half_ Class Reference
This reference page is linked to from the following overview topics: Math and Geometry Utilities.
Detailed Description
Half precision (16 bit) float class.
Supports fast conversion between float and half precision.
Format is compatible with openEXRs half precision type. 1 sign bit, 5 exponent bits (bias = 15), 10 fraction bits.
#include <image.h>
Public Member Functions |
|
| half_ () | |
| half_ (float f) | |
| operator float () const | |
| half_ & | operator= (float f) |
Constructor & Destructor Documentation
| half_ | ( | float | f | ) | [inline] |
Definition at line 41 of file image.h.
{
unsigned int x = *(unsigned int *)&f;
unsigned int frac = (x >> 13) & 0x03ff;
int exp = (int)((x >> 23) & 0xff) - 127;
m_bits = (x >> 16) & 0x8000; // extract the sign bit
if (exp > 16) { // handle overflows/underflows
exp = 16;
frac = 0x03ff;
} else if (exp <= -15) {
frac = 0;
exp = -15;
}
m_bits |= ((exp + 15) << 10) | frac;
}
Member Function Documentation
| operator float | ( | ) | const [inline] |
Definition at line 63 of file image.h.
{
int exp = (m_bits >> 10) & 0x1f;
unsigned int f = (m_bits & 0x8000) << 16; // extract the sign bit
unsigned int frac = (m_bits & 0x03ff) << 13;
if (exp == 0x1f) { // convert 16-bit FP inf/NaN to 32-bit inf/NaN value
exp = 0xff - ExponentBiasDelta;
} else if (exp == 0) {
exp = -ExponentBiasDelta; // convert 16-bit FP zero/denorm to 32-bit zero/denorm value
}
f |= ((exp + ExponentBiasDelta) << 23) | frac;
return *(float *)&f;
}
| half_& operator= | ( | float | f | ) | [inline] |
The documentation for this class was generated from the following file:
