Help: the overflow issue about fixed point data computing

wqyhandsome

Intellectual 520 points

Other Parts Discussed in Thread: TMS320C6416T

I am a fresher and hop to get your comments.

1. array1 x[i]/ ①value range:[0~1)/ ②32bit Q31 int

2. array1 y[i]/ ①value range:[0~1)/ ②16bit Q15 short int

3. array3 z[i]/ ①z[i]=x[i]*y[i] /②z[i],32bit Q31 int

I rewrite the z[i]=x[i]*y[i] in C language:

z[i]=x[i]*y[i]>>15;

the overflow happened at times in the elements of the array z. Could you share some comments to fix the issue.

P.S. TMS320C6416T,CCS3.3v

Thanks and Best Regards

over 15 years ago

0 wqyhandsome over 15 years ago

Intellectual 520 points

I think I have fixed the issue.

Solution: en-large the interger part of the data (enhance the precision), and divided the data into integer part and the fraction part; and then do the fixed-point computing...

Maybe someone have better ideas, hope to get suggestions.

thanks for your attention.

0 Yair Hougui over 15 years ago in reply to wqyhandsome

TI__Expert 5615 points

See MPYLIR instruction http://focus.ti.com/lit/ug/spru732j/spru732j.pdf.

To my understanding this is what you are looking for.

Yair

0 wqyhandsome over 15 years ago in reply to Yair Hougui

Intellectual 520 points

Hi Yair,

Thanks for your information.

Best Regards

0 wqyhandsome over 15 years ago in reply to wqyhandsome

Intellectual 520 points

update/F.Y.I

The long long (int) data type is also good in this case when using the C language

Best Regards

Processors

Processors forum

Help: the overflow issue about fixed point data computing