hipify-clang#
hipify-clang is a Clang-based tool for translating NVIDIA CUDA sources into HIP sources.
It translates CUDA source into an Abstract Syntax Tree (AST), which is traversed by transformation matchers. After applying all the matchers, the output HIP source is produced.
Advantages:
hipify-clangis a translator. It parses complex constructs successfully or else reports an error.It supports Clang options such as -I, -D, and –cuda-path.
The support for new CUDA versions is seamless, as the Clang front-end is statically linked into
hipify-clangand does all the syntactical parsing of a CUDA source to HIPIFY.It is very well supported as a compiler extension.
Disadvantages:
You must ensure that the input CUDA code is correct as incorrect code can’t be translated to HIP.
You must install CUDA and in case of multiple installations, specify using
--cuda-pathoption.You must provide all the
includesanddefinesto successfully translate the code.
Dependencies#
hipify-clang requires:
LLVM+Clang of at least version 4.0.0; the latest stable and recommended release: 19.1.6.
CUDA of at least version 7.0, the latest supported version is 12.6.3.
LLVM release version |
Latest supported CUDA version |
Windows |
Linux |
✅ |
✅ |
||
✅ |
✅ |
||
✅ |
✅ |
||
Works only with patch due to Clang bug 38811
|
❌ due to Clang bug 36384 |
||
Works only with patch due to Clang bug 38811
|
✅ |
||
✅ |
✅ |
||
✅ |
✅ |
||
Works only with patch due to Clang bug 47332
|
Works only with patch due to Clang bug 47332
|
||
✅ |
✅ |
||
Works only with patch due to Clang bug 47332
|
Works only with patch due to Clang bug 47332
|
||
✅ |
✅ |
||
✅ |
✅ |
||
Works only with patch due to Clang bug 54609
|
✅ |
||
14.0.5, 14.0.6, 15.0.0, 15.0.1, 15.0.2, 15.0.3, 15.0.4, 15.0.5, 15.0.6, 15.0.7 |
✅ |
✅ |
|
✅ |
✅ |
||
17.0.1, 17.0.2, 17.0.3, 17.0.4, 17.0.5, 17.0.6, 18.1.0, 18.1.1, 18.1.2, 18.1.3, 18.1.4, 18.1.5, 18.1.6, 18.1.7, 18.1.8 |
✅ |
✅ |
|
Latest stable config |
Latest stable config |
1 LLVM 3.x is no longer supported (but might still work).
2 Download the patch and unpack it into your LLVM distributive directory. This overwrites a few header files. You don’t need to rebuild LLVM.
3 Download the patch and unpack it into your LLVM source directory. This overwrites the Cuda.cpp file. You need to rebuild LLVM.
4 Represents the latest supported and recommended configuration.
In most cases, you can get a suitable version of LLVM+Clang with your package manager. However, you can also
download a release archive and build or install it. In case of multiple versions of LLVM installed, set
CMAKE_PREFIX_PATH so that
CMake can find the desired version of LLVM. For example, -DCMAKE_PREFIX_PATH=D:\LLVM\19.1.6\dist.
Usage#
To process a file, hipify-clang needs access to the same headers that are required to compile it
with Clang:
./hipify-clang square.cu --cuda-path=/usr/local/cuda-12.6 -I /usr/local/cuda-12.6/samples/common/inc
hipify-clang arguments are supplied first, followed by a separator -- and the arguments to be
passed to Clang for compiling the input file:
./hipify-clang cpp17.cu --cuda-path=/usr/local/cuda-12.6 -- -std=c++17
hipify-clang also supports the hipification of multiple files that can be specified in a single
command with absolute or relative paths:
./hipify-clang cpp17.cu ../../square.cu /home/user/cuda/intro.cu --cuda-path=/usr/local/cuda-12.6 -- -std=c++17
To use a specific version of LLVM during hipification, specify the hipify-clang option
--clang-resource-directory= to point to the Clang resource directory, which is the
parent directory for the include folder that contains __clang_cuda_runtime_wrapper.h and other
header files used during the hipification process:
./hipify-clang square.cu --cuda-path=/usr/local/cuda-12.6 --clang-resource-directory=/usr/llvm/19.1.6/dist/lib/clang/19
For more information, refer to the Clang manual for compiling CUDA.
