[FFmpeg-devel] [PATCH] avcodec/qpeldsp: copy less for the mc0x cases

[Michael Niedermayer]

On Mon, Jun 27, 2022 at 07:57:06AM +0200, Andreas Rheinhardt wrote:
> Michael Niedermayer:
> > Fixes: out of array access
> > Fixes: 47936/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MPEG4_fuzzer-5745039940124672
> > 
> > Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
> > Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
> > ---
> >  libavcodec/qpeldsp.c | 12 ++++++------
> >  1 file changed, 6 insertions(+), 6 deletions(-)
> > 
> > diff --git a/libavcodec/qpeldsp.c b/libavcodec/qpeldsp.c
> > index 2b9146ceb1..5f937f9d9e 100644
> > --- a/libavcodec/qpeldsp.c
> > +++ b/libavcodec/qpeldsp.c
> > @@ -199,7 +199,7 @@ static void OPNAME ## qpel8_mc01_c(uint8_t *dst, const uint8_t *src,          \
> >      uint8_t full[16 * 9];                                                     \
> >      uint8_t half[64];                                                         \
> >                                                                                \
> > -    copy_block9(full, src, 16, stride, 9);                                    \
> > +    copy_block8(full, src, 16, stride, 9);                                    \
> >      put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);                   \
> >      OPNAME ## pixels8_l2_8(dst, full, half, stride, 16, 8, 8);                \
> >  }                                                                             \
> > @@ -209,7 +209,7 @@ static void OPNAME ## qpel8_mc02_c(uint8_t *dst, const uint8_t *src,          \
> >  {                                                                             \
> >      uint8_t full[16 * 9];                                                     \
> >                                                                                \
> > -    copy_block9(full, src, 16, stride, 9);                                    \
> > +    copy_block8(full, src, 16, stride, 9);                                    \
> >      OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);                   \
> >  }                                                                             \
> >                                                                                \
> > @@ -219,7 +219,7 @@ static void OPNAME ## qpel8_mc03_c(uint8_t *dst, const uint8_t *src,          \
> >      uint8_t full[16 * 9];                                                     \
> >      uint8_t half[64];                                                         \
> >                                                                                \
> > -    copy_block9(full, src, 16, stride, 9);                                    \
> > +    copy_block8(full, src, 16, stride, 9);                                    \
> >      put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);                   \
> >      OPNAME ## pixels8_l2_8(dst, full + 16, half, stride, 16, 8, 8);           \
> >  }                                                                             \
> > @@ -459,7 +459,7 @@ static void OPNAME ## qpel16_mc01_c(uint8_t *dst, const uint8_t *src,         \
> >      uint8_t full[24 * 17];                                                    \
> >      uint8_t half[256];                                                        \
> >                                                                                \
> > -    copy_block17(full, src, 24, stride, 17);                                  \
> > +    copy_block16(full, src, 24, stride, 17);                                  \
> >      put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);                 \
> >      OPNAME ## pixels16_l2_8(dst, full, half, stride, 24, 16, 16);             \
> >  }                                                                             \
> > @@ -469,7 +469,7 @@ static void OPNAME ## qpel16_mc02_c(uint8_t *dst, const uint8_t *src,         \
> >  {                                                                             \
> >      uint8_t full[24 * 17];                                                    \
> >                                                                                \
> > -    copy_block17(full, src, 24, stride, 17);                                  \
> > +    copy_block16(full, src, 24, stride, 17);                                  \
> >      OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);                  \
> >  }                                                                             \
> >                                                                                \
> > @@ -479,7 +479,7 @@ static void OPNAME ## qpel16_mc03_c(uint8_t *dst, const uint8_t *src,         \
> >      uint8_t full[24 * 17];                                                    \
> >      uint8_t half[256];                                                        \
> >                                                                                \
> > -    copy_block17(full, src, 24, stride, 17);                                  \
> > +    copy_block16(full, src, 24, stride, 17);                                  \
> >      put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);                 \
> >      OPNAME ## pixels16_l2_8(dst, full + 24, half, stride, 24, 16, 16);        \
> >  }                                                                             \
> 
> Are the arch-specific dsp functions affected by this, too?

the fuzzer sample does not cause a anomaly in the x86 functions


> Do you happen to know why copy_block9/17 has been used in this code?

probably because there was asm anyway so it was unused code or maybe it
was oversight or some sort of code sharing/code cache use reduction but
these hypothesis dont fit entirely. so i dont really know/rememeber


> (After all, using copy_block8/16 should result in a slight speedup, so I
> using copy_block9/17 must have been intentional.)

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Homeopathy is like voting while filling the ballot out with transparent ink.
Sometimes the outcome one wanted occurs. Rarely its worse than filling out
a ballot properly.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20220627/9b8538d3/attachment.sig>