Skip to content

Latest commit

 

History

History
91 lines (60 loc) · 3.61 KB

README.md

File metadata and controls

91 lines (60 loc) · 3.61 KB
Raw

hipFFT

hipFFT is an FFT marshalling library that supports rocFFT and cuFFT backends.

hipFFT exports an interface that doesn't require the client to change, regardless of the chosen backend. It sits between your application and the backend FFT library, where it marshals inputs to the backend and marshals results back to your application.

Documentation

Note

The published hipFFT documentation is available at hipFFT in an organized, easy-to-read format, with search and a table of contents. The documentation source files reside in the hipFFT/docs folder of this repository. As with all ROCm projects, the documentation is open source. For more information, see Contribute to ROCm documentation.

To build our documentation locally, run the following code:

cd docs

pip3 install -r sphinx/requirements.txt

python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html

Build and install

You can download pre-built packages from the ROCm package servers.

If you're using Ubuntu, you can run: sudo apt update && sudo apt install hipfft.

Building from source

To build hipFFT from source, follow these steps:

  1. Install the library build dependencies:

    • On AMD platforms, you must install rocFFT.
    • On NVIDIA platforms, you must install cuFFT.

  2. Install the client build dependencies:

    • The clients (samples, tests, etc) included with the hipFFT source depend on hipRAND, FFTW and GoogleTest.

  3. Build hipFFT:

    To show all build options:

      mkdir build && cd build
  cmake -LH ..

Here are some CMake build examples:

  • AMD GPU
    • Case: Build a project using HIP language APIs + hipFFT with standard host compiler
      • Code: cmake -DCMAKE_CXX_COMPILER=g++ -DCMAKE_BUILD_TYPE=Release -L ..
    • Case: Build a project using HIP language APIs + hipFFT + device kernels with HIP-Clang
      • Code: cmake -DCMAKE_CXX_COMPILER=amdclang++ -DCMAKE_BUILD_TYPE=Release -DBUILD_CLIENTS=ON -L ..
  • NVIDIA GPU
    • Case: Build a project using HIP language APIs + hipFFT with standard host compiler
      • Code: cmake -DCMAKE_CXX_COMPILER=g++ -DCMAKE_BUILD_TYPE=Release -DBUILD_WITH_LIB=CUDA -L ..
    • Case: Build a project using HIP language APIs + hipFFT + device kernels with HIP-NVCC
      • Code: HIP_PLATFORM=nvidia cmake -DCMAKE_CXX_COMPILER=hipcc -DCMAKE_BUILD_TYPE=Release -DBUILD_CLIENTS=ON -L ..
The `-DBUILD_CLIENTS=ON` option is only allowed with the amdclang++ or HIPCC compilers.

Porting from CUDA

If you have existing CUDA code and want to transition to HIP, follow these steps:

  1. HIPIFY your code and fix all unsupported CUDA features and user-defined macros
  2. Build with HIP-NVCC to run on an NVIDIA device
  3. Build with HIP-Clang to run on an AMD device

More information about porting to HIP is available in the HIP porting guide.

Support

You can report bugs and feature requests through the GitHub issue tracker.

Contribute

If you want to contribute to hipFFT, you must follow our contribution guidelines.