• ltqin's avatar
    Example for conv2d backward weight fp16 (#106) · 7a9b93f4
    ltqin authored
    
    
    * add wrw reference
    
    * start device
    
    * raw not split version
    
    * run simple example
    
    * start to use atomic add
    
    * simple transform result correct
    
    * first version that can run
    
    * fix atomic and set operator choice
    
    * add check split-k
    
    * format
    
    * change input parameter
    
    * add pad for t total
    
    * rename example index
    Co-authored-by: default avatarltqin <letaoqin@amd.com>
    7a9b93f4
main.cpp 13.1 KB