Message ID | 20240130062626.98273-1-connorbworley@gmail.com |
---|---|
State | Accepted |
Commit | 1eeee68d8eb678665730bdd7862598be15867028 |
Headers | show |
Series | [FFmpeg-devel] lavc/dxv: fix incorrect back-reference index calculation in DXT5 decoding | expand |
Context | Check | Description |
---|---|---|
yinshiyou/make_loongarch64 | success | Make finished |
yinshiyou/make_fate_loongarch64 | success | Make fate finished |
andriy/make_x86 | success | Make finished |
andriy/make_fate_x86 | success | Make fate finished |
Attached are the input file used for comparison with Alley, and before-and-after images generated with `ffmpeg -i dice_dxt5.mov -f image2 out.png` On Mon, Jan 29, 2024 at 10:26 PM Connor Worley <connorbworley@gmail.com> wrote: > This bug causes the DXT5 decoder to produce incorrect block texture data. > After the fix, textures are visually correct and match data decoded by > Resolume Alley (extracted with Nvida Nsight for comparison). Current FATE > DXT5 > samples did not cover this case. > > Signed-off-by: Connor Worley <connorbworley@gmail.com> > --- > libavcodec/dxv.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/libavcodec/dxv.c b/libavcodec/dxv.c > index b29adf8ad9..be1216da86 100644 > --- a/libavcodec/dxv.c > +++ b/libavcodec/dxv.c > @@ -755,7 +755,7 @@ static int dxv_decompress_dxt5(AVCodecContext *avctx) > break; > case 2: > /* Copy two dwords from a previous index */ > - idx = 8 + bytestream2_get_le16(gbc); > + idx = 8 + 4 * bytestream2_get_le16(gbc); > if (idx > pos || (unsigned int)(pos - idx) + 2 > > ctx->tex_size / 4) > return AVERROR_INVALIDDATA; > prev = AV_RL32(ctx->tex_data + 4 * (pos - idx)); > -- > 2.40.1 > >
Got some confirmation that this patch fixes https://trac.ffmpeg.org/ticket/10264
On Thu, Feb 1, 2024 at 9:41 PM Connor Worley <connorbworley@gmail.com> wrote: > Got some confirmation that this patch fixes > https://trac.ffmpeg.org/ticket/10264 > > Thanks for the fix. I had encountered similar artifacts but failed to fix it, and then removed files long ago, but i got them from the web so will recheck if this indeed fixes it too. > -- > Connor Worley > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". >
Given that Paul is no longer a maintainer, can someone else please take a look?
diff --git a/libavcodec/dxv.c b/libavcodec/dxv.c index b29adf8ad9..be1216da86 100644 --- a/libavcodec/dxv.c +++ b/libavcodec/dxv.c @@ -755,7 +755,7 @@ static int dxv_decompress_dxt5(AVCodecContext *avctx) break; case 2: /* Copy two dwords from a previous index */ - idx = 8 + bytestream2_get_le16(gbc); + idx = 8 + 4 * bytestream2_get_le16(gbc); if (idx > pos || (unsigned int)(pos - idx) + 2 > ctx->tex_size / 4) return AVERROR_INVALIDDATA; prev = AV_RL32(ctx->tex_data + 4 * (pos - idx));
This bug causes the DXT5 decoder to produce incorrect block texture data. After the fix, textures are visually correct and match data decoded by Resolume Alley (extracted with Nvida Nsight for comparison). Current FATE DXT5 samples did not cover this case. Signed-off-by: Connor Worley <connorbworley@gmail.com> --- libavcodec/dxv.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)