[x264-devel] Getting SIGSEGV on Android port of the lib?

[Måns Rullgård]

BugMaster <BugMaster at narod.ru> writes:

> On Tue, 17 Jul 2012 17:05:48 -0400, Sean McGovern wrote:
>> On Tue, Jul 17, 2012 at 5:00 PM, BugMaster <BugMaster at narod.ru> wrote:
>>> On Tue, 17 Jul 2012 22:53:25 +0200, Marko Matic wrote:
>>>
>>>> So, now I disabled NEON? What are the side-effects of this fix?
>>>
>>> no, at least not all NEON optimazations but only memcpy_aligned
>>> function which looks like caused problem for some unknown (I still
>>> don't undestand why) reason.
>>>
>
>> Is it possible it's trying to memcpy between overlapping pointers?
>
>> -- Sean McG.
>
> Doubtful, at least I don't remember any place in code where this can
> happen. And the place it crashed in analyse.c definitely was not such
> case. The crash happened due pc register (instruction address) became
> zero. Dunno how this happen but doubt it was h->mc.memcpy_aligned
> function variable itself (only if it was somehow overwritten
> indirectly).

The debugger dump suggests it has a sane value.

> More probable is some stack problems inside it implementation of
> x264_memcpy_aligned_neon that cause jumping to zero address.

That function doesn't touch the stack at all, not even to save
registers, so a stack-related problem seems unlikely.

Can you post the output of readelf -a libx264.so?

-- 
Måns Rullgård
mans at mansr.com