[BN] Enable NHWC in OCL

[bghimireamd]

• Enable NHWC for Batch norm forward infer
• Initialize the driver and gtest with similar range of values.

[CAHEK7]

It's barely readable and probably redundant condition.
May I ask you to do the math and simplify it? (at least replace !(n < 3) with (n >= 3), but there are more simplifications possible)

[CAHEK7]

It's not that complex as the condition from src/solver/batchnorm/forward_spatial_multiple.cpp, but still can be simplified, since it contains redundant statements.

[CAHEK7]

One example of the transformations:

X == in_nhw < (32 * 1024 * 1024)
Y == in_cstride > 1024
Z == in_cstride > 512

!(X & Y) & !(X & Z) & !(!Z)
(!X | !Y) & (!X | !Z) & Z
(!X | !Y) & (!X & Z | !Z & Z) // !Z & Z -> false
(!X | !Y) & !X & Z
!X & Z | !Y & !X & Z
!X & Z & (1 | !Y) // (1 | !Y) -> true
!X & Z

So it basically means return (in_nhw >= (32 * 1024 * 1024)) && (in_cstride > 512);
Which probably can be simplified more, since in_nhw is n * in_cstride and we know for sure that in_cstride must be greater than 512.

If there are any doubts about those transformations, here is a proof (tested in excel, lol):
Using that fact that if Y is true, then Z must always be true we can even exclude few cases:

X	Y	Z	old	new	result
0	0	0	FALSE	FALSE	TRUE
1	0	0	FALSE	FALSE	TRUE
0	0	1	TRUE	TRUE	TRUE
1	0	1	FALSE	FALSE	TRUE
0	1	1	TRUE	TRUE	TRUE
1	1	1	FALSE	FALSE	TRUE

[CAHEK7]

Suggested change

	int n, c, h, w;
	size_t n, c, h, w;

[CAHEK7]

Suggested change

	unsigned int in_cstride = h * w;
	unsigned int in_nhw = n * in_cstride;
	size_t in_cstride = h * w;
	size_t in_nhw = n * in_cstride;

[CAHEK7]

But actually you can avoid all the prevois int-related comment by the following code:

Suggested change

	int n, c, h, w;
	std::tie(n, c, h, w) = tien<4>(problem.GetXDesc().GetLengths());
	unsigned int in_cstride = h * w;
	auto [n, c, h, w] = tien<4>(problem.GetXDesc().GetLengths());
	auto in_cstride = problem.GetXDesc().GetStrides()[1];

[CAHEK7]

See the comments from src/solver/batchnorm/backward_spatial_multiple.cpp

[CAHEK7]

Suggested change

	if(!(in_nhw < (32 * 1024 * 1024) && in_cstride > 1024) &&
	!(in_nhw < (32 * 1024 * 1024) && in_cstride > 512) && !(in_cstride <= 512))
	{
	return true;
	}
	else
	return false;
	}
	return !(in_nhw < (32 * 1024 * 1024) && in_cstride > 1024) &&
	!(in_nhw < (32 * 1024 * 1024) && in_cstride > 512) && !(in_cstride <= 512);