Add kokkos-fft blog post #158
Conversation
content/blog/blog-post-10.md
Outdated
|
|
||
| * wishing to integrate in-situ signal and image processing with FFTs. E.g., spectral analyses, low pass filtering, etc. | ||
|
|
||
| * __NOT__ willing to go through the documentation of de facto standard FFT libraries. They want to benefit from powerful vendor FFT libraries while working with a simple API like that of [`numpy.fft`](https://numpy.org/doc/stable/reference/routines.fft.html). |
There was a problem hiding this comment.
I don't get that point.
Is that about saying that the kokkos-fft API is easier to use than other C++ alternatives and inspired from numpy.fft?
There was a problem hiding this comment.
Yes
I would like to say that the kokkos-fft API is easier to use than those of vendor libraries and inspired from numpy.fft?
There was a problem hiding this comment.
I think that "easier" point is kind of subjective. It is certainly more intuitive to use for Python users, but I am not sure it is the case for hardcore fftw Fortran users.
There was a problem hiding this comment.
@cedricchevalier19 I understand your point. Actually, what I am claiming here is that our API is as simple as numpy.fft. As you may say, fftw experts could only prefer fftw APIs
content/blog/blog-post-10.md
Outdated
| # Prerequisites | ||
|
|
||
| To use kokkos-fft, we need the following: | ||
|
|
||
| * `CMake 3.22+` | ||
| * `Kokkos 4.4+` | ||
|
|
||
| Depending on your backend, you also need at least one of the following compilers (for the latest requirements, see [README](https://github.com/kokkos/kokkos-fft)): | ||
|
|
||
| * `gcc 8.3.0+` (CPUs) | ||
| * `IntelLLVM 2023.0.0+` (CPUs, Intel GPUs) | ||
| * `nvcc 11.0.0+` (NVIDIA GPUs) | ||
| * `rocm 5.3.0+` (AMD GPUs) | ||
|
|
||
| # Using kokkos-fft in your CMake project | ||
|
|
||
| For the moment, there are two ways to use kokkos-fft: include as a subdirectory in your CMake project or install as a library. Since kokkos-fft is a header-only library, it is easy to add it as a subdirectory in a CMake project. We will see how to do it. | ||
|
|
||
| It is assumed that both Kokkos and kokkos-fft are placed under `<project_directory>/tpls`. Here is the structure of a simple CMake project. | ||
|
|
||
| ``` | ||
| ---/ | ||
| | | ||
| └──<project_directory>/ | ||
| |──tpls/ | ||
| | |──kokkos/ | ||
| | └──kokkos-fft/ | ||
| |──CMakeLists.txt | ||
| └──hello.cpp | ||
| ``` | ||
|
|
||
| See [here](https://kokkos.org/kokkos-core-wiki/get-started/integrating-kokkos-into-your-cmake-project.html#embedded-kokkos-via-add-subdirectory-and-git-submodules) about embedding Kokkos via `add_subdirectory()` and Git Submodules in your CMake project. Then, you need to clone the repo of kokkos-fft into `<project_directory>/tpls/kokkos-fft`. | ||
|
|
||
| ```bash | ||
| git clone --recursive https://github.com/kokkos/kokkos-fft.git | ||
| ``` | ||
|
|
||
| The `CMakeLists.txt` file of the project would be | ||
|
|
||
| ```CMake | ||
| cmake_minimum_required(VERSION 3.23) | ||
| project(kokkos-fft-as-subdirectory LANGUAGES CXX) | ||
|
|
||
| add_subdirectory(tpls/kokkos) | ||
| add_subdirectory(tpls/kokkos-fft) | ||
|
|
||
| add_executable(hello-kokkos-fft hello.cpp) | ||
| target_link_libraries(hello-kokkos-fft PUBLIC Kokkos::kokkos KokkosFFT::fft) | ||
| ``` | ||
|
|
||
| To build the project, we basically rely on the CMake options for Kokkos. For example, the build steps for an A100 GPU based backend are as follows: | ||
|
|
||
| ```bash | ||
| cmake -B build \ | ||
| -DCMAKE_CXX_COMPILER=g++ \ | ||
| -DCMAKE_BUILD_TYPE=Release \ | ||
| -DKokkos_ENABLE_CUDA=ON \ | ||
| -DKokkos_ARCH_AMPERE80=ON | ||
| cmake --build build -j 8 | ||
| ``` | ||
|
|
||
| This way, all the FFT functionalities will be executed on the A100 GPU. For installation as a library, details are provided in the [documentation](https://kokkosfft.readthedocs.io/en/latest/intro/building.html#install-kokkosfft-as-a-library). |
There was a problem hiding this comment.
I would drop these two sections and defer to the kokkos-fft documentation.
You could say that the code for the following example can be found in the kokkos-fft repo and post a link.
There was a problem hiding this comment.
Makes sense. I just tried to make this blog post self-contained. If it is not necessary, I can drop this part
|
@dalg24 Thanks a lot for your comments. |
No hard deadline. But obviously we don't want to unnecessarily drag it too much. |
Good. I think we can safely make it before the end of next week |
|
Hi @dalg24 Since I have largely modified the source code, I made a benchmark again. |
yasahi-hpc
force-pushed
the
add-kokkos-fft-blog
branch
from
1e4dbca to
96238dc
Compare
|
@dalg24 |
This PR adds a blog post on kokkos-fft.