1 /*
2 * jidctint.c
3 *
4 * This file was part of the Independent JPEG Group's software:
5 * Copyright (C) 1991-1998, Thomas G. Lane.
6 * Modification developed 2002-2018 by Guido Vollbeding.
7 * libjpeg-turbo Modifications:
8 * Copyright (C) 2015, 2020, D. R. Commander.
9 * For conditions of distribution and use, see the accompanying README.ijg
10 * file.
11 *
12 * This file contains a slower but more accurate integer implementation of the
13 * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine
14 * must also perform dequantization of the input coefficients.
15 *
16 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
17 * on each row (or vice versa, but it's more convenient to emit a row at
18 * a time). Direct algorithms are also available, but they are much more
19 * complex and seem not to be any faster when reduced to code.
20 *
21 * This implementation is based on an algorithm described in
22 * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
23 * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
24 * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
25 * The primary algorithm described there uses 11 multiplies and 29 adds.
26 * We use their alternate method with 12 multiplies and 32 adds.
27 * The advantage of this method is that no data path contains more than one
28 * multiplication; this allows a very simple and accurate implementation in
29 * scaled fixed-point arithmetic, with a minimal number of shifts.
30 *
31 * We also provide IDCT routines with various output sample block sizes for
32 * direct resolution reduction or enlargement without additional resampling:
33 * NxN (N=1...16) pixels for one 8x8 input DCT block.
34 *
35 * For N<8 we simply take the corresponding low-frequency coefficients of
36 * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block
37 * to yield the downscaled outputs.
38 * This can be seen as direct low-pass downsampling from the DCT domain
39 * point of view rather than the usual spatial domain point of view,
40 * yielding significant computational savings and results at least
41 * as good as common bilinear (averaging) spatial downsampling.
42 *
43 * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as
44 * lower frequencies and higher frequencies assumed to be zero.
45 * It turns out that the computational effort is similar to the 8x8 IDCT
46 * regarding the output size.
47 * Furthermore, the scaling and descaling is the same for all IDCT sizes.
48 *
49 * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases
50 * since there would be too many additional constants to pre-calculate.
51 */
52
53 #define JPEG_INTERNALS
54 #include "jinclude.h"
55 #include "jpeglib.h"
56 #include "jdct.h" /* Private declarations for DCT subsystem */
57
58 #ifdef DCT_ISLOW_SUPPORTED
59
60
61 /*
62 * This module is specialized to the case DCTSIZE = 8.
63 */
64
65 #if DCTSIZE != 8
66 Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
67 #endif
68
69
70 /*
71 * The poop on this scaling stuff is as follows:
72 *
73 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
74 * larger than the true IDCT outputs. The final outputs are therefore
75 * a factor of N larger than desired; since N=8 this can be cured by
76 * a simple right shift at the end of the algorithm. The advantage of
77 * this arrangement is that we save two multiplications per 1-D IDCT,
78 * because the y0 and y4 inputs need not be divided by sqrt(N).
79 *
80 * We have to do addition and subtraction of the integer inputs, which
81 * is no problem, and multiplication by fractional constants, which is
82 * a problem to do in integer arithmetic. We multiply all the constants
83 * by CONST_SCALE and convert them to integer constants (thus retaining
84 * CONST_BITS bits of precision in the constants). After doing a
85 * multiplication we have to divide the product by CONST_SCALE, with proper
86 * rounding, to produce the correct output. This division can be done
87 * cheaply as a right shift of CONST_BITS bits. We postpone shifting
88 * as long as possible so that partial sums can be added together with
89 * full fractional precision.
90 *
91 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
92 * they are represented to better-than-integral precision. These outputs
93 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
94 * with the recommended scaling. (To scale up 12-bit sample data further, an
95 * intermediate JLONG array would be needed.)
96 *
97 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
98 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
99 * shows that the values given below are the most effective.
100 */
101
102 #if BITS_IN_JSAMPLE == 8
103 #define CONST_BITS 13
104 #define PASS1_BITS 2
105 #else
106 #define CONST_BITS 13
107 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
108 #endif
109
110 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
111 * causing a lot of useless floating-point operations at run time.
112 * To get around this we use the following pre-calculated constants.
113 * If you change CONST_BITS you may want to add appropriate values.
114 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
115 */
116
117 #if CONST_BITS == 13
118 #define FIX_0_298631336 ((JLONG)2446) /* FIX(0.298631336) */
119 #define FIX_0_390180644 ((JLONG)3196) /* FIX(0.390180644) */
120 #define FIX_0_541196100 ((JLONG)4433) /* FIX(0.541196100) */
121 #define FIX_0_765366865 ((JLONG)6270) /* FIX(0.765366865) */
122 #define FIX_0_899976223 ((JLONG)7373) /* FIX(0.899976223) */
123 #define FIX_1_175875602 ((JLONG)9633) /* FIX(1.175875602) */
124 #define FIX_1_501321110 ((JLONG)12299) /* FIX(1.501321110) */
125 #define FIX_1_847759065 ((JLONG)15137) /* FIX(1.847759065) */
126 #define FIX_1_961570560 ((JLONG)16069) /* FIX(1.961570560) */
127 #define FIX_2_053119869 ((JLONG)16819) /* FIX(2.053119869) */
128 #define FIX_2_562915447 ((JLONG)20995) /* FIX(2.562915447) */
129 #define FIX_3_072711026 ((JLONG)25172) /* FIX(3.072711026) */
130 #else
131 #define FIX_0_298631336 FIX(0.298631336)
132 #define FIX_0_390180644 FIX(0.390180644)
133 #define FIX_0_541196100 FIX(0.541196100)
134 #define FIX_0_765366865 FIX(0.765366865)
135 #define FIX_0_899976223 FIX(0.899976223)
136 #define FIX_1_175875602 FIX(1.175875602)
137 #define FIX_1_501321110 FIX(1.501321110)
138 #define FIX_1_847759065 FIX(1.847759065)
139 #define FIX_1_961570560 FIX(1.961570560)
140 #define FIX_2_053119869 FIX(2.053119869)
141 #define FIX_2_562915447 FIX(2.562915447)
142 #define FIX_3_072711026 FIX(3.072711026)
143 #endif
144
145
146 /* Multiply an JLONG variable by an JLONG constant to yield an JLONG result.
147 * For 8-bit samples with the recommended scaling, all the variable
148 * and constant values involved are no more than 16 bits wide, so a
149 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
150 * For 12-bit samples, a full 32-bit multiplication will be needed.
151 */
152
153 #if BITS_IN_JSAMPLE == 8
154 #define MULTIPLY(var, const) MULTIPLY16C16(var, const)
155 #else
156 #define MULTIPLY(var, const) ((var) * (const))
157 #endif
158
159
160 /* Dequantize a coefficient by multiplying it by the multiplier-table
161 * entry; produce an int result. In this module, both inputs and result
162 * are 16 bits or less, so either int or short multiply will work.
163 */
164
165 #define DEQUANTIZE(coef, quantval) (((ISLOW_MULT_TYPE)(coef)) * (quantval))
166
167
168 /*
169 * Perform dequantization and inverse DCT on one block of coefficients.
170 */
171
172 GLOBAL(void)
173 jpeg_idct_islow(j_decompress_ptr cinfo, jpeg_component_info *compptr,
174 JCOEFPTR coef_block, JSAMPARRAY output_buf,
175 JDIMENSION output_col)
176 {
177 JLONG tmp0, tmp1, tmp2, tmp3;
178 JLONG tmp10, tmp11, tmp12, tmp13;
179 JLONG z1, z2, z3, z4, z5;
180 JCOEFPTR inptr;
181 ISLOW_MULT_TYPE *quantptr;
182 int *wsptr;
183 JSAMPROW outptr;
184 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
185 int ctr;
186 int workspace[DCTSIZE2]; /* buffers data between passes */
187 SHIFT_TEMPS
188
189 /* Pass 1: process columns from input, store into work array. */
190 /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
191 /* furthermore, we scale the results by 2**PASS1_BITS. */
192
193 inptr = coef_block;
194 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
195 wsptr = workspace;
196 for (ctr = DCTSIZE; ctr > 0; ctr--) {
197 /* Due to quantization, we will usually find that many of the input
198 * coefficients are zero, especially the AC terms. We can exploit this
199 * by short-circuiting the IDCT calculation for any column in which all
200 * the AC terms are zero. In that case each output is equal to the
201 * DC coefficient (with scale factor as needed).
202 * With typical images and quantization tables, half or more of the
203 * column DCT calculations can be simplified this way.
204 */
205
206 if (inptr[DCTSIZE * 1] == 0 && inptr[DCTSIZE * 2] == 0 &&
207 inptr[DCTSIZE * 3] == 0 && inptr[DCTSIZE * 4] == 0 &&
208 inptr[DCTSIZE * 5] == 0 && inptr[DCTSIZE * 6] == 0 &&
209 inptr[DCTSIZE * 7] == 0) {
210 /* AC terms all zero */
211 int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE * 0],
212 quantptr[DCTSIZE * 0]), PASS1_BITS);
213
214 wsptr[DCTSIZE * 0] = dcval;
215 wsptr[DCTSIZE * 1] = dcval;
216 wsptr[DCTSIZE * 2] = dcval;
217 wsptr[DCTSIZE * 3] = dcval;
218 wsptr[DCTSIZE * 4] = dcval;
219 wsptr[DCTSIZE * 5] = dcval;
220 wsptr[DCTSIZE * 6] = dcval;
221 wsptr[DCTSIZE * 7] = dcval;
222
223 inptr++; /* advance pointers to next column */
224 quantptr++;
225 wsptr++;
226 continue;
227 }
228
229 /* Even part: reverse the even part of the forward DCT. */
230 /* The rotator is sqrt(2)*c(-6). */
231
232 z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
233 z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
234
235 z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
236 tmp2 = z1 + MULTIPLY(z3, -FIX_1_847759065);
237 tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
238
239 z2 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
240 z3 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
241
242 tmp0 = LEFT_SHIFT(z2 + z3, CONST_BITS);
243 tmp1 = LEFT_SHIFT(z2 - z3, CONST_BITS);
244
245 tmp10 = tmp0 + tmp3;
246 tmp13 = tmp0 - tmp3;
247 tmp11 = tmp1 + tmp2;
248 tmp12 = tmp1 - tmp2;
249
250 /* Odd part per figure 8; the matrix is unitary and hence its
251 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
252 */
253
254 tmp0 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
255 tmp1 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
256 tmp2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
257 tmp3 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
258
259 z1 = tmp0 + tmp3;
260 z2 = tmp1 + tmp2;
261 z3 = tmp0 + tmp2;
262 z4 = tmp1 + tmp3;
263 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
264
265 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
266 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
267 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
268 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
269 z1 = MULTIPLY(z1, -FIX_0_899976223); /* sqrt(2) * ( c7-c3) */
270 z2 = MULTIPLY(z2, -FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
271 z3 = MULTIPLY(z3, -FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
272 z4 = MULTIPLY(z4, -FIX_0_390180644); /* sqrt(2) * ( c5-c3) */
273
274 z3 += z5;
275 z4 += z5;
276
277 tmp0 += z1 + z3;
278 tmp1 += z2 + z4;
279 tmp2 += z2 + z3;
280 tmp3 += z1 + z4;
281
282 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
283
284 wsptr[DCTSIZE * 0] = (int)DESCALE(tmp10 + tmp3, CONST_BITS - PASS1_BITS);
285 wsptr[DCTSIZE * 7] = (int)DESCALE(tmp10 - tmp3, CONST_BITS - PASS1_BITS);
286 wsptr[DCTSIZE * 1] = (int)DESCALE(tmp11 + tmp2, CONST_BITS - PASS1_BITS);
287 wsptr[DCTSIZE * 6] = (int)DESCALE(tmp11 - tmp2, CONST_BITS - PASS1_BITS);
288 wsptr[DCTSIZE * 2] = (int)DESCALE(tmp12 + tmp1, CONST_BITS - PASS1_BITS);
289 wsptr[DCTSIZE * 5] = (int)DESCALE(tmp12 - tmp1, CONST_BITS - PASS1_BITS);
290 wsptr[DCTSIZE * 3] = (int)DESCALE(tmp13 + tmp0, CONST_BITS - PASS1_BITS);
291 wsptr[DCTSIZE * 4] = (int)DESCALE(tmp13 - tmp0, CONST_BITS - PASS1_BITS);
292
293 inptr++; /* advance pointers to next column */
294 quantptr++;
295 wsptr++;
296 }
297
298 /* Pass 2: process rows from work array, store into output array. */
299 /* Note that we must descale the results by a factor of 8 == 2**3, */
300 /* and also undo the PASS1_BITS scaling. */
301
302 wsptr = workspace;
303 for (ctr = 0; ctr < DCTSIZE; ctr++) {
304 outptr = output_buf[ctr] + output_col;
305 /* Rows of zeroes can be exploited in the same way as we did with columns.
306 * However, the column calculation has created many nonzero AC terms, so
307 * the simplification applies less often (typically 5% to 10% of the time).
308 * On machines with very fast multiplication, it's possible that the
309 * test takes more time than it's worth. In that case this section
310 * may be commented out.
311 */
312
313 #ifndef NO_ZERO_ROW_TEST
314 if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
315 wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
316 /* AC terms all zero */
317 JSAMPLE dcval = range_limit[(int)DESCALE((JLONG)wsptr[0],
318 PASS1_BITS + 3) & RANGE_MASK];
319
320 outptr[0] = dcval;
321 outptr[1] = dcval;
322 outptr[2] = dcval;
323 outptr[3] = dcval;
324 outptr[4] = dcval;
325 outptr[5] = dcval;
326 outptr[6] = dcval;
327 outptr[7] = dcval;
328
329 wsptr += DCTSIZE; /* advance pointer to next row */
330 continue;
331 }
332 #endif
333
334 /* Even part: reverse the even part of the forward DCT. */
335 /* The rotator is sqrt(2)*c(-6). */
336
337 z2 = (JLONG)wsptr[2];
338 z3 = (JLONG)wsptr[6];
339
340 z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
341 tmp2 = z1 + MULTIPLY(z3, -FIX_1_847759065);
342 tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
343
344 tmp0 = LEFT_SHIFT((JLONG)wsptr[0] + (JLONG)wsptr[4], CONST_BITS);
345 tmp1 = LEFT_SHIFT((JLONG)wsptr[0] - (JLONG)wsptr[4], CONST_BITS);
346
347 tmp10 = tmp0 + tmp3;
348 tmp13 = tmp0 - tmp3;
349 tmp11 = tmp1 + tmp2;
350 tmp12 = tmp1 - tmp2;
351
352 /* Odd part per figure 8; the matrix is unitary and hence its
353 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
354 */
355
356 tmp0 = (JLONG)wsptr[7];
357 tmp1 = (JLONG)wsptr[5];
358 tmp2 = (JLONG)wsptr[3];
359 tmp3 = (JLONG)wsptr[1];
360
361 z1 = tmp0 + tmp3;
362 z2 = tmp1 + tmp2;
363 z3 = tmp0 + tmp2;
364 z4 = tmp1 + tmp3;
365 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
366
367 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
368 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
369 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
370 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
371 z1 = MULTIPLY(z1, -FIX_0_899976223); /* sqrt(2) * ( c7-c3) */
372 z2 = MULTIPLY(z2, -FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
373 z3 = MULTIPLY(z3, -FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
374 z4 = MULTIPLY(z4, -FIX_0_390180644); /* sqrt(2) * ( c5-c3) */
375
376 z3 += z5;
377 z4 += z5;
378
379 tmp0 += z1 + z3;
380 tmp1 += z2 + z4;
381 tmp2 += z2 + z3;
382 tmp3 += z1 + z4;
383
384 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
385
386 outptr[0] = range_limit[(int)DESCALE(tmp10 + tmp3,
387 CONST_BITS + PASS1_BITS + 3) &
388 RANGE_MASK];
389 outptr[7] = range_limit[(int)DESCALE(tmp10 - tmp3,
390 CONST_BITS + PASS1_BITS + 3) &
391 RANGE_MASK];
392 outptr[1] = range_limit[(int)DESCALE(tmp11 + tmp2,
393 CONST_BITS + PASS1_BITS + 3) &
394 RANGE_MASK];
395 outptr[6] = range_limit[(int)DESCALE(tmp11 - tmp2,
396 CONST_BITS + PASS1_BITS + 3) &
397 RANGE_MASK];
398 outptr[2] = range_limit[(int)DESCALE(tmp12 + tmp1,
399 CONST_BITS + PASS1_BITS + 3) &
400 RANGE_MASK];
401 outptr[5] = range_limit[(int)DESCALE(tmp12 - tmp1,
402 CONST_BITS + PASS1_BITS + 3) &
403 RANGE_MASK];
404 outptr[3] = range_limit[(int)DESCALE(tmp13 + tmp0,
405 CONST_BITS + PASS1_BITS + 3) &
406 RANGE_MASK];
407 outptr[4] = range_limit[(int)DESCALE(tmp13 - tmp0,
408 CONST_BITS + PASS1_BITS + 3) &
409 RANGE_MASK];
410
411 wsptr += DCTSIZE; /* advance pointer to next row */
412 }
413 }
414
415 #ifdef IDCT_SCALING_SUPPORTED
416
417
418 /*
419 * Perform dequantization and inverse DCT on one block of coefficients,
420 * producing a reduced-size 7x7 output block.
421 *
422 * Optimized algorithm with 12 multiplications in the 1-D kernel.
423 * cK represents sqrt(2) * cos(K*pi/14).
424 */
425
426 GLOBAL(void)
jpeg_idct_7x7(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)427 jpeg_idct_7x7(j_decompress_ptr cinfo, jpeg_component_info *compptr,
428 JCOEFPTR coef_block, JSAMPARRAY output_buf,
429 JDIMENSION output_col)
430 {
431 JLONG tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
432 JLONG z1, z2, z3;
433 JCOEFPTR inptr;
434 ISLOW_MULT_TYPE *quantptr;
435 int *wsptr;
436 JSAMPROW outptr;
437 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
438 int ctr;
439 int workspace[7 * 7]; /* buffers data between passes */
440 SHIFT_TEMPS
441
442 /* Pass 1: process columns from input, store into work array. */
443
444 inptr = coef_block;
445 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
446 wsptr = workspace;
447 for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
448 /* Even part */
449
450 tmp13 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
451 tmp13 = LEFT_SHIFT(tmp13, CONST_BITS);
452 /* Add fudge factor here for final descale. */
453 tmp13 += ONE << (CONST_BITS - PASS1_BITS - 1);
454
455 z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
456 z2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
457 z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
458
459 tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
460 tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
461 tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
462 tmp0 = z1 + z3;
463 z2 -= tmp0;
464 tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
465 tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
466 tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
467 tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
468
469 /* Odd part */
470
471 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
472 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
473 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
474
475 tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
476 tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
477 tmp0 = tmp1 - tmp2;
478 tmp1 += tmp2;
479 tmp2 = MULTIPLY(z2 + z3, -FIX(1.378756276)); /* -c1 */
480 tmp1 += tmp2;
481 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
482 tmp0 += z2;
483 tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
484
485 /* Final output stage */
486
487 wsptr[7 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS);
488 wsptr[7 * 6] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS);
489 wsptr[7 * 1] = (int)RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS - PASS1_BITS);
490 wsptr[7 * 5] = (int)RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS - PASS1_BITS);
491 wsptr[7 * 2] = (int)RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS - PASS1_BITS);
492 wsptr[7 * 4] = (int)RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS - PASS1_BITS);
493 wsptr[7 * 3] = (int)RIGHT_SHIFT(tmp13, CONST_BITS - PASS1_BITS);
494 }
495
496 /* Pass 2: process 7 rows from work array, store into output array. */
497
498 wsptr = workspace;
499 for (ctr = 0; ctr < 7; ctr++) {
500 outptr = output_buf[ctr] + output_col;
501
502 /* Even part */
503
504 /* Add fudge factor here for final descale. */
505 tmp13 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
506 tmp13 = LEFT_SHIFT(tmp13, CONST_BITS);
507
508 z1 = (JLONG)wsptr[2];
509 z2 = (JLONG)wsptr[4];
510 z3 = (JLONG)wsptr[6];
511
512 tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
513 tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
514 tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
515 tmp0 = z1 + z3;
516 z2 -= tmp0;
517 tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
518 tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
519 tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
520 tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
521
522 /* Odd part */
523
524 z1 = (JLONG)wsptr[1];
525 z2 = (JLONG)wsptr[3];
526 z3 = (JLONG)wsptr[5];
527
528 tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
529 tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
530 tmp0 = tmp1 - tmp2;
531 tmp1 += tmp2;
532 tmp2 = MULTIPLY(z2 + z3, -FIX(1.378756276)); /* -c1 */
533 tmp1 += tmp2;
534 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
535 tmp0 += z2;
536 tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
537
538 /* Final output stage */
539
540 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0,
541 CONST_BITS + PASS1_BITS + 3) &
542 RANGE_MASK];
543 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0,
544 CONST_BITS + PASS1_BITS + 3) &
545 RANGE_MASK];
546 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1,
547 CONST_BITS + PASS1_BITS + 3) &
548 RANGE_MASK];
549 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1,
550 CONST_BITS + PASS1_BITS + 3) &
551 RANGE_MASK];
552 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12 + tmp2,
553 CONST_BITS + PASS1_BITS + 3) &
554 RANGE_MASK];
555 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp12 - tmp2,
556 CONST_BITS + PASS1_BITS + 3) &
557 RANGE_MASK];
558 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp13,
559 CONST_BITS + PASS1_BITS + 3) &
560 RANGE_MASK];
561
562 wsptr += 7; /* advance pointer to next row */
563 }
564 }
565
566
567 /*
568 * Perform dequantization and inverse DCT on one block of coefficients,
569 * producing a reduced-size 6x6 output block.
570 *
571 * Optimized algorithm with 3 multiplications in the 1-D kernel.
572 * cK represents sqrt(2) * cos(K*pi/12).
573 */
574
575 GLOBAL(void)
jpeg_idct_6x6(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)576 jpeg_idct_6x6(j_decompress_ptr cinfo, jpeg_component_info *compptr,
577 JCOEFPTR coef_block, JSAMPARRAY output_buf,
578 JDIMENSION output_col)
579 {
580 JLONG tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
581 JLONG z1, z2, z3;
582 JCOEFPTR inptr;
583 ISLOW_MULT_TYPE *quantptr;
584 int *wsptr;
585 JSAMPROW outptr;
586 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
587 int ctr;
588 int workspace[6 * 6]; /* buffers data between passes */
589 SHIFT_TEMPS
590
591 /* Pass 1: process columns from input, store into work array. */
592
593 inptr = coef_block;
594 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
595 wsptr = workspace;
596 for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
597 /* Even part */
598
599 tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
600 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
601 /* Add fudge factor here for final descale. */
602 tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1);
603 tmp2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
604 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
605 tmp1 = tmp0 + tmp10;
606 tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS - PASS1_BITS);
607 tmp10 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
608 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
609 tmp10 = tmp1 + tmp0;
610 tmp12 = tmp1 - tmp0;
611
612 /* Odd part */
613
614 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
615 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
616 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
617 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
618 tmp0 = tmp1 + LEFT_SHIFT(z1 + z2, CONST_BITS);
619 tmp2 = tmp1 + LEFT_SHIFT(z3 - z2, CONST_BITS);
620 tmp1 = LEFT_SHIFT(z1 - z2 - z3, PASS1_BITS);
621
622 /* Final output stage */
623
624 wsptr[6 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS);
625 wsptr[6 * 5] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS);
626 wsptr[6 * 1] = (int)(tmp11 + tmp1);
627 wsptr[6 * 4] = (int)(tmp11 - tmp1);
628 wsptr[6 * 2] = (int)RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS - PASS1_BITS);
629 wsptr[6 * 3] = (int)RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS - PASS1_BITS);
630 }
631
632 /* Pass 2: process 6 rows from work array, store into output array. */
633
634 wsptr = workspace;
635 for (ctr = 0; ctr < 6; ctr++) {
636 outptr = output_buf[ctr] + output_col;
637
638 /* Even part */
639
640 /* Add fudge factor here for final descale. */
641 tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
642 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
643 tmp2 = (JLONG)wsptr[4];
644 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
645 tmp1 = tmp0 + tmp10;
646 tmp11 = tmp0 - tmp10 - tmp10;
647 tmp10 = (JLONG)wsptr[2];
648 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
649 tmp10 = tmp1 + tmp0;
650 tmp12 = tmp1 - tmp0;
651
652 /* Odd part */
653
654 z1 = (JLONG)wsptr[1];
655 z2 = (JLONG)wsptr[3];
656 z3 = (JLONG)wsptr[5];
657 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
658 tmp0 = tmp1 + LEFT_SHIFT(z1 + z2, CONST_BITS);
659 tmp2 = tmp1 + LEFT_SHIFT(z3 - z2, CONST_BITS);
660 tmp1 = LEFT_SHIFT(z1 - z2 - z3, CONST_BITS);
661
662 /* Final output stage */
663
664 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0,
665 CONST_BITS + PASS1_BITS + 3) &
666 RANGE_MASK];
667 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0,
668 CONST_BITS + PASS1_BITS + 3) &
669 RANGE_MASK];
670 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1,
671 CONST_BITS + PASS1_BITS + 3) &
672 RANGE_MASK];
673 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1,
674 CONST_BITS + PASS1_BITS + 3) &
675 RANGE_MASK];
676 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12 + tmp2,
677 CONST_BITS + PASS1_BITS + 3) &
678 RANGE_MASK];
679 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp12 - tmp2,
680 CONST_BITS + PASS1_BITS + 3) &
681 RANGE_MASK];
682
683 wsptr += 6; /* advance pointer to next row */
684 }
685 }
686
687
688 /*
689 * Perform dequantization and inverse DCT on one block of coefficients,
690 * producing a reduced-size 5x5 output block.
691 *
692 * Optimized algorithm with 5 multiplications in the 1-D kernel.
693 * cK represents sqrt(2) * cos(K*pi/10).
694 */
695
696 GLOBAL(void)
jpeg_idct_5x5(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)697 jpeg_idct_5x5(j_decompress_ptr cinfo, jpeg_component_info *compptr,
698 JCOEFPTR coef_block, JSAMPARRAY output_buf,
699 JDIMENSION output_col)
700 {
701 JLONG tmp0, tmp1, tmp10, tmp11, tmp12;
702 JLONG z1, z2, z3;
703 JCOEFPTR inptr;
704 ISLOW_MULT_TYPE *quantptr;
705 int *wsptr;
706 JSAMPROW outptr;
707 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
708 int ctr;
709 int workspace[5 * 5]; /* buffers data between passes */
710 SHIFT_TEMPS
711
712 /* Pass 1: process columns from input, store into work array. */
713
714 inptr = coef_block;
715 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
716 wsptr = workspace;
717 for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
718 /* Even part */
719
720 tmp12 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
721 tmp12 = LEFT_SHIFT(tmp12, CONST_BITS);
722 /* Add fudge factor here for final descale. */
723 tmp12 += ONE << (CONST_BITS - PASS1_BITS - 1);
724 tmp0 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
725 tmp1 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
726 z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
727 z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
728 z3 = tmp12 + z2;
729 tmp10 = z3 + z1;
730 tmp11 = z3 - z1;
731 tmp12 -= LEFT_SHIFT(z2, 2);
732
733 /* Odd part */
734
735 z2 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
736 z3 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
737
738 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
739 tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
740 tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
741
742 /* Final output stage */
743
744 wsptr[5 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS);
745 wsptr[5 * 4] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS);
746 wsptr[5 * 1] = (int)RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS - PASS1_BITS);
747 wsptr[5 * 3] = (int)RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS - PASS1_BITS);
748 wsptr[5 * 2] = (int)RIGHT_SHIFT(tmp12, CONST_BITS - PASS1_BITS);
749 }
750
751 /* Pass 2: process 5 rows from work array, store into output array. */
752
753 wsptr = workspace;
754 for (ctr = 0; ctr < 5; ctr++) {
755 outptr = output_buf[ctr] + output_col;
756
757 /* Even part */
758
759 /* Add fudge factor here for final descale. */
760 tmp12 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
761 tmp12 = LEFT_SHIFT(tmp12, CONST_BITS);
762 tmp0 = (JLONG)wsptr[2];
763 tmp1 = (JLONG)wsptr[4];
764 z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
765 z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
766 z3 = tmp12 + z2;
767 tmp10 = z3 + z1;
768 tmp11 = z3 - z1;
769 tmp12 -= LEFT_SHIFT(z2, 2);
770
771 /* Odd part */
772
773 z2 = (JLONG)wsptr[1];
774 z3 = (JLONG)wsptr[3];
775
776 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
777 tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
778 tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
779
780 /* Final output stage */
781
782 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0,
783 CONST_BITS + PASS1_BITS + 3) &
784 RANGE_MASK];
785 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0,
786 CONST_BITS + PASS1_BITS + 3) &
787 RANGE_MASK];
788 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1,
789 CONST_BITS + PASS1_BITS + 3) &
790 RANGE_MASK];
791 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1,
792 CONST_BITS + PASS1_BITS + 3) &
793 RANGE_MASK];
794 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12,
795 CONST_BITS + PASS1_BITS + 3) &
796 RANGE_MASK];
797
798 wsptr += 5; /* advance pointer to next row */
799 }
800 }
801
802
803 /*
804 * Perform dequantization and inverse DCT on one block of coefficients,
805 * producing a reduced-size 3x3 output block.
806 *
807 * Optimized algorithm with 2 multiplications in the 1-D kernel.
808 * cK represents sqrt(2) * cos(K*pi/6).
809 */
810
811 GLOBAL(void)
jpeg_idct_3x3(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)812 jpeg_idct_3x3(j_decompress_ptr cinfo, jpeg_component_info *compptr,
813 JCOEFPTR coef_block, JSAMPARRAY output_buf,
814 JDIMENSION output_col)
815 {
816 JLONG tmp0, tmp2, tmp10, tmp12;
817 JCOEFPTR inptr;
818 ISLOW_MULT_TYPE *quantptr;
819 int *wsptr;
820 JSAMPROW outptr;
821 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
822 int ctr;
823 int workspace[3 * 3]; /* buffers data between passes */
824 SHIFT_TEMPS
825
826 /* Pass 1: process columns from input, store into work array. */
827
828 inptr = coef_block;
829 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
830 wsptr = workspace;
831 for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
832 /* Even part */
833
834 tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
835 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
836 /* Add fudge factor here for final descale. */
837 tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1);
838 tmp2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
839 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
840 tmp10 = tmp0 + tmp12;
841 tmp2 = tmp0 - tmp12 - tmp12;
842
843 /* Odd part */
844
845 tmp12 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
846 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
847
848 /* Final output stage */
849
850 wsptr[3 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS);
851 wsptr[3 * 2] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS);
852 wsptr[3 * 1] = (int)RIGHT_SHIFT(tmp2, CONST_BITS - PASS1_BITS);
853 }
854
855 /* Pass 2: process 3 rows from work array, store into output array. */
856
857 wsptr = workspace;
858 for (ctr = 0; ctr < 3; ctr++) {
859 outptr = output_buf[ctr] + output_col;
860
861 /* Even part */
862
863 /* Add fudge factor here for final descale. */
864 tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
865 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
866 tmp2 = (JLONG)wsptr[2];
867 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
868 tmp10 = tmp0 + tmp12;
869 tmp2 = tmp0 - tmp12 - tmp12;
870
871 /* Odd part */
872
873 tmp12 = (JLONG)wsptr[1];
874 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
875
876 /* Final output stage */
877
878 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0,
879 CONST_BITS + PASS1_BITS + 3) &
880 RANGE_MASK];
881 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0,
882 CONST_BITS + PASS1_BITS + 3) &
883 RANGE_MASK];
884 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp2,
885 CONST_BITS + PASS1_BITS + 3) &
886 RANGE_MASK];
887
888 wsptr += 3; /* advance pointer to next row */
889 }
890 }
891
892
893 /*
894 * Perform dequantization and inverse DCT on one block of coefficients,
895 * producing a 9x9 output block.
896 *
897 * Optimized algorithm with 10 multiplications in the 1-D kernel.
898 * cK represents sqrt(2) * cos(K*pi/18).
899 */
900
901 GLOBAL(void)
jpeg_idct_9x9(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)902 jpeg_idct_9x9(j_decompress_ptr cinfo, jpeg_component_info *compptr,
903 JCOEFPTR coef_block, JSAMPARRAY output_buf,
904 JDIMENSION output_col)
905 {
906 JLONG tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14;
907 JLONG z1, z2, z3, z4;
908 JCOEFPTR inptr;
909 ISLOW_MULT_TYPE *quantptr;
910 int *wsptr;
911 JSAMPROW outptr;
912 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
913 int ctr;
914 int workspace[8 * 9]; /* buffers data between passes */
915 SHIFT_TEMPS
916
917 /* Pass 1: process columns from input, store into work array. */
918
919 inptr = coef_block;
920 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
921 wsptr = workspace;
922 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
923 /* Even part */
924
925 tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
926 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
927 /* Add fudge factor here for final descale. */
928 tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1);
929
930 z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
931 z2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
932 z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
933
934 tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
935 tmp1 = tmp0 + tmp3;
936 tmp2 = tmp0 - tmp3 - tmp3;
937
938 tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
939 tmp11 = tmp2 + tmp0;
940 tmp14 = tmp2 - tmp0 - tmp0;
941
942 tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
943 tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
944 tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
945
946 tmp10 = tmp1 + tmp0 - tmp3;
947 tmp12 = tmp1 - tmp0 + tmp2;
948 tmp13 = tmp1 - tmp2 + tmp3;
949
950 /* Odd part */
951
952 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
953 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
954 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
955 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
956
957 z2 = MULTIPLY(z2, -FIX(1.224744871)); /* -c3 */
958
959 tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
960 tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
961 tmp0 = tmp2 + tmp3 - z2;
962 tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
963 tmp2 += z2 - tmp1;
964 tmp3 += z2 + tmp1;
965 tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
966
967 /* Final output stage */
968
969 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS);
970 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS);
971 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS - PASS1_BITS);
972 wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS - PASS1_BITS);
973 wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS - PASS1_BITS);
974 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS - PASS1_BITS);
975 wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS - PASS1_BITS);
976 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS - PASS1_BITS);
977 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp14, CONST_BITS - PASS1_BITS);
978 }
979
980 /* Pass 2: process 9 rows from work array, store into output array. */
981
982 wsptr = workspace;
983 for (ctr = 0; ctr < 9; ctr++) {
984 outptr = output_buf[ctr] + output_col;
985
986 /* Even part */
987
988 /* Add fudge factor here for final descale. */
989 tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
990 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
991
992 z1 = (JLONG)wsptr[2];
993 z2 = (JLONG)wsptr[4];
994 z3 = (JLONG)wsptr[6];
995
996 tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
997 tmp1 = tmp0 + tmp3;
998 tmp2 = tmp0 - tmp3 - tmp3;
999
1000 tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1001 tmp11 = tmp2 + tmp0;
1002 tmp14 = tmp2 - tmp0 - tmp0;
1003
1004 tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1005 tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1006 tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1007
1008 tmp10 = tmp1 + tmp0 - tmp3;
1009 tmp12 = tmp1 - tmp0 + tmp2;
1010 tmp13 = tmp1 - tmp2 + tmp3;
1011
1012 /* Odd part */
1013
1014 z1 = (JLONG)wsptr[1];
1015 z2 = (JLONG)wsptr[3];
1016 z3 = (JLONG)wsptr[5];
1017 z4 = (JLONG)wsptr[7];
1018
1019 z2 = MULTIPLY(z2, -FIX(1.224744871)); /* -c3 */
1020
1021 tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1022 tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1023 tmp0 = tmp2 + tmp3 - z2;
1024 tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1025 tmp2 += z2 - tmp1;
1026 tmp3 += z2 + tmp1;
1027 tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1028
1029 /* Final output stage */
1030
1031 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0,
1032 CONST_BITS + PASS1_BITS + 3) &
1033 RANGE_MASK];
1034 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0,
1035 CONST_BITS + PASS1_BITS + 3) &
1036 RANGE_MASK];
1037 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1,
1038 CONST_BITS + PASS1_BITS + 3) &
1039 RANGE_MASK];
1040 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1,
1041 CONST_BITS + PASS1_BITS + 3) &
1042 RANGE_MASK];
1043 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12 + tmp2,
1044 CONST_BITS + PASS1_BITS + 3) &
1045 RANGE_MASK];
1046 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp12 - tmp2,
1047 CONST_BITS + PASS1_BITS + 3) &
1048 RANGE_MASK];
1049 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp13 + tmp3,
1050 CONST_BITS + PASS1_BITS + 3) &
1051 RANGE_MASK];
1052 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp13 - tmp3,
1053 CONST_BITS + PASS1_BITS + 3) &
1054 RANGE_MASK];
1055 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp14,
1056 CONST_BITS + PASS1_BITS + 3) &
1057 RANGE_MASK];
1058
1059 wsptr += 8; /* advance pointer to next row */
1060 }
1061 }
1062
1063
1064 /*
1065 * Perform dequantization and inverse DCT on one block of coefficients,
1066 * producing a 10x10 output block.
1067 *
1068 * Optimized algorithm with 12 multiplications in the 1-D kernel.
1069 * cK represents sqrt(2) * cos(K*pi/20).
1070 */
1071
1072 GLOBAL(void)
jpeg_idct_10x10(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1073 jpeg_idct_10x10(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1074 JCOEFPTR coef_block, JSAMPARRAY output_buf,
1075 JDIMENSION output_col)
1076 {
1077 JLONG tmp10, tmp11, tmp12, tmp13, tmp14;
1078 JLONG tmp20, tmp21, tmp22, tmp23, tmp24;
1079 JLONG z1, z2, z3, z4, z5;
1080 JCOEFPTR inptr;
1081 ISLOW_MULT_TYPE *quantptr;
1082 int *wsptr;
1083 JSAMPROW outptr;
1084 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1085 int ctr;
1086 int workspace[8 * 10]; /* buffers data between passes */
1087 SHIFT_TEMPS
1088
1089 /* Pass 1: process columns from input, store into work array. */
1090
1091 inptr = coef_block;
1092 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
1093 wsptr = workspace;
1094 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1095 /* Even part */
1096
1097 z3 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
1098 z3 = LEFT_SHIFT(z3, CONST_BITS);
1099 /* Add fudge factor here for final descale. */
1100 z3 += ONE << (CONST_BITS - PASS1_BITS - 1);
1101 z4 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
1102 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1103 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1104 tmp10 = z3 + z1;
1105 tmp11 = z3 - z2;
1106
1107 tmp22 = RIGHT_SHIFT(z3 - LEFT_SHIFT(z1 - z2, 1),
1108 CONST_BITS - PASS1_BITS); /* c0 = (c4-c8)*2 */
1109
1110 z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
1111 z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
1112
1113 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1114 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1115 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1116
1117 tmp20 = tmp10 + tmp12;
1118 tmp24 = tmp10 - tmp12;
1119 tmp21 = tmp11 + tmp13;
1120 tmp23 = tmp11 - tmp13;
1121
1122 /* Odd part */
1123
1124 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
1125 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
1126 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
1127 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
1128
1129 tmp11 = z2 + z4;
1130 tmp13 = z2 - z4;
1131
1132 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1133 z5 = LEFT_SHIFT(z3, CONST_BITS);
1134
1135 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1136 z4 = z5 + tmp12;
1137
1138 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1139 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1140
1141 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1142 z4 = z5 - tmp12 - LEFT_SHIFT(tmp13, CONST_BITS - 1);
1143
1144 tmp12 = LEFT_SHIFT(z1 - tmp13 - z3, PASS1_BITS);
1145
1146 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1147 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1148
1149 /* Final output stage */
1150
1151 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS);
1152 wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS);
1153 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS);
1154 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS);
1155 wsptr[8 * 2] = (int)(tmp22 + tmp12);
1156 wsptr[8 * 7] = (int)(tmp22 - tmp12);
1157 wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS);
1158 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS);
1159 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS);
1160 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS);
1161 }
1162
1163 /* Pass 2: process 10 rows from work array, store into output array. */
1164
1165 wsptr = workspace;
1166 for (ctr = 0; ctr < 10; ctr++) {
1167 outptr = output_buf[ctr] + output_col;
1168
1169 /* Even part */
1170
1171 /* Add fudge factor here for final descale. */
1172 z3 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
1173 z3 = LEFT_SHIFT(z3, CONST_BITS);
1174 z4 = (JLONG)wsptr[4];
1175 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1176 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1177 tmp10 = z3 + z1;
1178 tmp11 = z3 - z2;
1179
1180 tmp22 = z3 - LEFT_SHIFT(z1 - z2, 1); /* c0 = (c4-c8)*2 */
1181
1182 z2 = (JLONG)wsptr[2];
1183 z3 = (JLONG)wsptr[6];
1184
1185 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1186 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1187 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1188
1189 tmp20 = tmp10 + tmp12;
1190 tmp24 = tmp10 - tmp12;
1191 tmp21 = tmp11 + tmp13;
1192 tmp23 = tmp11 - tmp13;
1193
1194 /* Odd part */
1195
1196 z1 = (JLONG)wsptr[1];
1197 z2 = (JLONG)wsptr[3];
1198 z3 = (JLONG)wsptr[5];
1199 z3 = LEFT_SHIFT(z3, CONST_BITS);
1200 z4 = (JLONG)wsptr[7];
1201
1202 tmp11 = z2 + z4;
1203 tmp13 = z2 - z4;
1204
1205 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1206
1207 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1208 z4 = z3 + tmp12;
1209
1210 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1211 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1212
1213 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1214 z4 = z3 - tmp12 - LEFT_SHIFT(tmp13, CONST_BITS - 1);
1215
1216 tmp12 = LEFT_SHIFT(z1 - tmp13, CONST_BITS) - z3;
1217
1218 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1219 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1220
1221 /* Final output stage */
1222
1223 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10,
1224 CONST_BITS + PASS1_BITS + 3) &
1225 RANGE_MASK];
1226 outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10,
1227 CONST_BITS + PASS1_BITS + 3) &
1228 RANGE_MASK];
1229 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11,
1230 CONST_BITS + PASS1_BITS + 3) &
1231 RANGE_MASK];
1232 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11,
1233 CONST_BITS + PASS1_BITS + 3) &
1234 RANGE_MASK];
1235 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12,
1236 CONST_BITS + PASS1_BITS + 3) &
1237 RANGE_MASK];
1238 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12,
1239 CONST_BITS + PASS1_BITS + 3) &
1240 RANGE_MASK];
1241 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13,
1242 CONST_BITS + PASS1_BITS + 3) &
1243 RANGE_MASK];
1244 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13,
1245 CONST_BITS + PASS1_BITS + 3) &
1246 RANGE_MASK];
1247 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14,
1248 CONST_BITS + PASS1_BITS + 3) &
1249 RANGE_MASK];
1250 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14,
1251 CONST_BITS + PASS1_BITS + 3) &
1252 RANGE_MASK];
1253
1254 wsptr += 8; /* advance pointer to next row */
1255 }
1256 }
1257
1258
1259 /*
1260 * Perform dequantization and inverse DCT on one block of coefficients,
1261 * producing an 11x11 output block.
1262 *
1263 * Optimized algorithm with 24 multiplications in the 1-D kernel.
1264 * cK represents sqrt(2) * cos(K*pi/22).
1265 */
1266
1267 GLOBAL(void)
jpeg_idct_11x11(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1268 jpeg_idct_11x11(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1269 JCOEFPTR coef_block, JSAMPARRAY output_buf,
1270 JDIMENSION output_col)
1271 {
1272 JLONG tmp10, tmp11, tmp12, tmp13, tmp14;
1273 JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1274 JLONG z1, z2, z3, z4;
1275 JCOEFPTR inptr;
1276 ISLOW_MULT_TYPE *quantptr;
1277 int *wsptr;
1278 JSAMPROW outptr;
1279 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1280 int ctr;
1281 int workspace[8 * 11]; /* buffers data between passes */
1282 SHIFT_TEMPS
1283
1284 /* Pass 1: process columns from input, store into work array. */
1285
1286 inptr = coef_block;
1287 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
1288 wsptr = workspace;
1289 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1290 /* Even part */
1291
1292 tmp10 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
1293 tmp10 = LEFT_SHIFT(tmp10, CONST_BITS);
1294 /* Add fudge factor here for final descale. */
1295 tmp10 += ONE << (CONST_BITS - PASS1_BITS - 1);
1296
1297 z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
1298 z2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
1299 z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
1300
1301 tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1302 tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1303 z4 = z1 + z3;
1304 tmp24 = MULTIPLY(z4, -FIX(1.155664402)); /* -(c2-c10) */
1305 z4 -= z2;
1306 tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1307 tmp21 = tmp20 + tmp23 + tmp25 -
1308 MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1309 tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1310 tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1311 tmp24 += tmp25;
1312 tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1313 tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1314 MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1315 tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1316
1317 /* Odd part */
1318
1319 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
1320 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
1321 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
1322 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
1323
1324 tmp11 = z1 + z2;
1325 tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1326 tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1327 tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1328 tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1329 tmp10 = tmp11 + tmp12 + tmp13 -
1330 MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1331 z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1332 tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1333 tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1334 z1 = MULTIPLY(z2 + z4, -FIX(1.798248910)); /* -(c1+c9) */
1335 tmp11 += z1;
1336 tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1337 tmp14 += MULTIPLY(z2, -FIX(1.467221301)) + /* -(c5+c9) */
1338 MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1339 MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1340
1341 /* Final output stage */
1342
1343 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS);
1344 wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS);
1345 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS);
1346 wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS);
1347 wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS);
1348 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS);
1349 wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS);
1350 wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS);
1351 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS);
1352 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS);
1353 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25, CONST_BITS - PASS1_BITS);
1354 }
1355
1356 /* Pass 2: process 11 rows from work array, store into output array. */
1357
1358 wsptr = workspace;
1359 for (ctr = 0; ctr < 11; ctr++) {
1360 outptr = output_buf[ctr] + output_col;
1361
1362 /* Even part */
1363
1364 /* Add fudge factor here for final descale. */
1365 tmp10 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
1366 tmp10 = LEFT_SHIFT(tmp10, CONST_BITS);
1367
1368 z1 = (JLONG)wsptr[2];
1369 z2 = (JLONG)wsptr[4];
1370 z3 = (JLONG)wsptr[6];
1371
1372 tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1373 tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1374 z4 = z1 + z3;
1375 tmp24 = MULTIPLY(z4, -FIX(1.155664402)); /* -(c2-c10) */
1376 z4 -= z2;
1377 tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1378 tmp21 = tmp20 + tmp23 + tmp25 -
1379 MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1380 tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1381 tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1382 tmp24 += tmp25;
1383 tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1384 tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1385 MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1386 tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1387
1388 /* Odd part */
1389
1390 z1 = (JLONG)wsptr[1];
1391 z2 = (JLONG)wsptr[3];
1392 z3 = (JLONG)wsptr[5];
1393 z4 = (JLONG)wsptr[7];
1394
1395 tmp11 = z1 + z2;
1396 tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1397 tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1398 tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1399 tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1400 tmp10 = tmp11 + tmp12 + tmp13 -
1401 MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1402 z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1403 tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1404 tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1405 z1 = MULTIPLY(z2 + z4, -FIX(1.798248910)); /* -(c1+c9) */
1406 tmp11 += z1;
1407 tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1408 tmp14 += MULTIPLY(z2, -FIX(1.467221301)) + /* -(c5+c9) */
1409 MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1410 MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1411
1412 /* Final output stage */
1413
1414 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10,
1415 CONST_BITS + PASS1_BITS + 3) &
1416 RANGE_MASK];
1417 outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10,
1418 CONST_BITS + PASS1_BITS + 3) &
1419 RANGE_MASK];
1420 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11,
1421 CONST_BITS + PASS1_BITS + 3) &
1422 RANGE_MASK];
1423 outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11,
1424 CONST_BITS + PASS1_BITS + 3) &
1425 RANGE_MASK];
1426 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12,
1427 CONST_BITS + PASS1_BITS + 3) &
1428 RANGE_MASK];
1429 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12,
1430 CONST_BITS + PASS1_BITS + 3) &
1431 RANGE_MASK];
1432 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13,
1433 CONST_BITS + PASS1_BITS + 3) &
1434 RANGE_MASK];
1435 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13,
1436 CONST_BITS + PASS1_BITS + 3) &
1437 RANGE_MASK];
1438 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14,
1439 CONST_BITS + PASS1_BITS + 3) &
1440 RANGE_MASK];
1441 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14,
1442 CONST_BITS + PASS1_BITS + 3) &
1443 RANGE_MASK];
1444 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25,
1445 CONST_BITS + PASS1_BITS + 3) &
1446 RANGE_MASK];
1447
1448 wsptr += 8; /* advance pointer to next row */
1449 }
1450 }
1451
1452
1453 /*
1454 * Perform dequantization and inverse DCT on one block of coefficients,
1455 * producing a 12x12 output block.
1456 *
1457 * Optimized algorithm with 15 multiplications in the 1-D kernel.
1458 * cK represents sqrt(2) * cos(K*pi/24).
1459 */
1460
1461 GLOBAL(void)
jpeg_idct_12x12(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1462 jpeg_idct_12x12(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1463 JCOEFPTR coef_block, JSAMPARRAY output_buf,
1464 JDIMENSION output_col)
1465 {
1466 JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1467 JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1468 JLONG z1, z2, z3, z4;
1469 JCOEFPTR inptr;
1470 ISLOW_MULT_TYPE *quantptr;
1471 int *wsptr;
1472 JSAMPROW outptr;
1473 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1474 int ctr;
1475 int workspace[8 * 12]; /* buffers data between passes */
1476 SHIFT_TEMPS
1477
1478 /* Pass 1: process columns from input, store into work array. */
1479
1480 inptr = coef_block;
1481 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
1482 wsptr = workspace;
1483 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1484 /* Even part */
1485
1486 z3 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
1487 z3 = LEFT_SHIFT(z3, CONST_BITS);
1488 /* Add fudge factor here for final descale. */
1489 z3 += ONE << (CONST_BITS - PASS1_BITS - 1);
1490
1491 z4 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
1492 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1493
1494 tmp10 = z3 + z4;
1495 tmp11 = z3 - z4;
1496
1497 z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
1498 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1499 z1 = LEFT_SHIFT(z1, CONST_BITS);
1500 z2 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
1501 z2 = LEFT_SHIFT(z2, CONST_BITS);
1502
1503 tmp12 = z1 - z2;
1504
1505 tmp21 = z3 + tmp12;
1506 tmp24 = z3 - tmp12;
1507
1508 tmp12 = z4 + z2;
1509
1510 tmp20 = tmp10 + tmp12;
1511 tmp25 = tmp10 - tmp12;
1512
1513 tmp12 = z4 - z1 - z2;
1514
1515 tmp22 = tmp11 + tmp12;
1516 tmp23 = tmp11 - tmp12;
1517
1518 /* Odd part */
1519
1520 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
1521 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
1522 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
1523 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
1524
1525 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1526 tmp14 = MULTIPLY(z2, -FIX_0_541196100); /* -c9 */
1527
1528 tmp10 = z1 + z3;
1529 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1530 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1531 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1532 tmp13 = MULTIPLY(z3 + z4, -FIX(1.045510580)); /* -(c7+c11) */
1533 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1534 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1535 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1536 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1537
1538 z1 -= z4;
1539 z2 -= z3;
1540 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1541 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1542 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1543
1544 /* Final output stage */
1545
1546 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS);
1547 wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS);
1548 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS);
1549 wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS);
1550 wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS);
1551 wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS);
1552 wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS);
1553 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS);
1554 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS);
1555 wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS);
1556 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS);
1557 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS);
1558 }
1559
1560 /* Pass 2: process 12 rows from work array, store into output array. */
1561
1562 wsptr = workspace;
1563 for (ctr = 0; ctr < 12; ctr++) {
1564 outptr = output_buf[ctr] + output_col;
1565
1566 /* Even part */
1567
1568 /* Add fudge factor here for final descale. */
1569 z3 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
1570 z3 = LEFT_SHIFT(z3, CONST_BITS);
1571
1572 z4 = (JLONG)wsptr[4];
1573 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1574
1575 tmp10 = z3 + z4;
1576 tmp11 = z3 - z4;
1577
1578 z1 = (JLONG)wsptr[2];
1579 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1580 z1 = LEFT_SHIFT(z1, CONST_BITS);
1581 z2 = (JLONG)wsptr[6];
1582 z2 = LEFT_SHIFT(z2, CONST_BITS);
1583
1584 tmp12 = z1 - z2;
1585
1586 tmp21 = z3 + tmp12;
1587 tmp24 = z3 - tmp12;
1588
1589 tmp12 = z4 + z2;
1590
1591 tmp20 = tmp10 + tmp12;
1592 tmp25 = tmp10 - tmp12;
1593
1594 tmp12 = z4 - z1 - z2;
1595
1596 tmp22 = tmp11 + tmp12;
1597 tmp23 = tmp11 - tmp12;
1598
1599 /* Odd part */
1600
1601 z1 = (JLONG)wsptr[1];
1602 z2 = (JLONG)wsptr[3];
1603 z3 = (JLONG)wsptr[5];
1604 z4 = (JLONG)wsptr[7];
1605
1606 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1607 tmp14 = MULTIPLY(z2, -FIX_0_541196100); /* -c9 */
1608
1609 tmp10 = z1 + z3;
1610 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1611 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1612 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1613 tmp13 = MULTIPLY(z3 + z4, -FIX(1.045510580)); /* -(c7+c11) */
1614 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1615 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1616 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1617 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1618
1619 z1 -= z4;
1620 z2 -= z3;
1621 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1622 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1623 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1624
1625 /* Final output stage */
1626
1627 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10,
1628 CONST_BITS + PASS1_BITS + 3) &
1629 RANGE_MASK];
1630 outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10,
1631 CONST_BITS + PASS1_BITS + 3) &
1632 RANGE_MASK];
1633 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11,
1634 CONST_BITS + PASS1_BITS + 3) &
1635 RANGE_MASK];
1636 outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11,
1637 CONST_BITS + PASS1_BITS + 3) &
1638 RANGE_MASK];
1639 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12,
1640 CONST_BITS + PASS1_BITS + 3) &
1641 RANGE_MASK];
1642 outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12,
1643 CONST_BITS + PASS1_BITS + 3) &
1644 RANGE_MASK];
1645 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13,
1646 CONST_BITS + PASS1_BITS + 3) &
1647 RANGE_MASK];
1648 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13,
1649 CONST_BITS + PASS1_BITS + 3) &
1650 RANGE_MASK];
1651 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14,
1652 CONST_BITS + PASS1_BITS + 3) &
1653 RANGE_MASK];
1654 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14,
1655 CONST_BITS + PASS1_BITS + 3) &
1656 RANGE_MASK];
1657 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15,
1658 CONST_BITS + PASS1_BITS + 3) &
1659 RANGE_MASK];
1660 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15,
1661 CONST_BITS + PASS1_BITS + 3) &
1662 RANGE_MASK];
1663
1664 wsptr += 8; /* advance pointer to next row */
1665 }
1666 }
1667
1668
1669 /*
1670 * Perform dequantization and inverse DCT on one block of coefficients,
1671 * producing a 13x13 output block.
1672 *
1673 * Optimized algorithm with 29 multiplications in the 1-D kernel.
1674 * cK represents sqrt(2) * cos(K*pi/26).
1675 */
1676
1677 GLOBAL(void)
jpeg_idct_13x13(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1678 jpeg_idct_13x13(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1679 JCOEFPTR coef_block, JSAMPARRAY output_buf,
1680 JDIMENSION output_col)
1681 {
1682 JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1683 JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1684 JLONG z1, z2, z3, z4;
1685 JCOEFPTR inptr;
1686 ISLOW_MULT_TYPE *quantptr;
1687 int *wsptr;
1688 JSAMPROW outptr;
1689 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1690 int ctr;
1691 int workspace[8 * 13]; /* buffers data between passes */
1692 SHIFT_TEMPS
1693
1694 /* Pass 1: process columns from input, store into work array. */
1695
1696 inptr = coef_block;
1697 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
1698 wsptr = workspace;
1699 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1700 /* Even part */
1701
1702 z1 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
1703 z1 = LEFT_SHIFT(z1, CONST_BITS);
1704 /* Add fudge factor here for final descale. */
1705 z1 += ONE << (CONST_BITS - PASS1_BITS - 1);
1706
1707 z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
1708 z3 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
1709 z4 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
1710
1711 tmp10 = z3 + z4;
1712 tmp11 = z3 - z4;
1713
1714 tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1715 tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1716
1717 tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1718 tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1719
1720 tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1721 tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1722
1723 tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1724 tmp25 = MULTIPLY(z2, -FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1725
1726 tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1727 tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1728
1729 tmp23 = MULTIPLY(z2, -FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1730 tmp24 = MULTIPLY(z2, -FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1731
1732 tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1733
1734 /* Odd part */
1735
1736 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
1737 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
1738 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
1739 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
1740
1741 tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1742 tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1743 tmp15 = z1 + z4;
1744 tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1745 tmp10 = tmp11 + tmp12 + tmp13 -
1746 MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1747 tmp14 = MULTIPLY(z2 + z3, -FIX(0.338443458)); /* -c11 */
1748 tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1749 tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1750 tmp14 = MULTIPLY(z2 + z4, -FIX(1.163874945)); /* -c5 */
1751 tmp11 += tmp14;
1752 tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1753 tmp14 = MULTIPLY(z3 + z4, -FIX(0.657217813)); /* -c9 */
1754 tmp12 += tmp14;
1755 tmp13 += tmp14;
1756 tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1757 tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1758 MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1759 z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1760 tmp14 += z1;
1761 tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1762 MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1763
1764 /* Final output stage */
1765
1766 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS);
1767 wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS);
1768 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS);
1769 wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS);
1770 wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS);
1771 wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS);
1772 wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS);
1773 wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS);
1774 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS);
1775 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS);
1776 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS);
1777 wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS);
1778 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26, CONST_BITS - PASS1_BITS);
1779 }
1780
1781 /* Pass 2: process 13 rows from work array, store into output array. */
1782
1783 wsptr = workspace;
1784 for (ctr = 0; ctr < 13; ctr++) {
1785 outptr = output_buf[ctr] + output_col;
1786
1787 /* Even part */
1788
1789 /* Add fudge factor here for final descale. */
1790 z1 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
1791 z1 = LEFT_SHIFT(z1, CONST_BITS);
1792
1793 z2 = (JLONG)wsptr[2];
1794 z3 = (JLONG)wsptr[4];
1795 z4 = (JLONG)wsptr[6];
1796
1797 tmp10 = z3 + z4;
1798 tmp11 = z3 - z4;
1799
1800 tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1801 tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1802
1803 tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1804 tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1805
1806 tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1807 tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1808
1809 tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1810 tmp25 = MULTIPLY(z2, -FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1811
1812 tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1813 tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1814
1815 tmp23 = MULTIPLY(z2, -FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1816 tmp24 = MULTIPLY(z2, -FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1817
1818 tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1819
1820 /* Odd part */
1821
1822 z1 = (JLONG)wsptr[1];
1823 z2 = (JLONG)wsptr[3];
1824 z3 = (JLONG)wsptr[5];
1825 z4 = (JLONG)wsptr[7];
1826
1827 tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1828 tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1829 tmp15 = z1 + z4;
1830 tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1831 tmp10 = tmp11 + tmp12 + tmp13 -
1832 MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1833 tmp14 = MULTIPLY(z2 + z3, -FIX(0.338443458)); /* -c11 */
1834 tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1835 tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1836 tmp14 = MULTIPLY(z2 + z4, -FIX(1.163874945)); /* -c5 */
1837 tmp11 += tmp14;
1838 tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1839 tmp14 = MULTIPLY(z3 + z4, -FIX(0.657217813)); /* -c9 */
1840 tmp12 += tmp14;
1841 tmp13 += tmp14;
1842 tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1843 tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1844 MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1845 z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1846 tmp14 += z1;
1847 tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1848 MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1849
1850 /* Final output stage */
1851
1852 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10,
1853 CONST_BITS + PASS1_BITS + 3) &
1854 RANGE_MASK];
1855 outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10,
1856 CONST_BITS + PASS1_BITS + 3) &
1857 RANGE_MASK];
1858 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11,
1859 CONST_BITS + PASS1_BITS + 3) &
1860 RANGE_MASK];
1861 outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11,
1862 CONST_BITS + PASS1_BITS + 3) &
1863 RANGE_MASK];
1864 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12,
1865 CONST_BITS + PASS1_BITS + 3) &
1866 RANGE_MASK];
1867 outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12,
1868 CONST_BITS + PASS1_BITS + 3) &
1869 RANGE_MASK];
1870 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13,
1871 CONST_BITS + PASS1_BITS + 3) &
1872 RANGE_MASK];
1873 outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13,
1874 CONST_BITS + PASS1_BITS + 3) &
1875 RANGE_MASK];
1876 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14,
1877 CONST_BITS + PASS1_BITS + 3) &
1878 RANGE_MASK];
1879 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14,
1880 CONST_BITS + PASS1_BITS + 3) &
1881 RANGE_MASK];
1882 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15,
1883 CONST_BITS + PASS1_BITS + 3) &
1884 RANGE_MASK];
1885 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15,
1886 CONST_BITS + PASS1_BITS + 3) &
1887 RANGE_MASK];
1888 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26,
1889 CONST_BITS + PASS1_BITS + 3) &
1890 RANGE_MASK];
1891
1892 wsptr += 8; /* advance pointer to next row */
1893 }
1894 }
1895
1896
1897 /*
1898 * Perform dequantization and inverse DCT on one block of coefficients,
1899 * producing a 14x14 output block.
1900 *
1901 * Optimized algorithm with 20 multiplications in the 1-D kernel.
1902 * cK represents sqrt(2) * cos(K*pi/28).
1903 */
1904
1905 GLOBAL(void)
jpeg_idct_14x14(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)1906 jpeg_idct_14x14(j_decompress_ptr cinfo, jpeg_component_info *compptr,
1907 JCOEFPTR coef_block, JSAMPARRAY output_buf,
1908 JDIMENSION output_col)
1909 {
1910 JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
1911 JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1912 JLONG z1, z2, z3, z4;
1913 JCOEFPTR inptr;
1914 ISLOW_MULT_TYPE *quantptr;
1915 int *wsptr;
1916 JSAMPROW outptr;
1917 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1918 int ctr;
1919 int workspace[8 * 14]; /* buffers data between passes */
1920 SHIFT_TEMPS
1921
1922 /* Pass 1: process columns from input, store into work array. */
1923
1924 inptr = coef_block;
1925 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
1926 wsptr = workspace;
1927 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1928 /* Even part */
1929
1930 z1 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
1931 z1 = LEFT_SHIFT(z1, CONST_BITS);
1932 /* Add fudge factor here for final descale. */
1933 z1 += ONE << (CONST_BITS - PASS1_BITS - 1);
1934 z4 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
1935 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
1936 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
1937 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
1938
1939 tmp10 = z1 + z2;
1940 tmp11 = z1 + z3;
1941 tmp12 = z1 - z4;
1942
1943 tmp23 = RIGHT_SHIFT(z1 - LEFT_SHIFT(z2 + z3 - z4, 1),
1944 CONST_BITS - PASS1_BITS); /* c0 = (c4+c12-c8)*2 */
1945
1946 z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
1947 z2 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
1948
1949 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
1950
1951 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
1952 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
1953 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
1954 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
1955
1956 tmp20 = tmp10 + tmp13;
1957 tmp26 = tmp10 - tmp13;
1958 tmp21 = tmp11 + tmp14;
1959 tmp25 = tmp11 - tmp14;
1960 tmp22 = tmp12 + tmp15;
1961 tmp24 = tmp12 - tmp15;
1962
1963 /* Odd part */
1964
1965 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
1966 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
1967 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
1968 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
1969 tmp13 = LEFT_SHIFT(z4, CONST_BITS);
1970
1971 tmp14 = z1 + z3;
1972 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
1973 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
1974 tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
1975 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
1976 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
1977 z1 -= z2;
1978 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */
1979 tmp16 += tmp15;
1980 z1 += z4;
1981 z4 = MULTIPLY(z2 + z3, -FIX(0.158341681)) - tmp13; /* -c13 */
1982 tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
1983 tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
1984 z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
1985 tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
1986 tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
1987
1988 tmp13 = LEFT_SHIFT(z1 - z3, PASS1_BITS);
1989
1990 /* Final output stage */
1991
1992 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS);
1993 wsptr[8 * 13] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS);
1994 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS);
1995 wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS);
1996 wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS);
1997 wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS);
1998 wsptr[8 * 3] = (int)(tmp23 + tmp13);
1999 wsptr[8 * 10] = (int)(tmp23 - tmp13);
2000 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS);
2001 wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS);
2002 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS);
2003 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS);
2004 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS - PASS1_BITS);
2005 wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS - PASS1_BITS);
2006 }
2007
2008 /* Pass 2: process 14 rows from work array, store into output array. */
2009
2010 wsptr = workspace;
2011 for (ctr = 0; ctr < 14; ctr++) {
2012 outptr = output_buf[ctr] + output_col;
2013
2014 /* Even part */
2015
2016 /* Add fudge factor here for final descale. */
2017 z1 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
2018 z1 = LEFT_SHIFT(z1, CONST_BITS);
2019 z4 = (JLONG)wsptr[4];
2020 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
2021 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
2022 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
2023
2024 tmp10 = z1 + z2;
2025 tmp11 = z1 + z3;
2026 tmp12 = z1 - z4;
2027
2028 tmp23 = z1 - LEFT_SHIFT(z2 + z3 - z4, 1); /* c0 = (c4+c12-c8)*2 */
2029
2030 z1 = (JLONG)wsptr[2];
2031 z2 = (JLONG)wsptr[6];
2032
2033 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
2034
2035 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
2036 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
2037 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
2038 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
2039
2040 tmp20 = tmp10 + tmp13;
2041 tmp26 = tmp10 - tmp13;
2042 tmp21 = tmp11 + tmp14;
2043 tmp25 = tmp11 - tmp14;
2044 tmp22 = tmp12 + tmp15;
2045 tmp24 = tmp12 - tmp15;
2046
2047 /* Odd part */
2048
2049 z1 = (JLONG)wsptr[1];
2050 z2 = (JLONG)wsptr[3];
2051 z3 = (JLONG)wsptr[5];
2052 z4 = (JLONG)wsptr[7];
2053 z4 = LEFT_SHIFT(z4, CONST_BITS);
2054
2055 tmp14 = z1 + z3;
2056 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
2057 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
2058 tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
2059 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
2060 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
2061 z1 -= z2;
2062 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */
2063 tmp16 += tmp15;
2064 tmp13 = MULTIPLY(z2 + z3, -FIX(0.158341681)) - z4; /* -c13 */
2065 tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
2066 tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
2067 tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
2068 tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
2069 tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
2070
2071 tmp13 = LEFT_SHIFT(z1 - z3, CONST_BITS) + z4;
2072
2073 /* Final output stage */
2074
2075 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10,
2076 CONST_BITS + PASS1_BITS + 3) &
2077 RANGE_MASK];
2078 outptr[13] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10,
2079 CONST_BITS + PASS1_BITS + 3) &
2080 RANGE_MASK];
2081 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11,
2082 CONST_BITS + PASS1_BITS + 3) &
2083 RANGE_MASK];
2084 outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11,
2085 CONST_BITS + PASS1_BITS + 3) &
2086 RANGE_MASK];
2087 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12,
2088 CONST_BITS + PASS1_BITS + 3) &
2089 RANGE_MASK];
2090 outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12,
2091 CONST_BITS + PASS1_BITS + 3) &
2092 RANGE_MASK];
2093 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13,
2094 CONST_BITS + PASS1_BITS + 3) &
2095 RANGE_MASK];
2096 outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13,
2097 CONST_BITS + PASS1_BITS + 3) &
2098 RANGE_MASK];
2099 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14,
2100 CONST_BITS + PASS1_BITS + 3) &
2101 RANGE_MASK];
2102 outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14,
2103 CONST_BITS + PASS1_BITS + 3) &
2104 RANGE_MASK];
2105 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15,
2106 CONST_BITS + PASS1_BITS + 3) &
2107 RANGE_MASK];
2108 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15,
2109 CONST_BITS + PASS1_BITS + 3) &
2110 RANGE_MASK];
2111 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26 + tmp16,
2112 CONST_BITS + PASS1_BITS + 3) &
2113 RANGE_MASK];
2114 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp26 - tmp16,
2115 CONST_BITS + PASS1_BITS + 3) &
2116 RANGE_MASK];
2117
2118 wsptr += 8; /* advance pointer to next row */
2119 }
2120 }
2121
2122
2123 /*
2124 * Perform dequantization and inverse DCT on one block of coefficients,
2125 * producing a 15x15 output block.
2126 *
2127 * Optimized algorithm with 22 multiplications in the 1-D kernel.
2128 * cK represents sqrt(2) * cos(K*pi/30).
2129 */
2130
2131 GLOBAL(void)
jpeg_idct_15x15(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)2132 jpeg_idct_15x15(j_decompress_ptr cinfo, jpeg_component_info *compptr,
2133 JCOEFPTR coef_block, JSAMPARRAY output_buf,
2134 JDIMENSION output_col)
2135 {
2136 JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
2137 JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2138 JLONG z1, z2, z3, z4;
2139 JCOEFPTR inptr;
2140 ISLOW_MULT_TYPE *quantptr;
2141 int *wsptr;
2142 JSAMPROW outptr;
2143 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2144 int ctr;
2145 int workspace[8 * 15]; /* buffers data between passes */
2146 SHIFT_TEMPS
2147
2148 /* Pass 1: process columns from input, store into work array. */
2149
2150 inptr = coef_block;
2151 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
2152 wsptr = workspace;
2153 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2154 /* Even part */
2155
2156 z1 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
2157 z1 = LEFT_SHIFT(z1, CONST_BITS);
2158 /* Add fudge factor here for final descale. */
2159 z1 += ONE << (CONST_BITS - PASS1_BITS - 1);
2160
2161 z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
2162 z3 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
2163 z4 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
2164
2165 tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2166 tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2167
2168 tmp12 = z1 - tmp10;
2169 tmp13 = z1 + tmp11;
2170 z1 -= LEFT_SHIFT(tmp11 - tmp10, 1); /* c0 = (c6-c12)*2 */
2171
2172 z4 = z2 - z3;
2173 z3 += z2;
2174 tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2175 tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2176 z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2177
2178 tmp20 = tmp13 + tmp10 + tmp11;
2179 tmp23 = tmp12 - tmp10 + tmp11 + z2;
2180
2181 tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2182 tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2183
2184 tmp25 = tmp13 - tmp10 - tmp11;
2185 tmp26 = tmp12 + tmp10 - tmp11 - z2;
2186
2187 tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2188 tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2189
2190 tmp21 = tmp12 + tmp10 + tmp11;
2191 tmp24 = tmp13 - tmp10 + tmp11;
2192 tmp11 += tmp11;
2193 tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2194 tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
2195
2196 /* Odd part */
2197
2198 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
2199 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
2200 z4 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
2201 z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2202 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
2203
2204 tmp13 = z2 - z4;
2205 tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2206 tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2207 tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2208
2209 tmp13 = MULTIPLY(z2, -FIX(0.831253876)); /* -c9 */
2210 tmp15 = MULTIPLY(z2, -FIX(1.344997024)); /* -c3 */
2211 z2 = z1 - z4;
2212 tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2213
2214 tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2215 tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2216 tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2217 z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2218 tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2219 tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2220
2221 /* Final output stage */
2222
2223 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS);
2224 wsptr[8 * 14] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS);
2225 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS);
2226 wsptr[8 * 13] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS);
2227 wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS);
2228 wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS);
2229 wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS);
2230 wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS);
2231 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS);
2232 wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS);
2233 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS);
2234 wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS);
2235 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS - PASS1_BITS);
2236 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS - PASS1_BITS);
2237 wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp27, CONST_BITS - PASS1_BITS);
2238 }
2239
2240 /* Pass 2: process 15 rows from work array, store into output array. */
2241
2242 wsptr = workspace;
2243 for (ctr = 0; ctr < 15; ctr++) {
2244 outptr = output_buf[ctr] + output_col;
2245
2246 /* Even part */
2247
2248 /* Add fudge factor here for final descale. */
2249 z1 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
2250 z1 = LEFT_SHIFT(z1, CONST_BITS);
2251
2252 z2 = (JLONG)wsptr[2];
2253 z3 = (JLONG)wsptr[4];
2254 z4 = (JLONG)wsptr[6];
2255
2256 tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2257 tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2258
2259 tmp12 = z1 - tmp10;
2260 tmp13 = z1 + tmp11;
2261 z1 -= LEFT_SHIFT(tmp11 - tmp10, 1); /* c0 = (c6-c12)*2 */
2262
2263 z4 = z2 - z3;
2264 z3 += z2;
2265 tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2266 tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2267 z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2268
2269 tmp20 = tmp13 + tmp10 + tmp11;
2270 tmp23 = tmp12 - tmp10 + tmp11 + z2;
2271
2272 tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2273 tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2274
2275 tmp25 = tmp13 - tmp10 - tmp11;
2276 tmp26 = tmp12 + tmp10 - tmp11 - z2;
2277
2278 tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2279 tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2280
2281 tmp21 = tmp12 + tmp10 + tmp11;
2282 tmp24 = tmp13 - tmp10 + tmp11;
2283 tmp11 += tmp11;
2284 tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2285 tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
2286
2287 /* Odd part */
2288
2289 z1 = (JLONG)wsptr[1];
2290 z2 = (JLONG)wsptr[3];
2291 z4 = (JLONG)wsptr[5];
2292 z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2293 z4 = (JLONG)wsptr[7];
2294
2295 tmp13 = z2 - z4;
2296 tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2297 tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2298 tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2299
2300 tmp13 = MULTIPLY(z2, -FIX(0.831253876)); /* -c9 */
2301 tmp15 = MULTIPLY(z2, -FIX(1.344997024)); /* -c3 */
2302 z2 = z1 - z4;
2303 tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2304
2305 tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2306 tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2307 tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2308 z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2309 tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2310 tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2311
2312 /* Final output stage */
2313
2314 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10,
2315 CONST_BITS + PASS1_BITS + 3) &
2316 RANGE_MASK];
2317 outptr[14] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10,
2318 CONST_BITS + PASS1_BITS + 3) &
2319 RANGE_MASK];
2320 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11,
2321 CONST_BITS + PASS1_BITS + 3) &
2322 RANGE_MASK];
2323 outptr[13] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11,
2324 CONST_BITS + PASS1_BITS + 3) &
2325 RANGE_MASK];
2326 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12,
2327 CONST_BITS + PASS1_BITS + 3) &
2328 RANGE_MASK];
2329 outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12,
2330 CONST_BITS + PASS1_BITS + 3) &
2331 RANGE_MASK];
2332 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13,
2333 CONST_BITS + PASS1_BITS + 3) &
2334 RANGE_MASK];
2335 outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13,
2336 CONST_BITS + PASS1_BITS + 3) &
2337 RANGE_MASK];
2338 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14,
2339 CONST_BITS + PASS1_BITS + 3) &
2340 RANGE_MASK];
2341 outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14,
2342 CONST_BITS + PASS1_BITS + 3) &
2343 RANGE_MASK];
2344 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15,
2345 CONST_BITS + PASS1_BITS + 3) &
2346 RANGE_MASK];
2347 outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15,
2348 CONST_BITS + PASS1_BITS + 3) &
2349 RANGE_MASK];
2350 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26 + tmp16,
2351 CONST_BITS + PASS1_BITS + 3) &
2352 RANGE_MASK];
2353 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp26 - tmp16,
2354 CONST_BITS + PASS1_BITS + 3) &
2355 RANGE_MASK];
2356 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp27,
2357 CONST_BITS + PASS1_BITS + 3) &
2358 RANGE_MASK];
2359
2360 wsptr += 8; /* advance pointer to next row */
2361 }
2362 }
2363
2364
2365 /*
2366 * Perform dequantization and inverse DCT on one block of coefficients,
2367 * producing a 16x16 output block.
2368 *
2369 * Optimized algorithm with 28 multiplications in the 1-D kernel.
2370 * cK represents sqrt(2) * cos(K*pi/32).
2371 */
2372
2373 GLOBAL(void)
jpeg_idct_16x16(j_decompress_ptr cinfo,jpeg_component_info * compptr,JCOEFPTR coef_block,JSAMPARRAY output_buf,JDIMENSION output_col)2374 jpeg_idct_16x16(j_decompress_ptr cinfo, jpeg_component_info *compptr,
2375 JCOEFPTR coef_block, JSAMPARRAY output_buf,
2376 JDIMENSION output_col)
2377 {
2378 JLONG tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
2379 JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2380 JLONG z1, z2, z3, z4;
2381 JCOEFPTR inptr;
2382 ISLOW_MULT_TYPE *quantptr;
2383 int *wsptr;
2384 JSAMPROW outptr;
2385 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2386 int ctr;
2387 int workspace[8 * 16]; /* buffers data between passes */
2388 SHIFT_TEMPS
2389
2390 /* Pass 1: process columns from input, store into work array. */
2391
2392 inptr = coef_block;
2393 quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table;
2394 wsptr = workspace;
2395 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2396 /* Even part */
2397
2398 tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]);
2399 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
2400 /* Add fudge factor here for final descale. */
2401 tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1);
2402
2403 z1 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]);
2404 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2405 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2406
2407 tmp10 = tmp0 + tmp1;
2408 tmp11 = tmp0 - tmp1;
2409 tmp12 = tmp0 + tmp2;
2410 tmp13 = tmp0 - tmp2;
2411
2412 z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]);
2413 z2 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]);
2414 z3 = z1 - z2;
2415 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2416 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2417
2418 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2419 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2420 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2421 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2422
2423 tmp20 = tmp10 + tmp0;
2424 tmp27 = tmp10 - tmp0;
2425 tmp21 = tmp12 + tmp1;
2426 tmp26 = tmp12 - tmp1;
2427 tmp22 = tmp13 + tmp2;
2428 tmp25 = tmp13 - tmp2;
2429 tmp23 = tmp11 + tmp3;
2430 tmp24 = tmp11 - tmp3;
2431
2432 /* Odd part */
2433
2434 z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]);
2435 z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]);
2436 z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]);
2437 z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]);
2438
2439 tmp11 = z1 + z3;
2440
2441 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2442 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2443 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2444 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2445 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2446 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2447 tmp0 = tmp1 + tmp2 + tmp3 -
2448 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2449 tmp13 = tmp10 + tmp11 + tmp12 -
2450 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2451 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2452 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2453 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2454 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2455 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2456 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2457 z2 += z4;
2458 z1 = MULTIPLY(z2, -FIX(0.666655658)); /* -c11 */
2459 tmp1 += z1;
2460 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2461 z2 = MULTIPLY(z2, -FIX(1.247225013)); /* -c5 */
2462 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2463 tmp12 += z2;
2464 z2 = MULTIPLY(z3 + z4, -FIX(1.353318001)); /* -c3 */
2465 tmp2 += z2;
2466 tmp3 += z2;
2467 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2468 tmp10 += z2;
2469 tmp11 += z2;
2470
2471 /* Final output stage */
2472
2473 wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS - PASS1_BITS);
2474 wsptr[8 * 15] = (int)RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS - PASS1_BITS);
2475 wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS - PASS1_BITS);
2476 wsptr[8 * 14] = (int)RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS - PASS1_BITS);
2477 wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS - PASS1_BITS);
2478 wsptr[8 * 13] = (int)RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS - PASS1_BITS);
2479 wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS - PASS1_BITS);
2480 wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS - PASS1_BITS);
2481 wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS - PASS1_BITS);
2482 wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS - PASS1_BITS);
2483 wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS - PASS1_BITS);
2484 wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS - PASS1_BITS);
2485 wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS - PASS1_BITS);
2486 wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS - PASS1_BITS);
2487 wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS - PASS1_BITS);
2488 wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS - PASS1_BITS);
2489 }
2490
2491 /* Pass 2: process 16 rows from work array, store into output array. */
2492
2493 wsptr = workspace;
2494 for (ctr = 0; ctr < 16; ctr++) {
2495 outptr = output_buf[ctr] + output_col;
2496
2497 /* Even part */
2498
2499 /* Add fudge factor here for final descale. */
2500 tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2));
2501 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
2502
2503 z1 = (JLONG)wsptr[4];
2504 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2505 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2506
2507 tmp10 = tmp0 + tmp1;
2508 tmp11 = tmp0 - tmp1;
2509 tmp12 = tmp0 + tmp2;
2510 tmp13 = tmp0 - tmp2;
2511
2512 z1 = (JLONG)wsptr[2];
2513 z2 = (JLONG)wsptr[6];
2514 z3 = z1 - z2;
2515 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2516 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2517
2518 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2519 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2520 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2521 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2522
2523 tmp20 = tmp10 + tmp0;
2524 tmp27 = tmp10 - tmp0;
2525 tmp21 = tmp12 + tmp1;
2526 tmp26 = tmp12 - tmp1;
2527 tmp22 = tmp13 + tmp2;
2528 tmp25 = tmp13 - tmp2;
2529 tmp23 = tmp11 + tmp3;
2530 tmp24 = tmp11 - tmp3;
2531
2532 /* Odd part */
2533
2534 z1 = (JLONG)wsptr[1];
2535 z2 = (JLONG)wsptr[3];
2536 z3 = (JLONG)wsptr[5];
2537 z4 = (JLONG)wsptr[7];
2538
2539 tmp11 = z1 + z3;
2540
2541 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2542 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2543 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2544 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2545 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2546 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2547 tmp0 = tmp1 + tmp2 + tmp3 -
2548 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2549 tmp13 = tmp10 + tmp11 + tmp12 -
2550 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2551 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2552 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2553 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2554 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2555 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2556 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2557 z2 += z4;
2558 z1 = MULTIPLY(z2, -FIX(0.666655658)); /* -c11 */
2559 tmp1 += z1;
2560 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2561 z2 = MULTIPLY(z2, -FIX(1.247225013)); /* -c5 */
2562 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2563 tmp12 += z2;
2564 z2 = MULTIPLY(z3 + z4, -FIX(1.353318001)); /* -c3 */
2565 tmp2 += z2;
2566 tmp3 += z2;
2567 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2568 tmp10 += z2;
2569 tmp11 += z2;
2570
2571 /* Final output stage */
2572
2573 outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp0,
2574 CONST_BITS + PASS1_BITS + 3) &
2575 RANGE_MASK];
2576 outptr[15] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp0,
2577 CONST_BITS + PASS1_BITS + 3) &
2578 RANGE_MASK];
2579 outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp1,
2580 CONST_BITS + PASS1_BITS + 3) &
2581 RANGE_MASK];
2582 outptr[14] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp1,
2583 CONST_BITS + PASS1_BITS + 3) &
2584 RANGE_MASK];
2585 outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp2,
2586 CONST_BITS + PASS1_BITS + 3) &
2587 RANGE_MASK];
2588 outptr[13] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp2,
2589 CONST_BITS + PASS1_BITS + 3) &
2590 RANGE_MASK];
2591 outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp3,
2592 CONST_BITS + PASS1_BITS + 3) &
2593 RANGE_MASK];
2594 outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp3,
2595 CONST_BITS + PASS1_BITS + 3) &
2596 RANGE_MASK];
2597 outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp10,
2598 CONST_BITS + PASS1_BITS + 3) &
2599 RANGE_MASK];
2600 outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp10,
2601 CONST_BITS + PASS1_BITS + 3) &
2602 RANGE_MASK];
2603 outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp11,
2604 CONST_BITS + PASS1_BITS + 3) &
2605 RANGE_MASK];
2606 outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp11,
2607 CONST_BITS + PASS1_BITS + 3) &
2608 RANGE_MASK];
2609 outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26 + tmp12,
2610 CONST_BITS + PASS1_BITS + 3) &
2611 RANGE_MASK];
2612 outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp26 - tmp12,
2613 CONST_BITS + PASS1_BITS + 3) &
2614 RANGE_MASK];
2615 outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp27 + tmp13,
2616 CONST_BITS + PASS1_BITS + 3) &
2617 RANGE_MASK];
2618 outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp27 - tmp13,
2619 CONST_BITS + PASS1_BITS + 3) &
2620 RANGE_MASK];
2621
2622 wsptr += 8; /* advance pointer to next row */
2623 }
2624 }
2625
2626 #endif /* IDCT_SCALING_SUPPORTED */
2627 #endif /* DCT_ISLOW_SUPPORTED */
2628