Context Navigation

source: rtems-graphics-toolkit/jpeg-7/jidctint.c @ 2abf143

base initial

Last change on this file since 2abf143 was 2abf143, checked in by Joel Sherrill <joel.sherrill@…>, on 12/20/09 at 00:59:55
Initial import of jpeg and tiff libraries.
Property mode set to `100644`
File size: 177.6 KB

Line
1	/*
2	* jidctint.c
3	*
4	* Copyright (C) 1991-1998, Thomas G. Lane.
5	* Modification developed 2002-2009 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.
8	*
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.
12	*
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.
17	*
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.
27	*
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.
32	*
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.
40	*
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.
46	*
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.
49	*/
50
51	#define JPEG_INTERNALS
52	#include "jinclude.h"
53	#include "jpeglib.h"
54	#include "jdct.h" /* Private declarations for DCT subsystem */
55
56	#ifdef DCT_ISLOW_SUPPORTED
57
58
59	/*
60	* This module is specialized to the case DCTSIZE = 8.
61	*/
62
63	#if DCTSIZE != 8
64	Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
65	#endif
66
67
68	/*
69	* The poop on this scaling stuff is as follows:
70	*
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).
77	*
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.
88	*
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.)
94	*
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.
98	*/
99
100	#if BITS_IN_JSAMPLE == 8
101	#define CONST_BITS 13
102	#define PASS1_BITS 2
103	#else
104	#define CONST_BITS 13
105	#define PASS1_BITS 1 /* lose a little precision to avoid overflow */
106	#endif
107
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...)
113	*/
114
115	#if CONST_BITS == 13
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) */
128	#else
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)
141	#endif
142
143
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.
149	*/
150
151	#if BITS_IN_JSAMPLE == 8
152	#define MULTIPLY(var,const) MULTIPLY16C16(var,const)
153	#else
154	#define MULTIPLY(var,const) ((var) * (const))
155	#endif
156
157
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.
161	*/
162
163	#define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))
164
165
166	/*
167	* Perform dequantization and inverse DCT on one block of coefficients.
168	*/
169
170	GLOBAL(void)
171	jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
172	JCOEFPTR coef_block,
173	JSAMPARRAY output_buf, JDIMENSION output_col)
174	{
175	INT32 tmp0, tmp1, tmp2, tmp3;
176	INT32 tmp10, tmp11, tmp12, tmp13;
177	INT32 z1, z2, z3;
178	JCOEFPTR inptr;
179	ISLOW_MULT_TYPE * quantptr;
180	int * wsptr;
181	JSAMPROW outptr;
182	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
183	int ctr;
184	int workspace[DCTSIZE2]; /* buffers data between passes */
185	SHIFT_TEMPS
186
187	/* Pass 1: process columns from input, store into work array. */
188	/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
189	/* furthermore, we scale the results by 2*PASS1_BITS. /
190
191	inptr = coef_block;
192	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
193	wsptr = workspace;
194	for (ctr = DCTSIZE; ctr > 0; ctr--) {
195	/* Due to quantization, we will usually find that many of the input
196	* coefficients are zero, especially the AC terms. We can exploit this
197	* by short-circuiting the IDCT calculation for any column in which all
198	* the AC terms are zero. In that case each output is equal to the
199	* DC coefficient (with scale factor as needed).
200	* With typical images and quantization tables, half or more of the
201	* column DCT calculations can be simplified this way.
202	*/
203
204	if (inptr[DCTSIZE1] == 0 && inptr[DCTSIZE2] == 0 &&
205	inptr[DCTSIZE3] == 0 && inptr[DCTSIZE4] == 0 &&
206	inptr[DCTSIZE5] == 0 && inptr[DCTSIZE6] == 0 &&
207	inptr[DCTSIZE*7] == 0) {
208	/* AC terms all zero */
209	int dcval = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]) << PASS1_BITS;
210
211	wsptr[DCTSIZE*0] = dcval;
212	wsptr[DCTSIZE*1] = dcval;
213	wsptr[DCTSIZE*2] = dcval;
214	wsptr[DCTSIZE*3] = dcval;
215	wsptr[DCTSIZE*4] = dcval;
216	wsptr[DCTSIZE*5] = dcval;
217	wsptr[DCTSIZE*6] = dcval;
218	wsptr[DCTSIZE*7] = dcval;
219
220	inptr++; /* advance pointers to next column */
221	quantptr++;
222	wsptr++;
223	continue;
224	}
225
226	/* Even part: reverse the even part of the forward DCT. */
227	/* The rotator is sqrt(2)c(-6). /
228
229	z2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
230	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
231
232	z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
233	tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
234	tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
235
236	z2 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
237	z3 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
238	z2 <<= CONST_BITS;
239	z3 <<= CONST_BITS;
240	/* Add fudge factor here for final descale. */
241	z2 += ONE << (CONST_BITS-PASS1_BITS-1);
242
243	tmp0 = z2 + z3;
244	tmp1 = z2 - z3;
245
246	tmp10 = tmp0 + tmp2;
247	tmp13 = tmp0 - tmp2;
248	tmp11 = tmp1 + tmp3;
249	tmp12 = tmp1 - tmp3;
250
251	/* Odd part per figure 8; the matrix is unitary and hence its
252	* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
253	*/
254
255	tmp0 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
256	tmp1 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
257	tmp2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
258	tmp3 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
259
260	z2 = tmp0 + tmp2;
261	z3 = tmp1 + tmp3;
262
263	z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
264	z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
265	z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
266	z2 += z1;
267	z3 += z1;
268
269	z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
270	tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
271	tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
272	tmp0 += z1 + z2;
273	tmp3 += z1 + z3;
274
275	z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
276	tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
277	tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
278	tmp1 += z1 + z3;
279	tmp2 += z1 + z2;
280
281	/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
282
283	wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
284	wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
285	wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
286	wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
287	wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
288	wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
289	wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
290	wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
291
292	inptr++; /* advance pointers to next column */
293	quantptr++;
294	wsptr++;
295	}
296
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. */
300
301	wsptr = workspace;
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.
310	*/
311
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)
317	& RANGE_MASK];
318
319	outptr[0] = dcval;
320	outptr[1] = dcval;
321	outptr[2] = dcval;
322	outptr[3] = dcval;
323	outptr[4] = dcval;
324	outptr[5] = dcval;
325	outptr[6] = dcval;
326	outptr[7] = dcval;
327
328	wsptr += DCTSIZE; /* advance pointer to next row */
329	continue;
330	}
331	#endif
332
333	/* Even part: reverse the even part of the forward DCT. */
334	/* The rotator is sqrt(2)c(-6). /
335
336	z2 = (INT32) wsptr[2];
337	z3 = (INT32) wsptr[6];
338
339	z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
340	tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
341	tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
342
343	/* Add fudge factor here for final descale. */
344	z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
345	z3 = (INT32) wsptr[4];
346
347	tmp0 = (z2 + z3) << CONST_BITS;
348	tmp1 = (z2 - z3) << CONST_BITS;
349
350	tmp10 = tmp0 + tmp2;
351	tmp13 = tmp0 - tmp2;
352	tmp11 = tmp1 + tmp3;
353	tmp12 = tmp1 - tmp3;
354
355	/* Odd part per figure 8; the matrix is unitary and hence its
356	* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
357	*/
358
359	tmp0 = (INT32) wsptr[7];
360	tmp1 = (INT32) wsptr[5];
361	tmp2 = (INT32) wsptr[3];
362	tmp3 = (INT32) wsptr[1];
363
364	z2 = tmp0 + tmp2;
365	z3 = tmp1 + tmp3;
366
367	z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
368	z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
369	z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
370	z2 += z1;
371	z3 += z1;
372
373	z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
374	tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
375	tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
376	tmp0 += z1 + z2;
377	tmp3 += z1 + z3;
378
379	z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
380	tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
381	tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
382	tmp1 += z1 + z3;
383	tmp2 += z1 + z2;
384
385	/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
386
387	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
388	CONST_BITS+PASS1_BITS+3)
389	& RANGE_MASK];
390	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
391	CONST_BITS+PASS1_BITS+3)
392	& RANGE_MASK];
393	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
394	CONST_BITS+PASS1_BITS+3)
395	& RANGE_MASK];
396	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
397	CONST_BITS+PASS1_BITS+3)
398	& RANGE_MASK];
399	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
400	CONST_BITS+PASS1_BITS+3)
401	& RANGE_MASK];
402	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
403	CONST_BITS+PASS1_BITS+3)
404	& RANGE_MASK];
405	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
406	CONST_BITS+PASS1_BITS+3)
407	& RANGE_MASK];
408	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
409	CONST_BITS+PASS1_BITS+3)
410	& RANGE_MASK];
411
412	wsptr += DCTSIZE; /* advance pointer to next row */
413	}
414	}
415
416	#ifdef IDCT_SCALING_SUPPORTED
417
418
419	/*
420	* Perform dequantization and inverse DCT on one block of coefficients,
421	* producing a 7x7 output block.
422	*
423	* Optimized algorithm with 12 multiplications in the 1-D kernel.
424	* cK represents sqrt(2) * cos(K*pi/14).
425	*/
426
427	GLOBAL(void)
428	jpeg_idct_7x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
429	JCOEFPTR coef_block,
430	JSAMPARRAY output_buf, JDIMENSION output_col)
431	{
432	INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
433	INT32 z1, z2, z3;
434	JCOEFPTR inptr;
435	ISLOW_MULT_TYPE * quantptr;
436	int * wsptr;
437	JSAMPROW outptr;
438	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
439	int ctr;
440	int workspace[77]; / buffers data between passes */
441	SHIFT_TEMPS
442
443	/* Pass 1: process columns from input, store into work array. */
444
445	inptr = coef_block;
446	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
447	wsptr = workspace;
448	for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
449	/* Even part */
450
451	tmp13 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
452	tmp13 <<= CONST_BITS;
453	/* Add fudge factor here for final descale. */
454	tmp13 += ONE << (CONST_BITS-PASS1_BITS-1);
455
456	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
457	z2 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
458	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
459
460	tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
461	tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
462	tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
463	tmp0 = z1 + z3;
464	z2 -= tmp0;
465	tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
466	tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
467	tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
468	tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
469
470	/* Odd part */
471
472	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
473	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
474	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
475
476	tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
477	tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
478	tmp0 = tmp1 - tmp2;
479	tmp1 += tmp2;
480	tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
481	tmp1 += tmp2;
482	z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
483	tmp0 += z2;
484	tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
485
486	/* Final output stage */
487
488	wsptr[7*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
489	wsptr[7*6] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
490	wsptr[7*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
491	wsptr[7*5] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
492	wsptr[7*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
493	wsptr[7*4] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
494	wsptr[7*3] = (int) RIGHT_SHIFT(tmp13, CONST_BITS-PASS1_BITS);
495	}
496
497	/* Pass 2: process 7 rows from work array, store into output array. */
498
499	wsptr = workspace;
500	for (ctr = 0; ctr < 7; ctr++) {
501	outptr = output_buf[ctr] + output_col;
502
503	/* Even part */
504
505	/* Add fudge factor here for final descale. */
506	tmp13 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
507	tmp13 <<= CONST_BITS;
508
509	z1 = (INT32) wsptr[2];
510	z2 = (INT32) wsptr[4];
511	z3 = (INT32) wsptr[6];
512
513	tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
514	tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
515	tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
516	tmp0 = z1 + z3;
517	z2 -= tmp0;
518	tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
519	tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
520	tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
521	tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
522
523	/* Odd part */
524
525	z1 = (INT32) wsptr[1];
526	z2 = (INT32) wsptr[3];
527	z3 = (INT32) wsptr[5];
528
529	tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
530	tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
531	tmp0 = tmp1 - tmp2;
532	tmp1 += tmp2;
533	tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
534	tmp1 += tmp2;
535	z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
536	tmp0 += z2;
537	tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
538
539	/* Final output stage */
540
541	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
542	CONST_BITS+PASS1_BITS+3)
543	& RANGE_MASK];
544	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
545	CONST_BITS+PASS1_BITS+3)
546	& RANGE_MASK];
547	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
548	CONST_BITS+PASS1_BITS+3)
549	& RANGE_MASK];
550	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
551	CONST_BITS+PASS1_BITS+3)
552	& RANGE_MASK];
553	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
554	CONST_BITS+PASS1_BITS+3)
555	& RANGE_MASK];
556	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
557	CONST_BITS+PASS1_BITS+3)
558	& RANGE_MASK];
559	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13,
560	CONST_BITS+PASS1_BITS+3)
561	& RANGE_MASK];
562
563	wsptr += 7; /* advance pointer to next row */
564	}
565	}
566
567
568	/*
569	* Perform dequantization and inverse DCT on one block of coefficients,
570	* producing a reduced-size 6x6 output block.
571	*
572	* Optimized algorithm with 3 multiplications in the 1-D kernel.
573	* cK represents sqrt(2) * cos(K*pi/12).
574	*/
575
576	GLOBAL(void)
577	jpeg_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
578	JCOEFPTR coef_block,
579	JSAMPARRAY output_buf, JDIMENSION output_col)
580	{
581	INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
582	INT32 z1, z2, z3;
583	JCOEFPTR inptr;
584	ISLOW_MULT_TYPE * quantptr;
585	int * wsptr;
586	JSAMPROW outptr;
587	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
588	int ctr;
589	int workspace[66]; / buffers data between passes */
590	SHIFT_TEMPS
591
592	/* Pass 1: process columns from input, store into work array. */
593
594	inptr = coef_block;
595	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
596	wsptr = workspace;
597	for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
598	/* Even part */
599
600	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
601	tmp0 <<= CONST_BITS;
602	/* Add fudge factor here for final descale. */
603	tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
604	tmp2 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
605	tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
606	tmp1 = tmp0 + tmp10;
607	tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
608	tmp10 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
609	tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
610	tmp10 = tmp1 + tmp0;
611	tmp12 = tmp1 - tmp0;
612
613	/* Odd part */
614
615	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
616	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
617	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
618	tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
619	tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
620	tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
621	tmp1 = (z1 - z2 - z3) << PASS1_BITS;
622
623	/* Final output stage */
624
625	wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
626	wsptr[6*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
627	wsptr[6*1] = (int) (tmp11 + tmp1);
628	wsptr[6*4] = (int) (tmp11 - tmp1);
629	wsptr[6*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
630	wsptr[6*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
631	}
632
633	/* Pass 2: process 6 rows from work array, store into output array. */
634
635	wsptr = workspace;
636	for (ctr = 0; ctr < 6; ctr++) {
637	outptr = output_buf[ctr] + output_col;
638
639	/* Even part */
640
641	/* Add fudge factor here for final descale. */
642	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
643	tmp0 <<= CONST_BITS;
644	tmp2 = (INT32) wsptr[4];
645	tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
646	tmp1 = tmp0 + tmp10;
647	tmp11 = tmp0 - tmp10 - tmp10;
648	tmp10 = (INT32) wsptr[2];
649	tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
650	tmp10 = tmp1 + tmp0;
651	tmp12 = tmp1 - tmp0;
652
653	/* Odd part */
654
655	z1 = (INT32) wsptr[1];
656	z2 = (INT32) wsptr[3];
657	z3 = (INT32) wsptr[5];
658	tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
659	tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
660	tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
661	tmp1 = (z1 - z2 - z3) << CONST_BITS;
662
663	/* Final output stage */
664
665	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
666	CONST_BITS+PASS1_BITS+3)
667	& RANGE_MASK];
668	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
669	CONST_BITS+PASS1_BITS+3)
670	& RANGE_MASK];
671	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
672	CONST_BITS+PASS1_BITS+3)
673	& RANGE_MASK];
674	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
675	CONST_BITS+PASS1_BITS+3)
676	& RANGE_MASK];
677	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
678	CONST_BITS+PASS1_BITS+3)
679	& RANGE_MASK];
680	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
681	CONST_BITS+PASS1_BITS+3)
682	& RANGE_MASK];
683
684	wsptr += 6; /* advance pointer to next row */
685	}
686	}
687
688
689	/*
690	* Perform dequantization and inverse DCT on one block of coefficients,
691	* producing a reduced-size 5x5 output block.
692	*
693	* Optimized algorithm with 5 multiplications in the 1-D kernel.
694	* cK represents sqrt(2) * cos(K*pi/10).
695	*/
696
697	GLOBAL(void)
698	jpeg_idct_5x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
699	JCOEFPTR coef_block,
700	JSAMPARRAY output_buf, JDIMENSION output_col)
701	{
702	INT32 tmp0, tmp1, tmp10, tmp11, tmp12;
703	INT32 z1, z2, z3;
704	JCOEFPTR inptr;
705	ISLOW_MULT_TYPE * quantptr;
706	int * wsptr;
707	JSAMPROW outptr;
708	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
709	int ctr;
710	int workspace[55]; / buffers data between passes */
711	SHIFT_TEMPS
712
713	/* Pass 1: process columns from input, store into work array. */
714
715	inptr = coef_block;
716	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
717	wsptr = workspace;
718	for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
719	/* Even part */
720
721	tmp12 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
722	tmp12 <<= CONST_BITS;
723	/* Add fudge factor here for final descale. */
724	tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
725	tmp0 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
726	tmp1 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
727	z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
728	z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
729	z3 = tmp12 + z2;
730	tmp10 = z3 + z1;
731	tmp11 = z3 - z1;
732	tmp12 -= z2 << 2;
733
734	/* Odd part */
735
736	z2 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
737	z3 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
738
739	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
740	tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
741	tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
742
743	/* Final output stage */
744
745	wsptr[5*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
746	wsptr[5*4] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
747	wsptr[5*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
748	wsptr[5*3] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
749	wsptr[5*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
750	}
751
752	/* Pass 2: process 5 rows from work array, store into output array. */
753
754	wsptr = workspace;
755	for (ctr = 0; ctr < 5; ctr++) {
756	outptr = output_buf[ctr] + output_col;
757
758	/* Even part */
759
760	/* Add fudge factor here for final descale. */
761	tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
762	tmp12 <<= CONST_BITS;
763	tmp0 = (INT32) wsptr[2];
764	tmp1 = (INT32) wsptr[4];
765	z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
766	z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
767	z3 = tmp12 + z2;
768	tmp10 = z3 + z1;
769	tmp11 = z3 - z1;
770	tmp12 -= z2 << 2;
771
772	/* Odd part */
773
774	z2 = (INT32) wsptr[1];
775	z3 = (INT32) wsptr[3];
776
777	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
778	tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
779	tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
780
781	/* Final output stage */
782
783	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
784	CONST_BITS+PASS1_BITS+3)
785	& RANGE_MASK];
786	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
787	CONST_BITS+PASS1_BITS+3)
788	& RANGE_MASK];
789	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
790	CONST_BITS+PASS1_BITS+3)
791	& RANGE_MASK];
792	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
793	CONST_BITS+PASS1_BITS+3)
794	& RANGE_MASK];
795	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
796	CONST_BITS+PASS1_BITS+3)
797	& RANGE_MASK];
798
799	wsptr += 5; /* advance pointer to next row */
800	}
801	}
802
803
804	/*
805	* Perform dequantization and inverse DCT on one block of coefficients,
806	* producing a reduced-size 4x4 output block.
807	*
808	* Optimized algorithm with 3 multiplications in the 1-D kernel.
809	* cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
810	*/
811
812	GLOBAL(void)
813	jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
814	JCOEFPTR coef_block,
815	JSAMPARRAY output_buf, JDIMENSION output_col)
816	{
817	INT32 tmp0, tmp2, tmp10, tmp12;
818	INT32 z1, z2, z3;
819	JCOEFPTR inptr;
820	ISLOW_MULT_TYPE * quantptr;
821	int * wsptr;
822	JSAMPROW outptr;
823	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
824	int ctr;
825	int workspace[44]; / buffers data between passes */
826	SHIFT_TEMPS
827
828	/* Pass 1: process columns from input, store into work array. */
829
830	inptr = coef_block;
831	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
832	wsptr = workspace;
833	for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
834	/* Even part */
835
836	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
837	tmp2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
838
839	tmp10 = (tmp0 + tmp2) << PASS1_BITS;
840	tmp12 = (tmp0 - tmp2) << PASS1_BITS;
841
842	/* Odd part */
843	/* Same rotation as in the even part of the 8x8 LL&M IDCT */
844
845	z2 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
846	z3 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
847
848	z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
849	/* Add fudge factor here for final descale. */
850	z1 += ONE << (CONST_BITS-PASS1_BITS-1);
851	tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
852	CONST_BITS-PASS1_BITS);
853	tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
854	CONST_BITS-PASS1_BITS);
855
856	/* Final output stage */
857
858	wsptr[4*0] = (int) (tmp10 + tmp0);
859	wsptr[4*3] = (int) (tmp10 - tmp0);
860	wsptr[4*1] = (int) (tmp12 + tmp2);
861	wsptr[4*2] = (int) (tmp12 - tmp2);
862	}
863
864	/* Pass 2: process 4 rows from work array, store into output array. */
865
866	wsptr = workspace;
867	for (ctr = 0; ctr < 4; ctr++) {
868	outptr = output_buf[ctr] + output_col;
869
870	/* Even part */
871
872	/* Add fudge factor here for final descale. */
873	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
874	tmp2 = (INT32) wsptr[2];
875
876	tmp10 = (tmp0 + tmp2) << CONST_BITS;
877	tmp12 = (tmp0 - tmp2) << CONST_BITS;
878
879	/* Odd part */
880	/* Same rotation as in the even part of the 8x8 LL&M IDCT */
881
882	z2 = (INT32) wsptr[1];
883	z3 = (INT32) wsptr[3];
884
885	z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
886	tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
887	tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
888
889	/* Final output stage */
890
891	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
892	CONST_BITS+PASS1_BITS+3)
893	& RANGE_MASK];
894	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
895	CONST_BITS+PASS1_BITS+3)
896	& RANGE_MASK];
897	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
898	CONST_BITS+PASS1_BITS+3)
899	& RANGE_MASK];
900	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
901	CONST_BITS+PASS1_BITS+3)
902	& RANGE_MASK];
903
904	wsptr += 4; /* advance pointer to next row */
905	}
906	}
907
908
909	/*
910	* Perform dequantization and inverse DCT on one block of coefficients,
911	* producing a reduced-size 3x3 output block.
912	*
913	* Optimized algorithm with 2 multiplications in the 1-D kernel.
914	* cK represents sqrt(2) * cos(K*pi/6).
915	*/
916
917	GLOBAL(void)
918	jpeg_idct_3x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
919	JCOEFPTR coef_block,
920	JSAMPARRAY output_buf, JDIMENSION output_col)
921	{
922	INT32 tmp0, tmp2, tmp10, tmp12;
923	JCOEFPTR inptr;
924	ISLOW_MULT_TYPE * quantptr;
925	int * wsptr;
926	JSAMPROW outptr;
927	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
928	int ctr;
929	int workspace[33]; / buffers data between passes */
930	SHIFT_TEMPS
931
932	/* Pass 1: process columns from input, store into work array. */
933
934	inptr = coef_block;
935	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
936	wsptr = workspace;
937	for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
938	/* Even part */
939
940	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
941	tmp0 <<= CONST_BITS;
942	/* Add fudge factor here for final descale. */
943	tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
944	tmp2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
945	tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
946	tmp10 = tmp0 + tmp12;
947	tmp2 = tmp0 - tmp12 - tmp12;
948
949	/* Odd part */
950
951	tmp12 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
952	tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
953
954	/* Final output stage */
955
956	wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
957	wsptr[3*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
958	wsptr[3*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
959	}
960
961	/* Pass 2: process 3 rows from work array, store into output array. */
962
963	wsptr = workspace;
964	for (ctr = 0; ctr < 3; ctr++) {
965	outptr = output_buf[ctr] + output_col;
966
967	/* Even part */
968
969	/* Add fudge factor here for final descale. */
970	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
971	tmp0 <<= CONST_BITS;
972	tmp2 = (INT32) wsptr[2];
973	tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
974	tmp10 = tmp0 + tmp12;
975	tmp2 = tmp0 - tmp12 - tmp12;
976
977	/* Odd part */
978
979	tmp12 = (INT32) wsptr[1];
980	tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
981
982	/* Final output stage */
983
984	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
985	CONST_BITS+PASS1_BITS+3)
986	& RANGE_MASK];
987	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
988	CONST_BITS+PASS1_BITS+3)
989	& RANGE_MASK];
990	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
991	CONST_BITS+PASS1_BITS+3)
992	& RANGE_MASK];
993
994	wsptr += 3; /* advance pointer to next row */
995	}
996	}
997
998
999	/*
1000	* Perform dequantization and inverse DCT on one block of coefficients,
1001	* producing a reduced-size 2x2 output block.
1002	*
1003	* Multiplication-less algorithm.
1004	*/
1005
1006	GLOBAL(void)
1007	jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1008	JCOEFPTR coef_block,
1009	JSAMPARRAY output_buf, JDIMENSION output_col)
1010	{
1011	INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
1012	ISLOW_MULT_TYPE * quantptr;
1013	JSAMPROW outptr;
1014	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1015	SHIFT_TEMPS
1016
1017	/* Pass 1: process columns from input. */
1018
1019	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1020
1021	/* Column 0 */
1022	tmp4 = DEQUANTIZE(coef_block[DCTSIZE0], quantptr[DCTSIZE0]);
1023	tmp5 = DEQUANTIZE(coef_block[DCTSIZE1], quantptr[DCTSIZE1]);
1024	/* Add fudge factor here for final descale. */
1025	tmp4 += ONE << 2;
1026
1027	tmp0 = tmp4 + tmp5;
1028	tmp2 = tmp4 - tmp5;
1029
1030	/* Column 1 */
1031	tmp4 = DEQUANTIZE(coef_block[DCTSIZE0+1], quantptr[DCTSIZE0+1]);
1032	tmp5 = DEQUANTIZE(coef_block[DCTSIZE1+1], quantptr[DCTSIZE1+1]);
1033
1034	tmp1 = tmp4 + tmp5;
1035	tmp3 = tmp4 - tmp5;
1036
1037	/* Pass 2: process 2 rows, store into output array. */
1038
1039	/* Row 0 */
1040	outptr = output_buf[0] + output_col;
1041
1042	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp0 + tmp1, 3) & RANGE_MASK];
1043	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp0 - tmp1, 3) & RANGE_MASK];
1044
1045	/* Row 1 */
1046	outptr = output_buf[1] + output_col;
1047
1048	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp2 + tmp3, 3) & RANGE_MASK];
1049	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2 - tmp3, 3) & RANGE_MASK];
1050	}
1051
1052
1053	/*
1054	* Perform dequantization and inverse DCT on one block of coefficients,
1055	* producing a reduced-size 1x1 output block.
1056	*
1057	* We hardly need an inverse DCT routine for this: just take the
1058	* average pixel value, which is one-eighth of the DC coefficient.
1059	*/
1060
1061	GLOBAL(void)
1062	jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1063	JCOEFPTR coef_block,
1064	JSAMPARRAY output_buf, JDIMENSION output_col)
1065	{
1066	int dcval;
1067	ISLOW_MULT_TYPE * quantptr;
1068	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1069	SHIFT_TEMPS
1070
1071	/* 1x1 is trivial: just take the DC coefficient divided by 8. */
1072	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1073	dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
1074	dcval = (int) DESCALE((INT32) dcval, 3);
1075
1076	output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
1077	}
1078
1079
1080	/*
1081	* Perform dequantization and inverse DCT on one block of coefficients,
1082	* producing a 9x9 output block.
1083	*
1084	* Optimized algorithm with 10 multiplications in the 1-D kernel.
1085	* cK represents sqrt(2) * cos(K*pi/18).
1086	*/
1087
1088	GLOBAL(void)
1089	jpeg_idct_9x9 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1090	JCOEFPTR coef_block,
1091	JSAMPARRAY output_buf, JDIMENSION output_col)
1092	{
1093	INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14;
1094	INT32 z1, z2, z3, z4;
1095	JCOEFPTR inptr;
1096	ISLOW_MULT_TYPE * quantptr;
1097	int * wsptr;
1098	JSAMPROW outptr;
1099	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1100	int ctr;
1101	int workspace[89]; / buffers data between passes */
1102	SHIFT_TEMPS
1103
1104	/* Pass 1: process columns from input, store into work array. */
1105
1106	inptr = coef_block;
1107	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1108	wsptr = workspace;
1109	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1110	/* Even part */
1111
1112	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
1113	tmp0 <<= CONST_BITS;
1114	/* Add fudge factor here for final descale. */
1115	tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
1116
1117	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
1118	z2 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
1119	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
1120
1121	tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
1122	tmp1 = tmp0 + tmp3;
1123	tmp2 = tmp0 - tmp3 - tmp3;
1124
1125	tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1126	tmp11 = tmp2 + tmp0;
1127	tmp14 = tmp2 - tmp0 - tmp0;
1128
1129	tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1130	tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1131	tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1132
1133	tmp10 = tmp1 + tmp0 - tmp3;
1134	tmp12 = tmp1 - tmp0 + tmp2;
1135	tmp13 = tmp1 - tmp2 + tmp3;
1136
1137	/* Odd part */
1138
1139	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
1140	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
1141	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
1142	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
1143
1144	z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
1145
1146	tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1147	tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1148	tmp0 = tmp2 + tmp3 - z2;
1149	tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1150	tmp2 += z2 - tmp1;
1151	tmp3 += z2 + tmp1;
1152	tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1153
1154	/* Final output stage */
1155
1156	wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
1157	wsptr[8*8] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
1158	wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
1159	wsptr[8*7] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
1160	wsptr[8*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
1161	wsptr[8*6] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
1162	wsptr[8*3] = (int) RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS-PASS1_BITS);
1163	wsptr[8*5] = (int) RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS-PASS1_BITS);
1164	wsptr[8*4] = (int) RIGHT_SHIFT(tmp14, CONST_BITS-PASS1_BITS);
1165	}
1166
1167	/* Pass 2: process 9 rows from work array, store into output array. */
1168
1169	wsptr = workspace;
1170	for (ctr = 0; ctr < 9; ctr++) {
1171	outptr = output_buf[ctr] + output_col;
1172
1173	/* Even part */
1174
1175	/* Add fudge factor here for final descale. */
1176	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1177	tmp0 <<= CONST_BITS;
1178
1179	z1 = (INT32) wsptr[2];
1180	z2 = (INT32) wsptr[4];
1181	z3 = (INT32) wsptr[6];
1182
1183	tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
1184	tmp1 = tmp0 + tmp3;
1185	tmp2 = tmp0 - tmp3 - tmp3;
1186
1187	tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1188	tmp11 = tmp2 + tmp0;
1189	tmp14 = tmp2 - tmp0 - tmp0;
1190
1191	tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1192	tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1193	tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1194
1195	tmp10 = tmp1 + tmp0 - tmp3;
1196	tmp12 = tmp1 - tmp0 + tmp2;
1197	tmp13 = tmp1 - tmp2 + tmp3;
1198
1199	/* Odd part */
1200
1201	z1 = (INT32) wsptr[1];
1202	z2 = (INT32) wsptr[3];
1203	z3 = (INT32) wsptr[5];
1204	z4 = (INT32) wsptr[7];
1205
1206	z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
1207
1208	tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1209	tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1210	tmp0 = tmp2 + tmp3 - z2;
1211	tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1212	tmp2 += z2 - tmp1;
1213	tmp3 += z2 + tmp1;
1214	tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1215
1216	/* Final output stage */
1217
1218	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1219	CONST_BITS+PASS1_BITS+3)
1220	& RANGE_MASK];
1221	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1222	CONST_BITS+PASS1_BITS+3)
1223	& RANGE_MASK];
1224	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
1225	CONST_BITS+PASS1_BITS+3)
1226	& RANGE_MASK];
1227	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
1228	CONST_BITS+PASS1_BITS+3)
1229	& RANGE_MASK];
1230	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
1231	CONST_BITS+PASS1_BITS+3)
1232	& RANGE_MASK];
1233	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
1234	CONST_BITS+PASS1_BITS+3)
1235	& RANGE_MASK];
1236	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp3,
1237	CONST_BITS+PASS1_BITS+3)
1238	& RANGE_MASK];
1239	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp3,
1240	CONST_BITS+PASS1_BITS+3)
1241	& RANGE_MASK];
1242	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp14,
1243	CONST_BITS+PASS1_BITS+3)
1244	& RANGE_MASK];
1245
1246	wsptr += 8; /* advance pointer to next row */
1247	}
1248	}
1249
1250
1251	/*
1252	* Perform dequantization and inverse DCT on one block of coefficients,
1253	* producing a 10x10 output block.
1254	*
1255	* Optimized algorithm with 12 multiplications in the 1-D kernel.
1256	* cK represents sqrt(2) * cos(K*pi/20).
1257	*/
1258
1259	GLOBAL(void)
1260	jpeg_idct_10x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1261	JCOEFPTR coef_block,
1262	JSAMPARRAY output_buf, JDIMENSION output_col)
1263	{
1264	INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1265	INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
1266	INT32 z1, z2, z3, z4, z5;
1267	JCOEFPTR inptr;
1268	ISLOW_MULT_TYPE * quantptr;
1269	int * wsptr;
1270	JSAMPROW outptr;
1271	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1272	int ctr;
1273	int workspace[810]; / buffers data between passes */
1274	SHIFT_TEMPS
1275
1276	/* Pass 1: process columns from input, store into work array. */
1277
1278	inptr = coef_block;
1279	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1280	wsptr = workspace;
1281	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1282	/* Even part */
1283
1284	z3 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
1285	z3 <<= CONST_BITS;
1286	/* Add fudge factor here for final descale. */
1287	z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1288	z4 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
1289	z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1290	z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1291	tmp10 = z3 + z1;
1292	tmp11 = z3 - z2;
1293
1294	tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1), /* c0 = (c4-c8)2 /
1295	CONST_BITS-PASS1_BITS);
1296
1297	z2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
1298	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
1299
1300	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1301	tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1302	tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1303
1304	tmp20 = tmp10 + tmp12;
1305	tmp24 = tmp10 - tmp12;
1306	tmp21 = tmp11 + tmp13;
1307	tmp23 = tmp11 - tmp13;
1308
1309	/* Odd part */
1310
1311	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
1312	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
1313	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
1314	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
1315
1316	tmp11 = z2 + z4;
1317	tmp13 = z2 - z4;
1318
1319	tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1320	z5 = z3 << CONST_BITS;
1321
1322	z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1323	z4 = z5 + tmp12;
1324
1325	tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1326	tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1327
1328	z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1329	z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));
1330
1331	tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;
1332
1333	tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1334	tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1335
1336	/* Final output stage */
1337
1338	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1339	wsptr[8*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1340	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1341	wsptr[8*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1342	wsptr[8*2] = (int) (tmp22 + tmp12);
1343	wsptr[8*7] = (int) (tmp22 - tmp12);
1344	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1345	wsptr[8*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1346	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1347	wsptr[8*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1348	}
1349
1350	/* Pass 2: process 10 rows from work array, store into output array. */
1351
1352	wsptr = workspace;
1353	for (ctr = 0; ctr < 10; ctr++) {
1354	outptr = output_buf[ctr] + output_col;
1355
1356	/* Even part */
1357
1358	/* Add fudge factor here for final descale. */
1359	z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1360	z3 <<= CONST_BITS;
1361	z4 = (INT32) wsptr[4];
1362	z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1363	z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1364	tmp10 = z3 + z1;
1365	tmp11 = z3 - z2;
1366
1367	tmp22 = z3 - ((z1 - z2) << 1); /* c0 = (c4-c8)2 /
1368
1369	z2 = (INT32) wsptr[2];
1370	z3 = (INT32) wsptr[6];
1371
1372	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1373	tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1374	tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1375
1376	tmp20 = tmp10 + tmp12;
1377	tmp24 = tmp10 - tmp12;
1378	tmp21 = tmp11 + tmp13;
1379	tmp23 = tmp11 - tmp13;
1380
1381	/* Odd part */
1382
1383	z1 = (INT32) wsptr[1];
1384	z2 = (INT32) wsptr[3];
1385	z3 = (INT32) wsptr[5];
1386	z3 <<= CONST_BITS;
1387	z4 = (INT32) wsptr[7];
1388
1389	tmp11 = z2 + z4;
1390	tmp13 = z2 - z4;
1391
1392	tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1393
1394	z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1395	z4 = z3 + tmp12;
1396
1397	tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1398	tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1399
1400	z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1401	z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));
1402
1403	tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;
1404
1405	tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1406	tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1407
1408	/* Final output stage */
1409
1410	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1411	CONST_BITS+PASS1_BITS+3)
1412	& RANGE_MASK];
1413	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1414	CONST_BITS+PASS1_BITS+3)
1415	& RANGE_MASK];
1416	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1417	CONST_BITS+PASS1_BITS+3)
1418	& RANGE_MASK];
1419	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1420	CONST_BITS+PASS1_BITS+3)
1421	& RANGE_MASK];
1422	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1423	CONST_BITS+PASS1_BITS+3)
1424	& RANGE_MASK];
1425	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1426	CONST_BITS+PASS1_BITS+3)
1427	& RANGE_MASK];
1428	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1429	CONST_BITS+PASS1_BITS+3)
1430	& RANGE_MASK];
1431	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1432	CONST_BITS+PASS1_BITS+3)
1433	& RANGE_MASK];
1434	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1435	CONST_BITS+PASS1_BITS+3)
1436	& RANGE_MASK];
1437	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1438	CONST_BITS+PASS1_BITS+3)
1439	& RANGE_MASK];
1440
1441	wsptr += 8; /* advance pointer to next row */
1442	}
1443	}
1444
1445
1446	/*
1447	* Perform dequantization and inverse DCT on one block of coefficients,
1448	* producing a 11x11 output block.
