[Frontend] Achieve norm optimization by scf dialect and intergrate into E2E pass #596
Conversation
This file contains hidden or 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
GuoningHuang
commented
Oct 27, 2025
Comment on lines
+2082
to
+2115
| # #Could choose to use scf.parallel to implement the loop_b or loop_i | ||
| # loop_b_parallel = scf.ParallelOp([], [c0.result], [dim1_index.result], [c1.result],[]) | ||
| # blk = loop_b_parallel.region.blocks.append() | ||
| # with ir.InsertionPoint(blk): | ||
| # b = blk.add_argument(ir.IndexType.get(), ir.Location.unknown()) | ||
| loop_b = scf.ForOp(c0.result, dim1_index.result, c1.result, []) | ||
| with ir.InsertionPoint(loop_b.body): | ||
| b = loop_b.induction_variable |
GuoningHuang
force-pushed
the
RMSnorm-scf
branch
from
0b266e0 to
d45fd33
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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 introduces an optimized implementation of Norm
It provides both frontend operator fusion and the corresponding lowering pass to support end-to-end execution.
Implemented frontend LayerNorm fusion, combining operations (reduce_sum, rsqrt, mul, add, etc.) into a single fused operator.
Added lowering pass from fused Norm to optimized vectorized IR.
Verified correctness on test_norm.py.
Can be directly integrated into the existing Deepseek R1 inference pipeline.
before opt:
after opt: