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
|
#include "config.h"
#include <arm_neon.h>
#include "AL/al.h"
#include "AL/alc.h"
#include "alMain.h"
#include "alu.h"
#include "hrtf.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)
{
ALuint c;
float32x4_t leftright4;
{
float32x2_t leftright2 = vdup_n_f32(0.0);
leftright2 = vset_lane_f32(left, leftright2, 0);
leftright2 = vset_lane_f32(right, leftright2, 1);
leftright4 = vcombine_f32(leftright2, leftright2);
}
for(c = 0;c < IrSize;c += 2)
{
const ALuint o0 = (Offset+c)&HRIR_MASK;
const ALuint o1 = (o0+1)&HRIR_MASK;
float32x4_t vals = vcombine_f32(vld1_f32((float32_t*)&Values[o0][0]),
vld1_f32((float32_t*)&Values[o1][0]));
float32x4_t coefs = vld1q_f32((float32_t*)&Coeffs[c][0]);
float32x4_t deltas = vld1q_f32(&CoeffStep[c][0]);
vals = vmlaq_f32(vals, coefs, leftright4);
coefs = vaddq_f32(coefs, deltas);
vst1_f32((float32_t*)&Values[o0][0], vget_low_f32(vals));
vst1_f32((float32_t*)&Values[o1][0], vget_high_f32(vals));
vst1q_f32(&Coeffs[c][0], coefs);
}
}
static inline void ApplyCoeffs(ALuint Offset, ALfloat (*restrict Values)[2],
const ALuint IrSize,
ALfloat (*restrict Coeffs)[2],
ALfloat left, ALfloat right)
{
ALuint c;
float32x4_t leftright4;
{
float32x2_t leftright2 = vdup_n_f32(0.0);
leftright2 = vset_lane_f32(left, leftright2, 0);
leftright2 = vset_lane_f32(right, leftright2, 1);
leftright4 = vcombine_f32(leftright2, leftright2);
}
for(c = 0;c < IrSize;c += 2)
{
const ALuint o0 = (Offset+c)&HRIR_MASK;
const ALuint o1 = (o0+1)&HRIR_MASK;
float32x4_t vals = vcombine_f32(vld1_f32((float32_t*)&Values[o0][0]),
vld1_f32((float32_t*)&Values[o1][0]));
float32x4_t coefs = vld1q_f32((float32_t*)&Coeffs[c][0]);
vals = vmlaq_f32(vals, coefs, leftright4);
vst1_f32((float32_t*)&Values[o0][0], vget_low_f32(vals));
vst1_f32((float32_t*)&Values[o1][0], vget_high_f32(vals));
}
}
#define MixHrtf MixHrtf_Neon
#define MixDirectHrtf MixDirectHrtf_Neon
#include "mixer_inc.c"
#undef MixHrtf
void Mix_Neon(const ALfloat *data, ALuint OutChans, ALfloat (*restrict OutBuffer)[BUFFERSIZE],
MixGains *Gains, ALuint Counter, ALuint OutPos, ALuint BufferSize)
{
ALfloat gain, step;
float32x4_t 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)
{
float32x4_t step4;
gain4 = vsetq_lane_f32(gain, gain4, 0);
gain4 = vsetq_lane_f32(gain + step, gain4, 1);
gain4 = vsetq_lane_f32(gain + step + step, gain4, 2);
gain4 = vsetq_lane_f32(gain + step + step + step, gain4, 3);
step4 = vdupq_n_f32(step + step + step + step);
do {
const float32x4_t val4 = vld1q_f32(&data[pos]);
float32x4_t dry4 = vld1q_f32(&OutBuffer[c][OutPos+pos]);
dry4 = vmlaq_f32(dry4, val4, gain4);
gain4 = vaddq_f32(gain4, step4);
vst1q_f32(&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 = vgetq_lane_f32(gain4, 0);
}
/* 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 = vdupq_n_f32(gain);
for(;BufferSize-pos > 3;pos += 4)
{
const float32x4_t val4 = vld1q_f32(&data[pos]);
float32x4_t dry4 = vld1q_f32(&OutBuffer[c][OutPos+pos]);
dry4 = vmlaq_f32(dry4, val4, gain4);
vst1q_f32(&OutBuffer[c][OutPos+pos], dry4);
}
for(;pos < BufferSize;pos++)
OutBuffer[c][OutPos+pos] += data[pos]*gain;
}
}
void MixRow_Neon(ALfloat *OutBuffer, const ALfloat *Gains, const ALfloat (*restrict data)[BUFFERSIZE], ALuint InChans, ALuint InPos, ALuint BufferSize)
{
float32x4_t gain4;
ALuint c;
for(c = 0;c < InChans;c++)
{
ALuint pos = 0;
ALfloat gain = Gains[c];
if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
continue;
gain4 = vdupq_n_f32(gain);
for(;BufferSize-pos > 3;pos += 4)
{
const float32x4_t val4 = vld1q_f32(&data[c][InPos+pos]);
float32x4_t dry4 = vld1q_f32(&OutBuffer[pos]);
dry4 = vmlaq_f32(dry4, val4, gain4);
vst1q_f32(&OutBuffer[pos], dry4);
}
for(;pos < BufferSize;pos++)
OutBuffer[pos] += data[c][InPos+pos]*gain;
}
}
|