1449	*
1450	* Optimized algorithm with 24 multiplications in the 1-D kernel.
1451	* cK represents sqrt(2) * cos(K*pi/22).
1452	*/
1453
1454	GLOBAL(void)
1455	jpeg_idct_11x11 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1456	JCOEFPTR coef_block,
1457	JSAMPARRAY output_buf, JDIMENSION output_col)
1458	{
1459	INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1460	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1461	INT32 z1, z2, z3, z4;
1462	JCOEFPTR inptr;
1463	ISLOW_MULT_TYPE * quantptr;
1464	int * wsptr;
1465	JSAMPROW outptr;
1466	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1467	int ctr;
1468	int workspace[811]; / buffers data between passes */
1469	SHIFT_TEMPS
1470
1471	/* Pass 1: process columns from input, store into work array. */
1472
1473	inptr = coef_block;
1474	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1475	wsptr = workspace;
1476	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1477	/* Even part */
1478
1479	tmp10 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
1480	tmp10 <<= CONST_BITS;
1481	/* Add fudge factor here for final descale. */
1482	tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
1483
1484	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
1485	z2 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
1486	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
1487
1488	tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1489	tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1490	z4 = z1 + z3;
1491	tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1492	z4 -= z2;
1493	tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1494	tmp21 = tmp20 + tmp23 + tmp25 -
1495	MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1496	tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1497	tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1498	tmp24 += tmp25;
1499	tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1500	tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1501	MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1502	tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1503
1504	/* Odd part */
1505
1506	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
1507	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
1508	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
1509	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
1510
1511	tmp11 = z1 + z2;
1512	tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1513	tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1514	tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1515	tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1516	tmp10 = tmp11 + tmp12 + tmp13 -
1517	MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2c9 /
1518	z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1519	tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3c9-c3 /
1520	tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1521	z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1522	tmp11 += z1;
1523	tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1524	tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1525	MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1526	MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1527
1528	/* Final output stage */
1529
1530	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1531	wsptr[8*10] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1532	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1533	wsptr[8*9] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1534	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1535	wsptr[8*8] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1536	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1537	wsptr[8*7] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1538	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1539	wsptr[8*6] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1540	wsptr[8*5] = (int) RIGHT_SHIFT(tmp25, CONST_BITS-PASS1_BITS);
1541	}
1542
1543	/* Pass 2: process 11 rows from work array, store into output array. */
1544
1545	wsptr = workspace;
1546	for (ctr = 0; ctr < 11; ctr++) {
1547	outptr = output_buf[ctr] + output_col;
1548
1549	/* Even part */
1550
1551	/* Add fudge factor here for final descale. */
1552	tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1553	tmp10 <<= CONST_BITS;
1554
1555	z1 = (INT32) wsptr[2];
1556	z2 = (INT32) wsptr[4];
1557	z3 = (INT32) wsptr[6];
1558
1559	tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1560	tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1561	z4 = z1 + z3;
1562	tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1563	z4 -= z2;
1564	tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1565	tmp21 = tmp20 + tmp23 + tmp25 -
1566	MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1567	tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1568	tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1569	tmp24 += tmp25;
1570	tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1571	tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1572	MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1573	tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1574
1575	/* Odd part */
1576
1577	z1 = (INT32) wsptr[1];
1578	z2 = (INT32) wsptr[3];
1579	z3 = (INT32) wsptr[5];
1580	z4 = (INT32) wsptr[7];
1581
1582	tmp11 = z1 + z2;
1583	tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1584	tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1585	tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1586	tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1587	tmp10 = tmp11 + tmp12 + tmp13 -
1588	MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2c9 /
1589	z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1590	tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3c9-c3 /
1591	tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1592	z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1593	tmp11 += z1;
1594	tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1595	tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1596	MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1597	MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1598
1599	/* Final output stage */
1600
1601	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1602	CONST_BITS+PASS1_BITS+3)
1603	& RANGE_MASK];
1604	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1605	CONST_BITS+PASS1_BITS+3)
1606	& RANGE_MASK];
1607	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1608	CONST_BITS+PASS1_BITS+3)
1609	& RANGE_MASK];
1610	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1611	CONST_BITS+PASS1_BITS+3)
1612	& RANGE_MASK];
1613	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1614	CONST_BITS+PASS1_BITS+3)
1615	& RANGE_MASK];
1616	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1617	CONST_BITS+PASS1_BITS+3)
1618	& RANGE_MASK];
1619	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1620	CONST_BITS+PASS1_BITS+3)
1621	& RANGE_MASK];
1622	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1623	CONST_BITS+PASS1_BITS+3)
1624	& RANGE_MASK];
1625	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1626	CONST_BITS+PASS1_BITS+3)
1627	& RANGE_MASK];
1628	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1629	CONST_BITS+PASS1_BITS+3)
1630	& RANGE_MASK];
1631	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25,
1632	CONST_BITS+PASS1_BITS+3)
1633	& RANGE_MASK];
1634
1635	wsptr += 8; /* advance pointer to next row */
1636	}
1637	}
1638
1639
1640	/*
1641	* Perform dequantization and inverse DCT on one block of coefficients,
1642	* producing a 12x12 output block.
1643	*
1644	* Optimized algorithm with 15 multiplications in the 1-D kernel.
1645	* cK represents sqrt(2) * cos(K*pi/24).
1646	*/
1647
1648	GLOBAL(void)
1649	jpeg_idct_12x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1650	JCOEFPTR coef_block,
1651	JSAMPARRAY output_buf, JDIMENSION output_col)
1652	{
1653	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1654	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1655	INT32 z1, z2, z3, z4;
1656	JCOEFPTR inptr;
1657	ISLOW_MULT_TYPE * quantptr;
1658	int * wsptr;
1659	JSAMPROW outptr;
1660	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1661	int ctr;
1662	int workspace[812]; / buffers data between passes */
1663	SHIFT_TEMPS
1664
1665	/* Pass 1: process columns from input, store into work array. */
1666
1667	inptr = coef_block;
1668	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1669	wsptr = workspace;
1670	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1671	/* Even part */
1672
1673	z3 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
1674	z3 <<= CONST_BITS;
1675	/* Add fudge factor here for final descale. */
1676	z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1677
1678	z4 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
1679	z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1680
1681	tmp10 = z3 + z4;
1682	tmp11 = z3 - z4;
1683
1684	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
1685	z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1686	z1 <<= CONST_BITS;
1687	z2 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
1688	z2 <<= CONST_BITS;
1689
1690	tmp12 = z1 - z2;
1691
1692	tmp21 = z3 + tmp12;
1693	tmp24 = z3 - tmp12;
1694
1695	tmp12 = z4 + z2;
1696
1697	tmp20 = tmp10 + tmp12;
1698	tmp25 = tmp10 - tmp12;
1699
1700	tmp12 = z4 - z1 - z2;
1701
1702	tmp22 = tmp11 + tmp12;
1703	tmp23 = tmp11 - tmp12;
1704
1705	/* Odd part */
1706
1707	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
1708	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
1709	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
1710	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
1711
1712	tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1713	tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1714
1715	tmp10 = z1 + z3;
1716	tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1717	tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1718	tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1719	tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1720	tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1721	tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1722	tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1723	MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1724
1725	z1 -= z4;
1726	z2 -= z3;
1727	z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1728	tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1729	tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1730
1731	/* Final output stage */
1732
1733	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1734	wsptr[8*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1735	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1736	wsptr[8*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1737	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1738	wsptr[8*9] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1739	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1740	wsptr[8*8] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1741	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1742	wsptr[8*7] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1743	wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1744	wsptr[8*6] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1745	}
1746
1747	/* Pass 2: process 12 rows from work array, store into output array. */
1748
1749	wsptr = workspace;
1750	for (ctr = 0; ctr < 12; ctr++) {
1751	outptr = output_buf[ctr] + output_col;
1752
1753	/* Even part */
1754
1755	/* Add fudge factor here for final descale. */
1756	z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1757	z3 <<= CONST_BITS;
1758
1759	z4 = (INT32) wsptr[4];
1760	z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1761
1762	tmp10 = z3 + z4;
1763	tmp11 = z3 - z4;
1764
1765	z1 = (INT32) wsptr[2];
1766	z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1767	z1 <<= CONST_BITS;
1768	z2 = (INT32) wsptr[6];
1769	z2 <<= CONST_BITS;
1770
1771	tmp12 = z1 - z2;
1772
1773	tmp21 = z3 + tmp12;
1774	tmp24 = z3 - tmp12;
1775
1776	tmp12 = z4 + z2;
1777
1778	tmp20 = tmp10 + tmp12;
1779	tmp25 = tmp10 - tmp12;
1780
1781	tmp12 = z4 - z1 - z2;
1782
1783	tmp22 = tmp11 + tmp12;
1784	tmp23 = tmp11 - tmp12;
1785
1786	/* Odd part */
1787
1788	z1 = (INT32) wsptr[1];
1789	z2 = (INT32) wsptr[3];
1790	z3 = (INT32) wsptr[5];
1791	z4 = (INT32) wsptr[7];
1792
1793	tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1794	tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1795
1796	tmp10 = z1 + z3;
1797	tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1798	tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1799	tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1800	tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1801	tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1802	tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1803	tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1804	MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1805
1806	z1 -= z4;
1807	z2 -= z3;
1808	z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1809	tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1810	tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1811
1812	/* Final output stage */
1813
1814	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1815	CONST_BITS+PASS1_BITS+3)
1816	& RANGE_MASK];
1817	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1818	CONST_BITS+PASS1_BITS+3)
1819	& RANGE_MASK];
1820	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1821	CONST_BITS+PASS1_BITS+3)
1822	& RANGE_MASK];
1823	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1824	CONST_BITS+PASS1_BITS+3)
1825	& RANGE_MASK];
1826	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1827	CONST_BITS+PASS1_BITS+3)
1828	& RANGE_MASK];
1829	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1830	CONST_BITS+PASS1_BITS+3)
1831	& RANGE_MASK];
1832	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1833	CONST_BITS+PASS1_BITS+3)
1834	& RANGE_MASK];
1835	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1836	CONST_BITS+PASS1_BITS+3)
1837	& RANGE_MASK];
1838	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1839	CONST_BITS+PASS1_BITS+3)
1840	& RANGE_MASK];
1841	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1842	CONST_BITS+PASS1_BITS+3)
1843	& RANGE_MASK];
1844	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1845	CONST_BITS+PASS1_BITS+3)
1846	& RANGE_MASK];
1847	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1848	CONST_BITS+PASS1_BITS+3)
1849	& RANGE_MASK];
1850
1851	wsptr += 8; /* advance pointer to next row */
1852	}
1853	}
1854
1855
1856	/*
1857	* Perform dequantization and inverse DCT on one block of coefficients,
1858	* producing a 13x13 output block.
1859	*
1860	* Optimized algorithm with 29 multiplications in the 1-D kernel.
1861	* cK represents sqrt(2) * cos(K*pi/26).
1862	*/
1863
1864	GLOBAL(void)
1865	jpeg_idct_13x13 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1866	JCOEFPTR coef_block,
1867	JSAMPARRAY output_buf, JDIMENSION output_col)
1868	{
1869	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1870	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1871	INT32 z1, z2, z3, z4;
1872	JCOEFPTR inptr;
1873	ISLOW_MULT_TYPE * quantptr;
1874	int * wsptr;
1875	JSAMPROW outptr;
1876	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1877	int ctr;
1878	int workspace[813]; / buffers data between passes */
1879	SHIFT_TEMPS
1880
1881	/* Pass 1: process columns from input, store into work array. */
1882
1883	inptr = coef_block;
1884	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1885	wsptr = workspace;
1886	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1887	/* Even part */
1888
1889	z1 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
1890	z1 <<= CONST_BITS;
1891	/* Add fudge factor here for final descale. */
1892	z1 += ONE << (CONST_BITS-PASS1_BITS-1);
1893
1894	z2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
1895	z3 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
1896	z4 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
1897
1898	tmp10 = z3 + z4;
1899	tmp11 = z3 - z4;
1900
1901	tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1902	tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1903
1904	tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1905	tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1906
1907	tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1908	tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1909
1910	tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1911	tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1912
1913	tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1914	tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1915
1916	tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1917	tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1918
1919	tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1920
1921	/* Odd part */
1922
1923	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
1924	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
1925	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
1926	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
1927
1928	tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1929	tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1930	tmp15 = z1 + z4;
1931	tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1932	tmp10 = tmp11 + tmp12 + tmp13 -
1933	MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1934	tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
1935	tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1936	tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1937	tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
1938	tmp11 += tmp14;
1939	tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1940	tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
1941	tmp12 += tmp14;
1942	tmp13 += tmp14;
1943	tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1944	tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1945	MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1946	z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1947	tmp14 += z1;
1948	tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1949	MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1950
1951	/* Final output stage */
1952
1953	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1954	wsptr[8*12] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1955	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1956	wsptr[8*11] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1957	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1958	wsptr[8*10] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1959	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1960	wsptr[8*9] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1961	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1962	wsptr[8*8] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1963	wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1964	wsptr[8*7] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1965	wsptr[8*6] = (int) RIGHT_SHIFT(tmp26, CONST_BITS-PASS1_BITS);
1966	}
1967
1968	/* Pass 2: process 13 rows from work array, store into output array. */
1969
1970	wsptr = workspace;
1971	for (ctr = 0; ctr < 13; ctr++) {
1972	outptr = output_buf[ctr] + output_col;
1973
1974	/* Even part */
1975
1976	/* Add fudge factor here for final descale. */
1977	z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1978	z1 <<= CONST_BITS;
1979
1980	z2 = (INT32) wsptr[2];
1981	z3 = (INT32) wsptr[4];
1982	z4 = (INT32) wsptr[6];
1983
1984	tmp10 = z3 + z4;
1985	tmp11 = z3 - z4;
1986
1987	tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1988	tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1989
1990	tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1991	tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1992
1993	tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1994	tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1995
1996	tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1997	tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1998
1999	tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
2000	tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
2001
2002	tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
2003	tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
2004
2005	tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
2006
2007	/* Odd part */
2008
2009	z1 = (INT32) wsptr[1];
2010	z2 = (INT32) wsptr[3];
2011	z3 = (INT32) wsptr[5];
2012	z4 = (INT32) wsptr[7];
2013
2014	tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
2015	tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
2016	tmp15 = z1 + z4;
2017	tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
2018	tmp10 = tmp11 + tmp12 + tmp13 -
2019	MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
2020	tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
2021	tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
2022	tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
2023	tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
2024	tmp11 += tmp14;
2025	tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
2026	tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
2027	tmp12 += tmp14;
2028	tmp13 += tmp14;
2029	tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
2030	tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
2031	MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
2032	z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
2033	tmp14 += z1;
2034	tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
2035	MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
2036
2037	/* Final output stage */
2038
2039	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2040	CONST_BITS+PASS1_BITS+3)
2041	& RANGE_MASK];
2042	outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2043	CONST_BITS+PASS1_BITS+3)
2044	& RANGE_MASK];
2045	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2046	CONST_BITS+PASS1_BITS+3)
2047	& RANGE_MASK];
2048	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2049	CONST_BITS+PASS1_BITS+3)
2050	& RANGE_MASK];
2051	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2052	CONST_BITS+PASS1_BITS+3)
2053	& RANGE_MASK];
2054	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2055	CONST_BITS+PASS1_BITS+3)
2056	& RANGE_MASK];
2057	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2058	CONST_BITS+PASS1_BITS+3)
2059	& RANGE_MASK];
2060	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2061	CONST_BITS+PASS1_BITS+3)
2062	& RANGE_MASK];
2063	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2064	CONST_BITS+PASS1_BITS+3)
2065	& RANGE_MASK];
2066	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2067	CONST_BITS+PASS1_BITS+3)
2068	& RANGE_MASK];
2069	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2070	CONST_BITS+PASS1_BITS+3)
2071	& RANGE_MASK];
2072	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2073	CONST_BITS+PASS1_BITS+3)
2074	& RANGE_MASK];
2075	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26,
2076	CONST_BITS+PASS1_BITS+3)
2077	& RANGE_MASK];
2078
2079	wsptr += 8; /* advance pointer to next row */
2080	}
2081	}
2082
2083
2084	/*
2085	* Perform dequantization and inverse DCT on one block of coefficients,
2086	* producing a 14x14 output block.
2087	*
2088	* Optimized algorithm with 20 multiplications in the 1-D kernel.
2089	* cK represents sqrt(2) * cos(K*pi/28).
2090	*/
2091
2092	GLOBAL(void)
2093	jpeg_idct_14x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2094	JCOEFPTR coef_block,
2095	JSAMPARRAY output_buf, JDIMENSION output_col)
2096	{
2097	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
2098	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
2099	INT32 z1, z2, z3, z4;
2100	JCOEFPTR inptr;
2101	ISLOW_MULT_TYPE * quantptr;
2102	int * wsptr;
2103	JSAMPROW outptr;
2104	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2105	int ctr;
2106	int workspace[814]; / buffers data between passes */
2107	SHIFT_TEMPS
2108
2109	/* Pass 1: process columns from input, store into work array. */
2110
2111	inptr = coef_block;
2112	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2113	wsptr = workspace;
2114	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2115	/* Even part */
2116
2117	z1 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
2118	z1 <<= CONST_BITS;
2119	/* Add fudge factor here for final descale. */
2120	z1 += ONE << (CONST_BITS-PASS1_BITS-1);
2121	z4 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
2122	z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
2123	z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
2124	z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
2125
2126	tmp10 = z1 + z2;
2127	tmp11 = z1 + z3;
2128	tmp12 = z1 - z4;
2129
2130	tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)2 /
2131	CONST_BITS-PASS1_BITS);
2132
2133	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
2134	z2 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
2135
2136	z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
2137
2138	tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
2139	tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
2140	tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
2141	MULTIPLY(z2, FIX(1.378756276)); /* c2 */
2142
2143	tmp20 = tmp10 + tmp13;
2144	tmp26 = tmp10 - tmp13;
2145	tmp21 = tmp11 + tmp14;
2146	tmp25 = tmp11 - tmp14;
2147	tmp22 = tmp12 + tmp15;
2148	tmp24 = tmp12 - tmp15;
2149
2150	/* Odd part */
2151
2152	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
2153	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
2154	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
2155	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
2156	tmp13 = z4 << CONST_BITS;
2157
2158	tmp14 = z1 + z3;
2159	tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
2160	tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
2161	tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
2162	tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
2163	tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
2164	z1 -= z2;
2165	tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */
2166	tmp16 += tmp15;
2167	z1 += z4;
2168	z4 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
2169	tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
2170	tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
2171	z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
2172	tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
2173	tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
2174
2175	tmp13 = (z1 - z3) << PASS1_BITS;
2176
2177	/* Final output stage */
2178
2179	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
2180	wsptr[8*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
2181	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
2182	wsptr[8*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
2183	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
2184	wsptr[8*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
2185	wsptr[8*3] = (int) (tmp23 + tmp13);
2186	wsptr[8*10] = (int) (tmp23 - tmp13);
2187	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
2188	wsptr[8*9] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
2189	wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
2190	wsptr[8*8] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
2191	wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
2192	wsptr[8*7] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
2193	}
2194
2195	/* Pass 2: process 14 rows from work array, store into output array. */
2196
2197	wsptr = workspace;
2198	for (ctr = 0; ctr < 14; ctr++) {
2199	outptr = output_buf[ctr] + output_col;
2200
2201	/* Even part */
2202
2203	/* Add fudge factor here for final descale. */
2204	z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2205	z1 <<= CONST_BITS;
2206	z4 = (INT32) wsptr[4];
2207	z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
2208	z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
2209	z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
2210
2211	tmp10 = z1 + z2;
2212	tmp11 = z1 + z3;
2213	tmp12 = z1 - z4;
2214
2215	tmp23 = z1 - ((z2 + z3 - z4) << 1); /* c0 = (c4+c12-c8)2 /
2216
2217	z1 = (INT32) wsptr[2];
2218	z2 = (INT32) wsptr[6];
2219
2220	z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
2221
2222	tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
2223	tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
2224	tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
2225	MULTIPLY(z2, FIX(1.378756276)); /* c2 */
2226
2227	tmp20 = tmp10 + tmp13;
2228	tmp26 = tmp10 - tmp13;
2229	tmp21 = tmp11 + tmp14;
2230	tmp25 = tmp11 - tmp14;
2231	tmp22 = tmp12 + tmp15;
2232	tmp24 = tmp12 - tmp15;
2233
2234	/* Odd part */
2235
2236	z1 = (INT32) wsptr[1];
2237	z2 = (INT32) wsptr[3];
2238	z3 = (INT32) wsptr[5];
2239	z4 = (INT32) wsptr[7];
2240	z4 <<= CONST_BITS;
2241
2242	tmp14 = z1 + z3;
2243	tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
2244	tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
2245	tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
2246	tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
2247	tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
2248	z1 -= z2;
2249	tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */
2250	tmp16 += tmp15;
2251	tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4; /* -c13 */
2252	tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
2253	tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
2254	tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
2255	tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
2256	tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
2257
2258	tmp13 = ((z1 - z3) << CONST_BITS) + z4;
2259
2260	/* Final output stage */
2261
2262	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2263	CONST_BITS+PASS1_BITS+3)
2264	& RANGE_MASK];
2265	outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2266	CONST_BITS+PASS1_BITS+3)
2267	& RANGE_MASK];
2268	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2269	CONST_BITS+PASS1_BITS+3)
2270	& RANGE_MASK];
2271	outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2272	CONST_BITS+PASS1_BITS+3)
2273	& RANGE_MASK];
2274	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2275	CONST_BITS+PASS1_BITS+3)
2276	& RANGE_MASK];
2277	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2278	CONST_BITS+PASS1_BITS+3)
2279	& RANGE_MASK];
2280	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2281	CONST_BITS+PASS1_BITS+3)
2282	& RANGE_MASK];
2283	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2284	CONST_BITS+PASS1_BITS+3)
2285	& RANGE_MASK];
2286	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2287	CONST_BITS+PASS1_BITS+3)
2288	& RANGE_MASK];
2289	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2290	CONST_BITS+PASS1_BITS+3)
2291	& RANGE_MASK];
2292	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2293	CONST_BITS+PASS1_BITS+3)
2294	& RANGE_MASK];
2295	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2296	CONST_BITS+PASS1_BITS+3)
2297	& RANGE_MASK];
2298	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
2299	CONST_BITS+PASS1_BITS+3)
2300	& RANGE_MASK];
2301	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
2302	CONST_BITS+PASS1_BITS+3)
2303	& RANGE_MASK];
2304
2305	wsptr += 8; /* advance pointer to next row */
2306	}
2307	}
2308
2309
2310	/*
2311	* Perform dequantization and inverse DCT on one block of coefficients,
2312	* producing a 15x15 output block.
2313	*
2314	* Optimized algorithm with 22 multiplications in the 1-D kernel.
2315	* cK represents sqrt(2) * cos(K*pi/30).
2316	*/
2317
2318	GLOBAL(void)
2319	jpeg_idct_15x15 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2320	JCOEFPTR coef_block,
2321	JSAMPARRAY output_buf, JDIMENSION output_col)
2322	{
2323	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
2324	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2325	INT32 z1, z2, z3, z4;
2326	JCOEFPTR inptr;
2327	ISLOW_MULT_TYPE * quantptr;
2328	int * wsptr;
2329	JSAMPROW outptr;
2330	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2331	int ctr;
2332	int workspace[815]; / buffers data between passes */
2333	SHIFT_TEMPS
2334
2335	/* Pass 1: process columns from input, store into work array. */
2336
2337	inptr = coef_block;
2338	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2339	wsptr = workspace;
2340	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2341	/* Even part */
2342
2343	z1 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
2344	z1 <<= CONST_BITS;
2345	/* Add fudge factor here for final descale. */
2346	z1 += ONE << (CONST_BITS-PASS1_BITS-1);
2347
2348	z2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
2349	z3 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
2350	z4 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
2351
2352	tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2353	tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2354
2355	tmp12 = z1 - tmp10;
2356	tmp13 = z1 + tmp11;
2357	z1 -= (tmp11 - tmp10) << 1; /* c0 = (c6-c12)2 /
2358
2359	z4 = z2 - z3;
2360	z3 += z2;
2361	tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2362	tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2363	z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2364
2365	tmp20 = tmp13 + tmp10 + tmp11;
2366	tmp23 = tmp12 - tmp10 + tmp11 + z2;
2367
2368	tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2369	tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2370
2371	tmp25 = tmp13 - tmp10 - tmp11;
2372	tmp26 = tmp12 + tmp10 - tmp11 - z2;
2373
2374	tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2375	tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2376
2377	tmp21 = tmp12 + tmp10 + tmp11;
2378	tmp24 = tmp13 - tmp10 + tmp11;
2379	tmp11 += tmp11;
2380	tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2381	tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)2 /
2382
2383	/* Odd part */
2384
2385	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
2386	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
2387	z4 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
2388	z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2389	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
2390
2391	tmp13 = z2 - z4;
2392	tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2393	tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2394	tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2395
2396	tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
2397	tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
2398	z2 = z1 - z4;
2399	tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2400
2401	tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2402	tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2403	tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2404	z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2405	tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2406	tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2407
2408	/* Final output stage */
2409
2410	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
2411	wsptr[8*14] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
2412	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
2413	wsptr[8*13] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
2414	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
2415	wsptr[8*12] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
2416	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
2417	wsptr[8*11] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
2418	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
2419	wsptr[8*10] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
2420	wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
2421	wsptr[8*9] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
2422	wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
2423	wsptr[8*8] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
2424	wsptr[8*7] = (int) RIGHT_SHIFT(tmp27, CONST_BITS-PASS1_BITS);
2425	}
2426
2427	/* Pass 2: process 15 rows from work array, store into output array. */
2428
2429	wsptr = workspace;
2430	for (ctr = 0; ctr < 15; ctr++) {
2431	outptr = output_buf[ctr] + output_col;
2432
2433	/* Even part */
2434
2435	/* Add fudge factor here for final descale. */
2436	z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2437	z1 <<= CONST_BITS;
2438
2439	z2 = (INT32) wsptr[2];
2440	z3 = (INT32) wsptr[4];
2441	z4 = (INT32) wsptr[6];
2442
2443	tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
2444	tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
2445
2446	tmp12 = z1 - tmp10;
2447	tmp13 = z1 + tmp11;
2448	z1 -= (tmp11 - tmp10) << 1; /* c0 = (c6-c12)2 /
2449
2450	z4 = z2 - z3;
2451	z3 += z2;
2452	tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
2453	tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
2454	z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
2455
2456	tmp20 = tmp13 + tmp10 + tmp11;
2457	tmp23 = tmp12 - tmp10 + tmp11 + z2;
2458
2459	tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
2460	tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
2461
2462	tmp25 = tmp13 - tmp10 - tmp11;
2463	tmp26 = tmp12 + tmp10 - tmp11 - z2;
2464
2465	tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
2466	tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
2467
2468	tmp21 = tmp12 + tmp10 + tmp11;
2469	tmp24 = tmp13 - tmp10 + tmp11;
2470	tmp11 += tmp11;
2471	tmp22 = z1 + tmp11; /* c10 = c6-c12 */
2472	tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)2 /
2473
2474	/* Odd part */
2475
2476	z1 = (INT32) wsptr[1];
2477	z2 = (INT32) wsptr[3];
2478	z4 = (INT32) wsptr[5];
2479	z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
2480	z4 = (INT32) wsptr[7];
2481
2482	tmp13 = z2 - z4;
2483	tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
2484	tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
2485	tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
2486
2487	tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
2488	tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
2489	z2 = z1 - z4;
2490	tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
2491
2492	tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
2493	tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
2494	tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
2495	z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
2496	tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
2497	tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
2498
2499	/* Final output stage */
2500
2501	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
2502	CONST_BITS+PASS1_BITS+3)
2503	& RANGE_MASK];
2504	outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
2505	CONST_BITS+PASS1_BITS+3)
2506	& RANGE_MASK];
2507	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
2508	CONST_BITS+PASS1_BITS+3)
2509	& RANGE_MASK];
2510	outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
2511	CONST_BITS+PASS1_BITS+3)
2512	& RANGE_MASK];
2513	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
2514	CONST_BITS+PASS1_BITS+3)
2515	& RANGE_MASK];
2516	outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
2517	CONST_BITS+PASS1_BITS+3)
2518	& RANGE_MASK];
2519	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
2520	CONST_BITS+PASS1_BITS+3)
2521	& RANGE_MASK];
2522	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
2523	CONST_BITS+PASS1_BITS+3)
2524	& RANGE_MASK];
2525	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
2526	CONST_BITS+PASS1_BITS+3)
2527	& RANGE_MASK];
2528	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
2529	CONST_BITS+PASS1_BITS+3)
2530	& RANGE_MASK];
2531	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
2532	CONST_BITS+PASS1_BITS+3)
2533	& RANGE_MASK];
2534	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
2535	CONST_BITS+PASS1_BITS+3)
2536	& RANGE_MASK];
2537	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
2538	CONST_BITS+PASS1_BITS+3)
2539	& RANGE_MASK];
2540	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
2541	CONST_BITS+PASS1_BITS+3)
2542	& RANGE_MASK];
2543	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27,
2544	CONST_BITS+PASS1_BITS+3)
2545	& RANGE_MASK];
2546
2547	wsptr += 8; /* advance pointer to next row */
2548	}
2549	}
2550
2551
2552	/*
2553	* Perform dequantization and inverse DCT on one block of coefficients,
2554	* producing a 16x16 output block.
2555	*
2556	* Optimized algorithm with 28 multiplications in the 1-D kernel.
2557	* cK represents sqrt(2) * cos(K*pi/32).
2558	*/
2559
2560	GLOBAL(void)
2561	jpeg_idct_16x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2562	JCOEFPTR coef_block,
2563	JSAMPARRAY output_buf, JDIMENSION output_col)
2564	{
2565	INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
2566	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2567	INT32 z1, z2, z3, z4;
2568	JCOEFPTR inptr;
2569	ISLOW_MULT_TYPE * quantptr;
2570	int * wsptr;
2571	JSAMPROW outptr;
2572	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2573	int ctr;
2574	int workspace[816]; / buffers data between passes */
2575	SHIFT_TEMPS
2576
2577	/* Pass 1: process columns from input, store into work array. */
2578
2579	inptr = coef_block;
2580	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2581	wsptr = workspace;
2582	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
2583	/* Even part */
2584
2585	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
2586	tmp0 <<= CONST_BITS;
2587	/* Add fudge factor here for final descale. */
2588	tmp0 += 1 << (CONST_BITS-PASS1_BITS-1);
2589
2590	z1 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
2591	tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2592	tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2593
2594	tmp10 = tmp0 + tmp1;
2595	tmp11 = tmp0 - tmp1;
2596	tmp12 = tmp0 + tmp2;
2597	tmp13 = tmp0 - tmp2;
2598
2599	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
2600	z2 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
2601	z3 = z1 - z2;
2602	z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2603	z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2604
2605	tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2606	tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2607	tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2608	tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2609
2610	tmp20 = tmp10 + tmp0;
2611	tmp27 = tmp10 - tmp0;
2612	tmp21 = tmp12 + tmp1;
2613	tmp26 = tmp12 - tmp1;
2614	tmp22 = tmp13 + tmp2;
2615	tmp25 = tmp13 - tmp2;
2616	tmp23 = tmp11 + tmp3;
2617	tmp24 = tmp11 - tmp3;
2618
2619	/* Odd part */
2620
2621	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
2622	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
2623	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
2624	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
2625
2626	tmp11 = z1 + z3;
2627
2628	tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2629	tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2630	tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2631	tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2632	tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2633	tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2634	tmp0 = tmp1 + tmp2 + tmp3 -
2635	MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2636	tmp13 = tmp10 + tmp11 + tmp12 -
2637	MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2638	z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2639	tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2640	tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2641	z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2642	tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2643	tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2644	z2 += z4;
2645	z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
2646	tmp1 += z1;
2647	tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2648	z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
2649	tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2650	tmp12 += z2;
2651	z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
2652	tmp2 += z2;
2653	tmp3 += z2;
2654	z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2655	tmp10 += z2;
2656	tmp11 += z2;
2657
2658	/* Final output stage */
2659
2660	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS-PASS1_BITS);
2661	wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS-PASS1_BITS);
2662	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS-PASS1_BITS);
2663	wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS-PASS1_BITS);
2664	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS-PASS1_BITS);
2665	wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS-PASS1_BITS);
2666	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS-PASS1_BITS);
2667	wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS-PASS1_BITS);
2668	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
2669	wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
2670	wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
2671	wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
2672	wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
2673	wsptr[8*9] = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
2674	wsptr[8*7] = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
2675	wsptr[8*8] = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
2676	}
2677
2678	/* Pass 2: process 16 rows from work array, store into output array. */
2679
2680	wsptr = workspace;
2681	for (ctr = 0; ctr < 16; ctr++) {
2682	outptr = output_buf[ctr] + output_col;
2683
2684	/* Even part */
2685
2686	/* Add fudge factor here for final descale. */
2687	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2688	tmp0 <<= CONST_BITS;
2689
2690	z1 = (INT32) wsptr[4];
2691	tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2692	tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2693
2694	tmp10 = tmp0 + tmp1;
2695	tmp11 = tmp0 - tmp1;
2696	tmp12 = tmp0 + tmp2;
2697	tmp13 = tmp0 - tmp2;
2698
2699	z1 = (INT32) wsptr[2];
2700	z2 = (INT32) wsptr[6];
2701	z3 = z1 - z2;
2702	z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2703	z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2704
2705	tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2706	tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2707	tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2708	tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2709
2710	tmp20 = tmp10 + tmp0;
2711	tmp27 = tmp10 - tmp0;
2712	tmp21 = tmp12 + tmp1;
2713	tmp26 = tmp12 - tmp1;
2714	tmp22 = tmp13 + tmp2;
2715	tmp25 = tmp13 - tmp2;
2716	tmp23 = tmp11 + tmp3;
2717	tmp24 = tmp11 - tmp3;
2718
2719	/* Odd part */
2720
2721	z1 = (INT32) wsptr[1];
2722	z2 = (INT32) wsptr[3];
2723	z3 = (INT32) wsptr[5];
2724	z4 = (INT32) wsptr[7];
2725
2726	tmp11 = z1 + z3;
2727
2728	tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
2729	tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
2730	tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
2731	tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
2732	tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
2733	tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
2734	tmp0 = tmp1 + tmp2 + tmp3 -
2735	MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
2736	tmp13 = tmp10 + tmp11 + tmp12 -
2737	MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
2738	z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
2739	tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
2740	tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
2741	z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
2742	tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
2743	tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
2744	z2 += z4;
2745	z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
2746	tmp1 += z1;
2747	tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
2748	z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
2749	tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
2750	tmp12 += z2;
2751	z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
2752	tmp2 += z2;
2753	tmp3 += z2;
2754	z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
2755	tmp10 += z2;
2756	tmp11 += z2;
2757
2758	/* Final output stage */
2759
2760	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
2761	CONST_BITS+PASS1_BITS+3)
2762	& RANGE_MASK];
2763	outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
2764	CONST_BITS+PASS1_BITS+3)
2765	& RANGE_MASK];
2766	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
2767	CONST_BITS+PASS1_BITS+3)
2768	& RANGE_MASK];
2769	outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
2770	CONST_BITS+PASS1_BITS+3)
2771	& RANGE_MASK];
2772	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
2773	CONST_BITS+PASS1_BITS+3)
2774	& RANGE_MASK];
2775	outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
2776	CONST_BITS+PASS1_BITS+3)
2777	& RANGE_MASK];
2778	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
2779	CONST_BITS+PASS1_BITS+3)
2780	& RANGE_MASK];
2781	outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
2782	CONST_BITS+PASS1_BITS+3)
2783	& RANGE_MASK];
2784	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
2785	CONST_BITS+PASS1_BITS+3)
2786	& RANGE_MASK];
2787	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
2788	CONST_BITS+PASS1_BITS+3)
2789	& RANGE_MASK];
2790	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
2791	CONST_BITS+PASS1_BITS+3)
2792	& RANGE_MASK];
2793	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
2794	CONST_BITS+PASS1_BITS+3)
2795	& RANGE_MASK];
2796	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
2797	CONST_BITS+PASS1_BITS+3)
2798	& RANGE_MASK];
2799	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
2800	CONST_BITS+PASS1_BITS+3)
2801	& RANGE_MASK];
2802	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
2803	CONST_BITS+PASS1_BITS+3)
2804	& RANGE_MASK];
2805	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
2806	CONST_BITS+PASS1_BITS+3)
2807	& RANGE_MASK];
2808
2809	wsptr += 8; /* advance pointer to next row */
2810	}
2811	}
2812
2813
2814	/*
2815	* Perform dequantization and inverse DCT on one block of coefficients,
2816	* producing a 16x8 output block.
2817	*
2818	* 8-point IDCT in pass 1 (columns), 16-point in pass 2 (rows).
2819	*/
2820
2821	GLOBAL(void)
2822	jpeg_idct_16x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
2823	JCOEFPTR coef_block,
2824	JSAMPARRAY output_buf, JDIMENSION output_col)
2825	{
2826	INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
2827	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
2828	INT32 z1, z2, z3, z4;
2829	JCOEFPTR inptr;
2830	ISLOW_MULT_TYPE * quantptr;
2831	int * wsptr;
2832	JSAMPROW outptr;
2833	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
2834	int ctr;
2835	int workspace[88]; / buffers data between passes */
2836	SHIFT_TEMPS
2837
2838	/* Pass 1: process columns from input, store into work array. */
2839	/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
2840	/* furthermore, we scale the results by 2*PASS1_BITS. /
2841
2842	inptr = coef_block;
2843	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
2844	wsptr = workspace;
2845	for (ctr = DCTSIZE; ctr > 0; ctr--) {
2846	/* Due to quantization, we will usually find that many of the input
2847	* coefficients are zero, especially the AC terms. We can exploit this
2848	* by short-circuiting the IDCT calculation for any column in which all
2849	* the AC terms are zero. In that case each output is equal to the
2850	* DC coefficient (with scale factor as needed).
2851	* With typical images and quantization tables, half or more of the
2852	* column DCT calculations can be simplified this way.
2853	*/
2854
2855	if (inptr[DCTSIZE1] == 0 && inptr[DCTSIZE2] == 0 &&
2856	inptr[DCTSIZE3] == 0 && inptr[DCTSIZE4] == 0 &&
2857	inptr[DCTSIZE5] == 0 && inptr[DCTSIZE6] == 0 &&
2858	inptr[DCTSIZE*7] == 0) {
2859	/* AC terms all zero */
2860	int dcval = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]) << PASS1_BITS;
2861
2862	wsptr[DCTSIZE*0] = dcval;
2863	wsptr[DCTSIZE*1] = dcval;
2864	wsptr[DCTSIZE*2] = dcval;
2865	wsptr[DCTSIZE*3] = dcval;
2866	wsptr[DCTSIZE*4] = dcval;
2867	wsptr[DCTSIZE*5] = dcval;
2868	wsptr[DCTSIZE*6] = dcval;
2869	wsptr[DCTSIZE*7] = dcval;
2870
2871	inptr++; /* advance pointers to next column */
2872	quantptr++;
2873	wsptr++;
2874	continue;
2875	}
2876
2877	/* Even part: reverse the even part of the forward DCT. */
2878	/* The rotator is sqrt(2)c(-6). /
2879
2880	z2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
2881	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
2882
2883	z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
2884	tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
2885	tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
2886
2887	z2 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
2888	z3 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
2889	z2 <<= CONST_BITS;
2890	z3 <<= CONST_BITS;
2891	/* Add fudge factor here for final descale. */
2892	z2 += ONE << (CONST_BITS-PASS1_BITS-1);
2893
2894	tmp0 = z2 + z3;
2895	tmp1 = z2 - z3;
2896
2897	tmp10 = tmp0 + tmp2;
2898	tmp13 = tmp0 - tmp2;
2899	tmp11 = tmp1 + tmp3;
2900	tmp12 = tmp1 - tmp3;
2901
2902	/* Odd part per figure 8; the matrix is unitary and hence its
2903	* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
2904	*/
2905
2906	tmp0 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
2907	tmp1 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
2908	tmp2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
2909	tmp3 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
2910
2911	z2 = tmp0 + tmp2;
2912	z3 = tmp1 + tmp3;
2913
2914	z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
2915	z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
2916	z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
2917	z2 += z1;
2918	z3 += z1;
2919
2920	z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
2921	tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
2922	tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
2923	tmp0 += z1 + z2;
2924	tmp3 += z1 + z3;
2925
2926	z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
2927	tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
2928	tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
2929	tmp1 += z1 + z3;
2930	tmp2 += z1 + z2;
2931
2932	/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
2933
2934	wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
2935	wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
2936	wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
2937	wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
2938	wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
2939	wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
2940	wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
2941	wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
2942
2943	inptr++; /* advance pointers to next column */
2944	quantptr++;
2945	wsptr++;
2946	}
2947
2948	/* Pass 2: process 8 rows from work array, store into output array.
2949	* 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
2950	*/
2951	wsptr = workspace;
2952	for (ctr = 0; ctr < 8; ctr++) {
2953	outptr = output_buf[ctr] + output_col;
2954
2955	/* Even part */
2956
2957	/* Add fudge factor here for final descale. */
2958	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
2959	tmp0 <<= CONST_BITS;
2960
2961	z1 = (INT32) wsptr[4];
2962	tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
2963	tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
2964
2965	tmp10 = tmp0 + tmp1;
2966	tmp11 = tmp0 - tmp1;
2967	tmp12 = tmp0 + tmp2;
2968	tmp13 = tmp0 - tmp2;
2969
2970	z1 = (INT32) wsptr[2];
2971	z2 = (INT32) wsptr[6];
2972	z3 = z1 - z2;
2973	z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
2974	z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
2975
2976	tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
2977	tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
2978	tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
2979	tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
2980
2981	tmp20 = tmp10 + tmp0;
2982	tmp27 = tmp10 - tmp0;
2983	tmp21 = tmp12 + tmp1;
2984	tmp26 = tmp12 - tmp1;
2985	tmp22 = tmp13 + tmp2;
2986	tmp25 = tmp13 - tmp2;
2987	tmp23 = tmp11 + tmp3;
2988	tmp24 = tmp11 - tmp3;
2989
2990	/* Odd part */
2991
2992	z1 = (INT32) wsptr[1];
2993	z2 = (INT32) wsptr[3];
2994	z3 = (INT32) wsptr[5];
2995	z4 = (INT32) wsptr[7];
2996
2997	tmp11 = z1 + z3;
2998
2999	tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
3000	tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
3001	tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
3002	tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
3003	tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
3004	tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
3005	tmp0 = tmp1 + tmp2 + tmp3 -
3006	MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
3007	tmp13 = tmp10 + tmp11 + tmp12 -
3008	MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
3009	z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
3010	tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
3011	tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
3012	z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
3013	tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
3014	tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
3015	z2 += z4;
3016	z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
3017	tmp1 += z1;
3018	tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
3019	z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
3020	tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
3021	tmp12 += z2;
3022	z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
3023	tmp2 += z2;
3024	tmp3 += z2;
3025	z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
3026	tmp10 += z2;
3027	tmp11 += z2;
3028
3029	/* Final output stage */
3030
3031	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
3032	CONST_BITS+PASS1_BITS+3)
3033	& RANGE_MASK];
3034	outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
3035	CONST_BITS+PASS1_BITS+3)
3036	& RANGE_MASK];
3037	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
3038	CONST_BITS+PASS1_BITS+3)
3039	& RANGE_MASK];
3040	outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
3041	CONST_BITS+PASS1_BITS+3)
3042	& RANGE_MASK];
3043	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
3044	CONST_BITS+PASS1_BITS+3)
3045	& RANGE_MASK];
3046	outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
3047	CONST_BITS+PASS1_BITS+3)
3048	& RANGE_MASK];
3049	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
3050	CONST_BITS+PASS1_BITS+3)
3051	& RANGE_MASK];
3052	outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
3053	CONST_BITS+PASS1_BITS+3)
3054	& RANGE_MASK];
3055	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
3056	CONST_BITS+PASS1_BITS+3)
3057	& RANGE_MASK];
3058	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
3059	CONST_BITS+PASS1_BITS+3)
3060	& RANGE_MASK];
3061	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
3062	CONST_BITS+PASS1_BITS+3)
3063	& RANGE_MASK];
3064	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
3065	CONST_BITS+PASS1_BITS+3)
3066	& RANGE_MASK];
3067	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
3068	CONST_BITS+PASS1_BITS+3)
3069	& RANGE_MASK];
3070	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
3071	CONST_BITS+PASS1_BITS+3)
3072	& RANGE_MASK];
3073	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
3074	CONST_BITS+PASS1_BITS+3)
3075	& RANGE_MASK];
3076	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
3077	CONST_BITS+PASS1_BITS+3)
3078	& RANGE_MASK];
3079
3080	wsptr += 8; /* advance pointer to next row */
3081	}
3082	}
3083
3084
3085	/*
3086	* Perform dequantization and inverse DCT on one block of coefficients,
3087	* producing a 14x7 output block.
3088	*
3089	* 7-point IDCT in pass 1 (columns), 14-point in pass 2 (rows).
3090	*/
3091
3092	GLOBAL(void)
3093	jpeg_idct_14x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3094	JCOEFPTR coef_block,
3095	JSAMPARRAY output_buf, JDIMENSION output_col)
3096	{
3097	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
3098	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
3099	INT32 z1, z2, z3, z4;
3100	JCOEFPTR inptr;
3101	ISLOW_MULT_TYPE * quantptr;
3102	int * wsptr;
3103	JSAMPROW outptr;
3104	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3105	int ctr;
3106	int workspace[87]; / buffers data between passes */
3107	SHIFT_TEMPS
3108
3109	/* Pass 1: process columns from input, store into work array.
3110	* 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
3111	*/
3112	inptr = coef_block;
3113	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3114	wsptr = workspace;
3115	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3116	/* Even part */
3117
3118	tmp23 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
3119	tmp23 <<= CONST_BITS;
3120	/* Add fudge factor here for final descale. */
3121	tmp23 += ONE << (CONST_BITS-PASS1_BITS-1);
3122
3123	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
3124	z2 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
3125	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
3126
3127	tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
3128	tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
3129	tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
3130	tmp10 = z1 + z3;
3131	z2 -= tmp10;
3132	tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
3133	tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
3134	tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
3135	tmp23 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
3136
3137	/* Odd part */
3138
3139	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
3140	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
3141	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
3142
3143	tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
3144	tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
3145	tmp10 = tmp11 - tmp12;
3146	tmp11 += tmp12;
3147	tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
3148	tmp11 += tmp12;
3149	z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
3150	tmp10 += z2;
3151	tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
3152
3153	/* Final output stage */
3154
3155	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
3156	wsptr[8*6] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
3157	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
3158	wsptr[8*5] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
3159	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
3160	wsptr[8*4] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
3161	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23, CONST_BITS-PASS1_BITS);
3162	}
3163
3164	/* Pass 2: process 7 rows from work array, store into output array.
3165	* 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
3166	*/
3167	wsptr = workspace;
3168	for (ctr = 0; ctr < 7; ctr++) {
3169	outptr = output_buf[ctr] + output_col;
3170
3171	/* Even part */
3172
3173	/* Add fudge factor here for final descale. */
3174	z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3175	z1 <<= CONST_BITS;
3176	z4 = (INT32) wsptr[4];
3177	z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
3178	z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
3179	z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
3180
3181	tmp10 = z1 + z2;
3182	tmp11 = z1 + z3;
3183	tmp12 = z1 - z4;
3184
3185	tmp23 = z1 - ((z2 + z3 - z4) << 1); /* c0 = (c4+c12-c8)2 /
3186
3187	z1 = (INT32) wsptr[2];
3188	z2 = (INT32) wsptr[6];
3189
3190	z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
3191
3192	tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
3193	tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
3194	tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
3195	MULTIPLY(z2, FIX(1.378756276)); /* c2 */
3196
3197	tmp20 = tmp10 + tmp13;
3198	tmp26 = tmp10 - tmp13;
3199	tmp21 = tmp11 + tmp14;
3200	tmp25 = tmp11 - tmp14;
3201	tmp22 = tmp12 + tmp15;
3202	tmp24 = tmp12 - tmp15;
3203
3204	/* Odd part */
3205
3206	z1 = (INT32) wsptr[1];
3207	z2 = (INT32) wsptr[3];
3208	z3 = (INT32) wsptr[5];
3209	z4 = (INT32) wsptr[7];
3210	z4 <<= CONST_BITS;
3211
3212	tmp14 = z1 + z3;
3213	tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
3214	tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
3215	tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
3216	tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
3217	tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
3218	z1 -= z2;
3219	tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */
3220	tmp16 += tmp15;
3221	tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4; /* -c13 */
3222	tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
3223	tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
3224	tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
3225	tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
3226	tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
3227
3228	tmp13 = ((z1 - z3) << CONST_BITS) + z4;
3229
3230	/* Final output stage */
3231
3232	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
3233	CONST_BITS+PASS1_BITS+3)
3234	& RANGE_MASK];
3235	outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
3236	CONST_BITS+PASS1_BITS+3)
3237	& RANGE_MASK];
3238	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
3239	CONST_BITS+PASS1_BITS+3)
3240	& RANGE_MASK];
3241	outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
3242	CONST_BITS+PASS1_BITS+3)
3243	& RANGE_MASK];
3244	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
3245	CONST_BITS+PASS1_BITS+3)
3246	& RANGE_MASK];
3247	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
3248	CONST_BITS+PASS1_BITS+3)
3249	& RANGE_MASK];
3250	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
3251	CONST_BITS+PASS1_BITS+3)
3252	& RANGE_MASK];
3253	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
3254	CONST_BITS+PASS1_BITS+3)
3255	& RANGE_MASK];
3256	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
3257	CONST_BITS+PASS1_BITS+3)
3258	& RANGE_MASK];
3259	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
3260	CONST_BITS+PASS1_BITS+3)
3261	& RANGE_MASK];
3262	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
3263	CONST_BITS+PASS1_BITS+3)
3264	& RANGE_MASK];
3265	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
3266	CONST_BITS+PASS1_BITS+3)
3267	& RANGE_MASK];
3268	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
3269	CONST_BITS+PASS1_BITS+3)
3270	& RANGE_MASK];
3271	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
3272	CONST_BITS+PASS1_BITS+3)
3273	& RANGE_MASK];
3274
3275	wsptr += 8; /* advance pointer to next row */
3276	}
3277	}
3278
3279
3280	/*
3281	* Perform dequantization and inverse DCT on one block of coefficients,
3282	* producing a 12x6 output block.
3283	*
3284	* 6-point IDCT in pass 1 (columns), 12-point in pass 2 (rows).
3285	*/
3286
3287	GLOBAL(void)
3288	jpeg_idct_12x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3289	JCOEFPTR coef_block,
3290	JSAMPARRAY output_buf, JDIMENSION output_col)
3291	{
3292	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
3293	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
3294	INT32 z1, z2, z3, z4;
3295	JCOEFPTR inptr;
3296	ISLOW_MULT_TYPE * quantptr;
3297	int * wsptr;
3298	JSAMPROW outptr;
3299	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3300	int ctr;
3301	int workspace[86]; / buffers data between passes */
3302	SHIFT_TEMPS
3303
3304	/* Pass 1: process columns from input, store into work array.
3305	* 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
3306	*/
3307	inptr = coef_block;
3308	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3309	wsptr = workspace;
3310	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3311	/* Even part */
3312
3313	tmp10 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
3314	tmp10 <<= CONST_BITS;
3315	/* Add fudge factor here for final descale. */
3316	tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
3317	tmp12 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
3318	tmp20 = MULTIPLY(tmp12, FIX(0.707106781)); /* c4 */
3319	tmp11 = tmp10 + tmp20;
3320	tmp21 = RIGHT_SHIFT(tmp10 - tmp20 - tmp20, CONST_BITS-PASS1_BITS);
3321	tmp20 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
3322	tmp10 = MULTIPLY(tmp20, FIX(1.224744871)); /* c2 */
3323	tmp20 = tmp11 + tmp10;
3324	tmp22 = tmp11 - tmp10;
3325
3326	/* Odd part */
3327
3328	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
3329	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
3330	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
3331	tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
3332	tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
3333	tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
3334	tmp11 = (z1 - z2 - z3) << PASS1_BITS;
3335
3336	/* Final output stage */
3337
3338	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
3339	wsptr[8*5] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
3340	wsptr[8*1] = (int) (tmp21 + tmp11);
3341	wsptr[8*4] = (int) (tmp21 - tmp11);
3342	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
3343	wsptr[8*3] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
3344	}
3345
3346	/* Pass 2: process 6 rows from work array, store into output array.
