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
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
|
/*
* fastmath.c
* Test for new math-routines
*
* Created on: 24.12.2011
* Author: holger
*/
// gcc -o fastmath fastmath.c -lm
#include "stdio.h"
#include <assert.h>
#include <math.h>
#include <time.h>
const float M_PI2 = M_PI*2;
//functions for fastmath without table lookup
/***********************************************************************
* fast sinus implementation *
***********************************************************************/
// schnelle Sinus Implementierung mit Inline Assembler (MASM-style)
//inline float _fastcall fastSinSSE(float x)
//{
// static const float _0_16666 = -1/6.0f;
// static const float _0_00833 = 1/120.0f;
// static const float _0_00019 = -1/5040.0f;
// static const float _0_0000027 = 1/362880.0f;
// _asm
// {
// movss xmm0, x
// comiss xmm0, PI // if (x > PI)
// jbe SHORT ln1
// addss xmm0, PI
// jmp SHORT ln2
//ln1:
// movss xmm1, _PI
// comiss xmm1, xmm0 // if ( x < -PI)
// jbe SHORT ln3
// subss xmm0, PI
//ln2:
// divss xmm0, PI2
// cvttss2si ecx, xmm0 // = z ohne rundung!
// cvtsi2ss xmm1, ecx // z wieder in float umwandeln
// movss xmm0, x
// mulss xmm1, PI2
// subss xmm0, xmm1
//ln3:
// // xmm0 beinhaltet immer das aktuelle ergebnis
// // xmm1 beinhaltet zu jeder zeit x^2
// // xmm2 beinhaltet zu jeder zeit x
// // x + x*x*x*(_0_16666 + x*x*(_0_00833 + x*x*(_0_00019f + _0_0000027*x*x))));
// movss xmm2, xmm0 // x in xmm0 legen
// movss xmm1, xmm0 // x in xmm1 legen
// mulss xmm1, xmm1 // xmm1 = x*x
// movss xmm0, _0_0000027
// mulss xmm0, xmm1 // xmm0 = _0_0000027*x*x
// addss xmm0, _0_00019
// mulss xmm0, xmm1 // xmm0 = x*x*(_0_00019f + _0_0000027*x*x)
// addss xmm0, _0_00833
// mulss xmm0, xmm1 // xmm0 = x*x*(_0_00833 + x*x*(_0_00019f + _0_0000027*x*x))
// addss xmm0, _0_16666
// mulss xmm0, xmm1 // xmm0 = x*x*(_0_16666 + x*x*(_0_00833 + x*x*(_0_00019 + _0_0000027*x*x)))
// mulss xmm0, xmm2
// addss xmm0, xmm2
// movss x, xmm0 // das ergebnis in x rüberschieben
// }
// return x;
//}
inline float fastsin(float x) {
if(M_PI < x) {
x = x-(int)((x+M_PI)/(M_PI2))*M_PI2;
}
else if(x < -M_PI) {
x = x-(int)((x-M_PI)/(M_PI2))*M_PI2;
}
return x*(1 - x*x*(0.16666667f - x*x*(0.00833333f - x*x*(0.0001984f - x*x*0.0000027f))));
}
/***********************************************************************
* fast cosinus implementation *
***********************************************************************/
// schnelle Kosinus Implementierung mit Inline Assembler (MASM-style)
//inline float _fastcall fastCosSSE(float x)
//{
// static const float _0_5 = -1/2.0f;
// static const float _0_0416 = 1/24.0f;
// static const float _0_001387 = -1/720.0f;
// static const float _0_0000248 = 1/40320.0f;
// static const float _0_000000275 = -1/3629000.0f;
// static const float _1 = 1.0f;
// // Wenn x groeßer oder kleiner als PI,
// // dann wird x auf das Intervall -PI bis PI zurückgerechnet
// _asm
// {
// movss xmm0, x
// comiss xmm0, PI // if (x > PI)
// jbe SHORT ln1
// addss xmm0, PI
// jmp SHORT ln2
//ln1:
// movss xmm1, _PI
// comiss xmm1, xmm0 // if ( x < -PI)
// jbe SHORT ln3
// subss xmm0, PI
//ln2:
// divss xmm0, PI2
// cvttss2si ecx, xmm0 // = z ohne rundung!
// cvtsi2ss xmm1, ecx // z wieder in float umwandeln
// movss xmm0, x
// mulss xmm1, PI2
// subss xmm0, xmm1
//ln3:
// // xmm0 beinhaltet immer das aktuelle ergebnis
// // xmm1 beinhaltet zu jeder zeit x^2
// // 1-x*x*(0.5f-x*x*(0.04166667f-x*x*(0.00138889f-x*x*(0.00002480f-x*x*0.000000275f))));
// movss xmm1, xmm0 // x in xmm1 legen
// mulss xmm1, xmm0 // xmm1 = x*x
// movss xmm0, _0_000000275
// mulss xmm0, xmm1 // xmm0 = _0_000000275*x*x
// addss xmm0, _0_0000248
// mulss xmm0, xmm1 // xmm0 = x*x*(_0_0000248 + _0_000000275*x*x)
// addss xmm0, _0_001387
// mulss xmm0, xmm1 // xmm0 = x*x*(_0_001387 + x*x*(_0_0000248 + _0_000000275*x*x))
// addss xmm0, _0_0416
// mulss xmm0, xmm1 // xmm0 = x*x*(_0_0416 + x*x*(_0_001387 + x*x*(_0_0000248 + _0_000000275*x*x)))
// addss xmm0, _0_5
// mulss xmm0, xmm1 // xmm0 = x*x*(_0_5+x*x*(_0_0416 + x*x*(_0_001387 + x*x*(_0_0000248 + _0_000000275*x*x))))
// addss xmm0, _1 // xmm0++
// movss x, xmm0 // das ergebnis in x rüberschieben
// }
// return x;
//}
inline float fastcos(float x) {
if(M_PI < x) {
x = x-(int)((x+M_PI)/(M_PI2))*M_PI2;
}
else if(x < -M_PI) {
x = x-(int)((x-M_PI)/(M_PI2))*M_PI2;
}
return 1-x*x*(0.5f-x*x*(0.04166667f-x*x*(0.00138889f-x*x*(0.00002480f-x*x*0.000000275f))));
}
const long iMaxTests = 10000000;
int main(int argc,char *argv[])
{
long i;
float s, c;
float sc, scr = 0;
unsigned long dwTickStart, dwTickEnd, dwDuration;
double d;
dwTickStart = clock();
for (i = - (iMaxTests/2) ; i < iMaxTests/2 ; i++)
{
d = (double)i;
s = (float)sin(d);
c = (float)cos(d);
// This exist only to force optimiser to not delete code
sc = s * c;
if (sc > scr)
{
scr = sc;
}
}
dwTickEnd = clock();
dwDuration = dwTickEnd - dwTickStart;
printf("%d sin and cos computed in %d ticks with standard math funcs\n", iMaxTests, dwDuration);
float f;
dwTickStart = clock();
for (i = - (iMaxTests/2) ; i < iMaxTests/2 ; i++)
{
f = (float)i;
s = fastsin(f);
c = fastcos(f);
// This exist only to force optimiser to not delete code
sc = s * c;
if (sc > scr)
{
scr = sc;
}
}
dwTickEnd = clock();
dwDuration = dwTickEnd - dwTickStart;
printf("%d sin and cos computed in %d ticks with fast[cos/sin]\n", iMaxTests, dwDuration);
return 0;
}
|
