Improve dpp reductions on navi

[pfultz2]

This will use ds_swizzle to compute reduction using all lanes in a wavefront. I am not sure what the best way is to write the inline assembly so for now it does the first 16 reductions in assembly and then does the swizzle using the intrinsic.

[migraphx-bot]

Test	Batch	Rate new 664b2f	Rate old 02f740	Diff	Compare
torchvision-resnet50	64	2,829.38	2,832.03	-0.09%	✅
torchvision-resnet50_fp16	64	6,497.38	6,501.16	-0.06%	✅
torchvision-densenet121	32	2,095.21	2,098.99	-0.18%	✅
torchvision-densenet121_fp16	32	3,669.75	3,666.47	0.09%	✅
torchvision-inceptionv3	32	1,586.49	1,584.68	0.11%	✅
torchvision-inceptionv3_fp16	32	2,574.64	2,569.37	0.21%	✅
cadene-inceptionv4	16	704.23	704.57	-0.05%	✅
cadene-resnext64x4	16	692.48	691.97	0.07%	✅
slim-mobilenet	64	8,330.36	8,324.07	0.08%	✅
slim-nasnetalarge	64	225.42	225.67	-0.11%	✅
slim-resnet50v2	64	2,665.03	2,665.32	-0.01%	✅
bert-mrpc-onnx	8	822.86	822.26	0.07%	✅
bert-mrpc-tf	1	388.97	386.92	0.53%	✅
pytorch-examples-wlang-gru	1	302.17	300.38	0.59%	✅
pytorch-examples-wlang-lstm	1	312.92	310.78	0.69%	✅
torchvision-resnet50_1	1	600.25	598.68	0.26%	✅
torchvision-inceptionv3_1	1	343.99	343.19	0.23%	✅
cadene-dpn92_1	1	400.93	401.44	-0.13%	✅
cadene-resnext101_1	1	324.71	324.52	0.06%	✅
slim-vgg16_1	1	459.86	459.52	0.07%	✅
slim-mobilenet_1	1	2,130.81	2,116.32	0.69%	✅
slim-inceptionv4_1	1	218.45	220.17	-0.78%	✅
onnx-taau-downsample	1	304.74	304.43	0.10%	✅
dlrm-criteoterabyte	1	21.62	21.60	0.08%	✅
dlrm-criteoterabyte_fp16	1	40.56	40.66	-0.26%	✅
agentmodel	1	5,708.90	5,930.88	-3.74%	🔴
unet_fp16	2	54.79	54.75	0.07%	✅
resnet50v1_fp16	1	939.69	959.61	-2.08%	✅
bert_base_cased_fp16	64	903.32	903.28	0.00%	✅
bert_large_uncased_fp16	32	285.63	285.65	-0.01%	✅
bert_large_fp16	1	166.87	166.48	0.23%	✅
distilgpt2_fp16	16	1,281.87	1,282.20	-0.03%	✅

This build is not recommended to 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

     ✅ torchvision-inceptionv3_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-dpn92_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-resnext101_1: PASSED: MIGraphX meets tolerance

     ✅ slim-vgg16_1: PASSED: MIGraphX meets tolerance

     ✅ slim-mobilenet_1: PASSED: MIGraphX meets tolerance

     ✅ slim-inceptionv4_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: FAILED: MIGraphX is not within tolerance - check verbose output

     ✅ bert_large_uncased_fp16: PASSED: MIGraphX meets tolerance

     ✅ bert_large: PASSED: MIGraphX meets tolerance

🔴distilgpt2_fp16: FAILED: MIGraphX is not within tolerance - check verbose output