This repository was archived by the owner on Nov 17, 2023. It is now read-only.

Port convolutions to cuDNN v8 API#20635

Merged

DickJC123 merged 17 commits intoapache:masterfrom mk-61:pr-cudnn-v8-api

Nov 15, 2021

+1,925-1,909

Commits on Nov 11, 2021

Add failsafe flag to StorageManager Alloc()

DickJC123
authored and
Vladimir Cherepanov
committed
Clear sticky cudaErrorMemoryAllocation errors

DickJC123
authored and
Vladimir Cherepanov
committed
Make Conv and Deconv cuDNN implementation use v8 API
Vladimir Cherepanov
committed
Downstandard some C++17 code to C++14 to accommodate CUDA 10
Vladimir Cherepanov
committed
Relax cuDNN version to 8.0.2
Vladimir Cherepanov
committed
Use newer cuDNN version in CI
Vladimir Cherepanov
committed
Dont's verify cmake.org certificate
Vladimir Cherepanov
committed
Disable mobilenet inference test
Vladimir Cherepanov
committed
Re-format with the new clang-format config
Vladimir Cherepanov
committed
Fix cpplint after clang-format
Vladimir Cherepanov
committed
Disable fprop eng:5 to fix test failure on M60
Vladimir Cherepanov
committed
Conv autotune workspaces released via DirectFree()

DickJC123
authored and
Vladimir Cherepanov
committed
Address review comments
Vladimir Cherepanov
committed
Pamper clang-format
Vladimir Cherepanov
committed
Fix default heuristics mode logic and document env var
Vladimir Cherepanov
committed
Add doc for MXNET_CUDNN_ALGO_VERBOSE_LEVEL

DickJC123
authored and
Vladimir Cherepanov
committed
More review comments
Vladimir Cherepanov
committed