Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
3 contributors

Users who have contributed to this file

@emankov @mangupta @ChrisKitching
377 lines (296 sloc) 20 KB

hipify-clang

hipify-clang is a clang-based tool to translate CUDA source code into portable HIP C++ automatically.

Table of Contents

Supported CUDA APIs

Dependencies

hipify-clang requires:

  1. LLVM+CLANG of at least version 3.8.0; the latest stable and recommended release: 6.0.1 on Windows, and 8.0.0 on Linux.

  2. CUDA at least version 7.0, the latest supported version is 9.0 on Windows, and 10.0 on Linux.

If the target CUDA is 9.1, 9.2 or 10.0, to work on Windows you may apply patches*: for LLVM 7.0.0, for LLVM 7.0.1, for LLVM 8.0.0.

LLVM release version CUDA latest supported version Windows Linux
3.8.0 7.5 + +
3.8.1 7.5 + +
3.9.0 7.5 + +
3.9.1 7.5 + +
4.0.0 8.0 + +
4.0.1 8.0 + +
5.0.0 8.0 + +
5.0.1 8.0 + +
5.0.2 8.0 + +
6.0.0 9.0 + +
6.0.1 9.0 +
LATEST STABLE RELEASE
+
7.0.0 9.2 -
not working due to
the clang's bug 38811
+
patch*
-
not working due to
the clang's bug 36384
7.0.1 9.2 -
not working due to
the clang's bug 38811
+
patch*
-
not working due to
the clang's bug 36384
7.1.0 9.2 (?) -
LLVM 7.1.0
is not yet released
-
LLVM 7.1.0
is not yet released
8.0.0 10.0 -
not working due to
the clang's bug 38811
+
patch*
+
LATEST STABLE RELEASE
8.0.1 10.1 (?) -
LLVM 8.0.1
is not yet released
-
LLVM 8.0.1
is not yet released

* Download the patch and unpack it into your LLVM distributive directory; a few header files will be overwritten; rebuilding of LLVM is not needed.

In most cases, you can get a suitable version of LLVM+CLANG with your package manager.

Failing that or having multiple versions of LLVM, you can download a release archive, build or install it, and set CMAKE_PREFIX_PATH so cmake can find it; for instance: -DCMAKE_PREFIX_PATH=f:\LLVM\6.0.1\dist

Build and install

Build

Assuming this repository is at ./HIP:

cd hipify-clang
mkdir build dist
cd build

cmake \
 -DCMAKE_INSTALL_PREFIX=../dist \
 -DCMAKE_BUILD_TYPE=Release \
 ..

make -j install

On Windows, the following option should be specified for cmake at first place: -G "Visual Studio 15 2017 Win64"; the generated hipify-clang.sln should be built by Visual Studio 15 2017 instead of make.

Debug build type -DCMAKE_BUILD_TYPE=Debug is also supported and tested; LLVM+CLANG should be built in Debug mode as well. 64-bit build mode -Thost=x64 is supported as well; LLVM+CLANG should be built in 64-bit mode as well.

The binary can then be found at ./dist/bin/hipify-clang.

Testing

hipify-clang has unit tests using LLVM lit/FileCheck.

LLVM+CLANG should be built from sources, pre-built binaries are not exhaustive for testing.

To run it:

  1. Download LLVM+CLANG sources.

  2. Build LLVM+CLANG:

    cd llvm
    mkdir build dist
    cd build
    • Linux:
         cmake \
          -DCMAKE_INSTALL_PREFIX=../dist \
          -DLLVM_SOURCE_DIR=../llvm \
          -DCMAKE_BUILD_TYPE=Release \
          ../llvm
         make -j install
    • Windows:
         cmake \
          -G "Visual Studio 15 2017 Win64" \
          -DCMAKE_INSTALL_PREFIX=../dist \
          -DLLVM_SOURCE_DIR=../llvm \
          -DCMAKE_BUILD_TYPE=Release \
          -Thost=x64 \
          ../llvm

                Run Visual Studio 15 2017, open the generated LLVM.sln, build all, build project INSTALL.

