We have been using v2.0 of Tiled-MM successfully for some time but when I recently tried updating to use v2.3 we started experiencing incorrect results from our application (https://github.com/RMGDFT/rmgdft). I determined that the problematic calls were occurring with transa and transb both 'N' while calls with transa='T' and transb='N' were producing the correct results. The matrices in question were tall and skinny (m,n,k) (46656,152,480). I tried running the testing app using these values but no error was reported so I'm not sure how to proceed. I looked over the commit history and 85331eb seens to be the only thing that could have changed the behavior.