1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
|
/*
* Mesa 3-D graphics library
*
* Copyright (C) 1999-2007 Brian Paul All Rights Reserved.
* Copyright (C) 2018-2019 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included
* in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
* OTHER DEALINGS IN THE SOFTWARE.
*/
#ifndef _HALF_FLOAT_H_
#define _HALF_FLOAT_H_
#include <stdbool.h>
#include <stdint.h>
#include <string.h>
#include "util/u_cpu_detect.h"
#if defined(USE_X86_64_ASM)
#include <immintrin.h>
#endif
#ifdef __cplusplus
extern "C" {
#endif
#define FP16_ONE ((uint16_t) 0x3c00)
#define FP16_ZERO ((uint16_t) 0)
uint16_t _mesa_float_to_half_slow(float val);
float _mesa_half_to_float_slow(uint16_t val);
uint8_t _mesa_half_to_unorm8(uint16_t v);
uint16_t _mesa_uint16_div_64k_to_half(uint16_t v);
/*
* _mesa_float_to_float16_rtz_slow is no more than a wrapper to the counterpart
* softfloat.h call. Still, softfloat.h conversion API is meant to be kept
* private. In other words, only use the API published here, instead of
* calling directly the softfloat.h one.
*/
uint16_t _mesa_float_to_float16_rtz_slow(float val);
static inline uint16_t
_mesa_float_to_half(float val)
{
#if defined(USE_X86_64_ASM)
if (util_get_cpu_caps()->has_f16c) {
__m128 in = {val};
__m128i out;
/* $0 = round to nearest */
__asm volatile("vcvtps2ph $0, %1, %0" : "=v"(out) : "v"(in));
return out[0];
}
#endif
return _mesa_float_to_half_slow(val);
}
static inline float
_mesa_half_to_float(uint16_t val)
{
#if defined(USE_X86_64_ASM)
if (util_get_cpu_caps()->has_f16c) {
__m128i in = {val};
__m128 out;
__asm volatile("vcvtph2ps %1, %0" : "=v"(out) : "v"(in));
return out[0];
}
#endif
return _mesa_half_to_float_slow(val);
}
static inline uint16_t
_mesa_float_to_float16_rtz(float val)
{
#if defined(USE_X86_64_ASM)
if (util_get_cpu_caps()->has_f16c) {
__m128 in = {val};
__m128i out;
/* $3 = round towards zero (truncate) */
__asm volatile("vcvtps2ph $3, %1, %0" : "=v"(out) : "v"(in));
return out[0];
}
#endif
return _mesa_float_to_float16_rtz_slow(val);
}
static inline uint16_t
_mesa_float_to_float16_rtne(float val)
{
return _mesa_float_to_half(val);
}
static inline bool
_mesa_half_is_negative(uint16_t h)
{
return !!(h & 0x8000);
}
#ifdef __cplusplus
/* Helper class for disambiguating fp16 from uint16_t in C++ overloads */
struct float16_t {
uint16_t bits;
float16_t(float f) : bits(_mesa_float_to_half(f)) {}
float16_t(double d) : bits(_mesa_float_to_half(d)) {}
float16_t(uint16_t raw_bits) : bits(raw_bits) {}
static float16_t one() { return float16_t(FP16_ONE); }
static float16_t zero() { return float16_t(FP16_ZERO); }
};
#endif
#ifdef __cplusplus
} /* extern C */
#endif
#endif /* _HALF_FLOAT_H_ */
|
