Benchmark based on DGEMM iterations in Jansik set

Fix convergence problem in y term, use e*e' instead of I for scalar matrix addition.

Fix convergence problem in y term, use e*e' instead of I for scalar matrix adwdition

Fix convergence problem in y term, use e*e' instead of I for scalar matrix adwdition

Fix the Gflop/s performance calculation

Fix rng to private in the openmp