AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge 'master' into 'amd-master'

Browse Source 
Change-Id: I85682ce0d64ef617489640f6f16bafff2a698585
This commit is contained in:
Jenkins
2019-08-30 18:18:48 -04:00
parent 1518155479 57760c6a57
commit 6c776daba1
36 changed files with 1271 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files
bin/hipcc
+1 -6
View File
@@ -383,7 +383,6 @@ if($HIP_PLATFORM eq "nvcc"){
my $toolArgs = ""; # arguments to pass to the hcc or nvcc tool
my $optArg = ""; # -O args
my $gArg = ""; # -g args
foreach $arg (@ARGV)
{
@@ -494,10 +493,6 @@ foreach $arg (@ARGV)
{
$optArg = $arg;
}
if($arg =~ m/^-g/)
{
$gArg = $arg;
}
## process linker response file for hip-clang
## extract object files from static library and pass them directly to
@@ -876,7 +871,7 @@ if ($needHipHcc) {
if ($HIP_PLATFORM eq "clang") {
# Set default optimization level to -O3 for hip-clang.
if ($optArg eq "" and $gArg ne "-g") {
if ($optArg eq "") {
$HIPCXXFLAGS .= " -O3";
$HIPLDFLAGS .= " -O3";
}
docs/markdown/CUDNN_API_supported_by_HIP.md
+9 -2
View File
@@ -233,9 +233,11 @@
| define |`CUDNN_SEQDATA_DIM_COUNT` | 9.0 | |
| struct |`cudnnSeqDataStruct` | 9.0 | |
| struct* |`cudnnSeqDataDescriptor_t` | 9.0 | |
| enum |***`cudnnAttnQueryMap_t`*** | 9.0 | |
| unsigned |***`cudnnAttnQueryMap_t`*** | 9.0 | |
| 0 |*`CUDNN_ATTN_QUERYMAP_ALL_TO_ONE`* | 9.0 | |
| 1 |*`CUDNN_ATTN_QUERYMAP_ONE_TO_ONE`* | 9.0 | |
| 1U << 0 |*`CUDNN_ATTN_QUERYMAP_ONE_TO_ONE`* | 9.0 | |
| 1 |*`CUDNN_ATTN_DISABLE_PROJ_BIASES`* | 10.1 Update 2 | |
| 1U << 1 |*`CUDNN_ATTN_ENABLE_PROJ_BIASES`* | 10.1 Update 2 | |
| struct |`cudnnAttnStruct` | 9.0 | |
| struct* |`cudnnAttnDescriptor_t` | 9.0 | |
| enum |***`cudnnMultiHeadAttnWeightKind_t`*** | 9.0 | |
@@ -243,6 +245,11 @@
| 1 |*`CUDNN_MH_ATTN_K_WEIGHTS`* | 9.0 | |
| 2 |*`CUDNN_MH_ATTN_V_WEIGHTS`* | 9.0 | |
| 3 |*`CUDNN_MH_ATTN_O_WEIGHTS`* | 9.0 | |
| 4 |*`CUDNN_MH_ATTN_Q_BIASES`* | 10.1 Update 2 | |
| 5 |*`CUDNN_MH_ATTN_K_BIASES`* | 10.1 Update 2 | |
| 6 |*`CUDNN_MH_ATTN_V_BIASES`* | 10.1 Update 2 | |
| 7 |*`CUDNN_MH_ATTN_O_BIASES`* | 10.1 Update 2 | |
| define 8 |`CUDNN_ATTN_WKIND_COUNT` | 10.1 Update 2 | |
| enum |***`cudnnWgradMode_t`*** | 9.0 | |
| 0 |*`CUDNN_WGRAD_MODE_ADD`* | 9.0 | |
| 1 |*`CUDNN_WGRAD_MODE_SET`* | 9.0 | |
hipify-clang/CMakeLists.txt
+20 -13
View File
@@ -65,24 +65,31 @@ target_link_libraries(hipify-clang PRIVATE
LLVMOption
LLVMCore)
if(WIN32)
target_link_libraries(hipify-clang PRIVATE version)
endif()
if ((LLVM_PACKAGE_VERSION VERSION_EQUAL "7") OR (LLVM_PACKAGE_VERSION VERSION_GREATER "7"))
if (LLVM_PACKAGE_VERSION VERSION_GREATER_EQUAL "7")
target_link_libraries(hipify-clang PRIVATE clangToolingInclusions)
endif()
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${EXTRA_CFLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${EXTRA_CFLAGS} -std=c++14")
if(MSVC)
if (MSVC)
target_link_libraries(hipify-clang PRIVATE version)
target_compile_options(hipify-clang PRIVATE "/Od /GR- /EHs- /EHc-")
set(CMAKE_CXX_LINK_FLAGS "${CMAKE_CXX_LINK_FLAGS} /SUBSYSTEM:WINDOWS")
set(StdCpp "/std:c++")
else()
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread -fno-rtti -fvisibility-inlines-hidden")
set(StdCpp "-std=c++")
endif()
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DHIPIFY_CLANG_RES=\\\"${LLVM_LIBRARY_DIRS}/clang/${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}.${LLVM_VERSION_PATCH}\\\"")
if (LLVM_PACKAGE_VERSION VERSION_GREATER_EQUAL "10.0")
string(APPEND StdCpp "14")
# MSVC starting from 1900 (VS 2015) supports only the following c++ std values: c++14|c++17|c++latest
elseif (MSVC)
set(StdCpp "")
else()
string(APPEND StdCpp "11")
endif()
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${EXTRA_CFLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${EXTRA_CFLAGS} ${StdCpp} -DHIPIFY_CLANG_RES=\\\"${LLVM_LIBRARY_DIRS}/clang/${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}.${LLVM_VERSION_PATCH}\\\"")
install(TARGETS hipify-clang DESTINATION bin)
@@ -111,15 +118,15 @@ if (HIPIFY_CLANG_TESTS)
(CUDA_VERSION VERSION_GREATER "9.2" AND LLVM_PACKAGE_VERSION VERSION_LESS "8.0") OR
(CUDA_VERSION VERSION_GREATER "10.0" AND LLVM_PACKAGE_VERSION VERSION_LESS "9.0"))
message(SEND_ERROR "CUDA ${CUDA_VERSION} is not supported by clang ${LLVM_PACKAGE_VERSION}.")
if (CUDA_VERSION VERSION_LESS "7.0")
if (CUDA_VERSION_MAJOR VERSION_LESS "7")
message(STATUS "Please install CUDA 7.0 or higher.")
elseif ((CUDA_VERSION VERSION_EQUAL "7.0") OR (CUDA_VERSION VERSION_EQUAL "7.5"))
elseif (CUDA_VERSION_MAJOR VERSION_LESS "8")
message(STATUS "Please install clang 3.8 or higher.")
elseif (CUDA_VERSION VERSION_EQUAL "8.0")
elseif (CUDA_VERSION_MAJOR VERSION_LESS "9")
message(STATUS "Please install clang 4.0 or higher.")
elseif (CUDA_VERSION VERSION_EQUAL "9.0")
message(STATUS "Please install clang 6.0 or higher.")
elseif ((CUDA_VERSION VERSION_EQUAL "9.1") OR (CUDA_VERSION VERSION_EQUAL "9.2"))
elseif (CUDA_VERSION_MAJOR VERSION_LESS "10")
message(STATUS "Please install clang 7.0 or higher.")
elseif (CUDA_VERSION VERSION_EQUAL "10.0")
message(STATUS "Please install clang 8.0 or higher.")
