1 /* Copyright (C) 2002 Jean-Marc Valin */
4 @brief Various analysis/synthesis filters (SSE version)
7 Redistribution and use in source and binary forms, with or without
8 modification, are permitted provided that the following conditions
11 - Redistributions of source code must retain the above copyright
12 notice, this list of conditions and the following disclaimer.
14 - Redistributions in binary form must reproduce the above copyright
15 notice, this list of conditions and the following disclaimer in the
16 documentation and/or other materials provided with the distribution.
18 - Neither the name of the Xiph.org Foundation nor the names of its
19 contributors may be used to endorse or promote products derived from
20 this software without specific prior written permission.
22 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
26 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
27 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
29 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
30 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
31 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
32 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <xmmintrin.h>
37 void filter_mem16_10(const float *x
, const float *_num
, const float *_den
, float *y
, int N
, int ord
, float *_mem
)
39 __m128 num
[3], den
[3], mem
[3];
43 /* Copy numerator, denominator and memory to aligned xmm */
46 mem
[i
] = _mm_loadu_ps(_mem
+4*i
);
47 num
[i
] = _mm_loadu_ps(_num
+4*i
);
48 den
[i
] = _mm_loadu_ps(_den
+4*i
);
50 mem
[2] = _mm_setr_ps(_mem
[8], _mem
[9], 0, 0);
51 num
[2] = _mm_setr_ps(_num
[8], _num
[9], 0, 0);
52 den
[2] = _mm_setr_ps(_den
[8], _den
[9], 0, 0);
58 /* Compute next filter result */
59 xx
= _mm_load_ps1(x
+i
);
60 yy
= _mm_add_ss(xx
, mem
[0]);
61 _mm_store_ss(y
+i
, yy
);
62 yy
= _mm_shuffle_ps(yy
, yy
, 0);
65 mem
[0] = _mm_move_ss(mem
[0], mem
[1]);
66 mem
[0] = _mm_shuffle_ps(mem
[0], mem
[0], 0x39);
68 mem
[0] = _mm_add_ps(mem
[0], _mm_mul_ps(xx
, num
[0]));
69 mem
[0] = _mm_sub_ps(mem
[0], _mm_mul_ps(yy
, den
[0]));
71 mem
[1] = _mm_move_ss(mem
[1], mem
[2]);
72 mem
[1] = _mm_shuffle_ps(mem
[1], mem
[1], 0x39);
74 mem
[1] = _mm_add_ps(mem
[1], _mm_mul_ps(xx
, num
[1]));
75 mem
[1] = _mm_sub_ps(mem
[1], _mm_mul_ps(yy
, den
[1]));
77 mem
[2] = _mm_shuffle_ps(mem
[2], mem
[2], 0xfd);
79 mem
[2] = _mm_add_ps(mem
[2], _mm_mul_ps(xx
, num
[2]));
80 mem
[2] = _mm_sub_ps(mem
[2], _mm_mul_ps(yy
, den
[2]));
82 /* Put memory back in its place */
83 _mm_storeu_ps(_mem
, mem
[0]);
84 _mm_storeu_ps(_mem
+4, mem
[1]);
85 _mm_store_ss(_mem
+8, mem
[2]);
86 mem
[2] = _mm_shuffle_ps(mem
[2], mem
[2], 0x55);
87 _mm_store_ss(_mem
+9, mem
[2]);
90 void filter_mem16_8(const float *x
, const float *_num
, const float *_den
, float *y
, int N
, int ord
, float *_mem
)
92 __m128 num
[2], den
[2], mem
[2];
96 /* Copy numerator, denominator and memory to aligned xmm */
