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 #include "util/half_float.h"
  34
  35 #ifdef __cplusplus
  36 extern "C" {
  37 #endif
  38
  39 /*
  40  * References for float <-> half conversions
  41  *
  42  *  http://fgiesen.wordpress.com/2012/03/28/half-to-float-done-quic/
  43  *  https://gist.github.com/2156668
  44  *  https://gist.github.com/2144712
  45  */
  46
  47 static inline uint16_t
  48 util_float_to_half(float f)
  49 {
  50    return _mesa_float_to_half(f);
  51 }
  52
  53 static inline uint16_t
  54 util_float_to_half_rtz(float f)
  55 {
  56    uint32_t sign_mask  = 0x80000000;
  57    uint32_t round_mask = ~0xfff;
  58    uint32_t f32inf = 0xff << 23;
  59    uint32_t f16inf = 0x1f << 23;
  60    uint32_t sign;
  61    union fi magic;
  62    union fi f32;
  63    uint16_t f16;
  64
  65    magic.ui = 0xf << 23;
  66
  67    f32.f = f;
  68
  69    /* Sign */
  70    sign = f32.ui & sign_mask;
  71    f32.ui ^= sign;
  72
  73    if (f32.ui == f32inf) {
  74       /* Inf */
  75       f16 = 0x7c00;
  76    } else if (f32.ui > f32inf) {
  77       /* NaN */
  78       f16 = 0x7e00;
  79    } else {
  80       /* Number */
  81       f32.ui &= round_mask;
  82       f32.f  *= magic.f;
  83       f32.ui -= round_mask;
  84       /*
  85        * XXX: The magic mul relies on denorms being available, otherwise
  86        * all f16 denorms get flushed to zero - hence when this is used
  87        * for tgsi_exec in softpipe we won't get f16 denorms.
  88        */
  89       /*
  90        * Clamp to max finite value if overflowed.
  91        * OpenGL has completely undefined rounding behavior for float to
  92        * half-float conversions, and this matches what is mandated for float
  93        * to fp11/fp10, which recommend round-to-nearest-finite too.
  94        * (d3d10 is deeply unhappy about flushing such values to infinity, and
  95        * while it also mandates round-to-zero it doesn't care nearly as much
  96        * about that.)
  97        */
  98       if (f32.ui > f16inf)
  99          f32.ui = f16inf - 1;
 100
 101       f16 = f32.ui >> 13;
 102    }
 103
 104    /* Sign */
 105    f16 |= sign >> 16;
 106
 107    return f16;
 108 }
 109
 110 static inline float
 111 util_half_to_float(uint16_t f16)
 112 {
 113    union fi infnan;
 114    union fi magic;
 115    union fi f32;
 116
 117    infnan.ui = 0x8f << 23;
 118    infnan.f = 65536.0f;
 119    magic.ui  = 0xef << 23;
 120
 121    /* Exponent / Mantissa */
 122    f32.ui = (f16 & 0x7fff) << 13;
 123
 124    /* Adjust */
 125    f32.f *= magic.f;
 126    /* XXX: The magic mul relies on denorms being available */
 127
 128    /* Inf / NaN */
 129    if (f32.f >= infnan.f)
 130       f32.ui |= 0xff << 23;
 131
 132    /* Sign */
 133    f32.ui |= (uint32_t)(f16 & 0x8000) << 16;
 134
 135    return f32.f;
 136 }
 137
 138 #ifdef __cplusplus
 139 }
 140 #endif
 141
 142 #endif /* U_HALF_H */
 143