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
|
/* OpenCL built-in library: exp2()
Copyright (c) 2015 Michal Babej / Tampere University of Technology
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.
*/
#include "hsail_templates.h"
#include "vml_constants.h"
float _CL_OVERLOADABLE exp2(float x)
{
// TODO: Check SLEEF 2.80 algorithm
// (in particular the improved-precision truncation)
// Rescale
float x0 = x;
// Round by adding, then subtracting again a large number
// Add a large number to move the mantissa bits to the right
int large = ((int)1 << PROPS_FLOAT_MANTISSA_BITS) + PROPS_FLOAT_EXPONENT_OFFSET;
float tmp = x + (float)large;
// tmp.barrier();
__asm__ ("");
float round_x = tmp - (float)large;
x -= round_x;
//VML_ASSERT(all(x >= (float)(-0.5) && x <= (float)(0.5)));
// Polynomial expansion
float r;
#ifdef VML_HAVE_FP_CONTRACT
// float, error=4.55549108005200277750378992345e-9
r = (float)(0.000154653240842602623787395880898);
r = mad(r, x, (float)(0.00133952915439234389712105060319));
r = mad(r, x, (float)(0.0096180399118156827664944870552));
r = mad(r, x, (float)(0.055503406540531310853149866446));
r = mad(r, x, (float)(0.240226511015459465468737123346));
r = mad(r, x, (float)(0.69314720007380208630542805293));
r = mad(r, x, (float)(0.99999999997182023878745628977));
#else
// float, error=1.62772721960621336664735896836e-7
r = (float)(0.00133952915439234389712105060319);
r = mad(r, x, (float)(0.009670773148229417605024318985));
r = mad(r, x, (float)(0.055503406540531310853149866446));
r = mad(r, x, (float)(0.240222115700585316818177639177));
r = mad(r, x, (float)(0.69314720007380208630542805293));
r = mad(r, x, (float)(1.00000005230745711373079206024));
#endif
// Use direct integer manipulation
// Extract integer as lowest mantissa bits (highest bits still
// contain offset, exponent, and sign)
int itmp = as_int(tmp);
// Construct scale factor by setting exponent (this shifts out the
// highest bits)
float scale = as_float(itmp << PROPS_FLOAT_MANTISSA_BITS);
r *= scale;
if (x0 < (float)PROPS_FLOAT_MIN_EXPONENT)
return 0.0f;
else
return r;
}
double _CL_OVERLOADABLE exp2(double x)
{
// TODO: Check SLEEF 2.80 algorithm
// (in particular the improved-precision truncation)
// Rescale
double x0 = x;
// Round by adding, then subtracting again a large number
// Add a large number to move the mantissa bits to the right
long large = ((long)1 << PROPS_DOUBLE_MANTISSA_BITS) + PROPS_DOUBLE_EXPONENT_OFFSET;
double tmp = x + (double)large;
// tmp.barrier();
__asm__ ("");
double round_x = tmp - (double)large;
x -= round_x;
// Polynomial expansion
double r;
#ifdef VML_HAVE_FP_CONTRACT
// double, error=9.32016781355638010975628074746e-18
r = (double)(4.45623165388261696886670014471e-10);
r = mad(r, x, (double)(7.0733589360775271430968224806e-9));
r = mad(r, x, (double)(1.01780540270960163558119510246e-7));
r = mad(r, x, (double)(1.3215437348041505269462510712e-6));
r = mad(r, x, (double)(0.000015252733849766201174247690629));
r = mad(r, x, (double)(0.000154035304541242555115696403795));
r = mad(r, x, (double)(0.00133335581463968601407096905671));
r = mad(r, x, (double)(0.0096181291075949686712855561931));
r = mad(r, x, (double)(0.055504108664821672870565883052));
r = mad(r, x, (double)(0.240226506959101382690753994082));
r = mad(r, x, (double)(0.69314718055994530864272481773));
r = mad(r, x, (double)(0.9999999999999999978508676375));
#else
// double, error=3.74939899823302048807873981077e-14
r = (double)(1.02072375599725694063203809188e-7);
r = mad(r, x, (double)(1.32573274434801314145133004073e-6));
r = mad(r, x, (double)(0.0000152526647170731944840736190013));
r = mad(r, x, (double)(0.000154034441925859828261898614555));
r = mad(r, x, (double)(0.00133335582175770747495287552557));
r = mad(r, x, (double)(0.0096181291794939392517233403183));
r = mad(r, x, (double)(0.055504108664525029438908798685));
r = mad(r, x, (double)(0.240226506957026959772247598695));
r = mad(r, x, (double)(0.6931471805599487321347668143));
r = mad(r, x, (double)(1.00000000000000942892870993489));
#endif
// Use direct integer manipulation
// Extract integer as lowest mantissa bits (highest bits still
// contain offset, exponent, and sign)
long itmp = as_long(tmp);
// Construct scale factor by setting exponent (this shifts out the
// highest bits)
double scale = as_double(itmp << PROPS_DOUBLE_MANTISSA_BITS);
r *= scale;
if (x0 < (float)PROPS_DOUBLE_MIN_EXPONENT)
return 0.0;
else
return r;
}
IMPLEMENT_VECWITHSCALARS(exp2, V_V, float, int)
IMPLEMENT_VECWITHSCALARS(exp2, V_V, double, int)
|
