#include "config.h"

#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"


// Obtain the next sample from the interpolator (SSE version).
static inline ALfloat bsinc32_sse(const BsincState *state, const ALfloat *in, const ALuint frac)
{
    const __m128 sf4 = _mm_set1_ps(state->sf);
    ALfloat pf, r;
    ALuint pi;

    // Calculate the phase index and factor.
#define FRAC_PHASE_BITDIFF (FRACTIONBITS-BSINC_PHASE_BITS)
    pi = frac >> FRAC_PHASE_BITDIFF;
    pf = (frac & ((1<<FRAC_PHASE_BITDIFF)-1)) * (1.0f/(1<<FRAC_PHASE_BITDIFF));
#undef FRAC_PHASE_BITDIFF

    {
        const ALuint m = state->m;
        const ALint l = state->l;
        const ALfloat *fil = state->coeffs[pi].filter;
        const ALfloat *scd = state->coeffs[pi].scDelta;
        const ALfloat *phd = state->coeffs[pi].phDelta;
        const ALfloat *spd = state->coeffs[pi].spDelta;
        const __m128 pf4 = _mm_set1_ps(pf);
        __m128 r4 = _mm_setzero_ps();
        ALuint j_f;
        ALint j_s;

        // Apply the scale and phase interpolated filter.
        for(j_f = 0,j_s = l;j_f < m;j_f+=4,j_s+=4)
        {
            const __m128 f4 = _mm_add_ps(
                _mm_add_ps(
                    _mm_load_ps(&fil[j_f]),
                    _mm_mul_ps(sf4, _mm_load_ps(&scd[j_f]))
                ),
                _mm_mul_ps(
                    pf4,
                    _mm_add_ps(
                        _mm_load_ps(&phd[j_f]),
                        _mm_mul_ps(sf4, _mm_load_ps(&spd[j_f]))
                    )
                )
            );
            r4 = _mm_add_ps(r4, _mm_mul_ps(f4, _mm_loadu_ps(&in[j_s])));
        }
        r4 = _mm_add_ps(r4, _mm_shuffle_ps(r4, r4, _MM_SHUFFLE(0, 1, 2, 3)));
        r4 = _mm_add_ps(r4, _mm_movehl_ps(r4, r4));
        r = _mm_cvtss_f32(r4);
    }
    return r;
}

const ALfloat *Resample_bsinc32_SSE(const BsincState *state, const ALfloat *src, ALuint frac,
                                    ALuint increment, ALfloat *restrict dst, ALuint dstlen)
{
    ALuint i;
    for(i = 0;i < dstlen;i++)
    {
        dst[i] = bsinc32_sse(state, src, frac);

        frac += increment;
        src  += frac>>FRACTIONBITS;
        frac &= FRACTIONMASK;
    }
    return dst;
}


static inline void SetupCoeffs(ALfloat (*restrict OutCoeffs)[2],
                               const HrtfParams *hrtfparams,
                               ALuint IrSize, ALuint Counter)
{
    const __m128 counter4 = _mm_set1_ps((float)Counter);
    __m128 coeffs, step4;
    ALuint i;

    for(i = 0;i < IrSize;i += 2)
    {
        step4  = _mm_load_ps(&hrtfparams->CoeffStep[i][0]);
        coeffs = _mm_load_ps(&hrtfparams->Coeffs[i][0]);
        coeffs = _mm_sub_ps(coeffs, _mm_mul_ps(step4, counter4));
        _mm_store_ps(&OutCoeffs[i][0], coeffs);
    }
}

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 MixHrtf MixHrtf_SSE
#include "mixer_inc.c"
#undef MixHrtf


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;
    ALuint c;

    for(c = 0;c < OutChans;c++)
    {
        ALuint pos = 0;
        gain = Gains[c].Current;
        step = Gains[c].Step;
        if(step != 0.0f && Counter > 0)
        {
            ALuint minsize = minu(BufferSize, Counter);
            /* Mix with applying gain steps in aligned multiples of 4. */
            if(minsize-pos > 3)
            {
                __m128 step4;
                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_add_ps(gain4, step4);
                    _mm_store_ps(&OutBuffer[c][OutPos+pos], dry4);
                    pos += 4;
                } while(minsize-pos > 3);
                /* NOTE: gain4 now represents the next four gains after the
                 * last four mixed samples, so the lowest element represents
                 * the next gain to apply.
                 */
                gain = _mm_cvtss_f32(gain4);
            }
            /* Mix with applying left over gain steps that aren't aligned multiples of 4. */
            for(;pos < minsize;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. */
            minsize = minu(BufferSize, (pos+3)&~3);
            for(;pos < minsize;pos++)
                OutBuffer[c][OutPos+pos] += data[pos]*gain;
        }

        if(!(fabsf(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;
    }
}