the SH4 has a stronger FPU unit clock-for-clock than the x87 math co-processor inside the x86. SSE is being used used because it works on FPU operations, MMX only deals with integer operations. Most likely 3dnow is being ignored because SSE is more prevelant (p3,p4,AMD XP and better) and SSE is available on hardware which can run Chankast at near full speed. Implementing 3dnow is alot of extra work for nothing -- even if were supported all the AMD-Tbird processors still too underpowered to run chankast at full speed.