This is the mail archive of the gcc-help@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Problems with new architecture


James Dessart <james at skwirl dot ca> writes:

> Hi, I'm busy setting up a new architecture for GCC, the m6809, running on
> the TRS-80 Color Computer series.  It'll be a cross-compiler, and I'm
> basing it on gcc 3.1.1.  However, despite it working on most files, it
> seems to choke at the end of long functions with 'unrecognizable insn'
> errors.  One of these files works when I remove -O2 compilation, but
> despite my best efforts, the other will not compile.
> 
> I know the first file, available at my site
> http://skwirl.ca/coco/gcc.html in libc-coco, named doprnt.c, worked for
> someone else running these config files under 2.95.3.  I had to make
> modifications to have it compile under 3.1.1, but I'm not sure I've fixed
> everything up properly.  The ChangeLogs are hard to go through, and it's
> almost impossible to find something unless you know specifically what it
> is you're looking for.  I managed to fix a couple of problems this way.
> 
> So, here's the unrecoverable error, it's from a file that comes from
> uip-0.6.0, a small TCP/IP stack:
> 
> [james at loki:coco] % make uip.o
> gcc09 -I../uip -I../apps/httpd -I. -fno-inline -c -o  uip.o
> ../uip/uip.c
> ../uip/uip.c: In function `uip_process':
> ../uip/uip.c:1023: unrecognizable insn:
> (insn 1898 1597 1896 (set (reg:QI 0 d)
>         (subreg:QI (mem:HI (plus:HI (reg/f:HI 3 u)
>                     (const_int -13 [0xfffffff3])) [0 S2 A8]) 1)) -1 (nil)
>     (nil))
> 
> Now, I'm not sure what's wrong here... u is a stack register, not used by
> gcc, as far as I know, for passing parameters.  Perhaps it's used for
> local storage, I'm not sure.  Is it possible that this is as a result of
> having unsupported local storage in the function?
> 
> Thanks for any help you can provide,
> James

I'm begining to think people with an in-depth knowledge of gcc's
    source do not post here. I suggest you try gcc at gcc dot gnu dot org, which is
    for the maintainers.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]