1 /* Copyright (c) 2013 The Chromium OS Authors. All rights reserved.
2  * Use of this source code is governed by a BSD-style license that can be
3  * found in the LICENSE file.
4  */
5 
6 #include <stdlib.h>
7 #include "eq2.h"
8 
9 struct eq2 {
10 	int n[2];
11 	struct biquad biquad[MAX_BIQUADS_PER_EQ2][2];
12 };
13 
eq2_new()14 struct eq2 *eq2_new()
15 {
16 	struct eq2 *eq2 = (struct eq2 *)calloc(1, sizeof(*eq2));
17 	int i, j;
18 
19 	/* Initialize all biquads to identity filter, so if two channels have
20 	 * different numbers of biquads, it still works. */
21 	for (i = 0; i < MAX_BIQUADS_PER_EQ2; i++)
22 		for (j = 0; j < 2; j++)
23 			biquad_set(&eq2->biquad[i][j], BQ_NONE, 0, 0, 0);
24 
25 	return eq2;
26 }
27 
eq2_free(struct eq2 * eq2)28 void eq2_free(struct eq2 *eq2)
29 {
30 	free(eq2);
31 }
32 
eq2_append_biquad(struct eq2 * eq2,int channel,enum biquad_type type,float freq,float Q,float gain)33 int eq2_append_biquad(struct eq2 *eq2, int channel,
34 		      enum biquad_type type, float freq, float Q, float gain)
35 {
36 	if (eq2->n[channel] >= MAX_BIQUADS_PER_EQ2)
37 		return -1;
38 	biquad_set(&eq2->biquad[eq2->n[channel]++][channel], type, freq, Q,
39 		   gain);
40 	return 0;
41 }
42 
eq2_append_biquad_direct(struct eq2 * eq2,int channel,const struct biquad * biquad)43 int eq2_append_biquad_direct(struct eq2 *eq2, int channel,
44 			     const struct biquad *biquad)
45 {
46 	if (eq2->n[channel] >= MAX_BIQUADS_PER_EQ2)
47 		return -1;
48 	eq2->biquad[eq2->n[channel]++][channel] = *biquad;
49 	return 0;
50 }
51 
eq2_process_one(struct biquad (* bq)[2],float * data0,float * data1,int count)52 static inline void eq2_process_one(struct biquad (*bq)[2],
53 				   float *data0, float *data1, int count)
54 {
55 	struct biquad *qL = &bq[0][0];
56 	struct biquad *qR = &bq[0][1];
57 
58 	float x1L = qL->x1;
59 	float x2L = qL->x2;
60 	float y1L = qL->y1;
61 	float y2L = qL->y2;
62 	float b0L = qL->b0;
63 	float b1L = qL->b1;
64 	float b2L = qL->b2;
65 	float a1L = qL->a1;
66 	float a2L = qL->a2;
67 
68 	float x1R = qR->x1;
69 	float x2R = qR->x2;
70 	float y1R = qR->y1;
71 	float y2R = qR->y2;
72 	float b0R = qR->b0;
73 	float b1R = qR->b1;
74 	float b2R = qR->b2;
75 	float a1R = qR->a1;
76 	float a2R = qR->a2;
77 
78 	int j;
79 	for (j = 0; j < count; j++) {
80 		float xL = data0[j];
81 		float xR = data1[j];
82 
83 		float yL = b0L*xL
84 			+ b1L*x1L + b2L*x2L
85 			- a1L*y1L - a2L*y2L;
86 		x2L = x1L;
87 		x1L = xL;
88 		y2L = y1L;
89 		y1L = yL;
90 
91 		float yR = b0R*xR
92 			+ b1R*x1R + b2R*x2R
93 			- a1R*y1R - a2R*y2R;
94 		x2R = x1R;
95 		x1R = xR;
96 		y2R = y1R;
97 		y1R = yR;
98 
99 		data0[j] = yL;
100 		data1[j] = yR;
101 	}
