• State Resolved
  • Locked Locked
  • Replies 2 replies
  • Answers 1 answer
  • Subscribers 98 subscribers
  • Views 682 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.

TMS320C6678: how to evaluate the precision of RSQRSP when calculate the Modulus of complex

sworder cool
Prodigy 60 points
Part Number: TMS320C6678

hi everyone ,

how to evaluate the precision of RSQRSP when calculate the Modulus of complex?

is it 2^(-8)?

after i used RSQRSP , I also use  Newton-Raphson method to get more accurate result，in this situation, how to evalute the presion?

i will appreciate it if anyone can answer me, this is very important to me 

thanks a lot 

  • 0
    TI__Guru**** 161600 points
    Hi,

    Have you checked the Optimizing Loops on the C66x DSP user guide (www.ti.com/.../sprabg7.pdf), section 3.1.4 One C64x+ Optimization & 3.1.5 Usage of Division Instructions (?). Section 3.1.5 mentions:
    "An important property of RSQRSP and RCPSP (and also of double-precision-type RSQRDP and RCPDP) is that these instructions provide the correct exponent, and the mantissa is accurate to the eighth binary position (therefore, mantissa error is less than 2^(-8)). To get higher precision results, Newton-Raphson interpolation is needed.
    For example, for RCPxP instructions to calculate 1/v, one Newton-Raphson interpolation x[n+1]=x[n]*(2 - v *x[n]) can improve the mantissa precision to 2^(-16), and one additional interpolation can improve the mantissa precision to 2^(-24) for single precision and to 2^(-32) for double-precision.

    Similarly for RSQRxP instructions to calculate 1/sqrt(v), one Newton-Raphson interpolation x[n+1]= x[n]*(1.5 - (v/2)*x[n]*x[n]) can improve the mantissa precision to 2^(-16), and one additional interpolation can improve the mantissa precision to 2^(-24) for single precision and to 2^(-32) for double-precision."

    Best Regards,
    Yordan