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 *restrict vals, ALuint UNUSED(frac))
  12 { return vals[0]; }
  13 static inline ALfloat lerp32(const ALfloat *restrict vals, ALuint frac)
  14 { return lerp(vals[0], vals[1], frac * (1.0f/FRACTIONONE)); }
  15 static inline ALfloat fir4_32(const ALfloat *restrict vals, ALuint frac)
  16 { return resample_fir4(vals[-1], vals[0], vals[1], vals[2], frac); }
  17 static inline ALfloat fir8_32(const ALfloat *restrict 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 InterpState* UNUSED(state),
  22   const ALfloat *restrict src, ALuint UNUSED(frac), ALint UNUSED(increment),
  23   ALfloat *restrict dst, ALsizei numsamples)
  24 {
  25 #if defined(HAVE_SSE) || defined(HAVE_NEON)
  26     /* Avoid copying the source data if it's aligned like the destination. */
  27     if((((intptr_t)src)&15) == (((intptr_t)dst)&15))
  28         return src;
  29 #endif
  30     memcpy(dst, src, numsamples*sizeof(ALfloat));
  31     return dst;
  32 }
  33
  34 #define DECL_TEMPLATE(Sampler)                                                \
  35 const ALfloat *Resample_##Sampler##_C(const InterpState* UNUSED(state),       \
  36   const ALfloat *restrict src, ALuint frac, ALint increment,                  \
  37   ALfloat *restrict dst, ALsizei numsamples)                                  \
  38 {                                                                             \
  39     ALsizei i;                                                                \
  40     for(i = 0;i < numsamples;i++)                                             \
  41     {                                                                         \
  42         dst[i] = Sampler(src, frac);                                          \
  43                                                                               \
  44         frac += increment;                                                    \
  45         src  += frac>>FRACTIONBITS;                                           \
  46         frac &= FRACTIONMASK;                                                 \
  47     }                                                                         \
  48     return dst;                                                               \
  49 }
  50
  51 DECL_TEMPLATE(point32)
  52 DECL_TEMPLATE(lerp32)
  53 DECL_TEMPLATE(fir4_32)
  54 DECL_TEMPLATE(fir8_32)
  55
  56 #undef DECL_TEMPLATE
  57
  58 const ALfloat *Resample_bsinc32_C(const InterpState *state, const ALfloat *restrict src,
  59                                   ALuint frac, ALint increment, ALfloat *restrict dst,
  60                                   ALsizei dstlen)
  61 {
  62     const ALfloat *fil, *scd, *phd, *spd;
  63     const ALfloat sf = state->bsinc.sf;
  64     const ALsizei m = state->bsinc.m;
  65     ALsizei j_f, pi, i;
  66     ALfloat pf, r;
  67
  68     src += state->bsinc.l;
  69     for(i = 0;i < dstlen;i++)
  70     {
  71         // Calculate the phase index and factor.
  72 #define FRAC_PHASE_BITDIFF (FRACTIONBITS-BSINC_PHASE_BITS)
  73         pi = frac >> FRAC_PHASE_BITDIFF;
  74         pf = (frac & ((1<<FRAC_PHASE_BITDIFF)-1)) * (1.0f/(1<<FRAC_PHASE_BITDIFF));
  75 #undef FRAC_PHASE_BITDIFF
  76
  77         fil = ASSUME_ALIGNED(state->bsinc.coeffs[pi].filter, 16);
  78         scd = ASSUME_ALIGNED(state->bsinc.coeffs[pi].scDelta, 16);
  79         phd = ASSUME_ALIGNED(state->bsinc.coeffs[pi].phDelta, 16);
  80         spd = ASSUME_ALIGNED(state->bsinc.coeffs[pi].spDelta, 16);
  81
  82         // Apply the scale and phase interpolated filter.
