Converting from 64-bit int to double fails to use floating point instructions

The customer builds with -mcpu=cortex-r5 -mfpu=vfpv3-d16. The function __floatdidf is called to convert a 64-bit integer to double. The variant of this function used in tiarmclang does not use the floating point instructions available on the VFP3 FPU. If it did, it would be much faster.