Using JSON compilation database#
For some hipification automation (starting from Clang 8.0.0), you can also provide a
Compilation Database in JSON format
in the compile_commands.json file:
-p <folder containing compile_commands.json> or
-p=<folder containing compile_commands.json>
You can provide the compilation database in the compile_commands.json file or generate using
Clang based on CMake. You can specify multiple source files as well.
To provide Clang options, use compile_commands.json file, whereas to provide hipify-clang options, use hipify-clang command line.
Note
Don’t use the options separator -- to avoid compilation error caused due to the hipify-clang options being
provided before the separator.
Here’s an
example
demonstrating the compile_commands.json usage:
[
{
"directory": "<test dir>",
"command": "hipify-clang \"<CUDA dir>\" -I./include -v",
"file": "cd_intro.cu"
}
]
Hipification statistics#
The options --print-stats and --print-stats-csv provide an overview of what is hipified and
what is not, and the hipification statistics:
hipify-clang intro.cu -cuda-path="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6" --print-stats
[HIPIFY] info: file 'intro.cu' statistics:
CONVERTED refs count: 40
UNCONVERTED refs count: 0
CONVERSION %: 100.0
REPLACED bytes: 604
[HIPIFY] info: file 'intro.cu' statistics:
CONVERTED refs count: 40
UNCONVERTED refs count: 0
CONVERSION %: 100.0
REPLACED bytes: 604
TOTAL bytes: 5794
CHANGED lines of code: 34
TOTAL lines of code: 174
CODE CHANGED (in bytes) %: 10.4
CODE CHANGED (in lines) %: 19.5
TIME ELAPSED s: 0.41
[HIPIFY] info: CONVERTED refs by type:
error: 2
device: 2
memory: 16
event: 9
thread: 1
include_cuda_main_header: 1
type: 2
numeric_literal: 7
[HIPIFY] info: CONVERTED refs by API:
CUDA Driver API: 1
CUDA RT API: 39
[HIPIFY] info: CONVERTED refs by names:
cuda.h: 1
cudaDeviceReset: 1
cudaError_t: 1
cudaEventCreate: 2
cudaEventElapsedTime: 1
cudaEventRecord: 3
cudaEventSynchronize: 3
cudaEvent_t: 1
cudaFree: 4
cudaFreeHost: 3
cudaGetDeviceCount: 1
cudaGetErrorString: 1
cudaGetLastError: 1
cudaMalloc: 3
cudaMemcpy: 6
cudaMemcpyDeviceToHost: 3
cudaMemcpyHostToDevice: 3
cudaSuccess: 1
cudaThreadSynchronize: 1
hipify-clang intro.cu -cuda-path="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6" --print-stats-csv
This generates intro.cu.csv file with statistics:
In case of multiple source files, the statistics are provided per file and in total.
For a list of hipify-clang options, run hipify-clang --help.
Building hipify-clang#
After cloning the HIPIFY repository (git clone https://github.com/ROCm/HIPIFY.git), run the following commands from the HIPIFY root folder.
cd .. \
mkdir build dist \
cd build
cmake \
-DCMAKE_INSTALL_PREFIX=../dist \
-DCMAKE_BUILD_TYPE=Release \
../hipify
make -j install
To ensure LLVM being found or in case of multiple LLVM instances, specify the path to the root folder containing the LLVM distributive:
-DCMAKE_PREFIX_PATH=/usr/llvm/19.1.6/dist
On Windows, specify the following option for CMake in the first place:
-G "Visual Studio 17 2022".
Build the generated hipify-clang.sln using
Visual Studio 17 2022 instead of Make. See Windows testing for the
supported tools for building.
As debug build type -DCMAKE_BUILD_TYPE=Debug is supported and tested, it is recommended to build LLVM+Clang
in debug mode.
Also, 64-bit build mode (-Thost=x64 on Windows) is supported, hence it is recommended to build LLVM+Clang in
64-bit mode.
You can find the binary at ./dist/hipify-clang or at the folder specified by the
-DCMAKE_INSTALL_PREFIX option.
Testing hipify-clang#
hipify-clang is equipped with unit tests using LLVM
lit or FileCheck.
Build LLVM+Clang from sources, as prebuilt binaries are not exhaustive for testing. Before
building, ensure that the
software required for building
belongs to an appropriate version.
LLVM <= 9.0.1#
Build LLVM+Clang:
cd .. \ mkdir build dist \ cd build
Linux:
cmake \ -DCMAKE_INSTALL_PREFIX=../dist \ -DLLVM_SOURCE_DIR=../llvm \ -DLLVM_TARGETS_TO_BUILD="X86" \ -DLLVM_INCLUDE_TESTS=OFF \ -DCMAKE_BUILD_TYPE=Release \ ../llvm make -j install
Windows:
cmake \ -G "Visual Studio 16 2019" \ -A x64 \ -Thost=x64 \ -DCMAKE_INSTALL_PREFIX=../dist \ -DLLVM_SOURCE_DIR=../llvm \ -DLLVM_TARGETS_TO_BUILD="" \ -DLLVM_INCLUDE_TESTS=OFF \ -DCMAKE_BUILD_TYPE=Release \ ../llvm
Run
Visual Studio 16 2019, open the generatedLLVM.sln, build all, and build theINSTALLproject.
LLVM >= 10.0.0#
Download LLVM project sources.
Build LLVM project:
cd .. \ mkdir build dist \ cd build
Linux:
cmake \ -DCMAKE_INSTALL_PREFIX=../dist \ -DLLVM_TARGETS_TO_BUILD="X86" \ -DLLVM_ENABLE_PROJECTS="clang" \ -DLLVM_INCLUDE_TESTS=OFF \ -DCMAKE_BUILD_TYPE=Release \ ../llvm-project/llvm make -j install
Windows:
cmake \ -G "Visual Studio 17 2022" \ -A x64 \ -Thost=x64 \ -DCMAKE_INSTALL_PREFIX=../dist \ -DLLVM_TARGETS_TO_BUILD="" \ -DLLVM_ENABLE_PROJECTS="clang" \ -DLLVM_INCLUDE_TESTS=OFF \ -DCMAKE_BUILD_TYPE=Release \ ../llvm-project/llvm
Run
Visual Studio 17 2022, open the generatedLLVM.sln, build all, and build projectINSTALL.Install CUDA version 7.0 or greater.
In case of multiple CUDA installations, specify the particular version using
DCUDA_TOOLKIT_ROOT_DIRoption:Linux:
-DCUDA_TOOLKIT_ROOT_DIR=/usr/includeWindows:
-DCUDA_TOOLKIT_ROOT_DIR="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6" -DCUDA_SDK_ROOT_DIR="C:/ProgramData/NVIDIA Corporation/CUDA Samples/v12.6"
[Optional] Install cuTensor:
To specify the path to cuTensor, use the
CUDA_TENSOR_ROOT_DIRoption:Linux:
-DCUDA_TENSOR_ROOT_DIR=/usr/includeWindows:
-DCUDA_TENSOR_ROOT_DIR=D:/CUDA/cuTensor/2.0.2.1
[Optional] Install cuDNN belonging to the version corresponding to the CUDA version:
To specify the path to cuDNN, use the
CUDA_DNN_ROOT_DIRoption:Linux:
-DCUDA_DNN_ROOT_DIR=/usr/includeWindows:
-DCUDA_DNN_ROOT_DIR=D:/CUDA/cuDNN/9.6.0
[Optional] Install CUB 1.9.8 for
CUDA < 11.0only; forCUDA >= 11.0, the CUB shipped with CUDA will be used for testing.To specify the path to CUB, use the
CUDA_CUB_ROOT_DIRoption (only forCUDA < 11.0):Linux:
-DCUDA_CUB_ROOT_DIR=/srv/git/CUBWindows:
-DCUDA_CUB_ROOT_DIR=D:/CUDA/CUB
Install Python version 3.0 or greater.
Install
litandFileCheck; these are distributed with LLVM.Install
litintoPython:Linux:
python /usr/llvm/19.1.6/llvm-project/llvm/utils/lit/setup.py install
Windows:
python D:/LLVM/19.1.6/llvm-project/llvm/utils/lit/setup.py install
In case of errors similar to
ModuleNotFoundError: No module named 'setuptools', upgrade thesetuptoolspackage:python -m pip install --upgrade pip setuptools
Starting with LLVM 6.0.1, specify the path to the
llvm-litPython script using theLLVM_EXTERNAL_LIToption:Linux:
-DLLVM_EXTERNAL_LIT=/usr/llvm/19.1.6/build/bin/llvm-litWindows:
-DLLVM_EXTERNAL_LIT=D:/LLVM/19.1.6/build/Release/bin/llvm-lit.pyFileCheck:Linux:
Copy from
/usr/llvm/19.1.6/build/bin/toCMAKE_INSTALL_PREFIX/dist/bin.Windows:
Copy from
D:/LLVM/19.1.6/build/Release/bintoCMAKE_INSTALL_PREFIX/dist/bin.Alternatively, specify the path to
FileCheckin theCMAKE_INSTALL_PREFIXoption.
To run OpenGL tests successfully on:
Linux:
Install GL headers.
On Ubuntu, use:
sudo apt-get install mesa-common-devWindows:
No installation required. All the required headers are shipped with the Windows SDK.
Set the
HIPIFY_CLANG_TESTSoption toON:-DHIPIFY_CLANG_TESTS=ONBuild and run tests.
Linux testing#
On Linux, the following configurations are tested:
Ubuntu 14: LLVM 4.0.0 - 7.1.0, CUDA 7.0 - 9.0, cuDNN 5.0.5 - 7.6.5
Ubuntu 16-19: LLVM 8.0.0 - 14.0.6, CUDA 7.0 - 10.2, cuDNN 5.1.10 - 8.0.5
Ubuntu 20-21: LLVM 9.0.0 - 19.1.6, CUDA 7.0 - 12.6.3, cuDNN 5.1.10 - 9.6.0, cuTensor 1.0.1.0 - 2.0.2.1
Ubuntu 22-23: LLVM 13.0.0 - 19.1.6, CUDA 7.0 - 12.6.3, cuDNN 8.0.5 - 9.6.0, cuTensor 1.0.1.0 - 2.0.2.1
Minimum build system requirements for the above configurations:
CMake 3.16.8, GNU C/C++ 9.2, Python 3.0.
Recommended build system requirements:
CMake 3.31.2, GNU C/C++ 13.2, Python 3.13.1.
Here’s how to build hipify-clang with testing support on Ubuntu 23.10.01:
cmake
-DHIPIFY_CLANG_TESTS=ON \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_INSTALL_PREFIX=../dist \
-DCMAKE_PREFIX_PATH=/usr/llvm/19.1.6/dist \
-DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-12.6.3 \
-DCUDA_DNN_ROOT_DIR=/usr/local/cudnn-9.6.0 \
-DCUDA_TENSOR_ROOT_DIR=/usr/local/cutensor-2.0.2.1 \
-DLLVM_EXTERNAL_LIT=/usr/llvm/19.1.6/build/bin/llvm-lit \
../hipify
The corresponding successful output is:
-- The C compiler identification is GNU 13.2.0
-- The CXX compiler identification is GNU 13.2.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- HIPIFY config:
-- - Build hipify-clang : ON
-- - Test hipify-clang : ON
-- - Is part of HIP SDK : OFF
-- - Install clang headers : ON
-- Found ZLIB: /usr/lib/x86_64-linux-gnu/libz.so (found version "1.2.13")
-- Found LLVM 19.1.6:
-- - CMake module path : /usr/llvm/19.1.6/dist/lib/cmake/llvm
-- - Clang include path : /usr/llvm/19.1.6/dist/include
-- - LLVM Include path : /usr/llvm/19.1.6/dist/include
-- - Binary path : /usr/llvm/19.1.6/dist/bin
-- Linker detection: GNU ld
-- ---- The below configuring for hipify-clang testing only ----
-- Found Python: /usr/bin/python3.13 (found suitable version "3.13.1", required range is "3.0...3.14") found components: Interpreter
-- Found lit: /usr/local/bin/lit
-- Found FileCheck: /GIT/LLVM/trunk/dist/FileCheck
-- Initial CUDA to configure:
-- - CUDA Toolkit path : /usr/local/cuda-12.6.3
-- - CUDA Samples path :
-- - cuDNN path : /usr/local/cudnn-9.6.0
-- - cuTENSOR path : /usr/local/cuTensor/2.0.2.1
-- - CUB path :
-- Found CUDAToolkit: /usr/local/cuda-12.6.3/targets/x86_64-linux/include (found version "12.6.85")
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Success
-- Found Threads: TRUE
-- Found CUDA config:
-- - CUDA Toolkit path : /usr/local/cuda-12.6.3
-- - CUDA Samples path : OFF
-- - cuDNN path : /usr/local/cudnn-9.6.0
-- - CUB path : /usr/local/cuda-12.6.3/include/cub
-- - cuTENSOR path : /usr/local/cuTensor/2.0.2.1
-- Configuring done (0.6s)
-- Generating done (0.0s)
-- Build files have been written to: /usr/hipify/build
make test-hipify
The corresponding successful output is:
Running HIPify regression tests
===============================================================
CUDA 12.6.85 - will be used for testing
LLVM 19.1.6 - will be used for testing
x86_64 - Platform architecture
Linux 6.5.0-15-generic - Platform OS
64 - hipify-clang binary bitness
64 - python 3.13.1 binary bitness
===============================================================
-- Testing: 106 tests, 12 threads --
Testing Time: 6.91s
Total Discovered Tests: 106
Passed: 106 (100.00%)
Windows testing#
Tested configurations:
LLVM |
CUDA |
cuDNN |
Visual Studio |
CMake |
Python |
|---|---|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 LLVM 14.x.x is the latest major release supporting Visual Studio 2017.
To build LLVM 14.x.x correctly using Visual Studio 2017, add -DLLVM_FORCE_USE_OLD_TOOLCHAIN=ON
to corresponding CMake command line.
You can also build LLVM < 14.x.x correctly using Visual Studio 2017 without the
LLVM_FORCE_USE_OLD_TOOLCHAIN option.
6 Note that LLVM 17.0.0 was withdrawn due to an issue; use 17.0.1 or newer instead.
7 Note that LLVM 18.0.0 has never been released; use 18.1.0 or newer instead.
Building with testing support using Visual Studio 17 2022 on Windows 11:
cmake
-G "Visual Studio 17 2022" \
-A x64 \
-Thost=x64 \
-DHIPIFY_CLANG_TESTS=ON \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_INSTALL_PREFIX=../dist \
-DCMAKE_PREFIX_PATH=D:/LLVM/19.1.6/dist \
-DCUDA_TOOLKIT_ROOT_DIR="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6" \
-DCUDA_SDK_ROOT_DIR="C:/ProgramData/NVIDIA Corporation/CUDA Samples/v12.5" \
-DCUDA_DNN_ROOT_DIR=D:/CUDA/cuDNN/9.6.0 \
-DCUDA_TENSOR_ROOT_DIR=D:/CUDA/cuTensor/2.0.2.1 \
-DLLVM_EXTERNAL_LIT=D:/LLVM/19.1.6/build/Release/bin/llvm-lit.py \
../hipify
The corresponding successful output is:
-- Selecting Windows SDK version 10.0.22621.0 to target Windows 10.0.22631.
-- The C compiler identification is MSVC 19.42.34435.0
-- The CXX compiler identification is MSVC 19.42.34435.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.42.34433/bin/Hostx64/x64/cl.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.42.34433/bin/Hostx64/x64/cl.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- HIPIFY config:
-- - Build hipify-clang : ON
-- - Test hipify-clang : ON
-- - Is part of HIP SDK : OFF
-- - Install clang headers : ON
-- Found LLVM 19.1.6:
-- - CMake module path : D:/LLVM/19.1.6/dist/lib/cmake/llvm
-- - Clang include path : D:/LLVM/19.1.6/dist/include
-- - LLVM Include path : D:/LLVM/19.1.6/dist/include
-- - Binary path : D:/LLVM/19.1.6/dist/bin
-- ---- The below configuring for hipify-clang testing only ----
-- Found Python: C:/Users/TT/AppData/Local/Programs/Python/Python313/python.exe (found suitable version "3.13.1", required range is "3.0...3.14") found components: Interpreter
-- Found lit: C:/Users/TT/AppData/Local/Programs/Python/Python313/Scripts/lit.exe
-- Found FileCheck: D:/LLVM/19.1.6/dist/bin/FileCheck.exe
-- Initial CUDA to configure:
-- - CUDA Toolkit path : C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6
-- - CUDA Samples path : C:/ProgramData/NVIDIA Corporation/CUDA Samples/v12.5
-- - cuDNN path : D:/CUDA/cuDNN/9.6.0
-- - cuTENSOR path : D:/CUDA/cuTensor/2.0.2.1
-- - CUB path :
-- Found CUDAToolkit: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6/include (found version "12.6.85")
-- Found CUDA config:
-- - CUDA Toolkit path : C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6
-- - CUDA Samples path : C:/ProgramData/NVIDIA Corporation/CUDA Samples/v12.5
-- - cuDNN path : D:/CUDA/cuDNN/9.6.0
-- - cuTENSOR path : D:/CUDA/cuTensor/2.0.2.1
-- - CUB path : C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.6/include/cub
-- Configuring done (4.4s)
-- Generating done (0.1s)
-- Build files have been written to: D:/HIPIFY/build
Run Visual Studio 17 2022, open the generated hipify-clang.sln, and build the project test-hipify.