Improvement to ck integration

[pfultz2]

• Add a CI job to test CK
• Add MIGRAPHX_TUNE_CK env variable to only do tuning for CK
• Continue tuning even when there is invalid configs
• Fix a bug with parallel compilation not using all available threads
• Add additional test for gemms using half types
• Removed int32 as supported type since it doesnt pass our test suite

[codecov]

Codecov Report

Merging #1859 (60e1f8e) into develop (1f827a7) will not change coverage.
The diff coverage is n/a.

❗ Current head 60e1f8e differs from pull request most recent head 6696d44. Consider uploading reports for the commit 6696d44 to get more accurate results

@@           Coverage Diff            @@
##           develop    #1859   +/-   ##
========================================
  Coverage    91.39%   91.39%           
========================================
  Files          419      419           
  Lines        15542    15542           
========================================
  Hits         14204    14204           
  Misses        1338     1338           

[umangyadav]

Suggested change

	auto d = value_of(MIGRAPHX_GPU_COMPILE_PARALLEL{});
	auto d = value_of(MIGRAPHX_GPU_COMPILE_PARALLEL{}, n);

[pfultz2]

That wont work, because the n will be cached when MIGRAPHX_GPU_COMPILE_PARALLEL is not set.

[migraphx-bot]

Test	Batch	Rate new 6696d4	Rate old d6aade	Diff	Compare
torchvision-resnet50	64	894.85	895.65	-0.09%	✅
torchvision-resnet50_fp16	64	5,313.14	5,315.31	-0.04%	✅
torchvision-densenet121	32	1,123.99	1,126.78	-0.25%	✅
torchvision-densenet121_fp16	32	3,275.44	3,282.20	-0.21%	✅
torchvision-inceptionv3	32	592.96	593.29	-0.06%	✅
torchvision-inceptionv3_fp16	32	2,494.23	2,517.31	-0.92%	✅
cadene-inceptionv4	16	328.70	329.04	-0.10%	✅
cadene-resnext64x4	16	396.92	397.15	-0.06%	✅
slim-mobilenet	64	7,132.06	7,136.84	-0.07%	✅
slim-nasnetalarge	64	159.84	160.01	-0.11%	✅
slim-resnet50v2	64	1,089.43	1,090.55	-0.10%	✅
bert-mrpc-onnx	8	718.90	719.50	-0.08%	✅
bert-mrpc-tf	1	369.81	370.16	-0.10%	✅
pytorch-examples-wlang-gru	1	294.66	304.54	-3.25%	🔴
pytorch-examples-wlang-lstm	1	298.45	308.91	-3.39%	🔴
torchvision-resnet50_1	1	91.44	91.91	-0.51%	✅
torchvision-inceptionv3_1	1	128.98	128.80	0.14%	✅
cadene-dpn92_1	1	333.88	336.62	-0.82%	✅
cadene-resnext101_1	1	236.49	237.34	-0.36%	✅
slim-vgg16_1	1	53.34	53.36	-0.04%	✅
slim-mobilenet_1	1	1,485.31	1,523.34	-2.50%	✅
slim-inceptionv4_1	1	101.96	101.62	0.34%	✅
onnx-taau-downsample	1	316.36	316.09	0.08%	✅
dlrm-criteoterabyte	1	21.42	21.42	0.01%	✅
dlrm-criteoterabyte_fp16	1	39.95	39.96	-0.03%	✅
agentmodel	1	5,413.13	5,733.29	-5.58%	🔴
unet_fp16	2	52.63	52.83	-0.37%	✅

This build is not recommended to 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

    :white_check_mark:torchvision-inceptionv3_1: PASSED: MIGraphX meets tolerance

🔴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

    :white_check_mark:slim-inceptionv4_1: PASSED: MIGraphX meets tolerance

    :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

[pfultz2]

@turneram Any feedback?

[turneram]

If we want to leave int32 enabled, then adding the requirement that m, n, and k are all divisible by 4 appears to maintain correctness. We could also investigate further to get more precise criteria and then add it back in another PR.