1 |
/*************************************************************************** |
2 |
* * |
3 |
* LinuxSampler - modular, streaming capable sampler * |
4 |
* * |
5 |
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
6 |
* Copyright (C) 2005 - 2007 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 |
// Note: the assembly code is currently disabled, as it doesn't fit into |
25 |
// the new synthesis core introduced by LS 0.4.0 |
26 |
|
27 |
#ifndef __LS_BIQUADFILTER_H__ |
28 |
#define __LS_BIQUADFILTER_H__ |
29 |
|
30 |
#include <math.h> |
31 |
|
32 |
#include "../../common/global.h" |
33 |
|
34 |
/// ln(2) / 2 |
35 |
#define LN_2_2 0.34657359f |
36 |
|
37 |
#ifndef LIMIT |
38 |
# define LIMIT(v,l,u) (v < l ? l : (v > u ? u : v)) |
39 |
#endif |
40 |
|
41 |
namespace LinuxSampler { |
42 |
|
43 |
typedef float bq_t; |
44 |
|
45 |
/** |
46 |
* Internal parameters of the biquad filter, which are actually the |
47 |
* final parameters of the filter's transfer function. This strucure is |
48 |
* only needed when these parameters should stored outside the |
49 |
* BiquadFilter class, e.g. to save calculation time by sharing them |
50 |
* between multiple filters. |
51 |
*/ |
52 |
struct biquad_param_t { |
53 |
bq_t b0; |
54 |
bq_t b1; |
55 |
bq_t b2; |
56 |
bq_t a1; |
57 |
bq_t a2; |
58 |
}; |
59 |
|
60 |
/** |
61 |
* Bi-quadratic filter |
62 |
* (adapted from lisp code by Eli Brandt, http://www.cs.cmu.edu/~eli/) |
63 |
*/ |
64 |
class BiquadFilter { |
65 |
protected: |
66 |
// following five variables are only used if no external biquad_param_t reference is used |
67 |
bq_t b0; |
68 |
bq_t b1; |
69 |
bq_t b2; |
70 |
bq_t a1; |
71 |
bq_t a2; |
72 |
// following four variables are used to buffer the feedback |
73 |
bq_t x1; |
74 |
bq_t x2; |
75 |
bq_t y1; |
76 |
bq_t y2; |
77 |
|
78 |
#if __GNUC__ >= 4 |
79 |
float fbc; |
80 |
#else |
81 |
const static float fbc = 0.98; |
82 |
#endif |
83 |
|
84 |
/** |
85 |
* Prevent \a f from going into denormal mode which would slow down |
86 |
* subsequent floating point calculations, we achieve that by setting |
87 |
* \a f to zero when it falls under the denormal threshold value. |
88 |
*/ |
89 |
inline void KillDenormal(bq_t& f) { |
90 |
// TODO: this is a generic solution for 32bit floats, should be replaced by CPU specific asm code |
91 |
f += 1e-18f; |
92 |
f -= 1e-18f; |
93 |
} |
94 |
public: |
95 |
BiquadFilter() { |
96 |
Reset(); |
97 |
#if __GNUC__ >= 4 |
98 |
fbc = 0.98f; |
99 |
#endif |
100 |
} |
101 |
|
102 |
void Reset() { |
103 |
x1 = 0.0f; |
104 |
x2 = 0.0f; |
105 |
y1 = 0.0f; |
106 |
y2 = 0.0f; |
107 |
} |
108 |
|
109 |
inline bq_t Apply(const bq_t x) { |
110 |
bq_t y; |
111 |
|
112 |
