Repository navigation
Add cuda-toolkit/13.4.3 with Linux and Windows arm64 support - #13
Merged
Merged
Conversation
danimtb
force-pushed
the
danimtb/cuda-toolkit-13.4-linux
branch
from
October 15, 2026 11:21
0b9cf9a to
0757764
Compare
danimtb
marked this pull request as ready for review
October 15, 2026 14:28
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Replaces the
13.4.0-previewrecipe (Windows only, based on the local installer) with the official 13.4.3 release from NVIDIA's redist. The new recipe starts from the 13.3 recipe and adds Windows arm64 support, including cross-building between x86_64 and arm64.Recipe changes (compared to 13.3)
windows-arm64redist platform (arch=armv8), includinglibcudla.cuda_compat. It is no longer part of the 13.4.3 redist (cuda_compat_orinis still packaged on linux-sbsa).cuda_cuobjdump,cuda_nvdisasmandcuda_nvprune, which were included in the preview package.nvcc.profileso nvcc takes the library path fromCONAN_CUDA_LIB_DIR, which the host package defines.CMAKE_VS_GLOBALS(CudaToolkitLibDir) so the Visual Studio generator links against the target architecture libraries.libdirs/bindirsare now per architecture (lib/{x64,arm64},bin/{x64,arm64},nvvm/lib/{x64,arm64}). The NVVM DLL is innvvm/bin/x64on x86_64 but innvvm/binon arm64.host_config.h.Other changes
config.yml:13.4.0-previewreplaced by13.4.3. The previewconandata.ymlis removed.13.4.3.resolve_prereleasesfrom the cuda-samples and llama-cpp steps.docs/building-for-windows-on-arm.md: remove the developer preview references and the prerelease flag.Testing (local)
conan createpasses.test_packagebuilds with the Visual Studio generator and runs.test_packagewas not built locally because there is no ARM64 MSVC toolset.conan createandtest_packagepass.test_packagebuilds are covered by CI.