All pull requests

Pull requests

• test(cuda.core): close leftover owned pools and run the teardown gc.collect() only for pool-owning tests

  CI/CDcuda.coreenhancementP0test
  #3045·Andy-Jost opened Oct 7, 2026Contributor·2 tasks done·

  ·

  · cuda.core 1.3.0
  6

  

• test(cuda.core): fold the pairwise cross-type object-protocol tests into one equality test

  cuda.coreenhancementP1
  #3044·Andy-Jost opened Oct 7, 2026Contributor·2 tasks done·

  ·

  · cuda.core 1.3.0
  1

  

• test(cuda.core): run the IPC modules and two slow modules single-threaded under pytest-run-parallel

  cuda.coreenhancementP1
  #3043·Andy-Jost opened Oct 7, 2026Contributor·2 tasks done·

  ·

  · cuda.core 1.3.0
  2

  

• test(cuda.core): compile the scalar-argument kernel once per module in test_launcher.py

  cuda.coreenhancementP1
  #3042·Andy-Jost opened Oct 7, 2026Contributor·2 tasks done·

  ·

  · cuda.core 1.3.0
  1

  

• test(cuda.core): freeze the collected heap once so the per-test gc.collect() stops walking it

  cuda.coreenhancementP0
  #3041·Andy-Jost opened Oct 7, 2026Contributor·2 tasks done·

  ·

  · cuda.core 1.3.0
  4

  

• Avoid nvJitLink cache crashes during incremental LTO

  awaiting-responsecuda.core
  #3040·GodlyDonuts opened Oct 7, 2026·

  ·
  3

  

• ci: retry artifact downloads once on transient infra failures

  CI/CD
  #3037·juenglin opened Oct 6, 2026Contributor·1 of 2 tasks·

  ·

  · cuda.core 1.3.0
  8

  

• cuda.core: narrow the memcpy update release note and test the host-operand boundary

  cuda.coredocumentationP1
  #3035·Andy-Jost opened Oct 6, 2026Contributor·

  ·

  · cuda.core 1.3.0
  1

  4

  

• Add a snippet-based release note approach

  CI/CDcuda.bindingscuda.corecuda.pathfinderPR review get-together
  #3034·mdboom opened Oct 6, 2026Contributor·

  ·

  · cuda.core next
  5

  

• cuda.core: experiment with torch 2.14+ stable PyObject->Tensor shim

  cuda.core
  #3033·leofang opened Oct 6, 2026Member·

  ·
  1

• Bump nightly PyTorch highest version to 2.14.1

  CI/CDcuda.core
  #3031·leofang opened Oct 6, 2026Member·

  ·

  · cuda.core 1.3.0
  4

  

• cuda.core: share VirtualMemoryResource buffers between processes

  cuda.corefeatureP0
  #3028·Andy-Jost opened Oct 6, 2026Contributor·

  ·

  · cuda.core 1.3.0
  8

  

• [WoA] Automate public cross-repository dispatch

  CI/CD
  #3026·isVoid opened Oct 5, 2026Contributor·5 of 9 tasks·

  ·
  1

• [WoA] Add manual cross-repository broker

  CI/CDfeature
  #3025·isVoid opened Oct 5, 2026Contributor·5 of 6 tasks·

  ·

  · cuda.core next
  5

  

• ci: switch to mozilla/sccache 0.18 and drop obsolete ACTIONS_ vars

  CI/CDenhancement
  #3024·juenglin opened Oct 5, 2026Contributor·2 tasks done·

  ·

  · cuda.core 1.3.0
  5

  

• chore: refresh pixi lockfiles

  CI/CDdependencies
  #3018·github-actions[bot] opened Oct 5, 2026Bot·

  ·

  · cuda.core 1.3.0
  2

  

• Remove xfail_if_mempool_oom and xfail_on_graph_mempool_oom helpers

  cuda.bindingscuda.coretest
  #3004·juenglin opened Oct 2, 2026Contributor·

  ·

  · cuda.core 1.3.0
  5

  

• ci: check documentation links on PRs and nightly with a shared cache

  CI/CDcuda.core
  #2993·rwgk opened Oct 1, 2026Contributor·2/3·

  ·

  · cuda.core next
  12

  

• test: skip CUDA-GL interop tests when the GL context is not on an NVIDIA GPU

  cuda.bindingscuda.coretest
  #2986·IvanGrigorik opened Oct 1, 2026·4 tasks done·

  ·

  · cuda.core 1.3.0
  1

  5

  

• [no-ci] Review-only: six layers of PR 2737

  CI/CDcuda.bindingscuda.core
  #2960·rwgk opened Sep 29, 2026Contributor·

  ·

  · cuda.bindings 13.5.0 & 12.9.10
  1

  

• Migrate samples

  CI/CDcuda.bindingscuda.corecuda.pathfinderexampleP0
  #2959·danielfrg opened Sep 28, 2026Contributor·1 of 2 tasks·

  ·

  · cuda.bindings 13.5.0 & 12.9.10
  1

  5

  

• feat(pathfinder): opt-in diagnostic logging for load_nvidia_dynamic_lib

  cuda.pathfinder
  #2951·u7k4rs6 opened Sep 25, 2026·1 of 2 tasks·

  ·
  1

  4

• Vendor CCCL and use it to reactivate tensormap

  CI/CDcuda.coreP1
  #2940·seberg opened Sep 23, 2026Contributor·2 tasks done·

  ·

  · cuda.core 1.3.0
  2

  

• Add host-callback contract tests from the stream-callback survey

  cuda.core
  #2925·0z5a opened Sep 20, 2026·

  ·
  4

• cuda.core: add a bounded asyncio wait for NVML device and system events

  bugcuda.coreenhancement
  #2924·0z5a opened Sep 19, 2026·

  ·

  · cuda.core 1.3.0
  31