3347	* 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
3348	*/
3349	wsptr = workspace;
3350	for (ctr = 0; ctr < 6; ctr++) {
3351	outptr = output_buf[ctr] + output_col;
3352
3353	/* Even part */
3354
3355	/* Add fudge factor here for final descale. */
3356	z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3357	z3 <<= CONST_BITS;
3358
3359	z4 = (INT32) wsptr[4];
3360	z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
3361
3362	tmp10 = z3 + z4;
3363	tmp11 = z3 - z4;
3364
3365	z1 = (INT32) wsptr[2];
3366	z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
3367	z1 <<= CONST_BITS;
3368	z2 = (INT32) wsptr[6];
3369	z2 <<= CONST_BITS;
3370
3371	tmp12 = z1 - z2;
3372
3373	tmp21 = z3 + tmp12;
3374	tmp24 = z3 - tmp12;
3375
3376	tmp12 = z4 + z2;
3377
3378	tmp20 = tmp10 + tmp12;
3379	tmp25 = tmp10 - tmp12;
3380
3381	tmp12 = z4 - z1 - z2;
3382
3383	tmp22 = tmp11 + tmp12;
3384	tmp23 = tmp11 - tmp12;
3385
3386	/* Odd part */
3387
3388	z1 = (INT32) wsptr[1];
3389	z2 = (INT32) wsptr[3];
3390	z3 = (INT32) wsptr[5];
3391	z4 = (INT32) wsptr[7];
3392
3393	tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
3394	tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
3395
3396	tmp10 = z1 + z3;
3397	tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
3398	tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
3399	tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
3400	tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
3401	tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
3402	tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
3403	tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
3404	MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
3405
3406	z1 -= z4;
3407	z2 -= z3;
3408	z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
3409	tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
3410	tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
3411
3412	/* Final output stage */
3413
3414	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
3415	CONST_BITS+PASS1_BITS+3)
3416	& RANGE_MASK];
3417	outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
3418	CONST_BITS+PASS1_BITS+3)
3419	& RANGE_MASK];
3420	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
3421	CONST_BITS+PASS1_BITS+3)
3422	& RANGE_MASK];
3423	outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
3424	CONST_BITS+PASS1_BITS+3)
3425	& RANGE_MASK];
3426	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
3427	CONST_BITS+PASS1_BITS+3)
3428	& RANGE_MASK];
3429	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
3430	CONST_BITS+PASS1_BITS+3)
3431	& RANGE_MASK];
3432	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
3433	CONST_BITS+PASS1_BITS+3)
3434	& RANGE_MASK];
3435	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
3436	CONST_BITS+PASS1_BITS+3)
3437	& RANGE_MASK];
3438	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
3439	CONST_BITS+PASS1_BITS+3)
3440	& RANGE_MASK];
3441	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
3442	CONST_BITS+PASS1_BITS+3)
3443	& RANGE_MASK];
3444	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
3445	CONST_BITS+PASS1_BITS+3)
3446	& RANGE_MASK];
3447	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
3448	CONST_BITS+PASS1_BITS+3)
3449	& RANGE_MASK];
3450
3451	wsptr += 8; /* advance pointer to next row */
3452	}
3453	}
3454
3455
3456	/*
3457	* Perform dequantization and inverse DCT on one block of coefficients,
3458	* producing a 10x5 output block.
3459	*
3460	* 5-point IDCT in pass 1 (columns), 10-point in pass 2 (rows).
3461	*/
3462
3463	GLOBAL(void)
3464	jpeg_idct_10x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3465	JCOEFPTR coef_block,
3466	JSAMPARRAY output_buf, JDIMENSION output_col)
3467	{
3468	INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
3469	INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
3470	INT32 z1, z2, z3, z4;
3471	JCOEFPTR inptr;
3472	ISLOW_MULT_TYPE * quantptr;
3473	int * wsptr;
3474	JSAMPROW outptr;
3475	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3476	int ctr;
3477	int workspace[85]; / buffers data between passes */
3478	SHIFT_TEMPS
3479
3480	/* Pass 1: process columns from input, store into work array.
3481	* 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
3482	*/
3483	inptr = coef_block;
3484	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3485	wsptr = workspace;
3486	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3487	/* Even part */
3488
3489	tmp12 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
3490	tmp12 <<= CONST_BITS;
3491	/* Add fudge factor here for final descale. */
3492	tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
3493	tmp13 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
3494	tmp14 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
3495	z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
3496	z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
3497	z3 = tmp12 + z2;
3498	tmp10 = z3 + z1;
3499	tmp11 = z3 - z1;
3500	tmp12 -= z2 << 2;
3501
3502	/* Odd part */
3503
3504	z2 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
3505	z3 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
3506
3507	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
3508	tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
3509	tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
3510
3511	/* Final output stage */
3512
3513	wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp13, CONST_BITS-PASS1_BITS);
3514	wsptr[8*4] = (int) RIGHT_SHIFT(tmp10 - tmp13, CONST_BITS-PASS1_BITS);
3515	wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp14, CONST_BITS-PASS1_BITS);
3516	wsptr[8*3] = (int) RIGHT_SHIFT(tmp11 - tmp14, CONST_BITS-PASS1_BITS);
3517	wsptr[8*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
3518	}
3519
3520	/* Pass 2: process 5 rows from work array, store into output array.
3521	* 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
3522	*/
3523	wsptr = workspace;
3524	for (ctr = 0; ctr < 5; ctr++) {
3525	outptr = output_buf[ctr] + output_col;
3526
3527	/* Even part */
3528
3529	/* Add fudge factor here for final descale. */
3530	z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3531	z3 <<= CONST_BITS;
3532	z4 = (INT32) wsptr[4];
3533	z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
3534	z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
3535	tmp10 = z3 + z1;
3536	tmp11 = z3 - z2;
3537
3538	tmp22 = z3 - ((z1 - z2) << 1); /* c0 = (c4-c8)2 /
3539
3540	z2 = (INT32) wsptr[2];
3541	z3 = (INT32) wsptr[6];
3542
3543	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
3544	tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
3545	tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
3546
3547	tmp20 = tmp10 + tmp12;
3548	tmp24 = tmp10 - tmp12;
3549	tmp21 = tmp11 + tmp13;
3550	tmp23 = tmp11 - tmp13;
3551
3552	/* Odd part */
3553
3554	z1 = (INT32) wsptr[1];
3555	z2 = (INT32) wsptr[3];
3556	z3 = (INT32) wsptr[5];
3557	z3 <<= CONST_BITS;
3558	z4 = (INT32) wsptr[7];
3559
3560	tmp11 = z2 + z4;
3561	tmp13 = z2 - z4;
3562
3563	tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
3564
3565	z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
3566	z4 = z3 + tmp12;
3567
3568	tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
3569	tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
3570
3571	z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
3572	z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));
3573
3574	tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;
3575
3576	tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
3577	tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
3578
3579	/* Final output stage */
3580
3581	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
3582	CONST_BITS+PASS1_BITS+3)
3583	& RANGE_MASK];
3584	outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
3585	CONST_BITS+PASS1_BITS+3)
3586	& RANGE_MASK];
3587	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
3588	CONST_BITS+PASS1_BITS+3)
3589	& RANGE_MASK];
3590	outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
3591	CONST_BITS+PASS1_BITS+3)
3592	& RANGE_MASK];
3593	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
3594	CONST_BITS+PASS1_BITS+3)
3595	& RANGE_MASK];
3596	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
3597	CONST_BITS+PASS1_BITS+3)
3598	& RANGE_MASK];
3599	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
3600	CONST_BITS+PASS1_BITS+3)
3601	& RANGE_MASK];
3602	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
3603	CONST_BITS+PASS1_BITS+3)
3604	& RANGE_MASK];
3605	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
3606	CONST_BITS+PASS1_BITS+3)
3607	& RANGE_MASK];
3608	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
3609	CONST_BITS+PASS1_BITS+3)
3610	& RANGE_MASK];
3611
3612	wsptr += 8; /* advance pointer to next row */
3613	}
3614	}
3615
3616
3617	/*
3618	* Perform dequantization and inverse DCT on one block of coefficients,
3619	* producing a 8x4 output block.
3620	*
3621	* 4-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
3622	*/
3623
3624	GLOBAL(void)
3625	jpeg_idct_8x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3626	JCOEFPTR coef_block,
3627	JSAMPARRAY output_buf, JDIMENSION output_col)
3628	{
3629	INT32 tmp0, tmp1, tmp2, tmp3;
3630	INT32 tmp10, tmp11, tmp12, tmp13;
3631	INT32 z1, z2, z3;
3632	JCOEFPTR inptr;
3633	ISLOW_MULT_TYPE * quantptr;
3634	int * wsptr;
3635	JSAMPROW outptr;
3636	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3637	int ctr;
3638	int workspace[84]; / buffers data between passes */
3639	SHIFT_TEMPS
3640
3641	/* Pass 1: process columns from input, store into work array.
3642	* 4-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
3643	*/
3644	inptr = coef_block;
3645	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3646	wsptr = workspace;
3647	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
3648	/* Even part */
3649
3650	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
3651	tmp2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
3652
3653	tmp10 = (tmp0 + tmp2) << PASS1_BITS;
3654	tmp12 = (tmp0 - tmp2) << PASS1_BITS;
3655
3656	/* Odd part */
3657	/* Same rotation as in the even part of the 8x8 LL&M IDCT */
3658
3659	z2 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
3660	z3 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
3661
3662	z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
3663	/* Add fudge factor here for final descale. */
3664	z1 += ONE << (CONST_BITS-PASS1_BITS-1);
3665	tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
3666	CONST_BITS-PASS1_BITS);
3667	tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
3668	CONST_BITS-PASS1_BITS);
3669
3670	/* Final output stage */
3671
3672	wsptr[8*0] = (int) (tmp10 + tmp0);
3673	wsptr[8*3] = (int) (tmp10 - tmp0);
3674	wsptr[8*1] = (int) (tmp12 + tmp2);
3675	wsptr[8*2] = (int) (tmp12 - tmp2);
3676	}
3677
3678	/* Pass 2: process rows from work array, store into output array. */
3679	/* Note that we must descale the results by a factor of 8 == 2*3, /
3680	/* and also undo the PASS1_BITS scaling. */
3681
3682	wsptr = workspace;
3683	for (ctr = 0; ctr < 4; ctr++) {
3684	outptr = output_buf[ctr] + output_col;
3685
3686	/* Even part: reverse the even part of the forward DCT. */
3687	/* The rotator is sqrt(2)c(-6). /
3688
3689	z2 = (INT32) wsptr[2];
3690	z3 = (INT32) wsptr[6];
3691
3692	z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
3693	tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
3694	tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
3695
3696	/* Add fudge factor here for final descale. */
3697	z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3698	z3 = (INT32) wsptr[4];
3699
3700	tmp0 = (z2 + z3) << CONST_BITS;
3701	tmp1 = (z2 - z3) << CONST_BITS;
3702
3703	tmp10 = tmp0 + tmp2;
3704	tmp13 = tmp0 - tmp2;
3705	tmp11 = tmp1 + tmp3;
3706	tmp12 = tmp1 - tmp3;
3707
3708	/* Odd part per figure 8; the matrix is unitary and hence its
3709	* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
3710	*/
3711
3712	tmp0 = (INT32) wsptr[7];
3713	tmp1 = (INT32) wsptr[5];
3714	tmp2 = (INT32) wsptr[3];
3715	tmp3 = (INT32) wsptr[1];
3716
3717	z2 = tmp0 + tmp2;
3718	z3 = tmp1 + tmp3;
3719
3720	z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
3721	z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
3722	z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
3723	z2 += z1;
3724	z3 += z1;
3725
3726	z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
3727	tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
3728	tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
3729	tmp0 += z1 + z2;
3730	tmp3 += z1 + z3;
3731
3732	z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
3733	tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
3734	tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
3735	tmp1 += z1 + z3;
3736	tmp2 += z1 + z2;
3737
3738	/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
3739
3740	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
3741	CONST_BITS+PASS1_BITS+3)
3742	& RANGE_MASK];
3743	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
3744	CONST_BITS+PASS1_BITS+3)
3745	& RANGE_MASK];
3746	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
3747	CONST_BITS+PASS1_BITS+3)
3748	& RANGE_MASK];
3749	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
3750	CONST_BITS+PASS1_BITS+3)
3751	& RANGE_MASK];
3752	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
3753	CONST_BITS+PASS1_BITS+3)
3754	& RANGE_MASK];
3755	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
3756	CONST_BITS+PASS1_BITS+3)
3757	& RANGE_MASK];
3758	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
3759	CONST_BITS+PASS1_BITS+3)
3760	& RANGE_MASK];
3761	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
3762	CONST_BITS+PASS1_BITS+3)
3763	& RANGE_MASK];
3764
3765	wsptr += DCTSIZE; /* advance pointer to next row */
3766	}
3767	}
3768
3769
3770	/*
3771	* Perform dequantization and inverse DCT on one block of coefficients,
3772	* producing a reduced-size 6x3 output block.
3773	*
3774	* 3-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
3775	*/
3776
3777	GLOBAL(void)
3778	jpeg_idct_6x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3779	JCOEFPTR coef_block,
3780	JSAMPARRAY output_buf, JDIMENSION output_col)
3781	{
3782	INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
3783	INT32 z1, z2, z3;
3784	JCOEFPTR inptr;
3785	ISLOW_MULT_TYPE * quantptr;
3786	int * wsptr;
3787	JSAMPROW outptr;
3788	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3789	int ctr;
3790	int workspace[63]; / buffers data between passes */
3791	SHIFT_TEMPS
3792
3793	/* Pass 1: process columns from input, store into work array.
3794	* 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
3795	*/
3796	inptr = coef_block;
3797	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3798	wsptr = workspace;
3799	for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
3800	/* Even part */
3801
3802	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
3803	tmp0 <<= CONST_BITS;
3804	/* Add fudge factor here for final descale. */
3805	tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
3806	tmp2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
3807	tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
3808	tmp10 = tmp0 + tmp12;
3809	tmp2 = tmp0 - tmp12 - tmp12;
3810
3811	/* Odd part */
3812
3813	tmp12 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
3814	tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
3815
3816	/* Final output stage */
3817
3818	wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
3819	wsptr[6*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
3820	wsptr[6*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
3821	}
3822
3823	/* Pass 2: process 3 rows from work array, store into output array.
3824	* 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
3825	*/
3826	wsptr = workspace;
3827	for (ctr = 0; ctr < 3; ctr++) {
3828	outptr = output_buf[ctr] + output_col;
3829
3830	/* Even part */
3831
3832	/* Add fudge factor here for final descale. */
3833	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
3834	tmp0 <<= CONST_BITS;
3835	tmp2 = (INT32) wsptr[4];
3836	tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
3837	tmp1 = tmp0 + tmp10;
3838	tmp11 = tmp0 - tmp10 - tmp10;
3839	tmp10 = (INT32) wsptr[2];
3840	tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
3841	tmp10 = tmp1 + tmp0;
3842	tmp12 = tmp1 - tmp0;
3843
3844	/* Odd part */
3845
3846	z1 = (INT32) wsptr[1];
3847	z2 = (INT32) wsptr[3];
3848	z3 = (INT32) wsptr[5];
3849	tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
3850	tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
3851	tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
3852	tmp1 = (z1 - z2 - z3) << CONST_BITS;
3853
3854	/* Final output stage */
3855
3856	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
3857	CONST_BITS+PASS1_BITS+3)
3858	& RANGE_MASK];
3859	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
3860	CONST_BITS+PASS1_BITS+3)
3861	& RANGE_MASK];
3862	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
3863	CONST_BITS+PASS1_BITS+3)
3864	& RANGE_MASK];
3865	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
3866	CONST_BITS+PASS1_BITS+3)
3867	& RANGE_MASK];
3868	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
3869	CONST_BITS+PASS1_BITS+3)
3870	& RANGE_MASK];
3871	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
3872	CONST_BITS+PASS1_BITS+3)
3873	& RANGE_MASK];
3874
3875	wsptr += 6; /* advance pointer to next row */
3876	}
3877	}
3878
3879
3880	/*
3881	* Perform dequantization and inverse DCT on one block of coefficients,
3882	* producing a 4x2 output block.
3883	*
3884	* 2-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
3885	*/
3886
3887	GLOBAL(void)
3888	jpeg_idct_4x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3889	JCOEFPTR coef_block,
3890	JSAMPARRAY output_buf, JDIMENSION output_col)
3891	{
3892	INT32 tmp0, tmp2, tmp10, tmp12;
3893	INT32 z1, z2, z3;
3894	JCOEFPTR inptr;
3895	ISLOW_MULT_TYPE * quantptr;
3896	INT32 * wsptr;
3897	JSAMPROW outptr;
3898	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3899	int ctr;
3900	INT32 workspace[42]; / buffers data between passes */
3901	SHIFT_TEMPS
3902
3903	/* Pass 1: process columns from input, store into work array. */
3904
3905	inptr = coef_block;
3906	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3907	wsptr = workspace;
3908	for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
3909	/* Even part */
3910
3911	tmp10 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
3912
3913	/* Odd part */
3914
3915	tmp0 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
3916
3917	/* Final output stage */
3918
3919	wsptr[4*0] = tmp10 + tmp0;
3920	wsptr[4*1] = tmp10 - tmp0;
3921	}
3922
3923	/* Pass 2: process 2 rows from work array, store into output array.
