An industrial smart camera are developing in our company. There are some problems encountered by us.
The sobel algorithms take almost the same time when scalar computing and vector computing( int4 or float4) with OpenCL are implemented respectively on our SabreLite board (kernel 3.05).
We believe that it is impossible. Did you test the efficiency with OpenCL on I.MX6Q? if so, can you tell us where demos can be find?
Thank you for your time.