Supose the following kernel:
kernel void kersample(double2 a[ ][ ], out double2 c<>, float2 i<> )
{ c = a[ i ] + 1; }
It would work properly on both, CPU and GPU backend.
However, if by mistake you type "double2 a[ ]" :
CPU backend results don't change
GPU backend returns only data from the first row of matrix 'a'
I think this different behavior could be a bit confusing when
developing in software mode.
