1 |
/*************************************************************************** |
2 |
* * |
3 |
* LinuxSampler - modular, streaming capable sampler * |
4 |
* * |
5 |
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
6 |
* Copyright (C) 2005 Christian Schoenebeck * |
7 |
* * |
8 |
* This program is free software; you can redistribute it and/or modify * |
9 |
* it under the terms of the GNU General Public License as published by * |
10 |
* the Free Software Foundation; either version 2 of the License, or * |
11 |
* (at your option) any later version. * |
12 |
* * |
13 |
* This program is distributed in the hope that it will be useful, * |
14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of * |
15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
16 |
* GNU General Public License for more details. * |
17 |
* * |
18 |
* You should have received a copy of the GNU General Public License * |
19 |
* along with this program; if not, write to the Free Software * |
20 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, * |
21 |
* MA 02111-1307 USA * |
22 |
***************************************************************************/ |
23 |
|
24 |
#ifndef __RT_MATH_H__ |
25 |
#define __RT_MATH_H__ |
26 |
|
27 |
#include <math.h> |
28 |
#include <stdint.h> |
29 |
#include "global.h" |
30 |
|
31 |
/// Needed for calculating frequency ratio used to pitch a sample |
32 |
#define TWELVEHUNDREDTH_ROOT_OF_TWO 1.000577789506555 |
33 |
|
34 |
enum implementation_t { |
35 |
CPP, |
36 |
ASM_X86_MMX_SSE |
37 |
}; |
38 |
|
39 |
/** @brief Real Time Math Base Class |
40 |
* |
41 |
* Math functions for real time operation. This base class contains all |
42 |
* non-template methods. |
43 |
*/ |
44 |
class RTMathBase { |
45 |
public: |
46 |
/** |
47 |
* Highly accurate time stamp. |
48 |
*/ |
49 |
typedef uint32_t time_stamp_t; |
50 |
|
51 |
/** |
52 |
* We read the processor's cycle count register as a reference |
53 |
* for the real time. These are of course only abstract values |
54 |
* with arbitrary time entity, but that's not a problem as long |
55 |
* as we calculate relatively. |
56 |
*/ |
57 |
static time_stamp_t CreateTimeStamp(); |
58 |
|
59 |
/** |
60 |
* Calculates the frequency ratio for a pitch value given in cents |
61 |
* (assuming equal tempered scale of course, divided into 12 |
62 |
* semitones per octave and 100 cents per semitone). |
63 |
* |
64 |
* Note: CONFIG_MAX_PITCH (defined in config.h) has to be defined to an |
65 |
* appropriate value, otherwise the behavior of this function is |
66 |
* undefined, but most probably if CONFIG_MAX_PITCH is too small, the |
67 |
* application will crash due to segmentation fault here. |
68 |
* |
69 |
* @param cents - pitch value in cents (+1200 cents means +1 octave) |
70 |
* @returns frequency ratio (e.g. +2.0 for +1 octave) |
71 |
*/ |
72 |
inline static double CentsToFreqRatio(double Cents) { |
73 |
int index_int = (int) (Cents); // integer index |
74 |
float index_fract = Cents - index_int; // fractional part of index |
75 |
return pCentsToFreqTable[index_int] + index_fract * (pCentsToFreqTable[index_int+1] - pCentsToFreqTable[index_int]); |
76 |
} |
77 |
|
78 |
private: |
79 |
static float CentsToFreqTable[CONFIG_MAX_PITCH * 1200 * 2 + 1]; |
80 |
static float* pCentsToFreqTable; |
81 |
|
82 |
static float* InitCentsToFreqTable(); |
