Refactor libtensor and backend into libtensor by ndgrigorian · Pull Request #3131 · IntelPython/dpnp

ndgrigorian · 2026-10-29T02:14:16Z

This PR proposes a refactor merging the dpnp backend pybind11 extensions into libtensor, moving libtensor to the dpnp directory (out from dpnp/tensor) and merging kernels from the dpnp backend into libtensor

This makes libtensor the sole home of all kernels and Python bindings, such that users of the libtensor library have access to the full suite of functionality from Python or C++

Slips in some small additional changes:

vm: the 2-D binary contig dispatch check now looks up both type ids (the row lookup could never be nullptr; the check still can't fire, since the earlier checks already guarantee a match) and fixes the fmax/fmin docstrings copied from fmod
lapack: cast the syevd eigenvalue output to RealT (a no-op for the supported T == RealT types); fix the "outpuy" typo and the gesvd error message naming gesv()
fix the "c-contiguos" typo (visible in the dpnp.interp error for non-contiguous input)
remove unused helpers: DispatchTable2 getters, local-memory-size overloads, and FourZeroOffsets_Indexer, FixedDimStridedIndexer and ThreeOffsets_FixedDimStridedIndexer from utils/offset_utils.hpp
CMake: remove dead branches and options, unused targets and variables, and no-op properties
coverage: the dpnp sources are listed first in the merged modules so coverage builds keep their instrumented copies of shared inline functions, and the shared static library is instrumented too
Fixes unnecessary copies of dependent event vectors

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?

github-actions · 2026-10-29T02:50:11Z

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

github-actions · 2026-10-29T03:02:23Z

Array API standard conformance tests for dpnp=0.22.0dev0=np2py314h8d9cdd5_19 ran successfully.
Passed: 1491
Failed: 0
Skipped: 6

coveralls · 2026-10-29T04:21:52Z

No base build to compare — feature/refactor-dpnp-backend-to-libtensor into master

Copilot

🟡 Changes recommended

Changing dpnp.get_include() leaves retained backend headers inaccessible through its returned path, causing existing extensions that rely on it to fail compilation.

1 open issue

Preserve access to legacy backend headers via the public include helper · New

What changed in this PR

This PR consolidates native kernels and pybind11 bindings under dpnp/libtensor, while keeping the Python tensor API in dpnp.tensor and retaining the legacy C backend.

Changes:

Redirects Python interfaces and tests to the consolidated native extensions.
Unifies kernel namespaces, shared elementwise helpers, and extension build configuration.
Updates header packaging and discovery, fixes VM dispatch lookup and LAPACK output casting, and refreshes diagnostics and changelog entries.

This pull request requires changes. The unresolved moderate issue at dpnp/dpnp_iface.py:441 (2 votes) breaks legacy header discovery: dpnp.get_include() no longer exposes the retained dpnp_iface.hpp and dpnp_iface_fptr.hpp headers. Preserve the existing lookup until deprecation, or expose those headers through the new include directory.

