How do I know my OpenCL kernel is running on GPU?

Question

How can I be 100% sure that my opencl kernel are actually running on the GPU and not on the CPU. I am not able to understand this because openCL kernel can also run on CPU. Any pointers here?

Answer 1

You have to choose between platforms and devices available on your computer when you create the OpenCL context. You can get some informations about platforms and devices with clGetPlatformInfo and clGetDeviceInfo. You can find samples codes in NVIDIA and AMD SDK to list platforms and devices.

For AMD (gDEBugger, Code XL), NVIDIA (Visual Profiler) and Intel (Intel GPA) devices, you can use profiling tools.

Answer 2

When creating your context, use clCreateContextFromType where you can pass the flag CL_DEVICE_TYPE_GPU, which specifies that you want GPU only. Something like this guarantees the use of a GPU (you can be 100% sure) :

cl_uint num_platforms = 0;
clGetPlatformIDs(0, NULL, &num_platforms);
cl_platform_id* platform = malloc(sizeof(cl_platform_id) * num_platforms);
cl_context_properties cprops[3] = { CL_CONTEXT_PLATFORM, (cl_context_properties)platform[0], 0 };
ctx = clCreateContextFromType(cprops, CL_DEVICE_TYPE_GPU, NULL, NULL, &err);

If you have already your context and devices set up, you can query the type of a device like so:

cl_device_type dev_type;
clGetDeviceInfo(my_device, CL_DEVICE_TYPE, sizeof(dev_type), &dev_type, NULL);
if (dev_type == CL_DEVICE_TYPE_GPU) {
    printf("I'm 100%% sure this device is a GPU");
}

Everything you will enqueue to a command queue created on this device will run on GPU.