2 * This file is part of the Independent JPEG Group's software.
4 * The authors make NO WARRANTY or representation, either express or implied,
5 * with respect to this software, its quality, accuracy, merchantability, or
6 * fitness for a particular purpose. This software is provided "AS IS", and
7 * you, its user, assume the entire risk as to its quality and accuracy.
9 * This software is copyright (C) 1994-1996, Thomas G. Lane.
10 * All Rights Reserved except as specified below.
12 * Permission is hereby granted to use, copy, modify, and distribute this
13 * software (or portions thereof) for any purpose, without fee, subject to
15 * (1) If any part of the source code for this software is distributed, then
16 * this README file must be included, with this copyright and no-warranty
17 * notice unaltered; and any additions, deletions, or changes to the original
18 * files must be clearly indicated in accompanying documentation.
19 * (2) If only executable code is distributed, then the accompanying
20 * documentation must state that "this software is based in part on the work
21 * of the Independent JPEG Group".
22 * (3) Permission for use of this software is granted only if the user accepts
23 * full responsibility for any undesirable consequences; the authors accept
24 * NO LIABILITY for damages of any kind.
26 * These conditions apply to any software derived from or based on the IJG
27 * code, not just to the unmodified library. If you use our work, you ought
30 * Permission is NOT granted for the use of any IJG author's name or company
31 * name in advertising or publicity relating to this software or products
32 * derived from it. This software may be referred to only as "the Independent
33 * JPEG Group's software".
35 * We specifically permit and encourage the use of this software as the basis
36 * of commercial products, provided that all warranty or liability claims are
37 * assumed by the product vendor.
39 * This file contains a fast, not so accurate integer implementation of the
40 * forward DCT (Discrete Cosine Transform).
42 * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
43 * on each column. Direct algorithms are also available, but they are
44 * much more complex and seem not to be any faster when reduced to code.
46 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
47 * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in
48 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
49 * JPEG textbook (see REFERENCES section in file README). The following code
50 * is based directly on figure 4-8 in P&M.
51 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
52 * possible to arrange the computation so that many of the multiplies are
53 * simple scalings of the final outputs. These multiplies can then be
54 * folded into the multiplications or divisions by the JPEG quantization
55 * table entries. The AA&N method leaves only 5 multiplies and 29 adds
56 * to be done in the DCT itself.
57 * The primary disadvantage of this method is that with fixed-point math,
58 * accuracy is lost due to imprecise representation of the scaled
59 * quantization values. The smaller the quantization table entry, the less
60 * precise the scaled value, so this implementation does worse with high-
61 * quality-setting files than with low-quality ones.
66 * Independent JPEG Group's fast AAN dct.
71 #include "libavutil/common.h"
76 #define RIGHT_SHIFT(x, n) ((x) >> (n))
79 * This module is specialized to the case DCTSIZE = 8.
83 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
87 /* Scaling decisions are generally the same as in the LL&M algorithm;
88 * see jfdctint.c for more details. However, we choose to descale
89 * (right shift) multiplication products as soon as they are formed,
90 * rather than carrying additional fractional bits into subsequent additions.
91 * This compromises accuracy slightly, but it lets us save a few shifts.
92 * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
93 * everywhere except in the multiplications proper; this saves a good deal
94 * of work on 16-bit-int machines.
96 * Again to save a few shifts, the intermediate results between pass 1 and
97 * pass 2 are not upscaled, but are represented only to integral precision.
99 * A final compromise is to represent the multiplicative constants to only
100 * 8 fractional bits, rather than 13. This saves some shifting work on some
101 * machines, and may also reduce the cost of multiplication (since there
102 * are fewer one-bits in the constants).
108 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
109 * causing a lot of useless floating-point operations at run time.
110 * To get around this we use the following pre-calculated constants.
111 * If you change CONST_BITS you may want to add appropriate values.
112 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
116 #define FIX_0_382683433 ((int32_t) 98) /* FIX(0.382683433) */
117 #define FIX_0_541196100 ((int32_t) 139) /* FIX(0.541196100) */
118 #define FIX_0_707106781 ((int32_t) 181) /* FIX(0.707106781) */
119 #define FIX_1_306562965 ((int32_t) 334) /* FIX(1.306562965) */
121 #define FIX_0_382683433 FIX(0.382683433)
122 #define FIX_0_541196100 FIX(0.541196100)
123 #define FIX_0_707106781 FIX(0.707106781)
124 #define FIX_1_306562965 FIX(1.306562965)
128 /* We can gain a little more speed, with a further compromise in accuracy,
129 * by omitting the addition in a descaling shift. This yields an incorrectly
130 * rounded result half the time...
133 #ifndef USE_ACCURATE_ROUNDING
135 #define DESCALE(x,n) RIGHT_SHIFT(x, n)
139 /* Multiply a int16_t variable by an int32_t constant, and immediately
140 * descale to yield a int16_t result.
143 #define MULTIPLY(var,const) ((int16_t) DESCALE((var) * (const), CONST_BITS))
145 static av_always_inline void row_fdct(int16_t * data){
146 int tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
147 int tmp10, tmp11, tmp12, tmp13;
148 int z1, z2, z3, z4, z5, z11, z13;
152 /* Pass 1: process rows. */
155 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
156 tmp0 = dataptr[0] + dataptr[7];
157 tmp7 = dataptr[0] - dataptr[7];
158 tmp1 = dataptr[1] + dataptr[6];
159 tmp6 = dataptr[1] - dataptr[6];
160 tmp2 = dataptr[2] + dataptr[5];
161 tmp5 = dataptr[2] - dataptr[5];
162 tmp3 = dataptr[3] + dataptr[4];
163 tmp4 = dataptr[3] - dataptr[4];
167 tmp10 = tmp0 + tmp3; /* phase 2 */
172 dataptr[0] = tmp10 + tmp11; /* phase 3 */
173 dataptr[4] = tmp10 - tmp11;
175 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
176 dataptr[2] = tmp13 + z1; /* phase 5 */
177 dataptr[6] = tmp13 - z1;
181 tmp10 = tmp4 + tmp5; /* phase 2 */
185 /* The rotator is modified from fig 4-8 to avoid extra negations. */
186 z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
187 z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
188 z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
189 z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
191 z11 = tmp7 + z3; /* phase 5 */
194 dataptr[5] = z13 + z2; /* phase 6 */
195 dataptr[3] = z13 - z2;
196 dataptr[1] = z11 + z4;
197 dataptr[7] = z11 - z4;
199 dataptr += DCTSIZE; /* advance pointer to next row */
204 * Perform the forward DCT on one block of samples.
208 ff_fdct_ifast (int16_t * data)
210 int tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
211 int tmp10, tmp11, tmp12, tmp13;
212 int z1, z2, z3, z4, z5, z11, z13;
218 /* Pass 2: process columns. */
221 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
222 tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
223 tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
224 tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
225 tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
226 tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
227 tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
228 tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
229 tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
233 tmp10 = tmp0 + tmp3; /* phase 2 */
238 dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
239 dataptr[DCTSIZE*4] = tmp10 - tmp11;
241 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
242 dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
243 dataptr[DCTSIZE*6] = tmp13 - z1;
247 tmp10 = tmp4 + tmp5; /* phase 2 */
251 /* The rotator is modified from fig 4-8 to avoid extra negations. */
252 z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
253 z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
254 z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
255 z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
257 z11 = tmp7 + z3; /* phase 5 */
260 dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */
261 dataptr[DCTSIZE*3] = z13 - z2;
262 dataptr[DCTSIZE*1] = z11 + z4;
263 dataptr[DCTSIZE*7] = z11 - z4;
265 dataptr++; /* advance pointer to next column */
270 * Perform the forward 2-4-8 DCT on one block of samples.
274 ff_fdct_ifast248 (int16_t * data)
276 int tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
277 int tmp10, tmp11, tmp12, tmp13;
284 /* Pass 2: process columns. */
287 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
288 tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*1];
289 tmp1 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];
290 tmp2 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5];
291 tmp3 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7];
292 tmp4 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*1];
293 tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];
294 tmp6 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5];
295 tmp7 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7];
304 dataptr[DCTSIZE*0] = tmp10 + tmp11;
305 dataptr[DCTSIZE*4] = tmp10 - tmp11;
307 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781);
308 dataptr[DCTSIZE*2] = tmp13 + z1;
309 dataptr[DCTSIZE*6] = tmp13 - z1;
316 dataptr[DCTSIZE*1] = tmp10 + tmp11;
317 dataptr[DCTSIZE*5] = tmp10 - tmp11;
319 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781);
320 dataptr[DCTSIZE*3] = tmp13 + z1;
321 dataptr[DCTSIZE*7] = tmp13 - z1;
323 dataptr++; /* advance pointer to next column */
331 #undef FIX_0_541196100
332 #undef FIX_1_306562965