Mercurial > cgi-bin > hgwebdir.cgi > PR > Applications > VSs > VSs__jpeg_decoder__Proj
view VSs_tinyjpeg/jidctflt.c @ 0:a8af8b3fc99d
initial commit
| author | Nina Engelhardt <nengel@mailbox.tu-berlin.de> |
|---|---|
| date | Thu, 05 Jul 2012 11:35:03 +0200 |
| parents | |
| children |
line source
1 /*
2 * jidctflt.c
3 *
4 * Copyright (C) 1994-1998, Thomas G. Lane.
5 * This file is part of the Independent JPEG Group's software.
6 *
7 * The authors make NO WARRANTY or representation, either express or implied,
8 * with respect to this software, its quality, accuracy, merchantability, or
9 * fitness for a particular purpose. This software is provided "AS IS", and you,
10 * its user, assume the entire risk as to its quality and accuracy.
11 *
12 * This software is copyright (C) 1991-1998, Thomas G. Lane.
13 * All Rights Reserved except as specified below.
14 *
15 * Permission is hereby granted to use, copy, modify, and distribute this
16 * software (or portions thereof) for any purpose, without fee, subject to these
17 * conditions:
18 * (1) If any part of the source code for this software is distributed, then this
19 * README file must be included, with this copyright and no-warranty notice
20 * unaltered; and any additions, deletions, or changes to the original files
21 * must be clearly indicated in accompanying documentation.
22 * (2) If only executable code is distributed, then the accompanying
23 * documentation must state that "this software is based in part on the work of
24 * the Independent JPEG Group".
25 * (3) Permission for use of this software is granted only if the user accepts
26 * full responsibility for any undesirable consequences; the authors accept
27 * NO LIABILITY for damages of any kind.
28 *
29 * These conditions apply to any software derived from or based on the IJG code,
30 * not just to the unmodified library. If you use our work, you ought to
31 * acknowledge us.
32 *
33 * Permission is NOT granted for the use of any IJG author's name or company name
34 * in advertising or publicity relating to this software or products derived from
35 * it. This software may be referred to only as "the Independent JPEG Group's
36 * software".
37 *
38 * We specifically permit and encourage the use of this software as the basis of
39 * commercial products, provided that all warranty or liability claims are
40 * assumed by the product vendor.
41 *
42 *
43 * This file contains a floating-point implementation of the
44 * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine
45 * must also perform dequantization of the input coefficients.
46 *
47 * This implementation should be more accurate than either of the integer
48 * IDCT implementations. However, it may not give the same results on all
49 * machines because of differences in roundoff behavior. Speed will depend
50 * on the hardware's floating point capacity.
51 *
52 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
53 * on each row (or vice versa, but it's more convenient to emit a row at
54 * a time). Direct algorithms are also available, but they are much more
55 * complex and seem not to be any faster when reduced to code.
56 *
57 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
58 * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in
59 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
60 * JPEG textbook (see REFERENCES section in file README). The following code
61 * is based directly on figure 4-8 in P&M.
62 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
63 * possible to arrange the computation so that many of the multiplies are
64 * simple scalings of the final outputs. These multiplies can then be
65 * folded into the multiplications or divisions by the JPEG quantization
66 * table entries. The AA&N method leaves only 5 multiplies and 29 adds
67 * to be done in the DCT itself.
68 * The primary disadvantage of this method is that with a fixed-point
69 * implementation, accuracy is lost due to imprecise representation of the
70 * scaled quantization values. However, that problem does not arise if
71 * we use floating point arithmetic.
72 */
74 #include <stdint.h>
75 #include "tinyjpeg-internal.h"
77 #define FAST_FLOAT float
78 #define DCTSIZE 8
79 #define DCTSIZE2 (DCTSIZE*DCTSIZE)
81 #define DEQUANTIZE(coef,quantval) (((FAST_FLOAT) (coef)) * (quantval))
83 static inline unsigned char descale_and_clamp(int x, int shift)
84 {
85 x += (1UL<<(shift-1));
86 if (x<0)
87 x = (x >> shift) | ((~(0UL)) << (32-(shift)));
88 else
89 x >>= shift;
90 x += 128;
91 if (x>255)
92 return 255;
93 else if (x<0)
94 return 0;
95 else
96 return x;
97 }
99 /*
100 * Perform dequantization and inverse DCT on one block of coefficients.
101 */
102 void tinyjpeg_idct_float (struct component *compptr, uint8_t *output_buf, int stride)
103 {
104 FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
105 FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
106 FAST_FLOAT z5, z10, z11, z12, z13;
107 int16_t *inptr;
108 FAST_FLOAT *quantptr;
109 FAST_FLOAT *wsptr;
110 uint8_t *outptr;
111 int ctr;
112 FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
114 /* Pass 1: process columns from input, store into work array. */
116 inptr = compptr->DCT;
117 quantptr = compptr->Q_table;
118 wsptr = workspace;
119 for (ctr = DCTSIZE; ctr > 0; ctr--) {
120 /* Due to quantization, we will usually find that many of the input
121 * coefficients are zero, especially the AC terms. We can exploit this
122 * by short-circuiting the IDCT calculation for any column in which all
123 * the AC terms are zero. In that case each output is equal to the
124 * DC coefficient (with scale factor as needed).
125 * With typical images and quantization tables, half or more of the
126 * column DCT calculations can be simplified this way.
127 */
129 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
130 inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
131 inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
132 inptr[DCTSIZE*7] == 0) {
133 /* AC terms all zero */
134 FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
136 wsptr[DCTSIZE*0] = dcval;
137 wsptr[DCTSIZE*1] = dcval;
138 wsptr[DCTSIZE*2] = dcval;
139 wsptr[DCTSIZE*3] = dcval;
140 wsptr[DCTSIZE*4] = dcval;
141 wsptr[DCTSIZE*5] = dcval;
142 wsptr[DCTSIZE*6] = dcval;
143 wsptr[DCTSIZE*7] = dcval;
145 inptr++; /* advance pointers to next column */
146 quantptr++;
147 wsptr++;
148 continue;
149 }
151 /* Even part */
153 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
154 tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
155 tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
156 tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
158 tmp10 = tmp0 + tmp2; /* phase 3 */
159 tmp11 = tmp0 - tmp2;
161 tmp13 = tmp1 + tmp3; /* phases 5-3 */
162 tmp12 = (tmp1 - tmp3) * ((FAST_FLOAT) 1.414213562) - tmp13; /* 2*c4 */
164 tmp0 = tmp10 + tmp13; /* phase 2 */
165 tmp3 = tmp10 - tmp13;
166 tmp1 = tmp11 + tmp12;
167 tmp2 = tmp11 - tmp12;
169 /* Odd part */
171 tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
172 tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
173 tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
174 tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
176 z13 = tmp6 + tmp5; /* phase 6 */
177 z10 = tmp6 - tmp5;
178 z11 = tmp4 + tmp7;
179 z12 = tmp4 - tmp7;
181 tmp7 = z11 + z13; /* phase 5 */
182 tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
184 z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
185 tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
186 tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
188 tmp6 = tmp12 - tmp7; /* phase 2 */
189 tmp5 = tmp11 - tmp6;
190 tmp4 = tmp10 + tmp5;
192 wsptr[DCTSIZE*0] = tmp0 + tmp7;
193 wsptr[DCTSIZE*7] = tmp0 - tmp7;
194 wsptr[DCTSIZE*1] = tmp1 + tmp6;
195 wsptr[DCTSIZE*6] = tmp1 - tmp6;
196 wsptr[DCTSIZE*2] = tmp2 + tmp5;
197 wsptr[DCTSIZE*5] = tmp2 - tmp5;
198 wsptr[DCTSIZE*4] = tmp3 + tmp4;
199 wsptr[DCTSIZE*3] = tmp3 - tmp4;
201 inptr++; /* advance pointers to next column */
202 quantptr++;
203 wsptr++;
204 }
206 /* Pass 2: process rows from work array, store into output array. */
207 /* Note that we must descale the results by a factor of 8 == 2**3. */
209 wsptr = workspace;
210 outptr = output_buf;
211 for (ctr = 0; ctr < DCTSIZE; ctr++) {
212 /* Rows of zeroes can be exploited in the same way as we did with columns.
213 * However, the column calculation has created many nonzero AC terms, so
214 * the simplification applies less often (typically 5% to 10% of the time).
215 * And testing floats for zero is relatively expensive, so we don't bother.
216 */
218 /* Even part */
220 tmp10 = wsptr[0] + wsptr[4];
221 tmp11 = wsptr[0] - wsptr[4];
223 tmp13 = wsptr[2] + wsptr[6];
224 tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT) 1.414213562) - tmp13;
226 tmp0 = tmp10 + tmp13;
227 tmp3 = tmp10 - tmp13;
228 tmp1 = tmp11 + tmp12;
229 tmp2 = tmp11 - tmp12;
231 /* Odd part */
233 z13 = wsptr[5] + wsptr[3];
234 z10 = wsptr[5] - wsptr[3];
235 z11 = wsptr[1] + wsptr[7];
236 z12 = wsptr[1] - wsptr[7];
238 tmp7 = z11 + z13;
239 tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562);
241 z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
242 tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
243 tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
245 tmp6 = tmp12 - tmp7;
246 tmp5 = tmp11 - tmp6;
247 tmp4 = tmp10 + tmp5;
249 /* Final output stage: scale down by a factor of 8 and range-limit */
251 outptr[0] = descale_and_clamp((int)(tmp0 + tmp7), 3);
252 outptr[7] = descale_and_clamp((int)(tmp0 - tmp7), 3);
253 outptr[1] = descale_and_clamp((int)(tmp1 + tmp6), 3);
254 outptr[6] = descale_and_clamp((int)(tmp1 - tmp6), 3);
255 outptr[2] = descale_and_clamp((int)(tmp2 + tmp5), 3);
256 outptr[5] = descale_and_clamp((int)(tmp2 - tmp5), 3);
257 outptr[4] = descale_and_clamp((int)(tmp3 + tmp4), 3);
258 outptr[3] = descale_and_clamp((int)(tmp3 - tmp4), 3);
261 wsptr += DCTSIZE; /* advance pointer to next row */
262 outptr += stride;
263 }
264 }