Minutes_2018_02_22

Numba Meeting: 2018-02-22

Attendees: Stan, Todd, Stefan, Stuart, Ehsan, Oscar, Siu

New:

numba#2757: Raise if AOT function arguments do not match original compiled signature
- Confused a user
numba#2758: The integer dtype used in array constructors need not be restricted to `intp`
numba#2759: prange fails to reuse accumulator for reduction
- Todd will look at next week
numba#2761: Support np.newaxis indexing
numba#2763: Numba/llvmlite code generation docs and tutorials
- Currently hard to onboard new developers
- Need a more advanced use case than what is in current tutorial
- Many functions are not well documented
- Need a tutorial that shows where the boundary between nopython and IR builder should be

Outstanding:

numba#2736: Using numpy functions in @guvectorize-d block
- Add to FAQ
numba#2742: Support ndarray.fill under parfors
- feature request
- Stuart will try this to see how easy it is
- Is there a way to streamline these ndarray methods with no equivalent numpy module function?

numba#2660: Support bools from cffi in nopython
- Waiting to see if Windows tests pass on build farm
numba#2727: Changes to enable vectorization in ParallelAccelerator
- Ready to merge?
- Still a few failing tests
numba#2748/llvmlite#326: Added Intel SVML optimizations as opt-out choice working by default
- Stuart will review
numba#2762: Support transpose with axes arguments
- Stuart will review
llvmlite#328: Build C files separately on Linux and support parallel make
- Seems fine (but only 13 CPP files in llvmlite)
- Subtle change in linker flags in patch. Need to see if that was intentional or not.
WIP: np.correlate, np.convolve
https://github.com/numba/pyculib/pull/14
- Merged right after call
https://github.com/numba/pyculib/pull/15
- Merged right after call

Gufunc rewrite:
- Detail writeup: https://gist.github.com/sklam/9fe5431672441e6a689e3ec2a121d104
- Current status
  - Basic example prototype: https://github.com/plures/gumath
  - Explanation: https://github.com/plures/gumath/blob/master/doc/experimental/dispatch.rst
Automate building of wheels
- Already nearly automated
Scipy switching symbols to hidden https://github.com/scipy/scipy/issues/8451
- Should we expose the function to get at pointers from cython modules?
- Seems like a good idea