y = this->b0 * x + this->b1 * this->x1 + this->b2 * this->x2 + |
113 |
this->a1 * this->y1 + this->a2 * this->y2; |
114 |
KillDenormal(y); |
115 |
this->x2 = this->x1; |
116 |
this->x1 = x; |
117 |
this->y2 = this->y1; |
118 |
this->y1 = y; |
119 |
|
120 |
return y; |
121 |
} |
122 |
|
123 |
inline bq_t Apply(biquad_param_t* param, const bq_t x) { |
124 |
bq_t y; |
125 |
|
126 |
y = param->b0 * x + param->b1 * this->x1 + param->b2 * this->x2 + |
127 |
param->a1 * this->y1 + param->a2 * this->y2; |
128 |
KillDenormal(y); |
129 |
this->x2 = this->x1; |
130 |
this->x1 = x; |
131 |
this->y2 = this->y1; |
132 |
this->y1 = y; |
133 |
|
134 |
return y; |
135 |
} |
136 |
|
137 |
#if 0 // CONFIG_ASM && ARCH_X86 |
138 |
// expects to find input in xmm0 (xmm0 stays unmodified) and finally leaves output in xmm6 |
139 |
inline void Apply4StepsSSE(biquad_param_t* param) { |
140 |
__asm__ __volatile__ ( |
141 |
"movss (%2),%%xmm4 # b0\n\t" |
142 |
"shufps $0x00,%%xmm4,%%xmm4 # copy b0 to other cells\n\t" |
143 |
"mulps %%xmm0,%%xmm4 # xmm4 = x*b0\n\t" |
144 |
"movups (%0),%%xmm2 # load b1,b2,a1,a2\n\t" |
145 |
"movups (%1),%%xmm5 # load x1,x2,y1,y2\n\t" |
146 |
/* sample 0 */ |
147 |
"movaps %%xmm5,%%xmm3\n\t" |
148 |
"mulps %%xmm2,%%xmm5 # xmm5 = [b1,b2,a1,a2] * [x1,x2,y1,y2]\n\t" |
149 |
"shufps $0x0a,%%xmm3,%%xmm3 # x2 = x1, y2 = y1\n\t" |
150 |
"movss %%xmm4,%%xmm6\n\t" |
151 |
"addss %%xmm5,%%xmm6\n\t" |
152 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
153 |
"addss %%xmm5,%%xmm6\n\t" |
154 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
155 |
"addss %%xmm5,%%xmm6\n\t" |
156 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
157 |
"addss %%xmm5,%%xmm6 # xmm6 = b0*x + b1*x1 + b2*x2 + a1*y1 + a2*y2\n\t" |
158 |
/* sample 1 */ |
159 |
"shufps $0x39,%%xmm4,%%xmm4 # rotate xmm4 down 1 cell\n\t" |
160 |
"movss %%xmm6,%%xmm3 # y1 = y\n\t" |
161 |
"shufps $0x4e,%%xmm3,%%xmm3 # rotate 2 cells\n\t" |
162 |
"movss %%xmm0,%%xmm3 # x1 = x\n\t" |
163 |
"shufps $0x93,%%xmm6,%%xmm6 # rotate output up 1 cell\n\t" |
164 |
"movaps %%xmm3,%%xmm5\n\t" |
165 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate input down 1 cell\n\t" |
166 |
"mulps %%xmm2,%%xmm5 # xmm5 = [b1,b2,a1,a2] * [x1,x2,y1,y2]\n\t" |
167 |
"movss %%xmm5,%%xmm6\n\t" |
168 |
"addss %%xmm4,%%xmm6\n\t" |
169 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
170 |
"addss %%xmm5,%%xmm6\n\t" |
171 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
172 |
"addss %%xmm5,%%xmm6\n\t" |
173 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
174 |
"addss %%xmm5,%%xmm6 # xmm6 = b0*x + b1*x1 + b2*x2 + a1*y1 + a2*y2\n\t" |
175 |
/* sample 2 */ |
176 |
"shufps $0x0a,%%xmm3,%%xmm3 # x2 = x1, y2 = y1\n\t" |
