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
|
#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 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 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 != 0.0f && 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_add_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(!(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;
}
}
|
