1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libjpeg/jidctint.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,2623 @@
1.4 +/*
1.5 + * jidctint.c
1.6 + *
1.7 + * Copyright (C) 1991-1998, Thomas G. Lane.
1.8 + * Modification developed 2002-2009 by Guido Vollbeding.
1.9 + * This file is part of the Independent JPEG Group's software.
1.10 + * For conditions of distribution and use, see the accompanying README file.
1.11 + *
1.12 + * This file contains a slow-but-accurate integer implementation of the
1.13 + * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine
1.14 + * must also perform dequantization of the input coefficients.
1.15 + *
1.16 + * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
1.17 + * on each row (or vice versa, but it's more convenient to emit a row at
1.18 + * a time). Direct algorithms are also available, but they are much more
1.19 + * complex and seem not to be any faster when reduced to code.
1.20 + *
1.21 + * This implementation is based on an algorithm described in
1.22 + * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
1.23 + * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
1.24 + * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
1.25 + * The primary algorithm described there uses 11 multiplies and 29 adds.
1.26 + * We use their alternate method with 12 multiplies and 32 adds.
1.27 + * The advantage of this method is that no data path contains more than one
1.28 + * multiplication; this allows a very simple and accurate implementation in
1.29 + * scaled fixed-point arithmetic, with a minimal number of shifts.
1.30 + *
1.31 + * We also provide IDCT routines with various output sample block sizes for
1.32 + * direct resolution reduction or enlargement without additional resampling:
1.33 + * NxN (N=1...16) pixels for one 8x8 input DCT block.
1.34 + *
1.35 + * For N<8 we simply take the corresponding low-frequency coefficients of
1.36 + * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block
1.37 + * to yield the downscaled outputs.
1.38 + * This can be seen as direct low-pass downsampling from the DCT domain
1.39 + * point of view rather than the usual spatial domain point of view,
1.40 + * yielding significant computational savings and results at least
1.41 + * as good as common bilinear (averaging) spatial downsampling.
1.42 + *
1.43 + * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as
1.44 + * lower frequencies and higher frequencies assumed to be zero.
1.45 + * It turns out that the computational effort is similar to the 8x8 IDCT
1.46 + * regarding the output size.
1.47 + * Furthermore, the scaling and descaling is the same for all IDCT sizes.
1.48 + *
1.49 + * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases
1.50 + * since there would be too many additional constants to pre-calculate.
1.51 + */
1.52 +
1.53 +#define JPEG_INTERNALS
1.54 +#include "jinclude.h"
1.55 +#include "jpeglib.h"
1.56 +#include "jdct.h" /* Private declarations for DCT subsystem */
1.57 +
1.58 +#ifdef DCT_ISLOW_SUPPORTED
1.59 +
1.60 +
1.61 +/*
1.62 + * This module is specialized to the case DCTSIZE = 8.
1.63 + */
1.64 +
1.65 +#if DCTSIZE != 8
1.66 + Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
1.67 +#endif
1.68 +
1.69 +
1.70 +/*
1.71 + * The poop on this scaling stuff is as follows:
1.72 + *
1.73 + * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
1.74 + * larger than the true IDCT outputs. The final outputs are therefore
1.75 + * a factor of N larger than desired; since N=8 this can be cured by
1.76 + * a simple right shift at the end of the algorithm. The advantage of
1.77 + * this arrangement is that we save two multiplications per 1-D IDCT,
1.78 + * because the y0 and y4 inputs need not be divided by sqrt(N).
1.79 + *
1.80 + * We have to do addition and subtraction of the integer inputs, which
1.81 + * is no problem, and multiplication by fractional constants, which is
1.82 + * a problem to do in integer arithmetic. We multiply all the constants
1.83 + * by CONST_SCALE and convert them to integer constants (thus retaining
1.84 + * CONST_BITS bits of precision in the constants). After doing a
1.85 + * multiplication we have to divide the product by CONST_SCALE, with proper
1.86 + * rounding, to produce the correct output. This division can be done
1.87 + * cheaply as a right shift of CONST_BITS bits. We postpone shifting
1.88 + * as long as possible so that partial sums can be added together with
1.89 + * full fractional precision.
