Move `_tensor_accumulation_impl `extension and use it for dpnp by vlad-perevezentsev · Pull Request #2791 · IntelPython/dpnp

vlad-perevezentsev · 2026-03-02T10:58:53Z

This PR completely moves _tensor_accumulation_impl pybind11 extension into dpctl_ext.tensor and extends dpctl_ext.tensor Python API with the functions cumulative_logsumexp, cumulative_prod and cumulative_sum reusing them in dpnp

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?

…axis

…_impl

github-actions · 2026-03-02T11:29:21Z

View rendered docs @ https://intelpython.github.io/dpnp/pull/2791/index.html

github-actions · 2026-03-02T11:39:58Z

Array API standard conformance tests for dpnp=0.20.0dev3=py313h509198e_122 ran successfully.
Passed: 1356
Failed: 1
Skipped: 16

ndgrigorian · 2026-03-03T21:23:00Z

dpctl_ext/tensor/libtensor/source/accumulators/accumulate_over_axis.hpp

+#pragma once
+
+#include <cstddef>
+#include <cstdint>


Suggested change

#include <cstdint>

believe it's unused

dpctl_ext/tensor/libtensor/source/accumulators/accumulate_over_axis.hpp

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_logsumexp.cpp

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_sum.cpp

antonwolfy · 2026-03-04T11:57:53Z

dpctl_ext/tensor/libtensor/source/accumulators/accumulate_over_axis.hpp

+
+#include <cstddef>
+#include <cstdint>
+#include <stdexcept>


Suggested change

#include <stdexcept>

#include <exception>

#include <iterator>

#include <stdexcept>

#include <tuple>

antonwolfy · 2026-03-04T11:58:24Z

dpctl_ext/tensor/libtensor/source/accumulators/accumulate_over_axis.hpp

+//===----------------------------------------------------------------------===//
+
+#pragma once
+


Suggested change

#include <algorithm>

antonwolfy · 2026-03-04T12:07:11Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_logsumexp.cpp

+                                  int trailing_dims_to_accumulate,
+                                  const arrayT &dst, sycl::queue &exec_q,
+                                  const event_vecT &depends = {}) {
+        using dpctl::tensor::py_internal::py_accumulate_over_axis;


Suggested change

using dpctl::tensor::py_internal::py_accumulate_over_axis;

antonwolfy · 2026-03-04T12:07:20Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_logsumexp.cpp

+            using dpctl::tensor::py_internal::
+                py_accumulate_final_axis_include_initial;


Suggested change

using dpctl::tensor::py_internal::

py_accumulate_final_axis_include_initial;

antonwolfy · 2026-03-04T12:07:28Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_logsumexp.cpp

+
+    auto cumlogsumexp_dtype_supported = [&](const py::dtype &input_dtype,
+                                            const py::dtype &output_dtype) {
+        using dpctl::tensor::py_internal::py_accumulate_dtype_supported;


Suggested change

using dpctl::tensor::py_internal::py_accumulate_dtype_supported;

antonwolfy · 2026-03-04T12:13:58Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_prod.cpp

+            using dpctl::tensor::py_internal::
+                py_accumulate_final_axis_include_initial;


Suggested change

using dpctl::tensor::py_internal::

py_accumulate_final_axis_include_initial;

antonwolfy · 2026-03-04T12:14:05Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_prod.cpp

+
+    auto cumprod_dtype_supported = [&](const py::dtype &input_dtype,
+                                       const py::dtype &output_dtype) {
+        using dpctl::tensor::py_internal::py_accumulate_dtype_supported;


Suggested change

using dpctl::tensor::py_internal::py_accumulate_dtype_supported;

antonwolfy · 2026-03-04T12:14:46Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_sum.cpp

+    auto cumsum_pyapi = [&](const arrayT &src, int trailing_dims_to_accumulate,
+                            const arrayT &dst, sycl::queue &exec_q,
+                            const event_vecT &depends = {}) {
+        using dpctl::tensor::py_internal::py_accumulate_over_axis;


Suggested change

using dpctl::tensor::py_internal::py_accumulate_over_axis;

antonwolfy · 2026-03-04T12:14:55Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_sum.cpp

+            using dpctl::tensor::py_internal::
+                py_accumulate_final_axis_include_initial;


Suggested change

using dpctl::tensor::py_internal::

py_accumulate_final_axis_include_initial;

antonwolfy · 2026-03-04T12:15:02Z

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_sum.cpp

+
+    auto cumsum_dtype_supported = [&](const py::dtype &input_dtype,
+                                      const py::dtype &output_dtype) {
+        using dpctl::tensor::py_internal::py_accumulate_dtype_supported;


Suggested change

using dpctl::tensor::py_internal::py_accumulate_dtype_supported;

vlad-perevezentsev added 7 commits February 27, 2026 08:57

Initialize _tensor_accumulation_impl extension and move _cumsum_over_…

4552e78

…axis

Move ti.cumulative_sum() and reuse it in dpnp

6b81e7a

Move _cumprod_over_axis to dpctl_ext.tensor._tensor_accumulation_impl

5af94c8

Add missing include

91547cc

Move ti.cumulative_prod() and reuse it in dpnp

668f3fb

Move _cumlogsumexp_over_axis to dpctl_ext.tensor._tensor_accumulation…

72d2109

…_impl

Move ti.cumulative_logsumexp() and reuse it in dpnp

d8c3680

vlad-perevezentsev self-assigned this Mar 2, 2026

vlad-perevezentsev changed the title ~~Move _tensor_accumulation_impl extensions and use it for dpnp~~ Move _tensor_accumulation_impl extension and use it for dpnp Mar 2, 2026

vlad-perevezentsev marked this pull request as ready for review March 2, 2026 20:31

vlad-perevezentsev requested review from antonwolfy and ndgrigorian as code owners March 2, 2026 20:31

vlad-perevezentsev mentioned this pull request Mar 2, 2026

Restructure dpctl, delegating all array functionality to dpnp #2743

Open

ndgrigorian reviewed Mar 3, 2026

View reviewed changes

dpctl_ext/tensor/libtensor/source/accumulators/accumulate_over_axis.hpp Show resolved Hide resolved

ndgrigorian reviewed Mar 3, 2026

View reviewed changes

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_logsumexp.cpp Show resolved Hide resolved

ndgrigorian reviewed Mar 3, 2026

View reviewed changes

dpctl_ext/tensor/libtensor/source/accumulators/cumulative_sum.cpp Show resolved Hide resolved

antonwolfy reviewed Mar 4, 2026

View reviewed changes

-#include <stdexcept>
+#include <exception>
+#include <iterator>
+#include <stdexcept>
+#include <tuple>

		//===----------------------------------------------------------------------===//

		#pragma once

		using dpctl::tensor::py_internal::
		py_accumulate_final_axis_include_initial;

Conversation

vlad-perevezentsev commented Mar 2, 2026

Uh oh!

github-actions bot commented Mar 2, 2026

Uh oh!

github-actions bot commented Mar 2, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants