Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update CHANGE_LOG for 0.51.0 #6084

Merged
merged 1 commit into from
Aug 5, 2020
Merged

Update CHANGE_LOG for 0.51.0 #6084

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
241 changes: 238 additions & 3 deletions CHANGE_LOG
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,242 @@
Version 0.51.0
--------------
Version 0.51.0 (August 12, 2020)
--------------------------------

This release continues to add new features to Numba and also contains a
significant number of bug fixes and stability improvements.

Highlights of core feature changes include:

* The compilation chain is now based on LLVM 10 (Valentin Haenel).
* Numba has internally switched to prefer non-literal types over literal ones so
as to reduce function over-specialisation, this with view of speeding up
compile times (Siu Kwan Lam).
* On the CUDA target: Support for CUDA Toolkit 11, Ampere, and Compute
Capability 8.0; Printing of ``SASS`` code for kernels; Callbacks to Python
functions can be inserted into CUDA streams, and streams are async awaitable;
Atomic ``nanmin`` and ``nanmax`` functions are added; Fixes for various
miscompilations and segfaults. (mostly Graham Markall; call backs on
streams by Peter Würtz).

Intel also kindly sponsored research and development that lead to some exciting
new features:

* Support for heterogeneous immutable lists and heterogeneous immutable string
key dictionaries. Also optional initial/construction value capturing for all
lists and dictionaries containing literal values (Stuart Archibald).
* A new pass-by-reference mutable structure extension type ``StructRef`` (Siu
Kwan Lam).
* Object mode blocks are now cacheable, with the side effect of numerous bug
fixes and performance improvements in caching. This also permits caching of
functions defined in closures (Siu Kwan Lam).

Deprecations to note:

To align with other targets, the ``argtypes`` and ``restypes`` kwargs to
``@cuda.jit`` are now deprecated, the ``bind`` kwarg is also deprecated.
Further the ``target`` kwarg to the ``numba.jit`` decorator family is
deprecated.

General Enhancements:

* PR #5463: Add str(int) impl
* PR #5526: Impl. np.asarray(literal)
* PR #5619: Add support for multi-output ufuncs
* PR #5711: Division with timedelta input
* PR #5763: Support minlength argument to np.bincount
* PR #5779: Return zero array from np.dot when the arguments are empty.
* PR #5796: Add implementation for np.positive
* PR #5849: Setitem for records when index is StringLiteral, including literal
unroll
* PR #5856: Add support for conversion of inplace_binop to parfor.
* PR #5893: Allocate 1D iteration space one at a time for more even
distribution.
* PR #5922: Reduce objmode and unpickling overhead
* PR #5944: re-enable OpenMP in wheels
* PR #5946: Implement literal dictionaries and lists.
* PR #5956: Update numba_sysinfo.py
* PR #5978: Add structref as a mutable struct that is pass-by-ref
* PR #5980: Deprecate target kwarg for numba.jit.
* PR #6058: Add prefer_literal option to overload API

Fixes:

* PR #5674: Fix #3955. Allow `with objmode` to be cached
* PR #5724: Initialize process lock lazily to prevent multiprocessing issue
* PR #5783: Make np.divide and np.remainder code more similar
* PR #5808: Fix 5665 Block jit(nopython=True, forceobj=True) and suppress
njit(forceobj=True)
* PR #5834: Fix the is operator on Ellipsis
* PR #5838: Ensure ``Dispatcher.__eq__`` always returns a bool
* PR #5841: cleanup: Use PythonAPI.bool_from_bool in more places
* PR #5862: Do not leak loop iteration variables into the numba.np.npyimpl
namespace
* PR #5869: Update repomap
* PR #5879: Fix erroneous input mutation in linalg routines
* PR #5882: Type check function in jit decorator
* PR #5925: Use np.inf and -np.inf for max and min float values respectively.
* PR #5935: Fix default arguments with multiprocessing
* PR #5952: Fix "Internal error ... local variable 'errstr' referenced before
assignment during BoundFunction(...)"
* PR #5962: Fix SVML tests with LLVM 10 and AVX512
* PR #5972: fix flake8 for numba/runtests.py
* PR #5995: Update setup.py with new llvmlite versions
* PR #5996: Set lower bound for llvmlite to 0.33
* PR #6004: Fix problem in branch pruning with LiteralStrKeyDict
* PR #6017: Fixing up numba_do_raise
* PR #6028: Fix #6023
* PR #6031: Continue 5821
* PR #6035: Fix overspecialize of literal
* PR #6046: Fixes statement reordering bug in maximize fusion step.
* PR #6056: Fix issue on invalid inlining of non-empty build_list by
inline_arraycall
* PR #6057: fix aarch64/python_3.8 failure on master
* PR #6070: Fix overspecialized containers
* PR #6071: Remove f-strings in setup.py
* PR #6072: Fix for #6005
* PR #6073: Fixes invalid C prototype in helper function.
* PR #6078: Duplicate NumPy's PyArray_DescrCheck macro
* PR #6081: Fix issue with cross drive use and relpath.
* PR #6083: Fix bug in initial value unify.

CUDA Enhancements/Fixes:

* PR #5359: Remove special-casing of 0d arrays
* PR #5709: CUDA: Refactoring of cuda.jit and kernel / dispatcher abstractions
* PR #5732: CUDA Docs: document ``forall`` method of kernels
* PR #5745: CUDA stream callbacks and async awaitable streams
* PR #5761: Add implmentation for int types for isnan and isinf for CUDA
* PR #5819: Add support for CUDA 11 and Ampere / CC 8.0
* PR #5826: CUDA: Add function to get SASS for kernels
* PR #5846: CUDA: Allow disabling NVVM optimizations, and fix debug issues
* PR #5851: CUDA EMM enhancements - add default get_ipc_handle implementation,
skip a test conditionally
* PR #5852: CUDA: Fix ``cuda.test()``
* PR #5857: CUDA docs: Add notes on resetting the EMM plugin
* PR #5859: CUDA: Fix reduce docs and style improvements
* PR #6016: Fixes change of list spelling in a cuda test.
* PR #6020: CUDA: Fix #5820, adding atomic nanmin / nanmax
* PR #6030: CUDA: Don't optimize IR before sending it to NVVM
* PR #6052: Fix dtype for atomic_add_double testsuite
* PR #6080: CUDA: Prevent auto-upgrade of atomic intrinsics

Documentation Updates:

* PR #5782: Host docs on Read the Docs
* PR #5830: doc: Mention that caching uses pickle
* PR #5963: Fix broken link to numpy ufunc signature docs
* PR #5975: restructure communication section
* PR #5981: Document bounds-checking behavior in python deviations page
* PR #5993: Docs for structref
* PR #6008: Small fix so bullet points are rendered by sphinx
* PR #6013: emphasize cuda kernel functions are asynchronous
* PR #6036: Update deprecation doc from numba.errors to numba.core.errors
* PR #6062: Change references to numba.pydata.org to https

CI updates:

* PR #5850: Updates the "New Issue" behaviour to better redirect users.
* PR #5940: Add discourse badge
* PR #5960: Setting mypy on CI

In development
Enhancements from user contributed PRs (with thanks!):

* Aisha Tammy added the ability to switch off TBB support at compile time in
#5821 (continued in #6031 by Stuart Archibald).
* Alexander Stiebing fixed a reference before assignment bug in #5952.
* Alexey Kozlov fixed a bug in tuple getitem for literals in #6028.
* Andrew Eckart updated the repomap in #5869, added support for Read the Docs
in #5782, fixed a bug in the ``np.dot`` implementation to correctly handle
empty arrays in #5779 and added support for ``minlength`` to ``np.bincount``
in #5763.
* ``@bitsisbits`` updated ``numba_sysinfo.py`` to handle HSA agents correctly in
#5956.
* Daichi Suzuo Fixed a bug in the threading backend initialisation sequence such
that it is now correctly a lazy lock in #5724.
* Eric Wieser contributed a number of patches, particularly in enhancing and
improving the ``ufunc`` capabilities:

