| 1 |
/*************************************************************************** |
| 2 |
* * |
| 3 |
* LinuxSampler - modular, streaming capable sampler * |
| 4 |
* * |
| 5 |
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
| 6 |
* * |
| 7 |
* This program is free software; you can redistribute it and/or modify * |
| 8 |
* it under the terms of the GNU General Public License as published by * |
| 9 |
* the Free Software Foundation; either version 2 of the License, or * |
| 10 |
* (at your option) any later version. * |
| 11 |
* * |
| 12 |
* This program is distributed in the hope that it will be useful, * |
| 13 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of * |
| 14 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
| 15 |
* GNU General Public License for more details. * |
| 16 |
* * |
| 17 |
* You should have received a copy of the GNU General Public License * |
| 18 |
* along with this program; if not, write to the Free Software * |
| 19 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, * |
| 20 |
* MA 02111-1307 USA * |
| 21 |
***************************************************************************/ |
| 22 |
|
| 23 |
#ifndef __LS_GIG_SYNTHESIZER_H__ |
| 24 |
#define __LS_GIG_SYNTHESIZER_H__ |
| 25 |
|
| 26 |
#include "../../common/global.h" |
| 27 |
#include "../../common/RTMath.h" |
| 28 |
#include "../common/Resampler.h" |
| 29 |
#include "../common/BiquadFilter.h" |
| 30 |
#include "Filter.h" |
| 31 |
#include "Voice.h" |
| 32 |
|
| 33 |
#define SYNTHESIS_MODE_SET_CONSTPITCH(iMode,bVal) if (bVal) iMode |= 0x01; else iMode &= ~0x01 /* (un)set mode bit 0 */ |
| 34 |
#define SYNTHESIS_MODE_SET_LOOP(iMode,bVal) if (bVal) iMode |= 0x02; else iMode &= ~0x02 /* (un)set mode bit 1 */ |
| 35 |
#define SYNTHESIS_MODE_SET_INTERPOLATE(iMode,bVal) if (bVal) iMode |= 0x04; else iMode &= ~0x04 /* (un)set mode bit 2 */ |
| 36 |
#define SYNTHESIS_MODE_SET_FILTER(iMode,bVal) if (bVal) iMode |= 0x08; else iMode &= ~0x08 /* (un)set mode bit 3 */ |
| 37 |
#define SYNTHESIS_MODE_SET_CHANNELS(iMode,bVal) if (bVal) iMode |= 0x10; else iMode &= ~0x10 /* (un)set mode bit 4 */ |
| 38 |
#define SYNTHESIS_MODE_SET_IMPLEMENTATION(iMode,bVal) if (bVal) iMode |= 0x20; else iMode &= ~0x20 /* (un)set mode bit 5 */ |
| 39 |
#define SYNTHESIS_MODE_SET_PROFILING(iMode,bVal) if (bVal) iMode |= 0x40; else iMode &= ~0x40 /* (un)set mode bit 6 */ |
| 40 |
|
| 41 |
#define SYNTHESIS_MODE_GET_CONSTPITCH(iMode) iMode & 0x01 |
| 42 |
#define SYNTHESIS_MODE_GET_LOOP(iMode) iMode & 0x02 |
| 43 |
#define SYNTHESIS_MODE_GET_INTERPOLATE(iMode) iMode & 0x04 |
| 44 |
#define SYNTHESIS_MODE_GET_FILTER(iMode) iMode & 0x08 |
| 45 |
#define SYNTHESIS_MODE_GET_CHANNELS(iMode) iMode & 0x10 |
| 46 |
#define SYNTHESIS_MODE_GET_IMPLEMENTATION(iMode) iMode & 0x20 |
| 47 |
|
| 48 |
// that's usually gig::Voice of course, but we make it a macro so we can |
| 49 |
// include this code for our synthesis benchmark which uses fake data |
| 50 |
// structures |
| 51 |
#ifndef VOICE |
| 52 |
# define VOICE Voice |
| 53 |
#endif // VOICE |
