Numba 0.27.0 released

[Stanley Seibert]

Hi Numba Users,

In preparation for SciPy 2016 kicking off today, we've released Numba version 0.27.0.  (In fact, there's a tutorial on Numba being taught at the SciPy conference this afternoon by Lorena Barba and Gil Forsyth.  I believe tutorials are generally recorded and posted to YouTube after the conference, so I'll be sure to post a link to the list when that is available.)

This Numba release requires llvmlite 0.12, which still depends on LLVM 3.7.  In our August release, we are planning to update to LLVM 3.8, and possibly CUDA 8, but didn't want to force everyone to upgrade just yet.

Highlights from the release include:

• Support for np.lib.stride_tricks.as_strided(), which is often used for doing nearest neighbor calculations in a more compact way
  
• Basic recursion support!  A frequent request for Numba, so we've added it for the case where the function calls itself directly, and a fixed type signature is given to the @jit decorator.  We plan to loosen up these restrictions in the future, but it will require a bit of work on our type inference pass.
  
• Support for getting cffi-compatible function objects from @cfunc decorated functions.
  
• np.linalg.lstsq, np.linalg.solve, np.linalg.pinv are supported in nopython mode.  These use the BLAS/LAPACK libraries that are linked against your scipy installation.  (If you aren't already using numpy/scipy linked to OpenBLAS or MKL, you definitely should consider it.)
  
• Various improvements to using print() from nopython mode
  
• Fixed handling of slice assignment when both source and destination are overlapping ranges in the same ndarray.
  
• NumPy ufunc calls work when doing ahead of time compilation
  
• Fixed a function caching regression when inside an IPython session
  
• 0d array constants are supported

Full release notes here: http://numba.pydata.org/numba-doc/latest/release-notes.html#version-0-27-0

To upgrade with conda, just type:

conda update numba=0.27.0

Otherwise, you can get the source tarball from PyPI:

https://pypi.python.org/pypi/numba/0.27.0

Thanks again to everyone who sent us feedback and bug reports!

[Zhang Richard]

Hi Stanley:

     I met a error in numba while transfer my data to device. I wrote my question in discussion group and I still haven't fix it. I don't know if I can remove the device data array manually or not. I think you might work with cuda functionality in numba. Can you give me some advice to solve this error ? 

     Thanks very much !

Richard

--
You received this message because you are subscribed to the Google Groups "Numba Public Discussion - Public" group.
To unsubscribe from this group and stop receiving emails from it, send an email to numba-users...@continuum.io.
To post to this group, send email to numba...@continuum.io.
To view this discussion on the web visit https://groups.google.com/a/continuum.io/d/msgid/numba-users/CADv3RKQO1dka8aWUqS3Fo1mvgoTqKCe7O%3DfZywaokeant3vDsg%40mail.gmail.com.
For more options, visit https://groups.google.com/a/continuum.io/d/optout.