[FFmpeg-devel] [PATCH] NellyMoser audio decoder

Reimar Döffinger Reimar.Doeffinger
Wed Sep 12 17:42:05 CEST 2007


Hello,
On Wed, Sep 12, 2007 at 10:50:23AM -0400, Daniel Serpell wrote:
> Well, with a 128 bit hash, and 2^20 pages (so you have 4GB virtual RAM),
> the probability of collision is (1-e(- (2^40) / (2^129))) = 1.6*10^-27

You can't do blind probability calculation for things that are very,
very far from random (the content of the pages on a PC in no way qualify
as random).
That is the same fallacy as the Intel FDIV bug that "should" not have
been noticed according to such calculations.
Nevertheless, this would entirely be an option before resorting to
things like an OOM killer, and in some applications it might be even
faster (namely when not using it to find duplicates among already
swapped out memory, but for finding duplicates in the current memory
before/instead of).
But before someone tries it out it is hard to tell if it makes any sense
in real-world situations...

Greetings,
Reimar D?ffinger




More information about the ffmpeg-devel mailing list