Fix remove rocblas tests

[tvukovic-amd]

• Added unsupported fp8 type for dot in case when MIGRAPHX_USE_ROCBLAS is OFF since mlir dot doesn't support fp8
• execute gemm tune tests when MIGRAPHX_USE_ROCBLAS is ON since they use rocBLAS

@apwojcik

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.82%. Comparing base (1f07af9) to head (c97572a).
Report is 150 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff            @@
##           develop    #3100   +/-   ##
========================================
  Coverage    91.82%   91.82%           
========================================
  Files          486      486           
  Lines        18991    18991           
========================================
  Hits         17438    17438           
  Misses        1553     1553           

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

[migraphx-bot]

Test	Batch	Rate new c97572	Rate old 93d77e	Diff	Compare
torchvision-resnet50	64	2,790.70	2,789.01	0.06%	✅
torchvision-resnet50_fp16	64	6,204.32	6,208.17	-0.06%	✅
torchvision-densenet121	32	2,094.36	2,090.07	0.21%	✅
torchvision-densenet121_fp16	32	3,615.08	3,614.88	0.01%	✅
torchvision-inceptionv3	32	1,595.53	1,601.73	-0.39%	✅
torchvision-inceptionv3_fp16	32	2,557.93	2,559.67	-0.07%	✅
cadene-inceptionv4	16	716.58	716.67	-0.01%	✅
cadene-resnext64x4	16	678.32	678.24	0.01%	✅
slim-mobilenet	64	5,814.53	5,817.12	-0.04%	✅
slim-nasnetalarge	64	154.28	154.27	0.01%	✅
slim-resnet50v2	64	2,579.22	2,577.22	0.08%	✅
bert-mrpc-onnx	8	968.90	969.57	-0.07%	✅
bert-mrpc-tf	1	405.88	408.71	-0.69%	✅
pytorch-examples-wlang-gru	1	391.77	390.29	0.38%	✅
pytorch-examples-wlang-lstm	1	368.35	374.69	-1.69%	✅
torchvision-resnet50_1	1	603.21	601.08	0.35%	✅
cadene-dpn92_1	1	383.69	384.30	-0.16%	✅
cadene-resnext101_1	1	325.01	323.16	0.57%	✅
onnx-taau-downsample	1	306.96	307.14	-0.06%	✅
dlrm-criteoterabyte	1	28.54	28.50	0.12%	✅
dlrm-criteoterabyte_fp16	1	47.17	47.20	-0.06%	✅
agentmodel	1	7,584.43	7,635.69	-0.67%	✅
unet_fp16	2	56.50	56.51	-0.02%	✅
resnet50v1_fp16	1	891.98	911.93	-2.19%	✅
resnet50v1_int8	1	794.69	783.20	1.47%	✅
bert_base_cased_fp16	64	1,021.04	1,021.20	-0.01%	✅
bert_large_uncased_fp16	32	299.00	298.94	0.02%	✅
bert_large_fp16	1	157.81	156.34	0.94%	✅
distilgpt2_fp16	16	1,830.77	1,831.72	-0.05%	✅
yolov5s	1	478.01	476.00	0.42%	✅
tinyllama	1	33.01	33.00	0.03%	✅
vicuna-fastchat	1	158.69	157.90	0.51%	✅
whisper-tiny-encoder	1	349.66	350.12	-0.13%	✅
whisper-tiny-decoder	1	401.76	401.76	0.00%	✅

This build is OK for 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

[umangyadav]

This shouldn't be necessary.

AMDMIGraphX/src/targets/gpu/fuse_mlir.cpp

Line 234 in 7011bda

return false;

MIGraphX doesn't pick rocMLIR for the FP8 dot/quant_dots. It would use rocBLAS for fp8 dots,quant_dots.

Do you have a case where rocBLAS is disabled in MIGraphX ?

[apwojcik]

Yes, that is required when rocBLAS is not used (see MIGRAPHX_USE_ROCBLAS=OFF).

[tvukovic-amd]

Yes, this is added for case where rocBLAS is disabled in MIGraphX. There is variable MIGRAPHX_USE_ROCBLAS in develop branch, so when it is set to OFF we need to add unsupported fp8 ops for dot and quant dot.

[pfultz2]

This just duplicating code for both branches. It would be better to have rocblas_fp8_available function that just always returns false when rocblas is disabled.