Fast vector norm sought

[Mark Hunter]

To compute the norm of a three-dimensional vector in PB (all variables are Single):

 v = Sqr(x*x + y*y + z*z)

To do it using the floating point processor:

 Macro Norm3(v,x,y,z)
 ! fld x                   'push X onto stack
 ! fld st(0)               'push it again
 ! fmul                    'stack top = product
 ! fld y                   'push Y onto stack
 ! fld st(0)               'push it again
 ! fmul                    'stack top = product
 ! fld z                   'push Z onto stack
 ! fld st(0)               'push it again
 ! fmul                    'stack top = product
 ! fadd                    'stack top = sum of last two
 ! fadd                    'sum of that with last
 ! fsqrt                   'stack top = sqr of that
 ! fstp v                  'copy stack top to result and pop stack
 End Macro

But this assembly code is literally no faster than PB. Is there a faster way?

[Mark Hunter]

This article indicates that using SSE and its reciprical square root (since I will be dividing anyway and I don't need much accuracy)
rsqrtss
will take 1/20 the time as FPU's fsqrt.

I've never used SSE before. One restriction is that the CPU hardware must support SSE, which didn't exist until 1999.

ADDED: PBCC4 doesn't recognize MMX, much less SSE. (Ditto for PBCC5 but I need this for version 4.)