I was experimenting with OpenCL and timed some commands. It seems to me that “non-blocking” commands, such as clEnqueueWriteBuffer with CL_FALSE for blocking_write, don’t return immediately, but only after the command has actually been executed, i.e. they actually are blocking…
The same problem has been discussed here before, but I wanted to know if anyone knows more about it and can confirm the problem.
It appears to be a limitation of the Nvidia implementation. Are there any plans to implement non-blocking commands? Or do other implementations (eg. by Apple or AMD) support non-blocking commands already?