Changelog

2023-05-02 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Bump xsimd dependency to 11.0.0, with a backported patch.

2023-04-29 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Improve performance of functions revieving scalar arguments. This changes the internal function call API.
• Improve performance of fix-stride slicing, using a new slice representation.
• Improve numpy.copyto performance, and detect copyto pattern usage.
• Force internal linkage of generated functions, which gives more optimization room to the C++ compiler.
• Provide entry points pythran.import_pythrancode and pythran.import_pythranfile, as a poor man JIT option.
• Optimize numpy.argmax(cst * val) into numpy.argmax(val) when cst is positive.
• Avoid copies upon numpy.array_split
• Get rid of unused functions C++ warnings
• Avoid generating a loop footer when the loop index is not used outside of the loop.

2023-01-05 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Bump xsimd depdency to 10.0.0
• Correctly raise KeyError when getting an unset dict item
• Fix negative binomial implementation
• Correctly convert strided array with offset
• Support np.clip(v, None, x) and np.clip(v, x, None)
• Support scipy.special.ndtr, scipy.special.ndtri, scipy.special.gammaincinv
• Fix set intersection
• Support numpy.fft.fftn
• Support axis parameter in numpy.argsort

2022-09-20 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Improve numpy expression computation speed
• Decent ICC support (fixing an issue from 2018!)
• Much faster C++ code generation
• Complete rework of constant folder
• Support C++-time evaluation of numpy.ndarray.ndim
• Improved omp declare reduction support
• Allow indexing of ndarray by integers of mixed types
• A lot of small pesty bug fixes in the C++ headers

2021-12-08 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Upgrade xsimd version to 8.0.5
• Performance fix on np.dot with transposed parameters
• Raise an error when no specs are given and we want to generate a native module
• Support more complex transpose expression
• Honor quotes in config files for cflags, ldflags etc
• Add pythran option -ftime-report to print out the time spent on optimizations
• Support isinstance when second argument is a tuple

2021-09-08 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Fix performance issue with assert handling
• Fix issue in libomp detection
• Support immediate value for some keyword parameters, esp; keep_dims parameter
• Better detection of generalized expression overlap
• And extra minor fixes :-)

2021-07-06 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Remove six, networkx and decorator dependency
• Bump gast and Beniget requirements to support python 3.10
• Bump xsimd to 7.5.0
• Minimal default support for non-linux, non-osx, now-windows platform
• Numpy improvements for np.bincount, np.transpose, np.searchsorted
• Restore (and test) cython compatibility
• Expose pythran.get_include for toolchain integration
• Improve error message on invalid spec
• Handle static dispatching based on keyword signature
• Raise Memory Error upon (too) large numpy alloc
• Support scalar case of scipy.special.binom
• Trim the number of warnings in pythonic codebase

2021-05-23 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Fix compatibility issue with python 3.10

2021-05-09 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Honor CXXFLAGS and LDFLAGS
• Generalize numpy.dot to higher dimenson (partial support)
• Fix important memory leak in handling of transposed matrices
• Fix several string interaction (str.split, f-string)
• Fix interaction with numpy.dtype.type
• Improve OpenMP detection
• Optimize some matrix transpose cases

2021-03-30 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Always honor $HOME for user configuration lookup (even on Windows)
• Default to clang-cl on windows
• Honor CFLAGS environment variable
• Correctly type tuples that are not bound to a variable
• Move to pocketfft for fft related operations
• Support numpy.vdot, numpy.dot between array of different dtype, improve numpy.copyto, numpy.ndarray constructor, numpy.ihfft, numpy.hfft, numpy.full and numpy.full_like
• Return a floating point type when computing builtins.pow, unless the exponenent is a positive integer literal
• Optimize shared reference of array expressions (not you again!)
• Introduce a specific type for strings of one element, aka chr
• Fix implementation of str.lstrip and str.rstrip, harden str.join
• Improve quality of the error report
• Detect divide by zero in debug mode
• Improve PythranBuildExt to support base class customization