File	Description
`setup.py`	Packages relocated libtensor headers.
`dpnp/tests/test_umath.py`	Updates VM extension import.
`dpnp/tests/test_scipy_sparse_linalg.py`	Updates sparse extension import.
`dpnp/tests/test_array_api_info.py`	Updates device-type helper import.
`dpnp/tests/tensor/test_usm_ndarray_reductions.py`	Updates index-type helper import.
`dpnp/tests/tensor/test_usm_ndarray_manipulation.py`	Updates tensor extension import.
`dpnp/tests/tensor/test_usm_ndarray_indexing.py`	Updates tensor extension import.
`dpnp/tests/tensor/test_tensor_statistical_functions.py`	Updates floating-type helper import.
`dpnp/tests/tensor/test_tensor_array_api_inspection.py`	Updates device-type helper imports.
`dpnp/tests/tensor/elementwise/test_divide.py`	Updates scalar-division binding import.
`dpnp/tensor/_utility_functions.py`	Redirects tensor and reduction imports.
`dpnp/tensor/_usmarray.pyx`	Redirects default dtype helper import.
`dpnp/tensor/_type_utils.py`	Redirects tensor extension import.
`dpnp/tensor/_statistical_functions.py`	Redirects statistical kernel imports.
`dpnp/tensor/_sorting.py`	Redirects sorting binding imports.
`dpnp/tensor/_set_functions.py`	Redirects set-operation kernel imports.
`dpnp/tensor/_searchsorted.py`	Redirects search, indexing, and copy imports.
`dpnp/tensor/_search_functions.py`	Redirects tensor extension import.
`dpnp/tensor/_reshape.py`	Redirects reshape binding imports.
`dpnp/tensor/_reduction.py`	Redirects reduction binding imports.
`dpnp/tensor/_print.py`	Redirects tensor extension import.
`dpnp/tensor/_manipulation_functions.py`	Redirects manipulation kernel imports.
`dpnp/tensor/_linear_algebra_functions.py`	Redirects linear-algebra kernel imports.
`dpnp/tensor/_indexing_functions.py`	Redirects indexing kernel imports.
`dpnp/tensor/_elementwise_funcs.py`	Redirects elementwise binding imports.
`dpnp/tensor/_elementwise_common.py`	Redirects shared tensor imports.
`dpnp/tensor/_data_types.py`	Redirects device dtype queries.
`dpnp/tensor/_ctors.py`	Redirects constructor binding imports.
`dpnp/tensor/_copy_utils.py`	Redirects copy binding imports.
`dpnp/tensor/_clip.py`	Redirects clipping kernel imports.
`dpnp/tensor/_array_api.py`	Redirects Array API dtype queries.
`dpnp/tensor/_accumulation.py`	Redirects accumulation binding imports.
`dpnp/scipy/sparse/linalg/_iterative.py`	Redirects BLAS extension import.
`dpnp/scipy/sparse/_csr.py`	Redirects sparse import and source-path comment.
`dpnp/scipy/linalg/_utils.py`	Redirects LAPACK and tensor imports.
`dpnp/linalg/dpnp_utils_linalg.py`	Redirects LAPACK and tensor imports.
`dpnp/linalg/dpnp_iface_linalg.py`	Redirects native exception import.
`dpnp/libtensor/source/zeros_ctor.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/zeros_ctor.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/window/kaiser.cpp`	Uses consolidated Kaiser kernel namespace.
`dpnp/libtensor/source/triul_ctor.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/triul_ctor.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/tensor_sorting.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/tensor_reductions.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/tensor_linalg.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/tensor_elementwise.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/tensor_accumulation.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/statistics/statistics_py.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/statistics/sliding_window1d.hpp`	Uses consolidated sliding-window kernel namespaces.
`dpnp/libtensor/source/statistics/histogram_common.hpp`	Uses consolidated histogram kernel namespace.
`dpnp/libtensor/source/sorting/topk.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/topk.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/searchsorted.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/searchsorted.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/radix_sort.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/radix_sort.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/radix_sort_support.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/radix_argsort.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/radix_argsort.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/py_sort_common.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/py_argsort_common.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/merge_sort.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/merge_sort.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/merge_argsort.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/merge_argsort.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/isin.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/sorting/isin.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/simplify_iteration_space.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/simplify_iteration_space.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/repeat.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/repeat.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/sum.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/sum.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/reduction_over_axis.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/reduction_common.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/reduction_common.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/reduction_atomic_support.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/reduce_hypot.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/reduce_hypot.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/prod.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/prod.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/min.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/min.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/max.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/max.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/logsumexp.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/logsumexp.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/argmin.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/argmin.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/argmax.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/argmax.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/any.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/any.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/all.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/reductions/all.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/putmask.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/linear_sequences.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/linear_sequences.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/linalg_functions/dot.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/linalg_functions/dot.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/linalg_functions/dot_dispatch.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/linalg_functions/dot_atomic_support.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/full_ctor.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/full_ctor.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/eye_ctor.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/eye_ctor.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/trunc.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/trunc.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/true_divide.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/true_divide.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/tanh.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/tanh.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/tan.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/tan.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/subtract.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/subtract.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/square.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/square.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/sqrt.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/sqrt.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/spacing.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/sinh.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/sinh.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/sinc.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/sin.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/sin.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/signbit.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/signbit.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/sign.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/sign.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/rsqrt.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/rsqrt.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/round.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/round.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/remainder.