4 * This file was part of the Independent JPEG Group's software.
5 * Copyright (C) 1991-1998, Thomas G. Lane.
6 * Modification developed 2002-2009 by Guido Vollbeding.
7 * libjpeg-turbo Modifications:
8 * Copyright (C) 2015, D. R. Commander
9 * For conditions of distribution and use, see the accompanying README file.
11 * This file contains a slow-but-accurate integer implementation of the
12 * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine
13 * must also perform dequantization of the input coefficients.
15 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
16 * on each row (or vice versa, but it's more convenient to emit a row at
17 * a time). Direct algorithms are also available, but they are much more
18 * complex and seem not to be any faster when reduced to code.
20 * This implementation is based on an algorithm described in
21 * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
22 * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
23 * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
24 * The primary algorithm described there uses 11 multiplies and 29 adds.
25 * We use their alternate method with 12 multiplies and 32 adds.
26 * The advantage of this method is that no data path contains more than one
27 * multiplication; this allows a very simple and accurate implementation in
28 * scaled fixed-point arithmetic, with a minimal number of shifts.
30 * We also provide IDCT routines with various output sample block sizes for
31 * direct resolution reduction or enlargement without additional resampling:
32 * NxN (N=1...16) pixels for one 8x8 input DCT block.
34 * For N<8 we simply take the corresponding low-frequency coefficients of
35 * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block
36 * to yield the downscaled outputs.
37 * This can be seen as direct low-pass downsampling from the DCT domain
38 * point of view rather than the usual spatial domain point of view,
39 * yielding significant computational savings and results at least
40 * as good as common bilinear (averaging) spatial downsampling.
42 * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as
43 * lower frequencies and higher frequencies assumed to be zero.
44 * It turns out that the computational effort is similar to the 8x8 IDCT
45 * regarding the output size.
46 * Furthermore, the scaling and descaling is the same for all IDCT sizes.
48 * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases
49 * since there would be too many additional constants to pre-calculate.
52 #define JPEG_INTERNALS
55 #include "jdct.h" /* Private declarations for DCT subsystem */
57 #ifdef DCT_ISLOW_SUPPORTED
61 * This module is specialized to the case DCTSIZE = 8.
65 Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
70 * The poop on this scaling stuff is as follows:
72 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
73 * larger than the true IDCT outputs. The final outputs are therefore
74 * a factor of N larger than desired; since N=8 this can be cured by
75 * a simple right shift at the end of the algorithm. The advantage of
76 * this arrangement is that we save two multiplications per 1-D IDCT,
77 * because the y0 and y4 inputs need not be divided by sqrt(N).
79 * We have to do addition and subtraction of the integer inputs, which
80 * is no problem, and multiplication by fractional constants, which is
81 * a problem to do in integer arithmetic. We multiply all the constants
82 * by CONST_SCALE and convert them to integer constants (thus retaining
83 * CONST_BITS bits of precision in the constants). After doing a
84 * multiplication we have to divide the product by CONST_SCALE, with proper
85 * rounding, to produce the correct output. This division can be done
86 * cheaply as a right shift of CONST_BITS bits. We postpone shifting
87 * as long as possible so that partial sums can be added together with
88 * full fractional precision.
90 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
91 * they are represented to better-than-integral precision. These outputs
92 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
93 * with the recommended scaling. (To scale up 12-bit sample data further, an
94 * intermediate INT32 array would be needed.)
96 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
97 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
98 * shows that the values given below are the most effective.
101 #if BITS_IN_JSAMPLE == 8
102 #define CONST_BITS 13
105 #define CONST_BITS 13
106 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
109 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
110 * causing a lot of useless floating-point operations at run time.
111 * To get around this we use the following pre-calculated constants.
112 * If you change CONST_BITS you may want to add appropriate values.
113 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
117 #define FIX_0_298631336 ((INT32) 2446) /* FIX(0.298631336) */
118 #define FIX_0_390180644 ((INT32) 3196) /* FIX(0.390180644) */
119 #define FIX_0_541196100 ((INT32) 4433) /* FIX(0.541196100) */
120 #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
121 #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
122 #define FIX_1_175875602 ((INT32) 9633) /* FIX(1.175875602) */
123 #define FIX_1_501321110 ((INT32) 12299) /* FIX(1.501321110) */
124 #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
125 #define FIX_1_961570560 ((INT32) 16069) /* FIX(1.961570560) */
126 #define FIX_2_053119869 ((INT32) 16819) /* FIX(2.053119869) */
127 #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
128 #define FIX_3_072711026 ((INT32) 25172) /* FIX(3.072711026) */
130 #define FIX_0_298631336 FIX(0.298631336)
131 #define FIX_0_390180644 FIX(0.390180644)
132 #define FIX_0_541196100 FIX(0.541196100)
133 #define FIX_0_765366865 FIX(0.765366865)
134 #define FIX_0_899976223 FIX(0.899976223)
135 #define FIX_1_175875602 FIX(1.175875602)
136 #define FIX_1_501321110 FIX(1.501321110)
137 #define FIX_1_847759065 FIX(1.847759065)
138 #define FIX_1_961570560 FIX(1.961570560)
139 #define FIX_2_053119869 FIX(2.053119869)
140 #define FIX_2_562915447 FIX(2.562915447)
141 #define FIX_3_072711026 FIX(3.072711026)
145 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
146 * For 8-bit samples with the recommended scaling, all the variable
147 * and constant values involved are no more than 16 bits wide, so a
148 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
149 * For 12-bit samples, a full 32-bit multiplication will be needed.
152 #if BITS_IN_JSAMPLE == 8
153 #define MULTIPLY(var,const) MULTIPLY16C16(var,const)
155 #define MULTIPLY(var,const) ((var) * (const))
159 /* Dequantize a coefficient by multiplying it by the multiplier-table
160 * entry; produce an int result. In this module, both inputs and result
161 * are 16 bits or less, so either int or short multiply will work.
164 #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))
168 * Perform dequantization and inverse DCT on one block of coefficients.
172 jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
174 JSAMPARRAY output_buf, JDIMENSION output_col)
176 INT32 tmp0, tmp1, tmp2, tmp3;
177 INT32 tmp10, tmp11, tmp12, tmp13;
178 INT32 z1, z2, z3, z4, z5;
180 ISLOW_MULT_TYPE * quantptr;
183 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
185 int workspace[DCTSIZE2]; /* buffers data between passes */
188 /* Pass 1: process columns from input, store into work array. */
189 /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
190 /* furthermore, we scale the results by 2**PASS1_BITS. */
193 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
195 for (ctr = DCTSIZE; ctr > 0; ctr--) {
196 /* Due to quantization, we will usually find that many of the input
197 * coefficients are zero, especially the AC terms. We can exploit this
198 * by short-circuiting the IDCT calculation for any column in which all
199 * the AC terms are zero. In that case each output is equal to the
200 * DC coefficient (with scale factor as needed).
