This is the release note of v8.0.0b2. See here for the complete list of solved issues and merged PRs.

We are planning to drop support for CUDA 8.0 / 9.1 (#3301) and cuDNN (#3303) in future v8 releases. If you have any concerns, please feel free to leave a comment in these issues.

New Features

• Add notification support for fallback_mode (#2279, thanks @Piyush-555!)
• Support multi-GPU cupy.cuda.cufft.Plan1d (#2644, thanks @leofang!)
• Add cupy.median (#3134, thanks @Harshan01!)
• Add cupy.flatiter (#3165)
• Add cupy.gcd and cupy.lcm (#3190, thanks @niteya-shah!)
• Support cusolverDn<t>gesvdj and cusolverDn<t>gesvdaStridedBatched (#3192)
• Add cupyx.scipy.ndimage.label (#3210)
• Add cupyx.scipy.ndimage.grey_erosion and cupyx.scipy.ndimage.grey_dilation (#3216)
• Add cupy.diag_indices and cupy.diag_indices_from (#3217, thanks @rushabh-v!)
• Support cusparse<t>csrgeam2 and cusparse<t>csrgemm2 (#3220)
• Add minimum_filter, maximum_filter, grey_closing, grey_opening to scipy.ndimage (#3239)
• Support cusolverDn<t>gesvdjBatched (#3247)
• Add cupy.kaiser (#3268, thanks @niteya-shah!)
• Support all dtypes in every sorting function in cupy.cuda.thrust (#3286, thanks @leofang!)

Enhancements

• Add R2C/C2R support to cupy.cuda.cufft.PlanNd (#3102, thanks @leofang!)

• Make RawKernel and RawModule aware of CUDA context (alt) (#3201, thanks @leofang!)

• Make diff return AxisError for an invalid axis (#3231, thanks @grlee77!)

• Improve the efficiency of cupy.pad for some simple cases (#3281, thanks @grlee77!)

• HIP

  • Support einsum with complex in HIP (#3203)
  • Add complex support to HIP Blas (#3206)

Performance Improvements

• Reduce list and tuple creation in _kernel and reduction (#2702)
• Remove unnecessary Arg instantiation in cuda/function.pyx (#3253)
• Improve norm (#3278)

Bug Fixes

• Fix: n-dimensional FFTs must preserve array contiguity when copying a view (#3034, thanks @grlee77!)
• Use larger type to represent index range in cupy.take (#3118)
• Fix byte buffer handling to support PyPy (#3225)
• Fix _reduce_dims call in reduction (#3262)
• Raise IndexError for R2C/C2R FFT with axes=() (#3264, thanks @leofang!)
• Code fix + bug fix for cupy.cuda.thrust (#3291, thanks @leofang!)

Code Fixes

• Remove cupy/cuda/_environment.py (#3145, thanks @leofang!)
• Fix cupy.fill_diagonal to implement with cupy.flatiter (#3207)
• Remove unreachable code (#3235)
• Refactor __array_function__ (#3236)
• Simplify TestEigenvalue (#3288)

Documentation

• Small typo/spelling fixes (#3243, thanks @svlandeg!)
• Use Sphinx 2.x on Read the Docs (#3272)

Tests

• Fix overfow in matmul test (#2403)
• Add cuTENSOR test (#3037)
• Rewrite some tests not use numpy_cupy_raises (#3155)
• Rewrite tests not use numpy_cupy_raises (#3256)