3dphotoshoot

annotate libs/libjpeg/jidctint.c @ 27:3d082c566b53

find changesets by author, revision, files, or words in the commit message
fixed all the bugs, pc version works
author John Tsiombikas <nuclear@member.fsf.org>
date Thu, 18 Jun 2015 04:32:25 +0300
parents
children
rev   line source
nuclear@14 1 /*
nuclear@14 2 * jidctint.c
nuclear@14 3 *
nuclear@14 4 * Copyright (C) 1991-1998, Thomas G. Lane.
nuclear@14 5 * This file is part of the Independent JPEG Group's software.
nuclear@14 6 * For conditions of distribution and use, see the accompanying README file.
nuclear@14 7 *
nuclear@14 8 * This file contains a slow-but-accurate integer implementation of the
nuclear@14 9 * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine
nuclear@14 10 * must also perform dequantization of the input coefficients.
nuclear@14 11 *
nuclear@14 12 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
nuclear@14 13 * on each row (or vice versa, but it's more convenient to emit a row at
nuclear@14 14 * a time). Direct algorithms are also available, but they are much more
nuclear@14 15 * complex and seem not to be any faster when reduced to code.
nuclear@14 16 *
nuclear@14 17 * This implementation is based on an algorithm described in
nuclear@14 18 * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
nuclear@14 19 * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
nuclear@14 20 * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
nuclear@14 21 * The primary algorithm described there uses 11 multiplies and 29 adds.
nuclear@14 22 * We use their alternate method with 12 multiplies and 32 adds.
nuclear@14 23 * The advantage of this method is that no data path contains more than one
nuclear@14 24 * multiplication; this allows a very simple and accurate implementation in
nuclear@14 25 * scaled fixed-point arithmetic, with a minimal number of shifts.
nuclear@14 26 */
nuclear@14 27
nuclear@14 28 #define JPEG_INTERNALS
nuclear@14 29 #include "jinclude.h"
nuclear@14 30 #include "jpeglib.h"
nuclear@14 31 #include "jdct.h" /* Private declarations for DCT subsystem */
nuclear@14 32
nuclear@14 33 #ifdef DCT_ISLOW_SUPPORTED
nuclear@14 34
nuclear@14 35
nuclear@14 36 /*
nuclear@14 37 * This module is specialized to the case DCTSIZE = 8.
nuclear@14 38 */
nuclear@14 39
nuclear@14 40 #if DCTSIZE != 8
nuclear@14 41 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
nuclear@14 42 #endif
nuclear@14 43
nuclear@14 44
nuclear@14 45 /*
nuclear@14 46 * The poop on this scaling stuff is as follows:
nuclear@14 47 *
nuclear@14 48 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
nuclear@14 49 * larger than the true IDCT outputs. The final outputs are therefore
nuclear@14 50 * a factor of N larger than desired; since N=8 this can be cured by
nuclear@14 51 * a simple right shift at the end of the algorithm. The advantage of
nuclear@14 52 * this arrangement is that we save two multiplications per 1-D IDCT,
nuclear@14 53 * because the y0 and y4 inputs need not be divided by sqrt(N).
nuclear@14 54 *
nuclear@14 55 * We have to do addition and subtraction of the integer inputs, which
nuclear@14 56 * is no problem, and multiplication by fractional constants, which is
nuclear@14 57 * a problem to do in integer arithmetic. We multiply all the constants
nuclear@14 58 * by CONST_SCALE and convert them to integer constants (thus retaining
nuclear@14 59 * CONST_BITS bits of precision in the constants). After doing a
nuclear@14 60 * multiplication we have to divide the product by CONST_SCALE, with proper
nuclear@14 61 * rounding, to produce the correct output. This division can be done
nuclear@14 62 * cheaply as a right shift of CONST_BITS bits. We postpone shifting
nuclear@14 63 * as long as possible so that partial sums can be added together with
nuclear@14 64 * full fractional precision.
nuclear@14 65 *
nuclear@14 66 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
nuclear@14 67 * they are represented to better-than-integral precision. These outputs
nuclear@14 68 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
nuclear@14 69 * with the recommended scaling. (To scale up 12-bit sample data further, an
nuclear@14 70 * intermediate INT32 array would be needed.)
