v5.8.2

CUDA v5.8.2

Diff since v5.8.1

Merged pull requests:

• Fix spdiagm with specified pairs (#2784) (@ErikQQY)
• Add diagm in CUBLAS (#2786) (@ErikQQY)

Closed issues:

• Where to host extension(s) (#2735)
• spdiagm doesn't support specified diagonal elements (#2783)
• CUDA failed to create a diagonal matrix of CuArray(u) (#2785)

v5.8.1

CUDA v5.8.1

Diff since v5.8.0

Merged pull requests:

• CUSPARSE: Bugfixes for sparse vector broadcast. (#2780) (@maleadt)

v5.8.0

CUDA v5.8.0

Diff since v5.7.3

Merged pull requests:

• SparseMatricesCSR Dispatch (#2720) (@Abdelrahman912)
• Very rough implementation of bcast for CuSparseVector (#2733) (@kshyatt)
• Possible fix for #2745, change args in call to cusparseCreateBsr (#2747) (@manuelbb-upb)
• Simple tests for check and explain_eltype (#2748) (@kshyatt)
• Test for printing OutOfGPUMemoryError (#2749) (@kshyatt)
• Fix log_message pileup (#2750) (@fps)
• Test for parse_limit (#2751) (@kshyatt)
• unsafe_wrap for symbols (#2753) (@vchuravy)
• Use thread adoption to handle log messages. (#2754) (@maleadt)
• Add pre-commit configuration (#2755) (@vchuravy)
• Broaden check for eltypes to make sure we don't allow invalid stuff (#2756) (@kshyatt)
• Prefer aligned_sizeof (#2757) (@vchuravy)
• More array tests (#2758) (@kshyatt)
• A few more tests for CUSOLVER Q mats (#2759) (@kshyatt)
• More tests for CuArrayPtr (#2760) (@kshyatt)
• [CUSOLVER] Update gesvdp! (#2763) (@amontoison)
• Get rid of unneeded version checks (#2765) (@kshyatt)
• Remove second import of aligned_sizeof (#2767) (@vchuravy)
• CUSPARSE SpGEMM: Support algorithms 2 and 3 (#2769) (@maleadt)
• Update to CUDA 12.9. (#2772) (@maleadt)
• Fix SPGEMM_ALGOS setup (#2773) (@jonas-schulze)
• Support new functionality from KA 0.9.32 (#2774) (@michel2323)
• cuTENSOR: Preserve storage type when multiplying (#2775) (@christiangnrd)
• Update subpackages. (#2776) (@maleadt)
• Remove the unnecessary reshape during mapreduce. (#2778) (@maleadt)

Closed issues:

• Type conversions in broadcast fails when compiling with always_inline=true (#2722)
• cuDNN loses memory to log messages in Pluto.jl context (#2743)
• Xgesvdp! failure when only requesting singular values (#2761)
• CUDA 5.7.3 fails to precompile on Julia 1.12.0-beta2 (#2762)
• aligned_sizeof with an existing identifier (#2766)
• CUSPARSE_SPGEMM_ALG2 not working (#2768)
• sum! throws dispatch error beyond a threshold number of rows (#2777)

v5.7.3

CUDA v5.7.3

Diff since v5.7.2

Merged pull requests:

• Merge CSC/CSR broadcast kernels (#2731) (@kshyatt)
• GPUToolbox v0.2 take 2 (#2736) (@christiangnrd)
• Add dispatches to access device matrix data via SparseArrays interface (#2738) (@termi-official)
• More tests for CuContext (#2739) (@kshyatt)
• Fill in missing KA functionality (KA.functional + sparse matrices adaption from CUDAbackend) (#2740) (@Abdelrahman912)
• Small tests and changes for coverage (#2742) (@kshyatt)
• More tests and better error type for cusparse generic (#2744) (@kshyatt)
• Restore the descriptors in CUSPARSE (#2746) (@amontoison)

v5.7.2

CUDA v5.7.2

Diff since v5.7.1

Merged pull requests:

