cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

t_verstraete
Journeyman III

profiling response is always 0s

Jump to solution

Hey,

I am just starting with OpenCl programming and as a starting project I took an algorithm that is fairly simple: gaussian filter.

Now I wanted to test out the profiling function so:

1 i added profiling in the command queue

     commandQueue = clCreateCommandQueue(context, devices[0], CL_QUEUE_PROFILING_ENABLE, NULL);

2. made a call back function that can read and output the profile values:

void CL_CALLBACK eventCallback(cl_event ev, cl_int ev_status, void *user_data)

{

        int evID = (long) user_data;

    cl_int errNum;

    cl_ulong ev_start_time = (cl_ulong) 0;

    cl_ulong ev_stop_time = (cl_ulong) 0;

    size_t return_bytes;

    double run_time;

    printf("PROFILING: Event callback %d %d ", (int) ev_status, evID);

    //read back the command event queued counter value

        errNum = clGetEventProfilingInfo(ev, CL_PROFILING_COMMAND_QUEUED, sizeof(cl_ulong), &ev_start_time, &return_bytes);

    if (errNum!=CL_SUCCESS)

    {

        printf("ERROR: get CL_PROFILING_COMMAND_QUEUED did not succeed");

    }

    //read back the command event end counter value

    errNum = clGetEventProfilingInfo(ev, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &ev_stop_time, &return_bytes);

    if (errNum!=CL_SUCCESS)

    {

        printf("ERROR: get CL_PROFILING_COMMAND_END did not succeed");

    }

    //calculate the run time from start and stop time

    run_time = (double) (ev_stop_time - ev_stop_time);

    //output the result

        printf("\n Kernel run time %f secs \n", run_time * 1.0e-9);

}

3. add the event to the kernel to call the callback function:

errNum = clEnqueueNDRangeKernel(commandQueue, kernel1, 2, NULL, globalWorkSize, localWorkSize, 0, NULL, &filter_order_event);

errNum = clSetEventCallback(filter_order_event, CL_COMPLETE, &eventCallback,(void *)ID);

Now i keep on getting 0s as a result and no actual values ... even though i see the GFX working for some time on the kernel. i also added clfinish(commandQueue) but same result. Am i missing something?

Thanks for helping,

kind regards,

Tim

0 Likes
1 Solution

Accepted Solutions
t_verstraete
Journeyman III

Re: profiling response is always 0s

Jump to solution

btw I have found the issue:

run_time = (double) (ev_stop_time - ev_stop_time);

should become

run_time = (double) (ev_stop_time - ev_start_time);

very stupid mistake 😉 ...

View solution in original post

0 Likes
1 Reply
t_verstraete
Journeyman III

Re: profiling response is always 0s

Jump to solution

btw I have found the issue:

run_time = (double) (ev_stop_time - ev_stop_time);

should become

run_time = (double) (ev_stop_time - ev_start_time);

very stupid mistake 😉 ...

View solution in original post

0 Likes