Program 1: Random matrix generation (randmat) This program fills a
matrix I with pseudo-random integers, given a seed. The output must be
independent of the number of threads/cores.

Program 2: Histogram thresholding (thresh) Given an integer matrix I
and a target percentage p, this program constructs a boolean matrix B
such that B_ij is set if, and only if, no more than p percent of the
values in I are bigger than I_ij.

Program 3: Weighted point selection (winnow) Each location where a
mask B (a boolean matrix) is true becomes a candidate point, with a
weight equal to the integer value in an integer matrix I at that
location and x and y coordinates equal to its row and column indices.
The program sorts these candidate points into increasing order by
weight, and selects nelts evenly-spaced points to create the result
vector X.

Program 4: Outer product (outer) This program turns a vector X
containing point positions into a dense, symmetric, diagonally
dominant matrix A by calculating the distances between each pair of
points. The diagonal of the matrix is set to the maximum element in
the respective row times the number of elements of the row. It also
constructs a real vector V whose values are the distance of each point
from the origin.

Program 5: Matrix-vector product (product) Given a matrix A and a
vector V, this program calculates the product A V.

Program 6: Chaining (chain) This problem chains all of the previous
problems together.

All the programs should understand a flag "--is_bench". This flag
means that only the size of the inputs is to be read, the inputs
should be generated on-the-fly, and the outputs should not be written
to the screen.

Inputs used for performance test: 20000x20000 matrices (randmat,
thresh, winnow), 1 percent (thresh), 10000 elements (winnow, outer),
10000x10000 matrix (product). Random integers interval is [0, 100)
(thresh). Speedup using 1-8 cores.