177 |
"shufps $0x39,%%xmm4,%%xmm4 # rotate xmm4 down 1 cell\n\t" |
178 |
"movss %%xmm6,%%xmm3 # y1 = y\n\t" |
179 |
"shufps $0x4e,%%xmm3,%%xmm3 # rotate 2 cells\n\t" |
180 |
"movss %%xmm0,%%xmm3 # x1 = x\n\t" |
181 |
"shufps $0x93,%%xmm6,%%xmm6 # rotate output up 1 cell\n\t" |
182 |
"movaps %%xmm3,%%xmm5\n\t" |
183 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate input down 1 cell\n\t" |
184 |
"mulps %%xmm2,%%xmm5 # xmm5 = [b1,b2,a1,a2] * [x1,x2,y1,y2]\n\t" |
185 |
"movss %%xmm5,%%xmm6\n\t" |
186 |
"addss %%xmm4,%%xmm6\n\t" |
187 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
188 |
"addss %%xmm5,%%xmm6\n\t" |
189 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
190 |
"addss %%xmm5,%%xmm6\n\t" |
191 |
"shufps $0x39,%%xmm5,%%xmm5\n\t" |
192 |
"addss %%xmm5,%%xmm6 # xmm6 = b0*x + b1*x1 + b2*x2 + a1*y1 + a2*y2\n\t" |
193 |
/* sample 3 */ |
194 |
"shufps $0x0a,%%xmm3,%%xmm3 # x2 = x1, y2 = y1\n\t" |
195 |
"shufps $0x39,%%xmm4,%%xmm4 # rotate xmm4 down 1 cell\n\t" |
196 |
"movss %%xmm6,%%xmm3 # y1 = y\n\t" |
197 |
"shufps $0x4e,%%xmm3,%%xmm3 # rotate 2 cells\n\t" |
198 |
"movss %%xmm0,%%xmm3 # x1 = x\n\t" |
199 |
"shufps $0x93,%%xmm6,%%xmm6 # rotate output up 1 cell\n\t" |
200 |
"mulps %%xmm3,%%xmm2 # xmm5 = [b1,b2,a1,a2] * [x1,x2,y1,y2]\n\t" |
201 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate input down 1 cell\n\t" |
202 |
"movss %%xmm2,%%xmm6\n\t" |
203 |
"shufps $0x39,%%xmm2,%%xmm2\n\t" |
204 |
"addss %%xmm2,%%xmm6\n\t" |
205 |
"shufps $0x39,%%xmm2,%%xmm2\n\t" |
206 |
"addss %%xmm2,%%xmm6\n\t" |
207 |
"shufps $0x39,%%xmm2,%%xmm2\n\t" |
208 |
"addss %%xmm2,%%xmm6\n\t" |
209 |
"addss %%xmm4,%%xmm6 # xmm6 = b0*x + b1*x1 + b2*x2 + a1*y1 + a2*y2\n\t" |
210 |
/* done */ |
211 |
"shufps $0x0a,%%xmm3,%%xmm3 # x2 = x1, y2 = y1\n\t" |
212 |
"movss %%xmm6,%%xmm3 # y1 = y\n\t" |
213 |
"shufps $0x4e,%%xmm3,%%xmm3 # rotate 2 cells\n\t" |
214 |
"movss %%xmm0,%%xmm3 # x1 = x\n\t" |
215 |
"shufps $0x1b,%%xmm6,%%xmm6 # swap output to correct order\n\t" |
216 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate input down 1 cell, to restore original input\n\t" |
217 |
"movups %%xmm3,(%1) # store x1,x2,y1,y2\n\t" |
218 |
: /* no output */ |
219 |
: "r" (¶m->b1), /* %0 - [b1,b2,a1,a2] */ |
220 |
"r" (&x1), /* %1 - [x1,x2,y1,y2] */ |
221 |
"r" (¶m->b0) /* %2 */ |
222 |
); |
223 |
} |
224 |
#endif // CONFIG_ASM && ARCH_X86 |
225 |
|
226 |
inline bq_t ApplyFB(bq_t x, const bq_t fb) { |
227 |
bq_t y; |
228 |
|
229 |
x += this->y1 * fb * 0.98; |
230 |
y = this->b0 * x + this->b1 * this->x1 + this->b2 * this->x2 + |
231 |
this->a1 * this->y1 + this->a2 * this->y2; |
232 |
KillDenormal(y); |
233 |
this->x2 = this->x1; |
234 |
this->x1 = x; |