• Support disabling implicit synchronization (#2662) (@vchuravy)
• More tests and bugfixes for CUSOLVER (#2707) (@kshyatt)
• Set neutral element to zero for sparse reduce (#2710) (@kshyatt)
• Bugfix and tests for cusolver/base (#2712) (@kshyatt)
• Small fixes and missed tests for CUTENSORNET (#2713) (@kshyatt)
• Even more tests and small fixes for CUTENSORNET (#2715) (@kshyatt)
• Tests for CUSTATEVEC errors (#2716) (@kshyatt)
• Add compat entries for recent devices and toolkits. (#2717) (@maleadt)
• Split out copyto for texture arrays and add more tests (#2719) (@kshyatt)
• Add a docstring for pointer (#2721) (@maleadt)
• More CUSOLVER dense tests (#2723) (@kshyatt)
• Tests for some helper functions (#2724) (@kshyatt)
• More tests and bugfixing for CUSPARSE (#2725) (@kshyatt)
• Add more methods for all versions to unstick tests (#2726) (@kshyatt)

Closed issues:

• Ability to opt out of / improved automatic synchronization between tasks for shared array usage (#2617)
• maximum(abs, CuSparseMatrixCSR) returns Inf (#2705)
• mapreduce(f, op, A) for sparse A is wrong if f(0) =/= 0 (#2709)

v5.7.1

CUDA v5.7.1

Diff since v5.7.0

Merged pull requests:

• Tests for MIME printing and indexing (#2686) (@kshyatt)
• Loosen VERSION check for sketchy test (#2688) (@kshyatt)
• CompatHelper: bump compat for GPUToolbox to 0.2, (keep existing compat) (#2689) (@github-actions[bot])
• Even more sparse printing and tril/triu tests (#2692) (@kshyatt)
• Even more sparse tests (#2695) (@kshyatt)
• More tests and a matmatmul fix (#2697) (@kshyatt)
• Sparse conversion tests (#2698) (@kshyatt)
• Tests for descriptors (#2700) (@kshyatt)
• More tests for some missing kron methods (#2701) (@kshyatt)
• Don't duplicate const defs (#2703) (@kshyatt)
• Exclude device-side sorting code from coverage (#2704) (@kshyatt)
• More tests for CuRef/CuRefArray (#2706) (@kshyatt)
• Update Project.toml (#2708) (@kshyatt)

Closed issues:

• GC corruption on 1.10 during cusparse/reduce tests (#2027)
• Launch bounds interface (#2674)
• Precompilation errors: ERROR: LoadError: invalid redefinition of constant CUSPARSE.CuSparseUpperOrUnitUpperTriangular (#2690)

v5.7.0

CUDA v5.7.0

Diff since v5.6.1

Merged pull requests:

