CuPy v7.0.0b2 を リリースしました。

[ecas...@preferred.jp]

CuPy v7.0.0b2 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 v7.0.0b2. See here for the complete list of solved issues and merged PRs.

Highlights

• cupy.cutensor has been introduced that wraps cuTENSOR, allowing high-performance tensor operations. Examples are available here.

Changes without compatibility

• cupy.load now specifies allow_pickle=False by default to follow the security fix made in NumPy 1.16.3 (see numpy/numpy #13359 and cupy/cupy #2290 for details). Most users should not be affected by this change; users loading ndarray serialized using pickle may need to explicitly specify allow_pickle=True.

New Features

• Support cuTENSOR (#2210)
• Add nansum and nanprod support (#2252, thanks @pentschev!)

Enhancements

• Raise error when no available algorithm found by cudnnFindConvolution* (#2234)
• Add NHWC layout support to batch normalization (#2235)
• Remove unused code (#2255)
• Support non-square matrices in lu_factor (#2286, thanks @econtal!)
• Remove cupy.ndarray.{nansum/nanprod} (#2292)
• Remove warning about nvcc absence (#2299)

Performance Improvements

• Add support for merge path algorithm (csrmvEx) when csr_matrix multiply with a dense vector (#2287, thanks @wonghang!)

Bug Fixes

• Add wrappers for can_cast, common_type and result_type functions (#2249, thanks @pentschev!)
• Make __cuda_array_interface__()['strides'] be tuple (#2260, thanks @leofang!)
• Avoid CUDNN_STATUS_BAD_PARAMS (#2261, thanks @himkt!)
• Support allow_pickle in cupy.load and cupy.save (#2290)

Documentation

• Update installation guide (#2184)
• Document interoperability with mpi4py (#2270, thanks @leofang!)

Installation

• Make nvcc code generation target configure by env var (#2293)

Tests

• Fix test failure when cuDNN is unavailable (#2284)
• Enhance tests for type routines (#2306)