0.4.3 (2019-06-11)

Fixes:

Fix max|min_index to behave like numo with CUMO_COMPATIBLE_MODE=ON

0.4.2 (2019-06-11)

Fixes:

cond_unary.c: add cudaDeviceSynchronize to avoid bus error
index.c: add cudaDeviceSynchronize to avoid bus error
cum.c: add cudaDeviceSynchronize to avoid bus error

0.4.1 (2019-05-06)

Fixes:

Fix fixed_batch_norm

0.4.0 (2019-05-04)

Released (same with 0.3.5)

0.3.5 (2019-05-04)

Fixes:

Fix each_with_index to synchronize on each element

0.3.4 (2019-05-04)

Enhancements:

Support cuDNN fixed_batch_norm (cudnnBatchNormalizationForwardInference)

0.3.3 (2019-05-02)

Fixes:

Fix each to synchronize on each element

0.3.2 (2019-05-02)

Fixes:

Fix max and max_index for sfloat and dfloat

0.3.1 (2019-04-16)

Fixes:

Fix batch_norm_backward
Fix scalar.dot(scalar)
Fix clip

0.3.0 (2019-04-10)

Enhancements:

Support cuDNN
- conv (cudnnConvolution)
- conv_transpose (cudnnConvolutionBackwardData)
- conv_grad_w (cudnnConvolutionBackwardFilter)
- batch_norm (cudnnBatchNormalizationForwardTraining)
- batch_norm_backward (cudnnBatchNormalizationBackward)
- avg_pool and max_pool (cudnnPoolingForward)
- avg_pool_backward and max_pool_backward (cudnnPoolingBackward)

0.2.5 (2019-03-04)

Enhancements:

Support arithmetic sequence, which is available in ruby >= 2.6.0 (thanks to naitoh)

0.2.4 (2018-11-21)

Changes:

Turn off CUMO_SHOW_WARNING as default

0.2.3 (2018-11-17)

Enhancements:

Add some missing synchronize workarounds

0.2.2 (2018-11-13)

Enhancements:

CUDA kernelize na_index_aref_naview
CUDA kernelize na_index_aref_nadata
CUDA kernelize diagonal
CUDA kernelize copy

0.2.1 (2018-11-12)

Enhancements:

Add CUMO_SHOW_WARNING and CUMO_SHOW_WARNING_ONCE environment variables to suppress cumo warnings (They are only for debug purpose, would be removed in future).

0.2.0 (2018-11-12)

All tests in red-chainer passed.

Fixes:

Fix advanced indexing
Fix accum_index reduction (max_index, and min_index)