You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2021/03/31 04:56:23 UTC

[GitHub] [tvm] junrushao1994 edited a comment on pull request #7765: [M1b] Scaffolding ScheduleState data structure

junrushao1994 edited a comment on pull request #7765:
URL: https://github.com/apache/tvm/pull/7765#issuecomment-810515117


   Although out of the scope of this PR, I am really glad that we have the discussion about block names.
   
   @comaniac brought up the point https://github.com/apache/tvm/pull/7765#discussion_r603657740:
   
   > This function makes me think that we should make root as a preserved block name, and we should not allow duplicated block names in every tree of a PrimFunc.
   
   I kinda agree with Cody about his points, but would love to hear more discussion on the block name. Particularly, we have three points to discuss:
   - A1. Block names need to be unique. The reason is that the canonical way of retrieving a block is to use its name, i.e. `schedule.get_block(name)`. Without a unique name, we are unable to even retrieve a block, which makes scheduling almost impossible. (of course, it is possible to retrieve a block by the buffer it produces or via a statement, but it is not the canonical way)
   - A2. We need reserved names for the root block. I am kinda in favor of this idea too, because we do provide syntactic sugar to auto complete the root block with the name "root". This could help us eliminate possible name conflicts.
   - A3. Users could specify the names of newly created blocks/loops. Yes, it is doable when implementing schedule primitives.
   


-- 
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