/src/freeimage-svn/FreeImage/trunk/Source/LibJPEG/jfdctint.c
Line | Count | Source |
1 | | /* |
2 | | * jfdctint.c |
3 | | * |
4 | | * Copyright (C) 1991-1996, Thomas G. Lane. |
5 | | * Modification developed 2003-2018 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 | | * forward DCT (Discrete Cosine Transform). |
11 | | * |
12 | | * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT |
13 | | * on each column. Direct algorithms are also available, but they are |
14 | | * much more complex and seem not to be any faster when reduced to code. |
15 | | * |
16 | | * This implementation is based on an algorithm described in |
17 | | * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT |
18 | | * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics, |
19 | | * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991. |
20 | | * The primary algorithm described there uses 11 multiplies and 29 adds. |
21 | | * We use their alternate method with 12 multiplies and 32 adds. |
22 | | * The advantage of this method is that no data path contains more than one |
23 | | * multiplication; this allows a very simple and accurate implementation in |
24 | | * scaled fixed-point arithmetic, with a minimal number of shifts. |
25 | | * |
26 | | * We also provide FDCT routines with various input sample block sizes for |
27 | | * direct resolution reduction or enlargement and for direct resolving the |
28 | | * common 2x1 and 1x2 subsampling cases without additional resampling: NxN |
29 | | * (N=1...16), 2NxN, and Nx2N (N=1...8) pixels for one 8x8 output DCT block. |
30 | | * |
31 | | * For N<8 we fill the remaining block coefficients with zero. |
32 | | * For N>8 we apply a partial N-point FDCT on the input samples, computing |
33 | | * just the lower 8 frequency coefficients and discarding the rest. |
34 | | * |
35 | | * We must scale the output coefficients of the N-point FDCT appropriately |
36 | | * to the standard 8-point FDCT level by 8/N per 1-D pass. This scaling |
37 | | * is folded into the constant multipliers (pass 2) and/or final/initial |
38 | | * shifting. |
39 | | * |
40 | | * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases |
41 | | * since there would be too many additional constants to pre-calculate. |
42 | | */ |
43 | | |
44 | | #define JPEG_INTERNALS |
45 | | #include "jinclude.h" |
46 | | #include "jpeglib.h" |
47 | | #include "jdct.h" /* Private declarations for DCT subsystem */ |
48 | | |
49 | | #ifdef DCT_ISLOW_SUPPORTED |
50 | | |
51 | | |
52 | | /* |
53 | | * This module is specialized to the case DCTSIZE = 8. |
54 | | */ |
55 | | |
56 | | #if DCTSIZE != 8 |
57 | | Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */ |
58 | | #endif |
59 | | |
60 | | |
61 | | /* |
62 | | * The poop on this scaling stuff is as follows: |
63 | | * |
64 | | * Each 1-D DCT step produces outputs which are a factor of sqrt(N) |
65 | | * larger than the true DCT outputs. The final outputs are therefore |
66 | | * a factor of N larger than desired; since N=8 this can be cured by |
67 | | * a simple right shift at the end of the algorithm. The advantage of |
68 | | * this arrangement is that we save two multiplications per 1-D DCT, |
69 | | * because the y0 and y4 outputs need not be divided by sqrt(N). |
70 | | * In the IJG code, this factor of 8 is removed by the quantization step |
71 | | * (in jcdctmgr.c), NOT in this module. |
72 | | * |
73 | | * We have to do addition and subtraction of the integer inputs, which |
74 | | * is no problem, and multiplication by fractional constants, which is |
75 | | * a problem to do in integer arithmetic. We multiply all the constants |
76 | | * by CONST_SCALE and convert them to integer constants (thus retaining |
77 | | * CONST_BITS bits of precision in the constants). After doing a |
78 | | * multiplication we have to divide the product by CONST_SCALE, with proper |
79 | | * rounding, to produce the correct output. This division can be done |
80 | | * cheaply as a right shift of CONST_BITS bits. We postpone shifting |
81 | | * as long as possible so that partial sums can be added together with |
82 | | * full fractional precision. |
83 | | * |
84 | | * The outputs of the first pass are scaled up by PASS1_BITS bits so that |
85 | | * they are represented to better-than-integral precision. These outputs |
86 | | * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word |
87 | | * with the recommended scaling. (For 12-bit sample data, the intermediate |
88 | | * array is INT32 anyway.) |
89 | | * |
90 | | * To avoid overflow of the 32-bit intermediate results in pass 2, we must |
91 | | * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis |
92 | | * shows that the values given below are the most effective. |
93 | | */ |
94 | | |
95 | | #if BITS_IN_JSAMPLE == 8 |
96 | 0 | #define CONST_BITS 13 |
97 | 0 | #define PASS1_BITS 2 |
98 | | #else |
99 | | #define CONST_BITS 13 |
100 | | #define PASS1_BITS 1 /* lose a little precision to avoid overflow */ |
101 | | #endif |
102 | | |
103 | | /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus |
104 | | * causing a lot of useless floating-point operations at run time. |
105 | | * To get around this we use the following pre-calculated constants. |
106 | | * If you change CONST_BITS you may want to add appropriate values. |
107 | | * (With a reasonable C compiler, you can just rely on the FIX() macro...) |
108 | | */ |
109 | | |
110 | | #if CONST_BITS == 13 |
111 | | #define FIX_0_298631336 ((INT32) 2446) /* FIX(0.298631336) */ |
112 | | #define FIX_0_390180644 ((INT32) 3196) /* FIX(0.390180644) */ |
113 | | #define FIX_0_541196100 ((INT32) 4433) /* FIX(0.541196100) */ |
114 | | #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */ |
115 | | #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */ |
116 | | #define FIX_1_175875602 ((INT32) 9633) /* FIX(1.175875602) */ |
117 | | #define FIX_1_501321110 ((INT32) 12299) /* FIX(1.501321110) */ |
118 | | #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */ |
119 | | #define FIX_1_961570560 ((INT32) 16069) /* FIX(1.961570560) */ |
120 | | #define FIX_2_053119869 ((INT32) 16819) /* FIX(2.053119869) */ |
121 | | #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */ |
122 | | #define FIX_3_072711026 ((INT32) 25172) /* FIX(3.072711026) */ |
123 | | #else |
124 | | #define FIX_0_298631336 FIX(0.298631336) |
125 | | #define FIX_0_390180644 FIX(0.390180644) |
126 | | #define FIX_0_541196100 FIX(0.541196100) |
127 | | #define FIX_0_765366865 FIX(0.765366865) |
128 | | #define FIX_0_899976223 FIX(0.899976223) |
129 | | #define FIX_1_175875602 FIX(1.175875602) |
130 | | #define FIX_1_501321110 FIX(1.501321110) |
131 | | #define FIX_1_847759065 FIX(1.847759065) |
132 | | #define FIX_1_961570560 FIX(1.961570560) |
133 | | #define FIX_2_053119869 FIX(2.053119869) |
134 | | #define FIX_2_562915447 FIX(2.562915447) |
135 | | #define FIX_3_072711026 FIX(3.072711026) |
136 | | #endif |
137 | | |
138 | | |
139 | | /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result. |
140 | | * For 8-bit samples with the recommended scaling, all the variable |
141 | | * and constant values involved are no more than 16 bits wide, so a |
142 | | * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. |
143 | | * For 12-bit samples, a full 32-bit multiplication will be needed. |
144 | | */ |
145 | | |
146 | | #if BITS_IN_JSAMPLE == 8 |
147 | 0 | #define MULTIPLY(var,const) MULTIPLY16C16(var,const) |
148 | | #else |
149 | | #define MULTIPLY(var,const) ((var) * (const)) |
150 | | #endif |
151 | | |
152 | | |
153 | | /* |
154 | | * Perform the forward DCT on one block of samples. |
155 | | */ |
156 | | |
157 | | GLOBAL(void) |
158 | | jpeg_fdct_islow (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
159 | 0 | { |
160 | 0 | INT32 tmp0, tmp1, tmp2, tmp3; |
161 | 0 | INT32 tmp10, tmp11, tmp12, tmp13; |
162 | 0 | INT32 z1; |
163 | 0 | DCTELEM *dataptr; |
164 | 0 | JSAMPROW elemptr; |
165 | 0 | int ctr; |
166 | 0 | SHIFT_TEMPS |
167 | | |
168 | | /* Pass 1: process rows. |
169 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
170 | | * furthermore, we scale the results by 2**PASS1_BITS. |
171 | | * cK represents sqrt(2) * cos(K*pi/16). |
172 | | */ |
173 | |
|
174 | 0 | dataptr = data; |
175 | 0 | for (ctr = 0; ctr < DCTSIZE; ctr++) { |
176 | 0 | elemptr = sample_data[ctr] + start_col; |
177 | | |
178 | | /* Even part per LL&M figure 1 --- note that published figure is faulty; |
179 | | * rotator "c1" should be "c6". |
180 | | */ |
181 | |
|
182 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]); |
183 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]); |
184 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]); |
185 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]); |
186 | |
|
187 | 0 | tmp10 = tmp0 + tmp3; |
188 | 0 | tmp12 = tmp0 - tmp3; |
189 | 0 | tmp11 = tmp1 + tmp2; |
190 | 0 | tmp13 = tmp1 - tmp2; |
191 | |
|
192 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]); |
193 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]); |
194 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]); |
195 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]); |
196 | | |
197 | | /* Apply unsigned->signed conversion. */ |
198 | 0 | dataptr[0] = (DCTELEM) ((tmp10 + tmp11 - 8 * CENTERJSAMPLE) << PASS1_BITS); |
199 | 0 | dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS); |
200 | |
|
201 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); /* c6 */ |
202 | | /* Add fudge factor here for final descale. */ |
203 | 0 | z1 += ONE << (CONST_BITS-PASS1_BITS-1); |
204 | |
|
205 | 0 | dataptr[2] = (DCTELEM) |
206 | 0 | RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865), /* c2-c6 */ |
207 | 0 | CONST_BITS-PASS1_BITS); |
208 | 0 | dataptr[6] = (DCTELEM) |
209 | 0 | RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065), /* c2+c6 */ |
210 | 0 | CONST_BITS-PASS1_BITS); |
211 | | |
212 | | /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
213 | | * i0..i3 in the paper are tmp0..tmp3 here. |
214 | | */ |
215 | |
|
216 | 0 | tmp12 = tmp0 + tmp2; |
217 | 0 | tmp13 = tmp1 + tmp3; |
218 | |
|
219 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /* c3 */ |
220 | | /* Add fudge factor here for final descale. */ |
221 | 0 | z1 += ONE << (CONST_BITS-PASS1_BITS-1); |
222 | |
|
223 | 0 | tmp12 = MULTIPLY(tmp12, - FIX_0_390180644); /* -c3+c5 */ |
224 | 0 | tmp13 = MULTIPLY(tmp13, - FIX_1_961570560); /* -c3-c5 */ |
225 | 0 | tmp12 += z1; |
226 | 0 | tmp13 += z1; |
227 | |
|
228 | 0 | z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */ |
229 | 0 | tmp0 = MULTIPLY(tmp0, FIX_1_501321110); /* c1+c3-c5-c7 */ |
230 | 0 | tmp3 = MULTIPLY(tmp3, FIX_0_298631336); /* -c1+c3+c5-c7 */ |
231 | 0 | tmp0 += z1 + tmp12; |
232 | 0 | tmp3 += z1 + tmp13; |
233 | |
|
234 | 0 | z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */ |
235 | 0 | tmp1 = MULTIPLY(tmp1, FIX_3_072711026); /* c1+c3+c5-c7 */ |
236 | 0 | tmp2 = MULTIPLY(tmp2, FIX_2_053119869); /* c1+c3-c5+c7 */ |
237 | 0 | tmp1 += z1 + tmp13; |
238 | 0 | tmp2 += z1 + tmp12; |
239 | |
|
240 | 0 | dataptr[1] = (DCTELEM) RIGHT_SHIFT(tmp0, CONST_BITS-PASS1_BITS); |
241 | 0 | dataptr[3] = (DCTELEM) RIGHT_SHIFT(tmp1, CONST_BITS-PASS1_BITS); |
242 | 0 | dataptr[5] = (DCTELEM) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS); |
243 | 0 | dataptr[7] = (DCTELEM) RIGHT_SHIFT(tmp3, CONST_BITS-PASS1_BITS); |
244 | |
|
245 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
246 | 0 | } |
247 | | |
248 | | /* Pass 2: process columns. |
249 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
250 | | * by an overall factor of 8. |
251 | | * cK represents sqrt(2) * cos(K*pi/16). |
252 | | */ |
253 | |
|
254 | 0 | dataptr = data; |
255 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
256 | | /* Even part per LL&M figure 1 --- note that published figure is faulty; |
257 | | * rotator "c1" should be "c6". |
258 | | */ |
259 | |
|
260 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7]; |
261 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6]; |
262 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5]; |
263 | 0 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4]; |
264 | | |
265 | | /* Add fudge factor here for final descale. */ |
266 | 0 | tmp10 = tmp0 + tmp3 + (ONE << (PASS1_BITS-1)); |
267 | 0 | tmp12 = tmp0 - tmp3; |
268 | 0 | tmp11 = tmp1 + tmp2; |
269 | 0 | tmp13 = tmp1 - tmp2; |
270 | |
|
271 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7]; |
272 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6]; |
273 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5]; |
274 | 0 | tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4]; |
275 | |
|
276 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp10 + tmp11, PASS1_BITS); |
277 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) RIGHT_SHIFT(tmp10 - tmp11, PASS1_BITS); |
278 | |
|
279 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); /* c6 */ |
280 | | /* Add fudge factor here for final descale. */ |
281 | 0 | z1 += ONE << (CONST_BITS+PASS1_BITS-1); |
282 | |
|
283 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
284 | 0 | RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865), /* c2-c6 */ |
285 | 0 | CONST_BITS+PASS1_BITS); |
286 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
287 | 0 | RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065), /* c2+c6 */ |
288 | 0 | CONST_BITS+PASS1_BITS); |
289 | | |
290 | | /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
291 | | * i0..i3 in the paper are tmp0..tmp3 here. |
292 | | */ |
293 | |
|
294 | 0 | tmp12 = tmp0 + tmp2; |
295 | 0 | tmp13 = tmp1 + tmp3; |
296 | |
|
297 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /* c3 */ |
298 | | /* Add fudge factor here for final descale. */ |
299 | 0 | z1 += ONE << (CONST_BITS+PASS1_BITS-1); |
300 | |
|
301 | 0 | tmp12 = MULTIPLY(tmp12, - FIX_0_390180644); /* -c3+c5 */ |
302 | 0 | tmp13 = MULTIPLY(tmp13, - FIX_1_961570560); /* -c3-c5 */ |
303 | 0 | tmp12 += z1; |
304 | 0 | tmp13 += z1; |
305 | |
|
306 | 0 | z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */ |
307 | 0 | tmp0 = MULTIPLY(tmp0, FIX_1_501321110); /* c1+c3-c5-c7 */ |
308 | 0 | tmp3 = MULTIPLY(tmp3, FIX_0_298631336); /* -c1+c3+c5-c7 */ |
309 | 0 | tmp0 += z1 + tmp12; |
310 | 0 | tmp3 += z1 + tmp13; |
311 | |
|
312 | 0 | z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */ |
313 | 0 | tmp1 = MULTIPLY(tmp1, FIX_3_072711026); /* c1+c3+c5-c7 */ |
314 | 0 | tmp2 = MULTIPLY(tmp2, FIX_2_053119869); /* c1+c3-c5+c7 */ |
315 | 0 | tmp1 += z1 + tmp13; |
316 | 0 | tmp2 += z1 + tmp12; |
317 | |
|
318 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) RIGHT_SHIFT(tmp0, CONST_BITS+PASS1_BITS); |
319 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) RIGHT_SHIFT(tmp1, CONST_BITS+PASS1_BITS); |
320 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) RIGHT_SHIFT(tmp2, CONST_BITS+PASS1_BITS); |
321 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) RIGHT_SHIFT(tmp3, CONST_BITS+PASS1_BITS); |
322 | |
|
323 | 0 | dataptr++; /* advance pointer to next column */ |
324 | 0 | } |
325 | 0 | } |
326 | | |
327 | | #ifdef DCT_SCALING_SUPPORTED |
328 | | |
329 | | |
330 | | /* |
331 | | * Perform the forward DCT on a 7x7 sample block. |
332 | | */ |
333 | | |
334 | | GLOBAL(void) |
335 | | jpeg_fdct_7x7 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
336 | 0 | { |
337 | 0 | INT32 tmp0, tmp1, tmp2, tmp3; |
338 | 0 | INT32 tmp10, tmp11, tmp12; |
339 | 0 | INT32 z1, z2, z3; |
340 | 0 | DCTELEM *dataptr; |
341 | 0 | JSAMPROW elemptr; |
342 | 0 | int ctr; |
343 | 0 | SHIFT_TEMPS |
344 | | |
345 | | /* Pre-zero output coefficient block. */ |
346 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
347 | | |
348 | | /* Pass 1: process rows. |
349 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
350 | | * furthermore, we scale the results by 2**PASS1_BITS. |
351 | | * cK represents sqrt(2) * cos(K*pi/14). |
352 | | */ |
353 | |
|
354 | 0 | dataptr = data; |
355 | 0 | for (ctr = 0; ctr < 7; ctr++) { |
356 | 0 | elemptr = sample_data[ctr] + start_col; |
357 | | |
358 | | /* Even part */ |
359 | |
|
360 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[6]); |
361 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[5]); |
362 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[4]); |
363 | 0 | tmp3 = GETJSAMPLE(elemptr[3]); |
364 | |
|
365 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[6]); |
366 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[5]); |
367 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[4]); |
368 | |
|
369 | 0 | z1 = tmp0 + tmp2; |
370 | | /* Apply unsigned->signed conversion. */ |
371 | 0 | dataptr[0] = (DCTELEM) |
372 | 0 | ((z1 + tmp1 + tmp3 - 7 * CENTERJSAMPLE) << PASS1_BITS); |
373 | 0 | tmp3 += tmp3; |
374 | 0 | z1 -= tmp3; |
375 | 0 | z1 -= tmp3; |
376 | 0 | z1 = MULTIPLY(z1, FIX(0.353553391)); /* (c2+c6-c4)/2 */ |
377 | 0 | z2 = MULTIPLY(tmp0 - tmp2, FIX(0.920609002)); /* (c2+c4-c6)/2 */ |
378 | 0 | z3 = MULTIPLY(tmp1 - tmp2, FIX(0.314692123)); /* c6 */ |
379 | 0 | dataptr[2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS-PASS1_BITS); |
380 | 0 | z1 -= z2; |
381 | 0 | z2 = MULTIPLY(tmp0 - tmp1, FIX(0.881747734)); /* c4 */ |
382 | 0 | dataptr[4] = (DCTELEM) |
383 | 0 | DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.707106781)), /* c2+c6-c4 */ |
384 | 0 | CONST_BITS-PASS1_BITS); |
385 | 0 | dataptr[6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS-PASS1_BITS); |
386 | | |
387 | | /* Odd part */ |
388 | |
|
389 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(0.935414347)); /* (c3+c1-c5)/2 */ |
390 | 0 | tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.170262339)); /* (c3+c5-c1)/2 */ |
391 | 0 | tmp0 = tmp1 - tmp2; |
392 | 0 | tmp1 += tmp2; |
393 | 0 | tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.378756276)); /* -c1 */ |
394 | 0 | tmp1 += tmp2; |
395 | 0 | tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.613604268)); /* c5 */ |
396 | 0 | tmp0 += tmp3; |
397 | 0 | tmp2 += tmp3 + MULTIPLY(tmp12, FIX(1.870828693)); /* c3+c1-c5 */ |
398 | |
|
399 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-PASS1_BITS); |
400 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-PASS1_BITS); |
401 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS-PASS1_BITS); |
402 | |
|
403 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
404 | 0 | } |
405 | | |
406 | | /* Pass 2: process columns. |
407 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
408 | | * by an overall factor of 8. |
409 | | * We must also scale the output by (8/7)**2 = 64/49, which we fold |
410 | | * into the constant multipliers: |
411 | | * cK now represents sqrt(2) * cos(K*pi/14) * 64/49. |
412 | | */ |
413 | |
|
414 | 0 | dataptr = data; |
415 | 0 | for (ctr = 0; ctr < 7; ctr++) { |
416 | | /* Even part */ |
417 | |
|
418 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*6]; |
419 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*5]; |
420 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*4]; |
421 | 0 | tmp3 = dataptr[DCTSIZE*3]; |
422 | |
|
423 | 0 | tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*6]; |
424 | 0 | tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*5]; |
425 | 0 | tmp12 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*4]; |
426 | |
|
427 | 0 | z1 = tmp0 + tmp2; |
428 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
429 | 0 | DESCALE(MULTIPLY(z1 + tmp1 + tmp3, FIX(1.306122449)), /* 64/49 */ |
430 | 0 | CONST_BITS+PASS1_BITS); |
431 | 0 | tmp3 += tmp3; |
432 | 0 | z1 -= tmp3; |
433 | 0 | z1 -= tmp3; |
434 | 0 | z1 = MULTIPLY(z1, FIX(0.461784020)); /* (c2+c6-c4)/2 */ |
435 | 0 | z2 = MULTIPLY(tmp0 - tmp2, FIX(1.202428084)); /* (c2+c4-c6)/2 */ |
436 | 0 | z3 = MULTIPLY(tmp1 - tmp2, FIX(0.411026446)); /* c6 */ |
437 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS+PASS1_BITS); |
438 | 0 | z1 -= z2; |
439 | 0 | z2 = MULTIPLY(tmp0 - tmp1, FIX(1.151670509)); /* c4 */ |
440 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
441 | 0 | DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.923568041)), /* c2+c6-c4 */ |
442 | 0 | CONST_BITS+PASS1_BITS); |
443 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS+PASS1_BITS); |
444 | | |
445 | | /* Odd part */ |
446 | |
|
447 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.221765677)); /* (c3+c1-c5)/2 */ |
448 | 0 | tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.222383464)); /* (c3+c5-c1)/2 */ |
449 | 0 | tmp0 = tmp1 - tmp2; |
450 | 0 | tmp1 += tmp2; |
451 | 0 | tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.800824523)); /* -c1 */ |
452 | 0 | tmp1 += tmp2; |
453 | 0 | tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.801442310)); /* c5 */ |
454 | 0 | tmp0 += tmp3; |
455 | 0 | tmp2 += tmp3 + MULTIPLY(tmp12, FIX(2.443531355)); /* c3+c1-c5 */ |
456 | |
|
457 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+PASS1_BITS); |
458 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+PASS1_BITS); |
459 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+PASS1_BITS); |
460 | |
|
461 | 0 | dataptr++; /* advance pointer to next column */ |
462 | 0 | } |
463 | 0 | } |
464 | | |
465 | | |
466 | | /* |
467 | | * Perform the forward DCT on a 6x6 sample block. |
468 | | */ |
469 | | |
470 | | GLOBAL(void) |
471 | | jpeg_fdct_6x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
472 | 0 | { |
473 | 0 | INT32 tmp0, tmp1, tmp2; |
474 | 0 | INT32 tmp10, tmp11, tmp12; |
475 | 0 | DCTELEM *dataptr; |
476 | 0 | JSAMPROW elemptr; |
477 | 0 | int ctr; |
478 | 0 | SHIFT_TEMPS |
479 | | |
480 | | /* Pre-zero output coefficient block. */ |
481 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
482 | | |
483 | | /* Pass 1: process rows. |
484 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
485 | | * furthermore, we scale the results by 2**PASS1_BITS. |
486 | | * cK represents sqrt(2) * cos(K*pi/12). |
487 | | */ |
488 | |
|
489 | 0 | dataptr = data; |
490 | 0 | for (ctr = 0; ctr < 6; ctr++) { |
491 | 0 | elemptr = sample_data[ctr] + start_col; |
492 | | |
493 | | /* Even part */ |
494 | |
|
495 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]); |
496 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]); |
497 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]); |
498 | |
|
499 | 0 | tmp10 = tmp0 + tmp2; |
500 | 0 | tmp12 = tmp0 - tmp2; |
501 | |
|
502 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]); |
503 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]); |
504 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]); |
505 | | |
506 | | /* Apply unsigned->signed conversion. */ |
507 | 0 | dataptr[0] = (DCTELEM) |
508 | 0 | ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << PASS1_BITS); |
509 | 0 | dataptr[2] = (DCTELEM) |
510 | 0 | DESCALE(MULTIPLY(tmp12, FIX(1.224744871)), /* c2 */ |
511 | 0 | CONST_BITS-PASS1_BITS); |
512 | 0 | dataptr[4] = (DCTELEM) |
513 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */ |
514 | 0 | CONST_BITS-PASS1_BITS); |
515 | | |
516 | | /* Odd part */ |
517 | |
|
518 | 0 | tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)), /* c5 */ |
519 | 0 | CONST_BITS-PASS1_BITS); |
520 | |
|
521 | 0 | dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << PASS1_BITS)); |
522 | 0 | dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << PASS1_BITS); |
523 | 0 | dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << PASS1_BITS)); |
524 | |
|
525 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
526 | 0 | } |
527 | | |
528 | | /* Pass 2: process columns. |
529 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
530 | | * by an overall factor of 8. |
531 | | * We must also scale the output by (8/6)**2 = 16/9, which we fold |
532 | | * into the constant multipliers: |
533 | | * cK now represents sqrt(2) * cos(K*pi/12) * 16/9. |
534 | | */ |
535 | |
|
536 | 0 | dataptr = data; |
537 | 0 | for (ctr = 0; ctr < 6; ctr++) { |
538 | | /* Even part */ |
539 | |
|
540 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5]; |
541 | 0 | tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4]; |
542 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3]; |
543 | |
|
544 | 0 | tmp10 = tmp0 + tmp2; |
545 | 0 | tmp12 = tmp0 - tmp2; |
546 | |
|
547 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5]; |
548 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4]; |
549 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3]; |
550 | |
|
551 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
552 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)), /* 16/9 */ |
553 | 0 | CONST_BITS+PASS1_BITS); |
