[clang-tools-extra] [openmp] [llvm] [clang] [PGO][OpenMP] Instrumentation for GPU devices (PR #76587)
https://github.com/EthanLuisMcDonough updated https://github.com/llvm/llvm-project/pull/76587 >From 530eb982b9770190377bb0bd09c5cb715f34d484 Mon Sep 17 00:00:00 2001 From: Ethan Luis McDonough Date: Fri, 15 Dec 2023 20:38:38 -0600 Subject: [PATCH 1/7] Add profiling functions to libomptarget --- .../include/llvm/Frontend/OpenMP/OMPKinds.def | 3 +++ openmp/libomptarget/DeviceRTL/CMakeLists.txt | 2 ++ .../DeviceRTL/include/Profiling.h | 21 +++ .../libomptarget/DeviceRTL/src/Profiling.cpp | 19 + 4 files changed, 45 insertions(+) create mode 100644 openmp/libomptarget/DeviceRTL/include/Profiling.h create mode 100644 openmp/libomptarget/DeviceRTL/src/Profiling.cpp diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def index d22d2a8e948b00..1d887d5cb58127 100644 --- a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def +++ b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def @@ -503,6 +503,9 @@ __OMP_RTL(__kmpc_barrier_simple_generic, false, Void, IdentPtr, Int32) __OMP_RTL(__kmpc_warp_active_thread_mask, false, Int64,) __OMP_RTL(__kmpc_syncwarp, false, Void, Int64) +__OMP_RTL(__llvm_profile_register_function, false, Void, VoidPtr) +__OMP_RTL(__llvm_profile_register_names_function, false, Void, VoidPtr, Int64) + __OMP_RTL(__last, false, Void, ) #undef __OMP_RTL diff --git a/openmp/libomptarget/DeviceRTL/CMakeLists.txt b/openmp/libomptarget/DeviceRTL/CMakeLists.txt index 1ce3e1e40a80ab..55ee15d068c67b 100644 --- a/openmp/libomptarget/DeviceRTL/CMakeLists.txt +++ b/openmp/libomptarget/DeviceRTL/CMakeLists.txt @@ -89,6 +89,7 @@ set(include_files ${include_directory}/Interface.h ${include_directory}/LibC.h ${include_directory}/Mapping.h + ${include_directory}/Profiling.h ${include_directory}/State.h ${include_directory}/Synchronization.h ${include_directory}/Types.h @@ -104,6 +105,7 @@ set(src_files ${source_directory}/Mapping.cpp ${source_directory}/Misc.cpp ${source_directory}/Parallelism.cpp + ${source_directory}/Profiling.cpp ${source_directory}/Reduction.cpp ${source_directory}/State.cpp ${source_directory}/Synchronization.cpp diff --git a/openmp/libomptarget/DeviceRTL/include/Profiling.h b/openmp/libomptarget/DeviceRTL/include/Profiling.h new file mode 100644 index 00..68c7744cd60752 --- /dev/null +++ b/openmp/libomptarget/DeviceRTL/include/Profiling.h @@ -0,0 +1,21 @@ +//=== Profiling.h - OpenMP interface -- C++ -*-===// +// +// 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 +// +//===--===// +// +// +//===--===// + +#ifndef OMPTARGET_DEVICERTL_PROFILING_H +#define OMPTARGET_DEVICERTL_PROFILING_H + +extern "C" { + +void __llvm_profile_register_function(void *ptr); +void __llvm_profile_register_names_function(void *ptr, long int i); +} + +#endif diff --git a/openmp/libomptarget/DeviceRTL/src/Profiling.cpp b/openmp/libomptarget/DeviceRTL/src/Profiling.cpp new file mode 100644 index 00..799477f5e47d27 --- /dev/null +++ b/openmp/libomptarget/DeviceRTL/src/Profiling.cpp @@ -0,0 +1,19 @@ +//===--- Profiling.cpp C++ ---===// +// +// 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 +// +//===--===// + +#include "Profiling.h" + +#pragma omp begin declare target device_type(nohost) + +extern "C" { + +void __llvm_profile_register_function(void *ptr) {} +void __llvm_profile_register_names_function(void *ptr, long int i) {} +} + +#pragma omp end declare target >From fb067d4ffe604fd68cf90b705db1942bce49dbb1 Mon Sep 17 00:00:00 2001 From: Ethan Luis McDonough Date: Sat, 16 Dec 2023 01:18:41 -0600 Subject: [PATCH 2/7] Fix PGO instrumentation for GPU targets --- clang/lib/CodeGen/CodeGenPGO.cpp | 10 -- .../lib/Transforms/Instrumentation/InstrProfiling.cpp | 11 --- 2 files changed, 16 insertions(+), 5 deletions(-) diff --git a/clang/lib/CodeGen/CodeGenPGO.cpp b/clang/lib/CodeGen/CodeGenPGO.cpp index 81bf8ea696b164..edae6885b528ac 100644 --- a/clang/lib/CodeGen/CodeGenPGO.cpp +++ b/clang/lib/CodeGen/CodeGenPGO.cpp @@ -959,8 +959,14 @@ void CodeGenPGO::emitCounterIncrement(CGBuilderTy &Builder, const Stmt *S, unsigned Counter = (*RegionCounterMap)[S]; - llvm::Value *Args[] = {FuncNameVar, - Builder.getInt64(FunctionHash), + // Make sure that pointer to global is passed in with zero addrspace + // This