hipify-clang/src/ArgParse.cpp
+5
View File
@@ -138,4 +138,9 @@ cl::list<std::string> MacroNames("D",
cl::Prefix,
cl::cat(ToolTemplateCategory));
cl::opt<bool> SkipExcludedPPConditionalBlocks("skip-excluded-preprocessor-conditional-blocks",
cl::desc("Enable default preprocessor behaviour by skipping undefined conditional blocks"),
cl::value_desc("skip-excluded-preprocessor-conditional-blocks"),
cl::cat(ToolTemplateCategory));
cl::extrahelp CommonHelp(ct::CommonOptionsParser::HelpMessage);
hipify-clang/src/ArgParse.h
+1
View File
@@ -52,3 +52,4 @@ extern cl::opt<bool> Examine;
extern cl::extrahelp CommonHelp;
extern cl::opt<bool> TranslateToRoc;
extern cl::opt<bool> DashDash;
extern cl::opt<bool> SkipExcludedPPConditionalBlocks;
hipify-clang/src/CUDA2HIP_DNN_API_types.cpp
+9 -2
View File
@@ -229,14 +229,21 @@ const std::map<llvm::StringRef, hipCounter> CUDA_DNN_TYPE_NAME_MAP{
{"CUDNN_SEQDATA_BATCH_DIM", {"HIPDNN_SEQDATA_BATCH_DIM", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 1
{"CUDNN_SEQDATA_BEAM_DIM", {"HIPDNN_SEQDATA_BEAM_DIM", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 2
{"CUDNN_SEQDATA_VECT_DIM", {"HIPDNN_SEQDATA_VECT_DIM", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 3
{"cudnnAttnQueryMap_t", {"hipdnnAttnQueryMap_t", "", CONV_TYPE, API_DNN, HIP_UNSUPPORTED}},
{"cudnnAttnQueryMap_t", {"hipdnnAttnQueryMap_t", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}},
{"CUDNN_ATTN_QUERYMAP_ALL_TO_ONE", {"HIPDNN_ATTN_QUERYMAP_ALL_TO_ONE", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 0
{"CUDNN_ATTN_QUERYMAP_ONE_TO_ONE", {"HIPDNN_ATTN_QUERYMAP_ONE_TO_ONE", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 1
{"CUDNN_ATTN_QUERYMAP_ONE_TO_ONE", {"HIPDNN_ATTN_QUERYMAP_ONE_TO_ONE", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 1U << 0
{"CUDNN_ATTN_DISABLE_PROJ_BIASES", {"HIPDNN_ATTN_DISABLE_PROJ_BIASES", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 0
{"CUDNN_ATTN_ENABLE_PROJ_BIASES", {"HIPDNN_ATTN_ENABLE_PROJ_BIASES", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 1U << 1
{"cudnnMultiHeadAttnWeightKind_t", {"hipdnnMultiHeadAttnWeightKind_t", "", CONV_TYPE, API_DNN, HIP_UNSUPPORTED}},
{"CUDNN_MH_ATTN_Q_WEIGHTS", {"HIPDNN_MH_ATTN_Q_WEIGHTS", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 0
{"CUDNN_MH_ATTN_K_WEIGHTS", {"HIPDNN_MH_ATTN_K_WEIGHTS", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 1
{"CUDNN_MH_ATTN_V_WEIGHTS", {"HIPDNN_MH_ATTN_V_WEIGHTS", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 2
{"CUDNN_MH_ATTN_O_WEIGHTS", {"HIPDNN_MH_ATTN_O_WEIGHTS", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 3
{"CUDNN_MH_ATTN_Q_BIASES", {"HIPDNN_MH_ATTN_Q_BIASES", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 4
{"CUDNN_MH_ATTN_K_BIASES", {"HIPDNN_MH_ATTN_K_BIASES", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 5
{"CUDNN_MH_ATTN_V_BIASES", {"HIPDNN_MH_ATTN_V_BIASES", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 6
{"CUDNN_MH_ATTN_O_BIASES", {"HIPDNN_MH_ATTN_O_BIASES", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 7
{"CUDNN_ATTN_WKIND_COUNT", {"HIPDNN_ATTN_WKIND_COUNT", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 8
{"cudnnWgradMode_t", {"hipdnnWgradMode_t", "", CONV_TYPE, API_DNN, HIP_UNSUPPORTED}},
{"CUDNN_WGRAD_MODE_ADD", {"HIPDNN_WGRAD_MODE_ADD", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 0
{"CUDNN_WGRAD_MODE_SET", {"HIPDNN_WGRAD_MODE_SET", "", CONV_NUMERIC_LITERAL, API_DNN, HIP_UNSUPPORTED}}, // 1
hipify-clang/src/HipifyAction.cpp
+5
View File
@@ -484,6 +484,11 @@ public:
}
bool HipifyAction::BeginInvocation(clang::CompilerInstance &CI) {
llcompat::RetainExcludedConditionalBlocks(CI);
return true;
}
void HipifyAction::ExecuteAction() {
clang::Preprocessor& PP = getCompilerInstance().getPreprocessor();
clang::SourceManager& SM = getCompilerInstance().getSourceManager();
hipify-clang/src/HipifyAction.h
+2
View File
@@ -91,6 +91,8 @@ protected:
void insertReplacement(const ct::Replacement& rep, const clang::FullSourceLoc& fullSL);
// FrontendAction entry point.
void ExecuteAction() override;
// Callback before starting processing a single input; used by hipify-clang for setting Preprocessor options.
bool BeginInvocation(clang::CompilerInstance &CI) override;
// Called at the start of each new file to process.
void EndSourceFileAction() override;
// MatchCallback API entry point. Called by the AST visitor while searching the AST for things we registered an interest for.
hipify-clang/src/LLVMCompat.cpp
+21
View File
@@ -20,8 +20,13 @@ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include "ArgParse.h"
#include "LLVMCompat.h"
#include "llvm/Support/Path.h"
#include "clang/Lex/PreprocessorOptions.h"
#include "clang/Frontend/CompilerInstance.h"
const std::string sHipify = "[HIPIFY] ", sConflict = "conflict: ", sError = "error: ", sWarning = "warning: ";
namespace llcompat {
@@ -122,4 +127,20 @@ bool pragma_once_outside_header() {
#endif
}
void RetainExcludedConditionalBlocks(clang::CompilerInstance &CI) {
#if LLVM_VERSION_MAJOR > 9
clang::PreprocessorOptions &PPOpts = CI.getPreprocessorOpts();
PPOpts.RetainExcludedConditionalBlocks = !SkipExcludedPPConditionalBlocks;
#endif
}
bool CheckCompatibility() {
#if LLVM_VERSION_MAJOR < 10
if (SkipExcludedPPConditionalBlocks) {
llvm::errs() << "\n" << sHipify << sWarning << "Option '" << SkipExcludedPPConditionalBlocks.ArgStr.str() << "' is supported starting from LLVM version 10.0\n";
}
#endif
return true;
}
} // namespace llcompat
hipify-clang/src/LLVMCompat.h
+6
View File
@@ -30,6 +30,8 @@ THE SOFTWARE.
namespace ct = clang::tooling;
extern const std::string sHipify, sConflict, sError, sWarning;
// Things for papering over the differences between different LLVM versions.
namespace llcompat {
@@ -83,4 +85,8 @@ std::error_code real_path(const Twine &path, SmallVectorImpl<char> &output,
bool pragma_once_outside_header();
void RetainExcludedConditionalBlocks(clang::CompilerInstance &CI);
bool CheckCompatibility();
} // namespace llcompat
hipify-clang/src/main.cpp
+12 -10
View File
@@ -37,7 +37,6 @@ THE SOFTWARE.
