Thanks for help. Now it runs, and I found that this command produces the best results:
sgminer -d 1 -k groestlcoin --difficulty-multiplier 0.0039062500 -o http://localhost:1441 -u u -p p --shaders 1280 --worksize 256 -g 1 --intensity 24
It produces (avg)2MH/s which is the half of the 4Mh/s I calculated earlier.
Does sgminer divides the Groestl-hash calculation number by 2? Although, It would be more reasonable.
Or something is really wrong, that It runs on half speed (exactly hald speed)?
sgminer -d 1 -k groestlcoin --difficulty-multiplier 0.0039062500 -o http://localhost:1441 -u u -p p --shaders 1280 --worksize 256 -g 1 --intensity 24
It produces (avg)2MH/s which is the half of the 4Mh/s I calculated earlier.
Does sgminer divides the Groestl-hash calculation number by 2? Although, It would be more reasonable.
Or something is really wrong, that It runs on half speed (exactly hald speed)?
Intensity 24 is too much, I'd stay between 20 and 22, otherwise you'll produce a lot of rejected shares (or orphans if solo mining).
The shaders option is ignored for groestl.
The hashrate should be calculated on the full computation, i.e. 2 chained hashes.
What kernel are you using?
You have to compile with it for every type of gcn cards multiplied by every Catalyst driver that was altered by AMD developers. My compiler only patches the binary into the .elf, the actual elf file is generated by the current Catalyst Driver of the currently selected gfx card. 
Check the second part of the optimizations: 
Then you can use the instruction cache as a precalc buffer as well. The advantage is that most gpu's can read from the inst cache in paralell to the level 1 cache.