moderato opened a new pull request #9058: URL: https://github.com/apache/tvm/pull/9058
Currently, the order of the AutoTVM function arguments might cause confusion in some cases. For example, for a 3x3 conv, the order of `arg_bufs` after calling `s, arg_bufs = task.instantiate(best_config)` is output, filter, input, while for a 1x1 conv, the order is output, input, filter, since it doesn't need padding and thus both input and filter are directly the input_tensors of conv, in which input comes first. This will significantly affect the use of the generated CUDA/asm code externally. This fix sort the arguments in topological order. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
