Add support for HipGraph

[pfultz2]

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.

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

Test	Batch	New Rate (0b1f870)	Old Rate (875a33)	Diff	Status
torchvision-resnet50	64	nan	3,163.45	nan	❌
torchvision-resnet50_fp16	64	nan	6,673.18	nan	❌
torchvision-densenet121	32	nan	2,709.43	nan	❌
torchvision-densenet121_fp16	32	nan	4,541.52	nan	❌
torchvision-inceptionv3	32	nan	1,800.68	nan	❌
torchvision-inceptionv3_fp16	32	nan	2,821.79	nan	❌
cadene-inceptionv4	16	nan	826.36	nan	❌
cadene-resnext64x4	16	nan	784.89	nan	❌
slim-mobilenet	64	nan	8,434.41	nan	❌
slim-nasnetalarge	64	nan	229.27	nan	❌
slim-resnet50v2	64	nan	3,331.37	nan	❌
bert-mrpc-onnx	8	nan	1,171.20	nan	❌
bert-mrpc-tf	1	nan	484.68	nan	❌
pytorch-examples-wlang-gru	1	nan	322.20	nan	❌
pytorch-examples-wlang-lstm	1	nan	449.24	nan	❌
torchvision-resnet50_1	1	nan	752.81	nan	❌
cadene-dpn92_1	1	nan	443.91	nan	❌
cadene-resnext101_1	1	nan	359.68	nan	❌
onnx-taau-downsample	1	nan	401.96	nan	❌
dlrm-criteoterabyte	1	nan	32.70	nan	❌
dlrm-criteoterabyte_fp16	1	nan	52.64	nan	❌
agentmodel	1	nan	8,419.42	nan	❌
unet_fp16	2	nan	57.18	nan	❌
resnet50v1_fp16	1	nan	932.86	nan	❌
resnet50v1_int8	1	nan	928.41	nan	❌
bert_base_cased_fp16	64	nan	1,103.03	nan	❌
bert_large_uncased_fp16	32	nan	347.67	nan	❌
bert_large_fp16	1	nan	205.48	nan	❌
distilgpt2_fp16	16	nan	2,092.07	nan	❌
yolov5s	1	nan	560.40	nan	❌
tinyllama	1	nan	46.03	nan	❌
vicuna-fastchat	1	nan	43.97	nan	❌
whisper-tiny-encoder	1	nan	420.65	nan	❌
whisper-tiny-decoder	1	nan	413.68	nan	❌
llama2_7b	1	nan	20.47	nan	❌
qwen1.5-7b	1	nan	23.68	nan	❌
phi3-3.8b	1	nan	27.00	nan	❌
llama3-8b	1	nan	21.85	nan	❌
whisper-large-encoder	1	nan	10.32	nan	❌
whisper-large-decoder	1	nan	104.23	nan	❌
mistral-7b	1	nan	23.85	nan	❌
FLUX.1-schnell	1	nan	749.64	nan	❌

