Alc/mixer.c

   1 /**
   2  * OpenAL cross platform audio library
   3  * Copyright (C) 1999-2007 by authors.
   4  * This library is free software; you can redistribute it and/or
   5  *  modify it under the terms of the GNU Library General Public
   6  *  License as published by the Free Software Foundation; either
   7  *  version 2 of the License, or (at your option) any later version.
   8  *
   9  * This library is distributed in the hope that it will be useful,
  10  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  *  Library General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU Library General Public
  15  *  License along with this library; if not, write to the
  16  *  Free Software Foundation, Inc.,
  17  *  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  * Or go to http://www.gnu.org/copyleft/lgpl.html
  19  */
  20
  21 #include "config.h"
  22
  23 #include <math.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <ctype.h>
  27 #include <assert.h>
  28
  29 #include "alMain.h"
  30 #include "AL/al.h"
  31 #include "AL/alc.h"
  32 #include "alSource.h"
  33 #include "alBuffer.h"
  34 #include "alListener.h"
  35 #include "alAuxEffectSlot.h"
  36 #include "alu.h"
  37
  38 #include "mixer_defs.h"
  39
  40
  41 extern inline void InitiatePositionArrays(ALuint frac, ALuint increment, ALuint *frac_arr, ALuint *pos_arr, ALuint size);
  42
  43 alignas(16) ALfloat CubicLUT[FRACTIONONE][4];
  44
  45
  46 void aluInitResamplers(void)
  47 {
  48     ALuint i;
  49     for(i = 0;i < FRACTIONONE;i++)
  50     {
  51         ALfloat mu = (ALfloat)i / FRACTIONONE;
  52         ALfloat mu2 = mu*mu, mu3 = mu*mu*mu;
  53         CubicLUT[i][0] = -0.5f*mu3 +       mu2 + -0.5f*mu;
  54         CubicLUT[i][1] =  1.5f*mu3 + -2.5f*mu2            + 1.0f;
  55         CubicLUT[i][2] = -1.5f*mu3 +  2.0f*mu2 +  0.5f*mu;
  56         CubicLUT[i][3] =  0.5f*mu3 + -0.5f*mu2;
  57     }
  58 }
  59
  60
  61 static inline HrtfMixerFunc SelectHrtfMixer(void)
  62 {
  63 #ifdef HAVE_SSE
  64     if((CPUCapFlags&CPU_CAP_SSE))
  65         return MixHrtf_SSE;
  66 #endif
  67 #ifdef HAVE_NEON
  68     if((CPUCapFlags&CPU_CAP_NEON))
  69         return MixHrtf_Neon;
  70 #endif
  71
  72     return MixHrtf_C;
  73 }
  74
  75 static inline MixerFunc SelectMixer(void)
  76 {
  77 #ifdef HAVE_SSE
  78     if((CPUCapFlags&CPU_CAP_SSE))
  79         return Mix_SSE;
  80 #endif
  81 #ifdef HAVE_NEON
  82     if((CPUCapFlags&CPU_CAP_NEON))
  83         return Mix_Neon;
  84 #endif
  85
  86     return Mix_C;
  87 }
  88
  89 static inline ResamplerFunc SelectResampler(enum Resampler resampler)
  90 {
  91     switch(resampler)
  92     {
  93         case PointResampler:
  94             return Resample_point32_C;
  95         case LinearResampler:
  96 #ifdef HAVE_SSE4_1
  97             if((CPUCapFlags&CPU_CAP_SSE4_1))
  98                 return Resample_lerp32_SSE41;
  99 #endif
 100 #ifdef HAVE_SSE2
 101             if((CPUCapFlags&CPU_CAP_SSE2))
 102                 return Resample_lerp32_SSE2;
 103 #endif
 104             return Resample_lerp32_C;
 105         case CubicResampler:
 106 #ifdef HAVE_SSE4_1
 107             if((CPUCapFlags&CPU_CAP_SSE4_1))
 108                 return Resample_cubic32_SSE41;
 109 #endif
 110 #ifdef HAVE_SSE2
 111             if((CPUCapFlags&CPU_CAP_SSE2))
 112                 return Resample_cubic32_SSE2;
 113 #endif
 114             return Resample_cubic32_C;
 115         case ResamplerMax:
 116             /* Shouldn't happen */
 117             break;
 118     }
 119
 120     return Resample_point32_C;
 121 }
 122
 123
 124 static inline ALfloat Sample_ALbyte(ALbyte val)
 125 { return val * (1.0f/127.0f); }
 126
 127 static inline ALfloat Sample_ALshort(ALshort val)
 128 { return val * (1.0f/32767.0f); }
 129
 130 static inline ALfloat Sample_ALfloat(ALfloat val)
