Timing execution of OpenCL kernels

Question

Is this a correct way of timing kernel execution time for OpenCL? I am quite keen on using the c++ wrapper (which unfortunately does not have many examples of timings).

cl::CommandQueue queue(context, device, CL_QUEUE_PROFILING_ENABLE, &err);
checkErr(err, "Cannot create the command queue");

/* Warm-up */
for (unsigned i = 0; i < NUMBER_OF_ITERATIONS; ++i)
{
    err = queue.enqueueNDRangeKernel(kernel, cl::NullRange, cl::NDRange(512), cl::NullRange, NULL, NULL);
    checkErr(err, "Cannot enqueue the kernel");
}
queue.finish();

/* Time kernels */
cl::Event start, stop;
queue.enqueueMarker(&start);
for (unsigned i = 0; i < NUMBER_OF_ITERATIONS; ++i)
{
    err = queue.enqueueNDRangeKernel(kernel, cl::NullRange, cl::NDRange(512), cl::NullRange, NULL, NULL);
    checkErr(err, "Cannot enqueue the kernel");
}
queue.enqueueMarker(&stop);

stop.wait();
cl_ulong time_start, time_end;
double total_time;
start.getProfilingInfo(CL_PROFILING_COMMAND_END, &time_start);
stop.getProfilingInfo(CL_PROFILING_COMMAND_START, &time_end);
total_time = time_end - time_start;

/* Results */
cout << "Execution time in milliseconds " << total_time / (float)10e6 / NUMBER_OF_ITERATIONS << endl;

Answer 1

I think your approach should work just fine (is it not?). Alternately, if you want to time each call, you can pass an event to enqueueNDRangeKernel and call getProfilingInfo on that enqueueNDRangeKernel.

cl::Event evt;
err = queue.enqueueNDRangeKernel(kernel, cl::NullRange, cl::NDRange(512), cl::NullRange, NULL, &evt);
evt.wait();
elapsed += evt.getProfilingInfo<CL_PROFILING_COMMAND_END>() -
            evt.getProfilingInfo<CL_PROFILING_COMMAND_START>();