203 lines
6.8 KiB
C
203 lines
6.8 KiB
C
#include "config.h"
|
|
|
|
#ifdef IN_IDE_PARSER
|
|
/* KDevelop's parser won't recognize these defines that get added by the -msse
|
|
* switch used to compile this source. Without them, xmmintrin.h fails to
|
|
* declare anything. */
|
|
#define __MMX__
|
|
#define __SSE__
|
|
#endif
|
|
#include <xmmintrin.h>
|
|
|
|
#include "AL/al.h"
|
|
#include "AL/alc.h"
|
|
#include "alMain.h"
|
|
#include "alu.h"
|
|
|
|
#include "alSource.h"
|
|
#include "alAuxEffectSlot.h"
|
|
#include "mixer_defs.h"
|
|
|
|
|
|
static inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*restrict Values)[2],
|
|
const ALuint IrSize,
|
|
ALfloat (*restrict Coeffs)[2],
|
|
const ALfloat (*restrict CoeffStep)[2],
|
|
ALfloat left, ALfloat right)
|
|
{
|
|
const __m128 lrlr = _mm_setr_ps(left, right, left, right);
|
|
__m128 coeffs, deltas, imp0, imp1;
|
|
__m128 vals = _mm_setzero_ps();
|
|
ALuint i;
|
|
|
|
if((Offset&1))
|
|
{
|
|
const ALuint o0 = Offset&HRIR_MASK;
|
|
const ALuint o1 = (Offset+IrSize-1)&HRIR_MASK;
|
|
|
|
coeffs = _mm_load_ps(&Coeffs[0][0]);
|
|
deltas = _mm_load_ps(&CoeffStep[0][0]);
|
|
vals = _mm_loadl_pi(vals, (__m64*)&Values[o0][0]);
|
|
imp0 = _mm_mul_ps(lrlr, coeffs);
|
|
coeffs = _mm_add_ps(coeffs, deltas);
|
|
vals = _mm_add_ps(imp0, vals);
|
|
_mm_store_ps(&Coeffs[0][0], coeffs);
|
|
_mm_storel_pi((__m64*)&Values[o0][0], vals);
|
|
for(i = 1;i < IrSize-1;i += 2)
|
|
{
|
|
const ALuint o2 = (Offset+i)&HRIR_MASK;
|
|
|
|
coeffs = _mm_load_ps(&Coeffs[i+1][0]);
|
|
deltas = _mm_load_ps(&CoeffStep[i+1][0]);
|
|
vals = _mm_load_ps(&Values[o2][0]);
|
|
imp1 = _mm_mul_ps(lrlr, coeffs);
|
|
coeffs = _mm_add_ps(coeffs, deltas);
|
|
imp0 = _mm_shuffle_ps(imp0, imp1, _MM_SHUFFLE(1, 0, 3, 2));
|
|
vals = _mm_add_ps(imp0, vals);
|
|
_mm_store_ps(&Coeffs[i+1][0], coeffs);
|
|
_mm_store_ps(&Values[o2][0], vals);
|
|
imp0 = imp1;
|
|
}
|
|
vals = _mm_loadl_pi(vals, (__m64*)&Values[o1][0]);
|
|
imp0 = _mm_movehl_ps(imp0, imp0);
|
|
vals = _mm_add_ps(imp0, vals);
|
|
_mm_storel_pi((__m64*)&Values[o1][0], vals);
|
|
}
|
|
else
|
|
{
|
|
for(i = 0;i < IrSize;i += 2)
|
|
{
|
|
const ALuint o = (Offset + i)&HRIR_MASK;
|
|
|
|
coeffs = _mm_load_ps(&Coeffs[i][0]);
|
|
deltas = _mm_load_ps(&CoeffStep[i][0]);
|
|
vals = _mm_load_ps(&Values[o][0]);
|
|
imp0 = _mm_mul_ps(lrlr, coeffs);
|
|
coeffs = _mm_add_ps(coeffs, deltas);
|
|
vals = _mm_add_ps(imp0, vals);
|
|
_mm_store_ps(&Coeffs[i][0], coeffs);
|
|
_mm_store_ps(&Values[o][0], vals);
|
|
}
|
|
}
|
|
}
|
|
|
|
static inline void ApplyCoeffs(ALuint Offset, ALfloat (*restrict Values)[2],
|
|
const ALuint IrSize,
|
|
ALfloat (*restrict Coeffs)[2],
|
|
ALfloat left, ALfloat right)
|
|
{
|
|
const __m128 lrlr = _mm_setr_ps(left, right, left, right);
|
|
__m128 vals = _mm_setzero_ps();
|
|
__m128 coeffs;
|
|
ALuint i;
|
|
|
|
if((Offset&1))
|
|
{
|
|
const ALuint o0 = Offset&HRIR_MASK;
|
|
const ALuint o1 = (Offset+IrSize-1)&HRIR_MASK;
|
|
__m128 imp0, imp1;
|
|
|
|
coeffs = _mm_load_ps(&Coeffs[0][0]);
|
|
