{{.+}}ST{{.+}}m{{.+}} = external global i32,
+struct St{
+ int a;
+};
+
+struct St1{
+ int a;
+ static int b;
+#pragma omp allocate(b) allocator(omp_default_mem_alloc)
+} d;
+
+int a, b, c;
+#pragma omp allocate(a) allocator(omp_large_cap_mem_alloc)
+#pragma omp allocate(b) allocator(omp_const_mem_alloc)
+#pragma omp allocate(d, c) allocator(omp_high_bw_mem_alloc)
+
+template <class T>
+struct ST {
+ static T m;
+ #pragma omp allocate(m) allocator(omp_low_lat_mem_alloc)
+};
+
+template <class T> T foo() {
+ T v;
+ #pragma omp allocate(v) allocator(omp_cgroup_mem_alloc)
+ v = ST<T>::m;
+ return v;
+}
+
+namespace ns{
+ int a;
+}
+#pragma omp allocate(ns::a) allocator(omp_pteam_mem_alloc)
+
+int main () {
+ static int a;
+#pragma omp allocate(a) allocator(omp_thread_mem_alloc)
+ a=2;
+ double b = 3;
+#pragma omp allocate(b)
+ return (foo<int>());
+}
+
+extern template int ST<int>::m;
+#pragma omp end declare target
+#endif
More information about the cfe-commits
mailing list