Since SUPERLU packs the system data into Harwell-Boeing format, and since SUPERLU is written in C, separate benchmark utilities were developed for benchmarking OPTIMQR and SUPERLU.
The two programs are as similar as they can possibly be, given the different functionalities of Fortran and C. The programs read the system definition from the regress.system file, generates a system matrix where the non-zero elements corrospond to the values from a Hilbert matrix, the right hand side matrix contains random numbers.
First the system is solved once, and the result is validated (calculating the residual). After that, two systems containing random numbers are solved 10.000 times each.
Simply measuring the elapsed time of the benchmark program will give us a reasonable performance number. Because of the large number of times the systems are solved, the initial matrix setup and validation will be insignificant in time compared to the repeated solutions.