Re: Optimization around mul and restoring r1

On 02/10/13 09:34, Oleksandr Redchuk wrote:
> 2013/10/2 Georg-Johann Lay <[email protected] <mailto:[email protected]>>
> 
>     David Brown schrieb:
> 
>         In general, it would be a good thing to delay the zeroing of r1,
>         and it
>         is worth searching the issue database for previous issues and
>         filing a
>         new "missed optimisation" issue if it is not there already:
> 
> 
>     IMHO everyone working on the avr back-end is aware of this problem.
> 
> 
> May be, reassigning __zero_reg__ to R2 and using R1:R0 as
> __temp_reg_hi__:__temp_reg__ will be better but I can't estimate both
> complexity of changes and code size/perfomance gain effect.
> 

I believe this has been discussed before, and the conclusion was that
this would have been a good idea if it had come in the early days of
avr-gcc, but it is not worth breaking the established ABI.  I think the
changes in gcc would be relatively simple - but the changes in the
support libraries in gcc and avr-libc would be much greater.  And of
course anyone with assembly code that works along with avr-gcc will be
in trouble too.

mvh.,

David