Minutes_2019_04_23

Attendees: Stuart, Stan, Todd, Siu, Ehsan

0. Feature Discussion

• LLVM 8 status
  • Windows problem "fixed"
  • Doing one more build and will push llvmdev to channel
• Integration test status
  • it's running now.
  • pending migration to numba org
• SSA
  • Stuart is making progress on this
  • IDOM works
  • DomTree works
  • DomFront in progress
  • To Phi or Not To Phi
• Starting Monday triage meeting next week

1. New issues

• #4008 - Formalise compiler passes
  • Todd suggested the use of futures to indicate analysis passes that need rerunning
• #4007 - @numba.njit(parallel=True) is NoOP for recarray view.
• #4006 - Bug? Expected behaviour? np.sum vs explicit loop
  • Need some ParallelAccelerator code transformations in the main compiler pipeline
• #4005 - Regression for structured arrays with titles
  • Siu will fix
• #4004 - Subclass primitive types as jitclass
  • Might run into limitations of dynamic dispatch
• #4003 - Support time functions like time.time()
  • Good idea
  • Don't think too hard about time
  • Really, don't think too hard about time
• #4002 - Hashing of jitclass doesn't work
  • Siu will investigate
• #4000 - LoweringError when using sqrt in CUDA
  • make sqrt(int) work, but emit a performance warning
• #3999 - segmentation fault on list insert
  • potential issue with list-of-unicode insertion and reflection behaviours.
• #3998 - Runtime control of environment variables (especially for Jupyter notebooks)
  • will need discussion
  • when are compiler-options baked in?
  • Need to investigate how refreshing values works
• #3997 - Error using nb.prange()
  • suspected issue with np.random.randn(a, 10)
  • need to check on list of side-effect free functions
• #3994 - ASCII flag for Numba Unicode
  • Good idea, see PR
• #3993 - Use of lists for shape args in empty(), zero(), Ones()
  • we should have a short guide of how the compiler work (Val volunteered)
• #3992 - Numba division zero error with numpy linalg functions
  • See PR
• #3989 - Support str as a construction method for strings
  • docs and user wants str(1)
  • see PR
• #3988 - Is there any way to write a "good" cuda reduce function?
• #3987 - numba -s doesn't work
• #3984 - Improve SIMD Vectorization docs
  • Yes, need to convert simd example notebook into a doc page
• #3983 - Passing jit'ed cuda device function pointer to compiled cuda code.
  • Good idea
• #3982 - Check status of "boundcheck"
  • not enabled

Already Closed

• #3986 - dict library can't import
• #3981 - Open access for numba ACM proceedings?

2. Open PRs

• #4010 - Support dict() and {}
• #4009 - Prevent zero division error in np.linalg.cond
• #4001 - ljust operation implementation
• #3996 - add is_ascii flag to UnicodeType
• #3985 - String to float conversion

already merged

• #3995 - Remove assert in type inference causing poor error message.
• #3991 - Permit Optionals in ufunc machinery
• #3990 - mention preprint repo in FAQ. Fixes #3981

Old Active

See https://github.com/numba/numba/projects/10

4. Next Release: Version 0.44, RC=May 20, Final=May 27 week

• LLVM 8
  • fixes twine bug
  • fixes common symbol bug
• Type-inferred Dict
• Caching improvements requirement gathering
  • transitive dependency
  • bug related to storing runtime-only info
  • may punt on jitclasses
  • may punt on issues of (g)ufunc
  • future: pre-filled/package-able cache
    • open questions:
      • numba version pinning(?)
      • regen cache for different numba versions
      • regen at post-install
• Add deprecation warnings
  • objmode
  • list/set reflection