Move demotion passes to GlobalOptimization. #14815
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The global optimizations really depend on the flow level demotion passes. Moves them to the right place. Also restores the ordering of the strip assertions pass so that it can guide optimizations. Renames old `-iree-flow-(demote|promote)-*` flags to `-iree-opt-(demote|promote)-*` flags now that they are just part of the rest of the global optimizations. Internally, renames "HighLevelOptimizations" to "GlobalOptimizations" for coherence. Fixes #14835.
stellaraccident
force-pushed
the
test_llama2_consteval
branch
from
f07ba0e
to
a0ee304
Compare
stellaraccident
changed the title
stellaraccident
requested review from
hanhanW,
MaheshRavishankar and
benvanik
as code owners
benvanik
approved these changes
Aug 26, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
The global optimizations really depend on the flow level demotion passes. Moves them to the right place. Also restores the ordering of the strip assertions pass so that it can guide optimizations.
Renames old
-iree-flow-(demote|promote)-*flags to-iree-opt-(demote|promote)-*flags now that they are just part of the rest of the global optimizations.Internally, renames "HighLevelOptimizations" to "GlobalOptimizations" for coherence.
This lets us enable consteval on llam2 7b qi4/f16 models and drops latency from 14.3ms -> 11.9ms. Mostly it is small scalar-level evaluations but it is also eliminating a 250MiB f16 transpose in both first/second that I expect is a main culprit.
Fixes #14835.