99 mem
[i
] = _mm_loadu_ps(_mem
+4*i
);
100 num
[i
] = _mm_loadu_ps(_num
+4*i
);
101 den
[i
] = _mm_loadu_ps(_den
+4*i
);
108 /* Compute next filter result */
109 xx
= _mm_load_ps1(x
+i
);
110 yy
= _mm_add_ss(xx
, mem
[0]);
111 _mm_store_ss(y
+i
, yy
);
112 yy
= _mm_shuffle_ps(yy
, yy
, 0);
115 mem
[0] = _mm_move_ss(mem
[0], mem
[1]);
116 mem
[0] = _mm_shuffle_ps(mem
[0], mem
[0], 0x39);
118 mem
[0] = _mm_add_ps(mem
[0], _mm_mul_ps(xx
, num
[0]));
119 mem
[0] = _mm_sub_ps(mem
[0], _mm_mul_ps(yy
, den
[0]));
121 mem
[1] = _mm_sub_ss(mem
[1], mem
[1]);
122 mem
[1] = _mm_shuffle_ps(mem
[1], mem
[1], 0x39);
124 mem
[1] = _mm_add_ps(mem
[1], _mm_mul_ps(xx
, num
[1]));
125 mem
[1] = _mm_sub_ps(mem
[1], _mm_mul_ps(yy
, den
[1]));
127 /* Put memory back in its place */
128 _mm_storeu_ps(_mem
, mem
[0]);
129 _mm_storeu_ps(_mem
+4, mem
[1]);
133 #define OVERRIDE_FILTER_MEM16
134 void filter_mem16(const float *x
, const float *_num
, const float *_den
, float *y
, int N
, int ord
, float *_mem
, char *stack
)
137 filter_mem16_10(x
, _num
, _den
, y
, N
, ord
, _mem
);
139 filter_mem16_8(x
, _num
, _den
, y
, N
, ord
, _mem
);
144 void iir_mem16_10(const float *x
, const float *_den
, float *y
, int N
, int ord
, float *_mem
)
146 __m128 den
[3], mem
[3];
150 /* Copy numerator, denominator and memory to aligned xmm */
153 mem
[i
] = _mm_loadu_ps(_mem
+4*i
);
154 den
[i
] = _mm_loadu_ps(_den
+4*i
);
156 mem
[2] = _mm_setr_ps(_mem
[8], _mem
[9], 0, 0);
157 den
[2] = _mm_setr_ps(_den
[8], _den
[9], 0, 0);
163 /* Compute next filter result */
164 xx
= _mm_load_ps1(x
+i
);
165 yy
= _mm_add_ss(xx
, mem
[0]);
166 _mm_store_ss(y
+i
, yy
);
167 yy
= _mm_shuffle_ps(yy
, yy
, 0);
170 mem
[0] = _mm_move_ss(mem
[0], mem
[1]);
171 mem
[0] = _mm_shuffle_ps(mem
[0], mem
[0], 0x39);
173 mem
[0] = _mm_sub_ps(mem
[0], _mm_mul_ps(yy
, den
[0]));
175 mem
[1] = _mm_move_ss(mem
[1], mem
[2]);
176 mem
[1] = _mm_shuffle_ps(mem
[1], mem
[1], 0x39);
178 mem
[1] = _mm_sub_ps(mem
[1], _mm_mul_ps(yy
, den
[1]));
180 mem
[2] = _mm_shuffle_ps(mem
[2], mem
[2], 0xfd);
182 mem
[2] = _mm_sub_ps(mem
[2], _mm_mul_ps(yy
, den
[2]));
184 /* Put memory back in its place */
185 _mm_storeu_ps(_mem
, mem
[0]);
186 _mm_storeu_ps(_mem
+4, mem
[1]);
187 _mm_store_ss(_mem
+8, mem
[2]);
188 mem
[2] = _mm_shuffle_ps(mem
[2], mem
[2], 0x55);
189 _mm_store_ss(_mem
+9, mem
[2]);
193 void iir_mem16_8(const float *x
, const float *_den
, float *y
, int N
, int ord
, float *_mem
)
195 __m128 den
[2], mem
[2];
199 /* Copy numerator, denominator and memory to aligned xmm */
202 mem
[i
] = _mm_loadu_ps(_mem
+4*i
);
203 den
[i
] = _mm_loadu_ps(_den
+4*i
);
210 /* Compute next filter result */
211 xx
= _mm_load_ps1(x
+i
);
212 yy
= _mm_add_ss(xx
, mem
[0]);
213 _mm_store_ss(y
+i
, yy
);
214 yy
= _mm_shuffle_ps(yy
, yy
, 0);
217 mem
[0] = _mm_move_ss(mem
[0], mem
[1]);
218 mem
[0] = _mm_shuffle_ps(mem
[0], mem
[0], 0x39);