1.90 + *
1.91 + * The outputs of the first pass are scaled up by PASS1_BITS bits so that
1.92 + * they are represented to better-than-integral precision. These outputs
1.93 + * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
1.94 + * with the recommended scaling. (To scale up 12-bit sample data further, an
1.95 + * intermediate INT32 array would be needed.)
1.96 + *
1.97 + * To avoid overflow of the 32-bit intermediate results in pass 2, we must
1.98 + * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis
1.99 + * shows that the values given below are the most effective.
1.100 + */
1.101 +
1.102 +#if BITS_IN_JSAMPLE == 8
1.103 +#define CONST_BITS 13
1.104 +#define PASS1_BITS 2
1.105 +#else
1.106 +#define CONST_BITS 13
1.107 +#define PASS1_BITS 1 /* lose a little precision to avoid overflow */
1.108 +#endif
1.109 +
1.110 +/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
1.111 + * causing a lot of useless floating-point operations at run time.
1.112 + * To get around this we use the following pre-calculated constants.
1.113 + * If you change CONST_BITS you may want to add appropriate values.
1.114 + * (With a reasonable C compiler, you can just rely on the FIX() macro...)
1.115 + */
1.116 +
1.117 +#if CONST_BITS == 13
1.118 +#define FIX_0_298631336 ((INT32) 2446) /* FIX(0.298631336) */
1.119 +#define FIX_0_390180644 ((INT32) 3196) /* FIX(0.390180644) */
1.120 +#define FIX_0_541196100 ((INT32) 4433) /* FIX(0.541196100) */
1.121 +#define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
1.122 +#define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
1.123 +#define FIX_1_175875602 ((INT32) 9633) /* FIX(1.175875602) */
1.124 +#define FIX_1_501321110 ((INT32) 12299) /* FIX(1.501321110) */
1.125 +#define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
1.126 +#define FIX_1_961570560 ((INT32) 16069) /* FIX(1.961570560) */
1.127 +#define FIX_2_053119869 ((INT32) 16819) /* FIX(2.053119869) */
1.128 +#define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
1.129 +#define FIX_3_072711026 ((INT32) 25172) /* FIX(3.072711026) */
1.130 +#else
1.131 +#define FIX_0_298631336 FIX(0.298631336)
1.132 +#define FIX_0_390180644 FIX(0.390180644)
1.133 +#define FIX_0_541196100 FIX(0.541196100)
1.134 +#define FIX_0_765366865 FIX(0.765366865)
1.135 +#define FIX_0_899976223 FIX(0.899976223)
1.136 +#define FIX_1_175875602 FIX(1.175875602)
1.137 +#define FIX_1_501321110 FIX(1.501321110)
1.138 +#define FIX_1_847759065 FIX(1.847759065)
1.139 +#define FIX_1_961570560 FIX(1.961570560)
1.140 +#define FIX_2_053119869 FIX(2.053119869)
1.141 +#define FIX_2_562915447 FIX(2.562915447)
1.142 +#define FIX_3_072711026 FIX(3.072711026)
1.143 +#endif
1.144 +
1.145 +
1.146 +/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
1.147 + * For 8-bit samples with the recommended scaling, all the variable
1.148 + * and constant values involved are no more than 16 bits wide, so a
1.149 + * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
1.150 + * For 12-bit samples, a full 32-bit multiplication will be needed.
1.151 + */
1.152 +
1.153 +#if BITS_IN_JSAMPLE == 8
1.154 +#define MULTIPLY(var,const) MULTIPLY16C16(var,const)
1.155 +#else
1.156 +#define MULTIPLY(var,const) ((var) * (const))
1.157 +#endif
1.158 +
1.159 +
1.160 +/* Dequantize a coefficient by multiplying it by the multiplier-table
1.161 + * entry; produce an int result. In this module, both inputs and result
1.162 + * are 16 bits or less, so either int or short multiply will work.
1.163 + */
1.164 +
1.165 +#define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))
1.166 +
1.167 +
1.168 +/*
1.169 + * Perform dequantization and inverse DCT on one block of coefficients.
