FFmpeg
iirfilter_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: Bojan Zivkovic (bojan@mips.com)
30  *
31  * IIR filter optimized for MIPS floating-point architecture
32  *
33  * This file is part of FFmpeg.
34  *
35  * FFmpeg is free software; you can redistribute it and/or
36  * modify it under the terms of the GNU Lesser General Public
37  * License as published by the Free Software Foundation; either
38  * version 2.1 of the License, or (at your option) any later version.
39  *
40  * FFmpeg is distributed in the hope that it will be useful,
41  * but WITHOUT ANY WARRANTY; without even the implied warranty of
42  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
43  * Lesser General Public License for more details.
44  *
45  * You should have received a copy of the GNU Lesser General Public
46  * License along with FFmpeg; if not, write to the Free Software
47  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
48  */
49 
50  /**
51  * @file
52  * Reference: libavcodec/iirfilter.c
53  */
54 
55 #include "config.h"
56 #include "libavcodec/iirfilter.h"
57 
58 #if HAVE_INLINE_ASM
59 #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
60 typedef struct FFIIRFilterCoeffs {
61  int order;
62  float gain;
63  int *cx;
64  float *cy;
66 
67 typedef struct FFIIRFilterState {
68  float x[1];
70 
71 static void iir_filter_flt_mips(const struct FFIIRFilterCoeffs *c,
72  struct FFIIRFilterState *s, int size,
73  const float *src, ptrdiff_t sstep, float *dst, ptrdiff_t dstep)
74 {
75  if (c->order == 2) {
76  int i;
77  const float *src0 = src;
78  float *dst0 = dst;
79  for (i = 0; i < size; i++) {
80  float in = *src0 * c->gain + s->x[0] * c->cy[0] + s->x[1] * c->cy[1];
81  *dst0 = s->x[0] + in + s->x[1] * c->cx[1];
82  s->x[0] = s->x[1];
83  s->x[1] = in;
84  src0 += sstep;
85  dst0 += dstep;
86  }
87  } else if (c->order == 4) {
88  int i;
89  const float *src0 = src;
90  float *dst0 = dst;
91  float four = 4.0;
92  float six = 6.0;
93  for (i = 0; i < size; i += 4) {
94  float in1, in2, in3, in4;
95  float res1, res2, res3, res4;
96  float *x = s->x;
97  float *cy = c->cy;
98  float gain = c->gain;
99  float src0_0 = src0[0 ];
100  float src0_1 = src0[sstep ];
101  float src0_2 = src0[2*sstep];
102  float src0_3 = src0[3*sstep];
103 
104  __asm__ volatile (
105  "lwc1 $f0, 0(%[cy]) \n\t"
106  "lwc1 $f4, 0(%[x]) \n\t"
107  "lwc1 $f5, 4(%[x]) \n\t"
108  "lwc1 $f6, 8(%[x]) \n\t"
109  "lwc1 $f7, 12(%[x]) \n\t"
110  "mul.s %[in1], %[src0_0], %[gain] \n\t"
111  "mul.s %[in2], %[src0_1], %[gain] \n\t"
112  "mul.s %[in3], %[src0_2], %[gain] \n\t"
113  "mul.s %[in4], %[src0_3], %[gain] \n\t"
114  "lwc1 $f1, 4(%[cy]) \n\t"
115  "madd.s %[in1], %[in1], $f0, $f4 \n\t"
116  "madd.s %[in2], %[in2], $f0, $f5 \n\t"
117  "madd.s %[in3], %[in3], $f0, $f6 \n\t"
118  "madd.s %[in4], %[in4], $f0, $f7 \n\t"
119  "lwc1 $f2, 8(%[cy]) \n\t"
120  "madd.s %[in1], %[in1], $f1, $f5 \n\t"
121  "madd.s %[in2], %[in2], $f1, $f6 \n\t"
122  "madd.s %[in3], %[in3], $f1, $f7 \n\t"
123  "lwc1 $f3, 12(%[cy]) \n\t"
124  "add.s $f8, $f5, $f7 \n\t"
125  "madd.s %[in1], %[in1], $f2, $f6 \n\t"
126  "madd.s %[in2], %[in2], $f2, $f7 \n\t"
127  "mul.s $f9, $f6, %[six] \n\t"
128  "mul.s $f10, $f7, %[six] \n\t"
129  "madd.s %[in1], %[in1], $f3, $f7 \n\t"
130  "madd.s %[in2], %[in2], $f3, %[in1] \n\t"
131  "madd.s %[in3], %[in3], $f2, %[in1] \n\t"
132  "madd.s %[in4], %[in4], $f1, %[in1] \n\t"
133  "add.s %[res1], $f4, %[in1] \n\t"
134  "swc1 %[in1], 0(%[x]) \n\t"
135  "add.s $f0, $f6, %[in1] \n\t"
136  "madd.s %[in3], %[in3], $f3, %[in2] \n\t"
137  "madd.s %[in4], %[in4], $f2, %[in2] \n\t"
138  "add.s %[res2], $f5, %[in2] \n\t"
139  "madd.s %[res1], %[res1], $f8, %[four] \n\t"
140  "add.s $f8, $f7, %[in2] \n\t"
141  "swc1 %[in2], 4(%[x]) \n\t"
142  "madd.s %[in4], %[in4], $f3, %[in3] \n\t"
143  "add.s %[res3], $f6, %[in3] \n\t"
144  "add.s %[res1], %[res1], $f9 \n\t"
145  "madd.s %[res2], %[res2], $f0, %[four] \n\t"
146  "swc1 %[in3], 8(%[x]) \n\t"
147  "add.s %[res4], $f7, %[in4] \n\t"
148  "madd.s %[res3], %[res3], $f8, %[four] \n\t"
149  "swc1 %[in4], 12(%[x]) \n\t"
150  "add.s %[res2], %[res2], $f10 \n\t"
151  "add.s $f8, %[in1], %[in3] \n\t"
152  "madd.s %[res3], %[res3], %[in1], %[six] \n\t"
153  "madd.s %[res4], %[res4], $f8, %[four] \n\t"
154  "madd.s %[res4], %[res4], %[in2], %[six] \n\t"
155 
156  : [in1]"=&f"(in1), [in2]"=&f"(in2),
157  [in3]"=&f"(in3), [in4]"=&f"(in4),
158  [res1]"=&f"(res1), [res2]"=&f"(res2),
159  [res3]"=&f"(res3), [res4]"=&f"(res4)
160  : [src0_0]"f"(src0_0), [src0_1]"f"(src0_1),
161  [src0_2]"f"(src0_2), [src0_3]"f"(src0_3),
162  [gain]"f"(gain), [x]"r"(x), [cy]"r"(cy),
163  [four]"f"(four), [six]"f"(six)
164  : "$f0", "$f1", "$f2", "$f3",
165  "$f4", "$f5", "$f6", "$f7",
166  "$f8", "$f9", "$f10",
167  "memory"
168  );
169 
170  dst0[0 ] = res1;
171  dst0[sstep ] = res2;
172  dst0[2*sstep] = res3;
173  dst0[3*sstep] = res4;
174 
175  src0 += 4*sstep;
176  dst0 += 4*dstep;
177  }
178  } else {
179  int i;
180  const float *src0 = src;
181  float *dst0 = dst;
182  for (i = 0; i < size; i++) {
183  int j;
184  float in, res;
185  in = *src0 * c->gain;
186  for(j = 0; j < c->order; j++)
187  in += c->cy[j] * s->x[j];
188  res = s->x[0] + in + s->x[c->order >> 1] * c->cx[c->order >> 1];
189  for(j = 1; j < c->order >> 1; j++)
190  res += (s->x[j] + s->x[c->order - j]) * c->cx[j];
191  for(j = 0; j < c->order - 1; j++)
192  s->x[j] = s->x[j + 1];
193  *dst0 = res;
194  s->x[c->order - 1] = in;
195  src0 += sstep;
196  dst0 += dstep;
197  }
198  }
199 }
200 #endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */
201 #endif /* HAVE_INLINE_ASM */
202 
204 #if HAVE_INLINE_ASM
205 #if !HAVE_MIPS32R6 && !HAVE_MIPS64R6
206  f->filter_flt = iir_filter_flt_mips;
207 #endif /* !HAVE_MIPS32R6 && !HAVE_MIPS64R6 */
208 #endif /* HAVE_INLINE_ASM */
209 }
ff_iir_filter_init_mips
void ff_iir_filter_init_mips(FFIIRFilterContext *f)
Definition: iirfilter_mips.c:203
FFIIRFilterState
IIR filter state.
Definition: iirfilter.c:47
iirfilter.h
FFIIRFilterCoeffs::gain
float gain
Definition: iirfilter.c:39
FFIIRFilterContext
Definition: iirfilter.h:51
FFIIRFilterCoeffs::cx
int * cx
Definition: iirfilter.c:40
FFIIRFilterCoeffs
IIR filter global parameters.
Definition: iirfilter.c:37
s
#define s(width, name)
Definition: cbs_vp9.c:256
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
f
f
Definition: af_crystalizer.c:122
size
int size
Definition: twinvq_data.h:10344
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:269
FFIIRFilterCoeffs::cy
float * cy
Definition: iirfilter.c:41
FFIIRFilterState::x
float x[1]
Definition: iirfilter.c:48
__asm__
__asm__(".macro parse_r var r\n\t" "\\var = -1\n\t" _IFC_REG(0) _IFC_REG(1) _IFC_REG(2) _IFC_REG(3) _IFC_REG(4) _IFC_REG(5) _IFC_REG(6) _IFC_REG(7) _IFC_REG(8) _IFC_REG(9) _IFC_REG(10) _IFC_REG(11) _IFC_REG(12) _IFC_REG(13) _IFC_REG(14) _IFC_REG(15) _IFC_REG(16) _IFC_REG(17) _IFC_REG(18) _IFC_REG(19) _IFC_REG(20) _IFC_REG(21) _IFC_REG(22) _IFC_REG(23) _IFC_REG(24) _IFC_REG(25) _IFC_REG(26) _IFC_REG(27) _IFC_REG(28) _IFC_REG(29) _IFC_REG(30) _IFC_REG(31) ".iflt \\var\n\t" ".error \"Unable to parse register name \\r\"\n\t" ".endif\n\t" ".endm")
FFIIRFilterCoeffs::order
int order
Definition: iirfilter.c:38
src0
const pixel *const src0
Definition: h264pred_template.c:420
src
INIT_CLIP pixel * src
Definition: h264pred_template.c:418