#define DEBUG_TYPE "cuda2hip"
std::string sHipify = "[HIPIFY] ", sConflict = "conflict: ", sError = "error: ";
namespace ct = clang::tooling;
std::string getAbsoluteFilePath(const std::string& sFile, std::error_code& EC) {
@@ -240,13 +239,16 @@ int main(int argc, const char **argv) {
}
llcompat::PrintStackTraceOnErrorSignal();
ct::CommonOptionsParser OptionsParser(argc, argv, ToolTemplateCategory, llvm::cl::ZeroOrMore);
if (!llcompat::CheckCompatibility()) {
return 1;
}
std::vector<std::string> fileSources = OptionsParser.getSourcePathList();
if (fileSources.empty() && !GeneratePerl && !GeneratePython) {
llvm::errs() << "\n" << sHipify << sError << "Must specify at least 1 positional argument for source file." << "\n";
llvm::errs() << "\n" << sHipify << sError << "Must specify at least 1 positional argument for source file" << "\n";
return 1;
}
if (!generatePerl(GeneratePerl)) {
llvm::errs() << "\n" << sHipify << sError << "hipify-perl generating failed." << "\n";
llvm::errs() << "\n" << sHipify << sError << "hipify-perl generating failed" << "\n";
return 1;
}
bool bToRoc = TranslateToRoc;
@@ -254,7 +256,7 @@ int main(int argc, const char **argv) {
bool bToPython = generatePython(GeneratePython);
TranslateToRoc = bToRoc;
if (!bToPython) {
llvm::errs() << "\n" << sHipify << sError << "hipify-python generating failed." << "\n";
llvm::errs() << "\n" << sHipify << sError << "hipify-python generating failed" << "\n";
return 1;
}
if (fileSources.empty()) {
@@ -268,15 +270,15 @@ int main(int argc, const char **argv) {
}
if (!dst.empty()) {
if (fileSources.size() > 1) {
llvm::errs() << sHipify << sConflict << "-o and multiple source files are specified.\n";
llvm::errs() << sHipify << sConflict << "-o and multiple source files are specified\n";
return 1;
}
if (Inplace) {
llvm::errs() << sHipify << sConflict << "both -o and -inplace options are specified.\n";
llvm::errs() << sHipify << sConflict << "both -o and -inplace options are specified\n";
return 1;
}
if (NoOutput) {
llvm::errs() << sHipify << sConflict << "both -no-output and -o options are specified.\n";
llvm::errs() << sHipify << sConflict << "both -no-output and -o options are specified\n";
return 1;
}
if (!dstDir.empty()) {
@@ -284,11 +286,11 @@ int main(int argc, const char **argv) {
}
}
if (NoOutput && Inplace) {
llvm::errs() << sHipify << sConflict << "both -no-output and -inplace options are specified.\n";
llvm::errs() << sHipify << sConflict << "both -no-output and -inplace options are specified\n";
return 1;
}
if (!dstDir.empty() && Inplace) {
llvm::errs() << sHipify << sConflict << "both -o-dir and -inplace options are specified.\n";
llvm::errs() << sHipify << sConflict << "both -o-dir and -inplace options are specified\n";
return 1;
}
if (Examine) {
@@ -425,8 +427,8 @@ int main(int argc, const char **argv) {
// Hipify _all_ the things!
if (Tool.runAndSave(&actionFactory)) {
currentStat.hasErrors = true;
LLVM_DEBUG(llvm::dbgs() << "Skipped some replacements.\n");
Result = 1;
LLVM_DEBUG(llvm::dbgs() << "Skipped some replacements.\n");
}
// Copy the tmpfile to the output
if (!NoOutput && !currentStat.hasErrors) {
include/hip/hcc_detail/device_functions.h
+21 -13
View File
@@ -748,28 +748,36 @@ unsigned long long int __ballot64(int predicate) {
// hip.amdgcn.bc - lanemask
__device__
inline
int64_t __lanemask_gt()
uint64_t __lanemask_gt()
{
int32_t activelane = __ockl_activelane_u32();
int64_t ballot = __ballot64(1);
if (activelane != 63) {
int64_t tmp = (~0ULL) << (activelane + 1);
return tmp & ballot;
}
return 0;
uint32_t lane = __ockl_lane_u32();
if (lane == 63)
return 0;
uint64_t ballot = __ballot64(1);
uint64_t mask = (~((uint64_t)0)) << (lane + 1);
return mask & ballot;
}
__device__
inline
int64_t __lanemask_lt()
uint64_t __lanemask_lt()
{
int32_t activelane = __ockl_activelane_u32();
uint32_t lane = __ockl_lane_u32();
int64_t ballot = __ballot64(1);
if (activelane == 0)
return 0;
return ballot;
uint64_t mask = ((uint64_t)1 << lane) - (uint64_t)1;
return mask & ballot;
}
__device__
inline
uint64_t __lanemask_eq()
{
uint32_t lane = __ockl_lane_u32();
int64_t mask = ((uint64_t)1 << lane);
return mask;
}
__device__ inline void* __local_to_generic(void* p) { return p; }
#ifdef __HIP_DEVICE_COMPILE__
include/hip/hcc_detail/device_library_decls.h
+10
View File
@@ -62,6 +62,16 @@ extern "C" __device__ __attribute__((const)) float __ocml_fmax_f32(float, float)
extern "C" __device__ __attribute__((convergent)) void __ockl_gws_init(uint nwm1, uint rid);
extern "C" __device__ __attribute__((convergent)) void __ockl_gws_barrier(uint nwm1, uint rid);
extern "C" __device__ __attribute__((const)) uint32_t __ockl_lane_u32();
extern "C" __device__ __attribute__((const)) int __ockl_grid_is_valid(void);
extern "C" __device__ __attribute__((convergent)) void __ockl_grid_sync(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_num_grids(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_grid_rank(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_size(void);
extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_thread_rank(void);
extern "C" __device__ __attribute__((const)) int __ockl_multi_grid_is_valid(void);
extern "C" __device__ __attribute__((convergent)) void __ockl_multi_grid_sync(void);
// Introduce local address space
#define __local __attribute__((address_space(3)))
include/hip/hcc_detail/functional_grid_launch.hpp
+4 -13
View File
@@ -129,7 +129,7 @@ void hipLaunchKernelGGLImpl(
template <typename F>
inline
void hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* blockSize,
hipError_t hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* blockSize,
F kernel, size_t dynSharedMemPerBlk, uint32_t blockSizeLimit) {
using namespace hip_impl;
@@ -138,13 +138,13 @@ void hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* blockSize,
auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
target_agent(0));
hipOccupancyMaxPotentialBlockSize(gridSize, blockSize, f,
return hipOccupancyMaxPotentialBlockSize(gridSize, blockSize, f,
dynSharedMemPerBlk, blockSizeLimit);
}
template <typename F>
inline
void hipOccupancyMaxActiveBlocksPerMultiprocessor(uint32_t* numBlocks, F kernel,
hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(uint32_t* numBlocks, F kernel,
uint32_t blockSize, size_t dynSharedMemPerBlk) {
using namespace hip_impl;
@@ -153,7 +153,7 @@ void hipOccupancyMaxActiveBlocksPerMultiprocessor(uint32_t* numBlocks, F kernel,
auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
target_agent(0));
hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, f, blockSize, dynSharedMemPerBlk);
return hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, f, blockSize, dynSharedMemPerBlk);
}
template <typename... Args, typename F = void (*)(Args...)>
@@ -177,13 +177,4 @@ void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
stream, &config[0]);
}
template <typename... Args, typename F = void (*)(hipLaunchParm, Args...)>
[[deprecated("hipLaunchKernel is deprecated and will be removed in the next "
"version of HIP; please upgrade to hipLaunchKernelGGL.")]]
inline void hipLaunchKernel(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
std::uint32_t groupMemBytes, hipStream_t stream, Args... args) {
hipLaunchKernelGGL(kernel, numBlocks, dimBlocks, groupMemBytes, stream, hipLaunchParm{},
std::move(args)...);
}
#pragma GCC visibility pop
include/hip/hcc_detail/hip_cooperative_groups.h
+216
View File
@@ -0,0 +1,216 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file hcc_detail/hip_cooperative_groups.h
*
* @brief Device side implementation of `Cooperative Group` feature.
*
* Defines new types and device API wrappers related to `Cooperative Group`
* feature, which the programmer can directly use in his kernel(s) in order to
* make use of this feature.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COOPERATIVE_GROUPS_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COOPERATIVE_GROUPS_H
#if __cplusplus
#include <hip/hcc_detail/hip_cooperative_groups_helper.h>
namespace cooperative_groups {
/** \brief The base type of all cooperative group types
*
* \details Holds the key properties of a constructed cooperative group type
* object, like the group type, its size, etc
*/
class thread_group {
protected:
uint32_t _type; // thread_group type
uint32_t _size; // total number of threads in the tread_group
uint64_t _mask; // Lanemask for coalesced and tiled partitioned group types,
// LSB represents lane 0, and MSB represents lane 63
// Construct a thread group, and set thread group type and other essential
// thread group properties. This generic thread group is directly constructed
// only when the group is supposed to contain only the calling the thread
// (throurh the API - `this_thread()`), and in all other cases, this thread
// group object is a sub-object of some other derived thread group object
__CG_QUALIFIER__ thread_group(internal::group_type type, uint32_t size,
uint64_t mask = (uint64_t)0) {
_type = type;
_size = size;
_mask = mask;
}
public:
// Total number of threads in the thread group, and this serves the purpose
// for all derived cooperative group types since their `size` is directly
// saved during the construction
__CG_QUALIFIER__ uint32_t size() const {
return _size;
}
// Rank of the calling thread within [0, size())
__CG_QUALIFIER__ uint32_t thread_rank() const;
// Is this cooperative group type valid?