Regressions detected 🔴

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

Test	Status	Result
bert-mrpc-onnx	❌	ERROR - check error output
bert-mrpc-tf	❌	ERROR - check error output
pytorch-examples-wlang-gru	✅	PASSED: MIGraphX meets tolerance
pytorch-examples-wlang-lstm	✅	PASSED: MIGraphX meets tolerance
dlrm-criteoterabyte	✅	PASSED: MIGraphX meets tolerance
agentmodel	❌	ERROR - check error output traceback 2026-06-17 11:54:14.119728 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): batch. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails. 2026-06-17 11:54:15.092273 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for hip::copy_to_gpu 2026-06-17 11:54:15.092308 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for hip::copy_to_gpu 2026-06-17 11:54:15.092326 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for hip::copy_to_gpu 2026-06-17 11:54:15.092344 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for hip::copy_to_gpu 2026-06-17 11:54:15.092361 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for hip::copy_to_gpu 2026-06-17 11:54:15.092381 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=multinomial[dtype=9],additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:54:15.092400 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=multinomial[dtype=9],additional_args=1,ignore_modules=0,output_shape=nullopt] 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 282, in main pred_migx = np.array(model.run(params)[-1]) RuntimeError: /data/src/program.cpp:523: operator(): Incorrect shape {float_type, {1, 21}, {483, 1}} for parameter: x4 should be: float_type, {1, 21}, {21, 1}
unet	❌	ERROR - check error output traceback 2026-06-17 11:55:49.373444 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=convolution[padding={0, 0, 0, 0},stride={1, 1},dilation={1, 1},group=1,padding_mode=0]],additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373485 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=convolution[padding={0, 0, 0, 0},stride={1, 1},dilation={1, 1},group=1,padding_mode=0]],additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373509 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=pointwise,additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373558 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=pointwise,additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373586 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=reshape[dims={1, 640, 32, 32}]],additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373614 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=reshape[dims={1, 640, 32, 32}]],additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373642 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=pointwise,additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373670 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=reshape[dims={1, 320, 64, 64}]],additional_args=1,ignore_modules=0,output_shape=nullopt] 2026-06-17 11:55:49.373698 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=reshape[dims={1, 320, 64, 64}]],additional_args=1,ignore_modules=0,output_shape=nullopt]
resnet50v1	❌	ERROR - check error output traceback 2026-06-17 11:59:22.111372 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): unk__616. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails.
bert_base_cased_fp16	❌	ERROR - check error output
bert_large_uncased_fp16	❌	ERROR - check error output
bert_large	❌	ERROR - check error output traceback 2026-06-17 12:01:02.104280 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): batch_size. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails.
yolov5s	✅	PASSED: MIGraphX meets tolerance
tinyllama	✅	PASSED: MIGraphX meets tolerance
vicuna-fastchat	❌	ERROR - check error output
whisper-tiny-encoder	❌	ERROR - check error output
whisper-tiny-decoder	❌	ERROR - check error output
distilgpt2_fp16	❌	ERROR - check error output
llama2_7b	❌	ERROR - check error output traceback 2026-06-17 12:05:41.679126 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): batch_size, sequence_length. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails.
qwen1.5-7b	❌	ERROR - check error output
phi3-3.8b	❌	ERROR - check error output traceback 2026-06-17 12:12:01.660180 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=dot],additional_args=1,ignore_modules=0,output_shape=nullopt]
llama3-8b	❌	ERROR - check error output traceback 2026-06-17 12:12:34.229875 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): batch_size, sequence_length. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails. 2026-06-17 12:14:43.234631 [WARN] [/data/src/adjust_allocation.cpp:67] output buffer doesn't match output for gpu::precompile_op[op=gpu::mlir_op[op=dot],additional_args=1,ignore_modules=0,output_shape=nullopt]
whisper-large-encoder	❌	ERROR - check error output traceback 2026-06-17 12:15:31.337237 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): batch_size, encoder_sequence_length, feature_size. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails. 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 224, in main model = migraphx.parse_onnx(model_name, default_dim_value=batch) RuntimeError: /data/src/include/migraphx/op/convolution.hpp:113: normalize_compute_shape: CONVOLUTION: mismatched channel numbers: input channels (1) != weights channels (80) * group (1)
whisper-large-decoder	❌	ERROR - check error output traceback 2026-06-17 12:15:34.933388 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): batch_size, decoder_sequence_length, encoder_sequence_length / 2. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails.
mistral-7b	❌	ERROR - check error output traceback 2026-06-17 12:17:03.195355 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): batch_size, sequence_length. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails.
FLUX.1-schnell	❌	ERROR - check error output traceback 2026-06-17 12:19:38.561791 [WARN] [/data/src/onnx/onnx_parser.cpp:282] Model has unbound symbolic dimension(s): B. These default to 1 and may cause unexpected behavior. Try setting --dim-param @<name> <value> or --input-dim @<input> <dims> if program compilation fails.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #4956      +/-   ##
===========================================
- Coverage    92.72%   92.72%   -0.00%     
===========================================
  Files          590      592       +2     
  Lines        31233    31256      +23     
===========================================
+ Hits         28960    28981      +21     
- Misses        2273     2275       +2     

see 4 files with indirect coverage changes

🚀 New features to boost your workflow:

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