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/remainder.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/reciprocal.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/reciprocal.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/real.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/real.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/radians.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/proj.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/proj.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/pow.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/pow.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/positive.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/positive.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/not_equal.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/not_equal.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/nextafter.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/nextafter.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/negative.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/negative.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/nan_to_num.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/multiply.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/multiply.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/modf.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/minimum.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/minimum.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/maximum.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/maximum.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_xor.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_xor.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_or.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_or.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_not.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_not.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_and.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logical_and.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logaddexp2.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/logaddexp.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/logaddexp.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log2.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log2.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log1p.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log1p.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log10.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log10.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/log.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/less.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/less.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/less_equal.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/less_equal.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/ldexp.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/lcm.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/isnan.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/isnan.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/isinf.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/isinf.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/isfinite.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/isfinite.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/isclose.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/interpolate.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/interpolate.cpp`	Consolidates binding and kernel namespaces.
`dpnp/libtensor/source/elementwise_functions/imag.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/imag.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/i0.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/hypot.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/hypot.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/heaviside.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/greater.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/greater.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/greater_equal.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/greater_equal.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/gcd.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/frexp.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/fmod.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/fmin.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/fmax.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/floor.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/floor.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/floor_divide.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/floor_divide.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/float_power.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/float_power.cpp`	Reuses shared elementwise type helper.
`dpnp/libtensor/source/elementwise_functions/fabs.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/expm1.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/expm1.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/exp2.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/exp2.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/exp.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/exp.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/erf_funcs.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/equal.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/equal.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/elementwise_functions_type_utils.hpp`	Declares two-output type lookup support.
`dpnp/libtensor/source/elementwise_functions/elementwise_functions_type_utils.cpp`	Implements two-output type lookup support.
`dpnp/libtensor/source/elementwise_functions/elementwise_common.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/divmod.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/degrees.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/cosh.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/cosh.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/cos.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/cos.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/copysign.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/copysign.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/conj.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/conj.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/ceil.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/ceil.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/cbrt.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/cbrt.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_xor.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_xor.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_right_shift.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_right_shift.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_or.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_or.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_left_shift.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_left_shift.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_invert.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_invert.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_count.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/elementwise_functions/bitwise_and.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/bitwise_and.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/atanh.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/atanh.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/atan2.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/atan2.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/atan.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/atan.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/asinh.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/asinh.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/asin.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/asin.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/angle.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/angle.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/add.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/add.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/acosh.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/acosh.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/acos.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/acos.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/abs.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/elementwise_functions/abs.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/device_support_queries.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/device_support_queries.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_numpy_ndarray_into_usm_ndarray.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_numpy_ndarray_into_usm_ndarray.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_for_roll.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_for_roll.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_for_reshape.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_for_reshape.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_as_contig.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_and_cast_usm_to_usm.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/copy_and_cast_usm_to_usm.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/choose.hpp`	Moves declaration into tensor binding namespace.
