This is faster than dodeca 1.0, but the speed up is very very different for different n values and ranges.
To obtain this I've eliminated almost all modular multiplications ( in the part when we see if "g" is good or not ). Now magic_constant=32000 is good for this version. I'll think what would be a good "default" value. But note that we are sieving also up to 32000 and one block length is also 32000.
Kosmaj can you test this version, I've checked only for n=44,47. And test your previous case 1 and case 2.
You can download exe for windows from:
http://www.robertgerbicz.tar.hu/dodeca_2_0.exe
Or see the attachment for the c code.