  1. Ensure CUDA of minimum version 7.0 is installed.

    • Having multiple CUDA installations to choose a particular version the DCUDA_TOOLKIT_ROOT_DIR option should be specified:

      • Linux: -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.0

      • Windows: -DCUDA_TOOLKIT_ROOT_DIR="c:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v9.0"

        -DCUDA_SDK_ROOT_DIR="c:/ProgramData/NVIDIA Corporation/CUDA Samples/v9.0"

  2. Ensure cuDNN of the version corresponding to CUDA's version is installed.

    • Path to cuDNN should be specified by the CUDA_DNN_ROOT_DIR option:

      • Linux: -DCUDA_DNN_ROOT_DIR=/srv/CUDNN/cudnn-10.0-v7.5.1.10

      • Windows: -DCUDA_DNN_ROOT_DIR=f:/CUDNN/cudnn-9.0-windows10-x64-v7.5.1.10

  3. Ensure python of minimum required version 2.7 is installed.

  4. Ensure lit and FileCheck are installed - these are distributed with LLVM.

    • Install lit into python:

      • Linux: python /srv/git/LLVM/8.0.0/llvm/utils/lit/setup.py install

      • Windows: python f:/LLVM/6.0.1/llvm/utils/lit/setup.py install

    • Starting with LLVM 6.0.1 path to llvm-lit python script should be specified by the LLVM_EXTERNAL_LIT option:

      • Linux: -DLLVM_EXTERNAL_LIT=/srv/git/LLVM/8.0.0/build/bin/llvm-lit

      • Windows: -DLLVM_EXTERNAL_LIT=f:/LLVM/6.0.1/build/Release/bin/llvm-lit.py

  5. Set HIPIFY_CLANG_TESTS option turned on: -DHIPIFY_CLANG_TESTS=1.

  6. Run cmake:

  7. Run tests:

    • Linux: make test-hipify.

    • Windows: run Visual Studio 15 2017, open the generated hipify-clang.sln, build project test-hipify.

Linux

On Linux the following configurations are tested:

Ubuntu 14: LLVM 5.0.0 - 6.0.1, CUDA 7.0 - 9.0, cudnn-5.0.5 - cudnn-7.5.1.10

Ubuntu 16-18: LLVM 8.0.0, CUDA 8.0 - 10.0, cudnn-5.1.10 - cudnn-7.5.1.10

Build system for the above configurations:

Python 2.7 (min), cmake 3.12.3 (min), GNU C/C++ 5.4.0 (min).

Here is an example of building hipify-clang with testing support on Ubuntu 16.04:

cmake
 -DHIPIFY_CLANG_TESTS=1 \
 -DCMAKE_BUILD_TYPE=Release \
 -DCMAKE_INSTALL_PREFIX=../dist \
 -DCMAKE_PREFIX_PATH=/srv/git/LLVM/8.0.0/dist \
 -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.0 \
 -DCUDA_DNN_ROOT_DIR=/srv/CUDNN/cudnn-10.0-v7.5.1.10 \
 -DLLVM_EXTERNAL_LIT=/srv/git/LLVM/8.0.0/build/bin/llvm-lit \
 ..

A corresponding successful output:

-- The C compiler identification is GNU 5.4.0
-- The CXX compiler identification is GNU 5.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found LLVM 8.0.0:
--    - CMake module path: /srv/git/LLVM/8.0.0/dist/lib/cmake/llvm
--    - Include path     : /srv/git/LLVM/8.0.0/dist/include
--    - Binary path      : /srv/git/LLVM/8.0.0/dist/bin
-- Linker detection: GNU ld
-- Found PythonInterp: /usr/bin/python2.7 (found suitable version "2.7.12", minimum required is "2.7")
-- Found lit: /usr/local/bin/lit
-- Found FileCheck: /srv/git/LLVM/8.0.0/dist/bin/FileCheck
-- Looking for pthread.h
-- Looking for pthread.h - found
-- Looking for pthread_create
-- Looking for pthread_create - not found
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - found
-- Found Threads: TRUE
-- Found CUDA: /usr/local/cuda-10.0 (found version "10.0")
-- Configuring done
-- Generating done
-- Build files have been written to: /srv/git/HIP/hipify-clang/build
make test-hipify

A corresponding successful output:

Running HIPify regression tests
========================================
CUDA 10.0 - will be used for testing
LLVM 8.0.0 - will be used for testing
x86_64 - Platform architecture
Linux 4.13.0-32-generic - Platform OS
64 - hipify-clang binary bitness
64 - python 2.7.12 binary bitness
========================================
-- Testing: 49 tests, 12 threads --
PASS: hipify :: unit_tests/headers/headers_test_01.cu (1 of 49)
PASS: hipify :: unit_tests/headers/headers_test_03.cu (2 of 49)
PASS: hipify :: unit_tests/headers/headers_test_02.cu (3 of 49)
PASS: hipify :: unit_tests/headers/headers_test_05.cu (4 of 49)
PASS: hipify :: unit_tests/headers/headers_test_06.cu (5 of 49)
PASS: hipify :: unit_tests/libraries/cuBLAS/cublas_0_based_indexing.cu (6 of 49)
PASS: hipify :: unit_tests/headers/headers_test_07.cu (7 of 49)
PASS: hipify :: unit_tests/libraries/CAFFE2/caffe2_02.cu (8 of 49)
PASS: hipify :: unit_tests/headers/headers_test_04.cu (9 of 49)
PASS: hipify :: unit_tests/headers/headers_test_08.cu (10 of 49)
PASS: hipify :: unit_tests/libraries/cuBLAS/cublas_1_based_indexing.cu (11 of 49)
PASS: hipify :: unit_tests/libraries/cuBLAS/rocBLAS/cublas_1_based_indexing_rocblas.cu (12 of 49)
PASS: hipify :: unit_tests/libraries/cuBLAS/cublas_sgemm_matrix_multiplication.cu (13 of 49)
PASS: hipify :: unit_tests/libraries/cuBLAS/rocBLAS/cublas_0_based_indexing_rocblas.cu (14 of 49)
PASS: hipify :: unit_tests/libraries/cuComplex/cuComplex_Julia.cu (15 of 49)
PASS: hipify :: unit_tests/libraries/cuDNN/cudnn_softmax.cu (16 of 49)
PASS: hipify :: unit_tests/libraries/cuFFT/simple_cufft.cu (17 of 49)
PASS: hipify :: unit_tests/libraries/cuBLAS/rocBLAS/cublas_sgemm_matrix_multiplication_rocblas.cu (18 of 49)
PASS: hipify :: unit_tests/libraries/CAFFE2/caffe2_01.cu (19 of 49)
PASS: hipify :: unit_tests/libraries/cuRAND/poisson_api_example.cu (20 of 49)
PASS: hipify :: unit_tests/libraries/cuDNN/cudnn_convolution_forward.cu (21 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_01.cu (22 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_02.cu (23 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_04.cu (24 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_03.cu (25 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_06.cu (26 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_05.cu (27 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_07.cu (28 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_08.cu (29 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_11.cu (30 of 49)
PASS: hipify :: unit_tests/headers/headers_test_09.cu (31 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_10.cu (32 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_09.cu (33 of 49)
PASS: hipify :: unit_tests/samples/2_Cookbook/0_MatrixTranspose/MatrixTranspose.cpp (34 of 49)
PASS: hipify :: unit_tests/samples/2_Cookbook/2_Profiler/Profiler.cpp (35 of 49)
PASS: hipify :: unit_tests/samples/2_Cookbook/1_hipEvent/hipEvent.cpp (36 of 49)
PASS: hipify :: unit_tests/samples/allocators.cu (37 of 49)
PASS: hipify :: unit_tests/samples/coalescing.cu (38 of 49)
PASS: hipify :: unit_tests/samples/dynamic_shared_memory.cu (39 of 49)
PASS: hipify :: unit_tests/samples/intro.cu (40 of 49)
PASS: hipify :: unit_tests/libraries/cuRAND/benchmark_curand_generate.cpp (41 of 49)
PASS: hipify :: unit_tests/samples/2_Cookbook/7_streams/stream.cpp (42 of 49)
PASS: hipify :: unit_tests/samples/square.cu (43 of 49)
PASS: hipify :: unit_tests/samples/static_shared_memory.cu (44 of 49)
PASS: hipify :: unit_tests/samples/axpy.cu (45 of 49)
PASS: hipify :: unit_tests/samples/vec_add.cu (46 of 49)
PASS: hipify :: unit_tests/samples/cudaRegister.cu (47 of 49)
PASS: hipify :: unit_tests/libraries/cuSPARSE/cuSPARSE_12.cu (48 of 49)
PASS: hipify :: unit_tests/libraries/cuRAND/benchmark_curand_kernel.cpp (49 of 49)
Testing Time: 2.82s
  Expected Passes    : 49
[100%] Built target test-hipify