235 |
this->y2 = this->y1; |
236 |
this->y1 = y; |
237 |
|
238 |
return y; |
239 |
} |
240 |
|
241 |
inline bq_t ApplyFB(biquad_param_t* param, bq_t x, const bq_t fb) { |
242 |
bq_t y; |
243 |
|
244 |
x += this->y1 * fb * 0.98; |
245 |
y = param->b0 * x + param->b1 * this->x1 + param->b2 * this->x2 + |
246 |
param->a1 * this->y1 + param->a2 * this->y2; |
247 |
KillDenormal(y); |
248 |
this->x2 = this->x1; |
249 |
this->x1 = x; |
250 |
this->y2 = this->y1; |
251 |
this->y1 = y; |
252 |
|
253 |
return y; |
254 |
} |
255 |
|
256 |
#if 0 // CONFIG_ASM && ARCH_X86 |
257 |
// expects to find input in xmm0 (xmm0 stays unmodified) and finally leaves output in xmm7 |
258 |
inline void ApplyFB4StepsSSE(biquad_param_t* param, const bq_t &fb) { |
259 |
float xs, ys; |
260 |
float t0, t1, t2, t3, t4, t5, t6, t7, t8; // temporary stack space |
261 |
__asm__ __volatile__ ( |
262 |
/* prepare input */ |
263 |
"movss %15,%%xmm5\n\t" |
264 |
"movss %%xmm0,(%14)\n\t" |
265 |
/* sample 0 */ |
266 |
"movss %0, %%xmm3\n\t" |
267 |
"movss %1, %%xmm4\n\t" |
268 |
"mulss %%xmm4, %%xmm5\n\t" |
269 |
"movss %%xmm3, %2\n\t" |
270 |
"movss %%xmm5, %16\n\t" |
271 |
"mulss %%xmm3, %%xmm5\n\t" |
272 |
"movss %19, %%xmm2\n\t" |
273 |
"movss %3, %%xmm6\n\t" |
274 |
"movss %21, %%xmm3\n\t" |
275 |
"addss %%xmm5, %%xmm6\n\t" |
276 |
"movss %%xmm2, %%xmm5\n\t" |
277 |
"movss %20, %%xmm4\n\t" |
278 |
"movss %%xmm6, %4\n\t" |
279 |
"mulss %%xmm6, %%xmm5\n\t" |
280 |
"movss %5, %%xmm6\n\t" |
281 |
"movss %%xmm2, %6\n\t" |
282 |
"movss %%xmm4, %7\n\t" |
283 |
"movss %%xmm3, %%xmm2\n\t" |
284 |
"mulss %%xmm6, %%xmm4\n\t" |
285 |
"mulss %8, %%xmm2\n\t" |
286 |
"movss %%xmm3, %9\n\t" |
287 |
"addss %%xmm4, %%xmm5\n\t" |
288 |
"movss %18, %%xmm3\n\t" |
289 |
"movss %17, %%xmm4\n\t" |
290 |
"addss %%xmm2, %%xmm5\n\t" |
291 |
"movss %%xmm4, %10\n\t" |
292 |
"movss %%xmm3, %%xmm2\n\t" |
293 |
"mulss %11, %%xmm4\n\t" |
294 |
"mulss %12, %%xmm2\n\t" |
295 |
"movss %%xmm3, %13\n\t" |
296 |
"addss %%xmm4, %%xmm5\n\t" |
297 |
"movss %11, %%xmm3\n\t" |
298 |
"movss %4, %%xmm4\n\t" |
299 |
"addss %%xmm2, %%xmm5\n\t" |
300 |
:: "m" (y1), /* %0 */ |
301 |
"m" (fbc), /* %1 */ |
302 |
"m" (t0), /* %2 */ |
303 |
"m" (xs), /* %3 */ |
304 |
"m" (t7), /* %4 */ |
305 |
"m" (x1), /* %5 */ |
306 |
"m" (t1), /* %6 */ |
307 |
"m" (t2), /* %7 */ |
308 |
"m" (x2), /* %8 */ |
309 |
"m" (t3), /* %9 */ |
310 |
"m" (t4), /* %10 */ |
311 |
"m" (t0), /* %11 */ |
312 |
"m" (y2), /* %12 */ |
313 |
"m" (t5), /* %13 */ |
314 |
"r" (&xs), /* %14 */ |
315 |
"m" (fb), /* %15 */ |
316 |
"m" (ys), /* %16 */ |
317 |
"m" (param->a1), /* %17 */ |
318 |
"m" (param->a2), /* %18 */ |
319 |
"m" (param->b0), /* %19 */ |
320 |
