[avr-gcc-list] Re: binutils/gas patch for ldi adiw ldd

[Klaus Rudolph]

For that problem I have also looked but this goes a bit deeper.
I have found a #define where it could be allowed to use unknown sections 
diffs.
This works and in the relocation stage I have a RELOC with the positiv
value but th negativ value is missing. Inbetween the stages I can see 
both symbols with
value 0 which is correct because the final link was not done until then.
I have no idea where the neagtiv value ist stored? Maybe it is stored 
nowhere and thats it.
As I saw there are only a few targets which support the undefined diffs 
and currently
it is for me much more importend to hint bugs instead of features :-)))
But maybe I have overseen one or two lines which hide the result...

We will see.

Regards,

  Klaus

> On Sat, Nov 20, 2004 at 01:49:13PM +0100, Klaus Rudolph wrote:
>  
>
> > P.S. I have no idea why offsets could not be created from "unknown"
> > sections. Here the gas is allways giveing a error message. that makes
> > no sense I think because the relocation is done later. Bug or feature?
> > Any idea how to fix it?
> >    
>
> Klaus,
>
>   A brief experiment with gas for the PowerPC shows that this is
> another avr-as "feature". PowerPC code which assembles fine in .text
> also gives no problems in section .frog:
>
>   .section ".sbss","a"
>   .align 2 
> alpha:
>   .skip 4
> beta: 
>  
>   .section ".frog","arx"
>  
> gamma:
>   stw   r4,beta - alpha(r7)        # Sorry, this is PowerPC assembler,
>   stw   r5,delta - gamma(r7)       # shown only to prove that "full
> delta:                              # pressure" gas can do it.
>
> This correctly produces:
>
> Disassembly of section .frog: 
>
> 00000000 <gamma>: /------------/------< Correct offsets of 4 and 8.
>   0: 90 87 00 04    stw   r4,4(r7)
>   4: 90 a7 00 08    stw   r5,8(r7) 
>
> I'll google for that GNU Internals Manual that you mentioned, but maybe
> assisting with the testsuite is a shorter term goal.
>
> Erik
>
>
>
>