[Bf-committers] arithb.c QuatToMat3 question

[Arne Schmitz]

Am Sonntag, 17. September 2006 07:36 schrieb Tom M:
> In QuatToMat3 and QuatToMat4
>
> we see q0= M_SQRT2 * q[0];
> 	q1= M_SQRT2 * q[1];
> 	q2= M_SQRT2 * q[2];
> 	q3= M_SQRT2 * q[3];
>
> this appears to be done to later save multiplications by two in the
> following
>
> 	qda= q0*q1;
> 	qdb= q0*q2;
> 	qdc= q0*q3;
> 	qaa= q1*q1;
> 	qab= q1*q2;
> 	qac= q1*q3;
> 	qbb= q2*q2;
> 	qbc= q2*q3;
> 	qcc= q3*q3;
>
> however I would think that if we leave it as multiply by 2, then the
> compiler will recognize it as a bitshift, which I would think would
> end up ultimately faster.

Floats cannot be bit-shifted. Nevertheless is the above code not optimal, 
since it is prone to rounding errors. Deleting the first four line and then 
writing:

  qda = 2.0f * q[0] * q[1];
  ...

should be more stable.

Cheers,

Arne

-- 
Dipl.-Inform. Arne Schmitz              Phone   +49 (0)241 80-21817
Computer Graphics Group                 Fax     +49 (0)241 80-22899
RWTH Aachen University                  http://www.rwth-graphics.de
Ahornstrasse 55, 52074 Aachen, Germany
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://projects.blender.org/pipermail/bf-committers/attachments/20060917/26d339d7/attachment.pgp