This is the release note of v9.0.0b1. See here for the complete list of solved issues and merged PRs.

Changes without compatibility

Deprecations

As announced in #4300, CuPy v9 no longer supports the following out-dated components:

• Python 3.5
• CUDA 9.0 (cupy-cuda90 will not be released for v9)
• cuDNN v7.5 (or earlier) and NCCL v2.3 (or earlier)
• NumPy 1.16 and SciPy 1.3

Removal of older pre-release packages from PyPI

As announced in #4360, we are going to remove pre-release wheels earlier than v6.0.0rc1 from PyPI on 2021-01-28. Those version wheels can be found at the GitHub release page of every version, and can be installed by specifying -f option:

pip install --pre cupy-cuda101 -f https://github.com/cupy/cupy/releases/v6.0.0rc1

Changes

New Features

• Adding cupyx.scipy.signal.fftconvolve (#3828)
• Add cupyx.scipy.sparse.linalg.gmres (#4236)
• Add cupyx.scipy.sparse.linalg.interface that exposes LinearOperator Feature (#4258)
• Add cupyx.scipy.sparse.csr_matrix.diagonal and cupyx.scipy.sparse.csr_matrix.setdiag (#4284)
• Support CUB-backed cupy.ReductionKernel and @cupy.fuse (#4289)
• Add batched cholesky solver (#4291)
• cupyx.ndimage.measurements: add center_of_mass, histogram, labeled_comprehension (#4311)
• Add cudaLaunchHostFunc (#4338)
• Add cupyx.scipy.sparse.linalg.spsolve_triangular (#4356)
• Add cupyx.scipy.ndimage.fourier_ellipsoid (#4361)
• Add cupyx.scipy.stats.entropy (#4369)
• Add cupy.quantile (#4370)
• Add cupyx.scipy.sparse.linalg.spsolve (#4375)
• Add tril, triu and find to cupyx.scipy.sparse (#4382)
• Add cupy.interp (#4418)
• Add support for LinearOperator to cg and gmres in cupyx.scipy.sparse.linalg (#4422)

Enhancements

• Adding C++ constructors for CArray and CIndexer (#3683)
• Fix ndarray binary operations with non-CuPy arrays to return NotImplemented (#4198)
• Change the wrapper arguments for cuBLAS L3 functions (#4263)
• Revise dtype handling in cupyx.scipy.ndimage.spline_filter (#4314)
• Update CUDA Array Interface to v3 - Part 1 (#4357)
• Inherit environment variable and detect cl.exe automatically (#4397)
• Fix poly1d (#4399)
• ndimage: support all interpolation boundary modes (#4400)
• ndimage: add grid_mode option to zoom (#4401)
• Add support for LinearOperator to eigsh and svds in cupyx.scipy.sparse.linalg (#4428)
• Update to use the new CArray and CIndexer constructors (#4463)
• Deprecations for v9 (#4479)
• Disable __cuda_array_interface__ for HIP (#4482)

Performance Improvements

• Improve scan_core in cupy/core/_routines_math.pyx (#4316)
• Improve scan in cupy/core/_routines_math.pyx (#4366)
• Fixing minor TODO in scan (#4464)

Bug Fixes

• Apply fixes to the FFT callback module (#4355)
• Fix rcond arg of linalg.lstsq (#4365)
• Fix cufft callback manager error in Windows (#4377)
• Fix linalg.lstsq for complex types (#4390)
• Fix cupy.searchsorted on HIP (#4437)
• Fix out-of-bound access in ndimage rank filters (#4439)
• Support complex types in solve_triangular (#4452)
• Eliminate read past end of array in percentile_weightnening (#4453)
• cupy.random.normal fix broadcasting of scale and loc arguments (#4457)
• Fix __cuda_array_interface__ for HIP (#4458)
• Fix overwritten issue in cumsum and cumprod (#4460)
• Fix dev info allocation (#4491)

Code Fixes

• Rename submodules under cupy.lib (#3713)
• Clean up cupy_backends (#4088)
• Replace stubs in ndimage by Jitify (#4264)
• Refactor cupyx.scipy.sparse.linalg.eigsh (#4275)
• Simplify cupy.linalg.inv (#4293)
• Make names of test classes start with Test (#4320)
• Refactor traceback in testing.helper (#4442)

Documentation

• Add scipy.fft module to the API comparison table (#4278)
• Fix docs of cupy.random functions/methods (#4319)
• Update links to forums in README (#4345)
• Incorrect environment variable description (#4410)
• Add spsolve_triangular to the reference (#4438)
• Fix doc of LinearOperator (#4441)

Installation

• Remove broken support for macOS (#3857)
• Fix parallel build (#4348)
• Add support for cuTENSOR 1.2.2 (#4404)
• Disentangle HIP from CUDA in the build script (#4416)

Tests

• Set __module__ attr to parameterized class (#4239)
• testing.numpy_cupy_allclose with per-dtype tolerance (#4269)
• Refactor fft_tests: unittest -> pytest (#4287)
• Fix testing.helper to work without unittest (#4304)
• Refactor ndimage_tests (#4307)
• Update [jenkins] requirement (#4325)
• Add broadcast test for cupy.vectorize (#4341)
• Parameterize sparse matrix tests (#4368)
• Remove travis (#4373)
• Fix some FFT test issues (#4385)
• Refactor test of linalg.lstsq (#4388)
• Require SciPy >= 1.4 in TestGmres (#4394)
• Run FFT callback test only on Linux (#4411)
• Refactor TestLinearOperator (#4413)
• Require SciPy 1.4 in TestLinearOperator (#4420)
• Fix RawKernel test (#4472)
• Exclude unsupported dtypes for TestOrderFilter (#4477)
• Fix TestEntropy: Requires scipy>=1.4.0 for axis parameter (#4481)
• Skip large size test for hip (#4461)

Others

• Drop support for Python 3.5 in CuPy v9 (#4340)
• Configure Mergify to check GitHub Actions instead of Travis (#4378)
• Bump version to v9.0.0b1 (#4499)

Contributors

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

@anaruse @carterbox @coderforlife @dmargala @grlee77 @leofang @mor2code @venkywonka @wphicks