C671x float multiplication results in rounded off decimal value

Jaiganesh Jayaraman76473

Hi all,

I am working on C671x DSP core and the following operation results in rounded off decimal value though the corresponding hex value is holding the precision.

y = m * x;

here,

m= 0.02( 0x3CA3D70A )

x = 9999.999 ( 0x461C3FFF)

Expected output: 199.99998, but the actual output observed in watch window is 200.0. But the hex value is 0x4347FFE and this corresponds to 199.99997. Why this rounding off to 200.0 happens and is there a way to prevent it?

Thank you

Jai

over 9 years ago

0 Rahul Prabhu over 9 years ago

TI__Guru** 116170 points

Please refer to the following wiki for some suggestions on modifying the compiler settings that may help in your evaluation.
processors.wiki.ti.com/.../Floating_Point_Optimization

Regards,
Rahul

0 Jaiganesh Jayaraman76473 over 9 years ago in reply to Rahul Prabhu

Prodigy 250 points

Thanks Rahul for the reply. I checked the compiler option is set to -mv6710. From wiki, I could see --fp_mode=strict can be used for IEEE-754 semantics. But, I don't see this option available for v7.4 compiler and i don't want to use --strict_ansi option at this point.

As I mentioned in previous post, the raw hex value holds the precision. Is the rounding off while converting to decimal intended by this compiler characteristics? Also, I verified the output using gcc compiler and I could see the result as 199.999969.

Please suggest if there is any other compiler option I can check.

Thank you,

Jai

Processors

Processors forum

C671x float multiplication results in rounded off decimal value