aboutsummaryrefslogtreecommitdiffstats
path: root/alc/mixer/mixer_c.cpp
blob: 8aa32e6bb51a1228d658425945688d92cf9dd4e6 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
#include "config.h"

#include <cassert>

#include <limits>

#include "alcmain.h"
#include "alu.h"

#include "defs.h"
#include "hrtfbase.h"


namespace {

inline float do_point(const InterpState&, const float *RESTRICT vals, const ALuint)
{ return vals[0]; }
inline float do_lerp(const InterpState&, const float *RESTRICT vals, const ALuint frac)
{ return lerp(vals[0], vals[1], static_cast<float>(frac)*(1.0f/FRACTIONONE)); }
inline float do_cubic(const InterpState&, const float *RESTRICT vals, const ALuint frac)
{ return cubic(vals[0], vals[1], vals[2], vals[3], static_cast<float>(frac)*(1.0f/FRACTIONONE)); }
inline float do_bsinc(const InterpState &istate, const float *RESTRICT vals, const ALuint frac)
{
    const size_t m{istate.bsinc.m};

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

    const float *fil{istate.bsinc.filter + m*pi*4};
    const float *phd{fil + m};
    const float *scd{phd + m};
    const float *spd{scd + m};

    // Apply the scale and phase interpolated filter.
    float r{0.0f};
    for(size_t j_f{0};j_f < m;j_f++)
        r += (fil[j_f] + istate.bsinc.sf*scd[j_f] + pf*(phd[j_f] + istate.bsinc.sf*spd[j_f])) * vals[j_f];
    return r;
}
inline float do_fastbsinc(const InterpState &istate, const float *RESTRICT vals, const ALuint frac)
{
    const size_t m{istate.bsinc.m};

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

    const float *fil{istate.bsinc.filter + m*pi*4};
    const float *phd{fil + m};

    // Apply the phase interpolated filter.
    float r{0.0f};
    for(size_t j_f{0};j_f < m;j_f++)
        r += (fil[j_f] + pf*phd[j_f]) * vals[j_f];
    return r;
}

using SamplerT = float(const InterpState&, const float*RESTRICT, const ALuint);
template<SamplerT &Sampler>
const float *DoResample(const InterpState *state, const float *RESTRICT src, ALuint frac,
    ALuint increment, const al::span<float> dst)
{
    const InterpState istate{*state};
    auto proc_sample = [&src,&frac,istate,increment]() -> ALfloat
    {
        const float ret{Sampler(istate, src, frac)};

        frac += increment;
        src  += frac>>FRACTIONBITS;
        frac &= FRACTIONMASK;

        return ret;
    };
    std::generate(dst.begin(), dst.end(), proc_sample);

    return dst.begin();
}

} // namespace

template<>
const ALfloat *Resample_<CopyTag,CTag>(const InterpState*, const ALfloat *RESTRICT src, ALuint,
    ALuint, const al::span<float> dst)
{
#if defined(HAVE_SSE) || defined(HAVE_NEON)
    /* Avoid copying the source data if it's aligned like the destination. */
    if((reinterpret_cast<intptr_t>(src)&15) == (reinterpret_cast<intptr_t>(dst.data())&15))
        return src;
#endif
    std::copy_n(src, dst.size(), dst.begin());
    return dst.begin();
}

template<>
const ALfloat *Resample_<PointTag,CTag>(const InterpState *state, const ALfloat *RESTRICT src,
    ALuint frac, ALuint increment, const al::span<float> dst)
{ return DoResample<do_point>(state, src, frac, increment, dst); }

template<>
const ALfloat *Resample_<LerpTag,CTag>(const InterpState *state, const ALfloat *RESTRICT src,
    ALuint frac, ALuint increment, const al::span<float> dst)
{ return DoResample<do_lerp>(state, src, frac, increment, dst); }

template<>
const ALfloat *Resample_<CubicTag,CTag>(const InterpState *state, const ALfloat *RESTRICT src,
    ALuint frac, ALuint increment, const al::span<float> dst)
{ return DoResample<do_cubic>(state, src-1, frac, increment, dst); }

template<>
const ALfloat *Resample_<BSincTag,CTag>(const InterpState *state, const ALfloat *RESTRICT src,
    ALuint frac, ALuint increment, const al::span<float> dst)
{ return DoResample<do_bsinc>(state, src-state->bsinc.l, frac, increment, dst); }

