GPGPU Programming for Games and Science

40 GPGPU Programming for Games and Science

}

else

{

// x = signaling NaN

}

// payload = t & 0x003fffff

}

}

LISTING 2.8: Decoding a 32-bit floating-point number.

The maximum (unbiased) exponent is e

max

= 127. The minimum (unbi-

ased) exponent is e

min

=1− e

max

= −126. The relationship between the

minimum and maximum exponents is required by the IEEE 754-2008 Stan-

dard. The number of bits in the significand is p = 24, which includes the

twenty-three bits of the trailing significand and the leading 1-valued bit for

normal numbers. The subnormal numbers have a leading 0-valued bit, so the

number of significant bits for subnormals is always smaller than p.

The encoding ...