1 | /* |
2 | * reserved comment block |
3 | * DO NOT REMOVE OR ALTER! |
4 | */ |
5 | /* |
6 | * jidctred.c |
7 | * |
8 | * Copyright (C) 1994-1998, Thomas G. Lane. |
9 | * This file is part of the Independent JPEG Group's software. |
10 | * For conditions of distribution and use, see the accompanying README file. |
11 | * |
12 | * This file contains inverse-DCT routines that produce reduced-size output: |
13 | * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block. |
14 | * |
15 | * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M) |
16 | * algorithm used in jidctint.c. We simply replace each 8-to-8 1-D IDCT step |
17 | * with an 8-to-4 step that produces the four averages of two adjacent outputs |
18 | * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output). |
19 | * These steps were derived by computing the corresponding values at the end |
20 | * of the normal LL&M code, then simplifying as much as possible. |
21 | * |
22 | * 1x1 is trivial: just take the DC coefficient divided by 8. |
23 | * |
24 | * See jidctint.c for additional comments. |
25 | */ |
26 | |
27 | #define JPEG_INTERNALS |
28 | #include "jinclude.h" |
29 | #include "jpeglib.h" |
30 | #include "jdct.h" /* Private declarations for DCT subsystem */ |
31 | |
32 | #ifdef IDCT_SCALING_SUPPORTED |
33 | |
34 | |
35 | /* |
36 | * This module is specialized to the case DCTSIZE = 8. |
37 | */ |
38 | |
39 | #if DCTSIZE != 8 |
40 | Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ |
41 | #endif |
42 | |
43 | |
44 | /* Scaling is the same as in jidctint.c. */ |
45 | |
46 | #if BITS_IN_JSAMPLE == 8 |
47 | #define CONST_BITS 13 |
48 | #define PASS1_BITS 2 |
49 | #else |
50 | #define CONST_BITS 13 |
51 | #define PASS1_BITS 1 /* lose a little precision to avoid overflow */ |
52 | #endif |
53 | |
54 | /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus |
55 | * causing a lot of useless floating-point operations at run time. |
56 | * To get around this we use the following pre-calculated constants. |
57 | * If you change CONST_BITS you may want to add appropriate values. |
58 | * (With a reasonable C compiler, you can just rely on the FIX() macro...) |
59 | */ |
60 | |
61 | #if CONST_BITS == 13 |
62 | #define FIX_0_211164243 ((INT32) 1730) /* FIX(0.211164243) */ |
63 | #define FIX_0_509795579 ((INT32) 4176) /* FIX(0.509795579) */ |
64 | #define FIX_0_601344887 ((INT32) 4926) /* FIX(0.601344887) */ |
65 | #define FIX_0_720959822 ((INT32) 5906) /* FIX(0.720959822) */ |
66 | #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */ |
67 | #define FIX_0_850430095 ((INT32) 6967) /* FIX(0.850430095) */ |
68 | #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */ |
69 | #define FIX_1_061594337 ((INT32) 8697) /* FIX(1.061594337) */ |
70 | #define FIX_1_272758580 ((INT32) 10426) /* FIX(1.272758580) */ |
71 | #define FIX_1_451774981 ((INT32) 11893) /* FIX(1.451774981) */ |
72 | #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */ |
73 | #define FIX_2_172734803 ((INT32) 17799) /* FIX(2.172734803) */ |
74 | #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */ |
75 | #define FIX_3_624509785 ((INT32) 29692) /* FIX(3.624509785) */ |
76 | #else |
77 | #define FIX_0_211164243 FIX(0.211164243) |
78 | #define FIX_0_509795579 FIX(0.509795579) |
79 | #define FIX_0_601344887 FIX(0.601344887) |
80 | #define FIX_0_720959822 FIX(0.720959822) |
81 | #define FIX_0_765366865 FIX(0.765366865) |
82 | #define FIX_0_850430095 FIX(0.850430095) |
83 | #define FIX_0_899976223 FIX(0.899976223) |
84 | #define FIX_1_061594337 FIX(1.061594337) |
85 | #define FIX_1_272758580 FIX(1.272758580) |
86 | #define FIX_1_451774981 FIX(1.451774981) |
87 | #define FIX_1_847759065 FIX(1.847759065) |
88 | #define FIX_2_172734803 FIX(2.172734803) |
89 | #define FIX_2_562915447 FIX(2.562915447) |
90 | #define FIX_3_624509785 FIX(3.624509785) |
91 | #endif |
92 | |
93 | |
94 | /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result. |
95 | * For 8-bit samples with the recommended scaling, all the variable |
96 | * and constant values involved are no more than 16 bits wide, so a |
97 | * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. |
98 | * For 12-bit samples, a full 32-bit multiplication will be needed. |
99 | */ |
100 | |
101 | #if BITS_IN_JSAMPLE == 8 |
102 | #define MULTIPLY(var,const) MULTIPLY16C16(var,const) |
103 | #else |
104 | #define MULTIPLY(var,const) ((var) * (const)) |
105 | #endif |
106 | |
107 | |
108 | /* Dequantize a coefficient by multiplying it by the multiplier-table |
109 | * entry; produce an int result. In this module, both inputs and result |
110 | * are 16 bits or less, so either int or short multiply will work. |
111 | */ |
112 | |
113 | #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval)) |
114 | |
115 | |
116 | /* |
117 | * Perform dequantization and inverse DCT on one block of coefficients, |
118 | * producing a reduced-size 4x4 output block. |
119 | */ |
120 | |
121 | GLOBAL(void) |
122 | jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
123 | JCOEFPTR coef_block, |
124 | JSAMPARRAY output_buf, JDIMENSION output_col) |
125 | { |
126 | INT32 tmp0, tmp2, tmp10, tmp12; |
127 | INT32 z1, z2, z3, z4; |
128 | JCOEFPTR inptr; |
129 | ISLOW_MULT_TYPE * quantptr; |
130 | int * wsptr; |
131 | JSAMPROW outptr; |
132 | JSAMPLE *range_limit = IDCT_range_limit(cinfo); |
133 | int ctr; |
134 | int workspace[DCTSIZE*4]; /* buffers data between passes */ |
135 | SHIFT_TEMPS |
136 | |
137 | /* Pass 1: process columns from input, store into work array. */ |
138 | |
139 | inptr = coef_block; |
140 | quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; |
141 | wsptr = workspace; |
142 | for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { |
143 | /* Don't bother to process column 4, because second pass won't use it */ |
144 | if (ctr == DCTSIZE-4) |
145 | continue; |
146 | if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 && |
147 | inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 && |
148 | inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) { |
149 | /* AC terms all zero; we need not examine term 4 for 4x4 output */ |
150 | int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS; |
151 | |
152 | wsptr[DCTSIZE*0] = dcval; |
153 | wsptr[DCTSIZE*1] = dcval; |
154 | wsptr[DCTSIZE*2] = dcval; |
155 | wsptr[DCTSIZE*3] = dcval; |
156 | |
157 | continue; |
158 | } |
159 | |
160 | /* Even part */ |
161 | |
162 | tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); |
163 | tmp0 <<= (CONST_BITS+1); |
164 | |
165 | z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]); |
166 | z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]); |
167 | |
168 | tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865); |
169 | |
170 | tmp10 = tmp0 + tmp2; |
171 | tmp12 = tmp0 - tmp2; |
172 | |
173 | /* Odd part */ |
174 | |
175 | z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); |
176 | z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); |
177 | z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); |
178 | z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); |
179 | |
180 | tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ |
181 | + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ |
182 | + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ |
183 | + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ |
184 | |
185 | tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ |
186 | + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ |
187 | + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ |
188 | + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ |
189 | |
190 | /* Final output stage */ |
191 | |
192 | wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1); |
193 | wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1); |
194 | wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1); |
195 | wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1); |
196 | } |
197 | |
198 | /* Pass 2: process 4 rows from work array, store into output array. */ |
199 | |
200 | wsptr = workspace; |
201 | for (ctr = 0; ctr < 4; ctr++) { |
202 | outptr = output_buf[ctr] + output_col; |
203 | /* It's not clear whether a zero row test is worthwhile here ... */ |
204 | |
205 | #ifndef NO_ZERO_ROW_TEST |
206 | if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && |
207 | wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) { |
208 | /* AC terms all zero */ |
209 | JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3) |
210 | & RANGE_MASK]; |
211 | |
212 | outptr[0] = dcval; |
213 | outptr[1] = dcval; |
214 | outptr[2] = dcval; |
215 | outptr[3] = dcval; |
216 | |
217 | wsptr += DCTSIZE; /* advance pointer to next row */ |
218 | continue; |
219 | } |
220 | #endif |
221 | |
222 | /* Even part */ |
223 | |
224 | tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1); |
225 | |
226 | tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065) |
227 | + MULTIPLY((INT32) wsptr[6], - FIX_0_765366865); |
228 | |
229 | tmp10 = tmp0 + tmp2; |
230 | tmp12 = tmp0 - tmp2; |
231 | |
232 | /* Odd part */ |
233 | |
234 | z1 = (INT32) wsptr[7]; |
235 | z2 = (INT32) wsptr[5]; |
236 | z3 = (INT32) wsptr[3]; |
237 | z4 = (INT32) wsptr[1]; |
238 | |
239 | tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ |
240 | + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ |
241 | + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ |
242 | + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ |
243 | |
244 | tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ |
245 | + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ |
246 | + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ |
247 | + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ |
248 | |
249 | /* Final output stage */ |
250 | |
251 | outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2, |
252 | CONST_BITS+PASS1_BITS+3+1) |
253 | & RANGE_MASK]; |
254 | outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2, |
255 | CONST_BITS+PASS1_BITS+3+1) |
256 | & RANGE_MASK]; |
257 | outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0, |
258 | CONST_BITS+PASS1_BITS+3+1) |
259 | & RANGE_MASK]; |
260 | outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0, |
261 | CONST_BITS+PASS1_BITS+3+1) |
262 | & RANGE_MASK]; |
263 | |
264 | wsptr += DCTSIZE; /* advance pointer to next row */ |
265 | } |
266 | } |
267 | |
268 | |
269 | /* |
270 | * Perform dequantization and inverse DCT on one block of coefficients, |
271 | * producing a reduced-size 2x2 output block. |
272 | */ |
273 | |
274 | GLOBAL(void) |
275 | jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
276 | JCOEFPTR coef_block, |
277 | JSAMPARRAY output_buf, JDIMENSION output_col) |
278 | { |
279 | INT32 tmp0, tmp10, z1; |
280 | JCOEFPTR inptr; |
281 | ISLOW_MULT_TYPE * quantptr; |
282 | int * wsptr; |
283 | JSAMPROW outptr; |
284 | JSAMPLE *range_limit = IDCT_range_limit(cinfo); |
285 | int ctr; |
286 | int workspace[DCTSIZE*2]; /* buffers data between passes */ |
287 | SHIFT_TEMPS |
288 | |
289 | /* Pass 1: process columns from input, store into work array. */ |
290 | |
291 | inptr = coef_block; |
292 | quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; |
293 | wsptr = workspace; |
294 | for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { |
295 | /* Don't bother to process columns 2,4,6 */ |
296 | if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6) |
297 | continue; |
298 | if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*3] == 0 && |
299 | inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) { |
300 | /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */ |
301 | int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS; |
302 | |
303 | wsptr[DCTSIZE*0] = dcval; |
304 | wsptr[DCTSIZE*1] = dcval; |
305 | |
306 | continue; |
307 | } |
308 | |
309 | /* Even part */ |
310 | |
311 | z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); |
312 | tmp10 = z1 << (CONST_BITS+2); |
313 | |
314 | /* Odd part */ |
315 | |
316 | z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); |
317 | tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */ |
318 | z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); |
319 | tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */ |
320 | z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); |
321 | tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */ |
322 | z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); |
323 | tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */ |
324 | |
325 | /* Final output stage */ |
326 | |
327 | wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2); |
328 | wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2); |
329 | } |
330 | |
331 | /* Pass 2: process 2 rows from work array, store into output array. */ |
332 | |
333 | wsptr = workspace; |
334 | for (ctr = 0; ctr < 2; ctr++) { |
335 | outptr = output_buf[ctr] + output_col; |
336 | /* It's not clear whether a zero row test is worthwhile here ... */ |
337 | |
338 | #ifndef NO_ZERO_ROW_TEST |
339 | if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) { |
340 | /* AC terms all zero */ |
341 | JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3) |
342 | & RANGE_MASK]; |
343 | |
344 | outptr[0] = dcval; |
345 | outptr[1] = dcval; |
346 | |
347 | wsptr += DCTSIZE; /* advance pointer to next row */ |
348 | continue; |
349 | } |
350 | #endif |
351 | |
352 | /* Even part */ |
353 | |
354 | tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2); |
355 | |
356 | /* Odd part */ |
357 | |
358 | tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */ |
359 | + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */ |
360 | + MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */ |
361 | + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */ |
362 | |
363 | /* Final output stage */ |
364 | |
365 | outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0, |
366 | CONST_BITS+PASS1_BITS+3+2) |
367 | & RANGE_MASK]; |
368 | outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0, |
369 | CONST_BITS+PASS1_BITS+3+2) |
370 | & RANGE_MASK]; |
371 | |
372 | wsptr += DCTSIZE; /* advance pointer to next row */ |
373 | } |
374 | } |
375 | |
376 | |
377 | /* |
378 | * Perform dequantization and inverse DCT on one block of coefficients, |
379 | * producing a reduced-size 1x1 output block. |
380 | */ |
381 | |
382 | GLOBAL(void) |
383 | jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr, |
384 | JCOEFPTR coef_block, |
385 | JSAMPARRAY output_buf, JDIMENSION output_col) |
386 | { |
387 | int dcval; |
388 | ISLOW_MULT_TYPE * quantptr; |
389 | JSAMPLE *range_limit = IDCT_range_limit(cinfo); |
390 | SHIFT_TEMPS |
391 | |
392 | /* We hardly need an inverse DCT routine for this: just take the |
393 | * average pixel value, which is one-eighth of the DC coefficient. |
394 | */ |
395 | quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; |
396 | dcval = DEQUANTIZE(coef_block[0], quantptr[0]); |
397 | dcval = (int) DESCALE((INT32) dcval, 3); |
398 | |
399 | output_buf[0][output_col] = range_limit[dcval & RANGE_MASK]; |
400 | } |
401 | |
402 | #endif /* IDCT_SCALING_SUPPORTED */ |
403 | |