Windows

On Windows 10 the following configurations are tested:

LLVM 5.0.0 - 5.0.2, CUDA 8.0, cudnn-5.1.10 - cudnn-7.1.4.18

LLVM 6.0.0 - 6.0.1, CUDA 9.0, cudnn-7.0.5.15 - cudnn-7.5.1.10

LLVM 7.0.0 - 8.0.0 (with patch*), CUDA 7.5 - 10.0, cudnn-7.0.5.15 - cudnn-7.5.1.10

Build system for the above configurations:

Python 3.6 (min), cmake 3.12.3 (min), Visual Studio 15.5.2 2017 - 16.0.4 2019.

Here is an example of building hipify-clang with testing support on Windows 10 by Visual Studio 15 2017:

cmake
 -G "Visual Studio 15 2017 Win64" \
 -DHIPIFY_CLANG_TESTS=1 \
 -DCMAKE_BUILD_TYPE=Release \
 -DCMAKE_INSTALL_PREFIX=../dist \
 -DCMAKE_PREFIX_PATH=f:/LLVM/6.0.1/dist \
 -DCUDA_TOOLKIT_ROOT_DIR="c:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v9.0" \
 -DCUDA_SDK_ROOT_DIR="c:/ProgramData/NVIDIA Corporation/CUDA Samples/v9.0" \
 -DCUDA_DNN_ROOT_DIR=f:/CUDNN/cudnn-9.0-windows10-x64-v7.5.1.10 \
 -DLLVM_EXTERNAL_LIT=f:/LLVM/6.0.1/build/Release/bin/llvm-lit.py \
 -Thost=x64
 ..

A corresponding successful output:

-- Found LLVM 6.0.1:
--    - CMake module path: F:/LLVM/6.0.1/dist/lib/cmake/llvm
--    - Include path     : F:/LLVM/6.0.1/dist/include
--    - Binary path      : F:/LLVM/6.0.1/dist/bin
-- Found PythonInterp: C:/Program Files/Python36/python.exe (found suitable version "3.6.4", minimum required is "3.6")
-- Found lit: C:/Program Files/Python36/Scripts/lit.exe
-- Found FileCheck: F:/LLVM/6.0.1/dist/bin/FileCheck.exe
-- Found CUDA: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v9.0 (found version "9.0")
-- Configuring done
-- Generating done
-- Build files have been written to: f:/HIP/hipify-clang/build

Running and using hipify-clang

To process a file, hipify-clang needs access to the same headers that would be needed to compile it with clang.

For example:

./hipify-clang square.cu --cuda-path=/usr/local/cuda-10.0 -I /usr/local/cuda-10.0/samples/common/inc

hipify-clang arguments are given first, followed by a separator, and then the arguments you'd pass to clang if you were compiling the input file. The Clang manual for compiling CUDA may be useful.

For a list of hipify-clang options, run hipify-clang --help.

Disclaimer

The information contained herein is for informational purposes only, and is subject to change without notice. While every precaution has been taken in the preparation of this document, it may contain technical inaccuracies, omissions and typographical errors, and AMD is under no obligation to update or otherwise correct this information. Advanced Micro Devices, Inc. makes no representations or warranties with respect to the accuracy or completeness of the contents of this document, and assumes no liability of any kind, including the implied warranties of noninfringement, merchantability or fitness for particular purposes, with respect to the operation or use of AMD hardware, software or other products described herein. No license, including implied or arising by estoppel, to any intellectual property rights is granted by this document. Terms and limitations applicable to the purchase or use of AMD's products are as set forth in a signed agreement between the parties or in AMD's Standard Terms and Conditions of Sale.

AMD, the AMD Arrow logo, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.

Copyright (c) 2014-2019 Advanced Micro Devices, Inc. All rights reserved.

You can’t perform that action at this time.