NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI · #409 most downloaded on PyPI
compiling Python code using LLVM
Last release 4 days ago
30 Sep 2026
Ships fairly regularly
a new release about every 3 weeks
Most releases are documented
notes for 43 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
14 years old
138 releases · first in 2012
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
One column per quarter.
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
PR #8964 _: fix missing nopython keyword in cuda random module (esc _)
Pull-Requests:
PR #8964: fix missing nopython keyword in cuda random module (esc)
PR #8965: fix return dtype for np.angle (guilhermeleobas esc)
PR #8982: Don't do the parfor diagnostics pass for the parfor gufunc. (DrTodd13)
PR #8958: resurrect the import, this time in the registry initialization (esc)
PR #8947: Introduce internal _isinstance_no_warn (guilhermeleobas esc)
PR #8895: CUDA: Enable caching functions that use CG (gmarkall)
PR #8976: Fix index URL for ptxcompiler/cubinlinker packages. (bdice)
PR #9004: Skip MVC test when libraries unavailable (gmarkall esc)
PR #9006: link to version support table instead of using explicit versions (esc)
PR #9005: Fix: Issue #8923 - avoid spurious device-to-host transfers in CUDA ufuncs (gmarkall)
Authors:
Nothing published for this version
Please note that this release contains a significant number of both deprecation and pending-deprecation notices with view of making it easier to devel…
This release continues to add new features, bug fixes and stability improvements to Numba. Please note that this release contains a significant number of both deprecation and pending-deprecation notices with view of making it easier to develop new technology for Numba in the future. Also note that this will be the last release to support Windows 32-bit packages produced by the Numba team.
Highlights of core dependency upgrades:
Support for Python 3.11 (minimum is moved to 3.8)
Support for NumPy 1.24 (minimum is moved to 1.21)
Python language support enhancements:
Exception classes now support arguments that are not compile time constant.
The built-in functions hasattr and getattr are supported for compile time constant attributes.
The built-in functions str and repr are now implemented similarly to their Python implementations. Custom __str__ and __repr__ functions can be associated with types and work as expected.
Numba's unicode functionality in str.startswith now supports kwargs start and end.
min and max now support boolean types.
Support is added for the dict(iterable) constructor.
NumPy features/enhancements:
The largest set of new features is within the numpy.random.Generator support, the vast majority of commonly used distributions are now supported. Namely:
Generator.beta
Generator.chisquare
Generator.exponential
Generator.f
Generator.gamma
Generator.geometric
Generator.integers
Generator.laplace
Generator.logistic
Generator.lognormal
Generator.logseries
Generator.negative_binomial
Generator.noncentral_chisquare
Generator.noncentral_f
Generator.normal
Generator.pareto
Generator.permutation
Generator.poisson
Generator.power
Generator.random
Generator.rayleigh
Generator.shuffle
Generator.standard_cauchy
Generator.standard_exponential
Generator.standard_gamma
Generator.standard_normal
Generator.standard_t
Generator.triangular
Generator.uniform
Generator.wald
Generator.weibull
Generator.zipf
The nbytes property on NumPy ndarray types is implemented.
Nesting of nested-array types is now supported.
datetime and timedelta types can be cast to int.
F-order iteration is supported in ufunc generation for increased performance when using combinations of predominantly F-order arrays.
The following functions are also now supported:
np.argpartition
np.isclose
np.nan_to_num
np.new_axis
np.union1d
Highlights of core changes:
A large amount of refactoring has taken place to convert many of Numba's internal implementations, of both Python and NumPy functions, from the low-level extension API to the high-level extension API (numba.extending).
The __repr__ method is supported for Numba types.
The default target for applicable functions in the extension API (numba.extending) is now "generic". This means that @overload* and @intrinsic functions will by default be accepted by both the CPU and CUDA targets.
The use of __getitem__ on Numba types is now supported in compiled code. i.e. types.float64[:, ::1] is now compilable.
Performance:
The performance of str.find() and str.rfind() has been improved.
Unicode support for __getitem__ now avoids allocation and returns a view.
The numba.typed.Dict dictionary now accepts an n_keys option to enable allocating the dictionary instance to a predetermined initial size (useful to avoid resizes!).
The Numba Run-time (NRT) has been improved in terms of performance and safety:
The NRT internal statistics counters are now off by default (removes atomic lock contentions).
Debug cache line filling is off by default.
The NRT is only compiled once a compilation starts opposed to at function decoration time, this improves import speed.
The NRT allocation calls are all made through a "checked" layer by default.
CUDA:
New NVIDIA hardware and software compatibility / support:
Toolkits: CUDA 11.8 and 12, with Minor Version Compatibility for 11.x.
Packaging: NVIDIA-packaged CUDA toolkit conda packages.
Hardware: Hopper, Ada Lovelace, and AGX Orin.
float16 support:
Arithmetic operations are now fully supported.
A new method, is_fp16_supported(), and device property, supports_float16, for checking the availability of float16 support.
Functionality:
The high-level extension API is now fully-supported in the CUDA target.
Eager compilation of multiple signatures, multiple outputs from generalized ufuncs, and specifying the return type of ufuncs are now supported.
A limited set of NumPy ufuncs (trigonometric functions) can now be called inside kernels.
Lineinfo quality improvement: enabling lineinfo no longer results in any changes to generated code.
Deprecations:
The numba.pycc module and everything in it is now pending deprecation.
The long awaited full deprecation of object mode fall-back is underway. This change means @jit with no keyword arguments will eventually alias @njit.
The @generated_jit decorator is deprecated as the Numba extension API provides a better supported superset of the same functionality, particularly through @numba.extending.overload.
Version support/dependency changes:
The setuptools package is now an optional run-time dependency opposed to a required run-time dependency.
The TBB threading-layer now requires version 2021.6 or later.
LLVM 14 is now supported on all platforms via llvmlite.
Pull-Requests:
PR #5113: Fix error handling in the Interval extending example (esc eric-wieser)
PR #7067: Implement np.isclose (guilhermeleobas)
PR #7255: CUDA: Support CUDA Toolkit conda packages from NVIDIA (gmarkall)
PR #7622: Support fortran loop ordering for ufunc generation (sklam)
PR #7733: fix for /tmp/tmp access issues (ChiCheng45)
PR #7884: Implement getattr builtin. (stuartarchibald)
PR #8010: Add support for fp16 comparison native operators (testhound)
PR #8024: Allow converting NumPy datetimes to int (apmasell)
PR #8038: Support for Numpy BitGenerators PR#2: Standard Distributions support (kc611)
PR #8040: Support for Numpy BitGenerators PR#3: Advanced Distributions Support. (kc611)
PR #8041: Support for Numpy BitGenerators PR#4: Generator().integers() Support. (kc611)
PR #8042: Support for NumPy BitGenerators PR#5: Generator Shuffling Methods. (kc611)
PR #8061: Migrate random glue_lowering to overload where easy (apmasell)
PR #8106: Remove injection of atomic JIT functions into NRT memsys. (stuartarchibald)
PR #8134: Support non-constant exception values in JIT (guilhermeleobas sklam)
PR #8147: Adds size variable at runtime for arrays that cannot be inferred (njriasan)
PR #8158: adding -pthread for linux-ppc64le in setup.py (esc)
PR #8164: remove myself from automatic reviewer assignment (esc)
PR #8167: CUDA: Facilitate and document passing arrays / pointers to foreign functions (gmarkall)
PR #8180: CUDA: Initial support for Minor Version Compatibility (gmarkall)
PR #8183: Add n_keys option to Dict.empty() (stefanfed gmarkall)
PR #8198: Update the release template to include updating the version table. (stuartarchibald)
PR #8200: Make the NRT use the "unsafe" allocation API by default. (stuartarchibald)
PR #8201: Bump llvmlite dependency to 0.40.dev0 for Numba 0.57.0dev0 (stuartarchibald)
PR #8212: release checklist: include a note to ping @RC_testers on discourse (esc)
PR #8216: chore: Set permissions for GitHub actions (naveensrinivasan)
PR #8217: Fix syntax in docs (jorgepiloto)
PR #8221: CUDA stubs docstring: Replace illegal escape sequence (gmarkall)
PR #8228: Fix typo in @vectorize docstring and a NumPy spelling. (stuartarchibald)
PR #8229: Remove mk_unique_var in inline_closurecall.py (sklam)
PR #8234: Replace @overload_glue by @overload for 20 NumPy functions (guilhermeleobas)
PR #8235: Make the NRT stats counters optional. (stuartarchibald)
PR #8240: Add get_shared_mem_per_block method to Dispatcher (testhound)
PR #8241: Reorder typeof checks to avoid infinite loops on StructrefProxy __hash__ (DannyWeitekamp)
PR #8243: Add a note to reference/numpysupported.rst ()
PR #8245: Fix links in CONTRIBUTING.md ()
PR #8247: Fix issue 8127 (bszollosinagy)
PR #8250: Fix issue 8161 (bszollosinagy)
PR #8253: CUDA: Verify NVVM IR prior to compilation (gmarkall)
PR #8255: CUDA: Make numba.cuda.tests.doc_examples.ffi a module to fix #8252 (gmarkall)
PR #8256: Migrate linear algebra functions from glue_lowering (apmasell)
PR #8258: refactor np.where to use overload (guilhermeleobas)
PR #8259: Add np.broadcast_to(scalar_array, ()) (guilhermeleobas)
PR #8264: remove mk_unique_var from parfor_lowering_utils.py (guilhermeleobas)
PR #8265: Remove mk_unique_var from array_analysis.py (guilhermeleobas)
PR #8266: Remove mk_unique_var in untyped_passes.py (guilhermeleobas)
PR #8267: Fix segfault for invalid axes in np.split (aseyboldt)
PR #8271: Implement some CUDA intrinsics with @overload, @overload_attribute, and @intrinsic (gmarkall)
PR #8274: Update version support table doc for 0.56. (stuartarchibald)
PR #8275: Update CHANGE_LOG for 0.56.0 final (stuartarchibald)
PR #8283: Clean up / remove support for old NumPy versions (gmarkall)
PR #8289: Revert #8265. (stuartarchibald)
PR #8290: CUDA: Replace use of deprecated NVVM IR features, questionable constructs (gmarkall)
PR #8295: Add get_const_mem_size method to Dispatcher (testhound gmarkall)
PR #8297: Add __name__ attribute to CUDAUFuncDispatcher and test case (testhound)
PR #8302: CUDA: Revert numba_nvvm intrinsic name workaround (gmarkall)
PR #8315: Add get_local_mem_per_thread method to Dispatcher (testhound)
PR #8319: Bump minimum supported Python version to 3.8 (esc stuartarchibald jamesobutler)
PR #8321: Fix literal_unroll pass erroneously exiting on non-conformant loop. (stuartarchibald)
PR #8325: Remove use of mk_unique_var in stencil.py (bszollosinagy)
PR #8326: Remove mk_unique_var from parfor_lowering.py (guilhermeleobas)
PR #8331: Extend docs with info on how to call C functions from Numba (guilhermeleobas)
PR #8334: Add dict(*iterable) constructor (guilhermeleobas)
PR #8335: Remove deprecated pycc script and related source. (stuartarchibald)
PR #8336: Fix typos of "Generalized" in GUFunc-related code (gmarkall)
PR #8338: Calculate reductions before fusion so that use of reduction vars can stop fusion. (DrTodd13)
PR #8339: Fix #8291 parfor leak of redtoset variable (sklam)
PR #8341: CUDA: Support multiple outputs for Generalized Ufuncs (gmarkall)
PR #8343: Eliminate references to type annotation in compile_ptx (testhound)
PR #8348: Add get_max_threads_per_block method to Dispatcher (testhound)
PR #8354: pin setuptools to < 65 and switch from mamba to conda on RTD (esc gmarkall)
PR #8357: Clean up the buildscripts directory. (stuartarchibald)
PR #8368: Remove glue_lowering in random math that requires IR (apmasell)
PR #8376: Fix issue 8370 (bszollosinagy)
PR #8387: Add support for compute capability in IR Lowering (testhound)