 131 { return val; }
 132
 133 #define DECL_TEMPLATE(T)                                                      \
 134 static void Load_##T(ALfloat *dst, const T *src, ALuint srcstep, ALuint samples)\
 135 {                                                                             \
 136     ALuint i;                                                                 \
 137     for(i = 0;i < samples;i++)                                                \
 138         dst[i] = Sample_##T(src[i*srcstep]);                                  \
 139 }
 140
 141 DECL_TEMPLATE(ALbyte)
 142 DECL_TEMPLATE(ALshort)
 143 DECL_TEMPLATE(ALfloat)
 144
 145 #undef DECL_TEMPLATE
 146
 147 static void LoadSamples(ALfloat *dst, const ALvoid *src, ALuint srcstep, enum FmtType srctype, ALuint samples)
 148 {
 149     switch(srctype)
 150     {
 151         case FmtByte:
 152             Load_ALbyte(dst, src, srcstep, samples);
 153             break;
 154         case FmtShort:
 155             Load_ALshort(dst, src, srcstep, samples);
 156             break;
 157         case FmtFloat:
 158             Load_ALfloat(dst, src, srcstep, samples);
 159             break;
 160     }
 161 }
 162
 163 static void SilenceSamples(ALfloat *dst, ALuint samples)
 164 {
 165     ALuint i;
 166     for(i = 0;i < samples;i++)
 167         dst[i] = 0.0f;
 168 }
 169
 170
 171 static const ALfloat *DoFilters(ALfilterState *lpfilter, ALfilterState *hpfilter,
 172                                 ALfloat *restrict dst, const ALfloat *restrict src,
 173                                 ALuint numsamples, enum ActiveFilters type)
 174 {
 175     ALuint i;
 176     switch(type)
 177     {
 178         case AF_None:
 179             break;
 180
 181         case AF_LowPass:
 182             ALfilterState_process(lpfilter, dst, src, numsamples);
 183             return dst;
 184         case AF_HighPass:
 185             ALfilterState_process(hpfilter, dst, src, numsamples);
 186             return dst;
 187
 188         case AF_BandPass:
 189             for(i = 0;i < numsamples;)
 190             {
 191                 ALfloat temp[64];
 192                 ALuint todo = minu(64, numsamples-i);
 193
 194                 ALfilterState_process(lpfilter, temp, src+i, todo);
 195                 ALfilterState_process(hpfilter, dst+i, temp, todo);
 196                 i += todo;
 197             }
 198             return dst;
 199     }
 200     return src;
 201 }
 202
 203
 204 ALvoid MixSource(ALvoice *voice, ALsource *Source, ALCdevice *Device, ALuint SamplesToDo)
 205 {
 206     MixerFunc Mix;
 207     HrtfMixerFunc HrtfMix;
 208     ResamplerFunc Resample;
 209     ALbufferlistitem *BufferListItem;
 210     ALuint DataPosInt, DataPosFrac;
 211     ALboolean isbformat = AL_FALSE;
 212     ALboolean Looping;
 213     ALuint increment;
 214     enum Resampler Resampler;
 215     ALenum State;
 216     ALuint OutPos;
 217     ALuint NumChannels;
 218     ALuint SampleSize;
 219     ALint64 DataSize64;
 220     ALuint IrSize;
 221     ALuint chan, j;
 222
 223     /* Get source info */
 224     State          = Source->state;
 225     BufferListItem = ATOMIC_LOAD(&Source->current_buffer);
 226     DataPosInt     = Source->position;
 227     DataPosFrac    = Source->position_fraction;
 228     Looping        = Source->Looping;
 229     Resampler      = Source->Resampler;
 230     NumChannels    = Source->NumChannels;
 231     SampleSize     = Source->SampleSize;
 232     increment      = voice->Step;
 233
 234     while(BufferListItem)
 235     {
 236         ALbuffer *buffer;
 237         if((buffer=BufferListItem->buffer) != NULL)
 238         {
 239             isbformat = (buffer->FmtChannels == FmtBFormat2D ||
 240                          buffer->FmtChannels == FmtBFormat3D);
 241             break;
