I am wondering if it is possible to have like one side of a 2D stream larger than 8192? Say if the other side is small enough such that L * W < 8192 * 8192 where L is the long side and W is the short side.
Hmm...yeah, I knew that. Just I am trying to slice a super domain up into chunks to feed to different GPUs and the original domain must be square for the problem I am working with. There is also the problem that I have to for each cell reference the cells around it. Referencing them becomes hairy when I flatten the stream out...But thank you.
And the time I tried simulating a 2D aray in a 1D array seemed to have a signifigant performance drop...
When my 1D streams are larger than 8192 and I try to add them up in a kernel using a '+' sign, I get output of 0.0 when I anticipate non-zero floating numbers based on my input, which I validate by printing out. Did anyone else have a similar problem? Is there any limitations that are not documented in the provided Brook+ specifications?