Re: Speed comparison with Intel compiler

[Toon Moene <toon at moene dot indiv dot nluug dot nl>]

Ignacio Fernández Galván wrote:

> I just tried running a piece of code I'm writing compiled with gfortran
> and with ifort, and the difference is worrying.
>
> ifort: 23 s
> gfortran: 165 s
> gfortran -O3 -ffast-math -funroll-loops -ftree-vectorize
> -march=athlon-xp -msse -mfpmath=sse,387: 164 s
>
> The code has nothing really special, it just converts some coordinates
> and quaternions into transformed coordinates, and the main part I guess
> is generating the rotation matrices and applying them (with MATMUL).
> There is also some (maybe too much) disk I/O)

Quaternions, like so:

There are at least two ways of representing quaternions as matrices,
in such a way that quaternion addition and multiplication correspond to 
matrix addition
and matrix multiplication (i.e., quaternion-matrix homomorphisms).
One is to use 2×2 complex matrices, and the other is to use 4×4 real 
matrices.

?

I betcha the Intel compiler expands the MATMULs inline and unrolls the 
loops completely - and gfortran doesn't.

Nice project for someone who knows the scalarizer.

--
Toon Moene - e-mail: toon@moene.indiv.nluug.nl - phone: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
At home: http://moene.indiv.nluug.nl/~toon/
Who's working on GNU Fortran: 
http://gcc.gnu.org/ml/gcc/2007-01/msg00059.html