 242         }
 243         BufferListItem = BufferListItem->next;
 244     }
 245
 246     IrSize = (Device->Hrtf ? GetHrtfIrSize(Device->Hrtf) : 0);
 247
 248     Mix = SelectMixer();
 249     HrtfMix = SelectHrtfMixer();
 250     Resample = ((increment == FRACTIONONE && DataPosFrac == 0) ?
 251                 Resample_copy32_C : SelectResampler(Resampler));
 252
 253     OutPos = 0;
 254     do {
 255         const ALuint BufferPrePadding = ResamplerPrePadding[Resampler];
 256         const ALuint BufferPadding = ResamplerPadding[Resampler];
 257         ALuint SrcBufferSize, DstBufferSize;
 258
 259         /* Figure out how many buffer samples will be needed */
 260         DataSize64  = SamplesToDo-OutPos;
 261         DataSize64 *= increment;
 262         DataSize64 += DataPosFrac+FRACTIONMASK;
 263         DataSize64 >>= FRACTIONBITS;
 264         DataSize64 += BufferPadding+BufferPrePadding;
 265
 266         SrcBufferSize = (ALuint)mini64(DataSize64, BUFFERSIZE);
 267
 268         /* Figure out how many samples we can actually mix from this. */
 269         DataSize64  = SrcBufferSize;
 270         DataSize64 -= BufferPadding+BufferPrePadding;
 271         DataSize64 <<= FRACTIONBITS;
 272         DataSize64 -= DataPosFrac;
 273
 274         DstBufferSize = (ALuint)((DataSize64+(increment-1)) / increment);
 275         DstBufferSize = minu(DstBufferSize, (SamplesToDo-OutPos));
 276
 277         /* Some mixers like having a multiple of 4, so try to give that unless
 278          * this is the last update. */
 279         if(OutPos+DstBufferSize < SamplesToDo)
 280             DstBufferSize &= ~3;
 281
 282         for(chan = 0;chan < NumChannels;chan++)
 283         {
 284             const ALfloat *ResampledData;
 285             ALfloat *SrcData = Device->SourceData;
 286             ALuint SrcDataSize = 0;
 287
 288             if(Source->SourceType == AL_STATIC)
 289             {
 290                 const ALbuffer *ALBuffer = BufferListItem->buffer;
 291                 const ALubyte *Data = ALBuffer->data;
 292                 ALuint DataSize;
 293                 ALuint pos;
 294
 295                 /* If current pos is beyond the loop range, do not loop */
 296                 if(Looping == AL_FALSE || DataPosInt >= (ALuint)ALBuffer->LoopEnd)
 297                 {
 298                     Looping = AL_FALSE;
 299
 300                     if(DataPosInt >= BufferPrePadding)
 301                         pos = DataPosInt - BufferPrePadding;
 302                     else
 303                     {
 304                         DataSize = BufferPrePadding - DataPosInt;
 305                         DataSize = minu(SrcBufferSize - SrcDataSize, DataSize);
 306
 307                         SilenceSamples(&SrcData[SrcDataSize], DataSize);
 308                         SrcDataSize += DataSize;
 309
 310                         pos = 0;
 311                     }
 312
 313                     /* Copy what's left to play in the source buffer, and clear the
 314                      * rest of the temp buffer */
 315                     DataSize = minu(SrcBufferSize - SrcDataSize, ALBuffer->SampleLen - pos);
 316
 317                     LoadSamples(&SrcData[SrcDataSize], &Data[(pos*NumChannels + chan)*SampleSize],
 318                                 NumChannels, ALBuffer->FmtType, DataSize);
 319                     SrcDataSize += DataSize;
 320
 321                     SilenceSamples(&SrcData[SrcDataSize], SrcBufferSize - SrcDataSize);
 322                     SrcDataSize += SrcBufferSize - SrcDataSize;
 323                 }
 324                 else
 325                 {
 326                     ALuint LoopStart = ALBuffer->LoopStart;
 327                     ALuint LoopEnd   = ALBuffer->LoopEnd;
 328
 329                     if(DataPosInt >= LoopStart)
 330                     {
 331                         pos = DataPosInt-LoopStart;
 332                         while(pos < BufferPrePadding)
 333                             pos += LoopEnd-LoopStart;
 334                         pos -= BufferPrePadding;
 335                         pos += LoopStart;
 336                     }
 337                     else if(DataPosInt >= BufferPrePadding)
 338                         pos = DataPosInt - BufferPrePadding;
 339                     else
 340                     {
 341                         DataSize = BufferPrePadding - DataPosInt;
 342                         DataSize = minu(SrcBufferSize - SrcDataSize, DataSize);
 343
 344                         SilenceSamples(&SrcData[SrcDataSize], DataSize);
 345                         SrcDataSize += DataSize;
 346
 347                         pos = 0;
 348                     }
 349
 350                     /* Copy what's left of this loop iteration, then copy repeats
 351                      * of the loop section */
 352                     DataSize = LoopEnd - pos;
 353                     DataSize = minu(SrcBufferSize - SrcDataSize, DataSize);
 354
 355                     LoadSamples(&SrcData[SrcDataSize], &Data[(pos*NumChannels + chan)*SampleSize],
 356                                 NumChannels, ALBuffer->FmtType, DataSize);
 357                     SrcDataSize += DataSize;
 358
 359                     DataSize = LoopEnd-LoopStart;
 360                     while(SrcBufferSize > SrcDataSize)
 361                     {
 362                         DataSize = minu(SrcBufferSize - SrcDataSize, DataSize);
 363
 364                         LoadSamples(&SrcData[SrcDataSize], &Data[(LoopStart*NumChannels + chan)*SampleSize],
 365                                     NumChannels, ALBuffer->FmtType, DataSize);
 366                         SrcDataSize += DataSize;
 367                     }
 368                 }
 369             }
 370             else
 371             {
 372                 /* Crawl the buffer queue to fill in the temp buffer */
 373                 ALbufferlistitem *tmpiter = BufferListItem;
 374                 ALuint pos;
 375
 376                 if(DataPosInt >= BufferPrePadding)
 377                     pos = DataPosInt - BufferPrePadding;
 378                 else
 379                 {
 380                     pos = BufferPrePadding - DataPosInt;
 381                     while(pos > 0)
 382                     {
 383                         ALbufferlistitem *prev;
 384                         if((prev=tmpiter->prev) != NULL)
 385                             tmpiter = prev;
 386                         else if(Looping)
 387                         {
 388                             while(tmpiter->next)
 389                                 tmpiter = tmpiter->next;
 390                         }
 391                         else
 392                         {
 393                             ALuint DataSize = minu(SrcBufferSize - SrcDataSize, pos);
 394
 395                             SilenceSamples(&SrcData[SrcDataSize], DataSize);
 396                             SrcDataSize += DataSize;
 397
 398                             pos = 0;
 399                             break;
 400                         }
 401
 402                         if(tmpiter->buffer)
 403                         {
 404                             if((ALuint)tmpiter->buffer->SampleLen > pos)
 405                             {
 406                                 pos = tmpiter->buffer->SampleLen - pos;
 407                                 break;
 408                             }
 409                             pos -= tmpiter->buffer->SampleLen;
 410                         }
 411                     }
 412                 }
 413
 414                 while(tmpiter && SrcBufferSize > SrcDataSize)
 415                 {
 416                     const ALbuffer *ALBuffer;
 417                     if((ALBuffer=tmpiter->buffer) != NULL)
 418                     {
 419                         const ALubyte *Data = ALBuffer->data;
 420                         ALuint DataSize = ALBuffer->SampleLen;
 421
 422                         /* Skip the data already played */
 423                         if(DataSize <= pos)
 424                             pos -= DataSize;
 425                         else
 426                         {
 427                             Data += (pos*NumChannels + chan)*SampleSize;
 428                             DataSize -= pos;
 429                             pos -= pos;
 430
 431                             DataSize = minu(SrcBufferSize - SrcDataSize, DataSize);
 432                             LoadSamples(&SrcData[SrcDataSize], Data, NumChannels,
 433                                         ALBuffer->FmtType, DataSize);
 434                             SrcDataSize += DataSize;
 435                         }
 436                     }
 437                     tmpiter = tmpiter->next;
 438                     if(!tmpiter && Looping)
 439                         tmpiter = ATOMIC_LOAD(&Source->queue);
 440                     else if(!tmpiter)
 441                     {
 442                         SilenceSamples(&SrcData[SrcDataSize], SrcBufferSize - SrcDataSize);
 443                         SrcDataSize += SrcBufferSize - SrcDataSize;
 444                     }
 445                 }
 446             }
 447
 448             /* Now resample, then filter and mix to the appropriate outputs. */
 449             ResampledData = Resample(
 450                 &SrcData[BufferPrePadding], DataPosFrac, increment,
 451                 Device->ResampledData, DstBufferSize
 452             );
 453             {
 454                 DirectParams *parms = &voice->Direct;
 455                 const ALfloat *samples;
 456
 457                 samples = DoFilters(
 458                     &parms->Filters[chan].LowPass, &parms->Filters[chan].HighPass,
 459                     Device->FilteredData, ResampledData, DstBufferSize,
 460                     parms->Filters[chan].ActiveType
 461                 );
 462                 if(!voice->IsHrtf)
 463                     Mix(samples, parms->OutChannels, parms->OutBuffer, parms->Gains[chan],
 464                         parms->Counter, OutPos, DstBufferSize);
 465                 else
 466                     HrtfMix(parms->OutBuffer, samples, parms->Counter, voice->Offset,
 467                             OutPos, IrSize, &parms->Hrtf.Params[chan],
 468                             &parms->Hrtf.State[chan], DstBufferSize);
 469             }
 470
 471             /* Only the first channel for B-Format buffers (W channel) goes to
 472              * the send paths. */
 473             if(chan > 0 && isbformat)
 474                 continue;
 475             for(j = 0;j < Device->NumAuxSends;j++)
 476             {
 477                 SendParams *parms = &voice->Send[j];
 478                 const ALfloat *samples;
 479
 480                 if(!parms->OutBuffer)
 481                     continue;
 482
 483                 samples = DoFilters(
 484                     &parms->Filters[chan].LowPass, &parms->Filters[chan].HighPass,
 485                     Device->FilteredData, ResampledData, DstBufferSize,
 486                     parms->Filters[chan].ActiveType
 487                 );
 488                 Mix(samples, 1, parms->OutBuffer, &parms->Gain,
 489                     parms->Counter, OutPos, DstBufferSize);
 490             }
 491         }
 492         /* Update positions */
 493         DataPosFrac += increment*DstBufferSize;
 494         DataPosInt  += DataPosFrac>>FRACTIONBITS;
 495         DataPosFrac &= FRACTIONMASK;
 496
 497         OutPos += DstBufferSize;
 498         voice->Offset += DstBufferSize;
 499         voice->Direct.Counter = maxu(voice->Direct.Counter, DstBufferSize) - DstBufferSize;
 500         for(j = 0;j < Device->NumAuxSends;j++)
 501             voice->Send[j].Counter = maxu(voice->Send[j].Counter, DstBufferSize) - DstBufferSize;
 502
 503         /* Handle looping sources */
 504         while(1)
 505         {
 506             const ALbuffer *ALBuffer;
 507             ALuint DataSize = 0;
 508             ALuint LoopStart = 0;
 509             ALuint LoopEnd = 0;
 510
 511             if((ALBuffer=BufferListItem->buffer) != NULL)
 512             {
 513                 DataSize = ALBuffer->SampleLen;
 514                 LoopStart = ALBuffer->LoopStart;
 515                 LoopEnd = ALBuffer->LoopEnd;
 516                 if(LoopEnd > DataPosInt)
 517                     break;
 518             }
 519
 520             if(Looping && Source->SourceType == AL_STATIC)
 521             {
 522                 assert(LoopEnd > LoopStart);
 523                 DataPosInt = ((DataPosInt-LoopStart)%(LoopEnd-LoopStart)) + LoopStart;
 524                 break;
 525             }
 526
 527             if(DataSize > DataPosInt)
 528                 break;
 529
 530             if(!(BufferListItem=BufferListItem->next))
 531             {
 532                 if(Looping)
 533                     BufferListItem = ATOMIC_LOAD(&Source->queue);
 534                 else
 535                 {
 536                     State = AL_STOPPED;
 537                     BufferListItem = NULL;
 538                     DataPosInt = 0;
 539                     DataPosFrac = 0;
 540                     break;
 541                 }
 542             }
 543
 544             DataPosInt -= DataSize;
 545         }
 546     } while(State == AL_PLAYING && OutPos < SamplesToDo);
 547
 548     /* Update source info */
 549     Source->state             = State;
 550     ATOMIC_STORE(&Source->current_buffer, BufferListItem);
 551     Source->position          = DataPosInt;
 552     Source->position_fraction = DataPosFrac;
 553 }