Re: benchmark between aranym and crest
On Fri, Oct 13, 2006 at 02:23:52PM +0200, Petr Stehlik wrote:
> The gcc-4.1.1-14 does not compile anything. It always crashes here
> (strace output):
>
> 7314 open("/tmp/ccoNE7AY.s", O_RDWR|O_CREAT|O_TRUNC, 0666) = 3
> 7314 fstat64(3, {st_mode=S_IFREG|0600, st_size=0, ...}) = 0
> 7314 old_mmap(NULL, 8192, PROT_READ|PROT_WRITE,
> MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xc033a000
> 7314 open("a.gcda", O_RDWR) = -1 ENOENT (No such file or
> directory)
> 7314 --- SIGILL (Illegal instruction) @ 0 (0) ---
This actually occurs in /usr/lib/gcc/.../cc1.
> g++ does this:
>
> 7390 open("/tmp/ccgruVK5.s", O_RDWR|O_CREAT|O_TRUNC, 0666) = 3
> 7390 fstat64(3, {st_mode=S_IFREG|0600, st_size=0, ...}) = 0
> 7390 old_mmap(NULL, 8192, PROT_READ|PROT_WRITE,
> MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xc033a000
> 7390 old_mmap(NULL, 16384, PROT_READ|PROT_WRITE,
> MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xc033c000
> 7390 open("a.gcda", O_RDWR) = -1 ENOENT (No such file or
> directory)
> 7390 old_mmap(NULL, 262144, PROT_READ|PROT_WRITE,
> MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xc0340000
> 7390 write(3, "#NO_APP\n\t.file\t\"a.c\"\n\t.text\n\t.al"..., 298) = 298
> 7390 close(3) = 0
>
>
> So g++ before opening the "a.gcda" file calls one more mmap than the gcc
> before the SIGSEGV. Does it mean anything? I don't feel like digging in
> gcc source code.
It much probably means g++ malloc() more memory than gcc. This usage of
mmap is typical of glibc malloc.
> I hope this problem appears on real m68k hardware - perhaps it's kernel
> or something else that's different than on crest and could cause this.
Well by any chance, could you compile a new kernel with both nfeth and
NFS support ? Crest run 2.4.33.
Cheers,
--
Bill. <ballombe@debian.org>
Imagine a large red swirl here.
Reply to: