Minutes_2018_08_16

Numba Meeting: 2018-08-16

Attendees: Ehsan, Todd, Siu, Stuart, Stan

1. New issues

• Some new segfaults
• #3229 - Frame injection for profiler can cause segfaults
  • Needs further investigation
  • Is there something wrong with our frame injection, Python specific issues?
• #3226 - CUDA JIT errors when multiple signatures provided
  • Probably an easy fix
• #3225 - Ellipsis indexing and assignment not working when needed number of : is zero
  • Array scalars should be implicitly cast
• #3223 - Importing time is taking too long
  • could hide import yaml parser inside function body
  • no other easy wins
• #3220 - Passing scalars by reference
  • Conclusion: not really possible in Python syntax
  • use tuples when possible for multiple return values
• #3219 - Note pyobject type exists in the Types and signatures
  • Easy fix to put into docs
• #3214 - Double while True loop causes strange errors
  • Root cause is fixed by #3222
  • Cascading problem resulted in infinite loop with no side effects
  • LLVM optimizer does crazy things
• #3210 - Improve serialization of AutoJitCUDAKernel
  • Results in slower execution when combining autojit CUDA kernel with distributed system like Dask

2. Open PRs

New

• 3230 - Fixes liveness analysis issue in looplifting
  • Siu will review
• 3228 - Reduce redundant module linking
  • functions linking against enormous list of global list of functions
  • Unfortunately, doesn't speed up compilation
  • Stuart will review
• 3222 - Fix #3214. Mishandling of POP_BLOCK in while True loop.
  • Stuart will review
• 3212 - Support for np.vander
  • Stuart will review
• 3211 - Handle unpacking in building tuple (BUILD_TUPLE_UNPACK opcode)
  • Siu will review
• 3209 - Support for np.tri, np.tril and np.triu
  • Waiting for author to respond to comments

Old

• 3202 - [WIP] TBB + backend refactor... DO NOT MERGE!
  • Close to done, but need some help with mystery segfaults in TBB
• 3199 Support inferring stencil index as constant in simple unary expressions
  • Need one test to verify error message when negative constant used outside
  • Triggered unrelated error in build farm. Not going to hold PR up for that, though.
• 3186 Support Records in CUDA Const Memory
  • needs Siu to check something
• 3172 Use float64 add Atomics, Where Available
  • needs re-review by Siu
• 3166 [WIP] Objmode with-block
  • Still in progress
• 3162 Support constant dtype string in nopython mode in functions like numpy.empty.
  • Need to resolve #3195
• 3160 First attempt at parallel diagnostics
  • Could merge for next release?
  • Will collecting the diagnostics add a maintenance burden?
  • How to allow other passes to added diagnostics?
  • Todd will continue review
• 3145 support for np.fill_diagonal
  • Ready to merge once CI passes
• 3142 Issue3139
  • Blocked on 3127
• 3134 [WIP] Cfunc x86 abi
  • Needs re-review
• 3127 Support for reductions on arrays.
  • Need to resolve conflict with master
  • Failure on Windows, Python 3.7
    • related to 32 vs 64 bit in arange?
  • Stuart will take a look
• 3124 Fix 3119, raise for 0d arrays in reductions
  • Need feedback from Ehsan and Todd
• 3093 [WIP] Singledispatch overload support for cuda array interface.
  • Needs review
• 3046 Pairwise sum implementation.
• #2999 Support LowLevelCallable
• #2983 [WIP] invert mapping b/w binop operators and the operator module
• #2950 Fix dispatcher to only consider contiguous-ness.
• #2942 Fix linkage nature (declspec(dllexport)) of some test functions
• #2894: [WIP] Implement jitclass default constructor arguments.
• #2817: [WIP] Emit LLVM optimization remarks

===========================

3. Feature Discussion

4. Next Release: Version 0.40, RC=Sept 3, 2018, Final=Sept 10, 2018

• Experimental python mode blocks
• Refactored threadpool interface
• AMD GPU backend
• Parallel diagnostics
• Usual collection of bug fixes