[FFmpeg-devel] [PATCH 4/4] tools/target_dec_fuzzer: Adjust max_pixels for IDCIN

Tomas Härdin tjoppen at acc.umu.se
Thu Aug 22 12:24:26 EEST 2019


tor 2019-08-22 klockan 10:34 +0200 skrev Michael Niedermayer:
> On Thu, Aug 22, 2019 at 10:23:40AM +0200, Paul B Mahol wrote:
> > On Thu, Aug 22, 2019 at 10:16 AM Michael Niedermayer <
> > michael at niedermayer.cc>
> > wrote:
> > 
> > > On Thu, Aug 22, 2019 at 10:09:30AM +0200, Paul B Mahol wrote:
> > > > Why not fix issues listed in decoder code instead of this hack?

I think there's better things we could have Michael do. That said:

> > Also all this old game codecs never used very high resolutions.
> > Max something vga, but not even that.
> 
> limiting to 640x480 instead of 4096x4096 would give a factor of about
> 55
> speedup, thats still a long way from 2000 so that solves also only
> part of it
> 
> anyway ill provide a patch that limits resolution so theres something
> that can be discussed. 

The probing in lavf/idcin.c already limits the resolution to 1024x1024.
The only sample we have is 320x240. The Quake 2 decoder (cl_cin.c)
limits CIN frames to 128 KiB compressed, from which the 1024x1024 limit
in idcinvideo.c likely derives (128 KiB == 1 Mib == 1 Mipx if always
using the shortest Huffman code). Zero-byte packets are also not
allowed.

If we're looking over this anyway we could toss in all these
limitations, and maybe limit audio to 11025, 22050 and 44100 Hz.

I also notice that idcinvideo.c contains code copy-pasted from cl_cin.c
in Quake 2 (SmallestNode1 -> huff_smallest_node). The copyright header
should reflect this.

/Tomas



More information about the ffmpeg-devel mailing list