Re: [eigen] Re: 4x4 matrix inverse

[Hauke Heibel <hauke.heibel@xxxxxxxxxxxxxx>]

On Tue, Dec 15, 2009 at 10:27 AM, Gael Guennebaud <gael.guennebaud@xxxxxxxxx> wrote:

sorry what I said is wrong. So currently we have by default EIGEN_FAST_MATH==1, and defining EIGEN_FAST_MATH==0 disable the vectorized version of sin and cos. If we want to be safe I suggest the following rules:

1-  set EIGEN_FAST_MATH==0 by default

2- if "-ffast-fast" and EIGEN_FAST_MATH is not defined then we set EIGEN_FAST_MATH==1

3- EIGEN_FAST_MATH==0 => no vectorized version of sin and cos and the use of _mm_sqrt_p* for ei_psqrt

4- EIGEN_FAST_MATH==1 => vectorized version of sin and cos + the use of an optimized version of ei_psqrt

Are the vectorized versions of sin and cos less precise? If not, why not always use vectorized sin and cos when vectorization is available and not explicitly disabled. I see ffast-fast and EIGEN_FAST_MATH as indications to go for approximating algorithms. Well, if that is already the case, then forget about this mail.

I am just asking since I think these (http://gruntthepeon.free.fr/ssemath/) SSE versions approximate only slightly but then they are still not IEEE compliant. Maybe in the end you are right and EIGEN_FAST_MATH indicates that we are not guaranteeing IEEE compliance anymore.

- Hauke