[TORCH] Add canonicalization of `aten.transpose.int` op

[Shukla-Gaurav]

This commit adds a canonicalization of aten.tranpose.int op when its operand is aten.permute op. aten.permute + aten.transpose.int -> aten.permute

Signed-Off-by: Gaurav Shukla [email protected]

[cathyzhyi]

Also I think this is only an optimization when the transpose is the only user of the original permute. Otherwise, we would be replacing one transpose with a listcontruct and an additional permute. Is this needed for performance optimization?

[Shukla-Gaurav]

Also I think this is only an optimization when the transpose is the only user of the original permute. Otherwise, we would be replacing one transpose with a listcontruct and an additional permute. Is this needed for performance optimization?

Yes, this is needed for the performance optimization of BERT. I will check for the users of permute op. That's a good point, Thanks!