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 librararies 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: jidctred.c /main/2 1996/05/09 03:51:41 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 inverse-DCT routines that produce reduced-size output:
32 * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block.
34 * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M)
35 * algorithm used in jidctint.c. We simply replace each 8-to-8 1-D IDCT step
36 * with an 8-to-4 step that produces the four averages of two adjacent outputs
37 * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output).
38 * These steps were derived by computing the corresponding values at the end
39 * of the normal LL&M code, then simplifying as much as possible.
41 * 1x1 is trivial: just take the DC coefficient divided by 8.
43 * See jidctint.c for additional comments.
46 #define JPEG_INTERNALS
49 #include "jdct.h" /* Private declarations for DCT subsystem */
51 #ifdef IDCT_SCALING_SUPPORTED
55 * This module is specialized to the case DCTSIZE = 8.
59 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
63 /* Scaling is the same as in jidctint.c. */
65 #if BITS_IN_JSAMPLE == 8
70 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
73 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
74 * causing a lot of useless floating-point operations at run time.
75 * To get around this we use the following pre-calculated constants.
76 * If you change CONST_BITS you may want to add appropriate values.
77 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
81 #define FIX_0_211164243 ((INT32) 1730) /* FIX(0.211164243) */
82 #define FIX_0_509795579 ((INT32) 4176) /* FIX(0.509795579) */
83 #define FIX_0_601344887 ((INT32) 4926) /* FIX(0.601344887) */
84 #define FIX_0_720959822 ((INT32) 5906) /* FIX(0.720959822) */
85 #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
86 #define FIX_0_850430095 ((INT32) 6967) /* FIX(0.850430095) */
87 #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
88 #define FIX_1_061594337 ((INT32) 8697) /* FIX(1.061594337) */
89 #define FIX_1_272758580 ((INT32) 10426) /* FIX(1.272758580) */
90 #define FIX_1_451774981 ((INT32) 11893) /* FIX(1.451774981) */
91 #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
92 #define FIX_2_172734803 ((INT32) 17799) /* FIX(2.172734803) */
93 #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
94 #define FIX_3_624509785 ((INT32) 29692) /* FIX(3.624509785) */
96 #define FIX_0_211164243 FIX(0.211164243)
97 #define FIX_0_509795579 FIX(0.509795579)
98 #define FIX_0_601344887 FIX(0.601344887)
99 #define FIX_0_720959822 FIX(0.720959822)
100 #define FIX_0_765366865 FIX(0.765366865)
101 #define FIX_0_850430095 FIX(0.850430095)
102 #define FIX_0_899976223 FIX(0.899976223)
103 #define FIX_1_061594337 FIX(1.061594337)
104 #define FIX_1_272758580 FIX(1.272758580)
105 #define FIX_1_451774981 FIX(1.451774981)
106 #define FIX_1_847759065 FIX(1.847759065)
107 #define FIX_2_172734803 FIX(2.172734803)
108 #define FIX_2_562915447 FIX(2.562915447)
109 #define FIX_3_624509785 FIX(3.624509785)
113 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
114 * For 8-bit samples with the recommended scaling, all the variable
115 * and constant values involved are no more than 16 bits wide, so a
116 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
117 * For 12-bit samples, a full 32-bit multiplication will be needed.
120 #if BITS_IN_JSAMPLE == 8
121 #define MULTIPLY(var,const) MULTIPLY16C16(var,const)
123 #define MULTIPLY(var,const) ((var) * (const))
127 /* Dequantize a coefficient by multiplying it by the multiplier-table
128 * entry; produce an int result. In this module, both inputs and result
129 * are 16 bits or less, so either int or short multiply will work.
132 #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))
136 * Perform dequantization and inverse DCT on one block of coefficients,
137 * producing a reduced-size 4x4 output block.
141 jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
143 JSAMPARRAY output_buf, JDIMENSION output_col)
145 INT32 tmp0, tmp2, tmp10, tmp12;
146 INT32 z1, z2, z3, z4;
148 ISLOW_MULT_TYPE * quantptr;
151 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
153 int workspace[DCTSIZE*4]; /* buffers data between passes */
156 /* Pass 1: process columns from input, store into work array. */
159 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
161 for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
162 /* Don't bother to process column 4, because second pass won't use it */
163 if (ctr == DCTSIZE-4)
165 if ((inptr[DCTSIZE*1] | inptr[DCTSIZE*2] | inptr[DCTSIZE*3] |
166 inptr[DCTSIZE*5] | inptr[DCTSIZE*6] | inptr[DCTSIZE*7]) == 0) {
167 /* AC terms all zero; we need not examine term 4 for 4x4 output */
168 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
170 wsptr[DCTSIZE*0] = dcval;
171 wsptr[DCTSIZE*1] = dcval;
172 wsptr[DCTSIZE*2] = dcval;
173 wsptr[DCTSIZE*3] = dcval;
180 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
181 tmp0 <<= (CONST_BITS+1);
183 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
184 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
186 tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865);
193 z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
194 z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
195 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
196 z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
198 tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
199 + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
200 + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
201 + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
203 tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
204 + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
205 + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
206 + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
208 /* Final output stage */
210 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1);
211 wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1);
212 wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1);
213 wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1);
216 /* Pass 2: process 4 rows from work array, store into output array. */
219 for (ctr = 0; ctr < 4; ctr++) {
220 outptr = output_buf[ctr] + output_col;
221 /* It's not clear whether a zero row test is worthwhile here ... */
223 #ifndef NO_ZERO_ROW_TEST
224 if ((wsptr[1] | wsptr[2] | wsptr[3] | wsptr[5] | wsptr[6] |
226 /* AC terms all zero */
227 JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
235 wsptr += DCTSIZE; /* advance pointer to next row */
242 tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1);
244 tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065)
245 + MULTIPLY((INT32) wsptr[6], - FIX_0_765366865);
252 z1 = (INT32) wsptr[7];
253 z2 = (INT32) wsptr[5];
254 z3 = (INT32) wsptr[3];
255 z4 = (INT32) wsptr[1];
257 tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
258 + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
259 + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
260 + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
262 tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
263 + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
264 + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
265 + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
267 /* Final output stage */
269 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2,
270 CONST_BITS+PASS1_BITS+3+1)
272 outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2,
273 CONST_BITS+PASS1_BITS+3+1)
275 outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0,
276 CONST_BITS+PASS1_BITS+3+1)
278 outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0,
279 CONST_BITS+PASS1_BITS+3+1)
282 wsptr += DCTSIZE; /* advance pointer to next row */
288 * Perform dequantization and inverse DCT on one block of coefficients,
289 * producing a reduced-size 2x2 output block.
293 jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
295 JSAMPARRAY output_buf, JDIMENSION output_col)
297 INT32 tmp0, tmp10, z1;
299 ISLOW_MULT_TYPE * quantptr;
302 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
304 int workspace[DCTSIZE*2]; /* buffers data between passes */
307 /* Pass 1: process columns from input, store into work array. */
310 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
312 for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
313 /* Don't bother to process columns 2,4,6 */
314 if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6)
316 if ((inptr[DCTSIZE*1] | inptr[DCTSIZE*3] |
317 inptr[DCTSIZE*5] | inptr[DCTSIZE*7]) == 0) {
318 /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */
319 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
321 wsptr[DCTSIZE*0] = dcval;
322 wsptr[DCTSIZE*1] = dcval;
329 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
330 tmp10 = z1 << (CONST_BITS+2);
334 z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
335 tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */
336 z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
337 tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */
338 z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
339 tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */
340 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
341 tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
343 /* Final output stage */
345 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2);
346 wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2);
349 /* Pass 2: process 2 rows from work array, store into output array. */
352 for (ctr = 0; ctr < 2; ctr++) {
353 outptr = output_buf[ctr] + output_col;
354 /* It's not clear whether a zero row test is worthwhile here ... */
356 #ifndef NO_ZERO_ROW_TEST
357 if ((wsptr[1] | wsptr[3] | wsptr[5] | wsptr[7]) == 0) {
358 /* AC terms all zero */
359 JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
365 wsptr += DCTSIZE; /* advance pointer to next row */
372 tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2);
376 tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */
377 + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */
378 + MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */
379 + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
381 /* Final output stage */
383 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0,
384 CONST_BITS+PASS1_BITS+3+2)
386 outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0,
387 CONST_BITS+PASS1_BITS+3+2)
390 wsptr += DCTSIZE; /* advance pointer to next row */
396 * Perform dequantization and inverse DCT on one block of coefficients,
397 * producing a reduced-size 1x1 output block.
401 jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
403 JSAMPARRAY output_buf, JDIMENSION output_col)
406 ISLOW_MULT_TYPE * quantptr;
407 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
410 /* We hardly need an inverse DCT routine for this: just take the
411 * average pixel value, which is one-eighth of the DC coefficient.
413 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
414 dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
415 dcval = (int) DESCALE((INT32) dcval, 3);
417 output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
420 #endif /* IDCT_SCALING_SUPPORTED */