102 
103 	qL->x1 = x1L;
104 	qL->x2 = x2L;
105 	qL->y1 = y1L;
106 	qL->y2 = y2L;
107 	qR->x1 = x1R;
108 	qR->x2 = x2R;
109 	qR->y1 = y1R;
110 	qR->y2 = y2R;
111 }
112 
113 #ifdef __ARM_NEON__
114 #include <arm_neon.h>
eq2_process_two_neon(struct biquad (* bq)[2],float * data0,float * data1,int count)115 static inline void eq2_process_two_neon(struct biquad (*bq)[2],
116 					float *data0, float *data1, int count)
117 {
118 	struct biquad *qL = &bq[0][0];
119 	struct biquad *rL = &bq[1][0];
120 	struct biquad *qR = &bq[0][1];
121 	struct biquad *rR = &bq[1][1];
122 
123 	float32x2_t x1 = {qL->x1, qR->x1};
124 	float32x2_t x2 = {qL->x2, qR->x2};
125 	float32x2_t y1 = {qL->y1, qR->y1};
126 	float32x2_t y2 = {qL->y2, qR->y2};
127 	float32x2_t qb0 = {qL->b0, qR->b0};
128 	float32x2_t qb1 = {qL->b1, qR->b1};
129 	float32x2_t qb2 = {qL->b2, qR->b2};
130 	float32x2_t qa1 = {qL->a1, qR->a1};
131 	float32x2_t qa2 = {qL->a2, qR->a2};
132 
133 	float32x2_t z1 = {rL->y1, rR->y1};
134 	float32x2_t z2 = {rL->y2, rR->y2};
135 	float32x2_t rb0 = {rL->b0, rR->b0};
136 	float32x2_t rb1 = {rL->b1, rR->b1};
137 	float32x2_t rb2 = {rL->b2, rR->b2};
138 	float32x2_t ra1 = {rL->a1, rR->a1};
139 	float32x2_t ra2 = {rL->a2, rR->a2};
140 
141 	__asm__ __volatile__(
142 		/* d0 = x, d1 = y, d2 = z */
143 		"1:                                     \n"
144 		"vmul.f32 d1, %P[qb1], %P[x1]           \n"
145 		"vld1.32 d0[0], [%[data0]]              \n"
146 		"vld1.32 d0[1], [%[data1]]              \n"
147 		"subs %[count], #1                      \n"
148 		"vmul.f32 d2, %P[rb1], %P[y1]           \n"
149 		"vmla.f32 d1, %P[qb0], d0               \n"
150 		"vmla.f32 d1, %P[qb2], %P[x2]           \n"
151 		"vmov.f32 %P[x2], %P[x1]                \n"
152 		"vmov.f32 %P[x1], d0                    \n"
153 		"vmls.f32 d1, %P[qa1], %P[y1]           \n"
154 		"vmls.f32 d1, %P[qa2], %P[y2]           \n"
155 		"vmla.f32 d2, %P[rb0], d1               \n"
156 		"vmla.f32 d2, %P[rb2], %P[y2]           \n"
157 		"vmov.f32 %P[y2], %P[y1]                \n"
158 		"vmov.f32 %P[y1], d1                    \n"
159 		"vmls.f32 d2, %P[ra1], %P[z1]           \n"
160 		"vmls.f32 d2, %P[ra2], %P[z2]           \n"
161 		"vmov.f32 %P[z2], %P[z1]                \n"
162 		"vmov.f32 %P[z1], d2                    \n"
163 		"vst1.f32 d2[0], [%[data0]]!            \n"
164 		"vst1.f32 d2[1], [%[data1]]!            \n"
165 		"bne 1b                                 \n"
166 		: /* output */
167 		  [data0]"+r"(data0),
168 		  [data1]"+r"(data1),
169 		  [count]"+r"(count),
170 		  [x1]"+w"(x1),
171 		  [x2]"+w"(x2),
172 		  [y1]"+w"(y1),
173 		  [y2]"+w"(y2),
174 		  [z1]"+w"(z1),
175 		  [z2]"+w"(z2)
176 		: /* input */
