On Wed, May 18, 2011 at 1:12 PM, Samuel Mimram <smim...@gmail.com> wrote:
>
>
> On Wed, May 18, 2011 at 10:09 PM, Samuel Mimram <smim...@gmail.com> wrote:
>>
>> On Wed, May 18, 2011 at 9:25 PM, Patrick Matthäi <pmatth...@debian.org>
>> wrote:
>>>
>>> Am 18.05.2011 21:22, schrieb Samuel Mimram:
>>> > On Wed, May 18, 2011 at 10:31 AM, Samuel Mimram <smim...@gmail.com
>>> > <mailto:smim...@gmail.com>> wrote:
>>> >
>>> > I can confirm that the bug still occurs with a non-hardened package
>>> > (rebuilt following your instructions).
>>> >
>>> >
>>> > For info, the full stacktrace is:
>>> >
>>> > #0 0x00007fffe5ec54cd in filter_line_sse2 (mode=0, dst=0x7fffde5fb7a0
>>> > "",
>>> > prev=0x7fffde6fd7a0
>>> > "\207\207\210\212\213\214\215\215\220\220\220\221\222\222\222\223\224\224\225\225\226\227\227\230\233\233\234\234\235\236\236\237\240\240\241\241\242\243\243\244\243\244\244\245\246\246\247\247\252\251\251\250\247\246\245\245\251\251\251\251\251\251\251\251\255\255\255\255\255\254\254\254\255\255\255\255\255\255\255\255\257\257\257\257\257\257\257\257\260\260\260\260\260\260\260\260\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\261\263\263\263\263\263\263\263\263\262\262\262\262\262\262\262\262\260\260\260\260\260\260\260\260\261\261\260\260\257\256\256\255\256\256\256\256\256\256\256\256\254\254\254\254\254\254\254\254\257\256\254\252\251\252\252\253\250\251\251\252\253\253\254\254\255\255\254\254\253\252\252\251\250\250\247\247\246\245\245\244"...,
>>>
>>> I see SSE2 foo, but this could not be the issue, because the i386
>>> packages (with that I tested it these days) are build without any CPU
>>>
>>> optimizations.
>>
>> Well, I'm on amd64 and SSE2 seems to be enabled here (I see -DUSE_SSE2 in
>> build logs...).
>
> Even better, when I add --disable-sse2 to the configure options on amd64, the
> problem vanishes!
>
I reproduced this on my Arch Linux box using gcc v4.6. Some
optimizations in -O1 and -O2 cause this SSE2 code to fail. I did
bisection testing of all of the specific options these enable and
isolated them:
# Since gcc 4.6, this optimization enabled with -O1 causes
filter_line_sse2 to crash.
echo "OPTIMISATIONS+=-fno-tree-dominator-opts"
# Since gcc 4.6, this optimization enabled with -O2 causes
filter_line_sse2 to crash.
echo "OPTIMISATIONS+=-fno-tree-pre"
The solution is in the next version or you can configure with
--disable-sse2 since only the YADIF deinterlacer uses SSE2 and it also
has a SSE version that still works.
--
+-DRD-+
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
