This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: State of gcc 3.3
On Sat, 10 Jan 2004, Gabriel Dos Reis wrote:
> Richard Guenther <rguenth@tat.physik.uni-tuebingen.de> writes:
>
> | Hi!
> |
> | The 3.3 branch seems to be in a quite bad state now, as I see several
> | pooma regressions compared to snapshots from november:
> |
> | -PASSED ... array_test1
> | -PASSED ... array_test7
> | -PASSED ... array_test21
> | -PASSED ... array_test27
> | -PASSED ... array_test28
> | +PASSED ... indirectionlist_test1
> | -PASSED ... ump_test5: guard cell fill test.
> | -PASSED ... ump_test7
> | -PASSED ... brickviewbase_test3
> | +FAILED ... brickviewbase_test3
> | -PASSED ... Gradient
> | -PASSED ... FileSetWriter
> | -PASSED ... FileSetWriterTest2
> | -PASSED ... ReduceOverContexts Test
> | +PASSED ... CollectFromContextsTest
> |
> | Standard checking options include -O2 -funroll-loops -g
>
> That is odd.
The failures all look very similar, all these are segfaults.
> | Anyone interested?
>
> I know of a C++ patch that I backported from mailine and that passed
> the regression test and is reported to cause a name lookup
> regression.
>
> I don't know whether what you see is related to that.
I don't think so, it should be not that recent (but I don't check the 3.3
branch that often). Maybe someone could run the automated regression test
searcher on the preprocessed testcase I put at
http://www.tat.physik.uni-tuebingen.de/~rguenth/gcc/array_test1.ii.gz
it fails (aka SIGSEGVs) at -O2 -funroll-loops and works (prints "PASSED
... array_test1") at -O2. This is on i686-pc-linux-gnu, with
g++-3.3 (GCC) 3.3.3 20040107 (prerelease). It's a regression towards
previous 3.3 (but I don't know which, around september it was at least ok,
I think).
Maybe optimization/13222 (which manifests on powerpc only) is also
related.
Richard.