Show kernel time when using gpu-driver

[pfultz2]

No description provided.

[codecov]

Codecov Report

Merging #1289 (62c707a) into develop (f253160) will not change coverage.
The diff coverage is n/a.

❗ Current head 62c707a differs from pull request most recent head 6ee87f9. Consider uploading reports for the commit 6ee87f9 to get more accurate results

@@           Coverage Diff            @@
##           develop    #1289   +/-   ##
========================================
  Coverage    93.14%   93.14%           
========================================
  Files          437      437           
  Lines        14297    14297           
========================================
  Hits         13317    13317           
  Misses         980      980           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f253160...6ee87f9. Read the comment docs.

[migraphx-bot]

Test	Rate new ce738c	Rate old bab950	Diff	Compare
torchvision-resnet50	2,220.43	2,224.44	-0.18%	✅
torchvision-resnet50_fp16	4,739.90	4,748.13	-0.17%	✅
torchvision-alexnet	4,971.68	4,979.89	-0.16%	✅
torchvision-alexnet_fp16	26,117.23	26,234.13	-0.45%	✅
torchvision-densenet121	1,637.34	1,633.89	0.21%	✅
torchvision-densenet121_fp16	2,528.15	2,530.44	-0.09%	✅
torchvision-inceptionv3	1,093.03	1,095.09	-0.19%	✅
torchvision-inceptionv3_fp16	1,985.59	1,984.58	0.05%	✅
torchvision-vgg16	894.81	895.54	-0.08%	✅
torchvision-vgg16_fp16	1,725.40	1,727.91	-0.15%	✅
cadene-inceptionv4	528.47	528.75	-0.05%	✅
cadene-resnext64x4	577.17	577.67	-0.09%	✅
slim-mobilenet	6,402.41	6,396.55	0.09%	✅
slim-nasnetalarge	203.20	203.33	-0.06%	✅
slim-resnet50v2	2,427.60	2,430.00	-0.10%	✅
bert-mrpc-onnx	639.04	639.18	-0.02%	✅
bert-mrpc-tf	296.87	294.95	0.65%	🔆
pytorch-examples-wlang-gru	230.14	232.24	-0.90%	✅
pytorch-examples-wlang-lstm	305.95	306.25	-0.10%	✅
torchvision-resnet50_1	513.80	512.38	0.28%	✅
torchvision-inceptionv3_1	303.28	303.38	-0.03%	✅
torchvision-vgg16_1	462.43	463.19	-0.16%	✅
cadene-dpn92_1	297.40	302.66	-1.74%	✅
cadene-resnext101_1	235.80	229.29	2.84%	✅
slim-vgg16_1	64.00	64.03	-0.05%	✅
slim-mobilenet_1	1,999.85	1,977.32	1.14%	🔆
slim-inceptionv4_1	196.00	195.67	0.17%	✅
onnx-taau-downsample	259.07	259.00	0.03%	✅

Check results before merge 🔆

[migraphx-bot]

Test	Rate new 3852e4	Rate old bab950	Diff	Compare
torchvision-resnet50	2,224.62	2,224.44	0.01%	✅
torchvision-resnet50_fp16	4,748.58	4,748.13	0.01%	✅
torchvision-alexnet	4,987.84	4,979.89	0.16%	✅
torchvision-alexnet_fp16	26,342.33	26,234.13	0.41%	✅
torchvision-densenet121	1,633.97	1,633.89	0.01%	✅
torchvision-densenet121_fp16	2,524.34	2,530.44	-0.24%	✅
torchvision-inceptionv3	1,096.54	1,095.09	0.13%	✅
torchvision-inceptionv3_fp16	1,982.15	1,984.58	-0.12%	✅
torchvision-vgg16	895.85	895.54	0.03%	✅
torchvision-vgg16_fp16	1,727.62	1,727.91	-0.02%	✅
cadene-inceptionv4	528.37	528.75	-0.07%	✅
cadene-resnext64x4	577.64	577.67	-0.01%	✅
slim-mobilenet	6,398.91	6,396.55	0.04%	✅
slim-nasnetalarge	203.26	203.33	-0.03%	✅
slim-resnet50v2	2,428.69	2,430.00	-0.05%	✅
bert-mrpc-onnx	638.32	639.18	-0.13%	✅
bert-mrpc-tf	295.64	294.95	0.23%	✅
pytorch-examples-wlang-gru	229.15	232.24	-1.33%	✅
pytorch-examples-wlang-lstm	306.78	306.25	0.17%	✅
torchvision-resnet50_1	513.06	512.38	0.13%	✅
torchvision-inceptionv3_1	302.97	303.38	-0.14%	✅
torchvision-vgg16_1	463.29	463.19	0.02%	✅
cadene-dpn92_1	311.18	302.66	2.82%	✅
cadene-resnext101_1	237.52	229.29	3.59%	✅
slim-vgg16_1	64.04	64.03	0.01%	✅
slim-mobilenet_1	1,987.67	1,977.32	0.52%	🔆
slim-inceptionv4_1	195.84	195.67	0.09%	✅
onnx-taau-downsample	259.08	259.00	0.03%	✅

Check results before merge 🔆

[kahmed10]

can we switch from milliseconds to microseconds? I know hipEventElapsedTime returns time in ms, but it's resolution is approximately 1 microsecond. So multiplying the result by 1000. This is more for readability as we could be dealing with smaller kernels or small improvements to a bigger kernel.

[pfultz2]

The host time is usually pretty large, and it would make the time inconsistent with the perf report.

[umangyadav]

@pfultz2 @causten can this be merged ?

[migraphx-bot]

Test	Rate new eba1e7	Rate old 875287	Diff	Compare
torchvision-resnet50	2,222.34	2,218.94	0.15%	✅
torchvision-resnet50_fp16	4,752.11	4,748.02	0.09%	✅
torchvision-alexnet	4,973.38	4,977.02	-0.07%	✅
torchvision-alexnet_fp16	26,144.65	26,216.08	-0.27%	✅
torchvision-densenet121	1,629.26	1,626.27	0.18%	✅
torchvision-densenet121_fp16	2,527.65	2,530.02	-0.09%	✅
torchvision-inceptionv3	1,096.43	1,096.56	-0.01%	✅
torchvision-inceptionv3_fp16	1,990.15	1,983.19	0.35%	✅
torchvision-vgg16	894.35	894.94	-0.07%	✅
torchvision-vgg16_fp16	1,724.77	1,725.04	-0.02%	✅
cadene-inceptionv4	528.86	528.41	0.08%	✅
cadene-resnext64x4	576.83	575.42	0.24%	✅
slim-mobilenet	6,396.48	6,390.86	0.09%	✅
slim-nasnetalarge	202.97	203.03	-0.03%	✅
slim-resnet50v2	2,425.22	nan	nan%	❌
bert-mrpc-onnx	638.68	684.31	-6.67%	✅
bert-mrpc-tf	296.93	308.71	-3.82%	✅
pytorch-examples-wlang-gru	228.92	242.24	-5.50%	✅
pytorch-examples-wlang-lstm	305.81	306.38	-0.19%	✅
torchvision-resnet50_1	510.62	510.54	0.01%	✅
torchvision-inceptionv3_1	303.04	292.48	3.61%	✅
torchvision-vgg16_1	463.32	463.42	-0.02%	✅
cadene-dpn92_1	311.28	307.72	1.15%	✅
cadene-resnext101_1	237.49	234.34	1.34%	✅
slim-vgg16_1	63.98	63.99	-0.02%	✅
slim-mobilenet_1	1,971.10	1,979.40	-0.42%	✅
slim-inceptionv4_1	195.63	184.11	6.26%	✅
onnx-taau-downsample	258.88	259.30	-0.16%	✅

This build is not recommended to merge 🔴