1.170 + */
1.171 +
1.172 +GLOBAL(void)
1.173 +jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.174 + JCOEFPTR coef_block,
1.175 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.176 +{
1.177 + INT32 tmp0, tmp1, tmp2, tmp3;
1.178 + INT32 tmp10, tmp11, tmp12, tmp13;
1.179 + INT32 z1, z2, z3, z4, z5;
1.180 + JCOEFPTR inptr;
1.181 + ISLOW_MULT_TYPE * quantptr;
1.182 + int * wsptr;
1.183 + JSAMPROW outptr;
1.184 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.185 + int ctr;
1.186 + int workspace[DCTSIZE2]; /* buffers data between passes */
1.187 + SHIFT_TEMPS
1.188 +
1.189 + /* Pass 1: process columns from input, store into work array. */
1.190 + /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
1.191 + /* furthermore, we scale the results by 2**PASS1_BITS. */
1.192 +
1.193 + inptr = coef_block;
1.194 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.195 + wsptr = workspace;
1.196 + for (ctr = DCTSIZE; ctr > 0; ctr--) {
1.197 + /* Due to quantization, we will usually find that many of the input
1.198 + * coefficients are zero, especially the AC terms. We can exploit this
1.199 + * by short-circuiting the IDCT calculation for any column in which all
1.200 + * the AC terms are zero. In that case each output is equal to the
1.201 + * DC coefficient (with scale factor as needed).
1.202 + * With typical images and quantization tables, half or more of the
1.203 + * column DCT calculations can be simplified this way.
1.204 + */
1.205 +
1.206 + if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
1.207 + inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
1.208 + inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
1.209 + inptr[DCTSIZE*7] == 0) {
1.210 + /* AC terms all zero */
1.211 + int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
1.212 +
1.213 + wsptr[DCTSIZE*0] = dcval;
1.214 + wsptr[DCTSIZE*1] = dcval;
1.215 + wsptr[DCTSIZE*2] = dcval;
1.216 + wsptr[DCTSIZE*3] = dcval;
1.217 + wsptr[DCTSIZE*4] = dcval;
1.218 + wsptr[DCTSIZE*5] = dcval;
1.219 + wsptr[DCTSIZE*6] = dcval;
1.220 + wsptr[DCTSIZE*7] = dcval;
1.221 +
1.222 + inptr++; /* advance pointers to next column */
1.223 + quantptr++;
1.224 + wsptr++;
1.225 + continue;
1.226 + }
1.227 +
1.228 + /* Even part: reverse the even part of the forward DCT. */
1.229 + /* The rotator is sqrt(2)*c(-6). */
1.230 +
1.231 + z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.232 + z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.233 +
1.234 + z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
1.235 + tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
1.236 + tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
1.237 +
1.238 + z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.239 + z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.240 +
1.241 + tmp0 = (z2 + z3) << CONST_BITS;
1.242 + tmp1 = (z2 - z3) << CONST_BITS;
1.243 +
1.244 + tmp10 = tmp0 + tmp3;
1.245 + tmp13 = tmp0 - tmp3;
1.246 + tmp11 = tmp1 + tmp2;
1.247 + tmp12 = tmp1 - tmp2;
1.248 +
1.249 + /* Odd part per figure 8; the matrix is unitary and hence its
1.250 + * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
1.251 + */
1.252 +
1.253 + tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.254 + tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.255 + tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.256 + tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.257 +
1.258 + z1 = tmp0 + tmp3;
1.259 + z2 = tmp1 + tmp2;
1.260 + z3 = tmp0 + tmp2;
1.261 + z4 = tmp1 + tmp3;
1.262 + z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
1.263 +
1.264 + tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
1.265 + tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
1.266 + tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
1.267 + tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
1.268 + z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
1.269 + z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
1.270 + z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
1.271 + z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
1.272 +
1.273 + z3 += z5;
1.274 + z4 += z5;
1.275 +
1.276 + tmp0 += z1 + z3;
1.277 + tmp1 += z2 + z4;
1.278 + tmp2 += z2 + z3;
1.279 + tmp3 += z1 + z4;
1.280 +
1.281 + /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
1.282 +
1.283 + wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
1.284 + wsptr[DCTSIZE*7] = (int) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
1.285 + wsptr[DCTSIZE*1] = (int) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
1.286 + wsptr[DCTSIZE*6] = (int) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
1.287 + wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
1.288 + wsptr[DCTSIZE*5] = (int) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
1.289 + wsptr[DCTSIZE*3] = (int) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
1.290 + wsptr[DCTSIZE*4] = (int) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
1.291 +
1.292 + inptr++; /* advance pointers to next column */
1.293 + quantptr++;
1.294 + wsptr++;
1.295 + }
1.296 +
1.297 + /* Pass 2: process rows from work array, store into output array. */
1.298 + /* Note that we must descale the results by a factor of 8 == 2**3, */
1.299 + /* and also undo the PASS1_BITS scaling. */
1.300 +
1.301 + wsptr = workspace;
1.302 + for (ctr = 0; ctr < DCTSIZE; ctr++) {
1.303 + outptr = output_buf[ctr] + output_col;
1.304 + /* Rows of zeroes can be exploited in the same way as we did with columns.
