Re: [boost] [fusion] More quanfusion

From: Andy Little (andy_at_[hidden])
Date: 2006-09-14 22:01:07

"Manfred Doudar" wrote

> If your dot product is indeed an inner product, then an inner_product
> is what you should call your op - being explicit rather than suggestive
> is always better (, sorry, just that my mathematics grounding makes me
> speak out on this one).

Dunno, but  FWIW , from  the source code:

http://tinyurl.com/e7729

involving transform of a homogeneous 2D coordinate (3 elements) by a 2D 
transform matrix (9 elements)
here's the resulting assembler of my latest CVS version  in VC8 with 
optimisation. I'm no expert but it looks pretty good to me

; 113  :     // do runtime calc
; 114  :     quan::fusion::dot_product dot;
; 115  :     result_type result(
; 116  :         dot(coordinate,col0),
; 117  :         dot(coordinate,col1),
; 118  :         dot(coordinate,col2)
; 119  :     );

  00164 dd 44 24 1c  fld  QWORD PTR _coordinate$[esp+304]
  00168 8d b4 24 d4 00
 00 00   lea  esi, DWORD PTR $T303578[esp+296]
  0016f dd 44 24 40  fld  QWORD PTR $T303558[esp+296]
  00173 d8 c9   fmul  ST(0), ST(1)
  00175 dd 44 24 14  fld  QWORD PTR _coordinate$[esp+296]
  00179 dd 44 24 28  fld  QWORD PTR $T303561[esp+296]
  0017d d8 c9   fmul  ST(0), ST(1)
  0017f de c2   faddp  ST(2), ST(0)
  00181 d9 ee   fldz
  00183 dc c2   fadd  ST(2), ST(0)
  00185 d9 cb   fxch  ST(3)
  00187 dc 4c 24 30  fmul  QWORD PTR $T303559[esp+296]
  0018b d9 c9   fxch  ST(1)
  0018d dc 8c 24 f0 00
 00 00   fmul  QWORD PTR _rotation$[esp+296]
  00194 de c1   faddp  ST(1), ST(0)
  00196 de c2   faddp  ST(2), ST(0)
  00198 d9 c9   fxch  ST(1)
  0019a dd 5c 24 14  fstp  QWORD PTR _result$[esp+296]
  0019e dd 5c 24 1c  fstp  QWORD PTR _result$[esp+304]

// start of output routine
  001a2 e8 00 00 00 00  call  tuple_delimiter

regards
Andy Little