Why not use floating-point hardware constants?
Because only a few are available, they are often not the right size, are not available on all processors, and most important are sometimes not accurate enough. Using these constants gives the best chance of portability. Intel X86 floating-point hardware, for example, includes instructions for providing floating-point constants pi and log base 2 (10). These are accurate to 64 significand bits, including guard and sticky bits, which is higher than for any constant read into a register or memory by the compiler, even from a more precise constant decimal digit string. Internally, the Intel X86 floating-point hardware holds all numbers in extended precision real format (64 significand bits compared to external double of 53 significand bits). Extra bits are mainly to support rounding mode correctly. Calculations performed using floating point register to register operations will be performed with this 64-bit accuracy. But operations that use external memory will have lower accuracy. The comp
