> @Rasterer Any update about this? I'm afraid I can not commit PR due to open source policy change in my working company. Welcome for code contribution if you are interesting in this.
In addition to my original proposal, I think padding performance can be improved further by changing fusion pattern. So far conv padding is inlined into the conv op, thus the condition switch cost in padding will happen more than once for each input element(due to repeated input access in conv). This can be avoided by fusing padding to the preceding op, and padding can be implemented by some offset output writing. -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/apache/incubator-tvm/issues/2682#issuecomment-562413289