`dpnp/libtensor/source/boolean_advanced_indexing.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/cumulative_sum.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/cumulative_sum.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/cumulative_prod.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/cumulative_prod.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/cumulative_logsumexp.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/cumulative_logsumexp.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/accumulators_common.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/accumulators_common.cpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators/accumulate_over_axis.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators.hpp`	Updates binding-reference comment.
`dpnp/libtensor/source/accumulators.cpp`	Updates binding-reference comment.
`dpnp/libtensor/include/utils/rich_comparisons.hpp`	Updates binding-reference comment.
`dpnp/libtensor/include/kernels/window/kaiser.hpp`	Consolidates Kaiser and Bessel namespaces.
`dpnp/libtensor/include/kernels/window/hanning.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/window/hamming.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/window/blackman.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/window/bartlett.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/statistics/sliding_window1d.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/statistics/histogram.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/sorting/sort_impl_fn_ptr_t.hpp`	Updates binding-reference comment.
`dpnp/libtensor/include/kernels/sorting/radix_sort.hpp`	Updates binding-reference comment.
`dpnp/libtensor/include/kernels/putmask.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/spacing.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/sinc.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/radians.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/nan_to_num.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/modf.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/logaddexp2.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/ldexp.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/lcm.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/isclose.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/interpolate.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/i0.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/heaviside.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/gcd.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/frexp.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/fmod.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/fmin.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/fmax.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/fabs.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/erfinv.hpp`	Consolidates error-function namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/erfcx.hpp`	Consolidates error-function namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/erf.hpp`	Consolidates error-function namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/divmod.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/degrees.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/elementwise_functions/bitwise_count.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/kernels/dpnp_tensor_types.hpp`	Updates binding-reference comment.
`dpnp/libtensor/include/kernels/choose.hpp`	Consolidates kernel namespace.
`dpnp/libtensor/include/ext/dispatch_table.hpp`	Removes individual supported-type accessors.
`dpnp/libtensor/include/ext/details/validation_utils_internal.hpp`	Corrects contiguity diagnostic spelling.
`dpnp/libtensor/extensions/vm/vm_py.cpp`	Updates binding-reference comment.
`dpnp/libtensor/extensions/vm/trunc.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/tanh.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/tan.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/sub.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/sqrt.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/sqr.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/sinh.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/sin.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/rint.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/pow.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/nextafter.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/mul.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/log2.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/log1p.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/log10.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/ln.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/inv.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/i0.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/hypot.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/fmod.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/fmin.cpp`	Reuses helpers and corrects binding description.
`dpnp/libtensor/extensions/vm/fmax.cpp`	Reuses helpers and corrects binding description.
`dpnp/libtensor/extensions/vm/floor.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/expm1.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/exp2.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/exp.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/erf_funcs.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/div.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/cosh.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/cos.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/copysign.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/conj.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/common.hpp`	Corrects binary dispatch-table indexing.
`dpnp/libtensor/extensions/vm/CMakeLists.txt`	Builds against shared sources and iteration-space helpers.
`dpnp/libtensor/extensions/vm/ceil.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/cbrt.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/atanh.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/atan2.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/atan.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/asinh.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/asin.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/arg.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/add.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/acosh.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/acos.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/vm/abs.cpp`	Reuses shared elementwise helpers.
`dpnp/libtensor/extensions/sparse/CMakeLists.txt`	Uses centralized extension installation and includes.
`dpnp/libtensor/extensions/lapack/syevd.cpp`	Corrects eigenvalue output-pointer cast.
`dpnp/libtensor/extensions/lapack/lapack_py.cpp`	Updates binding-reference comment.
`dpnp/libtensor/extensions/lapack/gesvd_common_utils.hpp`	Corrects SVD diagnostics.
`dpnp/libtensor/extensions/lapack/CMakeLists.txt`	Uses centralized extension installation and includes.
`dpnp/libtensor/extensions/fft/fft_py.cpp`	Updates binding-reference comment.
`dpnp/libtensor/extensions/fft/CMakeLists.txt`	Uses centralized extension installation.
`dpnp/libtensor/extensions/blas/CMakeLists.txt`	Uses centralized extension installation and includes.
`dpnp/libtensor/extensions/blas/blas_py.cpp`	Updates binding-reference comment.
`dpnp/fft/dpnp_utils_fft.py`	Redirects FFT and tensor imports.
`dpnp/dpnp_utils/dpnp_utils_linearalgebra.py`	Redirects BLAS and tensor imports.
`dpnp/dpnp_iface.py`	Redirects tensor import and changes public header discovery.
`dpnp/dpnp_iface_window.py`	Redirects window extension import.
`dpnp/dpnp_iface_statistics.py`	Redirects statistics and elementwise imports.
`dpnp/dpnp_iface_searching.py`	Redirects tensor extension import.
`dpnp/dpnp_iface_logic.py`	Uses consolidated scalar-tolerance comparison binding.
`dpnp/dpnp_iface_indexing.py`	Uses consolidated choose and putmask bindings.
`dpnp/dpnp_iface_histograms.py`	Redirects statistics extension import.
`dpnp/dpnp_iface_bitwise.py`	Uses consolidated bitwise-count bindings.
`dpnp/dpnp_algo/dpnp_fill.py`	Redirects fill binding imports.
`dpnp/dpnp_algo/dpnp_elementwise_common.py`	Redirects tensor and VM imports.
`dpnp/__main__.py`	Reports relocated tensor include directory.
`cmake/dpnp-config.cmake`	Finds relocated libtensor headers.
`CHANGELOG.md`	Records kernel and extension consolidation.

