Can opencl chain multiple passes without returning to CPU? -
i want auto scale data. so, want pass through data , find maximum extents of data. want go through data, calculations, , send results opengl rendering. type of multipass thing possible in opencl? or cpu have direct "find extents" calc, results, , direct other calc that?
it sounds need 2 opencl kernels, 1 calculating min , max , other scale data. using opencl command queues , events can queue these 2 kernels in order , store results first in global memory, reading results in second kernel. semantics of opencl command queues , events (assuming don't have out-of-order execution enabled) ensure 1 completes before other without interaction host application (see clenqueuendrangekernel).
Comments
Post a Comment