"m" (param->b1), /* %20 */ |
321 |
"m" (param->b2) /* %21 */ |
322 |
); |
323 |
__asm__ __volatile__ ( |
324 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate down one cell\n\t" |
325 |
"movss %%xmm5,%%xmm7\n\t" |
326 |
:: |
327 |
); |
328 |
/* sample 1 */ |
329 |
__asm__ __volatile__ ( |
330 |
"movss %0, %%xmm4\n\t" |
331 |
"movss %%xmm0, %%xmm3\n\t" |
332 |
"mulss %%xmm5, %%xmm4\n\t" |
333 |
"mulss %3, %%xmm6\n\t" |
334 |
"movss %5, %%xmm2\n\t" |
335 |
"addss %%xmm4, %%xmm3\n\t" |
336 |
"mulss %7, %%xmm2\n\t" |
337 |
"movss %6, %%xmm4\n\t" |
338 |
"movss %%xmm3, %8\n\t" |
339 |
"mulss %%xmm3, %%xmm4\n\t" |
340 |
"addss %%xmm2, %%xmm4\n\t" |
341 |
"movss %9, %%xmm3\n\t" |
342 |
"mulss %%xmm5, %%xmm3\n\t" |
343 |
"movss %10, %%xmm2\n\t" |
344 |
"addss %%xmm6, %%xmm4\n\t" |
345 |
"mulss %11, %%xmm2\n\t" |
346 |
"addss %%xmm3, %%xmm4\n\t" |
347 |
"addss %%xmm2, %%xmm4\n\t" |
348 |
:: "m" (ys), /* %0 */ |
349 |
"m" (fbc), /* %1 */ |
350 |
"m" (xs), /* %2 */ |
351 |
"m" (t3), /* %3 */ |
352 |
"m" (y2), /* %4 */ |
353 |
"m" (t2), /* %5 */ |
354 |
"m" (t1), /* %6 */ |
355 |
"m" (t7), /* %7 */ |
356 |
"m" (t8), /* %8 */ |
357 |
"m" (t4), /* %9 */ |
358 |
"m" (t5), /* %10 */ |
359 |
"m" (t0), /* %11 */ |
360 |
"m" (x2), /* %12 */ |
361 |
"m" (x1), /* %13 */ |
362 |
"m" (y1) /* %14 */ |
363 |
); |
364 |
__asm__ __volatile__ ( |
365 |
"shufps $0x93,%%xmm7,%%xmm7 # rotate up one cell\n\t" |
366 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate down one cell\n\t" |
367 |
"movss %%xmm4,%%xmm7\n\t" |
368 |
:: |
369 |
); |
370 |
/* sample 2 */ |
371 |
__asm__ __volatile__ ( |
372 |
"movss %2, %%xmm6\n\t" |
373 |
"movss %3, %%xmm3\n\t" |
374 |
"mulss %%xmm4, %%xmm6\n\t" |
375 |
"movss %4, %%xmm2\n\t" |
376 |
"mulss %9, %%xmm2\n\t" |
377 |
"addss %%xmm0, %%xmm6\n\t" |
378 |
"mulss %7, %%xmm5\n\t" |
379 |
"mulss %%xmm6, %%xmm3\n\t" |
380 |
"addss %%xmm2, %%xmm3\n\t" |
381 |
"movss %5, %%xmm2\n\t" |
382 |
"mulss %8, %%xmm2\n\t" |
383 |
"addss %%xmm2, %%xmm3\n\t" |
384 |
"movss %6, %%xmm2\n\t" |
385 |
"mulss %%xmm4, %%xmm2\n\t" |
386 |
"addss %%xmm5, %%xmm2\n\t" |
387 |
"addss %%xmm2, %%xmm3\n\t" |
388 |
:: "m" (xs), /* %0 */ |
389 |
"m" (fb), /* %1 */ |
390 |
"m" (ys), /* %2 */ |
391 |
"m" (t1), /* %3 */ |
392 |
"m" (t2), /* %4 */ |
393 |
"m" (t3), /* %5 */ |
394 |
"m" (t4), /* %6 */ |
395 |
"m" (t5), /* %7 */ |
396 |
"m" (t7), /* %8 */ |
397 |
"m" (t8), /* %9 */ |
398 |
"m" (x1), /* %10 */ |
399 |
"m" (x2), /* %11 */ |
400 |
"m" (y1), /* %12 */ |
401 |
"m" (y2) /* %13 */ |
402 |
); |
403 |
__asm__ __volatile__ ( |
404 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate down one cell\n\t" |
405 |
"shufps $0x93,%%xmm7,%%xmm7 # rotate up one cell\n\t" |
