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 "c99_math.h"
28 
29 #include <limits.h>
30 #include <stdint.h>
31 
32 #if defined(__SSE__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 1)) || defined(_M_X64)
33 #include <xmmintrin.h>
34 #include <emmintrin.h>
35 #endif
36 
37 #ifdef __SSE4_1__
38 #include <smmintrin.h>
39 #endif
40 
41 /* The C standard library has functions round()/rint()/nearbyint() that round
42  * their arguments according to the rounding mode set in the floating-point
43  * control register. While there are trunc()/ceil()/floor() functions that do
44  * a specific operation without modifying the rounding mode, there is no
45  * roundeven() in any version of C.
46  *
47  * Technical Specification 18661 (ISO/IEC TS 18661-1:2014) adds roundeven(),
48  * but it's unfortunately not implemented by glibc.
49  *
50  * This implementation differs in that it does not raise the inexact exception.
51  *
52  * We use rint() to implement these functions, with the assumption that the
53  * floating-point rounding mode has not been changed from the default Round
54  * to Nearest.
55  */
56 
57 /**
58  * \brief Rounds \c x to the nearest integer, with ties to the even integer.
59  */
60 static inline float
_mesa_roundevenf(float x)61 _mesa_roundevenf(float x)
62 {
63 #ifdef __SSE4_1__
64    float ret;
65    __m128 m = _mm_load_ss(&x);
66    m = _mm_round_ss(m, m, _MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC);
67    _mm_store_ss(&ret, m);
68    return ret;
69 #else
70    return rintf(x);
71 #endif
72 }
73 
74 /**
75  * \brief Rounds \c x to the nearest integer, with ties to the even integer.
76  */
77 static inline double
_mesa_roundeven(double x)78 _mesa_roundeven(double x)
79 {
80 #ifdef __SSE4_1__
81    double ret;
82    __m128d m = _mm_load_sd(&x);
83    m = _mm_round_sd(m, m, _MM_FROUND_CUR_DIRECTION | _MM_FROUND_NO_EXC);
84    _mm_store_sd(&ret, m);
85    return ret;
86 #else
87    return rint(x);
88 #endif
89 }
90 
91 /**
92  * \brief Rounds \c x to the nearest integer, with ties to the even integer,
93  * and returns the value as a long int.
94  */
95 static inline long
_mesa_lroundevenf(float x)96 _mesa_lroundevenf(float x)
97 {
98 #if defined(__SSE__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 1)) || defined(_M_X64)
99 #if LONG_MAX == INT64_MAX
100    return _mm_cvtss_si64(_mm_load_ss(&x));
101 #elif LONG_MAX == INT32_MAX
102    return _mm_cvtss_si32(_mm_load_ss(&x));
103 #else
104 #error "Unsupported long size"
105 #endif
106 #else
107    return lrintf(x);
108 #endif
109 }
110 
111 
112 /**
113  * \brief Rounds \c x to the nearest integer, with ties to the even integer,
114  * and returns the value as a long int.
115  */
116 static inline long
_mesa_lroundeven(double x)117 _mesa_lroundeven(double x)
118 {
119 #if defined(__SSE2__) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64)
120 #if LONG_MAX == INT64_MAX
121    return _mm_cvtsd_si64(_mm_load_sd(&x));
122 #elif LONG_MAX == INT32_MAX
123    return _mm_cvtsd_si32(_mm_load_sd(&x));
124 #else
125 #error "Unsupported long size"
126 #endif
127 #else
128    return lrint(x);
129 #endif
130 }
131 
132 /**
133  * \brief Rounds \c x to the nearest integer, with ties to the even integer,
134  * and returns the value as an int64_t.
135  */
136 static inline int64_t
_mesa_i64roundevenf(float x)137 _mesa_i64roundevenf(float x)
138 {
139 #if LONG_MAX == INT64_MAX
140    return _mesa_lroundevenf(x);
141 #elif LONG_MAX == INT32_MAX
142    return llrintf(x);
143 #else
144 #error "Unsupported long size"
145 #endif
146 }
147 
148 #endif
149