nuclear@14 71 *
nuclear@14 72 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
nuclear@14 73 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
nuclear@14 74 * shows that the values given below are the most effective.
nuclear@14 75 */
nuclear@14 76
nuclear@14 77 #if BITS_IN_JSAMPLE == 8
nuclear@14 78 #define CONST_BITS 13
nuclear@14 79 #define PASS1_BITS 2
nuclear@14 80 #else
nuclear@14 81 #define CONST_BITS 13
nuclear@14 82 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */
nuclear@14 83 #endif
nuclear@14 84
nuclear@14 85 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
nuclear@14 86 * causing a lot of useless floating-point operations at run time.
nuclear@14 87 * To get around this we use the following pre-calculated constants.
nuclear@14 88 * If you change CONST_BITS you may want to add appropriate values.
nuclear@14 89 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
nuclear@14 90 */
nuclear@14 91
nuclear@14 92 #if CONST_BITS == 13
nuclear@14 93 #define FIX_0_298631336 ((INT32) 2446) /* FIX(0.298631336) */
nuclear@14 94 #define FIX_0_390180644 ((INT32) 3196) /* FIX(0.390180644) */
nuclear@14 95 #define FIX_0_541196100 ((INT32) 4433) /* FIX(0.541196100) */
nuclear@14 96 #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
nuclear@14 97 #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
nuclear@14 98 #define FIX_1_175875602 ((INT32) 9633) /* FIX(1.175875602) */
nuclear@14 99 #define FIX_1_501321110 ((INT32) 12299) /* FIX(1.501321110) */
nuclear@14 100 #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
nuclear@14 101 #define FIX_1_961570560 ((INT32) 16069) /* FIX(1.961570560) */
nuclear@14 102 #define FIX_2_053119869 ((INT32) 16819) /* FIX(2.053119869) */
nuclear@14 103 #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
nuclear@14 104 #define FIX_3_072711026 ((INT32) 25172) /* FIX(3.072711026) */
nuclear@14 105 #else
nuclear@14 106 #define FIX_0_298631336 FIX(0.298631336)
nuclear@14 107 #define FIX_0_390180644 FIX(0.390180644)
nuclear@14 108 #define FIX_0_541196100 FIX(0.541196100)
nuclear@14 109 #define FIX_0_765366865 FIX(0.765366865)
nuclear@14 110 #define FIX_0_899976223 FIX(0.899976223)
nuclear@14 111 #define FIX_1_175875602 FIX(1.175875602)
nuclear@14 112 #define FIX_1_501321110 FIX(1.501321110)
nuclear@14 113 #define FIX_1_847759065 FIX(1.847759065)
nuclear@14 114 #define FIX_1_961570560 FIX(1.961570560)
nuclear@14 115 #define FIX_2_053119869 FIX(2.053119869)
nuclear@14 116 #define FIX_2_562915447 FIX(2.562915447)
nuclear@14 117 #define FIX_3_072711026 FIX(3.072711026)
nuclear@14 118 #endif
nuclear@14 119
nuclear@14 120
nuclear@14 121 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
nuclear@14 122 * For 8-bit samples with the recommended scaling, all the variable
nuclear@14 123 * and constant values involved are no more than 16 bits wide, so a
nuclear@14 124 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
nuclear@14 125 * For 12-bit samples, a full 32-bit multiplication will be needed.
nuclear@14 126 */
nuclear@14 127
nuclear@14 128 #if BITS_IN_JSAMPLE == 8
nuclear@14 129 #define MULTIPLY(var,const) MULTIPLY16C16(var,const)
nuclear@14 130 #else
nuclear@14 131 #define MULTIPLY(var,const) ((var) * (const))
nuclear@14 132 #endif
nuclear@14 133
nuclear@14 134
nuclear@14 135 /* Dequantize a coefficient by multiplying it by the multiplier-table
nuclear@14 136 * entry; produce an int result. In this module, both inputs and result
nuclear@14 137 * are 16 bits or less, so either int or short multiply will work.
nuclear@14 138 */
nuclear@14 139
nuclear@14 140 #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))
nuclear@14 141
nuclear@14 142
nuclear@14 143 /*
nuclear@14 144 * Perform dequantization and inverse DCT on one block of coefficients.