201 * With typical images and quantization tables, half or more of the
202 * column DCT calculations can be simplified this way.
205 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
206 inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
207 inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
208 inptr[DCTSIZE*7] == 0) {
209 /* AC terms all zero */
210 int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]),
213 wsptr[DCTSIZE*0] = dcval;
214 wsptr[DCTSIZE*1] = dcval;
215 wsptr[DCTSIZE*2] = dcval;
216 wsptr[DCTSIZE*3] = dcval;
217 wsptr[DCTSIZE*4] = dcval;
218 wsptr[DCTSIZE*5] = dcval;
219 wsptr[DCTSIZE*6] = dcval;
220 wsptr[DCTSIZE*7] = dcval;
222 inptr++; /* advance pointers to next column */
228 /* Even part: reverse the even part of the forward DCT. */
229 /* The rotator is sqrt(2)*c(-6). */
231 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
232 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
234 z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
235 tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
236 tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
238 z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
239 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
241 tmp0 = LEFT_SHIFT(z2 + z3, CONST_BITS);
242 tmp1 = LEFT_SHIFT(z2 - z3, CONST_BITS);
249 /* Odd part per figure 8; the matrix is unitary and hence its
250 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
253 tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
254 tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
255 tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
256 tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
262 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
264 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
265 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
266 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
267 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
268 z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
269 z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
270 z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
271 z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
281 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
283 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
284 wsptr[DCTSIZE*7] = (int) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
285 wsptr[DCTSIZE*1] = (int) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
286 wsptr[DCTSIZE*6] = (int) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
287 wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
288 wsptr[DCTSIZE*5] = (int) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
289 wsptr[DCTSIZE*3] = (int) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
290 wsptr[DCTSIZE*4] = (int) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
292 inptr++; /* advance pointers to next column */
297 /* Pass 2: process rows from work array, store into output array. */
298 /* Note that we must descale the results by a factor of 8 == 2**3, */
299 /* and also undo the PASS1_BITS scaling. */
302 for (ctr = 0; ctr < DCTSIZE; ctr++) {
303 outptr = output_buf[ctr] + output_col;
304 /* Rows of zeroes can be exploited in the same way as we did with columns.
305 * However, the column calculation has created many nonzero AC terms, so
306 * the simplification applies less often (typically 5% to 10% of the time).
307 * On machines with very fast multiplication, it's possible that the
308 * test takes more time than it's worth. In that case this section
309 * may be commented out.
312 #ifndef NO_ZERO_ROW_TEST
313 if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
314 wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
315 /* AC terms all zero */
316 JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
328 wsptr += DCTSIZE; /* advance pointer to next row */
333 /* Even part: reverse the even part of the forward DCT. */
334 /* The rotator is sqrt(2)*c(-6). */
336 z2 = (INT32) wsptr[2];
337 z3 = (INT32) wsptr[6];
339 z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
340 tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
341 tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
343 tmp0 = LEFT_SHIFT((INT32) wsptr[0] + (INT32) wsptr[4], CONST_BITS);
344 tmp1 = LEFT_SHIFT((INT32) wsptr[0] - (INT32) wsptr[4], CONST_BITS);
351 /* Odd part per figure 8; the matrix is unitary and hence its
352 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
355 tmp0 = (INT32) wsptr[7];
356 tmp1 = (INT32) wsptr[5];
357 tmp2 = (INT32) wsptr[3];
358 tmp3 = (INT32) wsptr[1];
364 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
366 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
367 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
368 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
369 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
370 z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
371 z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
372 z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
373 z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
383 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
385 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp3,
386 CONST_BITS+PASS1_BITS+3)
388 outptr[7] = range_limit[(int) DESCALE(tmp10 - tmp3,
389 CONST_BITS+PASS1_BITS+3)
391 outptr[1] = range_limit[(int) DESCALE(tmp11 + tmp2,
392 CONST_BITS+PASS1_BITS+3)
394 outptr[6] = range_limit[(int) DESCALE(tmp11 - tmp2,
395 CONST_BITS+PASS1_BITS+3)
397 outptr[2] = range_limit[(int) DESCALE(tmp12 + tmp1,
398 CONST_BITS+PASS1_BITS+3)
400 outptr[5] = range_limit[(int) DESCALE(tmp12 - tmp1,
401 CONST_BITS+PASS1_BITS+3)
403 outptr[3] = range_limit[(int) DESCALE(tmp13 + tmp0,
404 CONST_BITS+PASS1_BITS+3)
406 outptr[4] = range_limit[(int) DESCALE(tmp13 - tmp0,
407 CONST_BITS+PASS1_BITS+3)
410 wsptr += DCTSIZE; /* advance pointer to next row */
414 #ifdef IDCT_SCALING_SUPPORTED
418 * Perform dequantization and inverse DCT on one block of coefficients,
419 * producing a 7x7 output block.
421 * Optimized algorithm with 12 multiplications in the 1-D kernel.
422 * cK represents sqrt(2) * cos(K*pi/14).
426 jpeg_idct_7x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
428 JSAMPARRAY output_buf, JDIMENSION output_col)
430 INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
433 ISLOW_MULT_TYPE * quantptr;
436 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
438 int workspace[7*7]; /* buffers data between passes */
441 /* Pass 1: process columns from input, store into work array. */
444 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
446 for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
449 tmp13 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
450 tmp13 = LEFT_SHIFT(tmp13, CONST_BITS);
451 /* Add fudge factor here for final descale. */
452 tmp13 += ONE << (CONST_BITS-PASS1_BITS-1);
454 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
455 z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
456 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
458 tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
459 tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
460 tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
463 tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
464 tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
465 tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
466 tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
470 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
471 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
472 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
474 tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
475 tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
478 tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
480 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
482 tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
484 /* Final output stage */
486 wsptr[7*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
487 wsptr[7*6] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
488 wsptr[7*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
489 wsptr[7*5] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
490 wsptr[7*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
491 wsptr[7*4] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
492 wsptr[7*3] = (int) RIGHT_SHIFT(tmp13, CONST_BITS-PASS1_BITS);
495 /* Pass 2: process 7 rows from work array, store into output array. */
498 for (ctr = 0; ctr < 7; ctr++) {
499 outptr = output_buf[ctr] + output_col;
503 /* Add fudge factor here for final descale. */
504 tmp13 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
505 tmp13 = LEFT_SHIFT(tmp13, CONST_BITS);
507 z1 = (INT32) wsptr[2];
508 z2 = (INT32) wsptr[4];
509 z3 = (INT32) wsptr[6];
511 tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
512 tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
513 tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
516 tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
517 tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
518 tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
519 tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
523 z1 = (INT32) wsptr[1];
524 z2 = (INT32) wsptr[3];
525 z3 = (INT32) wsptr[5];
527 tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
528 tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
531 tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
533 z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
535 tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
537 /* Final output stage */
539 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
540 CONST_BITS+PASS1_BITS+3)
542 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
543 CONST_BITS+PASS1_BITS+3)
545 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
546 CONST_BITS+PASS1_BITS+3)
548 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
549 CONST_BITS+PASS1_BITS+3)
551 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
552 CONST_BITS+PASS1_BITS+3)
554 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
555 CONST_BITS+PASS1_BITS+3)
557 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13,
558 CONST_BITS+PASS1_BITS+3)
561 wsptr += 7; /* advance pointer to next row */
567 * Perform dequantization and inverse DCT on one block of coefficients,
568 * producing a reduced-size 6x6 output block.
570 * Optimized algorithm with 3 multiplications in the 1-D kernel.
571 * cK represents sqrt(2) * cos(K*pi/12).
575 jpeg_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
577 JSAMPARRAY output_buf, JDIMENSION output_col)
579 INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
582 ISLOW_MULT_TYPE * quantptr;
585 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
587 int workspace[6*6]; /* buffers data between passes */
590 /* Pass 1: process columns from input, store into work array. */
593 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
595 for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
598 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
599 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
600 /* Add fudge factor here for final descale. */
601 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
602 tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
603 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
605 tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
606 tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
607 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
613 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
614 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
615 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
616 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
617 tmp0 = tmp1 + LEFT_SHIFT(z1 + z2, CONST_BITS);
618 tmp2 = tmp1 + LEFT_SHIFT(z3 - z2, CONST_BITS);
619 tmp1 = LEFT_SHIFT(z1 - z2 - z3, PASS1_BITS);
621 /* Final output stage */
623 wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
624 wsptr[6*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
625 wsptr[6*1] = (int) (tmp11 + tmp1);
626 wsptr[6*4] = (int) (tmp11 - tmp1);
627 wsptr[6*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
628 wsptr[6*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
631 /* Pass 2: process 6 rows from work array, store into output array. */
634 for (ctr = 0; ctr < 6; ctr++) {
635 outptr = output_buf[ctr] + output_col;
639 /* Add fudge factor here for final descale. */
640 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
641 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
642 tmp2 = (INT32) wsptr[4];
643 tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
645 tmp11 = tmp0 - tmp10 - tmp10;
646 tmp10 = (INT32) wsptr[2];
647 tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
653 z1 = (INT32) wsptr[1];
654 z2 = (INT32) wsptr[3];
655 z3 = (INT32) wsptr[5];
656 tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
657 tmp0 = tmp1 + LEFT_SHIFT(z1 + z2, CONST_BITS);
658 tmp2 = tmp1 + LEFT_SHIFT(z3 - z2, CONST_BITS);
659 tmp1 = LEFT_SHIFT(z1 - z2 - z3, CONST_BITS);
661 /* Final output stage */
663 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
664 CONST_BITS+PASS1_BITS+3)
666 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
667 CONST_BITS+PASS1_BITS+3)
669 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
670 CONST_BITS+PASS1_BITS+3)
672 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
673 CONST_BITS+PASS1_BITS+3)
675 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
676 CONST_BITS+PASS1_BITS+3)
678 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
679 CONST_BITS+PASS1_BITS+3)
682 wsptr += 6; /* advance pointer to next row */
688 * Perform dequantization and inverse DCT on one block of coefficients,
689 * producing a reduced-size 5x5 output block.
691 * Optimized algorithm with 5 multiplications in the 1-D kernel.
692 * cK represents sqrt(2) * cos(K*pi/10).
696 jpeg_idct_5x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
698 JSAMPARRAY output_buf, JDIMENSION output_col)
700 INT32 tmp0, tmp1, tmp10, tmp11, tmp12;
703 ISLOW_MULT_TYPE * quantptr;
706 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
708 int workspace[5*5]; /* buffers data between passes */
711 /* Pass 1: process columns from input, store into work array. */
714 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
716 for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
719 tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
720 tmp12 = LEFT_SHIFT(tmp12, CONST_BITS);
721 /* Add fudge factor here for final descale. */
722 tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
723 tmp0 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
724 tmp1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
725 z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
726 z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
730 tmp12 -= LEFT_SHIFT(z2, 2);
734 z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
735 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
737 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
738 tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
739 tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
741 /* Final output stage */
743 wsptr[5*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
744 wsptr[5*4] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
745 wsptr[5*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
746 wsptr[5*3] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
747 wsptr[5*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
750 /* Pass 2: process 5 rows from work array, store into output array. */
753 for (ctr = 0; ctr < 5; ctr++) {
754 outptr = output_buf[ctr] + output_col;
758 /* Add fudge factor here for final descale. */
759 tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
760 tmp12 = LEFT_SHIFT(tmp12, CONST_BITS);
761 tmp0 = (INT32) wsptr[2];
762 tmp1 = (INT32) wsptr[4];
763 z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
764 z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
768 tmp12 -= LEFT_SHIFT(z2, 2);
772 z2 = (INT32) wsptr[1];
773 z3 = (INT32) wsptr[3];
775 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
776 tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
777 tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
779 /* Final output stage */
781 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
782 CONST_BITS+PASS1_BITS+3)
784 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
785 CONST_BITS+PASS1_BITS+3)
787 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
788 CONST_BITS+PASS1_BITS+3)
790 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
791 CONST_BITS+PASS1_BITS+3)
793 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
794 CONST_BITS+PASS1_BITS+3)
797 wsptr += 5; /* advance pointer to next row */
803 * Perform dequantization and inverse DCT on one block of coefficients,
804 * producing a reduced-size 3x3 output block.
806 * Optimized algorithm with 2 multiplications in the 1-D kernel.
807 * cK represents sqrt(2) * cos(K*pi/6).
811 jpeg_idct_3x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
813 JSAMPARRAY output_buf, JDIMENSION output_col)
815 INT32 tmp0, tmp2, tmp10, tmp12;
817 ISLOW_MULT_TYPE * quantptr;
820 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
822 int workspace[3*3]; /* buffers data between passes */
825 /* Pass 1: process columns from input, store into work array. */
828 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
830 for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
833 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
834 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
835 /* Add fudge factor here for final descale. */
836 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
837 tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
838 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
839 tmp10 = tmp0 + tmp12;
840 tmp2 = tmp0 - tmp12 - tmp12;
844 tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
845 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
847 /* Final output stage */
849 wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
850 wsptr[3*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
851 wsptr[3*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
854 /* Pass 2: process 3 rows from work array, store into output array. */
857 for (ctr = 0; ctr < 3; ctr++) {
858 outptr = output_buf[ctr] + output_col;
862 /* Add fudge factor here for final descale. */
863 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
864 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
865 tmp2 = (INT32) wsptr[2];
866 tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
867 tmp10 = tmp0 + tmp12;
868 tmp2 = tmp0 - tmp12 - tmp12;
872 tmp12 = (INT32) wsptr[1];
873 tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
875 /* Final output stage */
877 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
878 CONST_BITS+PASS1_BITS+3)
880 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
881 CONST_BITS+PASS1_BITS+3)
883 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
884 CONST_BITS+PASS1_BITS+3)
887 wsptr += 3; /* advance pointer to next row */
893 * Perform dequantization and inverse DCT on one block of coefficients,
894 * producing a 9x9 output block.
