Question:

I am trying to find the bottleneck in my OpenCL kernel, is it possible to profile OpenCL programms on mac os X? I found gDebugger on <a href="http://www.gremedy.com/" rel="nofollow">http://www.gremedy.com/</a>, but it requires 10.5 or 10.6 to run. AMD SDK supports only Linux and Windows.

Is there a profiler for Mountain Lion?

How detailed must your profiling information be? Is it okay to use the built-in internal profiler?<br /> OpenCL queues can be created with the CL_QUEUE_PROFILING_ENABLE flag.

This way you can see for each kernel you executed:<br /> When it has been

<ul><li>Enqueued</li> <li>Submitted to you OCL-Device</li> <li>Started</li> <li>Ended</li> </ul>

With <strong>C++-Bindings</strong>, the creation of the queue can look like this:

<pre class="lang-cpp prettyprint-override">_queue = new cl::CommandQueue(_context, _device, CL_QUEUE_PROFILING_ENABLE );

The extration of the profiling information looks like this:

1) Save the event object (in an array) delivered by the enqueued kernel you want to profile.

<pre class="lang-cpp prettyprint-override">cl::Event evt; _queue->enqueueNDRangeKernel( _kernel, cl::NullRange, _range, cl::NullRange, NULL, &evt);

2) After execution of the queue, extract the profiling information

<pre class="lang-cpp prettyprint-override">std::vector<cl::Event> evts; //add all events to this vector here //cl::Event evt; //_queue->enqueueNDRangeKernel( _kernel, cl::NullRange, _range, cl::NullRange, NULL, &evt); //evts.push_back(evt); uint64_t param; for (unsigned int i=0; i<evts.size(); i++) { evts[i].getProfilingInfo(CL_PROFILING_COMMAND_QUEUED, &param); printf("%u: %llu", i, param); evts[i].getProfilingInfo(CL_PROFILING_COMMAND_SUBMIT, &param); printf(" %llu", param); evts[i].getProfilingInfo(CL_PROFILING_COMMAND_START, &param); printf(" %llu", param); evts[i].getProfilingInfo(CL_PROFILING_COMMAND_END, &param); printf(" %llu\n", param); }