Add type annotations to algorithms #895
Conversation
alexfikl
force-pushed
the
type-algorithm
branch
3 times, most recently
from
4803d16 to
6ba057f
Compare
alexfikl
force-pushed
the
type-algorithm
branch
from
6ba057f to
ba5f344
Compare
| # FIXME: other scan_dtypes below use uint? | ||
| if len(ary) > np.iinfo(np.int32).max: | ||
| scan_dtype = np.int64 | ||
| scan_dtype = np.dtype(np.int64) | ||
| else: | ||
| scan_dtype = np.int32 | ||
| scan_dtype = np.dtype(np.int32) |
There was a problem hiding this comment.
Not quite sure if this was on purposes, so wanted to bring it up?
There was a problem hiding this comment.
I used to be pretty sloppy about "dtype" vs "type". Not opposed to cleaning that up.
There was a problem hiding this comment.
Ah, I meant that this is using np.int and the other places are using np.uint?
There was a problem hiding this comment.
Oh. For copy_if, I think uint will also be OK.
| index_dtype: DTypeLike = np.int32, | ||
| key_dtype: DTypeLike = np.uint32, |
There was a problem hiding this comment.
Are these dtypes always meant to be integers? I'm guessing yes because "radix sort". (need to update key_dtype attribute if yes)
pyopencl/algorithm.py
Outdated
| double_support=all( | ||
| has_double_support(dev) for dev in self.context.devices), |
There was a problem hiding this comment.
Should this be using self.devices instead of self.context.devices? Not quite sure what the use case is for setting self.devices to something different though..
| from pytools.persistent_dict import WriteOncePersistentDict | ||
|
|
||
| import pyopencl as cl | ||
| import pyopencl._mymako as mako |
There was a problem hiding this comment.
Was this on purpose? From a quick grep, this seems to be the only place that uses _mymako.
There was a problem hiding this comment.
Yeah. Everywhere else should switch to _mymako, too. Mako isn't a hard dependency, and all _mymako does is provide a nicer error message. For type checking, you could go
if TYPE_CHECKING:
import mako
else:
import pyopencl._mymako as mako| if neutral is None: | ||
| from warnings import warn | ||
| warn("not specifying 'neutral' is deprecated and will lead to " | ||
| "wrong results if your scan is not in-place or your " | ||
| "'output_statement' does something otherwise non-trivial", | ||
| stacklevel=2) | ||
| raise ValueError("must provide a 'neutral' element in scan") |
There was a problem hiding this comment.
Setting this to None would raise an exception below in _process_code_for_macro(neutral), so it doesn't seem like it's supported at all?
2 participants
No description provided.