Minutes_2018_03_08
Stan Seibert edited this page Mar 22, 2018
·
1 revision
Attendees: Stan, Siu, Stuart, Ehsan, Todd, Stefan
-
#2798: np.reshape(array, shape) is not supported whereas array.reshape(shape) is
- easy to fix
-
#2797: @guvectorize not accepting new size variable (i.e. () -> (n)) in output argument
- Doesn't look like this is allowed in the numpy spec, but could be supported?
- Not the first time someone has asked.
-
#2796: dictionary/map support request
- Common request
- Open question: how much needs to be supported to be useful
- Would scalar->scalar be enough?
- Feedback: YES
- Use list/set implementation as guide
-
#2795: Inline with LLVM IR
- Reply with information about how this could work
- Probably not going to do any enhancements to improve this
-
#2794: guvectorize and accumulate
- Looks like a NumPy ufunc limitation?
-
#2792: Add legend to HTML annotation tool.
- Led to open PR with annotation improvements
-
#2791: Reflected list (and other `reflected` types) need documenting
- Yes. Should be in FAQ and docs where list support is mentioned.
-
#2790: fill() method for DeviceNDArray
- Yes. Simple kernel
-
#2789: Array order (C/Fortran) is sometimes misinterpreted when using mmapped arrays
- Not clear what the issue is here? Race condition results in compiling for C or F order first non-determisitically.
-
#2788: Improve support on debugging Numba
- Good resource for ideas on improving debugging
-
#2787: Issue with np.concatenate
- Update docs
-
#2786: Types not recognized/supported for CuPy
- Need array protocol for GPU arrays
-
#2785: Errors on python-2.7
- Problem with FreeBSD port
- Need more info from reporter
-
#nonumber: Passing jitted functions as args.
- Need to put this into the FAQ
- Can we make this "just work"
- We should do the simple case. Siu says it is easy.
- #2793: Simplify and remove javascript from html_annotate templates.
- Needs some tests and help on other Python versions
- #2780 PowerPC reference counting fences & minor fixes
- Related: llvmlite#330 Add support for LLVM fence instruction
- Wait until LLVM 6.0
- #2779 Implement np.random.permutation
- Works, but uses Python style RNG, so hard to compare to NumPy
- #2778 Add More Device Array API Functions to CUDA Simulator
- merged
- #2777 Add support for np.correlate and np.convolve
- Needs review
- #2762 Support transpose with axes arguments.
- Very close. Testing blocker with varargs
- #2748 Added Intel SVML optimizations as opt-out choice working by default
- Stuart has built. Needs us to test further.
- #2727 Changes to enable vectorization in ParallelAccelerator.
- Stuart needs to fix a test
- Strange issues with error model handling?
- Related to open issue on non-propagation of error model (https://github.com/numba/numba/issues/2305)
- LLVM 6.0 released today
- C++ API changes:
- Fastmath IR changes that we can probably ignore for now:
- Notes from scipy-dev discussion on suitability for Numba as a core dependency:
- Interest in something that has decent performance and is easier to write than Cython.
- Concerns:
- Difficulty in debugging
- Size of dependency (30-50 MB for llvmlite)
- Lack of support for ARM
- Long term stability of project (APIs, developers, etc)
- AOT compilation needs to work for @vectorize/@guvectorize as well as @jit.
- If SciPy starts using Numba, want to make it a requirement, not just an option, to avoid having duplicate code paths.
- Other items:
- Need to write that blog post about llvmlite and static linking with LLVM.
- Numba annotate and HTML debug info
- Bunch of questions and contributions to enable numba annotate inside the Jupyter notebook
- Split annotate() so it doesn't need to write to disk?!
- First gufunc improvements
- Better SIMD generation (SVML + parfor fixes)
- LLVM 6.0
- Better debug/troubleshooting tools
- Improve docs and information as per community feedback
- Keep working through backlog of bugs and minor feature requests
- Pipeline manipulation (sklam)