1.305 + * However, the column calculation has created many nonzero AC terms, so
1.306 + * the simplification applies less often (typically 5% to 10% of the time).
1.307 + * On machines with very fast multiplication, it's possible that the
1.308 + * test takes more time than it's worth. In that case this section
1.309 + * may be commented out.
1.310 + */
1.311 +
1.312 +#ifndef NO_ZERO_ROW_TEST
1.313 + if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
1.314 + wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
1.315 + /* AC terms all zero */
1.316 + JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
1.317 + & RANGE_MASK];
1.318 +
1.319 + outptr[0] = dcval;
1.320 + outptr[1] = dcval;
1.321 + outptr[2] = dcval;
1.322 + outptr[3] = dcval;
1.323 + outptr[4] = dcval;
1.324 + outptr[5] = dcval;
1.325 + outptr[6] = dcval;
1.326 + outptr[7] = dcval;
1.327 +
1.328 + wsptr += DCTSIZE; /* advance pointer to next row */
1.329 + continue;
1.330 + }
1.331 +#endif
1.332 +
1.333 + /* Even part: reverse the even part of the forward DCT. */
1.334 + /* The rotator is sqrt(2)*c(-6). */
1.335 +
1.336 + z2 = (INT32) wsptr[2];
1.337 + z3 = (INT32) wsptr[6];
1.338 +
1.339 + z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
1.340 + tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
1.341 + tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
1.342 +
1.343 + tmp0 = ((INT32) wsptr[0] + (INT32) wsptr[4]) << CONST_BITS;
1.344 + tmp1 = ((INT32) wsptr[0] - (INT32) wsptr[4]) << CONST_BITS;
1.345 +
1.346 + tmp10 = tmp0 + tmp3;
1.347 + tmp13 = tmp0 - tmp3;
1.348 + tmp11 = tmp1 + tmp2;
1.349 + tmp12 = tmp1 - tmp2;
1.350 +
1.351 + /* Odd part per figure 8; the matrix is unitary and hence its
1.352 + * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
1.353 + */
1.354 +
1.355 + tmp0 = (INT32) wsptr[7];
1.356 + tmp1 = (INT32) wsptr[5];
1.357 + tmp2 = (INT32) wsptr[3];
1.358 + tmp3 = (INT32) wsptr[1];
1.359 +
1.360 + z1 = tmp0 + tmp3;
1.361 + z2 = tmp1 + tmp2;
1.362 + z3 = tmp0 + tmp2;
1.363 + z4 = tmp1 + tmp3;
1.364 + z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
1.365 +
1.366 + tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
1.367 + tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
1.368 + tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
1.369 + tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
1.370 + z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
1.371 + z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
1.372 + z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
1.373 + z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
1.374 +
1.375 + z3 += z5;
1.376 + z4 += z5;
1.377 +
1.378 + tmp0 += z1 + z3;
1.379 + tmp1 += z2 + z4;
1.380 + tmp2 += z2 + z3;
1.381 + tmp3 += z1 + z4;
1.382 +
1.383 + /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
1.384 +
1.385 + outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp3,
1.386 + CONST_BITS+PASS1_BITS+3)
1.387 + & RANGE_MASK];
1.388 + outptr[7] = range_limit[(int) DESCALE(tmp10 - tmp3,
1.389 + CONST_BITS+PASS1_BITS+3)
1.390 + & RANGE_MASK];
1.391 + outptr[1] = range_limit[(int) DESCALE(tmp11 + tmp2,
1.392 + CONST_BITS+PASS1_BITS+3)
1.393 + & RANGE_MASK];
1.394 + outptr[6] = range_limit[(int) DESCALE(tmp11 - tmp2,
1.395 + CONST_BITS+PASS1_BITS+3)
1.396 + & RANGE_MASK];
1.397 + outptr[2] = range_limit[(int) DESCALE(tmp12 + tmp1,
1.398 + CONST_BITS+PASS1_BITS+3)
1.399 + & RANGE_MASK];
1.400 + outptr[5] = range_limit[(int) DESCALE(tmp12 - tmp1,
1.401 + CONST_BITS+PASS1_BITS+3)
1.402 + & RANGE_MASK];
1.403 + outptr[3] = range_limit[(int) DESCALE(tmp13 + tmp0,
1.404 + CONST_BITS+PASS1_BITS+3)
1.405 + & RANGE_MASK];
1.406 + outptr[4] = range_limit[(int) DESCALE(tmp13 - tmp0,
1.407 + CONST_BITS+PASS1_BITS+3)
1.408 + & RANGE_MASK];
1.409 +
1.410 + wsptr += DCTSIZE; /* advance pointer to next row */
1.411 + }
1.412 +}
1.413 +
1.414 +#ifdef IDCT_SCALING_SUPPORTED
1.415 +
1.416 +
1.417 +/*
1.418 + * Perform dequantization and inverse DCT on one block of coefficients,
1.419 + * producing a 7x7 output block.
1.420 + *
1.421 + * Optimized algorithm with 12 multiplications in the 1-D kernel.
