deps: Update dependency cupy to >=14.2.0

[renovate]

This PR contains the following updates:

Package	Type	Update	Change
cupy (source)	feature-cuda-backends	minor	>=14.1.1 → >=14.2.0

---

Release Notes

cupy/cupy (cupy)

v14.2.0

CuPy v14.2.0 Release Note

This release for the CuPy v14 series introduces new features, enhancements, and bug fixes.

💬 Join the Matrix chat to talk with developers and users and ask quick questions!

🙌 Help us sustain the project by sponsoring CuPy!

✨ Highlights

SciPy-compatible Sparse Array Classes

We are happy to present to you the new sparse array classes, following SciPy counterparts for CPU. We encourage users to migrate from sparse matrices to sparse arrays following SciPy’s migration guide. New user code should stick to sparse arrays.

In this release, the supported formats include COO, CSR, CSC, and DIA. All formats support 2D and COO/CSR additionally support 1D. Both 32-bit and 64-bit indexing are supported. See our sparse array documentation for more information.

Windows on Arm Support (experimental)

CuPy now offers experimental support for CUDA 13.4 Developer Preview on RTX Spark. Building CuPy from source on WoA (Windows on Arm) is now possible. Experimental WoA wheels will be uploaded to PyPI in a few days after the release are available on PyPI (uploaded on 2026-09-01).

Initial integration with nvmath-python

As part of tighter coupling with NVIDIA CUDA Python platform, CuPy has begun the integration with nvmath-python, NVIDIA Math Libraries for the Python Ecosystem. CuPy can now be built from source against nvmath-python v1.0+ with the environment variable CUPY_USE_CUDA_PYTHON=1 to access CUDA math libraries. nvmath-python will become a default dependency in a future CuPy release.

Python 3.14 free-threaded Wheels for Windows

In addition to the Linux wheels provided in v14.1, CuPy v14.2 now offers Python 3.14t wheels for Windows.

📝 Changes

See here for the complete list of merged PRs.

New Features

• Use nvmath bindings for CUDA library bindings (#​10112)
• ENH: add explicit copy arg to reshape (#​10147)
• Add cumulative_sum and cumulative_prod (Array API) (#​10200)

Enhancements

• TST: Assert cupy.linalg.solve throws LinAlgError (#​9987)
• Define NULL in cupyx.scipy.spatial Delaunay kernel string (#​9991)
• Replace deprecated HIP API calls with their non-deprecated equivalents (#​9993)
• FeistelBijection RawKernel: select <stdint.h> on HIP, keep <cuda/std/cstdint> on CUDA (#​9994)
• Resolve symlinks when deriving ROCm install path from hipcc (#​9995)
• Test with and document support for cusparseLt v0.9.1 (#​10004)
• Provide HIPRTC fallback for cupy.ndarray.byteswap (#​10006)
• Update xsf submodule (#​10027)
• Add SciPy-compatible sparse array classes (sparray) [split 1/3] (#​10032)
• ENH: It seems cooperative groups do not need rdc (anymore) (#​10038)
• Make bspline QR helpers __device__ instead of __global__ (#​10071)
• Hip/fix rocsolver potrs (#​10072)
• Fix HIP-incompatible warp shuffle in cupyx.scipy.sparse CSR row-counting kernel (#​10073)
• Sparse array (sparray) scipy-parity behavior and validation [split 2/4] (#​10079)
• Clear runtime sticky error after any failed driver-API call (#​10082)
• Maintain F-contiguity in round-trip pickle (#​10087)
• Sparse array (sparray) error handling, robustness, and cleanup [split 3/4] (#​10103)
• Add 1-D COO/CSR sparse array support (coo_array, csr_array) [split 4/4] (#​10125)
• Support SciPy 1.17 (#​10150)
• Upgrade ROCm test environment to 7.14 (#​10159)

Performance Improvements

• Avoid extra host copy when uploading a non-contiguous NumPy array (#​10115)
• PERF: Add special cases for certain scalar powers (#​10118)
• Reject CUB block reduction for short (<128) partial reductions (#​10129)
• Re-organize numpy conversion to avoid double pinned-mem check (#​10142)
• Speed up dlpack export and ignore stream for CPU requests (#​10153)

