zenilib  0.5.3.0
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Properties Friends Macros Groups Pages
mixer_sse.c
Go to the documentation of this file.
1 #include "config.h"
2 
3 #ifdef HAVE_XMMINTRIN_H
4 #include <xmmintrin.h>
5 #endif
6 
7 #include "AL/al.h"
8 #include "AL/alc.h"
9 #include "alMain.h"
10 #include "alu.h"
11 
12 #include "alSource.h"
13 #include "alAuxEffectSlot.h"
14 #include "mixer_defs.h"
15 
16 
17 static __inline void ApplyCoeffsStep(ALuint Offset, ALfloat (*RESTRICT Values)[2],
18  const ALuint IrSize,
19  ALfloat (*RESTRICT Coeffs)[2],
20  const ALfloat (*RESTRICT CoeffStep)[2],
22 {
23  const __m128 lrlr = { left, right, left, right };
24  __m128 coeffs, deltas, imp0, imp1;
25  __m128 vals = _mm_setzero_ps();
26  ALuint i;
27 
28  if((Offset&1))
29  {
30  const ALuint o0 = Offset&HRIR_MASK;
31  const ALuint o1 = (Offset+IrSize-1)&HRIR_MASK;
32 
33  coeffs = _mm_load_ps(&Coeffs[0][0]);
34  deltas = _mm_load_ps(&CoeffStep[0][0]);
35  vals = _mm_loadl_pi(vals, (__m64*)&Values[o0][0]);
36  imp0 = _mm_mul_ps(lrlr, coeffs);
37  coeffs = _mm_add_ps(coeffs, deltas);
38  vals = _mm_add_ps(imp0, vals);
39  _mm_store_ps(&Coeffs[0][0], coeffs);
40  _mm_storel_pi((__m64*)&Values[o0][0], vals);
41  for(i = 1;i < IrSize-1;i += 2)
42  {
43  const ALuint o2 = (Offset+i)&HRIR_MASK;
44 
45  coeffs = _mm_load_ps(&Coeffs[i+1][0]);
46  deltas = _mm_load_ps(&CoeffStep[i+1][0]);
47  vals = _mm_load_ps(&Values[o2][0]);
48  imp1 = _mm_mul_ps(lrlr, coeffs);
49  coeffs = _mm_add_ps(coeffs, deltas);
50  imp0 = _mm_shuffle_ps(imp0, imp1, _MM_SHUFFLE(1, 0, 3, 2));
51  vals = _mm_add_ps(imp0, vals);
52  _mm_store_ps(&Coeffs[i+1][0], coeffs);
53  _mm_store_ps(&Values[o2][0], vals);
54  imp0 = imp1;
55  }
56  vals = _mm_loadl_pi(vals, (__m64*)&Values[o1][0]);
57  imp0 = _mm_movehl_ps(imp0, imp0);
58  vals = _mm_add_ps(imp0, vals);
59  _mm_storel_pi((__m64*)&Values[o1][0], vals);
60  }
61  else
62  {
63  for(i = 0;i < IrSize;i += 2)
64  {
65  const ALuint o = (Offset + i)&HRIR_MASK;
66 
67  coeffs = _mm_load_ps(&Coeffs[i][0]);
68  deltas = _mm_load_ps(&CoeffStep[i][0]);
69  vals = _mm_load_ps(&Values[o][0]);
70  imp0 = _mm_mul_ps(lrlr, coeffs);
71  coeffs = _mm_add_ps(coeffs, deltas);
72  vals = _mm_add_ps(imp0, vals);
73  _mm_store_ps(&Coeffs[i][0], coeffs);
74  _mm_store_ps(&Values[o][0], vals);
75  }
76  }
77 }
78 
79 static __inline void ApplyCoeffs(ALuint Offset, ALfloat (*RESTRICT Values)[2],
80  const ALuint IrSize,
81  ALfloat (*RESTRICT Coeffs)[2],
83 {
84  const __m128 lrlr = { left, right, left, right };
85  __m128 vals = _mm_setzero_ps();
86  __m128 coeffs;
87  ALuint i;
88 
89  if((Offset&1))
90  {
91  const ALuint o0 = Offset&HRIR_MASK;
92  const ALuint o1 = (Offset+IrSize-1)&HRIR_MASK;
93  __m128 imp0, imp1;
94 
95  coeffs = _mm_load_ps(&Coeffs[0][0]);
96  vals = _mm_loadl_pi(vals, (__m64*)&Values[o0][0]);
97  imp0 = _mm_mul_ps(lrlr, coeffs);
98  vals = _mm_add_ps(imp0, vals);
99  _mm_storel_pi((__m64*)&Values[o0][0], vals);
100  for(i = 1;i < IrSize-1;i += 2)
101  {
102  const ALuint o2 = (Offset+i)&HRIR_MASK;
103 
104  coeffs = _mm_load_ps(&Coeffs[i+1][0]);
105  vals = _mm_load_ps(&Values[o2][0]);
106  imp1 = _mm_mul_ps(lrlr, coeffs);
107  imp0 = _mm_shuffle_ps(imp0, imp1, _MM_SHUFFLE(1, 0, 3, 2));
108  vals = _mm_add_ps(imp0, vals);
109  _mm_store_ps(&Values[o2][0], vals);
110  imp0 = imp1;
111  }
112  vals = _mm_loadl_pi(vals, (__m64*)&Values[o1][0]);
113  imp0 = _mm_movehl_ps(imp0, imp0);
114  vals = _mm_add_ps(imp0, vals);
115  _mm_storel_pi((__m64*)&Values[o1][0], vals);
116  }
117  else
118  {
119  for(i = 0;i < IrSize;i += 2)
120  {
121  const ALuint o = (Offset + i)&HRIR_MASK;
122 
123  coeffs = _mm_load_ps(&Coeffs[i][0]);
124  vals = _mm_load_ps(&Values[o][0]);
