or has it totally been replaced by the fpu now?

I cut a routine to about 1/8th the time by replacing a single floating point value incremented and converted to integer to a fixed point once per pass. That was a differance of about 12MB/s versus 100MB/s fill rate.

I still use fixed point in most projects, especially VB projects where it can greatly speed up things. It looks like floating point multiply and divide instructions are getting faster with each new CPU version, but well, you still have MMX to work with fixed point. And also, I find it much harder to optimize floating point asm code than fixed point.

The FPU keeps getting faster, but so does the ALU. And with the notable exception of the P4, you usually have just as many (sometimes more) ALU pipelines as FPU pipelines. And there''s just plain less to do when doing integer math.

However all the 3D stuff uses floating point, so there''s not much point in storing fixed point numbers if you have to convert it all to floats.