v0.16.3 - cuda 12.9 support

See cuda release notes here: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

What's Changed

• Fix apparent use-after-free in nvrtc. by @caelunshun in #402
• Add support for cuda 12.9 by @coreylowman in #406

New Contributors

• @caelunshun made their first contribution in #402

Full Changelog: v0.16.2...v0.16.3

v0.16.2 - safe Unified memory API

What's Changed

• Adds UnifiedSlice - safe unified memory api by @coreylowman in #400

Full Changelog: v0.16.1...v0.16.2

v0.16.1 - Windows dynamic Loading fixes & better error handling in Drop

What's Changed

• Add Send & Sync for PinnedHostSlice by @torymur in #394
• Ignore cuCheckpoint api in bindings by @coreylowman in #398
• Save errors in Drop impls instead of .unwrap() by @coreylowman in #399

New Contributors

• @torymur made their first contribution in #394

Full Changelog: v0.16.0...v0.16.1

v0.16.0 - small cublaslt breaking change & allow disabling event tracking

What's Changed

• Add CudaContext::disable_event_tracking - a way to disable CudaEvent tracking for CudaSlice by @coreylowman in #391
• Expose cublaslt transc by @EricLBuehler in #390

New Contributors

• @EricLBuehler made their first contribution in #390

Full Changelog: v0.15.2...v0.16.0

v0.15.2 - optimizations for single stream use cases (most people)

What's Changed

• driver/safe test: fix 20% relative tolerance instead of absolute by @jedbrown in #384
• Only doing event recording if multiple streams are created by @coreylowman in #387
• Check ctx::get_current() before calling ctx::set_current() by @coreylowman in #388

New Contributors

• @jedbrown made their first contribution in #384

Full Changelog: v0.15.1...v0.15.2

v0.15.1 - Fixes for extraneous stream.synchronizations with host copies & cuSparse/cuSolver sys apis

What's Changed

• Add cuSparse and cuSolver unsafe FFI bindings (after v0.15 merge) by @ajz34 in #381
• Remove stream synchronizations with HostSlice by @coreylowman in #383

New Contributors

• @ajz34 made their first contribution in #381

Full Changelog: v0.15.0...v0.15.1

v0.15.0 - Support static-linking, dynamic-linking, and dynamic-loading

What's Changed

This is only breaking release if you depend on sys level APIs. No result or safe apis have been broken in this release

The breaking changes are that we are going back to the format of calling sys::<fn name>() rather than sys::lib().<fn name>().

Wonderful work by @Narsil to support all of this 🎉

• Rebased patch. by @Narsil in #371
• Fixes CudaSlice::leak by @coreylowman in #373
• Fixing missing symbols in static-linking by @coreylowman in #378
• Adding blocklist options for bindgen for curand/cublaslt by @coreylowman in #379
• Update bindings generation docs by @coreylowman in #380

Full Changelog: v0.14.0...v0.15.0

v0.14.0

What's Changed

See PRs for more info

• Export PinnedHostSlice as pub use by @aseyboldt in #339
• [Breaking] Add CudaContext, CudaEvent. Refactor CudaStream & launching by @coreylowman in #341
• [Breaking] Remove extra shared mem arg in occupancy functions by @coreylowman in #342
• [Breaking] Creating CudaRng requires CudaContext instead of CudaDevice by @coreylowman in #343
• CudaRng constructor now requires CudaStream by @coreylowman in #345
• [Breaking] nccl::Comm now constructed with CudaStream by @coreylowman in #346
• [Breaking] CudaBlas constructor requires CudaStream by @coreylowman in #347
• [Breaking] CudaBlasLT constructor requires CudaStream by @coreylowman in #348
• [Breaking] Cudnn now requires CUdaStream in constructor by @coreylowman in #349
• Changing stream ref methods to return &Arc<CudaStream> by @coreylowman in #350
• [Breaking] Remove CudaDevice by @coreylowman in #351
• CudaModule doesn't load CudaFunction on creation anymore by @coreylowman in #352
• Remove dependency on spin for no-std builds by @coreylowman in #353
• Reorg driver files by @coreylowman in #354
• [Breaking] View clones now maintain lifetimes instead of creating new lifetimes by @coreylowman in #355
• [Breaking] Adds name as option in nvrtc::CompileOptions by @coreylowman in #356
• Adding CudaGraph api by @coreylowman in #357
• Update docs by @coreylowman in #358
• [Breaking] Require stream synchronization to access device ptrs by @coreylowman in #359
• Adds SyncOnDrop & DevicePtr/Mut now return (CUdeviceptr, SyncOnDrop) by @coreylowman in #360
• PushKernelArg for scalar now takes reference by @coreylowman in #363

New Contributors

• @aseyboldt made their first contribution in #339

Full Changelog: v0.13.9...v0.14.0

v0.13.9 - fix for downstream crates building doc.rs

What's Changed

• make build.rs check if we are building for docs.rs by @jonaspleyer in #338

New Contributors

• @jonaspleyer made their first contribution in #338

Full Changelog: v0.13.8...v0.13.9

v0.13.8 - Add pinned memory apis

What's Changed

• Adds PinnedHostSlice to support pinned memory copies by @coreylowman in #336

Full Changelog: v0.13.7...v0.13.8