1.422 + * cK represents sqrt(2) * cos(K*pi/14).
1.423 + */
1.424 +
1.425 +GLOBAL(void)
1.426 +jpeg_idct_7x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.427 + JCOEFPTR coef_block,
1.428 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.429 +{
1.430 + INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
1.431 + INT32 z1, z2, z3;
1.432 + JCOEFPTR inptr;
1.433 + ISLOW_MULT_TYPE * quantptr;
1.434 + int * wsptr;
1.435 + JSAMPROW outptr;
1.436 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.437 + int ctr;
1.438 + int workspace[7*7]; /* buffers data between passes */
1.439 + SHIFT_TEMPS
1.440 +
1.441 + /* Pass 1: process columns from input, store into work array. */
1.442 +
1.443 + inptr = coef_block;
1.444 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.445 + wsptr = workspace;
1.446 + for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
1.447 + /* Even part */
1.448 +
1.449 + tmp13 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.450 + tmp13 <<= CONST_BITS;
1.451 + /* Add fudge factor here for final descale. */
1.452 + tmp13 += ONE << (CONST_BITS-PASS1_BITS-1);
1.453 +
1.454 + z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.455 + z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.456 + z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.457 +
1.458 + tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
1.459 + tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
1.460 + tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
1.461 + tmp0 = z1 + z3;
1.462 + z2 -= tmp0;
1.463 + tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
1.464 + tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
1.465 + tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
1.466 + tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
1.467 +
1.468 + /* Odd part */
1.469 +
1.470 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.471 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.472 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.473 +
1.474 + tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
1.475 + tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
1.476 + tmp0 = tmp1 - tmp2;
1.477 + tmp1 += tmp2;
1.478 + tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
1.479 + tmp1 += tmp2;
1.480 + z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
1.481 + tmp0 += z2;
1.482 + tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
1.483 +
1.484 + /* Final output stage */
1.485 +
1.486 + wsptr[7*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
1.487 + wsptr[7*6] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
1.488 + wsptr[7*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
1.489 + wsptr[7*5] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
1.490 + wsptr[7*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
1.491 + wsptr[7*4] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
1.492 + wsptr[7*3] = (int) RIGHT_SHIFT(tmp13, CONST_BITS-PASS1_BITS);
1.493 + }
1.494 +
1.495 + /* Pass 2: process 7 rows from work array, store into output array. */
1.496 +
1.497 + wsptr = workspace;
1.498 + for (ctr = 0; ctr < 7; ctr++) {
1.499 + outptr = output_buf[ctr] + output_col;
1.500 +
1.501 + /* Even part */
1.502 +
1.503 + /* Add fudge factor here for final descale. */
1.504 + tmp13 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.505 + tmp13 <<= CONST_BITS;
1.506 +
1.507 + z1 = (INT32) wsptr[2];
1.508 + z2 = (INT32) wsptr[4];
1.509 + z3 = (INT32) wsptr[6];
1.510 +
1.511 + tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */
1.512 + tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */
1.513 + tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
1.514 + tmp0 = z1 + z3;
1.515 + z2 -= tmp0;
1.516 + tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
1.517 + tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */
1.518 + tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */
1.519 + tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */
1.520 +
1.521 + /* Odd part */
1.522 +
1.523 + z1 = (INT32) wsptr[1];
1.524 + z2 = (INT32) wsptr[3];
1.525 + z3 = (INT32) wsptr[5];
1.526 +
1.527 + tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */
1.528 + tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */
1.529 + tmp0 = tmp1 - tmp2;
1.530 + tmp1 += tmp2;
1.531 + tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276)); /* -c1 */
1.532 + tmp1 += tmp2;
1.533 + z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */
1.534 + tmp0 += z2;
1.535 + tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */
1.536 +
1.537 + /* Final output stage */
1.538 +
1.539 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1.540 + CONST_BITS+PASS1_BITS+3)
1.541 + & RANGE_MASK];
1.542 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1.543 + CONST_BITS+PASS1_BITS+3)
1.544 + & RANGE_MASK];
1.545 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
1.546 + CONST_BITS+PASS1_BITS+3)
1.547 + & RANGE_MASK];
1.548 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
1.549 + CONST_BITS+PASS1_BITS+3)
1.550 + & RANGE_MASK];
1.551 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
1.552 + CONST_BITS+PASS1_BITS+3)
1.553 + & RANGE_MASK];
1.554 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
1.555 + CONST_BITS+PASS1_BITS+3)
1.556 + & RANGE_MASK];
1.557 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13,
1.558 + CONST_BITS+PASS1_BITS+3)
1.559 + & RANGE_MASK];
1.560 +
1.561 + wsptr += 7; /* advance pointer to next row */
1.562 + }
1.563 +}
1.564 +
1.565 +
1.566 +/*
1.567 + * Perform dequantization and inverse DCT on one block of coefficients,
1.568 + * producing a reduced-size 6x6 output block.
1.569 + *
1.570 + * Optimized algorithm with 3 multiplications in the 1-D kernel.
1.571 + * cK represents sqrt(2) * cos(K*pi/12).
1.572 + */
1.573 +
1.574 +GLOBAL(void)
1.575 +jpeg_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.576 + JCOEFPTR coef_block,
1.577 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.578 +{
1.579 + INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
1.580 + INT32 z1, z2, z3;
1.581 + JCOEFPTR inptr;
1.582 + ISLOW_MULT_TYPE * quantptr;
1.583 + int * wsptr;
1.584 + JSAMPROW outptr;
1.585 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.586 + int ctr;
1.587 + int workspace[6*6]; /* buffers data between passes */
1.588 + SHIFT_TEMPS
1.589 +
1.590 + /* Pass 1: process columns from input, store into work array. */
1.591 +
1.592 + inptr = coef_block;
1.593 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.594 + wsptr = workspace;
1.595 + for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
1.596 + /* Even part */
1.597 +
1.598 + tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.599 + tmp0 <<= CONST_BITS;
1.600 + /* Add fudge factor here for final descale. */
1.601 + tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
1.602 + tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.603 + tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
1.604 + tmp1 = tmp0 + tmp10;
1.605 + tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
1.606 + tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.607 + tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
1.608 + tmp10 = tmp1 + tmp0;
1.609 + tmp12 = tmp1 - tmp0;
1.610 +
1.611 + /* Odd part */
1.612 +
1.613 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.614 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.615 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.616 + tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
1.617 + tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
1.618 + tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
1.619 + tmp1 = (z1 - z2 - z3) << PASS1_BITS;
1.620 +
1.621 + /* Final output stage */
1.622 +
1.623 + wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
1.624 + wsptr[6*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
1.625 + wsptr[6*1] = (int) (tmp11 + tmp1);
1.626 + wsptr[6*4] = (int) (tmp11 - tmp1);
1.627 + wsptr[6*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
1.628 + wsptr[6*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
1.629 + }
1.630 +
1.631 + /* Pass 2: process 6 rows from work array, store into output array. */
1.632 +
1.633 + wsptr = workspace;
1.634 + for (ctr = 0; ctr < 6; ctr++) {
1.635 + outptr = output_buf[ctr] + output_col;
