This repository has been archived by the owner on Jan 24, 2024. It is now read-only.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This pr comb the code of op lowering.
Organize the generalized Lower process into four stages:
1.Call Compute and perform narrow Lower (i.e. translate graph to CINN IR).
2. Do Op level schedule.
3.Do group level schedule.
4.Post process, including processing of LoweredFunc arguments and temporary variables, application of low level pass, etc
Then, all types of group lower processes call these processes, abstracting only a small amount of different logic and using different callback functions for processing, to achieve the maximum uniformity of the generalized Lower process.
Specifically, due to significant differences in the Lower logic of CustomCall, it needs to be handled separately by LowerCustomCall().