|porting gcc firstname.lastname@example.org (2000-09-15)|
|Re: porting gcc email@example.com (Jeff Sturm) (2000-09-17)|
|Re: porting gcc firstname.lastname@example.org (Michael A. Sewell) (2000-09-17)|
|Re: porting gcc email@example.com (2000-09-17)|
|Re: porting gcc firstname.lastname@example.org (2000-09-28)|
|Re: porting gcc email@example.com (Tom Payne) (2000-10-08)|
|Date:||17 Sep 2000 23:01:58 -0400|
|Organization:||Mailgate.ORG Server - http://www.Mailgate.ORG|
> I hope I will be presented int n=*p++; in some abstract
> "pointer-plusplus"-token/struct, not in a "risc"
> MOVB R0,[R1]; ADD R1,4 ?
No, you are presented with elementary instructions. Then as far as I
remember the machine description can:
- use bottom-up (yacc like) algorithms to combine instructions and
- expand instructions (or sequences like above) to simpler, even more
RISCy sequences (e.g. x = ~x --> x = x ^ 0xFFFFFFFF)
- do peephole optimizations on the generated instructions.
> I guess what I am trying to ask is whether the backend programmer got
> full abstract C type information.
I don't know exactly, but there must be something because the compiler
does do alias analysis.
> - actually I want g++. assuming a g++ port, is the result any use for
> porting gcc? are they "RTL-compatible" ?
Yes, RTL is universal.
> and the "how to port gcc"
Search "Porting GCC for dummies" (not joking). A much more complicated
(and complete) guide can be found on any linux machine (`info gcc').
> please advice URLs.
Download gcc-core-2.95.2 (or something like that) from any GNU mirror.
Not really lean -- about 8 Megs. The 68000 machine description is
actually complex, better start with something like the PDP.
Return to the
Search the comp.compilers archives again.