-
Notifications
You must be signed in to change notification settings - Fork 1.1k
Minutes_2019_09_03
Siu Kwan Lam edited this page Oct 14, 2019
·
1 revision
Attendees: Siu, Stan, Aaron, Ehsan, Pearu, Stuart, Val, Todd, James, Tianhong
-
@lower_cast etc issues,
RefType
, and no implicit casting?!- forbidden implicit cast?
- still implicit for user
- re-use
Expr.cast
- better
- adding
RefType
likeiterator_impl
? - discussion at: https://github.com/numba/numba/issues/4494
- forbidden implicit cast?
-
should we fix anything on jitclass? use it in prod for e.g. numba-numba_scipy
- jitclass based API breaks "turn @jit off"
- we'd want to fix inconsistencies
- hopefully in a more fundamental level
- but also need quick fixes to address current user problems
-
compiler.py rewrite update PR#4513
- Why changing the passes?
- motivated by:
- lack of definition of a pass
- lack of contract (i.e. what is modified? what a pass depends on?)
- new code provides:
- structures; i.e. analysis pass, transformation pass
- strict ordering/sequence; i.e inter-pass dependency
- auto-recompute stale states
- motivated by:
- Why changing the compiler (pipeline)?
- tangled states
- needs better modularity
- Why changing the passes?
-
literally update PR#4467
- literal types for:
int
,str
,function
- literal types for:
Other open PRs:
- should be ready to merge. autojit https://github.com/numba/numba/pull/4331
- wip bounds checking https://github.com/numba/numba/pull/4432
- Stuart investigating trouble with ir.Loc
- Aaron proposed a way to raise exception base on runtime value
- still having some test failures.
- #4505 - Trouble converting between typed list and array with string values
- **** #4504 - Errors with various
reduce
calls - is fixing this going to break anything else? - should be fine - #4503 - Support the array_equals / array_equiv numpy function
- #4502 - Boxer doesn't handle dispatcher type
- #4500 - GUFuncs created in main are not serializable
- **** #4499 - Accept Numpy (1.17+) Generator and BitGenerator objects in jitted code
- when?!
- #4497 - Add unicode types in the documention
- #4496 - isinstance on jitclass instances returns False on parent class
- #4495 - Jitclass method call fails with keyword argument
-
#4494 -
lower_constant
andlower_cast
should probably return new references -
#4492 - No matching definition when using
deferred_type
- #4490 - Error while instanciating a jitclass with optional arguments in a jit function
- #4489 - Loops in Prange produces incorrect result
- #4488 - numba.errors.LoweringError: Failed in object mode pipeline (step: object mode backend) array.3
- #4487 - prange fail
- **** #4486 - [feature request] user defined priorities of automatic threading layer choice
- #4485 - Detect workqueue re-entrant to avoid deadlock
- **** #4506 - [WIP] Preparations to make
@lower_cast
and@lower_constant
return new references - #4501 - Add pre-commit hooks
- #4493 - Fix Overload Inliner wrt CUDA Intrinsics
- Finish pending rewrite passes
- Python 3.8 support
- 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