i get this error when i execute my kernel (in emu mode):
cutilCheckMsg() CUTIL CUDA error: Kernel execution failed in file <t.cu>, line 18 : invalid configuration argument.
dim3 grid(512,512,512); ..... kernel<<< grid , 512 >>>(result);
is this because the z variable in the grid is 512 ? The docs say z can be only 1 but i do not exceed 65536x65536 maximum grid size, why ‘z’ can’t be used? I need x,y and z to run within the 1-512 range, is there a way to do this without doing extra calculations? Right now the only solution seems to be running the kernel with ‘x’ from 1-512 and ‘y’ from 1-65536 and then divide the ‘y’ by 512 to simulate ‘z’, but the division takes many gpu cycles. How do you guys do it?
Thanks in advance for any ideas or suggestions.