one thing: it uses ALOT of floating point. And you say this happens:
[irqs off]
[timestamp1]
<ALOT of computing, only from memory, into memory, using CPU and FPU>
[timestamp2]
[irqs on]
and delta := timestamp2-timestamp1, and you say delta(DOS) is 17% better
than delta(Linux) ?
all code has the same alignment? And it's only this code that is
executing? And you say TLB cache isnt trashing?
Impossible!! ;) unless there is some memory chipset / CPU cache / FPU
setup trick done by DOS/Win/Win95.
-- mingo