554 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
555 | 0 | DESCALE(MULTIPLY(tmp12, FIX(2.177324216)), /* c2 */ |
556 | 0 | CONST_BITS+PASS1_BITS); |
557 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
558 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */ |
559 | 0 | CONST_BITS+PASS1_BITS); |
560 | | |
561 | | /* Odd part */ |
562 | |
|
563 | 0 | tmp10 = MULTIPLY(tmp0 + tmp2, FIX(0.650711829)); /* c5 */ |
564 | |
|
565 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
566 | 0 | DESCALE(tmp10 + MULTIPLY(tmp0 + tmp1, FIX(1.777777778)), /* 16/9 */ |
567 | 0 | CONST_BITS+PASS1_BITS); |
568 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
569 | 0 | DESCALE(MULTIPLY(tmp0 - tmp1 - tmp2, FIX(1.777777778)), /* 16/9 */ |
570 | 0 | CONST_BITS+PASS1_BITS); |
571 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
572 | 0 | DESCALE(tmp10 + MULTIPLY(tmp2 - tmp1, FIX(1.777777778)), /* 16/9 */ |
573 | 0 | CONST_BITS+PASS1_BITS); |
574 | |
|
575 | 0 | dataptr++; /* advance pointer to next column */ |
576 | 0 | } |
577 | 0 | } |
578 | | |
579 | | |
580 | | /* |
581 | | * Perform the forward DCT on a 5x5 sample block. |
582 | | */ |
583 | | |
584 | | GLOBAL(void) |
585 | | jpeg_fdct_5x5 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
586 | 0 | { |
587 | 0 | INT32 tmp0, tmp1, tmp2; |
588 | 0 | INT32 tmp10, tmp11; |
589 | 0 | DCTELEM *dataptr; |
590 | 0 | JSAMPROW elemptr; |
591 | 0 | int ctr; |
592 | 0 | SHIFT_TEMPS |
593 | | |
594 | | /* Pre-zero output coefficient block. */ |
595 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
596 | | |
597 | | /* Pass 1: process rows. |
598 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
599 | | * furthermore, we scale the results by 2**PASS1_BITS. |
600 | | * We scale the results further by 2 as part of output adaption |
601 | | * scaling for different DCT size. |
602 | | * cK represents sqrt(2) * cos(K*pi/10). |
603 | | */ |
604 | |
|
605 | 0 | dataptr = data; |
606 | 0 | for (ctr = 0; ctr < 5; ctr++) { |
607 | 0 | elemptr = sample_data[ctr] + start_col; |
608 | | |
609 | | /* Even part */ |
610 | |
|
611 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[4]); |
612 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[3]); |
613 | 0 | tmp2 = GETJSAMPLE(elemptr[2]); |
614 | |
|
615 | 0 | tmp10 = tmp0 + tmp1; |
616 | 0 | tmp11 = tmp0 - tmp1; |
617 | |
|
618 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[4]); |
619 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[3]); |
620 | | |
621 | | /* Apply unsigned->signed conversion. */ |
622 | 0 | dataptr[0] = (DCTELEM) |
623 | 0 | ((tmp10 + tmp2 - 5 * CENTERJSAMPLE) << (PASS1_BITS+1)); |
624 | 0 | tmp11 = MULTIPLY(tmp11, FIX(0.790569415)); /* (c2+c4)/2 */ |
625 | 0 | tmp10 -= tmp2 << 2; |
626 | 0 | tmp10 = MULTIPLY(tmp10, FIX(0.353553391)); /* (c2-c4)/2 */ |
627 | 0 | dataptr[2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS-PASS1_BITS-1); |
628 | 0 | dataptr[4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS-PASS1_BITS-1); |
629 | | |
630 | | /* Odd part */ |
631 | |
|
632 | 0 | tmp10 = MULTIPLY(tmp0 + tmp1, FIX(0.831253876)); /* c3 */ |
633 | |
|
634 | 0 | dataptr[1] = (DCTELEM) |
635 | 0 | DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.513743148)), /* c1-c3 */ |
636 | 0 | CONST_BITS-PASS1_BITS-1); |
637 | 0 | dataptr[3] = (DCTELEM) |
638 | 0 | DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.176250899)), /* c1+c3 */ |
639 | 0 | CONST_BITS-PASS1_BITS-1); |
640 | |
|
641 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
642 | 0 | } |
643 | | |
644 | | /* Pass 2: process columns. |
645 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
646 | | * by an overall factor of 8. |
647 | | * We must also scale the output by (8/5)**2 = 64/25, which we partially |
648 | | * fold into the constant multipliers (other part was done in pass 1): |
649 | | * cK now represents sqrt(2) * cos(K*pi/10) * 32/25. |
650 | | */ |
651 | |
|
652 | 0 | dataptr = data; |
653 | 0 | for (ctr = 0; ctr < 5; ctr++) { |
654 | | /* Even part */ |
655 | |
|
656 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*4]; |
657 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*3]; |
658 | 0 | tmp2 = dataptr[DCTSIZE*2]; |
659 | |
|
660 | 0 | tmp10 = tmp0 + tmp1; |
661 | 0 | tmp11 = tmp0 - tmp1; |
662 | |
|
663 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*4]; |
664 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*3]; |
665 | |
|
666 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
667 | 0 | DESCALE(MULTIPLY(tmp10 + tmp2, FIX(1.28)), /* 32/25 */ |
668 | 0 | CONST_BITS+PASS1_BITS); |
669 | 0 | tmp11 = MULTIPLY(tmp11, FIX(1.011928851)); /* (c2+c4)/2 */ |
670 | 0 | tmp10 -= tmp2 << 2; |
671 | 0 | tmp10 = MULTIPLY(tmp10, FIX(0.452548340)); /* (c2-c4)/2 */ |
672 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS+PASS1_BITS); |
673 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS+PASS1_BITS); |
674 | | |
675 | | /* Odd part */ |
676 | |
|
677 | 0 | tmp10 = MULTIPLY(tmp0 + tmp1, FIX(1.064004961)); /* c3 */ |
678 | |
|
679 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
680 | 0 | DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.657591230)), /* c1-c3 */ |
681 | 0 | CONST_BITS+PASS1_BITS); |
682 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
683 | 0 | DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.785601151)), /* c1+c3 */ |
684 | 0 | CONST_BITS+PASS1_BITS); |
685 | |
|
686 | 0 | dataptr++; /* advance pointer to next column */ |
687 | 0 | } |
688 | 0 | } |
689 | | |
690 | | |
691 | | /* |
692 | | * Perform the forward DCT on a 4x4 sample block. |
693 | | */ |
694 | | |
695 | | GLOBAL(void) |
696 | | jpeg_fdct_4x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
697 | 0 | { |
698 | 0 | INT32 tmp0, tmp1; |
699 | 0 | INT32 tmp10, tmp11; |
700 | 0 | DCTELEM *dataptr; |
701 | 0 | JSAMPROW elemptr; |
702 | 0 | int ctr; |
703 | 0 | SHIFT_TEMPS |
704 | | |
705 | | /* Pre-zero output coefficient block. */ |
706 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
707 | | |
708 | | /* Pass 1: process rows. |
709 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
710 | | * furthermore, we scale the results by 2**PASS1_BITS. |
711 | | * We must also scale the output by (8/4)**2 = 2**2, which we add here. |
712 | | * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. |
713 | | */ |
714 | |
|
715 | 0 | dataptr = data; |
716 | 0 | for (ctr = 0; ctr < 4; ctr++) { |
717 | 0 | elemptr = sample_data[ctr] + start_col; |
718 | | |
719 | | /* Even part */ |
720 | |
|
721 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]); |
722 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]); |
723 | |
|
724 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]); |
725 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]); |
726 | | |
727 | | /* Apply unsigned->signed conversion. */ |
728 | 0 | dataptr[0] = (DCTELEM) |
729 | 0 | ((tmp0 + tmp1 - 4 * CENTERJSAMPLE) << (PASS1_BITS+2)); |
730 | 0 | dataptr[2] = (DCTELEM) ((tmp0 - tmp1) << (PASS1_BITS+2)); |
731 | | |
732 | | /* Odd part */ |
733 | |
|
734 | 0 | tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100); /* c6 */ |
735 | | /* Add fudge factor here for final descale. */ |
736 | 0 | tmp0 += ONE << (CONST_BITS-PASS1_BITS-3); |
737 | |
|
738 | 0 | dataptr[1] = (DCTELEM) |
739 | 0 | RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */ |
740 | 0 | CONST_BITS-PASS1_BITS-2); |
741 | 0 | dataptr[3] = (DCTELEM) |
742 | 0 | RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */ |
743 | 0 | CONST_BITS-PASS1_BITS-2); |
744 | |
|
745 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
746 | 0 | } |
747 | | |
748 | | /* Pass 2: process columns. |
749 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
750 | | * by an overall factor of 8. |
751 | | * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. |
752 | | */ |
753 | |
|
754 | 0 | dataptr = data; |
755 | 0 | for (ctr = 0; ctr < 4; ctr++) { |
756 | | /* Even part */ |
757 | | |
758 | | /* Add fudge factor here for final descale. */ |
759 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*3] + (ONE << (PASS1_BITS-1)); |
760 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*2]; |
761 | |
|
762 | 0 | tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*3]; |
763 | 0 | tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*2]; |
764 | |
|
765 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp0 + tmp1, PASS1_BITS); |
766 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) RIGHT_SHIFT(tmp0 - tmp1, PASS1_BITS); |
767 | | |
768 | | /* Odd part */ |
769 | |
|
770 | 0 | tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100); /* c6 */ |
771 | | /* Add fudge factor here for final descale. */ |
772 | 0 | tmp0 += ONE << (CONST_BITS+PASS1_BITS-1); |
773 | |
|
774 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
775 | 0 | RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */ |
776 | 0 | CONST_BITS+PASS1_BITS); |
777 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
778 | 0 | RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */ |
779 | 0 | CONST_BITS+PASS1_BITS); |
780 | |
|
781 | 0 | dataptr++; /* advance pointer to next column */ |
782 | 0 | } |
783 | 0 | } |
784 | | |
785 | | |
786 | | /* |
787 | | * Perform the forward DCT on a 3x3 sample block. |
788 | | */ |
789 | | |
790 | | GLOBAL(void) |
791 | | jpeg_fdct_3x3 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
792 | 0 | { |
793 | 0 | INT32 tmp0, tmp1, tmp2; |
794 | 0 | DCTELEM *dataptr; |
795 | 0 | JSAMPROW elemptr; |
796 | 0 | int ctr; |
797 | 0 | SHIFT_TEMPS |
798 | | |
799 | | /* Pre-zero output coefficient block. */ |
800 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
801 | | |
802 | | /* Pass 1: process rows. |
803 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
804 | | * furthermore, we scale the results by 2**PASS1_BITS. |
805 | | * We scale the results further by 2**2 as part of output adaption |
806 | | * scaling for different DCT size. |
807 | | * cK represents sqrt(2) * cos(K*pi/6). |
808 | | */ |
809 | |
|
810 | 0 | dataptr = data; |
811 | 0 | for (ctr = 0; ctr < 3; ctr++) { |
812 | 0 | elemptr = sample_data[ctr] + start_col; |
813 | | |
814 | | /* Even part */ |
815 | |
|
816 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[2]); |
817 | 0 | tmp1 = GETJSAMPLE(elemptr[1]); |
818 | |
|
819 | 0 | tmp2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[2]); |
820 | | |
821 | | /* Apply unsigned->signed conversion. */ |
822 | 0 | dataptr[0] = (DCTELEM) |
823 | 0 | ((tmp0 + tmp1 - 3 * CENTERJSAMPLE) << (PASS1_BITS+2)); |
824 | 0 | dataptr[2] = (DCTELEM) |
825 | 0 | DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(0.707106781)), /* c2 */ |
826 | 0 | CONST_BITS-PASS1_BITS-2); |
827 | | |
828 | | /* Odd part */ |
829 | |
|
830 | 0 | dataptr[1] = (DCTELEM) |
831 | 0 | DESCALE(MULTIPLY(tmp2, FIX(1.224744871)), /* c1 */ |
832 | 0 | CONST_BITS-PASS1_BITS-2); |
833 | |
|
834 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
835 | 0 | } |
836 | | |
837 | | /* Pass 2: process columns. |
838 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
839 | | * by an overall factor of 8. |
840 | | * We must also scale the output by (8/3)**2 = 64/9, which we partially |
841 | | * fold into the constant multipliers (other part was done in pass 1): |
842 | | * cK now represents sqrt(2) * cos(K*pi/6) * 16/9. |
843 | | */ |
844 | |
|
845 | 0 | dataptr = data; |
846 | 0 | for (ctr = 0; ctr < 3; ctr++) { |
847 | | /* Even part */ |
848 | |
|
849 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*2]; |
850 | 0 | tmp1 = dataptr[DCTSIZE*1]; |
851 | |
|
852 | 0 | tmp2 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*2]; |
853 | |
|
854 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
855 | 0 | DESCALE(MULTIPLY(tmp0 + tmp1, FIX(1.777777778)), /* 16/9 */ |
856 | 0 | CONST_BITS+PASS1_BITS); |
857 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
858 | 0 | DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(1.257078722)), /* c2 */ |
859 | 0 | CONST_BITS+PASS1_BITS); |
860 | | |
861 | | /* Odd part */ |
862 | |
|
863 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
864 | 0 | DESCALE(MULTIPLY(tmp2, FIX(2.177324216)), /* c1 */ |
865 | 0 | CONST_BITS+PASS1_BITS); |
866 | |
|
867 | 0 | dataptr++; /* advance pointer to next column */ |
868 | 0 | } |
869 | 0 | } |
870 | | |
871 | | |
872 | | /* |
873 | | * Perform the forward DCT on a 2x2 sample block. |
874 | | */ |
875 | | |
876 | | GLOBAL(void) |
877 | | jpeg_fdct_2x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
878 | 0 | { |
879 | 0 | DCTELEM tmp0, tmp1, tmp2, tmp3; |
880 | 0 | JSAMPROW elemptr; |
881 | | |
882 | | /* Pre-zero output coefficient block. */ |
883 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
884 | | |
885 | | /* Pass 1: process rows. |
886 | | * Note results are scaled up by sqrt(8) compared to a true DCT. |
887 | | */ |
888 | | |
889 | | /* Row 0 */ |
890 | 0 | elemptr = sample_data[0] + start_col; |
891 | |
|
892 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[1]); |
893 | 0 | tmp1 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[1]); |
894 | | |
895 | | /* Row 1 */ |
896 | 0 | elemptr = sample_data[1] + start_col; |
897 | |
|
898 | 0 | tmp2 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[1]); |
899 | 0 | tmp3 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[1]); |
900 | | |
901 | | /* Pass 2: process columns. |
902 | | * We leave the results scaled up by an overall factor of 8. |
903 | | * We must also scale the output by (8/2)**2 = 2**4. |
904 | | */ |
905 | | |
906 | | /* Column 0 */ |
907 | | /* Apply unsigned->signed conversion. */ |
908 | 0 | data[DCTSIZE*0] = (tmp0 + tmp2 - 4 * CENTERJSAMPLE) << 4; |
909 | 0 | data[DCTSIZE*1] = (tmp0 - tmp2) << 4; |
910 | | |
911 | | /* Column 1 */ |
912 | 0 | data[DCTSIZE*0+1] = (tmp1 + tmp3) << 4; |
913 | 0 | data[DCTSIZE*1+1] = (tmp1 - tmp3) << 4; |
914 | 0 | } |
915 | | |
916 | | |
917 | | /* |
918 | | * Perform the forward DCT on a 1x1 sample block. |
919 | | */ |
920 | | |
921 | | GLOBAL(void) |
922 | | jpeg_fdct_1x1 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
923 | 0 | { |
924 | 0 | DCTELEM dcval; |
925 | | |
926 | | /* Pre-zero output coefficient block. */ |
927 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
928 | |
|
929 | 0 | dcval = GETJSAMPLE(sample_data[0][start_col]); |
930 | | |
931 | | /* We leave the result scaled up by an overall factor of 8. */ |
932 | | /* We must also scale the output by (8/1)**2 = 2**6. */ |
933 | | /* Apply unsigned->signed conversion. */ |
934 | 0 | data[0] = (dcval - CENTERJSAMPLE) << 6; |
935 | 0 | } |
936 | | |
937 | | |
938 | | /* |
939 | | * Perform the forward DCT on a 9x9 sample block. |
940 | | */ |
941 | | |
942 | | GLOBAL(void) |
943 | | jpeg_fdct_9x9 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
944 | 0 | { |
945 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4; |
946 | 0 | INT32 tmp10, tmp11, tmp12, tmp13; |
947 | 0 | INT32 z1, z2; |
948 | 0 | DCTELEM workspace[8]; |
949 | 0 | DCTELEM *dataptr; |
950 | 0 | DCTELEM *wsptr; |
951 | 0 | JSAMPROW elemptr; |
952 | 0 | int ctr; |
953 | 0 | SHIFT_TEMPS |
954 | | |
955 | | /* Pass 1: process rows. |
956 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
957 | | * we scale the results further by 2 as part of output adaption |
958 | | * scaling for different DCT size. |
959 | | * cK represents sqrt(2) * cos(K*pi/18). |
960 | | */ |
961 | |
|
962 | 0 | dataptr = data; |
963 | 0 | ctr = 0; |
964 | 0 | for (;;) { |
965 | 0 | elemptr = sample_data[ctr] + start_col; |
966 | | |
967 | | /* Even part */ |
968 | |
|
969 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[8]); |
970 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[7]); |
971 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[6]); |
972 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[5]); |
973 | 0 | tmp4 = GETJSAMPLE(elemptr[4]); |
974 | |
|
975 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[8]); |
976 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[7]); |
977 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[6]); |
978 | 0 | tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[5]); |
979 | |
|
980 | 0 | z1 = tmp0 + tmp2 + tmp3; |
981 | 0 | z2 = tmp1 + tmp4; |
982 | | /* Apply unsigned->signed conversion. */ |
983 | 0 | dataptr[0] = (DCTELEM) ((z1 + z2 - 9 * CENTERJSAMPLE) << 1); |
984 | 0 | dataptr[6] = (DCTELEM) |
985 | 0 | DESCALE(MULTIPLY(z1 - z2 - z2, FIX(0.707106781)), /* c6 */ |
986 | 0 | CONST_BITS-1); |
987 | 0 | z1 = MULTIPLY(tmp0 - tmp2, FIX(1.328926049)); /* c2 */ |
988 | 0 | z2 = MULTIPLY(tmp1 - tmp4 - tmp4, FIX(0.707106781)); /* c6 */ |
989 | 0 | dataptr[2] = (DCTELEM) |
990 | 0 | DESCALE(MULTIPLY(tmp2 - tmp3, FIX(1.083350441)) /* c4 */ |
991 | 0 | + z1 + z2, CONST_BITS-1); |
992 | 0 | dataptr[4] = (DCTELEM) |
993 | 0 | DESCALE(MULTIPLY(tmp3 - tmp0, FIX(0.245575608)) /* c8 */ |
994 | 0 | + z1 - z2, CONST_BITS-1); |
995 | | |
996 | | /* Odd part */ |
997 | |
|
998 | 0 | dataptr[3] = (DCTELEM) |
999 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12 - tmp13, FIX(1.224744871)), /* c3 */ |
1000 | 0 | CONST_BITS-1); |
1001 | |
|
1002 | 0 | tmp11 = MULTIPLY(tmp11, FIX(1.224744871)); /* c3 */ |
1003 | 0 | tmp0 = MULTIPLY(tmp10 + tmp12, FIX(0.909038955)); /* c5 */ |
1004 | 0 | tmp1 = MULTIPLY(tmp10 + tmp13, FIX(0.483689525)); /* c7 */ |
1005 | |
|
1006 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp11 + tmp0 + tmp1, CONST_BITS-1); |
1007 | |
|
1008 | 0 | tmp2 = MULTIPLY(tmp12 - tmp13, FIX(1.392728481)); /* c1 */ |
1009 | |
|
1010 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp0 - tmp11 - tmp2, CONST_BITS-1); |
1011 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp1 - tmp11 + tmp2, CONST_BITS-1); |
1012 | |
|
1013 | 0 | ctr++; |
1014 | |
|
1015 | 0 | if (ctr != DCTSIZE) { |
1016 | 0 | if (ctr == 9) |
1017 | 0 | break; /* Done. */ |
1018 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
1019 | 0 | } else |
1020 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
1021 | 0 | } |
1022 | | |
1023 | | /* Pass 2: process columns. |
1024 | | * We leave the results scaled up by an overall factor of 8. |
1025 | | * We must also scale the output by (8/9)**2 = 64/81, which we partially |
1026 | | * fold into the constant multipliers and final/initial shifting: |
1027 | | * cK now represents sqrt(2) * cos(K*pi/18) * 128/81. |
1028 | | */ |
1029 | |
|
1030 | 0 | dataptr = data; |
1031 | 0 | wsptr = workspace; |
1032 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
1033 | | /* Even part */ |
1034 | |
|
1035 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*0]; |
1036 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*7]; |
1037 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*6]; |
1038 | 0 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*5]; |
1039 | 0 | tmp4 = dataptr[DCTSIZE*4]; |
1040 | |
|
1041 | 0 | tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*0]; |
1042 | 0 | tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*7]; |
1043 | 0 | tmp12 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*6]; |
1044 | 0 | tmp13 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*5]; |
1045 | |
|
1046 | 0 | z1 = tmp0 + tmp2 + tmp3; |
1047 | 0 | z2 = tmp1 + tmp4; |
1048 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
1049 | 0 | DESCALE(MULTIPLY(z1 + z2, FIX(1.580246914)), /* 128/81 */ |
1050 | 0 | CONST_BITS+2); |
1051 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
1052 | 0 | DESCALE(MULTIPLY(z1 - z2 - z2, FIX(1.117403309)), /* c6 */ |
1053 | 0 | CONST_BITS+2); |
1054 | 0 | z1 = MULTIPLY(tmp0 - tmp2, FIX(2.100031287)); /* c2 */ |
1055 | 0 | z2 = MULTIPLY(tmp1 - tmp4 - tmp4, FIX(1.117403309)); /* c6 */ |
1056 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
1057 | 0 | DESCALE(MULTIPLY(tmp2 - tmp3, FIX(1.711961190)) /* c4 */ |
1058 | 0 | + z1 + z2, CONST_BITS+2); |
1059 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
1060 | 0 | DESCALE(MULTIPLY(tmp3 - tmp0, FIX(0.388070096)) /* c8 */ |
1061 | 0 | + z1 - z2, CONST_BITS+2); |
1062 | | |
1063 | | /* Odd part */ |
1064 | |
|
1065 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
1066 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12 - tmp13, FIX(1.935399303)), /* c3 */ |
1067 | 0 | CONST_BITS+2); |
1068 | |
|
1069 | 0 | tmp11 = MULTIPLY(tmp11, FIX(1.935399303)); /* c3 */ |
1070 | 0 | tmp0 = MULTIPLY(tmp10 + tmp12, FIX(1.436506004)); /* c5 */ |
1071 | 0 | tmp1 = MULTIPLY(tmp10 + tmp13, FIX(0.764348879)); /* c7 */ |
1072 | |
|
1073 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
1074 | 0 | DESCALE(tmp11 + tmp0 + tmp1, CONST_BITS+2); |
1075 | |
|
1076 | 0 | tmp2 = MULTIPLY(tmp12 - tmp13, FIX(2.200854883)); /* c1 */ |
1077 | |
|
1078 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
1079 | 0 | DESCALE(tmp0 - tmp11 - tmp2, CONST_BITS+2); |
1080 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) |
1081 | 0 | DESCALE(tmp1 - tmp11 + tmp2, CONST_BITS+2); |
1082 | |
|
1083 | 0 | dataptr++; /* advance pointer to next column */ |
1084 | 0 | wsptr++; /* advance pointer to next column */ |
1085 | 0 | } |
1086 | 0 | } |
1087 | | |
1088 | | |
1089 | | /* |
1090 | | * Perform the forward DCT on a 10x10 sample block. |
1091 | | */ |
1092 | | |
1093 | | GLOBAL(void) |
1094 | | jpeg_fdct_10x10 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
1095 | 0 | { |
1096 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4; |
1097 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14; |
1098 | 0 | DCTELEM workspace[8*2]; |
1099 | 0 | DCTELEM *dataptr; |
1100 | 0 | DCTELEM *wsptr; |
1101 | 0 | JSAMPROW elemptr; |
1102 | 0 | int ctr; |
1103 | 0 | SHIFT_TEMPS |
1104 | | |
1105 | | /* Pass 1: process rows. |
1106 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
1107 | | * we scale the results further by 2 as part of output adaption |
1108 | | * scaling for different DCT size. |
1109 | | * cK represents sqrt(2) * cos(K*pi/20). |
1110 | | */ |
1111 | |
|
1112 | 0 | dataptr = data; |
1113 | 0 | ctr = 0; |
1114 | 0 | for (;;) { |
1115 | 0 | elemptr = sample_data[ctr] + start_col; |
1116 | | |
1117 | | /* Even part */ |
1118 | |
|
1119 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[9]); |
1120 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[8]); |
1121 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[7]); |
1122 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[6]); |
1123 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[5]); |
1124 | |
|
1125 | 0 | tmp10 = tmp0 + tmp4; |
1126 | 0 | tmp13 = tmp0 - tmp4; |
1127 | 0 | tmp11 = tmp1 + tmp3; |
1128 | 0 | tmp14 = tmp1 - tmp3; |
1129 | |
|
1130 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[9]); |
1131 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[8]); |
1132 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[7]); |
1133 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[6]); |
1134 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[5]); |
1135 | | |
1136 | | /* Apply unsigned->signed conversion. */ |
1137 | 0 | dataptr[0] = (DCTELEM) |
1138 | 0 | ((tmp10 + tmp11 + tmp12 - 10 * CENTERJSAMPLE) << 1); |
1139 | 0 | tmp12 += tmp12; |
1140 | 0 | dataptr[4] = (DCTELEM) |
1141 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.144122806)) - /* c4 */ |
1142 | 0 | MULTIPLY(tmp11 - tmp12, FIX(0.437016024)), /* c8 */ |
1143 | 0 | CONST_BITS-1); |
1144 | 0 | tmp10 = MULTIPLY(tmp13 + tmp14, FIX(0.831253876)); /* c6 */ |
1145 | 0 | dataptr[2] = (DCTELEM) |
1146 | 0 | DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.513743148)), /* c2-c6 */ |
1147 | 0 | CONST_BITS-1); |
1148 | 0 | dataptr[6] = (DCTELEM) |
1149 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.176250899)), /* c2+c6 */ |
1150 | 0 | CONST_BITS-1); |
1151 | | |
1152 | | /* Odd part */ |
1153 | |
|
1154 | 0 | tmp10 = tmp0 + tmp4; |
1155 | 0 | tmp11 = tmp1 - tmp3; |
1156 | 0 | dataptr[5] = (DCTELEM) ((tmp10 - tmp11 - tmp2) << 1); |
1157 | 0 | tmp2 <<= CONST_BITS; |
1158 | 0 | dataptr[1] = (DCTELEM) |
1159 | 0 | DESCALE(MULTIPLY(tmp0, FIX(1.396802247)) + /* c1 */ |
1160 | 0 | MULTIPLY(tmp1, FIX(1.260073511)) + tmp2 + /* c3 */ |
1161 | 0 | MULTIPLY(tmp3, FIX(0.642039522)) + /* c7 */ |
1162 | 0 | MULTIPLY(tmp4, FIX(0.221231742)), /* c9 */ |
1163 | 0 | CONST_BITS-1); |
1164 | 0 | tmp12 = MULTIPLY(tmp0 - tmp4, FIX(0.951056516)) - /* (c3+c7)/2 */ |
1165 | 0 | MULTIPLY(tmp1 + tmp3, FIX(0.587785252)); /* (c1-c9)/2 */ |
1166 | 0 | tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.309016994)) + /* (c3-c7)/2 */ |
1167 | 0 | (tmp11 << (CONST_BITS - 1)) - tmp2; |
