|
5
|
1 /*
|
|
|
2 * SpanDSP - a series of DSP components for telephony
|
|
|
3 *
|
|
|
4 * vector_int.c - Integer vector arithmetic
|
|
|
5 *
|
|
|
6 * Written by Steve Underwood <steveu@coppice.org>
|
|
|
7 *
|
|
|
8 * Copyright (C) 2006 Steve Underwood
|
|
|
9 *
|
|
|
10 * All rights reserved.
|
|
|
11 *
|
|
|
12 * This program is free software; you can redistribute it and/or modify
|
|
|
13 * it under the terms of the GNU General Public License version 2, as
|
|
|
14 * published by the Free Software Foundation.
|
|
|
15 *
|
|
|
16 * This program is distributed in the hope that it will be useful,
|
|
|
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
19 * GNU General Public License for more details.
|
|
|
20 *
|
|
|
21 * You should have received a copy of the GNU General Public License
|
|
|
22 * along with this program; if not, write to the Free Software
|
|
|
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
|
24 *
|
|
|
25 * $Id: vector_int.c,v 1.5 2006/11/19 14:07:26 steveu Exp $
|
|
|
26 */
|
|
|
27
|
|
|
28 /*! \file */
|
|
|
29
|
|
|
30 #ifdef HAVE_CONFIG_H
|
|
|
31 #include <config.h>
|
|
|
32 #endif
|
|
|
33
|
|
|
34 #include <inttypes.h>
|
|
|
35 #include <stdlib.h>
|
|
|
36 #include <stdio.h>
|
|
|
37 #include <string.h>
|
|
|
38 #if defined(HAVE_TGMATH_H)
|
|
|
39 #include <tgmath.h>
|
|
|
40 #endif
|
|
|
41 #if defined(HAVE_MATH_H)
|
|
|
42 #include <math.h>
|
|
|
43 #endif
|
|
|
44 #include <assert.h>
|
|
|
45
|
|
|
46 #include "spandsp/telephony.h"
|
|
|
47 #include "spandsp/logging.h"
|
|
|
48 #include "spandsp/vector_int.h"
|
|
|
49
|
|
|
50 int32_t vec_dot_prodi16(const int16_t x[], const int16_t y[], int n)
|
|
|
51 {
|
|
|
52 int32_t z;
|
|
|
53
|
|
|
54 #if defined(__GNUC__) && defined(__i386__)
|
|
|
55 __asm__ __volatile__(
|
|
|
56 " emms;\n"
|
|
|
57 " pxor %%mm0,%%mm0;\n"
|
|
|
58 " leal -32(%%esi,%%eax,2),%%edx;\n" /* edx = top - 32 */
|
|
|
59
|
|
|
60 " cmpl %%edx,%%esi;\n"
|
|
|
61 " ja 1f;\n"
|
|
|
62
|
|
|
63 /* Work in blocks of 16 int16_t's until we are near the end */
|
|
|
64 " .p2align 2;\n"
|
|
|
65 "2:\n"
|
|
|
66 " movq (%%edi),%%mm1;\n"
|
|
|
67 " movq (%%esi),%%mm2;\n"
|
|
|
68 " pmaddwd %%mm2,%%mm1;\n"
|
|
|
69 " paddd %%mm1,%%mm0;\n"
|
|
|
70 " movq 8(%%edi),%%mm1;\n"
|
|
|
71 " movq 8(%%esi),%%mm2;\n"
|
|
|
72 " pmaddwd %%mm2,%%mm1;\n"
|
|
|
73 " paddd %%mm1,%%mm0;\n"
|
|
|
74 " movq 16(%%edi),%%mm1;\n"
|
|
|
75 " movq 16(%%esi),%%mm2;\n"
|
|
|
76 " pmaddwd %%mm2,%%mm1;\n"
|
|
|
77 " paddd %%mm1,%%mm0;\n"
|
|
|
78 " movq 24(%%edi),%%mm1;\n"
|
|
|
79 " movq 24(%%esi),%%mm2;\n"
|
|
|
80 " pmaddwd %%mm2,%%mm1;\n"
|
|
|
81 " paddd %%mm1,%%mm0;\n"
|
|
|
82
|
|
|
83 " addl $32,%%esi;\n"
|
|
|
84 " addl $32,%%edi;\n"
|
|
|
85 " cmpl %%edx,%%esi;\n"
|
|
|
86 " jbe 2b;\n"
|
|
|
87
|
|
|
88 " .p2align 2;\n"
|
|
|
89 "1:\n"
|
|
|
90 " addl $24,%%edx;\n" /* now edx = top - 8 */
|
|
|
91 " cmpl %%edx,%%esi;\n"
|
|
|
92 " ja 3f;\n"
|
|
|
93
|
|
|
94 /* Work in blocks of 4 int16_t's until we are near the end */
|
|
|
95 " .p2align 2;\n"
|
|
|
96 "4:\n"
|
|
|
97 " movq (%%edi),%%mm1;\n"
|
|
|
98 " movq (%%esi),%%mm2;\n"
|
|
|
99 " pmaddwd %%mm2,%%mm1;\n"
|
|
|
100 " paddd %%mm1,%%mm0;\n"
|
|
|
101
|
|
|
102 " addl $8,%%esi;\n"
|
|
|
103 " addl $8,%%edi;\n"
|
|
|
104 " cmpl %%edx,%%esi;"
|
|
|
105 " jbe 4b;\n"
|
|
|
106
|
|
|
107 " .p2align 2;\n"
|
|
|
108 "3:\n"
|
|
|
109 " addl $4,%%edx;\n" /* now edx = top - 4 */
|
|
|
110 " cmpl %%edx,%%esi;\n"
|
|
|
111 " ja 5f;\n"
|
|
|
112
|
|
|
113 /* Work in a block of 2 int16_t's */
|
|
|
114 " movd (%%edi),%%mm1;\n"
|
|
|
115 " movd (%%esi),%%mm2;\n"
|
|
|
116 " pmaddwd %%mm2,%%mm1;\n"
|
|
|
117 " paddd %%mm1,%%mm0;\n"
|
|
|
118
|
|
|
119 " addl $4,%%esi;\n"
|
|
|
120 " addl $4,%%edi;\n"
|
|
|
121
|
|
|
122 " .p2align 2;\n"
|
|
|
123 "5:\n"
|
|
|
124 " addl $2,%%edx;\n" /* now edx = top - 2 */
|
|
|
125 " cmpl %%edx,%%esi;\n"
|
|
|
126 " ja 6f;\n"
|
|
|
127
|
|
|
128 /* Deal with the very last int16_t, when n is odd */
|
|
|
129 " movswl (%%edi),%%eax;\n"
|
|
|
130 " andl $65535,%%eax;\n"
|
|
|
131 " movd %%eax,%%mm1;\n"
|
|
|
132 " movswl (%%esi),%%eax;\n"
|
|
|
133 " andl $65535,%%eax;\n"
|
|
|
134 " movd %%eax,%%mm2;\n"
|
|
|
135 " pmaddwd %%mm2,%%mm1;\n"
|
|
|
136 " paddd %%mm1,%%mm0;\n"
|
|
|
137
|
|
|
138 " .p2align 2;\n"
|
|
|
139 "6:\n"
|
|
|
140 /* Merge the pieces of the answer */
|
|
|
141 " movq %%mm0,%%mm1;\n"
|
|
|
142 " punpckhdq %%mm0,%%mm1;\n"
|
|
|
143 " paddd %%mm1,%%mm0;\n"
|
|
|
144 /* et voila, eax has the final result */
|
|
|
145 " movd %%mm0,%%eax;\n"
|
|
|
146
|
|
|
147 " emms;\n"
|
|
|
148 : "=a" (z)
|
|
|
149 : "S" (x), "D" (y), "a" (n)
|
|
|
150 : "cc"
|
|
|
151 );
|
|
|
152 #else
|
|
|
153 int i;
|
|
|
154
|
|
|
155 z = 0;
|
|
|
156 for (i = 0; i < n; i++)
|
|
|
157 z += x[i]*y[i];
|
|
|
158 #endif
|
|
|
159 return z;
|
|
|
160 }
|
|
|
161 /*- End of function --------------------------------------------------------*/
|
|
|
162
|
|
|
163 int32_t vec_min_maxi16(const int16_t x[], int n, int16_t out[])
|
|
|
164 {
|
|
|
165 #if defined(__GNUC__) && defined(__i386__)
|
|
|
166 static const int32_t lower_bound = 0x80008000;
|
|
|
167 static const int32_t upper_bound = 0x7FFF7FFF;
|
|
|
168 int32_t max;
|
|
|
169
|
|
|
170 __asm__ __volatile__(
|
|
|
171 " emms;\n"
|
|
|
172 " pushl %%edx;\n"
|
|
|
173 " leal -8(%%esi,%%eax,2),%%edx;\n"
|
|
|
174
|
|
|
175 " cmpl %%edx,%%esi;\n"
|
|
|
176 " jbe 2f;\n"
|
|
|
177 " movd %[lower],%%mm0;\n"
|
|
|
178 " movd %[upper],%%mm1;\n"
|
|
|
179 " jmp 1f;\n"
|
|
|
180
|
|
|
181 " .p2align 2;\n"
|
|
|
182 "2:\n"
|
|
|
183 " movq (%%esi),%%mm0;\n" /* mm0 will be max's */
|
|
|
184 " movq %%mm0,%%mm1;\n" /* mm1 will be min's */
|
|
|
185 " addl $8,%%esi;\n"
|
|
|
186 " cmpl %%edx,%%esi;\n"
|
|
|
187 " ja 4f;\n"
|
|
|
188
|
|
|
189 " .p2align 2;\n"
|
|
|
190 "3:\n"
|
|
|
191 " movq (%%esi),%%mm2;\n"
|
|
|
192
|
|
|
193 " movq %%mm2,%%mm3;\n"
|
|
|
194 " pcmpgtw %%mm0,%%mm3;\n" /* mm3 is bitmask for words where mm2 > mm0 */
|
|
|
195 " movq %%mm3,%%mm4;\n"
|
|
|
196 " pand %%mm2,%%mm3;\n" /* mm3 is mm2 masked to new max's */
|
|
|
197 " pandn %%mm0,%%mm4;\n" /* mm4 is mm0 masked to its max's */
|
|
|
198 " por %%mm3,%%mm4;\n"
|
|
|
199 " movq %%mm4,%%mm0;\n" /* Now mm0 is updated max's */
|
|
|
200
|
|
|
201 " movq %%mm1,%%mm3;\n"
|
|
|
202 " pcmpgtw %%mm2,%%mm3;\n" /* mm3 is bitmask for words where mm2 < mm1 */
|
|
|
203 " pand %%mm3,%%mm2;\n" /* mm2 is mm2 masked to new min's */
|
|
|
204 " pandn %%mm1,%%mm3;\n" /* mm3 is mm1 masked to its min's */
|
|
|
205 " por %%mm3,%%mm2;\n"
|
|
|
206 " movq %%mm2,%%mm1;\n" /* now mm1 is updated min's */
|
|
|
207
|
|
|
208 " addl $8,%%esi;\n"
|
|
|
209 " cmpl %%edx,%%esi;\n"
|
|
|
210 " jbe 3b;\n"
|
|
|
211
|
|
|
212 " .p2align 2;\n"
|
|
|
213 "4:\n"
|
|
|
214 /* Merge down the 4-word max/mins to lower 2 words */
|
|
|
215 " movq %%mm0,%%mm2;\n"
|
|
|
216 " psrlq $32,%%mm2;\n"
|
|
|
217 " movq %%mm2,%%mm3;\n"
|
|
|
218 " pcmpgtw %%mm0,%%mm3;\n" /* mm3 is bitmask for words where mm2 > mm0 */
|
|
|
219 " pand %%mm3,%%mm2;\n" /* mm2 is mm2 masked to new max's */
|
|
|
220 " pandn %%mm0,%%mm3;\n" /* mm3 is mm0 masked to its max's */
|
|
|
221 " por %%mm3,%%mm2;\n"
|
|
|
222 " movq %%mm2,%%mm0;\n" /* now mm0 is updated max's */
|
|
|
223
|
|
|
224 " movq %%mm1,%%mm2;\n"
|
|
|
225 " psrlq $32,%%mm2;\n"
|
|
|
226 " movq %%mm1,%%mm3;\n"
|
|
|
227 " pcmpgtw %%mm2,%%mm3;\n" /* mm3 is bitmask for words where mm2 < mm1 */
|
|
|
228 " pand %%mm3,%%mm2;\n" /* mm2 is mm2 masked to new min's */
|
|
|
229 " pandn %%mm1,%%mm3;\n" /* mm3 is mm1 masked to its min's */
|
|
|
230 " por %%mm3,%%mm2;\n"
|
|
|
231 " movq %%mm2,%%mm1;\n" /* now mm1 is updated min's */
|
|
|
232
|
|
|
233 " .p2align 2;\n"
|
|
|
234 "1:\n"
|
|
|
235 " addl $4,%%edx;\n" /* now dx = top-4 */
|
|
|
236 " cmpl %%edx,%%esi;\n"
|
|
|
237 " ja 5f;\n"
|
|
|
238 /* Here, there are >= 2 words of input remaining */
|
|
|
239 " movd (%%esi),%%mm2;\n"
|
|
|
240
|
|
|
241 " movq %%mm2,%%mm3;\n"
|
|
|
242 " pcmpgtw %%mm0,%%mm3;\n" /* mm3 is bitmask for words where mm2 > mm0 */
|
|
|
243 " movq %%mm3,%%mm4;\n"
|
|
|
244 " pand %%mm2,%%mm3;\n" /* mm3 is mm2 masked to new max's */
|
|
|
245 " pandn %%mm0,%%mm4;\n" /* mm4 is mm0 masked to its max's */
|
|
|
246 " por %%mm3,%%mm4;\n"
|
|
|
247 " movq %%mm4,%%mm0;\n" /* now mm0 is updated max's */
|
|
|
248
|
|
|
249 " movq %%mm1,%%mm3;\n"
|
|
|
250 " pcmpgtw %%mm2,%%mm3;\n" /* mm3 is bitmask for words where mm2 < mm1 */
|
|
|
251 " pand %%mm3,%%mm2;\n" /* mm2 is mm2 masked to new min's */
|
|
|
252 " pandn %%mm1,%%mm3;\n" /* mm3 is mm1 masked to its min's */
|
|
|
253 " por %%mm3,%%mm2;\n"
|
|
|
254 " movq %%mm2,%%mm1;\n" /* now mm1 is updated min's */
|
|
|
255
|
|
|
256 " addl $4,%%esi;\n"
|
|
|
257
|
|
|
258 " .p2align 2;\n"
|
|
|
259 "5:\n"
|
|
|
260 /* Merge down the 2-word max/mins to 1 word */
|
|
|
261 " movq %%mm0,%%mm2;\n"
|
|
|
262 " psrlq $16,%%mm2;\n"
|
|
|
263 " movq %%mm2,%%mm3;\n"
|
|
|
264 " pcmpgtw %%mm0,%%mm3;\n" /* mm3 is bitmask for words where mm2 > mm0 */
|
|
|
265 " pand %%mm3,%%mm2;\n" /* mm2 is mm2 masked to new max's */
|
|
|
266 " pandn %%mm0,%%mm3;\n" /* mm3 is mm0 masked to its max's */
|
|
|
267 " por %%mm3,%%mm2;\n"
|
|
|
268 " movd %%mm2,%%ecx;\n" /* cx is max so far */
|
|
|
269
|
|
|
270 " movq %%mm1,%%mm2;\n"
|
|
|
271 " psrlq $16,%%mm2;\n"
|
|
|
272 " movq %%mm1,%%mm3;\n"
|
|
|
273 " pcmpgtw %%mm2,%%mm3;\n" /* mm3 is bitmask for words where mm2 < mm1 */
|
|
|
274 " pand %%mm3,%%mm2;\n" /* mm2 is mm2 masked to new min's */
|
|
|
275 " pandn %%mm1,%%mm3;\n" /* mm3 is mm1 masked to its min's */
|
|
|
276 " por %%mm3,%%mm2;\n"
|
|
|
277 " movd %%mm2,%%eax;\n" /* ax is min so far */
|
|
|
278
|
|
|
279 " addl $2,%%edx;\n" /* now dx = top-2 */
|
|
|
280 " cmpl %%edx,%%esi;\n"
|
|
|
281 " ja 6f;\n"
|
|
|
282
|
|
|
283 /* Here, there is one word of input left */
|
|
|
284 " cmpw (%%esi),%%cx;\n"
|
|
|
285 " jge 9f;\n"
|
|
|
286 " movw (%%esi),%%cx;\n"
|
|
|
287 " .p2align 2;\n"
|
|
|
288 "9:\n"
|
|
|
289 " cmpw (%%esi),%%ax;\n"
|
|
|
290 " jle 6f;\n"
|
|
|
291 " movw (%%esi),%%ax;\n"
|
|
|
292
|
|
|
293 " .p2align 2;\n"
|
|
|
294 "6:\n"
|
|
|
295 /* (finally!) cx is the max, ax the min */
|
|
|
296 " movswl %%cx,%%ecx;\n"
|
|
|
297 " movswl %%ax,%%eax;\n"
|
|
|
298
|
|
|
299 " popl %%edx;\n" /* ptr to output max,min vals */
|
|
|
300 " andl %%edx,%%edx;\n"
|
|
|
301 " jz 7f;\n"
|
|
|
302 " movw %%cx,(%%edx);\n" /* max */
|
|
|
303 " movw %%ax,2(%%edx);\n" /* min */
|
|
|
304 " .p2align 2;\n"
|
|
|
305 "7:\n"
|
|
|
306 /* Now calculate max absolute value */
|
|
|
307 " negl %%eax;\n"
|
|
|
308 " cmpl %%ecx,%%eax;\n"
|
|
|
309 " jge 8f;\n"
|
|
|
310 " movl %%ecx,%%eax;\n"
|
|
|
311 " .p2align 2;\n"
|
|
|
312 "8:\n"
|
|
|
313 " emms;\n"
|
|
|
314 : "=a" (max)
|
|
|
315 : "S" (x), "a" (n), "d" (out), [lower] "m" (lower_bound), [upper] "m" (upper_bound)
|
|
|
316 : "ecx"
|
|
|
317 );
|
|
|
318 return max;
|
|
|
319 #else
|
|
|
320 int i;
|
|
|
321 int16_t min;
|
|
|
322 int16_t max;
|
|
|
323 int16_t temp;
|
|
|
324 int32_t z;
|
|
|
325
|
|
|
326 max = INT16_MIN;
|
|
|
327 min = INT16_MAX;
|
|
|
328 for (i = 0; i < n; i++)
|
|
|
329 {
|
|
|
330 temp = x[i];
|
|
|
331 if (temp > max)
|
|
|
332 max = temp;
|
|
|
333 /*endif*/
|
|
|
334 if (temp < min)
|
|
|
335 min = temp;
|
|
|
336 /*endif*/
|
|
|
337 }
|
|
|
338 /*endfor*/
|
|
|
339 out[0] = max;
|
|
|
340 out[1] = min;
|
|
|
341 z = abs(min);
|
|
|
342 if (z > max)
|
|
|
343 return z;
|
|
|
344 return max;
|
|
|
345 #endif
|
|
|
346 }
|
|
|
347 /*- End of function --------------------------------------------------------*/
|
|
|
348 /*- End of file ------------------------------------------------------------*/
|
