[dependabot]

Updates the requirements on CUDACore to permit the latest version.

Release notes

Sourced from CUDACore's releases.

v6.4.0

CUDA v6.4.0

Diff since v6.3.1

Merged pull requests:

• Add UE5M3 to cudaDataType enum (#3226) (@​AntonOresten)
• [cuSPARSE] fix Float16 spmm (#3253) (@​simeonschaub)
• Fix profiling and cuDNN tests on Tegra (#3254) (@​maleadt)
• Restore best-effort CUDA 10.2 and 11 support (#3255) (@​maleadt)
• Gate CUPTI MARKER_DATA activity records on CUDA 11 (#3256) (@​maleadt)
• Implement LinearAlgebra's storage-specific mul! methods (#3257) (@​maleadt)
• cuSOLVER: Materialize QR's Q with orgqr (#3258) (@​maleadt)
• Annotate CuErrors with the driver error log (#3259) (@​maleadt)
• Finish the binary-stack rework, and fix Tegra profiling (#3260) (@​maleadt)
• Separate PTX and cubin target compatibility (#3261) (@​maleadt)
• Fix cached toolkit incompatibility and local Tegra profiling (#3262) (@​maleadt)
• Unify CUDA driver and library logging (#3263) (@​maleadt)
• Hook the Enzyme kernel rules into the @​cuda launch pipeline (#3264) (@​vchuravy)
• Fix BitonicSortImpl Int/Int32 method overwrite on 32-bit Julia (#3265) (@​ChrisRackauckas-Claude)
• Adapt to the library-based NVPTX_LLVM_Backend_jll 23 (#3267) (@​maleadt)
• Default CUDA atomics to device scope and expose explicit scopes (#3268) (@​maleadt)
• Update to CUDA 13.4 GA. (#3269) (@​maleadt)
• Improve Tegra compatibility (best-effort) (#3270) (@​maleadt)
• Gate the pinned-memory system-scope atomic test on host native atomics (#3271) (@​maleadt)

Closed issues:

• NVIDIA GeForce GTX 1080 GPU (compute capability 6.1) is not fully supported by CUDA 12.9.0. (#3187)
• Could not locate cudnn_engines_tensor_ir64_9.dll (#3205)
• Instantiating F.Q errors on a very large, thin A input to F = qr!(A). (#3234)

Changelog

Sourced from CUDACore's changelog.

CUDA.jl release history

This document lists the noteworthy changes in every release of CUDA.jl, newest first. It is a hand-written summary aimed at users; for the complete list of merged pull requests, see the GitHub releases.

CUDA.jl follows semantic versioning, so intentional breaking changes are reserved for major releases. Some minor releases nevertheless contain changes that are technically breaking, such as stricter validation of previously invalid uses, corrections to unsupported behavior, or changes to support floors; those are labeled separately below. Patch releases are listed as subsections of the minor release they belong to.

v6.4 (unreleased)

NVIDIA's Jetson boards are now a supported platform, across every JetPack generation still in the field, and the binary stack that picks the driver and the CUDA toolkit was rebuilt to make that selection reliable.

Pkg.add("CUDA") on a Jetson selects the cuda_platform=jetson toolkit builds automatically. On JetPack 7 that is all there is to it: CUDA 13 artifacts on the system driver, with every vendor library and the integrated profiler. On JetPack 5 and 6, CUDA_Driver_jll bundles NVIDIA's L4T forward-compatibility driver for the board's kernel-mode driver generation, verifies that it initializes and supports every device present, and only then loads it; the toolkit is then sized on that driver rather than on the system one. An AGX Xavier goes from CUDA 11.4 to CUDA 12.5 without reflashing. On JetPack 4, CUDA 10.2 is now fully artifact-based, so a Jetson Nano no longer needs a local toolkit.

The three JLLs involved now have one responsibility each. CUDA_Driver_jll decides which driver to load and exports libcuda; CUDA_Runtime_jll and CUDA_Compiler_jll decide which toolkit to use by querying that driver directly during artifact selection. Previously the driver JLL hosted a toolkit-selection library that the other two called across a package boundary, which was unreliable in Pkg's artifact-selection subprocess and meant every toolkit addition needed a driver release. Requires CUDA_Driver_jll 13.3.4, CUDA_Runtime_jll 0.24.4 and CUDA_Compiler_jll 0.6.2.

Technically breaking changes:

• CUDA.jl atomic operations now default to device scope, matching the scope of CUDA C's atomicX functions. LLVM 22 began honoring system scope for CAS, causing the Windows driver to reject Pascal kernels containing these instructions with error 801, including bounds-checking exception paths (#3187). The low-level atomic_*! functions accept a trailing Val(:block), Val(:device), or Val(:system); CUDA.@atomic uses device scope. These functions report a

... (truncated)

Commits

• 65f6ddd Bump version to v6.4.
• 0e22d5c Fix BitonicSortImpl Int/Int32 method overwrite on 32-bit Julia (#3265)
• 6bca074 Hook the Enzyme kernel rules into the @​cuda launch pipeline (#3264)
• 0cc3d37 Merge pull request #3270 from JuliaGPU/tb/tegra-fixes
• bc72d58 Gate the pinned-memory system-scope atomic test on host native atomics (#3271)
• 5365af7 Respect device and toolkit limits in core tests
• 1ee5da1 Report unavailable optional libraries and skip their tests
• 3830fba Preserve stream-capture errors during kernel loading
• fc125cf Throw when SASS inspection is unavailable
• ce7f8ab Require CUPTI 11 for integrated profiling
• Additional commits viewable in compare view

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

---

Dependabot commands and options

You can trigger Dependabot actions by commenting on this PR:

• @dependabot rebase will rebase this PR
• @dependabot recreate will recreate this PR, overwriting any edits that have been made to it
• @dependabot show <dependency name> ignore conditions will show all of the ignore conditions of the specified dependency
• @dependabot ignore this major version will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this minor version will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself)
• @dependabot ignore this dependency will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself)

[dependabot]

Superseded by #11.