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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
|
/* Test of rounding towards zero.
Copyright (C) 2007-2010 Free Software Foundation, Inc.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>. */
/* Written by Bruno Haible <bruno@clisp.org>, 2007. */
#include <config.h>
#include <math.h>
#include <float.h>
#include <stdbool.h>
#include <stdint.h>
#include <stdio.h>
#include "isnand-nolibm.h"
#include "macros.h"
/* The reference implementation, taken from lib/trunc.c. */
#define DOUBLE double
#define MANT_DIG DBL_MANT_DIG
#define L_(literal) literal
/* 2^(MANT_DIG-1). */
static const DOUBLE TWO_MANT_DIG =
/* Assume MANT_DIG <= 5 * 31.
Use the identity
n = floor(n/5) + floor((n+1)/5) + ... + floor((n+4)/5). */
(DOUBLE) (1U << ((MANT_DIG - 1) / 5))
* (DOUBLE) (1U << ((MANT_DIG - 1 + 1) / 5))
* (DOUBLE) (1U << ((MANT_DIG - 1 + 2) / 5))
* (DOUBLE) (1U << ((MANT_DIG - 1 + 3) / 5))
* (DOUBLE) (1U << ((MANT_DIG - 1 + 4) / 5));
DOUBLE
trunc_reference (DOUBLE x)
{
/* The use of 'volatile' guarantees that excess precision bits are dropped
at each addition step and before the following comparison at the caller's
site. It is necessary on x86 systems where double-floats are not IEEE
compliant by default, to avoid that the results become platform and compiler
option dependent. 'volatile' is a portable alternative to gcc's
-ffloat-store option. */
volatile DOUBLE y = x;
volatile DOUBLE z = y;
if (z > L_(0.0))
{
/* Avoid rounding errors for values near 2^k, where k >= MANT_DIG-1. */
if (z < TWO_MANT_DIG)
{
/* Round to the next integer (nearest or up or down, doesn't matter). */
z += TWO_MANT_DIG;
z -= TWO_MANT_DIG;
/* Enforce rounding down. */
if (z > y)
z -= L_(1.0);
}
}
else if (z < L_(0.0))
{
/* Avoid rounding errors for values near -2^k, where k >= MANT_DIG-1. */
if (z > - TWO_MANT_DIG)
{
/* Round to the next integer (nearest or up or down, doesn't matter). */
z -= TWO_MANT_DIG;
z += TWO_MANT_DIG;
/* Enforce rounding up. */
if (z < y)
z += L_(1.0);
}
}
return z;
}
/* Test for equality. */
static int
equal (DOUBLE x, DOUBLE y)
{
return (isnand (x) ? isnand (y) : x == y);
}
/* Test whether the result for a given argument is correct. */
static bool
correct_result_p (DOUBLE x, DOUBLE result)
{
return
(x >= 0
? (x < 1 ? result == L_(0.0) :
x - 1 < x ? result <= x && result >= x - 1 && x - result < 1 :
equal (result, x))
: (x > -1 ? result == L_(0.0) :
x + 1 > x ? result >= x && result <= x + 1 && result - x < 1 :
equal (result, x)));
}
/* Test the function for a given argument. */
static int
check (double x)
{
/* If the reference implementation is incorrect, bail out immediately. */
double reference = trunc_reference (x);
ASSERT (correct_result_p (x, reference));
/* If the actual implementation is wrong, return an error code. */
{
double result = trunc (x);
if (correct_result_p (x, result))
return 0;
else
{
fprintf (stderr, "trunc %g(%a) = %g(%a) or %g(%a)?\n",
x, x, reference, reference, result, result);
return 1;
}
}
}
#define NUM_HIGHBITS 13
#define NUM_LOWBITS 4
int
main ()
{
unsigned int highbits;
unsigned int lowbits;
int error = 0;
for (highbits = 0; highbits < (1 << NUM_HIGHBITS); highbits++)
for (lowbits = 0; lowbits < (1 << NUM_LOWBITS); lowbits++)
{
/* Combine highbits and lowbits into a floating-point number,
sign-extending the lowbits to 32-NUM_HIGHBITS bits. */
union { double f; uint64_t i; } janus;
janus.i = ((uint64_t) highbits << (64 - NUM_HIGHBITS))
| ((uint64_t) ((int64_t) ((uint64_t) lowbits << (64 - NUM_LOWBITS))
>> (64 - NUM_LOWBITS - NUM_HIGHBITS))
>> NUM_HIGHBITS);
error |= check (janus.f);
}
return (error ? 1 : 0);
}
|
