worksize - it is ocl localworksize of your workgroup. 64 is kind of optimal for amd on most algos(not for equihash i think, this algo is more complex than just compute hash). For amd it will vary from 1 to 256. Best sizes if it is divisible by 64. Or at least 2. But again it depends from algo and its implementation.
rawintensity - when miner compute hashes, it check nonce from 0x00000000 to 0xFFFFFFFF. You can't say to gpu: compute this whole range at once, that wouldn't be effective. So you should split this range in pieces(in this case it would be global worksize in opencl terms. You can think about it like a number of hashes that sgminer would compute before verifying gpu's work on cpu and moving on to a next piece of nonces). rawintensity would be the size of such piece. It can be selected based on hardware properties of your videocard, or to be chosen experimentally(this is what most miners do, lol). xIntensity or intensity arguments allow you to set this rawintensity without computing it by yourself, as i remember intensity will set globalworksize to 2^intensity. I don't know if you can apply this to equihash impl in sgminer, i'm talking based on exp with other algos. I think Zawawa can tell more, or someone who ported his impl into sgminer-gm.

UPD: actually, here is your guide: https://github.com/genesismining/sgminer-gm/blob/master/doc/MINING.md
with intensity and worksize description.