Re: Fix twolf -funroll-loops -O3 miscompilation (a semi-latent web.c bug)

[Bin.Cheng]

On Wed, Oct 17, 2012 at 6:54 AM, Steven Bosscher <stevenb....@gmail.com> wrote:
> On Tue, Oct 16, 2012 at 1:38 PM, Paolo Bonzini wrote:
>>>> Bottom line: This is a bug in df_kill_notes.
>>
>> Yep, and it could cause wrong code generation, couldn't it?  Because the
>> new (reg:DI 72) is obviously not equal to (plus:DI (reg:DI 72)
>> (const_int 24)), but it could be propagated to subsequent insns.
>
> Right.
>
>
>> So I think this patch should be backported to all release branches after
>> regstrapping.
>
> Agreed, but let's wait a few days at least to see what happens with
> the patch on the trunk.
>
>
>> Ok after bootstrap, regtest and checking that it actually fixes the bug.
>
> I made sure of that before posting the patch ;-)
>
> Bootstrap&regtest is running on x86_64-unknown-linux-gnu and
> powerpc64-unknown-linux-gnu. I'll commit the patch tomorrow if there
> are no test suite regressions.
>
> Well done to us all for probing until we found the root cause of this bug!
>

Hi Jan,

For the test case:

main()
{
  configure2();
}

Please make main return ZERO by default.

When cross testing with qemu/simulator, the wrap exit checks whether
the case by verifying the return value.
For ARM target, R0 is checked while it may be clobbered(thus non-ZERO)
if main does not return any value, which causes failure of test case.

Thanks very much.

-- 
Best Regards.