  83         r = 0.0f;
  84         for(j_f = 0;j_f < m;j_f++)
  85             r += (fil[j_f] + sf*scd[j_f] + pf*(phd[j_f] + sf*spd[j_f])) *
  86                     src[j_f];
  87         dst[i] = r;
  88
  89         frac += increment;
  90         src  += frac>>FRACTIONBITS;
  91         frac &= FRACTIONMASK;
  92     }
  93     return dst;
  94 }
  95
  96
  97 void ALfilterState_processC(ALfilterState *filter, ALfloat *restrict dst, const ALfloat *restrict src, ALsizei numsamples)
  98 {
  99     ALsizei i;
 100     if(numsamples > 1)
 101     {
 102         dst[0] = filter->b0 * src[0] +
 103                  filter->b1 * filter->x[0] +
 104                  filter->b2 * filter->x[1] -
 105                  filter->a1 * filter->y[0] -
 106                  filter->a2 * filter->y[1];
 107         dst[1] = filter->b0 * src[1] +
 108                  filter->b1 * src[0] +
 109                  filter->b2 * filter->x[0] -
 110                  filter->a1 * dst[0] -
 111                  filter->a2 * filter->y[0];
 112         for(i = 2;i < numsamples;i++)
 113             dst[i] = filter->b0 * src[i] +
 114                      filter->b1 * src[i-1] +
 115                      filter->b2 * src[i-2] -
 116                      filter->a1 * dst[i-1] -
 117                      filter->a2 * dst[i-2];
 118         filter->x[0] = src[i-1];
 119         filter->x[1] = src[i-2];
 120         filter->y[0] = dst[i-1];
 121         filter->y[1] = dst[i-2];
 122     }
 123     else if(numsamples == 1)
 124     {
 125         dst[0] = filter->b0 * src[0] +
 126                  filter->b1 * filter->x[0] +
 127                  filter->b2 * filter->x[1] -
 128                  filter->a1 * filter->y[0] -
 129                  filter->a2 * filter->y[1];
 130         filter->x[1] = filter->x[0];
 131         filter->x[0] = src[0];
 132         filter->y[1] = filter->y[0];
 133         filter->y[0] = dst[0];
 134     }
 135 }
 136
 137
 138 static inline void ApplyCoeffsStep(ALsizei Offset, ALfloat (*restrict Values)[2],
 139                                    const ALsizei IrSize,
 140                                    ALfloat (*restrict Coeffs)[2],
 141                                    const ALfloat (*restrict CoeffStep)[2],
 142                                    ALfloat left, ALfloat right)
 143 {
 144     ALsizei c;
 145     for(c = 0;c < IrSize;c++)
 146     {
 147         const ALsizei off = (Offset+c)&HRIR_MASK;
 148         Values[off][0] += Coeffs[c][0] * left;
 149         Values[off][1] += Coeffs[c][1] * right;
 150         Coeffs[c][0] += CoeffStep[c][0];
 151         Coeffs[c][1] += CoeffStep[c][1];
 152     }
 153 }
 154
 155 static inline void ApplyCoeffs(ALsizei Offset, ALfloat (*restrict Values)[2],
 156                                const ALsizei IrSize,
 157                                ALfloat (*restrict Coeffs)[2],
 158                                ALfloat left, ALfloat right)
 159 {
 160     ALsizei c;
 161     for(c = 0;c < IrSize;c++)
 162     {
 163         const ALsizei off = (Offset+c)&HRIR_MASK;
 164         Values[off][0] += Coeffs[c][0] * left;
 165         Values[off][1] += Coeffs[c][1] * right;
 166     }
 167 }
 168
 169 #define MixHrtf MixHrtf_C
 170 #define MixDirectHrtf MixDirectHrtf_C
 171 #include "mixer_inc.c"
 172 #undef MixHrtf
 173
 174
 175 void Mix_C(const ALfloat *data, ALsizei OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
 176            ALfloat *CurrentGains, const ALfloat *TargetGains, ALsizei Counter, ALsizei OutPos,
 177            ALsizei BufferSize)
 178 {
 179     ALfloat gain, delta, step;
 180     ALsizei c;
 181
 182     delta = (Counter > 0) ? 1.0f/(ALfloat)Counter : 0.0f;
 183
 184     for(c = 0;c < OutChans;c++)
 185     {
 186         ALsizei pos = 0;
 187         gain = CurrentGains[c];
 188         step = (TargetGains[c] - gain) * delta;
 189         if(fabsf(step) > FLT_EPSILON)
 190         {
 191             ALsizei minsize = mini(BufferSize, Counter);
 192             for(;pos < minsize;pos++)
 193             {
 194                 OutBuffer[c][OutPos+pos] += data[pos]*gain;
 195                 gain += step;
 196             }
 197             if(pos == Counter)
 198                 gain = TargetGains[c];
 199             CurrentGains[c] = gain;
 200         }
 201
 202         if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
 203             continue;
 204         for(;pos < BufferSize;pos++)
 205             OutBuffer[c][OutPos+pos] += data[pos]*gain;
 206     }
 207 }
 208
 209 /* Basically the inverse of the above. Rather than one input going to multiple
 210  * outputs (each with its own gain), it's multiple inputs (each with its own
 211  * gain) going to one output. This applies one row (vs one column) of a matrix
 212  * transform. And as the matrices are more or less static once set up, no
 213  * stepping is necessary.
 214  */
 215 void MixRow_C(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restrict data)[BUFFERSIZE], ALsizei InChans, ALsizei InPos, ALsizei BufferSize)
 216 {
 217     ALsizei c, i;
 218
 219     for(c = 0;c < InChans;c++)
 220     {
 221         ALfloat gain = Gains[c];
 222         if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
 223             continue;
 224
 225         for(i = 0;i < BufferSize;i++)
 226             OutBuffer[i] += data[c][InPos+i] * gain;
 227     }
 228 }