How to see actual cuda file generated by tvm?

Questions
1

Hi, I’m new to TVM. I’d like to know how to see actual cuda file generated by tvm?

2

You can get the code from the device module as in the Tensor Expression tutorial.

Best regards

Thomas

3

@t-vi Thanks for reply :slight_smile: what i mean is , as described in relay quick start, by using graph_runtime.create() we can get the module, then we can use module.get_output() get the result. But, I want to know what the excatly schedule optimization tvm has done after compiled the DL model. Is there any approach for graph_runtime() object to get source code ? Best regrads

4

I could be wrong (and I don’t always have access to cuda to check), but my impression was that the library you pass to graph_runtime has a specialization to the precise schedule.