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:  Nedeljko Babic (nbabic@mips.com)
30  *
31  * This file is part of FFmpeg.
32  *
33  * FFmpeg is free software; you can redistribute it and/or
34  * modify it under the terms of the GNU Lesser General Public
35  * License as published by the Free Software Foundation; either
36  * version 2.1 of the License, or (at your option) any later version.
37  *
38  * FFmpeg is distributed in the hope that it will be useful,
39  * but WITHOUT ANY WARRANTY; without even the implied warranty of
40  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
41  * Lesser General Public License for more details.
42  *
43  * You should have received a copy of the GNU Lesser General Public
44  * License along with FFmpeg; if not, write to the Free Software
45  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
46  */
47 
48 #ifndef AVUTIL_FIXED_DSP_H
49 #define AVUTIL_FIXED_DSP_H
50 
51 #include <stdint.h>
52 #include "attributes.h"
53 #include "common.h"
54 #include "libavcodec/mathops.h"
55 
56 typedef struct AVFixedDSPContext {
57     /* Assume len is a multiple of 16, and arrays are 32-byte aligned */
58     /* Results of multiplications are scaled down by 31 bit (and rounded) if not
59      * stated otherwise */
60 
61     /**
62      * Overlap/add with window function.
63      * Result is scaled down by "bits" bits.
64      * Used primarily by MDCT-based audio codecs.
65      * Source and destination vectors must overlap exactly or not at all.
66      *
67      * @param dst  result vector
68      *             constraints: 16-byte aligned
69      * @param src0 first source vector
70      *             constraints: 16-byte aligned
71      * @param src1 second source vector
72      *             constraints: 16-byte aligned
73      * @param win  half-window vector
74      *             constraints: 16-byte aligned
75      * @param len  length of vector
76      *             constraints: multiple of 4
77      * @param bits scaling parameter
78      *
79      */
80     void (*vector_fmul_window_scaled)(int16_t *dst, const int32_t *src0, const int32_t *src1, const int32_t *win, int len, uint8_t bits);
81 
82     /**
83      * Overlap/add with window function.
84      * Used primarily by MDCT-based audio codecs.
85      * Source and destination vectors must overlap exactly or not at all.
86      *
87      * @param dst  result vector
88      *             constraints: 32-byte aligned
89      * @param src0 first source vector
90      *             constraints: 16-byte aligned
91      * @param src1 second source vector
92      *             constraints: 16-byte aligned
93      * @param win  half-window vector
94      *             constraints: 16-byte aligned
95      * @param len  length of vector
96      *             constraints: multiple of 4
97      */
98     void (*vector_fmul_window)(int32_t *dst, const int32_t *src0, const int32_t *src1, const int32_t *win, int len);
99 
100     /**
101      * Fixed-point multiplication that calculates the entry wise product of two
102      * vectors of integers and stores the result in a vector of integers.
103      *
104      * @param dst  output vector
105      *             constraints: 32-byte aligned
106      * @param src0 first input vector
107      *             constraints: 32-byte aligned
108      * @param src1 second input vector
109      *             constraints: 32-byte aligned
110      * @param len  number of elements in the input
111      *             constraints: multiple of 16
112      */
113     void (*vector_fmul)(int *dst, const int *src0, const int *src1,
114                         int len);
115 
116     void (*vector_fmul_reverse)(int *dst, const int *src0, const int *src1, int len);
117     /**
118      * Calculate the entry wise product of two vectors of integers, add a third vector of
119      * integers and store the result in a vector of integers.
120      *
121      * @param dst  output vector
122      *             constraints: 32-byte aligned
123      * @param src0 first input vector
124      *             constraints: 32-byte aligned
125      * @param src1 second input vector
126      *             constraints: 32-byte aligned
127      * @param src2 third input vector
128      *             constraints: 32-byte aligned
129      * @param len  number of elements in the input
130      *             constraints: multiple of 16
131      */
132     void (*vector_fmul_add)(int *dst, const int *src0, const int *src1,
133                             const int *src2, int len);
134 
135     /**
136      * Calculate the scalar product of two vectors of integers.
137      *
138      * @param v1  first vector, 16-byte aligned
139      * @param v2  second vector, 16-byte aligned
140      * @param len length of vectors, multiple of 4
141      *
142      * @return sum of elementwise products
143      */
144     int (*scalarproduct_fixed)(const int *v1, const int *v2, int len);
145 
146     /**
147      * Calculate the sum and difference of two vectors of integers.
148      *
149      * @param v1  first input vector, sum output, 16-byte aligned
150      * @param v2  second input vector, difference output, 16-byte aligned
151      * @param len length of vectors, multiple of 4
152      */
153     void (*butterflies_fixed)(int *av_restrict v1, int *av_restrict v2, int len);
154 } AVFixedDSPContext;
155 
156 /**
157  * Allocate and initialize a fixed DSP context.
158  * note: should be freed with a av_free call when no longer needed.
159  *
160  * @param strict  setting to non-zero avoids using functions which may not be IEEE-754 compliant
161  */
162 AVFixedDSPContext * avpriv_alloc_fixed_dsp(int strict);
163 
164 void ff_fixed_dsp_init_x86(AVFixedDSPContext *fdsp);
165 
166 /**
167  * Calculate the square root
168  *
169  * @param x    input fixed point number
170  *
171  * @param bits format of fixed point number (32 - bits).bits
172  *
173  * note: input is normalized to (0, 1) fixed point value
174  */
175 
fixed_sqrt(int x,int bits)176 static av_always_inline int fixed_sqrt(int x, int bits)
177 {
178     int retval, bit_mask, guess, square, i;
179     int64_t accu;
180     int shift1 = 30 - bits;
181     int shift2 = bits - 15;
182 
183     if (shift1 > 0) retval = ff_sqrt(x << shift1);
184     else retval = ff_sqrt(x >> -shift1);
185 
186     if (shift2 > 0) {
187         retval = retval << shift2;
188         bit_mask = (1 << (shift2 - 1));
189 
190         for (i=0; i<shift2; i++){
191             guess = retval + bit_mask;
192             accu = (int64_t)guess * guess;
193             square = (int)((accu + bit_mask) >> bits);
194             if (x >= square)
195                 retval += bit_mask;
196             bit_mask >>= 1;
197         }
198 
199     }
200     else retval >>= (-shift2);
201 
202     return retval;
203 }
204 
205 #endif /* AVUTIL_FIXED_DSP_H */
206