Use nvcomp defaults for algo options.

[bdice]

This PR updates the nvcomp bindings to use nvcomp's defaults rather than hardcode the default options. The defaults changed in nvcomp 4.0.1, so this is needed for #449.

[jakirkham]

@madsbk @Alexey-Kamenev @thomcom could you please look over these changes when you have a moment? 🙂

[vuule]

Semantically the change looks good to me, not qualified to evaluate the (Python) code quality.

[Alexey-Kamenev]

LGTM with minor comment.

[jakirkham]

After Alexey observed the f-string typo. Saw a few more. Flagged them below

[jakirkham]

As it appears we do tests constructing compressors with default arguments, this should already be covered in the existing test suite

kvikio/python/kvikio/tests/test_nvcomp.py

Lines 195 to 217 in 3ce8bcc

	@pytest.mark.parametrize(
	"inputs",
	[
	{},
	{"chunk_size": 1 << 16, "algo": 0, "device_id": 0},
	{
	"chunk_size": 1 << 16,
	},
	{
	"algo": 0,
	},
	{
	"device_id": 0,
	},
	],
	)
	def test_gdeflate_inputs(inputs):
	size = 10000
	dtype = inputs.get("data_type") if inputs.get("data_type") else np.int8
	data = cupy.array(np.arange(0, size // dtype(0).itemsize, dtype=dtype))
	compressor = libnvcomp.GdeflateManager(**inputs)
	final = compressor.compress(data)
	assert_compression_size(len(final), LEN["Gdeflate"])

Though please let me know if I've missed anything here

[madsbk]

Looks good, I only have some type hint suggestions

[wence-]

Non-blocking nits

[wence-]

hmm, how did this ever work? I guess cython is less strict with parsing.

[jakirkham]

Both are valid, the type format we had before is called a ctuple and has been around in Cython longer

Support for typing.Tuple is relatively new

Here's an example in the docs (please see 2nd entry)

[jakirkham]

Cython already checks whether the value is an int or None based on the function signature. So the exceptions can be dropped. If we wanted to be extra careful, we could add an assert. Though likely that is unneeded

Example of this behavior demonstrated below:

In [1]: %load_ext cython

In [2]: %%cython
   ...: 
   ...: from typing import Optional
   ...: 
   ...: class Class:
   ...:     def __init__(self, v: Optional[int] = None):
   ...:         print(f"v = {v}")
   ...:         self.v = v
   ...: 

In [3]: Class()
v = None
Out[3]: <_cython_magic_96fdfa8e6c04d6dbb7c0c7398f039f6decfcbe5b.Class at 0x10524ad10>

In [4]: Class(1)
v = 1
Out[4]: <_cython_magic_96fdfa8e6c04d6dbb7c0c7398f039f6decfcbe5b.Class at 0x104f4c610>

In [5]: Class(3.14)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[5], line 1
----> 1 Class(3.14)

TypeError: Argument 'v' has incorrect type (expected int, got float)

[vuule]

Thanks everyone for the reviews! I'll wait another work day to make sure folks are happy with the way the comments like #450 (comment) have been resolved.

[bdice]

Thanks for the reviews and fixes. I am happy with this.

[bdice]

/merge