Commits on Apr 15, 2022

1. Autotune the workspace_size_limit in conv. (PaddlePaddle#40338) …

   * Using the maximum workspace_size of all alogirhms to limit the workspace size in exhaustive search mode.
   
   * Use the system cudaMalloc and cudaFree to allocate workspace during searching.
   
   * Enable switch of two kind of workspace setting methods.
   
   Co-authored-by: Liu Yiqun <liuyiqun01@baidu.com>

   

   JamesLim-sy and Xreki committed Apr 15, 2022
   Configuration menu

   • View commit details
   • Copy full SHA for 32d0fc3
   • Browse repository at this point

   Copy the full SHA

   32d0fc3 View commit details

   Browse the repository at this point in the history

2. Optimize the finding of max workspace size. (PaddlePaddle#41741)

   

   Xreki committed Apr 15, 2022
   Configuration menu

   • View commit details
   • Copy full SHA for e2faadc
   • Browse repository at this point

   Copy the full SHA

   e2faadc View commit details

   Browse the repository at this point in the history

3. Change cuDNN Conv kernel for auto tune feature (PaddlePaddle#41313) …

   * change cudnn helper for auto-tune
   
   * Add FLAGS_use_autotune to set the global status of autotune and change the order of choosing algorithm.
   
   * Fix the bug in calculating and printing current step cache hit rate.
   
   * Improve the autotune cache and fix unittest.
   
   * Change the key from AlgorithmType to int64_t.
   
   * Fix unittest for cpu-only env.
   
   * change ChooseAlgoByWorkspace for heuristic mode
   
   Co-authored-by: Liu Yiqun <liuyiqun01@baidu.com>

   

   JamesLim-sy and Xreki committed Apr 15, 2022
   Configuration menu

   • View commit details
   • Copy full SHA for 8bf90c2
   • Browse repository at this point

   Copy the full SHA

   8bf90c2 View commit details

   Browse the repository at this point in the history