177 		  [qb0]"w"(qb0),
178 		  [qb1]"w"(qb1),
179 		  [qb2]"w"(qb2),
180 		  [qa1]"w"(qa1),
181 		  [qa2]"w"(qa2),
182 		  [rb0]"w"(rb0),
183 		  [rb1]"w"(rb1),
184 		  [rb2]"w"(rb2),
185 		  [ra1]"w"(ra1),
186 		  [ra2]"w"(ra2)
187 		: /* clobber */
188 		  "d0", "d1", "d2", "memory", "cc"
189 		);
190 
191 	qL->x1 = x1[0];
192 	qL->x2 = x2[0];
193 	qL->y1 = y1[0];
194 	qL->y2 = y2[0];
195 	rL->y1 = z1[0];
196 	rL->y2 = z2[0];
197 	qR->x1 = x1[1];
198 	qR->x2 = x2[1];
199 	qR->y1 = y1[1];
200 	qR->y2 = y2[1];
201 	rR->y1 = z1[1];
202 	rR->y2 = z2[1];
203 }
204 #endif
205 
206 #if defined(__SSE3__) && defined(__x86_64__)
207 #include <emmintrin.h>
eq2_process_two_sse3(struct biquad (* bq)[2],float * data0,float * data1,int count)208 static inline void eq2_process_two_sse3(struct biquad (*bq)[2],
209 					float *data0, float *data1, int count)
210 {
211 	struct biquad *qL = &bq[0][0];
212 	struct biquad *rL = &bq[1][0];
213 	struct biquad *qR = &bq[0][1];
214 	struct biquad *rR = &bq[1][1];
215 
216 	__m128 x1 = {qL->x1, qR->x1};
217 	__m128 x2 = {qL->x2, qR->x2};
218 	__m128 y1 = {qL->y1, qR->y1};
219 	__m128 y2 = {qL->y2, qR->y2};
220 	__m128 qb0 = {qL->b0, qR->b0};
221 	__m128 qb1 = {qL->b1, qR->b1};
222 	__m128 qb2 = {qL->b2, qR->b2};
223 	__m128 qa1 = {qL->a1, qR->a1};
224 	__m128 qa2 = {qL->a2, qR->a2};
225 
226 	__m128 z1 = {rL->y1, rR->y1};
227 	__m128 z2 = {rL->y2, rR->y2};
228 	__m128 rb0 = {rL->b0, rR->b0};
229 	__m128 rb1 = {rL->b1, rR->b1};
230 	__m128 rb2 = {rL->b2, rR->b2};
231 	__m128 ra1 = {rL->a1, rR->a1};
232 	__m128 ra2 = {rL->a2, rR->a2};
233 
234 	__asm__ __volatile__(
235 		"1:                                     \n"
236 		"movss (%[data0]), %%xmm2               \n"
237 		"movss (%[data1]), %%xmm1               \n"
238 		"unpcklps %%xmm1, %%xmm2                \n"
239 		"mulps %[qb2],%[x2]                     \n"
240 		"lddqu %[qb0],%%xmm0                    \n"
241 		"mulps %[ra2],%[z2]                     \n"
242 		"lddqu %[qb1],%%xmm1                    \n"
243 		"mulps %%xmm2,%%xmm0                    \n"
244 		"mulps %[x1],%%xmm1                     \n"
245 		"addps %%xmm1,%%xmm0                    \n"
246 		"movaps %[qa1],%%xmm1                   \n"
247 		"mulps %[y1],%%xmm1                     \n"
248 		"addps %[x2],%%xmm0                     \n"
249 		"movaps %[rb1],%[x2]                    \n"
250 		"mulps %[y1],%[x2]                      \n"
251 		"subps %%xmm1,%%xmm0                    \n"
252 		"movaps %[qa2],%%xmm1                   \n"
253 		"mulps %[y2],%%xmm1                     \n"
254 		"mulps %[rb2],%[y2]                     \n"
255 		"subps %%xmm1,%%xmm0                    \n"