1168 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS-1); |
1169 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS-1); |
1170 | |
|
1171 | 0 | ctr++; |
1172 | |
|
1173 | 0 | if (ctr != DCTSIZE) { |
1174 | 0 | if (ctr == 10) |
1175 | 0 | break; /* Done. */ |
1176 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
1177 | 0 | } else |
1178 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
1179 | 0 | } |
1180 | | |
1181 | | /* Pass 2: process columns. |
1182 | | * We leave the results scaled up by an overall factor of 8. |
1183 | | * We must also scale the output by (8/10)**2 = 16/25, which we partially |
1184 | | * fold into the constant multipliers and final/initial shifting: |
1185 | | * cK now represents sqrt(2) * cos(K*pi/20) * 32/25. |
1186 | | */ |
1187 | |
|
1188 | 0 | dataptr = data; |
1189 | 0 | wsptr = workspace; |
1190 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
1191 | | /* Even part */ |
1192 | |
|
1193 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*1]; |
1194 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*0]; |
1195 | 0 | tmp12 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*7]; |
1196 | 0 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*6]; |
1197 | 0 | tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5]; |
1198 | |
|
1199 | 0 | tmp10 = tmp0 + tmp4; |
1200 | 0 | tmp13 = tmp0 - tmp4; |
1201 | 0 | tmp11 = tmp1 + tmp3; |
1202 | 0 | tmp14 = tmp1 - tmp3; |
1203 | |
|
1204 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*1]; |
1205 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*0]; |
1206 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*7]; |
1207 | 0 | tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*6]; |
1208 | 0 | tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5]; |
1209 | |
|
1210 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
1211 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(1.28)), /* 32/25 */ |
1212 | 0 | CONST_BITS+2); |
1213 | 0 | tmp12 += tmp12; |
1214 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
1215 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.464477191)) - /* c4 */ |
1216 | 0 | MULTIPLY(tmp11 - tmp12, FIX(0.559380511)), /* c8 */ |
1217 | 0 | CONST_BITS+2); |
1218 | 0 | tmp10 = MULTIPLY(tmp13 + tmp14, FIX(1.064004961)); /* c6 */ |
1219 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
1220 | 0 | DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.657591230)), /* c2-c6 */ |
1221 | 0 | CONST_BITS+2); |
1222 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
1223 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.785601151)), /* c2+c6 */ |
1224 | 0 | CONST_BITS+2); |
1225 | | |
1226 | | /* Odd part */ |
1227 | |
|
1228 | 0 | tmp10 = tmp0 + tmp4; |
1229 | 0 | tmp11 = tmp1 - tmp3; |
1230 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
1231 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp2, FIX(1.28)), /* 32/25 */ |
1232 | 0 | CONST_BITS+2); |
1233 | 0 | tmp2 = MULTIPLY(tmp2, FIX(1.28)); /* 32/25 */ |
1234 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
1235 | 0 | DESCALE(MULTIPLY(tmp0, FIX(1.787906876)) + /* c1 */ |
1236 | 0 | MULTIPLY(tmp1, FIX(1.612894094)) + tmp2 + /* c3 */ |
1237 | 0 | MULTIPLY(tmp3, FIX(0.821810588)) + /* c7 */ |
1238 | 0 | MULTIPLY(tmp4, FIX(0.283176630)), /* c9 */ |
1239 | 0 | CONST_BITS+2); |
1240 | 0 | tmp12 = MULTIPLY(tmp0 - tmp4, FIX(1.217352341)) - /* (c3+c7)/2 */ |
1241 | 0 | MULTIPLY(tmp1 + tmp3, FIX(0.752365123)); /* (c1-c9)/2 */ |
1242 | 0 | tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.395541753)) + /* (c3-c7)/2 */ |
1243 | 0 | MULTIPLY(tmp11, FIX(0.64)) - tmp2; /* 16/25 */ |
1244 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS+2); |
1245 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS+2); |
1246 | |
|
1247 | 0 | dataptr++; /* advance pointer to next column */ |
1248 | 0 | wsptr++; /* advance pointer to next column */ |
1249 | 0 | } |
1250 | 0 | } |
1251 | | |
1252 | | |
1253 | | /* |
1254 | | * Perform the forward DCT on an 11x11 sample block. |
1255 | | */ |
1256 | | |
1257 | | GLOBAL(void) |
1258 | | jpeg_fdct_11x11 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
1259 | 0 | { |
1260 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5; |
1261 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14; |
1262 | 0 | INT32 z1, z2, z3; |
1263 | 0 | DCTELEM workspace[8*3]; |
1264 | 0 | DCTELEM *dataptr; |
1265 | 0 | DCTELEM *wsptr; |
1266 | 0 | JSAMPROW elemptr; |
1267 | 0 | int ctr; |
1268 | 0 | SHIFT_TEMPS |
1269 | | |
1270 | | /* Pass 1: process rows. |
1271 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
1272 | | * we scale the results further by 2 as part of output adaption |
1273 | | * scaling for different DCT size. |
1274 | | * cK represents sqrt(2) * cos(K*pi/22). |
1275 | | */ |
1276 | |
|
1277 | 0 | dataptr = data; |
1278 | 0 | ctr = 0; |
1279 | 0 | for (;;) { |
1280 | 0 | elemptr = sample_data[ctr] + start_col; |
1281 | | |
1282 | | /* Even part */ |
1283 | |
|
1284 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[10]); |
1285 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[9]); |
1286 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[8]); |
1287 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[7]); |
1288 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[6]); |
1289 | 0 | tmp5 = GETJSAMPLE(elemptr[5]); |
1290 | |
|
1291 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[10]); |
1292 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[9]); |
1293 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[8]); |
1294 | 0 | tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[7]); |
1295 | 0 | tmp14 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[6]); |
1296 | | |
1297 | | /* Apply unsigned->signed conversion. */ |
1298 | 0 | dataptr[0] = (DCTELEM) |
1299 | 0 | ((tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5 - 11 * CENTERJSAMPLE) << 1); |
1300 | 0 | tmp5 += tmp5; |
1301 | 0 | tmp0 -= tmp5; |
1302 | 0 | tmp1 -= tmp5; |
1303 | 0 | tmp2 -= tmp5; |
1304 | 0 | tmp3 -= tmp5; |
1305 | 0 | tmp4 -= tmp5; |
1306 | 0 | z1 = MULTIPLY(tmp0 + tmp3, FIX(1.356927976)) + /* c2 */ |
1307 | 0 | MULTIPLY(tmp2 + tmp4, FIX(0.201263574)); /* c10 */ |
1308 | 0 | z2 = MULTIPLY(tmp1 - tmp3, FIX(0.926112931)); /* c6 */ |
1309 | 0 | z3 = MULTIPLY(tmp0 - tmp1, FIX(1.189712156)); /* c4 */ |
1310 | 0 | dataptr[2] = (DCTELEM) |
1311 | 0 | DESCALE(z1 + z2 - MULTIPLY(tmp3, FIX(1.018300590)) /* c2+c8-c6 */ |
1312 | 0 | - MULTIPLY(tmp4, FIX(1.390975730)), /* c4+c10 */ |
1313 | 0 | CONST_BITS-1); |
1314 | 0 | dataptr[4] = (DCTELEM) |
1315 | 0 | DESCALE(z2 + z3 + MULTIPLY(tmp1, FIX(0.062335650)) /* c4-c6-c10 */ |
1316 | 0 | - MULTIPLY(tmp2, FIX(1.356927976)) /* c2 */ |
1317 | 0 | + MULTIPLY(tmp4, FIX(0.587485545)), /* c8 */ |
1318 | 0 | CONST_BITS-1); |
1319 | 0 | dataptr[6] = (DCTELEM) |
1320 | 0 | DESCALE(z1 + z3 - MULTIPLY(tmp0, FIX(1.620527200)) /* c2+c4-c6 */ |
1321 | 0 | - MULTIPLY(tmp2, FIX(0.788749120)), /* c8+c10 */ |
1322 | 0 | CONST_BITS-1); |
1323 | | |
1324 | | /* Odd part */ |
1325 | |
|
1326 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.286413905)); /* c3 */ |
1327 | 0 | tmp2 = MULTIPLY(tmp10 + tmp12, FIX(1.068791298)); /* c5 */ |
1328 | 0 | tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.764581576)); /* c7 */ |
1329 | 0 | tmp0 = tmp1 + tmp2 + tmp3 - MULTIPLY(tmp10, FIX(1.719967871)) /* c7+c5+c3-c1 */ |
1330 | 0 | + MULTIPLY(tmp14, FIX(0.398430003)); /* c9 */ |
1331 | 0 | tmp4 = MULTIPLY(tmp11 + tmp12, - FIX(0.764581576)); /* -c7 */ |
1332 | 0 | tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(1.399818907)); /* -c1 */ |
1333 | 0 | tmp1 += tmp4 + tmp5 + MULTIPLY(tmp11, FIX(1.276416582)) /* c9+c7+c1-c3 */ |
1334 | 0 | - MULTIPLY(tmp14, FIX(1.068791298)); /* c5 */ |
1335 | 0 | tmp10 = MULTIPLY(tmp12 + tmp13, FIX(0.398430003)); /* c9 */ |
1336 | 0 | tmp2 += tmp4 + tmp10 - MULTIPLY(tmp12, FIX(1.989053629)) /* c9+c5+c3-c7 */ |
1337 | 0 | + MULTIPLY(tmp14, FIX(1.399818907)); /* c1 */ |
1338 | 0 | tmp3 += tmp5 + tmp10 + MULTIPLY(tmp13, FIX(1.305598626)) /* c1+c5-c9-c7 */ |
1339 | 0 | - MULTIPLY(tmp14, FIX(1.286413905)); /* c3 */ |
1340 | |
|
1341 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-1); |
1342 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-1); |
1343 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS-1); |
1344 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp3, CONST_BITS-1); |
1345 | |
|
1346 | 0 | ctr++; |
1347 | |
|
1348 | 0 | if (ctr != DCTSIZE) { |
1349 | 0 | if (ctr == 11) |
1350 | 0 | break; /* Done. */ |
1351 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
1352 | 0 | } else |
1353 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
1354 | 0 | } |
1355 | | |
1356 | | /* Pass 2: process columns. |
1357 | | * We leave the results scaled up by an overall factor of 8. |
1358 | | * We must also scale the output by (8/11)**2 = 64/121, which we partially |
1359 | | * fold into the constant multipliers and final/initial shifting: |
1360 | | * cK now represents sqrt(2) * cos(K*pi/22) * 128/121. |
1361 | | */ |
1362 | |
|
1363 | 0 | dataptr = data; |
1364 | 0 | wsptr = workspace; |
1365 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
1366 | | /* Even part */ |
1367 | |
|
1368 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*2]; |
1369 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*1]; |
1370 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*0]; |
1371 | 0 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*7]; |
1372 | 0 | tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*6]; |
1373 | 0 | tmp5 = dataptr[DCTSIZE*5]; |
1374 | |
|
1375 | 0 | tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*2]; |
1376 | 0 | tmp11 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*1]; |
1377 | 0 | tmp12 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*0]; |
1378 | 0 | tmp13 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*7]; |
1379 | 0 | tmp14 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*6]; |
1380 | |
|
1381 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
1382 | 0 | DESCALE(MULTIPLY(tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5, |
1383 | 0 | FIX(1.057851240)), /* 128/121 */ |
1384 | 0 | CONST_BITS+2); |
1385 | 0 | tmp5 += tmp5; |
1386 | 0 | tmp0 -= tmp5; |
1387 | 0 | tmp1 -= tmp5; |
1388 | 0 | tmp2 -= tmp5; |
1389 | 0 | tmp3 -= tmp5; |
1390 | 0 | tmp4 -= tmp5; |
1391 | 0 | z1 = MULTIPLY(tmp0 + tmp3, FIX(1.435427942)) + /* c2 */ |
1392 | 0 | MULTIPLY(tmp2 + tmp4, FIX(0.212906922)); /* c10 */ |
1393 | 0 | z2 = MULTIPLY(tmp1 - tmp3, FIX(0.979689713)); /* c6 */ |
1394 | 0 | z3 = MULTIPLY(tmp0 - tmp1, FIX(1.258538479)); /* c4 */ |
1395 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
1396 | 0 | DESCALE(z1 + z2 - MULTIPLY(tmp3, FIX(1.077210542)) /* c2+c8-c6 */ |
1397 | 0 | - MULTIPLY(tmp4, FIX(1.471445400)), /* c4+c10 */ |
1398 | 0 | CONST_BITS+2); |
1399 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
1400 | 0 | DESCALE(z2 + z3 + MULTIPLY(tmp1, FIX(0.065941844)) /* c4-c6-c10 */ |
1401 | 0 | - MULTIPLY(tmp2, FIX(1.435427942)) /* c2 */ |
1402 | 0 | + MULTIPLY(tmp4, FIX(0.621472312)), /* c8 */ |
1403 | 0 | CONST_BITS+2); |
1404 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
1405 | 0 | DESCALE(z1 + z3 - MULTIPLY(tmp0, FIX(1.714276708)) /* c2+c4-c6 */ |
1406 | 0 | - MULTIPLY(tmp2, FIX(0.834379234)), /* c8+c10 */ |
1407 | 0 | CONST_BITS+2); |
1408 | | |
1409 | | /* Odd part */ |
1410 | |
|
1411 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.360834544)); /* c3 */ |
1412 | 0 | tmp2 = MULTIPLY(tmp10 + tmp12, FIX(1.130622199)); /* c5 */ |
1413 | 0 | tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.808813568)); /* c7 */ |
1414 | 0 | tmp0 = tmp1 + tmp2 + tmp3 - MULTIPLY(tmp10, FIX(1.819470145)) /* c7+c5+c3-c1 */ |
1415 | 0 | + MULTIPLY(tmp14, FIX(0.421479672)); /* c9 */ |
1416 | 0 | tmp4 = MULTIPLY(tmp11 + tmp12, - FIX(0.808813568)); /* -c7 */ |
1417 | 0 | tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(1.480800167)); /* -c1 */ |
1418 | 0 | tmp1 += tmp4 + tmp5 + MULTIPLY(tmp11, FIX(1.350258864)) /* c9+c7+c1-c3 */ |
1419 | 0 | - MULTIPLY(tmp14, FIX(1.130622199)); /* c5 */ |
1420 | 0 | tmp10 = MULTIPLY(tmp12 + tmp13, FIX(0.421479672)); /* c9 */ |
1421 | 0 | tmp2 += tmp4 + tmp10 - MULTIPLY(tmp12, FIX(2.104122847)) /* c9+c5+c3-c7 */ |
1422 | 0 | + MULTIPLY(tmp14, FIX(1.480800167)); /* c1 */ |
1423 | 0 | tmp3 += tmp5 + tmp10 + MULTIPLY(tmp13, FIX(1.381129125)) /* c1+c5-c9-c7 */ |
1424 | 0 | - MULTIPLY(tmp14, FIX(1.360834544)); /* c3 */ |
1425 | |
|
1426 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+2); |
1427 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+2); |
1428 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+2); |
1429 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+2); |
1430 | |
|
1431 | 0 | dataptr++; /* advance pointer to next column */ |
1432 | 0 | wsptr++; /* advance pointer to next column */ |
1433 | 0 | } |
1434 | 0 | } |
1435 | | |
1436 | | |
1437 | | /* |
1438 | | * Perform the forward DCT on a 12x12 sample block. |
1439 | | */ |
1440 | | |
1441 | | GLOBAL(void) |
1442 | | jpeg_fdct_12x12 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
1443 | 0 | { |
1444 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5; |
1445 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15; |
1446 | 0 | DCTELEM workspace[8*4]; |
1447 | 0 | DCTELEM *dataptr; |
1448 | 0 | DCTELEM *wsptr; |
1449 | 0 | JSAMPROW elemptr; |
1450 | 0 | int ctr; |
1451 | 0 | SHIFT_TEMPS |
1452 | | |
1453 | | /* Pass 1: process rows. |
1454 | | * Note results are scaled up by sqrt(8) compared to a true DCT. |
1455 | | * cK represents sqrt(2) * cos(K*pi/24). |
1456 | | */ |
1457 | |
|
1458 | 0 | dataptr = data; |
1459 | 0 | ctr = 0; |
1460 | 0 | for (;;) { |
1461 | 0 | elemptr = sample_data[ctr] + start_col; |
1462 | | |
1463 | | /* Even part */ |
1464 | |
|
1465 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[11]); |
1466 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[10]); |
1467 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[9]); |
1468 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[8]); |
1469 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[7]); |
1470 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[6]); |
1471 | |
|
1472 | 0 | tmp10 = tmp0 + tmp5; |
1473 | 0 | tmp13 = tmp0 - tmp5; |
1474 | 0 | tmp11 = tmp1 + tmp4; |
1475 | 0 | tmp14 = tmp1 - tmp4; |
1476 | 0 | tmp12 = tmp2 + tmp3; |
1477 | 0 | tmp15 = tmp2 - tmp3; |
1478 | |
|
1479 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[11]); |
1480 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[10]); |
1481 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[9]); |
1482 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[8]); |
1483 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[7]); |
1484 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[6]); |
1485 | | |
1486 | | /* Apply unsigned->signed conversion. */ |
1487 | 0 | dataptr[0] = (DCTELEM) (tmp10 + tmp11 + tmp12 - 12 * CENTERJSAMPLE); |
1488 | 0 | dataptr[6] = (DCTELEM) (tmp13 - tmp14 - tmp15); |
1489 | 0 | dataptr[4] = (DCTELEM) |
1490 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.224744871)), /* c4 */ |
1491 | 0 | CONST_BITS); |
1492 | 0 | dataptr[2] = (DCTELEM) |
1493 | 0 | DESCALE(tmp14 - tmp15 + MULTIPLY(tmp13 + tmp15, FIX(1.366025404)), /* c2 */ |
1494 | 0 | CONST_BITS); |
1495 | | |
1496 | | /* Odd part */ |
1497 | |
|
1498 | 0 | tmp10 = MULTIPLY(tmp1 + tmp4, FIX_0_541196100); /* c9 */ |
1499 | 0 | tmp14 = tmp10 + MULTIPLY(tmp1, FIX_0_765366865); /* c3-c9 */ |
1500 | 0 | tmp15 = tmp10 - MULTIPLY(tmp4, FIX_1_847759065); /* c3+c9 */ |
1501 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.121971054)); /* c5 */ |
1502 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.860918669)); /* c7 */ |
1503 | 0 | tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.580774953)) /* c5+c7-c1 */ |
1504 | 0 | + MULTIPLY(tmp5, FIX(0.184591911)); /* c11 */ |
1505 | 0 | tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.184591911)); /* -c11 */ |
1506 | 0 | tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.339493912)) /* c1+c5-c11 */ |
1507 | 0 | + MULTIPLY(tmp5, FIX(0.860918669)); /* c7 */ |
1508 | 0 | tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.725788011)) /* c1+c11-c7 */ |
1509 | 0 | - MULTIPLY(tmp5, FIX(1.121971054)); /* c5 */ |
1510 | 0 | tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.306562965)) /* c3 */ |
1511 | 0 | - MULTIPLY(tmp2 + tmp5, FIX_0_541196100); /* c9 */ |
1512 | |
|
1513 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS); |
1514 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS); |
1515 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS); |
1516 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS); |
1517 | |
|
1518 | 0 | ctr++; |
1519 | |
|
1520 | 0 | if (ctr != DCTSIZE) { |
1521 | 0 | if (ctr == 12) |
1522 | 0 | break; /* Done. */ |
1523 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
1524 | 0 | } else |
1525 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
1526 | 0 | } |
1527 | | |
1528 | | /* Pass 2: process columns. |
1529 | | * We leave the results scaled up by an overall factor of 8. |
1530 | | * We must also scale the output by (8/12)**2 = 4/9, which we partially |
1531 | | * fold into the constant multipliers and final shifting: |
1532 | | * cK now represents sqrt(2) * cos(K*pi/24) * 8/9. |
1533 | | */ |
1534 | |
|
1535 | 0 | dataptr = data; |
1536 | 0 | wsptr = workspace; |
1537 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
1538 | | /* Even part */ |
1539 | |
|
1540 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*3]; |
1541 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*2]; |
1542 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*1]; |
1543 | 0 | tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*0]; |
1544 | 0 | tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*7]; |
1545 | 0 | tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*6]; |
1546 | |
|
1547 | 0 | tmp10 = tmp0 + tmp5; |
1548 | 0 | tmp13 = tmp0 - tmp5; |
1549 | 0 | tmp11 = tmp1 + tmp4; |
1550 | 0 | tmp14 = tmp1 - tmp4; |
1551 | 0 | tmp12 = tmp2 + tmp3; |
1552 | 0 | tmp15 = tmp2 - tmp3; |
1553 | |
|
1554 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*3]; |
1555 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*2]; |
1556 | 0 | tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*1]; |
1557 | 0 | tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*0]; |
1558 | 0 | tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*7]; |
1559 | 0 | tmp5 = dataptr[DCTSIZE*5] - dataptr[DCTSIZE*6]; |
1560 | |
|
1561 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
1562 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(0.888888889)), /* 8/9 */ |
1563 | 0 | CONST_BITS+1); |
1564 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
1565 | 0 | DESCALE(MULTIPLY(tmp13 - tmp14 - tmp15, FIX(0.888888889)), /* 8/9 */ |
1566 | 0 | CONST_BITS+1); |
1567 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
1568 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.088662108)), /* c4 */ |
1569 | 0 | CONST_BITS+1); |
1570 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
1571 | 0 | DESCALE(MULTIPLY(tmp14 - tmp15, FIX(0.888888889)) + /* 8/9 */ |
1572 | 0 | MULTIPLY(tmp13 + tmp15, FIX(1.214244803)), /* c2 */ |
1573 | 0 | CONST_BITS+1); |
1574 | | |
1575 | | /* Odd part */ |
1576 | |
|
1577 | 0 | tmp10 = MULTIPLY(tmp1 + tmp4, FIX(0.481063200)); /* c9 */ |
1578 | 0 | tmp14 = tmp10 + MULTIPLY(tmp1, FIX(0.680326102)); /* c3-c9 */ |
1579 | 0 | tmp15 = tmp10 - MULTIPLY(tmp4, FIX(1.642452502)); /* c3+c9 */ |
1580 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(0.997307603)); /* c5 */ |
1581 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.765261039)); /* c7 */ |
1582 | 0 | tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.516244403)) /* c5+c7-c1 */ |
1583 | 0 | + MULTIPLY(tmp5, FIX(0.164081699)); /* c11 */ |
1584 | 0 | tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.164081699)); /* -c11 */ |
1585 | 0 | tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.079550144)) /* c1+c5-c11 */ |
1586 | 0 | + MULTIPLY(tmp5, FIX(0.765261039)); /* c7 */ |
1587 | 0 | tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.645144899)) /* c1+c11-c7 */ |
1588 | 0 | - MULTIPLY(tmp5, FIX(0.997307603)); /* c5 */ |
1589 | 0 | tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.161389302)) /* c3 */ |
1590 | 0 | - MULTIPLY(tmp2 + tmp5, FIX(0.481063200)); /* c9 */ |
1591 | |
|
1592 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+1); |
1593 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+1); |
1594 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+1); |
1595 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+1); |
1596 | |
|
1597 | 0 | dataptr++; /* advance pointer to next column */ |
1598 | 0 | wsptr++; /* advance pointer to next column */ |
1599 | 0 | } |
1600 | 0 | } |
1601 | | |
1602 | | |
1603 | | /* |
1604 | | * Perform the forward DCT on a 13x13 sample block. |
1605 | | */ |
1606 | | |
1607 | | GLOBAL(void) |
1608 | | jpeg_fdct_13x13 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
1609 | 0 | { |
1610 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6; |
1611 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15; |
1612 | 0 | INT32 z1, z2; |
1613 | 0 | DCTELEM workspace[8*5]; |
1614 | 0 | DCTELEM *dataptr; |
1615 | 0 | DCTELEM *wsptr; |
1616 | 0 | JSAMPROW elemptr; |
1617 | 0 | int ctr; |
1618 | 0 | SHIFT_TEMPS |
1619 | | |
1620 | | /* Pass 1: process rows. |
1621 | | * Note results are scaled up by sqrt(8) compared to a true DCT. |
1622 | | * cK represents sqrt(2) * cos(K*pi/26). |
1623 | | */ |
1624 | |
|
1625 | 0 | dataptr = data; |
1626 | 0 | ctr = 0; |
1627 | 0 | for (;;) { |
1628 | 0 | elemptr = sample_data[ctr] + start_col; |
1629 | | |
1630 | | /* Even part */ |
1631 | |
|
1632 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[12]); |
1633 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[11]); |
1634 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[10]); |
1635 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[9]); |
1636 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[8]); |
1637 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[7]); |
1638 | 0 | tmp6 = GETJSAMPLE(elemptr[6]); |
1639 | |
|
1640 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[12]); |
1641 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[11]); |
1642 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[10]); |
1643 | 0 | tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[9]); |
1644 | 0 | tmp14 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[8]); |
1645 | 0 | tmp15 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[7]); |
1646 | | |
1647 | | /* Apply unsigned->signed conversion. */ |
1648 | 0 | dataptr[0] = (DCTELEM) |
1649 | 0 | (tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5 + tmp6 - 13 * CENTERJSAMPLE); |
1650 | 0 | tmp6 += tmp6; |
1651 | 0 | tmp0 -= tmp6; |
1652 | 0 | tmp1 -= tmp6; |
1653 | 0 | tmp2 -= tmp6; |
1654 | 0 | tmp3 -= tmp6; |
1655 | 0 | tmp4 -= tmp6; |
1656 | 0 | tmp5 -= tmp6; |
1657 | 0 | dataptr[2] = (DCTELEM) |
1658 | 0 | DESCALE(MULTIPLY(tmp0, FIX(1.373119086)) + /* c2 */ |
1659 | 0 | MULTIPLY(tmp1, FIX(1.058554052)) + /* c6 */ |
1660 | 0 | MULTIPLY(tmp2, FIX(0.501487041)) - /* c10 */ |
1661 | 0 | MULTIPLY(tmp3, FIX(0.170464608)) - /* c12 */ |
1662 | 0 | MULTIPLY(tmp4, FIX(0.803364869)) - /* c8 */ |
1663 | 0 | MULTIPLY(tmp5, FIX(1.252223920)), /* c4 */ |
1664 | 0 | CONST_BITS); |
1665 | 0 | z1 = MULTIPLY(tmp0 - tmp2, FIX(1.155388986)) - /* (c4+c6)/2 */ |
1666 | 0 | MULTIPLY(tmp3 - tmp4, FIX(0.435816023)) - /* (c2-c10)/2 */ |
1667 | 0 | MULTIPLY(tmp1 - tmp5, FIX(0.316450131)); /* (c8-c12)/2 */ |
1668 | 0 | z2 = MULTIPLY(tmp0 + tmp2, FIX(0.096834934)) - /* (c4-c6)/2 */ |
1669 | 0 | MULTIPLY(tmp3 + tmp4, FIX(0.937303064)) + /* (c2+c10)/2 */ |
1670 | 0 | MULTIPLY(tmp1 + tmp5, FIX(0.486914739)); /* (c8+c12)/2 */ |
1671 | |
|
1672 | 0 | dataptr[4] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS); |
1673 | 0 | dataptr[6] = (DCTELEM) DESCALE(z1 - z2, CONST_BITS); |
1674 | | |
1675 | | /* Odd part */ |
1676 | |
|
1677 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.322312651)); /* c3 */ |
1678 | 0 | tmp2 = MULTIPLY(tmp10 + tmp12, FIX(1.163874945)); /* c5 */ |
1679 | 0 | tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.937797057)) + /* c7 */ |
1680 | 0 | MULTIPLY(tmp14 + tmp15, FIX(0.338443458)); /* c11 */ |
1681 | 0 | tmp0 = tmp1 + tmp2 + tmp3 - |
