zpaq 1.02 update

[Matt Mahoney]

I released a minor update, zpaq v1.02. There is no change in compression or compatibility with versions back to 1.00.
http://cs.fit.edu/~mmahoney/compression/

It fixes a minor bug in both zpaq and unzpaq where extracted files were not closed until the program exited. This could cause it to fail to extract very large archives because of too many open files.

The new version is compiled with g++ 4.4 instead of g++ 3.4.2. It is about 20% faster using max.cfg. I also removed g++ 4.4 warnings and some (not all) Borland and Mars compiler warnings. I checked that those changes didn't change the executable.

[LovePimple]

Thanks Matt!

[joerg]

compiled with g++ 4.4 instead of g++ 3.4.2

--- about 20% faster using max.cfg

WOW

wonderful!

[Sportman]

I don't get compression started with syntax:
zpaq cmax.cfg @outfile @infile

[LovePimple]

Its working OK here.

[Sportman]

Its working OK here.

Oops I was sleeping here, it work after adding c before cfg name...

Speed improvements are good:

http://www.metacompressor.com/upload...estfile=file39
http://www.metacompressor.com/top.aspx?testfile=file39

http://www.metacompressor.com/upload...estfile=file40
http://www.metacompressor.com/top.aspx?testfile=file40

[Skymmer]

I released a minor update, zpaq v1.02.

Thank you, Matt.

Speed improvements are good:

Very impressive speedup on your quad-core!
23.9% for max.cfg
33.4% for mid.cfg
13.1% for min.cfg

But I can't say the same for my modest AMD64 4000+ (Single core)
Tested on my lovely PAQ_TestBed.tar

           |   max   |   mid  |  min  |
--------------------------------------|
zPAQ v1.00 | 200.684 | 73.569 | 9.711 |
zPAQ v1.02 | 192.538 | 66.584 | 9.365 |
--------------------------------------'

So for my system:
4.1% for max.cfg
9.5% for mid.cfg
3.6% for min.cfg

Interesting that in your and my cases the biggest speedup is for mid.cfg. I wonder why ?

[Bulat Ziganshin]

compiler optimizations may improve arithmetic calculations but not mem access

[Matt Mahoney]

Not sure the reason why mid.cfg improves the most. It makes more use of hash table lookups I think. Also, very interesting that zpaq beats paq8o8 on the image files even without a BMP model. (zpaq would support it with a context depending on the image width).

[Rugxulo]

I guess I'm the only one here who uses OpenWatcom. Anyways, OW 1.8 compiles ZPAQ incorrectly for Win32 target (Vista doesn't like it!) when using -ox (aka, -obmiler -s) yet default (mix between -ot and -os) and -ot both work okay. Just FYI.

[LovePimple]

I had the same problem under Windows 2000 until I compiled without the -s (remove stack overflow checks). The attached build was compiled with -obmilert optimization settings. It works OK under Windows 2000, but its very slow compared to the original GCC 4.4 build.

[Skymmer]

Here is the test for WATCOM compile of zPAQ v1.02

           |   max   |   mid   |   min  |
----------------------------------------|
  original | 192.538 |  66.584 |  9.365 |
    watcom | 349.607 | 131.768 | 23.602 |
----------------------------------------'

Output files are identical to original compilation.