Minutes_2019_05_21
Siu Kwan Lam edited this page Jun 11, 2019
·
1 revision
Attendees: Stuart, Todd, Ehsan, Siu
- 0.44 RC status
- llvmlite RC building in the farm (almost done)
- numba pending changelog merge
-
#4098 - numpy.empty() TypingError with structured dtype
- need better error message for when
dtype
is not understood
- need better error message for when
- **** #4097 - np.kron fails to compile under njit with call signature
- reshape() should handle non-contiguous arrays
- rabbit hole of partially specified behavior
- likely we have interpreted the arguments incorrectly
- candidate for conversion from builder syntax to overload
- discussion:
- there are other numpy functions that need to be updated to use
@overload
. - we should ask help from the community to rewrite/update them.
- Stuart: write ticket for this
- there are other numpy functions that need to be updated to use
- **** #4093 - Memory leak when allocating 1D and 2D numpy arrays inside jitted functions
- several possible approaches
- Stuart: Fix it to postproc on end of the array const inliner function.
- Ticket to fix it properly.
- Investigate when ir.Del is used (direct and indirect)
- Stuart: Legalize just ahead of lowering check input == output, i.e. no change
- Lightweight check at tail of each pass to check for e.g. double ir.Del
-
#4092 - Using a non-default device in CuPy leads to error in cuPointerGetAttribute
- fails in function that is supposed to safeguard against moving pointers between contexts
- Siu will investigate, but needs a system with GPUs on different PCI-E domains
-
#4087 - Python 2.7 heterogeneous list unbox problem
- low priority because both Python 2.7 and this style of list support is going away
-
#4086 - Question: Efficiently shuffle array of fixed-width strings
- Stan can respond with permute np.arange() idea
- #4100 - Fix alignment check on 32-bit.
- #4099 - Fix record alignment test
- **** #4095 - Support ir.Global/FreeVar in find_const()
- add tests specially around branch-pruning
-
#4090 - Update to LLVM8 memset/memcpy intrinsic
- Worry about post 0.44 release
- **** #4088 - Resolves issue4075.
- **** #4085 - Resolves #3314.
- #4096 - Update env-vars for CUDA libraries lookup
- #4094 - Fix function definition finding logic for commented def
- #4091 - Deprecate the use of iternext_impl without RefType
- #4089 - Make the warnings fixer flush work for warning comparing on type.
- #4084 - Fix missing incref on optional return None
- 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
- open questions:
- Add deprecation warnings
- objmode
- list/set reflection