10 Pull requests merged by 7 people

• Remove the unnecessary reshape during mapreduce.

  #2778 merged May 12, 2025

• cuTENSOR: Preserve storage type when multiplying

  #2775 merged May 9, 2025

• Fix SPGEMM_ALGOS setup

  #2773 merged May 8, 2025

• SparseMatricesCSR Dispatch

  #2720 merged May 8, 2025

• Update to CUDA 12.9.

  #2772 merged May 8, 2025

• Support new functionality from KA 0.9.32

  #2774 merged May 8, 2025

• unsafe_wrap for symbols

  #2753 merged May 8, 2025

• Remove second import of aligned_sizeof

  #2767 merged May 7, 2025

• CUSPARSE SpGEMM: Support algorithms 2 and 3

  #2769 merged May 7, 2025

• Get rid of unneeded version checks

  #2765 merged May 7, 2025

2 Pull requests opened by 2 people

• Update subpackages.

  #2776 opened May 9, 2025

• add KA.get_backend(dev)

  #2779 opened May 12, 2025

3 Issues closed by 1 person

• `sum!` throws dispatch error beyond a threshold number of rows

  #2777 closed May 12, 2025

• aligned_sizeof with an existing identifier

  #2766 closed May 7, 2025

• CUSPARSE_SPGEMM_ALG2 not working

  #2768 closed May 7, 2025

1 Issue opened by 1 person

• Error in testset libraries/cublas/level3/gemm and core/device/intrinsics/atomics

  #2770 opened May 5, 2025

48 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Prefix scan using warp primitives and fallback for non primitive types

  #287 commented on May 6, 2025 • 0 new comments

• Constant memory support

  #552 commented on May 6, 2025 • 0 new comments

• Add GPU to CPU RPC mechanism

  #567 commented on May 6, 2025 • 0 new comments

• add custom sum tutorial

  #666 commented on May 6, 2025 • 0 new comments

• Implement sync_threads using an unaligned barrier.

  #798 commented on May 6, 2025 • 0 new comments

• Add support for strided cufft

  #912 commented on May 6, 2025 • 0 new comments

• updating docs to include matrix-vector multiply example

  #918 commented on May 6, 2025 • 0 new comments

• Latency improvements.

  #1066 commented on May 6, 2025 • 0 new comments

• Register BFloat16

  #1092 commented on May 6, 2025 • 0 new comments

• Fixed reducedim operations for CuQRPackedQ

  #1118 commented on May 6, 2025 • 0 new comments

• Add a hostcall interface

  #1140 commented on May 6, 2025 • 0 new comments

• Calling similar(...) on AbstractCuSparseArray with dims

  #1184 commented on May 6, 2025 • 0 new comments

• 2-arg `show`

  #1197 commented on May 6, 2025 • 0 new comments

• improve sparse matrix conversions

  #1215 commented on May 6, 2025 • 0 new comments

• [POC] Support for cuFile.

  #1235 commented on May 6, 2025 • 0 new comments

• Add explicit strides to CuArray and CuDeviceArray.

  #1322 commented on May 6, 2025 • 0 new comments

• Support ordering argument for atomics

  #1393 commented on May 6, 2025 • 0 new comments

• WMMA TensorFloat32 (TF32)

  #1419 commented on May 6, 2025 • 0 new comments

• WMMA BFloat16 (BF16)

  #1425 commented on May 6, 2025 • 0 new comments

• WMMA Float64

  #1426 commented on May 6, 2025 • 0 new comments

• CUSPARSE: Better error msg for unsupported sparse mm

  #1467 commented on May 6, 2025 • 0 new comments

• limit csc/csr/bsr sparse conversion index to be cint & fix a few conversion bugs

  #1563 commented on May 6, 2025 • 0 new comments

• Add scoped atomic_thread_fence

  #1644 commented on May 6, 2025 • 0 new comments

• Switch secrets management to cryptic.

  #1687 commented on May 6, 2025 • 0 new comments

• Support for qr of strided inputs (non-contiguous views)

  #1764 commented on May 6, 2025 • 0 new comments

• Adding copyto for non-contiguous matrices and vectors

  #1778 commented on May 6, 2025 • 0 new comments

• Use Atomix

  #1790 commented on May 6, 2025 • 0 new comments

• Add wrappers for NVPERF

  #1823 commented on May 6, 2025 • 0 new comments

• Attempted to add method definitions for supporting the SubArray type …

  #1830 commented on May 6, 2025 • 0 new comments

• Add an experimental opaque closure type.

  #1853 commented on May 6, 2025 • 0 new comments

• WIP: Add an index typevar to CuDeviceArray.

  #1895 commented on May 6, 2025 • 0 new comments

• Add contract through FastmathOverlays.jl

  #2037 commented on May 6, 2025 • 0 new comments

• Support FFT adjoint plans and test

  #2073 commented on May 6, 2025 • 0 new comments

• Use TaskLocalValues

  #2075 commented on May 6, 2025 • 0 new comments

• docs: perf tips: deemphasize `assume` in favor of UnsafeAssume.jl

  #2181 commented on May 6, 2025 • 0 new comments

• Add a dispatch for LinearAlgebra.norm2

  #2302 commented on May 6, 2025 • 0 new comments

• Use PrecompileTools to warmup CUDA.jl

  #2325 commented on May 6, 2025 • 0 new comments

• High Level Wrapper for Fused Matmul + Bias + Activation

  #2360 commented on May 6, 2025 • 0 new comments

• WIP: Native I/O.

  #2485 commented on May 6, 2025 • 0 new comments

• [CUSPARSE] Fix constructor of sparse empty matrices

  #2575 commented on May 6, 2025 • 0 new comments

• Directed rounding

  #2576 commented on May 6, 2025 • 0 new comments

• make CUDA randn work with Zygote

  #2581 commented on May 6, 2025 • 0 new comments

• Allow disabling the linking of libdevice in CUDACompilerParams

  #2611 commented on May 6, 2025 • 0 new comments

• Try fast linear indexes for KA

  #2612 commented on May 6, 2025 • 0 new comments

• Wrap and test some more Float16 intrinsics

  #2644 commented on May 6, 2025 • 0 new comments

• Use invariant.load for ldg

  #2655 commented on May 6, 2025 • 0 new comments

• Added new api and fixed type errors in cuStateVec

  #2728 commented on May 6, 2025 • 0 new comments

• add fastmath flag

  #2732 commented on May 6, 2025 • 0 new comments