Heterogeneous ScaLAPACK (HeteroScaLAPACK) 1.0.5-BETA

Performance models of the following routines are considerably slimmed down and optimized.
1). {pcgetrf, pdgetrf, psgetrf, pzgetrf}
2). {pcpotrf, pdpotrf, pspotrf, pzpotrf}
3). {pcposv, pdposv, psposv, pzposv}

The benchmark code used to refresh the speeds of the processors now gets the number of nodes and uses it to get the values for the data dimensions to use.

For example, Consider PDGEMM, if the number of available nodes is 6, it determines the processor arrangement as (p,q)=(srqt(6), 6/sqrt(6)), the benchmark code used is multiplication of m/p x b and b x n/q matrices (DGEMM rank-b update).

This package contains an important fix in mpC too. Due to this fix, the communication times are now taken into consideration during the estimation procedure.

Plus there is support for Open MPI.