-- simple float benchmark ---

gcc3.4.4

16,     35.463203,      46.811821,
32,     91.106583,      48.834839,
64,     101.137962,     50.832504,
128,    105.591942,     51.285336,
256,    106.863571,     51.886494,
512,    107.682907,     52.154556,
1024,   108.634781,     52.469364,
2048,   108.469238,     53.138008,
4096,   105.251945,     52.899765,
8192,   102.261131,     52.217560,
16384,  60.678284,      42.913631,
32768,  57.499565,      41.799500,
65536,  57.266228,      41.547661,

gcc 4.2.1

16,     123.808571,     68.840912,
32,     134.663010,     75.793777,
64,     139.243759,     81.315186,
128,    160.630875,     81.588379,
256,    166.071762,     84.110275,
512,    168.058945,     83.314697,
1024,   164.680878,     83.799026,
2048,   169.331482,     79.088135,
4096,   167.259445,     80.079590,
8192,   153.391693,     78.866806,
16384,  77.417122,      56.227489,
32768,  72.475601,      55.724892,
65536,  70.863876,      55.093391,

options: 

-mabi=aapcs-linux -mfpu=vfp -mfloat-abi=softfp -ffast-math -mcpu=arm1136j-s -mtune=arm1136j-s -O3