Re: [avr-libc-dev] Double precision floating point routines

[Martin Cibulski]

>> > As Jörg Wunsch wrote:
>> >
>> > It would probably be very nice to combine all this into the
>> > compiler/library, but I believe that is not possible using the
>> > outlined stack engine.
>> > -- 
>>
>> I could modify my routines to use a stack in memory
>
>My guess would be that, to include these routines, you would have to use
>the normal data stack, just like other subroutines in the library. Even if
>it uses a lot.

>Would this be possible for you to do?

Eric,

it would be possible to use a separate data stack.
Can you please explain a little more how it works ?
I read that the IRC compiler uses a data stack pointed
by the Y register. Does AVRGCC do the same ?
If yes, how is it used (post/pre increment/decrement, which pointer
register) ?

How are parameters passed in AVRGCC ?
In the mean time I looked at the documentation of
the library and the GCC config files (for compilation of GCC itself, I
think).
It seems to me that GCC generates calls like
'result = add (parameter1, parameter2)'
and combines these to calculate more complex expressions.

In the sourcecode of the single precision math routines
I found both parameters passed in registers, not on the stack.
Would AVRGCC pass 8 byte wide double parameters in registers too ?
Perhaps you or someone else could do a little test
 with simulated 8 byte wide doubles like the following

typedef struct {
    char c[8];
} DOUBLE;

DOUBLE test (DOUBLE a, DOUBLE b)
{
    return a;
}

DOUBLE b;    // global

main()
{
    DOUBLE b, c;    // local

    c = test (a, b);
}

The generated assembler code might
answer some of my questions.

Thanks for help
Martin