compilation:

1. git clone https://github.com/afanasyev-ilya/mars.git

2. cd mars

3. make CXX=nvcc

For non-default CUDA Toolkit install location, please specify CUDA_DIR variable in the makefile.

run:

1. ./bin/mars_cuda_cu -dim 20 -mtx test_mat.csv

to run on matrix with(or without) h vector provided. In the case when h vector not provided, it is set to all zeroes.

2. ./bin/mars_cuda_cu -dim 100

to run on a larger random matrix (100x100 in this example)

3. ./bin/mars_cuda_cu -help

to get information about all available parameters

4. ./bin/mars_cuda_cu -dim 20 -mtx ./cmake-build-debug/test_mat.csv -batch ./cmake-build-debug/batch.txt -nstreams 5 -ngpu 2

to do a run on multiple input parameters {tmin, tmax, cstep, alpha} using the same matrix and h vector.
Batch file has to be provided, example:
0 10 0.001 0.3
0 10 0.001 0.4
0 5 0.001 0.8
10 15 0.001 0.2
10 14 0.001 0.5
20 24 0.001 0.5
0 2 0.0001 0.2
0 12 0.001 0.3
0 1 0.001 0.4
4 8 0.001 0.8
10 12 0.001 0.3
1 7 0.001 0.5
1 2 0.001 0.5
0 2 0.0001 0.3