_{Review effort: Balanced}

antonwolfy · 2026-11-01T22:37:51Z

Build time and memory: master vs this PR

Measured on a Panther Lake H Linux box (8 cores, 14 GB RAM, Xe3 iGPU) with icpx 2026.2: python scripts/build_locally.py --clean, CMAKE_BUILD_PARALLEL_LEVEL=6. The PR head and its master base were built in place in two separate trees.

	master	PR	Δ
Clean build, wall time	978.7 s	1035.0 s	+56 s (+5.8%)
Clean build, CPU time (user+sys)	3264 s	3253 s	−0.3%
Peak RSS of the largest process (`ld.lld` of `_tensor_impl`)	3216 MiB	3479 MiB	+263 MiB (+8%)
Peak RSS of all build processes (1 s sampling)	4271 MiB	5704 MiB	+1433 MiB (+34%)
Rebuild `_tensor_impl` (+ `_indexing_impl` on master)	577 s	636 s	+58.5 s (+10%)
Rebuild `_tensor_elementwise_impl` (+ `_ufunc_impl` on master)	505 s	575 s	+70.5 s (+14%)
Rebuild `_statistics_impl` + `_window_impl`	46.5 s	46.5 s	0
Installed `.so` total	192.70 MB	192.16 MB	−0.28%
Device images in the `.so` files	98.96 MB	98.97 MB	+0.01%
Shipped headers	114 files, 1.83 MB	153 files, 2.11 MB	+39 files
RSS after `import dpnp`	400.3 MiB	403.7 MiB	+0.9%
RSS after 12 calls on the L0 GPU	632.8 MiB	637.6 MiB	+0.8%
GPU memory (`drm-total-gtt`) after the same calls	118.2 MiB	106.7 MiB	−11.5 MiB

Method: clean builds are one run per side. Each rebuild deleted the modules' objects and .so and ran ninja -j6 four times, in the order master, PR, PR, master; repeats agree within 1 s. The runtime rows are medians of at least 10 fresh processes per side, run interleaved.
The slowdown is all link time: compile time is the same on both sides (object total 2643 s vs 2625 s). The six large SYCL links start only after all compile jobs are done, and the _tensor_impl link is the last to finish on both sides. choose.cpp and putmask.cpp make that link 58 s longer, which accounts for the whole delta. Merging _ufunc_impl makes the _tensor_elementwise_impl link 70 s longer than the two separate links running in parallel, but that link doesn't finish last.
The +1.4 GiB total peak depends on timing: the longer elementwise link now ends at the same time as the _tensor_reductions_impl link, so their two ld.lld processes (2.5 and 2.9 GiB) overlap. Whether that happens depends on core count and job count. The +263 MiB of the largest process happens regardless.
No runtime cost: import time and first-call (JIT) latency of the moved functions are unchanged within noise.

Copilot

🔵 Needs a closer look

Cross-module SYCL/C++ consolidation changes linking, initialization, and installed-package layout, requiring human review and runtime validation of imports and CPU/GPU kernels that source inspection alone cannot provide.

1 open issue

Preserve access to legacy backend headers via the public include helper

_{Review effort: Balanced}

- vm: check the 2-D binary contig dispatch table with both type ids, the row lookup could never be nullptr - vm: fix fmax/fmin docstrings copied from fmod - lapack: cast syevd eigenvalues output to RealT, fix gesvd typo and error message naming gesv - ext: fix "c-contiguos" typo, drop unused DispatchTable2 getters and unused queue/reserve overloads of the local memory size helpers - offset_utils: drop unused FourZeroOffsets_Indexer, FixedDimStridedIndexer and ThreeOffsets_FixedDimStridedIndexer - cmake: drop the dead BUILD_DPNP_EXT_SYCL branch, unused variables, unused *_headers interface targets, a no-op PIC property, the empty COMPILE_DEFINITIONS loop and commented-out include dirs; stop adding -fno-fast-math twice to elementwise_functions_type_utils.cpp

- Move the C++ guide to dpnp/libtensor/AGENTS.md and update the layout, ufunc steps and paths in the other AGENTS.md files - Add changelog entries for the moved headers, dpnp.get_include and the removed indexers; fix the get_include docstring and test it - Keep the dpnp sources first in the merged modules and instrument the shared helpers so coverage builds measure them; note the mixed FP flags - Drop the dead SYCL option of build_dpnp_tensor_ext and the no-op CMAKE_POSITION_INDEPENDENT_CODE property of dpnp_backend_c

simplify_iteration_space and elementwise_functions_type_utils.cpp are compiled with -fsycl in Coverage builds. Instrumenting their device code too made sycl-link fail with "Appending variables with different address spaces need to be linked!" for every module that links them.

antonwolfy

Thanks for addressing all the comments. LGTM.

