I wont keep quoting the above but badman's develop branch is working well for me now with 4-5% speedups over the v5_0 branch.

@bullus I'm also running 7950s. What are you settings and hashrates? I'm currently pulling 2424-2435 khs per card @ 1025 engine / 1250 mem on x15 using:

"hamsi-expand-big" : "7",
"hamsi-short" : true,
"blake-compact" : true,
"keccak-unroll" : "8",
"luffa-parallel" : true,

Not sure if these are the right settings for this card or not. It's about 5.2% faster than the v5_0 branch, but there might be room for optimization. I'll probably fiddle around with it throughout the week.