CHIP-SPV

CHIP-SPV makes HIP and CUDA applications portable to platforms which support SPIR-V as the device intermediate representation. Currently it supports OpenCL and Level Zero as the low-level runtime alternatives.

For User documentation, read this. For Developer documentation, read this. For a list of (un)supported features, read this.

This project is an integration of HIPCL and HIPLZ projects.

Prerequisites

Cmake >= 3.16.0
Clang 14, 15 or 16
- Can be installed, for example, by adding the LLVM's Debian/Ubuntu repository and installing packages 'clang-15 llvm-15 clang-tools-15'.
- NOTE: Some features currently require patches that are not yet upstreamed to LLVM. They are on top of the branch here.
SPIRV-LLVM-Translator from a branch matching to the clang version: (e.g. llvm_release_150 for Clang 15.0) llvm-spirv.
- For best results, install CHIP-SPV's LLVM 15 branch or CHIP-SPV's LLVM 16 branch which have fixes that are not yet in upstream.
For Level Zero Backend
- Intel Compute Runtime or oneAPI
- oneAPI Level Zero Loader
For OpenCL Backend
- OpenCL 2.0 or 3.0 implementation with coarse grained Shared Virtual Memory and SPIR-V input supported.
For HIP-SYCL and HIP-MKL Interoperability
- oneAPI

Downloading Sources

git clone https://github.com/CHIP-SPV/chip-spv.git
cd chip-spv
git submodule update --init --recursive

Building

mkdir build
cd build
cmake .. \ 
    -DLLVM_CONFIG=/path/to/llvm-config # optional, if not in PATH or if only versioned binary is available i.e. llvm-config-15
    -DCMAKE_INSTALL_PREFIX=/path/to/install # optional, default is <build_dir>/install
make
make install

Useful options:

-DCMAKE_BUILD_TYPE=<Debug(default), Release, RelWithDebInfo>
-DBUILD_SAMPLES=<ON(default), OFF> # Samples are built by default, unless you set this to OFF

The documentation will be placed in doxygen/html.

Building & Running Unit Tests

make build_tests
make check # runs only tests that are expected to work

Building documentation

Make sure you have doxygen installed. Then:

....
cd build
cmake .....
make gendocs

The documentation will be placed in doxygen/html.

Troubleshooting

Missing Double Precision Support

When running the tests on OpenCL devices which do not support double precision floats, there will be multiple tests that will error out.

It might be possible to enable software emulation of double precision floats for Intel iGPUs by setting two environment variables to make kernels using doubles work but with the major overhead of software emulation:

export IGC_EnableDPEmulation=1
export OverrideDefaultFP64Settings=1

colleeneb/chip-spv