896 * Optimized algorithm with 10 multiplications in the 1-D kernel.
897 * cK represents sqrt(2) * cos(K*pi/18).
901 jpeg_idct_9x9 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
903 JSAMPARRAY output_buf, JDIMENSION output_col)
905 INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14;
906 INT32 z1, z2, z3, z4;
908 ISLOW_MULT_TYPE * quantptr;
911 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
913 int workspace[8*9]; /* buffers data between passes */
916 /* Pass 1: process columns from input, store into work array. */
919 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
921 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
924 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
925 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
926 /* Add fudge factor here for final descale. */
927 tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
929 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
930 z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
931 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
933 tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
935 tmp2 = tmp0 - tmp3 - tmp3;
937 tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
939 tmp14 = tmp2 - tmp0 - tmp0;
941 tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
942 tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
943 tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
945 tmp10 = tmp1 + tmp0 - tmp3;
946 tmp12 = tmp1 - tmp0 + tmp2;
947 tmp13 = tmp1 - tmp2 + tmp3;
951 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
952 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
953 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
954 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
956 z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
958 tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
959 tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
960 tmp0 = tmp2 + tmp3 - z2;
961 tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
964 tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
966 /* Final output stage */
968 wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
969 wsptr[8*8] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
970 wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
971 wsptr[8*7] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
972 wsptr[8*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
973 wsptr[8*6] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
974 wsptr[8*3] = (int) RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS-PASS1_BITS);
975 wsptr[8*5] = (int) RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS-PASS1_BITS);
976 wsptr[8*4] = (int) RIGHT_SHIFT(tmp14, CONST_BITS-PASS1_BITS);
979 /* Pass 2: process 9 rows from work array, store into output array. */
982 for (ctr = 0; ctr < 9; ctr++) {
983 outptr = output_buf[ctr] + output_col;
987 /* Add fudge factor here for final descale. */
988 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
989 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
991 z1 = (INT32) wsptr[2];
992 z2 = (INT32) wsptr[4];
993 z3 = (INT32) wsptr[6];
995 tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
997 tmp2 = tmp0 - tmp3 - tmp3;
999 tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1000 tmp11 = tmp2 + tmp0;
1001 tmp14 = tmp2 - tmp0 - tmp0;
1003 tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1004 tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1005 tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1007 tmp10 = tmp1 + tmp0 - tmp3;
1008 tmp12 = tmp1 - tmp0 + tmp2;
1009 tmp13 = tmp1 - tmp2 + tmp3;
1013 z1 = (INT32) wsptr[1];
1014 z2 = (INT32) wsptr[3];
1015 z3 = (INT32) wsptr[5];
1016 z4 = (INT32) wsptr[7];
1018 z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
1020 tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1021 tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1022 tmp0 = tmp2 + tmp3 - z2;
1023 tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1026 tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1028 /* Final output stage */
1030 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1031 CONST_BITS+PASS1_BITS+3)
1033 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1034 CONST_BITS+PASS1_BITS+3)
1036 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
1037 CONST_BITS+PASS1_BITS+3)
1039 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
1040 CONST_BITS+PASS1_BITS+3)
1042 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
1043 CONST_BITS+PASS1_BITS+3)
1045 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
1046 CONST_BITS+PASS1_BITS+3)
1048 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp3,
1049 CONST_BITS+PASS1_BITS+3)
1051 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp3,
1052 CONST_BITS+PASS1_BITS+3)
1054 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp14,
1055 CONST_BITS+PASS1_BITS+3)
1058 wsptr += 8; /* advance pointer to next row */
1064 * Perform dequantization and inverse DCT on one block of coefficients,
1065 * producing a 10x10 output block.
1067 * Optimized algorithm with 12 multiplications in the 1-D kernel.
1068 * cK represents sqrt(2) * cos(K*pi/20).
1072 jpeg_idct_10x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1073 JCOEFPTR coef_block,
1074 JSAMPARRAY output_buf, JDIMENSION output_col)
1076 INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1077 INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
1078 INT32 z1, z2, z3, z4, z5;
1080 ISLOW_MULT_TYPE * quantptr;
1083 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1085 int workspace[8*10]; /* buffers data between passes */
1088 /* Pass 1: process columns from input, store into work array. */
1091 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1093 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1096 z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1097 z3 = LEFT_SHIFT(z3, CONST_BITS);
1098 /* Add fudge factor here for final descale. */
1099 z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1100 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1101 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1102 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1106 tmp22 = RIGHT_SHIFT(z3 - LEFT_SHIFT(z1 - z2, 1),
1107 CONST_BITS-PASS1_BITS); /* c0 = (c4-c8)*2 */
1109 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1110 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1112 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1113 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1114 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1116 tmp20 = tmp10 + tmp12;
1117 tmp24 = tmp10 - tmp12;
1118 tmp21 = tmp11 + tmp13;
1119 tmp23 = tmp11 - tmp13;
1123 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1124 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1125 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1126 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1131 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1132 z5 = LEFT_SHIFT(z3, CONST_BITS);
1134 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1137 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1138 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1140 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1141 z4 = z5 - tmp12 - LEFT_SHIFT(tmp13, CONST_BITS - 1);
1143 tmp12 = LEFT_SHIFT(z1 - tmp13 - z3, PASS1_BITS);
1145 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1146 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1148 /* Final output stage */
1150 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1151 wsptr[8*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1152 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1153 wsptr[8*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1154 wsptr[8*2] = (int) (tmp22 + tmp12);
1155 wsptr[8*7] = (int) (tmp22 - tmp12);
1156 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1157 wsptr[8*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1158 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1159 wsptr[8*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1162 /* Pass 2: process 10 rows from work array, store into output array. */
1165 for (ctr = 0; ctr < 10; ctr++) {
1166 outptr = output_buf[ctr] + output_col;
1170 /* Add fudge factor here for final descale. */
1171 z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1172 z3 = LEFT_SHIFT(z3, CONST_BITS);
1173 z4 = (INT32) wsptr[4];
1174 z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1175 z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1179 tmp22 = z3 - LEFT_SHIFT(z1 - z2, 1); /* c0 = (c4-c8)*2 */
1181 z2 = (INT32) wsptr[2];
1182 z3 = (INT32) wsptr[6];
1184 z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1185 tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1186 tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1188 tmp20 = tmp10 + tmp12;
1189 tmp24 = tmp10 - tmp12;
1190 tmp21 = tmp11 + tmp13;
1191 tmp23 = tmp11 - tmp13;
1195 z1 = (INT32) wsptr[1];
1196 z2 = (INT32) wsptr[3];
1197 z3 = (INT32) wsptr[5];
1198 z3 = LEFT_SHIFT(z3, CONST_BITS);
1199 z4 = (INT32) wsptr[7];
1204 tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1206 z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1209 tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1210 tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1212 z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1213 z4 = z3 - tmp12 - LEFT_SHIFT(tmp13, CONST_BITS - 1);
1215 tmp12 = LEFT_SHIFT(z1 - tmp13, CONST_BITS) - z3;
1217 tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1218 tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1220 /* Final output stage */
1222 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1223 CONST_BITS+PASS1_BITS+3)
1225 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1226 CONST_BITS+PASS1_BITS+3)
1228 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1229 CONST_BITS+PASS1_BITS+3)
1231 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1232 CONST_BITS+PASS1_BITS+3)
1234 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1235 CONST_BITS+PASS1_BITS+3)
1237 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1238 CONST_BITS+PASS1_BITS+3)
1240 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1241 CONST_BITS+PASS1_BITS+3)
1243 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1244 CONST_BITS+PASS1_BITS+3)
1246 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1247 CONST_BITS+PASS1_BITS+3)
1249 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1250 CONST_BITS+PASS1_BITS+3)
1253 wsptr += 8; /* advance pointer to next row */
1259 * Perform dequantization and inverse DCT on one block of coefficients,
1260 * producing a 11x11 output block.
1262 * Optimized algorithm with 24 multiplications in the 1-D kernel.
1263 * cK represents sqrt(2) * cos(K*pi/22).
1267 jpeg_idct_11x11 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1268 JCOEFPTR coef_block,
1269 JSAMPARRAY output_buf, JDIMENSION output_col)
1271 INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1272 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1273 INT32 z1, z2, z3, z4;
1275 ISLOW_MULT_TYPE * quantptr;
1278 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1280 int workspace[8*11]; /* buffers data between passes */
1283 /* Pass 1: process columns from input, store into work array. */
1286 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1288 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1291 tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1292 tmp10 = LEFT_SHIFT(tmp10, CONST_BITS);
1293 /* Add fudge factor here for final descale. */
1294 tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
1296 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1297 z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1298 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1300 tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1301 tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1303 tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1305 tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1306 tmp21 = tmp20 + tmp23 + tmp25 -
1307 MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1308 tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1309 tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1311 tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1312 tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1313 MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1314 tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1318 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1319 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1320 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1321 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1324 tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1325 tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1326 tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1327 tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1328 tmp10 = tmp11 + tmp12 + tmp13 -
1329 MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1330 z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1331 tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1332 tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1333 z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1335 tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1336 tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1337 MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1338 MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1340 /* Final output stage */
1342 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1343 wsptr[8*10] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1344 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1345 wsptr[8*9] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1346 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1347 wsptr[8*8] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1348 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1349 wsptr[8*7] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1350 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1351 wsptr[8*6] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1352 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25, CONST_BITS-PASS1_BITS);
1355 /* Pass 2: process 11 rows from work array, store into output array. */
1358 for (ctr = 0; ctr < 11; ctr++) {
1359 outptr = output_buf[ctr] + output_col;
1363 /* Add fudge factor here for final descale. */
1364 tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1365 tmp10 = LEFT_SHIFT(tmp10, CONST_BITS);
1367 z1 = (INT32) wsptr[2];
1368 z2 = (INT32) wsptr[4];
1369 z3 = (INT32) wsptr[6];
1371 tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1372 tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1374 tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1376 tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1377 tmp21 = tmp20 + tmp23 + tmp25 -
1378 MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1379 tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1380 tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1382 tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1383 tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1384 MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1385 tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1389 z1 = (INT32) wsptr[1];
1390 z2 = (INT32) wsptr[3];
1391 z3 = (INT32) wsptr[5];
1392 z4 = (INT32) wsptr[7];
1395 tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1396 tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1397 tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1398 tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1399 tmp10 = tmp11 + tmp12 + tmp13 -
1400 MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1401 z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1402 tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1403 tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1404 z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1406 tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1407 tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1408 MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1409 MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1411 /* Final output stage */
1413 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1414 CONST_BITS+PASS1_BITS+3)
1416 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1417 CONST_BITS+PASS1_BITS+3)
1419 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1420 CONST_BITS+PASS1_BITS+3)
1422 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1423 CONST_BITS+PASS1_BITS+3)
1425 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1426 CONST_BITS+PASS1_BITS+3)
1428 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1429 CONST_BITS+PASS1_BITS+3)
1431 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1432 CONST_BITS+PASS1_BITS+3)
1434 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1435 CONST_BITS+PASS1_BITS+3)
1437 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1438 CONST_BITS+PASS1_BITS+3)
1440 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1441 CONST_BITS+PASS1_BITS+3)
1443 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25,
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 12x12 output block.
1456 * Optimized algorithm with 15 multiplications in the 1-D kernel.
1457 * cK represents sqrt(2) * cos(K*pi/24).
1461 jpeg_idct_12x12 (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, tmp15;
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*12]; /* 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 z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1486 z3 = LEFT_SHIFT(z3, CONST_BITS);
1487 /* Add fudge factor here for final descale. */
1488 z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1490 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1491 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1496 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1497 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1498 z1 = LEFT_SHIFT(z1, CONST_BITS);
1499 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1500 z2 = LEFT_SHIFT(z2, CONST_BITS);
1509 tmp20 = tmp10 + tmp12;
1510 tmp25 = tmp10 - tmp12;
1512 tmp12 = z4 - z1 - z2;
1514 tmp22 = tmp11 + tmp12;
1515 tmp23 = tmp11 - tmp12;
1519 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1520 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1521 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1522 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1524 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1525 tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1528 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1529 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1530 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1531 tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1532 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1533 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1534 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1535 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1539 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1540 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1541 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1543 /* Final output stage */
1545 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1546 wsptr[8*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1547 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1548 wsptr[8*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1549 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1550 wsptr[8*9] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1551 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1552 wsptr[8*8] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1553 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1554 wsptr[8*7] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1555 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1556 wsptr[8*6] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1559 /* Pass 2: process 12 rows from work array, store into output array. */
1562 for (ctr = 0; ctr < 12; ctr++) {
1563 outptr = output_buf[ctr] + output_col;
1567 /* Add fudge factor here for final descale. */
1568 z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1569 z3 = LEFT_SHIFT(z3, CONST_BITS);
1571 z4 = (INT32) wsptr[4];
1572 z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1577 z1 = (INT32) wsptr[2];
1578 z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1579 z1 = LEFT_SHIFT(z1, CONST_BITS);
1580 z2 = (INT32) wsptr[6];
1581 z2 = LEFT_SHIFT(z2, CONST_BITS);
1590 tmp20 = tmp10 + tmp12;
1591 tmp25 = tmp10 - tmp12;
1593 tmp12 = z4 - z1 - z2;
1595 tmp22 = tmp11 + tmp12;
1596 tmp23 = tmp11 - tmp12;
1600 z1 = (INT32) wsptr[1];
1601 z2 = (INT32) wsptr[3];
1602 z3 = (INT32) wsptr[5];
1603 z4 = (INT32) wsptr[7];
1605 tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1606 tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1609 tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1610 tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1611 tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1612 tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1613 tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1614 tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1615 tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1616 MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1620 z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1621 tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1622 tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1624 /* Final output stage */
1626 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1627 CONST_BITS+PASS1_BITS+3)
1629 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1630 CONST_BITS+PASS1_BITS+3)
1632 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1633 CONST_BITS+PASS1_BITS+3)
1635 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1636 CONST_BITS+PASS1_BITS+3)
1638 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1639 CONST_BITS+PASS1_BITS+3)
1641 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1642 CONST_BITS+PASS1_BITS+3)
1644 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1645 CONST_BITS+PASS1_BITS+3)
1647 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1648 CONST_BITS+PASS1_BITS+3)
1650 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1651 CONST_BITS+PASS1_BITS+3)
1653 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1654 CONST_BITS+PASS1_BITS+3)
1656 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1657 CONST_BITS+PASS1_BITS+3)
1659 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1660 CONST_BITS+PASS1_BITS+3)
1663 wsptr += 8; /* advance pointer to next row */
1669 * Perform dequantization and inverse DCT on one block of coefficients,
1670 * producing a 13x13 output block.
1672 * Optimized algorithm with 29 multiplications in the 1-D kernel.
1673 * cK represents sqrt(2) * cos(K*pi/26).
1677 jpeg_idct_13x13 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1678 JCOEFPTR coef_block,
1679 JSAMPARRAY output_buf, JDIMENSION output_col)
1681 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1682 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1683 INT32 z1, z2, z3, z4;
1685 ISLOW_MULT_TYPE * quantptr;
1688 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1690 int workspace[8*13]; /* buffers data between passes */
1693 /* Pass 1: process columns from input, store into work array. */
1696 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1698 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1701 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1702 z1 = LEFT_SHIFT(z1, CONST_BITS);
1703 /* Add fudge factor here for final descale. */
1704 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
1706 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1707 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1708 z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1713 tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1714 tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1716 tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1717 tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1719 tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1720 tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1722 tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1723 tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1725 tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1726 tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1728 tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1729 tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1731 tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1735 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1736 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1737 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1738 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1740 tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1741 tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1743 tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1744 tmp10 = tmp11 + tmp12 + tmp13 -
1745 MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1746 tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
1747 tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1748 tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1749 tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
1751 tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1752 tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
1755 tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1756 tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1757 MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1758 z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1760 tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1761 MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1763 /* Final output stage */
1765 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1766 wsptr[8*12] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1767 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1768 wsptr[8*11] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1769 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1770 wsptr[8*10] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1771 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1772 wsptr[8*9] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1773 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1774 wsptr[8*8] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1775 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1776 wsptr[8*7] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1777 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26, CONST_BITS-PASS1_BITS);
1780 /* Pass 2: process 13 rows from work array, store into output array. */
1783 for (ctr = 0; ctr < 13; ctr++) {
1784 outptr = output_buf[ctr] + output_col;
1788 /* Add fudge factor here for final descale. */
1789 z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1790 z1 = LEFT_SHIFT(z1, CONST_BITS);
1792 z2 = (INT32) wsptr[2];
1793 z3 = (INT32) wsptr[4];
1794 z4 = (INT32) wsptr[6];
1799 tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1800 tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1802 tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1803 tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1805 tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1806 tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1808 tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1809 tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1811 tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1812 tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1814 tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1815 tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1817 tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1821 z1 = (INT32) wsptr[1];
1822 z2 = (INT32) wsptr[3];
1823 z3 = (INT32) wsptr[5];
1824 z4 = (INT32) wsptr[7];
1826 tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1827 tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1829 tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1830 tmp10 = tmp11 + tmp12 + tmp13 -
1831 MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1832 tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
1833 tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1834 tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1835 tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
1837 tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1838 tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
1841 tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1842 tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1843 MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1844 z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1846 tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1847 MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1849 /* Final output stage */
1851 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1852 CONST_BITS+PASS1_BITS+3)
1854 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1855 CONST_BITS+PASS1_BITS+3)
1857 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1858 CONST_BITS+PASS1_BITS+3)
1860 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1861 CONST_BITS+PASS1_BITS+3)
1863 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1864 CONST_BITS+PASS1_BITS+3)
1866 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1867 CONST_BITS+PASS1_BITS+3)
1869 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1870 CONST_BITS+PASS1_BITS+3)
1872 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1873 CONST_BITS+PASS1_BITS+3)
1875 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1876 CONST_BITS+PASS1_BITS+3)
1878 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1879 CONST_BITS+PASS1_BITS+3)
1881 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1882 CONST_BITS+PASS1_BITS+3)
1884 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1885 CONST_BITS+PASS1_BITS+3)
1887 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26,
1888 CONST_BITS+PASS1_BITS+3)
1891 wsptr += 8; /* advance pointer to next row */
1897 * Perform dequantization and inverse DCT on one block of coefficients,
1898 * producing a 14x14 output block.
1900 * Optimized algorithm with 20 multiplications in the 1-D kernel.
1901 * cK represents sqrt(2) * cos(K*pi/28).
1905 jpeg_idct_14x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1906 JCOEFPTR coef_block,
1907 JSAMPARRAY output_buf, JDIMENSION output_col)
1909 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
1910 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1911 INT32 z1, z2, z3, z4;
1913 ISLOW_MULT_TYPE * quantptr;
1916 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1918 int workspace[8*14]; /* buffers data between passes */
1921 /* Pass 1: process columns from input, store into work array. */
1924 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1926 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1929 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1930 z1 = LEFT_SHIFT(z1, CONST_BITS);
1931 /* Add fudge factor here for final descale. */
1932 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
1933 z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1934 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
1935 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
1936 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
1942 tmp23 = RIGHT_SHIFT(z1 - LEFT_SHIFT(z2 + z3 - z4, 1),
1943 CONST_BITS-PASS1_BITS); /* c0 = (c4+c12-c8)*2 */
1945 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1946 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1948 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
1950 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
1951 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
1952 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
1953 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
1955 tmp20 = tmp10 + tmp13;
1956 tmp26 = tmp10 - tmp13;
1957 tmp21 = tmp11 + tmp14;
1958 tmp25 = tmp11 - tmp14;
1959 tmp22 = tmp12 + tmp15;
1960 tmp24 = tmp12 - tmp15;
1964 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1965 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1966 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1967 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1968 tmp13 = LEFT_SHIFT(z4, CONST_BITS);
1971 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
1972 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
1973 tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
1974 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
1975 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
1977 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */
1980 z4 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
1981 tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
1982 tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
1983 z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
1984 tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
1985 tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
1987 tmp13 = LEFT_SHIFT(z1 - z3, PASS1_BITS);
1989 /* Final output stage */
1991 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1992 wsptr[8*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1993 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1994 wsptr[8*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1995 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1996 wsptr[8*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1997 wsptr[8*3] = (int) (tmp23 + tmp13);
1998 wsptr[8*10] = (int) (tmp23 - tmp13);
1999 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
2000 wsptr[8*9] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
2001 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
2002 wsptr[8*8] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
2003 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
2004 wsptr[8*7] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
2007 /* Pass 2: process 14 rows from work array, store into output array. */
2010 for (ctr = 0; ctr < 14; ctr++) {
2011 outptr = output_buf[ctr] + output_col;
2015 /* Add fudge factor here for final descale. */
2016 z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2017 z1 = LEFT_SHIFT(z1, CONST_BITS);
2018 z4 = (INT32) wsptr[4];
2019 z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
2020 z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
2021 z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
2027 tmp23 = z1 - LEFT_SHIFT(z2 + z3 - z4, 1); /* c0 = (c4+c12-c8)*2 */
2029 z1 = (INT32) wsptr[2];
2030 z2 = (INT32) wsptr[6];
2032 z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
2034 tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
2035 tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
2036 tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
2037 MULTIPLY(z2, FIX(1.378756276)); /* c2 */
2039 tmp20 = tmp10 + tmp13;
2040 tmp26 = tmp10 - tmp13;
2041 tmp21 = tmp11 + tmp14;
2042 tmp25 = tmp11 - tmp14;
2043 tmp22 = tmp12 + tmp15;
2044 tmp24 = tmp12 - tmp15;
2048 z1 = (INT32) wsptr[1];
2049 z2 = (INT32) wsptr[3];
2050 z3 = (INT32) wsptr[5];
2051 z4 = (INT32) wsptr[7];
2052 z4 = LEFT_SHIFT(z4, CONST_BITS);
2055 tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
2056 tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
2057 tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
2058 tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
2059 tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
2061 tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */
2063 tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4; /* -c13 */
2064 tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
2065 tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
2066 tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
2067 tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
2068 tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
2070 tmp13 = LEFT_SHIFT(z1 - z3, CONST_BITS) + z4;
2072 /* Final output stage */
2074 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2075 CONST_BITS+PASS1_BITS+3)
2077 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2078 CONST_BITS+PASS1_BITS+3)
2080 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2081 CONST_BITS+PASS1_BITS+3)
2083 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2084 CONST_BITS+PASS1_BITS+3)
2086 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2087 CONST_BITS+PASS1_BITS+3)
2089 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2090 CONST_BITS+PASS1_BITS+3)
2092 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2093 CONST_BITS+PASS1_BITS+3)
2095 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2096 CONST_BITS+PASS1_BITS+3)
2098 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2099 CONST_BITS+PASS1_BITS+3)
2101 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2102 CONST_BITS+PASS1_BITS+3)
2104 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2105 CONST_BITS+PASS1_BITS+3)
2107 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2108 CONST_BITS+PASS1_BITS+3)
2110 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
2111 CONST_BITS+PASS1_BITS+3)
2113 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
2114 CONST_BITS+PASS1_BITS+3)
2117 wsptr += 8; /* advance pointer to next row */
2123 * Perform dequantization and inverse DCT on one block of coefficients,
2124 * producing a 15x15 output block.
2126 * Optimized algorithm with 22 multiplications in the 1-D kernel.
2127 * cK represents sqrt(2) * cos(K*pi/30).
2131 jpeg_idct_15x15 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2132 JCOEFPTR coef_block,
2133 JSAMPARRAY output_buf, JDIMENSION output_col)
2135 INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
2136 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2137 INT32 z1, z2, z3, z4;
2139 ISLOW_MULT_TYPE * quantptr;
2142 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2144 int workspace[8*15]; /* buffers data between passes */
2147 /* Pass 1: process columns from input, store into work array. */
2150 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2152 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2155 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
2156 z1 = LEFT_SHIFT(z1, CONST_BITS);
2157 /* Add fudge factor here for final descale. */
2158 z1 += ONE << (CONST_BITS-PASS1_BITS-1);
2160 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
2161 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
2162 z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
2164 tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2165 tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2169 z1 -= LEFT_SHIFT(tmp11 - tmp10, 1); /* c0 = (c6-c12)*2 */
2173 tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2174 tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2175 z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2177 tmp20 = tmp13 + tmp10 + tmp11;
2178 tmp23 = tmp12 - tmp10 + tmp11 + z2;
2180 tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2181 tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2183 tmp25 = tmp13 - tmp10 - tmp11;
2184 tmp26 = tmp12 + tmp10 - tmp11 - z2;
2186 tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2187 tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2189 tmp21 = tmp12 + tmp10 + tmp11;
2190 tmp24 = tmp13 - tmp10 + tmp11;
2192 tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2193 tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
2197 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
2198 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
2199 z4 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
2200 z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2201 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
2204 tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2205 tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2206 tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2208 tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
2209 tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
2211 tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2213 tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2214 tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2215 tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2216 z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2217 tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2218 tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2220 /* Final output stage */
2222 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
2223 wsptr[8*14] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
2224 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
2225 wsptr[8*13] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
2226 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
2227 wsptr[8*12] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
2228 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
2229 wsptr[8*11] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
2230 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
2231 wsptr[8*10] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
2232 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
2233 wsptr[8*9] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
2234 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
2235 wsptr[8*8] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
2236 wsptr[8*7] = (int) RIGHT_SHIFT(tmp27, CONST_BITS-PASS1_BITS);
2239 /* Pass 2: process 15 rows from work array, store into output array. */
2242 for (ctr = 0; ctr < 15; ctr++) {
2243 outptr = output_buf[ctr] + output_col;
2247 /* Add fudge factor here for final descale. */
2248 z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2249 z1 = LEFT_SHIFT(z1, CONST_BITS);
2251 z2 = (INT32) wsptr[2];
2252 z3 = (INT32) wsptr[4];
2253 z4 = (INT32) wsptr[6];
2255 tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2256 tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2260 z1 -= LEFT_SHIFT(tmp11 - tmp10, 1); /* c0 = (c6-c12)*2 */
2264 tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2265 tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2266 z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2268 tmp20 = tmp13 + tmp10 + tmp11;
2269 tmp23 = tmp12 - tmp10 + tmp11 + z2;
2271 tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2272 tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2274 tmp25 = tmp13 - tmp10 - tmp11;
2275 tmp26 = tmp12 + tmp10 - tmp11 - z2;
2277 tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2278 tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2280 tmp21 = tmp12 + tmp10 + tmp11;
2281 tmp24 = tmp13 - tmp10 + tmp11;
2283 tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2284 tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
2288 z1 = (INT32) wsptr[1];
2289 z2 = (INT32) wsptr[3];
2290 z4 = (INT32) wsptr[5];
2291 z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2292 z4 = (INT32) wsptr[7];
2295 tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2296 tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2297 tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2299 tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
2300 tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
2302 tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2304 tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2305 tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2306 tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2307 z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2308 tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2309 tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2311 /* Final output stage */
2313 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2314 CONST_BITS+PASS1_BITS+3)
2316 outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2317 CONST_BITS+PASS1_BITS+3)
2319 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2320 CONST_BITS+PASS1_BITS+3)
2322 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2323 CONST_BITS+PASS1_BITS+3)
2325 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2326 CONST_BITS+PASS1_BITS+3)
2328 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2329 CONST_BITS+PASS1_BITS+3)
2331 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2332 CONST_BITS+PASS1_BITS+3)
2334 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2335 CONST_BITS+PASS1_BITS+3)
2337 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2338 CONST_BITS+PASS1_BITS+3)
2340 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2341 CONST_BITS+PASS1_BITS+3)
2343 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2344 CONST_BITS+PASS1_BITS+3)
2346 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2347 CONST_BITS+PASS1_BITS+3)
2349 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
2350 CONST_BITS+PASS1_BITS+3)
2352 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
2353 CONST_BITS+PASS1_BITS+3)
2355 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27,
2356 CONST_BITS+PASS1_BITS+3)
2359 wsptr += 8; /* advance pointer to next row */
2365 * Perform dequantization and inverse DCT on one block of coefficients,
2366 * producing a 16x16 output block.
2368 * Optimized algorithm with 28 multiplications in the 1-D kernel.
2369 * cK represents sqrt(2) * cos(K*pi/32).
2373 jpeg_idct_16x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2374 JCOEFPTR coef_block,
2375 JSAMPARRAY output_buf, JDIMENSION output_col)
2377 INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
2378 INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2379 INT32 z1, z2, z3, z4;
2381 ISLOW_MULT_TYPE * quantptr;
2384 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2386 int workspace[8*16]; /* buffers data between passes */
2389 /* Pass 1: process columns from input, store into work array. */
2392 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2394 for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2397 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
2398 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
2399 /* Add fudge factor here for final descale. */
2400 tmp0 += 1 << (CONST_BITS-PASS1_BITS-1);
2402 z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
2403 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2404 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2406 tmp10 = tmp0 + tmp1;
2407 tmp11 = tmp0 - tmp1;
2408 tmp12 = tmp0 + tmp2;
2409 tmp13 = tmp0 - tmp2;
2411 z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
2412 z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
2414 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2415 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2417 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2418 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2419 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2420 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2422 tmp20 = tmp10 + tmp0;
2423 tmp27 = tmp10 - tmp0;
2424 tmp21 = tmp12 + tmp1;
2425 tmp26 = tmp12 - tmp1;
2426 tmp22 = tmp13 + tmp2;
2427 tmp25 = tmp13 - tmp2;
2428 tmp23 = tmp11 + tmp3;
2429 tmp24 = tmp11 - tmp3;
2433 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
2434 z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
2435 z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
2436 z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
2440 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2441 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2442 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2443 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2444 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2445 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2446 tmp0 = tmp1 + tmp2 + tmp3 -
2447 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2448 tmp13 = tmp10 + tmp11 + tmp12 -
2449 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2450 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2451 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2452 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2453 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2454 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2455 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2457 z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
2459 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2460 z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
2461 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2463 z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
2466 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2470 /* Final output stage */
2472 wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS-PASS1_BITS);
2473 wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS-PASS1_BITS);
2474 wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS-PASS1_BITS);
2475 wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS-PASS1_BITS);
2476 wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS-PASS1_BITS);
2477 wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS-PASS1_BITS);
2478 wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS-PASS1_BITS);
2479 wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS-PASS1_BITS);
2480 wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
2481 wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
2482 wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
2483 wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
2484 wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
2485 wsptr[8*9] = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
2486 wsptr[8*7] = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
2487 wsptr[8*8] = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
2490 /* Pass 2: process 16 rows from work array, store into output array. */
2493 for (ctr = 0; ctr < 16; ctr++) {
2494 outptr = output_buf[ctr] + output_col;
2498 /* Add fudge factor here for final descale. */
2499 tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2500 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS);
2502 z1 = (INT32) wsptr[4];
2503 tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2504 tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2506 tmp10 = tmp0 + tmp1;
2507 tmp11 = tmp0 - tmp1;
2508 tmp12 = tmp0 + tmp2;
2509 tmp13 = tmp0 - tmp2;
2511 z1 = (INT32) wsptr[2];
2512 z2 = (INT32) wsptr[6];
2514 z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2515 z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2517 tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2518 tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2519 tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2520 tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2522 tmp20 = tmp10 + tmp0;
2523 tmp27 = tmp10 - tmp0;
2524 tmp21 = tmp12 + tmp1;
2525 tmp26 = tmp12 - tmp1;
2526 tmp22 = tmp13 + tmp2;
2527 tmp25 = tmp13 - tmp2;
2528 tmp23 = tmp11 + tmp3;
2529 tmp24 = tmp11 - tmp3;
2533 z1 = (INT32) wsptr[1];
2534 z2 = (INT32) wsptr[3];
2535 z3 = (INT32) wsptr[5];
2536 z4 = (INT32) wsptr[7];
2540 tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2541 tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2542 tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2543 tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2544 tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2545 tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2546 tmp0 = tmp1 + tmp2 + tmp3 -
2547 MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2548 tmp13 = tmp10 + tmp11 + tmp12 -
2549 MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2550 z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2551 tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2552 tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2553 z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2554 tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2555 tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2557 z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
2559 tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2560 z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
2561 tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2563 z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
2566 z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2570 /* Final output stage */
2572 outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
2573 CONST_BITS+PASS1_BITS+3)
2575 outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
2576 CONST_BITS+PASS1_BITS+3)
2578 outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
2579 CONST_BITS+PASS1_BITS+3)
2581 outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
2582 CONST_BITS+PASS1_BITS+3)
2584 outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
2585 CONST_BITS+PASS1_BITS+3)
2587 outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
2588 CONST_BITS+PASS1_BITS+3)
2590 outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
2591 CONST_BITS+PASS1_BITS+3)
2593 outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
2594 CONST_BITS+PASS1_BITS+3)
2596 outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
2597 CONST_BITS+PASS1_BITS+3)
2599 outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
2600 CONST_BITS+PASS1_BITS+3)
2602 outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
2603 CONST_BITS+PASS1_BITS+3)
2605 outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
2606 CONST_BITS+PASS1_BITS+3)
2608 outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
2609 CONST_BITS+PASS1_BITS+3)
2611 outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
2612 CONST_BITS+PASS1_BITS+3)
2614 outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
2615 CONST_BITS+PASS1_BITS+3)
2617 outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
2618 CONST_BITS+PASS1_BITS+3)
2621 wsptr += 8; /* advance pointer to next row */
2625 #endif /* IDCT_SCALING_SUPPORTED */
2626 #endif /* DCT_ISLOW_SUPPORTED */