Hi,
Does any tool to profile OpenCL on R-car H3 ?
Because I need to know the OpenCL is whether to reach the operational limit.
Thank for your help.
boyu
Hi Boyu,
I don’t fully understand your question. Can you clarify what you are trying to do?
You can use clinfo on the device to check the OpenCL options available on the board.
You can use computecpp_info to check the SYCL options available on the board.
You may be able to use some of the tools described here to profile your code on the board https://developer.codeplay.com/products/computecpp/ce/guides/sycl-guide/debugging
Hi rod,
I am sorry that I did not explain clearly.
Because I currently need to execute OpenCL program on R-Car H3, but it is only one ms time for one kernel running time to multiply two 256*256 matrices (not including the data from the host side to the gpu end). Because the kernel is just a simple multiplication of the matrix, it will ask if it is a valid evaluation tool.
In addition, is there a neural network architecture similar to yolo on R-Car H3? The opencl sdk provided by Renesas has caffe face recognition, but the performance is very poor.
Thank your help!
boyu
Hi Boyu,
Are you running the kernel we provide with the release? The example we provide is simply to test the environment set up is working, this is not an optimized algorithm and in fact running a trivial 256x256 matrix multiply on any hardware the performance will result in slow execution. Tiling is a technique that can be used to get better performance.
I assume the OpenCL SDK you are referring to is for the PowerVR GPU? I’m afraid we don’t know much about this since we provide OpenCL support for the Renesas CVengine processors.
In terms of neural networks, the V3M and V3H boards are more suitable for running these sorts of applications.