1682 | 0 | MULTIPLY(tmp10, FIX(2.020082300)) + /* c3+c5+c7-c1 */ |
1683 | 0 | MULTIPLY(tmp14, FIX(0.318774355)); /* c9-c11 */ |
1684 | 0 | tmp4 = MULTIPLY(tmp14 - tmp15, FIX(0.937797057)) - /* c7 */ |
1685 | 0 | MULTIPLY(tmp11 + tmp12, FIX(0.338443458)); /* c11 */ |
1686 | 0 | tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(1.163874945)); /* -c5 */ |
1687 | 0 | tmp1 += tmp4 + tmp5 + |
1688 | 0 | MULTIPLY(tmp11, FIX(0.837223564)) - /* c5+c9+c11-c3 */ |
1689 | 0 | MULTIPLY(tmp14, FIX(2.341699410)); /* c1+c7 */ |
1690 | 0 | tmp6 = MULTIPLY(tmp12 + tmp13, - FIX(0.657217813)); /* -c9 */ |
1691 | 0 | tmp2 += tmp4 + tmp6 - |
1692 | 0 | MULTIPLY(tmp12, FIX(1.572116027)) + /* c1+c5-c9-c11 */ |
1693 | 0 | MULTIPLY(tmp15, FIX(2.260109708)); /* c3+c7 */ |
1694 | 0 | tmp3 += tmp5 + tmp6 + |
1695 | 0 | MULTIPLY(tmp13, FIX(2.205608352)) - /* c3+c5+c9-c7 */ |
1696 | 0 | MULTIPLY(tmp15, FIX(1.742345811)); /* c1+c11 */ |
1697 | |
|
1698 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS); |
1699 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS); |
1700 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS); |
1701 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp3, CONST_BITS); |
1702 | |
|
1703 | 0 | ctr++; |
1704 | |
|
1705 | 0 | if (ctr != DCTSIZE) { |
1706 | 0 | if (ctr == 13) |
1707 | 0 | break; /* Done. */ |
1708 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
1709 | 0 | } else |
1710 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
1711 | 0 | } |
1712 | | |
1713 | | /* Pass 2: process columns. |
1714 | | * We leave the results scaled up by an overall factor of 8. |
1715 | | * We must also scale the output by (8/13)**2 = 64/169, which we partially |
1716 | | * fold into the constant multipliers and final shifting: |
1717 | | * cK now represents sqrt(2) * cos(K*pi/26) * 128/169. |
1718 | | */ |
1719 | |
|
1720 | 0 | dataptr = data; |
1721 | 0 | wsptr = workspace; |
1722 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
1723 | | /* Even part */ |
1724 | |
|
1725 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*4]; |
1726 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*3]; |
1727 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*2]; |
1728 | 0 | tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*1]; |
1729 | 0 | tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*0]; |
1730 | 0 | tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*7]; |
1731 | 0 | tmp6 = dataptr[DCTSIZE*6]; |
1732 | |
|
1733 | 0 | tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*4]; |
1734 | 0 | tmp11 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*3]; |
1735 | 0 | tmp12 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*2]; |
1736 | 0 | tmp13 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*1]; |
1737 | 0 | tmp14 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*0]; |
1738 | 0 | tmp15 = dataptr[DCTSIZE*5] - dataptr[DCTSIZE*7]; |
1739 | |
|
1740 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
1741 | 0 | DESCALE(MULTIPLY(tmp0 + tmp1 + tmp2 + tmp3 + tmp4 + tmp5 + tmp6, |
1742 | 0 | FIX(0.757396450)), /* 128/169 */ |
1743 | 0 | CONST_BITS+1); |
1744 | 0 | tmp6 += tmp6; |
1745 | 0 | tmp0 -= tmp6; |
1746 | 0 | tmp1 -= tmp6; |
1747 | 0 | tmp2 -= tmp6; |
1748 | 0 | tmp3 -= tmp6; |
1749 | 0 | tmp4 -= tmp6; |
1750 | 0 | tmp5 -= tmp6; |
1751 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
1752 | 0 | DESCALE(MULTIPLY(tmp0, FIX(1.039995521)) + /* c2 */ |
1753 | 0 | MULTIPLY(tmp1, FIX(0.801745081)) + /* c6 */ |
1754 | 0 | MULTIPLY(tmp2, FIX(0.379824504)) - /* c10 */ |
1755 | 0 | MULTIPLY(tmp3, FIX(0.129109289)) - /* c12 */ |
1756 | 0 | MULTIPLY(tmp4, FIX(0.608465700)) - /* c8 */ |
1757 | 0 | MULTIPLY(tmp5, FIX(0.948429952)), /* c4 */ |
1758 | 0 | CONST_BITS+1); |
1759 | 0 | z1 = MULTIPLY(tmp0 - tmp2, FIX(0.875087516)) - /* (c4+c6)/2 */ |
1760 | 0 | MULTIPLY(tmp3 - tmp4, FIX(0.330085509)) - /* (c2-c10)/2 */ |
1761 | 0 | MULTIPLY(tmp1 - tmp5, FIX(0.239678205)); /* (c8-c12)/2 */ |
1762 | 0 | z2 = MULTIPLY(tmp0 + tmp2, FIX(0.073342435)) - /* (c4-c6)/2 */ |
1763 | 0 | MULTIPLY(tmp3 + tmp4, FIX(0.709910013)) + /* (c2+c10)/2 */ |
1764 | 0 | MULTIPLY(tmp1 + tmp5, FIX(0.368787494)); /* (c8+c12)/2 */ |
1765 | |
|
1766 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS+1); |
1767 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 - z2, CONST_BITS+1); |
1768 | | |
1769 | | /* Odd part */ |
1770 | |
|
1771 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.001514908)); /* c3 */ |
1772 | 0 | tmp2 = MULTIPLY(tmp10 + tmp12, FIX(0.881514751)); /* c5 */ |
1773 | 0 | tmp3 = MULTIPLY(tmp10 + tmp13, FIX(0.710284161)) + /* c7 */ |
1774 | 0 | MULTIPLY(tmp14 + tmp15, FIX(0.256335874)); /* c11 */ |
1775 | 0 | tmp0 = tmp1 + tmp2 + tmp3 - |
1776 | 0 | MULTIPLY(tmp10, FIX(1.530003162)) + /* c3+c5+c7-c1 */ |
1777 | 0 | MULTIPLY(tmp14, FIX(0.241438564)); /* c9-c11 */ |
1778 | 0 | tmp4 = MULTIPLY(tmp14 - tmp15, FIX(0.710284161)) - /* c7 */ |
1779 | 0 | MULTIPLY(tmp11 + tmp12, FIX(0.256335874)); /* c11 */ |
1780 | 0 | tmp5 = MULTIPLY(tmp11 + tmp13, - FIX(0.881514751)); /* -c5 */ |
1781 | 0 | tmp1 += tmp4 + tmp5 + |
1782 | 0 | MULTIPLY(tmp11, FIX(0.634110155)) - /* c5+c9+c11-c3 */ |
1783 | 0 | MULTIPLY(tmp14, FIX(1.773594819)); /* c1+c7 */ |
1784 | 0 | tmp6 = MULTIPLY(tmp12 + tmp13, - FIX(0.497774438)); /* -c9 */ |
1785 | 0 | tmp2 += tmp4 + tmp6 - |
1786 | 0 | MULTIPLY(tmp12, FIX(1.190715098)) + /* c1+c5-c9-c11 */ |
1787 | 0 | MULTIPLY(tmp15, FIX(1.711799069)); /* c3+c7 */ |
1788 | 0 | tmp3 += tmp5 + tmp6 + |
1789 | 0 | MULTIPLY(tmp13, FIX(1.670519935)) - /* c3+c5+c9-c7 */ |
1790 | 0 | MULTIPLY(tmp15, FIX(1.319646532)); /* c1+c11 */ |
1791 | |
|
1792 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+1); |
1793 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+1); |
1794 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+1); |
1795 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+1); |
1796 | |
|
1797 | 0 | dataptr++; /* advance pointer to next column */ |
1798 | 0 | wsptr++; /* advance pointer to next column */ |
1799 | 0 | } |
1800 | 0 | } |
1801 | | |
1802 | | |
1803 | | /* |
1804 | | * Perform the forward DCT on a 14x14 sample block. |
1805 | | */ |
1806 | | |
1807 | | GLOBAL(void) |
1808 | | jpeg_fdct_14x14 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
1809 | 0 | { |
1810 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6; |
1811 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16; |
1812 | 0 | DCTELEM workspace[8*6]; |
1813 | 0 | DCTELEM *dataptr; |
1814 | 0 | DCTELEM *wsptr; |
1815 | 0 | JSAMPROW elemptr; |
1816 | 0 | int ctr; |
1817 | 0 | SHIFT_TEMPS |
1818 | | |
1819 | | /* Pass 1: process rows. |
1820 | | * Note results are scaled up by sqrt(8) compared to a true DCT. |
1821 | | * cK represents sqrt(2) * cos(K*pi/28). |
1822 | | */ |
1823 | |
|
1824 | 0 | dataptr = data; |
1825 | 0 | ctr = 0; |
1826 | 0 | for (;;) { |
1827 | 0 | elemptr = sample_data[ctr] + start_col; |
1828 | | |
1829 | | /* Even part */ |
1830 | |
|
1831 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[13]); |
1832 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[12]); |
1833 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[11]); |
1834 | 0 | tmp13 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[10]); |
1835 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[9]); |
1836 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[8]); |
1837 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[7]); |
1838 | |
|
1839 | 0 | tmp10 = tmp0 + tmp6; |
1840 | 0 | tmp14 = tmp0 - tmp6; |
1841 | 0 | tmp11 = tmp1 + tmp5; |
1842 | 0 | tmp15 = tmp1 - tmp5; |
1843 | 0 | tmp12 = tmp2 + tmp4; |
1844 | 0 | tmp16 = tmp2 - tmp4; |
1845 | |
|
1846 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[13]); |
1847 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[12]); |
1848 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[11]); |
1849 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[10]); |
1850 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[9]); |
1851 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[8]); |
1852 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[7]); |
1853 | | |
1854 | | /* Apply unsigned->signed conversion. */ |
1855 | 0 | dataptr[0] = (DCTELEM) |
1856 | 0 | (tmp10 + tmp11 + tmp12 + tmp13 - 14 * CENTERJSAMPLE); |
1857 | 0 | tmp13 += tmp13; |
1858 | 0 | dataptr[4] = (DCTELEM) |
1859 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.274162392)) + /* c4 */ |
1860 | 0 | MULTIPLY(tmp11 - tmp13, FIX(0.314692123)) - /* c12 */ |
1861 | 0 | MULTIPLY(tmp12 - tmp13, FIX(0.881747734)), /* c8 */ |
1862 | 0 | CONST_BITS); |
1863 | |
|
1864 | 0 | tmp10 = MULTIPLY(tmp14 + tmp15, FIX(1.105676686)); /* c6 */ |
1865 | |
|
1866 | 0 | dataptr[2] = (DCTELEM) |
1867 | 0 | DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.273079590)) /* c2-c6 */ |
1868 | 0 | + MULTIPLY(tmp16, FIX(0.613604268)), /* c10 */ |
1869 | 0 | CONST_BITS); |
1870 | 0 | dataptr[6] = (DCTELEM) |
1871 | 0 | DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.719280954)) /* c6+c10 */ |
1872 | 0 | - MULTIPLY(tmp16, FIX(1.378756276)), /* c2 */ |
1873 | 0 | CONST_BITS); |
1874 | | |
1875 | | /* Odd part */ |
1876 | |
|
1877 | 0 | tmp10 = tmp1 + tmp2; |
1878 | 0 | tmp11 = tmp5 - tmp4; |
1879 | 0 | dataptr[7] = (DCTELEM) (tmp0 - tmp10 + tmp3 - tmp11 - tmp6); |
1880 | 0 | tmp3 <<= CONST_BITS; |
1881 | 0 | tmp10 = MULTIPLY(tmp10, - FIX(0.158341681)); /* -c13 */ |
1882 | 0 | tmp11 = MULTIPLY(tmp11, FIX(1.405321284)); /* c1 */ |
1883 | 0 | tmp10 += tmp11 - tmp3; |
1884 | 0 | tmp11 = MULTIPLY(tmp0 + tmp2, FIX(1.197448846)) + /* c5 */ |
1885 | 0 | MULTIPLY(tmp4 + tmp6, FIX(0.752406978)); /* c9 */ |
1886 | 0 | dataptr[5] = (DCTELEM) |
1887 | 0 | DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(2.373959773)) /* c3+c5-c13 */ |
1888 | 0 | + MULTIPLY(tmp4, FIX(1.119999435)), /* c1+c11-c9 */ |
1889 | 0 | CONST_BITS); |
1890 | 0 | tmp12 = MULTIPLY(tmp0 + tmp1, FIX(1.334852607)) + /* c3 */ |
1891 | 0 | MULTIPLY(tmp5 - tmp6, FIX(0.467085129)); /* c11 */ |
1892 | 0 | dataptr[3] = (DCTELEM) |
1893 | 0 | DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.424103948)) /* c3-c9-c13 */ |
1894 | 0 | - MULTIPLY(tmp5, FIX(3.069855259)), /* c1+c5+c11 */ |
1895 | 0 | CONST_BITS); |
1896 | 0 | dataptr[1] = (DCTELEM) |
1897 | 0 | DESCALE(tmp11 + tmp12 + tmp3 + tmp6 - |
1898 | 0 | MULTIPLY(tmp0 + tmp6, FIX(1.126980169)), /* c3+c5-c1 */ |
1899 | 0 | CONST_BITS); |
1900 | |
|
1901 | 0 | ctr++; |
1902 | |
|
1903 | 0 | if (ctr != DCTSIZE) { |
1904 | 0 | if (ctr == 14) |
1905 | 0 | break; /* Done. */ |
1906 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
1907 | 0 | } else |
1908 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
1909 | 0 | } |
1910 | | |
1911 | | /* Pass 2: process columns. |
1912 | | * We leave the results scaled up by an overall factor of 8. |
1913 | | * We must also scale the output by (8/14)**2 = 16/49, which we partially |
1914 | | * fold into the constant multipliers and final shifting: |
1915 | | * cK now represents sqrt(2) * cos(K*pi/28) * 32/49. |
1916 | | */ |
1917 | |
|
1918 | 0 | dataptr = data; |
1919 | 0 | wsptr = workspace; |
1920 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
1921 | | /* Even part */ |
1922 | |
|
1923 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*5]; |
1924 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*4]; |
1925 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*3]; |
1926 | 0 | tmp13 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*2]; |
1927 | 0 | tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*1]; |
1928 | 0 | tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*0]; |
1929 | 0 | tmp6 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7]; |
1930 | |
|
1931 | 0 | tmp10 = tmp0 + tmp6; |
1932 | 0 | tmp14 = tmp0 - tmp6; |
1933 | 0 | tmp11 = tmp1 + tmp5; |
1934 | 0 | tmp15 = tmp1 - tmp5; |
1935 | 0 | tmp12 = tmp2 + tmp4; |
1936 | 0 | tmp16 = tmp2 - tmp4; |
1937 | |
|
1938 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*5]; |
1939 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*4]; |
1940 | 0 | tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*3]; |
1941 | 0 | tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*2]; |
1942 | 0 | tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*1]; |
1943 | 0 | tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*0]; |
1944 | 0 | tmp6 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7]; |
1945 | |
|
1946 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
1947 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12 + tmp13, |
1948 | 0 | FIX(0.653061224)), /* 32/49 */ |
1949 | 0 | CONST_BITS+1); |
1950 | 0 | tmp13 += tmp13; |
1951 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
1952 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(0.832106052)) + /* c4 */ |
1953 | 0 | MULTIPLY(tmp11 - tmp13, FIX(0.205513223)) - /* c12 */ |
1954 | 0 | MULTIPLY(tmp12 - tmp13, FIX(0.575835255)), /* c8 */ |
1955 | 0 | CONST_BITS+1); |
1956 | |
|
1957 | 0 | tmp10 = MULTIPLY(tmp14 + tmp15, FIX(0.722074570)); /* c6 */ |
1958 | |
|
1959 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
1960 | 0 | DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.178337691)) /* c2-c6 */ |
1961 | 0 | + MULTIPLY(tmp16, FIX(0.400721155)), /* c10 */ |
1962 | 0 | CONST_BITS+1); |
1963 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
1964 | 0 | DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.122795725)) /* c6+c10 */ |
1965 | 0 | - MULTIPLY(tmp16, FIX(0.900412262)), /* c2 */ |
1966 | 0 | CONST_BITS+1); |
1967 | | |
1968 | | /* Odd part */ |
1969 | |
|
1970 | 0 | tmp10 = tmp1 + tmp2; |
1971 | 0 | tmp11 = tmp5 - tmp4; |
1972 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) |
1973 | 0 | DESCALE(MULTIPLY(tmp0 - tmp10 + tmp3 - tmp11 - tmp6, |
1974 | 0 | FIX(0.653061224)), /* 32/49 */ |
1975 | 0 | CONST_BITS+1); |
1976 | 0 | tmp3 = MULTIPLY(tmp3 , FIX(0.653061224)); /* 32/49 */ |
1977 | 0 | tmp10 = MULTIPLY(tmp10, - FIX(0.103406812)); /* -c13 */ |
1978 | 0 | tmp11 = MULTIPLY(tmp11, FIX(0.917760839)); /* c1 */ |
1979 | 0 | tmp10 += tmp11 - tmp3; |
1980 | 0 | tmp11 = MULTIPLY(tmp0 + tmp2, FIX(0.782007410)) + /* c5 */ |
1981 | 0 | MULTIPLY(tmp4 + tmp6, FIX(0.491367823)); /* c9 */ |
1982 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
1983 | 0 | DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(1.550341076)) /* c3+c5-c13 */ |
1984 | 0 | + MULTIPLY(tmp4, FIX(0.731428202)), /* c1+c11-c9 */ |
1985 | 0 | CONST_BITS+1); |
1986 | 0 | tmp12 = MULTIPLY(tmp0 + tmp1, FIX(0.871740478)) + /* c3 */ |
1987 | 0 | MULTIPLY(tmp5 - tmp6, FIX(0.305035186)); /* c11 */ |
1988 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
1989 | 0 | DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.276965844)) /* c3-c9-c13 */ |
1990 | 0 | - MULTIPLY(tmp5, FIX(2.004803435)), /* c1+c5+c11 */ |
1991 | 0 | CONST_BITS+1); |
1992 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
1993 | 0 | DESCALE(tmp11 + tmp12 + tmp3 |
1994 | 0 | - MULTIPLY(tmp0, FIX(0.735987049)) /* c3+c5-c1 */ |
1995 | 0 | - MULTIPLY(tmp6, FIX(0.082925825)), /* c9-c11-c13 */ |
1996 | 0 | CONST_BITS+1); |
1997 | |
|
1998 | 0 | dataptr++; /* advance pointer to next column */ |
1999 | 0 | wsptr++; /* advance pointer to next column */ |
2000 | 0 | } |
2001 | 0 | } |
2002 | | |
2003 | | |
2004 | | /* |
2005 | | * Perform the forward DCT on a 15x15 sample block. |
2006 | | */ |
2007 | | |
2008 | | GLOBAL(void) |
2009 | | jpeg_fdct_15x15 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
2010 | 0 | { |
2011 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
2012 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16; |
2013 | 0 | INT32 z1, z2, z3; |
2014 | 0 | DCTELEM workspace[8*7]; |
2015 | 0 | DCTELEM *dataptr; |
2016 | 0 | DCTELEM *wsptr; |
2017 | 0 | JSAMPROW elemptr; |
2018 | 0 | int ctr; |
2019 | 0 | SHIFT_TEMPS |
2020 | | |
2021 | | /* Pass 1: process rows. |
2022 | | * Note results are scaled up by sqrt(8) compared to a true DCT. |
2023 | | * cK represents sqrt(2) * cos(K*pi/30). |
2024 | | */ |
2025 | |
|
2026 | 0 | dataptr = data; |
2027 | 0 | ctr = 0; |
2028 | 0 | for (;;) { |
2029 | 0 | elemptr = sample_data[ctr] + start_col; |
2030 | | |
2031 | | /* Even part */ |
2032 | |
|
2033 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[14]); |
2034 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[13]); |
2035 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[12]); |
2036 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[11]); |
2037 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[10]); |
2038 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[9]); |
2039 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[8]); |
2040 | 0 | tmp7 = GETJSAMPLE(elemptr[7]); |
2041 | |
|
2042 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[14]); |
2043 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[13]); |
2044 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[12]); |
2045 | 0 | tmp13 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[11]); |
2046 | 0 | tmp14 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[10]); |
2047 | 0 | tmp15 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[9]); |
2048 | 0 | tmp16 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[8]); |
2049 | |
|
2050 | 0 | z1 = tmp0 + tmp4 + tmp5; |
2051 | 0 | z2 = tmp1 + tmp3 + tmp6; |
2052 | 0 | z3 = tmp2 + tmp7; |
2053 | | /* Apply unsigned->signed conversion. */ |
2054 | 0 | dataptr[0] = (DCTELEM) (z1 + z2 + z3 - 15 * CENTERJSAMPLE); |
2055 | 0 | z3 += z3; |
2056 | 0 | dataptr[6] = (DCTELEM) |
2057 | 0 | DESCALE(MULTIPLY(z1 - z3, FIX(1.144122806)) - /* c6 */ |
2058 | 0 | MULTIPLY(z2 - z3, FIX(0.437016024)), /* c12 */ |
2059 | 0 | CONST_BITS); |
2060 | 0 | tmp2 += ((tmp1 + tmp4) >> 1) - tmp7 - tmp7; |
2061 | 0 | z1 = MULTIPLY(tmp3 - tmp2, FIX(1.531135173)) - /* c2+c14 */ |
2062 | 0 | MULTIPLY(tmp6 - tmp2, FIX(2.238241955)); /* c4+c8 */ |
2063 | 0 | z2 = MULTIPLY(tmp5 - tmp2, FIX(0.798468008)) - /* c8-c14 */ |
2064 | 0 | MULTIPLY(tmp0 - tmp2, FIX(0.091361227)); /* c2-c4 */ |
2065 | 0 | z3 = MULTIPLY(tmp0 - tmp3, FIX(1.383309603)) + /* c2 */ |
2066 | 0 | MULTIPLY(tmp6 - tmp5, FIX(0.946293579)) + /* c8 */ |
2067 | 0 | MULTIPLY(tmp1 - tmp4, FIX(0.790569415)); /* (c6+c12)/2 */ |
2068 | |
|
2069 | 0 | dataptr[2] = (DCTELEM) DESCALE(z1 + z3, CONST_BITS); |
2070 | 0 | dataptr[4] = (DCTELEM) DESCALE(z2 + z3, CONST_BITS); |
2071 | | |
2072 | | /* Odd part */ |
2073 | |
|
2074 | 0 | tmp2 = MULTIPLY(tmp10 - tmp12 - tmp13 + tmp15 + tmp16, |
2075 | 0 | FIX(1.224744871)); /* c5 */ |
2076 | 0 | tmp1 = MULTIPLY(tmp10 - tmp14 - tmp15, FIX(1.344997024)) + /* c3 */ |
2077 | 0 | MULTIPLY(tmp11 - tmp13 - tmp16, FIX(0.831253876)); /* c9 */ |
2078 | 0 | tmp12 = MULTIPLY(tmp12, FIX(1.224744871)); /* c5 */ |
2079 | 0 | tmp4 = MULTIPLY(tmp10 - tmp16, FIX(1.406466353)) + /* c1 */ |
2080 | 0 | MULTIPLY(tmp11 + tmp14, FIX(1.344997024)) + /* c3 */ |
2081 | 0 | MULTIPLY(tmp13 + tmp15, FIX(0.575212477)); /* c11 */ |
2082 | 0 | tmp0 = MULTIPLY(tmp13, FIX(0.475753014)) - /* c7-c11 */ |
2083 | 0 | MULTIPLY(tmp14, FIX(0.513743148)) + /* c3-c9 */ |
2084 | 0 | MULTIPLY(tmp16, FIX(1.700497885)) + tmp4 + tmp12; /* c1+c13 */ |
2085 | 0 | tmp3 = MULTIPLY(tmp10, - FIX(0.355500862)) - /* -(c1-c7) */ |
2086 | 0 | MULTIPLY(tmp11, FIX(2.176250899)) - /* c3+c9 */ |
2087 | 0 | MULTIPLY(tmp15, FIX(0.869244010)) + tmp4 - tmp12; /* c11+c13 */ |
2088 | |
|
2089 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS); |
2090 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS); |
2091 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS); |
2092 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp3, CONST_BITS); |
2093 | |
|
2094 | 0 | ctr++; |
2095 | |
|
2096 | 0 | if (ctr != DCTSIZE) { |
2097 | 0 | if (ctr == 15) |
2098 | 0 | break; /* Done. */ |
2099 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
2100 | 0 | } else |
2101 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
2102 | 0 | } |
2103 | | |
2104 | | /* Pass 2: process columns. |
2105 | | * We leave the results scaled up by an overall factor of 8. |
2106 | | * We must also scale the output by (8/15)**2 = 64/225, which we partially |
2107 | | * fold into the constant multipliers and final shifting: |
2108 | | * cK now represents sqrt(2) * cos(K*pi/30) * 256/225. |
2109 | | */ |
2110 | |
|
2111 | 0 | dataptr = data; |
2112 | 0 | wsptr = workspace; |
2113 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
2114 | | /* Even part */ |
2115 | |
|
2116 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*6]; |
2117 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*5]; |
2118 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*4]; |
2119 | 0 | tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*3]; |
2120 | 0 | tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*2]; |
2121 | 0 | tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*1]; |
2122 | 0 | tmp6 = dataptr[DCTSIZE*6] + wsptr[DCTSIZE*0]; |
2123 | 0 | tmp7 = dataptr[DCTSIZE*7]; |
2124 | |
|
2125 | 0 | tmp10 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*6]; |
2126 | 0 | tmp11 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*5]; |
2127 | 0 | tmp12 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*4]; |
2128 | 0 | tmp13 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*3]; |
2129 | 0 | tmp14 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*2]; |
2130 | 0 | tmp15 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*1]; |
2131 | 0 | tmp16 = dataptr[DCTSIZE*6] - wsptr[DCTSIZE*0]; |
2132 | |
|
2133 | 0 | z1 = tmp0 + tmp4 + tmp5; |
2134 | 0 | z2 = tmp1 + tmp3 + tmp6; |
2135 | 0 | z3 = tmp2 + tmp7; |
2136 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
2137 | 0 | DESCALE(MULTIPLY(z1 + z2 + z3, FIX(1.137777778)), /* 256/225 */ |
2138 | 0 | CONST_BITS+2); |
2139 | 0 | z3 += z3; |
2140 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
2141 | 0 | DESCALE(MULTIPLY(z1 - z3, FIX(1.301757503)) - /* c6 */ |
2142 | 0 | MULTIPLY(z2 - z3, FIX(0.497227121)), /* c12 */ |
2143 | 0 | CONST_BITS+2); |
2144 | 0 | tmp2 += ((tmp1 + tmp4) >> 1) - tmp7 - tmp7; |
2145 | 0 | z1 = MULTIPLY(tmp3 - tmp2, FIX(1.742091575)) - /* c2+c14 */ |
2146 | 0 | MULTIPLY(tmp6 - tmp2, FIX(2.546621957)); /* c4+c8 */ |
2147 | 0 | z2 = MULTIPLY(tmp5 - tmp2, FIX(0.908479156)) - /* c8-c14 */ |
2148 | 0 | MULTIPLY(tmp0 - tmp2, FIX(0.103948774)); /* c2-c4 */ |
2149 | 0 | z3 = MULTIPLY(tmp0 - tmp3, FIX(1.573898926)) + /* c2 */ |
2150 | 0 | MULTIPLY(tmp6 - tmp5, FIX(1.076671805)) + /* c8 */ |
2151 | 0 | MULTIPLY(tmp1 - tmp4, FIX(0.899492312)); /* (c6+c12)/2 */ |
2152 | |
|
2153 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z3, CONST_BITS+2); |
2154 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(z2 + z3, CONST_BITS+2); |
2155 | | |
2156 | | /* Odd part */ |
2157 | |
|
2158 | 0 | tmp2 = MULTIPLY(tmp10 - tmp12 - tmp13 + tmp15 + tmp16, |
2159 | 0 | FIX(1.393487498)); /* c5 */ |
2160 | 0 | tmp1 = MULTIPLY(tmp10 - tmp14 - tmp15, FIX(1.530307725)) + /* c3 */ |
2161 | 0 | MULTIPLY(tmp11 - tmp13 - tmp16, FIX(0.945782187)); /* c9 */ |
2162 | 0 | tmp12 = MULTIPLY(tmp12, FIX(1.393487498)); /* c5 */ |
2163 | 0 | tmp4 = MULTIPLY(tmp10 - tmp16, FIX(1.600246161)) + /* c1 */ |
2164 | 0 | MULTIPLY(tmp11 + tmp14, FIX(1.530307725)) + /* c3 */ |
2165 | 0 | MULTIPLY(tmp13 + tmp15, FIX(0.654463974)); /* c11 */ |
2166 | 0 | tmp0 = MULTIPLY(tmp13, FIX(0.541301207)) - /* c7-c11 */ |
2167 | 0 | MULTIPLY(tmp14, FIX(0.584525538)) + /* c3-c9 */ |
2168 | 0 | MULTIPLY(tmp16, FIX(1.934788705)) + tmp4 + tmp12; /* c1+c13 */ |
2169 | 0 | tmp3 = MULTIPLY(tmp10, - FIX(0.404480980)) - /* -(c1-c7) */ |
2170 | 0 | MULTIPLY(tmp11, FIX(2.476089912)) - /* c3+c9 */ |
2171 | 0 | MULTIPLY(tmp15, FIX(0.989006518)) + tmp4 - tmp12; /* c11+c13 */ |
2172 | |
|
2173 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+2); |
2174 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+2); |
2175 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+2); |
2176 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+2); |
2177 | |
|
2178 | 0 | dataptr++; /* advance pointer to next column */ |
2179 | 0 | wsptr++; /* advance pointer to next column */ |
2180 | 0 | } |
2181 | 0 | } |
2182 | | |
2183 | | |
2184 | | /* |
2185 | | * Perform the forward DCT on a 16x16 sample block. |
2186 | | */ |
2187 | | |
2188 | | GLOBAL(void) |
2189 | | jpeg_fdct_16x16 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
2190 | 0 | { |
2191 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
2192 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16, tmp17; |
