Allow Python scalars as the search values in `dpt.searchsorted` by ndgrigorian · Pull Request #2225 · IntelPython/dpctl

ndgrigorian · 2026-01-06T05:00:25Z

This PR proposes permitting the search values of dpt.searchsorted to be Python scalars, a change coming to the 2026 array API spec

Resolves #2224

Have you provided a meaningful PR description?
Have you added a test, reproducer or referred to an issue with a reproducer?
Have you tested your changes locally for CPU and GPU devices?
Have you made sure that new changes do not introduce compiler warnings?
Have you checked performance impact of proposed changes?
Have you added documentation for your changes, if necessary?
Have you added your changes to the changelog?
If this PR is a work in progress, are you opening the PR as a draft?

github-actions · 2026-01-06T05:27:32Z

View rendered docs @ https://intelpython.github.io/dpctl/pulls/2225/index.html

github-actions · 2026-01-06T05:36:10Z

Array API standard conformance tests for dpctl=0.22.0dev0=py310h93fe807_104 ran successfully.
Passed: 1114
Failed: 44
Skipped: 82

github-actions · 2026-01-06T07:07:22Z

Array API standard conformance tests for dpctl=0.22.0dev0=py310h93fe807_104 ran successfully.
Passed: 1115
Failed: 43
Skipped: 82

github-actions · 2026-01-06T07:56:22Z

Array API standard conformance tests for dpctl=0.22.0dev0=py310h93fe807_105 ran successfully.
Passed: 1115
Failed: 43
Skipped: 82

coveralls · 2026-01-06T07:57:25Z

coverage: 86.248% (+0.003%) from 86.245%
when pulling 65d2651 on allow-scalars-searchsorted-search-values
into 4c57ee7 on master.

github-actions · 2026-01-07T00:04:25Z

Array API standard conformance tests for dpctl=0.22.0dev0=py310h93fe807_106 ran successfully.
Passed: 1115
Failed: 43
Skipped: 82

github-actions · 2026-01-07T00:52:27Z

Array API standard conformance tests for dpctl=0.22.0dev0=py310h93fe807_106 ran successfully.
Passed: 1114
Failed: 44
Skipped: 82

antonwolfy · 2026-01-08T12:00:41Z

dpctl/tensor/_searchsorted.py

+
+    if not isinstance(x2, usm_ndarray):
+        x2 = dpt.asarray(x2, dtype=dt2, usm_type=res_usm_type, sycl_queue=q)
+    if x2.dtype != dt:


Suggested change

if x2.dtype != dt:

elif x2.dtype != dt:

this is not meant to be an elif. x2 is taken to dt2 first if not usm_ndarray then is cast to dt which is the resolved dtype. This way, it passes through dt2 under all circumstances.

Then it's unclear for me, why do we need to run 2 kernels in that case?

to convert a scalar to a temporary usm_ndarray and to copy back to the result

to cast the array to dt

What is a drawback to pass dt directly to asarray call?

Also, in the sequence:

cast x1 to dt

cast x2 to usm_ndarray if any

cast x2 to dt

We have to make step 2 as the 1st or as the last in the sequence, but not in the middle, because currently step 2 depends on events from step 1, but shouldn't.
And from other side, step 2 produces depending events which we must pass to step 3, but we didn't (asarray must be completed before we can start casting x2 to dt).

Then it's unclear for me, why do we need to run 2 kernels in that case?

to convert a scalar to a temporary usm_ndarray and to copy back to the result

to cast the array to dt

What is a drawback to pass dt directly to asarray call?

with these Python scalars, it's what we've been doing in general (see: BinaryElementwiseFunc for example) and on top of this, it avoids possible undefined behavior. I can't immediately come up with examples, but it can be the case that

a = dpt.asarray(sc, dtype=dt)

is not necessarily the same as

tmp = dpt.asarray(sc, dtype=o_dt) a = dpt.astype(tmp, dt)

but practically speaking we can reassess if any of these cases come up now that we've made some changes (like with unsigned ints)

Also, in the sequence:

cast x1 to dt

cast x2 to usm_ndarray if any

cast x2 to dt

We have to make step 2 as the 1st or as the last in the sequence, but not in the middle, because currently step 2 depends on events from step 1, but shouldn't. And from other side, step 2 produces depending events which we must pass to step 3, but we didn't (asarray must be completed before we can start casting x2 to dt).

great point, I'll adjust the dependency logic

dpctl/tensor/_searchsorted.py

dpctl/tests/test_usm_ndarray_searchsorted.py

github-actions · 2026-01-12T08:07:47Z

Array API standard conformance tests for dpctl=0.22.0dev0=py310h93fe807_110 ran successfully.
Passed: 1114
Failed: 44
Skipped: 82

dpctl/tensor/_searchsorted.py

antonwolfy · 2026-01-12T13:56:20Z

dpctl/tensor/_searchsorted.py

+
+    if not isinstance(x2, usm_ndarray):
+        x2 = dpt.asarray(x2, dtype=dt2, usm_type=res_usm_type, sycl_queue=q)
+    if x2.dtype != dt:


Then it's unclear for me, why do we need to run 2 kernels in that case?

to convert a scalar to a temporary usm_ndarray and to copy back to the result

to cast the array to dt

What is a drawback to pass dt directly to asarray call?

antonwolfy · 2026-01-12T14:02:00Z

dpctl/tensor/_searchsorted.py

+
+    if not isinstance(x2, usm_ndarray):
+        x2 = dpt.asarray(x2, dtype=dt2, usm_type=res_usm_type, sycl_queue=q)
+    if x2.dtype != dt:


Also, in the sequence:

cast x1 to dt

cast x2 to usm_ndarray if any

cast x2 to dt

We have to make step 2 as the 1st or as the last in the sequence, but not in the middle, because currently step 2 depends on events from step 1, but shouldn't.
And from other side, step 2 produces depending events which we must pass to step 3, but we didn't (asarray must be completed before we can start casting x2 to dt).

permit Python scalars as second argument to dpt.searchsorted

e85940e

ndgrigorian force-pushed the allow-scalars-searchsorted-search-values branch from b7de201 to e85940e Compare January 6, 2026 06:31

fix searchsorted validation test

fff27a8

add test for python scalar second argument for searchsorted

b7c56e1

ndgrigorian force-pushed the allow-scalars-searchsorted-search-values branch from bda9809 to b7c56e1 Compare January 7, 2026 00:16

ndgrigorian marked this pull request as ready for review January 7, 2026 03:22

ndgrigorian requested review from antonwolfy and vlad-perevezentsev as code owners January 7, 2026 03:22

antonwolfy reviewed Jan 8, 2026

View reviewed changes

ndgrigorian added 2 commits January 11, 2026 23:26

do not sequence copy and cast for x1 and x2 in searchsorted

1b9c2ec

parametrize zeros in searchsorted python scalar test

65d2651

ndgrigorian requested a review from antonwolfy January 12, 2026 07:32

antonwolfy reviewed Jan 12, 2026

View reviewed changes

Conversation

ndgrigorian commented Jan 6, 2026

Uh oh!

github-actions bot commented Jan 6, 2026

Uh oh!

github-actions bot commented Jan 6, 2026

Uh oh!

github-actions bot commented Jan 6, 2026

Uh oh!

github-actions bot commented Jan 6, 2026

Uh oh!

coveralls commented Jan 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Jan 7, 2026

Uh oh!

github-actions bot commented Jan 7, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Jan 12, 2026

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

coveralls commented Jan 6, 2026 •

edited

Loading