[Shootout-list] Organisation for numerical tests

[Sebastien Loisel]

> sounds reasonable to me. this might be because i do not know anything
> about numerical algorithm tests.

Ok, here's some scientific computing tests. Please contribute any of
your suggestions.

First, the most commonly cited numerical benchmark is scimark.
http://www.intel.com/cd/ids/developer/asmo-na/eng/microprocessors/ia32/xeon=
/206112.htm

   1. FFT - performs a complex 1D fast Fourier transform
   2. SOR - solves the Laplace equation in 2D by successive over-relaxation
   3. MC - computes pi by Monte Carlo integration
   4. MM - multiplies two sparse matrices in compressed-row format
   5. LU - computes the LU factorization of a dense N x N matrix

LAPack includes some benchmarks
http://www.netlib.org/lapack/lug/node71.html. They are  basically
* dense LU
* computing all eigenvalues/eigenvectors
* singular value decomposition

The famous linpack benchmark is an LU decomposition.
http://www.netlib.org/benchmark/linpackjava/

Blitz++ has some different benchmarks.
http://www.oonumerics.org/blitz/benchmarks/ including

* pointwise vector operations
http://www.oonumerics.org/blitz/benchmarks/loops.html
* acoustic PDE in 2d
http://www.oonumerics.org/blitz/benchmarks/acoustic.html and 3d.

Some other ideas that aren't in benchmarks I know:

* multigrid
* multipole
* GMRES
* Conjugate Gradient
* Newton iteration
* Automatic differentiation
* ode integration

Sebastien Loisel