Implement LPPool

[hieule88]

• Added LPPool 1D 2D forward and backward.

• Added driver test and gtest for LPPool.

• New API is guarded by MIOPEN_BETA_API macro.

• Average over all cases:

LPPool1D

Type	Forward	Backward
float16	2.35	4.12
float32	2.37	3.51
bfloat16	2.40	4.17

LPPool1D FP16

op_name	dtype	input_size	p	kernel_size	stride	contiguous	direction	rocm_kernel_avg	MIOpen	MIOpen_over_rocm
LPPool1d	float16	[16 672 32]	2	2	1	contiguous	fwd	58000	28675	2.022667829
LPPool1d	float16	[16 672 32]	2	2	1	contiguous	bwd	101184	36888	2.743005856
LPPool1d	float16	[16 672 32]	2	2	1	noncontiguous	fwd	69536	29955	2.32134869
LPPool1d	float16	[16 672 32]	2	2	1	noncontiguous	bwd	139455	38150	3.655439056
LPPool1d	float16	[16 960 32]	2	2	1	contiguous	fwd	62048	37812	1.640960542
LPPool1d	float16	[16 960 32]	2	2	1	contiguous	bwd	108479	51519	2.105611522
LPPool1d	float16	[16 960 32]	2	2	1	noncontiguous	fwd	75680	39217	1.929775353
LPPool1d	float16	[16 960 32]	2	2	1	noncontiguous	bwd	155887	52621	2.962448452
LPPool1d	float16	[3 2048 64]	2	2	1	contiguous	fwd	58944	32177	1.831867483
LPPool1d	float16	[3 2048 64]	2	2	1	contiguous	bwd	102911	41528	2.478111154
LPPool1d	float16	[3 2048 64]	2	2	1	noncontiguous	fwd	73840	33475	2.205825243
LPPool1d	float16	[3 2048 64]	2	2	1	noncontiguous	bwd	151647	43696	3.470500732
LPPool1d	float16	[64 2208 7]	2	2	1	contiguous	fwd	75408	62665	1.203351153
LPPool1d	float16	[64 2208 7]	2	2	1	contiguous	bwd	130208	95785	1.359377773
LPPool1d	float16	[64 2208 7]	2	2	1	noncontiguous	fwd	108944	66683	1.633759729
LPPool1d	float16	[64 2208 7]	2	2	1	noncontiguous	bwd	222895	99625	2.237340025

LPPool1D FP32

op_name	dtype	input_size	p	kernel_size	stride	contiguous	direction	rocm_kernel_avg	MIOpen	MIOpen_over_rocm
LPPool1d	float32	[16 672 32]	2	2	1	contiguous	fwd	59728	28870	2.068860409
LPPool1d	float32	[16 672 32]	2	2	1	contiguous	bwd	103023	37386	2.755657198
LPPool1d	float32	[16 672 32]	2	2	1	noncontiguous	fwd	74128	32195	2.302469328
LPPool1d	float32	[16 672 32]	2	2	1	noncontiguous	bwd	146720	41013	3.577402287
LPPool1d	float32	[16 960 32]	2	2	1	contiguous	fwd	66448	37653	1.764746501
LPPool1d	float32	[16 960 32]	2	2	1	contiguous	bwd	115967	50826	2.281647188
LPPool1d	float32	[16 960 32]	2	2	1	noncontiguous	fwd	84736	42595	1.989341472
LPPool1d	float32	[16 960 32]	2	2	1	noncontiguous	bwd	170623	54328	3.140608894
LPPool1d	float32	[3 2048 64]	2	2	1	contiguous	fwd	61136	32266	1.894749892
LPPool1d	float32	[3 2048 64]	2	2	1	contiguous	bwd	106160	41368	2.566234771
LPPool1d	float32	[3 2048 64]	2	2	1	noncontiguous	fwd	77904	33617	2.317398935
LPPool1d	float32	[3 2048 64]	2	2	1	noncontiguous	bwd	157215	44000	3.573068182
LPPool1d	float32	[64 2208 7]	2	2	1	contiguous	fwd	87072	62666	1.38946159
LPPool1d	float32	[64 2208 7]	2	2	1	contiguous	bwd	163584	95608	1.710986528
LPPool1d	float32	[64 2208 7]	2	2	1	noncontiguous	fwd	119376	69137	1.726658663
LPPool1d	float32	[64 2208 7]	2	2	1	noncontiguous	bwd	264847	100800	2.627450397