PR #8388: Remove more references to the pycc binary. (stuartarchibald)
PR #8389: Make C++ extensions compile with correct compiler (apmasell)
PR #8390: Use NumPy logic for lessthan in sort to move NaNs to the back. (sklam)
PR #8415: Refactor numba.np.arraymath methods from lower_builtins to overloads (kc611)
PR #8421: Update release checklist: add a task to check dependency pinnings on subsequent releases (e.g. PATCH) (esc)
PR #8422: Switch public CI builds to use gdb from conda packages. (stuartarchibald)
PR #8423: Remove public facing and CI references to 32 bit linux support. (stuartarchibald, in addition, we are grateful for the contribution of jamesobutler towards a similar goal in PR #8319)
PR #8427: Shorten the time to verify test discovery. (stuartarchibald)
PR #8431: Replace @overload_glue by @overload for np.linspace and np.take (guilhermeleobas)
PR #8432: Refactor carray/farray to use @overload (guilhermeleobas)
PR #8435: Migrate np.atleast_? functions from glue_lowering to overload (apmasell)
PR #8438: Make the initialisation of the NRT more lazy for the njit decorator. (stuartarchibald)
PR #8439: Update the contributing docs to include a policy on formatting changes. (stuartarchibald)
PR #8440: [DOC]: Replaces icc_rt with intel-cmplr-lib-rt (oleksandr-pavlyk)
PR #8442: Implement hasattr(), str() and repr(). (stuartarchibald)
PR #8446: add version info in ImportError's (raybellwaves)
PR #8450: remove GitHub username from changelog generation script (esc)
PR #8467: Convert implementations using generated_jit to overload (gmarkall)
PR #8468: Reference test suite in installation documentation (apmasell)
PR #8469: Correctly handle optional types in parfors lowering (apmasell)
PR #8473: change the include style in _pymodule.h and remove unused or duplicate headers in two header files ()
PR #8476: Make setuptools optional at runtime. (stuartarchibald)
PR #8490: Restore installing SciPy from defaults instead of conda-forge on public CI (esc)
PR #8494: Remove context.compile_internal where easy on numba/cpython/cmathimpl.py (guilhermeleobas)
PR #8495: Removes context.compile_internal where easy on numba/cpython/listobj.py (guilhermeleobas)
PR #8496: Rewrite most of the set API to use overloads (guilhermeleobas)
PR #8499: Deprecate numba.generated_jit (stuartarchibald)
PR #8508: This updates the release checklists to capture some more checks. (stuartarchibald)
PR #8517: make some typedlist C-APIs public ()
PR #8518: Adjust stencil tests to use hardcoded python source opposed to AST. (stuartarchibald)
PR #8520: Added noncentral-chisquared, noncentral-f and logseries distributions (kc611)
PR #8522: Import jitclass from numba.experimental in jitclass documentation (armgabrielyan)
PR #8524: Fix grammar in stencil.rst (armgabrielyan)
PR #8526: Fix broken url (Nimrod0901)
PR #8527: Fix grammar in troubleshoot.rst (armgabrielyan)
PR #8539: Fix #8534, np.broadcast_to should update array size attr. (stuartarchibald)
PR #8541: Remove restoration of "free" channel in Azure CI windows builds. (stuartarchibald)
PR #8542: CUDA: Make arg optional for Stream.add_callback() (gmarkall)
PR #8544: Remove reliance on npy_<impl> ufunc loops. (stuartarchibald)
PR #8547: [Unicode] Add more string view usages for unicode operations ()
PR #8549: Fix rstcheck in Azure CI builds, update sphinx dep and docs to match (stuartarchibald)
PR #8550: Changes how tests are split between test instances (apmasell)
PR #8554: Make target for @overload have 'generic' as default. (stuartarchibald gmarkall)
PR #8557: [Unicode] support startswith with args, start and end. ()
PR #8566: Update workqueue abort message on concurrent access. (stuartarchibald)
PR #8572: CUDA: Reduce memory pressure from local memory tests (gmarkall)
PR #8579: CUDA: Add CUDA 11.8 / Hopper support and required fixes (gmarkall)
PR #8580: adding note about doing a wheel test build prior to tagging (esc)
PR #8583: Skip tests that contribute to M1 RuntimeDyLd Assertion error (sklam)
PR #8587: Remove unused refcount removal code, clean core/cpu.py module. (stuartarchibald)
PR #8588: Remove lowering extension hooks, replace with pass infrastructure. (stuartarchibald)
PR #8592: fix failure of test_cache_invalidate due to read-only install (tpwrules)
PR #8593: Adjusted ULP precesion for noncentral distribution test (kc611)
PR #8597: Prevent use of NumPy's MaskedArray. (stuartarchibald)
PR #8600: Chrome trace timestamp should be in microseconds not seconds. (sklam)
PR #8602: Throw error for unsupported dunder methods (apmasell)
PR #8605: Support for CUDA fp16 math functions (part 1) (testhound)
PR #8606: [Doc] Make the RewriteArrayExprs doc more precise ()
PR #8619: Added flat iteration logic for random distributions (kc611)
PR #8623: Adds support for np.nan_to_num (thomasjpfan)
PR #8624: DOC: Add guvectorize scalar return example (Matt711)
PR #8626: [unicode-PERF]: use optmized BM algorithm to replace the brute-force finder (dlee992)
PR #8630: Fix #8628: Don't test math.trunc with non-float64 NumPy scalars (gmarkall)
PR #8639: Python 3.11 - fix majority of remaining test failures. (stuartarchibald)
PR #8649: Remove numba.core.overload_glue module. (apmasell)
PR #8661: Make external compiler discovery lazy in the test suite. (stuartarchibald)
PR #8662: Add support for .nbytes accessor for numpy arrays (alanhdu)
PR #8666: Updates for Python 3.8 baseline/Python 3.11 migration (stuartarchibald)
PR #8673: Enable the CUDA simulator tests on Windows builds in Azure CI. (stuartarchibald)
PR #8675: Make always_run test decorator a tag and improve shard tests. (stuartarchibald)
PR #8677: Add support for min and max on boolean types. (DrTodd13)
PR #8680: Adjust flake8 config to be compatible with flake8=6.0.0 (thomasjpfan)
PR #8701: Relaxed ULP testing precision for NumPy Generator tests across all systems (kc611)
PR #8702: Supply concrete timeline for objmode fallback deprecation/removal. (stuartarchibald)
PR #8711: Python 3.11 tracing support (continuation of #8670). (AndrewVallette sklam)
PR #8716: CI: Use set -e in "Before Install" step and fix install (gmarkall)
PR #8723: Check for void return type in cuda.compile_ptx (brandonwillard)
PR #8726: Make Numba dependency check run ahead of Numba internal imports. (stuartarchibald)
PR #8728: Fix flake8 checks since upgrade to flake8=6.x (stuartarchibald)
PR #8729: Run flake8 CI step in multiple processes. (stuartarchibald)
PR #8732: Add numpy argpartition function support ()
PR #8735: Update bot to close PRs waiting on authors for more than 3 months (guilhermeleobas)
PR #8736: Implement np.lib.stride_tricks.sliding_window_view ()
PR #8744: Update CtypesLinker::add_cu error message to include fp16 usage (testhound gmarkall)
PR #8764: CUDA tidy-up: remove some unneeded methods (gmarkall)
PR #8765: BLD: remove distutils (fangchenli)
PR #8766: Stale bot: Use abandoned - stale label for closed PRs (gmarkall)
PR #8771: Update vendored Versioneer from 0.14 to 0.28 (oscargus gmarkall)
PR #8780: Improved documentation for Atomic CAS (MiloniAtal)
PR #8781: Ensure gc.collect() is called before checking refcount in tests. (sklam)
PR #8782: Changed wording of the escape error (MiloniAtal)
PR #8788: CUDA: Fix returned dtype of vectorized functions (Issue #8400) (gmarkall)
PR #8790: CUDA compare and swap with index (ianthomas23)
PR #8795: Add pending-deprecation warnings for numba.pycc (stuartarchibald)
PR #8802: Move the minimum supported NumPy version to 1.21 (stuartarchibald)
PR #8803: Attempted fix to #8789 by changing compile_ptx to accept a signature instead of argument tuple (KyanCheung)
PR #8819: Support "static" __getitem__ on Numba types in @njit code. (stuartarchibald)
PR #8822: Merge py3.11 branch to main (esc AndrewVallette stuartarchibald sklam)
PR #8826: CUDA CFFI test: conditionally require cffi module (gmarkall)
PR #8833: Fix typeguard import hook location. (stuartarchibald)
PR #8836: Fix failing typeguard test. (stuartarchibald)
PR #8837: Update AzureCI matrix for Python 3.11/NumPy 1.21..1.24 (stuartarchibald)
PR #8842: Fix buildscripts, setup.py, docs for setuptools becoming optional. (stuartarchibald)
PR #8843: Pin typeguard to 3.0.1 in AzureCI. (stuartarchibald)
PR #8852: Disable SLP vectorisation due to miscompilations. (stuartarchibald)
PR #8855: DOC: pip into double backticks in installing.rst (F3eQnxN3RriK)
PR #8856: Update TBB to use >= 2021.6 by default. (kozlov-alexey stuartarchibald)
PR #8858: Update deprecation notice for objmode fallback RE @jit use. (stuartarchibald)
PR #8869: Update CHANGE_LOG for 0.57.0rc1 (stuartarchibald esc gmarkall)
PR #8870: Fix opcode "spelling" change since Python 3.11 in CUDA debug test. (stuartarchibald)
PR #8879: Remove use of compile_isolated from generator tests. (stuartarchibald)
PR #8880: Fix missing dependency guard on pyyaml in test_azure_config. (stuartarchibald)
PR #8881: Replace use of compile_isolated in test_obj_lifetime (sklam)
PR #8887: Update PyPI supported version tags (bryant1410)
PR #8896: Remove codecov install (now deleted from PyPI) (gmarkall)
PR #8907: Work around issue #8898. Defer exp2 (and log2) calls to Numba internal symbols. (stuartarchibald)
PR #8909: Fix #8903. NumbaDeprecationWarning``s raised from ``@{gu,}vectorize. (stuartarchibald)
PR #8929: Update CHANGE_LOG for 0.57.0 final. (stuartarchibald)
Authors:
Nothing published for this version
This is a bugfix release to fix a regression in the CUDA target in relation to the .view() method on CUDA device arrays that is present when using Num
This is a bugfix release to fix a regression in the CUDA target in relation to the .view() method on CUDA device arrays that is present when using NumPy version 1.23.0 or later.
Pull-Requests:
PR #8537: Make ol_compatible_view accessible on all targets (gmarkall)
PR #8552: Update version support table for 0.56.4. (stuartarchibald)
PR #8553: Update CHANGE_LOG for 0.56.4 (stuartarchibald)
PR #8570: Release 0.56 branch: Fix overloads with target="generic" for CUDA (gmarkall)
PR #8571: Additional update to CHANGE_LOG for 0.56.4 (stuartarchibald)
Authors:
This is a bugfix release to remove the version restriction applied to the setuptools package and to fix a bug in the CUDA target in relation to copyin
This is a bugfix release to remove the version restriction applied to the setuptools package and to fix a bug in the CUDA target in relation to copying zero length device arrays to zero length host arrays.
Pull-Requests:
PR #8482: Fix #8477: Allow copies with different strides for 0-length data (gmarkall)
PR #8486: Restrict the TBB development package to supported version in Azure. (stuartarchibald)
PR #8503: Update version support table for 0.56.3 (stuartarchibald)
PR #8504: Update CHANGE_LOG for 0.56.3 (stuartarchibald)
Authors:
This is a bugfix release that supports NumPy 1.23 and fixes CUDA function caching.
This is a bugfix release that supports NumPy 1.23 and fixes CUDA function caching.
Pull-Requests:
PR #8239: Add decorator to run a test in a subprocess (stuartarchibald)
PR #8276: Move Azure to use macos-11 (stuartarchibald)
PR #8310: CUDA: Fix Issue #8309 - atomics don't work on complex components (Graham Markall)
PR #8342: Upgrade to ubuntu-20.04 for azure pipeline CI (jamesobutler)
PR #8356: Update setup.py, buildscripts, CI and docs to require setuptools<60 (stuartarchibald)
PR #8374: Don't pickle LLVM IR for CUDA code libraries (Graham Markall)
PR #8377: Add support for NumPy 1.23 (stuartarchibald)
PR #8384: Move strace() check into tests that actually need it (stuartarchibald)
PR #8386: Fix the docs for numba.get_thread_id (stuartarchibald)
PR #8407: Pin NumPy version to 1.18-1.24 (Andre Masella)
PR #8412: Create changelog for 0.56.1 (Andre Masella)
PR #8413: Fix Azure CI for NumPy 1.23 and use conda-forge scipy (Siu Kwan Lam)
PR #8414: Hotfix for 0.56.2 (Siu Kwan Lam)
Authors:
PR #7755 _: CUDA: Deprecate support for CC < 5.3 and CTK < 10.2 (Graham Markall _)
This release continues to add new features, bug fixes and stability improvements to Numba. Please note that this will be the last release that has support for Python 3.7 as the next release series (Numba 0.57) will support Python 3.11! Also note that, this will be the last release to support linux-32 packages produced by the Numba team.
Python language support enhancements:
Previously missing support for large, in-line dictionaries and internal calls to functions with large numbers of keyword arguments in Python 3.10 has been added.
operator.mul now works for list s.
Literal slices, e.g. slice(1, 10, 2) can be returned from nopython mode functions.
The len function now works on dict_keys, dict_values and dict_items .
Numba's set implementation now supports reference counted items e.g. strings.
Numba specific feature enhancements:
The experimental jitclass feature gains support for a large number of builtin methods e.g. declaring __hash__ or __getitem__ for a jitclass type.
It's now possible to use @vectorize on an already @jit family decorated function.
Name mangling has been updated to emit compiled function names that exactly match the function name in Python. This means debuggers, like GDB, can be set to break directly on Python function names.
A GDB "pretty printing" support module has been added, when loaded into GDB Numba's internal representations of Python/NumPy types are rendered inside GDB as they would be in Python.
An experimental option is added to the @jit family decorators to entirely turn off LLVM's optimisation passes for a given function (see _dbg_optnone kwarg in the @jit decorator family).
A new environment variable is added NUMBA_EXTEND_VARIABLE_LIFETIMES, which if set will extend the lifetime of variables to the end of their basic block, this to permit a debugging experience in GDB similar to that found in compiled C/C++/Fortran code.
NumPy features/enhancements:
Initial support for passing, using and returning numpy.random.Generator instances has been added, this currently includes support for the random distribution.
The broadcasting functions np.broadcast_shapes and np.broadcast_arrays are now supported.
The min and max functions now work with np.timedelta64 and np.datetime64 types.
Sorting multi-dimensional arrays along the last axis is now supported in np.sort().
The np.clip function is updated to accept NumPy arrays for the a_min and a_max arguments.
The NumPy allocation routines (np.empty , np.ones etc.) support shape arguments specified using members of enum.IntEnum s.
The function np.random.noncentral_chisquare is now supported.
The performance of functions np.full and np.ones has been improved.
Parallel Accelerator enhancements:
The parallel=True functionality is enhanced through the addition of the functions numba.set_parallel_chunksize and numba.get_parallel_chunksize to permit a more fine grained scheduling of work defined in a parallel region. There is also support for adjusting the chunksize via a context manager.
The ID of a thread is now defined to be predictable and within a known range, it is available through calling the function numba.get_thread_id.
The performance of @stencil s has been improved in both serial and parallel execution.
CUDA enhancements:
New functionality:
Self-recursive device functions.
Vector type support (float4, int2, etc.).
Shared / local arrays of extension types can now be created.
Support for linking CUDA C / C++ device functions into Python kernels.
PTX generation for Compute Capabilities 8.6 and 8.7 - e.g. RTX A series, GTX 3000 series.
Comparison operations for float16 types.
Performance improvements:
Context queries are no longer made during launch configuration.
Launch configurations are now LRU cached.
On-disk caching of CUDA kernels is now supported.
Documentation: many new examples added.
Docs:
Numba now has an official "mission statement".
There's now a "version support table" in the documentation to act as an easy to use, single reference point, for looking up information about Numba releases and their required/supported dependencies.
General Enhancements:
Numba imports more quickly in environments with large numbers of packages as it now uses importlib-metadata for querying other packages.
Emission of chrome tracing output is now supported for the internal compilation event handling system.
This release is tested and known to work when using the Pyston Python interpreter.
Pull-Requests:
PR #5209: Use importlib to load numba extensions (Stepan Rakitin Graham Markall stuartarchibald)
PR #5877: Jitclass builtin methods (Ethan Pronovost Graham Markall)
PR #6490: Stencil output allocated with np.empty now and new code to initialize the borders. (Todd A. Anderson)
PR #7005: Make numpy.searchsorted match NumPy when first argument is unsorted (Brandon T. Willard)
PR #7363: Update cuda.local.array to clarify "simple constant expression" (e.g. no NumPy ints) (Sterling Baird)
PR #7364: Removes an instance of signed integer overflow undefined behaviour. (Tobias Sargeant)
PR #7537: Add chrome tracing (Hadia Ahmed Siu Kwan Lam)
PR #7556: Testhound/fp16 comparison (Michael Collison Graham Markall)
PR #7586: Support for len on dict.keys, dict.values, and dict.items (Nick Riasanovsky)
PR #7617: Numba gdb-python extension for printing (stuartarchibald)
PR #7619: CUDA: Fix linking with PTX when compiling lazily (Graham Markall)
PR #7621: Add support for linking CUDA C / C++ with @cuda.jit kernels (Graham Markall)
PR #7625: Combined parfor chunking and caching PRs. (stuartarchibald Todd A. Anderson Siu Kwan Lam)
PR #7660: Add support for np.broadcast_arrays (Guilherme Leobas)
PR #7664: Flatten mangling dicts into a single dict (Graham Markall)
PR #7680: CUDA Docs: include example calling slow matmul (Graham Markall)
PR #7682: performance improvements to np.full and np.ones (Rishi Kulkarni)
PR #7684: DOC: remove incorrect warning in np.random reference (Rishi Kulkarni)
PR #7685: Don't convert setitems that have dimension mismatches to parfors. (Todd A. Anderson)
PR #7690: Implemented np.random.noncentral_chisquare for all size arguments (Rishi Kulkarni)
PR #7695: IntEnumMember support for np.empty, np.zeros, and np.ones (Benjamin Graham)
PR #7699: CUDA: Provide helpful error if the return type is missing for declare_device (Graham Markall)
PR #7700: Support for scalar arguments in Np.ascontiguousarray (Dhruv Patel)
PR #7703: Ignore unsupported types in ShapeEquivSet._getnames() (Benjamin Graham)
PR #7704: Move the type annotation pass to post legalization. (stuartarchibald)
PR #7709: CUDA: Fixes missing type annotation pass following #7704 (stuartarchibald)
PR #7712: Fixing issue 7693 (stuartarchibald Graham Markall luk-f-a)
PR #7714: Support for boxing SliceLiteral type (Nick Riasanovsky)
PR #7718: Bump llvmlite dependency to 0.39.0dev0 for Numba 0.56.0dev0 (stuartarchibald)
PR #7724: Update URLs in error messages to refer to RTD docs. (stuartarchibald)
PR #7728: Document that AOT-compiled functions do not check arg types (Graham Markall)
PR #7729: Handle Omitted/OmittedArgDataModel in DI generation. (stuartarchibald)
PR #7732: update release checklist following 0.55.0 RC1 (esc)
PR #7736: Update CHANGE_LOG for 0.55.0 final. (stuartarchibald)
PR #7740: CUDA Python 11.6 support (Graham Markall)
PR #7744: Fix issues with locating/parsing source during DebugInfo emission. (stuartarchibald)
PR #7745: Fix the release year for Numba 0.55 change log entry. (stuartarchibald)
PR #7748: Fix #7713: Ensure _prng_random_hash return has correct bitwidth (Graham Markall)
PR #7749: Refactor threading layer priority tests to not use stdout/stderr (stuartarchibald)
PR #7752: Fix #7751: Use original filename for array exprs (Graham Markall)
PR #7755: CUDA: Deprecate support for CC < 5.3 and CTK < 10.2 (Graham Markall)
PR #7763: Update Read the Docs configuration (automatic) (readthedocs-assistant)
PR #7764: Add dbg_optnone and dbg_extend_lifetimes flags (Siu Kwan Lam)
PR #7771: Move function unique ID to abi-tags (stuartarchibald Siu Kwan Lam)
PR #7772: CUDA: Add Support to Creating StructModel Array (Michael Wang)
PR #7776: Updates coverage.py config (stuartarchibald)
PR #7777: Remove reference existing issue from GH template. (stuartarchibald)
PR #7778: Remove long deprecated flags from the CLI. (stuartarchibald)
PR #7780: Fix sets with reference counted items (Benjamin Graham)
PR #7786: Remove dependency on intel-openmp for OSX (stuartarchibald)
PR #7788: Avoid issue with DI gen for arrayexprs. (stuartarchibald)
PR #7805: Enhance source line finding logic for debuginfo (Siu Kwan Lam)
PR #7809: Updates the gdb configuration to accept a binary name or a path. (stuartarchibald)
PR #7813: Extend parfors test timeout for aarch64. (stuartarchibald)
PR #7814: CUDA Dispatcher refactor (Graham Markall)
PR #7815: CUDA Dispatcher refactor 2: inherit from dispatcher.Dispatcher (Graham Markall)
PR #7817: Update intersphinx URLs for NumPy and llvmlite. (stuartarchibald)
PR #7823: Add renamed vars to callee scope such that it is self consistent. (stuartarchibald)
PR #7829: CUDA: Support Enum/IntEnum in Kernel (Michael Wang)
PR #7833: Add version support information table to docs. (stuartarchibald)
PR #7835: Fix pickling error when module cannot be imported (idorrington)
PR #7836: min() and max() support for np.datetime and np.timedelta (Benjamin Graham)
PR #7837: Initial refactoring of parfor reduction lowering (Siu Kwan Lam)
PR #7845: change time.time() to time.perf_counter() in docs (Nopileos2)
PR #7846: Fix CUDA enum vectorize test on Windows (Graham Markall)
PR #7848: Support for int * list (Nick Riasanovsky)
PR #7850: CUDA: Pass fastmath compiler flag down to compile_ptx and compile_device; Improve fastmath tests (Michael Wang)
PR #7855: Ensure np.argmin/no.argmax return type is intp (stuartarchibald)
PR #7858: CUDA: Deprecate ptx Attribute and Update Tests (Graham Markall Michael Wang)
PR #7861: Fix a spelling mistake in README (Zizheng Guo)
PR #7864: Fix cross_iter_dep check. (Todd A. Anderson)
PR #7865: Remove add_user_function (Graham Markall)
PR #7866: Support for large numbers of args/kws with Python 3.10 (Nick Riasanovsky)
PR #7878: CUDA: Remove some deprecated support, add CC 8.6 and 8.7 (Graham Markall)
PR #7893: Use uuid.uuid4() as the key in serialization. (stuartarchibald)
PR #7895: Remove use of llvmlite.llvmpy (Andre Masella)
PR #7898: Skip test_ptds under cuda-memcheck (Graham Markall)
PR #7901: Pyston compatibility for the test suite (Kevin Modzelewski)
PR #7911: added sys import (Nightfurex)
PR #7915: CUDA: Fix test checking debug info rendering. (stuartarchibald)
PR #7918: Add JIT examples to CUDA docs (brandon-b-miller Graham Markall)
PR #7919: Disallow //= reductions in pranges. (Todd A. Anderson)
PR #7924: Retain non-modified index tuple components. (Todd A. Anderson)
PR #7939: Fix rendering in feature request template. (stuartarchibald)
PR #7940: Implemented np.allclose in numba/np/arraymath.py (Gagandeep Singh)
PR #7941: Remove debug dump output from closure inlining pass. (stuartarchibald)
PR #7946: instructions for creating a build environment were outdated (esc)
PR #7949: Add Cuda Vector Types (Michael Wang)
PR #7956: Stop using pip for 3.10 on public ci (Revert "start testing Python 3.10 on public CI") (esc)
PR #7957: Use cloudpickle for disk caches (Siu Kwan Lam)
PR #7958: numpy.clip accept numpy.array for a_min, a_max (Gagandeep Singh)
PR #7959: Permit a new array model to have a super set of array model fields. (stuartarchibald)
PR #7961: numba.typed.typeddict.Dict.get uses castedkey to avoid returning default value even if the key is present (Gagandeep Singh)
PR #7963: remove the roadmap from the sphinx based docs (esc)
PR #7964: Support for large constant dictionaries in Python 3.10 (Nick Riasanovsky)
PR #7965: Use uuid4 instead of PID in cache temp name to prevent collisions. (stuartarchibald)
PR #7971: lru cache for configure call (Tingkai Liu)
PR #7972: Fix fp16 support for cuda shared array (Michael Collison Graham Markall)
PR #7986: Small caching refactor to support target cache implementations (Graham Markall)
PR #7994: Supporting multidimensional arrays in quick sort (Gagandeep Singh Siu Kwan Lam)
PR #7996: Fix binding logic in @overload_glue. (stuartarchibald)
PR #7999: Remove @overload_glue for NumPy allocators. (stuartarchibald)
PR #8003: Add np.broadcast_shapes (Guilherme Leobas)
PR #8004: CUDA fixes for Windows (Graham Markall)
PR #8014: Fix support for {real,imag} array attrs in Parfors. (stuartarchibald)
PR #8016: [Docs] [Very Minor] Make numba.jit boundscheck doc line consistent (Kyle Martin)
PR #8017: Update FAQ to include details about using debug-only option (Guilherme Leobas)
PR #8027: Support for NumPy 1.22 (stuartarchibald)
PR #8031: Support for Numpy BitGenerators PR#1 - Core Generator Support (Kaustubh)
PR #8035: Fix a couple of typos RE implementation (stuartarchibald)
PR #8037: CUDA self-recursion tests (Graham Markall)
PR #8044: Make Python 3.10 kwarg peephole less restrictive (Nick Riasanovsky)
PR #8046: Fix caching test failures (Siu Kwan Lam)
PR #8052: Ensure pthread is linked in when building for ppc64le. (Siu Kwan Lam)
PR #8056: Move caching tests from test_dispatcher to test_caching (Graham Markall)
PR #8057: Fix coverage checking (Graham Markall)
PR #8064: Rename "nb:run_pass" to "numba:run_pass" and document it. (Siu Kwan Lam)
PR #8065: Fix PyLowering mishandling starargs (Siu Kwan Lam)
PR #8077: change return type of np.broadcast_shapes to a tuple (Guilherme Leobas)
PR #8080: Fix windows test failure due to timeout when the machine is slow poss… (Siu Kwan Lam)
PR #8081: Fix erroneous array count in parallel gufunc kernel generation. (stuartarchibald)
PR #8089: Support on-disk caching in the CUDA target (Graham Markall)
PR #8099: Fix Py_DECREF use in case of error state (for devicearray). (stuartarchibald)
PR #8102: Combine numpy run_constrained in meta.yaml to the run requirements (Siu Kwan Lam)
PR #8109: Pin TBB support with respect to incompatible 2021.6 API. (stuartarchibald)
PR #8123: Fix CUDA print tests on Windows (Graham Markall)
PR #8124: Add explicit checks to all allocators in the NRT. (stuartarchibald)
PR #8126: Mark gufuncs as having mutable outputs (Andre Masella)
PR #8133: Fix #8132. Regression in Record.make_c_struct for handling nestedarray (Siu Kwan Lam)
PR #8137: CUDA: Fix #7806, Division by zero stops the kernel (Graham Markall)
PR #8142: CUDA: Fix some missed changes from dropping 9.2 (Graham Markall)
PR #8144: Fix NumPy capitalisation in docs. (stuartarchibald)
PR #8145: Allow ufunc builder to use previously JITed function (Andre Masella)
PR #8163: CUDA: Remove context query in launch config (Graham Markall)
PR #8165: Restrict strace based tests to be linux only via support feature. (stuartarchibald)
PR #8170: CUDA: Fix missing space in low occupancy warning (Graham Markall)
PR #8187: Update CHANGE_LOG for 0.55.2 (stuartarchibald esc)
PR #8189: updated version support information for 0.55.2/0.57 (esc)
PR #8191: CUDA: Update deprecation notes for 0.56. (Graham Markall)
PR #8192: Update CHANGE_LOG for 0.56.0 (stuartarchibald esc Siu Kwan Lam)
PR #8195: Make the workqueue threading backend once again fork safe. (stuartarchibald)
PR #8196: Fix numerical tolerance in parfors caching test. (stuartarchibald)
PR #8197: Fix isinstance warning check test. (stuartarchibald)
PR #8255: CUDA: Make numba.cuda.tests.doc_examples.ffi a module to fix #8252 (Graham Markall)
PR #8274: Update version support table doc for 0.56. (stuartarchibald)
PR #8275: Update CHANGE_LOG for 0.56.0 final (stuartarchibald)
Authors:
Nothing published for this version
This is a maintenance release to support NumPy 1.22 and Apple M1.
