Implement reference op for rotary embedding

[music-dino]

Motivation

Create a separate reference implementation for rotary embedding to be used with GQA and Sparse Attention. The reference implementation is accompanied by an op builder, which is to be used instead of the reference op directly, with the idea being to implement rotary embedding via operator composition sometime down the line.

Technical Details

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.

[migraphx-bot]

Test	Batch	Rate new 79c234	Rate old 38fdc6	Diff	Compare
torchvision-resnet50	64	3,157.45	3,173.83	-0.52%	✅
torchvision-resnet50_fp16	64	6,590.55	6,613.50	-0.35%	✅
torchvision-densenet121	32	2,437.43	2,445.12	-0.31%	✅
torchvision-densenet121_fp16	32	4,116.15	4,132.06	-0.38%	✅
torchvision-inceptionv3	32	1,665.57	1,673.47	-0.47%	✅
torchvision-inceptionv3_fp16	32	2,589.12	2,596.39	-0.28%	✅
cadene-inceptionv4	16	794.28	797.27	-0.38%	✅
cadene-resnext64x4	16	802.11	806.10	-0.49%	✅
slim-mobilenet	64	8,197.09	8,232.04	-0.42%	✅
slim-nasnetalarge	64	221.72	222.86	-0.51%	✅
slim-resnet50v2	64	3,295.32	3,305.41	-0.31%	✅
bert-mrpc-onnx	8	1,132.24	1,144.06	-1.03%	✅
bert-mrpc-tf	1	486.43	486.42	0.00%	✅
pytorch-examples-wlang-gru	1	317.25	309.90	2.37%	✅
pytorch-examples-wlang-lstm	1	450.92	387.21	16.45%	🔆
torchvision-resnet50_1	1	807.28	745.35	8.31%	🔆
cadene-dpn92_1	1	436.44	428.94	1.75%	✅
cadene-resnext101_1	1	368.60	369.63	-0.28%	✅
onnx-taau-downsample	1	398.07	399.22	-0.29%	✅
dlrm-criteoterabyte	1	31.92	32.03	-0.36%	✅
dlrm-criteoterabyte_fp16	1	51.00	51.11	-0.21%	✅
agentmodel	1	9,826.08	9,659.41	1.73%	✅
unet_fp16	2	59.03	59.19	-0.28%	✅
resnet50v1_fp16	1	993.33	991.39	0.20%	✅
resnet50v1_int8	1	992.10	971.32	2.14%	✅
bert_base_cased_fp16	64	1,099.32	1,104.24	-0.45%	✅
bert_large_uncased_fp16	32	343.81	345.64	-0.53%	✅
bert_large_fp16	1	197.82	198.02	-0.10%	✅
distilgpt2_fp16	16	2,076.27	2,085.17	-0.43%	✅
yolov5s	1	587.50	588.82	-0.23%	✅
tinyllama	1	43.80	43.95	-0.35%	✅
vicuna-fastchat	1	45.04	45.27	-0.51%	✅
whisper-tiny-encoder	1	410.03	410.98	-0.23%	✅
whisper-tiny-decoder	1	414.22	415.37	-0.28%	✅
llama2_7b	1	19.11	19.15	-0.24%	✅
qwen1.5-7b	1	23.43	23.53	-0.42%	✅
phi3-3.8b	1	26.57	26.70	-0.48%	✅
mask-rcnn	1	12.15	12.24	-0.72%	✅
llama3-8b	1	21.65	21.74	-0.40%	✅
whisper-large-encoder	1	10.17	10.22	-0.51%	✅
whisper-large-decoder	1	98.90	99.83	-0.94%	✅
mistral-7b	1	23.65	23.74	-0.35%	✅
FLUX.1-schnell	1	726.34	721.05	0.73%	✅
nan	nan	nan	nan	nan%	❌

This build is not recommended to merge 🔴

[migraphx-bot]

     ✅ bert-mrpc-onnx: PASSED: MIGraphX meets tolerance

❌bert-mrpc-tf: ERROR - check error output

2025-09-24 09:25:13.803844: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
Traceback (most recent call last):
File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 359, in
main()
File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 306, in main
graph = load_tf_graph(model_name)
File "/src/AMDMIGraphX/tools/accuracy/accuracy_checker.py", line 300, in load_tf_graph
graph_def.ParseFromString(f.read())
File "/usr/local/lib/python3.10/dist-packages/tensorflow/python/lib/io/file_io.py", line 116, in read
self._preread_check()
File "/usr/local/lib/python3.10/dist-packages/tensorflow/python/lib/io/file_io.py", line 77, in _preread_check
self._read_buf = _pywrap_file_io.BufferedInputStream(
tensorflow.python.framework.errors_impl.UnimplementedError: File system scheme '[local]' not implemented (file: '/new-saved-models/tf-misc/bert_mrpc1.pb')

     ✅ 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: PASSED: MIGraphX meets tolerance

     ✅ llama2_7b: PASSED: MIGraphX meets tolerance

     ✅ qwen1.5-7b: PASSED: MIGraphX meets tolerance

     ✅ phi3-3.8b: PASSED: MIGraphX meets tolerance

🔴mask-rcnn: FAILED: MIGraphX is not within tolerance - check verbose output

     ✅ llama3-8b: PASSED: MIGraphX meets tolerance

     ✅ whisper-large-decoder: PASSED: MIGraphX meets tolerance

     ✅ mistral-7b: PASSED: MIGraphX meets tolerance

     ✅ FLUX.1-schnell: PASSED: MIGraphX meets tolerance

[codecov]

Codecov Report

❌ Patch coverage is 86.31579% with 13 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/op/builder/rotary_embedding.cpp	0.00%	12 Missing ⚠️
src/include/migraphx/op/rotary_embedding.hpp	98.80%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #4315      +/-   ##
===========================================
+ Coverage    92.23%   92.24%   +0.02%     
===========================================
  Files          557      562       +5     
  Lines        25924    26453     +529     
===========================================
+ Hits         23909    24401     +492     
- Misses        2015     2052      +37     

Files with missing lines	Coverage Δ
src/include/migraphx/op/rotary_embedding.hpp	98.80% <98.80%> (ø)
src/op/builder/rotary_embedding.cpp	0.00% <0.00%> (ø)

... and 24 files with indirect coverage changes

🚀 New features to boost your workflow:

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