[FFmpeg-devel] [PATCH] postproc: fix unaligned access

Andreas Cadhalpun andreas.cadhalpun at googlemail.com
Fri Jun 19 01:49:22 CEST 2015


On 19.06.2015 01:34, Michael Niedermayer wrote:
> On Fri, Jun 19, 2015 at 12:05:50AM +0200, Andreas Cadhalpun wrote:
>> On 18.06.2015 22:26, Michael Niedermayer wrote:
>>> On Thu, Jun 18, 2015 at 08:29:06PM +0200, Andreas Cadhalpun wrote:
>>>> QP_store is only 8-bit-aligned, so accessing it as uint32_t causes
>>>> SIGILL crashes on sparc.
>>>>
>>>> Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun at googlemail.com>
>>>> ---
>>>>  libpostproc/postprocess.c | 5 +----
>>>>  1 file changed, 1 insertion(+), 4 deletions(-)
>>>>
>>>> diff --git a/libpostproc/postprocess.c b/libpostproc/postprocess.c
>>>> index 92cc436..e4ea515 100644
>>>> --- a/libpostproc/postprocess.c
>>>> +++ b/libpostproc/postprocess.c
>>>> @@ -996,10 +996,7 @@ void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],
>>>>          if (QPStride >= 0){
>>>>              int i;
>>>>              const int count= FFMAX(mbHeight * QPStride, mbWidth);
>>>> -            for(i=0; i<(count>>2); i++){
>>>> -                ((uint32_t*)c->nonBQPTable)[i] = ((const uint32_t*)QP_store)[i] & 0x3F3F3F3F;
>>>> -            }
>>>> -            for(i<<=2; i<count; i++){
>>>
>>> maybe this can be left under #if HAVE_FAST_UNALIGNED with
>>> AV_RN32 AV_WN32
>>
>> Yes, updated patch attached.
>>
>> Best regards,
>> Andreas
> 
>>  postprocess.c |    3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>> 09c6ba371eb89ee2767b642b731ed6d192e9bcf3  0001-postproc-fix-unaligned-access.patch
>> From 92662bfcebb5a49b3f6728d18582f1ae7205bcb3 Mon Sep 17 00:00:00 2001
>> From: Andreas Cadhalpun <Andreas.Cadhalpun at googlemail.com>
>> Date: Thu, 18 Jun 2015 20:15:12 +0200
>> Subject: [PATCH] postproc: fix unaligned access
>>
>> QP_store is only 8-bit-aligned, so accessing it as uint32_t causes
>> SIGBUS crashes on sparc.
>> The AV_RN32/AV_WN32 macros only do unaligned access in the
>> HAVE_FAST_UNALIGNED case.
> 
> LGTM

Pushed.

Best regards,
Andreas



More information about the ffmpeg-devel mailing list