• Bugfix for batched gemv (#2481) (@kose-y)
• Split out level 3 gemm tests (#2610) (@kshyatt)
• Switch CUBLAS to device-side pointer mode (#2616) (@kshyatt)
• Elide bounds checks when kernels contains manual ones. (#2621) (@maleadt)
• Support passing symbols as arguments (#2624) (@vchuravy)
• Remove eager synchronization with HtoD copies. (#2625) (@maleadt)
• Don't prefetch on multi-device systems (#2626) (@vchuravy)
• Cooperative groups: add a boundscheck to avoid confusing inexact errors. (#2631) (@maleadt)
• NFC fixes (#2632) (@maleadt)
• Update to CUDA 12.8 (#2634) (@maleadt)
• [CUSOLVER] Update the test of syevBatched! (#2636) (@amontoison)
• Improve NSight Systems activation by inspecting the session list. (#2638) (@maleadt)
• [CUSPARSE] Support CuSparseMatrixBSR in the generic mm! (#2639) (@amontoison)
• [CUSOLVER] Support symmetric factorization without pivoting (#2640) (@amontoison)
• Wrap the Givens rotation methods (#2642) (@kshyatt)
• Remove kron methods and use those in GPUArrays (#2643) (@kshyatt)
• Add a simpler CuRefValue. (#2645) (@maleadt)
• Use GPUToolbox.jl (#2646) (@christiangnrd)
• DtoH copies: perform a nonblocking sync before calling into libcuda. (#2648) (@maleadt)
• Support Adjoint/Transpose -> COO (#2649) (@kshyatt)
• Support cuTENSOR contractors for 1D views (#2650) (@kshyatt)
• Re-enable mixed precision sparse mv (#2651) (@kshyatt)
• Proper support for similar on CuSparseMats (#2652) (@kshyatt)
• Test error throw for accumulate (#2656) (@kshyatt)
• Lots more tests for CUBLAS (#2657) (@kshyatt)
• MORE tests for CUBLAS and a bugfix (#2659) (@kshyatt)
• Add tests for gemmEx in fast math mode (#2660) (@kshyatt)
• More tests/better coverage for CUSPARSE (#2663) (@kshyatt)
• Fixes and tests for CuStateVec (#2664) (@kshyatt)
• Re-enable NVTX on Windows. (#2665) (@maleadt)
• Protect against occupancy calculations with very large numbers. (#2666) (@maleadt)
• Fixes and tests for COO indexing, exclude more kernels from coverage (#2668) (@kshyatt)
• Exclude lib*jl from coverage also for CUSTATEVEC, CUTENSOR, and CUTENSORNET (#2669) (@kshyatt)
• Even MORE tests and cov for CUBLAS (#2670) (@kshyatt)
• Fix and test for mgpu batch measure (#2671) (@kshyatt)
• Remove some invalid conversions and test more (#2673) (@kshyatt)
• Exclude more device side code in CUSPARSE (#2676) (@kshyatt)
• More tests, better errors, more exclusions for CUSPARSE (#2677) (@kshyatt)
• Try re-enabling the convolution tests (#2678) (@kshyatt)
• Fix Markdown formatting in overview.md (#2680) (@singularitti)
• Even more CUSPARSE tests (#2682) (@kshyatt)
• Fix inference of FFT plan creation (#2683) (@jipolanco)
• Some cudadrv tests (#2684) (@kshyatt)

Closed issues:

• Batched strided GEMM tests fail (#151)
• CuArrays.CURAND.curand missing methods (#141)
• Rationals behave badly (#118)
• Matrix inversion for CuArray (#116)
• Dot product of a complex CuArray with a real CuArray performance (#668)
• Sporadic cudnn/convolution test failures (#725)
• Support for LinearAlgebra.pinv (#883)
• Update mv!, mm!, sv! and sm! with the future release of CUPARSE (#1610)
• [CUSPARSE] changing size in similar returns a cpu array (#1667)
• Mix precision sparse mul is not dispatched correctly (#1760)
• Make CuRef(Value) behave more like Ref (#1803)
• [cuTENSOR] Issue when contracting views of CuArrays with cuTENSOR (#2407)
• versioninfo broken on Jetson Orin due to NVML lookup failure (#2542)
• CUBLAS: Improve concurrency using device pointer mode (#2571)
• NVML issues on Jetson Nano Orin (#2580)
• Passing Symbol as a an argument fails (#2590)
• Remove kron functionality (#2602)
• Disable or make automatic prefecthing of unified memory optional (#2618)
• Circular dependency in CUDA with Julia 1.10 (#2622)
• Regression with nsys profile and CUDA.@profile (#2629)
• PrecompileTools.jl with CUDA.jl causes kernels to fail to run on 1.11 (#2637)
• Support Adjoint Sparse Matrices for CuSparseMatrixCOO (#2647)
• Implicit stream sync in tasks serialise kernel execution (#2654)
• Broadcasting on arrays larger than typemax(Int32) yields truncation error (#2658)
• Problem with function in CUDA (#2667)
• CUDA.limit errors with invalid argument (code 1, ERROR_INVALID_VALUE) (#2672)
• CUDA.jl does not support tuples of UInt128 (#2675)
• Can not permutedims! CuArray with length larger that typemax(Int32) (#2679)
• Support for older GPUs (#2685)

v5.6.1

CUDA v5.6.1

Diff since v5.6.0

Merged pull requests:

