CuPy v8.0.0b1 をリリースしました

[ecas...@preferred.jp]

CuPy v8.0.0a1 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 v8.0.0b1. See here for the complete list of solved issues and merged PRs.

Highlights

CuPy gets faster and more stable towards its v8.0.0 release. This version adds a handful of new routines, adds library wide performance improvements and corrects several bugs.

Changes without compatibility

• Removed cupy.scatter_add, which had been deprecated since CuPy v4. Use cupyx.scatter_add instead.

New Features

• Add get_global() to cupy.RawModule (#2510, thanks @leofang!)
• Support multi-GPU in cupy.cuda.cufft.Plan1d (#2644, thanks @leofang!)
• Add hstack, vstack, and bmat to cupyx.scipy.sparse (#2665, thanks @cjnolet!)
• Add cupy.require (#3083, thanks @niteya-shah!)
• Add cupy.compress (#3103, thanks @Harshan01!)
• Add cupy.ravel_multi_index (#3104, thanks @grlee77!)
• Add cupy.extract (#3109, thanks @Harshan01!)
• Add cupy.bitwise_not as alias to invert (#3120, thanks @Harshan01!)
• Add cupy.argwhere (#3135, thanks @rushabh-v!)
• Add cupy.select (#3138, thanks @niteya-shah!)
• Add cupy.cuda.ExternalStream (#3141)
• Add cupy.array_equal (#3189, thanks @rushabh-v!)

Enhancements

• Add ndarray variants AND inplace support in fallback_mode (#2391, thanks @Piyush-555!)
• Support array-like start/stop and add axis argument to linspace (#2461, thanks @grlee77!)
• Add fp16 support of CUB (#2600, thanks @y1r!)
• Raise errors instead of assertion on array type checks (#2795)
• Drop support for NumPy 1.15 or earlier (#2938)
• Import using_allocator in cupy.cuda (#2951, thanks @jakirkham!)
• Remove __future__ imports (#2995)
• Support CUB prod (#3067, thanks @leofang!)
• Remove deprecated cupy.scatter_add (#3074)
• Update cupy.pad to use cupy.linspace instead of numpy.linspace internally (#3101, thanks @grlee77!)
• Histogram update: support range, weights and density (#3124, thanks @grlee77!)
• Add support for ord = 2, -2, and 'nuc' in cupy.linalg.norm (#3130, thanks @rushabh-v!)
• Use ElementwiseKernel in cupy.fill_diagonal (#3139)
• Allow dia_matrix creation from SciPy equivalent (#3160, thanks @jakirkham!)
• Add labels in the benchmark and add kwargs to repeat (#3172, thanks @rushabh-v!)
• Add out parameter to cupy.concatenate and cupy.stack (#2983)
• Fix reshape to raise ValueError for order 'K' (#3123)

Performance Improvements

• Improve cuDNN performance when using deterministic mode (#1380)
• Improve performance of cumsum and cumprod (#2907)
• Improve ndimage convolve and correlate (#3179)
• Add check of c_contiguous when indexing CArray (#3191)

Bug Fixes

• Fix an issue using non-existing attribute in cub.pyx (#2985)
• Use size_t nbytes in __cuda_array_interface__ (#3009, thanks @jakirkham!)
• Fix fill_diagonal (#3011)
• Fix cupy.random.multivariate_normal (#3018, thanks @espg!)
• Use Python scalar as random seed (#3054)
• Properly decrement total bytes in memory pool (#3068)
• Fix condition to use slice copy in ndarray.__setitem__ (#3088)
• Fix compiler bug when building cupy.cuda.cub with CUDA < 9.2 (#3089, thanks @leofang!)
• Fix cub_reduction for CUPY_CUB_MIN and float16 arrays (#3100)
• Use time.process_time instead of time.clock (#3128, thanks @rushabh-v!)
• Add support for 0 sized matrices in svd (#3140, thanks @rushabh-v!)
• Fix CUB-based cupy.prod for half precision (#3148, thanks @leofang!)
• Fix error type and message in coo_matrix (#3150)
• Allow MatDescriptor to be pickle-able (#3157, thanks @jakirkham!)
• Fix erfinv & erfcinv in cupyx.scipy.special (#3159, thanks @leofang!)
• Remove some xfails in sorting tests (#3167)
• Fix Event.__del__ behavior on shutdown` (#3176)
• Add the missing initialization value in the reduction test (#3194, thanks @leofang!)

Code Fixes

• Clean up internal.pyx (get_contiguous_strides) (#1950)
• Remove custom tempdir context manager (#3003)
• Use intptr_t instead of size_t for cuSPARSE and cuBLAS handles (#3081, thanks @Harshan01!)
• Use intptr_t for cuDNN handles (#3082, thanks @Harshan01!)
• Minor fix to using_allocator (#3094)
• Remove IndexOrValueError (#3096)
• Remove unused argument in fill_diagonal (#3171)
• Silence sign comparison warning (cont'd) (#3181, thanks @leofang!)
• Avoid enum comparison (-Wenum-compare) (#3182, thanks @leofang!)
• Remove unused, deprecated fields from cudaPointerAttributes (#3183, thanks @leofang!)

Documentation

• Update installation guide for conda-forge (#3052, thanks @leofang!)
• Include UnownedMemory in the API docs (#3086, thanks @jakirkham!)
• Fix gencode example in the doc (#3147, thanks @leofang!)
• Document convolve and correlate (#3161, thanks @jakirkham!)

Examples

• Add mpi4py examples (#3049, thanks @leofang!)

Tests

• Added a compute capability check for testing grid sync (#3051)
• Fix tolerance of fft tests (#3056)
• Skip irfft tests for compute capability != 7 (#3084)
• Rewrite tests not to use numpy_cupy_raises cupyx.* tests (#3099)
• Rewrite manipulation tests not to use numpy_cupy_raises (#3122)
• Remove python 2.7 builds (#3162)

Others

• Add copyright notice for Random Kit (#3107)
• Bump version to v8.0.0b1 (#3204)