Problem with OMAP3530 ARM Floating Point to integer type cast operation - Major Issue

I am using the OMAP3530 to perform some floating point math and I notice that the typecast operation is incorrect on the OMAP. I use the VFPV3 to do floating point and I recompiled the RTS libraries for eabi and VFPV3.

Please look at the following test code -  and the output from using the OMAP3530 and running it on the desktop using the libraries with the GCC compiler

int main()
{
    double ffreq = 4092002.55;
    double fnominalNco = 262.40025;
    double fncoval = ffreq*fnominalNco;
    unsigned int fval = ((double)fncoval+0.5);
    printf("Nco Val %f %f %f %x %d\r\n",
     ffreq,
     fnominalNco,
     fncoval,
     fval,
     fval);

    return 0;
}

Output from GCC:

Nco Val 4092002.550000 262.400250 1073742492.120638 4000029c 1073742492

Output from OMAP3530:

Nco Val 4092002.550000 262.400250 1073742492.120638 40000280 1073742464

Please note the huge discrepancy

I looked at the dissambly code

0x81C54C7C:   EEB71BC2 DD              S2, D2
0x81C54C80:   EEF61A00 VMOVS           S3, #5.000000e-01
0x81C54C84:   EE311A81 FADDS           S2, S3, S2
0x81C54C88:   EEBC1AC1 FTOUIZS         S2, S2
0x81C54C8C:   EE11CA10 FMRS            R12, S2

Looks like it is all done using processor - also of note is that just converting double to the float by itself results in the incorrect result.

Please advise.