Graham Markall

Remove `resolve_argument_type()` from typing context

2

It was only used by the CUDA target, and contained CUDA-specific code in the core of Numba, which was a bit of a violation of the target abstraction. The CUDA...

Reproducer: ```python from numba import njit import numpy as np @njit def f(x): return x + 1 arr = np.array('2010', dtype='datetime64[10Y]') f(arr) ``` gives ``` $ python repro.py Traceback (most...

CUDA: Refactor handling of target options

5

The broad aim of this PR is to handle target options in CUDA more closely to how they are handled by the CPU target, with the longer-term aim of fixing...

Uses LLVM 15 by default in CI and 16 experimentally. Testing on CI for now to check for and resolve any issues.

Creating this issue to keep track of additional requests on PR #1046 and to continue the discussion - points raised, with some light editing: > * I think it might...

I can't push to #986 so this PR is for testing #986 on CI. Initially this consists of the `ohu/crt` branch with `main` from https://github.com/numba/llvmlite merged in. CC @oliverhu

Remove numba dependency

4

**EDIT:** This now removes the numba dependency, which comes transitively from cuDF. **Original description:** numba-cuda is the NVIDIA-maintained CUDA target for Numba, which depends on the numba package. This PR...

CUDA: Add overloads generated by specialization to the current dispatcher.

7

Adding overloads to the current dispatcher when specializing saves recreating the overload for subsequent calls to the unspecialized dispatcher that would have used the overload. This also has the side...

[CI TEST] Test with Opaque Pointers

3

To predict the effect of numba/llvmlite#1082. cc @rj-jesus

Migrate gpuCI to RAPIDS Ops Github Actions + self-hosted runners

1

gpuCI is being retired in favour of self-hosted runners with Github Actions. This mean the CUDA CI needs to be migrated over to use it. A rough checklist of steps:...