• State Resolved
  • Locked Locked
  • Replies 2 replies
  • Subscribers 98 subscribers
  • Views 953 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.

DSPLIB FFT Problem

Tunc Alkanat
Prodigy 210 points


Hi,

I am trying to implement a code that takes the FFT of the input on EVM6678 using CCSv5.4 and DSPLIB 3.1.0.0. I did a similar work for the c67x series and it performed smoothly using DSPLIB v200 but when I try to directly run that code on C66x, I have noticed that DSPF_sp_SPxSP function does not perform as before. I am pretty sure that when given the same inputs, this function performs differently for DSPLIB versions and using MATLAB, it turns out that the result found using the latest version of DSPLIB and C66x was wrong where my initial configuration performs OK. Is this a known issue? Did TI revised the above mentioned function? If neither, can anyone inform me about the possible causes of this incompatibility?

Current system:

CCSv5.4

bios_6_35_01_29

dsplib_c66x_3_1_0_0

mcsdk_2_01_02_06

xdctools_3_25_00_48

Thanks in advance.

Regards,

Tunc.

  • 0
    TI__Intellectual 1870 points

    Hi Tunc,

    The DSPLIB on C66 are optimized for C66x core.  They have better performance but the results should be the same.  Coud you try the examples that come with DSPLIB first?

    Xiaohui

  • 0 in reply to Xiaohui Li
    Prodigy 210 points

    Hi, 

    I have managed to solve the problem. Just so you know for further questions, it turns out that the very same named function, DSPF_sp_SPxSP, is implemented differently in DSPLIB 2.0 and 3.1 versions. The different part is the code below the compiler directive,

    #if _LITTLE_ENDIAN

     line in the implementation in DSPLIB 3.1.0.0. It  is different for the two versions, actually the 2.0 version do not include such directive, which forced me to change my twiddle generator code in order to obtain the same results. The si1 to si3 and co1 to co3 variables are assigned differently. You might want to take a look.  Thank you for your response, suggestions that would further optimize the way I solve the problem is greatly appreciated given the way I solved it.

    Regards,

    Tunc.