406 |
"movss %%xmm3,%%xmm7\n\t" |
407 |
:: |
408 |
); |
409 |
/* sample 3 */ |
410 |
__asm__ __volatile__ ( |
411 |
"movss %1, %%xmm2\n\t" |
412 |
"mulss %7, %%xmm4\n\t" |
413 |
"mulss %%xmm3, %%xmm2\n\t" |
414 |
"movss %3, %%xmm5\n\t" |
415 |
"movss %%xmm6, %11\n\t" |
416 |
"addss %%xmm0, %%xmm2\n\t" |
417 |
"movss %%xmm3, %13\n\t" |
418 |
"mulss %%xmm2, %%xmm5\n\t" |
419 |
"mulss %4, %%xmm6\n\t" |
420 |
"movss %%xmm2, %10\n\t" |
421 |
"addss %%xmm6, %%xmm5\n\t" |
422 |
"movss %5, %%xmm2\n\t" |
423 |
"mulss %9, %%xmm2\n\t" |
424 |
"mulss %6, %%xmm3\n\t" |
425 |
"addss %%xmm2, %%xmm5\n\t" |
426 |
"addss %%xmm3, %%xmm4\n\t" |
427 |
"addss %%xmm4, %%xmm5\n\t" |
428 |
"movss %%xmm5, %12\n\t" |
429 |
:: "m" (xs), /* %0 */ |
430 |
"m" (ys), /* %1 */ |
431 |
"m" (fbc), /* %2 */ |
432 |
"m" (t1), /* %3 */ |
433 |
"m" (t2), /* %4 */ |
434 |
"m" (t3), /* %5 */ |
435 |
"m" (t4), /* %6 */ |
436 |
"m" (t5), /* %7 */ |
437 |
"m" (t6), /* %8 */ |
438 |
"m" (t8), /* %9 */ |
439 |
"m" (x1), /* %10 */ |
440 |
"m" (x2), /* %11 */ |
441 |
"m" (y1), /* %12 */ |
442 |
"m" (y2) /* %13 */ |
443 |
); |
444 |
__asm__ __volatile__ ( |
445 |
"shufps $0x93,%%xmm7,%%xmm7 # rotate up one cell\n\t" |
446 |
"shufps $0x39,%%xmm0,%%xmm0 # rotate down one cell to restore original input\n\t" |
447 |
"movss %%xmm5,%%xmm7\n\t" |
448 |
"shufps $0x1b,%%xmm7,%%xmm7 # swap output to correct order\n\t" |
449 |
:: |
450 |
); |
451 |
} |
452 |
#endif // CONFIG_ASM && ARCH_X86 |
453 |
}; |
454 |
|
455 |
/** @brief Lowpass Filter |
456 |
* |
457 |
* Lowpass filter based on biquad filter implementation. |
458 |
*/ |
459 |
class LowpassFilter : public BiquadFilter { |
460 |
public: |
461 |
inline LowpassFilter() : BiquadFilter() {} |
462 |
|
463 |
inline void SetParameters(bq_t fc, bq_t bw, bq_t fs) { |
464 |
bq_t omega = 2.0 * M_PI * fc / fs; |
465 |
bq_t sn = sin(omega); |
466 |
bq_t cs = cos(omega); |
467 |
bq_t alpha = sn * sinh(M_LN2 / 2.0 * bw * omega / sn); |
468 |
|
469 |
const float a0r = 1.0 / (1.0 + alpha); |
470 |
this->b0 = a0r * (1.0 - cs) * 0.5; |
471 |
this->b1 = a0r * (1.0 - cs); |
472 |
this->b2 = a0r * (1.0 - cs) * 0.5; |
473 |
this->a1 = a0r * (2.0 * cs); |
474 |
this->a2 = a0r * (alpha - 1.0); |
475 |
} |
476 |
|
477 |
inline void SetParameters(biquad_param_t* param, bq_t fc, bq_t bw, bq_t fs) { |
478 |
bq_t omega = 2.0 * M_PI * fc / fs; |
479 |
bq_t sn = sin(omega); |
480 |
bq_t cs = cos(omega); |
481 |
bq_t alpha = sn * sinh(M_LN2 / 2.0 * bw * omega / sn); |
482 |
|
483 |
const float a0r = 1.0 / (1.0 + alpha); |
484 |
param->b0 = a0r * (1.0 - cs) * 0.5; |
485 |
param->b1 = a0r * (1.0 - cs); |
486 |
param->b2 = a0r * (1.0 - cs) * 0.5; |
487 |
param->a1 = a0r * (2.0 * cs); |
