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