This is a maintenance release to support NumPy 1.22 and Apple M1.
Pull-Requests:
PR #8067: Backport #8027: Support for NumPy 1.22 (stuartarchibald)
PR #8078: Backport #7804: update local references from master -> main (esc)
PR #8082: Backport #8080: fix windows failure due to timeout (Siu Kwan Lam)
PR #8084: Pin meta.yaml to llvmlite 0.38 series (Siu Kwan Lam)
PR #8094: Backport #8052 Ensure pthread is linked in when building for ppc64le. (Siu Kwan Lam)
PR #8098: Backport #8097: Exclude libopenblas 0.3.20 on osx-arm64 (esc)
PR #8100: Backport #7786 for 0.55.2: Remove dependency on intel-openmp for OSX (stuartarchibald)
PR #8103: Backport #8102 to fix numpy requirements (Siu Kwan Lam)
PR #8114: Backport #8109 Pin TBB support with respect to incompatible 2021.6 API. (stuartarchibald)
Total PRs: 12
Authors:
Total authors: 3
CUDA target deprecation notices:
This is a bugfix release that closes all the remaining issues from the accelerated release of 0.55.0 and also any release critical regressions discovered since then.
CUDA target deprecation notices:
Support for CUDA toolkits < 10.2 is deprecated and will be removed in Numba 0.56.
Support for devices with Compute Capability < 5.3 is deprecated and will be removed in Numba 0.56.
Pull-Requests:
PR #7755: CUDA: Deprecate support for CC < 5.3 and CTK < 10.2 (Graham Markall)
PR #7749: Refactor threading layer priority tests to not use stdout/stderr (stuartarchibald)
PR #7744: Fix issues with locating/parsing source during DebugInfo emission. (stuartarchibald)
PR #7712: Fixing issue 7693 (Graham Markall luk-f-a stuartarchibald)
PR #7729: Handle Omitted/OmittedArgDataModel in DI generation. (stuartarchibald)
PR #7788: Avoid issue with DI gen for arrayexprs. (stuartarchibald)
PR #7752: Fix #7751: Use original filename for array exprs (Graham Markall)
PR #7748: Fix #7713: Ensure _prng_random_hash return has correct bitwidth (Graham Markall)
PR #7745: Fix the release year for Numba 0.55 change log entry. (stuartarchibald)
PR #7740: CUDA Python 11.6 support (Graham Markall)
PR #7724: Update URLs in error messages to refer to RTD docs. (stuartarchibald)
PR #7709: CUDA: Fixes missing type annotation pass following #7704 (stuartarchibald)
PR #7704: Move the type annotation pass to post legalization. (stuartarchibald)
PR #7619: CUDA: Fix linking with PTX when compiling lazily (Graham Markall)
Authors:
NOTE: Due to NumPy CVE-2021-33430 this release has bypassed the usual release process so as to promptly provide a Numba release that supports NumPy 1.…
This release includes a significant number important dependency upgrades along with a number of new features and bug fixes.
NOTE: Due to NumPy CVE-2021-33430 this release has bypassed the usual release process so as to promptly provide a Numba release that supports NumPy 1.21. A single release candidate (RC1) was made and a few issues were reported, these are summarised as follows and will be fixed in a subsequent 0.55.1 release.
Known issues with this release:
Incorrect result copying array-typed field of structured array (#7693)
Compilation failure for hash of floating point values on 32 bit Windows when using Python 3.10 (#7713).
Highlights of core dependency upgrades:
Support for Python 3.10
Support for NumPy 1.21
Python language support enhancements:
Experimental support for isinstance.
NumPy features/enhancements:
The following functions are now supported:
np.broadcast_to
np.float_power
np.cbrt
np.logspace
np.take_along_axis
np.average
np.argmin gains support for the axis kwarg.
np.ndarray.astype gains support for types expressed as literal strings.
Highlights of core changes:
For users of the Numba extension API, Numba now has a new error handling mode whereby it will treat all exceptions that do not inherit from numba.errors.NumbaException as a "hard error" and immediately unwind the stack. This makes it much easier to debug when writing @overloads etc from the extension API as there's now no confusion between Python errors and Numba errors. This feature can be enabled by setting the environment variable: NUMBA_CAPTURED_ERRORS='new_style'.
The threading layer selection priority can now be changed via the environment variable NUMBA_THREADING_LAYER_PRIORITY.
Highlights of changes for the CUDA target:
Support for NVIDIA's CUDA Python bindings.
Support for 16-bit floating point numbers and their basic operations via intrinsics.
Streams are provided in the Stream.async_done result, making it easier to implement asynchronous work queues.
Support for structured types in device arrays, character sequences in NumPy arrays, and some array operations on nested arrays.
Much underlying refactoring to align the CUDA target more closely with the CPU target, which lays the groudwork for supporting the high level extension API in CUDA in future releases.
Intel also kindly sponsored research and development into native debug (DWARF) support and handling per-function compilation flags:
Line number/location tracking is much improved.
Numba's internal representation of containers (e.g. tuples, arrays) are now encoded as structures.
Numba's per-function compilation flags are encoded into the ABI field of the mangled name of the function such that it's possible to compile and differentiate between versions of the same function with different flags set.
General deprecation notices:
There are no new general deprecations.
CUDA target deprecation notices:
There are no new CUDA target deprecations.
Version support/dependency changes:
Python 3.10 is supported.
NumPy version 1.21 is supported.
The minimum supported NumPy version is raised to 1.18 for runtime (compilation however remains compatible with NumPy 1.11).
