René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Julia JIT support for device APIs","articleBody":"As header-only libraries based on CUTLASS, these only work in C++.\n\nIt would be nice to have e.g. the new DGEMM via IMMA (https://github.com/NVIDIA/CUDALibrarySamples/tree/master/MathDx/cuBLASDx/16_dgemm_emulation) inside Julia kernels.\n\nUnfortunately, Warp/numba can only JIT compile a subset of python types, and not the (more general) Julia types that we require (e.g. differential equation solvers).\n\nA potential way forward would be to compile libmathdx to PTX, link it in via LLVM, and get full performance via LTO. However, these often contain NVVM IR, which LLVM cannot handle.\n\nRelated discussion: https://discourse.julialang.org/t/using-cublasdx-in-julia/125527","author":{"url":"https://github.com/ggkountouras","@type":"Person","name":"ggkountouras"},"datePublished":"2025-06-14T15:27:34.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":2},"url":"https://github.com/32/nvmath-python/issues/32"}