I have a question about the domain of execution in a scatter output, for example, if I have this code:
float var1
float var2;
For i
For j
For k
var2 = var1[y+offset];
For k
sort var2;
For k
var1[y+other_offset] = var2;
Can you run the domain over just x and y dimensions without running over the z dimension, since the entire z dimension is going to be used in every thread?
So, basically, I just want the domain to be over the first two For loops, i and j, not k. I will unroll k inside the kernel.