Use pointer macro to prevent integer oveflow

[luszczek]

An integer overflow occurs and causes segmentation fault
when 2D row/col index is linearized using 32-bit integers.
PBLAS Mptr() macro should be used instead.

This fix only fixes address calculations with row, column, and leading
dimension. There might be other places where an overflow occurs but
they may not matter for practical purposes because they would require
very large matrices or very long vectors within a single process.

[langou]

Thanks @luszczek !