| 54 |
|
| 55 |
namespace LinuxSampler { namespace gig { |
| 56 |
|
| 57 |
typedef void SynthesizeFragment_Fn(VOICE&, uint, sample_t*, uint); |
| 58 |
|
| 59 |
void* GetSynthesisFunction(const int SynthesisMode); |
| 60 |
void RunSynthesisFunction(const int SynthesisMode, VOICE& voice, uint Samples, sample_t* pSrc, uint Skip); |
| 61 |
|
| 62 |
enum channels_t { |
| 63 |
MONO, |
| 64 |
STEREO |
| 65 |
}; |
| 66 |
|
| 67 |
template<implementation_t IMPLEMENTATION, channels_t CHANNELS, bool USEFILTER, bool INTERPOLATE, bool DOLOOP, bool CONSTPITCH> |
| 68 |
class Synthesizer : public __RTMath<IMPLEMENTATION>, public LinuxSampler::Resampler<INTERPOLATE> { |
| 69 |
public: |
| 70 |
template<typename VOICE_T> |
| 71 |
inline static void SynthesizeFragment(VOICE_T& Voice, uint Samples, sample_t* pSrc, uint i) { |
| 72 |
if (IMPLEMENTATION == ASM_X86_MMX_SSE) { |
| 73 |
float fPos = (float) Voice.Pos; |
| 74 |
SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount, |
| 75 |
Voice.pSample->LoopStart, |
| 76 |
Voice.pSample->LoopEnd, |
| 77 |
Voice.pSample->LoopSize, |
| 78 |
Voice.LoopCyclesLeft, |
| 79 |
(void *)&fPos, |
| 80 |
Voice.PitchBase, |
| 81 |
Voice.PitchBend); |
| 82 |
#if ARCH_X86 |
| 83 |
if (INTERPOLATE) EMMS; |
| 84 |
#endif |
| 85 |
Voice.Pos = (double) fPos; |
| 86 |
} else { |
| 87 |
SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount, |
| 88 |
Voice.pSample->LoopStart, |
| 89 |
Voice.pSample->LoopEnd, |
| 90 |
Voice.pSample->LoopSize, |
| 91 |
Voice.LoopCyclesLeft, |
| 92 |
(void *)&Voice.Pos, |
| 93 |
Voice.PitchBase, |
| 94 |
Voice.PitchBend); |
| 95 |
} |
| 96 |
} |
| 97 |
|
| 98 |
//protected: |
| 99 |
|
| 100 |
template<typename VOICE_T> |
| 101 |
inline static void SynthesizeFragment(VOICE_T& Voice, uint Samples, sample_t* pSrc, uint& i, uint& LoopPlayCount, uint LoopStart, uint LoopEnd, uint LoopSize, uint& LoopCyclesLeft, void* Pos, float& PitchBase, float& PitchBend) { |
| 102 |
const float loopEnd = Float(LoopEnd); |
| 103 |
const float PBbyPB = Mul(PitchBase, PitchBend); |
| 104 |
const float f_LoopStart = Float(LoopStart); |
| 105 |
const float f_LoopSize = Float(LoopSize); |
| 106 |
if (DOLOOP) { |
| 107 |
if (LoopPlayCount) { |
| 108 |
// render loop (loop count limited) |
| 109 |
while (i < Samples && LoopCyclesLeft) { |
| 110 |
if (CONSTPITCH) { |
| 111 |
const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd,Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up |
| 112 |
while (i < processEnd) Synthesize(Voice, Pos, pSrc, i); |
| 113 |
} |
| 114 |
else Synthesize(Voice, Pos, pSrc, i); |
| 115 |
if (WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos)) LoopCyclesLeft--; |
| 116 |
} |
| 117 |
// render on without loop |
| 118 |
while (i < Samples) Synthesize(Voice, Pos, pSrc, i); |
| 119 |
} |
| 120 |
else { // render loop (endless loop) |
| 121 |
while (i < Samples) { |
| 122 |
if (CONSTPITCH) { |
| 123 |
const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd, Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up |
| 124 |