LPPool1D BFP16

op_name	dtype	input_size	p	kernel_size	stride	contiguous	direction	rocm_kernel_avg	MIOpen	MIOpen_over_rocm
LPPool1d	bfloat16	[16 672 32]	2	2	1	contiguous	fwd	60016	28871	2.078764158
LPPool1d	bfloat16	[16 672 32]	2	2	1	contiguous	bwd	108479	36924	2.937899469
LPPool1d	bfloat16	[16 672 32]	2	2	1	noncontiguous	fwd	71936	29937	2.402912784
LPPool1d	bfloat16	[16 672 32]	2	2	1	noncontiguous	bwd	144095	38524	3.740395598
LPPool1d	bfloat16	[16 960 32]	2	2	1	contiguous	fwd	64448	37902	1.700385204
LPPool1d	bfloat16	[16 960 32]	2	2	1	contiguous	bwd	114207	51022	2.238387362
LPPool1d	bfloat16	[16 960 32]	2	2	1	noncontiguous	fwd	77840	39217	1.984853507
LPPool1d	bfloat16	[16 960 32]	2	2	1	noncontiguous	bwd	159198	52853	3.012090137
LPPool1d	bfloat16	[3 2048 64]	2	2	1	contiguous	fwd	60816	32302	1.882731719
LPPool1d	bfloat16	[3 2048 64]	2	2	1	contiguous	bwd	107967	41671	2.590938542
LPPool1d	bfloat16	[3 2048 64]	2	2	1	noncontiguous	fwd	75615	33635	2.248104653
LPPool1d	bfloat16	[3 2048 64]	2	2	1	noncontiguous	bwd	157103	43591	3.604023766
LPPool1d	bfloat16	[64 2208 7]	2	2	1	contiguous	fwd	77359	62826	1.231321427
LPPool1d	bfloat16	[64 2208 7]	2	2	1	contiguous	bwd	138367	96213	1.438132061
LPPool1d	bfloat16	[64 2208 7]	2	2	1	noncontiguous	fwd	109663	66755	1.642768332

LPPool 2D

Type	Forward	Backward
float16	1.25	1.47
float32	1.36	1.68
bfloat16	1.35	1.56

LPPool2D FP16

op_name	dtype	input_size	p	kernel_size	stride	contiguous	direction	rocm_kernel_avg	MIOpen	MIOpen_over_rocm
LPPool2d	float16	[256 256 6 6]	2	[1 1]	[1 1]	noncontiguous	fwd	210288	136549	1.540018601
LPPool2d	float16	[256 256 6 6]	2	[1 1]	[1 1]	noncontiguous	bwd	456046	164619	2.770312054
LPPool2d	float16	[16 72 64 64]	2	[3 3]	[1 1]	noncontiguous	fwd	482078	422766	1.140295104
LPPool2d	float16	[16 72 64 64]	2	[3 3]	[1 1]	noncontiguous	bwd	2646486	598016	4.425443466
LPPool2d	float16	[16 120 64 64]	2	[2 2]	[1 1]	noncontiguous	fwd	880077	578603	1.521037741
LPPool2d	float16	[16 120 64 64]	2	[2 2]	[1 1]	noncontiguous	bwd	3723649	1044300	3.565688978
LPPool2d	float16	[16 480 32 32]	2	[4 4]	[1 1]	noncontiguous	fwd	840396	730067	1.151121746
LPPool2d	float16	[16 480 32 32]	2	[4 4]	[1 1]	noncontiguous	bwd	2259911	1167730	1.935302681

LPPool2D FP32

op_name	dtype	input_size	p	kernel_size	stride	contiguous	direction	rocm_kernel_avg	MIOpen	MIOpen_over_rocm
LPPool2d	float32	[256 256 6 6]	2	[1 1]	[1 1]	noncontiguous	fwd	249567	136692	1.825761566
LPPool2d	float32	[256 256 6 6]	2	[1 1]	[1 1]	noncontiguous	bwd	627934	168639	3.723539632
LPPool2d	float32	[16 72 64 64]	2	[3 3]	[1 1]	noncontiguous	fwd	806060	646911	1.246013748
LPPool2d	float32	[16 72 64 64]	2	[3 3]	[1 1]	noncontiguous	bwd	1934392	1134820	1.704580462
LPPool2d	float32	[16 120 64 64]	2	[2 2]	[1 1]	noncontiguous	fwd	1337402	891175	1.500717592
LPPool2d	float32	[16 120 64 64]	2	[2 2]	[1 1]	noncontiguous	bwd	3362195	1787430	1.881021914
LPPool2d	float32	[16 480 32 32]	2	[4 4]	[1 1]	noncontiguous	fwd	1313339	1198500	1.09581894
LPPool2d	float32	[16 480 32 32]	2	[4 4]	[1 1]	noncontiguous	bwd	3545698	3306250	1.072422836

LPPool2D BFP16

op_name	dtype	input_size	p	kernel_size	stride	contiguous	direction	rocm_kernel_avg	MIOpen	MIOpen_over_rocm
LPPool2d	bfloat16	[256 256 6 6]	2	[1 1]	[1 1]	noncontiguous	fwd	221390	137280	1.612689394
LPPool2d	bfloat16	[256 256 6 6]	2	[1 1]	[1 1]	noncontiguous	bwd	490428	166275	2.949499323
LPPool2d	bfloat16	[16 72 64 64]	2	[3 3]	[1 1]	noncontiguous	fwd	502860	424123	1.185646617
LPPool2d	bfloat16	[16 72 64 64]	2	[3 3]	[1 1]	noncontiguous	bwd	2771849	599962	4.620040936
LPPool2d	bfloat16	[16 120 64 64]	2	[2 2]	[1 1]	noncontiguous	fwd	915832	580460	1.577769355
LPPool2d	bfloat16	[16 120 64 64]	2	[2 2]	[1 1]	noncontiguous	bwd	3039143	1033949	2.939354842
LPPool2d	bfloat16	[16 480 32 32]	2	[4 4]	[1 1]	noncontiguous	fwd	871049	731553	1.190684749
LPPool2d	bfloat16	[16 480 32 32]	2	[4 4]	[1 1]	noncontiguous	bwd	3307300	1170540	2.825448084

[BradPepersAMD]

MIOpen is moving to the new monorepo setup and all older unmerged PR's are being closed. Please re-open this as part of the new repo if these changes are still needed.