3924	* 4-point IDCT kernel,
3925	* cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
3926	*/
3927	wsptr = workspace;
3928	for (ctr = 0; ctr < 2; ctr++) {
3929	outptr = output_buf[ctr] + output_col;
3930
3931	/* Even part */
3932
3933	/* Add fudge factor here for final descale. */
3934	tmp0 = wsptr[0] + (ONE << 2);
3935	tmp2 = wsptr[2];
3936
3937	tmp10 = (tmp0 + tmp2) << CONST_BITS;
3938	tmp12 = (tmp0 - tmp2) << CONST_BITS;
3939
3940	/* Odd part */
3941	/* Same rotation as in the even part of the 8x8 LL&M IDCT */
3942
3943	z2 = wsptr[1];
3944	z3 = wsptr[3];
3945
3946	z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
3947	tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
3948	tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
3949
3950	/* Final output stage */
3951
3952	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
3953	CONST_BITS+3)
3954	& RANGE_MASK];
3955	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
3956	CONST_BITS+3)
3957	& RANGE_MASK];
3958	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
3959	CONST_BITS+3)
3960	& RANGE_MASK];
3961	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
3962	CONST_BITS+3)
3963	& RANGE_MASK];
3964
3965	wsptr += 4; /* advance pointer to next row */
3966	}
3967	}
3968
3969
3970	/*
3971	* Perform dequantization and inverse DCT on one block of coefficients,
3972	* producing a 2x1 output block.
3973	*
3974	* 1-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
3975	*/
3976
3977	GLOBAL(void)
3978	jpeg_idct_2x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
3979	JCOEFPTR coef_block,
3980	JSAMPARRAY output_buf, JDIMENSION output_col)
3981	{
3982	INT32 tmp0, tmp10;
3983	ISLOW_MULT_TYPE * quantptr;
3984	JSAMPROW outptr;
3985	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
3986	SHIFT_TEMPS
3987
3988	/* Pass 1: empty. */
3989
3990	/* Pass 2: process 1 row from input, store into output array. */
3991
3992	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
3993	outptr = output_buf[0] + output_col;
3994
3995	/* Even part */
3996
3997	tmp10 = DEQUANTIZE(coef_block[0], quantptr[0]);
3998	/* Add fudge factor here for final descale. */
3999	tmp10 += ONE << 2;
4000
4001	/* Odd part */
4002
4003	tmp0 = DEQUANTIZE(coef_block[1], quantptr[1]);
4004
4005	/* Final output stage */
4006
4007	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, 3) & RANGE_MASK];
4008	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, 3) & RANGE_MASK];
4009	}
4010
4011
4012	/*
4013	* Perform dequantization and inverse DCT on one block of coefficients,
4014	* producing a 8x16 output block.
4015	*
4016	* 16-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
4017	*/
4018
4019	GLOBAL(void)
4020	jpeg_idct_8x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4021	JCOEFPTR coef_block,
4022	JSAMPARRAY output_buf, JDIMENSION output_col)
4023	{
4024	INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
4025	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
4026	INT32 z1, z2, z3, z4;
4027	JCOEFPTR inptr;
4028	ISLOW_MULT_TYPE * quantptr;
4029	int * wsptr;
4030	JSAMPROW outptr;
4031	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4032	int ctr;
4033	int workspace[816]; / buffers data between passes */
4034	SHIFT_TEMPS
4035
4036	/* Pass 1: process columns from input, store into work array.
4037	* 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
4038	*/
4039	inptr = coef_block;
4040	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4041	wsptr = workspace;
4042	for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
4043	/* Even part */
4044
4045	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
4046	tmp0 <<= CONST_BITS;
4047	/* Add fudge factor here for final descale. */
4048	tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
4049
4050	z1 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
4051	tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
4052	tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
4053
4054	tmp10 = tmp0 + tmp1;
4055	tmp11 = tmp0 - tmp1;
4056	tmp12 = tmp0 + tmp2;
4057	tmp13 = tmp0 - tmp2;
4058
4059	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
4060	z2 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
4061	z3 = z1 - z2;
4062	z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
4063	z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
4064
4065	tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
4066	tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
4067	tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
4068	tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
4069
4070	tmp20 = tmp10 + tmp0;
4071	tmp27 = tmp10 - tmp0;
4072	tmp21 = tmp12 + tmp1;
4073	tmp26 = tmp12 - tmp1;
4074	tmp22 = tmp13 + tmp2;
4075	tmp25 = tmp13 - tmp2;
4076	tmp23 = tmp11 + tmp3;
4077	tmp24 = tmp11 - tmp3;
4078
4079	/* Odd part */
4080
4081	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
4082	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
4083	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
4084	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
4085
4086	tmp11 = z1 + z3;
4087
4088	tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
4089	tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
4090	tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
4091	tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
4092	tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
4093	tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
4094	tmp0 = tmp1 + tmp2 + tmp3 -
4095	MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
4096	tmp13 = tmp10 + tmp11 + tmp12 -
4097	MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
4098	z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
4099	tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
4100	tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
4101	z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
4102	tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
4103	tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
4104	z2 += z4;
4105	z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
4106	tmp1 += z1;
4107	tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
4108	z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
4109	tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
4110	tmp12 += z2;
4111	z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
4112	tmp2 += z2;
4113	tmp3 += z2;
4114	z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
4115	tmp10 += z2;
4116	tmp11 += z2;
4117
4118	/* Final output stage */
4119
4120	wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS-PASS1_BITS);
4121	wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS-PASS1_BITS);
4122	wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS-PASS1_BITS);
4123	wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS-PASS1_BITS);
4124	wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS-PASS1_BITS);
4125	wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS-PASS1_BITS);
4126	wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS-PASS1_BITS);
4127	wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS-PASS1_BITS);
4128	wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
4129	wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
4130	wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
4131	wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
4132	wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
4133	wsptr[8*9] = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
4134	wsptr[8*7] = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
4135	wsptr[8*8] = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
4136	}
4137
4138	/* Pass 2: process rows from work array, store into output array. */
4139	/* Note that we must descale the results by a factor of 8 == 2*3, /
4140	/* and also undo the PASS1_BITS scaling. */
4141
4142	wsptr = workspace;
4143	for (ctr = 0; ctr < 16; ctr++) {
4144	outptr = output_buf[ctr] + output_col;
4145
4146	/* Even part: reverse the even part of the forward DCT. */
4147	/* The rotator is sqrt(2)c(-6). /
4148
4149	z2 = (INT32) wsptr[2];
4150	z3 = (INT32) wsptr[6];
4151
4152	z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
4153	tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
4154	tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
4155
4156	/* Add fudge factor here for final descale. */
4157	z2 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4158	z3 = (INT32) wsptr[4];
4159
4160	tmp0 = (z2 + z3) << CONST_BITS;
4161	tmp1 = (z2 - z3) << CONST_BITS;
4162
4163	tmp10 = tmp0 + tmp2;
4164	tmp13 = tmp0 - tmp2;
4165	tmp11 = tmp1 + tmp3;
4166	tmp12 = tmp1 - tmp3;
4167
4168	/* Odd part per figure 8; the matrix is unitary and hence its
4169	* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
4170	*/
4171
4172	tmp0 = (INT32) wsptr[7];
4173	tmp1 = (INT32) wsptr[5];
4174	tmp2 = (INT32) wsptr[3];
4175	tmp3 = (INT32) wsptr[1];
4176
4177	z2 = tmp0 + tmp2;
4178	z3 = tmp1 + tmp3;
4179
4180	z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
4181	z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
4182	z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
4183	z2 += z1;
4184	z3 += z1;
4185
4186	z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
4187	tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
4188	tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
4189	tmp0 += z1 + z2;
4190	tmp3 += z1 + z3;
4191
4192	z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
4193	tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
4194	tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
4195	tmp1 += z1 + z3;
4196	tmp2 += z1 + z2;
4197
4198	/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
4199
4200	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
4201	CONST_BITS+PASS1_BITS+3)
4202	& RANGE_MASK];
4203	outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
4204	CONST_BITS+PASS1_BITS+3)
4205	& RANGE_MASK];
4206	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
4207	CONST_BITS+PASS1_BITS+3)
4208	& RANGE_MASK];
4209	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
4210	CONST_BITS+PASS1_BITS+3)
4211	& RANGE_MASK];
4212	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
4213	CONST_BITS+PASS1_BITS+3)
4214	& RANGE_MASK];
4215	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
4216	CONST_BITS+PASS1_BITS+3)
4217	& RANGE_MASK];
4218	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
4219	CONST_BITS+PASS1_BITS+3)
4220	& RANGE_MASK];
4221	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
4222	CONST_BITS+PASS1_BITS+3)
4223	& RANGE_MASK];
4224
4225	wsptr += DCTSIZE; /* advance pointer to next row */
4226	}
4227	}
4228
4229
4230	/*
4231	* Perform dequantization and inverse DCT on one block of coefficients,
4232	* producing a 7x14 output block.
4233	*
4234	* 14-point IDCT in pass 1 (columns), 7-point in pass 2 (rows).
4235	*/
4236
4237	GLOBAL(void)
4238	jpeg_idct_7x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4239	JCOEFPTR coef_block,
4240	JSAMPARRAY output_buf, JDIMENSION output_col)
4241	{
4242	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
4243	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
4244	INT32 z1, z2, z3, z4;
4245	JCOEFPTR inptr;
4246	ISLOW_MULT_TYPE * quantptr;
4247	int * wsptr;
4248	JSAMPROW outptr;
4249	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4250	int ctr;
4251	int workspace[714]; / buffers data between passes */
4252	SHIFT_TEMPS
4253
4254	/* Pass 1: process columns from input, store into work array.
4255	* 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
4256	*/
4257	inptr = coef_block;
4258	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4259	wsptr = workspace;
4260	for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
4261	/* Even part */
4262
4263	z1 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
4264	z1 <<= CONST_BITS;
4265	/* Add fudge factor here for final descale. */
4266	z1 += ONE << (CONST_BITS-PASS1_BITS-1);
4267	z4 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
4268	z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
4269	z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
4270	z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
4271
4272	tmp10 = z1 + z2;
4273	tmp11 = z1 + z3;
4274	tmp12 = z1 - z4;
4275
4276	tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)2 /
4277	CONST_BITS-PASS1_BITS);
4278
4279	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
4280	z2 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
4281
4282	z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
4283
4284	tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
4285	tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
4286	tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
4287	MULTIPLY(z2, FIX(1.378756276)); /* c2 */
4288
4289	tmp20 = tmp10 + tmp13;
4290	tmp26 = tmp10 - tmp13;
4291	tmp21 = tmp11 + tmp14;
4292	tmp25 = tmp11 - tmp14;
4293	tmp22 = tmp12 + tmp15;
4294	tmp24 = tmp12 - tmp15;
4295
4296	/* Odd part */
4297
4298	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
4299	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
4300	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
4301	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
4302	tmp13 = z4 << CONST_BITS;
4303
4304	tmp14 = z1 + z3;
4305	tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
4306	tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
4307	tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
4308	tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
4309	tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
4310	z1 -= z2;
4311	tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */
4312	tmp16 += tmp15;
4313	z1 += z4;
4314	z4 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
4315	tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
4316	tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
4317	z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
4318	tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
4319	tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
4320
4321	tmp13 = (z1 - z3) << PASS1_BITS;
4322
4323	/* Final output stage */
4324
4325	wsptr[7*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
4326	wsptr[7*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
4327	wsptr[7*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
4328	wsptr[7*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
4329	wsptr[7*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
4330	wsptr[7*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
4331	wsptr[7*3] = (int) (tmp23 + tmp13);
4332	wsptr[7*10] = (int) (tmp23 - tmp13);
4333	wsptr[7*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
4334	wsptr[7*9] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
4335	wsptr[7*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
4336	wsptr[7*8] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
4337	wsptr[7*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
4338	wsptr[7*7] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
4339	}
4340
4341	/* Pass 2: process 14 rows from work array, store into output array.
4342	* 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
4343	*/
4344	wsptr = workspace;
4345	for (ctr = 0; ctr < 14; ctr++) {
4346	outptr = output_buf[ctr] + output_col;
4347
4348	/* Even part */
4349
4350	/* Add fudge factor here for final descale. */
4351	tmp23 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4352	tmp23 <<= CONST_BITS;
4353
4354	z1 = (INT32) wsptr[2];
4355	z2 = (INT32) wsptr[4];
4356	z3 = (INT32) wsptr[6];
4357
4358	tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
4359	tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
4360	tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
4361	tmp10 = z1 + z3;
4362	z2 -= tmp10;
4363	tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
4364	tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
4365	tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
4366	tmp23 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
4367
4368	/* Odd part */
4369
4370	z1 = (INT32) wsptr[1];
4371	z2 = (INT32) wsptr[3];
4372	z3 = (INT32) wsptr[5];
4373
4374	tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
4375	tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
4376	tmp10 = tmp11 - tmp12;
4377	tmp11 += tmp12;
4378	tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
4379	tmp11 += tmp12;
4380	z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
4381	tmp10 += z2;
4382	tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
4383
4384	/* Final output stage */
4385
4386	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
4387	CONST_BITS+PASS1_BITS+3)
4388	& RANGE_MASK];
4389	outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
4390	CONST_BITS+PASS1_BITS+3)
4391	& RANGE_MASK];
4392	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
4393	CONST_BITS+PASS1_BITS+3)
4394	& RANGE_MASK];
4395	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
4396	CONST_BITS+PASS1_BITS+3)
4397	& RANGE_MASK];
4398	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
4399	CONST_BITS+PASS1_BITS+3)
4400	& RANGE_MASK];
4401	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
4402	CONST_BITS+PASS1_BITS+3)
4403	& RANGE_MASK];
4404	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23,
4405	CONST_BITS+PASS1_BITS+3)
4406	& RANGE_MASK];
4407
4408	wsptr += 7; /* advance pointer to next row */
4409	}
4410	}
4411
4412
4413	/*
4414	* Perform dequantization and inverse DCT on one block of coefficients,
4415	* producing a 6x12 output block.
4416	*
4417	* 12-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
4418	*/
4419
4420	GLOBAL(void)
4421	jpeg_idct_6x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4422	JCOEFPTR coef_block,
4423	JSAMPARRAY output_buf, JDIMENSION output_col)
4424	{
4425	INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
4426	INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
4427	INT32 z1, z2, z3, z4;
4428	JCOEFPTR inptr;
4429	ISLOW_MULT_TYPE * quantptr;
4430	int * wsptr;
4431	JSAMPROW outptr;
4432	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4433	int ctr;
4434	int workspace[612]; / buffers data between passes */
4435	SHIFT_TEMPS
4436
4437	/* Pass 1: process columns from input, store into work array.
4438	* 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
4439	*/
4440	inptr = coef_block;
4441	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4442	wsptr = workspace;
4443	for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
4444	/* Even part */
4445
4446	z3 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
4447	z3 <<= CONST_BITS;
4448	/* Add fudge factor here for final descale. */
4449	z3 += ONE << (CONST_BITS-PASS1_BITS-1);
4450
4451	z4 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
4452	z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
4453
4454	tmp10 = z3 + z4;
4455	tmp11 = z3 - z4;
4456
4457	z1 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
4458	z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
4459	z1 <<= CONST_BITS;
4460	z2 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
4461	z2 <<= CONST_BITS;
4462
4463	tmp12 = z1 - z2;
4464
4465	tmp21 = z3 + tmp12;
4466	tmp24 = z3 - tmp12;
4467
4468	tmp12 = z4 + z2;
4469
4470	tmp20 = tmp10 + tmp12;
4471	tmp25 = tmp10 - tmp12;
4472
4473	tmp12 = z4 - z1 - z2;
4474
4475	tmp22 = tmp11 + tmp12;
4476	tmp23 = tmp11 - tmp12;
4477
4478	/* Odd part */
4479
4480	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
4481	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
4482	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
4483	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
4484
4485	tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
4486	tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
4487
4488	tmp10 = z1 + z3;
4489	tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
4490	tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
4491	tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
4492	tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
4493	tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
4494	tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
4495	tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
4496	MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
4497
4498	z1 -= z4;
4499	z2 -= z3;
4500	z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
4501	tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
4502	tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
4503
4504	/* Final output stage */
4505
4506	wsptr[6*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
4507	wsptr[6*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
4508	wsptr[6*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
4509	wsptr[6*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
4510	wsptr[6*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
4511	wsptr[6*9] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
4512	wsptr[6*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
4513	wsptr[6*8] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
4514	wsptr[6*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
4515	wsptr[6*7] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
4516	wsptr[6*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
4517	wsptr[6*6] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
4518	}
4519
4520	/* Pass 2: process 12 rows from work array, store into output array.
4521	* 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
4522	*/
4523	wsptr = workspace;
4524	for (ctr = 0; ctr < 12; ctr++) {
4525	outptr = output_buf[ctr] + output_col;
4526
4527	/* Even part */
4528
4529	/* Add fudge factor here for final descale. */
4530	tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4531	tmp10 <<= CONST_BITS;
4532	tmp12 = (INT32) wsptr[4];
4533	tmp20 = MULTIPLY(tmp12, FIX(0.707106781)); /* c4 */
4534	tmp11 = tmp10 + tmp20;
4535	tmp21 = tmp10 - tmp20 - tmp20;
4536	tmp20 = (INT32) wsptr[2];
4537	tmp10 = MULTIPLY(tmp20, FIX(1.224744871)); /* c2 */
4538	tmp20 = tmp11 + tmp10;
4539	tmp22 = tmp11 - tmp10;
4540
4541	/* Odd part */
4542
4543	z1 = (INT32) wsptr[1];
4544	z2 = (INT32) wsptr[3];
4545	z3 = (INT32) wsptr[5];
4546	tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
4547	tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
4548	tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
4549	tmp11 = (z1 - z2 - z3) << CONST_BITS;
4550
4551	/* Final output stage */
4552
4553	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
4554	CONST_BITS+PASS1_BITS+3)
4555	& RANGE_MASK];
4556	outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
4557	CONST_BITS+PASS1_BITS+3)
4558	& RANGE_MASK];
4559	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
4560	CONST_BITS+PASS1_BITS+3)
4561	& RANGE_MASK];
4562	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
4563	CONST_BITS+PASS1_BITS+3)
4564	& RANGE_MASK];
4565	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
4566	CONST_BITS+PASS1_BITS+3)
4567	& RANGE_MASK];
4568	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
4569	CONST_BITS+PASS1_BITS+3)
4570	& RANGE_MASK];
4571
4572	wsptr += 6; /* advance pointer to next row */
4573	}
4574	}
4575
4576
4577	/*
4578	* Perform dequantization and inverse DCT on one block of coefficients,
4579	* producing a 5x10 output block.
