Skip FP8 dot/gemm for MLIR and offload them to rocBLAS

[umangyadav]

MIGraphX has verify tests where it tests FP8 * FP8 = FP8 case.
rocBLAS has support for it but not the MLIR. rocBLAS internally converts the output of the GEMM to fp8.

FP8 * FP8 = FP8 case is probably unlikely to happen for the gemm/dot operations. It should mostly be fp8 * fp8 = fp32.

FP8 tests are run on MI300 hardware and therefore Jenkins has not seen any failures but 1416489 broke unit-tests for FP8 * FP8 = FP8 case on MI300 because with that change all the gemms with k<=1028 are now being offloaded to MLIR.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.83%. Comparing base (f0cb545) to head (c0fc875).

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #2924   +/-   ##
========================================
  Coverage    91.83%   91.83%           
========================================
  Files          479      479           
  Lines        18340    18340           
========================================
  Hits         16842    16842           
  Misses        1498     1498           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[migraphx-bot]

Test	Batch	Rate new c0fc87	Rate old f0cb54	Diff	Compare
torchvision-resnet50	64	2,819.41	2,825.79	-0.23%	✅
torchvision-resnet50_fp16	64	6,543.00	6,546.40	-0.05%	✅
torchvision-densenet121	32	2,090.71	2,100.61	-0.47%	✅
torchvision-densenet121_fp16	32	3,686.38	3,687.46	-0.03%	✅
torchvision-inceptionv3	32	1,601.12	1,600.62	0.03%	✅
torchvision-inceptionv3_fp16	32	2,551.69	2,551.41	0.01%	✅
cadene-inceptionv4	16	715.33	714.73	0.08%	✅
cadene-resnext64x4	16	678.81	678.24	0.08%	✅
slim-mobilenet	64	5,843.68	5,844.96	-0.02%	✅
slim-nasnetalarge	64	152.96	152.88	0.05%	✅
slim-resnet50v2	64	2,583.29	2,582.46	0.03%	✅
bert-mrpc-onnx	8	917.29	916.77	0.06%	✅
bert-mrpc-tf	1	437.36	436.75	0.14%	✅
pytorch-examples-wlang-gru	1	422.20	426.30	-0.96%	✅
pytorch-examples-wlang-lstm	1	408.09	394.12	3.54%	🔆
torchvision-resnet50_1	1	609.04	604.22	0.80%	✅
cadene-dpn92_1	1	392.08	391.28	0.20%	✅
cadene-resnext101_1	1	331.12	331.08	0.01%	✅
onnx-taau-downsample	1	306.13	304.34	0.59%	✅
dlrm-criteoterabyte	1	28.72	28.73	-0.03%	✅
dlrm-criteoterabyte_fp16	1	48.37	48.31	0.13%	✅
agentmodel	1	7,524.74	7,604.45	-1.05%	✅
unet_fp16	2	57.59	57.62	-0.04%	✅
resnet50v1_fp16	1	907.17	907.72	-0.06%	✅
resnet50v1_int8	1	824.24	807.49	2.07%	✅
bert_base_cased_fp16	64	1,052.21	1,052.67	-0.04%	✅
bert_large_uncased_fp16	32	300.24	300.18	0.02%	✅
bert_large_fp16	1	158.49	158.43	0.04%	✅
distilgpt2_fp16	16	1,856.20	1,860.28	-0.22%	✅
yolov5s	1	475.36	476.75	-0.29%	✅
tinyllama	1	32.86	32.85	0.02%	✅
vicuna-fastchat	1	158.35	158.09	0.16%	✅
whisper-tiny-encoder	1	346.94	348.19	-0.36%	✅
whisper-tiny-decoder	1	400.45	401.19	-0.18%	✅

Check results before merge 🔆

[migraphx-bot]

     ✅ 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

     ✅ torchvision-resnet50_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-dpn92_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-resnext101_1: 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