Skip to content

Conversation

@causten
Copy link
Collaborator

@causten causten commented Oct 6, 2025

Motivation

7.1 branch cut missed important fixes

  • CI Improvements
  • Fix TopK operator
  • Added Python 3.13 support
  • Update dependencies

Technical Details

894f16022 Update onnxruntime main be655f69f6c8623eebcac094626d0c8545951e6d (#4348)
94bea56d4 Lower lrn to pooling (#4294)
fd0ad324e Updated SD3 example for change in optimum-onnx[onnxruntime] (#4344)
b3b6dfdc2 Fuse GEMM+GEMM with rocMLIR (#4261)
b92bf17d3 Bump onnx to 1.18.0 (#4323)
53a7447fe Brevitas MXFP4 quantization parse (#4301)
825ebca38 Fix MXFP4 bugs (#4324)
14678120b Add video and render groups in docker for CI (#4340)
8b591436e Use --input-dim for specifying dynamic shapes at driver runtime (#4342)
3c4378fa7 TopK exception bugfix (#4329)
12dc8308d Bump SQlite3 to 3.50.4 (#4322)
3a688d339 Update build and test script and dockerfile to add in onnxrt pai laun… (#4336)

Changelog Category

    • Added: New functionality.
    • Changed: Changes to existing functionality.
    • Removed: Functionality or support that has been removed. (Compared to a previous release)
    • Optimized: Component performance that has been optimized or improved.
    • Resolved Issues: Known issues from a previous version that have been resolved.
    • Not Applicable: This PR is not to be included in the changelog.

TedThemistokleous and others added 12 commits September 30, 2025 19:25
#4336)

Update build and test script and dockerfile to add in onnxrt pai launcher scripts
Push pai launcher scripts to Onnxrutime to let us reuse scripts from prior run
Related to change from - #4321
* propagate_constant ignores `unpack_fp4` instructions now
* `match_find_mx_quantizable_ops` from `simplify_qdq` updated to not require non-constant scales. The scales can be literals.
* transpose, reshape, and broadcast instructions propagated on scale instructions when going to `quant_dot` and `quant_conv`
* `raw_data` `operator<<` updated to use `fallback_visit` to also handle non-computable types
rocmlir supports GEMM+GEMM fusion, so MIGraphX needs to support the fusion on our side.
Solves #4230
This PR implements a lowering transformation that converts LRN operations into a series of pooling and arithmetic operations.
@causten causten requested a review from a team as a code owner October 6, 2025 19:21
@causten causten changed the title Merge 71 [7.1] bulk merge Oct 6, 2025
| Default: No tuning is done for composable kernels.

* - | ``MIGRAPHX_REWRITE_LRN``
| Turns on LRN-to-pooling lowering in the rewrite_pooling pass.
Copy link
Contributor

Choose a reason for hiding this comment

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

Part of me thinks rewrite_pooling should be in code text: rewrite_pooling

I would check how the passes are written in the rest of the text and keep the same formatting.

Copy link
Contributor

@spolifroni-amd spolifroni-amd left a comment

Choose a reason for hiding this comment

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

This looks OK to me. I had only one small comment and it's up to you whether you want to change this.

@codecov
Copy link

codecov bot commented Oct 8, 2025

Codecov Report

❌ Patch coverage is 95.78313% with 7 lines in your changes missing coverage. Please review.

Files with missing lines Patch % Lines
src/onnx/parse_dynamicscale.cpp 90.91% 5 Missing ⚠️
src/onnx/parse_quantizelinear.cpp 95.00% 1 Missing ⚠️
src/simplify_qdq.cpp 95.65% 1 Missing ⚠️
Additional details and impacted files
@@                   Coverage Diff                   @@
##             release/rocm-rel-7.1    #4358   +/-   ##
=======================================================
  Coverage                        ?   92.26%           
=======================================================
  Files                           ?      560           
  Lines                           ?    26358           
  Branches                        ?        0           
=======================================================
  Hits                            ?    24319           
  Misses                          ?     2039           
  Partials                        ?        0           
Files with missing lines Coverage Δ
src/include/migraphx/module.hpp 100.00% <ø> (ø)
src/include/migraphx/raw_data.hpp 96.67% <100.00%> (ø)
src/module.cpp 87.12% <100.00%> (ø)
src/onnx/onnx_parser.cpp 88.92% <100.00%> (ø)
src/propagate_constant.cpp 83.93% <100.00%> (ø)
src/rewrite_pooling.cpp 99.16% <100.00%> (ø)
src/onnx/parse_quantizelinear.cpp 94.23% <95.00%> (ø)
src/simplify_qdq.cpp 96.14% <95.65%> (ø)
src/onnx/parse_dynamicscale.cpp 90.91% <90.91%> (ø)
🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@jharryma jharryma merged commit 7dfa665 into release/rocm-rel-7.1 Oct 10, 2025
14 checks passed
@causten causten deleted the merge_71 branch October 10, 2025 17:51
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.