220 mem
[0] = _mm_sub_ps(mem
[0], _mm_mul_ps(yy
, den
[0]));
222 mem
[1] = _mm_sub_ss(mem
[1], mem
[1]);
223 mem
[1] = _mm_shuffle_ps(mem
[1], mem
[1], 0x39);
225 mem
[1] = _mm_sub_ps(mem
[1], _mm_mul_ps(yy
, den
[1]));
227 /* Put memory back in its place */
228 _mm_storeu_ps(_mem
, mem
[0]);
229 _mm_storeu_ps(_mem
+4, mem
[1]);
232 #define OVERRIDE_IIR_MEM16
233 void iir_mem16(const float *x
, const float *_den
, float *y
, int N
, int ord
, float *_mem
, char *stack
)
236 iir_mem16_10(x
, _den
, y
, N
, ord
, _mem
);
238 iir_mem16_8(x
, _den
, y
, N
, ord
, _mem
);
242 void fir_mem16_10(const float *x
, const float *_num
, float *y
, int N
, int ord
, float *_mem
)
244 __m128 num
[3], mem
[3];
248 /* Copy numerator, denominator and memory to aligned xmm */
251 mem
[i
] = _mm_loadu_ps(_mem
+4*i
);
252 num
[i
] = _mm_loadu_ps(_num
+4*i
);
254 mem
[2] = _mm_setr_ps(_mem
[8], _mem
[9], 0, 0);
255 num
[2] = _mm_setr_ps(_num
[8], _num
[9], 0, 0);
261 /* Compute next filter result */
262 xx
= _mm_load_ps1(x
+i
);
263 yy
= _mm_add_ss(xx
, mem
[0]);
264 _mm_store_ss(y
+i
, yy
);
265 yy
= _mm_shuffle_ps(yy
, yy
, 0);
268 mem
[0] = _mm_move_ss(mem
[0], mem
[1]);
269 mem
[0] = _mm_shuffle_ps(mem
[0], mem
[0], 0x39);
271 mem
[0] = _mm_add_ps(mem
[0], _mm_mul_ps(xx
, num
[0]));
273 mem
[1] = _mm_move_ss(mem
[1], mem
[2]);
274 mem
[1] = _mm_shuffle_ps(mem
[1], mem
[1], 0x39);
276 mem
[1] = _mm_add_ps(mem
[1], _mm_mul_ps(xx
, num
[1]));
278 mem
[2] = _mm_shuffle_ps(mem
[2], mem
[2], 0xfd);
280 mem
[2] = _mm_add_ps(mem
[2], _mm_mul_ps(xx
, num
[2]));
282 /* Put memory back in its place */
283 _mm_storeu_ps(_mem
, mem
[0]);
284 _mm_storeu_ps(_mem
+4, mem
[1]);
285 _mm_store_ss(_mem
+8, mem
[2]);
286 mem
[2] = _mm_shuffle_ps(mem
[2], mem
[2], 0x55);
287 _mm_store_ss(_mem
+9, mem
[2]);
290 void fir_mem16_8(const float *x
, const float *_num
, float *y
, int N
, int ord
, float *_mem
)
292 __m128 num
[2], mem
[2];
296 /* Copy numerator, denominator and memory to aligned xmm */
299 mem
[i
] = _mm_loadu_ps(_mem
+4*i
);
300 num
[i
] = _mm_loadu_ps(_num
+4*i
);
307 /* Compute next filter result */
308 xx
= _mm_load_ps1(x
+i
);
309 yy
= _mm_add_ss(xx
, mem
[0]);
310 _mm_store_ss(y
+i
, yy
);
311 yy
= _mm_shuffle_ps(yy
, yy
, 0);
314 mem
[0] = _mm_move_ss(mem
[0], mem
[1]);
315 mem
[0] = _mm_shuffle_ps(mem
[0], mem
[0], 0x39);
317 mem
[0] = _mm_add_ps(mem
[0], _mm_mul_ps(xx
, num
[0]));
319 mem
[1] = _mm_sub_ss(mem
[1], mem
[1]);
320 mem
[1] = _mm_shuffle_ps(mem
[1], mem
[1], 0x39);
322 mem
[1] = _mm_add_ps(mem
[1], _mm_mul_ps(xx
, num
[1]));
324 /* Put memory back in its place */
325 _mm_storeu_ps(_mem
, mem
[0]);
326 _mm_storeu_ps(_mem
+4, mem
[1]);
329 #define OVERRIDE_FIR_MEM16
330 void fir_mem16(const float *x
, const float *_num
, float *y
, int N
, int ord
, float *_mem
, char *stack
)
333 fir_mem16_10(x
, _num
, y
, N
, ord
, _mem
);
335 fir_mem16_8(x
, _num
, y
, N
, ord
, _mem
);