2193 | 0 | DCTELEM workspace[DCTSIZE2]; |
2194 | 0 | DCTELEM *dataptr; |
2195 | 0 | DCTELEM *wsptr; |
2196 | 0 | JSAMPROW elemptr; |
2197 | 0 | int ctr; |
2198 | 0 | SHIFT_TEMPS |
2199 | | |
2200 | | /* Pass 1: process rows. |
2201 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
2202 | | * furthermore, we scale the results by 2**PASS1_BITS. |
2203 | | * cK represents sqrt(2) * cos(K*pi/32). |
2204 | | */ |
2205 | |
|
2206 | 0 | dataptr = data; |
2207 | 0 | ctr = 0; |
2208 | 0 | for (;;) { |
2209 | 0 | elemptr = sample_data[ctr] + start_col; |
2210 | | |
2211 | | /* Even part */ |
2212 | |
|
2213 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[15]); |
2214 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[14]); |
2215 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[13]); |
2216 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[12]); |
2217 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[11]); |
2218 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[10]); |
2219 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[9]); |
2220 | 0 | tmp7 = GETJSAMPLE(elemptr[7]) + GETJSAMPLE(elemptr[8]); |
2221 | |
|
2222 | 0 | tmp10 = tmp0 + tmp7; |
2223 | 0 | tmp14 = tmp0 - tmp7; |
2224 | 0 | tmp11 = tmp1 + tmp6; |
2225 | 0 | tmp15 = tmp1 - tmp6; |
2226 | 0 | tmp12 = tmp2 + tmp5; |
2227 | 0 | tmp16 = tmp2 - tmp5; |
2228 | 0 | tmp13 = tmp3 + tmp4; |
2229 | 0 | tmp17 = tmp3 - tmp4; |
2230 | |
|
2231 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[15]); |
2232 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[14]); |
2233 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[13]); |
2234 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[12]); |
2235 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[11]); |
2236 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[10]); |
2237 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[9]); |
2238 | 0 | tmp7 = GETJSAMPLE(elemptr[7]) - GETJSAMPLE(elemptr[8]); |
2239 | | |
2240 | | /* Apply unsigned->signed conversion. */ |
2241 | 0 | dataptr[0] = (DCTELEM) |
2242 | 0 | ((tmp10 + tmp11 + tmp12 + tmp13 - 16 * CENTERJSAMPLE) << PASS1_BITS); |
2243 | 0 | dataptr[4] = (DCTELEM) |
2244 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */ |
2245 | 0 | MULTIPLY(tmp11 - tmp12, FIX_0_541196100), /* c12[16] = c6[8] */ |
2246 | 0 | CONST_BITS-PASS1_BITS); |
2247 | |
|
2248 | 0 | tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) + /* c14[16] = c7[8] */ |
2249 | 0 | MULTIPLY(tmp14 - tmp16, FIX(1.387039845)); /* c2[16] = c1[8] */ |
2250 | |
|
2251 | 0 | dataptr[2] = (DCTELEM) |
2252 | 0 | DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982)) /* c6+c14 */ |
2253 | 0 | + MULTIPLY(tmp16, FIX(2.172734804)), /* c2+c10 */ |
2254 | 0 | CONST_BITS-PASS1_BITS); |
2255 | 0 | dataptr[6] = (DCTELEM) |
2256 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243)) /* c2-c6 */ |
2257 | 0 | - MULTIPLY(tmp17, FIX(1.061594338)), /* c10+c14 */ |
2258 | 0 | CONST_BITS-PASS1_BITS); |
2259 | | |
2260 | | /* Odd part */ |
2261 | |
|
2262 | 0 | tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) + /* c3 */ |
2263 | 0 | MULTIPLY(tmp6 - tmp7, FIX(0.410524528)); /* c13 */ |
2264 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) + /* c5 */ |
2265 | 0 | MULTIPLY(tmp5 + tmp7, FIX(0.666655658)); /* c11 */ |
2266 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) + /* c7 */ |
2267 | 0 | MULTIPLY(tmp4 - tmp7, FIX(0.897167586)); /* c9 */ |
2268 | 0 | tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) + /* c15 */ |
2269 | 0 | MULTIPLY(tmp6 - tmp5, FIX(1.407403738)); /* c1 */ |
2270 | 0 | tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) + /* -c11 */ |
2271 | 0 | MULTIPLY(tmp4 + tmp6, - FIX(1.247225013)); /* -c5 */ |
2272 | 0 | tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) + /* -c3 */ |
2273 | 0 | MULTIPLY(tmp5 - tmp4, FIX(0.410524528)); /* c13 */ |
2274 | 0 | tmp10 = tmp11 + tmp12 + tmp13 - |
2275 | 0 | MULTIPLY(tmp0, FIX(2.286341144)) + /* c7+c5+c3-c1 */ |
2276 | 0 | MULTIPLY(tmp7, FIX(0.779653625)); /* c15+c13-c11+c9 */ |
2277 | 0 | tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */ |
2278 | 0 | - MULTIPLY(tmp6, FIX(1.663905119)); /* c7+c13+c1-c5 */ |
2279 | 0 | tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */ |
2280 | 0 | + MULTIPLY(tmp5, FIX(1.227391138)); /* c9-c11+c1-c13 */ |
2281 | 0 | tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */ |
2282 | 0 | + MULTIPLY(tmp4, FIX(2.167985692)); /* c1+c13+c5-c9 */ |
2283 | |
|
2284 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS); |
2285 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS); |
2286 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS); |
2287 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS); |
2288 | |
|
2289 | 0 | ctr++; |
2290 | |
|
2291 | 0 | if (ctr != DCTSIZE) { |
2292 | 0 | if (ctr == DCTSIZE * 2) |
2293 | 0 | break; /* Done. */ |
2294 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
2295 | 0 | } else |
2296 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
2297 | 0 | } |
2298 | | |
2299 | | /* Pass 2: process columns. |
2300 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
2301 | | * by an overall factor of 8. |
2302 | | * We must also scale the output by (8/16)**2 = 1/2**2. |
2303 | | * cK represents sqrt(2) * cos(K*pi/32). |
2304 | | */ |
2305 | |
|
2306 | 0 | dataptr = data; |
2307 | 0 | wsptr = workspace; |
2308 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
2309 | | /* Even part */ |
2310 | |
|
2311 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*7]; |
2312 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*6]; |
2313 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*5]; |
2314 | 0 | tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*4]; |
2315 | 0 | tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*3]; |
2316 | 0 | tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*2]; |
2317 | 0 | tmp6 = dataptr[DCTSIZE*6] + wsptr[DCTSIZE*1]; |
2318 | 0 | tmp7 = dataptr[DCTSIZE*7] + wsptr[DCTSIZE*0]; |
2319 | |
|
2320 | 0 | tmp10 = tmp0 + tmp7; |
2321 | 0 | tmp14 = tmp0 - tmp7; |
2322 | 0 | tmp11 = tmp1 + tmp6; |
2323 | 0 | tmp15 = tmp1 - tmp6; |
2324 | 0 | tmp12 = tmp2 + tmp5; |
2325 | 0 | tmp16 = tmp2 - tmp5; |
2326 | 0 | tmp13 = tmp3 + tmp4; |
2327 | 0 | tmp17 = tmp3 - tmp4; |
2328 | |
|
2329 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*7]; |
2330 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*6]; |
2331 | 0 | tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*5]; |
2332 | 0 | tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*4]; |
2333 | 0 | tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*3]; |
2334 | 0 | tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*2]; |
2335 | 0 | tmp6 = dataptr[DCTSIZE*6] - wsptr[DCTSIZE*1]; |
2336 | 0 | tmp7 = dataptr[DCTSIZE*7] - wsptr[DCTSIZE*0]; |
2337 | |
|
2338 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
2339 | 0 | DESCALE(tmp10 + tmp11 + tmp12 + tmp13, PASS1_BITS+2); |
2340 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
2341 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */ |
2342 | 0 | MULTIPLY(tmp11 - tmp12, FIX_0_541196100), /* c12[16] = c6[8] */ |
2343 | 0 | CONST_BITS+PASS1_BITS+2); |
2344 | |
|
2345 | 0 | tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) + /* c14[16] = c7[8] */ |
2346 | 0 | MULTIPLY(tmp14 - tmp16, FIX(1.387039845)); /* c2[16] = c1[8] */ |
2347 | |
|
2348 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
2349 | 0 | DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982)) /* c6+c14 */ |
2350 | 0 | + MULTIPLY(tmp16, FIX(2.172734804)), /* c2+10 */ |
2351 | 0 | CONST_BITS+PASS1_BITS+2); |
2352 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
2353 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243)) /* c2-c6 */ |
2354 | 0 | - MULTIPLY(tmp17, FIX(1.061594338)), /* c10+c14 */ |
2355 | 0 | CONST_BITS+PASS1_BITS+2); |
2356 | | |
2357 | | /* Odd part */ |
2358 | |
|
2359 | 0 | tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) + /* c3 */ |
2360 | 0 | MULTIPLY(tmp6 - tmp7, FIX(0.410524528)); /* c13 */ |
2361 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) + /* c5 */ |
2362 | 0 | MULTIPLY(tmp5 + tmp7, FIX(0.666655658)); /* c11 */ |
2363 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) + /* c7 */ |
2364 | 0 | MULTIPLY(tmp4 - tmp7, FIX(0.897167586)); /* c9 */ |
2365 | 0 | tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) + /* c15 */ |
2366 | 0 | MULTIPLY(tmp6 - tmp5, FIX(1.407403738)); /* c1 */ |
2367 | 0 | tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) + /* -c11 */ |
2368 | 0 | MULTIPLY(tmp4 + tmp6, - FIX(1.247225013)); /* -c5 */ |
2369 | 0 | tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) + /* -c3 */ |
2370 | 0 | MULTIPLY(tmp5 - tmp4, FIX(0.410524528)); /* c13 */ |
2371 | 0 | tmp10 = tmp11 + tmp12 + tmp13 - |
2372 | 0 | MULTIPLY(tmp0, FIX(2.286341144)) + /* c7+c5+c3-c1 */ |
2373 | 0 | MULTIPLY(tmp7, FIX(0.779653625)); /* c15+c13-c11+c9 */ |
2374 | 0 | tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */ |
2375 | 0 | - MULTIPLY(tmp6, FIX(1.663905119)); /* c7+c13+c1-c5 */ |
2376 | 0 | tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */ |
2377 | 0 | + MULTIPLY(tmp5, FIX(1.227391138)); /* c9-c11+c1-c13 */ |
2378 | 0 | tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */ |
2379 | 0 | + MULTIPLY(tmp4, FIX(2.167985692)); /* c1+c13+c5-c9 */ |
2380 | |
|
2381 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+PASS1_BITS+2); |
2382 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+PASS1_BITS+2); |
2383 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+PASS1_BITS+2); |
2384 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+PASS1_BITS+2); |
2385 | |
|
2386 | 0 | dataptr++; /* advance pointer to next column */ |
2387 | 0 | wsptr++; /* advance pointer to next column */ |
2388 | 0 | } |
2389 | 0 | } |
2390 | | |
2391 | | |
2392 | | /* |
2393 | | * Perform the forward DCT on a 16x8 sample block. |
2394 | | * |
2395 | | * 16-point FDCT in pass 1 (rows), 8-point in pass 2 (columns). |
2396 | | */ |
2397 | | |
2398 | | GLOBAL(void) |
2399 | | jpeg_fdct_16x8 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
2400 | 0 | { |
2401 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
2402 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16, tmp17; |
2403 | 0 | INT32 z1; |
2404 | 0 | DCTELEM *dataptr; |
2405 | 0 | JSAMPROW elemptr; |
2406 | 0 | int ctr; |
2407 | 0 | SHIFT_TEMPS |
2408 | | |
2409 | | /* Pass 1: process rows. |
2410 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
2411 | | * furthermore, we scale the results by 2**PASS1_BITS. |
2412 | | * 16-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/32). |
2413 | | */ |
2414 | |
|
2415 | 0 | dataptr = data; |
2416 | 0 | ctr = 0; |
2417 | 0 | for (ctr = 0; ctr < DCTSIZE; ctr++) { |
2418 | 0 | elemptr = sample_data[ctr] + start_col; |
2419 | | |
2420 | | /* Even part */ |
2421 | |
|
2422 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[15]); |
2423 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[14]); |
2424 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[13]); |
2425 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[12]); |
2426 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[11]); |
2427 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[10]); |
2428 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[9]); |
2429 | 0 | tmp7 = GETJSAMPLE(elemptr[7]) + GETJSAMPLE(elemptr[8]); |
2430 | |
|
2431 | 0 | tmp10 = tmp0 + tmp7; |
2432 | 0 | tmp14 = tmp0 - tmp7; |
2433 | 0 | tmp11 = tmp1 + tmp6; |
2434 | 0 | tmp15 = tmp1 - tmp6; |
2435 | 0 | tmp12 = tmp2 + tmp5; |
2436 | 0 | tmp16 = tmp2 - tmp5; |
2437 | 0 | tmp13 = tmp3 + tmp4; |
2438 | 0 | tmp17 = tmp3 - tmp4; |
2439 | |
|
2440 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[15]); |
2441 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[14]); |
2442 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[13]); |
2443 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[12]); |
2444 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[11]); |
2445 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[10]); |
2446 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[9]); |
2447 | 0 | tmp7 = GETJSAMPLE(elemptr[7]) - GETJSAMPLE(elemptr[8]); |
2448 | | |
2449 | | /* Apply unsigned->signed conversion. */ |
2450 | 0 | dataptr[0] = (DCTELEM) |
2451 | 0 | ((tmp10 + tmp11 + tmp12 + tmp13 - 16 * CENTERJSAMPLE) << PASS1_BITS); |
2452 | 0 | dataptr[4] = (DCTELEM) |
2453 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */ |
2454 | 0 | MULTIPLY(tmp11 - tmp12, FIX_0_541196100), /* c12[16] = c6[8] */ |
2455 | 0 | CONST_BITS-PASS1_BITS); |
2456 | |
|
2457 | 0 | tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) + /* c14[16] = c7[8] */ |
2458 | 0 | MULTIPLY(tmp14 - tmp16, FIX(1.387039845)); /* c2[16] = c1[8] */ |
2459 | |
|
2460 | 0 | dataptr[2] = (DCTELEM) |
2461 | 0 | DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982)) /* c6+c14 */ |
2462 | 0 | + MULTIPLY(tmp16, FIX(2.172734804)), /* c2+c10 */ |
2463 | 0 | CONST_BITS-PASS1_BITS); |
2464 | 0 | dataptr[6] = (DCTELEM) |
2465 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243)) /* c2-c6 */ |
2466 | 0 | - MULTIPLY(tmp17, FIX(1.061594338)), /* c10+c14 */ |
2467 | 0 | CONST_BITS-PASS1_BITS); |
2468 | | |
2469 | | /* Odd part */ |
2470 | |
|
2471 | 0 | tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) + /* c3 */ |
2472 | 0 | MULTIPLY(tmp6 - tmp7, FIX(0.410524528)); /* c13 */ |
2473 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) + /* c5 */ |
2474 | 0 | MULTIPLY(tmp5 + tmp7, FIX(0.666655658)); /* c11 */ |
2475 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) + /* c7 */ |
2476 | 0 | MULTIPLY(tmp4 - tmp7, FIX(0.897167586)); /* c9 */ |
2477 | 0 | tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) + /* c15 */ |
2478 | 0 | MULTIPLY(tmp6 - tmp5, FIX(1.407403738)); /* c1 */ |
2479 | 0 | tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) + /* -c11 */ |
2480 | 0 | MULTIPLY(tmp4 + tmp6, - FIX(1.247225013)); /* -c5 */ |
2481 | 0 | tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) + /* -c3 */ |
2482 | 0 | MULTIPLY(tmp5 - tmp4, FIX(0.410524528)); /* c13 */ |
2483 | 0 | tmp10 = tmp11 + tmp12 + tmp13 - |
2484 | 0 | MULTIPLY(tmp0, FIX(2.286341144)) + /* c7+c5+c3-c1 */ |
2485 | 0 | MULTIPLY(tmp7, FIX(0.779653625)); /* c15+c13-c11+c9 */ |
2486 | 0 | tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */ |
2487 | 0 | - MULTIPLY(tmp6, FIX(1.663905119)); /* c7+c13+c1-c5 */ |
2488 | 0 | tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */ |
2489 | 0 | + MULTIPLY(tmp5, FIX(1.227391138)); /* c9-c11+c1-c13 */ |
2490 | 0 | tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */ |
2491 | 0 | + MULTIPLY(tmp4, FIX(2.167985692)); /* c1+c13+c5-c9 */ |
2492 | |
|
2493 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS); |
2494 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS); |
2495 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS); |
2496 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS); |
2497 | |
|
2498 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
2499 | 0 | } |
2500 | | |
2501 | | /* Pass 2: process columns. |
2502 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
2503 | | * by an overall factor of 8. |
2504 | | * We must also scale the output by 8/16 = 1/2. |
2505 | | * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16). |
2506 | | */ |
2507 | |
|
2508 | 0 | dataptr = data; |
2509 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
2510 | | /* Even part per LL&M figure 1 --- note that published figure is faulty; |
2511 | | * rotator "c1" should be "c6". |
2512 | | */ |
2513 | |
|
2514 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7]; |
2515 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6]; |
2516 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5]; |
2517 | 0 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4]; |
2518 | |
|
2519 | 0 | tmp10 = tmp0 + tmp3; |
2520 | 0 | tmp12 = tmp0 - tmp3; |
2521 | 0 | tmp11 = tmp1 + tmp2; |
2522 | 0 | tmp13 = tmp1 - tmp2; |
2523 | |
|
2524 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7]; |
2525 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6]; |
2526 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5]; |
2527 | 0 | tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4]; |
2528 | |
|
2529 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS+1); |
2530 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS+1); |
2531 | |
|
2532 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); /* c6 */ |
2533 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
2534 | 0 | DESCALE(z1 + MULTIPLY(tmp12, FIX_0_765366865), /* c2-c6 */ |
2535 | 0 | CONST_BITS+PASS1_BITS+1); |
2536 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
2537 | 0 | DESCALE(z1 - MULTIPLY(tmp13, FIX_1_847759065), /* c2+c6 */ |
2538 | 0 | CONST_BITS+PASS1_BITS+1); |
2539 | | |
2540 | | /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
2541 | | * i0..i3 in the paper are tmp0..tmp3 here. |
2542 | | */ |
2543 | |
|
2544 | 0 | tmp12 = tmp0 + tmp2; |
2545 | 0 | tmp13 = tmp1 + tmp3; |
2546 | |
|
2547 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /* c3 */ |
2548 | 0 | tmp12 = MULTIPLY(tmp12, - FIX_0_390180644); /* -c3+c5 */ |
2549 | 0 | tmp13 = MULTIPLY(tmp13, - FIX_1_961570560); /* -c3-c5 */ |
2550 | 0 | tmp12 += z1; |
2551 | 0 | tmp13 += z1; |
2552 | |
|
2553 | 0 | z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */ |
2554 | 0 | tmp0 = MULTIPLY(tmp0, FIX_1_501321110); /* c1+c3-c5-c7 */ |
2555 | 0 | tmp3 = MULTIPLY(tmp3, FIX_0_298631336); /* -c1+c3+c5-c7 */ |
2556 | 0 | tmp0 += z1 + tmp12; |
2557 | 0 | tmp3 += z1 + tmp13; |
2558 | |
|
2559 | 0 | z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */ |
2560 | 0 | tmp1 = MULTIPLY(tmp1, FIX_3_072711026); /* c1+c3+c5-c7 */ |
2561 | 0 | tmp2 = MULTIPLY(tmp2, FIX_2_053119869); /* c1+c3-c5+c7 */ |
2562 | 0 | tmp1 += z1 + tmp13; |
2563 | 0 | tmp2 += z1 + tmp12; |
2564 | |
|
2565 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+PASS1_BITS+1); |
2566 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+PASS1_BITS+1); |
2567 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+PASS1_BITS+1); |
2568 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp3, CONST_BITS+PASS1_BITS+1); |
2569 | |
|
2570 | 0 | dataptr++; /* advance pointer to next column */ |
2571 | 0 | } |
2572 | 0 | } |
2573 | | |
2574 | | |
2575 | | /* |
2576 | | * Perform the forward DCT on a 14x7 sample block. |
2577 | | * |
2578 | | * 14-point FDCT in pass 1 (rows), 7-point in pass 2 (columns). |
2579 | | */ |
2580 | | |
2581 | | GLOBAL(void) |
2582 | | jpeg_fdct_14x7 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
2583 | 0 | { |
2584 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6; |
2585 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16; |
2586 | 0 | INT32 z1, z2, z3; |
2587 | 0 | DCTELEM *dataptr; |
2588 | 0 | JSAMPROW elemptr; |
2589 | 0 | int ctr; |
2590 | 0 | SHIFT_TEMPS |
2591 | | |
2592 | | /* Zero bottom row of output coefficient block. */ |
2593 | 0 | MEMZERO(&data[DCTSIZE*7], SIZEOF(DCTELEM) * DCTSIZE); |
2594 | | |
2595 | | /* Pass 1: process rows. |
2596 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
2597 | | * furthermore, we scale the results by 2**PASS1_BITS. |
2598 | | * 14-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/28). |
2599 | | */ |
2600 | |
|
2601 | 0 | dataptr = data; |
2602 | 0 | for (ctr = 0; ctr < 7; ctr++) { |
2603 | 0 | elemptr = sample_data[ctr] + start_col; |
2604 | | |
2605 | | /* Even part */ |
2606 | |
|
2607 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[13]); |
2608 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[12]); |
2609 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[11]); |
2610 | 0 | tmp13 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[10]); |
2611 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[9]); |
2612 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[8]); |
2613 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) + GETJSAMPLE(elemptr[7]); |
2614 | |
|
2615 | 0 | tmp10 = tmp0 + tmp6; |
2616 | 0 | tmp14 = tmp0 - tmp6; |
2617 | 0 | tmp11 = tmp1 + tmp5; |
2618 | 0 | tmp15 = tmp1 - tmp5; |
2619 | 0 | tmp12 = tmp2 + tmp4; |
2620 | 0 | tmp16 = tmp2 - tmp4; |
2621 | |
|
2622 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[13]); |
2623 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[12]); |
2624 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[11]); |
2625 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[10]); |
2626 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[9]); |
2627 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[8]); |
2628 | 0 | tmp6 = GETJSAMPLE(elemptr[6]) - GETJSAMPLE(elemptr[7]); |
2629 | | |
2630 | | /* Apply unsigned->signed conversion. */ |
2631 | 0 | dataptr[0] = (DCTELEM) |
2632 | 0 | ((tmp10 + tmp11 + tmp12 + tmp13 - 14 * CENTERJSAMPLE) << PASS1_BITS); |
2633 | 0 | tmp13 += tmp13; |
2634 | 0 | dataptr[4] = (DCTELEM) |
2635 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.274162392)) + /* c4 */ |
2636 | 0 | MULTIPLY(tmp11 - tmp13, FIX(0.314692123)) - /* c12 */ |
2637 | 0 | MULTIPLY(tmp12 - tmp13, FIX(0.881747734)), /* c8 */ |
2638 | 0 | CONST_BITS-PASS1_BITS); |
2639 | |
|
2640 | 0 | tmp10 = MULTIPLY(tmp14 + tmp15, FIX(1.105676686)); /* c6 */ |
2641 | |
|
2642 | 0 | dataptr[2] = (DCTELEM) |
2643 | 0 | DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.273079590)) /* c2-c6 */ |
2644 | 0 | + MULTIPLY(tmp16, FIX(0.613604268)), /* c10 */ |
2645 | 0 | CONST_BITS-PASS1_BITS); |
2646 | 0 | dataptr[6] = (DCTELEM) |
2647 | 0 | DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.719280954)) /* c6+c10 */ |
2648 | 0 | - MULTIPLY(tmp16, FIX(1.378756276)), /* c2 */ |
2649 | 0 | CONST_BITS-PASS1_BITS); |
2650 | | |
2651 | | /* Odd part */ |
2652 | |
|
2653 | 0 | tmp10 = tmp1 + tmp2; |
2654 | 0 | tmp11 = tmp5 - tmp4; |
2655 | 0 | dataptr[7] = (DCTELEM) ((tmp0 - tmp10 + tmp3 - tmp11 - tmp6) << PASS1_BITS); |
2656 | 0 | tmp3 <<= CONST_BITS; |
2657 | 0 | tmp10 = MULTIPLY(tmp10, - FIX(0.158341681)); /* -c13 */ |
2658 | 0 | tmp11 = MULTIPLY(tmp11, FIX(1.405321284)); /* c1 */ |
2659 | 0 | tmp10 += tmp11 - tmp3; |
2660 | 0 | tmp11 = MULTIPLY(tmp0 + tmp2, FIX(1.197448846)) + /* c5 */ |
2661 | 0 | MULTIPLY(tmp4 + tmp6, FIX(0.752406978)); /* c9 */ |
2662 | 0 | dataptr[5] = (DCTELEM) |
2663 | 0 | DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(2.373959773)) /* c3+c5-c13 */ |
2664 | 0 | + MULTIPLY(tmp4, FIX(1.119999435)), /* c1+c11-c9 */ |
2665 | 0 | CONST_BITS-PASS1_BITS); |
2666 | 0 | tmp12 = MULTIPLY(tmp0 + tmp1, FIX(1.334852607)) + /* c3 */ |
2667 | 0 | MULTIPLY(tmp5 - tmp6, FIX(0.467085129)); /* c11 */ |
2668 | 0 | dataptr[3] = (DCTELEM) |
2669 | 0 | DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.424103948)) /* c3-c9-c13 */ |
2670 | 0 | - MULTIPLY(tmp5, FIX(3.069855259)), /* c1+c5+c11 */ |
2671 | 0 | CONST_BITS-PASS1_BITS); |
2672 | 0 | dataptr[1] = (DCTELEM) |
2673 | 0 | DESCALE(tmp11 + tmp12 + tmp3 + tmp6 - |
2674 | 0 | MULTIPLY(tmp0 + tmp6, FIX(1.126980169)), /* c3+c5-c1 */ |
2675 | 0 | CONST_BITS-PASS1_BITS); |
2676 | |
|
2677 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
2678 | 0 | } |
2679 | | |
2680 | | /* Pass 2: process columns. |
2681 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
2682 | | * by an overall factor of 8. |
2683 | | * We must also scale the output by (8/14)*(8/7) = 32/49, which we |
2684 | | * partially fold into the constant multipliers and final shifting: |
2685 | | * 7-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/14) * 64/49. |
2686 | | */ |
2687 | |
|
2688 | 0 | dataptr = data; |
2689 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
2690 | | /* Even part */ |
2691 | |
|
2692 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*6]; |
2693 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*5]; |
2694 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*4]; |
2695 | 0 | tmp3 = dataptr[DCTSIZE*3]; |
2696 | |
|
2697 | 0 | tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*6]; |
2698 | 0 | tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*5]; |
2699 | 0 | tmp12 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*4]; |
2700 | |
|
2701 | 0 | z1 = tmp0 + tmp2; |
2702 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
2703 | 0 | DESCALE(MULTIPLY(z1 + tmp1 + tmp3, FIX(1.306122449)), /* 64/49 */ |
2704 | 0 | CONST_BITS+PASS1_BITS+1); |
2705 | 0 | tmp3 += tmp3; |
2706 | 0 | z1 -= tmp3; |