The build files changed after my build-cost measurements only in coverage builds (plus source order and dead code), so those numbers still stand: +56 s clean build (+5.8%) and +263 MiB peak ld.lld RSS. Both come from the _tensor_impl link, which got longer with choose/putmask and is the critical path of the build.

Possible follow-up (not for this PR): research ways to reduce the build time. Each item needs measuring (clean build time, peak link memory, .so size, first-call JIT time) before changing anything:

Parallel device link: try -fsycl-max-parallel-link-jobs=<N> on the big modules. With -fsycl-device-code-split=per_kernel, most of the link is spent in the per-kernel device images, which are processed one after another by default.
Device code split mode: compare per_kernel with per_source/auto for link time against runtime JIT and memory.
--offload-compress cost: measure how much of the link time it takes and whether a lower compression level helps.
Link scheduling: the big links start only after all compiles finish. Find out what serializes them, and try a CMake JOB_POOL_LINK so the heavy lld runs overlap with compiles without raising peak memory.
Module granularity: splitting the largest modules (e.g. _tensor_impl) into smaller ones would let their links run in parallel instead of in one long critical-path link, at the cost of more modules.
CI caching: look into ccache/sccache for the compile step in GitHub Actions and Jenkins.

ndgrigorian · 2026-11-03T15:44:13Z

@antonwolfy
I can't merge for some reason ("failing merge requirements") so please merge when you have a chance

This PR proposes a refactor merging the dpnp backend pybind11 extensions into libtensor, moving libtensor to the dpnp directory (out from dpnp/tensor) and merging kernels from the dpnp backend into libtensor This makes libtensor the sole home of all kernels and Python bindings, such that users of the libtensor library have access to the full suite of functionality from Python or C++ Slips in some small additional changes: * vm: the 2-D binary contig dispatch check now looks up both type ids (the row lookup could never be `nullptr`; the check still can't fire, since the earlier checks already guarantee a match) and fixes the `fmax`/`fmin` docstrings copied from `fmod` * lapack: cast the `syevd` eigenvalue output to `RealT` (a no-op for the supported `T == RealT` types); fix the "outpuy" typo and the gesvd error message naming `gesv()` * fix the "c-contiguos" typo (visible in the `dpnp.interp` error for non-contiguous input) * remove unused helpers: `DispatchTable2` getters, local-memory-size overloads, and `FourZeroOffsets_Indexer`, `FixedDimStridedIndexer` and `ThreeOffsets_FixedDimStridedIndexer` from `utils/offset_utils.hpp` * CMake: remove dead branches and options, unused targets and variables, and no-op properties * coverage: the dpnp sources are listed first in the merged modules so coverage builds keep their instrumented copies of shared inline functions, and the shared static library is instrumented too * Fixes unnecessary copies of dependent event vectors bc5d6ed

ndgrigorian added a commit that referenced this pull request Oct 29, 2026

Add #3131 to changelog

5e84d56

antonwolfy added this to the 0.22.0 release milestone Oct 29, 2026

ndgrigorian added a commit that referenced this pull request Oct 30, 2026

Add #3131 to changelog

8afda54

ndgrigorian force-pushed the feature/refactor-dpnp-backend-to-libtensor branch from d6424db to 56d8d97 Compare October 30, 2026 07:55

ndgrigorian marked this pull request as ready for review October 30, 2026 08:39

ndgrigorian requested review from antonwolfy and vlad-perevezentsev as code owners October 30, 2026 08:39

ndgrigorian added a commit that referenced this pull request Oct 31, 2026

Add #3131 to changelog

0945704

ndgrigorian force-pushed the feature/refactor-dpnp-backend-to-libtensor branch from 56d8d97 to c8c00ac Compare October 31, 2026 08:59

