fast PPM
Dmitry Shkarin
Fast implementation of PPMII scheme is placed at
http://www.compression.ru/ds/ppmdj.rar
Matt Mahoney
Very nice improvement. ppmd compresses as follows in less than 1
second on a 2.2 GHz Athlon-64 in 32 bit mode (XP home), using less than
2 MB memory:
25,443 BIB.pmd
216,021 BOOK1.pmd
149,404 BOOK2.pmd
55,403 GEO.pmd
109,407 NEWS.pmd
9,419 OBJ1.pmd
72,551 OBJ2.pmd
15,004 PAPER1.pmd
22,912 PAPER2.pmd
50,095 PIC.pmd
11,332 PROGC.pmd
14,751 PROGL.pmd
10,193 PROGP.pmd
16,965 TRANS.pmd
778,900 bytes
ppmonstr compresses this in 11 seconds using 15 MB:
22,870 BIB.pmm
203,247 BOOK1.pmm
131,191 BOOK2.pmm
47,950 GEO.pmm
96,990 NEWS.pmm
8,191 OBJ1.pmm
54,078 OBJ2.pmm
13,965 PAPER1.pmm
21,550 PAPER2.pmm
31,984 PIC.pmm
10,255 PROGC.pmm
11,867 PROGL.pmm
8,214 PROGP.pmm
13,133 TRANS.pmm
675,485 bytes
These are for default options. Playing with the -o and -m options
brings this down a bit.
-- Matt Mahoney
Sachin Garg
The ratios are better, it uses slightly less memory and speed seems to
almost the same. It sure would have been hard to improve perfection.
Great work. :-)
I guess it will be even faster on 64-bit processors as now the data is
64-bit aligned on those plaforms (thatz the only 64-bit thing I
noticed, did I missed something?).
btw, ppmdj doesnt seems to be backward compatible with ppmdi1. Now that
ppmdi1 is used as a standard (like in zip format), I think improvements
(like speed/64-bit support/threading support) to ppmdi1 code will help
too.
Sachin Garg [India]
http://www.sachingarg.com
johan....@gmail.com
compared to variant I:
GIMP Source test:
6.882.401 <- 7.448.321
Lossless Audio test:
39.229.385 <- 40.408.849
Scourge test:
10.830.524 <- 12.076.088
Fallon
implementation of PPMII', -ppmonster-, when durilca seems to perform
better?
And to take up one of Sachin Garg's remarks: will ppmdi1, as in
winzip10, now be improved at some point?
Fallon
are separate issues, but if you care to say something about future
plans.... :)
Malcolm Taylor
As far as I know, durilca is based on PPMonstr. In fact I expect the new
PPMonstr is the same PPM engine from durilca with a few small improvements.
Durilca gets better compression in some cases by using filtering and
pre-computed models.
If you look at the benchmarks, PPMonstr gets better results than durilca
except where durilca is able to leverage the filters (eg. EXE and DLLs).
As to WinZIP 10, I certainly hope they do not include var. J as it will
just screw up backwards compatibility for a very small gain in compression.
Malcolm
Dmitry Shkarin
F> From a curious nitwit: what's the significance of working on this
F> 'fast
F> implementation of PPMII', -ppmonster-, when durilca seems to perform
F> better?
DURLICA = Filters + PPMonstr
DURLICA`light = Filters + PPMd
As You can see here is clear reason to improve PPMonstr.
Another question is that monstrous compressors are nearly useless in
real life (not for research, competing etc.), so You would better look at
RAR, 7-zip etc.
Fallon
> As You can see here is clear reason to improve PPMonstr.
Yes, the filters, it dawned on me now.
> Another question is that monstrous compressors are nearly useless in
>real life (not for research, competing etc.)
I was planning my personal look at the results. :)
> You would better look at RAR, 7-zip etc
Your guess is right, Dmitry, I use 7-zip and WinRAR for most archiving.
Apart from that WinRK, UHA and Bee http://www.beegui.org/
>From a tourist with an interest to follow your work...
Thank you both.