Skip to content

Update mlir-air and mlir-aie to latest wheels (LLVM 278dba37)#30

Merged
erwei-xilinx merged 2 commits into
mainfrom
update-mlir-air-aie-versions
Mar 27, 2026
Merged

Update mlir-air and mlir-aie to latest wheels (LLVM 278dba37)#30
erwei-xilinx merged 2 commits into
mainfrom
update-mlir-air-aie-versions

Conversation

@erwei-xilinx

Copy link
Copy Markdown
Collaborator

Summary

  • Bump mlir-air to 5798472 and mlir-aie to 3a1aba2 (both timestamp 2026032704) to pick up the LLVM 278dba37 update
  • Replace removed transform.apply_patterns.vector.lower_multi_reduction with its three successor ops (reorder_multi_reduction_dims, multi_reduction_flattening, multi_reduction_unrolling) across the transform library and 6 example transform scripts
  • Add explicit --air-runtime-loop-tiling-sizes=4 4 to aircc invocation in driver.py to match the new default (changed from [4,4] to [] in mlir-air #1470)

Test plan

  • vec-add: PASS (all sizes)
  • relu: PASS (all sizes)
  • axpy: PASS
  • leaky_relu: PASS
  • sigmoid: PASS
  • silu: PASS
  • swiglu: PASS
  • autotune-matmul: PASS
  • matmul: PASS (all 27 size combinations including previously failing M=4096)
  • test_softmax: PASS
  • average_pool: PASS
  • test_layernorm: PASS

🤖 Generated with Claude Code

Copilot AI review requested due to automatic review settings March 27, 2026 20:26
Bump mlir-air to 5798472 (2026032704) and mlir-aie to 3a1aba2
(2026032704) to pick up LLVM 278dba37 update.

Two compatibility fixes for the new LLVM:

1. Replace `transform.apply_patterns.vector.lower_multi_reduction`
   with the three ops it was split into: `reorder_multi_reduction_dims`,
   `multi_reduction_flattening`, and `multi_reduction_unrolling`.

2. Add explicit `--air-runtime-loop-tiling-sizes=4` to the aircc
   command in driver.py since mlir-air #1470 changed the default
   from [4,4] to [] (no tiling).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Updates the repo’s pinned mlir-air / mlir-aie wheels to pick up an LLVM revision bump, and adjusts the transform scripts + backend driver to remain compatible with upstream API/default changes.

Changes:

  • Bump pinned wheel commit hashes/timestamps for mlir-air and mlir-aie.
  • Replace removed transform.apply_patterns.vector.lower_multi_reduction with its successor patterns across the transform library and example scripts.
  • Add explicit --air-runtime-loop-tiling-sizes arguments to the aircc invocation to match the prior default behavior.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file
File Description
utils/mlir-air-hash.txt Update pinned mlir-air commit/timestamp.
utils/mlir-aie-hash.txt Update pinned mlir-aie commit/timestamp.
examples/test_layernorm/transform_aie2p.mlir Swap deprecated multi-reduction lowering pattern for successor patterns.
examples/test_layernorm/transform_aie2.mlir Swap deprecated multi-reduction lowering pattern for successor patterns.
examples/rms_norm/transform_aie2p.mlir Swap deprecated multi-reduction lowering pattern for successor patterns.
examples/average_pool/transform_aie2p.mlir Swap deprecated multi-reduction lowering pattern for successor patterns.
examples/average_pool/transform_aie2.mlir Swap deprecated multi-reduction lowering pattern for successor patterns.
amd_triton_npu/backend/transform_library/vectorization.mlir Update reduction cleanup sequence to use successor multi-reduction lowering patterns.
amd_triton_npu/backend/driver.py Add explicit runtime loop tiling sizing flags to aircc invocation.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread amd_triton_npu/backend/driver.py Outdated
Comment thread amd_triton_npu/backend/transform_library/vectorization.mlir Outdated
Comment thread amd_triton_npu/backend/transform_library/vectorization.mlir Outdated
- Fix driver.py comment to say "to [4,4]" matching the actual CLI args
- Put pattern + lowering_strategy on single lines in vectorization.mlir
  for consistency with the rest of the codebase
- Update comment to match actual pattern ordering (cast_away first,
  then multi_reduction lowering)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@erwei-xilinx erwei-xilinx merged commit c3c665a into main Mar 27, 2026
8 of 9 checks passed
@erwei-xilinx erwei-xilinx deleted the update-mlir-air-aie-versions branch March 27, 2026 22:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants