Implement Normalize Backward

[littlecutebird]

• Add NormalizeBackward API wrapped by MIOPEN_BETA_API
• Add solver and kernel for cases: contiguous tensor, reduce dimension is last dimension, reduce size % 256 = 0 (256 is chosen as block size for kernel), outer_size >= number of CUs. Those cases has performance improvement compared to ROCm pytorch. Please note that bfp16 is not supported because bfp16 calculation in HIP is calculated after convert to fp32, which reduce perf.
• Added driver and gtest for NormalizeBackward.

type	Forward
float32	2.32
float16	2.52

fp32

input_size	num_elem	p	reduce_dim	ROCm	MIOpen	Improvement
11 37 11 11 1280	63036160	2	4	4434010	1283040	3.46
11 37 11 11 1280	63036160	2.5	4	4923864	3440900	1.43
13 2 3221 1024	85755904	2	3	6017754	1752640	3.43
13 2 3221 1024	85755904	2.5	3	6631930	4684560	1.42
131 97 3584	45541888	2	2	3220495	931529	3.46
131 97 3584	45541888	2.5	2	3562568	2492200	1.43
14387 2 3 256	22098432	2	3	1604645	557403	2.88
14387 2 3 256	22098432	2.5	3	1793928	1312380	1.37
172 2 2 2 35584	48963584	2	4	3453336	1001470	3.45
172 2 2 2 35584	48963584	2.5	4	3831136	2683940	1.43
1877 2 23 2 512	88414208	2	4	6173956	1875100	3.29
1877 2 23 2 512	88414208	2.5	4	6831684	4907600	1.39
19 3 17 2 29696	57550848	2	4	4051365	1170830	3.46
19 3 17 2 29696	57550848	2.5	4	4496589	3149270	1.43
19 5 4027 256	97936640	2	3	6998025	2403470	2.91
19 5 4027 256	97936640	2.5	3	7737972	5755220	1.34
2 29 3 2 141056	49087488	2	4	3479493	1022340	3.40
2 29 3 2 141056	49087488	2.5	4	3843096	2702680	1.42
2 311 31 2048	39489536	2	3	2767901	802871	3.45
2 311 31 2048	39489536	2.5	3	3097582	2146900	1.44
2 37957 2 256	38867968	2	3	2779540	955178	2.91
2 37957 2 256	38867968	2.5	3	3088572	2277070	1.36
2 49363 3 256	75821568	2	3	5452614	1849560	2.95
2 49363 3 256	75821568	2.5	3	6012118	4430710	1.36
2 63541 3 256	97598976	2	3	6971395	2374200	2.94
2 63541 3 256	97598976	2.5	3	7712790	5699290	1.35
2 9818 2 512	20107264	2	3	1441360	437402	3.30
2 9818 2 512	20107264	2.5	3	1614396	1117440	1.44
20549 7 2 256	73647616	2	3	5284847	1812170	2.92
20549 7 2 256	73647616	2.5	3	5880898	4340240	1.35
4 3881 21 256	83457024	2	3	5968965	2058190	2.90
4 3881 21 256	83457024	2.5	3	6619226	4902430	1.35
5 11 5 5 54272	74624000	2	4	5238577	1501170	3.49
5 11 5 5 54272	74624000	2.5	4	5789439	4060110	1.43
5 3 211 3840	12153600	2	3	895535	263778	3.40
5 3 211 3840	12153600	2.5	3	1003046	677144	1.48
61613 1280	78864640	2	1	5543941	1599840	3.47
61613 1280	78864640	2.5	1	6118706	4306930	1.42
651 133888	87161088	2	1	6092961	1761470	3.46
651 133888	87161088	2.5	1	6715880	4758320	1.41
7 59 2 100864	83313664	2	3	5852554	1672450	3.50
7 59 2 100864	83313664	2.5	3	6480370	4542550	1.43
78533 256	20104448	2	1	1468958	508019	2.89
78533 256	20104448	2.5	1	1649928	1196430	1.38

fp16

opname	dtype	input_size	num_elem	p	reduce_dim	ROCm	MIOpen	Improvement
Normalize	float16	11 37 11 11 1280	63036160	2	4	3196507	1083980	2.95
Normalize	float16	11 37 11 11 1280	63036160	2.5	4	3142656	1245010	2.52
Normalize	float16	13 2 3221 1024	85755904	2	3	4319363	1501150	2.88
Normalize	float16	13 2 3221 1024	85755904	2.5	3	4227689	1719890	2.46
Normalize	float16	131 97 3584	45541888	2	2	2332048	781748	2.98
Normalize	float16	131 97 3584	45541888	2.5	2	2285526	898533	2.54
Normalize	float16	14387 2 3 256	22098432	2	3	1233432	516762	2.39
Normalize	float16	14387 2 3 256	22098432	2.5	3	1230406	572195	2.15
Normalize	float16	172 2 2 2 35584	48963584	2	4	2509903	862319	2.91
Normalize	float16	172 2 2 2 35584	48963584	2.5	4	2459371	986305	2.49
Normalize	float16	1877 2 23 2 512	88414208	2	4	4514497	1691070	2.67
Normalize	float16	1877 2 23 2 512	88414208	2.5	4	4431526	1915170	2.31
Normalize	float16	19 3 17 2 29696	57550848	2	4	2931277	1013720	2.89
Normalize	float16	19 3 17 2 29696	57550848	2.5	4	2882340	1158760	2.49
Normalize	float16	19 5 4027 256	97936640	2	3	5168268	2234190	2.31
Normalize	float16	19 5 4027 256	97936640	2.5	3	5068704	2483460	2.04
Normalize	float16	2 29 3 2 141056	49087488	2	4	2519934	909612	2.77
Normalize	float16	2 29 3 2 141056	49087488	2.5	4	2465106	1034510	2.38
Normalize	float16	2 311 31 2048	39489536	2	3	2037826	674707	3.02
Normalize	float16	2 311 31 2048	39489536	2.5	3	2011590	773217	2.60
Normalize	float16	2 37957 2 256	38867968	2	3	2095106	887283	2.36
Normalize	float16	2 37957 2 256	38867968	2.5	3	2068248	985811	2.10
Normalize	float16	2 49363 3 256	75821568	2	3	4035139	1717550	2.35
Normalize	float16	2 49363 3 256	75821568	2.5	3	3960294	1906930	2.08
Normalize	float16	2 63541 3 256	97598976	2	3	5156972	2205620	2.34
Normalize	float16	2 63541 3 256	97598976	2.5	3	5052478	2449970	2.06
Normalize	float16	2 9818 2 512	20107264	2	3	1087337	395676	2.75
Normalize	float16	2 9818 2 512	20107264	2.5	3	1076638	446629	2.41
Normalize	float16	20549 7 2 256	73647616	2	3	3906774	1686440	2.32
Normalize	float16	20549 7 2 256	73647616	2.5	3	3844973	1872940	2.05
Normalize	float16	4 3881 21 256	83457024	2	3	4418577	1906380	2.32
Normalize	float16	4 3881 21 256	83457024	2.5	3	4324286	2113760	2.05
Normalize	float16	5 11 5 5 54272	74624000	2	4	3764358	1305290	2.88
Normalize	float16	5 11 5 5 54272	74624000	2.5	4	3685521	1493160	2.47
Normalize	float16	5 3 211 3840	12153600	2	3	680443	222692	3.06
Normalize	float16	5 3 211 3840	12153600	2.5	3	694004	253591	2.74
Normalize	float16	61613 1280	78864640	2	1	3968165	1357260	2.92
Normalize	float16	61613 1280	78864640	2.5	1	3892473	1556220	2.50
Normalize	float16	651 133888	87161088	2	1	4380067	1553620	2.82
Normalize	float16	651 133888	87161088	2.5	1	4282347	1778860	2.41
Normalize	float16	7 59 2 100864	83313664	2	3	4206771	1451800	2.90
Normalize	float16	7 59 2 100864	83313664	2.5	3	4119342	1660810	2.48
Normalize	float16	78533 256	20104448	2	1	1127065	471875	2.39
Normalize	float16	78533 256	20104448	2.5	1	1126806	523236	2.15