This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] Aliasing fixes, take 3
- From: Jan Hubicka <hubicka at ucw dot cz>
- To: Diego Novillo <dnovillo at redhat dot com>
- Cc: Jan Hubicka <hubicka at ucw dot cz>,"gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>,Richard Henderson <rth at redhat dot com>, Jeff Law <law at redhat dot com>
- Date: Sat, 22 Nov 2003 18:18:37 +0100
- Subject: Re: [tree-ssa] Aliasing fixes, take 3
- References: <20031122141228.GE8644@kam.mff.cuni.cz> <1069520382.6132.186.camel@frodo.toronto.redhat.com>
> On Sat, 2003-11-22 at 09:12, Jan Hubicka wrote:
>
> > Per Jeff's request, I've re-tested my patches with i686-pc-linux-gnu, instead
> > of i386-linux.
> >
> There shouldn't be much of a difference. How is it that you were
> getting no regressions on i386-linux? The branch is not completely
> target-independent, but it's odd for two almost identical targets to
> work differently.
It looks more like difference in libc. We had random memory overflow in
parser.c caused by wrong aliasing. (missing TREE_ADDRESABLE bit after
repated inlining caused dominator1 pass to do wrong changes before
address has been completely elliminated)
This caused context to be sometimes wrong and parser didn't find
function declarations resultins in random failures during libstdc++
build. This is what caused regressions with my tail calls I wasn't able
to reproduce and vice versa. This is fixed
gone by this patch, I now get same results for i386-linux on Debian
machine and i686-linux on SuSE and tese match RH results reported to
lists.
>
> > This ineed makes clean tree to bootstrap. Additionally I've
> > also bootstrapped i386-linux and x86-64-linux (without libjava).
> >
> Why no libjava? It should work.
It works now, I simply didn't re-tested and my script still contained
hack to disable libjava. I did run full test in meantime with following
results:
Differences:
46,47d45
< FAIL: gcc.c-torture/execute/20010403-1.c execution, -O3 -fomit-frame-pointer
< FAIL: gcc.c-torture/execute/20010403-1.c execution, -O3 -g
157,158c155,156
< # of expected passes 24054
< # of unexpected failures 74
---
> # of expected passes 24056
> # of unexpected failures 72
715c713
< FAIL: Thread_Monitor output - gij test
---
> FAIL: Thread_Monitor execution - gij test
735a734
> WARNING: program timed out.
1205c1204
< # of expected passes 2218
---
> # of expected passes 2217
1208c1207
< # of untested testcases 61
---
> # of untested testcases 62
1328a1328
> FAIL: 24_iterators/iterator.cc execution test
1334,1335c1334,1335
< # of expected passes 2281
< # of unexpected failures 2
---
> # of expected passes 2280
> # of unexpected failures 3
The extra timeout looks more like an random noise.
Many libjava tests takes ages to fail, rerunning the tests brings
differences in different tests, so I hope I can interpret this as pass.
Honza