I have test some examples of SDK 3.2 on both 835 and 820( both are Open-Q board), and get the lower performance of 835 than that of 820.
# this is result from 820
calling gaussian7x7_Gaussian7x7u8 on a 1920x1080 image...
run time of gaussian7x7_Gaussian7x7u8: 198342 microseconds for 100 iterations
return value from gaussian7x7_Gaussian7x7u8: 0
calling dspCV_deinitQ6()...
return value from dspCV_deinitQ6(): 0
- success
# this is result from 835
calling gaussian7x7_Gaussian7x7u8 on a 1920x1080 image...
run time of gaussian7x7_Gaussian7x7u8: 329994 microseconds for 100 iterations
return value from gaussian7x7_Gaussian7x7u8: 0
calling dspCV_deinitQ6()...
return value from dspCV_deinitQ6(): 0
- success
I'm dault the performance of Hexagon 682, is that normal?