125  vals = _mm_add_ps(vals, _mm_mul_ps(lrlr, coeffs));
126  _mm_store_ps(&Values[o][0], vals);
127  }
128  }
129 }
130 
131 #define SUFFIX SSE
132 #include "mixer_inc.c"
133 #undef SUFFIX
134 
135 
136 void MixDirect_SSE(const DirectParams *params, const ALfloat *RESTRICT data, ALuint srcchan,
137  ALuint OutPos, ALuint SamplesToDo, ALuint BufferSize)
138 {
139  ALfloat (*RESTRICT DryBuffer)[BUFFERSIZE] = params->OutBuffer;
140  ALfloat *RESTRICT ClickRemoval = params->ClickRemoval;
141  ALfloat *RESTRICT PendingClicks = params->PendingClicks;
142  ALfloat DrySend;
143  ALuint pos;
144  ALuint c;
145 
146  for(c = 0;c < MaxChannels;c++)
147  {
148  __m128 gain;
149 
150  DrySend = params->Gains[srcchan][c];
151  if(DrySend < 0.00001f)
152  continue;
153 
154  if(OutPos == 0)
155  ClickRemoval[c] -= data[0]*DrySend;
156 
157  gain = _mm_set1_ps(DrySend);
158  for(pos = 0;pos < BufferSize-3;pos += 4)
159  {
160  const __m128 val4 = _mm_load_ps(&data[pos]);
161  __m128 dry4 = _mm_load_ps(&DryBuffer[c][OutPos+pos]);
162  dry4 = _mm_add_ps(dry4, _mm_mul_ps(val4, gain));
163  _mm_store_ps(&DryBuffer[c][OutPos+pos], dry4);
164  }
165  for(;pos < BufferSize;pos++)
166  DryBuffer[c][OutPos+pos] += data[pos]*DrySend;
167 
168  if(OutPos+pos == SamplesToDo)
169  PendingClicks[c] += data[pos]*DrySend;
170  }
171 }
172 
173 
174 void MixSend_SSE(const SendParams *params, const ALfloat *RESTRICT data,
175  ALuint OutPos, ALuint SamplesToDo, ALuint BufferSize)
176 {
177  ALeffectslot *Slot = params->Slot;
178  ALfloat (*RESTRICT WetBuffer)[BUFFERSIZE] = Slot->WetBuffer;
179  ALfloat *RESTRICT WetClickRemoval = Slot->ClickRemoval;
180  ALfloat *RESTRICT WetPendingClicks = Slot->PendingClicks;
181  const ALfloat WetGain = params->Gain;
182  __m128 gain;
183  ALuint pos;
184 
185  if(WetGain < 0.00001f)
186  return;
187 
188  if(OutPos == 0)
189  WetClickRemoval[0] -= data[0] * WetGain;
190 
191  gain = _mm_set1_ps(WetGain);
192  for(pos = 0;pos < BufferSize-3;pos+=4)
193  {
194  const __m128 val4 = _mm_load_ps(&data[pos]);
195  __m128 wet4 = _mm_load_ps(&WetBuffer[0][OutPos+pos]);
196  wet4 = _mm_add_ps(wet4, _mm_mul_ps(val4, gain));
197  _mm_store_ps(&WetBuffer[0][OutPos+pos], wet4);
198  }
199  for(;pos < BufferSize;pos++)
200  WetBuffer[0][OutPos+pos] += data[pos] * WetGain;
201 
202  if(OutPos+pos == SamplesToDo)
203  WetPendingClicks[0] += data[pos] * WetGain;
204 }
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)
Definition: mixer_sse.c:17
ALfloat(* OutBuffer)[BUFFERSIZE]
Definition: alSource.h:50
void MixDirect_SSE(const struct DirectParams *, const ALfloat *RESTRICT, ALuint, ALuint, ALuint, ALuint)
GLint left
Definition: glew.h:7291
GLclampf f
Definition: glew.h:3390
void MixSend_SSE(const struct SendParams *, const ALfloat *RESTRICT, ALuint, ALuint, ALuint)
static __inline void ApplyCoeffs(ALuint Offset, ALfloat(*RESTRICT Values)[2], const ALuint IrSize, ALfloat(*RESTRICT Coeffs)[2], ALfloat left, ALfloat right)
Definition: mixer_sse.c:79
float ALfloat
Definition: al.h:68
GLenum GLvoid ** params
Definition: gl2ext.h:806
ALfloat PendingClicks[1]
struct ALeffectslot * Slot
Definition: alSource.h:70
GLint GLenum GLsizei GLsizei GLsizei GLint GLsizei const GLvoid * data
Definition: gl2ext.h:848
const GLfloat * c
Definition: glew.h:14913
ALfloat Gain
Definition: alSource.h:74
ALfloat Gains[MaxChannels][MaxChannels]
Definition: alSource.h:62
unsigned int ALuint
Definition: al.h:59
ALfloat ClickRemoval[1]
ALfloat * PendingClicks
Definition: alSource.h:52
#define HRIR_MASK
Definition: alMain.h:777
int i
Definition: pngrutil.c:1377
GLfloat right
Definition: glew.h:13816
#define BUFFERSIZE
Definition: alMain.h:556
ALfloat * ClickRemoval
Definition: alSource.h:51
GLenum GLenum const GLfloat * coeffs
Definition: glew.h:12405