Fix for gh-1468 in arithmetic reduction when type promotion is needed #1470
Conversation
Removed mention of dtype kwarg in usage line
Function _reduce_over_axis promotes input array to requested result data type and carries out reduction computation in that data type. This is done in dtype if implementation supports it. If implementation does not support the requested dtype, we reduce in the default_dtype, and cast to the request dtype afterwards.
|
View rendered docs @ https://intelpython.github.io/dpctl/pulls/1470/index.html |
|
Array API standard conformance tests for dpctl=0.15.1dev1=py310ha25a700_4 ran successfully. |
|
Array API standard conformance tests for dpctl=0.15.1dev1=py310ha25a700_3 ran successfully. |
oleksandr-pavlyk
force-pushed
the
fix-gh-1468-reduction
branch
from
d5438d4 to
ff9b5eb
Compare
|
Array API standard conformance tests for dpctl=0.15.1dev1=py310ha25a700_4 ran successfully. |
dpctl/tensor/_reduction.py
Outdated
| @@ -118,7 +118,7 @@ def _reduction_over_axis( | |||
| dpt.full( | |||
| res_shape, | |||
| _identity, | |||
| dtype=_default_reduction_type_fn(inp_dt, q), | |||
| dtype=dtype, | |||
There was a problem hiding this comment.
With this change, the call to astype is no longer necessary.
It also means that when logsumexp or reduce_hypot reduce over an empty axis or array (e.g., dpt.logsumexp(dpt.ones((1, 0, 1), dtype="i4"), axis=1, dtype="i4")) you get OverflowError instead of silently casting the identity to the output type.
For now, the astype can be removed. I've experimented with removing this branch in #1465 too.
There was a problem hiding this comment.
Yes, and it should not be dtype, it should be res_dt.
|
Array API standard conformance tests for dpctl=0.15.1dev1=py310ha25a700_5 ran successfully. |
Closes gh-1468
This PR changes
_reduce_over_axiswhen an implementation to reduce using requesteddtypefor given input type does not exist. The change casts input array into a temporary array of requesteddtypeand calls reduction on the temporary (silent assumption that implementation to reduce input array using the same input dtype is available).