This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

how to improve performance

Other Parts Discussed in Thread: OMAPL138, OMAP-L138

hello, engineers
now i incouter a touch issue. i want running my CNN classifier for character recognition on omapl138 as fast as it running on PC. my code is

written in C++, and i have tried everything i known. now i adhere my code for you, please give me some advices for improving the performance,

thank you all! my ccs version is 3.3.82, now in simulator it costs about 30ms to recognize a character while on PC only 3ms.

best wishes

eric3757.CNN520.rar

 

  • Eric,

      Also - you have to consider that your PC may be running at a much high clock frequency than the TI device. Typically modern PC operate in the GHz range, the OMAP-L138 only operates up to 456 MHz. If you are comparing a PC to an OMAP-L138 you must also factor in CPU architecture differences, cache differences, power consumption, clock frequency, etc.

    Aside from this, there are many optimization techniques that can be used with the C6000 Compilers to increase efficiency. Please review the wiki to determine if there are any additional techniques that you can take advantage of with your code.

  • thanks for replying, Drew.

    since there are many differences between my PC and OMAP-L138, now i want to know is it possible that my algorithm could run on OMAP-L138 (actrually on DSP only)as fast as on PC through optimization techniques??? 

    thanks,

    eric

  • There is no guarantee due to the above differences, because the performance comparison you are trying to achieve is dependent on too many variables.

     

    You can use the Simulator in CCS to determine the theoretical speed of execution on the DSP given your the configuration of the device.