Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
8 years ago

Unexplained performance difference for same kernels

I have develop two identical kernels, with a single difference where the size of the for loop in one of them is higher than the other. Here are my two kernels, Kernel# 1 and Kernel# 2. First Ke...
Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
8 years ago

Here is the part of the code where deploys the opencl kernel onto the device: 

   Event evKernel (alg.name);
   err = clEnqueueNDRangeKernel (queue, kernel, 1, NULL,
                                                         globalWorkSize, localWorkSize,
                                                         0, NULL, &evKernel.CLEvent());
   CL_CHECK_ERROR (err);
   err = clWaitForEvents (1, &evKernel.CLEvent());
   CL_CHECK_ERROR (err);

I feel like I already wait for the kernel to finish. So, do I still need clFlush and clFinish?

BTW, I have fixed the issue you have mentioned earlier, and I've seen the effect on the GPU at least. The performance has dropped which seems to be reasonable. I will go on and try them on the FPGA too and see how it'll behave.