256 		"movaps %[rb0],%%xmm1                   \n"
257 		"mulps %%xmm0,%%xmm1                    \n"
258 		"addps %[x2],%%xmm1                     \n"
259 		"movaps %[x1],%[x2]                     \n"
260 		"movaps %%xmm2,%[x1]                    \n"
261 		"addps %[y2],%%xmm1                     \n"
262 		"movaps %[ra1],%[y2]                    \n"
263 		"mulps %[z1],%[y2]                      \n"
264 		"subps %[y2],%%xmm1                     \n"
265 		"movaps %[y1],%[y2]                     \n"
266 		"movaps %%xmm0,%[y1]                    \n"
267 		"subps %[z2],%%xmm1                     \n"
268 		"movaps %[z1],%[z2]                     \n"
269 		"movaps %%xmm1,%[z1]                    \n"
270 		"movss %%xmm1, (%[data0])               \n"
271 		"shufps $1, %%xmm1, %%xmm1              \n"
272 		"movss %%xmm1, (%[data1])               \n"
273 		"add $4, %[data0]                       \n"
274 		"add $4, %[data1]                       \n"
275 		"sub $1, %[count]                       \n"
276 		"jnz 1b                                 \n"
277 		: /* output */
278 		  [data0]"+r"(data0),
279 		  [data1]"+r"(data1),
280 		  [count]"+r"(count),
281 		  [x1]"+x"(x1),
282 		  [x2]"+x"(x2),
283 		  [y1]"+x"(y1),
284 		  [y2]"+x"(y2),
285 		  [z1]"+x"(z1),
286 		  [z2]"+x"(z2)
287 		: /* input */
288 		  [qb0]"m"(qb0),
289 		  [qb1]"m"(qb1),
290 		  [qb2]"m"(qb2),
291 		  [qa1]"x"(qa1),
292 		  [qa2]"x"(qa2),
293 		  [rb0]"x"(rb0),
294 		  [rb1]"x"(rb1),
295 		  [rb2]"x"(rb2),
296 		  [ra1]"x"(ra1),
297 		  [ra2]"x"(ra2)
298 		: /* clobber */
299 		  "xmm0", "xmm1", "xmm2", "memory", "cc"
300 		);
301 
302 	qL->x1 = x1[0];
303 	qL->x2 = x2[0];
304 	qL->y1 = y1[0];
305 	qL->y2 = y2[0];
306 	rL->y1 = z1[0];
307 	rL->y2 = z2[0];
308 	qR->x1 = x1[1];
309 	qR->x2 = x2[1];
310 	qR->y1 = y1[1];
311 	qR->y2 = y2[1];
312 	rR->y1 = z1[1];
313 	rR->y2 = z2[1];
314 }
315 #endif
316 
eq2_process(struct eq2 * eq2,float * data0,float * data1,int count)317 void eq2_process(struct eq2 *eq2, float *data0, float *data1, int count)
318 {
319 	int i;
320 	int n;
321 	if (!count)
322 		return;
323 	n = eq2->n[0];
324 	if (eq2->n[1] > n)
325 		n = eq2->n[1];
326 	for (i = 0; i < n; i += 2) {
327 		if (i + 1 == n) {
328 			eq2_process_one(&eq2->biquad[i], data0, data1, count);
329 		} else {
330 #if defined(__ARM_NEON__)
331 			eq2_process_two_neon(&eq2->biquad[i], data0, data1,
332 					     count);
333 #elif defined(__SSE3__) && defined(__x86_64__)
334 			eq2_process_two_sse3(&eq2->biquad[i], data0, data1,
335 					     count);
336 #else
337 			eq2_process_one(&eq2->biquad[i], data0, data1, count);
338 			eq2_process_one(&eq2->biquad[i+1], data0, data1, count);
339 #endif
340 		}
341 	}
342 }
343