* #5359: Remove special-casing of 0d arrays
* #5834: Fix the is operator on Ellipsis
* #5619: Add support for multi-output ufuncs
* #5841: cleanup: Use PythonAPI.bool_from_bool in more places
* #5862: Do not leak loop iteration variables into the numba.np.npyimpl
namespace
* #5838: Ensure ``Dispatcher.__eq__`` always returns a bool
* #5830: doc: Mention that caching uses pickle
* #5783: Make np.divide and np.remainder code more similar

* Ethan Pronovost added a guard to prevent the common mistake of applying a jit
decorator to the same function twice in #5881.
* Graham Markall contributed many patches to the CUDA target, as follows:

* #6052: Fix dtype for atomic_add_double tests
* #6030: CUDA: Don't optimize IR before sending it to NVVM
* #5846: CUDA: Allow disabling NVVM optimizations, and fix debug issues
* #5826: CUDA: Add function to get SASS for kernels
* #5851: CUDA EMM enhancements - add default get_ipc_handle implementation,
skip a test conditionally
* #5709: CUDA: Refactoring of cuda.jit and kernel / dispatcher abstractions
* #5819: Add support for CUDA 11 and Ampere / CC 8.0
* #6020: CUDA: Fix #5820, adding atomic nanmin / nanmax
* #5857: CUDA docs: Add notes on resetting the EMM plugin
* #5859: CUDA: Fix reduce docs and style improvements
* #5852: CUDA: Fix ``cuda.test()``
* #5732: CUDA Docs: document ``forall`` method of kernels

* Guilherme Leobas added support for ``str(int)`` in #5463 and
``np.asarray(literal value)``` in #5526.
* Hameer Abbasi deprecated the ``target`` kwarg for ``numba.jit`` in #5980.
* Hannes Pahl added a badge to the Numba github page linking to the new
discourse forum in #5940 and also fixed a bug that permitted illegal
combinations of flags to be passed into ``@jit`` in #5808.
* Kayran Schmidt emphasized that CUDA kernel functions are asynchronous in the
documentation in #6013.
* Leonardo Uieda fixed a broken link to the NumPy ufunc signature docs in #5963.
* Lucio Fernandez-Arjona added mypy to CI and started adding type annotations to
the code base in #5960, also fixed a (de)serialization problem on the
dispatcher in #5935, improved the undefined variable error message in #5876,
added support for division with timedelta input in #5711 and implemented
``setitem`` for records when the index is a ``StringLiteral`` in #5849.
* Ludovic Tiako documented Numba's bounds-checking behavior in the python
deviations page in #5981.
* Matt Roeschke changed all ``http`` references ``https`` in #6062.
* ``@niteya-shah`` implemented ``isnan`` and ``isinf`` for integer types on the
CUDA target in #5761 and implemented ``np.positive`` in #5796.
* Peter Würtz added CUDA stream callbacks and async awaitable streams in #5745.
* ``@rht`` fixed an invalid import referred to in the deprecation documentation
in #6036.
* Sergey Pokhodenko updated the SVML tests for LLVM 10 in #5962.
* Shyam Saladi fixed a Sphinx rendering bug in #6008.

Authors:

* Aisha Tammy
* Alexander Stiebing
* Alexey Kozlov
* Andrew Eckart
* ``@bitsisbits``
* Daichi Suzuo
* Eric Wieser
* Ethan Pronovost
* Graham Markall
* Guilherme Leobas
* Hameer Abbasi
* Hannes Pahl
* Kayran Schmidt
* Kozlov, Alexey
* Leonardo Uieda
* Lucio Fernandez-Arjona
* Ludovic Tiako
* Matt Roeschke
* ``@niteya-shah``
* Peter Würtz
* Sergey Pokhodenko
* Shyam Saladi
* ``@rht``
* Siu Kwan Lam (core dev)
* Stuart Archibald (core dev)
* Todd A. Anderson (core dev)
* Valentin Haenel (core dev)


Version 0.50.1 (Jun 24, 2020)
Expand Down