View Full Version : profiling

09-09-2009, 03:57 AM
I use the command clGetEventProfilingInfo with CL_PROFILING_COMMAND_END and CL_PROFILING_COMMAND_START to get the time by the kernel on the gpu device or the memory transfer ?
But how can I measure the consumed time by the command clEnqueueNDRangeKernel on the CPU and GPU ?

Should I add the user time CPU (e.g. returned by getrusage) and the time returned by clGetEventProfilingInfo for the event associated with clEnqueueNDRangeKernel?


09-09-2009, 02:15 PM
You can use clGetEventProfilingInfo with CL_PROFILING_COMMAND_QUEUED and CL_PROFILING_COMMAND_SUBMIT to find out delta between when command was enqueued by the application and when the command was submitted to the device. Similarly, the delta between CL_PROFILING_COMMAND_SUBMIT and CL_PROFILING_COMMAND_START will tell you the time delta between when the command was submitted to the device and when it actually started executing on the device.