[vtk-developers] Speeding up raycasting on x86

[Volpe, Christopher R (Research)]

> -----Original Message-----
> From: David Gobbi [mailto:dgobbi at imaging.robarts.ca]

> And all of this is necessary just because the engineers that designed
> the 386 didn't include a 'truncate to int' instruction, only 
> a 'convert
> float to int according to prevailing rounding mode' where 
> every operating
> system uses 'round to nearest' as the default mode.

Actually, this would be useful even if Intel *did* provide such an instruction. I've used inline
assembler to invoke the X86 FISTP instruction (without changing the FPU rounding mode) when I didn't
care that it would round, and this pure-C++ method is still faster.

-Chris