Alc/mixer_c.c

   1 #include "config.h"
   2
   3 #include <assert.h>
   4
   5 #include "alMain.h"
   6 #include "alu.h"
   7 #include "alSource.h"
   8 #include "alAuxEffectSlot.h"
   9
  10
  11 static inline ALfloat point32(const ALfloat *vals, ALuint UNUSED(frac))
  12 { return vals[0]; }
  13 static inline ALfloat lerp32(const ALfloat *vals, ALuint frac)
  14 { return lerp(vals[0], vals[1], frac * (1.0f/FRACTIONONE)); }
  15 static inline ALfloat fir4_32(const ALfloat *vals, ALuint frac)
  16 { return resample_fir4(vals[-1], vals[0], vals[1], vals[2], frac); }
  17 static inline ALfloat fir8_32(const ALfloat *vals, ALuint frac)
  18 { return resample_fir8(vals[-3], vals[-2], vals[-1], vals[0], vals[1], vals[2], vals[3], vals[4], frac); }
  19
  20
  21 const ALfloat *Resample_copy32_C(const BsincState* UNUSED(state), const ALfloat *src, ALuint UNUSED(frac),
  22   ALuint UNUSED(increment), ALfloat *restrict dst, ALuint numsamples)
  23 {
  24 #if defined(HAVE_SSE) || defined(HAVE_NEON)
  25     /* Avoid copying the source data if it's aligned like the destination. */
  26     if((((intptr_t)src)&15) == (((intptr_t)dst)&15))
  27         return src;
  28 #endif
  29     memcpy(dst, src, numsamples*sizeof(ALfloat));
  30     return dst;
  31 }
  32
  33 #define DECL_TEMPLATE(Sampler)                                                \
  34 const ALfloat *Resample_##Sampler##_C(const BsincState* UNUSED(state),        \
  35   const ALfloat *src, ALuint frac, ALuint increment,                          \
  36   ALfloat *restrict dst, ALuint numsamples)                                   \
  37 {                                                                             \
  38     ALuint i;                                                                 \
  39     for(i = 0;i < numsamples;i++)                                             \
  40     {                                                                         \
  41         dst[i] = Sampler(src, frac);                                          \
  42                                                                               \
  43         frac += increment;                                                    \
  44         src  += frac>>FRACTIONBITS;                                           \
  45         frac &= FRACTIONMASK;                                                 \
  46     }                                                                         \
  47     return dst;                                                               \
  48 }
  49
  50 DECL_TEMPLATE(point32)
  51 DECL_TEMPLATE(lerp32)
  52 DECL_TEMPLATE(fir4_32)
  53 DECL_TEMPLATE(fir8_32)
  54
  55 #undef DECL_TEMPLATE
  56
  57 const ALfloat *Resample_bsinc32_C(const BsincState *state, const ALfloat *src, ALuint frac,
  58                                   ALuint increment, ALfloat *restrict dst, ALuint dstlen)
  59 {
  60     const ALfloat *fil, *scd, *phd, *spd;
  61     const ALfloat sf = state->sf;
  62     const ALuint m = state->m;
  63     const ALint l = state->l;
  64     ALuint j_f, pi, i;
  65     ALfloat pf, r;
  66     ALint j_s;
  67
  68     for(i = 0;i < dstlen;i++)
  69     {
  70         // Calculate the phase index and factor.
  71 #define FRAC_PHASE_BITDIFF (FRACTIONBITS-BSINC_PHASE_BITS)
  72         pi = frac >> FRAC_PHASE_BITDIFF;
  73         pf = (frac & ((1<<FRAC_PHASE_BITDIFF)-1)) * (1.0f/(1<<FRAC_PHASE_BITDIFF));
  74 #undef FRAC_PHASE_BITDIFF
  75
  76         fil = state->coeffs[pi].filter;
  77         scd = state->coeffs[pi].scDelta;
  78         phd = state->coeffs[pi].phDelta;
  79         spd = state->coeffs[pi].spDelta;
  80
  81         // Apply the scale and phase interpolated filter.
