CuPy v8.0.0b2 をリリースしました

2 views
Skip to first unread message

ecas...@preferred.jp

unread,
Apr 23, 2020, 4:08:13 AM4/23/20
to CuPy Japanese User Group

CuPy v8.0.0b2 をリリースしました! リリースノートは以下の通りです。



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_filtermaximum_filtergrey_closinggrey_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

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)
Reply all
Reply to author
Forward
0 new messages