SSD static sidestep

[CharlieL7]

Motivation

• Attempt to get static shape rewrite of SSD model.
• I don't think this would be a good approach even if I cleaned it up. It's equivalent to the pad-slice method with a single maximum kernel size but significantly more convoluted because it's trying to avoid dynamic shapes.

Technical Details

• Basically the majority of the post-processing step of the SSD model needed to be rewritten.
• The NMS -> Gather -> TopK needs a Where mask.
• The k calculation for the TopK needed to be moved past the Gathers after TopK and done as multiple dynamic slices at the end of the model.
• The TopK needed to be changed to have the k value set to the maximum literal concatenated before the reduce_min of the k calculation. Otherwise it was a massive number that didn't compile.
• Removed the MIGRAPHX_USE_DYNAMIC_NMS environment variable.

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.

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

Test	Batch	New Rate (3fe3b6)	Old Rate (4bcfe7)	Diff	Status
torchvision-resnet50	64	807.52	3,158.93	-74.44%	🔴
torchvision-resnet50_fp16	64	1,105.52	6,668.55	-83.42%	🔴
torchvision-densenet121	32	607.90	2,707.64	-77.55%	🔴
torchvision-densenet121_fp16	32	4,307.84	4,558.06	-5.49%	🔴
torchvision-inceptionv3	32	748.55	1,797.18	-58.35%	🔴
torchvision-inceptionv3_fp16	32	388.30	2,825.28	-86.26%	🔴
cadene-inceptionv4	16	154.93	822.57	-81.17%	🔴
cadene-resnext64x4	16	498.52	783.23	-36.35%	🔴
slim-mobilenet	64	7,116.21	8,423.81	-15.52%	🔴
slim-nasnetalarge	64	217.83	229.41	-5.05%	🔴
slim-resnet50v2	64	3,176.30	3,176.86	-0.02%	✅
bert-mrpc-onnx	8	1,172.24	1,170.11	0.18%	✅
bert-mrpc-tf	1	490.85	494.35	-0.71%	✅
pytorch-examples-wlang-gru	1	448.07	598.80	-25.17%	🔴
pytorch-examples-wlang-lstm	1	321.05	389.11	-17.49%	🔴
torchvision-resnet50_1	1	768.38	750.26	2.42%	✅
cadene-dpn92_1	1	332.78	438.99	-24.19%	🔴
cadene-resnext101_1	1	193.63	364.33	-46.85%	🔴
onnx-taau-downsample	1	399.49	401.95	-0.61%	✅
dlrm-criteoterabyte	1	32.28	32.67	-1.20%	✅
dlrm-criteoterabyte_fp16	1	52.23	52.59	-0.68%	✅
agentmodel	1	11,724.59	7,471.94	56.91%	🔆
unet_fp16	2	50.39	57.41	-12.23%	🔴
resnet50v1_fp16	1	948.54	940.42	0.86%	✅
resnet50v1_int8	1	396.81	950.24	-58.24%	🔴
bert_base_cased_fp16	64	1,104.38	1,102.43	0.18%	✅
bert_large_uncased_fp16	32	347.72	346.73	0.29%	✅
bert_large_fp16	1	204.01	203.03	0.48%	✅
distilgpt2_fp16	16	2,100.31	2,092.22	0.39%	✅
yolov5s	1	559.48	559.72	-0.04%	✅
tinyllama	1	45.97	46.00	-0.07%	✅
vicuna-fastchat	1	44.13	44.11	0.04%	✅
whisper-tiny-encoder	1	415.25	415.41	-0.04%	✅
whisper-tiny-decoder	1	412.66	410.21	0.60%	✅
llama2_7b	1	20.99	20.95	0.16%	✅
qwen1.5-7b	1	23.68	23.67	0.06%	✅
phi3-3.8b	1	26.88	26.84	0.17%	✅
llama3-8b	1	21.90	21.85	0.23%	✅
whisper-large-encoder	1	10.22	10.21	0.15%	✅
whisper-large-decoder	1	106.94	104.30	2.53%	✅
mistral-7b	1	23.92	23.89	0.15%	✅
FLUX.1-schnell	1	250.08	747.04	-66.52%	🔴

Regressions detected 🔴

[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