Re: [avr-gcc-list] 32 bit double precision

[Rolf Ebert]

On Tue, 21 Jun 2005 19:13:20 -0700, intiha Ho gai <address@hidden> wrote:

> Hi,
> I am planning to do floating point calculations. I tried to look at
> the precision of doulbe on the avr-gcc manual, but other than that it
> is 32 bit I couldnt find the number of bits for the mantissa/exponent.
> Also, is there a good tutorial on optimizing performance under such
> precision constraints ( I know 32 bit are not doing the job for me
> directly :( ).

32 bit floats have 23 bits mantissa, which corresponds roughly to 6  
decimal digits.

One is usually spoiled by the 15 decimal digits of doubles on desktop  
computers.  I wonder which algorithm that you want to run on an AVR  
depends on 15 digits.  Most algorithms work quite well in 6 digits.

    Rolf