2020-11-09 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Basic f-string support
• Optimize shared reference of array expressions (again)
• np.nan{min,max}, np.around, np.wrap implementation fixes
• PYTHONOPTIMIZE={1,2} compatibility
• Support list.sort with key argument

2020-09-22 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Improve computation of contiguous slices when bounds are positive
• Optimize shared reference of array expressions
• Speedup complex combined types computation
• Fix pythran-config with cl.exe or clang-cl.exe
• Support Python 3.9

2020-08-01 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• First release only supporting Python3
• Allow to disable blas usage through blas=none setting
• Improve range value analysis
• Change numpy expression evaluator
• Change sorting algorithm from std::sort to pdqsort
• Bug fixes in various numpy.* implementation
• Freeze gast and beniget version
• Diligently use and document assert behavior
• Many fix related to static conditions
• Improve np.sort and np.median argument support
• Improve scope computation in presence of if/else
• Refuse assert with side effect
• Provide sane default for OpenMP reduction on complex
• Upgrade boost version to 1.72
• Allow %{ext} substitution in output filename
• Prettier pythran syntax error
• Faster compilation (in some cases!)
• Partial np.tofile implementation
• Allow to specify –config ‘pythran.optimizations=’ on the CLI
• Fix interaction between OpenMP collapse and pythran-generated loops

2019-12-31 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Last release supporting both Python2 and Python3
• Vectorized version of numpy.arange
• Support more numpy type conversion operators
• Improve translation time
• Version bump for xsimd dependency
• Fix compile without ENABLE_PYTHON_MODULE
• Various fixes for bug introduced in previous revision ^^!

2019-10-30 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Support Python up to 3.8 (included)
• clang-cl on Windows support
• PyPy3.6 support
• Fix bug involving is None and loops
• Support numpy.heaviside, numpy.cross
• Significant improvement of numpy.random.* support
• Fix memory leak when converting a dict to python
• New optimization: dead function elimination
• Support for kwonly arguments
• More pattern transformations: numpy.cbrt, numpy.sqrt and variant of abssqr
• Support dtype argument for numpy.linspace
• Improve constant folding & forward substitution
• Extend range analysis to tuples
• Basic support for scipy.special.spherical_jn and scipy.special.spherical_yn
• Support isinstance builtin
• Support the type() builtin
• New command line parameter: –config= as an alternative to .pythranrc
• Various fixes and doc upgrade I’m too lazy to list here ;-)

2019-08-19 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Warn users about deprecation of python2
• Fix incompatibility with numpy 1.17.0
• Allow pythran-openblas as a fallback dependency for openblas
• Better 32bit arch support
• Better support of OpenMP collapse clause
• Upgrade boost dep to 1.68 and xsimd dep to 7.2.2
• Use static shape information for more efficient broadcasting
• Allow / options on Windows platform in addition to Unix-style options
• Fix typing issue for r-value dict/set/list
• Allow to pass slice to pythran-exported functions
• Fix np.arange for integral numbers
• Fix static if support
• Support tuple arguments for np.concatenate
• Support default arguments for np.randint
• Support kind argument for np.sort

2019-05-05 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Better support for is None pattern
• Support numpy.byte in code and annotation
• Fix string slice assignment
• Support numpy.interp
• Improved Windows support
• Fix numpy.fft in presence of threading
• Better error message upon type mismatch
• Extended support for numpy.append
• Fix ndarray printing
• Property report error on Elippsis
• Optimize away some uses of np.array
• Properly support keyword-only arguments from Python3
• Improved fixed-size array broadcasting
• New annotation syntax for default argument through the ‘?’ qualifier
• Stricter type checking between Python and Pythran result type
• Rely on beniget <https://github.com/serge-sans-paille/beniget> for some analyses
• Fix dtype inference for OSX and Windows, wrt 32/64 bits
• Generate code compatible with OpenMP collapse clause
• Fix np.bincount, np.angle, np.fromiter implementation
• Improved (but still incomplete) support of slicing of an array through an array
• Allow specification of memory layout for 2D array parameters in annotations
• Avoid useless copies for some dict operations
• Support np.expand_dims, np.correlate, np.convolve, np.setdiff1d
• Detect and specialize code for fixed-size list
• Support more dot-idiom-to-blas forwarding
• Fix important memory leak in numpy_gexpr to_python

2019-01-29 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Fix np.transpose regression
• Upgrade xsimd to 7.1.2
• Fix setup.py test target

2019-01-18 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Honor PYTHRANRC environment variable for config file lookup
• Stricter type checking for export parameters
• Allow some kind of list to tuple conversion
• Lazy slicing of broadcasted expression and transposed expression
• Support numpy.stack, numpy.rollaxis, numpy.broadcast_to, numpy.ndarray.dtype.type
• Better support of array of complex numbers
• Verbose mode in pythran-config to debug compiler backend issues
• Config file linting
• Evaluate numpy.arange lazily when valid
• Faster PRNG, namely pcg
• Favor Python3 support in various places
• Fix Numpy.remainder implementation
• Better support for importing user modules
• More vectorized operations support

2018-11-06 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Moving to xsimd as vectorization engine, requires -DUSE_XSIMD
• Better support of view conversion from Python to Pythran
• Improved Cython integration
• Improved documentation, add example section
• Updated ArchLinux Packaging
• Remove useless warnings during compilation
• Faster abs(x**2) for complex numbers
• IPython magic now accepts most optimization flags
• Automatic detection of partially (or fully) constant shape for arrays
• Add ignoreflags settings to .pythranrc to skip some compiler flags
• Quad number (aka long double) basic support
• And many minor bugfixes, as usual :-)

2018-09-16 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Decent support for is None pattern
• Bessel functions, aka scipy.special.stuff
• __builtin__.slice support
• np.unravel_index, np.(u)intc support
• Cleanup and fix issues around sdist / build_ext and stuff
• O(1) View conversion to Python
• Dropped big int support
• Speed improvement for np.argmin, np.argmax and square of complex
• Allow partially (or fully) constant shape for arrays
• Deterministic header inclusion (toward reproducible build?)
• Better error report on argument mismatch
• Better support for view arguments (but still lotta missing :-/)
• Better Windows support (fix compiler-specific bugs) including OpenMP

2018-06-06 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Much better support of fancy indexing, still not 100% support though
• Better distutils integration (extra_compile_args)
• Better support for np.unique, functor, np.Inf, np.arange, it.repeat
• Better support of user import from pythranized module (still no globals)
• Better OSX support (fix compiler-specific bugs)
• Most tests now work on Python 2 and Python 3
• Many bugfixes triggered by the bug report of Yann Diorcet, Jean Laroche and David Menéndez Hurtado, Fabien Rozar, C. Claus

2018-04-23 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• numpy.fft support (thanks to Jean Laroche)
• Faster generalized expression
• Faster numpy.transpose, numpy.argmax, numpy reduction
• Sphinx-compatible generated docstring (thanks to Pierre Augier)
• Python output through -P (thanks to Pierre Augier)
• Many bugfixes and numpy improvements (thanks to Yann Diorecet and Jean Laroche)

2018-02-05 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Slimmer generated binaries
• Preliminary native Windows support for Python3
• Several numpy-related function improvements
• @ operator support
• Better negative index support and range detection
• Glimpses of OpenMP4 support, min/max support in OpenMP reductions
• Python Capsule support
• Work around GCC-7 parsing bug

2017-10-14 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Moving to networkx 2.0
• List/Tuple partial constant folding
• Minor notebook integration fixes
• Minor cython integration fixes
• Memory leak removal
• Support out field in numpy.sum