Bug Fixes

• Fix bincount with empty input array (#​9986)
• BUG: Avoid _symiirorder1_nd/NaN issues with type-casting (#​9988)
• Fix delete incompatibilities with NumPy (#​10010)
• sparse: respect copy=False for DIA offsets array (#​10011)
• BUG: Fix real sign(NaN) -> NaN and complex sign(0) -> 0 (#​10056)
• Enable NumPy 2.5 support (shape deprecation warning) (#​10057)
• BUG: ndimage: binary_erosion vs. broadcasted input (#​10068)
• Fix incorrect output for ndimage filters when output aliases input (#​10105)
• Use Python trampolines for cupyx.cutensor and cusolver (#​10128)
• Fix libcufft_static.a lookup when the CUDA path is a targets directory (#​10134)
• ENH: Add stride support to upfirdn kernels (#​10157)
• Fix leading-zero sensitivity in cupyx.scipy.signal.bilinear (#​10160)
• CI: give ci-trigger hygiene job pull-requests:write for PR labels (#​10186)
• BUG: Random integer ranges should use uint internally (#​10194)
• Relax ndimage label output dtype (#​10195)
• Fix small FT issue and CUDA 13.3 spgeam compilation/use (#​10196)
• Fix kernel cache with name expressions (#​10203)
• BUG: Fix weak integer comparison logic in old fusion path (#​10208)

Code Fixes

• fix: raise AttributeError for NumPy 2.0 removed type aliases (#​9985)
• Clean up two cupy-owned compiler warnings (rocsolver typedef + dead CAI helper) (#​10000)
• Use cuSPARSE type aliases consistently (#​10007)
• TST: Migrate fusion_tests to using pytest (#​10022)
• Add pytest support to @​testing.for_contiguous_axes decorator (#​10026)
• Remove legacy cuSparse bindings from CUDA build (#​10042)
• Cython 3.2.6 is broken, don't use it (#​10046)
• Clean up cuSPARSE shims for CUDA 12+ and remove stale APIs (#​10047)

Documentation

• Document cupy.lib.stride_tricks.sliding_window_view in API reference (#​10155)
• DOC: Add sparse arrays to rst docs (and see also) (#​10182)
• DOC: Sparse array docs follow-up, and document NumPy 2.4 / SciPy 1.17 support (#​10204)

Installation

• Add Windows ARM64 support (#​10197)
• Bump version to v14.2.0 (#​10209)

Tests

• Drop stale xfail on TestRand{int,omIntegers}2.test_goodness_of_fit_2 (#​9989)
• Skip TestUnivariateSpline on HIP (spsolve / csrlsvqr unavailable) (#​9990)
• Drop stale xfail on TestChoiceChi.test_goodness_of_fit_2 (#​9992)
• test_delaunay::test_pathological[dataset0]: skip on HIP (SIGABRT) (#​9996)
• test_lfilter_zi_with_zeros: skip zeros=(1,2) on HIP (SIGABRT in SVD) (#​9997)
• test_affine_transform_int: extend HIP xfail to matrix_shape (3, 3) and (2,3) (#​9998)
• TST: Make new pytest versions happy (#​10021)
• Migrate test_raw.py to using pytest (#​10065)
• Bump typing action versions at least a bit and fix things (#​10119)
• Unskip batched posv test with nrhs > 1 and drop stale docstring note (#​10137)
• CI: add cuda-python job with cuda13x (#​10140)
• CI: build cupy-cuda{12,13}x wheels on GH-hosted runners, gated behind /test (#​10178)
• Restore the pinned memory pool in TestAsarray instead of disabling it (#​10189)
• CI: pin rapidsai/sccache to v0.16.0-rapids.4 (#​10191)
• [v14] Skip scipy 1.17-removed tests (#​10198)
• CI: wire .pfnci test stage to GHA-built wheels (#​10201)
• CI: Add .gitattributes for generated files (#​10206)
• CI: PFN/FlexCI follow-ups on top of #​10058 (#​10207)

Others

• Narrow CODEOWNERS notifications to important metadata files (#​10025)

👥 Contributors

The CuPy Team would like to thank all those who contributed to this release!

@​acosmicflamingo @​asi1024 @​DanielCohenn @​dufp2002 @​eriknw @​gdaisukesuzuki @​gpinkert @​isVoid @​jcrist @​kmaehashi @​leofang @​mfep @​seberg @​somtri @​steppi @​thakoreh @​ViSaReVe @​yangcal

---

Configuration

📅 Schedule: (UTC)

• Branch creation
  • At any time (no schedule defined)
• Automerge
  • At any time (no schedule defined)

🚦 Automerge: Enabled.

♻ Rebasing: Whenever PR is behind base branch, or you tick the rebase/retry checkbox.

🔕 Ignore: Close this PR and you won't be reminded about this update again.

---

• If you want to rebase/retry this PR, check this box

---

This PR was generated by Mend Renovate. View the repository job log.