[AIMIGRAPHX-1005] [AIMIGRAPHX-1227] [AIMIGRAPHX-1228]

[bdevorem]

Motivation

Technical Details

Changelog Category

Add a CHANGELOG.md entry for any option other than Not Applicable

• • 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.

Follow the LLVM AI Tool Use Policy for contributions using AI.

[Copilot]

Pull request overview

This PR strengthens the fuse_attention flash-decoding path by making flash-decoding submodule rebuilding more robust in the presence of additional submodule inputs (e.g., masks), constants/materialized literals, and non-trivial instruction ordering. It also expands the unit tests to cover these additional graph shapes and edge cases, and makes existing tests less sensitive to host environment configuration.

Changes:

• Add test coverage for flash decoding with embedded @literal, extra @param mask inputs, unary ops affecting broadcast shape inference, outline nodes, and rebuild ordering constraints.
• Update flash-decoding submodule rebuild to handle @literal/@outline, extra score-shaped params, and dependency-aware rebuild ordering (incl. broadcast shape inference).
• Ensure certain tests run fuse_attention without inheriting flash-decoding settings from environment variables.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
test/fuse_attention.cpp	Adds new flash-decoding tests and introduces an “isolated” pass runner to avoid environment-dependent behavior.
src/fuse_attention.cpp	Improves flash-decoding submodule reconstruction (params handling, literals/outlines, topo-ish rebuild, broadcast shape handling).

[codecov]

Codecov Report

❌ Patch coverage is 93.37017% with 12 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/fuse_attention.cpp	93.37%	12 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #5114      +/-   ##
===========================================
- Coverage    93.26%   93.22%   -0.04%     
===========================================
  Files          623      623              
  Lines        33097    33191      +94     
===========================================
+ Hits         30866    30942      +76     
- Misses        2231     2249      +18     

Files with missing lines	Coverage Δ
src/fuse_attention.cpp	94.95% <93.37%> (-2.55%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[gh-app-migraphx-bot-pr-write]

Test	Batch	New Rate (38fc3e)	Old Rate (3a503c)*	Diff	Status
torchvision-resnet50	64	3,325.83	3,264.92	1.87%	✅
torchvision-resnet50_fp16	64	7,879.49	7,548.67	4.38%	✅
torchvision-densenet121	32	2,491.18	2,483.99	0.29%	✅
torchvision-densenet121_fp16	32	5,042.87	5,004.24	0.77%	✅
torchvision-inceptionv3	32	2,077.09	2,058.51	0.90%	✅
torchvision-inceptionv3_fp16	32	4,491.64	4,416.99	1.69%	✅
cadene-inceptionv4	16	819.91	820.61	-0.08%	✅
cadene-resnext64x4	16	780.48	782.78	-0.29%	✅
slim-mobilenet	64	8,388.56	8,386.36	0.03%	✅
slim-nasnetalarge	64	229.03	228.86	0.08%	✅
slim-resnet50v2	64	3,223.87	3,180.91	1.35%	✅
bert-mrpc-onnx	8	1,172.60	1,168.84	0.32%	✅
bert-mrpc-tf	1	493.56	498.63	-1.02%	✅
pytorch-examples-wlang-gru	1	474.38	473.35	0.22%	✅
pytorch-examples-wlang-lstm	1	384.37	384.83	-0.12%	✅
torchvision-resnet50_1	1	1,052.69	1,046.63	0.58%	✅
cadene-dpn92_1	1	451.31	437.32	3.20%	✅
cadene-resnext101_1	1	364.75	365.89	-0.31%	✅
onnx-taau-downsample	1	843.63	844.09	-0.05%	✅
dlrm-criteoterabyte	1	32.38	32.42	-0.11%	✅
dlrm-criteoterabyte_fp16	1	51.89	51.80	0.17%	✅
agentmodel	1	9,729.42	9,209.12	5.65%	🔆
unet_fp16	2	39.61	58.80	-32.63%	🔴
resnet50v1_fp16	1	1,400.75	1,366.11	2.54%	✅
resnet50v1_int8	1	1,183.84	1,883.96	-37.16%	🔴
bert_base_cased_fp16	64	914.58	1,098.16	-16.72%	🔴
bert_large_uncased_fp16	32	345.49	345.59	-0.03%	✅
bert_large_fp16	1	207.41	206.59	0.40%	✅
distilgpt2_fp16	16	2,090.77	2,092.89	-0.10%	✅
yolov5s	1	284.16	558.33	-49.11%	🔴
tinyllama	1	45.85	45.83	0.04%	✅
vicuna-fastchat	1	44.12	44.20	-0.17%	✅
whisper-tiny-encoder	1	412.63	411.87	0.18%	✅
whisper-tiny-decoder	1	189.83	408.48	-53.53%	🔴
llama2_7b	1	8.66	20.84	-58.45%	🔴
qwen1.5-7b	1	23.49	23.58	-0.37%	✅
phi3-3.8b	1	23.46	26.72	-12.17%	🔴
llama3-8b	1	13.82	21.80	-36.64%	🔴
whisper-large-encoder	1	10.18	10.18	0.02%	✅
whisper-large-decoder	1	103.05	105.30	-2.13%	✅
mistral-7b	1	5.89	23.78	-75.24%	🔴
FLUX.1-schnell	1	104.69	755.22	-86.14%	🔴

Regressions detected 🔴

* No develop baseline was found for this PR's branch point; compared against the latest available develop run instead.

[gh-app-migraphx-bot-pr-write]

Test	Status	Result
bert-mrpc-onnx	✅	PASSED: MIGraphX meets tolerance
bert-mrpc-tf	✅	PASSED: MIGraphX meets tolerance
pytorch-examples-wlang-gru	✅	PASSED: MIGraphX meets tolerance
pytorch-examples-wlang-lstm	✅	PASSED: MIGraphX meets tolerance
dlrm-criteoterabyte	✅	PASSED: MIGraphX meets tolerance
agentmodel	✅	PASSED: MIGraphX meets tolerance
unet	✅	PASSED: MIGraphX meets tolerance
resnet50v1	✅	PASSED: MIGraphX meets tolerance
bert_base_cased_fp16	✅	PASSED: MIGraphX meets tolerance
bert_large_uncased_fp16	🔴	FAILED: MIGraphX is not within tolerance - check verbose output
bert_large	✅	PASSED: MIGraphX meets tolerance
yolov5s	✅	PASSED: MIGraphX meets tolerance
tinyllama	✅	PASSED: MIGraphX meets tolerance
vicuna-fastchat	✅	PASSED: MIGraphX meets tolerance
whisper-tiny-encoder	✅	PASSED: MIGraphX meets tolerance
whisper-tiny-decoder	✅	PASSED: MIGraphX meets tolerance
distilgpt2_fp16	🔴	FAILED: MIGraphX is not within tolerance - check verbose output
llama2_7b	✅	PASSED: MIGraphX meets tolerance
qwen1.5-7b	✅	PASSED: MIGraphX meets tolerance
phi3-3.8b	✅	PASSED: MIGraphX meets tolerance
llama3-8b	✅	PASSED: MIGraphX meets tolerance
whisper-large-encoder	✅	PASSED: MIGraphX meets tolerance
whisper-large-decoder	✅	PASSED: MIGraphX meets tolerance
mistral-7b	✅	PASSED: MIGraphX meets tolerance
FLUX.1-schnell	✅	PASSED: MIGraphX meets tolerance