2707 | 0 | z1 -= tmp3; |
2708 | 0 | z1 = MULTIPLY(z1, FIX(0.461784020)); /* (c2+c6-c4)/2 */ |
2709 | 0 | z2 = MULTIPLY(tmp0 - tmp2, FIX(1.202428084)); /* (c2+c4-c6)/2 */ |
2710 | 0 | z3 = MULTIPLY(tmp1 - tmp2, FIX(0.411026446)); /* c6 */ |
2711 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS+PASS1_BITS+1); |
2712 | 0 | z1 -= z2; |
2713 | 0 | z2 = MULTIPLY(tmp0 - tmp1, FIX(1.151670509)); /* c4 */ |
2714 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
2715 | 0 | DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.923568041)), /* c2+c6-c4 */ |
2716 | 0 | CONST_BITS+PASS1_BITS+1); |
2717 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS+PASS1_BITS+1); |
2718 | | |
2719 | | /* Odd part */ |
2720 | |
|
2721 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(1.221765677)); /* (c3+c1-c5)/2 */ |
2722 | 0 | tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.222383464)); /* (c3+c5-c1)/2 */ |
2723 | 0 | tmp0 = tmp1 - tmp2; |
2724 | 0 | tmp1 += tmp2; |
2725 | 0 | tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.800824523)); /* -c1 */ |
2726 | 0 | tmp1 += tmp2; |
2727 | 0 | tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.801442310)); /* c5 */ |
2728 | 0 | tmp0 += tmp3; |
2729 | 0 | tmp2 += tmp3 + MULTIPLY(tmp12, FIX(2.443531355)); /* c3+c1-c5 */ |
2730 | |
|
2731 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp0, CONST_BITS+PASS1_BITS+1); |
2732 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp1, CONST_BITS+PASS1_BITS+1); |
2733 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp2, CONST_BITS+PASS1_BITS+1); |
2734 | |
|
2735 | 0 | dataptr++; /* advance pointer to next column */ |
2736 | 0 | } |
2737 | 0 | } |
2738 | | |
2739 | | |
2740 | | /* |
2741 | | * Perform the forward DCT on a 12x6 sample block. |
2742 | | * |
2743 | | * 12-point FDCT in pass 1 (rows), 6-point in pass 2 (columns). |
2744 | | */ |
2745 | | |
2746 | | GLOBAL(void) |
2747 | | jpeg_fdct_12x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
2748 | 0 | { |
2749 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5; |
2750 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15; |
2751 | 0 | DCTELEM *dataptr; |
2752 | 0 | JSAMPROW elemptr; |
2753 | 0 | int ctr; |
2754 | 0 | SHIFT_TEMPS |
2755 | | |
2756 | | /* Zero 2 bottom rows of output coefficient block. */ |
2757 | 0 | MEMZERO(&data[DCTSIZE*6], SIZEOF(DCTELEM) * DCTSIZE * 2); |
2758 | | |
2759 | | /* Pass 1: process rows. |
2760 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
2761 | | * furthermore, we scale the results by 2**PASS1_BITS. |
2762 | | * 12-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/24). |
2763 | | */ |
2764 | |
|
2765 | 0 | dataptr = data; |
2766 | 0 | for (ctr = 0; ctr < 6; ctr++) { |
2767 | 0 | elemptr = sample_data[ctr] + start_col; |
2768 | | |
2769 | | /* Even part */ |
2770 | |
|
2771 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[11]); |
2772 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[10]); |
2773 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[9]); |
2774 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[8]); |
2775 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[7]); |
2776 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) + GETJSAMPLE(elemptr[6]); |
2777 | |
|
2778 | 0 | tmp10 = tmp0 + tmp5; |
2779 | 0 | tmp13 = tmp0 - tmp5; |
2780 | 0 | tmp11 = tmp1 + tmp4; |
2781 | 0 | tmp14 = tmp1 - tmp4; |
2782 | 0 | tmp12 = tmp2 + tmp3; |
2783 | 0 | tmp15 = tmp2 - tmp3; |
2784 | |
|
2785 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[11]); |
2786 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[10]); |
2787 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[9]); |
2788 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[8]); |
2789 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[7]); |
2790 | 0 | tmp5 = GETJSAMPLE(elemptr[5]) - GETJSAMPLE(elemptr[6]); |
2791 | | |
2792 | | /* Apply unsigned->signed conversion. */ |
2793 | 0 | dataptr[0] = (DCTELEM) |
2794 | 0 | ((tmp10 + tmp11 + tmp12 - 12 * CENTERJSAMPLE) << PASS1_BITS); |
2795 | 0 | dataptr[6] = (DCTELEM) ((tmp13 - tmp14 - tmp15) << PASS1_BITS); |
2796 | 0 | dataptr[4] = (DCTELEM) |
2797 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.224744871)), /* c4 */ |
2798 | 0 | CONST_BITS-PASS1_BITS); |
2799 | 0 | dataptr[2] = (DCTELEM) |
2800 | 0 | DESCALE(tmp14 - tmp15 + MULTIPLY(tmp13 + tmp15, FIX(1.366025404)), /* c2 */ |
2801 | 0 | CONST_BITS-PASS1_BITS); |
2802 | | |
2803 | | /* Odd part */ |
2804 | |
|
2805 | 0 | tmp10 = MULTIPLY(tmp1 + tmp4, FIX_0_541196100); /* c9 */ |
2806 | 0 | tmp14 = tmp10 + MULTIPLY(tmp1, FIX_0_765366865); /* c3-c9 */ |
2807 | 0 | tmp15 = tmp10 - MULTIPLY(tmp4, FIX_1_847759065); /* c3+c9 */ |
2808 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.121971054)); /* c5 */ |
2809 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.860918669)); /* c7 */ |
2810 | 0 | tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.580774953)) /* c5+c7-c1 */ |
2811 | 0 | + MULTIPLY(tmp5, FIX(0.184591911)); /* c11 */ |
2812 | 0 | tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.184591911)); /* -c11 */ |
2813 | 0 | tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.339493912)) /* c1+c5-c11 */ |
2814 | 0 | + MULTIPLY(tmp5, FIX(0.860918669)); /* c7 */ |
2815 | 0 | tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.725788011)) /* c1+c11-c7 */ |
2816 | 0 | - MULTIPLY(tmp5, FIX(1.121971054)); /* c5 */ |
2817 | 0 | tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.306562965)) /* c3 */ |
2818 | 0 | - MULTIPLY(tmp2 + tmp5, FIX_0_541196100); /* c9 */ |
2819 | |
|
2820 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS); |
2821 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS); |
2822 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS); |
2823 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS); |
2824 | |
|
2825 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
2826 | 0 | } |
2827 | | |
2828 | | /* Pass 2: process columns. |
2829 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
2830 | | * by an overall factor of 8. |
2831 | | * We must also scale the output by (8/12)*(8/6) = 8/9, which we |
2832 | | * partially fold into the constant multipliers and final shifting: |
2833 | | * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12) * 16/9. |
2834 | | */ |
2835 | |
|
2836 | 0 | dataptr = data; |
2837 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
2838 | | /* Even part */ |
2839 | |
|
2840 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5]; |
2841 | 0 | tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4]; |
2842 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3]; |
2843 | |
|
2844 | 0 | tmp10 = tmp0 + tmp2; |
2845 | 0 | tmp12 = tmp0 - tmp2; |
2846 | |
|
2847 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5]; |
2848 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4]; |
2849 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3]; |
2850 | |
|
2851 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
2852 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)), /* 16/9 */ |
2853 | 0 | CONST_BITS+PASS1_BITS+1); |
2854 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
2855 | 0 | DESCALE(MULTIPLY(tmp12, FIX(2.177324216)), /* c2 */ |
2856 | 0 | CONST_BITS+PASS1_BITS+1); |
2857 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
2858 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */ |
2859 | 0 | CONST_BITS+PASS1_BITS+1); |
2860 | | |
2861 | | /* Odd part */ |
2862 | |
|
2863 | 0 | tmp10 = MULTIPLY(tmp0 + tmp2, FIX(0.650711829)); /* c5 */ |
2864 | |
|
2865 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
2866 | 0 | DESCALE(tmp10 + MULTIPLY(tmp0 + tmp1, FIX(1.777777778)), /* 16/9 */ |
2867 | 0 | CONST_BITS+PASS1_BITS+1); |
2868 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
2869 | 0 | DESCALE(MULTIPLY(tmp0 - tmp1 - tmp2, FIX(1.777777778)), /* 16/9 */ |
2870 | 0 | CONST_BITS+PASS1_BITS+1); |
2871 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
2872 | 0 | DESCALE(tmp10 + MULTIPLY(tmp2 - tmp1, FIX(1.777777778)), /* 16/9 */ |
2873 | 0 | CONST_BITS+PASS1_BITS+1); |
2874 | |
|
2875 | 0 | dataptr++; /* advance pointer to next column */ |
2876 | 0 | } |
2877 | 0 | } |
2878 | | |
2879 | | |
2880 | | /* |
2881 | | * Perform the forward DCT on a 10x5 sample block. |
2882 | | * |
2883 | | * 10-point FDCT in pass 1 (rows), 5-point in pass 2 (columns). |
2884 | | */ |
2885 | | |
2886 | | GLOBAL(void) |
2887 | | jpeg_fdct_10x5 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
2888 | 0 | { |
2889 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4; |
2890 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14; |
2891 | 0 | DCTELEM *dataptr; |
2892 | 0 | JSAMPROW elemptr; |
2893 | 0 | int ctr; |
2894 | 0 | SHIFT_TEMPS |
2895 | | |
2896 | | /* Zero 3 bottom rows of output coefficient block. */ |
2897 | 0 | MEMZERO(&data[DCTSIZE*5], SIZEOF(DCTELEM) * DCTSIZE * 3); |
2898 | | |
2899 | | /* Pass 1: process rows. |
2900 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
2901 | | * furthermore, we scale the results by 2**PASS1_BITS. |
2902 | | * 10-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/20). |
2903 | | */ |
2904 | |
|
2905 | 0 | dataptr = data; |
2906 | 0 | for (ctr = 0; ctr < 5; ctr++) { |
2907 | 0 | elemptr = sample_data[ctr] + start_col; |
2908 | | |
2909 | | /* Even part */ |
2910 | |
|
2911 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[9]); |
2912 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[8]); |
2913 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[7]); |
2914 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[6]); |
2915 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) + GETJSAMPLE(elemptr[5]); |
2916 | |
|
2917 | 0 | tmp10 = tmp0 + tmp4; |
2918 | 0 | tmp13 = tmp0 - tmp4; |
2919 | 0 | tmp11 = tmp1 + tmp3; |
2920 | 0 | tmp14 = tmp1 - tmp3; |
2921 | |
|
2922 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[9]); |
2923 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[8]); |
2924 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[7]); |
2925 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[6]); |
2926 | 0 | tmp4 = GETJSAMPLE(elemptr[4]) - GETJSAMPLE(elemptr[5]); |
2927 | | |
2928 | | /* Apply unsigned->signed conversion. */ |
2929 | 0 | dataptr[0] = (DCTELEM) |
2930 | 0 | ((tmp10 + tmp11 + tmp12 - 10 * CENTERJSAMPLE) << PASS1_BITS); |
2931 | 0 | tmp12 += tmp12; |
2932 | 0 | dataptr[4] = (DCTELEM) |
2933 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.144122806)) - /* c4 */ |
2934 | 0 | MULTIPLY(tmp11 - tmp12, FIX(0.437016024)), /* c8 */ |
2935 | 0 | CONST_BITS-PASS1_BITS); |
2936 | 0 | tmp10 = MULTIPLY(tmp13 + tmp14, FIX(0.831253876)); /* c6 */ |
2937 | 0 | dataptr[2] = (DCTELEM) |
2938 | 0 | DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.513743148)), /* c2-c6 */ |
2939 | 0 | CONST_BITS-PASS1_BITS); |
2940 | 0 | dataptr[6] = (DCTELEM) |
2941 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.176250899)), /* c2+c6 */ |
2942 | 0 | CONST_BITS-PASS1_BITS); |
2943 | | |
2944 | | /* Odd part */ |
2945 | |
|
2946 | 0 | tmp10 = tmp0 + tmp4; |
2947 | 0 | tmp11 = tmp1 - tmp3; |
2948 | 0 | dataptr[5] = (DCTELEM) ((tmp10 - tmp11 - tmp2) << PASS1_BITS); |
2949 | 0 | tmp2 <<= CONST_BITS; |
2950 | 0 | dataptr[1] = (DCTELEM) |
2951 | 0 | DESCALE(MULTIPLY(tmp0, FIX(1.396802247)) + /* c1 */ |
2952 | 0 | MULTIPLY(tmp1, FIX(1.260073511)) + tmp2 + /* c3 */ |
2953 | 0 | MULTIPLY(tmp3, FIX(0.642039522)) + /* c7 */ |
2954 | 0 | MULTIPLY(tmp4, FIX(0.221231742)), /* c9 */ |
2955 | 0 | CONST_BITS-PASS1_BITS); |
2956 | 0 | tmp12 = MULTIPLY(tmp0 - tmp4, FIX(0.951056516)) - /* (c3+c7)/2 */ |
2957 | 0 | MULTIPLY(tmp1 + tmp3, FIX(0.587785252)); /* (c1-c9)/2 */ |
2958 | 0 | tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.309016994)) + /* (c3-c7)/2 */ |
2959 | 0 | (tmp11 << (CONST_BITS - 1)) - tmp2; |
2960 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS-PASS1_BITS); |
2961 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS-PASS1_BITS); |
2962 | |
|
2963 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
2964 | 0 | } |
2965 | | |
2966 | | /* Pass 2: process columns. |
2967 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
2968 | | * by an overall factor of 8. |
2969 | | * We must also scale the output by (8/10)*(8/5) = 32/25, which we |
2970 | | * fold into the constant multipliers: |
2971 | | * 5-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/10) * 32/25. |
2972 | | */ |
2973 | |
|
2974 | 0 | dataptr = data; |
2975 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
2976 | | /* Even part */ |
2977 | |
|
2978 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*4]; |
2979 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*3]; |
2980 | 0 | tmp2 = dataptr[DCTSIZE*2]; |
2981 | |
|
2982 | 0 | tmp10 = tmp0 + tmp1; |
2983 | 0 | tmp11 = tmp0 - tmp1; |
2984 | |
|
2985 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*4]; |
2986 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*3]; |
2987 | |
|
2988 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
2989 | 0 | DESCALE(MULTIPLY(tmp10 + tmp2, FIX(1.28)), /* 32/25 */ |
2990 | 0 | CONST_BITS+PASS1_BITS); |
2991 | 0 | tmp11 = MULTIPLY(tmp11, FIX(1.011928851)); /* (c2+c4)/2 */ |
2992 | 0 | tmp10 -= tmp2 << 2; |
2993 | 0 | tmp10 = MULTIPLY(tmp10, FIX(0.452548340)); /* (c2-c4)/2 */ |
2994 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS+PASS1_BITS); |
2995 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS+PASS1_BITS); |
2996 | | |
2997 | | /* Odd part */ |
2998 | |
|
2999 | 0 | tmp10 = MULTIPLY(tmp0 + tmp1, FIX(1.064004961)); /* c3 */ |
3000 | |
|
3001 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
3002 | 0 | DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.657591230)), /* c1-c3 */ |
3003 | 0 | CONST_BITS+PASS1_BITS); |
3004 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
3005 | 0 | DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.785601151)), /* c1+c3 */ |
3006 | 0 | CONST_BITS+PASS1_BITS); |
3007 | |
|
3008 | 0 | dataptr++; /* advance pointer to next column */ |
3009 | 0 | } |
3010 | 0 | } |
3011 | | |
3012 | | |
3013 | | /* |
3014 | | * Perform the forward DCT on an 8x4 sample block. |
3015 | | * |
3016 | | * 8-point FDCT in pass 1 (rows), 4-point in pass 2 (columns). |
3017 | | */ |
3018 | | |
3019 | | GLOBAL(void) |
3020 | | jpeg_fdct_8x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3021 | 0 | { |
3022 | 0 | INT32 tmp0, tmp1, tmp2, tmp3; |
3023 | 0 | INT32 tmp10, tmp11, tmp12, tmp13; |
3024 | 0 | INT32 z1; |
3025 | 0 | DCTELEM *dataptr; |
3026 | 0 | JSAMPROW elemptr; |
3027 | 0 | int ctr; |
3028 | 0 | SHIFT_TEMPS |
3029 | | |
3030 | | /* Zero 4 bottom rows of output coefficient block. */ |
3031 | 0 | MEMZERO(&data[DCTSIZE*4], SIZEOF(DCTELEM) * DCTSIZE * 4); |
3032 | | |
3033 | | /* Pass 1: process rows. |
3034 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
3035 | | * furthermore, we scale the results by 2**PASS1_BITS. |
3036 | | * We must also scale the output by 8/4 = 2, which we add here. |
3037 | | * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16). |
3038 | | */ |
3039 | |
|
3040 | 0 | dataptr = data; |
3041 | 0 | for (ctr = 0; ctr < 4; ctr++) { |
3042 | 0 | elemptr = sample_data[ctr] + start_col; |
3043 | | |
3044 | | /* Even part per LL&M figure 1 --- note that published figure is faulty; |
3045 | | * rotator "c1" should be "c6". |
3046 | | */ |
3047 | |
|
3048 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]); |
3049 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]); |
3050 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]); |
3051 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]); |
3052 | |
|
3053 | 0 | tmp10 = tmp0 + tmp3; |
3054 | 0 | tmp12 = tmp0 - tmp3; |
3055 | 0 | tmp11 = tmp1 + tmp2; |
3056 | 0 | tmp13 = tmp1 - tmp2; |
3057 | |
|
3058 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]); |
3059 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]); |
3060 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]); |
3061 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]); |
3062 | | |
3063 | | /* Apply unsigned->signed conversion. */ |
3064 | 0 | dataptr[0] = (DCTELEM) |
3065 | 0 | ((tmp10 + tmp11 - 8 * CENTERJSAMPLE) << (PASS1_BITS+1)); |
3066 | 0 | dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << (PASS1_BITS+1)); |
3067 | |
|
3068 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); /* c6 */ |
3069 | | /* Add fudge factor here for final descale. */ |
3070 | 0 | z1 += ONE << (CONST_BITS-PASS1_BITS-2); |
3071 | |
|
3072 | 0 | dataptr[2] = (DCTELEM) |
3073 | 0 | RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865), /* c2-c6 */ |
3074 | 0 | CONST_BITS-PASS1_BITS-1); |
3075 | 0 | dataptr[6] = (DCTELEM) |
3076 | 0 | RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065), /* c2+c6 */ |
3077 | 0 | CONST_BITS-PASS1_BITS-1); |
3078 | | |
3079 | | /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
3080 | | * i0..i3 in the paper are tmp0..tmp3 here. |
3081 | | */ |
3082 | |
|
3083 | 0 | tmp12 = tmp0 + tmp2; |
3084 | 0 | tmp13 = tmp1 + tmp3; |
3085 | |
|
3086 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /* c3 */ |
3087 | | /* Add fudge factor here for final descale. */ |
3088 | 0 | z1 += ONE << (CONST_BITS-PASS1_BITS-2); |
3089 | |
|
3090 | 0 | tmp12 = MULTIPLY(tmp12, - FIX_0_390180644); /* -c3+c5 */ |
3091 | 0 | tmp13 = MULTIPLY(tmp13, - FIX_1_961570560); /* -c3-c5 */ |
3092 | 0 | tmp12 += z1; |
3093 | 0 | tmp13 += z1; |
3094 | |
|
3095 | 0 | z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */ |
3096 | 0 | tmp0 = MULTIPLY(tmp0, FIX_1_501321110); /* c1+c3-c5-c7 */ |
3097 | 0 | tmp3 = MULTIPLY(tmp3, FIX_0_298631336); /* -c1+c3+c5-c7 */ |
3098 | 0 | tmp0 += z1 + tmp12; |
3099 | 0 | tmp3 += z1 + tmp13; |
3100 | |
|
3101 | 0 | z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */ |
3102 | 0 | tmp1 = MULTIPLY(tmp1, FIX_3_072711026); /* c1+c3+c5-c7 */ |
3103 | 0 | tmp2 = MULTIPLY(tmp2, FIX_2_053119869); /* c1+c3-c5+c7 */ |
3104 | 0 | tmp1 += z1 + tmp13; |
3105 | 0 | tmp2 += z1 + tmp12; |
3106 | |
|
3107 | 0 | dataptr[1] = (DCTELEM) RIGHT_SHIFT(tmp0, CONST_BITS-PASS1_BITS-1); |
3108 | 0 | dataptr[3] = (DCTELEM) RIGHT_SHIFT(tmp1, CONST_BITS-PASS1_BITS-1); |
3109 | 0 | dataptr[5] = (DCTELEM) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS-1); |
3110 | 0 | dataptr[7] = (DCTELEM) RIGHT_SHIFT(tmp3, CONST_BITS-PASS1_BITS-1); |
3111 | |
|
3112 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
3113 | 0 | } |
3114 | | |
3115 | | /* Pass 2: process columns. |
3116 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
3117 | | * by an overall factor of 8. |
3118 | | * 4-point FDCT kernel, |
3119 | | * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. |
3120 | | */ |
3121 | |
|
3122 | 0 | dataptr = data; |
3123 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
3124 | | /* Even part */ |
3125 | | |
3126 | | /* Add fudge factor here for final descale. */ |
3127 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*3] + (ONE << (PASS1_BITS-1)); |
3128 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*2]; |
3129 | |
|
3130 | 0 | tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*3]; |
3131 | 0 | tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*2]; |
3132 | |
|
3133 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp0 + tmp1, PASS1_BITS); |
3134 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) RIGHT_SHIFT(tmp0 - tmp1, PASS1_BITS); |
3135 | | |
3136 | | /* Odd part */ |
3137 | |
|
3138 | 0 | tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100); /* c6 */ |
3139 | | /* Add fudge factor here for final descale. */ |
3140 | 0 | tmp0 += ONE << (CONST_BITS+PASS1_BITS-1); |
3141 | |
|
3142 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
3143 | 0 | RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */ |
3144 | 0 | CONST_BITS+PASS1_BITS); |
3145 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
3146 | 0 | RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */ |
3147 | 0 | CONST_BITS+PASS1_BITS); |
3148 | |
|
3149 | 0 | dataptr++; /* advance pointer to next column */ |
3150 | 0 | } |
3151 | 0 | } |
3152 | | |
3153 | | |
3154 | | /* |
3155 | | * Perform the forward DCT on a 6x3 sample block. |
3156 | | * |
3157 | | * 6-point FDCT in pass 1 (rows), 3-point in pass 2 (columns). |
3158 | | */ |
3159 | | |
3160 | | GLOBAL(void) |
3161 | | jpeg_fdct_6x3 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3162 | 0 | { |
3163 | 0 | INT32 tmp0, tmp1, tmp2; |
3164 | 0 | INT32 tmp10, tmp11, tmp12; |
3165 | 0 | DCTELEM *dataptr; |
3166 | 0 | JSAMPROW elemptr; |
3167 | 0 | int ctr; |
3168 | 0 | SHIFT_TEMPS |
3169 | | |
3170 | | /* Pre-zero output coefficient block. */ |
3171 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
3172 | | |
3173 | | /* Pass 1: process rows. |
3174 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
3175 | | * furthermore, we scale the results by 2**PASS1_BITS. |
3176 | | * We scale the results further by 2 as part of output adaption |
3177 | | * scaling for different DCT size. |
3178 | | * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12). |
3179 | | */ |
3180 | |
|
3181 | 0 | dataptr = data; |
3182 | 0 | for (ctr = 0; ctr < 3; ctr++) { |
3183 | 0 | elemptr = sample_data[ctr] + start_col; |
3184 | | |
3185 | | /* Even part */ |
3186 | |
|
3187 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]); |
3188 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]); |
3189 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]); |
3190 | |
|
3191 | 0 | tmp10 = tmp0 + tmp2; |
3192 | 0 | tmp12 = tmp0 - tmp2; |
3193 | |
|
3194 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]); |
3195 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]); |
3196 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]); |
3197 | | |
3198 | | /* Apply unsigned->signed conversion. */ |
3199 | 0 | dataptr[0] = (DCTELEM) |
3200 | 0 | ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << (PASS1_BITS+1)); |
3201 | 0 | dataptr[2] = (DCTELEM) |
3202 | 0 | DESCALE(MULTIPLY(tmp12, FIX(1.224744871)), /* c2 */ |
3203 | 0 | CONST_BITS-PASS1_BITS-1); |
3204 | 0 | dataptr[4] = (DCTELEM) |
3205 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */ |
3206 | 0 | CONST_BITS-PASS1_BITS-1); |
3207 | | |
3208 | | /* Odd part */ |
3209 | |
|
3210 | 0 | tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)), /* c5 */ |
3211 | 0 | CONST_BITS-PASS1_BITS-1); |
3212 | |
|
3213 | 0 | dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << (PASS1_BITS+1))); |
3214 | 0 | dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << (PASS1_BITS+1)); |
3215 | 0 | dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << (PASS1_BITS+1))); |
3216 | |
|
3217 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
3218 | 0 | } |
3219 | | |
3220 | | /* Pass 2: process columns. |
3221 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
3222 | | * by an overall factor of 8. |
3223 | | * We must also scale the output by (8/6)*(8/3) = 32/9, which we partially |
3224 | | * fold into the constant multipliers (other part was done in pass 1): |
3225 | | * 3-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/6) * 16/9. |
3226 | | */ |
3227 | |
|
3228 | 0 | dataptr = data; |
3229 | 0 | for (ctr = 0; ctr < 6; ctr++) { |
3230 | | /* Even part */ |
3231 | |
|
3232 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*2]; |
3233 | 0 | tmp1 = dataptr[DCTSIZE*1]; |
3234 | |
|
3235 | 0 | tmp2 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*2]; |
3236 | |
|
3237 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
3238 | 0 | DESCALE(MULTIPLY(tmp0 + tmp1, FIX(1.777777778)), /* 16/9 */ |
3239 | 0 | CONST_BITS+PASS1_BITS); |
3240 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
3241 | 0 | DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(1.257078722)), /* c2 */ |
3242 | 0 | CONST_BITS+PASS1_BITS); |
3243 | | |
3244 | | /* Odd part */ |
3245 | |
|
3246 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
3247 | 0 | DESCALE(MULTIPLY(tmp2, FIX(2.177324216)), /* c1 */ |
3248 | 0 | CONST_BITS+PASS1_BITS); |
3249 | |
|
3250 | 0 | dataptr++; /* advance pointer to next column */ |
3251 | 0 | } |
3252 | 0 | } |
3253 | | |
3254 | | |
3255 | | /* |
3256 | | * Perform the forward DCT on a 4x2 sample block. |
3257 | | * |
3258 | | * 4-point FDCT in pass 1 (rows), 2-point in pass 2 (columns). |
3259 | | */ |
3260 | | |
3261 | | GLOBAL(void) |
3262 | | jpeg_fdct_4x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3263 | 0 | { |
3264 | 0 | DCTELEM tmp0, tmp2, tmp10, tmp12, tmp4, tmp5; |
3265 | 0 | INT32 tmp1, tmp3, tmp11, tmp13; |
