I try to speed up an opencl kernel, using vector data types.
If I run the kernel on cpu, I get a speed up.
but on gpu, the kernel is slower than with scalar data types.
Is the use of vector data types on gpu unusefull?
That is not necessary, but it may be true sometimes.
It is not possible if your kernel falls in that case or you are doing anything wrong unless you post the kernel.
Also post your system configuration:CPU,GPU<SDK<DRIVER<OS.
This might be interesting to you:http://forums.amd.com/forum/messageview.cfm?catid=390&threadid=136119