Update arithmetic_device_uvector_t (device_span_t) to support int8_t and int16_t types.

[seunghwak]

This updates type erased vectors/spans (arithmetic_device_uvector_t and arithmetic_device_span_t) to support int8_t/int16_t.

This is a per-requisite to support int16_t edge type in graph creation.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[seunghwak]

/ok to test 9f1d033

[seunghwak]

/ok to test 3f4bd9b

[seunghwak]

1
FAILED tests/internals/test_replicate_edgelist_mg.py::test_mg_replicate_edgelist[False-True-False-False-dataset2] - RuntimeError: non-success value returned from replicate_edgelist: CUGRAPH_UNKNOWN_ERROR ERROR: Data type INT32 not allowed for weight type (valid types: FLOAT32, FLOAT64).

Do we use int32_t for weight_t?

[seunghwak]

/ok to test 88bc842

[ChuckHastings]

1
FAILED tests/internals/test_replicate_edgelist_mg.py::test_mg_replicate_edgelist[False-True-False-False-dataset2] - RuntimeError: non-success value returned from replicate_edgelist: CUGRAPH_UNKNOWN_ERROR ERROR: Data type INT32 not allowed for weight type (valid types: FLOAT32, FLOAT64).

Do we use int32_t for weight_t?

I suspect this was not just not converted and was passing because we didn't have the check you just added.

At one time we used the edge weight column to pass edge ids. Perhaps this is leftover from that test?

[seunghwak]

/ok to test acc6cd0

[seunghwak]

def _mg_call_plc_replicate(
    client: dask.distributed.client.Client,
    sID: bytes,
    dask_object: dict,
    input_type: str,
    col_names: list,
) -> Union[dask_cudf.DataFrame, dask_cudf.Series]:
    if input_type == "dataframe":
        result = [
            client.submit(
                _call_plc_replicate_dataframe,
                sID,
                edata,
                workers=[w],
                allow_other_workers=False,
                pure=False,
            )
            for w, edata in dask_object.items()
        ]
    elif input_type == "dataframe":
        result = [
            client.submit(
                _call_plc_replicate_series,
                sID,
                edata,
                workers=[w],
                allow_other_workers=False,
                pure=False,
            )
            for w, edata in dask_object.items()
        ]

Not directly related to this PR, but this code looks suspicious as well.

    elif input_type == "dataframe":

Shouldn't "dataframe" here be "series"?