3266 | 0 | INT32 z1, z2, z3; |
3267 | 0 | JSAMPROW elemptr; |
3268 | 0 | SHIFT_TEMPS |
3269 | | |
3270 | | /* Pre-zero output coefficient block. */ |
3271 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
3272 | | |
3273 | | /* Pass 1: process rows. |
3274 | | * Note results are scaled up by sqrt(8) compared to a true DCT. |
3275 | | * 4-point FDCT kernel, |
3276 | | * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. |
3277 | | */ |
3278 | | |
3279 | | /* Row 0 */ |
3280 | 0 | elemptr = sample_data[0] + start_col; |
3281 | | |
3282 | | /* Even part */ |
3283 | |
|
3284 | 0 | tmp4 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]); |
3285 | 0 | tmp5 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]); |
3286 | |
|
3287 | 0 | tmp0 = tmp4 + tmp5; |
3288 | 0 | tmp2 = tmp4 - tmp5; |
3289 | | |
3290 | | /* Odd part */ |
3291 | |
|
3292 | 0 | z2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]); |
3293 | 0 | z3 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]); |
3294 | |
|
3295 | 0 | z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */ |
3296 | | /* Add fudge factor here for final descale. */ |
3297 | 0 | z1 += ONE << (CONST_BITS-3-1); |
3298 | 0 | tmp1 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */ |
3299 | 0 | tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */ |
3300 | | |
3301 | | /* Row 1 */ |
3302 | 0 | elemptr = sample_data[1] + start_col; |
3303 | | |
3304 | | /* Even part */ |
3305 | |
|
3306 | 0 | tmp4 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]); |
3307 | 0 | tmp5 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]); |
3308 | |
|
3309 | 0 | tmp10 = tmp4 + tmp5; |
3310 | 0 | tmp12 = tmp4 - tmp5; |
3311 | | |
3312 | | /* Odd part */ |
3313 | |
|
3314 | 0 | z2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]); |
3315 | 0 | z3 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]); |
3316 | |
|
3317 | 0 | z1 = MULTIPLY(z2 + z3, FIX_0_541196100); /* c6 */ |
3318 | 0 | tmp11 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */ |
3319 | 0 | tmp13 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */ |
3320 | | |
3321 | | /* Pass 2: process columns. |
3322 | | * We leave the results scaled up by an overall factor of 8. |
3323 | | * We must also scale the output by (8/4)*(8/2) = 2**3. |
3324 | | */ |
3325 | | |
3326 | | /* Column 0 */ |
3327 | | /* Apply unsigned->signed conversion. */ |
3328 | 0 | data[DCTSIZE*0] = (tmp0 + tmp10 - 8 * CENTERJSAMPLE) << 3; |
3329 | 0 | data[DCTSIZE*1] = (tmp0 - tmp10) << 3; |
3330 | | |
3331 | | /* Column 1 */ |
3332 | 0 | data[DCTSIZE*0+1] = (DCTELEM) RIGHT_SHIFT(tmp1 + tmp11, CONST_BITS-3); |
3333 | 0 | data[DCTSIZE*1+1] = (DCTELEM) RIGHT_SHIFT(tmp1 - tmp11, CONST_BITS-3); |
3334 | | |
3335 | | /* Column 2 */ |
3336 | 0 | data[DCTSIZE*0+2] = (tmp2 + tmp12) << 3; |
3337 | 0 | data[DCTSIZE*1+2] = (tmp2 - tmp12) << 3; |
3338 | | |
3339 | | /* Column 3 */ |
3340 | 0 | data[DCTSIZE*0+3] = (DCTELEM) RIGHT_SHIFT(tmp3 + tmp13, CONST_BITS-3); |
3341 | 0 | data[DCTSIZE*1+3] = (DCTELEM) RIGHT_SHIFT(tmp3 - tmp13, CONST_BITS-3); |
3342 | 0 | } |
3343 | | |
3344 | | |
3345 | | /* |
3346 | | * Perform the forward DCT on a 2x1 sample block. |
3347 | | * |
3348 | | * 2-point FDCT in pass 1 (rows), 1-point in pass 2 (columns). |
3349 | | */ |
3350 | | |
3351 | | GLOBAL(void) |
3352 | | jpeg_fdct_2x1 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3353 | 0 | { |
3354 | 0 | DCTELEM tmp0, tmp1; |
3355 | 0 | JSAMPROW elemptr; |
3356 | | |
3357 | | /* Pre-zero output coefficient block. */ |
3358 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
3359 | |
|
3360 | 0 | elemptr = sample_data[0] + start_col; |
3361 | |
|
3362 | 0 | tmp0 = GETJSAMPLE(elemptr[0]); |
3363 | 0 | tmp1 = GETJSAMPLE(elemptr[1]); |
3364 | | |
3365 | | /* We leave the results scaled up by an overall factor of 8. |
3366 | | * We must also scale the output by (8/2)*(8/1) = 2**5. |
3367 | | */ |
3368 | | |
3369 | | /* Even part */ |
3370 | | |
3371 | | /* Apply unsigned->signed conversion. */ |
3372 | 0 | data[0] = (tmp0 + tmp1 - 2 * CENTERJSAMPLE) << 5; |
3373 | | |
3374 | | /* Odd part */ |
3375 | |
|
3376 | 0 | data[1] = (tmp0 - tmp1) << 5; |
3377 | 0 | } |
3378 | | |
3379 | | |
3380 | | /* |
3381 | | * Perform the forward DCT on an 8x16 sample block. |
3382 | | * |
3383 | | * 8-point FDCT in pass 1 (rows), 16-point in pass 2 (columns). |
3384 | | */ |
3385 | | |
3386 | | GLOBAL(void) |
3387 | | jpeg_fdct_8x16 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3388 | 0 | { |
3389 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
3390 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16, tmp17; |
3391 | 0 | INT32 z1; |
3392 | 0 | DCTELEM workspace[DCTSIZE2]; |
3393 | 0 | DCTELEM *dataptr; |
3394 | 0 | DCTELEM *wsptr; |
3395 | 0 | JSAMPROW elemptr; |
3396 | 0 | int ctr; |
3397 | 0 | SHIFT_TEMPS |
3398 | | |
3399 | | /* Pass 1: process rows. |
3400 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
3401 | | * furthermore, we scale the results by 2**PASS1_BITS. |
3402 | | * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16). |
3403 | | */ |
3404 | |
|
3405 | 0 | dataptr = data; |
3406 | 0 | ctr = 0; |
3407 | 0 | for (;;) { |
3408 | 0 | elemptr = sample_data[ctr] + start_col; |
3409 | | |
3410 | | /* Even part per LL&M figure 1 --- note that published figure is faulty; |
3411 | | * rotator "c1" should be "c6". |
3412 | | */ |
3413 | |
|
3414 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[7]); |
3415 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[6]); |
3416 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[5]); |
3417 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) + GETJSAMPLE(elemptr[4]); |
3418 | |
|
3419 | 0 | tmp10 = tmp0 + tmp3; |
3420 | 0 | tmp12 = tmp0 - tmp3; |
3421 | 0 | tmp11 = tmp1 + tmp2; |
3422 | 0 | tmp13 = tmp1 - tmp2; |
3423 | |
|
3424 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[7]); |
3425 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[6]); |
3426 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[5]); |
3427 | 0 | tmp3 = GETJSAMPLE(elemptr[3]) - GETJSAMPLE(elemptr[4]); |
3428 | | |
3429 | | /* Apply unsigned->signed conversion. */ |
3430 | 0 | dataptr[0] = (DCTELEM) ((tmp10 + tmp11 - 8 * CENTERJSAMPLE) << PASS1_BITS); |
3431 | 0 | dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS); |
3432 | |
|
3433 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); /* c6 */ |
3434 | 0 | dataptr[2] = (DCTELEM) |
3435 | 0 | DESCALE(z1 + MULTIPLY(tmp12, FIX_0_765366865), /* c2-c6 */ |
3436 | 0 | CONST_BITS-PASS1_BITS); |
3437 | 0 | dataptr[6] = (DCTELEM) |
3438 | 0 | DESCALE(z1 - MULTIPLY(tmp13, FIX_1_847759065), /* c2+c6 */ |
3439 | 0 | CONST_BITS-PASS1_BITS); |
3440 | | |
3441 | | /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
3442 | | * i0..i3 in the paper are tmp0..tmp3 here. |
3443 | | */ |
3444 | |
|
3445 | 0 | tmp12 = tmp0 + tmp2; |
3446 | 0 | tmp13 = tmp1 + tmp3; |
3447 | |
|
3448 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /* c3 */ |
3449 | 0 | tmp12 = MULTIPLY(tmp12, - FIX_0_390180644); /* -c3+c5 */ |
3450 | 0 | tmp13 = MULTIPLY(tmp13, - FIX_1_961570560); /* -c3-c5 */ |
3451 | 0 | tmp12 += z1; |
3452 | 0 | tmp13 += z1; |
3453 | |
|
3454 | 0 | z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */ |
3455 | 0 | tmp0 = MULTIPLY(tmp0, FIX_1_501321110); /* c1+c3-c5-c7 */ |
3456 | 0 | tmp3 = MULTIPLY(tmp3, FIX_0_298631336); /* -c1+c3+c5-c7 */ |
3457 | 0 | tmp0 += z1 + tmp12; |
3458 | 0 | tmp3 += z1 + tmp13; |
3459 | |
|
3460 | 0 | z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */ |
3461 | 0 | tmp1 = MULTIPLY(tmp1, FIX_3_072711026); /* c1+c3+c5-c7 */ |
3462 | 0 | tmp2 = MULTIPLY(tmp2, FIX_2_053119869); /* c1+c3-c5+c7 */ |
3463 | 0 | tmp1 += z1 + tmp13; |
3464 | 0 | tmp2 += z1 + tmp12; |
3465 | |
|
3466 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-PASS1_BITS); |
3467 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-PASS1_BITS); |
3468 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS-PASS1_BITS); |
3469 | 0 | dataptr[7] = (DCTELEM) DESCALE(tmp3, CONST_BITS-PASS1_BITS); |
3470 | |
|
3471 | 0 | ctr++; |
3472 | |
|
3473 | 0 | if (ctr != DCTSIZE) { |
3474 | 0 | if (ctr == DCTSIZE * 2) |
3475 | 0 | break; /* Done. */ |
3476 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
3477 | 0 | } else |
3478 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
3479 | 0 | } |
3480 | | |
3481 | | /* Pass 2: process columns. |
3482 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
3483 | | * by an overall factor of 8. |
3484 | | * We must also scale the output by 8/16 = 1/2. |
3485 | | * 16-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/32). |
3486 | | */ |
3487 | |
|
3488 | 0 | dataptr = data; |
3489 | 0 | wsptr = workspace; |
3490 | 0 | for (ctr = DCTSIZE-1; ctr >= 0; ctr--) { |
3491 | | /* Even part */ |
3492 | |
|
3493 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*7]; |
3494 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*6]; |
3495 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*5]; |
3496 | 0 | tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*4]; |
3497 | 0 | tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*3]; |
3498 | 0 | tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*2]; |
3499 | 0 | tmp6 = dataptr[DCTSIZE*6] + wsptr[DCTSIZE*1]; |
3500 | 0 | tmp7 = dataptr[DCTSIZE*7] + wsptr[DCTSIZE*0]; |
3501 | |
|
3502 | 0 | tmp10 = tmp0 + tmp7; |
3503 | 0 | tmp14 = tmp0 - tmp7; |
3504 | 0 | tmp11 = tmp1 + tmp6; |
3505 | 0 | tmp15 = tmp1 - tmp6; |
3506 | 0 | tmp12 = tmp2 + tmp5; |
3507 | 0 | tmp16 = tmp2 - tmp5; |
3508 | 0 | tmp13 = tmp3 + tmp4; |
3509 | 0 | tmp17 = tmp3 - tmp4; |
3510 | |
|
3511 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*7]; |
3512 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*6]; |
3513 | 0 | tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*5]; |
3514 | 0 | tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*4]; |
3515 | 0 | tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*3]; |
3516 | 0 | tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*2]; |
3517 | 0 | tmp6 = dataptr[DCTSIZE*6] - wsptr[DCTSIZE*1]; |
3518 | 0 | tmp7 = dataptr[DCTSIZE*7] - wsptr[DCTSIZE*0]; |
3519 | |
|
3520 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
3521 | 0 | DESCALE(tmp10 + tmp11 + tmp12 + tmp13, PASS1_BITS+1); |
3522 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
3523 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(1.306562965)) + /* c4[16] = c2[8] */ |
3524 | 0 | MULTIPLY(tmp11 - tmp12, FIX_0_541196100), /* c12[16] = c6[8] */ |
3525 | 0 | CONST_BITS+PASS1_BITS+1); |
3526 | |
|
3527 | 0 | tmp10 = MULTIPLY(tmp17 - tmp15, FIX(0.275899379)) + /* c14[16] = c7[8] */ |
3528 | 0 | MULTIPLY(tmp14 - tmp16, FIX(1.387039845)); /* c2[16] = c1[8] */ |
3529 | |
|
3530 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
3531 | 0 | DESCALE(tmp10 + MULTIPLY(tmp15, FIX(1.451774982)) /* c6+c14 */ |
3532 | 0 | + MULTIPLY(tmp16, FIX(2.172734804)), /* c2+c10 */ |
3533 | 0 | CONST_BITS+PASS1_BITS+1); |
3534 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
3535 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(0.211164243)) /* c2-c6 */ |
3536 | 0 | - MULTIPLY(tmp17, FIX(1.061594338)), /* c10+c14 */ |
3537 | 0 | CONST_BITS+PASS1_BITS+1); |
3538 | | |
3539 | | /* Odd part */ |
3540 | |
|
3541 | 0 | tmp11 = MULTIPLY(tmp0 + tmp1, FIX(1.353318001)) + /* c3 */ |
3542 | 0 | MULTIPLY(tmp6 - tmp7, FIX(0.410524528)); /* c13 */ |
3543 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(1.247225013)) + /* c5 */ |
3544 | 0 | MULTIPLY(tmp5 + tmp7, FIX(0.666655658)); /* c11 */ |
3545 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(1.093201867)) + /* c7 */ |
3546 | 0 | MULTIPLY(tmp4 - tmp7, FIX(0.897167586)); /* c9 */ |
3547 | 0 | tmp14 = MULTIPLY(tmp1 + tmp2, FIX(0.138617169)) + /* c15 */ |
3548 | 0 | MULTIPLY(tmp6 - tmp5, FIX(1.407403738)); /* c1 */ |
3549 | 0 | tmp15 = MULTIPLY(tmp1 + tmp3, - FIX(0.666655658)) + /* -c11 */ |
3550 | 0 | MULTIPLY(tmp4 + tmp6, - FIX(1.247225013)); /* -c5 */ |
3551 | 0 | tmp16 = MULTIPLY(tmp2 + tmp3, - FIX(1.353318001)) + /* -c3 */ |
3552 | 0 | MULTIPLY(tmp5 - tmp4, FIX(0.410524528)); /* c13 */ |
3553 | 0 | tmp10 = tmp11 + tmp12 + tmp13 - |
3554 | 0 | MULTIPLY(tmp0, FIX(2.286341144)) + /* c7+c5+c3-c1 */ |
3555 | 0 | MULTIPLY(tmp7, FIX(0.779653625)); /* c15+c13-c11+c9 */ |
3556 | 0 | tmp11 += tmp14 + tmp15 + MULTIPLY(tmp1, FIX(0.071888074)) /* c9-c3-c15+c11 */ |
3557 | 0 | - MULTIPLY(tmp6, FIX(1.663905119)); /* c7+c13+c1-c5 */ |
3558 | 0 | tmp12 += tmp14 + tmp16 - MULTIPLY(tmp2, FIX(1.125726048)) /* c7+c5+c15-c3 */ |
3559 | 0 | + MULTIPLY(tmp5, FIX(1.227391138)); /* c9-c11+c1-c13 */ |
3560 | 0 | tmp13 += tmp15 + tmp16 + MULTIPLY(tmp3, FIX(1.065388962)) /* c15+c3+c11-c7 */ |
3561 | 0 | + MULTIPLY(tmp4, FIX(2.167985692)); /* c1+c13+c5-c9 */ |
3562 | |
|
3563 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+PASS1_BITS+1); |
3564 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+PASS1_BITS+1); |
3565 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+PASS1_BITS+1); |
3566 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+PASS1_BITS+1); |
3567 | |
|
3568 | 0 | dataptr++; /* advance pointer to next column */ |
3569 | 0 | wsptr++; /* advance pointer to next column */ |
3570 | 0 | } |
3571 | 0 | } |
3572 | | |
3573 | | |
3574 | | /* |
3575 | | * Perform the forward DCT on a 7x14 sample block. |
3576 | | * |
3577 | | * 7-point FDCT in pass 1 (rows), 14-point in pass 2 (columns). |
3578 | | */ |
3579 | | |
3580 | | GLOBAL(void) |
3581 | | jpeg_fdct_7x14 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3582 | 0 | { |
3583 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6; |
3584 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16; |
3585 | 0 | INT32 z1, z2, z3; |
3586 | 0 | DCTELEM workspace[8*6]; |
3587 | 0 | DCTELEM *dataptr; |
3588 | 0 | DCTELEM *wsptr; |
3589 | 0 | JSAMPROW elemptr; |
3590 | 0 | int ctr; |
3591 | 0 | SHIFT_TEMPS |
3592 | | |
3593 | | /* Pre-zero output coefficient block. */ |
3594 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
3595 | | |
3596 | | /* Pass 1: process rows. |
3597 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
3598 | | * furthermore, we scale the results by 2**PASS1_BITS. |
3599 | | * 7-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/14). |
3600 | | */ |
3601 | |
|
3602 | 0 | dataptr = data; |
3603 | 0 | ctr = 0; |
3604 | 0 | for (;;) { |
3605 | 0 | elemptr = sample_data[ctr] + start_col; |
3606 | | |
3607 | | /* Even part */ |
3608 | |
|
3609 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[6]); |
3610 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[5]); |
3611 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[4]); |
3612 | 0 | tmp3 = GETJSAMPLE(elemptr[3]); |
3613 | |
|
3614 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[6]); |
3615 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[5]); |
3616 | 0 | tmp12 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[4]); |
3617 | |
|
3618 | 0 | z1 = tmp0 + tmp2; |
3619 | | /* Apply unsigned->signed conversion. */ |
3620 | 0 | dataptr[0] = (DCTELEM) |
3621 | 0 | ((z1 + tmp1 + tmp3 - 7 * CENTERJSAMPLE) << PASS1_BITS); |
3622 | 0 | tmp3 += tmp3; |
3623 | 0 | z1 -= tmp3; |
3624 | 0 | z1 -= tmp3; |
3625 | 0 | z1 = MULTIPLY(z1, FIX(0.353553391)); /* (c2+c6-c4)/2 */ |
3626 | 0 | z2 = MULTIPLY(tmp0 - tmp2, FIX(0.920609002)); /* (c2+c4-c6)/2 */ |
3627 | 0 | z3 = MULTIPLY(tmp1 - tmp2, FIX(0.314692123)); /* c6 */ |
3628 | 0 | dataptr[2] = (DCTELEM) DESCALE(z1 + z2 + z3, CONST_BITS-PASS1_BITS); |
3629 | 0 | z1 -= z2; |
3630 | 0 | z2 = MULTIPLY(tmp0 - tmp1, FIX(0.881747734)); /* c4 */ |
3631 | 0 | dataptr[4] = (DCTELEM) |
3632 | 0 | DESCALE(z2 + z3 - MULTIPLY(tmp1 - tmp3, FIX(0.707106781)), /* c2+c6-c4 */ |
3633 | 0 | CONST_BITS-PASS1_BITS); |
3634 | 0 | dataptr[6] = (DCTELEM) DESCALE(z1 + z2, CONST_BITS-PASS1_BITS); |
3635 | | |
3636 | | /* Odd part */ |
3637 | |
|
3638 | 0 | tmp1 = MULTIPLY(tmp10 + tmp11, FIX(0.935414347)); /* (c3+c1-c5)/2 */ |
3639 | 0 | tmp2 = MULTIPLY(tmp10 - tmp11, FIX(0.170262339)); /* (c3+c5-c1)/2 */ |
3640 | 0 | tmp0 = tmp1 - tmp2; |
3641 | 0 | tmp1 += tmp2; |
3642 | 0 | tmp2 = MULTIPLY(tmp11 + tmp12, - FIX(1.378756276)); /* -c1 */ |
3643 | 0 | tmp1 += tmp2; |
3644 | 0 | tmp3 = MULTIPLY(tmp10 + tmp12, FIX(0.613604268)); /* c5 */ |
3645 | 0 | tmp0 += tmp3; |
3646 | 0 | tmp2 += tmp3 + MULTIPLY(tmp12, FIX(1.870828693)); /* c3+c1-c5 */ |
3647 | |
|
3648 | 0 | dataptr[1] = (DCTELEM) DESCALE(tmp0, CONST_BITS-PASS1_BITS); |
3649 | 0 | dataptr[3] = (DCTELEM) DESCALE(tmp1, CONST_BITS-PASS1_BITS); |
3650 | 0 | dataptr[5] = (DCTELEM) DESCALE(tmp2, CONST_BITS-PASS1_BITS); |
3651 | |
|
3652 | 0 | ctr++; |
3653 | |
|
3654 | 0 | if (ctr != DCTSIZE) { |
3655 | 0 | if (ctr == 14) |
3656 | 0 | break; /* Done. */ |
3657 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
3658 | 0 | } else |
3659 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
3660 | 0 | } |
3661 | | |
3662 | | /* Pass 2: process columns. |
3663 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
3664 | | * by an overall factor of 8. |
3665 | | * We must also scale the output by (8/7)*(8/14) = 32/49, which we |
3666 | | * fold into the constant multipliers: |
3667 | | * 14-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/28) * 32/49. |
3668 | | */ |
3669 | |
|
3670 | 0 | dataptr = data; |
3671 | 0 | wsptr = workspace; |
3672 | 0 | for (ctr = 0; ctr < 7; ctr++) { |
3673 | | /* Even part */ |
3674 | |
|
3675 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*5]; |
3676 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*4]; |
3677 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*3]; |
3678 | 0 | tmp13 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*2]; |
3679 | 0 | tmp4 = dataptr[DCTSIZE*4] + wsptr[DCTSIZE*1]; |
3680 | 0 | tmp5 = dataptr[DCTSIZE*5] + wsptr[DCTSIZE*0]; |
3681 | 0 | tmp6 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7]; |
3682 | |
|
3683 | 0 | tmp10 = tmp0 + tmp6; |
3684 | 0 | tmp14 = tmp0 - tmp6; |
3685 | 0 | tmp11 = tmp1 + tmp5; |
3686 | 0 | tmp15 = tmp1 - tmp5; |
3687 | 0 | tmp12 = tmp2 + tmp4; |
3688 | 0 | tmp16 = tmp2 - tmp4; |
3689 | |
|
3690 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*5]; |
3691 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*4]; |
3692 | 0 | tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*3]; |
3693 | 0 | tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*2]; |
3694 | 0 | tmp4 = dataptr[DCTSIZE*4] - wsptr[DCTSIZE*1]; |
3695 | 0 | tmp5 = dataptr[DCTSIZE*5] - wsptr[DCTSIZE*0]; |
3696 | 0 | tmp6 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7]; |
3697 | |
|
3698 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
3699 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12 + tmp13, |
3700 | 0 | FIX(0.653061224)), /* 32/49 */ |
3701 | 0 | CONST_BITS+PASS1_BITS); |
3702 | 0 | tmp13 += tmp13; |
3703 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
3704 | 0 | DESCALE(MULTIPLY(tmp10 - tmp13, FIX(0.832106052)) + /* c4 */ |
3705 | 0 | MULTIPLY(tmp11 - tmp13, FIX(0.205513223)) - /* c12 */ |
3706 | 0 | MULTIPLY(tmp12 - tmp13, FIX(0.575835255)), /* c8 */ |
3707 | 0 | CONST_BITS+PASS1_BITS); |
3708 | |
|
3709 | 0 | tmp10 = MULTIPLY(tmp14 + tmp15, FIX(0.722074570)); /* c6 */ |
3710 | |
|
3711 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
3712 | 0 | DESCALE(tmp10 + MULTIPLY(tmp14, FIX(0.178337691)) /* c2-c6 */ |
3713 | 0 | + MULTIPLY(tmp16, FIX(0.400721155)), /* c10 */ |
3714 | 0 | CONST_BITS+PASS1_BITS); |
3715 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
3716 | 0 | DESCALE(tmp10 - MULTIPLY(tmp15, FIX(1.122795725)) /* c6+c10 */ |
3717 | 0 | - MULTIPLY(tmp16, FIX(0.900412262)), /* c2 */ |
3718 | 0 | CONST_BITS+PASS1_BITS); |
3719 | | |
3720 | | /* Odd part */ |
3721 | |
|
3722 | 0 | tmp10 = tmp1 + tmp2; |
3723 | 0 | tmp11 = tmp5 - tmp4; |
3724 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) |
3725 | 0 | DESCALE(MULTIPLY(tmp0 - tmp10 + tmp3 - tmp11 - tmp6, |
3726 | 0 | FIX(0.653061224)), /* 32/49 */ |
3727 | 0 | CONST_BITS+PASS1_BITS); |
3728 | 0 | tmp3 = MULTIPLY(tmp3 , FIX(0.653061224)); /* 32/49 */ |
3729 | 0 | tmp10 = MULTIPLY(tmp10, - FIX(0.103406812)); /* -c13 */ |
3730 | 0 | tmp11 = MULTIPLY(tmp11, FIX(0.917760839)); /* c1 */ |
3731 | 0 | tmp10 += tmp11 - tmp3; |
3732 | 0 | tmp11 = MULTIPLY(tmp0 + tmp2, FIX(0.782007410)) + /* c5 */ |
3733 | 0 | MULTIPLY(tmp4 + tmp6, FIX(0.491367823)); /* c9 */ |
3734 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
3735 | 0 | DESCALE(tmp10 + tmp11 - MULTIPLY(tmp2, FIX(1.550341076)) /* c3+c5-c13 */ |
3736 | 0 | + MULTIPLY(tmp4, FIX(0.731428202)), /* c1+c11-c9 */ |
3737 | 0 | CONST_BITS+PASS1_BITS); |
3738 | 0 | tmp12 = MULTIPLY(tmp0 + tmp1, FIX(0.871740478)) + /* c3 */ |
3739 | 0 | MULTIPLY(tmp5 - tmp6, FIX(0.305035186)); /* c11 */ |
3740 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
3741 | 0 | DESCALE(tmp10 + tmp12 - MULTIPLY(tmp1, FIX(0.276965844)) /* c3-c9-c13 */ |
3742 | 0 | - MULTIPLY(tmp5, FIX(2.004803435)), /* c1+c5+c11 */ |
3743 | 0 | CONST_BITS+PASS1_BITS); |
3744 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
3745 | 0 | DESCALE(tmp11 + tmp12 + tmp3 |
3746 | 0 | - MULTIPLY(tmp0, FIX(0.735987049)) /* c3+c5-c1 */ |
3747 | 0 | - MULTIPLY(tmp6, FIX(0.082925825)), /* c9-c11-c13 */ |
3748 | 0 | CONST_BITS+PASS1_BITS); |
3749 | |
|
3750 | 0 | dataptr++; /* advance pointer to next column */ |
3751 | 0 | wsptr++; /* advance pointer to next column */ |
3752 | 0 | } |
3753 | 0 | } |
3754 | | |
3755 | | |
3756 | | /* |
3757 | | * Perform the forward DCT on a 6x12 sample block. |
3758 | | * |
3759 | | * 6-point FDCT in pass 1 (rows), 12-point in pass 2 (columns). |
3760 | | */ |
3761 | | |
3762 | | GLOBAL(void) |
3763 | | jpeg_fdct_6x12 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3764 | 0 | { |
3765 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5; |
3766 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15; |
3767 | 0 | DCTELEM workspace[8*4]; |
3768 | 0 | DCTELEM *dataptr; |
3769 | 0 | DCTELEM *wsptr; |
3770 | 0 | JSAMPROW elemptr; |
3771 | 0 | int ctr; |
3772 | 0 | SHIFT_TEMPS |
3773 | | |
3774 | | /* Pre-zero output coefficient block. */ |
3775 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
3776 | | |
3777 | | /* Pass 1: process rows. |
3778 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
3779 | | * furthermore, we scale the results by 2**PASS1_BITS. |
3780 | | * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12). |
3781 | | */ |
3782 | |
|
3783 | 0 | dataptr = data; |
3784 | 0 | ctr = 0; |
3785 | 0 | for (;;) { |
3786 | 0 | elemptr = sample_data[ctr] + start_col; |
3787 | | |
3788 | | /* Even part */ |
3789 | |
|
3790 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[5]); |
3791 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[4]); |
3792 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) + GETJSAMPLE(elemptr[3]); |
3793 | |
|
3794 | 0 | tmp10 = tmp0 + tmp2; |
3795 | 0 | tmp12 = tmp0 - tmp2; |
3796 | |
|
3797 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[5]); |
3798 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[4]); |
3799 | 0 | tmp2 = GETJSAMPLE(elemptr[2]) - GETJSAMPLE(elemptr[3]); |
3800 | | |
3801 | | /* Apply unsigned->signed conversion. */ |
3802 | 0 | dataptr[0] = (DCTELEM) |
3803 | 0 | ((tmp10 + tmp11 - 6 * CENTERJSAMPLE) << PASS1_BITS); |
3804 | 0 | dataptr[2] = (DCTELEM) |
3805 | 0 | DESCALE(MULTIPLY(tmp12, FIX(1.224744871)), /* c2 */ |
3806 | 0 | CONST_BITS-PASS1_BITS); |
3807 | 0 | dataptr[4] = (DCTELEM) |
3808 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(0.707106781)), /* c4 */ |
3809 | 0 | CONST_BITS-PASS1_BITS); |
3810 | | |
3811 | | /* Odd part */ |
3812 | |
|
3813 | 0 | tmp10 = DESCALE(MULTIPLY(tmp0 + tmp2, FIX(0.366025404)), /* c5 */ |
3814 | 0 | CONST_BITS-PASS1_BITS); |
3815 | |
|
3816 | 0 | dataptr[1] = (DCTELEM) (tmp10 + ((tmp0 + tmp1) << PASS1_BITS)); |
3817 | 0 | dataptr[3] = (DCTELEM) ((tmp0 - tmp1 - tmp2) << PASS1_BITS); |
3818 | 0 | dataptr[5] = (DCTELEM) (tmp10 + ((tmp2 - tmp1) << PASS1_BITS)); |
3819 | |
|
3820 | 0 | ctr++; |
3821 | |
|
3822 | 0 | if (ctr != DCTSIZE) { |
3823 | 0 | if (ctr == 12) |
3824 | 0 | break; /* Done. */ |
3825 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
3826 | 0 | } else |
3827 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
3828 | 0 | } |
3829 | | |
3830 | | /* Pass 2: process columns. |
3831 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
3832 | | * by an overall factor of 8. |
3833 | | * We must also scale the output by (8/6)*(8/12) = 8/9, which we |