Pull-Requests:
PR #6075: add np.float_power and np.cbrt (Guilherme Leobas)
PR #7047: Support __hash__ for numpy.datetime64 (Guilherme Leobas stuartarchibald)
PR #7057: Fix #7041: Add charseq registry to CUDA target (Graham Markall stuartarchibald)
PR #7082: Added Add/Sub between datetime64 array and timedelta64 scalar (Nick Riasanovsky stuartarchibald)
PR #7119: Add support for np.broadcast_to (Guilherme Leobas)
PR #7129: Add support for axis keyword argument to np.argmin() (Itamar Turner-Trauring)
PR #7132: gh #7131 Support for astype with literal strings (Nick Riasanovsky)
PR #7177: Add debug infomation support based on datamodel. (stuartarchibald)
PR #7185: Add get_impl_key as abstract method to types.Callable (Alexey Kozlov)
PR #7186: Add support for np.logspace. (Guoqiang QI)
PR #7189: CUDA: Skip IPC tests on ARM (Graham Markall)
PR #7190: CUDA: Fix test_pinned on Jetson (Graham Markall)
PR #7192: Fix missing import in array.argsort impl and add more tests. (stuartarchibald)
PR #7196: Fixes for lineinfo emission (stuartarchibald)
PR #7197: don't post to python announce on the first RC (esc)
PR #7202: Initial implementation of np.take_along_axis (Itamar Turner-Trauring)
PR #7216: Update CHANGE_LOG for 0.54.0rc2 (stuartarchibald)
PR #7219: bump llvmlite dependency to 0.38.0dev0 for Numba 0.55.0dev0 (esc)
PR #7221: Show GPU UUIDs in cuda.detect() output (Graham Markall)
PR #7222: CUDA: Warn when debug=True and opt=True (Graham Markall)
PR #7223: Replace assertion errors on IR assumption violation (Siu Kwan Lam)
PR #7226: Add support for structured types in Device Arrays (Michael Collison)
PR #7227: FIX: Typo (Srinath Kailasa)
PR #7230: PR #7171 bugfix only (stuartarchibald Todd A. Anderson)
PR #7234: add THREADING_LAYER_PRIORITY & NUMBA_THREADING_LAYER_PRIORITY (Kolen Cheung)
PR #7235: replace wordings of WIP by draft PR (Kolen Cheung)
PR #7236: CUDA: Skip managed alloc tests on ARM (Graham Markall)
PR #7237: fix a typo in a string (Kolen Cheung)
PR #7241: Set aliasing information for inplace_binops.. (Todd A. Anderson)
PR #7242: FIX: typo (Srinath Kailasa)
PR #7244: Implement partial literal propagation pass (support 'isinstance') (Guilherme Leobas stuartarchibald)
PR #7247: Solve memory leak to fix issue #7210 (Siu Kwan Lam Graham Markall ysheffer)
PR #7251: Fix #6001: typed.List ignores ctor arguments with JIT disabled (Graham Markall)
PR #7256: Fix link to the discourse forum in README (Kenichi Maehashi)
PR #7257: Use normal list constructor in List.__new__() (Graham Markall)
PR #7260: Support typed lists in heapq (Graham Markall)
PR #7263: Updated issue URL for error messages #7261 (DeviousLab)
PR #7265: Fix linspace to use np.divide and clamp to stop. (stuartarchibald)
PR #7266: CUDA: Skip multi-GPU copy test with peer access disabled (Graham Markall)
PR #7267: Fix #7258. Bug in SROA optimization (Siu Kwan Lam)
PR #7271: Update 3rd party license text. (stuartarchibald)
PR #7272: Allow annotations in njit-ed functions (LunarLanding)
PR #7273: Update CHANGE_LOG for 0.54.0rc3. (stuartarchibald)
PR #7283: Added NPM to Glossary and linked to mentions (Nihal Shetty)
PR #7285: CUDA: Fix OOB in test_kernel_arg (Graham Markall)
PR #7288: Handle cval as a np attr in stencil generation. (stuartarchibald)
PR #7294: Continuation of PR #7280, fixing lifetime of TBB task_scheduler_handle (Sergey Pokhodenko stuartarchibald)
PR #7296: Fix generator lowering not casting to the actual yielded type (Siu Kwan Lam)
PR #7298: Use CBC to pin GCC to 7 on most linux and 9 on aarch64. (stuartarchibald)
PR #7304: Continue PR#3655: add support for np.average (Hadia Ahmed slnguyen)
PR #7307: Prevent mutation of arrays in global tuples. (stuartarchibald)
PR #7309: Update MapConstraint to handle type coercion for typed.Dict correctly. (stuartarchibald)
PR #7312: Fix #7302. Workaround missing pthread problem on ppc64le (Siu Kwan Lam)
PR #7315: Link ELF obj as DSO for radare2 disassembly CFG (stuartarchibald)
PR #7316: Use float64 for consistent typing in heapq tests. (stuartarchibald)
PR #7317: In TBB tsh test switch os.fork for mp fork ctx (stuartarchibald)
PR #7319: Update CHANGE_LOG for 0.54.0 final. (stuartarchibald)
PR #7329: Improve documentation in reference to CUDA local memory (Sterling Baird)
PR #7330: Cuda matmul docs (Sterling Baird)
PR #7340: Add size_t and ssize_t types (Bruce Merry)
PR #7345: Add check for ipykernel file in IPython cache locator (Sahil Gupta)
PR #7347: fix:updated url for error report and feature rquest using issue template (DEBARGHA SAHA)
PR #7349: Allow arbitrary walk-back in reduction nodes to find inplace_binop. (Todd A. Anderson)
PR #7359: Extend support for nested arrays inside numpy records (Graham Markall luk-f-a)
PR #7375: CUDA: Run doctests as part of numba.cuda.tests and fix test_cg (Graham Markall)
PR #7395: Fix #7394 and #6550 & Added test & improved error message (MegaIng)
PR #7397: Add option to catch only Numba numba.core.errors derived exceptions. (stuartarchibald)
PR #7398: Add support for arrayanalysis of tuple args. (Todd A. Anderson)
PR #7403: Fix for issue 7402: implement missing numpy ufunc interface (Guilherme Leobas)
PR #7422: Update Omitted Type to use Hashable Values as Keys for Caching (Nick Riasanovsky)
PR #7429: Update CHANGE_LOG for 0.54.1 (stuartarchibald)
PR #7440: Refactor TargetConfig naming. (stuartarchibald)
PR #7441: Permit any string as a key in literalstrkeydict type. (stuartarchibald)
PR #7442: Add some diagnostics to SVML test failures. (stuartarchibald)
PR #7443: Refactor template selection logic for targets. (stuartarchibald)
PR #7453: CUDA: Provide stream in async_done result (Graham Markall)
PR #7456: Fix invalid codegen for #7451. (stuartarchibald)
PR #7457: Factor out target registry selection logic (stuartarchibald)
PR #7459: Include compiler flags in symbol mangling (Siu Kwan Lam)
PR #7460: Add FP16 support for CUDA (Michael Collison Graham Markall)
PR #7461: Support NVIDIA's CUDA Python bindings (Graham Markall)
PR #7465: Update changelog for 0.54.1 release (Siu Kwan Lam)
PR #7477: Fix unicode operator.eq handling of Optional types. (stuartarchibald)
PR #7479: CUDA: Print format string and warn for > 32 print() args (Graham Markall)
PR #7483: NumPy 1.21 support (Sebastian Berg stuartarchibald)
PR #7484: Fixed outgoing link to nvidia documentation. (Dhruv Patel)
PR #7493: Consolidate TLS stacks in target configuration (Siu Kwan Lam)
PR #7496: CUDA: Use a single dispatcher class for all kinds of functions (Graham Markall)
PR #7498: refactor with-detection logic (stuartarchibald esc)
PR #7499: Add build scripts for CUDA testing on gpuCI (Charles Blackmon-Luca Graham Markall)
PR #7500: Update parallel.rst (Julius Bier Kirkegaard)
PR #7506: Enhance Flags mangling/demangling (Siu Kwan Lam)
PR #7514: Fixup cuda debuginfo emission for 7177 (Siu Kwan Lam)
PR #7525: Make sure` demangle()` returns str type. (Siu Kwan Lam)
PR #7538: Fix @overload_glue performance regression. (stuartarchibald)
PR #7539: Fix str decode issue from merge #7525/#7506 (stuartarchibald)
PR #7546: Fix handling of missing const key in LiteralStrKeyDict (Siu Kwan Lam stuartarchibald)
PR #7547: Remove 32bit linux scipy installation. (stuartarchibald)
PR #7548: Correct evaluation order in assert statement (Graham Markall)
PR #7552: Prepend the inlined function name to inlined variables. (stuartarchibald)
PR #7557: Python3.10 v2 (stuartarchibald esc)
PR #7560: Refactor with detection py310 (Siu Kwan Lam esc)
PR #7561: fix a typo (Kolen Cheung)
PR #7567: Update docs to note meetings are public. (stuartarchibald)
PR #7570: Update the docs and error message for errors when importing Numba. (stuartarchibald)
PR #7580: Fix #7507. catch NotImplementedError in .get_function() (Siu Kwan Lam)
PR #7581: Add support for casting from int enums (Michael Collison)
PR #7583: Make numba.types.Optional __str__ less verbose. (stuartarchibald)
PR #7588: Fix casting of start/stop in linspace (stuartarchibald)
PR #7591: Remove deprecations (Graham Markall)
PR #7596: Fix max symbol match length for r2 (stuartarchibald)
PR #7597: Update gdb docs for new DWARF enhancements. (stuartarchibald)
PR #7603: Fix list.insert() for refcounted values (Ehsan Totoni)
PR #7605: Fix TBB 2021 DSO names on OSX/Win and make TBB reporting consistent (stuartarchibald)
PR #7606: Ensure a prescribed threading layer can load in CI. (stuartarchibald)
PR #7610: Fix #7609. Type should not be mutated. (Siu Kwan Lam)
PR #7618: Fix the doc build: docutils 0.18 not compatible with pinned sphinx (stuartarchibald)
PR #7626: Fix issues with package dependencies. (stuartarchibald esc)
PR #7627: PR 7321 continued (stuartarchibald Eric Wieser)
PR #7628: Move to using windows-2019 images in Azure (stuartarchibald)
PR #7632: Capture output in CUDA matmul doctest (Graham Markall)
PR #7636: Copy prange loop header to after the parfor. (Todd A. Anderson)
PR #7637: Increase the timeout on the SVML tests for loaded machines. (stuartarchibald)
PR #7645: In debuginfo, do not add noinline to functions marked alwaysinline (stuartarchibald)
PR #7650: Move Azure builds to OSX 10.15 (stuartarchibald esc Siu Kwan Lam)
Authors:
Nothing published for this version
This is a bugfix release for 0.54.0. It fixes a regression in structured array type handling, a potential leak on initialization failure in the CUDA t
This is a bugfix release for 0.54.0. It fixes a regression in structured array type handling, a potential leak on initialization failure in the CUDA target, a regression caused by Numba's vendored cloudpickle module resetting dynamic classes and a few minor testing/infrastructure related problems.
PR #7348: test_inspect_cli: Decode exception with default (utf-8) codec (Graham Markall)
PR #7360: CUDA: Fix potential leaks when initialization fails (Graham Markall)
PR #7386: Ensure the NRT is initialized prior to use in external NRT tests. (stuartarchibald)
PR #7388: Patch cloudpickle to not reset dynamic class each time it is unpickled (Siu Kwan Lam)
PR #7393: skip azure pipeline test if file not present (esc)
PR #7428: Fix regression #7355: cannot set items in structured array data types (Siu Kwan Lam)
Authors:
Nothing published for this version
CUDA target deprecations and breaking changes:
This release includes a significant number of new features, important refactoring, critical bug fixes and a number of dependency upgrades.
Python language support enhancements:
Basic support for f-strings.
dict comprehensions are now supported.
The sum built-in function is implemented.
NumPy features/enhancements:
The following functions are now supported:
np.clip
np.iscomplex
np.iscomplexobj
np.isneginf
np.isposinf
np.isreal
np.isrealobj
np.isscalar
np.random.dirichlet
np.rot90
np.swapaxes
Also np.argmax has gained support for the axis keyword argument and it's now possible to use 0d NumPy arrays as scalars in __setitem__ calls.
Internal changes:
Debugging support through DWARF has been fixed and enhanced.
Numba now optimises the way in which locals are emitted to help reduce time spent in LLVM's SROA passes.
CUDA target changes:
Support for emitting lineinfo to be consumed by profiling tools such as Nsight Compute
Improved fastmath code generation for various trig, division, and other functions
Faster compilation using lazy addition of libdevice to compiled units
Support for IPC on Windows
Support for passing tuples to CUDA ufuncs
Performance warnings:
When making implicit copies by calling a kernel on arrays in host memory
When occupancy is poor due to kernel or ufunc/gufunc configuration
Support for implementing warp-aggregated intrinsics:
Using support for more CUDA functions: activemask(), lanemask_lt()
The ffs() function now works correctly!
Support for @overload in the CUDA target
Intel kindly sponsored research and development that lead to a number of new features and internal support changes:
Dispatchers can now be retargetted to a new target via a user defined context manager.
Support for custom NumPy array subclasses has been added (including an overloadable memory allocator).
An inheritance based model for targets that permits targets to share @overload implementations.
Per function compiler flags with inheritance behaviours.
The extension API now has support for overloading class methods via the @overload_classmethod decorator.
Deprecations:
The ROCm target (for AMD ROC GPUs) has been moved to an "unmaintained" status and a seperate repository stub has been created for it at: https://github.com/numba/numba-rocm
CUDA target deprecations and breaking changes:
Relaxed strides checking is now the default when computing the contiguity of device arrays.
The inspect_ptx() method is deprecated. For use cases that obtain PTX for further compilation outside of Numba, use compile_ptx() instead.
Eager compilation of device functions (the case when device=True and a signature is provided) is deprecated.
Version support/dependency changes:
LLVM 11 is now supported on all platforms via llvmlite.
The minimum supported Python version is raised to 3.7.
NumPy version 1.20 is supported.
The minimum supported NumPy version is raised to 1.17 for runtime (compilation however remains compatible with NumPy 1.11).
Vendor cloudpickle v1.6.0 -- now used for all pickle operations.
TBB >= 2021 is now supported and all prior versions are unsupported (not easily possible to maintain the ABI breaking changes).
