• State TI Thinks Resolved
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 99 subscribers
  • Views 409 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

Why dm648 runs slower than dm642?

James Chen Z
Intellectual 280 points

Hi,

we are going to use dm648 to replace dm642,but when testing arithmetic we found that dm648 using about 25 millisecond while dm642 use less than 10 millisecond .

i think maybe L2 CACHE is not woking, so i move all data and code segment to sram , but still use more than 20 millisecond .

i checked the execution graph , and found KNL_swi and PRD_swi used most of cpu time, while my arithmetic task used a little time . but i dont know how much i can trust it.

though dm642 can run the arithmetic in 10 millisecond , i think dm648 should not need more time . so what can i do ?

thanks for your help

James

 

  • 0
    TI__Mastermind 18320 points

    Although this post mentioned BIOS, I think it is a broader question about DM648 performance, so I will move it accordingly from the BIOS forum to the DM64x forum..

  • 0
    Expert 1575 points

    Hi James,

    do you know that you have to setup a lot more things on DM648 than on DM642? For example the core PLL - the processor runs really slow unless you configure the PLL multiplicators.Did you configure the PLLs for DSP core and DDR Ram?A fast test could be to write a PRD function that increments a global variable every ms. Run your app for 10 seconds and you'll see if it has counted up to 10000. If it is somewere around 500, your clock is not setup correctly.You have to enable more or less every hardware unit you use of DM648 manually ("PSC modules", PINMUX...). Its quite a confusing device to start when you know the DM642 (which was quite a confusing device when coming from the C6414... ;-)

    bye,

    Thomas

  • 0 in reply to Thomas9070
    Intellectual 280 points

    thank you.

    i found the reason. the time we use is related to the data source. when i load a real graph data, instead a undefined buffer, it works well.