v0.30.0

What's Changed

• Test cuDF third party test test_groupby_apply_return_reindexed_series by @mroeschke in #823
• Move cuda-pathfinder dependency to core by @ZzEeKkAa in #835
• Fix cache invalidation logic. by @tpn in #800
• Implement launch config infrastructure. by @tpn in #804
• Use cuda-pathfinder to locate nvdisasm by @Jlisowskyy in #842
• Fix CABI calling convention: skip env global and escape mangled names by @isVoid in #844
• Bump Version to 0.30.0 by @isVoid in #848

New Contributors

• @mroeschke made their first contribution in #823

Full Changelog: v0.29.0...v0.30.0

v0.29.0

What's Changed

• Extend dbg.value coverage to loadvar for scalar kernel parameters by @jiel-nv in #813
• Fix FP8 uint64 cast flake on Windows by @cpcloud in #829
• Use dbg.declare for scalar kernel parameters by @cpcloud in #828
• Fix mixed-IR liveness for inline overload DCE by @cpcloud in #795
• Use cuda-python for nvvm bindings by @brandon-b-miller in #818
• fix(ci): cudaRoundMode typing failure in FP8 test by @kaeun97 in #834
• Support cuda_bindings FastEnum by @mdboom in #837
• Support cuda.core.GraphBuilder as a kernel-launch stream by @Andy-Jost in #836
• fix: normalize numpy integer types to python int to prevent overflow errors by @kaeun97 in #774
• Bump Version to 0.29.0 by @isVoid in #838

New Contributors

• @Andy-Jost made their first contribution in #836

Full Changelog: v0.28.2...v0.29.0

v0.28.2

What's Changed

• build(deps): bump the actions-monthly group with 2 updates by @dependabot[bot] in #815
• Swallow DLPack conversion error by @leofang in #825
• Fix empty DW_AT_location for reused loop variables by @jiel-nv in #811
• Bump version to 0.28.2 by @brandon-b-miller in #827

Full Changelog: v0.28.1...v0.28.2

v0.28.1

What's Changed

• Support Interop with External Objects Via DLPack by @isVoid in #790
• skip flaky test for now by @leofang in #820
• Bump version to 0.28.1 by @leofang in #821

Full Changelog: v0.28.0...v0.28.1

v0.28.0

What's Changed

• Fix CI checks job to also fail on dependency failures by @kkraus14 in #777
• Remove enums and unused ctypes code that required it by @brandon-b-miller in #775
• Clear rtsys during context reset by @brandon-b-miller in #783
• Fix pnputil to only restart NVIDIA display adapters by @kkraus14 in #779
• Bump cudf version to 26.02 in thirdparty tests by @brandon-b-miller in #785
• Remove unused cudf patch by @gmarkall in #786
• Fix np.dtype overload signature drift by @cpcloud in #797
• Use only public .handle to access cuda-core objects by @brandon-b-miller in #794
• Remove most of drvapi.py in favor of direct cuda-python usage by @brandon-b-miller in #784
• Find CUDA headers via pathfinder by @brandon-b-miller in #771
• Fix incorrect DWARF type encodings for i8 and discriminator types by @jiel-nv in #806
• Fix int64 elements DWARF encoding in UniTuple types by @jiel-nv in #807
• Unpin nvjitlink from ctk version by @ZzEeKkAa in #809
• Disable legilize return type for inline always by @ZzEeKkAa in #812
• Fix mixed C ABI / Numba ABI internal-call lowering (#781, #789) by @isVoid in #782
• Aligned NumPy dtypes caching by @Jlisowskyy in #792
• fix(errors): restore CUDA exception hierarchy to avoid slow string compilation by @cpcloud in #796
• Revert exception hierarchy unification (#634) and its fix (#796) by @cpcloud in #816
• Bump version to 0.28.0 by @gmarkall in #817

New Contributors

• @Jlisowskyy made their first contribution in #792

Full Changelog: v0.27.0...v0.28.0

v0.27.0

What's Changed

• remove super args by @cpcloud in #763
• test(refactor): clean up run_in_subprocess by @cpcloud in #762
• Disable automatic review trigger for Greptile by @gmarkall in #743
• Enable apt proxy caching; skip hosted Windows builds by @kkraus14 in #766
• build(deps): bump actions/setup-python from 6.1.0 to 6.2.0 in the actions-monthly group across 1 directory by @dependabot[bot] in #768
• Add cuda-core to oldest tests by @brandon-b-miller in #769
• Generate line info for PHI exporters in terminator block by @jiel-nv in #756
• Move CallConv from CUDAContext to FunctionDescriptor by @isVoid in #717
• feat: Add documentation for debugging Numba CUDA programs with CUDA GDB and VSCode by @mmason-nvidia in #665
• Remove unused rtapi.py by @brandon-b-miller in #773
• fix: fix boolean return type mismatch in C ABI wrapper by @kaeun97 in #770
• Add CUDA FP8 type + conversion bindings (E5M2/E4M3/E8M0), HW-accel detection, and comprehensive tests by @isVoid in #686
• Bump version to 0.27.0 by @gmarkall in #776

Full Changelog: v0.26.0...v0.27.0

v0.26.0

What's Changed

• Eliminate duplicate DWARF entries for boolean kernel parameters by @jiel-nv in #749
• feat: accept cuda.core.Buffer and cuda.core.utils.StridedMemoryView as kernel inputs by @cpcloud in #751
• Replace legacy wheels-build.yaml with build-wheel.yml in publish workflow [no-ci] by @kkraus14 in #760
• MatMul test: Move from unittest to fully pytest by @maifeeulasad in #754
• ci: move benchmarks to single function calls so that the units and results are easier to interpret by @cpcloud in #759
• CI cleanup, extend no-ci skip logic, and add GitHub Release uploads by @kkraus14 in #761
• bump pixi version and relock by @cpcloud in #757
• Bump version to 0.26.0 by @kkraus14 in #764

New Contributors

• @maifeeulasad made their first contribution in #754

Full Changelog: v0.25.0...v0.26.0

v0.25.0

What's Changed

• build(deps): bump the actions-monthly group across 1 directory with 8 updates by @dependabot[bot] in #704
• chore(dev): build pixi using rattler by @cpcloud in #713
• [feat] Initial version of the Numba CUDA GDB pretty-printer by @mmason-nvidia in #692
• revert: chore(dev): build pixi using rattler (#713) by @cpcloud in #719
• Fix DISubprogram line number to point to function definition line by @jiel-nv in #695
• chore(deps): regenerate pixi lockfile by @cpcloud in #722
• Disable per-PR nvmath tests + follow same test practice by @leofang in #723
• Adding pixi run test and pixi run test-par support by @rparolin in #724
• CI: Add CUDA 13.1 testing support by @Copilot in #705
• Use pathfinder for dynamic libraries by @brandon-b-miller in #308
• ci: remove rapids containers from conda ci by @cpcloud in #737
• Pass the -numba-debug flag to libnvvm by @mmason-nvidia in #681
• Fix compatibility with NumPy 2.4: np.trapz and np.in1d removed by @kkraus14 in #739
• feat: users can pass shared_memory_carveout to @cuda.jit by @kaeun97 in #642
• ci: run tests in parallel by @cpcloud in #740
• fix: Fix race condition in CUDA Simulator by @ccam80 in #690
• fix: enable flake8-bugbear lints and fix found problems by @cpcloud in #708
• chore(deps): add cuda-pathfinder to pixi deps by @cpcloud in #741
• Fix: Pass correct flags to linker when debugging in the presence of LTOIR code by @mmason-nvidia in #698
• Fix missing line info in Jupyter notebooks by @jiel-nv in #742
• Fix kernel return type in DISubroutineType debug metadata by @jiel-nv in #745
• Fix prologue debug line info pointing to decorator instead of def line by @jiel-nv in #746
• Fix max block size computation in forall by @brandon-b-miller in #744
• feat: swap out internal device array usage with StridedMemoryView by @cpcloud in #703
• Add Python 3.14 to the wheel publishing matrix by @gmarkall in #750

New Contributors

• @mmason-nvidia made their first contribution in #692
• @ccam80 made their first contribution in #690

Full Changelog: v0.24.0...v0.25.0

v0.24.0

What's Changed

• Set up a new VM-based CI infrastructure by @leofang in #604
• chore(dev-deps): remove ipython and pyinstrument by @cpcloud in #670
• Use rapidsai/sccache in CI by @trxcllnt in #674
• Remove dangling references to NUMBA_CUDA_ENABLE_MINOR_VERSION_COMPATIBILITY by @brandon-b-miller in #675
• Drop experimental from cuda.core namespace imports by @brandon-b-miller in #676
• Remove customized address space tracking and address class emission in debug info by @jiel-nv in #669
• Support python 3.14 by @brandon-b-miller in #599
• fix: use freethreading-supported _PySet_NextItemRef where possible by @cpcloud in #682
• chore(deps): bump deps in pixi lockfile by @cpcloud in #693
• chore: perf lint by @cpcloud in #697
• perf: let CAI fall through instead of calling from_cuda_array_interface by @cpcloud in #694
• perf: remove some exception control flow and buffer-exception penalization for arrays by @cpcloud in #700
• Fix test_wheel_deps_wheels.sh to actually uninstall nvvm and nvrtc packages for CUDA 13 by @brandon-b-miller in #701
• Dropping bits in the old CI & Propagating recent changes from cuda-python by @leofang in #683
• chore(deps): bump numba-cuda version and relock pixi by @cpcloud in #707
• ci: remove redundant conda build in ci by @cpcloud in #711
• ci: relock pixi by @cpcloud in #712
• chore: disable locked flag to bypass prefix-dev/pixi#5256 by @cpcloud in #714
• Add arch specific target support by @ZzEeKkAa in #549

New Contributors

• @trxcllnt made their first contribution in #674

Full Changelog: v0.23.0...v0.24.0

v0.23.0

What's Changed

• refactor: remove devicearray code to reduce complexity by @cpcloud in #600
• chore: bump version in pixi.toml by @cpcloud in #641
• feat: add set_shared_memory_carveout by @kaeun97 in #629
• Migrate numba-cuda driver to use cuda.core.launch API by @rparolin in #609
• refactor: cull dead linker objects by @cpcloud in #649
• Add support for dependabot by @jpascucci-nv in #647
• Fix false negative NRT link decision when NRT was previously toggled on by @brandon-b-miller in #650
• test: fix bogus self argument to Context by @cpcloud in #656
• Only run dependabot monthly and open fewer PRs by @jpascucci-nv in #658
• feat: add print support for int64 tuples by @cpcloud in #663
• Do not manually set DUMP_ASSEMBLY in nvjitlink tests by @brandon-b-miller in #662
• Test RAPIDS 25.12 by @brandon-b-miller in #661
• build(deps): bump actions/upload-artifact from 4 to 5 by @dependabot[bot] in #652
• build(deps): bump actions/setup-python from 5.6.0 to 6.1.0 by @dependabot[bot] in #655
• feat: allow printing nested tuples by @cpcloud in #667
• Fix Issue #588: separate compilation of NVVM IR modules when generating debuginfo by @gmarkall in #591
• Fix #624: Accept Numba IR nodes in all places Numba-CUDA IR nodes are expected by @gmarkall in #643
• Capture global device arrays in kernels and device functions by @shwina in #666

New Contributors

• @jpascucci-nv made their first contribution in #647
• @dependabot[bot] made their first contribution in #652
• @shwina made their first contribution in #666

Full Changelog: v0.22.1...v0.23.0