• Support GPUArrays allocations cache (#2593) (@pxl-th)
• Fix resize! when pool=none is in use (#2613) (@luraess)
• Update to new alloc cache interface. (#2614) (@maleadt)
• Work around NVML issue on Jetson Orin. (#2620) (@maleadt)

Closed issues:

• Add strides, implement CUDA Array Interface (#1298)
• Restore broken CUBLAS test (#2584)
• Issues with multiple GPUs on a single node (#2615)

v5.6.0

CUDA v5.6.0

Diff since v5.5.2

CUDA.jl v5.6 is a relatively minor release, which the most important change being behind the scenes: GPUArrays.jl v11 has switched to KernelAbstractions.jl (#2524).

Features

• Update to CUDA 12.6.2 (#2512)
• CUSOLVER: support for Xgeev! (#2513), XsyevBatched (#2577), gesv! and gels! (#2406)
• CUBLAS: added multiplication of transpose / adjoint matrices by diagonal matrices (#2518, #2538)
• Improve handle cache performance in the presence of many short-lived tasks (#2583)
• CUFFT: Pre-allocate the buffer required for complex-to-real FFTs only once (#2578)
• Improved batched pointer conversion for very large batches (#2608)

Bug fixes

• Fix findall with an empty CuArray (#2554)
• CUBLAS: Fix use of level 1 methods with strided arrays (#2528)
• CUSOLVER: Fix Xgesvdr! (#2556)
• Preserve the array buffer type with more linear algebra operations (#2534)
  Work around LinearAlgebra.jl breakage in Julia 1.11.2 concerning generic triangular (l/r)mul! - (#2585)
• Fix ambiguity of LinearAlgebra.dot (#2569)
• Native RNG: Fixes when working with very large arrays (#2561)
• Avoid a deadlock due do union splitting in the mapreduce kernel (#2595)
• Fix pinning of resized CPU memory by automatically re-pinning (#2599)

Merged pull requests:

• [CUSOLVER] Interface gesv! and gels! (#2406) (@amontoison)
• Update wrappers for CUDA v12.6.2 (#2512) (@amontoison)
• [CUSOLVER] Interface Xgeev! (#2513) (@amontoison)
• Added multiplication of transpose / adjoint matrices by diagonal matrices (#2518) (@amontoison)
• CompatHelper: bump compat for GPUCompiler to 1, (keep existing compat) (#2521) (@github-actions[bot])
• Adapt to GPUArrays.jl transition to KernelAbstractions.jl. (#2524) (@maleadt)
• Switch CI to 1.11. (#2525) (@maleadt)
• CUTENSOR: Reduce amount of broadcasts compiled during tests. (#2527) (@maleadt)
• CUBLAS: Don't use BLAS1 wrappers for strided arrays, only vectors. (#2528) (@maleadt)
• Clarify the synchronize(ctx)/device_synchronize() docstrings (#2532) (@JamesWrigley)
• Issue #2533: Preserving the buffer type in linear algebra (#2534) (@kmp5VT)
• Clarify description of how LocalPreferences.toml is generated in the docs (#2535) (@glwagner)
• Adapt to JuliaGPU/GPUArrays.jl#567. (#2537) (@maleadt)
• Removed allocations for transpose/adjoint - diagonal multiplications (#2538) (@RedRussianBear)
• Consistent use of Nsight Compute (#2541) (@huiyuxie)
• Fix formatting in profiling docs page (#2543) (@efaulhaber)
• Fix typo in EnzymeCoreExt.jl (#2550) (@wsmoses)
• Enhance warning under a profiler (#2552) (@huiyuxie)
• Fix findall with an empty CuArray of Bool (#2554) (@amontoison)
• [CUSOLVER] Fix Xgesvdr! (#2556) (@amontoison)
• Test restore Enzyme.jl (#2557) (@wsmoses)
• Native RNG fixes for very large arrays (#2561) (@maleadt)
• [Enzyme] Mark launch_configuration as inactive (#2563) (@wsmoses)
• Update EnzymeCoreExt.jl (#2565) (@simenhu)
• Fix ambiguity of LinearAlgebra.dot (#2569) (@amontoison)
• [CUSOLVER] Add more tests for the dense SVD (#2574) (@amontoison)
• [CUSOLVER] Interface XsyevBatched (#2577) (@amontoison)
• [CUFFT] Preallocate a buffer for complex-to-real FFT (#2578) (@amontoison)
• Run the GC when failing to find a handle, but lots are active. (#2583) (@maleadt)
• Work around LinearAlgebra.jl breakage in 1.11.2. (#2585) (@maleadt)
• mapreduce: avoid deadlock by forcing the accumulator type. (#2596) (@maleadt)
• Switch to GitHub Actions-based benchmarks. (#2597) (@maleadt)
• Re-pin variable sized memory (#2599) (@jipolanco)
• Enzyme: add make_zero of cuarrays (#2600) (@wsmoses)
• Update cache.jl (#2604) (@jarbus)
• Enzyme: mark device_sync as non-differentiable [only downstream] (#2605) (@wsmoses)
• Move strided batch pointer conversion to GPU (#2608) (@THargreaves)
• Split linalg tests into multiple files (#2609) (@kshyatt)

