Minutes_2019_09_24

Attendees: Siu, Stan, Stuart, Aaron, Pearu,

0. Feature Discussion

• Numba 0.46, llvmlite 0.30, numba-scipy 0.2 readiness.
  • PR state
  • Build matrix state (NumPy 117, numba-scipy)
  • New llvmdev builds?
    • no
    • just need to fix fPIC issue in llvmlite

1. New issues

• **** #4605 - [FEA] Supporting LLVM 9
  • to investigate for showstopper
  • to check list:
    • MCJIT/ORCJITv1/ORCJITv2
    • VS2017
    • Manylinux???? wheel
    • PPC
• #4603 - Property setter ignored
  • label: jitclass bug
  • setter typed incorrectly
• **** #4602 - Inserting global njit, numpy function into an njit function fails
  • Siu Q: should njit(np.sum) be supported via the overloaded version?
    • A: maybe, current behavior is an inconsistency; or, accept (overloaded) supported function as argument.
    • TODO: open new issue about it
• #4600 - Inconsistent error reporting for out of range indexing with parallel vs serial and lists vs arrays
  • pending on boundchecks
• #4596 - Trying Numba with Grakel (Graph kernel, Random Walk)
  • label: no action required
• #4590 - Cache numba.cuda functions on repeated deserialization
  • Stan self-assigned
• #4589 - Use different cuOccupancyMaxPotentialBlockSize call?
  • PR: https://github.com/numba/numba/pull/4592
  • need to check smoketest
• **** #4587 - Support njitting partial functions
  • seems like a big feature to write
  • should we provide numba.partial base on numba.generated_jit?
    • numba.partial(func, **kwargs) => newfunc = lambda *args: func(*args, **kwargs) but jittable
• #4585 - Strange NumbaPerformanceWarning for numpy @ operator.
  • M[:, :] forcing "any" layout
• **** #4584 - Typed list of arrays much slower than a reflected list
  • call <c-api> not hoisted
  • try to make it inlineable; maybe reuse the reflected list impl.
  • needs to do the perf analysis
• #4580 - Incompatible shape for array on vector function
  • missing .reshape() that requires copy
• #4579 - parallel list comprehension corruption
  • label: parfor, bug
  • Todds is working on this
• #4577 - Set of tuples with string crashes
  • WIP PR #4586

Already Closed

• #4598 - Numba build failed on MacOS
• #4597 - Numba failed
• #4594 - Scalar float type signature not respected for both inputs and outputs
• #4591 - Iterating over 3D array
  • label: need triage

2. New Open PRs

• #4606 - {WIP} Enable support for count() method for unicode string
• #4604 - add sometrue and alltrue
  • to check if they still exist
• #4601 - Add triangular indices functions
• **** #4599 - WIP: Permits closures to become functions, enables map(), filter()
  • maybe worth splitting it so we have the closure=>functions ready earlier.
• #4595 - [WIP] Implement numba.cross2d
• #4592 - Fix #4589. Just pass NULL for b2d_func for constant dynamic sharedme…
• **** #4588 - Let dead code removal handle removing more of the unneeded code after prange conversion to parfor.
• **** #4586 - [WIP] Fix Set for Unicode items
• #4583 - np.cumproduct
• **** #4582 - Convert sub to add and div to mul when doing the reduction across the per-thread reduction array.
• #4578 - Clarify numba ufunc supported features

Already merged

• #4593 - Enable Azure CI batching
• #4581 - Prevent taking the GIL in ForAll
• #4576 - Replace use of np.prod with functools.reduce for computing size from shape

4. Next Release: Version 0.46.0, RC=Sept 30

• Finish pending rewrite passes
• Document best practices for constructing new compiler pipelines
• Define autodiscovery system for Numba extensions (like numba_scipy or HPAT) that don't need direct user import
• Allow opt-in dispatching of functions by literal value
• Making caching aware of transitive dependencies
• Define declarative typing system for @overload (to be used in future releases)
• Numba Runtime C API for extensions to register reference-counted memory with the runtime.
• Start using new CI system in parallel with existing one
• Priority bug fixes:
  • Low performance of JIT method calls (requested by Pandas devs)
  • Others TBD