__CG_QUALIFIER__ bool is_valid() const;
// synchronize the threads in the thread group
__CG_QUALIFIER__ void sync() const;
};
/** \brief The multi-grid cooperative group type
*
* \details Represents an inter-device cooperative group type where the
* participating threads within the group spans across multple
* devices, running the (same) kernel on these devices
*/
class multi_grid_group : public thread_group {
// Only these friend functions are allowed to construct an object of this class
// and access its resources
friend __CG_QUALIFIER__ multi_grid_group this_multi_grid();
protected:
// Construct mutli-grid thread group (through the API this_multi_grid())
explicit __CG_QUALIFIER__ multi_grid_group(uint32_t size)
: thread_group(internal::cg_multi_grid, size) { }
public:
// Number of invocations participating in this multi-grid group. In other
// words, the number of GPUs
__CG_QUALIFIER__ uint32_t num_grids() {
return internal::multi_grid::num_grids();
}
// Rank of this invocation. In other words, an ID number within the range
// [0, num_grids()) of the GPU, this kernel is running on
__CG_QUALIFIER__ uint32_t grid_rank() {
return internal::multi_grid::grid_rank();
}
__CG_QUALIFIER__ uint32_t thread_rank() const {
return internal::multi_grid::thread_rank();
}
__CG_QUALIFIER__ bool is_valid() const {
return internal::multi_grid::is_valid();
}
__CG_QUALIFIER__ void sync() const {
internal::multi_grid::sync();
}
};
/** \brief User exposed API interface to construct multi-grid cooperative
* group type object - `multi_grid_group`
*
* \details User is not allowed to directly construct an object of type
* `multi_grid_group`. Instead, he should construct it through this
* API function
*/
__CG_QUALIFIER__ multi_grid_group
this_multi_grid() {
return multi_grid_group(internal::multi_grid::size());
}
/** \brief The grid cooperative group type
*
* \details Represents an inter-workgroup cooperative group type where the
* participating threads within the group spans across multiple
* workgroups running the (same) kernel on the same device
*/
class grid_group : public thread_group {
// Only these friend functions are allowed to construct an object of this class
// and access its resources
friend __CG_QUALIFIER__ grid_group this_grid();
protected:
// Construct grid thread group (through the API this_grid())
explicit __CG_QUALIFIER__ grid_group(uint32_t size)
: thread_group(internal::cg_grid, size) { }
public:
__CG_QUALIFIER__ uint32_t thread_rank() const {
return internal::grid::thread_rank();
}
__CG_QUALIFIER__ bool is_valid() const {
return internal::grid::is_valid();
}
__CG_QUALIFIER__ void sync() const {
internal::grid::sync();
}
};
/** \brief User exposed API interface to construct grid cooperative group type
* object - `grid_group`
*
* \details User is not allowed to directly construct an object of type
* `multi_grid_group`. Instead, he should construct it through this
* API function
*/
__CG_QUALIFIER__ grid_group
this_grid() {
return grid_group(internal::grid::size());
}
/**
* Implemenation of all publicly exposed base class APIs
*/
__CG_QUALIFIER__ uint32_t thread_group::thread_rank() const {
switch (this->_type) {
case internal::cg_multi_grid: {
return (static_cast<const multi_grid_group*>(this)->thread_rank());
}
case internal::cg_grid: {
return (static_cast<const grid_group*>(this)->thread_rank());
}
default: {
return 0; //TODO(mahesha)
}
}
}
__CG_QUALIFIER__ bool thread_group::is_valid() const {
switch (this->_type) {
case internal::cg_multi_grid: {
return (static_cast<const multi_grid_group*>(this)->is_valid());
}
case internal::cg_grid: {
return (static_cast<const grid_group*>(this)->is_valid());
}
default: {
return false;
}
}
}
__CG_QUALIFIER__ void thread_group::sync() const {
switch (this->_type) {
case internal::cg_multi_grid: {
static_cast<const multi_grid_group*>(this)->sync();
break;
}
case internal::cg_grid: {
static_cast<const grid_group*>(this)->sync();
break;
}
}
}
} // namespace cooperative_groups
#endif // __cplusplus
#endif // HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COOPERATIVE_GROUPS_H
include/hip/hcc_detail/hip_cooperative_groups_helper.h
+144
View File
@@ -0,0 +1,144 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file hcc_detail/hip_cooperative_groups_helper.h
*
* @brief Device side implementation of cooperative group feature.
*
* Defines helper constructs and APIs which aid the types and device API
* wrappers defined within `hcc_detail/hip_cooperative_groups.h`.
*/
#ifndef HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COOPERATIVE_GROUPS_HELPER_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COOPERATIVE_GROUPS_HELPER_H
#if __cplusplus
#include <hip/hcc_detail/hip_runtime_api.h>
#include <hip/hcc_detail/device_functions.h>
#if !defined(__align__)
#define __align__(x) __attribute__((aligned(x)))
#endif
#if !defined(__CG_QUALIFIER__)
#define __CG_QUALIFIER__ __device__ __forceinline__
#endif
#if !defined(__CG_STATIC_QUALIFIER__)
#define __CG_STATIC_QUALIFIER__ __device__ static __forceinline__
#endif
#if !defined(WAVEFRONT_SIZE)
#define WAVEFRONT_SIZE 64
#endif
namespace cooperative_groups {
namespace internal {
/** \brief Enums representing different cooperative group types
*/
typedef enum {
cg_invalid,
cg_multi_grid,
cg_grid
} group_type;
/**
* Functionalities related to multi-grid cooperative group type
*/
namespace multi_grid {
__CG_STATIC_QUALIFIER__ uint32_t num_grids() {
return (uint32_t)__ockl_multi_grid_num_grids();
}
__CG_STATIC_QUALIFIER__ uint32_t grid_rank() {
return (uint32_t)__ockl_multi_grid_grid_rank();
}
__CG_STATIC_QUALIFIER__ uint32_t size() {
return (uint32_t)__ockl_multi_grid_size();
}
__CG_STATIC_QUALIFIER__ uint32_t thread_rank() {
return (uint32_t)__ockl_multi_grid_thread_rank();
}
__CG_STATIC_QUALIFIER__ bool is_valid() {
return (bool)__ockl_multi_grid_is_valid();
}
__CG_STATIC_QUALIFIER__ void sync() {
__ockl_multi_grid_sync();
}
} // namespace multi_grid
/**
* Functionalities related to grid cooperative group type
*/
namespace grid {
__CG_STATIC_QUALIFIER__ uint32_t size() {
return (uint32_t)((hipBlockDim_z * hipGridDim_z) *
(hipBlockDim_y * hipGridDim_y) *
(hipBlockDim_x * hipGridDim_x));
}
__CG_STATIC_QUALIFIER__ uint32_t thread_rank() {
// Compute global id of the workgroup to which the current threads belongs to
uint32_t blkIdx =
(uint32_t)((hipBlockIdx_z * hipGridDim_y * hipGridDim_x) +
(hipBlockIdx_y * hipGridDim_x) +
(hipBlockIdx_x));
// Compute total number of threads being passed to reach current workgroup
// within grid
uint32_t num_threads_till_current_workgroup =
(uint32_t)(blkIdx * (hipBlockIdx_x * hipBlockIdx_y * hipBlockIdx_z));
// Compute thread local rank within current workgroup
uint32_t local_thread_rank =
(uint32_t)((hipThreadIdx_z * hipBlockDim_y * hipBlockDim_x) +
(hipThreadIdx_y * hipBlockDim_x) +
(hipThreadIdx_x));
return (num_threads_till_current_workgroup + local_thread_rank);
}
__CG_STATIC_QUALIFIER__ bool is_valid() {
return (bool)__ockl_grid_is_valid();
}
__CG_STATIC_QUALIFIER__ void sync() {
__ockl_grid_sync();
}
} // namespace grid
} // namespace internal
} // namespace cooperative_groups
#endif // __cplusplus
#endif // HIP_INCLUDE_HIP_HCC_DETAIL_HIP_COOPERATIVE_GROUPS_HELPER_H
include/hip/hcc_detail/hip_runtime_api.h
+28
View File
@@ -1520,6 +1520,34 @@ hipError_t hipMemcpyToSymbol(void*, const void*, size_t, size_t, hipMemcpyKind,
} // Namespace hip_impl.
