Grid and Thread size problem...

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,
I do a simple test :

global void test(float* diff)
{
diff[0] = 3.33;
}

(…)

dim3 grid(4,4,D);
dim3 threads(16,16,1);
test<<<grid,threads>>>(diff_gpu);

It works with D=1 but not with D>1 :s. My program is more complicated than this one and actually, this is just a test to debug my program.
Thx for your help,
Vince

2

The grid of blocks could be only 2D (as written in the manual).
A block could be 3D.

3

Hum… Ok I see! But, In my test, when I try

dim3 grid(4,4,1);
dim3 threads(16,16,D);

with D>1, it doesn’t work… Strange no?

4

Pb fixed!

When I use blocks of size 16x16x1 and 16x16x2 it works.
But with 16x16x3…, it doesn’t works because I have more than 512 threads per block!
Thx for your help
Vince