488 |
param->a2 = a0r * (alpha - 1.0); |
489 |
} |
490 |
}; |
491 |
|
492 |
/** @brief Bandpass Filter |
493 |
* |
494 |
* Bandpass filter based on biquad filter implementation. |
495 |
*/ |
496 |
class BandpassFilter : public BiquadFilter { |
497 |
public: |
498 |
inline BandpassFilter() : BiquadFilter() {} |
499 |
|
500 |
inline void SetParameters(bq_t fc, bq_t bw, bq_t fs) { |
501 |
bq_t omega = 2.0 * M_PI * fc / fs; |
502 |
bq_t sn = sin(omega); |
503 |
bq_t cs = cos(omega); |
504 |
bq_t alpha = sn * sinh(M_LN2 / 2.0 * bw * omega / sn); |
505 |
|
506 |
const float a0r = 1.0 / (1.0 + alpha); |
507 |
this->b0 = a0r * sn * 0.71; |
508 |
this->b1 = 0.0; |
509 |
this->b2 = a0r * -sn * 0.71; |
510 |
this->a1 = a0r * (2.0 * cs); |
511 |
this->a2 = a0r * (alpha - 1.0); |
512 |
} |
513 |
|
514 |
inline void SetParameters(biquad_param_t* param, bq_t fc, bq_t bw, bq_t fs) { |
515 |
bq_t omega = 2.0 * M_PI * fc / fs; |
516 |
bq_t sn = sin(omega); |
517 |
bq_t cs = cos(omega); |
518 |
bq_t alpha = sn * sinh(M_LN2 / 2.0 * bw * omega / sn); |
519 |
|
520 |
const float a0r = 1.0 / (1.0 + alpha); |
521 |
param->b0 = a0r * sn * 0.71; |
522 |
param->b1 = 0.0; |
523 |
param->b2 = a0r * -sn * 0.71; |
524 |
param->a1 = a0r * (2.0 * cs); |
525 |
param->a2 = a0r * (alpha - 1.0); |
526 |
} |
527 |
}; |
528 |
|
529 |
/** @brief Highpass Filter |
530 |
* |
531 |
* Highpass filter based on biquad filter implementation. |
532 |
*/ |
533 |
class HighpassFilter : public BiquadFilter { |
534 |
public: |
535 |
inline HighpassFilter() : BiquadFilter() {} |
536 |
|
537 |
inline void SetParameters(bq_t fc, bq_t bw, bq_t fs) { |
538 |
bq_t omega = 2.0 * M_PI * fc / fs; |
539 |
bq_t sn = sin(omega); |
540 |
bq_t cs = cos(omega); |
541 |
bq_t alpha = sn * sinh(M_LN2 / 2.0 * bw * omega / sn); |
542 |
|
543 |
const float a0r = 1.0 / (1.0 + alpha); |
544 |
this->b0 = a0r * (1.0 + cs) * 0.5; |
545 |
this->b1 = a0r * -(1.0 + cs); |
546 |
this->b2 = a0r * (1.0 + cs) * 0.5; |
547 |
this->a1 = a0r * (2.0 * cs); |
548 |
this->a2 = a0r * (alpha - 1.0); |
549 |
} |
550 |
|
551 |
inline void SetParameters(biquad_param_t* param, bq_t fc, bq_t bw, bq_t fs) { |
552 |
bq_t omega = 2.0 * M_PI * fc / fs; |
553 |
bq_t sn = sin(omega); |
554 |
bq_t cs = cos(omega); |
555 |
bq_t alpha = sn * sinh(M_LN2 / 2.0 * bw * omega / sn); |
556 |
|
557 |
const float a0r = 1.0 / (1.0 + alpha); |
558 |
param->b0 = a0r * (1.0 + cs) * 0.5; |
559 |
param->b1 = a0r * -(1.0 + cs); |
560 |
param->b2 = a0r * (1.0 + cs) * 0.5; |
561 |
param->a1 = a0r * (2.0 * cs); |
562 |
param->a2 = a0r * (alpha - 1.0); |
563 |
} |
564 |
}; |
565 |
|
566 |
} // namespace LinuxSampler |
567 |
|
568 |
#endif // __LS_BIQUADFILTER_H__ |