The ScaLAPACK software relies as much as possible on the BLAS for efficiency and portability. Consequently, the local processor flop rate can be best approximated (from the user's viewpoint) by the performance of the BLAS. The user is therefore strongly urged to use, whenever possible, the most efficient BLAS implementation available. Not using a machine-optimized BLAS implementation may substantially lower the peak flop rate that the hardware can achieve.