4580	*
4581	* 10-point IDCT in pass 1 (columns), 5-point in pass 2 (rows).
4582	*/
4583
4584	GLOBAL(void)
4585	jpeg_idct_5x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4586	JCOEFPTR coef_block,
4587	JSAMPARRAY output_buf, JDIMENSION output_col)
4588	{
4589	INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
4590	INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
4591	INT32 z1, z2, z3, z4, z5;
4592	JCOEFPTR inptr;
4593	ISLOW_MULT_TYPE * quantptr;
4594	int * wsptr;
4595	JSAMPROW outptr;
4596	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4597	int ctr;
4598	int workspace[510]; / buffers data between passes */
4599	SHIFT_TEMPS
4600
4601	/* Pass 1: process columns from input, store into work array.
4602	* 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
4603	*/
4604	inptr = coef_block;
4605	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4606	wsptr = workspace;
4607	for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
4608	/* Even part */
4609
4610	z3 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
4611	z3 <<= CONST_BITS;
4612	/* Add fudge factor here for final descale. */
4613	z3 += ONE << (CONST_BITS-PASS1_BITS-1);
4614	z4 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
4615	z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
4616	z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
4617	tmp10 = z3 + z1;
4618	tmp11 = z3 - z2;
4619
4620	tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1), /* c0 = (c4-c8)2 /
4621	CONST_BITS-PASS1_BITS);
4622
4623	z2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
4624	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
4625
4626	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
4627	tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
4628	tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
4629
4630	tmp20 = tmp10 + tmp12;
4631	tmp24 = tmp10 - tmp12;
4632	tmp21 = tmp11 + tmp13;
4633	tmp23 = tmp11 - tmp13;
4634
4635	/* Odd part */
4636
4637	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
4638	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
4639	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
4640	z4 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
4641
4642	tmp11 = z2 + z4;
4643	tmp13 = z2 - z4;
4644
4645	tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
4646	z5 = z3 << CONST_BITS;
4647
4648	z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
4649	z4 = z5 + tmp12;
4650
4651	tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
4652	tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
4653
4654	z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
4655	z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));
4656
4657	tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;
4658
4659	tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
4660	tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
4661
4662	/* Final output stage */
4663
4664	wsptr[5*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
4665	wsptr[5*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
4666	wsptr[5*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
4667	wsptr[5*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
4668	wsptr[5*2] = (int) (tmp22 + tmp12);
4669	wsptr[5*7] = (int) (tmp22 - tmp12);
4670	wsptr[5*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
4671	wsptr[5*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
4672	wsptr[5*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
4673	wsptr[5*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
4674	}
4675
4676	/* Pass 2: process 10 rows from work array, store into output array.
4677	* 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
4678	*/
4679	wsptr = workspace;
4680	for (ctr = 0; ctr < 10; ctr++) {
4681	outptr = output_buf[ctr] + output_col;
4682
4683	/* Even part */
4684
4685	/* Add fudge factor here for final descale. */
4686	tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4687	tmp12 <<= CONST_BITS;
4688	tmp13 = (INT32) wsptr[2];
4689	tmp14 = (INT32) wsptr[4];
4690	z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
4691	z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
4692	z3 = tmp12 + z2;
4693	tmp10 = z3 + z1;
4694	tmp11 = z3 - z1;
4695	tmp12 -= z2 << 2;
4696
4697	/* Odd part */
4698
4699	z2 = (INT32) wsptr[1];
4700	z3 = (INT32) wsptr[3];
4701
4702	z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
4703	tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
4704	tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
4705
4706	/* Final output stage */
4707
4708	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp13,
4709	CONST_BITS+PASS1_BITS+3)
4710	& RANGE_MASK];
4711	outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp13,
4712	CONST_BITS+PASS1_BITS+3)
4713	& RANGE_MASK];
4714	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp14,
4715	CONST_BITS+PASS1_BITS+3)
4716	& RANGE_MASK];
4717	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp14,
4718	CONST_BITS+PASS1_BITS+3)
4719	& RANGE_MASK];
4720	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
4721	CONST_BITS+PASS1_BITS+3)
4722	& RANGE_MASK];
4723
4724	wsptr += 5; /* advance pointer to next row */
4725	}
4726	}
4727
4728
4729	/*
4730	* Perform dequantization and inverse DCT on one block of coefficients,
4731	* producing a 4x8 output block.
4732	*
4733	* 8-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
4734	*/
4735
4736	GLOBAL(void)
4737	jpeg_idct_4x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4738	JCOEFPTR coef_block,
4739	JSAMPARRAY output_buf, JDIMENSION output_col)
4740	{
4741	INT32 tmp0, tmp1, tmp2, tmp3;
4742	INT32 tmp10, tmp11, tmp12, tmp13;
4743	INT32 z1, z2, z3;
4744	JCOEFPTR inptr;
4745	ISLOW_MULT_TYPE * quantptr;
4746	int * wsptr;
4747	JSAMPROW outptr;
4748	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4749	int ctr;
4750	int workspace[48]; / buffers data between passes */
4751	SHIFT_TEMPS
4752
4753	/* Pass 1: process columns from input, store into work array. */
4754	/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
4755	/* furthermore, we scale the results by 2*PASS1_BITS. /
4756
4757	inptr = coef_block;
4758	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4759	wsptr = workspace;
4760	for (ctr = 4; ctr > 0; ctr--) {
4761	/* Due to quantization, we will usually find that many of the input
4762	* coefficients are zero, especially the AC terms. We can exploit this
4763	* by short-circuiting the IDCT calculation for any column in which all
4764	* the AC terms are zero. In that case each output is equal to the
4765	* DC coefficient (with scale factor as needed).
4766	* With typical images and quantization tables, half or more of the
4767	* column DCT calculations can be simplified this way.
4768	*/
4769
4770	if (inptr[DCTSIZE1] == 0 && inptr[DCTSIZE2] == 0 &&
4771	inptr[DCTSIZE3] == 0 && inptr[DCTSIZE4] == 0 &&
4772	inptr[DCTSIZE5] == 0 && inptr[DCTSIZE6] == 0 &&
4773	inptr[DCTSIZE*7] == 0) {
4774	/* AC terms all zero */
4775	int dcval = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]) << PASS1_BITS;
4776
4777	wsptr[4*0] = dcval;
4778	wsptr[4*1] = dcval;
4779	wsptr[4*2] = dcval;
4780	wsptr[4*3] = dcval;
4781	wsptr[4*4] = dcval;
4782	wsptr[4*5] = dcval;
4783	wsptr[4*6] = dcval;
4784	wsptr[4*7] = dcval;
4785
4786	inptr++; /* advance pointers to next column */
4787	quantptr++;
4788	wsptr++;
4789	continue;
4790	}
4791
4792	/* Even part: reverse the even part of the forward DCT. */
4793	/* The rotator is sqrt(2)c(-6). /
4794
4795	z2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
4796	z3 = DEQUANTIZE(inptr[DCTSIZE6], quantptr[DCTSIZE6]);
4797
4798	z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
4799	tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);
4800	tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);
4801
4802	z2 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
4803	z3 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
4804	z2 <<= CONST_BITS;
4805	z3 <<= CONST_BITS;
4806	/* Add fudge factor here for final descale. */
4807	z2 += ONE << (CONST_BITS-PASS1_BITS-1);
4808
4809	tmp0 = z2 + z3;
4810	tmp1 = z2 - z3;
4811
4812	tmp10 = tmp0 + tmp2;
4813	tmp13 = tmp0 - tmp2;
4814	tmp11 = tmp1 + tmp3;
4815	tmp12 = tmp1 - tmp3;
4816
4817	/* Odd part per figure 8; the matrix is unitary and hence its
4818	* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
4819	*/
4820
4821	tmp0 = DEQUANTIZE(inptr[DCTSIZE7], quantptr[DCTSIZE7]);
4822	tmp1 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
4823	tmp2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
4824	tmp3 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
4825
4826	z2 = tmp0 + tmp2;
4827	z3 = tmp1 + tmp3;
4828
4829	z1 = MULTIPLY(z2 + z3, FIX_1_175875602); /* sqrt(2) * c3 */
4830	z2 = MULTIPLY(z2, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
4831	z3 = MULTIPLY(z3, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
4832	z2 += z1;
4833	z3 += z1;
4834
4835	z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
4836	tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
4837	tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
4838	tmp0 += z1 + z2;
4839	tmp3 += z1 + z3;
4840
4841	z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
4842	tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
4843	tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
4844	tmp1 += z1 + z3;
4845	tmp2 += z1 + z2;
4846
4847	/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
4848
4849	wsptr[4*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
4850	wsptr[4*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
4851	wsptr[4*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
4852	wsptr[4*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
4853	wsptr[4*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
4854	wsptr[4*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
4855	wsptr[4*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
4856	wsptr[4*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
4857
4858	inptr++; /* advance pointers to next column */
4859	quantptr++;
4860	wsptr++;
4861	}
4862
4863	/* Pass 2: process 8 rows from work array, store into output array.
4864	* 4-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
4865	*/
4866	wsptr = workspace;
4867	for (ctr = 0; ctr < 8; ctr++) {
4868	outptr = output_buf[ctr] + output_col;
4869
4870	/* Even part */
4871
4872	/* Add fudge factor here for final descale. */
4873	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4874	tmp2 = (INT32) wsptr[2];
4875
4876	tmp10 = (tmp0 + tmp2) << CONST_BITS;
4877	tmp12 = (tmp0 - tmp2) << CONST_BITS;
4878
4879	/* Odd part */
4880	/* Same rotation as in the even part of the 8x8 LL&M IDCT */
4881
4882	z2 = (INT32) wsptr[1];
4883	z3 = (INT32) wsptr[3];
4884
4885	z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
4886	tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
4887	tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
4888
4889	/* Final output stage */
4890
4891	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
4892	CONST_BITS+PASS1_BITS+3)
4893	& RANGE_MASK];
4894	outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
4895	CONST_BITS+PASS1_BITS+3)
4896	& RANGE_MASK];
4897	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
4898	CONST_BITS+PASS1_BITS+3)
4899	& RANGE_MASK];
4900	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
4901	CONST_BITS+PASS1_BITS+3)
4902	& RANGE_MASK];
4903
4904	wsptr += 4; /* advance pointer to next row */
4905	}
4906	}
4907
4908
4909	/*
4910	* Perform dequantization and inverse DCT on one block of coefficients,
4911	* producing a reduced-size 3x6 output block.
4912	*
4913	* 6-point IDCT in pass 1 (columns), 3-point in pass 2 (rows).
4914	*/
4915
4916	GLOBAL(void)
4917	jpeg_idct_3x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
4918	JCOEFPTR coef_block,
4919	JSAMPARRAY output_buf, JDIMENSION output_col)
4920	{
4921	INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
4922	INT32 z1, z2, z3;
4923	JCOEFPTR inptr;
4924	ISLOW_MULT_TYPE * quantptr;
4925	int * wsptr;
4926	JSAMPROW outptr;
4927	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
4928	int ctr;
4929	int workspace[36]; / buffers data between passes */
4930	SHIFT_TEMPS
4931
4932	/* Pass 1: process columns from input, store into work array.
4933	* 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
4934	*/
4935	inptr = coef_block;
4936	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
4937	wsptr = workspace;
4938	for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
4939	/* Even part */
4940
4941	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
4942	tmp0 <<= CONST_BITS;
4943	/* Add fudge factor here for final descale. */
4944	tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
4945	tmp2 = DEQUANTIZE(inptr[DCTSIZE4], quantptr[DCTSIZE4]);
4946	tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
4947	tmp1 = tmp0 + tmp10;
4948	tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
4949	tmp10 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
4950	tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
4951	tmp10 = tmp1 + tmp0;
4952	tmp12 = tmp1 - tmp0;
4953
4954	/* Odd part */
4955
4956	z1 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
4957	z2 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
4958	z3 = DEQUANTIZE(inptr[DCTSIZE5], quantptr[DCTSIZE5]);
4959	tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
4960	tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
4961	tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
4962	tmp1 = (z1 - z2 - z3) << PASS1_BITS;
4963
4964	/* Final output stage */
4965
4966	wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
4967	wsptr[3*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
4968	wsptr[3*1] = (int) (tmp11 + tmp1);
4969	wsptr[3*4] = (int) (tmp11 - tmp1);
4970	wsptr[3*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
4971	wsptr[3*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
4972	}
4973
4974	/* Pass 2: process 6 rows from work array, store into output array.
4975	* 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
4976	*/
4977	wsptr = workspace;
4978	for (ctr = 0; ctr < 6; ctr++) {
4979	outptr = output_buf[ctr] + output_col;
4980
4981	/* Even part */
4982
4983	/* Add fudge factor here for final descale. */
4984	tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
4985	tmp0 <<= CONST_BITS;
4986	tmp2 = (INT32) wsptr[2];
4987	tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
4988	tmp10 = tmp0 + tmp12;
4989	tmp2 = tmp0 - tmp12 - tmp12;
4990
4991	/* Odd part */
4992
4993	tmp12 = (INT32) wsptr[1];
4994	tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
4995
4996	/* Final output stage */
4997
4998	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
4999	CONST_BITS+PASS1_BITS+3)
5000	& RANGE_MASK];
5001	outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
5002	CONST_BITS+PASS1_BITS+3)
5003	& RANGE_MASK];
5004	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
5005	CONST_BITS+PASS1_BITS+3)
5006	& RANGE_MASK];
5007
5008	wsptr += 3; /* advance pointer to next row */
5009	}
5010	}
5011
5012
5013	/*
5014	* Perform dequantization and inverse DCT on one block of coefficients,
5015	* producing a 2x4 output block.
5016	*
5017	* 4-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
5018	*/
5019
5020	GLOBAL(void)
5021	jpeg_idct_2x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
5022	JCOEFPTR coef_block,
5023	JSAMPARRAY output_buf, JDIMENSION output_col)
5024	{
5025	INT32 tmp0, tmp2, tmp10, tmp12;
5026	INT32 z1, z2, z3;
5027	JCOEFPTR inptr;
5028	ISLOW_MULT_TYPE * quantptr;
5029	INT32 * wsptr;
5030	JSAMPROW outptr;
5031	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
5032	int ctr;
5033	INT32 workspace[24]; / buffers data between passes */
5034	SHIFT_TEMPS
5035
5036	/* Pass 1: process columns from input, store into work array.
5037	* 4-point IDCT kernel,
5038	* cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
5039	*/
5040	inptr = coef_block;
5041	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
5042	wsptr = workspace;
5043	for (ctr = 0; ctr < 2; ctr++, inptr++, quantptr++, wsptr++) {
5044	/* Even part */
5045
5046	tmp0 = DEQUANTIZE(inptr[DCTSIZE0], quantptr[DCTSIZE0]);
5047	tmp2 = DEQUANTIZE(inptr[DCTSIZE2], quantptr[DCTSIZE2]);
5048
5049	tmp10 = (tmp0 + tmp2) << CONST_BITS;
5050	tmp12 = (tmp0 - tmp2) << CONST_BITS;
5051
5052	/* Odd part */
5053	/* Same rotation as in the even part of the 8x8 LL&M IDCT */
5054
5055	z2 = DEQUANTIZE(inptr[DCTSIZE1], quantptr[DCTSIZE1]);
5056	z3 = DEQUANTIZE(inptr[DCTSIZE3], quantptr[DCTSIZE3]);
5057
5058	z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */
5059	tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
5060	tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */
5061
5062	/* Final output stage */
5063
5064	wsptr[2*0] = tmp10 + tmp0;
5065	wsptr[2*3] = tmp10 - tmp0;
5066	wsptr[2*1] = tmp12 + tmp2;
5067	wsptr[2*2] = tmp12 - tmp2;
5068	}
5069
5070	/* Pass 2: process 4 rows from work array, store into output array. */
5071
5072	wsptr = workspace;
5073	for (ctr = 0; ctr < 4; ctr++) {
5074	outptr = output_buf[ctr] + output_col;
5075
5076	/* Even part */
5077
5078	/* Add fudge factor here for final descale. */
5079	tmp10 = wsptr[0] + (ONE << (CONST_BITS+2));
5080
5081	/* Odd part */
5082
5083	tmp0 = wsptr[1];
5084
5085	/* Final output stage */
5086
5087	outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS+3)
5088	& RANGE_MASK];
5089	outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS+3)
5090	& RANGE_MASK];
5091
5092	wsptr += 2; /* advance pointer to next row */
5093	}
5094	}
5095
5096
5097	/*
5098	* Perform dequantization and inverse DCT on one block of coefficients,
5099	* producing a 1x2 output block.
5100	*
5101	* 2-point IDCT in pass 1 (columns), 1-point in pass 2 (rows).
5102	*/
5103
5104	GLOBAL(void)
5105	jpeg_idct_1x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
5106	JCOEFPTR coef_block,
5107	JSAMPARRAY output_buf, JDIMENSION output_col)
5108	{
5109	INT32 tmp0, tmp10;
5110	ISLOW_MULT_TYPE * quantptr;
5111	JSAMPLE *range_limit = IDCT_range_limit(cinfo);
5112	SHIFT_TEMPS
5113
5114	/* Process 1 column from input, store into output array. */
5115
5116	quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
5117
5118	/* Even part */
5119
5120	tmp10 = DEQUANTIZE(coef_block[DCTSIZE0], quantptr[DCTSIZE0]);
5121	/* Add fudge factor here for final descale. */
5122	tmp10 += ONE << 2;
5123
5124	/* Odd part */
5125
5126	tmp0 = DEQUANTIZE(coef_block[DCTSIZE1], quantptr[DCTSIZE1]);
5127
5128	/* Final output stage */
5129
5130	output_buf[0][output_col] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, 3)
5131	& RANGE_MASK];
5132	output_buf[1][output_col] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, 3)
5133	& RANGE_MASK];
5134	}
5135
5136	#endif /* IDCT_SCALING_SUPPORTED */
5137	#endif /* DCT_ISLOW_SUPPORTED */

Note: See TracBrowser for help on using the repository browser.

Download in other formats: