Sebastian Berg
Sebastian Berg
return_inverse returns something different even for `axis=0`. This was likely a mistake in NumPy 2, and we will probably revert it (not for `axis=None`). If you care for working correctly...
This implements distributed sorting for cudf-polars, it should work but is missing new sorting tests and I am also not quite sure that local tests exercised the `rapidsmpf` paths properly....
~**Draft: But only because it depends on gh-22566**~ What is the fallout of this? No idea, lets find out! The main thing that magically stops working, are comparisons like `timedelta64(3,...
The scalar methods still don't use `METH_FASTCALL` for their forwarding. Changing this would improve performance at least if kwargs are involved (maybe just moderately, these are slow anyway due to...
Sorry, I don't know the discussion, a question for @leofang or @tqchen probably. The docs for the stream say that: > If stream is -1, the value may be used...
I implemented dlpack v1 for CuPy (see https://github.com/cupy/cupy/pull/8683), and there are two choices that are important for other implementations and maybe the spec: 1. We chose to export the `cudaManaged`...
As promised, you can now [1] do a runtime check for `PyArray_RUNTIME_VERSION >= NPY_2_3_API_VERSION`. If this is the case, `NPY_MAXARGS` is irrelevant (integer number of operands supported). NumPy always raised...
`napari` uses the entry-point value for non-object references, i.e. a file path to load metadata from (i.e. as `module:file_path`). This pattern seemed like a good idea for certain use cases....
Thought a small PR is maybe good, this is **start** to test with [`pytest-run-parallel`](https://github.com/Quansight-Labs/pytest-run-parallel). The `cupy_tests/core_tests` (mostly?) run with this. I did free-threading earlier, but for now it doesn't really...
I was using sanitizers and tried compiling with clang, which did not like `-fno-gnu-unique`. Based on the history, this seems to have been to avoid symbol clashes via cub/thrust. Based...