From:	Andreas <andreas-e AT foni DOT net>
To:	pgcc AT delorie DOT com
Date:	Sun, 26 Mar 2000 18:33:08 +0200
X-Mailer:	KMail [version 1.0.28]
MIME-Version:	1.0
Message-Id:	<00032618333700.00583@localhost.localdomain>
X-MIME-Autoconverted:	from quoted-printable to 8bit by delorie.com id OAA04161
Reply-To:	pgcc AT delorie DOT com

Hi, there,

some time ago, I posted a bug report concerning the mmx flag of the pgcc
compiler. As I did not receive much feedback, I would like to post my
problem again, maybe some of the developers might come back to me. From a
project of which I received only the binaries, I remember that fine tuning
some of the code with respect to mmx instructions (maybe manually) resulted
in a speed-up of 100%, so it might well be worth spending some time with mmx.

I am trying to compile a project of mine with the cvs snapshot of a few days
ago, using the flags "-march=pentium -mmx". For several of my c files, I get
error messages like the following one:

gcc  vinci_screen.c -c -march=pentium -mmx
vinci_screen.c: In function `print_hyperplanes':
vinci_screen.c:41: internal error--insn does not satisfy its constraints:
(insn 183 128 186 (set (reg:SI 17 %mm0)
        (reg:SI 0 %eax)) 48 {movsi+2} (nil)
    (nil))

Even worse, compiling only these files without the mmx-flag and linking all
files together results in a binary which does not crash, but which does not
produce the desired result either.

If desired, I can post my incriminated files again.

Best regards,

Andreas


PS: The version descriptor string does not seem to be changed in the cvs
versions; "gcc -v" still returns
Reading specs from /usr/local/lib/gcc-lib/i586-pc-linux-gnu/pgcc-2.95.2/specs 
gcc version pgcc-2.95.2 19991024 (release)
Could this be changed, maybe, to reflect the correct installed version?

- Raw text -