Minutes_2018_04_05
Stan Seibert edited this page Apr 5, 2018
·
1 revision
Attendees: Ehsan, Stan, Stuart, Siu
- #2879: Rename NUMBAPRO cuda related environment variables + refactor.
- Seems like a good idea, but need to keep old names for compat for a bit (unfortunately)
- #2876: Wrong type declaration
- reproduces, but needs triage to understand root cause
- low priority due to weird behavior in object mode, but needs looking at
- #2875: CUDA device reset segfaults
- Close, build farm messed up
- #2873: Dead-code eliminator removes a call with side-effect.
- Already open PR to fix this issue
- #2872: Out-dated docs: RNG is now threadsafe
- Easy fix. Leftover notes from before RNG was using thread-local storage
- (0.38)
- #2870: Numba configuration methods
- Numba has so many config options, could use a file to set defaults
- #2868: cache=True & generators?
- Fix is non-trivial
- Would really like a generic "hash this function" implementation (incl args, body, globals, etc)
- #2866: NotImplementedError/LoweringError when using jit on a function containing an inner function
- Inliner fails with unused inner function (seems very contrived?)
- Weird situation, best fix would be dead code elimination
- TODO Ask if there's a real usecase
- #2863: Compiler warnings from in helplib et al.
- Already open PR to fix this
- #2862: Type inference issue with numpy.full.
- np.full doesn't look at the type of the element
- Fix will shrink code as well
- #2861: Turn off MKL threading when used in parfor context.
- Points to needing to get TBB interface improved in thread pool
- Need to check on how to enable TBB backend in MKL
- llvmlite#337: Add support for optimization remarks in pass managers.
- Not ready to merge (0.39)
- #2878: [Testing] Pr/2865
- Test fixes to SELP crashing on Windows
- Stuart will see if he can figure out the buildbot weird behavior
- #2880: This adds a github issue template.
- Needs review
- #2877: Add performance tips documentation.
- (0.38)
- Ready to merge once #2748 is merged
- #2874: Fix dead code eliminator not to remove a call with side-effect -- np.…
- (0.38)
- Needs test, then ready to merge
- #2869: WIP: use pairwise summation in `sum`
- (0.39)
- #2867: Full support for CUDA device attributes
- (0.38)
- Spot checked enum. Ready to merge.
- #2865: CUDA SELP patch is failing tests
- (0.38)
- #2864: Correct C code to prevent compiler warnings.
- (0.38)
- Ready to merge.
- #2860 [WIP] __cuda_array_interface__
- (0.39)
- #2859 Add support for improved error reporting.
- (0.38)
- Maybe a str/unicode issue with IPython on Py27?
- #2847 Replace signed array access indices in unsiged prange loop body
- (0.38)
- Ready to merge.
- #2843: [WIP] Custom pipeline
- (0.38)
- Needs docs and review. Design looks good.
- #2840: [WIP] Support list of refcounted types
- (0.39)
- #2839: Support for np.transpose and np.reshape.
- (0.38)
- Ready to merge.
- #2824: Add CUDA Primitives for Population Count
- (0.38)
- Ready to merge once CUDA test machines are happy again?
- #2817: [WIP] Emit LLVM optimization remarks
- (0.39)
- #2793: Simplify and remove javascript from html_annotate templates.
- (0.39)
- #2780: PowerPC reference counting fences & minor fixes
- (0.39)
- #2779: Implement np.random.permutation
- (0.38)
- already merged
- #2777: Add support for np.correlate and np.convolve
- (0.38)
- ready to merge
- #2748: Added Intel SVML optimizations as opt-out choice working by default
- (0.38)
- Ready to merge?
===========================
- LLVM 6 patches merged
- Twine fix is merged
- Need to fix SELP for CUDA builders
- SVML main blocker. Ready, needs icc_rt packages in numba channel.
- Merge other critical PRs from above
- 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)