while (i < processEnd) Synthesize(Voice, Pos, pSrc, i); |
| 125 |
} |
| 126 |
else Synthesize(Voice, Pos, pSrc, i); |
| 127 |
WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos); |
| 128 |
} |
| 129 |
} |
| 130 |
} |
| 131 |
else { // no looping |
| 132 |
while (i < Samples) { Synthesize(Voice, Pos, pSrc, i);} |
| 133 |
} |
| 134 |
} |
| 135 |
|
| 136 |
template<typename VOICE_T> |
| 137 |
inline static void Synthesize(VOICE_T& Voice, void* Pos, sample_t* pSrc, uint& i) { |
| 138 |
Synthesize(pSrc, Pos, |
| 139 |
Voice.pEngine->pSynthesisParameters[Event::destination_vco][i], |
| 140 |
Voice.pEngine->pOutputLeft, |
| 141 |
Voice.pEngine->pOutputRight, |
| 142 |
i, |
| 143 |
Voice.pEngine->pSynthesisParameters[Event::destination_vca], |
| 144 |
&Voice.PanLeft, |
| 145 |
&Voice.PanRight, |
| 146 |
Voice.FilterLeft, |
| 147 |
Voice.FilterRight, |
| 148 |
Voice.pEngine->pBasicFilterParameters[i], |
| 149 |
Voice.pEngine->pMainFilterParameters[i]); |
| 150 |
} |
| 151 |
|
| 152 |
inline static int DiffToLoopEnd(const float& LoopEnd, const void* Pos, const float& Pitch) { |
| 153 |
switch (IMPLEMENTATION) { |
| 154 |
// pure C++ implementation (thus platform independent) |
| 155 |
case CPP: { |
| 156 |
return uint((LoopEnd - *((double *)Pos)) / Pitch); |
| 157 |
} |
| 158 |
#if ARCH_X86 |
| 159 |
case ASM_X86_MMX_SSE: { |
| 160 |
int result; |
| 161 |
__asm__ __volatile__ ( |
| 162 |
"movss (%1), %%xmm0 #read loopend\n\t" |
| 163 |
"subss (%2), %%xmm0 #sub pos\n\t" |
| 164 |
"divss (%3), %%xmm0 #div by pitch\n\t" |
| 165 |
"cvtss2si %%xmm0, %0 #convert to int\n\t" |
| 166 |
: "=r" (result) /* %0 */ |
| 167 |
: "r" (&LoopEnd), /* %1 */ |
| 168 |
"r" (Pos), /* %2 */ |
| 169 |
"r" (&Pitch) /* %3 */ |
| 170 |
); |
| 171 |
return result; |
| 172 |
} |
| 173 |
#endif // ARCH_X86 |
| 174 |
} |
| 175 |
} |
| 176 |
|
| 177 |
inline static int WrapLoop(const float& LoopStart, const float& LoopSize, const float& LoopEnd, void* vPos) { |
| 178 |
switch (IMPLEMENTATION) { |
| 179 |
// pure C++ implementation (thus platform independent) |
| 180 |
case CPP: { |
| 181 |
double * Pos = (double *)vPos; |
| 182 |
if (*Pos < LoopEnd) return 0; |
| 183 |
*Pos = fmod(*Pos - LoopEnd, LoopSize) + LoopStart; |
| 184 |
return 1; |
| 185 |
} |
| 186 |
#if ARCH_X86 |
| 187 |
case ASM_X86_MMX_SSE: { |
| 188 |
int result = 0; |
| 189 |
__asm__ __volatile__ ( |
| 190 |
"movss (%2), %%xmm0 # load LoopEnd\n\t" |
| 191 |
"movss (%1), %%xmm1 # load Pos\n\t" |
| 192 |
"comiss %%xmm0, %%xmm1 # LoopEnd <> Pos\n\t" |
| 193 |
"jb 1f # jump if no work needs to be done\n\t" |
| 194 |
"movss (%3), %%xmm2 # load LoopSize\n\t" |
| 195 |
"subss %%xmm0, %%xmm1 # Pos - LoopEnd\n\t" |
| 196 |
//now the fmodf |
| 197 |
"movss %%xmm1, %%xmm3 # xmm3 = (Pos - LoopEnd)\n\t" |
| 198 |
"divss %%xmm2, %%xmm1 # (Pos - LoopEnd) / LoopSize\n\t" |
| 199 |
"cvttss2si %%xmm1, %%eax # convert to int\n\t" |
| 200 |
"cvtsi2ss %%eax, %%xmm1 # convert back to float\n\t" |
| 201 |
"movss (%4), %%xmm0 # load LoopStart\n\t" |
| 202 |
"mulss %%xmm2, %%xmm1 # LoopSize * int((Pos-LoopEnd)/LoopSize)\n\t" |
| 203 |
"subss %%xmm1, %%xmm3 # xmm2 = fmodf(Pos - LoopEnd, LoopSize)\n\t" |
| 204 |
//done with fmodf |
| 205 |
"addss %%xmm0, %%xmm3 # add LoopStart\n\t" |
| 206 |
"movss %%xmm3, (%1) # update Pos\n\t" |
| 207 |
"movl $1, (%0) # result = 1\n\t" |
| 208 |
".balign 16 \n\t" |
| 209 |
"1:\n\t" |
| 210 |
:: "r" (&result), /* %0 */ |
| 211 |
"r" (vPos), /* %1 */ |
| 212 |
"r" (&LoopEnd), /* %2 */ |
| 213 |
"r" (&LoopSize), /* %3 */ |
| 214 |
"r" (&LoopStart) /* %4 */ |
| 215 |
); |
| 216 |
return result; |
| 217 |
} |
| 218 |
#endif // ARCH_X86 |
| 219 |
} |
| 220 |
} |
| 221 |
|
| 222 |
inline static void Synthesize(sample_t* pSrc, void* Pos, float& Pitch, float* pOutL, float* pOutR, uint& i, float* Volume, float* PanL, float* PanR, Filter& FilterL, Filter& FilterR, biquad_param_t& bqBase, biquad_param_t& bqMain) { |
| 223 |
switch (IMPLEMENTATION) { |
| 224 |
// pure C++ implementation (thus platform independent) |
| 225 |
case CPP: { |
| 226 |
switch (CHANNELS) { |
| 227 |
case MONO: { |
| 228 |
float samplePoint = GetNextSampleMonoCPP(pSrc, (double *)Pos, Pitch); |
| 229 |
if (USEFILTER) samplePoint = FilterL.Apply(&bqBase, &bqMain, samplePoint); |
| 230 |
pOutL[i] += samplePoint * Volume[i] * *PanL; |
| 231 |
pOutR[i] += samplePoint * Volume[i] * *PanR; |
| 232 |
i++; |
| 233 |
break; |
| 234 |
} |
| 235 |
case STEREO: { |
| 236 |
stereo_sample_t samplePoint = GetNextSampleStereoCPP(pSrc, (double *)Pos, Pitch); |
| 237 |
if (USEFILTER) { |
| 238 |
samplePoint.left = FilterL.Apply(&bqBase, &bqMain, samplePoint.left); |
| 239 |
samplePoint.right = FilterR.Apply(&bqBase, &bqMain, samplePoint.right); |
| 240 |
} |
| 241 |
pOutL[i] += samplePoint.left * Volume[i] * *PanL; |
| 242 |
pOutR[i] += samplePoint.right * Volume[i] * *PanR; |
| 243 |
i++; |
| 244 |
break; |
| 245 |
} |
| 246 |
} |
| 247 |
break; |
| 248 |
} |
| 249 |
#if ARCH_X86 |
| 250 |
// Assembly optimization using the MMX & SSE(1) instruction set (thus only for x86) |
| 251 |
case ASM_X86_MMX_SSE: { |
| 252 |
const int ii = i & 0xfffffffc; |
| 253 |
i += 4; |
| 254 |
switch (CHANNELS) { |
| 255 |
case MONO: { |
| 256 |
GetNext4SamplesMonoMMXSSE(pSrc, (float *)Pos, Pitch); // outputs samples in xmm2 |
| 257 |
if (USEFILTER) { |
| 258 |
/* prepare filter input */ |
| 259 |
__asm__ __volatile__ ( |
| 260 |
"movaps %xmm2,%xmm0" |
| 261 |
); |
| 262 |
FilterL.Apply4StepsSSE(&bqBase, &bqMain); // xmm0 input, xmm7 output |
| 263 |
__asm__ __volatile__ ( |
| 264 |
"movaps %xmm7,%xmm2 # mono filter result -> xmm2" |
| 265 |
); |
| 266 |
} |
| 267 |
/* apply panorama and volume factors */ |
| 268 |
__asm__ __volatile__ ( |
| 269 |
"movss (%1),%%xmm0 # load pan left\n\t" |
| 270 |
"movss (%2),%%xmm1 # load pan right\n\t" |
| 271 |
"movaps (%0),%%xmm4 # load vca\n\t" |
| 272 |
"shufps $0x00,%%xmm0,%%xmm0 # copy pan left to the other 3 cells\n\t" |
| 273 |
"shufps $0x00,%%xmm1,%%xmm1 # copy pan right to the other 3 cells\n\t" |
| 274 |
"mulps %%xmm2,%%xmm0 # left = sample * pan_left\n\t" |
| 275 |
"mulps %%xmm2,%%xmm1 # right = sample * pan_right\n\t" |
| 276 |
"mulps %%xmm4,%%xmm0 # left = vca * (sample * pan_left)\n\t" |
| 277 |
"mulps %%xmm4,%%xmm1 # right = vca * (sample * pan_right)\n\t" |
| 278 |
: /* no output */ |
| 279 |
: "r" (&Volume[ii]), /* %0 */ |
| 280 |
"r" (PanL), /* %1 */ |
| 281 |
"r" (PanR) /* %2 */ |
| 282 |
: "xmm0", /* holds final left sample (for the 4 samples) at the end */ |
| 283 |
"xmm1" /* holds final right sample (for the 4 samples) at the end */ |
| 284 |
); |
| 285 |
break; |
| 286 |
} |
| 287 |
case STEREO: { |
| 288 |
GetNext4SamplesStereoMMXSSE(pSrc, (float *)Pos, Pitch); // outputs samples in xmm2 (left channel) and xmm3 (right channel) |
| 289 |
if (USEFILTER) { |
| 290 |
__asm__ __volatile__ ( |
| 291 |
"movaps %xmm2,%xmm0 # prepare left channel for filter\n\t" |
| 292 |
"movaps %xmm3,%xmm1 # save right channel not to get overwritten by filter algorithms\n\t" |
| 293 |
); |
| 294 |
FilterL.Apply4StepsSSE(&bqBase, &bqMain); // xmm0 input, xmm7 output |
| 295 |
__asm__ __volatile__ ( |
| 296 |
"movaps %xmm1,%xmm0 # prepare right channel for filter\n\t" |
| 297 |
"movaps %xmm7,%xmm1 # save filter output for left channel\n\t" |
| 298 |
); |
| 299 |
FilterR.Apply4StepsSSE(&bqBase, &bqMain); // xmm0 input, xmm7 output |
| 300 |
__asm__ __volatile__ ( |
| 301 |
"movaps %xmm1,%xmm2 # result left channel -> xmm2\n\t" |
| 302 |
"movaps %xmm7,%xmm3 # result right channel -> xmm3\n\t" |
| 303 |
); |
| 304 |
} |
| 305 |
/* apply panorama and volume factors */ |
| 306 |
__asm__ __volatile__ ( |
| 307 |
"movss (%1),%%xmm0 # load pan left\n\t" |
| 308 |
"movss (%2),%%xmm1 # load pan right\n\t" |
| 309 |
"movaps (%0),%%xmm4 # load vca\n\t" |
| 310 |
"shufps $0x00,%%xmm0,%%xmm0 # copy pan left to the other 3 cells\n\t" |
| 311 |
"shufps $0x00,%%xmm1,%%xmm1 # copy pan right to the other 3 cells\n\t" |
| 312 |
"mulps %%xmm2,%%xmm0 # left = sample_left * pan_left\n\t" |
| 313 |
"mulps %%xmm3,%%xmm1 # right = sample_right * pan_right\n\t" |
| 314 |
"mulps %%xmm4,%%xmm0 # left = vca * (sample_left * pan_left)\n\t" |
| 315 |
"mulps %%xmm4,%%xmm1 # right = vca * (sample_right * pan_right)\n\t" |
| 316 |
: /* no output */ |
| 317 |
: "r" (&Volume[ii]), /* %0 */ |
| 318 |
"r" (PanL), /* %1 */ |
| 319 |
"r" (PanR) /* %2 */ |
| 320 |
); |
| 321 |
break; |
| 322 |
} |
| 323 |
} |
| 324 |
/* mix the 4 samples to the output channels */ |
| 325 |
__asm__ __volatile__ ( |
| 326 |
"addps (%0),%%xmm0 # mix calculated sample(s) to output left\n\t" |
| 327 |
"movaps %%xmm0,(%0) # output to left channel\n\t" |
| 328 |
"addps (%1),%%xmm1 # mix calculated sample(s) to output right\n\t" |
| 329 |
"movaps %%xmm1,(%1) # output to right channel\n\t" |
| 330 |
: /* no output */ |
| 331 |
: "r" (&pOutL[ii]), /* %0 - must be 16 byte aligned ! */ |
| 332 |
"r" (&pOutR[ii]) /* %1 - must be 16 byte aligned ! */ |
| 333 |
); |
| 334 |
} |
| 335 |
#endif // ARCH_X86 |
| 336 |
} |
| 337 |
} |
| 338 |
}; |
| 339 |
|
| 340 |
}} // namespace LinuxSampler::gig |
| 341 |
|
| 342 |
#endif // __LS_GIG_SYNTHESIZER_H__ |
Parent Directory
| 
Revision Log