read_imagef : CL_RGBA vs CL_R


Is it normal that reading a RGBA+FLOAT image is much slower than reading a R+FLOAT?
In my case, it’s about 3 times slower. I understand that in the first case I read 4 floats while I read only 1 float for the second case.
I expected to have a slight increase in reading time but not that much…



Did you had a look at the coalescing memory accesses? there will be 3 times less memory accesses when all acceces are coalesced, isn’t it?