Lanczos algorithm for Hermitian matrices

Algorithm description : Wikipedia

How to compile

• Change the comments in the Makefile according to the compiler you want.

$ make

• Build the executable for testing.

$ make test

How to run it

Three parameters, matrix dimension n , subspace m and starting $\beta_0 \neq 0$ value

$  ./lanczos  -n [...]  -m [...]  -b [...]

To run the tests, take care to set the value for n to less than 10.

$  ./test  -n [...]

To run the benchmark add -B.

$  ./lanczos  -n [...]  -m [...]  -b [...] -B [benchmark number]

• 1 : Parallel benchmark for fixed matrix size and fixed m with a number of threads ranging from 1 to 40.
• 2 : Sequential benchmark for a fixed matrix size and m ranges from 5 to 50

Plot the result

You can plot the results of multiple runs

• To obtain the graph of a sequential execution with m varying from 5 to 50.

$  cd result
$  python3 plot_m.py [file_1.txt] [file_2.txt] ...

• To obtain the graph of a parallel execution with a number of threads from 1 to 40

$  cd result
$  python3 plot_speedup.py [file_1.txt] [file_2.txt] ...