nuclear@14 145 */
nuclear@14 146
nuclear@14 147 GLOBAL(void)
nuclear@14 148 jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
nuclear@14 149 JCOEFPTR coef_block,
nuclear@14 150 JSAMPARRAY output_buf, JDIMENSION output_col)
nuclear@14 151 {
nuclear@14 152 INT32 tmp0, tmp1, tmp2, tmp3;
nuclear@14 153 INT32 tmp10, tmp11, tmp12, tmp13;
nuclear@14 154 INT32 z1, z2, z3, z4, z5;
nuclear@14 155 JCOEFPTR inptr;
nuclear@14 156 ISLOW_MULT_TYPE * quantptr;
nuclear@14 157 int * wsptr;
nuclear@14 158 JSAMPROW outptr;
nuclear@14 159 JSAMPLE *range_limit = IDCT_range_limit(cinfo);
nuclear@14 160 int ctr;
nuclear@14 161 int workspace[DCTSIZE2]; /* buffers data between passes */
nuclear@14 162 SHIFT_TEMPS
nuclear@14 163
nuclear@14 164 /* Pass 1: process columns from input, store into work array. */
nuclear@14 165 /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
nuclear@14 166 /* furthermore, we scale the results by 2**PASS1_BITS. */
nuclear@14 167
nuclear@14 168 inptr = coef_block;
nuclear@14 169 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
nuclear@14 170 wsptr = workspace;
nuclear@14 171 for (ctr = DCTSIZE; ctr > 0; ctr--) {
nuclear@14 172 /* Due to quantization, we will usually find that many of the input
nuclear@14 173 * coefficients are zero, especially the AC terms. We can exploit this
nuclear@14 174 * by short-circuiting the IDCT calculation for any column in which all
nuclear@14 175 * the AC terms are zero. In that case each output is equal to the
nuclear@14 176 * DC coefficient (with scale factor as needed).
nuclear@14 177 * With typical images and quantization tables, half or more of the
nuclear@14 178 * column DCT calculations can be simplified this way.
nuclear@14 179 */
nuclear@14 180
nuclear@14 181 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
nuclear@14 182 inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
nuclear@14 183 inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
nuclear@14 184 inptr[DCTSIZE*7] == 0) {
nuclear@14 185 /* AC terms all zero */
nuclear@14 186 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
nuclear@14 187
nuclear@14 188 wsptr[DCTSIZE*0] = dcval;
nuclear@14 189 wsptr[DCTSIZE*1] = dcval;
nuclear@14 190 wsptr[DCTSIZE*2] = dcval;
nuclear@14 191 wsptr[DCTSIZE*3] = dcval;
nuclear@14 192 wsptr[DCTSIZE*4] = dcval;
nuclear@14 193 wsptr[DCTSIZE*5] = dcval;
nuclear@14 194 wsptr[DCTSIZE*6] = dcval;
nuclear@14 195 wsptr[DCTSIZE*7] = dcval;
nuclear@14 196
nuclear@14 197 inptr++; /* advance pointers to next column */
nuclear@14 198 quantptr++;
nuclear@14 199 wsptr++;
nuclear@14 200 continue;
nuclear@14 201 }
nuclear@14 202
nuclear@14 203 /* Even part: reverse the even part of the forward DCT. */
nuclear@14 204 /* The rotator is sqrt(2)*c(-6). */
nuclear@14 205
nuclear@14 206 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
nuclear@14 207 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
nuclear@14 208
nuclear@14 209 z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
nuclear@14 210 tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
nuclear@14 211 tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
nuclear@14 212
nuclear@14 213 z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
nuclear@14 214 z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
nuclear@14 215
nuclear@14 216 tmp0 = (z2 + z3) << CONST_BITS;
nuclear@14 217 tmp1 = (z2 - z3) << CONST_BITS;
nuclear@14 218
nuclear@14 219 tmp10 = tmp0 + tmp3;
nuclear@14 220 tmp13 = tmp0 - tmp3;
nuclear@14 221 tmp11 = tmp1 + tmp2;
nuclear@14 222 tmp12 = tmp1 - tmp2;
nuclear@14 223
nuclear@14 224 /* Odd part per figure 8; the matrix is unitary and hence its
nuclear@14 225 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
nuclear@14 226 */
nuclear@14 227
nuclear@14 228 tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
nuclear@14 229 tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
nuclear@14 230 tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
nuclear@14 231 tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
nuclear@14 232
nuclear@14 233 z1 = tmp0 + tmp3;
nuclear@14 234 z2 = tmp1 + tmp2;
nuclear@14 235 z3 = tmp0 + tmp2;
nuclear@14 236 z4 = tmp1 + tmp3;
nuclear@14 237 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
nuclear@14 238
nuclear@14 239 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
nuclear@14 240 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
nuclear@14 241 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
nuclear@14 242 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
nuclear@14 243 z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
nuclear@14 244 z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
nuclear@14 245 z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
nuclear@14 246 z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
nuclear@14 247
nuclear@14 248 z3 += z5;
nuclear@14 249 z4 += z5;
nuclear@14 250
nuclear@14 251 tmp0 += z1 + z3;
nuclear@14 252 tmp1 += z2 + z4;
nuclear@14 253 tmp2 += z2 + z3;
nuclear@14 254 tmp3 += z1 + z4;
nuclear@14 255
nuclear@14 256 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
nuclear@14 257
nuclear@14 258 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
nuclear@14 259 wsptr[DCTSIZE*7] = (int) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
nuclear@14 260 wsptr[DCTSIZE*1] = (int) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
nuclear@14 261 wsptr[DCTSIZE*6] = (int) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
nuclear@14 262 wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
nuclear@14 263 wsptr[DCTSIZE*5] = (int) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
nuclear@14 264 wsptr[DCTSIZE*3] = (int) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
nuclear@14 265 wsptr[DCTSIZE*4] = (int) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
nuclear@14 266
nuclear@14 267 inptr++; /* advance pointers to next column */
nuclear@14 268 quantptr++;
nuclear@14 269 wsptr++;
nuclear@14 270 }
nuclear@14 271
nuclear@14 272 /* Pass 2: process rows from work array, store into output array. */
nuclear@14 273 /* Note that we must descale the results by a factor of 8 == 2**3, */
nuclear@14 274 /* and also undo the PASS1_BITS scaling. */
nuclear@14 275
nuclear@14 276 wsptr = workspace;
nuclear@14 277 for (ctr = 0; ctr < DCTSIZE; ctr++) {
nuclear@14 278 outptr = output_buf[ctr] + output_col;
nuclear@14 279 /* Rows of zeroes can be exploited in the same way as we did with columns.
nuclear@14 280 * However, the column calculation has created many nonzero AC terms, so
nuclear@14 281 * the simplification applies less often (typically 5% to 10% of the time).
nuclear@14 282 * On machines with very fast multiplication, it's possible that the
nuclear@14 283 * test takes more time than it's worth. In that case this section
nuclear@14 284 * may be commented out.
nuclear@14 285 */
nuclear@14 286
nuclear@14 287 #ifndef NO_ZERO_ROW_TEST
nuclear@14 288 if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
nuclear@14 289 wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
nuclear@14 290 /* AC terms all zero */
nuclear@14 291 JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
nuclear@14 292 & RANGE_MASK];
nuclear@14 293
nuclear@14 294 outptr[0] = dcval;
nuclear@14 295 outptr[1] = dcval;
nuclear@14 296 outptr[2] = dcval;
nuclear@14 297 outptr[3] = dcval;
nuclear@14 298 outptr[4] = dcval;
nuclear@14 299 outptr[5] = dcval;
nuclear@14 300 outptr[6] = dcval;