Pull-Requests:
PR #4516: Make setitem accept 0d np-arrays (Guilherme Leobas)
PR #4610: Implement np.is* functions (Guilherme Leobas)
PR #5984: Handle idx and size unification in wrap_index manually. (Todd A. Anderson)
PR #6468: Access replace_functions_map via PreParforPass instance (Sergey Pokhodenko Reazul Hoque)
PR #6469: Add address space in pointer type (Sergey Pokhodenko Reazul Hoque)
PR #6608: Support f-strings for common cases (Ehsan Totoni)
PR #6619: Improved fastmath code generation for trig, log, and exp/pow. (Graham Markall Michael Collison)
PR #6681: Explicitly catch with..as and raise error. (stuartarchibald)
PR #6689: Fix setup.py build command detection (Hannes Pahl)
PR #6695: Enable negative indexing for cuda atomic operations (Ashutosh Varma)
PR #6696: flake8: made more files flake8 compliant (Ashutosh Varma)
PR #6698: Fix #6697: Wrong dtype when using np.asarray on DeviceNDArray (Ashutosh Varma)
PR #6700: Add UUID to CUDA devices (Graham Markall)
PR #6709: Block matplotlib in test examples (Graham Markall)
PR #6718: doc: fix typo in rewrites.rst (extra iterates) (Alexander-Makaryev)
PR #6720: Faster compile (Siu Kwan Lam)
PR #6730: Fix Typeguard error (Graham Markall)
PR #6731: Add CUDA-specific pipeline (Graham Markall)
PR #6735: CUDA: Don't parse IR for modules with llvmlite (Graham Markall)
PR #6736: Support for dict comprehension (stuartarchibald)
PR #6742: Do not add overload function definitions to index. (stuartarchibald)
PR #6750: Bump to llvmlite 0.37 series (Siu Kwan Lam)
PR #6751: Suppress typeguard warnings that affect testing. (Siu Kwan Lam)
PR #6753: The check for internal types in RewriteArrayExprs (Alexander-Makaryev)
PR #6758: patch to compile _devicearray.cpp with c++11 (esc)
PR #6760: Fix scheduler bug where it rounds to 0 divisions for a chunk. (Todd A. Anderson)
PR #6762: Glue wrappers to create @overload from split typing and lowering. (stuartarchibald Siu Kwan Lam)
PR #6766: Fix DeviceNDArray null shape issue (Michael Collison)
PR #6769: CUDA: Replace CachedPTX and CachedCUFunction with CUDACodeLibrary functionality (Graham Markall)
PR #6776: Fix issue with TBB interface causing warnings and parfors counting them (stuartarchibald)
PR #6779: Fix wrap_index type unification. (Todd A. Anderson)
PR #6786: Fix gufunc kwargs support (Siu Kwan Lam)
PR #6788: Add support for fastmath 32-bit floating point divide (Michael Collison)
PR #6789: Fix warnings struct ref typeguard (stuartarchibald Siu Kwan Lam esc)
PR #6794: refactor and move create_temp_module into numba.tests.support (Alexander-Makaryev)
PR #6795: CUDA: Lazily add libdevice to compilation units (Graham Markall)
PR #6798: CUDA: Add optional Driver API argument logging (Graham Markall)
PR #6799: Print Numba and llvmlite versions in sysinfo (Graham Markall)
PR #6800: Make a common standard API for querying ufunc impl (Sergey Pokhodenko Siu Kwan Lam)
PR #6801: ParallelAccelerator no long will convert StaticSetItem to SetItem because record arrays require StaticSetItems. (Todd A. Anderson)
PR #6802: Add lineinfo flag to PTX and SASS compilation (Graham Markall Max Katz)
PR #6808: #3468 continued: Add support for np.clip (Graham Markall Aaron Russell Voelker)
PR #6810: Fix tiny formatting error in ROC kernel docs (Felix Divo)
PR #6811: CUDA: Remove test of runtime being a supported version (Graham Markall)
PR #6813: Mostly CUDA: Replace llvmpy API usage with llvmlite APIs (Graham Markall)
PR #6814: Improving context stack (stuartarchibald Siu Kwan Lam)
PR #6818: CUDA: Support IPC on Windows (Graham Markall)
PR #6822: Add support for np.rot90 (stuartarchibald Daniel Nagel)
PR #6829: Fix accuracy of np.arange and np.linspace (stuartarchibald)
PR #6830: CUDA: Use relaxed strides checking to compute contiguity (Graham Markall)
PR #6833: Raise TypeError exception if numpy array is cast to scalar (Michael Collison)
PR #6834: Remove illegal "debug" kw argument (Shaun Cutts)
PR #6836: CUDA: Documentation updates (Graham Markall)
PR #6840: CUDA: Remove items deprecated in 0.53 + simulator test fixes (Graham Markall)
PR #6841: CUDA: Fix source location on kernel entry and enable breakpoints to be set on kernels by mangled name (Graham Markall)
PR #6844: CUDA: Remove NUMBAPRO env var warnings, envvars.py + other small tidy-ups (Graham Markall)
PR #6848: Ignore .ycm_extra_conf.py (Graham Markall)
PR #6849: Add __hash__ for IntEnum (Hannes Pahl)
PR #6850: Fix up more internal warnings (stuartarchibald)
PR #6854: PR 6096 continued (stuartarchibald Ivan Butygin)
PR #6869: Implement builtin sum() (stuartarchibald)
PR #6870: Add support for dispatcher retargeting using with-context (stuartarchibald Siu Kwan Lam)
PR #6871: Force text-align:left when using Annotate (Guilherme Leobas)
PR #6873: docs: Update reference to @jitclass location (David Nadlinger)
PR #6876: Add trailing slashes to dir paths in CODEOWNERS (Graham Markall)
PR #6877: Add doc for recent target extension features (Siu Kwan Lam)
PR #6878: CUDA: Support passing tuples to ufuncs (Graham Markall)
PR #6879: CUDA: NumPy and string dtypes for local and shared arrays (Graham Markall)
PR #6880: Add attribute lower_extension to CPUContext (Reazul Hoque)
PR #6883: Add support of np.swapaxes #4074 (Daniel Nagel)
PR #6885: CUDA: Explicitly specify objmode + looplifting for jit functions in cuda.random (Graham Markall)
PR #6886: CUDA: Fix parallel testing for all testsuite submodules (Graham Markall)
PR #6888: Get overload to consider compiler flags in cache lookup (Siu Kwan Lam)
PR #6889: Address guvectorize too slow for cuda target (Michael Collison)
PR #6898: Work on overloading by hardware target. (stuartarchibald)
PR #6911: CUDA: Add support for activemask(), lanemask_lt(), and nanosleep() (Graham Markall)
PR #6912: Prevent use of varargs in closure calls. (stuartarchibald)
PR #6913: Add runtests option to gitdiff on the common ancestor (Siu Kwan Lam)
PR #6915: Update _Intrinsic for sphinx to capture the inner docstring (Guilherme Leobas)
PR #6917: Add type conversion for StringLiteral to unicode_type and test. (stuartarchibald)
PR #6918: Start section on commonly encounted unsupported parfors code. (stuartarchibald)
PR #6924: CUDA: Fix ffs (Graham Markall)
PR #6928: Add support for axis keyword arg to numpy.argmax() (stuartarchibald Itamar Turner-Trauring)
PR #6929: Fix CI failure when gitpython is missing. (Siu Kwan Lam)
PR #6936: CUDA: Implement support for PTDS globally (Graham Markall)
PR #6937: Fix memory leak in bytes boxing (stuartarchibald)
PR #6940: Fix function resolution for intrinsics across hardware. (stuartarchibald)
PR #6941: ABC the target descriptor and make consistent throughout. (stuartarchibald)
PR #6944: CUDA: Support for @overload (Graham Markall)
PR #6945: Fix issue with array analysis tests needing scipy. (stuartarchibald)
PR #6948: Refactor registry init. (stuartarchibald Graham Markall Siu Kwan Lam)
PR #6953: CUDA: Fix and deprecate inspect_ptx(), fix NVVM option setup for device functions (Graham Markall)
PR #6958: Inconsistent behavior of reshape between numpy and numba/cuda device array (Lauren Arnett)
PR #6961: Update overload glue to deal with typing_key (stuartarchibald)
PR #6964: Move minimum supported Python version to 3.7 (stuartarchibald)
PR #6966: Fix issue with TBB test detecting forks from incorrect state. (stuartarchibald)
PR #6971: Fix CUDA @intrinsic use (stuartarchibald)
PR #6977: Vendor cloudpickle (Siu Kwan Lam)
PR #6978: Implement operator.contains for empty Tuples (Brandon T. Willard)
PR #6981: Fix LLVM IR parsing error on use of np.bool_ in globals (stuartarchibald)
PR #6983: Support Optional types in ufuncs. (stuartarchibald)
PR #6985: Implement static set/get items on records with integer index (stuartarchibald)
PR #6990: Refactor hardware extension API to refer to "target" instead. (stuartarchibald)
PR #6991: Move ROCm target status to "unmaintained". (stuartarchibald)
PR #6995: Resolve issue where nan was being assigned to int type numpy array (Michael Collison)
PR #6996: Add constant lowering support for SliceType`s (`Brandon T. Willard)
PR #6997: CUDA: Remove catch of NotImplementedError in target.py (Graham Markall)
PR #6999: Fix errors introduced by the cloudpickle patch (Siu Kwan Lam)
PR #7003: More mainline fixes (stuartarchibald Graham Markall Siu Kwan Lam)
PR #7004: Test extending the CUDA target (Graham Markall)
PR #7007: Made stencil compilation not fail for arrays of conflicting types. (MegaIng)
PR #7008: Added support for np.random.dirichlet with all size arguments (Rishi Kulkarni)
PR #7016: Docs: Add DALI to list of CAI-supporting libraries (Graham Markall)
PR #7018: Remove cu{blas,sparse,rand,fft} from library checks (Graham Markall)
PR #7019: Support NumPy 1.20 (stuartarchibald)
PR #7020: Fix #7017. Adds util class PickleCallableByPath (Siu Kwan Lam)
PR #7024: fixed llvmir usage in create_module method (stuartarchibald Kalyan)
PR #7031: Fix inliner to use a single scope for all blocks (Alexey Kozlov Siu Kwan Lam)
PR #7040: Add Github action to mark issues as stale (Graham Markall)
PR #7044: Fixes for LLVM 11 (stuartarchibald)
PR #7049: Make NumPy random module use @overload_glue (stuartarchibald)
PR #7050: Add overload_classmethod (Siu Kwan Lam)
PR #7052: Fix string support in CUDA target (Graham Markall)
PR #7056: Change prange conversion approach to reuse header block. (Todd A. Anderson)
PR #7061: Add ndarray allocator classmethod (stuartarchibald Siu Kwan Lam)
PR #7064: Testhound/host array performance warning (Michael Collison)
PR #7066: Fix #7065: Add expected exception messages for NumPy 1.20 to tests (Graham Markall)
PR #7068: Enhancing docs about PRNG seeding (Jérome Eertmans)
PR #7070: Improve the issue templates and pull request template. (Guoqiang QI)
PR #7080: Fix __eq__ for Flags and cpu_options classes (Siu Kwan Lam)
PR #7087: Add note to docs about zero-initialization of variables. (stuartarchibald)
PR #7088: Initialize NUMBA_DEFAULT_NUM_THREADS with a batch scheduler aware value (Thomas VINCENT)
PR #7100: Replace deprecated call to cuDeviceComputeCapability (Graham Markall)
PR #7113: Temporarily disable debug env export. (stuartarchibald)
PR #7114: CUDA: Deprecate eager compilation of device functions (Graham Markall)
PR #7116: Fix various issues with dwarf emission: (stuartarchibald vlad-perevezentsev)
PR #7118: Remove print to stdout (stuartarchibald)
PR #7121: Continue work on numpy subclasses (Todd A. Anderson Siu Kwan Lam)
PR #7134: Move minimum LLVM version to 11. (stuartarchibald)
PR #7137: skip pycc test on Python 3.7 + macOS because of distutils issue (esc)
PR #7138: Update the Azure default linux image to Ubuntu 18.04 (stuartarchibald)
PR #7141: Require llvmlite 0.37 as minimum supported. (stuartarchibald)
PR #7143: Update version checks in __init__ for np 1.17 (stuartarchibald)
PR #7145: Fix mainline (stuartarchibald)
PR #7146: Fix inline_closurecall may not be imported (Siu Kwan Lam)
PR #7147: Revert "Workaround gitpython 3.1.18 dependency issue" (stuartarchibald)
PR #7149: Fix issue in bytecode analysis where target and next are same. (stuartarchibald)
PR #7152: Fix iterators in CUDA (Graham Markall)
PR #7156: Fix ir_utils._max_label being updated incorrectly (Siu Kwan Lam)
PR #7160: Split parfors tests (stuartarchibald)
PR #7161: Update README for 0.54 (stuartarchibald)
PR #7162: CUDA: Fix linkage of device functions when compiling for debug (Graham Markall)
PR #7163: Split legalization pass to consider IR and features separately. (stuartarchibald)
PR #7165: Fix use of np.clip where out is not provided. (stuartarchibald)
PR #7189: CUDA: Skip IPC tests on ARM (Graham Markall)
PR #7190: CUDA: Fix test_pinned on Jetson (Graham Markall)
PR #7192: Fix missing import in array.argsort impl and add more tests. (stuartarchibald)
PR #7196: Fixes for lineinfo emission. (stuartarchibald)
PR #7216: Update CHANGE_LOG for 0.54.0rc2. (stuartarchibald)
PR #7223: Replace assertion errors on IR assumption violation (Siu Kwan Lam)
PR #7230: PR #7171 bugfix only (Todd A. Anderson stuartarchibald)
PR #7236: CUDA: Skip managed alloc tests on ARM (Graham Markall)
PR #7267: Fix #7258. Bug in SROA optimization (Siu Kwan Lam)
PR #7271: Update 3rd party license text. (stuartarchibald)
PR #7272: Allow annotations in njit-ed functions (LunarLanding)
PR #7273: Update CHANGE_LOG for 0.54.0rc3. (stuartarchibald)
PR #7285: CUDA: Fix OOB in test_kernel_arg (Graham Markall)
PR #7294: Continuation of PR #7280, fixing lifetime of TBB task_scheduler_handle (Sergey Pokhodenko stuartarchibald)
PR #7298: Use CBC to pin GCC to 7 on most linux and 9 on aarch64. (stuartarchibald)
PR #7312: Fix #7302. Workaround missing pthread problem on ppc64le (Siu Kwan Lam)
PR #7317: In TBB tsh test switch os.fork for mp fork ctx (stuartarchibald)
PR #7319: Update CHANGE_LOG for 0.54.0 final. (stuartarchibald)
Authors:
Nothing published for this version
Nothing published for this version
This is a bugfix release for 0.53.0. It contains the following four pull-requests which fix two critical regressions and two build failures reported b
This is a bugfix release for 0.53.0. It contains the following four pull-requests which fix two critical regressions and two build failures reported by the openSuSe team:
PR #6826 Fix regression on gufunc serialization
PR #6828 Fix regression in CUDA: Set stream in mapped and managed array device_setup
PR #6837 Ignore warnings from packaging module when testing import behaviour.
