Double-Precision Inaccuracy

Double-precision calculations are not accurate.

Example (running on an ADSP-21587 compiled with CCES 2.8.3.0):

long double a = 1.16023373864628e-20;

long double ldtemp;

ldtemp = (2.0/3.0) / a;

The processor/compiler yields ldtemp = 5.745968802051255e+19

The correct answer is              ldtemp = 5.745968630808046e+19

The generated result is only accurate to 32 bits.