[FFmpeg-devel] [PATCH] avcodec/mpeg4videodec: Align idct-block appropriately

[Andreas Rheinhardt]

Sean McGovern:
> Andreas:
> 
> On Tue, Mar 12, 2024 at 8:53 PM Andreas Rheinhardt
> <andreas.rheinhardt at outlook.com> wrote:
>>
>> It is accessed via AV_RN64() in ff_simple_idct_put_int32_10bit().
>> Should fix the UBSan failures in the mpeg4-simple-studio-profile
>> test here:
>> https://fate.ffmpeg.org/report.cgi?time=20240312011016&slot=ppc-linux-gcc-13.2-ubsan-altivec-qemu
>>
>> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt at outlook.com>
>> ---
>>  libavcodec/mpeg4videodec.h | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/libavcodec/mpeg4videodec.h b/libavcodec/mpeg4videodec.h
>> index c0e6ec6592..4a26d18987 100644
>> --- a/libavcodec/mpeg4videodec.h
>> +++ b/libavcodec/mpeg4videodec.h
>> @@ -29,6 +29,7 @@
>>  #include "mpegvideo.h"
>>  #include "mpeg4videodsp.h"
>>
>> +#include "libavutil/mem_internal.h"
>>
>>  typedef struct Mpeg4DecContext {
>>      MpegEncContext m;
>> @@ -83,7 +84,7 @@ typedef struct Mpeg4DecContext {
>>
>>      Mpeg4VideoDSPContext mdsp;
>>
>> -    int32_t block32[12][64];
>> +    DECLARE_ALIGNED(8, int32_t, block32)[12][64];
>>      // 0 = DCT, 1 = DPCM top to bottom scan, -1 = DPCM bottom to top scan
>>      int dpcm_direction;
>>      int16_t dpcm_macroblock[3][256];
>> --
>> 2.40.1
>>
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> 
> Confirming that this fixes fate-mpeg4-simple-studio-profile on PowerPC
> QEMU with GCC UBsan as noted.
> 
> It was not failing on POWER7 (ppc64) or POWER9 (ppc64le).
> 

Not surprising: The Mpeg4VideoDSPContext immediately before block32
contains (function) pointers and therefore has alignment 8 on 64bit
machines, so that block32 happens to be properly aligned already.

- Andreas