  82         r = 0.0f;
  83         for(j_f = 0,j_s = l;j_f < m;j_f++,j_s++)
  84             r += (fil[j_f] + sf*scd[j_f] + pf*(phd[j_f] + sf*spd[j_f])) *
  85                     src[j_s];
  86         dst[i] = r;
  87
  88         frac += increment;
  89         src  += frac>>FRACTIONBITS;
  90         frac &= FRACTIONMASK;
  91     }
  92     return dst;
  93 }
  94
  95
  96 void ALfilterState_processC(ALfilterState *filter, ALfloat *restrict dst, const ALfloat *restrict src, ALuint numsamples)
  97 {
  98     ALuint i;
  99     if(numsamples > 1)
 100     {
 101         dst[0] = filter->b0 * src[0] +
 102                  filter->b1 * filter->x[0] +
 103                  filter->b2 * filter->x[1] -
 104                  filter->a1 * filter->y[0] -
 105                  filter->a2 * filter->y[1];
 106         dst[1] = filter->b0 * src[1] +
 107                  filter->b1 * src[0] +
 108                  filter->b2 * filter->x[0] -
 109                  filter->a1 * dst[0] -
 110                  filter->a2 * filter->y[0];
 111         for(i = 2;i < numsamples;i++)
 112             dst[i] = filter->b0 * src[i] +
 113                      filter->b1 * src[i-1] +
 114                      filter->b2 * src[i-2] -
 115                      filter->a1 * dst[i-1] -
 116                      filter->a2 * dst[i-2];
 117         filter->x[0] = src[i-1];
 118         filter->x[1] = src[i-2];
 119         filter->y[0] = dst[i-1];
 120         filter->y[1] = dst[i-2];
 121     }
 122     else if(numsamples == 1)
 123     {
 124         dst[0] = filter->b0 * src[0] +
 125                  filter->b1 * filter->x[0] +
 126                  filter->b2 * filter->x[1] -
 127                  filter->a1 * filter->y[0] -
 128                  filter->a2 * filter->y[1];
 129         filter->x[1] = filter->x[0];
 130         filter->x[0] = src[0];
 131         filter->y[1] = filter->y[0];
 132         filter->y[0] = dst[0];
 133     }
 134 }
 135
 136
 137 static inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*restrict Values)[2],
 138                                    const ALuint IrSize,
 139                                    ALfloat (*restrict Coeffs)[2],
 140                                    const ALfloat (*restrict CoeffStep)[2],
 141                                    ALfloat left, ALfloat right)
 142 {
 143     ALuint c;
 144     for(c = 0;c < IrSize;c++)
 145     {
 146         const ALuint off = (Offset+c)&HRIR_MASK;
 147         Values[off][0] += Coeffs[c][0] * left;
 148         Values[off][1] += Coeffs[c][1] * right;
 149         Coeffs[c][0] += CoeffStep[c][0];
 150         Coeffs[c][1] += CoeffStep[c][1];
 151     }
 152 }
 153
 154 static inline void ApplyCoeffs(ALuint Offset, ALfloat (*restrict Values)[2],
 155                                const ALuint IrSize,
 156                                ALfloat (*restrict Coeffs)[2],
 157                                ALfloat left, ALfloat right)
 158 {
 159     ALuint c;
 160     for(c = 0;c < IrSize;c++)
 161     {
 162         const ALuint off = (Offset+c)&HRIR_MASK;
 163         Values[off][0] += Coeffs[c][0] * left;
 164         Values[off][1] += Coeffs[c][1] * right;
 165     }
 166 }
 167
 168 #define MixHrtf MixHrtf_C
 169 #include "mixer_inc.c"
 170 #undef MixHrtf
 171
 172
 173 void Mix_C(const ALfloat *data, ALuint OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
 174            MixGains *Gains, ALuint Counter, ALuint OutPos, ALuint BufferSize)
 175 {
 176     ALfloat gain, step;
 177     ALuint c;
 178
 179     for(c = 0;c < OutChans;c++)
 180     {
 181         ALuint pos = 0;
 182         gain = Gains[c].Current;
 183         step = Gains[c].Step;
 184         if(step != 0.0f && Counter > 0)
 185         {
 186             ALuint minsize = minu(BufferSize, Counter);
 187             for(;pos < minsize;pos++)
 188             {
 189                 OutBuffer[c][OutPos+pos] += data[pos]*gain;
 190                 gain += step;
 191             }
 192             if(pos == Counter)
 193                 gain = Gains[c].Target;
 194             Gains[c].Current = gain;
 195         }
 196
 197         if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
 198             continue;
 199         for(;pos < BufferSize;pos++)
 200             OutBuffer[c][OutPos+pos] += data[pos]*gain;
 201     }
 202 }
 203
 204 /* Basically the inverse of the above. Rather than one input going to multiple
 205  * outputs (each with its own gain), it's multiple inputs (each with its own
 206  * gain) going to one output. This applies one row (vs one column) of a matrix
 207  * transform. And as the matrices are more or less static once set up, no
 208  * stepping is necessary.
 209  */
 210 void MixRow_C(ALfloat *OutBuffer, const ALfloat *Mtx, ALfloat (*restrict data)[BUFFERSIZE], ALuint InChans, ALuint BufferSize)
 211 {
 212     ALuint c, i;
 213
 214     for(c = 0;c < InChans;c++)
 215     {
 216         ALfloat gain = Mtx[c];
 217         if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
 218             continue;
 219
 220         for(i = 0;i < BufferSize;i++)
 221             OutBuffer[i] += data[c][i] * gain;
 222     }
 223 }