#endif
#ifdef __cplusplus
extern "C" {
#endif
/**
* @brief C compliant kernel launch API
*
* @param [in] function_address - kernel function pointer.
* @param [in] numBlocks - number of blocks
* @param [in] dimBlocks - dimension of a block
* @param [in] args - kernel arguments
* @param [in] sharedMemBytes - Amount of dynamic shared memory to allocate for this kernel. The
* Kernel can access this with HIP_DYNAMIC_SHARED.
* @param [in] stream - Stream where the kernel should be dispatched. May be 0, in which case th
* default stream is used with associated synchronization rules.
*
* @returns #hipSuccess, #hipErrorInvalidValue, hipInvalidDevice
*
*/
hipError_t hipLaunchKernel(const void* function_address,
dim3 numBlocks, dim3 dimBlocks, void** args,
size_t sharedMemBytes, hipStream_t stream);
#ifdef __cplusplus
}
#endif
#if defined(__cplusplus)
extern "C" {
#endif
include/hip/hip_cooperative_groups.h
+43
View File
@@ -0,0 +1,43 @@
/*
Copyright (c) 2015 - present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
/**
* @file hip_cooperative_groups.h
*
* @brief Defines new types and device API wrappers for `Cooperative Group`
* feature.
*/
#ifndef HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
#define HIP_INCLUDE_HIP_HIP_VECTOR_TYPES_H
#if defined(__HIP_PLATFORM_HCC__) && !defined(__HIP_PLATFORM_NVCC__)
#if __cplusplus
#include <hip/hcc_detail/hip_cooperative_groups.h>
#endif
#elif defined(__HIP_PLATFORM_NVCC__) && !defined(__HIP_PLATFORM_HCC__)
#include <cooperative_groups.h>
#else
#error("Must define exactly one of __HIP_PLATFORM_HCC__ or __HIP_PLATFORM_NVCC__");
#endif
#endif // HIP_INCLUDE_HIP_HIP_COOPERATIVE_GROUP_H
include/hip/nvcc_detail/hip_runtime.h
-5
View File
@@ -31,11 +31,6 @@ THE SOFTWARE.
typedef int hipLaunchParm;
#define hipLaunchKernel(kernelName, numblocks, numthreads, memperblock, streamId, ...) \
do { \
kernelName<<<numblocks, numthreads, memperblock, streamId>>>(0, ##__VA_ARGS__); \
} while (0)
#define hipLaunchKernelGGL(kernelName, numblocks, numthreads, memperblock, streamId, ...) \
do { \
kernelName<<<numblocks, numthreads, memperblock, streamId>>>(__VA_ARGS__); \
include/hip/nvcc_detail/hip_runtime_api.h
+69 -20
View File
@@ -216,38 +216,80 @@ inline static hipError_t hipCUDAErrorTohipError(cudaError_t cuError) {
switch (cuError) {
case cudaSuccess:
return hipSuccess;
case cudaErrorProfilerDisabled:
return hipErrorProfilerDisabled;
case cudaErrorProfilerNotInitialized:
return hipErrorProfilerNotInitialized;
case cudaErrorProfilerAlreadyStarted:
return hipErrorProfilerAlreadyStarted;
case cudaErrorProfilerAlreadyStopped:
return hipErrorProfilerAlreadyStopped;
case cudaErrorInsufficientDriver:
return hipErrorInsufficientDriver;
case cudaErrorUnsupportedLimit:
return hipErrorUnsupportedLimit;
case cudaErrorPeerAccessUnsupported:
return hipErrorPeerAccessUnsupported;
case cudaErrorInvalidGraphicsContext:
return hipErrorInvalidGraphicsContext;
case cudaErrorSharedObjectSymbolNotFound:
return hipErrorSharedObjectSymbolNotFound;
case cudaErrorSharedObjectInitFailed:
return hipErrorSharedObjectInitFailed;
case cudaErrorOperatingSystem:
return hipErrorOperatingSystem;
case cudaErrorSetOnActiveProcess:
return hipErrorSetOnActiveProcess;
case cudaErrorIllegalAddress:
return hipErrorIllegalAddress;
case cudaErrorInvalidSymbol:
return hipErrorInvalidSymbol;
case cudaErrorMissingConfiguration:
return hipErrorMissingConfiguration;
case cudaErrorMemoryAllocation:
return hipErrorMemoryAllocation;
case cudaErrorLaunchOutOfResources:
return hipErrorLaunchOutOfResources;
case cudaErrorInvalidValue:
return hipErrorInvalidValue;
case cudaErrorInvalidResourceHandle:
return hipErrorInvalidResourceHandle;
case cudaErrorInvalidDevice:
return hipErrorInvalidDevice;
case cudaErrorInvalidMemcpyDirection:
return hipErrorInvalidMemcpyDirection;
case cudaErrorInvalidDevicePointer:
return hipErrorInvalidDevicePointer;
case cudaErrorInitializationError:
return hipErrorInitializationError;
case cudaErrorNoDevice:
return hipErrorNoDevice;
case cudaErrorNotReady:
return hipErrorNotReady;
case cudaErrorLaunchFailure:
return hipErrorLaunchFailure;
case cudaErrorPriorLaunchFailure:
return hipErrorPriorLaunchFailure;
case cudaErrorLaunchOutOfResources:
return hipErrorLaunchOutOfResources;
case cudaErrorInvalidDeviceFunction:
return hipErrorInvalidDeviceFunction;
case cudaErrorInvalidConfiguration:
return hipErrorInvalidConfiguration;
case cudaErrorInvalidDevice:
return hipErrorInvalidDevice;
case cudaErrorInvalidValue:
return hipErrorInvalidValue;
case cudaErrorInvalidDevicePointer:
return hipErrorInvalidDevicePointer;
case cudaErrorInvalidMemcpyDirection:
return hipErrorInvalidMemcpyDirection;
case cudaErrorUnknown:
return hipErrorUnknown;
case cudaErrorPeerAccessNotEnabled:
return hipErrorPeerAccessNotEnabled;
case cudaErrorInvalidResourceHandle:
return hipErrorInvalidResourceHandle;
case cudaErrorNotReady:
return hipErrorNotReady;
case cudaErrorNoDevice:
return hipErrorNoDevice;
case cudaErrorPeerAccessAlreadyEnabled:
return hipErrorPeerAccessAlreadyEnabled;
case cudaErrorPeerAccessNotEnabled:
return hipErrorPeerAccessNotEnabled;
case cudaErrorHostMemoryAlreadyRegistered:
return hipErrorHostMemoryAlreadyRegistered;
case cudaErrorHostMemoryNotRegistered:
return hipErrorHostMemoryNotRegistered;
case cudaErrorUnsupportedLimit:
return hipErrorUnsupportedLimit;
case cudaErrorMapBufferObjectFailed:
return hipErrorMapBufferObjectFailed;
case cudaErrorAssert:
return hipErrorAssert;
case cudaErrorNotSupported:
return hipErrorNotSupported;
default:
return hipErrorUnknown; // Note - translated error.
}
@@ -1241,6 +1283,13 @@ inline static hipError_t hipModuleLoadDataEx(hipModule_t* module, const void* im
cuModuleLoadDataEx(module, image, numOptions, options, optionValues));
}
inline static hipError_t hipLaunchKernel(const void* function_address, dim3 numBlocks,
dim3 dimBlocks, void** args, size_t sharedMemBytes,
hipStream_t stream)
{
return hipCUDAErrorTohipError(cudaLaunchKernel(function_address,numBlocks,dimBlocks,args,sharedMemBytes,stream));
}
inline static hipError_t hipModuleLaunchKernel(hipFunction_t f, unsigned int gridDimX,
unsigned int gridDimY, unsigned int gridDimZ,
unsigned int blockDimX, unsigned int blockDimY,
samples/2_Cookbook/11_texture_driver/tex2dKernel.cpp
+6 -1
View File
@@ -21,7 +21,12 @@ THE SOFTWARE.
*/
#include "hip/hip_runtime.h"
extern texture<float, 2, hipReadModeElementType> tex;
#if __HIP__
__hip_pinned_shadow__
#else
extern
#endif
texture<float, 2, hipReadModeElementType> tex;
extern "C" __global__ void tex2dKernel(float* outputData, int width, int height) {
int x = hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x;
samples/2_Cookbook/13_occupancy/Makefile
+20
View File
@@ -0,0 +1,20 @@
HIP_PATH?= $(wildcard /opt/rocm/hip)
ifeq (,$(HIP_PATH))
HIP_PATH=../../..
endif
HIPCC=$(HIP_PATH)/bin/hipcc
EXE=./occupancy
.PHONY: test
all: test
$(EXE): occupancy.cpp
$(HIPCC) $^ -o $@
test: $(EXE)
$(EXE)
clean:
rm -f *.o $(EXE)
samples/2_Cookbook/13_occupancy/occupancy.cpp
+176
View File
@@ -0,0 +1,176 @@
/*
Copyright (c) 2015-Present Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANNTY OF ANY KIND, EXPRESS OR
IMPLIED, INNCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANNY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER INN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR INN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.
*/
#include "hip/hip_runtime.h"
#include <iostream>
#define NUM 1000000
#define HIP_CHECK(status) \
if (status != hipSuccess) { \
std::cout << "Got Status: " << status << " at Line: " << __LINE__ << std::endl; \
exit(0); \
}
// Device (Kernel) function
__global__ void multiply(float* C, float* A, float* B, int N){
int tx = blockDim.x*blockIdx.x+threadIdx.x;
if (tx < N){
C[tx] = A[tx] * B[tx];
}
}
// CPU implementation
void multiplyCPU(float* C, float* A, float* B, int N){
for(unsigned int i=0; i<N; i++){
C[i] = A[i] * B[i];
}
}
void launchKernel(float* C, float* A, float* B, bool manual){
hipDeviceProp_t devProp;
HIP_CHECK(hipGetDeviceProperties(&devProp, 0));
hipEvent_t start, stop;
HIP_CHECK(hipEventCreate(&start));
HIP_CHECK(hipEventCreate(&stop));
float eventMs = 1.0f;
const unsigned threadsperblock = 32;
const unsigned blocks = (NUM/threadsperblock)+1;
uint32_t mingridSize = 0;
uint32_t gridSize = 0;
uint32_t blockSize = 0;
if (manual){
blockSize = threadsperblock;
gridSize = blocks;
std::cout << std::endl << "Manual Configuration with block size " << blockSize << std::endl;
}
else{
HIP_CHECK(hipOccupancyMaxPotentialBlockSize(&mingridSize, &blockSize, multiply, 0, 0));
std::cout << std::endl << "Automatic Configuation based on hipOccupancyMaxPotentialBlockSize " << std::endl;
std::cout << "Suggested blocksize is " << blockSize << ", Minimum gridsize is " << mingridSize << std::endl;
gridSize = (NUM/blockSize)+1;
}
// Record the start event
HIP_CHECK(hipEventRecord(start, NULL));
// Launching the Kernel from Host
hipLaunchKernelGGL(multiply, dim3(gridSize), dim3(blockSize), 0, 0, C, A, B, NUM);
// Record the stop event
HIP_CHECK(hipEventRecord(stop, NULL));
HIP_CHECK(hipEventSynchronize(stop));
HIP_CHECK(hipEventElapsedTime(&eventMs, start, stop));
printf("kernel Execution time = %6.3fms\n", eventMs);
//Calculate Occupancy
uint32_t numBlock = 0;
HIP_CHECK(hipOccupancyMaxActiveBlocksPerMultiprocessor(&numBlock, multiply, blockSize, 0));
if(devProp.maxThreadsPerMultiProcessor){
std::cout << "Theoretical Occupancy is " << (double)numBlock* blockSize/devProp.maxThreadsPerMultiProcessor * 100 << "%" << std::endl;
}
}
int main() {
float *A, *B, *C0, *C1, *cpuC;
float *Ad, *Bd, *C0d, *C1d;
int errors=0;
int i;
// initialize the input data
A = (float *)malloc(NUM * sizeof(float));
B = (float *)malloc(NUM * sizeof(float));
C0 = (float *)malloc(NUM * sizeof(float));
C1 = (float *)malloc(NUM * sizeof(float));
cpuC = (float *)malloc(NUM * sizeof(float));
for(i=0; i< NUM; i++){
A[i] = i;
B[i] = i;
}
// allocate the memory on the device side
HIP_CHECK(hipMalloc((void**)&Ad, NUM * sizeof(float)));
HIP_CHECK(hipMalloc((void**)&Bd, NUM * sizeof(float)));
HIP_CHECK(hipMalloc((void**)&C0d, NUM * sizeof(float)));
HIP_CHECK(hipMalloc((void**)&C1d, NUM * sizeof(float)));
// Memory transfer from host to device
HIP_CHECK(hipMemcpy(Ad,A,NUM * sizeof(float), hipMemcpyHostToDevice));
HIP_CHECK(hipMemcpy(Bd,B,NUM * sizeof(float), hipMemcpyHostToDevice));
//Kernel launch with manual/default block size
launchKernel(C0d, Ad, Bd, 1);
//Kernel launch with the block size suggested by hipOccupancyMaxPotentialBlockSize
launchKernel(C1d, Ad, Bd, 0);
// Memory transfer from device to host
HIP_CHECK(hipMemcpy(C0,C0d, NUM * sizeof(float), hipMemcpyDeviceToHost));
HIP_CHECK(hipMemcpy(C1,C1d, NUM * sizeof(float), hipMemcpyDeviceToHost));
// CPU computation
multiplyCPU(cpuC, A, B, NUM);
//verify the results
double eps = 1.0E-6;
for (i = 0; i < NUM; i++) {
if (std::abs(C0[i] - cpuC[i]) > eps) {
errors++;
}
}
if (errors != 0){
printf("\nManual Test FAILED: %d errors\n", errors);
errors=0;
} else {
printf("\nManual Test PASSED!\n");
}
for (i = 0; i < NUM; i++) {
if (std::abs(C1[i] - cpuC[i]) > eps) {
errors++;
}
}
if (errors != 0){
printf("\n Automatic Test FAILED: %d errors\n", errors);
} else {
printf("\nAutomatic Test PASSED!\n");
}
HIP_CHECK(hipFree(Ad));
HIP_CHECK(hipFree(Bd));
HIP_CHECK(hipFree(C0d));
HIP_CHECK(hipFree(C1d));
free(A);
free(B);
free(C0);
free(C1);
free(cpuC);
}
src/hip_hcc.cpp
+10
View File
@@ -1671,6 +1671,8 @@ const char* ihipErrorString(hipError_t hip_error) {
return "hipErrorProfilerAlreadyStarted";
case hipErrorProfilerAlreadyStopped:
return "hipErrorProfilerAlreadyStopped";
case hipErrorInsufficientDriver:
return "hipErrorInsufficientDriver";
case hipErrorInvalidImage:
return "hipErrorInvalidImage";
case hipErrorInvalidContext:
@@ -1725,6 +1727,8 @@ const char* ihipErrorString(hipError_t hip_error) {
return "hipErrorNotFound";
case hipErrorIllegalAddress:
return "hipErrorIllegalAddress";
case hipErrorInvalidSymbol:
return "hipErrorInvalidSymbol";
case hipErrorMissingConfiguration:
return "hipErrorMissingConfiguration";
@@ -1773,6 +1777,12 @@ const char* ihipErrorString(hipError_t hip_error) {
return "hipErrorHostMemoryAlreadyRegistered";
case hipErrorHostMemoryNotRegistered:
return "hipErrorHostMemoryNotRegistered";
case hipErrorMapBufferObjectFailed:
return "hipErrorMapBufferObjectFailed";
case hipErrorAssert:
return "hipErrorAssert";
case hipErrorNotSupported:
return "hipErrorNotSupported";
case hipErrorTbd:
return "hipErrorTbd";
default:
src/hip_module.cpp
+28
View File
@@ -1238,3 +1238,31 @@ hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
return ihipLogStatus(ihipOccupancyMaxActiveBlocksPerMultiprocessor(
tls, numBlocks, f, blockSize, dynSharedMemPerBlk));
}
hipError_t hipLaunchKernel(
const void* func_addr, dim3 numBlocks, dim3 dimBlocks, void** args,
size_t sharedMemBytes, hipStream_t stream)
{
HIP_INIT_API(hipLaunchKernel,func_addr,numBlocks,dimBlocks,args,sharedMemBytes,stream);
hipFunction_t kd = hip_impl::get_program_state().kernel_descriptor((std::uintptr_t)func_addr,
hip_impl::target_agent(stream));
if(kd == nullptr || kd->_header == nullptr)
return ihipLogStatus(hipErrorInvalidValue);
size_t szKernArg = kd->_header->kernarg_segment_byte_size;
if(args == NULL && szKernArg != 0)
return ihipLogStatus(hipErrorInvalidValue);
void* config[]{
HIP_LAUNCH_PARAM_BUFFER_POINTER,
args,
HIP_LAUNCH_PARAM_BUFFER_SIZE,
&szKernArg,
HIP_LAUNCH_PARAM_END};
return ihipLogStatus(ihipModuleLaunchKernel(tls, kd, numBlocks.x * dimBlocks.x, numBlocks.y * dimBlocks.y, numBlocks.z * dimBlocks.z,
dimBlocks.x, dimBlocks.y, dimBlocks.z, sharedMemBytes, stream, nullptr, (void**)&config, nullptr, nullptr, 0));
}
tests/hipify-clang/lit.cfg
+4
View File
@@ -42,6 +42,10 @@ if config.cuda_version_major < 10:
config.excludes.append('cuSPARSE_10.cu')
config.excludes.append('cuSPARSE_11.cu')
if config.llvm_version_major < 10:
config.excludes.append('pp_if_else_conditionals_LLVM_10.cu')
config.excludes.append('pp_if_else_conditionals_01_LLVM_10.cu')
# name: The name of this test suite.
config.name = 'hipify'
tests/hipify-clang/lit.site.cfg.in
+1
View File
@@ -3,6 +3,7 @@ import os
config.pointer_size = @CMAKE_SIZEOF_VOID_P@
config.llvm_version = "@LLVM_PACKAGE_VERSION@"
config.llvm_version_major = int("@LLVM_VERSION_MAJOR@")
config.llvm_tools_dir = "@LLVM_TOOLS_BINARY_DIR@"
config.obj_root = "@CMAKE_CURRENT_BINARY_DIR@"
config.cuda_root = "@CUDA_TOOLKIT_ROOT_DIR@"
tests/hipify-clang/unit_tests/libraries/cuComplex/cuComplex_Julia.cu
+1 -1
View File
@@ -1,4 +1,4 @@
// RUN: %run_test hipify "%s" "%t" %hipify_args %clang_args
// RUN: %run_test hipify "%s" "%t" %hipify_args "--skip-excluded-preprocessor-conditional-blocks" %clang_args
// CHECK: #include <hip/hip_runtime.h>
// CHECK: #include "hip/hip_complex.h"
tests/hipify-clang/unit_tests/libraries/cuSPARSE/cuSPARSE_03.cu
+1 -1
View File
@@ -1,4 +1,4 @@
// RUN: %run_test hipify "%s" "%t" %hipify_args %clang_args
// RUN: %run_test hipify "%s" "%t" %hipify_args "--skip-excluded-preprocessor-conditional-blocks" %clang_args
#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
tests/hipify-clang/unit_tests/pp/pp_if_else_conditionals.cu
+30
View File
@@ -0,0 +1,30 @@
// RUN: %run_test hipify "%s" "%t" %hipify_args "--skip-excluded-preprocessor-conditional-blocks" %clang_args
// CHECK: #include <hip/hip_runtime.h>
#include <cuda.h>
__global__ void axpy_kernel(float a, float* x, float* y) {
y[threadIdx.x] = a * x[threadIdx.x];
}
void axpy(float a, float* x, float* y) {
#ifdef SOME_MACRO
// CHECK: axpy_kernel <<<1, 1>>> (a, y, x);
axpy_kernel <<<1, 1>>> (a, y, x);
#endif
#ifndef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(2), 0, 0, a, y, x);
axpy_kernel <<<1, 2>>> (a, y, x);
#endif
#ifdef SOME_MACRO
// CHECK: axpy_kernel <<<1, 3>>> (a, y, x);
axpy_kernel <<<1, 3>>> (a, y, x);
#else
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(4), 0, 0, a, x, y);
axpy_kernel <<<1, 4>>> (a, x, y);
#endif
}
tests/hipify-clang/unit_tests/pp/pp_if_else_conditionals_01.cu
+52
View File
@@ -0,0 +1,52 @@
// RUN: %run_test hipify "%s" "%t" %hipify_args "--skip-excluded-preprocessor-conditional-blocks" %clang_args
// CHECK: #include <hip/hip_runtime.h>
__global__ void axpy_kernel(float a, float* x, float* y) {
y[threadIdx.x] = a * x[threadIdx.x];
}
void axpy(float a, float* x, float* y) {
float* y_new = nullptr;
#ifdef SOME_MACRO
y_new = x;
// CHECK: axpy_kernel <<<1, 1>>> (a, y_new, x);
axpy_kernel <<<1, 1>>> (a, y_new, x);
#endif
#ifndef SOME_MACRO
y_new = y;
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(2), 0, 0, a, y_new, x);
axpy_kernel <<<1, 2>>> (a, y_new, x);
#endif
#ifdef SOME_MACRO
// CHECK: axpy_kernel <<<1, 3>>> (a, y, x);
axpy_kernel <<<1, 3>>> (a, y, x);
#else
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(4), 0, 0, a, x, y);
axpy_kernel <<<1, 4>>> (a, x, y);
#endif
#ifdef SOME_MACRO
// CHECK: axpy_kernel <<<1, 5>>> (a, y, x);
axpy_kernel <<<1, 5>>> (a, y, x);
#elif defined SOME_MACRO_1
// CHECK: axpy_kernel <<<1, 6>>> (a, x, y);
axpy_kernel <<<1, 6>>> (a, x, y);
#else
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(7), 0, 0, a, x, y);
axpy_kernel <<<1, 7>>> (a, x, y);
#endif
#ifndef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(8), 0, 0, a, y, x);
axpy_kernel <<<1, 8>>> (a, y, x);
#elif !defined(SOME_MACRO_1)
// CHECK: axpy_kernel <<<1, 9>>> (a, x, y);
axpy_kernel <<<1, 9>>> (a, x, y);
#else
// CHECK: axpy_kernel <<<1, 10>>> (a, x, y);
axpy_kernel <<<1, 10>>> (a, x, y);
#endif
}
tests/hipify-clang/unit_tests/pp/pp_if_else_conditionals_01_LLVM_10.cu
+52
View File
@@ -0,0 +1,52 @@
// RUN: %run_test hipify "%s" "%t" %hipify_args %clang_args
// CHECK: #include <hip/hip_runtime.h>
__global__ void axpy_kernel(float a, float* x, float* y) {
y[threadIdx.x] = a * x[threadIdx.x];
}
void axpy(float a, float* x, float* y) {
float* y_new = nullptr;
#ifdef SOME_MACRO
y_new = x;
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(1), 0, 0, a, y_new, x);
axpy_kernel <<<1, 1>>> (a, y_new, x);
#endif
#ifndef SOME_MACRO
y_new = y;
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(2), 0, 0, a, y_new, x);
axpy_kernel <<<1, 2>>> (a, y_new, x);
#endif
#ifdef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(3), 0, 0, a, y, x);
axpy_kernel <<<1, 3>>> (a, y, x);
#else
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(4), 0, 0, a, x, y);
axpy_kernel <<<1, 4>>> (a, x, y);
#endif
#ifdef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(5), 0, 0, a, y, x);
axpy_kernel <<<1, 5>>> (a, y, x);
#elif defined SOME_MACRO_1
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(6), 0, 0, a, x, y);
axpy_kernel <<<1, 6>>> (a, x, y);
#else
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(7), 0, 0, a, x, y);
axpy_kernel <<<1, 7>>> (a, x, y);
#endif
#ifndef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(8), 0, 0, a, y, x);
axpy_kernel <<<1, 8>>> (a, y, x);
#elif !defined(SOME_MACRO_1)
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(9), 0, 0, a, x, y);
axpy_kernel <<<1, 9>>> (a, x, y);
#else
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(10), 0, 0, a, x, y);
axpy_kernel <<<1, 10>>> (a, x, y);
#endif
}
tests/hipify-clang/unit_tests/pp/pp_if_else_conditionals_LLVM_10.cu
+30
View File
@@ -0,0 +1,30 @@
// RUN: %run_test hipify "%s" "%t" %hipify_args %clang_args
// CHECK: #include <hip/hip_runtime.h>
#include <cuda.h>
__global__ void axpy_kernel(float a, float* x, float* y) {
y[threadIdx.x] = a * x[threadIdx.x];
}
void axpy(float a, float* x, float* y) {
#ifdef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(1), 0, 0, a, y, x);
axpy_kernel <<<1, 1>>> (a, y, x);
#endif
#ifndef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(2), 0, 0, a, y, x);
axpy_kernel <<<1, 2>>> (a, y, x);
#endif
#ifdef SOME_MACRO
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(3), 0, 0, a, y, x);
axpy_kernel <<<1, 3>>> (a, y, x);
#else
// CHECK: hipLaunchKernelGGL(axpy_kernel, dim3(1), dim3(4), 0, 0, a, x, y);
axpy_kernel <<<1, 4>>> (a, x, y);
#endif
}
tests/src/gcc/LaunchKernel.c
+165
View File
@@ -0,0 +1,165 @@
/* Copyright (c) 2019-Present Advanced Micro Devices, Inc. All rights reserved.
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANNTY OF ANY KIND, EXPRESS OR
* IMPLIED, INNCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANNY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER INN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR INN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
/* HIT_START
* BUILD_CMD: gpu.o %hc -I%hip-path/include -g -c %S/gpu.cpp -o %T/gpu.o EXCLUDE_HIP_PLATFORM nvcc
* BUILD_CMD: launchkernel.o %cc -D__HIP_PLATFORM_HCC__ -g -I%hip-path/include -c %S/LaunchKernel.c -o %T/launchkernel.o EXCLUDE_HIP_PLATFORM nvcc
* BUILD_CMD: LaunchKernel %hc %T/launchkernel.o %T/gpu.o -g -Wl,--rpath=%hip-path/lib %hip-path/lib/libhip_hcc.so -o %T/%t DEPENDS gpu.o launchkernel.o EXCLUDE_HIP_PLATFORM nvcc
* TEST: %t EXCLUDE_HIP_PLATFORM nvcc
* HIT_END
*/
#include "hip/hip_runtime.h"
#include <stdio.h>
#include "LaunchKernel.h"
#define HIPCHECK(error) \
{ \
hipError_t localError = error; \
if ((localError != hipSuccess) && (localError != hipErrorPeerAccessAlreadyEnabled)) { \
printf("%serror: '%s'(%d) from %s at %s:%d%s\n", "\x1B[31m", hipGetErrorString(localError), \
localError, #error, __FILE__, __LINE__, "\x1B[0m"); \
return false; \
} \
}
bool LaunchKernelArg()
{
dim3 blocks = {1,1,1};
dim3 threads = {1,1,1};
HIPCHECK(hipLaunchKernel(kernel, blocks, threads,NULL, 0, 0));
return true;
}
bool LaunchKernelArg1()
{
int A = 0;
int *A_d = NULL;
dim3 blocks = {1,1,1};
dim3 threads = {1,1,1};
// Allocate Device memory
HIPCHECK(hipMalloc((void**)&A_d, sizeof(int)));
void* Args[]={A_d};
HIPCHECK(hipLaunchKernel(kernel1, blocks, threads, Args,0,0));
// Get the result back to host memory
HIPCHECK(hipMemcpy(&A, A_d, sizeof(int), hipMemcpyDeviceToHost));
HIPCHECK(hipFree(A_d));
if(A != 333)
return false;
return true;
}
bool LaunchKernelArg2()
{
int A = 0;
int B = 123;
int *A_d = NULL;
int *B_d = NULL;
dim3 blocks = {1,1,1};
dim3 threads = {1,1,1};
// Allocate Device memory
HIPCHECK(hipMalloc((void**)&A_d, sizeof(int)));
HIPCHECK(hipMalloc((void**)&B_d, sizeof(int)));
// Copy data from host memory to device memory
HIPCHECK(hipMemcpy(B_d,&B, sizeof(int), hipMemcpyHostToDevice));
void* Args[]={A_d,B_d};
HIPCHECK(hipLaunchKernel(kernel2, blocks, threads, Args,0,0));
// Get the result back to host memory
HIPCHECK(hipMemcpy(&A, A_d, sizeof(int), hipMemcpyDeviceToHost));
HIPCHECK(hipFree(A_d));
HIPCHECK(hipFree(B_d));
if(A != 123)
return false;
return true;
}
bool LaunchKernelArg3()
{
int A = 321;
int B = 123;
int C = 0;
int *A_d = NULL;
int *B_d = NULL;
int *C_d = NULL;
dim3 blocks = {1,1,1};
dim3 threads = {1,1,1};
// Allocate Device memory
HIPCHECK(hipMalloc((void**)&A_d, sizeof(int)));
HIPCHECK(hipMalloc((void**)&B_d, sizeof(int)));
HIPCHECK(hipMalloc((void**)&C_d, sizeof(int)));
// Copy data from host memory to device memory
HIPCHECK(hipMemcpy(A_d,&A, sizeof(int), hipMemcpyHostToDevice));
HIPCHECK(hipMemcpy(B_d,&B, sizeof(int), hipMemcpyHostToDevice));
void* Args[]={A_d,B_d,C_d};
HIPCHECK(hipLaunchKernel(kernel3, blocks, threads, Args,0,0));
// Get the result back to host memory
HIPCHECK(hipMemcpy(&C, C_d, sizeof(int), hipMemcpyDeviceToHost));
HIPCHECK(hipFree(A_d));
HIPCHECK(hipFree(B_d));
HIPCHECK(hipFree(C_d));
if(C != 444)
return false;
return true;
}
int main()
{
if( LaunchKernelArg() &&
LaunchKernelArg1() &&
LaunchKernelArg2() &&
LaunchKernelArg3())
{
printf("PASSED!\n");
}
else
printf("FAILED\n");
}
tests/src/gcc/LaunchKernel.h
+23
View File
@@ -0,0 +1,23 @@
/* Copyright (c) 2019-Present Advanced Micro Devices, Inc. All rights reserved.
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANNTY OF ANY KIND, EXPRESS OR
* IMPLIED, INNCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANNY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER INN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR INN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
extern __global__ void kernel();
extern __global__ void kernel1(int*);
extern __global__ void kernel2(int*,int*);
extern __global__ void kernel3(int*,int*,int*);
tests/src/gcc/gpu.cpp
+46
View File
@@ -0,0 +1,46 @@
/*
* Copyright (c) 2019-Present Advanced Micro Devices, Inc. All rights reserved.
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANNTY OF ANY KIND, EXPRESS OR
* IMPLIED, INNCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANNY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER INN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR INN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
* */
#include<hip/hip_runtime.h>
extern "C"
{
__global__ void kernel()
{
int a = 20;
}
__global__ void kernel1(int *a)
{
*a = 333;
}
__global__ void kernel2(int *a, int*b)
{
*a = *b;
}
__global__ void kernel3(int *a, int*b, int* c)
{
*c = *a+*b;
}
}//extern "C"