new sr2

[Piotr Tarsa]

hi!

i'm rewriting sr2 to structural (c- style) code and incorporating fpaqd code.

currently it's in early form. here's the code: http://phpfi.com/286283 . only model is written, coder is not present yet. any ideas on speed improvements are welcomed.

[LovePimple]

Excellent!

[Nania Francesco]

Please rewrite SR3 to new fpaqd code!

[Piotr Tarsa]

nania:
i'll leave that task for you
my code is ugly and there is big chance that you won't benefit from it.

i've added coder but program now crashes (i don't know why). it doesn't even print anything on console.

http://phpfi.com/286307

i'm working on pelles c version 4.50.113

[Nania Francesco]

Pelles C is more faster compared with Dev-cpp ?

[Nania Francesco]

Detected The Error:
Is impossible to allocate
U64 QInv[2 << 12];
U32 T4[1 << 20], StateMap[(1024 + 64) * 258];
U16 Stack[10 * 65536];
U8 Output[2 * 65536];
in the MAIN procedure!

these so great variables in the C or C++ declare him to the beginning of the code !

[Piotr Tarsa]

thanks francesco!

now it doesnt crash, but dont work properly

http://phpfi.com/286326

Pelles C is more faster compared with Dev-cpp ?

i dont know (i dont even care) but it has imo much better ide (code formatting in dev- cpp annoys me).

[Piotr Tarsa]

corrected some errors (still works improperly):
http://phpfi.com/286329

[Piotr Tarsa]

hopefully it produces correct output now:
http://phpfi.com/286331

huh, time to write decoder

[Black_Fox1]

at school we were presented with a recommended choice between dev-cpp and code::blocks... all have chosen c::b

[Piotr Tarsa]

huh, it seems that i won't make the decoder (ie. have too many ideas about encoder and too little time for decoder)

but there are cotinuous improvements

http://phpfi.com/286340

[Nania Francesco]

THANKS DONKEY7! VERY GOOD AND FAST JOB!

[Piotr Tarsa]

current version (i think it's enough for now):
http://phpfi.com/286344

implemented optimization from: http://www.encode.su/forums/index.php?action=vthre ad&forum=1&topic=589&page=1#msg7299

removed lookup tables for learning rates (statefactor). used direct calculations instead.

i have only old version of gcc (almost three years old). it produces slow executables. i'm curious about performance of sr2v2enc compared to sr2 (compiled in good compilers). i know sr2v2 is slower than sr2 but i can't estimate how much using my current compilers

[Simon Berger]

It seems to be a huge gap. I compiled with Visual C++ 7. Didn?t compiled sr2v1 for myselve. But that shouldn?t make a big difference

timer srv2 test.bin output.svr22

Timer 3.01 Copyright (c) 2002-2003 Igor Pavlov 2003-07-10
29705864 -> 7471130 in 5.13 sec.

Kernel Time = 0.031 = 00:00:00.031 = 0%
User Time = 5.078 = 00:00:05.078 = 99%
Process Time = 5.109 = 00:00:05.109 = 99%
Global Time = 5.125 = 00:00:05.125 = 100%

-------------------------------------------------- ----------------------------------

timer sr2 test.bin output.svr21

Timer 3.01 Copyright (c) 2002-2003 Igor Pavlov 2003-07-10
29705864 -> 7476944 in 1.56 sec.

Kernel Time = 0.046 = 00:00:00.046 = 3%
User Time = 1.500 = 00:00:01.500 = 96%
Process Time = 1.546 = 00:00:01.546 = 99%
Global Time = 1.562 = 00:00:01.562 = 100%

[Piotr Tarsa]

on my system there's no huge difference

i've written decoder. fortunately there wasn't many problems with it.

it's here (together with compiled executables) - please mirror it if posible:
http://asembler.republika.pl/misc/sr2v2.zip (hope i included right files)

compression is much slower, decompression is little slower (than orignal sr2). maybe this will change when using better compilers.

cheers!

[LovePimple]

Thanks donkey7!

Mirror: Download