1 /*
2 * Copyright © 2015 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #ifndef _ROUNDING_H
25 #define _ROUNDING_H
26
27 #include <limits.h>
28 #include <stdint.h>
29 #include <math.h>
30
31 #if defined(__SSE__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 1)) || (defined(_M_X64) && !defined(_M_ARM64EC))
32 #include <xmmintrin.h>
33 #include <emmintrin.h>
34 #endif
35
36 #ifdef __SSE4_1__
37 #include <smmintrin.h>
38 #endif
39
40 /* The C standard library has functions round()/rint()/nearbyint() that round
41 * their arguments according to the rounding mode set in the floating-point
42 * control register. While there are trunc()/ceil()/floor() functions that do
43 * a specific operation without modifying the rounding mode, there is no
44 * roundeven() in any version of C.
45 *
46 * Technical Specification 18661 (ISO/IEC TS 18661-1:2014) adds roundeven(),
47 * but it's unfortunately not implemented by glibc.
48 *
49 * This implementation differs in that it does not raise the inexact exception.
50 *
51 * We use rint() to implement these functions, with the assumption that the
52 * floating-point rounding mode has not been changed from the default Round
53 * to Nearest.
54 */
55
56 /**
57 * \brief Rounds \c x to the nearest integer, with ties to the even integer.
58 */
59 static inline float
_mesa_roundevenf(float x)60 _mesa_roundevenf(float x)
61 {
62 #ifdef __SSE4_1__
63 float ret;
64 __m128 m = _mm_load_ss(&x);
65 m = _mm_round_ss(m, m, _MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC);
66 _mm_store_ss(&ret, m);
67 return ret;
68 #else
69 return rintf(x);
70 #endif
71 }
72
73 /**
74 * \brief Rounds \c x to the nearest integer, with ties to the even integer.
75 */
76 static inline double
_mesa_roundeven(double x)77 _mesa_roundeven(double x)
78 {
79 #ifdef __SSE4_1__
80 double ret;
81 __m128d m = _mm_load_sd(&x);
82 m = _mm_round_sd(m, m, _MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC);
83 _mm_store_sd(&ret, m);
84 return ret;
85 #else
86 return rint(x);
87 #endif
88 }
89
90 /**
91 * \brief Rounds \c x to the nearest integer, with ties to the even integer,
92 * and returns the value as a long int.
93 */
94 static inline long
_mesa_lroundevenf(float x)95 _mesa_lroundevenf(float x)
96 {
97 #if defined(__SSE__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 1)) || (defined(_M_X64) && !defined(_M_ARM64EC))
98 #if LONG_MAX == INT64_MAX
99 return _mm_cvtss_si64(_mm_load_ss(&x));
100 #elif LONG_MAX == INT32_MAX
101 return _mm_cvtss_si32(_mm_load_ss(&x));
102 #else
103 #error "Unsupported long size"
104 #endif
105 #else
106 return lrintf(x);
107 #endif
108 }
109
110 /**
111 * \brief Rounds \c x to the nearest integer, with ties to the even integer,
112 * and returns the value as a long int.
113 */
114 static inline long
_mesa_lroundeven(double x)115 _mesa_lroundeven(double x)
116 {
117 #if defined(__SSE2__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || (defined(_M_X64) && !defined(_M_ARM64EC))
118 #if LONG_MAX == INT64_MAX
119 return _mm_cvtsd_si64(_mm_load_sd(&x));
120 #elif LONG_MAX == INT32_MAX
121 return _mm_cvtsd_si32(_mm_load_sd(&x));
122 #else
123 #error "Unsupported long size"
124 #endif
125 #else
126 return lrint(x);
127 #endif
128 }
129
130 /**
131 * \brief Rounds \c x to the nearest integer, with ties to the even integer,
132 * and returns the value as an int64_t.
133 */
134 static inline int64_t
_mesa_i64roundevenf(float x)135 _mesa_i64roundevenf(float x)
136 {
137 #if LONG_MAX == INT64_MAX
138 return _mesa_lroundevenf(x);
139 #elif LONG_MAX == INT32_MAX
140 return llrintf(x);
141 #else
142 #error "Unsupported long size"
143 #endif
144 }
145
146 #endif
147