vals = _mm_loadl_pi(vals, (__m64*)&Values[o0][0]);
|
|
imp0 = _mm_mul_ps(lrlr, coeffs);
|
|
vals = _mm_add_ps(imp0, vals);
|
|
_mm_storel_pi((__m64*)&Values[o0][0], vals);
|
|
for(i = 1;i < IrSize-1;i += 2)
|
|
{
|
|
const ALuint o2 = (Offset+i)&HRIR_MASK;
|
|
|
|
coeffs = _mm_load_ps(&Coeffs[i+1][0]);
|
|
vals = _mm_load_ps(&Values[o2][0]);
|
|
imp1 = _mm_mul_ps(lrlr, coeffs);
|
|
imp0 = _mm_shuffle_ps(imp0, imp1, _MM_SHUFFLE(1, 0, 3, 2));
|
|
vals = _mm_add_ps(imp0, vals);
|
|
_mm_store_ps(&Values[o2][0], vals);
|
|
imp0 = imp1;
|
|
}
|
|
vals = _mm_loadl_pi(vals, (__m64*)&Values[o1][0]);
|
|
imp0 = _mm_movehl_ps(imp0, imp0);
|
|
vals = _mm_add_ps(imp0, vals);
|
|
_mm_storel_pi((__m64*)&Values[o1][0], vals);
|
|
}
|
|
else
|
|
{
|
|
for(i = 0;i < IrSize;i += 2)
|
|
{
|
|
const ALuint o = (Offset + i)&HRIR_MASK;
|
|
|
|
coeffs = _mm_load_ps(&Coeffs[i][0]);
|
|
vals = _mm_load_ps(&Values[o][0]);
|
|
vals = _mm_add_ps(vals, _mm_mul_ps(lrlr, coeffs));
|
|
_mm_store_ps(&Values[o][0], vals);
|
|
}
|
|
}
|
|
}
|
|
|
|
#define SUFFIX SSE
|
|
#include "mixer_inc.c"
|
|
#undef SUFFIX
|
|
|
|
|
|
void Mix_SSE(const ALfloat *data, ALuint OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
|
|
MixGains *Gains, ALuint Counter, ALuint OutPos, ALuint BufferSize)
|
|
{
|
|
ALfloat gain, step;
|
|
__m128 gain4, step4;
|
|
ALuint c;
|
|
|
|
for(c = 0;c < OutChans;c++)
|
|
{
|
|
ALuint pos = 0;
|
|
gain = Gains[c].Current;
|
|
step = Gains[c].Step;
|
|
if(step != 1.0f && Counter > 0)
|
|
{
|
|
/* Mix with applying gain steps in aligned multiples of 4. */
|
|
if(BufferSize-pos > 3 && Counter-pos > 3)
|
|
{
|
|
gain4 = _mm_setr_ps(
|
|
gain,
|
|
gain * step,
|
|
gain * step * step,
|
|
gain * step * step * step
|
|
);
|
|
step4 = _mm_set1_ps(step * step * step * step);
|
|
do {
|
|
const __m128 val4 = _mm_load_ps(&data[pos]);
|
|
__m128 dry4 = _mm_load_ps(&OutBuffer[c][OutPos+pos]);
|
|
dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
|
|
gain4 = _mm_mul_ps(gain4, step4);
|
|
_mm_store_ps(&OutBuffer[c][OutPos+pos], dry4);
|
|
pos += 4;
|
|
} while(BufferSize-pos > 3 && Counter-pos > 3);
|
|
gain = _mm_cvtss_f32(gain4);
|
|
}
|
|
/* Mix with applying left over gain steps that aren't aligned multiples of 4. */
|
|
for(;pos < BufferSize && pos < Counter;pos++)
|
|
{
|
|
OutBuffer[c][OutPos+pos] += data[pos]*gain;
|
|
gain *= step;
|
|
}
|
|
if(pos == Counter)
|
|
gain = Gains[c].Target;
|
|
Gains[c].Current = gain;
|
|
/* Mix until pos is aligned with 4 or the mix is done. */
|
|
for(;pos < BufferSize && (pos&3) != 0;pos++)
|
|
OutBuffer[c][OutPos+pos] += data[pos]*gain;
|
|
}
|
|
|
|
if(!(gain > GAIN_SILENCE_THRESHOLD))
|
|
continue;
|
|
gain4 = _mm_set1_ps(gain);
|
|
for(;BufferSize-pos > 3;pos += 4)
|
|
{
|
|
const __m128 val4 = _mm_load_ps(&data[pos]);
|
|
__m128 dry4 = _mm_load_ps(&OutBuffer[c][OutPos+pos]);
|
|
dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain4));
|
|
_mm_store_ps(&OutBuffer[c][OutPos+pos], dry4);
|
|
}
|
|
for(;pos < BufferSize;pos++)
|
|
OutBuffer[c][OutPos+pos] += data[pos]*gain;
|
|
}
|
|
}
|