[clang-tools-extra] [openmp] [llvm] [clang] [PGO][OpenMP] Instrumentation for GPU devices (PR #76587)
https://github.com/EthanLuisMcDonough updated https://github.com/llvm/llvm-project/pull/76587 >From 530eb982b9770190377bb0bd09c5cb715f34d484 Mon Sep 17 00:00:00 2001 From: Ethan Luis McDonough Date: Fri, 15 Dec 2023 20:38:38 -0600 Subject: [PATCH 1/8] Add profiling functions to libomptarget --- .../include/llvm/Frontend/OpenMP/OMPKinds.def | 3 +++ openmp/libomptarget/DeviceRTL/CMakeLists.txt | 2 ++ .../DeviceRTL/include/Profiling.h | 21 +++ .../libomptarget/DeviceRTL/src/Profiling.cpp | 19 + 4 files changed, 45 insertions(+) create mode 100644 openmp/libomptarget/DeviceRTL/include/Profiling.h create mode 100644 openmp/libomptarget/DeviceRTL/src/Profiling.cpp diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def index d22d2a8e948b00..1d887d5cb58127 100644 --- a/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def +++ b/llvm/include/llvm/Frontend/OpenMP/OMPKinds.def @@ -503,6 +503,9 @@ __OMP_RTL(__kmpc_barrier_simple_generic, false, Void, IdentPtr, Int32) __OMP_RTL(__kmpc_warp_active_thread_mask, false, Int64,) __OMP_RTL(__kmpc_syncwarp, false, Void, Int64) +__OMP_RTL(__llvm_profile_register_function, false, Void, VoidPtr) +__OMP_RTL(__llvm_profile_register_names_function, false, Void, VoidPtr, Int64) + __OMP_RTL(__last, false, Void, ) #undef __OMP_RTL diff --git a/openmp/libomptarget/DeviceRTL/CMakeLists.txt b/openmp/libomptarget/DeviceRTL/CMakeLists.txt index 1ce3e1e40a80ab..55ee15d068c67b 100644 --- a/openmp/libomptarget/DeviceRTL/CMakeLists.txt +++ b/openmp/libomptarget/DeviceRTL/CMakeLists.txt @@ -89,6 +89,7 @@ set(include_files ${include_directory}/Interface.h ${include_directory}/LibC.h ${include_directory}/Mapping.h + ${include_directory}/Profiling.h ${include_directory}/State.h ${include_directory}/Synchronization.h ${include_directory}/Types.h @@ -104,6 +105,7 @@ set(src_files ${source_directory}/Mapping.cpp ${source_directory}/Misc.cpp ${source_directory}/Parallelism.cpp + ${source_directory}/Profiling.cpp ${source_directory}/Reduction.cpp ${source_directory}/State.cpp ${source_directory}/Synchronization.cpp diff --git a/openmp/libomptarget/DeviceRTL/include/Profiling.h b/openmp/libomptarget/DeviceRTL/include/Profiling.h new file mode 100644 index 00..68c7744cd60752 --- /dev/null +++ b/openmp/libomptarget/DeviceRTL/include/Profiling.h @@ -0,0 +1,21 @@ +//=== Profiling.h - OpenMP interface -- C++ -*-===// +// +// 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 +// +//===--===// +// +// +//===--===// + +#ifndef OMPTARGET_DEVICERTL_PROFILING_H +#define OMPTARGET_DEVICERTL_PROFILING_H + +extern "C" { + +void __llvm_profile_register_function(void *ptr); +void __llvm_profile_register_names_function(void *ptr, long int i); +} + +#endif diff --git a/openmp/libomptarget/DeviceRTL/src/Profiling.cpp b/openmp/libomptarget/DeviceRTL/src/Profiling.cpp new file mode 100644 index 00..799477f5e47d27 --- /dev/null +++ b/openmp/libomptarget/DeviceRTL/src/Profiling.cpp @@ -0,0 +1,19 @@ +//===--- Profiling.cpp C++ ---===// +// +// 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 +// +//===--===// + +#include "Profiling.h" + +#pragma omp begin declare target device_type(nohost) + +extern "C" { + +void __llvm_profile_register_function(void *ptr) {} +void __llvm_profile_register_names_function(void *ptr, long int i) {} +} + +#pragma omp end declare target >From fb067d4ffe604fd68cf90b705db1942bce49dbb1 Mon Sep 17 00:00:00 2001 From: Ethan Luis McDonough Date: Sat, 16 Dec 2023 01:18:41 -0600 Subject: [PATCH 2/8] Fix PGO instrumentation for GPU targets --- clang/lib/CodeGen/CodeGenPGO.cpp | 10 -- .../lib/Transforms/Instrumentation/InstrProfiling.cpp | 11 --- 2 files changed, 16 insertions(+), 5 deletions(-) diff --git a/clang/lib/CodeGen/CodeGenPGO.cpp b/clang/lib/CodeGen/CodeGenPGO.cpp index 81bf8ea696b164..edae6885b528ac 100644 --- a/clang/lib/CodeGen/CodeGenPGO.cpp +++ b/clang/lib/CodeGen/CodeGenPGO.cpp @@ -959,8 +959,14 @@ void CodeGenPGO::emitCounterIncrement(CGBuilderTy &Builder, const Stmt *S, unsigned Counter = (*RegionCounterMap)[S]; - llvm::Value *Args[] = {FuncNameVar, - Builder.getInt64(FunctionHash), + // Make sure that pointer to global is passed in with zero addrspace + // This