2 * CDE - Common Desktop Environment
4 * Copyright (c) 1993-2012, The Open Group. All rights reserved.
6 * These libraries and programs are free software; you can
7 * redistribute them and/or modify them under the terms of the GNU
8 * Lesser General Public License as published by the Free Software
9 * Foundation; either version 2 of the License, or (at your option)
12 * These libraries and programs are distributed in the hope that
13 * they will be useful, but WITHOUT ANY WARRANTY; without even the
14 * implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
15 * PURPOSE. See the GNU Lesser General Public License for more
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with these libraries and programs; if not, write
20 * to the Free Software Foundation, Inc., 51 Franklin Street, Fifth
21 * Floor, Boston, MA 02110-1301 USA
23 /* $XConsortium: jidctflt.c /main/2 1996/05/09 03:50:55 drk $ */
27 * Copyright (C) 1994-1996, Thomas G. Lane.
28 * This file is part of the Independent JPEG Group's software.
29 * For conditions of distribution and use, see the accompanying README file.
31 * This file contains a floating-point implementation of the
32 * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine
33 * must also perform dequantization of the input coefficients.
35 * This implementation should be more accurate than either of the integer
36 * IDCT implementations. However, it may not give the same results on all
37 * machines because of differences in roundoff behavior. Speed will depend
38 * on the hardware's floating point capacity.
40 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
41 * on each row (or vice versa, but it's more convenient to emit a row at
42 * a time). Direct algorithms are also available, but they are much more
43 * complex and seem not to be any faster when reduced to code.
45 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
46 * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in
47 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
48 * JPEG textbook (see REFERENCES section in file README). The following code
49 * is based directly on figure 4-8 in P&M.
50 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
51 * possible to arrange the computation so that many of the multiplies are
52 * simple scalings of the final outputs. These multiplies can then be
53 * folded into the multiplications or divisions by the JPEG quantization
54 * table entries. The AA&N method leaves only 5 multiplies and 29 adds
55 * to be done in the DCT itself.
56 * The primary disadvantage of this method is that with a fixed-point
57 * implementation, accuracy is lost due to imprecise representation of the
58 * scaled quantization values. However, that problem does not arise if
59 * we use floating point arithmetic.
62 #define JPEG_INTERNALS
65 #include "jdct.h" /* Private declarations for DCT subsystem */
67 #ifdef DCT_FLOAT_SUPPORTED
71 * This module is specialized to the case DCTSIZE = 8.
75 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
79 /* Dequantize a coefficient by multiplying it by the multiplier-table
80 * entry; produce a float result.
83 #define DEQUANTIZE(coef,quantval) (((FAST_FLOAT) (coef)) * (quantval))
87 * Perform dequantization and inverse DCT on one block of coefficients.
91 jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
93 JSAMPARRAY output_buf, JDIMENSION output_col)
95 FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
96 FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
97 FAST_FLOAT z5, z10, z11, z12, z13;
99 FLOAT_MULT_TYPE * quantptr;
102 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
104 FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
107 /* Pass 1: process columns from input, store into work array. */
110 quantptr = (FLOAT_MULT_TYPE *) compptr->dct_table;
112 for (ctr = DCTSIZE; ctr > 0; ctr--) {
113 /* Due to quantization, we will usually find that many of the input
114 * coefficients are zero, especially the AC terms. We can exploit this
115 * by short-circuiting the IDCT calculation for any column in which all
116 * the AC terms are zero. In that case each output is equal to the
117 * DC coefficient (with scale factor as needed).
118 * With typical images and quantization tables, half or more of the
119 * column DCT calculations can be simplified this way.
122 if ((inptr[DCTSIZE*1] | inptr[DCTSIZE*2] | inptr[DCTSIZE*3] |
123 inptr[DCTSIZE*4] | inptr[DCTSIZE*5] | inptr[DCTSIZE*6] |
124 inptr[DCTSIZE*7]) == 0) {
125 /* AC terms all zero */
126 FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
128 wsptr[DCTSIZE*0] = dcval;
129 wsptr[DCTSIZE*1] = dcval;
130 wsptr[DCTSIZE*2] = dcval;
131 wsptr[DCTSIZE*3] = dcval;
132 wsptr[DCTSIZE*4] = dcval;
133 wsptr[DCTSIZE*5] = dcval;
134 wsptr[DCTSIZE*6] = dcval;
135 wsptr[DCTSIZE*7] = dcval;
137 inptr++; /* advance pointers to next column */
145 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
146 tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
147 tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
148 tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
150 tmp10 = tmp0 + tmp2; /* phase 3 */
153 tmp13 = tmp1 + tmp3; /* phases 5-3 */
154 tmp12 = (tmp1 - tmp3) * ((FAST_FLOAT) 1.414213562) - tmp13; /* 2*c4 */
156 tmp0 = tmp10 + tmp13; /* phase 2 */
157 tmp3 = tmp10 - tmp13;
158 tmp1 = tmp11 + tmp12;
159 tmp2 = tmp11 - tmp12;
163 tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
164 tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
165 tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
166 tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
168 z13 = tmp6 + tmp5; /* phase 6 */
173 tmp7 = z11 + z13; /* phase 5 */
174 tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
176 z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
177 tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
178 tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
180 tmp6 = tmp12 - tmp7; /* phase 2 */
184 wsptr[DCTSIZE*0] = tmp0 + tmp7;
185 wsptr[DCTSIZE*7] = tmp0 - tmp7;
186 wsptr[DCTSIZE*1] = tmp1 + tmp6;
187 wsptr[DCTSIZE*6] = tmp1 - tmp6;
188 wsptr[DCTSIZE*2] = tmp2 + tmp5;
189 wsptr[DCTSIZE*5] = tmp2 - tmp5;
190 wsptr[DCTSIZE*4] = tmp3 + tmp4;
191 wsptr[DCTSIZE*3] = tmp3 - tmp4;
193 inptr++; /* advance pointers to next column */
198 /* Pass 2: process rows from work array, store into output array. */
199 /* Note that we must descale the results by a factor of 8 == 2**3. */
202 for (ctr = 0; ctr < DCTSIZE; ctr++) {
203 outptr = output_buf[ctr] + output_col;
204 /* Rows of zeroes can be exploited in the same way as we did with columns.
205 * However, the column calculation has created many nonzero AC terms, so
206 * the simplification applies less often (typically 5% to 10% of the time).
207 * And testing floats for zero is relatively expensive, so we don't bother.
212 tmp10 = wsptr[0] + wsptr[4];
213 tmp11 = wsptr[0] - wsptr[4];
215 tmp13 = wsptr[2] + wsptr[6];
216 tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT) 1.414213562) - tmp13;
218 tmp0 = tmp10 + tmp13;
219 tmp3 = tmp10 - tmp13;
220 tmp1 = tmp11 + tmp12;
221 tmp2 = tmp11 - tmp12;
225 z13 = wsptr[5] + wsptr[3];
226 z10 = wsptr[5] - wsptr[3];
227 z11 = wsptr[1] + wsptr[7];
228 z12 = wsptr[1] - wsptr[7];
231 tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562);
233 z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
234 tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
235 tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
241 /* Final output stage: scale down by a factor of 8 and range-limit */
243 outptr[0] = range_limit[(int) DESCALE((INT32) (tmp0 + tmp7), 3)
245 outptr[7] = range_limit[(int) DESCALE((INT32) (tmp0 - tmp7), 3)
247 outptr[1] = range_limit[(int) DESCALE((INT32) (tmp1 + tmp6), 3)
249 outptr[6] = range_limit[(int) DESCALE((INT32) (tmp1 - tmp6), 3)
251 outptr[2] = range_limit[(int) DESCALE((INT32) (tmp2 + tmp5), 3)
253 outptr[5] = range_limit[(int) DESCALE((INT32) (tmp2 - tmp5), 3)
255 outptr[4] = range_limit[(int) DESCALE((INT32) (tmp3 + tmp4), 3)
257 outptr[3] = range_limit[(int) DESCALE((INT32) (tmp3 - tmp4), 3)
260 wsptr += DCTSIZE; /* advance pointer to next row */
264 #endif /* DCT_FLOAT_SUPPORTED */