[GitHub] [incubator-tvm] cloudhan commented on issue #4118: [RFC] Dynamic Shape Support - Graph Dispatching

[GitBox <gi...@apache.org>]

cloudhan commented on issue #4118:
URL: https://github.com/apache/incubator-tvm/issues/4118#issuecomment-638269770


   @kevinthesun Any timeframe?
   
   Off topic, I want to mention TensorRT supports dynamic shape from 7.0. To provide better performance, it supports multiple optimization profiles for different shape range. Say your input is 1d ranged from 1 to 1024. You can create profiles whatever shape you specified, overlap or non-overlap, e.g.
   ```
   (1, 8, 16)    -> this profile is valid for 1 to 16, specifically optimized of 8 
   (16, 24, 32)
   (33, 48, 63)
   ...
   ```
   Users can manually select profile to use. In terms of https://sampl.cs.washington.edu/tvmconf/slides/2019/Jared-Roesch-Haichen-Shen-RelayVM.pdf I have some questions, `Representing Dynamism` is solved by shape function, but how do I manually represent a constrained dynamism? That is our input is only valid from 1 to 1024 or 16 to 32.
   This dynamic shape may be n-d varying, how do you want to handle overlapping (in this case, unavoidable)?


----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org