Closed issues:

• Inference failure with sort(::CuMatrix) after loading MLDatasets (#2258)
• Kron Support for CuSparseMatrixCSC (#2370)
• Broadcasting a function returning an anonymous function with a constructor over CUDA arrays fails to compile, "not isbits" (#2514)
• CuArray view has different variable type outside x inside the cuda kernel (#2516)
• Can't build cuDNN on centos7.8 (#2517)
• Precompile errors (#2519)
• Precompile errors (#2520)
• Error returned from CUDA function in CUDA-aware MPI multi-GPU test (#2522)
• Broadcasting over random static array errors on Julia 1.11 (#2523)
• gemm_strided_batched only using strided CUDA kernel when first matrix is transposed (#2529)
• CUDA runtime libraries are loaded from a system path due to LD_LIBRARY_PATH being set (#2530)
• [Bug] UnifiedMemory buffer changes during LinearAlgebra operations (#2533)
• Improve system library warning when running under profiler (#2540)
• Local CUDA settings not propagated to Pkg.test (#2545)
• Out of Memory when working with Distributed for Small Matricies (#2548)
• findall is not working with an empty vector of bool (#2553)
• CUDA code does not return when running under VSC Debugging mode (#2558)
• dot is quite slow in multinest Arrays (#2559)
• UndefVarError: backend not defined in GPUArrays (#2564)
• view() returns CuArray instead of view for 1-D CuArrays (#2566)
• dot ambiguity (#2568)
• InvalidIRError thrown only if critical function is not previously compiled (#2573)
• circular dependency during precompilation (#2579)
• Sparse MatVec Is Nondeterministic? (#2582)
• CUDA triggers long Circular dependency list (#2586)
• Release v5.5.3 for GPUArray v11? (#2587)
• 'dot' gives different answers when viewing rather than slicing multidimensional arrays (#2589)
• Scalar indexing when performing kron on two CuVectors (#2591)
• Faster strided-batched to batched wrapper (#2592)
• Error when copying data to pinned and resized CPU array (#2594)
• mapreducedim! size-dependent fail when narrowing float element types (#2595)
• Missing Enzyme.make_zero in Enzyme extension leads to incorrect behaviour (#2598)
• 'ArgumentError: array must be non-empty' when attempting to pop idle handles from HandleCache (#2603)
• Do a release as current one doesn't support GPUArrays v11 (#2606)

v5.5.2

CUDA v5.5.2

Diff since v5.5.1

Merged pull requests:

• Fix type of AbstractFFTs.Plan for real-complex FFTs (#2504) (@jipolanco)
• Profiler: Demangle kernel names. (#2505) (@maleadt)
• Bump CUDNN. (#2507) (@maleadt)
• Restore Enzyme checks (#2508) (@wsmoses)