PR #6851 set non-reported llvm timing values to 0.0
Authors:
Ben Greiner
Graham Markall
Siu Kwan Lam
Stuart Archibald
There are no new general deprecations.
This release continues to add new features, bug fixes and stability improvements to Numba.
Highlights of core changes:
Support for Python 3.9 (Stuart Archibald).
Function sub-typing (Lucio Fernandez-Arjona).
Initial support for dynamic gufuncs (i.e. from @guvectorize) (Guilherme Leobas).
Parallel Accelerator (@njit(parallel=True) now supports Fortran ordered arrays (Todd A. Anderson and Siu Kwan Lam).
Intel also kindly sponsored research and development that lead to two new features:
Exposing LLVM compilation pass timings for diagnostic purposes (Siu Kwan Lam).
An event system for broadcasting compiler events (Siu Kwan Lam).
Highlights of changes for the CUDA target:
CUDA 11.2 onwards (versions of the toolkit using NVVM IR 1.6 / LLVM IR 7.0.1) are now supported (Graham Markall).
A fast cube root function is added (Michael Collison).
Support for atomic xor, increment, decrement, exchange, are added, and compare-and-swap is extended to support 64-bit integers (Michael Collison).
Addition of cuda.is_supported_version() to check if the CUDA runtime version is supported (Graham Markall).
The CUDA dispatcher now shares infrastructure with the CPU dispatcher, improving launch times for lazily-compiled kernels (Graham Markall).
The CUDA Array Interface is updated to version 3, with support for streams added (Graham Markall).
Tuples and namedtuples can now be passed to kernels (Graham Markall).
Initial support for Cooperative Groups is added, with support for Grid Groups and Grid Sync (Graham Markall and Nick White).
Support for math.log2 and math.remainder is added (Guilherme Leobas).
General deprecation notices:
There are no new general deprecations.
CUDA target deprecation notices:
CUDA support on macOS is deprecated with this release (it still works, it is just unsupported).
The argtypes, restypes, and bind keyword arguments to the cuda.jit decorator, deprecated since 0.51.0, are removed
The Device.COMPUTE_CAPABILITY property, deprecated since 2014, has been removed (use compute_capability instead).
The to_host method of device arrays is removed (use copy_to_host instead).
General Enhancements:
PR #4769: objmode complex type spelling (Siu Kwan Lam)
PR #5579: Function subtyping (Lucio Fernandez-Arjona)
PR #5659: Add support for parfors creating 'F'ortran layout Numpy arrays. (Todd A. Anderson)
PR #5936: Improve array analysis for user-defined data types. (Todd A. Anderson)
PR #5938: Initial support for dynamic gufuncs (Guilherme Leobas)
PR #5958: Making typed.List a typing Generic (Lucio Fernandez-Arjona)
PR #6334: Support attribute access from other modules (Farah Hariri)
PR #6373: Allow Dispatchers to be cached (Eric Wieser)
PR #6519: Avoid unnecessary ir.Del generation and removal (Ehsan Totoni)
PR #6545: Refactoring ParforDiagnostics (Elena Totmenina)
PR #6560: Add LLVM pass timer (Siu Kwan Lam)
PR #6573: Improve __str__ for typed.List when invoked from IPython shell (Amin Sadeghi)
PR #6575: Avoid temp variable assignments (Ehsan Totoni)
PR #6578: Add support for numpy intersect1d and basic test cases (@caljrobe)
PR #6579: Python 3.9 support. (Stuart Archibald)
PR #6580: Store partial typing errors in compiler state (Ehsan Totoni)
PR #6626: A simple event system to broadcast compiler events (Siu Kwan Lam)
PR #6635: Try to resolve dynamic getitems as static post unroll transform. (Stuart Archibald)
PR #6636: Adds llvm_lock event (Siu Kwan Lam)
PR #6664: Adds tests for PR 5659 (Siu Kwan Lam)
PR #6680: Allow getattr to work in objmode output type spec (Siu Kwan Lam)
Fixes:
PR #6176: Remove references to deprecated numpy globals (Eric Wieser)
PR #6374: Use Python 3 style OSError handling (Eric Wieser)
PR #6402: Fix typed.Dict and typed.List crashing on parametrized types (Andreas Sodeur)
PR #6403: Add types.ListType.key (Andreas Sodeur)
PR #6410: Fixes issue #6386 (Danny Weitekamp)
PR #6425: Fix unicode join for issue #6405 (Teugea Ioan-Teodor)
PR #6437: Don't pass reduction variables known in an outer parfor to inner parfors when analyzing reductions. (Todd A. Anderson)
PR #6453: Keep original variable names in metadata to improve diagnostics (Ehsan Totoni)
PR #6454: FIX: Fixes for literals (Eric Larson)
PR #6463: Bump llvmlite to 0.36 series (Stuart Archibald)
PR #6466: Remove the misspelling of finalize_dynamic_globals (Sergey Pokhodenko)
PR #6489: Improve the error message for unsupported Buffer in Buffer situation. (Stuart Archibald)
PR #6503: Add test to ensure Numba imports without warnings. (Stuart Archibald)
PR #6508: Defer requirements to setup.py (Siu Kwan Lam)
PR #6521: Skip annotated jitclass test if typeguard is running. (Stuart Archibald)
PR #6524: Fix typed.List return value (Lucio Fernandez-Arjona)
PR #6562: Correcting typo in numba sysinfo output (Nick Sutcliffe)
PR #6574: Run parfor fusion if 2 or more parfors (Ehsan Totoni)
PR #6582: Fix typed dict error with uninitialized padding bytes (Siu Kwan Lam)
PR #6584: Remove jitclass from __init__ __all__. (Stuart Archibald)
PR #6586: Run closure inlining ahead of branch pruning in case of nonlocal (Stuart Archibald)
PR #6591: Fix inlineasm test failure. (Siu Kwan Lam)
PR #6622: Fix 6534, handle unpack of assign-like tuples. (Stuart Archibald)
PR #6652: Simplify PR-6334 (Siu Kwan Lam)
PR #6653: Fix get_numba_envvar (Siu Kwan Lam)
PR #6654: Fix #6632 support alternative dtype string spellings (Stuart Archibald)
PR #6685: Add Python 3.9 to classifiers. (Stuart Archibald)
PR #6693: patch to compile _devicearray.cpp with c++11 (Valentin Haenel)
PR #6716: Consider assignment lhs live if used in rhs (Fixes #6715) (Ehsan Totoni)
PR #6727: Avoid errors in array analysis for global tuples with non-int (Ehsan Totoni)
PR #6733: Fix segfault and errors in #6668 (Siu Kwan Lam)
PR #6741: Enable SSA in IR inliner (Ehsan Totoni)
PR #6763: use an alternative constraint for the conda packages (Valentin Haenel)
PR #6786: Fix gufunc kwargs support (Siu Kwan Lam)
CUDA Enhancements/Fixes:
PR #5162: Specify synchronization semantics of CUDA Array Interface (Graham Markall)
PR #6245: CUDA Cooperative grid groups (Graham Markall and Nick White)
PR #6333: Remove dead _Kernel.__call__ (Graham Markall)
PR #6343: CUDA: Add support for passing tuples and namedtuples to kernels (Graham Markall)
PR #6349: Refactor Dispatcher to remove unnecessary indirection (Graham Markall)
PR #6358: Add log2 and remainder implementations for cuda (Guilherme Leobas)
PR #6376: Added a fixed seed in test_atomics.py for issue #6370 (Teugea Ioan-Teodor)
PR #6377: CUDA: Fix various issues in test suite (Graham Markall)
PR #6409: Implement cuda atomic xor (Michael Collison)
PR #6422: CUDA: Remove deprecated items, expect CUDA 11.1 (Graham Markall)
PR #6427: Remove duplicate repeated definition of gufunc (Amit Kumar)
PR #6432: CUDA: Use _dispatcher.Dispatcher as base Dispatcher class (Graham Markall)
PR #6447: CUDA: Add get_regs_per_thread method to Dispatcher (Graham Markall)
PR #6499: CUDA atomic increment, decrement, exchange and compare and swap (Michael Collison)
PR #6510: CUDA: Make device array assignment synchronous where necessary (Graham Markall)
PR #6517: CUDA: Add NVVM test of all 8-bit characters (Graham Markall)
PR #6567: Refactor llvm replacement code into separate function (Michael Collison)
PR #6642: Testhound/cuda cuberoot (Michael Collison)
PR #6661: CUDA: Support NVVM70 / CUDA 11.2 (Graham Markall)
PR #6663: Fix error caused by missing "-static" libraries defined for some platforms (Siu Kwan Lam)
PR #6666: CUDA: Add a function to query whether the runtime version is supported. (Graham Markall)
PR #6725: CUDA: Fix compile to PTX with debug for CUDA 11.2 (Graham Markall)
Documentation Updates:
PR #5740: Add FAQ entry on how to create a MWR. (Stuart Archibald)
PR #6346: DOC: add where to get dev builds from to FAQ (Eyal Trabelsi)
PR #6418: docs: use https for homepage (@imba-tjd)
PR #6430: CUDA docs: Add RNG example with 3D grid and strided loops (Graham Markall)
PR #6436: docs: remove typo in Deprecation Notices (Thibault Ballier)
PR #6440: Add note about performance of typed containers from the interpreter. (Stuart Archibald)
PR #6457: Link to read the docs instead of numba homepage (Hannes Pahl)
PR #6470: Adding PyCon Sweden 2020 talk on numba (Ankit Mahato)
PR #6472: Document numba.extending.is_jitted (Stuart Archibald)
PR #6495: Fix typo in literal list docs. (Stuart Archibald)
PR #6501: Add doc entry on Numba's limited resources and how to help. (Stuart Archibald)
PR #6502: Add CODEOWNERS file. (Stuart Archibald)
PR #6531: Update canonical URL. (Stuart Archibald)
PR #6544: Minor typo / grammar fixes to 5 minute guide (Ollin Boer Bohan)
PR #6599: docs: fix simple typo, consevatively -> conservatively (Tim Gates)
PR #6609: Recommend miniforge instead of c4aarch64 (Isuru Fernando)
PR #6671: Update environment creation example to python 3.8 (Lucio Fernandez-Arjona)
PR #6676: Update hardware and software versions in various docs. (Stuart Archibald)
PR #6682: Update deprecation notices for 0.53 (Stuart Archibald)
CI/Infrastructure Updates:
PR #6458: Enable typeguard in CI (Siu Kwan Lam)
PR #6500: Update bug and feature request templates. (Stuart Archibald)
PR #6516: Fix RTD build by using conda. (Stuart Archibald)
PR #6587: Add zenodo badge (Siu Kwan Lam)
Authors:
Amin Sadeghi
Amit Kumar
Andreas Sodeur
Ankit Mahato
Chris Barnes
Danny Weitekamp
Ehsan Totoni (core dev)
Eric Larson
Eric Wieser
Eyal Trabelsi
Farah Hariri
Graham Markall
Guilherme Leobas
Hannes Pahl
Isuru Fernando
Lucio Fernandez-Arjona
Michael Collison
Nick Sutcliffe
Nick White
Ollin Boer Bohan
Sergey Pokhodenko
Siu Kwan Lam (core dev)
Stuart Archibald (core dev)
Teugea Ioan-Teodor
Thibault Ballier
Tim Gates
Todd A. Anderson (core dev)
Valentin Haenel (core dev)
@caljrobe
@imba-tjd
Nothing published for this version
Nothing published for this version
Nothing published for this version
There are no new deprecations. However, note that "compatibility" mode, which was added some 40 releases ago to help transition from 0.11 to 0.12+, ha…
This release focuses on performance improvements, but also adds some new features and contains numerous bug fixes and stability improvements.