template<>
const ALfloat *Resample_<FastBSincTag,CTag>(const InterpState *state, const ALfloat *RESTRICT src,
    ALuint frac, ALuint increment, const al::span<float> dst)
{ return DoResample<do_fastbsinc>(state, src-state->bsinc.l, frac, increment, dst); }


static inline void ApplyCoeffs(size_t /*Offset*/, float2 *RESTRICT Values, const ALuint IrSize,
    const HrirArray &Coeffs, const float left, const float right)
{
    ASSUME(IrSize >= 4);
    for(ALuint c{0};c < IrSize;++c)
    {
        Values[c][0] += Coeffs[c][0] * left;
        Values[c][1] += Coeffs[c][1] * right;
    }
}

template<>
void MixHrtf_<CTag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
    const float *InSamples, float2 *AccumSamples, const size_t OutPos, const ALuint IrSize,
    MixHrtfFilter *hrtfparams, const size_t BufferSize)
{
    MixHrtfBase<ApplyCoeffs>(LeftOut, RightOut, InSamples, AccumSamples, OutPos, IrSize,
        hrtfparams, BufferSize);
}

template<>
void MixHrtfBlend_<CTag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
    const float *InSamples, float2 *AccumSamples, const size_t OutPos, const ALuint IrSize,
    const HrtfFilter *oldparams, MixHrtfFilter *newparams, const size_t BufferSize)
{
    MixHrtfBlendBase<ApplyCoeffs>(LeftOut, RightOut, InSamples, AccumSamples, OutPos, IrSize,
        oldparams, newparams, BufferSize);
}

template<>
void MixDirectHrtf_<CTag>(FloatBufferLine &LeftOut, FloatBufferLine &RightOut,
    const al::span<const FloatBufferLine> InSamples, float2 *AccumSamples, DirectHrtfState *State,
    const size_t BufferSize)
{ MixDirectHrtfBase<ApplyCoeffs>(LeftOut, RightOut, InSamples, AccumSamples, State, BufferSize); }


template<>
void Mix_<CTag>(const al::span<const float> InSamples, const al::span<FloatBufferLine> OutBuffer,
    float *CurrentGains, const float *TargetGains, const size_t Counter, const size_t OutPos)
{
    const ALfloat delta{(Counter > 0) ? 1.0f / static_cast<ALfloat>(Counter) : 0.0f};
    const bool reached_target{InSamples.size() >= Counter};
    const auto min_end = reached_target ? InSamples.begin() + Counter : InSamples.end();
    for(FloatBufferLine &output : OutBuffer)
    {
        ALfloat *RESTRICT dst{al::assume_aligned<16>(output.data()+OutPos)};
        ALfloat gain{*CurrentGains};
        const ALfloat diff{*TargetGains - gain};

        auto in_iter = InSamples.begin();
        if(std::fabs(diff) > std::numeric_limits<float>::epsilon())
        {
            const ALfloat step{diff * delta};
            ALfloat step_count{0.0f};
            while(in_iter != min_end)
            {
                *(dst++) += *(in_iter++) * (gain + step*step_count);
                step_count += 1.0f;
            }
            if(reached_target)
                gain = *TargetGains;
            else
                gain += step*step_count;
            *CurrentGains = gain;
        }
        ++CurrentGains;
        ++TargetGains;

        if(!(std::fabs(gain) > GAIN_SILENCE_THRESHOLD))
            continue;
        while(in_iter != InSamples.end())
            *(dst++) += *(in_iter++) * gain;
    }
}

/* Basically the inverse of the above. Rather than one input going to multiple
 * outputs (each with its own gain), it's multiple inputs (each with its own
 * gain) going to one output. This applies one row (vs one column) of a matrix
 * transform. And as the matrices are more or less static once set up, no
 * stepping is necessary.
 */
template<>
void MixRow_<CTag>(const al::span<float> OutBuffer, const al::span<const float> Gains,
    const float *InSamples, const size_t InStride)
{
    for(const float gain : Gains)
    {
        const float *RESTRICT input{InSamples};
        InSamples += InStride;

        if(!(std::fabs(gain) > GAIN_SILENCE_THRESHOLD))
            continue;

        auto do_mix = [gain](const float cur, const float src) noexcept -> float
        { return cur + src*gain; };
        std::transform(OutBuffer.begin(), OutBuffer.end(), input, OutBuffer.begin(), do_mix);
    }
}