I've moved into Neoscrypt with 280X cards, 14.6 Catalyst and Driver, and the recent Nicehash neoscrypt kernel.
I have my 280X's running 1000gpuclock / 1500memclock and 340KH/s   

However, I see that 400KH/s or better is doable, so hack at the kernel I must.         

I have already found some weird things with the compiler, such as crashing on perfectly good code,
and there are places where adding meaningless instructions actually makes it faster, so I assume there
is some fitment of code into a core size which I need to get a feel for.

Would anyone care to be so nice as to point me to a particular section of the kernel where I should
focus my attention, or at least a push in the right direction?  Several unrolls and coding functions inline
have resulted in just a few KH/s

I will start it off.....adding a temp variable to neoscrypt_blkmix makes it go 15KH/s faster.