2017-09-12 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Significant compilation time improvements (again)
• Improved cython cooperation
• Many OSX and Python3 integration fixes
• Revive pure C++ mode
• Exported functions now support keyword passing style

2017-07-04 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Significant compilation time improvements
• Support for separated .pythran files
• Many bug fixes and perf improvement

2017-01-05 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Python 3 support
• (unsound) Type Checker
• Various bug fixes and perf improvement, as usual

2016-07-05 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Fix install / setup minor issues
• Restore OpenMP support
• Fix GMP installation issue

2016-06-13 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Better Jupyter Note book integration
• Numpy Broadcasting support
• Improved value binding analysis
• Simple inlining optimization
• Type engine improvement
• Less fat in the generated modules
• More and better support for various Numpy functions
• Various performance improvement
• Global variable handling, as constants only though

2016-01-05 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• IPython’s magic for pythran now supports extra compile flags
• Pythran’s C++ output is compatible with Python3 and pythran3 can compile it!
• More syntax checks (and less template traceback)
• Improved UI (multiline pythran exports, better setup.py…)
• Pythonic leaning / bugfixing (this tends to be a permanent item)
• More generic support for numpy’s dtype
• Simpler install (no more boost.python deps, nor nt2 configuration)
• Faster compilation (no more boost.python deps, smarter pass manager)
• Better testing (gcc + clang)

2015-10-13 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Significantly decrease compilation time
• Faster execution of numpy generalized slicing
• Tentative conda support
• Tentative Windows support (using Win Python)
• Preserve original docstrings
• Add __pythran__ global variable to pythran generated modules
• Faster implementation of various itertools functions
• Rely on distutils for module code compilation
• Support most of numpy.random
• Remove git and make dependency to install nt2
• Proper pip support instead of distuils
• Remove dependency to boost.python
• Remove dependency to tcmalloc
• Pythonic library cleaning (less dependencies / header / splitting / mrpropering)
• More lazy computations
• More numpy function support (including dot on matrices, linalg.norm, mean)
• Lot of code cleaning / refactoring (both in Python and C++)
• Many bugfixes, thanks to all the bug reporters!

2015-04-06 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Various numpy.* function implementation improvement (incl. concatenate, str.join, itertools.combinations)
• Better error detection during install step
• 32 bit compatibility
• Complete rewrite of the expression engine
• Improved support of numpy extended expression
• Better user feedback on invalid pythran spec
• More efficient support of string literals
• Faster exponentiation when index is an integer
• NT2 revision bump
• No-copy list as numpy expression parameters
• Accept C and fortran layout for input arrays
• Range value analysis and boundcheck removal
• Newaxis style indexing
• Better array-of-complex support
• Glimpses of python3 support
• Support for importing user defined modules
• Archlinux support
• Accept strided array as exported function input

2014-10-22 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Full SIMD support! Almost all numpy expressions are vectorized
• Better memory management at the Python/C++ layer, esp. when sharing
• Support named parameters
• Better complex numbers support
• A lot of internal code cleaning
• Better code generation for regular loops
• MacOS install guide & ArchLinux packages
• Travis run the test suite, w and w/ SIMD, w and w/ OpenMP
• Many performance improvements at the numpy expression level
• Faster array copies, including slices
• Much better constant folding
• Distutils support through a PythranExtension
• Improve implementation of many numpy functions
• Improve forward substitution
• Use most recent nt2 version
• Make dependency on libgomp optional

2014-05-17 Serge Guelton <serge.guelton@telecom-bretagne.eu>

• Improved C++ compilation time (twice as fast)
• Efficient extended slicing
• Support most numpy dtype ([u]int8,…, [u]int64, float32, float64)
• Support indexing array through boolean array
• Add a nice Pythran logo :-)
• Improve validation a lot
• Reduce native module loading overhead
• Forward substitution implementation
• More numpy support and many bug fixes!
• Remove array auto vectorization/parallelization