nuclear@14 301 outptr[7] = dcval;
nuclear@14 302
nuclear@14 303 wsptr += DCTSIZE; /* advance pointer to next row */
nuclear@14 304 continue;
nuclear@14 305 }
nuclear@14 306 #endif
nuclear@14 307
nuclear@14 308 /* Even part: reverse the even part of the forward DCT. */
nuclear@14 309 /* The rotator is sqrt(2)*c(-6). */
nuclear@14 310
nuclear@14 311 z2 = (INT32) wsptr[2];
nuclear@14 312 z3 = (INT32) wsptr[6];
nuclear@14 313
nuclear@14 314 z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
nuclear@14 315 tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
nuclear@14 316 tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
nuclear@14 317
nuclear@14 318 tmp0 = ((INT32) wsptr[0] + (INT32) wsptr[4]) << CONST_BITS;
nuclear@14 319 tmp1 = ((INT32) wsptr[0] - (INT32) wsptr[4]) << CONST_BITS;
nuclear@14 320
nuclear@14 321 tmp10 = tmp0 + tmp3;
nuclear@14 322 tmp13 = tmp0 - tmp3;
nuclear@14 323 tmp11 = tmp1 + tmp2;
nuclear@14 324 tmp12 = tmp1 - tmp2;
nuclear@14 325
nuclear@14 326 /* Odd part per figure 8; the matrix is unitary and hence its
nuclear@14 327 * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
nuclear@14 328 */
nuclear@14 329
nuclear@14 330 tmp0 = (INT32) wsptr[7];
nuclear@14 331 tmp1 = (INT32) wsptr[5];
nuclear@14 332 tmp2 = (INT32) wsptr[3];
nuclear@14 333 tmp3 = (INT32) wsptr[1];
nuclear@14 334
nuclear@14 335 z1 = tmp0 + tmp3;
nuclear@14 336 z2 = tmp1 + tmp2;
nuclear@14 337 z3 = tmp0 + tmp2;
nuclear@14 338 z4 = tmp1 + tmp3;
nuclear@14 339 z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
nuclear@14 340
nuclear@14 341 tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
nuclear@14 342 tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
nuclear@14 343 tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
nuclear@14 344 tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
nuclear@14 345 z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
nuclear@14 346 z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
nuclear@14 347 z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
nuclear@14 348 z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
nuclear@14 349
nuclear@14 350 z3 += z5;
nuclear@14 351 z4 += z5;
nuclear@14 352
nuclear@14 353 tmp0 += z1 + z3;
nuclear@14 354 tmp1 += z2 + z4;
nuclear@14 355 tmp2 += z2 + z3;
nuclear@14 356 tmp3 += z1 + z4;
nuclear@14 357
nuclear@14 358 /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
nuclear@14 359
nuclear@14 360 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp3,
nuclear@14 361 CONST_BITS+PASS1_BITS+3)
nuclear@14 362 & RANGE_MASK];
nuclear@14 363 outptr[7] = range_limit[(int) DESCALE(tmp10 - tmp3,
nuclear@14 364 CONST_BITS+PASS1_BITS+3)
nuclear@14 365 & RANGE_MASK];
nuclear@14 366 outptr[1] = range_limit[(int) DESCALE(tmp11 + tmp2,
nuclear@14 367 CONST_BITS+PASS1_BITS+3)
nuclear@14 368 & RANGE_MASK];
nuclear@14 369 outptr[6] = range_limit[(int) DESCALE(tmp11 - tmp2,
nuclear@14 370 CONST_BITS+PASS1_BITS+3)
nuclear@14 371 & RANGE_MASK];
nuclear@14 372 outptr[2] = range_limit[(int) DESCALE(tmp12 + tmp1,
nuclear@14 373 CONST_BITS+PASS1_BITS+3)
nuclear@14 374 & RANGE_MASK];
nuclear@14 375 outptr[5] = range_limit[(int) DESCALE(tmp12 - tmp1,
nuclear@14 376 CONST_BITS+PASS1_BITS+3)
nuclear@14 377 & RANGE_MASK];
nuclear@14 378 outptr[3] = range_limit[(int) DESCALE(tmp13 + tmp0,
nuclear@14 379 CONST_BITS+PASS1_BITS+3)
nuclear@14 380 & RANGE_MASK];
nuclear@14 381 outptr[4] = range_limit[(int) DESCALE(tmp13 - tmp0,
nuclear@14 382 CONST_BITS+PASS1_BITS+3)
nuclear@14 383 & RANGE_MASK];
nuclear@14 384
nuclear@14 385 wsptr += DCTSIZE; /* advance pointer to next row */
nuclear@14 386 }
nuclear@14 387 }
nuclear@14 388
nuclear@14 389 #endif /* DCT_ISLOW_SUPPORTED */