Fix invalid concat in find_conv_concat_split_fuse for Topaz flow models

[urpetkov-amd]

Motivation

Topaz flow U-Net ONNX models (e.g. thf-v4-fnet-fp16-256x352, thd-v3-fnet-fp16-256x352) fail GPU compile in MIGraphX during the simplify_algebra pass. The failure blocks compilation via migraphx-driver and any execution provider that uses the MIGraphX GPU pipeline (including the AMDGPU EP). The only workaround today is disabling the entire pass with MIGRAPHX_DISABLE_PASSES=simplify_algebra.

This PR fixes the compile crash so these models compile without disabling simplify_algebra.

Technical Details

Root cause: In find_conv_concat_split_fuse (src/simplify_algebra.cpp), horizontal conv fusion through a channel concat can build a suffix concat(axis=1) over the remaining concat inputs without verifying that spatial dimensions match. On Topaz-style U-Net graphs—decoder feature + encoder skip concat, conv on the concat, parallel stride-2 conv on the skip branch—this can produce an invalid concat after rewrite_convolution has rewritten transposed convs, triggering:

Error simplify_algebra: .../concat.hpp:97: normalize_compute_shape:
CONCAT: all input dimensions should match in axis 2

Changes:

• Hoist is_valid_concat() to file scope (same logic already used by find_concat_op) so both transforms share one helper.
• In find_conv_concat_split_fuse, compute suffix remaining inputs before inserting any fusion IR.
• Skip fusion when remaining.size() > 1 and is_valid_concat(remaining, 1) is false.

Valid fusion paths (e.g. existing conv_horizontal_fuse in test/simplify_algebra_test.cpp) are unchanged. Invalid paths skip fusion instead of aborting compile.

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.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #5156   +/-   ##
========================================
  Coverage    93.13%   93.13%           
========================================
  Files          625      625           
  Lines        33252    33254    +2     
========================================
+ Hits         30967    30969    +2     
  Misses        2285     2285           

Files with missing lines	Coverage Δ
src/simplify_algebra.cpp	97.48% <100.00%> (+<0.01%)	⬆️

🚀 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 (16fba2)	Old Rate (3a503c)*	Diff	Status
torchvision-resnet50	64	3,322.17	3,264.92	1.75%	✅
torchvision-resnet50_fp16	64	7,872.47	7,548.67	4.29%	✅
torchvision-densenet121	32	2,489.48	2,483.99	0.22%	✅
torchvision-densenet121_fp16	32	5,023.22	5,004.24	0.38%	✅
torchvision-inceptionv3	32	2,072.70	2,058.51	0.69%	✅
torchvision-inceptionv3_fp16	32	4,463.06	4,416.99	1.04%	✅
cadene-inceptionv4	16	817.23	820.61	-0.41%	✅
cadene-resnext64x4	16	781.35	782.78	-0.18%	✅
slim-mobilenet	64	8,380.78	8,386.36	-0.07%	✅
slim-nasnetalarge	64	228.18	228.86	-0.30%	✅
slim-resnet50v2	64	3,238.03	3,180.91	1.80%	✅
bert-mrpc-onnx	8	1,168.94	1,168.84	0.01%	✅
bert-mrpc-tf	1	496.95	498.63	-0.34%	✅
pytorch-examples-wlang-gru	1	497.86	473.35	5.18%	🔆
pytorch-examples-wlang-lstm	1	409.57	384.83	6.43%	🔆
torchvision-resnet50_1	1	1,052.59	1,046.63	0.57%	✅
cadene-dpn92_1	1	441.41	437.32	0.94%	✅
cadene-resnext101_1	1	365.55	365.89	-0.09%	✅
onnx-taau-downsample	1	847.39	844.09	0.39%	✅
dlrm-criteoterabyte	1	32.24	32.42	-0.56%	✅
dlrm-criteoterabyte_fp16	1	51.56	51.80	-0.46%	✅
agentmodel	1	13,953.05	9,209.12	51.51%	🔆
unet_fp16	2	58.33	58.80	-0.80%	✅
resnet50v1_fp16	1	1,431.14	1,366.11	4.76%	✅
resnet50v1_int8	1	1,811.04	1,883.96	-3.87%	✅
bert_base_cased_fp16	64	1,097.39	1,098.16	-0.07%	✅
bert_large_uncased_fp16	32	345.86	345.59	0.08%	✅
bert_large_fp16	1	206.17	206.59	-0.20%	✅
distilgpt2_fp16	16	2,090.66	2,092.89	-0.11%	✅
yolov5s	1	566.27	558.33	1.42%	✅
tinyllama	1	45.82	45.83	-0.01%	✅
vicuna-fastchat	1	44.33	44.20	0.30%	✅
whisper-tiny-encoder	1	412.74	411.87	0.21%	✅
whisper-tiny-decoder	1	410.73	408.48	0.55%	✅
llama2_7b	1	20.88	20.84	0.19%	✅
qwen1.5-7b	1	23.65	23.58	0.31%	✅
phi3-3.8b	1	26.74	26.72	0.10%	✅
llama3-8b	1	21.72	21.80	-0.37%	✅
whisper-large-encoder	1	10.17	10.18	-0.08%	✅
whisper-large-decoder	1	107.29	105.30	1.89%	✅
mistral-7b	1	23.78	23.78	-0.00%	✅
FLUX.1-schnell	1	780.87	755.22	3.40%	✅

Check flagged results 🔆

* 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	❌	ERROR - check error output traceback Traceback (most recent call last): File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 377, in main() File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 313, in main import tensorflow as tf File "/usr/local/lib/python3.12/dist-packages/tensorflow/init.py", line 40, in from tensorflow.python import pywrap_tensorflow as _pywrap_tensorflow # pylint: disable=unused-import ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/pywrap_tensorflow.py", line 37, in self_check.preload_check() File "/usr/local/lib/python3.12/dist-packages/tensorflow/python/platform/self_check.py", line 63, in preload_check from tensorflow.python.platform import _pywrap_cpu_feature_guard ImportError: libnuma.so.1: cannot open shared object file: No such file or directory
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

[TedThemistokleous]

Looks good. Add a unit test that triggers the invalid concat so we can capture the failure that this change fixes.