Switch CUDA group numbering to new ordering

[semi-h]

[semi-h]

Relevant to #131

[semi-h]

The logic is also hardcoded here

x3d2/src/backend/cuda/backend.f90

Lines 772 to 780 in c02672e

	if (x%dir == DIR_X) then
	n = dims(1); n_j = dims(2); n_i = dims(3); n_i_pad = dims_padded(3)
	else if (x%dir == DIR_Y) then
	n = dims(2); n_j = dims(1); n_i = dims(3); n_i_pad = dims_padded(3)
	else if (x%dir == DIR_Z) then
	n = dims(3); n_j = dims(2); n_i = dims(1); n_i_pad = dims_padded(1)
	else
	error stop 'scalar_product_cuda does not support DIR_C fields!'
	end if

x3d2/src/backend/cuda/backend.f90

Lines 846 to 854 in c02672e

	if (f%dir == DIR_X) then
	n = dims(1); n_j = dims(2); n_i = dims(3); n_i_pad = dims_padded(3)
	else if (f%dir == DIR_Y) then
	n = dims(2); n_j = dims(1); n_i = dims(3); n_i_pad = dims_padded(3)
	else if (f%dir == DIR_Z) then
	n = dims(3); n_j = dims(2); n_i = dims(1); n_i_pad = dims_padded(1)
	else
	error stop 'field_max_mean does not support DIR_C fields!'
	end if

Just copy the below OpenMP mapping rules back to CUDA

x3d2/src/backend/omp/backend.f90

Lines 679 to 687 in c02672e

	if (f%dir == DIR_X) then
	n = dims(1); n_j = dims(2); n_i = dims(3); n_i_pad = dims_padded(3)
	else if (f%dir == DIR_Y) then
	n = dims(2); n_j = dims(1); n_i = dims(3); n_i_pad = dims_padded(3)
	else if (f%dir == DIR_Z) then
	n = dims(3); n_j = dims(1); n_i = dims(2); n_i_pad = dims_padded(2)
	else
	error stop 'field_max_mean does not support DIR_C fields!'
	end if

[semi-h]

sum_yintox and sum_zintox kernels also hardcode the mapping, but identical to reorder_y2x and reorder_z2x rules anyways.