src/gallium/auxiliary/util/u_half.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2010 Luca Barbieri
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining
   6  * a copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sublicense, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the
  14  * next paragraph) shall be included in all copies or substantial
  15  * portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  18  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  20  * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  21  * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  22  * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  23  * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  *
  25  **************************************************************************/
  26
  27
  28 #ifndef U_HALF_H
  29 #define U_HALF_H
  30
  31 #include "pipe/p_compiler.h"
  32 #include "util/u_math.h"
  33
  34 #ifdef __cplusplus
  35 extern "C" {
  36 #endif
  37
  38 /*
  39  * References for float <-> half conversions
  40  *
  41  *  http://fgiesen.wordpress.com/2012/03/28/half-to-float-done-quic/
  42  *  https://gist.github.com/2156668
  43  *  https://gist.github.com/2144712
  44  */
  45
  46 static inline uint16_t
  47 util_float_to_half(float f)
  48 {
  49    uint32_t sign_mask  = 0x80000000;
  50    uint32_t round_mask = ~0xfff;
  51    uint32_t f32inf = 0xff << 23;
  52    uint32_t f16inf = 0x1f << 23;
  53    uint32_t sign;
  54    union fi magic;
  55    union fi f32;
  56    uint16_t f16;
  57
  58    magic.ui = 0xf << 23;
  59
  60    f32.f = f;
  61
  62    /* Sign */
  63    sign = f32.ui & sign_mask;
  64    f32.ui ^= sign;
  65
  66    if (f32.ui == f32inf) {
  67       /* Inf */
  68       f16 = 0x7c00;
  69    } else if (f32.ui > f32inf) {
  70       /* NaN */
  71       f16 = 0x7e00;
  72    } else {
  73       /* Number */
  74       f32.ui &= round_mask;
  75       f32.f  *= magic.f;
  76       f32.ui -= round_mask;
  77       /*
  78        * XXX: The magic mul relies on denorms being available, otherwise
  79        * all f16 denorms get flushed to zero - hence when this is used
  80        * for tgsi_exec in softpipe we won't get f16 denorms.
  81        */
  82       /*
  83        * Clamp to max finite value if overflowed.
  84        * OpenGL has completely undefined rounding behavior for float to
  85        * half-float conversions, and this matches what is mandated for float
  86        * to fp11/fp10, which recommend round-to-nearest-finite too.
  87        * (d3d10 is deeply unhappy about flushing such values to infinity, and
  88        * while it also mandates round-to-zero it doesn't care nearly as much
  89        * about that.)
  90        */
  91       if (f32.ui > f16inf)
  92          f32.ui = f16inf - 1;
  93
  94       f16 = f32.ui >> 13;
  95    }
  96
  97    /* Sign */
  98    f16 |= sign >> 16;
  99
 100    return f16;
 101 }
 102
 103 static inline float
 104 util_half_to_float(uint16_t f16)
 105 {
 106    union fi infnan;
 107    union fi magic;
 108    union fi f32;
 109
 110    infnan.ui = 0x8f << 23;
 111    infnan.f = 65536.0f;
 112    magic.ui  = 0xef << 23;
 113
 114    /* Exponent / Mantissa */
 115    f32.ui = (f16 & 0x7fff) << 13;
 116
 117    /* Adjust */
 118    f32.f *= magic.f;
 119    /* XXX: The magic mul relies on denorms being available */
 120
 121    /* Inf / NaN */
 122    if (f32.f >= infnan.f)
 123       f32.ui |= 0xff << 23;
 124
 125    /* Sign */
 126    f32.ui |= (f16 & 0x8000) << 16;
 127
 128    return f32.f;
 129 }
 130
 131 #ifdef __cplusplus
 132 }
 133 #endif
 134
 135 #endif /* U_HALF_H */
 136