This is a MAGMA 0.3 SGEMM Routine for Fermi GPUs.
In this version sizes have to be divisible by 96.

Results for C2050 
[peak is 1030 GFlop/s, achieved up to 639 GFlop/s, i.e.,62% of peak]
====================================================================


    N	magmablas0.3 GFLops/s	cudablas-3.1 GFlops/s    error
===================================================================
  864        541.31141              381.75350         0.000000e+00
 1440        580.47900              549.29801         0.000000e+00
 2016        608.41614              398.67361         0.000000e+00
 2592        618.15163              406.85144         0.000000e+00
 3168        628.97641              408.18766         0.000000e+00
 3744        631.77247              409.44035         0.000000e+00
 4320        633.38166              582.16613         0.000000e+00
 4896        634.55076              410.75138         0.000000e+00
 5472        633.91444              411.12302         0.000000e+00
 6048        635.32870              409.17803         0.000000e+00


Results for GTX480
[peak is 1344 GFlop/s, achieved up to 835 GFlop/s, i.e., 62% of peak]
on ig
    N	magmablas0.3 GFLops/s	cudablas-3.1 GFlops/s    error
=====================================================================
  864	653.136753417721	483.850370592648	0.000000e+00
 1728	788.354522841864	535.357994604690	0.000000e+00
 2592	814.969051291651	532.619433500023	0.000000e+00
 3456	821.187924681448	540.960256023485	0.000000e+00
 4320	829.939500625376	761.018774955399	0.000000e+00
 5184	829.171445005238	541.043372321028	0.000000e+00
 6048	831.286043156492	534.770138504436	0.000000e+00
 6912	832.299197199339	541.931472106343	0.000000e+00
 7776	833.684231704949	539.058926429062	0.000000e+00
 8640	834.519990192355	763.036403709580	0.000000e+00
