[GitHub] [tvm] wrongtest-intellif commented on issue #12298: [Bug][TIR] Split Error in Schedule Verification

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

wrongtest-intellif commented on issue #12298:
URL: https://github.com/apache/tvm/issues/12298#issuecomment-1240095516

   @vinx13 I agree here the invocation to standalone `canonical_simplify` is quite strange. However actually I try to increase the steps but could not fix the issue. The order of `rewrite_simplify` and `canonical_simplify` matters such that if we first `rewrite_simplify` (the first step of Simplify), we will never be able to proof some cases, no matter how many times of iters, and vice versa according to current issue.
   
   Whether could I add two code path here as a middle term solution? 
   - Step 1. Try directly use `Simplify` here, the most cases should be covered.
   - Step 2. If it fails unfortunately, try `canonical_simplify` first.
   
   For a long term solution to be more robust, I notice that there do exist rules to handle patterns like in your mentioned cases. But unfortunately, for example, if we have a cancel pattern `R(a - b) -> simplified`, current system could not ensure for `a + c - b` the pattern could be matched with associative rule, unless the term order is luckily changed (`canonical_simplify` sometimes achieve this). It seems that a O(n^2) summation term matching is required to solve this issue. 
   


-- 
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: commits-unsubscribe@tvm.apache.org

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