just before going to one week holiday, I converted shade-c variance shadowmapping into exponential shadowmapping. linear depth, filtering in log space. it seems to be a better choice. fast but not perfect yet, first shot:

http://www.mediafire.com/view/cicrr82mr82u60t/ESM_shot0.jpg

sivan@ looks so promissing dude, could this go as a birthday gift for me?

yes of course happy birthday 3run!

sivan@ thank you very much man

little update: I think ESM ready with tweakable parameters, including 2 selectable blur type. it will replace VSM released in MB 2.2 and 2.21. throw them out

shadow map resolution 1366^2, are covered 3000^2 quants, 16bit float depthmap, 5 sample box blur, beside max object shader quality settings:

sivan@ it just looks perfect, nothing left to say.. but what about performance? is it much slower than original shade-c (0.91) shadows?

nice, are there any good information or code snippets about ESM that you could provide?

edit: '5 sample box blur'
do you mean 5 samples in total or 5 samples per axis (25 which would be a bit inefficient...)?

it is a bit faster and nicer than the original shade-c vsm. normally I get at least 40 fps, but it is level dependent, and there are a lot of things to tweak performance in the corresponding shader package too. I get about 30 fps with my RTS pathfinder having 490 units moving with vertex animation in a terrain only 500x500 tile sized scene. previously it was about 25 fps.
the simple blur takes only the actual pixel itself, and the 4 adjacent ones (but sampling distance can be adjusted). I made a poisson blur too, taking 12 pixels in total. probably will make a gaussian too (sample 2x5 or similar).
I used this little blog for my implementation: http://www.olhovsky.com/2011/07/exponential-shadow-map-filtering-in-hlsl/ (Boh_Havok also read it, if you check some outcommented lines in shade-c evo shaders), and read a short nvidia presentation, and some others I have not understood...

Sorry for the late response and thanks again!
It's basically the flood fill approach but I'm currently having a few troubles with the unit-unit collision detection, especially when there are many units they start to penetrate each other or even stockpile. However, my three approaches so far which stop that lead to a chunky unit movement, I guess this will take a while before it works smoothly and solid.

@sivan: Why did you go for a 1366^2 shadow map resolution (camera size/ texture size?)? Seems a little inappropriate to me.

Never test with your friends 2

http://www.youtube.com/watch?v=GVXOpK1V47E
(you can turn on english subtitles)