Highlights of core performance improvements include:
Intel kindly sponsored research and development into producing a new reference count pruning pass. This pass operates at the LLVM level and can prune a number of common reference counting patterns. This will improve performance for two primary reasons:
There will be less pressure on the atomic locks used to do the reference counting.
Removal of reference counting operations permits more inlining and the optimisation passes can in general do more with what is present.
(Siu Kwan Lam).
Intel also sponsored work to improve the performance of the numba.typed.List container, particularly in the case of __getitem__ and iteration (Stuart Archibald).
Superword-level parallelism vectorization is now switched on and the optimisation pipeline has been lightly analysed and tuned so as to be able to vectorize more and more often (Stuart Archibald).
Highlights of core feature changes include:
The inspect_cfg method on the JIT dispatcher object has been significantly enhanced and now includes highlighted output and interleaved line markers and Python source (Stuart Archibald).
The BSD operating system is now unofficially supported (Stuart Archibald).
Numerous features/functionality improvements to NumPy support, including support for:
np.asfarray (Guilherme Leobas)
"subtyping" in record arrays (Lucio Fernandez-Arjona)
np.split and np.array_split (Isaac Virshup)
operator.contains with ndarray (@mugoh).
np.asarray_chkfinite (Rishabh Varshney).
NumPy 1.19 (Stuart Archibald).
the ndarray allocators, empty, ones and zeros, accepting a dtype specified as a string literal (Stuart Archibald).
Booleans are now supported as literal types (Alexey Kozlov).
On the CUDA target:
CUDA 9.0 is now the minimum supported version (Graham Markall).
Support for Unified Memory has been added (Max Katz).
Kernel launch overhead is reduced (Graham Markall).
Cudasim support for mapped array, memcopies and memset has been added (Mike Williams).
Access has been wired in to all libdevice functions (Graham Markall).
Additional CUDA atomic operations have been added (Michael Collison).
Additional math library functions (frexp, ldexp, isfinite) (Zhihao Yuan).
Support for power on complex numbers (Graham Markall).
Deprecations to note:
There are no new deprecations. However, note that "compatibility" mode, which was added some 40 releases ago to help transition from 0.11 to 0.12+, has been removed! Also, the shim to permit the import of jitclass from Numba's top level namespace has now been removed as per the deprecation schedule.
General Enhancements:
PR #5418: Add np.asfarray impl (Guilherme Leobas)
PR #5560: Record subtyping (Lucio Fernandez-Arjona)
PR #5609: Jitclass Infer Spec from Type Annotations (Ethan Pronovost)
PR #5699: Implement np.split and np.array_split (Isaac Virshup)
PR #6015: Adding BooleanLiteral type (Alexey Kozlov)
PR #6027: Support operators inlining in InlineOverloads (Alexey Kozlov)
PR #6038: Closes #6037, fixing FreeBSD compilation (László Károlyi)
PR #6086: Add more accessible version information (Stuart Archibald)
PR #6157: Add pipeline_class argument to @cfunc as supported by @jit. (Arthur Peters)
PR #6262: Support dtype from str literal. (Stuart Archibald)
PR #6271: Support ndarray contains (@mugoh)
PR #6295: Enhance inspect_cfg (Stuart Archibald)
PR #6304: Support NumPy 1.19 (Stuart Archibald)
PR #6309: Add suitable file search path for BSDs. (Stuart Archibald)
PR #6341: Re roll 6279 (Rishabh Varshney and Valentin Haenel)
Performance Enhancements:
PR #6145: Patch to fingerprint namedtuples. (Stuart Archibald)
PR #6202: Speed up str(int) (Stuart Archibald)
PR #6261: Add np.ndarray.ptp() support. (Stuart Archibald)
PR #6266: Use custom LLVM refcount pruning pass (Siu Kwan Lam)
PR #6275: Switch on SLP vectorize. (Stuart Archibald)
PR #6278: Improve typed list performance. (Stuart Archibald)
PR #6335: Split optimisation passes. (Stuart Archibald)
PR #6455: Fix refprune on obfuscated refs and stabilize optimisation WRT wrappers. (Stuart Archibald)
Fixes:
PR #5639: Make UnicodeType inherit from Hashable (Stuart Archibald)
PR #6006: Resolves incorrectly hoisted list in parfor. (Todd A. Anderson)
PR #6126: fix version_info if version can not be determined (Valentin Haenel)
PR #6137: Remove references to Python 2's long (Eric Wieser)
PR #6139: Use direct syntax instead of the add_metaclass decorator (Eric Wieser)
PR #6140: Replace calls to utils.iteritems(d) with d.items() (Eric Wieser)
PR #6141: Fix #6130 objmode cache segfault (Siu Kwan Lam)
PR #6156: Remove callers of reraise in favor of using with_traceback directly (Eric Wieser)
PR #6162: Move charseq support out of init (Stuart Archibald)
PR #6165: #5425 continued (Amos Bird and Stuart Archibald)
PR #6166: Remove Python 2 compatibility from numba.core.utils (Eric Wieser)
PR #6185: Better error message on NotDefinedError (Luiz Almeida)
PR #6194: Remove recursion from traverse_types (Radu Popovici)
PR #6200: Workaround #5973 (Stuart Archibald)
PR #6203: Make find_callname only lookup functions that are likely part of NumPy. (Stuart Archibald)
PR #6204: Fix unicode kind selection for getitem. (Stuart Archibald)
PR #6206: Build all extension modules with -g -Wall -Werror on Linux x86, provide -O0 flag option (Graham Markall)
PR #6212: Fix for objmode recompilation issue (Alexey Kozlov)
PR #6213: Fix #6177. Remove AOT dependency on the Numba package (Siu Kwan Lam)
PR #6224: Add support for tuple concatenation to array analysis. (#5396 continued) (Todd A. Anderson)
PR #6231: Remove compatibility mode (Graham Markall)
PR #6254: Fix win-32 hashing bug (from Stuart Archibald) (Ray Donnelly)
PR #6265: Fix #6260 (Stuart Archibald)
PR #6267: speed up a couple of really slow unittests (Stuart Archibald)
PR #6281: Remove numba.jitclass shim as per deprecation schedule. (Stuart Archibald)
PR #6294: Make return type propagate to all return variables (Andreas Sodeur)
PR #6300: Un-skip tests that were skipped because of #4026. (Owen Anderson)
PR #6307: Remove restrictions on SVML version due to bug in LLVM SVML CC (Stuart Archibald)
PR #6316: Make IR inliner tests not self mutating. (Stuart Archibald)
PR #6318: PR #5892 continued (Todd A. Anderson, via Stuart Archibald)
PR #6319: Permit switching off boundschecking when debug is on. (Stuart Archibald)
PR #6324: PR 6208 continued (Ivan Butygin and Stuart Archibald)
PR #6337: Implements key on types.TypeRef (Andreas Sodeur)
PR #6354: Bump llvmlite to 0.35. series. (Stuart Archibald)
PR #6357: Fix enumerate invalid decref (Siu Kwan Lam)
PR #6359: Fixes typed list indexing on 32bit (Stuart Archibald)
PR #6378: Fix incorrect CPU override in vectorization test. (Stuart Archibald)
PR #6379: Use O0 to enable inline and not affect loop-vectorization by later O3... (Siu Kwan Lam)
PR #6384: Fix failing tests to match on platform invariant int spelling. (Stuart Archibald)
PR #6390: Updates inspect_cfg (Stuart Archibald)
PR #6396: Remove hard dependency on tbb package. (Stuart Archibald)
PR #6408: Don't do array analysis for tuples that contain arrays. (Todd A. Anderson)
PR #6441: Fix ASCII flag in Unicode slicing (0.52.0rc2 regression) (Ehsan Totoni)
PR #6442: Fix array analysis regression in 0.52 RC2 for tuple of 1D arrays (Ehsan Totoni)
PR #6446: Fix #6444: pruner issues with reference stealing functions (Siu Kwan Lam)
PR #6450: Fix asfarray kwarg default handling. (Stuart Archibald)
PR #6486: fix abstract base class import (Valentin Haenel)
PR #6487: Restrict maximum version of python (Siu Kwan Lam)
PR #6527: setup.py: fix py version guard (Chris Barnes)
CUDA Enhancements/Fixes:
PR #5465: Remove macro expansion and replace uses with FE typing + BE lowering (Graham Markall)
PR #5741: CUDA: Add two-argument implementation of round() (Graham Markall)
PR #5900: Enable CUDA Unified Memory (Max Katz)
PR #6042: CUDA: Lower launch overhead by launching kernel directly (Graham Markall)
PR #6064: Lower math.frexp and math.ldexp in numba.cuda (Zhihao Yuan)
PR #6066: Lower math.isfinite in numba.cuda (Zhihao Yuan)
PR #6092: CUDA: Add mapped_array_like and pinned_array_like (Graham Markall)
PR #6127: Fix race in reduction kernels on Volta, require CUDA 9, add syncwarp with default mask (Graham Markall)
PR #6129: Extend Cudasim to support most of the memory functionality. (Mike Williams)
PR #6150: CUDA: Turn on flake8 for cudadrv and fix errors (Graham Markall)
PR #6152: CUDA: Provide wrappers for all libdevice functions, and fix typing of math function (#4618) (Graham Markall)
PR #6227: Raise exception when no supported architectures are found (Jacob Tomlinson)
PR #6244: CUDA Docs: Make workflow using simulator more explicit (Graham Markall)
PR #6248: Add support for CUDA atomic subtract operations (Michael Collison)
PR #6289: Refactor atomic test cases to reduce code duplication (Michael Collison)
PR #6290: CUDA: Add support for complex power (Graham Markall)
PR #6296: Fix flake8 violations in numba.cuda module (Graham Markall)
PR #6297: Fix flake8 violations in numba.cuda.tests.cudapy module (Graham Markall)
PR #6298: Fix flake8 violations in numba.cuda.tests.cudadrv (Graham Markall)
PR #6299: Fix flake8 violations in numba.cuda.simulator (Graham Markall)
PR #6306: Fix flake8 in cuda atomic test from merge. (Stuart Archibald)
PR #6325: Refactor code for atomic operations (Michael Collison)
PR #6329: Flake8 fix for a CUDA test (Stuart Archibald)
PR #6331: Explicitly state that NUMBA_ENABLE_CUDASIM needs to be set before import (Graham Markall)
PR #6340: CUDA: Fix #6339, performance regression launching specialized kernels (Graham Markall)
PR #6380: Only test managed allocations on Linux (Graham Markall)
Documentation Updates:
PR #6090: doc: Add doc on direct creation of Numba typed-list (@rht)
PR #6110: Update CONTRIBUTING.md (Stuart Archibald)
PR #6128: CUDA Docs: Restore Dispatcher.forall() docs (Graham Markall)
PR #6277: fix: cross2d wrong doc. reference (issue #6276) (@jeertmans)
PR #6282: Remove docs on Python 2(.7) EOL. (Stuart Archibald)
PR #6283: Add note on how public CI is impl and what users can do to help. (Stuart Archibald)
PR #6292: Document support for structured array attribute access (Graham Markall)
PR #6310: Declare unofficial *BSD support (Stuart Archibald)
PR #6342: Fix docs on literally usage. (Stuart Archibald)
PR #6348: doc: fix typo in jitclass.rst ("initilising" -> "initialising") (@muxator)
PR #6362: Move llvmlite support in README to 0.35 (Stuart Archibald)
PR #6363: Note that reference counted types are not permitted in set(). (Stuart Archibald)
PR #6364: Move deprecation schedules for 0.52 (Stuart Archibald)
CI/Infrastructure Updates:
PR #6252: Show channel URLs (Siu Kwan Lam)
PR #6338: Direct user questions to Discourse instead of the Google Group. (Stan Seibert)
PR #6474: Add skip on PPC64LE for tests causing SIGABRT in LLVM. (Stuart Archibald)
Authors:
Alexey Kozlov
Amos Bird
Andreas Sodeur
Arthur Peters
Chris Barnes
Ehsan Totoni (core dev)
Eric Wieser
Ethan Pronovost
Graham Markall
Guilherme Leobas
Isaac Virshup
Ivan Butygin
Jacob Tomlinson
Luiz Almeida
László Károlyi
Lucio Fernandez-Arjona
Max Katz
Michael Collison
Mike Williams
Owen Anderson
Radu Popovici
Ray Donnelly
Rishabh Varshney
Siu Kwan Lam (core dev)
Stan Seibert (core dev)
Stuart Archibald (core dev)
Todd A. Anderson (core dev)
Valentin Haenel (core dev)
Zhihao Yuan
@jeertmans
@mugoh
@muxator
@rht
Nothing published for this version
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →