4 * Copyright (C) 1991-1998, Thomas G. Lane.
5 * Modification developed 2002-2013 by Guido Vollbeding.
6 * This file is part of the Independent JPEG Group's software.
7 * For conditions of distribution and use, see the accompanying README file.
9 * This file contains a slow-but-accurate integer implementation of the
10 * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine
11 * must also perform dequantization of the input coefficients.
13 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
14 * on each row (or vice versa, but it's more convenient to emit a row at
15 * a time). Direct algorithms are also available, but they are much more
16 * complex and seem not to be any faster when reduced to code.
18 * This implementation is based on an algorithm described in
19 * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
20 * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
21 * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
22 * The primary algorithm described there uses 11 multiplies and 29 adds.
23 * We use their alternate method with 12 multiplies and 32 adds.
24 * The advantage of this method is that no data path contains more than one
25 * multiplication; this allows a very simple and accurate implementation in
26 * scaled fixed-point arithmetic, with a minimal number of shifts.
28 * We also provide IDCT routines with various output sample block sizes for
29 * direct resolution reduction or enlargement and for direct resolving the
30 * common 2x1 and 1x2 subsampling cases without additional resampling: NxN
31 * (N=1...16), 2NxN, and Nx2N (N=1...8) pixels for one 8x8 input DCT block.
33 * For N<8 we simply take the corresponding low-frequency coefficients of
34 * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block
35 * to yield the downscaled outputs.
36 * This can be seen as direct low-pass downsampling from the DCT domain
37 * point of view rather than the usual spatial domain point of view,
38 * yielding significant computational savings and results at least
39 * as good as common bilinear (averaging) spatial downsampling.
41 * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as
42 * lower frequencies and higher frequencies assumed to be zero.
43 * It turns out that the computational effort is similar to the 8x8 IDCT
44 * regarding the output size.
45 * Furthermore, the scaling and descaling is the same for all IDCT sizes.
47 * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases
48 * since there would be too many additional constants to pre-calculate.
51 #define JPEG_INTERNALS
54 #include "jdct.h" /* Private declarations for DCT subsystem */
56 #ifdef DCT_ISLOW_SUPPORTED
60 * This module is specialized to the case DCTSIZE = 8.
64 Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
69 * The poop on this scaling stuff is as follows:
71 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
72 * larger than the true IDCT outputs. The final outputs are therefore
73 * a factor of N larger than desired; since N=8 this can be cured by
74 * a simple right shift at the end of the algorithm. The advantage of
75 * this arrangement is that we save two multiplications per 1-D IDCT,
76 * because the y0 and y4 inputs need not be divided by sqrt(N).
78 * We have to do addition and subtraction of the integer inputs, which
79 * is no problem, and multiplication by fractional constants, which is
80 * a problem to do in integer arithmetic. We multiply all the constants
81 * by CONST_SCALE and convert them to integer constants (thus retaining
82 * CONST_BITS bits of precision in the constants). After doing a
83 * multiplication we have to divide the product by CONST_SCALE, with proper
84 * rounding, to produce the correct output. This division can be done
85 * cheaply as a right shift of CONST_BITS bits. We postpone shifting
86 * as long as possible so that partial sums can be added together with
87 * full fractional precision.
89 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
90 * they are represented to better-than-integral precision. These outputs
91 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
92 * with the recommended scaling. (To scale up 12-bit sample data further, an
93 * intermediate INT32 array would be needed.)
95 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
96 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
97 * shows that the values given below are the most effective.
100 #if BITS_IN_JSAMPLE == 8
101 #define CONST_BITS 13
104 #define CONST_BITS 13
105 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
108 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
109 * causing a lot of useless floating-point operations at run time.
110 * To get around this we use the following pre-calculated constants.
111 * If you change CONST_BITS you may want to add appropriate values.
112 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
116 #define FIX_0_298631336 ((INT32) 2446) /* FIX(0.298631336) */
117 #define FIX_0_390180644 ((INT32) 3196) /* FIX(0.390180644) */
118 #define FIX_0_541196100 ((INT32) 4433) /* FIX(0.541196100) */
119 #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
120 #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
121 #define FIX_1_175875602 ((INT32) 9633) /* FIX(1.175875602) */
122 #define FIX_1_501321110 ((INT32) 12299) /* FIX(1.501321110) */
123 #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
124 #define FIX_1_961570560 ((INT32) 16069) /* FIX(1.961570560) */
125 #define FIX_2_053119869 ((INT32) 16819) /* FIX(2.053119869) */
126 #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
127 #define FIX_3_072711026 ((INT32) 25172) /* FIX(3.072711026) */
129 #define FIX_0_298631336 FIX(0.298631336)
130 #define FIX_0_390180644 FIX(0.390180644)
131 #define FIX_0_541196100 FIX(0.541196100)
132 #define FIX_0_765366865 FIX(0.765366865)
133 #define FIX_0_899976223 FIX(0.899976223)
134 #define FIX_1_175875602 FIX(1.175875602)
135 #define FIX_1_501321110 FIX(1.501321110)
136 #define FIX_1_847759065 FIX(1.847759065)
137 #define FIX_1_961570560 FIX(1.961570560)
138 #define FIX_2_053119869 FIX(2.053119869)
139 #define FIX_2_562915447 FIX(2.562915447)
140 #define FIX_3_072711026 FIX(3.072711026)
144 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
145 * For 8-bit samples with the recommended scaling, all the variable
146 * and constant values involved are no more than 16 bits wide, so a
147 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
148 * For 12-bit samples, a full 32-bit multiplication will be needed.
151 #if BITS_IN_JSAMPLE == 8
152 #define MULTIPLY(var,const) MULTIPLY16C16(var,const)
154 #define MULTIPLY(var,const) ((var) * (const))
158 /* Dequantize a coefficient by multiplying it by the multiplier-table
159 * entry; produce an int result. In this module, both inputs and result
160 * are 16 bits or less, so either int or short multiply will work.
163 #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))
167 * Perform dequantization and inverse DCT on one block of coefficients.
169 * cK represents sqrt(2) * cos(K*pi/16).
173 jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
175 JSAMPARRAY output_buf, JDIMENSION output_col)
177 INT32 tmp0, tmp1, tmp2, tmp3;
178 INT32 tmp10, tmp11, tmp12, tmp13;
181 ISLOW_MULT_TYPE * quantptr;
184 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
186 int workspace[DCTSIZE2]; /* buffers data between passes */
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.
195 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
197 for (ctr = DCTSIZE; ctr > 0; ctr--) {
198 /* Due to quantization, we will usually find that many of the input
199 * coefficients are zero, especially the AC terms. We can exploit this
200 * by short-circuiting the IDCT calculation for any column in which all
201 * the AC terms are zero. In that case each output is equal to the
202 * DC coefficient (with scale factor as needed).
203 * With typical images and quantization tables, half or more of the
204 * column DCT calculations can be simplified this way.
207 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
208 inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
209 inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
210 inptr[DCTSIZE*7] == 0) {
211 /* AC terms all zero */
212 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
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;
223 inptr++; /* advance pointers to next column */
229 /* Even part: reverse the even part of the forward DCT.
230 * The rotator is c(-6).
233 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
234 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
236 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
237 tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
238 tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
240 z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
241 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
244 /* Add fudge factor here for final descale. */
245 z2 += ONE << (CONST_BITS-PASS1_BITS-1);
255 /* Odd part per figure 8; the matrix is unitary and hence its
256 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
259 tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
260 tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
261 tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
262 tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
267 z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* c3 */
268 z2 = MULTIPLY(z2, - FIX_1_961570560); /* -c3-c5 */
269 z3 = MULTIPLY(z3, - FIX_0_390180644); /* -c3+c5 */
273 z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
274 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* -c1+c3+c5-c7 */
275 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* c1+c3-c5-c7 */
279 z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
280 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* c1+c3-c5+c7 */
281 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* c1+c3+c5-c7 */
285 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
287 wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
288 wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
289 wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
290 wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
291 wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
292 wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
293 wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
294 wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
296 inptr++; /* advance pointers to next column */
301 /* Pass 2: process rows from work array, store into output array.
302 * Note that we must descale the results by a factor of 8 == 2**3,
303 * and also undo the PASS1_BITS scaling.
307 for (ctr = 0; ctr < DCTSIZE; ctr++) {
308 outptr = output_buf[ctr] + output_col;
309 /* Rows of zeroes can be exploited in the same way as we did with columns.
310 * However, the column calculation has created many nonzero AC terms, so
311 * the simplification applies less often (typically 5% to 10% of the time).
312 * On machines with very fast multiplication, it's possible that the
313 * test takes more time than it's worth. In that case this section
314 * may be commented out.
317 #ifndef NO_ZERO_ROW_TEST
318 if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
319 wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
320 /* AC terms all zero */
321 JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
333 wsptr += DCTSIZE; /* advance pointer to next row */
338 /* Even part: reverse the even part of the forward DCT.
339 * The rotator is c(-6).
342 z2 = (INT32) wsptr[2];
343 z3 = (INT32) wsptr[6];
345 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
346 tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
347 tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
349 /* Add fudge factor here for final descale. */
350 z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
351 z3 = (INT32) wsptr[4];
353 tmp0 = (z2 + z3) << CONST_BITS;
354 tmp1 = (z2 - z3) << CONST_BITS;
361 /* Odd part per figure 8; the matrix is unitary and hence its
362 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
365 tmp0 = (INT32) wsptr[7];
366 tmp1 = (INT32) wsptr[5];
367 tmp2 = (INT32) wsptr[3];
368 tmp3 = (INT32) wsptr[1];
373 z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* c3 */
374 z2 = MULTIPLY(z2, - FIX_1_961570560); /* -c3-c5 */
375 z3 = MULTIPLY(z3, - FIX_0_390180644); /* -c3+c5 */
379 z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
380 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* -c1+c3+c5-c7 */
381 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* c1+c3-c5-c7 */
385 z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
386 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* c1+c3-c5+c7 */
387 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* c1+c3+c5-c7 */
391 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
393 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
394 CONST_BITS+PASS1_BITS+3)
396 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
397 CONST_BITS+PASS1_BITS+3)
399 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
400 CONST_BITS+PASS1_BITS+3)
402 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
403 CONST_BITS+PASS1_BITS+3)
405 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
406 CONST_BITS+PASS1_BITS+3)
408 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
409 CONST_BITS+PASS1_BITS+3)
411 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
412 CONST_BITS+PASS1_BITS+3)
414 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
415 CONST_BITS+PASS1_BITS+3)
418 wsptr += DCTSIZE; /* advance pointer to next row */
422 #ifdef IDCT_SCALING_SUPPORTED
426 * Perform dequantization and inverse DCT on one block of coefficients,
427 * producing a 7x7 output block.
429 * Optimized algorithm with 12 multiplications in the 1-D kernel.
430 * cK represents sqrt(2) * cos(K*pi/14).
434 jpeg_idct_7x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
436 JSAMPARRAY output_buf, JDIMENSION output_col)
438 INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
441 ISLOW_MULT_TYPE * quantptr;
444 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
446 int workspace[7*7]; /* buffers data between passes */
449 /* Pass 1: process columns from input, store into work array. */
452 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
454 for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
457 tmp13 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
458 tmp13 <<= CONST_BITS;
459 /* Add fudge factor here for final descale. */
460 tmp13 += ONE << (CONST_BITS-PASS1_BITS-1);
462 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
463 z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
464 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
466 tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
467 tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
468 tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
471 tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
472 tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
473 tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
474 tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
478 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
479 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
480 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
482 tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
483 tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
486 tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
488 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
490 tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
492 /* Final output stage */
494 wsptr[7*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
495 wsptr[7*6] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
496 wsptr[7*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
497 wsptr[7*5] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
498 wsptr[7*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
499 wsptr[7*4] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
500 wsptr[7*3] = (int) RIGHT_SHIFT(tmp13, CONST_BITS-PASS1_BITS);
503 /* Pass 2: process 7 rows from work array, store into output array. */
506 for (ctr = 0; ctr < 7; ctr++) {
507 outptr = output_buf[ctr] + output_col;
511 /* Add fudge factor here for final descale. */
512 tmp13 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
513 tmp13 <<= CONST_BITS;
515 z1 = (INT32) wsptr[2];
516 z2 = (INT32) wsptr[4];
517 z3 = (INT32) wsptr[6];
519 tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
520 tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
521 tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
524 tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
525 tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
526 tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
527 tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
531 z1 = (INT32) wsptr[1];
532 z2 = (INT32) wsptr[3];
533 z3 = (INT32) wsptr[5];
535 tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
536 tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
539 tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
541 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
543 tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
545 /* Final output stage */
547 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
548 CONST_BITS+PASS1_BITS+3)
550 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
551 CONST_BITS+PASS1_BITS+3)
553 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
554 CONST_BITS+PASS1_BITS+3)
556 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
557 CONST_BITS+PASS1_BITS+3)
559 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
560 CONST_BITS+PASS1_BITS+3)
562 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
563 CONST_BITS+PASS1_BITS+3)
565 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13,
566 CONST_BITS+PASS1_BITS+3)
569 wsptr += 7; /* advance pointer to next row */
575 * Perform dequantization and inverse DCT on one block of coefficients,
576 * producing a reduced-size 6x6 output block.
578 * Optimized algorithm with 3 multiplications in the 1-D kernel.
579 * cK represents sqrt(2) * cos(K*pi/12).
583 jpeg_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
585 JSAMPARRAY output_buf, JDIMENSION output_col)
587 INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
590 ISLOW_MULT_TYPE * quantptr;
593 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
595 int workspace[6*6]; /* buffers data between passes */
598 /* Pass 1: process columns from input, store into work array. */
601 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
603 for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
606 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
608 /* Add fudge factor here for final descale. */
609 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
610 tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
611 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
613 tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
614 tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
615 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
621 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
622 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
623 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
624 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
625 tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
626 tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
627 tmp1 = (z1 - z2 - z3) << PASS1_BITS;
629 /* Final output stage */
631 wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
632 wsptr[6*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
633 wsptr[6*1] = (int) (tmp11 + tmp1);
634 wsptr[6*4] = (int) (tmp11 - tmp1);
635 wsptr[6*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
636 wsptr[6*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
639 /* Pass 2: process 6 rows from work array, store into output array. */
642 for (ctr = 0; ctr < 6; ctr++) {
643 outptr = output_buf[ctr] + output_col;
647 /* Add fudge factor here for final descale. */
648 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
650 tmp2 = (INT32) wsptr[4];
651 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
653 tmp11 = tmp0 - tmp10 - tmp10;
654 tmp10 = (INT32) wsptr[2];
655 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
661 z1 = (INT32) wsptr[1];
662 z2 = (INT32) wsptr[3];
663 z3 = (INT32) wsptr[5];
664 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
665 tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
666 tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
667 tmp1 = (z1 - z2 - z3) << CONST_BITS;
669 /* Final output stage */
671 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
672 CONST_BITS+PASS1_BITS+3)
674 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
675 CONST_BITS+PASS1_BITS+3)
677 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
678 CONST_BITS+PASS1_BITS+3)
680 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
681 CONST_BITS+PASS1_BITS+3)
683 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
684 CONST_BITS+PASS1_BITS+3)
686 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
687 CONST_BITS+PASS1_BITS+3)
690 wsptr += 6; /* advance pointer to next row */
696 * Perform dequantization and inverse DCT on one block of coefficients,
697 * producing a reduced-size 5x5 output block.
699 * Optimized algorithm with 5 multiplications in the 1-D kernel.
700 * cK represents sqrt(2) * cos(K*pi/10).
704 jpeg_idct_5x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
706 JSAMPARRAY output_buf, JDIMENSION output_col)
708 INT32 tmp0, tmp1, tmp10, tmp11, tmp12;
711 ISLOW_MULT_TYPE * quantptr;
714 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
716 int workspace[5*5]; /* buffers data between passes */
719 /* Pass 1: process columns from input, store into work array. */
722 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
724 for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
727 tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
728 tmp12 <<= CONST_BITS;
729 /* Add fudge factor here for final descale. */
730 tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
731 tmp0 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
732 tmp1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
733 z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
734 z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
742 z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
743 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
745 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
746 tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
747 tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
749 /* Final output stage */
751 wsptr[5*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
752 wsptr[5*4] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
753 wsptr[5*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
754 wsptr[5*3] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
755 wsptr[5*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
758 /* Pass 2: process 5 rows from work array, store into output array. */
761 for (ctr = 0; ctr < 5; ctr++) {
762 outptr = output_buf[ctr] + output_col;
766 /* Add fudge factor here for final descale. */
767 tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
768 tmp12 <<= CONST_BITS;
769 tmp0 = (INT32) wsptr[2];
770 tmp1 = (INT32) wsptr[4];
771 z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
772 z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
780 z2 = (INT32) wsptr[1];
781 z3 = (INT32) wsptr[3];
783 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
784 tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
785 tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
787 /* Final output stage */
789 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
790 CONST_BITS+PASS1_BITS+3)
792 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
793 CONST_BITS+PASS1_BITS+3)
795 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
796 CONST_BITS+PASS1_BITS+3)
798 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
799 CONST_BITS+PASS1_BITS+3)
801 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
802 CONST_BITS+PASS1_BITS+3)
805 wsptr += 5; /* advance pointer to next row */
811 * Perform dequantization and inverse DCT on one block of coefficients,
812 * producing a reduced-size 4x4 output block.
814 * Optimized algorithm with 3 multiplications in the 1-D kernel.
815 * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
819 jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
821 JSAMPARRAY output_buf, JDIMENSION output_col)
823 INT32 tmp0, tmp2, tmp10, tmp12;
826 ISLOW_MULT_TYPE * quantptr;
829 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
831 int workspace[4*4]; /* buffers data between passes */
834 /* Pass 1: process columns from input, store into work array. */
837 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
839 for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
842 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
843 tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
845 tmp10 = (tmp0 + tmp2) << PASS1_BITS;
846 tmp12 = (tmp0 - tmp2) << PASS1_BITS;
849 /* Same rotation as in the even part of the 8x8 LL&M IDCT */
851 z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
852 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
854 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
855 /* Add fudge factor here for final descale. */
856 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
857 tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
858 CONST_BITS-PASS1_BITS);
859 tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
860 CONST_BITS-PASS1_BITS);
862 /* Final output stage */
864 wsptr[4*0] = (int) (tmp10 + tmp0);
865 wsptr[4*3] = (int) (tmp10 - tmp0);
866 wsptr[4*1] = (int) (tmp12 + tmp2);
867 wsptr[4*2] = (int) (tmp12 - tmp2);
870 /* Pass 2: process 4 rows from work array, store into output array. */
873 for (ctr = 0; ctr < 4; ctr++) {
874 outptr = output_buf[ctr] + output_col;
878 /* Add fudge factor here for final descale. */
879 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
880 tmp2 = (INT32) wsptr[2];
882 tmp10 = (tmp0 + tmp2) << CONST_BITS;
883 tmp12 = (tmp0 - tmp2) << CONST_BITS;
886 /* Same rotation as in the even part of the 8x8 LL&M IDCT */
888 z2 = (INT32) wsptr[1];
889 z3 = (INT32) wsptr[3];
891 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
892 tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
893 tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
895 /* Final output stage */
897 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
898 CONST_BITS+PASS1_BITS+3)
900 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
901 CONST_BITS+PASS1_BITS+3)
903 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
904 CONST_BITS+PASS1_BITS+3)
906 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
907 CONST_BITS+PASS1_BITS+3)
910 wsptr += 4; /* advance pointer to next row */
916 * Perform dequantization and inverse DCT on one block of coefficients,
917 * producing a reduced-size 3x3 output block.
919 * Optimized algorithm with 2 multiplications in the 1-D kernel.
920 * cK represents sqrt(2) * cos(K*pi/6).
924 jpeg_idct_3x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
926 JSAMPARRAY output_buf, JDIMENSION output_col)
928 INT32 tmp0, tmp2, tmp10, tmp12;
930 ISLOW_MULT_TYPE * quantptr;
933 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
935 int workspace[3*3]; /* buffers data between passes */
938 /* Pass 1: process columns from input, store into work array. */
941 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
943 for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
946 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
948 /* Add fudge factor here for final descale. */
949 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
950 tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
951 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
952 tmp10 = tmp0 + tmp12;
953 tmp2 = tmp0 - tmp12 - tmp12;
957 tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
958 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
960 /* Final output stage */
962 wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
963 wsptr[3*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
964 wsptr[3*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
967 /* Pass 2: process 3 rows from work array, store into output array. */
970 for (ctr = 0; ctr < 3; ctr++) {
971 outptr = output_buf[ctr] + output_col;
975 /* Add fudge factor here for final descale. */
976 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
978 tmp2 = (INT32) wsptr[2];
979 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
980 tmp10 = tmp0 + tmp12;
981 tmp2 = tmp0 - tmp12 - tmp12;
985 tmp12 = (INT32) wsptr[1];
986 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
988 /* Final output stage */
990 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
991 CONST_BITS+PASS1_BITS+3)
993 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
994 CONST_BITS+PASS1_BITS+3)
996 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
997 CONST_BITS+PASS1_BITS+3)
1000 wsptr += 3; /* advance pointer to next row */
1006 * Perform dequantization and inverse DCT on one block of coefficients,
1007 * producing a reduced-size 2x2 output block.
1009 * Multiplication-less algorithm.
1013 jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1014 JCOEFPTR coef_block,
1015 JSAMPARRAY output_buf, JDIMENSION output_col)
1017 INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
1018 ISLOW_MULT_TYPE * quantptr;
1020 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1023 /* Pass 1: process columns from input. */
1025 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1028 tmp4 = DEQUANTIZE(coef_block[DCTSIZE*0], quantptr[DCTSIZE*0]);
1029 tmp5 = DEQUANTIZE(coef_block[DCTSIZE*1], quantptr[DCTSIZE*1]);
1030 /* Add fudge factor here for final descale. */
1037 tmp4 = DEQUANTIZE(coef_block[DCTSIZE*0+1], quantptr[DCTSIZE*0+1]);
1038 tmp5 = DEQUANTIZE(coef_block[DCTSIZE*1+1], quantptr[DCTSIZE*1+1]);
1043 /* Pass 2: process 2 rows, store into output array. */
1046 outptr = output_buf[0] + output_col;
1048 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp0 + tmp1, 3) & RANGE_MASK];
1049 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp0 - tmp1, 3) & RANGE_MASK];
1052 outptr = output_buf[1] + output_col;
1054 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp2 + tmp3, 3) & RANGE_MASK];
1055 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2 - tmp3, 3) & RANGE_MASK];
1060 * Perform dequantization and inverse DCT on one block of coefficients,
1061 * producing a reduced-size 1x1 output block.
1063 * We hardly need an inverse DCT routine for this: just take the
1064 * average pixel value, which is one-eighth of the DC coefficient.
1068 jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1069 JCOEFPTR coef_block,
1070 JSAMPARRAY output_buf, JDIMENSION output_col)
1073 ISLOW_MULT_TYPE * quantptr;
1074 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1077 /* 1x1 is trivial: just take the DC coefficient divided by 8. */
1078 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1079 dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
1080 dcval = (int) DESCALE((INT32) dcval, 3);
1082 output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
1087 * Perform dequantization and inverse DCT on one block of coefficients,
1088 * producing a 9x9 output block.
1090 * Optimized algorithm with 10 multiplications in the 1-D kernel.
1091 * cK represents sqrt(2) * cos(K*pi/18).
1095 jpeg_idct_9x9 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1096 JCOEFPTR coef_block,
1097 JSAMPARRAY output_buf, JDIMENSION output_col)
1099 INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14;
1100 INT32 z1, z2, z3, z4;
1102 ISLOW_MULT_TYPE * quantptr;
1105 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1107 int workspace[8*9]; /* buffers data between passes */
1110 /* Pass 1: process columns from input, store into work array. */
1113 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1115 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1118 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1119 tmp0 <<= CONST_BITS;
1120 /* Add fudge factor here for final descale. */
1121 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
1123 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1124 z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1125 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1127 tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
1129 tmp2 = tmp0 - tmp3 - tmp3;
1131 tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1132 tmp11 = tmp2 + tmp0;
1133 tmp14 = tmp2 - tmp0 - tmp0;
1135 tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1136 tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1137 tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1139 tmp10 = tmp1 + tmp0 - tmp3;
1140 tmp12 = tmp1 - tmp0 + tmp2;
1141 tmp13 = tmp1 - tmp2 + tmp3;
1145 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1146 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1147 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1148 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1150 z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
1152 tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1153 tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1154 tmp0 = tmp2 + tmp3 - z2;
1155 tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1158 tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1160 /* Final output stage */
1162 wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
1163 wsptr[8*8] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
1164 wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
1165 wsptr[8*7] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
1166 wsptr[8*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
1167 wsptr[8*6] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
1168 wsptr[8*3] = (int) RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS-PASS1_BITS);
1169 wsptr[8*5] = (int) RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS-PASS1_BITS);
1170 wsptr[8*4] = (int) RIGHT_SHIFT(tmp14, CONST_BITS-PASS1_BITS);
1173 /* Pass 2: process 9 rows from work array, store into output array. */
1176 for (ctr = 0; ctr < 9; ctr++) {
1177 outptr = output_buf[ctr] + output_col;
1181 /* Add fudge factor here for final descale. */
1182 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1183 tmp0 <<= CONST_BITS;
1185 z1 = (INT32) wsptr[2];
1186 z2 = (INT32) wsptr[4];
1187 z3 = (INT32) wsptr[6];
1189 tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
1191 tmp2 = tmp0 - tmp3 - tmp3;
1193 tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1194 tmp11 = tmp2 + tmp0;
1195 tmp14 = tmp2 - tmp0 - tmp0;
1197 tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1198 tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1199 tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1201 tmp10 = tmp1 + tmp0 - tmp3;
1202 tmp12 = tmp1 - tmp0 + tmp2;
1203 tmp13 = tmp1 - tmp2 + tmp3;
1207 z1 = (INT32) wsptr[1];
1208 z2 = (INT32) wsptr[3];
1209 z3 = (INT32) wsptr[5];
1210 z4 = (INT32) wsptr[7];
1212 z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
1214 tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1215 tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1216 tmp0 = tmp2 + tmp3 - z2;
1217 tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1220 tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1222 /* Final output stage */
1224 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1225 CONST_BITS+PASS1_BITS+3)
1227 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1228 CONST_BITS+PASS1_BITS+3)
1230 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
1231 CONST_BITS+PASS1_BITS+3)
1233 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
1234 CONST_BITS+PASS1_BITS+3)
1236 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
1237 CONST_BITS+PASS1_BITS+3)
1239 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
1240 CONST_BITS+PASS1_BITS+3)
1242 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp3,
1243 CONST_BITS+PASS1_BITS+3)
1245 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp3,
1246 CONST_BITS+PASS1_BITS+3)
1248 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp14,
1249 CONST_BITS+PASS1_BITS+3)
1252 wsptr += 8; /* advance pointer to next row */
1258 * Perform dequantization and inverse DCT on one block of coefficients,
1259 * producing a 10x10 output block.
1261 * Optimized algorithm with 12 multiplications in the 1-D kernel.
1262 * cK represents sqrt(2) * cos(K*pi/20).
1266 jpeg_idct_10x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1267 JCOEFPTR coef_block,
1268 JSAMPARRAY output_buf, JDIMENSION output_col)
1270 INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1271 INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
1272 INT32 z1, z2, z3, z4, z5;
1274 ISLOW_MULT_TYPE * quantptr;
1277 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1279 int workspace[8*10]; /* buffers data between passes */
1282 /* Pass 1: process columns from input, store into work array. */
1285 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1287 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1290 z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1292 /* Add fudge factor here for final descale. */
1293 z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1294 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1295 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1296 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1300 tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1), /* c0 = (c4-c8)*2 */
1301 CONST_BITS-PASS1_BITS);
1303 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1304 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1306 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1307 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1308 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1310 tmp20 = tmp10 + tmp12;
1311 tmp24 = tmp10 - tmp12;
1312 tmp21 = tmp11 + tmp13;
1313 tmp23 = tmp11 - tmp13;
1317 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1318 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1319 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1320 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1325 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1326 z5 = z3 << CONST_BITS;
1328 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1331 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1332 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1334 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1335 z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));
1337 tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;
1339 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1340 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1342 /* Final output stage */
1344 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1345 wsptr[8*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1346 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1347 wsptr[8*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1348 wsptr[8*2] = (int) (tmp22 + tmp12);
1349 wsptr[8*7] = (int) (tmp22 - tmp12);
1350 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1351 wsptr[8*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1352 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1353 wsptr[8*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1356 /* Pass 2: process 10 rows from work array, store into output array. */
1359 for (ctr = 0; ctr < 10; ctr++) {
1360 outptr = output_buf[ctr] + output_col;
1364 /* Add fudge factor here for final descale. */
1365 z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1367 z4 = (INT32) wsptr[4];
1368 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1369 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1373 tmp22 = z3 - ((z1 - z2) << 1); /* c0 = (c4-c8)*2 */
1375 z2 = (INT32) wsptr[2];
1376 z3 = (INT32) wsptr[6];
1378 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1379 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1380 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1382 tmp20 = tmp10 + tmp12;
1383 tmp24 = tmp10 - tmp12;
1384 tmp21 = tmp11 + tmp13;
1385 tmp23 = tmp11 - tmp13;
1389 z1 = (INT32) wsptr[1];
1390 z2 = (INT32) wsptr[3];
1391 z3 = (INT32) wsptr[5];
1393 z4 = (INT32) wsptr[7];
1398 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1400 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1403 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1404 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1406 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1407 z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));
1409 tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;
1411 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1412 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1414 /* Final output stage */
1416 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1417 CONST_BITS+PASS1_BITS+3)
1419 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1420 CONST_BITS+PASS1_BITS+3)
1422 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1423 CONST_BITS+PASS1_BITS+3)
1425 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1426 CONST_BITS+PASS1_BITS+3)
1428 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1429 CONST_BITS+PASS1_BITS+3)
1431 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1432 CONST_BITS+PASS1_BITS+3)
1434 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1435 CONST_BITS+PASS1_BITS+3)
1437 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1438 CONST_BITS+PASS1_BITS+3)
1440 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1441 CONST_BITS+PASS1_BITS+3)
1443 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1444 CONST_BITS+PASS1_BITS+3)
1447 wsptr += 8; /* advance pointer to next row */
1453 * Perform dequantization and inverse DCT on one block of coefficients,
1454 * producing a 11x11 output block.
1456 * Optimized algorithm with 24 multiplications in the 1-D kernel.
1457 * cK represents sqrt(2) * cos(K*pi/22).
1461 jpeg_idct_11x11 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1462 JCOEFPTR coef_block,
1463 JSAMPARRAY output_buf, JDIMENSION output_col)
1465 INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1466 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1467 INT32 z1, z2, z3, z4;
1469 ISLOW_MULT_TYPE * quantptr;
1472 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1474 int workspace[8*11]; /* buffers data between passes */
1477 /* Pass 1: process columns from input, store into work array. */
1480 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1482 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1485 tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1486 tmp10 <<= CONST_BITS;
1487 /* Add fudge factor here for final descale. */
1488 tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
1490 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1491 z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1492 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1494 tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1495 tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1497 tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1499 tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1500 tmp21 = tmp20 + tmp23 + tmp25 -
1501 MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1502 tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1503 tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1505 tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1506 tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1507 MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1508 tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1512 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1513 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1514 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1515 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1518 tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1519 tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1520 tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1521 tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1522 tmp10 = tmp11 + tmp12 + tmp13 -
1523 MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1524 z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1525 tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1526 tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1527 z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1529 tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1530 tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1531 MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1532 MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1534 /* Final output stage */
1536 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1537 wsptr[8*10] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1538 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1539 wsptr[8*9] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1540 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1541 wsptr[8*8] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1542 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1543 wsptr[8*7] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1544 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1545 wsptr[8*6] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1546 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25, CONST_BITS-PASS1_BITS);
1549 /* Pass 2: process 11 rows from work array, store into output array. */
1552 for (ctr = 0; ctr < 11; ctr++) {
1553 outptr = output_buf[ctr] + output_col;
1557 /* Add fudge factor here for final descale. */
1558 tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1559 tmp10 <<= CONST_BITS;
1561 z1 = (INT32) wsptr[2];
1562 z2 = (INT32) wsptr[4];
1563 z3 = (INT32) wsptr[6];
1565 tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1566 tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1568 tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1570 tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1571 tmp21 = tmp20 + tmp23 + tmp25 -
1572 MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1573 tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1574 tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1576 tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1577 tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1578 MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1579 tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1583 z1 = (INT32) wsptr[1];
1584 z2 = (INT32) wsptr[3];
1585 z3 = (INT32) wsptr[5];
1586 z4 = (INT32) wsptr[7];
1589 tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1590 tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1591 tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1592 tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1593 tmp10 = tmp11 + tmp12 + tmp13 -
1594 MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1595 z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1596 tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1597 tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1598 z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1600 tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1601 tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1602 MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1603 MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1605 /* Final output stage */
1607 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1608 CONST_BITS+PASS1_BITS+3)
1610 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1611 CONST_BITS+PASS1_BITS+3)
1613 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1614 CONST_BITS+PASS1_BITS+3)
1616 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1617 CONST_BITS+PASS1_BITS+3)
1619 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1620 CONST_BITS+PASS1_BITS+3)
1622 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1623 CONST_BITS+PASS1_BITS+3)
1625 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1626 CONST_BITS+PASS1_BITS+3)
1628 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1629 CONST_BITS+PASS1_BITS+3)
1631 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1632 CONST_BITS+PASS1_BITS+3)
1634 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1635 CONST_BITS+PASS1_BITS+3)
1637 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25,
1638 CONST_BITS+PASS1_BITS+3)
1641 wsptr += 8; /* advance pointer to next row */
1647 * Perform dequantization and inverse DCT on one block of coefficients,
1648 * producing a 12x12 output block.
1650 * Optimized algorithm with 15 multiplications in the 1-D kernel.
1651 * cK represents sqrt(2) * cos(K*pi/24).
1655 jpeg_idct_12x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1656 JCOEFPTR coef_block,
1657 JSAMPARRAY output_buf, JDIMENSION output_col)
1659 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1660 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1661 INT32 z1, z2, z3, z4;
1663 ISLOW_MULT_TYPE * quantptr;
1666 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1668 int workspace[8*12]; /* buffers data between passes */
1671 /* Pass 1: process columns from input, store into work array. */
1674 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1676 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1679 z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1681 /* Add fudge factor here for final descale. */
1682 z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1684 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1685 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1690 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1691 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1693 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1703 tmp20 = tmp10 + tmp12;
1704 tmp25 = tmp10 - tmp12;
1706 tmp12 = z4 - z1 - z2;
1708 tmp22 = tmp11 + tmp12;
1709 tmp23 = tmp11 - tmp12;
1713 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1714 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1715 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1716 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1718 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1719 tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1722 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1723 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1724 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1725 tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1726 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1727 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1728 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1729 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1733 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1734 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1735 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1737 /* Final output stage */
1739 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1740 wsptr[8*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1741 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1742 wsptr[8*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1743 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1744 wsptr[8*9] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1745 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1746 wsptr[8*8] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1747 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1748 wsptr[8*7] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1749 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1750 wsptr[8*6] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1753 /* Pass 2: process 12 rows from work array, store into output array. */
1756 for (ctr = 0; ctr < 12; ctr++) {
1757 outptr = output_buf[ctr] + output_col;
1761 /* Add fudge factor here for final descale. */
1762 z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1765 z4 = (INT32) wsptr[4];
1766 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1771 z1 = (INT32) wsptr[2];
1772 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1774 z2 = (INT32) wsptr[6];
1784 tmp20 = tmp10 + tmp12;
1785 tmp25 = tmp10 - tmp12;
1787 tmp12 = z4 - z1 - z2;
1789 tmp22 = tmp11 + tmp12;
1790 tmp23 = tmp11 - tmp12;
1794 z1 = (INT32) wsptr[1];
1795 z2 = (INT32) wsptr[3];
1796 z3 = (INT32) wsptr[5];
1797 z4 = (INT32) wsptr[7];
1799 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1800 tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1803 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1804 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1805 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1806 tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1807 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1808 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1809 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1810 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1814 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1815 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1816 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1818 /* Final output stage */
1820 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1821 CONST_BITS+PASS1_BITS+3)
1823 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1824 CONST_BITS+PASS1_BITS+3)
1826 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1827 CONST_BITS+PASS1_BITS+3)
1829 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1830 CONST_BITS+PASS1_BITS+3)
1832 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1833 CONST_BITS+PASS1_BITS+3)
1835 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1836 CONST_BITS+PASS1_BITS+3)
1838 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1839 CONST_BITS+PASS1_BITS+3)
1841 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1842 CONST_BITS+PASS1_BITS+3)
1844 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1845 CONST_BITS+PASS1_BITS+3)
1847 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1848 CONST_BITS+PASS1_BITS+3)
1850 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1851 CONST_BITS+PASS1_BITS+3)
1853 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1854 CONST_BITS+PASS1_BITS+3)
1857 wsptr += 8; /* advance pointer to next row */
1863 * Perform dequantization and inverse DCT on one block of coefficients,
1864 * producing a 13x13 output block.
1866 * Optimized algorithm with 29 multiplications in the 1-D kernel.
1867 * cK represents sqrt(2) * cos(K*pi/26).
1871 jpeg_idct_13x13 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1872 JCOEFPTR coef_block,
1873 JSAMPARRAY output_buf, JDIMENSION output_col)
1875 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1876 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1877 INT32 z1, z2, z3, z4;
1879 ISLOW_MULT_TYPE * quantptr;
1882 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1884 int workspace[8*13]; /* buffers data between passes */
1887 /* Pass 1: process columns from input, store into work array. */
1890 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1892 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1895 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1897 /* Add fudge factor here for final descale. */
1898 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
1900 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1901 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1902 z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1907 tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1908 tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1910 tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1911 tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1913 tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1914 tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1916 tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1917 tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1919 tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1920 tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1922 tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1923 tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1925 tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1929 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1930 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1931 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1932 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1934 tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1935 tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1937 tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1938 tmp10 = tmp11 + tmp12 + tmp13 -
1939 MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1940 tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
1941 tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1942 tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1943 tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
1945 tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1946 tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
1949 tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1950 tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1951 MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1952 z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1954 tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1955 MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1957 /* Final output stage */
1959 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1960 wsptr[8*12] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1961 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1962 wsptr[8*11] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1963 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1964 wsptr[8*10] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1965 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1966 wsptr[8*9] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1967 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1968 wsptr[8*8] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1969 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1970 wsptr[8*7] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1971 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26, CONST_BITS-PASS1_BITS);
1974 /* Pass 2: process 13 rows from work array, store into output array. */
1977 for (ctr = 0; ctr < 13; ctr++) {
1978 outptr = output_buf[ctr] + output_col;
1982 /* Add fudge factor here for final descale. */
1983 z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1986 z2 = (INT32) wsptr[2];
1987 z3 = (INT32) wsptr[4];
1988 z4 = (INT32) wsptr[6];
1993 tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1994 tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1996 tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1997 tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1999 tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
2000 tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
2002 tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
2003 tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
2005 tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
2006 tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
2008 tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
2009 tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
2011 tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
2015 z1 = (INT32) wsptr[1];
2016 z2 = (INT32) wsptr[3];
2017 z3 = (INT32) wsptr[5];
2018 z4 = (INT32) wsptr[7];
2020 tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
2021 tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
2023 tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
2024 tmp10 = tmp11 + tmp12 + tmp13 -
2025 MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
2026 tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
2027 tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
2028 tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
2029 tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
2031 tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
2032 tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
2035 tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
2036 tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
2037 MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
2038 z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
2040 tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
2041 MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
2043 /* Final output stage */
2045 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2046 CONST_BITS+PASS1_BITS+3)
2048 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2049 CONST_BITS+PASS1_BITS+3)
2051 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2052 CONST_BITS+PASS1_BITS+3)
2054 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2055 CONST_BITS+PASS1_BITS+3)
2057 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2058 CONST_BITS+PASS1_BITS+3)
2060 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2061 CONST_BITS+PASS1_BITS+3)
2063 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2064 CONST_BITS+PASS1_BITS+3)
2066 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2067 CONST_BITS+PASS1_BITS+3)
2069 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2070 CONST_BITS+PASS1_BITS+3)
2072 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2073 CONST_BITS+PASS1_BITS+3)
2075 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2076 CONST_BITS+PASS1_BITS+3)
2078 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2079 CONST_BITS+PASS1_BITS+3)
2081 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26,
2082 CONST_BITS+PASS1_BITS+3)
2085 wsptr += 8; /* advance pointer to next row */
2091 * Perform dequantization and inverse DCT on one block of coefficients,
2092 * producing a 14x14 output block.
2094 * Optimized algorithm with 20 multiplications in the 1-D kernel.
2095 * cK represents sqrt(2) * cos(K*pi/28).
2099 jpeg_idct_14x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2100 JCOEFPTR coef_block,
2101 JSAMPARRAY output_buf, JDIMENSION output_col)
2103 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
2104 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
2105 INT32 z1, z2, z3, z4;
2107 ISLOW_MULT_TYPE * quantptr;
2110 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2112 int workspace[8*14]; /* buffers data between passes */
2115 /* Pass 1: process columns from input, store into work array. */
2118 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2120 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2123 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
2125 /* Add fudge factor here for final descale. */
2126 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
2127 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
2128 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
2129 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
2130 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
2136 tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)*2 */
2137 CONST_BITS-PASS1_BITS);
2139 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
2140 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
2142 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
2144 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
2145 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
2146 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
2147 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
2149 tmp20 = tmp10 + tmp13;
2150 tmp26 = tmp10 - tmp13;
2151 tmp21 = tmp11 + tmp14;
2152 tmp25 = tmp11 - tmp14;
2153 tmp22 = tmp12 + tmp15;
2154 tmp24 = tmp12 - tmp15;
2158 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
2159 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
2160 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
2161 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
2162 tmp13 = z4 << CONST_BITS;
2165 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
2166 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
2167 tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
2168 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
2169 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
2171 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */
2174 z4 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
2175 tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
2176 tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
2177 z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
2178 tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
2179 tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
2181 tmp13 = (z1 - z3) << PASS1_BITS;
2183 /* Final output stage */
2185 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
2186 wsptr[8*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
2187 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
2188 wsptr[8*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
2189 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
2190 wsptr[8*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
2191 wsptr[8*3] = (int) (tmp23 + tmp13);
2192 wsptr[8*10] = (int) (tmp23 - tmp13);
2193 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
2194 wsptr[8*9] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
2195 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
2196 wsptr[8*8] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
2197 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
2198 wsptr[8*7] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
2201 /* Pass 2: process 14 rows from work array, store into output array. */
2204 for (ctr = 0; ctr < 14; ctr++) {
2205 outptr = output_buf[ctr] + output_col;
2209 /* Add fudge factor here for final descale. */
2210 z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2212 z4 = (INT32) wsptr[4];
2213 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
2214 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
2215 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
2221 tmp23 = z1 - ((z2 + z3 - z4) << 1); /* c0 = (c4+c12-c8)*2 */
2223 z1 = (INT32) wsptr[2];
2224 z2 = (INT32) wsptr[6];
2226 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
2228 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
2229 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
2230 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
2231 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
2233 tmp20 = tmp10 + tmp13;
2234 tmp26 = tmp10 - tmp13;
2235 tmp21 = tmp11 + tmp14;
2236 tmp25 = tmp11 - tmp14;
2237 tmp22 = tmp12 + tmp15;
2238 tmp24 = tmp12 - tmp15;
2242 z1 = (INT32) wsptr[1];
2243 z2 = (INT32) wsptr[3];
2244 z3 = (INT32) wsptr[5];
2245 z4 = (INT32) wsptr[7];
2249 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
2250 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
2251 tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
2252 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
2253 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
2255 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */
2257 tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4; /* -c13 */
2258 tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
2259 tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
2260 tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
2261 tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
2262 tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
2264 tmp13 = ((z1 - z3) << CONST_BITS) + z4;
2266 /* Final output stage */
2268 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2269 CONST_BITS+PASS1_BITS+3)
2271 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2272 CONST_BITS+PASS1_BITS+3)
2274 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2275 CONST_BITS+PASS1_BITS+3)
2277 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2278 CONST_BITS+PASS1_BITS+3)
2280 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2281 CONST_BITS+PASS1_BITS+3)
2283 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2284 CONST_BITS+PASS1_BITS+3)
2286 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2287 CONST_BITS+PASS1_BITS+3)
2289 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2290 CONST_BITS+PASS1_BITS+3)
2292 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2293 CONST_BITS+PASS1_BITS+3)
2295 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2296 CONST_BITS+PASS1_BITS+3)
2298 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2299 CONST_BITS+PASS1_BITS+3)
2301 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2302 CONST_BITS+PASS1_BITS+3)
2304 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
2305 CONST_BITS+PASS1_BITS+3)
2307 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
2308 CONST_BITS+PASS1_BITS+3)
2311 wsptr += 8; /* advance pointer to next row */
2317 * Perform dequantization and inverse DCT on one block of coefficients,
2318 * producing a 15x15 output block.
2320 * Optimized algorithm with 22 multiplications in the 1-D kernel.
2321 * cK represents sqrt(2) * cos(K*pi/30).
2325 jpeg_idct_15x15 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2326 JCOEFPTR coef_block,
2327 JSAMPARRAY output_buf, JDIMENSION output_col)
2329 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
2330 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2331 INT32 z1, z2, z3, z4;
2333 ISLOW_MULT_TYPE * quantptr;
2336 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2338 int workspace[8*15]; /* buffers data between passes */
2341 /* Pass 1: process columns from input, store into work array. */
2344 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2346 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2349 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
2351 /* Add fudge factor here for final descale. */
2352 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
2354 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
2355 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
2356 z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
2358 tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2359 tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2363 z1 -= (tmp11 - tmp10) << 1; /* c0 = (c6-c12)*2 */
2367 tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2368 tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2369 z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2371 tmp20 = tmp13 + tmp10 + tmp11;
2372 tmp23 = tmp12 - tmp10 + tmp11 + z2;
2374 tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2375 tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2377 tmp25 = tmp13 - tmp10 - tmp11;
2378 tmp26 = tmp12 + tmp10 - tmp11 - z2;
2380 tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2381 tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2383 tmp21 = tmp12 + tmp10 + tmp11;
2384 tmp24 = tmp13 - tmp10 + tmp11;
2386 tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2387 tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
2391 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
2392 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
2393 z4 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
2394 z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2395 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
2398 tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2399 tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2400 tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2402 tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
2403 tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
2405 tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2407 tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2408 tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2409 tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2410 z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2411 tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2412 tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2414 /* Final output stage */
2416 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
2417 wsptr[8*14] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
2418 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
2419 wsptr[8*13] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
2420 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
2421 wsptr[8*12] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
2422 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
2423 wsptr[8*11] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
2424 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
2425 wsptr[8*10] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
2426 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
2427 wsptr[8*9] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
2428 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
2429 wsptr[8*8] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
2430 wsptr[8*7] = (int) RIGHT_SHIFT(tmp27, CONST_BITS-PASS1_BITS);
2433 /* Pass 2: process 15 rows from work array, store into output array. */
2436 for (ctr = 0; ctr < 15; ctr++) {
2437 outptr = output_buf[ctr] + output_col;
2441 /* Add fudge factor here for final descale. */
2442 z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2445 z2 = (INT32) wsptr[2];
2446 z3 = (INT32) wsptr[4];
2447 z4 = (INT32) wsptr[6];
2449 tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2450 tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2454 z1 -= (tmp11 - tmp10) << 1; /* c0 = (c6-c12)*2 */
2458 tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2459 tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2460 z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2462 tmp20 = tmp13 + tmp10 + tmp11;
2463 tmp23 = tmp12 - tmp10 + tmp11 + z2;
2465 tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2466 tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2468 tmp25 = tmp13 - tmp10 - tmp11;
2469 tmp26 = tmp12 + tmp10 - tmp11 - z2;
2471 tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2472 tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2474 tmp21 = tmp12 + tmp10 + tmp11;
2475 tmp24 = tmp13 - tmp10 + tmp11;
2477 tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2478 tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
2482 z1 = (INT32) wsptr[1];
2483 z2 = (INT32) wsptr[3];
2484 z4 = (INT32) wsptr[5];
2485 z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2486 z4 = (INT32) wsptr[7];
2489 tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2490 tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2491 tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2493 tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
2494 tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
2496 tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2498 tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2499 tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2500 tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2501 z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2502 tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2503 tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2505 /* Final output stage */
2507 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2508 CONST_BITS+PASS1_BITS+3)
2510 outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2511 CONST_BITS+PASS1_BITS+3)
2513 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2514 CONST_BITS+PASS1_BITS+3)
2516 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2517 CONST_BITS+PASS1_BITS+3)
2519 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2520 CONST_BITS+PASS1_BITS+3)
2522 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2523 CONST_BITS+PASS1_BITS+3)
2525 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2526 CONST_BITS+PASS1_BITS+3)
2528 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2529 CONST_BITS+PASS1_BITS+3)
2531 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2532 CONST_BITS+PASS1_BITS+3)
2534 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2535 CONST_BITS+PASS1_BITS+3)
2537 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2538 CONST_BITS+PASS1_BITS+3)
2540 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2541 CONST_BITS+PASS1_BITS+3)
2543 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
2544 CONST_BITS+PASS1_BITS+3)
2546 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
2547 CONST_BITS+PASS1_BITS+3)
2549 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27,
2550 CONST_BITS+PASS1_BITS+3)
2553 wsptr += 8; /* advance pointer to next row */
2559 * Perform dequantization and inverse DCT on one block of coefficients,
2560 * producing a 16x16 output block.
2562 * Optimized algorithm with 28 multiplications in the 1-D kernel.
2563 * cK represents sqrt(2) * cos(K*pi/32).
2567 jpeg_idct_16x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2568 JCOEFPTR coef_block,
2569 JSAMPARRAY output_buf, JDIMENSION output_col)
2571 INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
2572 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2573 INT32 z1, z2, z3, z4;
2575 ISLOW_MULT_TYPE * quantptr;
2578 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2580 int workspace[8*16]; /* buffers data between passes */
2583 /* Pass 1: process columns from input, store into work array. */
2586 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2588 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2591 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
2592 tmp0 <<= CONST_BITS;
2593 /* Add fudge factor here for final descale. */
2594 tmp0 += 1 << (CONST_BITS-PASS1_BITS-1);
2596 z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
2597 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2598 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2600 tmp10 = tmp0 + tmp1;
2601 tmp11 = tmp0 - tmp1;
2602 tmp12 = tmp0 + tmp2;
2603 tmp13 = tmp0 - tmp2;
2605 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
2606 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
2608 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2609 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2611 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2612 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2613 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2614 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2616 tmp20 = tmp10 + tmp0;
2617 tmp27 = tmp10 - tmp0;
2618 tmp21 = tmp12 + tmp1;
2619 tmp26 = tmp12 - tmp1;
2620 tmp22 = tmp13 + tmp2;
2621 tmp25 = tmp13 - tmp2;
2622 tmp23 = tmp11 + tmp3;
2623 tmp24 = tmp11 - tmp3;
2627 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
2628 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
2629 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
2630 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
2634 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2635 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2636 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2637 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2638 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2639 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2640 tmp0 = tmp1 + tmp2 + tmp3 -
2641 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2642 tmp13 = tmp10 + tmp11 + tmp12 -
2643 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2644 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2645 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2646 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2647 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2648 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2649 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2651 z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
2653 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2654 z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
2655 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2657 z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
2660 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2664 /* Final output stage */
2666 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS-PASS1_BITS);
2667 wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS-PASS1_BITS);
2668 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS-PASS1_BITS);
2669 wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS-PASS1_BITS);
2670 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS-PASS1_BITS);
2671 wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS-PASS1_BITS);
2672 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS-PASS1_BITS);
2673 wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS-PASS1_BITS);
2674 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
2675 wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
2676 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
2677 wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
2678 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
2679 wsptr[8*9] = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
2680 wsptr[8*7] = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
2681 wsptr[8*8] = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
2684 /* Pass 2: process 16 rows from work array, store into output array. */
2687 for (ctr = 0; ctr < 16; ctr++) {
2688 outptr = output_buf[ctr] + output_col;
2692 /* Add fudge factor here for final descale. */
2693 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2694 tmp0 <<= CONST_BITS;
2696 z1 = (INT32) wsptr[4];
2697 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2698 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2700 tmp10 = tmp0 + tmp1;
2701 tmp11 = tmp0 - tmp1;
2702 tmp12 = tmp0 + tmp2;
2703 tmp13 = tmp0 - tmp2;
2705 z1 = (INT32) wsptr[2];
2706 z2 = (INT32) wsptr[6];
2708 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2709 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2711 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2712 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2713 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2714 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2716 tmp20 = tmp10 + tmp0;
2717 tmp27 = tmp10 - tmp0;
2718 tmp21 = tmp12 + tmp1;
2719 tmp26 = tmp12 - tmp1;
2720 tmp22 = tmp13 + tmp2;
2721 tmp25 = tmp13 - tmp2;
2722 tmp23 = tmp11 + tmp3;
2723 tmp24 = tmp11 - tmp3;
2727 z1 = (INT32) wsptr[1];
2728 z2 = (INT32) wsptr[3];
2729 z3 = (INT32) wsptr[5];
2730 z4 = (INT32) wsptr[7];
2734 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2735 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2736 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2737 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2738 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2739 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2740 tmp0 = tmp1 + tmp2 + tmp3 -
2741 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2742 tmp13 = tmp10 + tmp11 + tmp12 -
2743 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2744 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2745 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2746 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2747 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2748 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2749 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2751 z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
2753 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2754 z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
2755 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2757 z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
2760 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2764 /* Final output stage */
2766 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
2767 CONST_BITS+PASS1_BITS+3)
2769 outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
2770 CONST_BITS+PASS1_BITS+3)
2772 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
2773 CONST_BITS+PASS1_BITS+3)
2775 outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
2776 CONST_BITS+PASS1_BITS+3)
2778 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
2779 CONST_BITS+PASS1_BITS+3)
2781 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
2782 CONST_BITS+PASS1_BITS+3)
2784 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
2785 CONST_BITS+PASS1_BITS+3)
2787 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
2788 CONST_BITS+PASS1_BITS+3)
2790 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
2791 CONST_BITS+PASS1_BITS+3)
2793 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
2794 CONST_BITS+PASS1_BITS+3)
2796 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
2797 CONST_BITS+PASS1_BITS+3)
2799 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
2800 CONST_BITS+PASS1_BITS+3)
2802 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
2803 CONST_BITS+PASS1_BITS+3)
2805 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
2806 CONST_BITS+PASS1_BITS+3)
2808 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
2809 CONST_BITS+PASS1_BITS+3)
2811 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
2812 CONST_BITS+PASS1_BITS+3)
2815 wsptr += 8; /* advance pointer to next row */
2821 * Perform dequantization and inverse DCT on one block of coefficients,
2822 * producing a 16x8 output block.
2824 * 8-point IDCT in pass 1 (columns), 16-point in pass 2 (rows).
2828 jpeg_idct_16x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2829 JCOEFPTR coef_block,
2830 JSAMPARRAY output_buf, JDIMENSION output_col)
2832 INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
2833 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2834 INT32 z1, z2, z3, z4;
2836 ISLOW_MULT_TYPE * quantptr;
2839 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2841 int workspace[8*8]; /* buffers data between passes */
2844 /* Pass 1: process columns from input, store into work array.
2845 * Note results are scaled up by sqrt(8) compared to a true IDCT;
2846 * furthermore, we scale the results by 2**PASS1_BITS.
2847 * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
2851 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2853 for (ctr = DCTSIZE; ctr > 0; ctr--) {
2854 /* Due to quantization, we will usually find that many of the input
2855 * coefficients are zero, especially the AC terms. We can exploit this
2856 * by short-circuiting the IDCT calculation for any column in which all
2857 * the AC terms are zero. In that case each output is equal to the
2858 * DC coefficient (with scale factor as needed).
2859 * With typical images and quantization tables, half or more of the
2860 * column DCT calculations can be simplified this way.
2863 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
2864 inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
2865 inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
2866 inptr[DCTSIZE*7] == 0) {
2867 /* AC terms all zero */
2868 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
2870 wsptr[DCTSIZE*0] = dcval;
2871 wsptr[DCTSIZE*1] = dcval;
2872 wsptr[DCTSIZE*2] = dcval;
2873 wsptr[DCTSIZE*3] = dcval;
2874 wsptr[DCTSIZE*4] = dcval;
2875 wsptr[DCTSIZE*5] = dcval;
2876 wsptr[DCTSIZE*6] = dcval;
2877 wsptr[DCTSIZE*7] = dcval;
2879 inptr++; /* advance pointers to next column */
2885 /* Even part: reverse the even part of the forward DCT.
2886 * The rotator is c(-6).
2889 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
2890 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
2892 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
2893 tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
2894 tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
2896 z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
2897 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
2900 /* Add fudge factor here for final descale. */
2901 z2 += ONE << (CONST_BITS-PASS1_BITS-1);
2906 tmp10 = tmp0 + tmp2;
2907 tmp13 = tmp0 - tmp2;
2908 tmp11 = tmp1 + tmp3;
2909 tmp12 = tmp1 - tmp3;
2911 /* Odd part per figure 8; the matrix is unitary and hence its
2912 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
2915 tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
2916 tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
2917 tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
2918 tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
2923 z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* c3 */
2924 z2 = MULTIPLY(z2, - FIX_1_961570560); /* -c3-c5 */
2925 z3 = MULTIPLY(z3, - FIX_0_390180644); /* -c3+c5 */
2929 z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
2930 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* -c1+c3+c5-c7 */
2931 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* c1+c3-c5-c7 */
2935 z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
2936 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* c1+c3-c5+c7 */
2937 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* c1+c3+c5-c7 */
2941 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
2943 wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
2944 wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
2945 wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
2946 wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
2947 wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
2948 wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
2949 wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
2950 wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
2952 inptr++; /* advance pointers to next column */
2957 /* Pass 2: process 8 rows from work array, store into output array.
2958 * 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
2962 for (ctr = 0; ctr < 8; ctr++) {
2963 outptr = output_buf[ctr] + output_col;
2967 /* Add fudge factor here for final descale. */
2968 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2969 tmp0 <<= CONST_BITS;
2971 z1 = (INT32) wsptr[4];
2972 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2973 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2975 tmp10 = tmp0 + tmp1;
2976 tmp11 = tmp0 - tmp1;
2977 tmp12 = tmp0 + tmp2;
2978 tmp13 = tmp0 - tmp2;
2980 z1 = (INT32) wsptr[2];
2981 z2 = (INT32) wsptr[6];
2983 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2984 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2986 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2987 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2988 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2989 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2991 tmp20 = tmp10 + tmp0;
2992 tmp27 = tmp10 - tmp0;
2993 tmp21 = tmp12 + tmp1;
2994 tmp26 = tmp12 - tmp1;
2995 tmp22 = tmp13 + tmp2;
2996 tmp25 = tmp13 - tmp2;
2997 tmp23 = tmp11 + tmp3;
2998 tmp24 = tmp11 - tmp3;
3002 z1 = (INT32) wsptr[1];
3003 z2 = (INT32) wsptr[3];
3004 z3 = (INT32) wsptr[5];
3005 z4 = (INT32) wsptr[7];
3009 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
3010 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
3011 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
3012 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
3013 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
3014 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
3015 tmp0 = tmp1 + tmp2 + tmp3 -
3016 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
3017 tmp13 = tmp10 + tmp11 + tmp12 -
3018 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
3019 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
3020 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
3021 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
3022 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
3023 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
3024 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
3026 z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
3028 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
3029 z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
3030 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
3032 z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
3035 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
3039 /* Final output stage */
3041 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
3042 CONST_BITS+PASS1_BITS+3)
3044 outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
3045 CONST_BITS+PASS1_BITS+3)
3047 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
3048 CONST_BITS+PASS1_BITS+3)
3050 outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
3051 CONST_BITS+PASS1_BITS+3)
3053 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
3054 CONST_BITS+PASS1_BITS+3)
3056 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
3057 CONST_BITS+PASS1_BITS+3)
3059 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
3060 CONST_BITS+PASS1_BITS+3)
3062 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
3063 CONST_BITS+PASS1_BITS+3)
3065 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
3066 CONST_BITS+PASS1_BITS+3)
3068 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
3069 CONST_BITS+PASS1_BITS+3)
3071 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
3072 CONST_BITS+PASS1_BITS+3)
3074 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
3075 CONST_BITS+PASS1_BITS+3)
3077 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
3078 CONST_BITS+PASS1_BITS+3)
3080 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
3081 CONST_BITS+PASS1_BITS+3)
3083 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
3084 CONST_BITS+PASS1_BITS+3)
3086 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
3087 CONST_BITS+PASS1_BITS+3)
3090 wsptr += 8; /* advance pointer to next row */
3096 * Perform dequantization and inverse DCT on one block of coefficients,
3097 * producing a 14x7 output block.
3099 * 7-point IDCT in pass 1 (columns), 14-point in pass 2 (rows).
3103 jpeg_idct_14x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3104 JCOEFPTR coef_block,
3105 JSAMPARRAY output_buf, JDIMENSION output_col)
3107 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
3108 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
3109 INT32 z1, z2, z3, z4;
3111 ISLOW_MULT_TYPE * quantptr;
3114 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3116 int workspace[8*7]; /* buffers data between passes */
3119 /* Pass 1: process columns from input, store into work array.
3120 * 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
3124 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3126 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3129 tmp23 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
3130 tmp23 <<= CONST_BITS;
3131 /* Add fudge factor here for final descale. */
3132 tmp23 += ONE << (CONST_BITS-PASS1_BITS-1);
3134 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
3135 z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
3136 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
3138 tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
3139 tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
3140 tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
3143 tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
3144 tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
3145 tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
3146 tmp23 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
3150 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
3151 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
3152 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
3154 tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
3155 tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
3156 tmp10 = tmp11 - tmp12;
3158 tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
3160 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
3162 tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
3164 /* Final output stage */
3166 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
3167 wsptr[8*6] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
3168 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
3169 wsptr[8*5] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
3170 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
3171 wsptr[8*4] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
3172 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23, CONST_BITS-PASS1_BITS);
3175 /* Pass 2: process 7 rows from work array, store into output array.
3176 * 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
3180 for (ctr = 0; ctr < 7; ctr++) {
3181 outptr = output_buf[ctr] + output_col;
3185 /* Add fudge factor here for final descale. */
3186 z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3188 z4 = (INT32) wsptr[4];
3189 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
3190 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
3191 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
3197 tmp23 = z1 - ((z2 + z3 - z4) << 1); /* c0 = (c4+c12-c8)*2 */
3199 z1 = (INT32) wsptr[2];
3200 z2 = (INT32) wsptr[6];
3202 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
3204 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
3205 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
3206 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
3207 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
3209 tmp20 = tmp10 + tmp13;
3210 tmp26 = tmp10 - tmp13;
3211 tmp21 = tmp11 + tmp14;
3212 tmp25 = tmp11 - tmp14;
3213 tmp22 = tmp12 + tmp15;
3214 tmp24 = tmp12 - tmp15;
3218 z1 = (INT32) wsptr[1];
3219 z2 = (INT32) wsptr[3];
3220 z3 = (INT32) wsptr[5];
3221 z4 = (INT32) wsptr[7];
3225 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
3226 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
3227 tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
3228 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
3229 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
3231 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */
3233 tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4; /* -c13 */
3234 tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
3235 tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
3236 tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
3237 tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
3238 tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
3240 tmp13 = ((z1 - z3) << CONST_BITS) + z4;
3242 /* Final output stage */
3244 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
3245 CONST_BITS+PASS1_BITS+3)
3247 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
3248 CONST_BITS+PASS1_BITS+3)
3250 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
3251 CONST_BITS+PASS1_BITS+3)
3253 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
3254 CONST_BITS+PASS1_BITS+3)
3256 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
3257 CONST_BITS+PASS1_BITS+3)
3259 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
3260 CONST_BITS+PASS1_BITS+3)
3262 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
3263 CONST_BITS+PASS1_BITS+3)
3265 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
3266 CONST_BITS+PASS1_BITS+3)
3268 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
3269 CONST_BITS+PASS1_BITS+3)
3271 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
3272 CONST_BITS+PASS1_BITS+3)
3274 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
3275 CONST_BITS+PASS1_BITS+3)
3277 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
3278 CONST_BITS+PASS1_BITS+3)
3280 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
3281 CONST_BITS+PASS1_BITS+3)
3283 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
3284 CONST_BITS+PASS1_BITS+3)
3287 wsptr += 8; /* advance pointer to next row */
3293 * Perform dequantization and inverse DCT on one block of coefficients,
3294 * producing a 12x6 output block.
3296 * 6-point IDCT in pass 1 (columns), 12-point in pass 2 (rows).
3300 jpeg_idct_12x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3301 JCOEFPTR coef_block,
3302 JSAMPARRAY output_buf, JDIMENSION output_col)
3304 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
3305 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
3306 INT32 z1, z2, z3, z4;
3308 ISLOW_MULT_TYPE * quantptr;
3311 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3313 int workspace[8*6]; /* buffers data between passes */
3316 /* Pass 1: process columns from input, store into work array.
3317 * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
3321 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3323 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3326 tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
3327 tmp10 <<= CONST_BITS;
3328 /* Add fudge factor here for final descale. */
3329 tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
3330 tmp12 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
3331 tmp20 = MULTIPLY(tmp12, FIX(0.707106781)); /* c4 */
3332 tmp11 = tmp10 + tmp20;
3333 tmp21 = RIGHT_SHIFT(tmp10 - tmp20 - tmp20, CONST_BITS-PASS1_BITS);
3334 tmp20 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
3335 tmp10 = MULTIPLY(tmp20, FIX(1.224744871)); /* c2 */
3336 tmp20 = tmp11 + tmp10;
3337 tmp22 = tmp11 - tmp10;
3341 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
3342 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
3343 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
3344 tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
3345 tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
3346 tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
3347 tmp11 = (z1 - z2 - z3) << PASS1_BITS;
3349 /* Final output stage */
3351 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
3352 wsptr[8*5] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
3353 wsptr[8*1] = (int) (tmp21 + tmp11);
3354 wsptr[8*4] = (int) (tmp21 - tmp11);
3355 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
3356 wsptr[8*3] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
3359 /* Pass 2: process 6 rows from work array, store into output array.
3360 * 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
3364 for (ctr = 0; ctr < 6; ctr++) {
3365 outptr = output_buf[ctr] + output_col;
3369 /* Add fudge factor here for final descale. */
3370 z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3373 z4 = (INT32) wsptr[4];
3374 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
3379 z1 = (INT32) wsptr[2];
3380 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
3382 z2 = (INT32) wsptr[6];
3392 tmp20 = tmp10 + tmp12;
3393 tmp25 = tmp10 - tmp12;
3395 tmp12 = z4 - z1 - z2;
3397 tmp22 = tmp11 + tmp12;
3398 tmp23 = tmp11 - tmp12;
3402 z1 = (INT32) wsptr[1];
3403 z2 = (INT32) wsptr[3];
3404 z3 = (INT32) wsptr[5];
3405 z4 = (INT32) wsptr[7];
3407 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
3408 tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
3411 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
3412 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
3413 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
3414 tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
3415 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
3416 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
3417 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
3418 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
3422 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
3423 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
3424 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
3426 /* Final output stage */
3428 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
3429 CONST_BITS+PASS1_BITS+3)
3431 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
3432 CONST_BITS+PASS1_BITS+3)
3434 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
3435 CONST_BITS+PASS1_BITS+3)
3437 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
3438 CONST_BITS+PASS1_BITS+3)
3440 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
3441 CONST_BITS+PASS1_BITS+3)
3443 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
3444 CONST_BITS+PASS1_BITS+3)
3446 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
3447 CONST_BITS+PASS1_BITS+3)
3449 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
3450 CONST_BITS+PASS1_BITS+3)
3452 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
3453 CONST_BITS+PASS1_BITS+3)
3455 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
3456 CONST_BITS+PASS1_BITS+3)
3458 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
3459 CONST_BITS+PASS1_BITS+3)
3461 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
3462 CONST_BITS+PASS1_BITS+3)
3465 wsptr += 8; /* advance pointer to next row */
3471 * Perform dequantization and inverse DCT on one block of coefficients,
3472 * producing a 10x5 output block.
3474 * 5-point IDCT in pass 1 (columns), 10-point in pass 2 (rows).
3478 jpeg_idct_10x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3479 JCOEFPTR coef_block,
3480 JSAMPARRAY output_buf, JDIMENSION output_col)
3482 INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
3483 INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
3484 INT32 z1, z2, z3, z4;
3486 ISLOW_MULT_TYPE * quantptr;
3489 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3491 int workspace[8*5]; /* buffers data between passes */
3494 /* Pass 1: process columns from input, store into work array.
3495 * 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
3499 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3501 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3504 tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
3505 tmp12 <<= CONST_BITS;
3506 /* Add fudge factor here for final descale. */
3507 tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
3508 tmp13 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
3509 tmp14 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
3510 z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
3511 z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
3519 z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
3520 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
3522 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
3523 tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
3524 tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
3526 /* Final output stage */
3528 wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp13, CONST_BITS-PASS1_BITS);
3529 wsptr[8*4] = (int) RIGHT_SHIFT(tmp10 - tmp13, CONST_BITS-PASS1_BITS);
3530 wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp14, CONST_BITS-PASS1_BITS);
3531 wsptr[8*3] = (int) RIGHT_SHIFT(tmp11 - tmp14, CONST_BITS-PASS1_BITS);
3532 wsptr[8*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
3535 /* Pass 2: process 5 rows from work array, store into output array.
3536 * 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
3540 for (ctr = 0; ctr < 5; ctr++) {
3541 outptr = output_buf[ctr] + output_col;
3545 /* Add fudge factor here for final descale. */
3546 z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3548 z4 = (INT32) wsptr[4];
3549 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
3550 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
3554 tmp22 = z3 - ((z1 - z2) << 1); /* c0 = (c4-c8)*2 */
3556 z2 = (INT32) wsptr[2];
3557 z3 = (INT32) wsptr[6];
3559 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
3560 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
3561 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
3563 tmp20 = tmp10 + tmp12;
3564 tmp24 = tmp10 - tmp12;
3565 tmp21 = tmp11 + tmp13;
3566 tmp23 = tmp11 - tmp13;
3570 z1 = (INT32) wsptr[1];
3571 z2 = (INT32) wsptr[3];
3572 z3 = (INT32) wsptr[5];
3574 z4 = (INT32) wsptr[7];
3579 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
3581 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
3584 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
3585 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
3587 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
3588 z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));
3590 tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;
3592 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
3593 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
3595 /* Final output stage */
3597 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
3598 CONST_BITS+PASS1_BITS+3)
3600 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
3601 CONST_BITS+PASS1_BITS+3)
3603 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
3604 CONST_BITS+PASS1_BITS+3)
3606 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
3607 CONST_BITS+PASS1_BITS+3)
3609 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
3610 CONST_BITS+PASS1_BITS+3)
3612 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
3613 CONST_BITS+PASS1_BITS+3)
3615 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
3616 CONST_BITS+PASS1_BITS+3)
3618 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
3619 CONST_BITS+PASS1_BITS+3)
3621 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
3622 CONST_BITS+PASS1_BITS+3)
3624 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
3625 CONST_BITS+PASS1_BITS+3)
3628 wsptr += 8; /* advance pointer to next row */
3634 * Perform dequantization and inverse DCT on one block of coefficients,
3635 * producing a 8x4 output block.
3637 * 4-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
3641 jpeg_idct_8x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3642 JCOEFPTR coef_block,
3643 JSAMPARRAY output_buf, JDIMENSION output_col)
3645 INT32 tmp0, tmp1, tmp2, tmp3;
3646 INT32 tmp10, tmp11, tmp12, tmp13;
3649 ISLOW_MULT_TYPE * quantptr;
3652 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3654 int workspace[8*4]; /* buffers data between passes */
3657 /* Pass 1: process columns from input, store into work array.
3658 * 4-point IDCT kernel,
3659 * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
3663 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3665 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3668 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
3669 tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
3671 tmp10 = (tmp0 + tmp2) << PASS1_BITS;
3672 tmp12 = (tmp0 - tmp2) << PASS1_BITS;
3675 /* Same rotation as in the even part of the 8x8 LL&M IDCT */
3677 z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
3678 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
3680 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
3681 /* Add fudge factor here for final descale. */
3682 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
3683 tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
3684 CONST_BITS-PASS1_BITS);
3685 tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
3686 CONST_BITS-PASS1_BITS);
3688 /* Final output stage */
3690 wsptr[8*0] = (int) (tmp10 + tmp0);
3691 wsptr[8*3] = (int) (tmp10 - tmp0);
3692 wsptr[8*1] = (int) (tmp12 + tmp2);
3693 wsptr[8*2] = (int) (tmp12 - tmp2);
3696 /* Pass 2: process rows from work array, store into output array.
3697 * Note that we must descale the results by a factor of 8 == 2**3,
3698 * and also undo the PASS1_BITS scaling.
3699 * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
3703 for (ctr = 0; ctr < 4; ctr++) {
3704 outptr = output_buf[ctr] + output_col;
3706 /* Even part: reverse the even part of the forward DCT.
3707 * The rotator is c(-6).
3710 z2 = (INT32) wsptr[2];
3711 z3 = (INT32) wsptr[6];
3713 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
3714 tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
3715 tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
3717 /* Add fudge factor here for final descale. */
3718 z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3719 z3 = (INT32) wsptr[4];
3721 tmp0 = (z2 + z3) << CONST_BITS;
3722 tmp1 = (z2 - z3) << CONST_BITS;
3724 tmp10 = tmp0 + tmp2;
3725 tmp13 = tmp0 - tmp2;
3726 tmp11 = tmp1 + tmp3;
3727 tmp12 = tmp1 - tmp3;
3729 /* Odd part per figure 8; the matrix is unitary and hence its
3730 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
3733 tmp0 = (INT32) wsptr[7];
3734 tmp1 = (INT32) wsptr[5];
3735 tmp2 = (INT32) wsptr[3];
3736 tmp3 = (INT32) wsptr[1];
3741 z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* c3 */
3742 z2 = MULTIPLY(z2, - FIX_1_961570560); /* -c3-c5 */
3743 z3 = MULTIPLY(z3, - FIX_0_390180644); /* -c3+c5 */
3747 z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
3748 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* -c1+c3+c5-c7 */
3749 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* c1+c3-c5-c7 */
3753 z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
3754 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* c1+c3-c5+c7 */
3755 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* c1+c3+c5-c7 */
3759 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
3761 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
3762 CONST_BITS+PASS1_BITS+3)
3764 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
3765 CONST_BITS+PASS1_BITS+3)
3767 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
3768 CONST_BITS+PASS1_BITS+3)
3770 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
3771 CONST_BITS+PASS1_BITS+3)
3773 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
3774 CONST_BITS+PASS1_BITS+3)
3776 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
3777 CONST_BITS+PASS1_BITS+3)
3779 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
3780 CONST_BITS+PASS1_BITS+3)
3782 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
3783 CONST_BITS+PASS1_BITS+3)
3786 wsptr += DCTSIZE; /* advance pointer to next row */
3792 * Perform dequantization and inverse DCT on one block of coefficients,
3793 * producing a reduced-size 6x3 output block.
3795 * 3-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
3799 jpeg_idct_6x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3800 JCOEFPTR coef_block,
3801 JSAMPARRAY output_buf, JDIMENSION output_col)
3803 INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
3806 ISLOW_MULT_TYPE * quantptr;
3809 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3811 int workspace[6*3]; /* buffers data between passes */
3814 /* Pass 1: process columns from input, store into work array.
3815 * 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
3819 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3821 for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
3824 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
3825 tmp0 <<= CONST_BITS;
3826 /* Add fudge factor here for final descale. */
3827 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
3828 tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
3829 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
3830 tmp10 = tmp0 + tmp12;
3831 tmp2 = tmp0 - tmp12 - tmp12;
3835 tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
3836 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
3838 /* Final output stage */
3840 wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
3841 wsptr[6*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
3842 wsptr[6*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
3845 /* Pass 2: process 3 rows from work array, store into output array.
3846 * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
3850 for (ctr = 0; ctr < 3; ctr++) {
3851 outptr = output_buf[ctr] + output_col;
3855 /* Add fudge factor here for final descale. */
3856 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3857 tmp0 <<= CONST_BITS;
3858 tmp2 = (INT32) wsptr[4];
3859 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
3860 tmp1 = tmp0 + tmp10;
3861 tmp11 = tmp0 - tmp10 - tmp10;
3862 tmp10 = (INT32) wsptr[2];
3863 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
3864 tmp10 = tmp1 + tmp0;
3865 tmp12 = tmp1 - tmp0;
3869 z1 = (INT32) wsptr[1];
3870 z2 = (INT32) wsptr[3];
3871 z3 = (INT32) wsptr[5];
3872 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
3873 tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
3874 tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
3875 tmp1 = (z1 - z2 - z3) << CONST_BITS;
3877 /* Final output stage */
3879 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
3880 CONST_BITS+PASS1_BITS+3)
3882 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
3883 CONST_BITS+PASS1_BITS+3)
3885 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
3886 CONST_BITS+PASS1_BITS+3)
3888 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
3889 CONST_BITS+PASS1_BITS+3)
3891 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
3892 CONST_BITS+PASS1_BITS+3)
3894 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
3895 CONST_BITS+PASS1_BITS+3)
3898 wsptr += 6; /* advance pointer to next row */
3904 * Perform dequantization and inverse DCT on one block of coefficients,
3905 * producing a 4x2 output block.
3907 * 2-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
3911 jpeg_idct_4x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3912 JCOEFPTR coef_block,
3913 JSAMPARRAY output_buf, JDIMENSION output_col)
3915 INT32 tmp0, tmp2, tmp10, tmp12;
3918 ISLOW_MULT_TYPE * quantptr;
3921 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3923 INT32 workspace[4*2]; /* buffers data between passes */
3926 /* Pass 1: process columns from input, store into work array. */
3929 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3931 for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
3934 tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
3938 tmp0 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
3940 /* Final output stage */
3942 wsptr[4*0] = tmp10 + tmp0;
3943 wsptr[4*1] = tmp10 - tmp0;
3946 /* Pass 2: process 2 rows from work array, store into output array.
3947 * 4-point IDCT kernel,
3948 * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
3952 for (ctr = 0; ctr < 2; ctr++) {
3953 outptr = output_buf[ctr] + output_col;
3957 /* Add fudge factor here for final descale. */
3958 tmp0 = wsptr[0] + (ONE << 2);
3961 tmp10 = (tmp0 + tmp2) << CONST_BITS;
3962 tmp12 = (tmp0 - tmp2) << CONST_BITS;
3965 /* Same rotation as in the even part of the 8x8 LL&M IDCT */
3970 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
3971 tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
3972 tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
3974 /* Final output stage */
3976 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
3979 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
3982 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
3985 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
3989 wsptr += 4; /* advance pointer to next row */
3995 * Perform dequantization and inverse DCT on one block of coefficients,
3996 * producing a 2x1 output block.
3998 * 1-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
4002 jpeg_idct_2x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4003 JCOEFPTR coef_block,
4004 JSAMPARRAY output_buf, JDIMENSION output_col)
4007 ISLOW_MULT_TYPE * quantptr;
4009 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4012 /* Pass 1: empty. */
4014 /* Pass 2: process 1 row from input, store into output array. */
4016 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4017 outptr = output_buf[0] + output_col;
4021 tmp0 = DEQUANTIZE(coef_block[0], quantptr[0]);
4022 /* Add fudge factor here for final descale. */
4027 tmp1 = DEQUANTIZE(coef_block[1], quantptr[1]);
4029 /* Final output stage */
4031 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp0 + tmp1, 3) & RANGE_MASK];
4032 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp0 - tmp1, 3) & RANGE_MASK];
4037 * Perform dequantization and inverse DCT on one block of coefficients,
4038 * producing a 8x16 output block.
4040 * 16-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
4044 jpeg_idct_8x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4045 JCOEFPTR coef_block,
4046 JSAMPARRAY output_buf, JDIMENSION output_col)
4048 INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
4049 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
4050 INT32 z1, z2, z3, z4;
4052 ISLOW_MULT_TYPE * quantptr;
4055 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4057 int workspace[8*16]; /* buffers data between passes */
4060 /* Pass 1: process columns from input, store into work array.
4061 * 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
4065 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4067 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
4070 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
4071 tmp0 <<= CONST_BITS;
4072 /* Add fudge factor here for final descale. */
4073 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
4075 z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
4076 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
4077 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
4079 tmp10 = tmp0 + tmp1;
4080 tmp11 = tmp0 - tmp1;
4081 tmp12 = tmp0 + tmp2;
4082 tmp13 = tmp0 - tmp2;
4084 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
4085 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
4087 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
4088 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
4090 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
4091 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
4092 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
4093 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
4095 tmp20 = tmp10 + tmp0;
4096 tmp27 = tmp10 - tmp0;
4097 tmp21 = tmp12 + tmp1;
4098 tmp26 = tmp12 - tmp1;
4099 tmp22 = tmp13 + tmp2;
4100 tmp25 = tmp13 - tmp2;
4101 tmp23 = tmp11 + tmp3;
4102 tmp24 = tmp11 - tmp3;
4106 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
4107 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
4108 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
4109 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
4113 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
4114 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
4115 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
4116 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
4117 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
4118 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
4119 tmp0 = tmp1 + tmp2 + tmp3 -
4120 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
4121 tmp13 = tmp10 + tmp11 + tmp12 -
4122 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
4123 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
4124 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
4125 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
4126 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
4127 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
4128 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
4130 z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
4132 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
4133 z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
4134 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
4136 z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
4139 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
4143 /* Final output stage */
4145 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS-PASS1_BITS);
4146 wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS-PASS1_BITS);
4147 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS-PASS1_BITS);
4148 wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS-PASS1_BITS);
4149 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS-PASS1_BITS);
4150 wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS-PASS1_BITS);
4151 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS-PASS1_BITS);
4152 wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS-PASS1_BITS);
4153 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
4154 wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
4155 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
4156 wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
4157 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
4158 wsptr[8*9] = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
4159 wsptr[8*7] = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
4160 wsptr[8*8] = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
4163 /* Pass 2: process rows from work array, store into output array.
4164 * Note that we must descale the results by a factor of 8 == 2**3,
4165 * and also undo the PASS1_BITS scaling.
4166 * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
4170 for (ctr = 0; ctr < 16; ctr++) {
4171 outptr = output_buf[ctr] + output_col;
4173 /* Even part: reverse the even part of the forward DCT.
4174 * The rotator is c(-6).
4177 z2 = (INT32) wsptr[2];
4178 z3 = (INT32) wsptr[6];
4180 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
4181 tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
4182 tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
4184 /* Add fudge factor here for final descale. */
4185 z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4186 z3 = (INT32) wsptr[4];
4188 tmp0 = (z2 + z3) << CONST_BITS;
4189 tmp1 = (z2 - z3) << CONST_BITS;
4191 tmp10 = tmp0 + tmp2;
4192 tmp13 = tmp0 - tmp2;
4193 tmp11 = tmp1 + tmp3;
4194 tmp12 = tmp1 - tmp3;
4196 /* Odd part per figure 8; the matrix is unitary and hence its
4197 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
4200 tmp0 = (INT32) wsptr[7];
4201 tmp1 = (INT32) wsptr[5];
4202 tmp2 = (INT32) wsptr[3];
4203 tmp3 = (INT32) wsptr[1];
4208 z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* c3 */
4209 z2 = MULTIPLY(z2, - FIX_1_961570560); /* -c3-c5 */
4210 z3 = MULTIPLY(z3, - FIX_0_390180644); /* -c3+c5 */
4214 z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
4215 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* -c1+c3+c5-c7 */
4216 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* c1+c3-c5-c7 */
4220 z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
4221 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* c1+c3-c5+c7 */
4222 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* c1+c3+c5-c7 */
4226 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
4228 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
4229 CONST_BITS+PASS1_BITS+3)
4231 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
4232 CONST_BITS+PASS1_BITS+3)
4234 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
4235 CONST_BITS+PASS1_BITS+3)
4237 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
4238 CONST_BITS+PASS1_BITS+3)
4240 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
4241 CONST_BITS+PASS1_BITS+3)
4243 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
4244 CONST_BITS+PASS1_BITS+3)
4246 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
4247 CONST_BITS+PASS1_BITS+3)
4249 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
4250 CONST_BITS+PASS1_BITS+3)
4253 wsptr += DCTSIZE; /* advance pointer to next row */
4259 * Perform dequantization and inverse DCT on one block of coefficients,
4260 * producing a 7x14 output block.
4262 * 14-point IDCT in pass 1 (columns), 7-point in pass 2 (rows).
4266 jpeg_idct_7x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4267 JCOEFPTR coef_block,
4268 JSAMPARRAY output_buf, JDIMENSION output_col)
4270 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
4271 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
4272 INT32 z1, z2, z3, z4;
4274 ISLOW_MULT_TYPE * quantptr;
4277 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4279 int workspace[7*14]; /* buffers data between passes */
4282 /* Pass 1: process columns from input, store into work array.
4283 * 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
4287 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4289 for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
4292 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
4294 /* Add fudge factor here for final descale. */
4295 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
4296 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
4297 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
4298 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
4299 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
4305 tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)*2 */
4306 CONST_BITS-PASS1_BITS);
4308 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
4309 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
4311 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
4313 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
4314 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
4315 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
4316 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
4318 tmp20 = tmp10 + tmp13;
4319 tmp26 = tmp10 - tmp13;
4320 tmp21 = tmp11 + tmp14;
4321 tmp25 = tmp11 - tmp14;
4322 tmp22 = tmp12 + tmp15;
4323 tmp24 = tmp12 - tmp15;
4327 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
4328 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
4329 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
4330 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
4331 tmp13 = z4 << CONST_BITS;
4334 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
4335 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
4336 tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
4337 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
4338 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
4340 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */
4343 z4 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
4344 tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
4345 tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
4346 z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
4347 tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
4348 tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
4350 tmp13 = (z1 - z3) << PASS1_BITS;
4352 /* Final output stage */
4354 wsptr[7*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
4355 wsptr[7*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
4356 wsptr[7*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
4357 wsptr[7*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
4358 wsptr[7*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
4359 wsptr[7*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
4360 wsptr[7*3] = (int) (tmp23 + tmp13);
4361 wsptr[7*10] = (int) (tmp23 - tmp13);
4362 wsptr[7*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
4363 wsptr[7*9] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
4364 wsptr[7*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
4365 wsptr[7*8] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
4366 wsptr[7*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
4367 wsptr[7*7] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
4370 /* Pass 2: process 14 rows from work array, store into output array.
4371 * 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
4375 for (ctr = 0; ctr < 14; ctr++) {
4376 outptr = output_buf[ctr] + output_col;
4380 /* Add fudge factor here for final descale. */
4381 tmp23 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4382 tmp23 <<= CONST_BITS;
4384 z1 = (INT32) wsptr[2];
4385 z2 = (INT32) wsptr[4];
4386 z3 = (INT32) wsptr[6];
4388 tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
4389 tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
4390 tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
4393 tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
4394 tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
4395 tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
4396 tmp23 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
4400 z1 = (INT32) wsptr[1];
4401 z2 = (INT32) wsptr[3];
4402 z3 = (INT32) wsptr[5];
4404 tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
4405 tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
4406 tmp10 = tmp11 - tmp12;
4408 tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
4410 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
4412 tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
4414 /* Final output stage */
4416 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
4417 CONST_BITS+PASS1_BITS+3)
4419 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
4420 CONST_BITS+PASS1_BITS+3)
4422 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
4423 CONST_BITS+PASS1_BITS+3)
4425 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
4426 CONST_BITS+PASS1_BITS+3)
4428 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
4429 CONST_BITS+PASS1_BITS+3)
4431 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
4432 CONST_BITS+PASS1_BITS+3)
4434 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23,
4435 CONST_BITS+PASS1_BITS+3)
4438 wsptr += 7; /* advance pointer to next row */
4444 * Perform dequantization and inverse DCT on one block of coefficients,
4445 * producing a 6x12 output block.
4447 * 12-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
4451 jpeg_idct_6x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4452 JCOEFPTR coef_block,
4453 JSAMPARRAY output_buf, JDIMENSION output_col)
4455 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
4456 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
4457 INT32 z1, z2, z3, z4;
4459 ISLOW_MULT_TYPE * quantptr;
4462 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4464 int workspace[6*12]; /* buffers data between passes */
4467 /* Pass 1: process columns from input, store into work array.
4468 * 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
4472 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4474 for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
4477 z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
4479 /* Add fudge factor here for final descale. */
4480 z3 += ONE << (CONST_BITS-PASS1_BITS-1);
4482 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
4483 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
4488 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
4489 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
4491 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
4501 tmp20 = tmp10 + tmp12;
4502 tmp25 = tmp10 - tmp12;
4504 tmp12 = z4 - z1 - z2;
4506 tmp22 = tmp11 + tmp12;
4507 tmp23 = tmp11 - tmp12;
4511 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
4512 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
4513 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
4514 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
4516 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
4517 tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
4520 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
4521 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
4522 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
4523 tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
4524 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
4525 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
4526 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
4527 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
4531 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
4532 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
4533 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
4535 /* Final output stage */
4537 wsptr[6*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
4538 wsptr[6*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
4539 wsptr[6*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
4540 wsptr[6*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
4541 wsptr[6*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
4542 wsptr[6*9] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
4543 wsptr[6*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
4544 wsptr[6*8] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
4545 wsptr[6*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
4546 wsptr[6*7] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
4547 wsptr[6*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
4548 wsptr[6*6] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
4551 /* Pass 2: process 12 rows from work array, store into output array.
4552 * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
4556 for (ctr = 0; ctr < 12; ctr++) {
4557 outptr = output_buf[ctr] + output_col;
4561 /* Add fudge factor here for final descale. */
4562 tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4563 tmp10 <<= CONST_BITS;
4564 tmp12 = (INT32) wsptr[4];
4565 tmp20 = MULTIPLY(tmp12, FIX(0.707106781)); /* c4 */
4566 tmp11 = tmp10 + tmp20;
4567 tmp21 = tmp10 - tmp20 - tmp20;
4568 tmp20 = (INT32) wsptr[2];
4569 tmp10 = MULTIPLY(tmp20, FIX(1.224744871)); /* c2 */
4570 tmp20 = tmp11 + tmp10;
4571 tmp22 = tmp11 - tmp10;
4575 z1 = (INT32) wsptr[1];
4576 z2 = (INT32) wsptr[3];
4577 z3 = (INT32) wsptr[5];
4578 tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
4579 tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
4580 tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
4581 tmp11 = (z1 - z2 - z3) << CONST_BITS;
4583 /* Final output stage */
4585 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
4586 CONST_BITS+PASS1_BITS+3)
4588 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
4589 CONST_BITS+PASS1_BITS+3)
4591 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
4592 CONST_BITS+PASS1_BITS+3)
4594 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
4595 CONST_BITS+PASS1_BITS+3)
4597 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
4598 CONST_BITS+PASS1_BITS+3)
4600 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
4601 CONST_BITS+PASS1_BITS+3)
4604 wsptr += 6; /* advance pointer to next row */
4610 * Perform dequantization and inverse DCT on one block of coefficients,
4611 * producing a 5x10 output block.
4613 * 10-point IDCT in pass 1 (columns), 5-point in pass 2 (rows).
4617 jpeg_idct_5x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4618 JCOEFPTR coef_block,
4619 JSAMPARRAY output_buf, JDIMENSION output_col)
4621 INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
4622 INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
4623 INT32 z1, z2, z3, z4, z5;
4625 ISLOW_MULT_TYPE * quantptr;
4628 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4630 int workspace[5*10]; /* buffers data between passes */
4633 /* Pass 1: process columns from input, store into work array.
4634 * 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
4638 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4640 for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
4643 z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
4645 /* Add fudge factor here for final descale. */
4646 z3 += ONE << (CONST_BITS-PASS1_BITS-1);
4647 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
4648 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
4649 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
4653 tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1), /* c0 = (c4-c8)*2 */
4654 CONST_BITS-PASS1_BITS);
4656 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
4657 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
4659 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
4660 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
4661 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
4663 tmp20 = tmp10 + tmp12;
4664 tmp24 = tmp10 - tmp12;
4665 tmp21 = tmp11 + tmp13;
4666 tmp23 = tmp11 - tmp13;
4670 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
4671 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
4672 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
4673 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
4678 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
4679 z5 = z3 << CONST_BITS;
4681 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
4684 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
4685 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
4687 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
4688 z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));
4690 tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;
4692 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
4693 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
4695 /* Final output stage */
4697 wsptr[5*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
4698 wsptr[5*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
4699 wsptr[5*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
4700 wsptr[5*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
4701 wsptr[5*2] = (int) (tmp22 + tmp12);
4702 wsptr[5*7] = (int) (tmp22 - tmp12);
4703 wsptr[5*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
4704 wsptr[5*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
4705 wsptr[5*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
4706 wsptr[5*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
4709 /* Pass 2: process 10 rows from work array, store into output array.
4710 * 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
4714 for (ctr = 0; ctr < 10; ctr++) {
4715 outptr = output_buf[ctr] + output_col;
4719 /* Add fudge factor here for final descale. */
4720 tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4721 tmp12 <<= CONST_BITS;
4722 tmp13 = (INT32) wsptr[2];
4723 tmp14 = (INT32) wsptr[4];
4724 z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
4725 z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
4733 z2 = (INT32) wsptr[1];
4734 z3 = (INT32) wsptr[3];
4736 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
4737 tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
4738 tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
4740 /* Final output stage */
4742 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp13,
4743 CONST_BITS+PASS1_BITS+3)
4745 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp13,
4746 CONST_BITS+PASS1_BITS+3)
4748 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp14,
4749 CONST_BITS+PASS1_BITS+3)
4751 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp14,
4752 CONST_BITS+PASS1_BITS+3)
4754 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
4755 CONST_BITS+PASS1_BITS+3)
4758 wsptr += 5; /* advance pointer to next row */
4764 * Perform dequantization and inverse DCT on one block of coefficients,
4765 * producing a 4x8 output block.
4767 * 8-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
4771 jpeg_idct_4x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4772 JCOEFPTR coef_block,
4773 JSAMPARRAY output_buf, JDIMENSION output_col)
4775 INT32 tmp0, tmp1, tmp2, tmp3;
4776 INT32 tmp10, tmp11, tmp12, tmp13;
4779 ISLOW_MULT_TYPE * quantptr;
4782 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4784 int workspace[4*8]; /* buffers data between passes */
4787 /* Pass 1: process columns from input, store into work array.
4788 * Note results are scaled up by sqrt(8) compared to a true IDCT;
4789 * furthermore, we scale the results by 2**PASS1_BITS.
4790 * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
4794 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4796 for (ctr = 4; ctr > 0; ctr--) {
4797 /* Due to quantization, we will usually find that many of the input
4798 * coefficients are zero, especially the AC terms. We can exploit this
4799 * by short-circuiting the IDCT calculation for any column in which all
4800 * the AC terms are zero. In that case each output is equal to the
4801 * DC coefficient (with scale factor as needed).
4802 * With typical images and quantization tables, half or more of the
4803 * column DCT calculations can be simplified this way.
4806 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
4807 inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
4808 inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
4809 inptr[DCTSIZE*7] == 0) {
4810 /* AC terms all zero */
4811 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
4822 inptr++; /* advance pointers to next column */
4828 /* Even part: reverse the even part of the forward DCT.
4829 * The rotator is c(-6).
4832 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
4833 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
4835 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
4836 tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
4837 tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
4839 z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
4840 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
4843 /* Add fudge factor here for final descale. */
4844 z2 += ONE << (CONST_BITS-PASS1_BITS-1);
4849 tmp10 = tmp0 + tmp2;
4850 tmp13 = tmp0 - tmp2;
4851 tmp11 = tmp1 + tmp3;
4852 tmp12 = tmp1 - tmp3;
4854 /* Odd part per figure 8; the matrix is unitary and hence its
4855 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
4858 tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
4859 tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
4860 tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
4861 tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
4866 z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* c3 */
4867 z2 = MULTIPLY(z2, - FIX_1_961570560); /* -c3-c5 */
4868 z3 = MULTIPLY(z3, - FIX_0_390180644); /* -c3+c5 */
4872 z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
4873 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* -c1+c3+c5-c7 */
4874 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* c1+c3-c5-c7 */
4878 z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
4879 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* c1+c3-c5+c7 */
4880 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* c1+c3+c5-c7 */
4884 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
4886 wsptr[4*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
4887 wsptr[4*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
4888 wsptr[4*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
4889 wsptr[4*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
4890 wsptr[4*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
4891 wsptr[4*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
4892 wsptr[4*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
4893 wsptr[4*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
4895 inptr++; /* advance pointers to next column */
4900 /* Pass 2: process 8 rows from work array, store into output array.
4901 * 4-point IDCT kernel,
4902 * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
4906 for (ctr = 0; ctr < 8; ctr++) {
4907 outptr = output_buf[ctr] + output_col;
4911 /* Add fudge factor here for final descale. */
4912 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4913 tmp2 = (INT32) wsptr[2];
4915 tmp10 = (tmp0 + tmp2) << CONST_BITS;
4916 tmp12 = (tmp0 - tmp2) << CONST_BITS;
4919 /* Same rotation as in the even part of the 8x8 LL&M IDCT */
4921 z2 = (INT32) wsptr[1];
4922 z3 = (INT32) wsptr[3];
4924 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
4925 tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
4926 tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
4928 /* Final output stage */
4930 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
4931 CONST_BITS+PASS1_BITS+3)
4933 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
4934 CONST_BITS+PASS1_BITS+3)
4936 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
4937 CONST_BITS+PASS1_BITS+3)
4939 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
4940 CONST_BITS+PASS1_BITS+3)
4943 wsptr += 4; /* advance pointer to next row */
4949 * Perform dequantization and inverse DCT on one block of coefficients,
4950 * producing a reduced-size 3x6 output block.
4952 * 6-point IDCT in pass 1 (columns), 3-point in pass 2 (rows).
4956 jpeg_idct_3x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4957 JCOEFPTR coef_block,
4958 JSAMPARRAY output_buf, JDIMENSION output_col)
4960 INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
4963 ISLOW_MULT_TYPE * quantptr;
4966 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4968 int workspace[3*6]; /* buffers data between passes */
4971 /* Pass 1: process columns from input, store into work array.
4972 * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
4976 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4978 for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
4981 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
4982 tmp0 <<= CONST_BITS;
4983 /* Add fudge factor here for final descale. */
4984 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
4985 tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
4986 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
4987 tmp1 = tmp0 + tmp10;
4988 tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
4989 tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
4990 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
4991 tmp10 = tmp1 + tmp0;
4992 tmp12 = tmp1 - tmp0;
4996 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
4997 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
4998 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
4999 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
5000 tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
5001 tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
5002 tmp1 = (z1 - z2 - z3) << PASS1_BITS;
5004 /* Final output stage */
5006 wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
5007 wsptr[3*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
5008 wsptr[3*1] = (int) (tmp11 + tmp1);
5009 wsptr[3*4] = (int) (tmp11 - tmp1);
5010 wsptr[3*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
5011 wsptr[3*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
5014 /* Pass 2: process 6 rows from work array, store into output array.
5015 * 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
5019 for (ctr = 0; ctr < 6; ctr++) {
5020 outptr = output_buf[ctr] + output_col;
5024 /* Add fudge factor here for final descale. */
5025 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
5026 tmp0 <<= CONST_BITS;
5027 tmp2 = (INT32) wsptr[2];
5028 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
5029 tmp10 = tmp0 + tmp12;
5030 tmp2 = tmp0 - tmp12 - tmp12;
5034 tmp12 = (INT32) wsptr[1];
5035 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
5037 /* Final output stage */
5039 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
5040 CONST_BITS+PASS1_BITS+3)
5042 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
5043 CONST_BITS+PASS1_BITS+3)
5045 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
5046 CONST_BITS+PASS1_BITS+3)
5049 wsptr += 3; /* advance pointer to next row */
5055 * Perform dequantization and inverse DCT on one block of coefficients,
5056 * producing a 2x4 output block.
5058 * 4-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
5062 jpeg_idct_2x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
5063 JCOEFPTR coef_block,
5064 JSAMPARRAY output_buf, JDIMENSION output_col)
5066 INT32 tmp0, tmp2, tmp10, tmp12;
5069 ISLOW_MULT_TYPE * quantptr;
5072 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
5074 INT32 workspace[2*4]; /* buffers data between passes */
5077 /* Pass 1: process columns from input, store into work array.
5078 * 4-point IDCT kernel,
5079 * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
5083 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
5085 for (ctr = 0; ctr < 2; ctr++, inptr++, quantptr++, wsptr++) {
5088 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
5089 tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
5091 tmp10 = (tmp0 + tmp2) << CONST_BITS;
5092 tmp12 = (tmp0 - tmp2) << CONST_BITS;
5095 /* Same rotation as in the even part of the 8x8 LL&M IDCT */
5097 z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
5098 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
5100 z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
5101 tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
5102 tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
5104 /* Final output stage */
5106 wsptr[2*0] = tmp10 + tmp0;
5107 wsptr[2*3] = tmp10 - tmp0;
5108 wsptr[2*1] = tmp12 + tmp2;
5109 wsptr[2*2] = tmp12 - tmp2;
5112 /* Pass 2: process 4 rows from work array, store into output array. */
5115 for (ctr = 0; ctr < 4; ctr++) {
5116 outptr = output_buf[ctr] + output_col;
5120 /* Add fudge factor here for final descale. */
5121 tmp10 = wsptr[0] + (ONE << (CONST_BITS+2));
5127 /* Final output stage */
5129 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS+3)
5131 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS+3)
5134 wsptr += 2; /* advance pointer to next row */
5140 * Perform dequantization and inverse DCT on one block of coefficients,
5141 * producing a 1x2 output block.
5143 * 2-point IDCT in pass 1 (columns), 1-point in pass 2 (rows).
5147 jpeg_idct_1x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
5148 JCOEFPTR coef_block,
5149 JSAMPARRAY output_buf, JDIMENSION output_col)
5152 ISLOW_MULT_TYPE * quantptr;
5153 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
5156 /* Process 1 column from input, store into output array. */
5158 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
5162 tmp0 = DEQUANTIZE(coef_block[DCTSIZE*0], quantptr[DCTSIZE*0]);
5163 /* Add fudge factor here for final descale. */
5168 tmp1 = DEQUANTIZE(coef_block[DCTSIZE*1], quantptr[DCTSIZE*1]);
5170 /* Final output stage */
5172 output_buf[0][output_col] = range_limit[(int) RIGHT_SHIFT(tmp0 + tmp1, 3)
5174 output_buf[1][output_col] = range_limit[(int) RIGHT_SHIFT(tmp0 - tmp1, 3)
5178 #endif /* IDCT_SCALING_SUPPORTED */
5179 #endif /* DCT_ISLOW_SUPPORTED */