Minutes_2018_07_05

Numba Meeting: 2018-07-05

Attendees: Ehsan, Siu, Stuart, Todd

#3083: numba.cuda.cudadrv.driver.LinkerError: [999] Call to cuLinkAddData results in CUDA_ERROR_UNKNOWN
#3082: Cuda array interface should support singledispatch-based overloads. feature_request
#3081: Segfault in NRT from use of generator bug
#3078: LoweringError: Failed at object (object mode frontend) bug
#3072: Segfault from test_gil_reacquire_deadlock test.
- not our fault segfault w/ workaround
#3071: Performance issue with lists of arrays?
- Siu will look at this
#3070: np.median doesn't support read-only arrays bug
- Stuart will fix
#3069: prange fails when array are changed with augmented assignment ParallelAccelerator bug
- Todd will be looking at this
#3068: Trying to initialize np.empty gives "All templates rejected" error
- ignoring issue for user
- not hard to fix
- but need changes in many places

3080 Compute np.var and np.std correctly for complex types.
- involves parfors; Ehsan/Todd will review
3079 Implement +/-(types.boolean) Fix #2624
- Siu wil review
3077 Adds optional number of processes for -m in testing
- Merged
- Need to fix --failed-first option
3067 Abort simulator threads on exception in thread block.
- Siu will review
3062 Fix cuda.In freeing devary before the kernel launch
- Merged
- Stuart - open issue to look for tool to check for package dependencies
3060 Adjust attribute error for NumPy functions.
- Depends on 3059
- Tripping over unrelated optional typing bug that needs to be resolved
3059 Handler for known LoweringErrors.
- Same unrelated problem as 3060

3046 Pairwise sum implementation.
3045 Fix #3027 - Numba reassigns sys.stdout
3038 Fix random integer generation to match results from NumPy.
- Stuart reviewed, asked for comments to explain reasoning
3033 Add support for multidimensional CFFI arrays
- Stuart reviewed, needs a few more tests
3023 Support for AMDGCN/ROCm.
- Stuart reviewed, needs Siu review
3017 Add facility to support with-contexts
#2999 Support LowLevelCallable
#2983 [WIP] invert mapping b/w binop operators and the operator module
#2977 Added docstring coding convention and specified line terminator to developer docs
#2976 [WIP] Expansion of developer documentation.
#2968 WIP: Adding xnd kernels
#2959 np.unique return_counts support
#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.
- Need to give some feedback to contributor on approach
#2817: [WIP] Emit LLVM optimization remarks
- Want to refactor to not require temp file on disk

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

NumPy accepted ARM support patch. What to do about compilers/toolchains!?
- No wheel standard for ARM
- Only requests we've had are for Raspberry Pi (ARMv7) or Jetson TX1/2 + Ubuntu 16.04 (ARMv8), so we can rely on system compilers in that case
- Berryconda provides ARMv7 package environment, and Stan has very experimental port to ARMv8 (called Jetconda to tell them apart)