FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
float_dsp_mips.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2012
3  * MIPS Technologies, Inc., California.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  * notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  * notice, this list of conditions and the following disclaimer in the
12  * documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14  * contributors may be used to endorse or promote products derived from
15  * this software without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED. IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * Author: Branimir Vasic (bvasic@mips.com)
30  * Author: Zoran Lukic (zoranl@mips.com)
31  *
32  * This file is part of FFmpeg.
33  *
34  * FFmpeg is free software; you can redistribute it and/or
35  * modify it under the terms of the GNU Lesser General Public
36  * License as published by the Free Software Foundation; either
37  * version 2.1 of the License, or (at your option) any later version.
38  *
39  * FFmpeg is distributed in the hope that it will be useful,
40  * but WITHOUT ANY WARRANTY; without even the implied warranty of
41  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
42  * Lesser General Public License for more details.
43  *
44  * You should have received a copy of the GNU Lesser General Public
45  * License along with FFmpeg; if not, write to the Free Software
46  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
47  */
48 
49 /**
50  * @file
51  * Reference: libavutil/float_dsp.c
52  */
53 
54 #include "config.h"
55 #include "libavutil/float_dsp.h"
56 
57 #if HAVE_INLINE_ASM && HAVE_MIPSFPU
58 static void vector_fmul_mips(float *dst, const float *src0, const float *src1,
59  int len)
60 {
61  int i;
62 
63  if (len & 3) {
64  for (i = 0; i < len; i++)
65  dst[i] = src0[i] * src1[i];
66  } else {
67  float *d = (float *)dst;
68  float *d_end = d + len;
69  float *s0 = (float *)src0;
70  float *s1 = (float *)src1;
71 
72  float src0_0, src0_1, src0_2, src0_3;
73  float src1_0, src1_1, src1_2, src1_3;
74 
75  __asm__ volatile (
76  "1: \n\t"
77  "lwc1 %[src0_0], 0(%[s0]) \n\t"
78  "lwc1 %[src1_0], 0(%[s1]) \n\t"
79  "lwc1 %[src0_1], 4(%[s0]) \n\t"
80  "lwc1 %[src1_1], 4(%[s1]) \n\t"
81  "lwc1 %[src0_2], 8(%[s0]) \n\t"
82  "lwc1 %[src1_2], 8(%[s1]) \n\t"
83  "lwc1 %[src0_3], 12(%[s0]) \n\t"
84  "lwc1 %[src1_3], 12(%[s1]) \n\t"
85  "mul.s %[src0_0], %[src0_0], %[src1_0] \n\t"
86  "mul.s %[src0_1], %[src0_1], %[src1_1] \n\t"
87  "mul.s %[src0_2], %[src0_2], %[src1_2] \n\t"
88  "mul.s %[src0_3], %[src0_3], %[src1_3] \n\t"
89  "swc1 %[src0_0], 0(%[d]) \n\t"
90  "swc1 %[src0_1], 4(%[d]) \n\t"
91  "swc1 %[src0_2], 8(%[d]) \n\t"
92  "swc1 %[src0_3], 12(%[d]) \n\t"
93  "addiu %[s0], %[s0], 16 \n\t"
94  "addiu %[s1], %[s1], 16 \n\t"
95  "addiu %[d], %[d], 16 \n\t"
96  "bne %[d], %[d_end], 1b \n\t"
97 
98  : [src0_0]"=&f"(src0_0), [src0_1]"=&f"(src0_1),
99  [src0_2]"=&f"(src0_2), [src0_3]"=&f"(src0_3),
100  [src1_0]"=&f"(src1_0), [src1_1]"=&f"(src1_1),
101  [src1_2]"=&f"(src1_2), [src1_3]"=&f"(src1_3),
102  [d]"+r"(d), [s0]"+r"(s0), [s1]"+r"(s1)
103  : [d_end]"r"(d_end)
104  : "memory"
105  );
106  }
107 }
108 
109 static void vector_fmul_scalar_mips(float *dst, const float *src, float mul,
110  int len)
111 {
112  float temp0, temp1, temp2, temp3;
113  float *local_src = (float*)src;
114  float *end = local_src + len;
115 
116  /* loop unrolled 4 times */
117  __asm__ volatile(
118  ".set push \n\t"
119  ".set noreorder \n\t"
120  "1: \n\t"
121  "lwc1 %[temp0], 0(%[src]) \n\t"
122  "lwc1 %[temp1], 4(%[src]) \n\t"
123  "lwc1 %[temp2], 8(%[src]) \n\t"
124  "lwc1 %[temp3], 12(%[src]) \n\t"
125  "addiu %[dst], %[dst], 16 \n\t"
126  "mul.s %[temp0], %[temp0], %[mul] \n\t"
127  "mul.s %[temp1], %[temp1], %[mul] \n\t"
128  "mul.s %[temp2], %[temp2], %[mul] \n\t"
129  "mul.s %[temp3], %[temp3], %[mul] \n\t"
130  "addiu %[src], %[src], 16 \n\t"
131  "swc1 %[temp0], -16(%[dst]) \n\t"
132  "swc1 %[temp1], -12(%[dst]) \n\t"
133  "swc1 %[temp2], -8(%[dst]) \n\t"
134  "bne %[src], %[end], 1b \n\t"
135  " swc1 %[temp3], -4(%[dst]) \n\t"
136  ".set pop \n\t"
137 
138  : [temp0]"=&f"(temp0), [temp1]"=&f"(temp1),
139  [temp2]"=&f"(temp2), [temp3]"=&f"(temp3),
140  [dst]"+r"(dst), [src]"+r"(local_src)
141  : [end]"r"(end), [mul]"f"(mul)
142  : "memory"
143  );
144 }
145 
146 static void vector_fmul_window_mips(float *dst, const float *src0,
147  const float *src1, const float *win, int len)
148 {
149  int i, j;
150  /*
151  * variables used in inline assembler
152  */
153  float * dst_i, * dst_j, * dst_i2, * dst_j2;
154  float temp, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
155 
156  dst += len;
157  win += len;
158  src0 += len;
159 
160  for (i = -len, j = len - 1; i < 0; i += 8, j -= 8) {
161 
162  dst_i = dst + i;
163  dst_j = dst + j;
164 
165  dst_i2 = dst + i + 4;
166  dst_j2 = dst + j - 4;
167 
168  __asm__ volatile (
169  "mul.s %[temp], %[s1], %[wi] \n\t"
170  "mul.s %[temp1], %[s1], %[wj] \n\t"
171  "mul.s %[temp2], %[s11], %[wi1] \n\t"
172  "mul.s %[temp3], %[s11], %[wj1] \n\t"
173 
174  "msub.s %[temp], %[temp], %[s0], %[wj] \n\t"
175  "madd.s %[temp1], %[temp1], %[s0], %[wi] \n\t"
176  "msub.s %[temp2], %[temp2], %[s01], %[wj1] \n\t"
177  "madd.s %[temp3], %[temp3], %[s01], %[wi1] \n\t"
178 
179  "swc1 %[temp], 0(%[dst_i]) \n\t" /* dst[i] = s0*wj - s1*wi; */
180  "swc1 %[temp1], 0(%[dst_j]) \n\t" /* dst[j] = s0*wi + s1*wj; */
181  "swc1 %[temp2], 4(%[dst_i]) \n\t" /* dst[i+1] = s01*wj1 - s11*wi1; */
182  "swc1 %[temp3], -4(%[dst_j]) \n\t" /* dst[j-1] = s01*wi1 + s11*wj1; */
183 
184  "mul.s %[temp4], %[s12], %[wi2] \n\t"
185  "mul.s %[temp5], %[s12], %[wj2] \n\t"
186  "mul.s %[temp6], %[s13], %[wi3] \n\t"
187  "mul.s %[temp7], %[s13], %[wj3] \n\t"
188 
189  "msub.s %[temp4], %[temp4], %[s02], %[wj2] \n\t"
190  "madd.s %[temp5], %[temp5], %[s02], %[wi2] \n\t"
191  "msub.s %[temp6], %[temp6], %[s03], %[wj3] \n\t"
192  "madd.s %[temp7], %[temp7], %[s03], %[wi3] \n\t"
193 
194  "swc1 %[temp4], 8(%[dst_i]) \n\t" /* dst[i+2] = s02*wj2 - s12*wi2; */
195  "swc1 %[temp5], -8(%[dst_j]) \n\t" /* dst[j-2] = s02*wi2 + s12*wj2; */
196  "swc1 %[temp6], 12(%[dst_i]) \n\t" /* dst[i+2] = s03*wj3 - s13*wi3; */
197  "swc1 %[temp7], -12(%[dst_j]) \n\t" /* dst[j-3] = s03*wi3 + s13*wj3; */
198  : [temp]"=&f"(temp), [temp1]"=&f"(temp1), [temp2]"=&f"(temp2),
199  [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), [temp5]"=&f"(temp5),
200  [temp6]"=&f"(temp6), [temp7]"=&f"(temp7)
201  : [dst_j]"r"(dst_j), [dst_i]"r" (dst_i),
202  [s0] "f"(src0[i]), [wj] "f"(win[j]), [s1] "f"(src1[j]),
203  [wi] "f"(win[i]), [s01]"f"(src0[i + 1]),[wj1]"f"(win[j - 1]),
204  [s11]"f"(src1[j - 1]), [wi1]"f"(win[i + 1]), [s02]"f"(src0[i + 2]),
205  [wj2]"f"(win[j - 2]), [s12]"f"(src1[j - 2]),[wi2]"f"(win[i + 2]),
206  [s03]"f"(src0[i + 3]), [wj3]"f"(win[j - 3]), [s13]"f"(src1[j - 3]),
207  [wi3]"f"(win[i + 3])
208  : "memory"
209  );
210 
211  __asm__ volatile (
212  "mul.s %[temp], %[s1], %[wi] \n\t"
213  "mul.s %[temp1], %[s1], %[wj] \n\t"
214  "mul.s %[temp2], %[s11], %[wi1] \n\t"
215  "mul.s %[temp3], %[s11], %[wj1] \n\t"
216 
217  "msub.s %[temp], %[temp], %[s0], %[wj] \n\t"
218  "madd.s %[temp1], %[temp1], %[s0], %[wi] \n\t"
219  "msub.s %[temp2], %[temp2], %[s01], %[wj1] \n\t"
220  "madd.s %[temp3], %[temp3], %[s01], %[wi1] \n\t"
221 
222  "swc1 %[temp], 0(%[dst_i2]) \n\t" /* dst[i] = s0*wj - s1*wi; */
223  "swc1 %[temp1], 0(%[dst_j2]) \n\t" /* dst[j] = s0*wi + s1*wj; */
224  "swc1 %[temp2], 4(%[dst_i2]) \n\t" /* dst[i+1] = s01*wj1 - s11*wi1; */
225  "swc1 %[temp3], -4(%[dst_j2]) \n\t" /* dst[j-1] = s01*wi1 + s11*wj1; */
226 
227  "mul.s %[temp4], %[s12], %[wi2] \n\t"
228  "mul.s %[temp5], %[s12], %[wj2] \n\t"
229  "mul.s %[temp6], %[s13], %[wi3] \n\t"
230  "mul.s %[temp7], %[s13], %[wj3] \n\t"
231 
232  "msub.s %[temp4], %[temp4], %[s02], %[wj2] \n\t"
233  "madd.s %[temp5], %[temp5], %[s02], %[wi2] \n\t"
234  "msub.s %[temp6], %[temp6], %[s03], %[wj3] \n\t"
235  "madd.s %[temp7], %[temp7], %[s03], %[wi3] \n\t"
236 
237  "swc1 %[temp4], 8(%[dst_i2]) \n\t" /* dst[i+2] = s02*wj2 - s12*wi2; */
238  "swc1 %[temp5], -8(%[dst_j2]) \n\t" /* dst[j-2] = s02*wi2 + s12*wj2; */
239  "swc1 %[temp6], 12(%[dst_i2]) \n\t" /* dst[i+2] = s03*wj3 - s13*wi3; */
240  "swc1 %[temp7], -12(%[dst_j2]) \n\t" /* dst[j-3] = s03*wi3 + s13*wj3; */
241  : [temp]"=&f"(temp),
242  [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), [temp3]"=&f"(temp3),
243  [temp4]"=&f"(temp4), [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
244  [temp7] "=&f" (temp7)
245  : [dst_j2]"r"(dst_j2), [dst_i2]"r"(dst_i2),
246  [s0] "f"(src0[i + 4]), [wj] "f"(win[j - 4]), [s1] "f"(src1[j - 4]),
247  [wi] "f"(win[i + 4]), [s01]"f"(src0[i + 5]),[wj1]"f"(win[j - 5]),
248  [s11]"f"(src1[j - 5]), [wi1]"f"(win[i + 5]), [s02]"f"(src0[i + 6]),
249  [wj2]"f"(win[j - 6]), [s12]"f"(src1[j - 6]),[wi2]"f"(win[i + 6]),
250  [s03]"f"(src0[i + 7]), [wj3]"f"(win[j - 7]), [s13]"f"(src1[j - 7]),
251  [wi3]"f"(win[i + 7])
252  : "memory"
253  );
254  }
255 }
256 
257 static void butterflies_float_mips(float *av_restrict v1, float *av_restrict v2,
258  int len)
259 {
260  float temp0, temp1, temp2, temp3, temp4;
261  float temp5, temp6, temp7, temp8, temp9;
262  float temp10, temp11, temp12, temp13, temp14, temp15;
263  int pom;
264  pom = (len >> 2)-1;
265 
266  /* loop unrolled 4 times */
267  __asm__ volatile (
268  "lwc1 %[temp0], 0(%[v1]) \n\t"
269  "lwc1 %[temp1], 4(%[v1]) \n\t"
270  "lwc1 %[temp2], 8(%[v1]) \n\t"
271  "lwc1 %[temp3], 12(%[v1]) \n\t"
272  "lwc1 %[temp4], 0(%[v2]) \n\t"
273  "lwc1 %[temp5], 4(%[v2]) \n\t"
274  "lwc1 %[temp6], 8(%[v2]) \n\t"
275  "lwc1 %[temp7], 12(%[v2]) \n\t"
276  "beq %[pom], $zero, 2f \n\t"
277  "1: \n\t"
278  "sub.s %[temp8], %[temp0], %[temp4] \n\t"
279  "add.s %[temp9], %[temp0], %[temp4] \n\t"
280  "sub.s %[temp10], %[temp1], %[temp5] \n\t"
281  "add.s %[temp11], %[temp1], %[temp5] \n\t"
282  "sub.s %[temp12], %[temp2], %[temp6] \n\t"
283  "add.s %[temp13], %[temp2], %[temp6] \n\t"
284  "sub.s %[temp14], %[temp3], %[temp7] \n\t"
285  "add.s %[temp15], %[temp3], %[temp7] \n\t"
286  "addiu %[v1], %[v1], 16 \n\t"
287  "addiu %[v2], %[v2], 16 \n\t"
288  "addiu %[pom], %[pom], -1 \n\t"
289  "lwc1 %[temp0], 0(%[v1]) \n\t"
290  "lwc1 %[temp1], 4(%[v1]) \n\t"
291  "lwc1 %[temp2], 8(%[v1]) \n\t"
292  "lwc1 %[temp3], 12(%[v1]) \n\t"
293  "lwc1 %[temp4], 0(%[v2]) \n\t"
294  "lwc1 %[temp5], 4(%[v2]) \n\t"
295  "lwc1 %[temp6], 8(%[v2]) \n\t"
296  "lwc1 %[temp7], 12(%[v2]) \n\t"
297  "swc1 %[temp9], -16(%[v1]) \n\t"
298  "swc1 %[temp8], -16(%[v2]) \n\t"
299  "swc1 %[temp11], -12(%[v1]) \n\t"
300  "swc1 %[temp10], -12(%[v2]) \n\t"
301  "swc1 %[temp13], -8(%[v1]) \n\t"
302  "swc1 %[temp12], -8(%[v2]) \n\t"
303  "swc1 %[temp15], -4(%[v1]) \n\t"
304  "swc1 %[temp14], -4(%[v2]) \n\t"
305  "bgtz %[pom], 1b \n\t"
306  "2: \n\t"
307  "sub.s %[temp8], %[temp0], %[temp4] \n\t"
308  "add.s %[temp9], %[temp0], %[temp4] \n\t"
309  "sub.s %[temp10], %[temp1], %[temp5] \n\t"
310  "add.s %[temp11], %[temp1], %[temp5] \n\t"
311  "sub.s %[temp12], %[temp2], %[temp6] \n\t"
312  "add.s %[temp13], %[temp2], %[temp6] \n\t"
313  "sub.s %[temp14], %[temp3], %[temp7] \n\t"
314  "add.s %[temp15], %[temp3], %[temp7] \n\t"
315  "swc1 %[temp9], 0(%[v1]) \n\t"
316  "swc1 %[temp8], 0(%[v2]) \n\t"
317  "swc1 %[temp11], 4(%[v1]) \n\t"
318  "swc1 %[temp10], 4(%[v2]) \n\t"
319  "swc1 %[temp13], 8(%[v1]) \n\t"
320  "swc1 %[temp12], 8(%[v2]) \n\t"
321  "swc1 %[temp15], 12(%[v1]) \n\t"
322  "swc1 %[temp14], 12(%[v2]) \n\t"
323 
324  : [v1]"+r"(v1), [v2]"+r"(v2), [pom]"+r"(pom), [temp0] "=&f" (temp0),
325  [temp1]"=&f"(temp1), [temp2]"=&f"(temp2), [temp3]"=&f"(temp3),
326  [temp4]"=&f"(temp4), [temp5]"=&f"(temp5), [temp6]"=&f"(temp6),
327  [temp7]"=&f"(temp7), [temp8]"=&f"(temp8), [temp9]"=&f"(temp9),
328  [temp10]"=&f"(temp10), [temp11]"=&f"(temp11), [temp12]"=&f"(temp12),
329  [temp13]"=&f"(temp13), [temp14]"=&f"(temp14), [temp15]"=&f"(temp15)
330  :
331  : "memory"
332  );
333 }
334 
335 static void vector_fmul_reverse_mips(float *dst, const float *src0, const float *src1, int len){
336  int i;
337  float temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
338  src1 += len-1;
339 
340  for(i=0; i<(len>>2); i++)
341  {
342  /* loop unrolled 4 times */
343  __asm__ volatile(
344  "lwc1 %[temp0], 0(%[src0]) \n\t"
345  "lwc1 %[temp1], 0(%[src1]) \n\t"
346  "lwc1 %[temp2], 4(%[src0]) \n\t"
347  "lwc1 %[temp3], -4(%[src1]) \n\t"
348  "lwc1 %[temp4], 8(%[src0]) \n\t"
349  "lwc1 %[temp5], -8(%[src1]) \n\t"
350  "lwc1 %[temp6], 12(%[src0]) \n\t"
351  "lwc1 %[temp7], -12(%[src1]) \n\t"
352  "mul.s %[temp0], %[temp1], %[temp0] \n\t"
353  "mul.s %[temp2], %[temp3], %[temp2] \n\t"
354  "mul.s %[temp4], %[temp5], %[temp4] \n\t"
355  "mul.s %[temp6], %[temp7], %[temp6] \n\t"
356  "addiu %[src0], %[src0], 16 \n\t"
357  "addiu %[src1], %[src1], -16 \n\t"
358  "addiu %[dst], %[dst], 16 \n\t"
359  "swc1 %[temp0], -16(%[dst]) \n\t"
360  "swc1 %[temp2], -12(%[dst]) \n\t"
361  "swc1 %[temp4], -8(%[dst]) \n\t"
362  "swc1 %[temp6], -4(%[dst]) \n\t"
363 
364  : [dst]"+r"(dst), [src0]"+r"(src0), [src1]"+r"(src1),
365  [temp0]"=&f"(temp0), [temp1]"=&f"(temp1),[temp2]"=&f"(temp2),
366  [temp3]"=&f"(temp3), [temp4]"=&f"(temp4), [temp5]"=&f"(temp5),
367  [temp6]"=&f"(temp6), [temp7]"=&f"(temp7)
368  :
369  : "memory"
370  );
371  }
372 }
373 #endif /* HAVE_INLINE_ASM && HAVE_MIPSFPU */
374 
376 #if HAVE_INLINE_ASM && HAVE_MIPSFPU
377  fdsp->vector_fmul = vector_fmul_mips;
378  fdsp->vector_fmul_scalar = vector_fmul_scalar_mips;
379  fdsp->vector_fmul_window = vector_fmul_window_mips;
380  fdsp->butterflies_float = butterflies_float_mips;
381  fdsp->vector_fmul_reverse = vector_fmul_reverse_mips;
382 #endif /* HAVE_INLINE_ASM && HAVE_MIPSFPU */
383 }