Nnabla-ext-cuda

release-note-bugfix
+ [Fix: copy function was missing in MaxPoolingCudnn](https://github.com/sony/nnabla-ext-cuda/pull/391)

release-note-core
+ [Fused weight decay into Solver update / Fix SgdW / Add Lamb solver](https://github.com/sony/nnabla-ext-cuda/pull/388)
+ [Enable customizing std types and functions with memory allocation](https://github.com/sony/nnabla-ext-cuda/pull/385)

release-note-op-layer
+ [Feature/20210608 cholesky decomposition](https://github.com/sony/nnabla-ext-cuda/pull/390)
+ [Speedup SortCuda with cub::DeviceSegmentedRadixSort](https://github.com/sony/nnabla-ext-cuda/pull/384)

release-note-utility
+ [Auto restart DALI iterator at StopIteration](https://github.com/sony/nnabla-ext-cuda/pull/389)
+ [Support NVTX range events](https://github.com/sony/nnabla-ext-cuda/pull/386)

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda102 For CUDA version 10.2 users


You can also install the cuda extension with specific versions from one of the following. [See also FAQ](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-build
+ [Replace pow to products](https://github.com/sony/nnabla-ext-cuda/pull/382)
+ [drop support python3.6 and CUDA-cuDNN 10.0/7](https://github.com/sony/nnabla-ext-cuda/pull/378)
+ [Support dynamic load mpi library](https://github.com/sony/nnabla-ext-cuda/pull/375)
+ [Feature/20211203 unify dockerfiles](https://github.com/sony/nnabla-ext-cuda/pull/374)

release-note-core
+ [restrict condition with dst for communicator reduce](https://github.com/sony/nnabla-ext-cuda/pull/380)
+ [Add computation graph active inputs handling.](https://github.com/sony/nnabla-ext-cuda/pull/373)

release-note-op-layer
+ [Change random_choice to use faster cumulative summing](https://github.com/sony/nnabla-ext-cuda/pull/379)

release-note-utility
+ [caching device properties for avoiding slow recall](https://github.com/sony/nnabla-ext-cuda/pull/381)

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda102 For CUDA version 10.2 users


You can also install the cuda extension with specific versions from one of the following. [See also FAQ](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-bugfix
+ [Fix FusedBN not to fall back to cpu implementation always](https://github.com/sony/nnabla-ext-cuda/pull/372)
+ [Update deformable_convolution.cu](https://github.com/sony/nnabla-ext-cuda/pull/369)
+ [Add AssignCuda not to perform host-device copy](https://github.com/sony/nnabla-ext-cuda/pull/366)
+ [Use get_current_device_attribute rather than properties](https://github.com/sony/nnabla-ext-cuda/pull/364)

release-note-build
+ [upgrade g++ from 4.9.2 to 6.5.0 and enable stdc++14](https://github.com/sony/nnabla-ext-cuda/pull/363)

release-note-doc
+ [Fix spelling "Currently"](https://github.com/sony/nnabla-ext-cuda/pull/365)

release-note-examples
+ [make corresponding modification for the files in reconstruction directories](https://github.com/sony/nnabla-ext-cuda/pull/370)

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda102 For CUDA version 10.2 users


You can also install the cuda extension with specific versions from one of the following. [See also FAQ](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-bugfix
+ [Remove the argument "output_mask" from Dropout](https://github.com/sony/nnabla-ext-cuda/pull/361)
+ [fix multithread issues](https://github.com/sony/nnabla-ext-cuda/pull/359)

release-note-build
+ [Add PYTEST_OPTS for pytest parallel execution](https://github.com/sony/nnabla-ext-cuda/pull/356)
+ [support python3.9](https://github.com/sony/nnabla-ext-cuda/pull/353)

release-note-op-layer
+ [Support broadcast in instance norm kernel to improve performance](https://github.com/sony/nnabla-ext-cuda/pull/358)
+ [Make ISTFT consistent with PyTorch implementation (NOLA condition)](https://github.com/sony/nnabla-ext-cuda/pull/357)
+ [Add linspace function](https://github.com/sony/nnabla-ext-cuda/pull/355)
+ [Optimize CumProd/CumSum](https://github.com/sony/nnabla-ext-cuda/pull/354)

release-note-utility
+ [Default disable watchdog and enable it with environment variable](https://github.com/sony/nnabla-ext-cuda/pull/362)

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda102 For CUDA version 10.2 users


You can also install the cuda extension with specific versions from one of the following. [See also FAQ](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-bugfix
+ [Need explicit Shape_t initialization since CUDA 11.1](https://github.com/sony/nnabla-ext-cuda/pull/352)

release-note-build
+ [Revert "replace centos with ubuntu"](https://github.com/sony/nnabla-ext-cuda/pull/351)
+ [Add pytest-xdist options](https://github.com/sony/nnabla-ext-cuda/pull/347)
+ [Apply new pyyaml loader()](https://github.com/sony/nnabla-ext-cuda/pull/346)

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda102 For CUDA version 10.2 users


You can also install the cuda extension with specific versions from one of the following. [See also FAQ](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)

release-note-break-compat
+ [Disable TF32 by default](https://github.com/sony/nnabla-ext-cuda/pull/340)

release-note-bugfix
+ [Use generic transpose kernel if the grid y-dimension exceeds the CUDA limit.](https://github.com/sony/nnabla-ext-cuda/pull/341)
+ [ Bugfix the higher-order gradient of gather_nd](https://github.com/sony/nnabla-ext-cuda/pull/339)

release-note-build
+ [replace centos with ubuntu and use PEP600 tag](https://github.com/sony/nnabla-ext-cuda/pull/342)

release-note-core
+ [Improve ioctl time on A100 by cudaDeviceGetAttribute](https://github.com/sony/nnabla-ext-cuda/pull/345)

release-note-op-layer
+ [Fast reduction](https://github.com/sony/nnabla-ext-cuda/pull/344)
+ [Optimize normalization functions](https://github.com/sony/nnabla-ext-cuda/pull/343)

Install the latest nnabla by:

pip install nnabla
pip install nnabla-ext-cuda102 For CUDA version 10.2 users


You can also install the cuda extension with specific versions from one of the following. [See also FAQ](https://nnabla.readthedocs.io/en/latest/python/install_on_linux.html#faq)