Base automatically changed from feature/implement-radix-select-topk to master October 31, 2026 11:31

ndgrigorian added a commit that referenced this pull request Oct 31, 2026

Add #3131 to changelog

d1635a7

ndgrigorian force-pushed the feature/refactor-dpnp-backend-to-libtensor branch from c8c00ac to ff6221b Compare October 31, 2026 19:42

antonwolfy requested a balanced review from Copilot November 1, 2026 17:09

Copilot started reviewing on behalf of antonwolfy November 1, 2026 17:09

View session

Copilot AI reviewed Nov 1, 2026

View reviewed changes

Comment thread dpnp/dpnp_iface.py Resolved

antonwolfy reviewed Nov 1, 2026

View reviewed changes

Comment thread dpnp/libtensor/source/CMakeLists.txt Resolved

Comment thread dpnp/dpnp_iface.py Resolved

Comment thread CHANGELOG.md Resolved

Comment thread dpnp/CMakeLists.txt Resolved

Comment thread dpnp/__main__.py Resolved

Comment thread dpnp/libtensor/source/CMakeLists.txt Resolved

antonwolfy reviewed Nov 1, 2026

View reviewed changes

Comment thread dpnp/libtensor/source/CMakeLists.txt Resolved

antonwolfy reviewed Nov 1, 2026

View reviewed changes

Comment thread dpnp/libtensor/extensions/vm/common.hpp Resolved

ndgrigorian requested a review from antonwolfy November 2, 2026 06:19

ndgrigorian added a commit that referenced this pull request Nov 2, 2026

Add #3131 to changelog

2da3462

ndgrigorian force-pushed the feature/refactor-dpnp-backend-to-libtensor branch from e24ca7b to 2df1e76 Compare November 2, 2026 15:06

antonwolfy requested a balanced review from Copilot November 2, 2026 19:06

Copilot started reviewing on behalf of antonwolfy November 2, 2026 19:07

View session

antonwolfy reviewed Nov 2, 2026

View reviewed changes

Comment thread dpnp/libtensor/source/elementwise_functions/elementwise_functions.hpp Resolved

Comment thread dpnp/libtensor/source/elementwise_functions/elementwise_functions.hpp Resolved

Copilot AI reviewed Nov 2, 2026

View reviewed changes

antonwolfy reviewed Nov 2, 2026

View reviewed changes

Comment thread dpnp/libtensor/source/elementwise_functions/elementwise_functions.hpp Resolved

ndgrigorian added a commit that referenced this pull request Nov 2, 2026

Add #3131 to changelog

6dfe902

ndgrigorian force-pushed the feature/refactor-dpnp-backend-to-libtensor branch from 2df1e76 to 186068e Compare November 2, 2026 20:19

ndgrigorian added 7 commits November 2, 2026 13:40

Refactor libtensor and backend into libtensor

73ef7f1

Add #3131 to changelog

43ec458

Keep const-ref depends and original year in elementwise_functions.hpp

0d9a046

Fix unnecessary copies of vectors of dependent events

43ca63d

ndgrigorian force-pushed the feature/refactor-dpnp-backend-to-libtensor branch from 236c2e1 to 43ca63d Compare November 2, 2026 21:41

antonwolfy approved these changes Nov 3, 2026

View reviewed changes

antonwolfy merged commit bc5d6ed into master Nov 3, 2026
105 of 109 checks passed

antonwolfy deleted the feature/refactor-dpnp-backend-to-libtensor branch November 3, 2026 16:31

Conversation

ndgrigorian commented Oct 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

coveralls commented Oct 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

antonwolfy commented Nov 1, 2026

Build time and memory: master vs this PR

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🔵 Needs a closer look

Uh oh!

Uh oh!

antonwolfy left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ndgrigorian commented Nov 3, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

ndgrigorian commented Oct 29, 2026 •

edited

Loading

github-actions Bot commented Oct 29, 2026 •

edited

Loading

github-actions Bot commented Oct 29, 2026 •

edited

Loading

coveralls commented Oct 29, 2026 •

edited

Loading

antonwolfy left a comment •

edited

Loading