3834 | | * fold into the constant multipliers: |
3835 | | * 12-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/24) * 8/9. |
3836 | | */ |
3837 | |
|
3838 | 0 | dataptr = data; |
3839 | 0 | wsptr = workspace; |
3840 | 0 | for (ctr = 0; ctr < 6; ctr++) { |
3841 | | /* Even part */ |
3842 | |
|
3843 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*3]; |
3844 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*2]; |
3845 | 0 | tmp2 = dataptr[DCTSIZE*2] + wsptr[DCTSIZE*1]; |
3846 | 0 | tmp3 = dataptr[DCTSIZE*3] + wsptr[DCTSIZE*0]; |
3847 | 0 | tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*7]; |
3848 | 0 | tmp5 = dataptr[DCTSIZE*5] + dataptr[DCTSIZE*6]; |
3849 | |
|
3850 | 0 | tmp10 = tmp0 + tmp5; |
3851 | 0 | tmp13 = tmp0 - tmp5; |
3852 | 0 | tmp11 = tmp1 + tmp4; |
3853 | 0 | tmp14 = tmp1 - tmp4; |
3854 | 0 | tmp12 = tmp2 + tmp3; |
3855 | 0 | tmp15 = tmp2 - tmp3; |
3856 | |
|
3857 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*3]; |
3858 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*2]; |
3859 | 0 | tmp2 = dataptr[DCTSIZE*2] - wsptr[DCTSIZE*1]; |
3860 | 0 | tmp3 = dataptr[DCTSIZE*3] - wsptr[DCTSIZE*0]; |
3861 | 0 | tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*7]; |
3862 | 0 | tmp5 = dataptr[DCTSIZE*5] - dataptr[DCTSIZE*6]; |
3863 | |
|
3864 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
3865 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(0.888888889)), /* 8/9 */ |
3866 | 0 | CONST_BITS+PASS1_BITS); |
3867 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
3868 | 0 | DESCALE(MULTIPLY(tmp13 - tmp14 - tmp15, FIX(0.888888889)), /* 8/9 */ |
3869 | 0 | CONST_BITS+PASS1_BITS); |
3870 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
3871 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.088662108)), /* c4 */ |
3872 | 0 | CONST_BITS+PASS1_BITS); |
3873 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
3874 | 0 | DESCALE(MULTIPLY(tmp14 - tmp15, FIX(0.888888889)) + /* 8/9 */ |
3875 | 0 | MULTIPLY(tmp13 + tmp15, FIX(1.214244803)), /* c2 */ |
3876 | 0 | CONST_BITS+PASS1_BITS); |
3877 | | |
3878 | | /* Odd part */ |
3879 | |
|
3880 | 0 | tmp10 = MULTIPLY(tmp1 + tmp4, FIX(0.481063200)); /* c9 */ |
3881 | 0 | tmp14 = tmp10 + MULTIPLY(tmp1, FIX(0.680326102)); /* c3-c9 */ |
3882 | 0 | tmp15 = tmp10 - MULTIPLY(tmp4, FIX(1.642452502)); /* c3+c9 */ |
3883 | 0 | tmp12 = MULTIPLY(tmp0 + tmp2, FIX(0.997307603)); /* c5 */ |
3884 | 0 | tmp13 = MULTIPLY(tmp0 + tmp3, FIX(0.765261039)); /* c7 */ |
3885 | 0 | tmp10 = tmp12 + tmp13 + tmp14 - MULTIPLY(tmp0, FIX(0.516244403)) /* c5+c7-c1 */ |
3886 | 0 | + MULTIPLY(tmp5, FIX(0.164081699)); /* c11 */ |
3887 | 0 | tmp11 = MULTIPLY(tmp2 + tmp3, - FIX(0.164081699)); /* -c11 */ |
3888 | 0 | tmp12 += tmp11 - tmp15 - MULTIPLY(tmp2, FIX(2.079550144)) /* c1+c5-c11 */ |
3889 | 0 | + MULTIPLY(tmp5, FIX(0.765261039)); /* c7 */ |
3890 | 0 | tmp13 += tmp11 - tmp14 + MULTIPLY(tmp3, FIX(0.645144899)) /* c1+c11-c7 */ |
3891 | 0 | - MULTIPLY(tmp5, FIX(0.997307603)); /* c5 */ |
3892 | 0 | tmp11 = tmp15 + MULTIPLY(tmp0 - tmp3, FIX(1.161389302)) /* c3 */ |
3893 | 0 | - MULTIPLY(tmp2 + tmp5, FIX(0.481063200)); /* c9 */ |
3894 | |
|
3895 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10, CONST_BITS+PASS1_BITS); |
3896 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp11, CONST_BITS+PASS1_BITS); |
3897 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12, CONST_BITS+PASS1_BITS); |
3898 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp13, CONST_BITS+PASS1_BITS); |
3899 | |
|
3900 | 0 | dataptr++; /* advance pointer to next column */ |
3901 | 0 | wsptr++; /* advance pointer to next column */ |
3902 | 0 | } |
3903 | 0 | } |
3904 | | |
3905 | | |
3906 | | /* |
3907 | | * Perform the forward DCT on a 5x10 sample block. |
3908 | | * |
3909 | | * 5-point FDCT in pass 1 (rows), 10-point in pass 2 (columns). |
3910 | | */ |
3911 | | |
3912 | | GLOBAL(void) |
3913 | | jpeg_fdct_5x10 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
3914 | 0 | { |
3915 | 0 | INT32 tmp0, tmp1, tmp2, tmp3, tmp4; |
3916 | 0 | INT32 tmp10, tmp11, tmp12, tmp13, tmp14; |
3917 | 0 | DCTELEM workspace[8*2]; |
3918 | 0 | DCTELEM *dataptr; |
3919 | 0 | DCTELEM *wsptr; |
3920 | 0 | JSAMPROW elemptr; |
3921 | 0 | int ctr; |
3922 | 0 | SHIFT_TEMPS |
3923 | | |
3924 | | /* Pre-zero output coefficient block. */ |
3925 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
3926 | | |
3927 | | /* Pass 1: process rows. |
3928 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
3929 | | * furthermore, we scale the results by 2**PASS1_BITS. |
3930 | | * 5-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/10). |
3931 | | */ |
3932 | |
|
3933 | 0 | dataptr = data; |
3934 | 0 | ctr = 0; |
3935 | 0 | for (;;) { |
3936 | 0 | elemptr = sample_data[ctr] + start_col; |
3937 | | |
3938 | | /* Even part */ |
3939 | |
|
3940 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[4]); |
3941 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[3]); |
3942 | 0 | tmp2 = GETJSAMPLE(elemptr[2]); |
3943 | |
|
3944 | 0 | tmp10 = tmp0 + tmp1; |
3945 | 0 | tmp11 = tmp0 - tmp1; |
3946 | |
|
3947 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[4]); |
3948 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[3]); |
3949 | | |
3950 | | /* Apply unsigned->signed conversion. */ |
3951 | 0 | dataptr[0] = (DCTELEM) |
3952 | 0 | ((tmp10 + tmp2 - 5 * CENTERJSAMPLE) << PASS1_BITS); |
3953 | 0 | tmp11 = MULTIPLY(tmp11, FIX(0.790569415)); /* (c2+c4)/2 */ |
3954 | 0 | tmp10 -= tmp2 << 2; |
3955 | 0 | tmp10 = MULTIPLY(tmp10, FIX(0.353553391)); /* (c2-c4)/2 */ |
3956 | 0 | dataptr[2] = (DCTELEM) DESCALE(tmp11 + tmp10, CONST_BITS-PASS1_BITS); |
3957 | 0 | dataptr[4] = (DCTELEM) DESCALE(tmp11 - tmp10, CONST_BITS-PASS1_BITS); |
3958 | | |
3959 | | /* Odd part */ |
3960 | |
|
3961 | 0 | tmp10 = MULTIPLY(tmp0 + tmp1, FIX(0.831253876)); /* c3 */ |
3962 | |
|
3963 | 0 | dataptr[1] = (DCTELEM) |
3964 | 0 | DESCALE(tmp10 + MULTIPLY(tmp0, FIX(0.513743148)), /* c1-c3 */ |
3965 | 0 | CONST_BITS-PASS1_BITS); |
3966 | 0 | dataptr[3] = (DCTELEM) |
3967 | 0 | DESCALE(tmp10 - MULTIPLY(tmp1, FIX(2.176250899)), /* c1+c3 */ |
3968 | 0 | CONST_BITS-PASS1_BITS); |
3969 | |
|
3970 | 0 | ctr++; |
3971 | |
|
3972 | 0 | if (ctr != DCTSIZE) { |
3973 | 0 | if (ctr == 10) |
3974 | 0 | break; /* Done. */ |
3975 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
3976 | 0 | } else |
3977 | 0 | dataptr = workspace; /* switch pointer to extended workspace */ |
3978 | 0 | } |
3979 | | |
3980 | | /* Pass 2: process columns. |
3981 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
3982 | | * by an overall factor of 8. |
3983 | | * We must also scale the output by (8/5)*(8/10) = 32/25, which we |
3984 | | * fold into the constant multipliers: |
3985 | | * 10-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/20) * 32/25. |
3986 | | */ |
3987 | |
|
3988 | 0 | dataptr = data; |
3989 | 0 | wsptr = workspace; |
3990 | 0 | for (ctr = 0; ctr < 5; ctr++) { |
3991 | | /* Even part */ |
3992 | |
|
3993 | 0 | tmp0 = dataptr[DCTSIZE*0] + wsptr[DCTSIZE*1]; |
3994 | 0 | tmp1 = dataptr[DCTSIZE*1] + wsptr[DCTSIZE*0]; |
3995 | 0 | tmp12 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*7]; |
3996 | 0 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*6]; |
3997 | 0 | tmp4 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5]; |
3998 | |
|
3999 | 0 | tmp10 = tmp0 + tmp4; |
4000 | 0 | tmp13 = tmp0 - tmp4; |
4001 | 0 | tmp11 = tmp1 + tmp3; |
4002 | 0 | tmp14 = tmp1 - tmp3; |
4003 | |
|
4004 | 0 | tmp0 = dataptr[DCTSIZE*0] - wsptr[DCTSIZE*1]; |
4005 | 0 | tmp1 = dataptr[DCTSIZE*1] - wsptr[DCTSIZE*0]; |
4006 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*7]; |
4007 | 0 | tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*6]; |
4008 | 0 | tmp4 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5]; |
4009 | |
|
4010 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
4011 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11 + tmp12, FIX(1.28)), /* 32/25 */ |
4012 | 0 | CONST_BITS+PASS1_BITS); |
4013 | 0 | tmp12 += tmp12; |
4014 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
4015 | 0 | DESCALE(MULTIPLY(tmp10 - tmp12, FIX(1.464477191)) - /* c4 */ |
4016 | 0 | MULTIPLY(tmp11 - tmp12, FIX(0.559380511)), /* c8 */ |
4017 | 0 | CONST_BITS+PASS1_BITS); |
4018 | 0 | tmp10 = MULTIPLY(tmp13 + tmp14, FIX(1.064004961)); /* c6 */ |
4019 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
4020 | 0 | DESCALE(tmp10 + MULTIPLY(tmp13, FIX(0.657591230)), /* c2-c6 */ |
4021 | 0 | CONST_BITS+PASS1_BITS); |
4022 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
4023 | 0 | DESCALE(tmp10 - MULTIPLY(tmp14, FIX(2.785601151)), /* c2+c6 */ |
4024 | 0 | CONST_BITS+PASS1_BITS); |
4025 | | |
4026 | | /* Odd part */ |
4027 | |
|
4028 | 0 | tmp10 = tmp0 + tmp4; |
4029 | 0 | tmp11 = tmp1 - tmp3; |
4030 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
4031 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp2, FIX(1.28)), /* 32/25 */ |
4032 | 0 | CONST_BITS+PASS1_BITS); |
4033 | 0 | tmp2 = MULTIPLY(tmp2, FIX(1.28)); /* 32/25 */ |
4034 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
4035 | 0 | DESCALE(MULTIPLY(tmp0, FIX(1.787906876)) + /* c1 */ |
4036 | 0 | MULTIPLY(tmp1, FIX(1.612894094)) + tmp2 + /* c3 */ |
4037 | 0 | MULTIPLY(tmp3, FIX(0.821810588)) + /* c7 */ |
4038 | 0 | MULTIPLY(tmp4, FIX(0.283176630)), /* c9 */ |
4039 | 0 | CONST_BITS+PASS1_BITS); |
4040 | 0 | tmp12 = MULTIPLY(tmp0 - tmp4, FIX(1.217352341)) - /* (c3+c7)/2 */ |
4041 | 0 | MULTIPLY(tmp1 + tmp3, FIX(0.752365123)); /* (c1-c9)/2 */ |
4042 | 0 | tmp13 = MULTIPLY(tmp10 + tmp11, FIX(0.395541753)) + /* (c3-c7)/2 */ |
4043 | 0 | MULTIPLY(tmp11, FIX(0.64)) - tmp2; /* 16/25 */ |
4044 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp12 + tmp13, CONST_BITS+PASS1_BITS); |
4045 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp12 - tmp13, CONST_BITS+PASS1_BITS); |
4046 | |
|
4047 | 0 | dataptr++; /* advance pointer to next column */ |
4048 | 0 | wsptr++; /* advance pointer to next column */ |
4049 | 0 | } |
4050 | 0 | } |
4051 | | |
4052 | | |
4053 | | /* |
4054 | | * Perform the forward DCT on a 4x8 sample block. |
4055 | | * |
4056 | | * 4-point FDCT in pass 1 (rows), 8-point in pass 2 (columns). |
4057 | | */ |
4058 | | |
4059 | | GLOBAL(void) |
4060 | | jpeg_fdct_4x8 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
4061 | 0 | { |
4062 | 0 | INT32 tmp0, tmp1, tmp2, tmp3; |
4063 | 0 | INT32 tmp10, tmp11, tmp12, tmp13; |
4064 | 0 | INT32 z1; |
4065 | 0 | DCTELEM *dataptr; |
4066 | 0 | JSAMPROW elemptr; |
4067 | 0 | int ctr; |
4068 | 0 | SHIFT_TEMPS |
4069 | | |
4070 | | /* Pre-zero output coefficient block. */ |
4071 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
4072 | | |
4073 | | /* Pass 1: process rows. |
4074 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
4075 | | * furthermore, we scale the results by 2**PASS1_BITS. |
4076 | | * We must also scale the output by 8/4 = 2, which we add here. |
4077 | | * 4-point FDCT kernel, |
4078 | | * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. |
4079 | | */ |
4080 | |
|
4081 | 0 | dataptr = data; |
4082 | 0 | for (ctr = 0; ctr < DCTSIZE; ctr++) { |
4083 | 0 | elemptr = sample_data[ctr] + start_col; |
4084 | | |
4085 | | /* Even part */ |
4086 | |
|
4087 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[3]); |
4088 | 0 | tmp1 = GETJSAMPLE(elemptr[1]) + GETJSAMPLE(elemptr[2]); |
4089 | |
|
4090 | 0 | tmp10 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[3]); |
4091 | 0 | tmp11 = GETJSAMPLE(elemptr[1]) - GETJSAMPLE(elemptr[2]); |
4092 | | |
4093 | | /* Apply unsigned->signed conversion. */ |
4094 | 0 | dataptr[0] = (DCTELEM) |
4095 | 0 | ((tmp0 + tmp1 - 4 * CENTERJSAMPLE) << (PASS1_BITS+1)); |
4096 | 0 | dataptr[2] = (DCTELEM) ((tmp0 - tmp1) << (PASS1_BITS+1)); |
4097 | | |
4098 | | /* Odd part */ |
4099 | |
|
4100 | 0 | tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100); /* c6 */ |
4101 | | /* Add fudge factor here for final descale. */ |
4102 | 0 | tmp0 += ONE << (CONST_BITS-PASS1_BITS-2); |
4103 | |
|
4104 | 0 | dataptr[1] = (DCTELEM) |
4105 | 0 | RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */ |
4106 | 0 | CONST_BITS-PASS1_BITS-1); |
4107 | 0 | dataptr[3] = (DCTELEM) |
4108 | 0 | RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */ |
4109 | 0 | CONST_BITS-PASS1_BITS-1); |
4110 | |
|
4111 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
4112 | 0 | } |
4113 | | |
4114 | | /* Pass 2: process columns. |
4115 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
4116 | | * by an overall factor of 8. |
4117 | | * 8-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/16). |
4118 | | */ |
4119 | |
|
4120 | 0 | dataptr = data; |
4121 | 0 | for (ctr = 0; ctr < 4; ctr++) { |
4122 | | /* Even part per LL&M figure 1 --- note that published figure is faulty; |
4123 | | * rotator "c1" should be "c6". |
4124 | | */ |
4125 | |
|
4126 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7]; |
4127 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6]; |
4128 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5]; |
4129 | 0 | tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4]; |
4130 | | |
4131 | | /* Add fudge factor here for final descale. */ |
4132 | 0 | tmp10 = tmp0 + tmp3 + (ONE << (PASS1_BITS-1)); |
4133 | 0 | tmp12 = tmp0 - tmp3; |
4134 | 0 | tmp11 = tmp1 + tmp2; |
4135 | 0 | tmp13 = tmp1 - tmp2; |
4136 | |
|
4137 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7]; |
4138 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6]; |
4139 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5]; |
4140 | 0 | tmp3 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4]; |
4141 | |
|
4142 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) RIGHT_SHIFT(tmp10 + tmp11, PASS1_BITS); |
4143 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) RIGHT_SHIFT(tmp10 - tmp11, PASS1_BITS); |
4144 | |
|
4145 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); /* c6 */ |
4146 | | /* Add fudge factor here for final descale. */ |
4147 | 0 | z1 += ONE << (CONST_BITS+PASS1_BITS-1); |
4148 | |
|
4149 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
4150 | 0 | RIGHT_SHIFT(z1 + MULTIPLY(tmp12, FIX_0_765366865), /* c2-c6 */ |
4151 | 0 | CONST_BITS+PASS1_BITS); |
4152 | 0 | dataptr[DCTSIZE*6] = (DCTELEM) |
4153 | 0 | RIGHT_SHIFT(z1 - MULTIPLY(tmp13, FIX_1_847759065), /* c2+c6 */ |
4154 | 0 | CONST_BITS+PASS1_BITS); |
4155 | | |
4156 | | /* Odd part per figure 8 --- note paper omits factor of sqrt(2). |
4157 | | * i0..i3 in the paper are tmp0..tmp3 here. |
4158 | | */ |
4159 | |
|
4160 | 0 | tmp12 = tmp0 + tmp2; |
4161 | 0 | tmp13 = tmp1 + tmp3; |
4162 | |
|
4163 | 0 | z1 = MULTIPLY(tmp12 + tmp13, FIX_1_175875602); /* c3 */ |
4164 | | /* Add fudge factor here for final descale. */ |
4165 | 0 | z1 += ONE << (CONST_BITS+PASS1_BITS-1); |
4166 | |
|
4167 | 0 | tmp12 = MULTIPLY(tmp12, - FIX_0_390180644); /* -c3+c5 */ |
4168 | 0 | tmp13 = MULTIPLY(tmp13, - FIX_1_961570560); /* -c3-c5 */ |
4169 | 0 | tmp12 += z1; |
4170 | 0 | tmp13 += z1; |
4171 | |
|
4172 | 0 | z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */ |
4173 | 0 | tmp0 = MULTIPLY(tmp0, FIX_1_501321110); /* c1+c3-c5-c7 */ |
4174 | 0 | tmp3 = MULTIPLY(tmp3, FIX_0_298631336); /* -c1+c3+c5-c7 */ |
4175 | 0 | tmp0 += z1 + tmp12; |
4176 | 0 | tmp3 += z1 + tmp13; |
4177 | |
|
4178 | 0 | z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */ |
4179 | 0 | tmp1 = MULTIPLY(tmp1, FIX_3_072711026); /* c1+c3+c5-c7 */ |
4180 | 0 | tmp2 = MULTIPLY(tmp2, FIX_2_053119869); /* c1+c3-c5+c7 */ |
4181 | 0 | tmp1 += z1 + tmp13; |
4182 | 0 | tmp2 += z1 + tmp12; |
4183 | |
|
4184 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) RIGHT_SHIFT(tmp0, CONST_BITS+PASS1_BITS); |
4185 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) RIGHT_SHIFT(tmp1, CONST_BITS+PASS1_BITS); |
4186 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) RIGHT_SHIFT(tmp2, CONST_BITS+PASS1_BITS); |
4187 | 0 | dataptr[DCTSIZE*7] = (DCTELEM) RIGHT_SHIFT(tmp3, CONST_BITS+PASS1_BITS); |
4188 | |
|
4189 | 0 | dataptr++; /* advance pointer to next column */ |
4190 | 0 | } |
4191 | 0 | } |
4192 | | |
4193 | | |
4194 | | /* |
4195 | | * Perform the forward DCT on a 3x6 sample block. |
4196 | | * |
4197 | | * 3-point FDCT in pass 1 (rows), 6-point in pass 2 (columns). |
4198 | | */ |
4199 | | |
4200 | | GLOBAL(void) |
4201 | | jpeg_fdct_3x6 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
4202 | 0 | { |
4203 | 0 | INT32 tmp0, tmp1, tmp2; |
4204 | 0 | INT32 tmp10, tmp11, tmp12; |
4205 | 0 | DCTELEM *dataptr; |
4206 | 0 | JSAMPROW elemptr; |
4207 | 0 | int ctr; |
4208 | 0 | SHIFT_TEMPS |
4209 | | |
4210 | | /* Pre-zero output coefficient block. */ |
4211 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
4212 | | |
4213 | | /* Pass 1: process rows. |
4214 | | * Note results are scaled up by sqrt(8) compared to a true DCT; |
4215 | | * furthermore, we scale the results by 2**PASS1_BITS. |
4216 | | * We scale the results further by 2 as part of output adaption |
4217 | | * scaling for different DCT size. |
4218 | | * 3-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/6). |
4219 | | */ |
4220 | |
|
4221 | 0 | dataptr = data; |
4222 | 0 | for (ctr = 0; ctr < 6; ctr++) { |
4223 | 0 | elemptr = sample_data[ctr] + start_col; |
4224 | | |
4225 | | /* Even part */ |
4226 | |
|
4227 | 0 | tmp0 = GETJSAMPLE(elemptr[0]) + GETJSAMPLE(elemptr[2]); |
4228 | 0 | tmp1 = GETJSAMPLE(elemptr[1]); |
4229 | |
|
4230 | 0 | tmp2 = GETJSAMPLE(elemptr[0]) - GETJSAMPLE(elemptr[2]); |
4231 | | |
4232 | | /* Apply unsigned->signed conversion. */ |
4233 | 0 | dataptr[0] = (DCTELEM) |
4234 | 0 | ((tmp0 + tmp1 - 3 * CENTERJSAMPLE) << (PASS1_BITS+1)); |
4235 | 0 | dataptr[2] = (DCTELEM) |
4236 | 0 | DESCALE(MULTIPLY(tmp0 - tmp1 - tmp1, FIX(0.707106781)), /* c2 */ |
4237 | 0 | CONST_BITS-PASS1_BITS-1); |
4238 | | |
4239 | | /* Odd part */ |
4240 | |
|
4241 | 0 | dataptr[1] = (DCTELEM) |
4242 | 0 | DESCALE(MULTIPLY(tmp2, FIX(1.224744871)), /* c1 */ |
4243 | 0 | CONST_BITS-PASS1_BITS-1); |
4244 | |
|
4245 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
4246 | 0 | } |
4247 | | |
4248 | | /* Pass 2: process columns. |
4249 | | * We remove the PASS1_BITS scaling, but leave the results scaled up |
4250 | | * by an overall factor of 8. |
4251 | | * We must also scale the output by (8/6)*(8/3) = 32/9, which we partially |
4252 | | * fold into the constant multipliers (other part was done in pass 1): |
4253 | | * 6-point FDCT kernel, cK represents sqrt(2) * cos(K*pi/12) * 16/9. |
4254 | | */ |
4255 | |
|
4256 | 0 | dataptr = data; |
4257 | 0 | for (ctr = 0; ctr < 3; ctr++) { |
4258 | | /* Even part */ |
4259 | |
|
4260 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*5]; |
4261 | 0 | tmp11 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*4]; |
4262 | 0 | tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3]; |
4263 | |
|
4264 | 0 | tmp10 = tmp0 + tmp2; |
4265 | 0 | tmp12 = tmp0 - tmp2; |
4266 | |
|
4267 | 0 | tmp0 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*5]; |
4268 | 0 | tmp1 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*4]; |
4269 | 0 | tmp2 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3]; |
4270 | |
|
4271 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) |
4272 | 0 | DESCALE(MULTIPLY(tmp10 + tmp11, FIX(1.777777778)), /* 16/9 */ |
4273 | 0 | CONST_BITS+PASS1_BITS); |
4274 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) |
4275 | 0 | DESCALE(MULTIPLY(tmp12, FIX(2.177324216)), /* c2 */ |
4276 | 0 | CONST_BITS+PASS1_BITS); |
4277 | 0 | dataptr[DCTSIZE*4] = (DCTELEM) |
4278 | 0 | DESCALE(MULTIPLY(tmp10 - tmp11 - tmp11, FIX(1.257078722)), /* c4 */ |
4279 | 0 | CONST_BITS+PASS1_BITS); |
4280 | | |
4281 | | /* Odd part */ |
4282 | |
|
4283 | 0 | tmp10 = MULTIPLY(tmp0 + tmp2, FIX(0.650711829)); /* c5 */ |
4284 | |
|
4285 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
4286 | 0 | DESCALE(tmp10 + MULTIPLY(tmp0 + tmp1, FIX(1.777777778)), /* 16/9 */ |
4287 | 0 | CONST_BITS+PASS1_BITS); |
4288 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
4289 | 0 | DESCALE(MULTIPLY(tmp0 - tmp1 - tmp2, FIX(1.777777778)), /* 16/9 */ |
4290 | 0 | CONST_BITS+PASS1_BITS); |
4291 | 0 | dataptr[DCTSIZE*5] = (DCTELEM) |
4292 | 0 | DESCALE(tmp10 + MULTIPLY(tmp2 - tmp1, FIX(1.777777778)), /* 16/9 */ |
4293 | 0 | CONST_BITS+PASS1_BITS); |
4294 | |
|
4295 | 0 | dataptr++; /* advance pointer to next column */ |
4296 | 0 | } |
4297 | 0 | } |
4298 | | |
4299 | | |
4300 | | /* |
4301 | | * Perform the forward DCT on a 2x4 sample block. |
4302 | | * |
4303 | | * 2-point FDCT in pass 1 (rows), 4-point in pass 2 (columns). |
4304 | | */ |
4305 | | |
4306 | | GLOBAL(void) |
4307 | | jpeg_fdct_2x4 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
4308 | 0 | { |
4309 | 0 | INT32 tmp0, tmp1; |
4310 | 0 | INT32 tmp10, tmp11; |
4311 | 0 | DCTELEM *dataptr; |
4312 | 0 | JSAMPROW elemptr; |
4313 | 0 | int ctr; |
4314 | 0 | SHIFT_TEMPS |
4315 | | |
4316 | | /* Pre-zero output coefficient block. */ |
4317 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
4318 | | |
4319 | | /* Pass 1: process rows. |
4320 | | * Note results are scaled up by sqrt(8) compared to a true DCT. |
4321 | | */ |
4322 | |
|
4323 | 0 | dataptr = data; |
4324 | 0 | for (ctr = 0; ctr < 4; ctr++) { |
4325 | 0 | elemptr = sample_data[ctr] + start_col; |
4326 | | |
4327 | | /* Even part */ |
4328 | |
|
4329 | 0 | tmp0 = GETJSAMPLE(elemptr[0]); |
4330 | 0 | tmp1 = GETJSAMPLE(elemptr[1]); |
4331 | | |
4332 | | /* Apply unsigned->signed conversion. */ |
4333 | 0 | dataptr[0] = (DCTELEM) (tmp0 + tmp1 - 2 * CENTERJSAMPLE); |
4334 | | |
4335 | | /* Odd part */ |
4336 | |
|
4337 | 0 | dataptr[1] = (DCTELEM) (tmp0 - tmp1); |
4338 | |
|
4339 | 0 | dataptr += DCTSIZE; /* advance pointer to next row */ |
4340 | 0 | } |
4341 | | |
4342 | | /* Pass 2: process columns. |
4343 | | * We leave the results scaled up by an overall factor of 8. |
4344 | | * We must also scale the output by (8/2)*(8/4) = 2**3. |
4345 | | * 4-point FDCT kernel, |
4346 | | * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point FDCT]. |
4347 | | */ |
4348 | |
|
4349 | 0 | dataptr = data; |
4350 | 0 | for (ctr = 0; ctr < 2; ctr++) { |
4351 | | /* Even part */ |
4352 | |
|
4353 | 0 | tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*3]; |
4354 | 0 | tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*2]; |
4355 | |
|
4356 | 0 | tmp10 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*3]; |
4357 | 0 | tmp11 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*2]; |
4358 | |
|
4359 | 0 | dataptr[DCTSIZE*0] = (DCTELEM) ((tmp0 + tmp1) << 3); |
4360 | 0 | dataptr[DCTSIZE*2] = (DCTELEM) ((tmp0 - tmp1) << 3); |
4361 | | |
4362 | | /* Odd part */ |
4363 | |
|
4364 | 0 | tmp0 = MULTIPLY(tmp10 + tmp11, FIX_0_541196100); /* c6 */ |
4365 | | /* Add fudge factor here for final descale. */ |
4366 | 0 | tmp0 += ONE << (CONST_BITS-3-1); |
4367 | |
|
4368 | 0 | dataptr[DCTSIZE*1] = (DCTELEM) |
4369 | 0 | RIGHT_SHIFT(tmp0 + MULTIPLY(tmp10, FIX_0_765366865), /* c2-c6 */ |
4370 | 0 | CONST_BITS-3); |
4371 | 0 | dataptr[DCTSIZE*3] = (DCTELEM) |
4372 | 0 | RIGHT_SHIFT(tmp0 - MULTIPLY(tmp11, FIX_1_847759065), /* c2+c6 */ |
4373 | 0 | CONST_BITS-3); |
4374 | |
|
4375 | 0 | dataptr++; /* advance pointer to next column */ |
4376 | 0 | } |
4377 | 0 | } |
4378 | | |
4379 | | |
4380 | | /* |
4381 | | * Perform the forward DCT on a 1x2 sample block. |
4382 | | * |
4383 | | * 1-point FDCT in pass 1 (rows), 2-point in pass 2 (columns). |
4384 | | */ |
4385 | | |
4386 | | GLOBAL(void) |
4387 | | jpeg_fdct_1x2 (DCTELEM * data, JSAMPARRAY sample_data, JDIMENSION start_col) |
4388 | 0 | { |
4389 | 0 | DCTELEM tmp0, tmp1; |
4390 | | |
4391 | | /* Pre-zero output coefficient block. */ |
4392 | 0 | MEMZERO(data, SIZEOF(DCTELEM) * DCTSIZE2); |
4393 | | |
4394 | | /* Pass 1: empty. */ |
4395 | | |
4396 | | /* Pass 2: process columns. |
4397 | | * We leave the results scaled up by an overall factor of 8. |
4398 | | * We must also scale the output by (8/1)*(8/2) = 2**5. |
4399 | | */ |
4400 | | |
4401 | | /* Even part */ |
4402 | |
|
4403 | 0 | tmp0 = GETJSAMPLE(sample_data[0][start_col]); |
4404 | 0 | tmp1 = GETJSAMPLE(sample_data[1][start_col]); |
4405 | | |
4406 | | /* Apply unsigned->signed conversion. */ |
4407 | 0 | data[DCTSIZE*0] = (tmp0 + tmp1 - 2 * CENTERJSAMPLE) << 5; |
4408 | | |
4409 | | /* Odd part */ |
4410 | |
|
4411 | 0 | data[DCTSIZE*1] = (tmp0 - tmp1) << 5; |
4412 | 0 | } |
4413 | | |
4414 | | #endif /* DCT_SCALING_SUPPORTED */ |
4415 | | #endif /* DCT_ISLOW_SUPPORTED */ |