83 |
}; |
84 |
|
85 |
/** @brief Real Time Math |
86 |
* |
87 |
* This is a template which provides customized methods for the desired low |
88 |
* level implementation. The ASM_X86_MMX_SSE implementation of each method |
89 |
* for example doesn't use 387 FPU instruction. This is needed for MMX |
90 |
* algorithms which do not allow mixed MMX and 387 instructions. |
91 |
*/ |
92 |
template<implementation_t IMPL = CPP> |
93 |
class __RTMath : public RTMathBase { |
94 |
public: |
95 |
// conversion using truncate |
96 |
inline static int Int(const float a) { |
97 |
switch (IMPL) { |
98 |
#if CONFIG_ASM && ARCH_X86 |
99 |
case ASM_X86_MMX_SSE: { |
100 |
int ret; |
101 |
asm ( |
102 |
"cvttss2si %1, %0 # convert to int\n\t" |
103 |
: "=r" (ret) |
104 |
: "m" (a) |
105 |
); |
106 |
return ret; |
107 |
} |
108 |
#endif // CONFIG_ASM && ARCH_X86 |
109 |
default: { |
110 |
return (int) a; |
111 |
} |
112 |
} |
113 |
} |
114 |
|
115 |
//for doubles and everything else except floats |
116 |
template<class T_a> inline static int Int(const T_a a) { |
117 |
return (int) a; |
118 |
} |
119 |
|
120 |
inline static float Float(const int a) { |
121 |
switch (IMPL) { |
122 |
#if CONFIG_ASM && ARCH_X86 |
123 |
case ASM_X86_MMX_SSE: { |
124 |
float ret; |
125 |
asm ( |
126 |
"cvtsi2ss %1, %%xmm0 # convert to float\n\t" |
127 |
"movss %%xmm0,%0 # output\n\t" |
128 |
: "=m" (ret) |
129 |
: "r" (a) |
130 |
); |
131 |
return ret; |
132 |
} |
133 |
#endif // CONFIG_ASM && ARCH_X86 |
134 |
default: { |
135 |
return (float) a; |
136 |
} |
137 |
} |
138 |
} |
139 |
|
140 |
#if 0 |
141 |
//for everything except ints |
142 |
template<class T_a> inline static float Float(T_a a) { |
143 |
return (float) a; |
144 |
} |
145 |
#endif |
146 |
|
147 |
inline static float Sum(const float& a, const float& b) { |
148 |
switch (IMPL) { |
149 |
#if CONFIG_ASM && ARCH_X86 |
150 |
case ASM_X86_MMX_SSE: { |
151 |
float ret; |
152 |
asm ( |
153 |
"movss %1, %%xmm0 # load a\n\t" |
154 |
"addss %2, %%xmm0 # a + b\n\t" |
155 |
"movss %%xmm0, %0 # output\n\t" |
156 |
: "=m" (ret) |
157 |
: "m" (a), "m" (b) |
158 |
); |
159 |
return ret; |
160 |
} |
161 |
#endif // CONFIG_ASM && ARCH_X86 |
162 |
default: { |
163 |
return (a + b); |
164 |
} |
165 |
} |
166 |
} |
167 |
|
168 |
template<class T_a, class T_b> inline static T_a Sum(const T_a a, const T_b b) { |
169 |
return (a + b); |
170 |
} |
171 |
|
172 |
inline static float Sub(const float& a, const float& b) { |
173 |
switch (IMPL) { |
174 |
#if CONFIG_ASM && ARCH_X86 |
175 |
case ASM_X86_MMX_SSE: { |
176 |
float ret; |
177 |
asm ( |
178 |
"movss %1, %%xmm0 # load a\n\t" |
179 |
"subss %2, %%xmm0 # a - b\n\t" |
180 |
"movss %%xmm0, %0 # output\n\t" |
181 |
: "=m" (ret) |
182 |
: "m" (a), "m" (b) |
183 |
); |
184 |
return ret; |
185 |
} |
186 |
#endif // CONFIG_ASM && ARCH_X86 |
187 |
default: { |
188 |
return (a - b); |
189 |
} |
190 |
} |
191 |
} |
192 |
|
193 |
template<class T_a, class T_b> inline static T_a Sub(const T_a a, const T_b b) { |
194 |
return (a - b); |
195 |
} |
196 |
|
197 |
inline static float Mul(const float a, const float b) { |
198 |
switch (IMPL) { |
199 |
#if CONFIG_ASM && ARCH_X86 |
200 |
case ASM_X86_MMX_SSE: { |
201 |
float ret; |
202 |
asm ( |
203 |
"movss %1, %%xmm0 # load a\n\t" |
204 |
"mulss %2, %%xmm0 # a * b\n\t" |
205 |
"movss %%xmm0, %0 # output\n\t" |
206 |
: "=m" (ret) |
207 |
: "m" (a), "m" (b) |
208 |
); |
209 |
return ret; |
210 |
} |
211 |
#endif // CONFIG_ASM && ARCH_X86 |
212 |
default: { |
213 |
return (a * b); |
214 |
} |
215 |
} |
216 |
} |
217 |
|
218 |
template<class T_a, class T_b> inline static T_a Mul(const T_a a, const T_b b) { |
219 |
return (a * b); |
220 |
} |
221 |
|
222 |
inline static float Div(const float a, const float b) { |
223 |
switch (IMPL) { |
224 |
#if CONFIG_ASM && ARCH_X86 |
225 |
case ASM_X86_MMX_SSE: { |
226 |
float ret; |
227 |
asm ( |
228 |
"movss %1, %%xmm0 # load a\n\t" |
229 |
"divss %2, %%xmm0 # a / b\n\t" |
230 |
"movss %%xmm0, %0 # output\n\t" |
231 |
: "=m" (ret) |
232 |
: "m" (a), "m" (b) |
233 |
); |
234 |
return ret; |
235 |
} |
236 |
#endif // CONFIG_ASM && ARCH_X86 |
237 |
default: { |
238 |
return (a / b); |
239 |
} |
240 |
} |
241 |
} |
242 |
|
243 |
template<class T_a, class T_b> inline static T_a Div(const T_a a, const T_b b) { |
244 |
return (a / b); |
245 |
} |
246 |
|
247 |
inline static float Min(const float a, const float b) { |
248 |
switch (IMPL) { |
249 |
#if CONFIG_ASM && ARCH_X86 |
250 |
case ASM_X86_MMX_SSE: { |
251 |
float ret; |
252 |
asm ( |
253 |
"movss %1, %%xmm0 # load a\n\t" |
254 |
"minss %2, %%xmm0 # Minimum(a, b)\n\t" |
255 |
"movss %%xmm0, %0 # output\n\t" |
256 |
: "=m" (ret) |
257 |
: "m" (a), "m" (b) |
258 |
); |
259 |
return ret; |
260 |
} |
261 |
#endif // CONFIG_ASM && ARCH_X86 |
262 |
default: { |
263 |
return std::min(a, b); |
264 |
} |
265 |
} |
266 |
} |
267 |
|
268 |
template<class T_a, class T_b> inline static T_a Min(const T_a a, const T_b b) { |
269 |
return (b < a) ? b : a; |
270 |
} |
271 |
|
272 |
inline static float Max(const float a, const float b) { |
273 |
switch (IMPL) { |
274 |
#if CONFIG_ASM && ARCH_X86 |
275 |
case ASM_X86_MMX_SSE: { |
276 |
float ret; |
277 |
asm ( |
278 |
"movss %1, %%xmm0 # load a\n\t" |
279 |
"maxss %2, %%xmm0 # Maximum(a, b)\n\t" |
280 |
"movss %%xmm0, %0 # output\n\t" |
281 |
: "=m" (ret) |
282 |
: "m" (a), "m" (b) |
283 |
); |
284 |
return ret; |
285 |
} |
286 |
#endif // CONFIG_ASM && ARCH_X86 |
287 |
default: { |
288 |
return std::max(a, b); |
289 |
} |
290 |
} |
291 |
} |
292 |
|
293 |
template<class T_a, class T_b> inline static T_a Max(const T_a a, const T_b b) { |
294 |
return (b > a) ? b : a; |
295 |
} |
296 |
|
297 |
inline static float Fmodf(const float &a, const float &b) { |
298 |
switch (IMPL) { |
299 |
#if CONFIG_ASM && ARCH_X86 |
300 |
case ASM_X86_MMX_SSE: { |
301 |
float ret; |
302 |
asm ( |
303 |
"movss %1, %%xmm0 # load a\n\t" |
304 |
"movss %2, %%xmm1 # load b\n\t" |
305 |
"movss %%xmm0,%%xmm2\n\t" |
306 |
"divss %%xmm1, %%xmm2 # xmm2 = a / b\n\t" |
307 |
"cvttss2si %%xmm2, %%ecx #convert to int\n\t" |
308 |
"cvtsi2ss %%ecx, %%xmm2 #convert back to float\n\t" |
309 |
"mulss %%xmm1, %%xmm2 # xmm2 = b * int(a/b)\n\t" |
310 |
"subss %%xmm2, %%xmm0 #sub a\n\t" |
311 |
"movss %%xmm0, %0 # output\n\t" |
312 |
: "=m" (ret) |
313 |
: "m" (a), "m" (b) |
314 |
: "%ecx" |
315 |
); |
316 |
return ret; |
317 |
} |
318 |
#endif // CONFIG_ASM && ARCH_X86 |
319 |
default: { |
320 |
return fmodf(a, b); |
321 |
} |
322 |
} |
323 |
} |
324 |
}; |
325 |
|
326 |
/// convenience typedef for using the default implementation (which is CPP) |
327 |
typedef __RTMath<> RTMath; |
328 |
|
329 |
#endif // __RT_MATH_H__ |