avr-gcc-list
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

AW: [avr-gcc-list] How often does your code not fit?


From: Haase Bjoern (PT-BEU/EMT) *
Subject: AW: [avr-gcc-list] How often does your code not fit?
Date: Mon, 11 Jul 2005 09:27:37 +0200

I have analyzed for quite a while how to reduce code size with gcc. Present 
status is that gcc 4.0.0 with -Os -mcall-prologues is best. It yields about 4 % 
smaller results than 3.4.4 for my applications. 

Issues for further reducing code size for gcc, IMO, are 
1.) Reducing register pressure by lowering Simode and Himode subregs at expand.
2.) Adding support for multiple memory spaces (there has been posted a method 
recently that, however, presently is not integrated in gcc mainline.)

Possibly one could also try to reconfigure the linker so that it could choose 
rjmp/rcall out/ld if at link time it comes out that the smaller alternative 
could be used.

Yours,

Björn 

-----Urspr√ľngliche Nachricht-----
Von: address@hidden [mailto:address@hidden Im Auftrag von Keith Gudger
Gesendet: Samstag, 9. Juli 2005 16:59
An: Ben L. Titzer
Cc: AVR GCC List
Betreff: Re: [avr-gcc-list] How often does your code not fit?

Ben:

Regarding space optimization, this is one area I've been meaning to
investigate.  I posted a message about how IAR does so much better on my
code, and here is a response I've been saving (I hope Neil Johnson doesn't
mind my reposting his reply):

> The space optimization tricks that IAR does have seems like an
> interesting thread to follow. Since avr-gcc and Atmel's AVRs are here to
> stay, it certainly would pay off to get those integrated.
                                                                                
Do a google on "Procedural Abstraction" and "Cross Linking", or hunt out
papers at citeseer.com.
                                                                                
There is a standalone tool (aiPop) that does quite aggressive post-link
optimization on C16x/ST10 code:
                                                                                
        http://www.absint.com
                                                                                
where they use the term "functional abstraction".
                                                                                
I'm not sure how well proc.abs has yet been implemented in GCC.  There are
several goes in academic circles, but so far as I know nothing in the main
GCC distribution.
                                                                                
Neil
--
Neil Johnson :: Computer Laboratory :: University of Cambridge ::
http://www.njohnson.co.uk          http://www.cl.cam.ac.uk/~nej22
----  IEE Cambridge Branch: http://www.iee-cambridge.org.uk  ----
                                                                                
Keith Gudger

On Fri, 8 Jul 2005, Ben L. Titzer wrote:


  [NON-Text Body part not included]



_______________________________________________
AVR-GCC-list mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/avr-gcc-list







reply via email to

[Prev in Thread] Current Thread [Next in Thread]