This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [RFA] expand from SSA form (1/2)
On Mon, Apr 27, 2009 at 12:57 PM, David Edelsohn <dje.gcc@gmail.com> wrote:
> On Mon, Apr 27, 2009 at 11:51 AM, Michael Matz <matz@suse.de> wrote:
>> Hi,
>>
>> On Mon, 27 Apr 2009, David Edelsohn wrote:
>>
>>> > This confusion with the numbers seems indeed related to PR39922. ?If
>>> > you can test the below patch on powerpc that would be nice. ?I know
>>> > that it fixes the PR on i686-linux where something similar happens.
>>>
>>> A quick test of the patch does not fix the failure on AIX. ?I will try a
>>> complete bootstrap in case the bug was causing other miscompilations.
>>
>> Hmpf. ?There's another problem (promoted parameters), which might also
>> cause this. ?The patch I just sent to Andreas Krebbel should fix that one.
>> If it doesn't help AIX/darwin I fear I need a testcase. ?There probably
>> are some failures in execute.exp which also would be miscompiled, that
>> might be easier to extract than the miscompilation itself.
>
> Bootstrapping from scratch with the patch you sent me still shows the
> problem. ?I will try with the patch you sent to Andreas.
>
> As I mentioned in the PR, for AIX one can reproduce the problem by
> compiling libobjc/linking.m. ?That is a fairly small file and produces
> invalid assembly due to ctor priority of -1.
Unfortunately, the second patch does not fix the PowerPC failure either.
I am running the C testsuite to see what new failures occur.
David