Conversation
|
It would be better to have the patches in the standard way rather than in the build_tarballs.jl. |
Sorry, but I am not quite familiar with BinaryBuilder. May I ask what is the "standard way"? Where can I find the source to learn about it? Thanks! |
|
If you look inside the Yggdrasil repo, you will see that patches are in |
- Extract inline sed patch to bundled/patches/fix_log2_ambiguity.patch - Extract heredoc dct_shim.cu to bundled/dct_shim.cu - Use DirectorySource and atomic_patch following Yggdrasil conventions
|
Done. |
|
I just noticed this is a GPU package. Take a look at S/SuiteSparse/SuiteSparse_GPU@7 to see how the GPU libraries need to be built. You need newer GCC, and an agumented platform block. It would be good to build the JLLs locally and test them. You can git grep in the repo for |
|
Done. Thanks for pointing the idiomatic way. |
3 participants
This PR adds a new recipe for VkDCT, a C++ shim library that wraps VkFFT to expose high-performance 3D DCT-I transforms for CUDA backends.
This library is a binary dependency for AcceleratedDCTs.jl. It enables users to perform GPU-accelerated Discrete Cosine Transforms without requiring a manual CUDA compilation step.