Enable MLIR by default in migraphx builds

[pfultz2]

No description provided.

[codecov]

Codecov Report

Merging #2037 (ac60ff6) into develop (065d06a) will not change coverage.
The diff coverage is n/a.

❗ Current head ac60ff6 differs from pull request most recent head e9bd9ac. Consider uploading reports for the commit e9bd9ac to get more accurate results

@@           Coverage Diff            @@
##           develop    #2037   +/-   ##
========================================
  Coverage    91.40%   91.40%           
========================================
  Files          422      422           
  Lines        15645    15645           
========================================
  Hits         14301    14301           
  Misses        1344     1344           

[umangyadav]

what's the purpose behind having it enabled by default ?
I see that we already have Jenkins job that builds MIGraphX with MLIR.

[causten]

ORT needs 3.26.4 would it make sense to bump this to 3.26.4 too?

[pfultz2]

no, we should make sure our stuff builds with the cmake in ubuntu 22.04.

[causten]

builds and runs as expected.

[pfultz2]

The tidy error requires a fix in rocm-cmake: ROCm/rocm-cmake#140

[migraphx-bot]

Test	Batch	Rate new e9bd9a	Rate old 065d06	Diff	Compare
torchvision-resnet50	64	2,275.64	2,276.51	-0.04%	✅
torchvision-resnet50_fp16	64	5,341.92	5,340.77	0.02%	✅
torchvision-densenet121	32	1,833.33	1,837.64	-0.23%	✅
torchvision-densenet121_fp16	32	3,382.42	3,375.49	0.21%	✅
torchvision-inceptionv3	32	1,343.84	1,343.71	0.01%	✅
torchvision-inceptionv3_fp16	32	2,529.36	2,520.84	0.34%	✅
cadene-inceptionv4	16	675.95	677.79	-0.27%	✅
cadene-resnext64x4	16	589.31	591.12	-0.31%	✅
slim-mobilenet	64	7,213.80	7,207.63	0.09%	✅
slim-nasnetalarge	64	236.14	236.08	0.03%	✅
slim-resnet50v2	64	2,521.58	2,519.90	0.07%	✅
bert-mrpc-onnx	8	719.26	717.63	0.23%	✅
bert-mrpc-tf	1	363.29	364.68	-0.38%	✅
pytorch-examples-wlang-gru	1	315.74	312.99	0.88%	✅
pytorch-examples-wlang-lstm	1	318.16	315.97	0.69%	✅
torchvision-resnet50_1	1	562.83	556.65	1.11%	✅
torchvision-inceptionv3_1	1	306.32	305.74	0.19%	✅
cadene-dpn92_1	1	358.96	354.11	1.37%	✅
cadene-resnext101_1	1	219.74	219.79	-0.02%	✅
slim-vgg16_1	1	223.67	223.39	0.13%	✅
slim-mobilenet_1	1	1,486.87	1,449.30	2.59%	✅
slim-inceptionv4_1	1	222.07	223.03	-0.43%	✅
onnx-taau-downsample	1	321.11	320.68	0.14%	✅
dlrm-criteoterabyte	1	21.67	21.65	0.10%	✅
dlrm-criteoterabyte_fp16	1	40.55	40.57	-0.06%	✅
agentmodel	1	5,761.80	5,856.41	-1.62%	✅
unet_fp16	2	54.99	55.00	-0.03%	✅

This build is OK for merge ✅

[migraphx-bot]

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

    :white_check_mark:bert-mrpc-tf: PASSED: MIGraphX meets tolerance

    :white_check_mark:pytorch-examples-wlang-gru: PASSED: MIGraphX meets tolerance

    :white_check_mark:pytorch-examples-wlang-lstm: PASSED: MIGraphX meets tolerance

    :white_check_mark:torchvision-resnet50_1: PASSED: MIGraphX meets tolerance

🔴torchvision-inceptionv3_1: FAILED: MIGraphX is not within tolerance - check verbose output

🔴cadene-dpn92_1: FAILED: MIGraphX is not within tolerance - check verbose output

    :white_check_mark:cadene-resnext101_1: PASSED: MIGraphX meets tolerance

    :white_check_mark:slim-vgg16_1: PASSED: MIGraphX meets tolerance

    :white_check_mark:slim-mobilenet_1: PASSED: MIGraphX meets tolerance

🔴slim-inceptionv4_1: FAILED: MIGraphX is not within tolerance - check verbose output

    :white_check_mark:dlrm-criteoterabyte: PASSED: MIGraphX meets tolerance

    :white_check_mark:agentmodel: PASSED: MIGraphX meets tolerance

    :white_check_mark:unet: PASSED: MIGraphX meets tolerance

[causten]

what's the purpose behind having it enabled by default ? I see that we already have Jenkins job that builds MIGraphX with MLIR.

The goal has been to use rocMLIR to aid in performance optimizations. Just waiting until rocm5.7 was branched to turn it on. Now all teams can simple turn on a runtime flag rather then rebuilding migraphx manually