1.636 +
1.637 + /* Even part */
1.638 +
1.639 + /* Add fudge factor here for final descale. */
1.640 + tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.641 + tmp0 <<= CONST_BITS;
1.642 + tmp2 = (INT32) wsptr[4];
1.643 + tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */
1.644 + tmp1 = tmp0 + tmp10;
1.645 + tmp11 = tmp0 - tmp10 - tmp10;
1.646 + tmp10 = (INT32) wsptr[2];
1.647 + tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */
1.648 + tmp10 = tmp1 + tmp0;
1.649 + tmp12 = tmp1 - tmp0;
1.650 +
1.651 + /* Odd part */
1.652 +
1.653 + z1 = (INT32) wsptr[1];
1.654 + z2 = (INT32) wsptr[3];
1.655 + z3 = (INT32) wsptr[5];
1.656 + tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
1.657 + tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
1.658 + tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
1.659 + tmp1 = (z1 - z2 - z3) << CONST_BITS;
1.660 +
1.661 + /* Final output stage */
1.662 +
1.663 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1.664 + CONST_BITS+PASS1_BITS+3)
1.665 + & RANGE_MASK];
1.666 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1.667 + CONST_BITS+PASS1_BITS+3)
1.668 + & RANGE_MASK];
1.669 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
1.670 + CONST_BITS+PASS1_BITS+3)
1.671 + & RANGE_MASK];
1.672 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
1.673 + CONST_BITS+PASS1_BITS+3)
1.674 + & RANGE_MASK];
1.675 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
1.676 + CONST_BITS+PASS1_BITS+3)
1.677 + & RANGE_MASK];
1.678 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
1.679 + CONST_BITS+PASS1_BITS+3)
1.680 + & RANGE_MASK];
1.681 +
1.682 + wsptr += 6; /* advance pointer to next row */
1.683 + }
1.684 +}
1.685 +
1.686 +
1.687 +/*
1.688 + * Perform dequantization and inverse DCT on one block of coefficients,
1.689 + * producing a reduced-size 5x5 output block.
1.690 + *
1.691 + * Optimized algorithm with 5 multiplications in the 1-D kernel.
1.692 + * cK represents sqrt(2) * cos(K*pi/10).
1.693 + */
1.694 +
1.695 +GLOBAL(void)
1.696 +jpeg_idct_5x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.697 + JCOEFPTR coef_block,
1.698 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.699 +{
1.700 + INT32 tmp0, tmp1, tmp10, tmp11, tmp12;
1.701 + INT32 z1, z2, z3;
1.702 + JCOEFPTR inptr;
1.703 + ISLOW_MULT_TYPE * quantptr;
1.704 + int * wsptr;
1.705 + JSAMPROW outptr;
1.706 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.707 + int ctr;
1.708 + int workspace[5*5]; /* buffers data between passes */
1.709 + SHIFT_TEMPS
1.710 +
1.711 + /* Pass 1: process columns from input, store into work array. */
1.712 +
1.713 + inptr = coef_block;
1.714 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.715 + wsptr = workspace;
1.716 + for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
1.717 + /* Even part */
1.718 +
1.719 + tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.720 + tmp12 <<= CONST_BITS;
1.721 + /* Add fudge factor here for final descale. */
1.722 + tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
1.723 + tmp0 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.724 + tmp1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.725 + z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
1.726 + z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
1.727 + z3 = tmp12 + z2;
1.728 + tmp10 = z3 + z1;
1.729 + tmp11 = z3 - z1;
1.730 + tmp12 -= z2 << 2;
1.731 +
1.732 + /* Odd part */
1.733 +
1.734 + z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.735 + z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.736 +
1.737 + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
1.738 + tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
1.739 + tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
1.740 +
1.741 + /* Final output stage */
1.742 +
1.743 + wsptr[5*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
1.744 + wsptr[5*4] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
1.745 + wsptr[5*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
1.746 + wsptr[5*3] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
1.747 + wsptr[5*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
1.748 + }
1.749 +
1.750 + /* Pass 2: process 5 rows from work array, store into output array. */
1.751 +
1.752 + wsptr = workspace;
1.753 + for (ctr = 0; ctr < 5; ctr++) {
1.754 + outptr = output_buf[ctr] + output_col;
1.755 +
1.756 + /* Even part */
1.757 +
1.758 + /* Add fudge factor here for final descale. */
1.759 + tmp12 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.760 + tmp12 <<= CONST_BITS;
1.761 + tmp0 = (INT32) wsptr[2];
1.762 + tmp1 = (INT32) wsptr[4];
1.763 + z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
1.764 + z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
1.765 + z3 = tmp12 + z2;
1.766 + tmp10 = z3 + z1;
1.767 + tmp11 = z3 - z1;
1.768 + tmp12 -= z2 << 2;
1.769 +
1.770 + /* Odd part */
1.771 +
1.772 + z2 = (INT32) wsptr[1];
1.773 + z3 = (INT32) wsptr[3];
1.774 +
1.775 + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */
1.776 + tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */
1.777 + tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */
1.778 +
1.779 + /* Final output stage */
1.780 +
1.781 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1.782 + CONST_BITS+PASS1_BITS+3)
1.783 + & RANGE_MASK];
1.784 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1.785 + CONST_BITS+PASS1_BITS+3)
1.786 + & RANGE_MASK];
1.787 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
1.788 + CONST_BITS+PASS1_BITS+3)
1.789 + & RANGE_MASK];
1.790 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
1.791 + CONST_BITS+PASS1_BITS+3)
1.792 + & RANGE_MASK];
1.793 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
1.794 + CONST_BITS+PASS1_BITS+3)
1.795 + & RANGE_MASK];
1.796 +
1.797 + wsptr += 5; /* advance pointer to next row */
1.798 + }
1.799 +}
1.800 +
1.801 +
1.802 +/*
1.803 + * Perform dequantization and inverse DCT on one block of coefficients,
1.804 + * producing a reduced-size 3x3 output block.
1.805 + *
1.806 + * Optimized algorithm with 2 multiplications in the 1-D kernel.
1.807 + * cK represents sqrt(2) * cos(K*pi/6).
1.808 + */
1.809 +
1.810 +GLOBAL(void)
1.811 +jpeg_idct_3x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.812 + JCOEFPTR coef_block,
1.813 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.814 +{
1.815 + INT32 tmp0, tmp2, tmp10, tmp12;
1.816 + JCOEFPTR inptr;
1.817 + ISLOW_MULT_TYPE * quantptr;
1.818 + int * wsptr;
1.819 + JSAMPROW outptr;
1.820 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.821 + int ctr;
1.822 + int workspace[3*3]; /* buffers data between passes */
1.823 + SHIFT_TEMPS
1.824 +
1.825 + /* Pass 1: process columns from input, store into work array. */
1.826 +
1.827 + inptr = coef_block;
1.828 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.829 + wsptr = workspace;
1.830 + for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
1.831 + /* Even part */
1.832 +
1.833 + tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.834 + tmp0 <<= CONST_BITS;
1.835 + /* Add fudge factor here for final descale. */
1.836 + tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
1.837 + tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.838 + tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
1.839 + tmp10 = tmp0 + tmp12;
1.840 + tmp2 = tmp0 - tmp12 - tmp12;
1.841 +
1.842 + /* Odd part */
1.843 +
1.844 + tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.845 + tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
1.846 +
1.847 + /* Final output stage */
1.848 +
1.849 + wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
1.850 + wsptr[3*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
1.851 + wsptr[3*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
1.852 + }
1.853 +
1.854 + /* Pass 2: process 3 rows from work array, store into output array. */
1.855 +
1.856 + wsptr = workspace;
1.857 + for (ctr = 0; ctr < 3; ctr++) {
1.858 + outptr = output_buf[ctr] + output_col;
1.859 +
1.860 + /* Even part */
1.861 +
1.862 + /* Add fudge factor here for final descale. */
1.863 + tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.864 + tmp0 <<= CONST_BITS;
1.865 + tmp2 = (INT32) wsptr[2];
1.866 + tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
1.867 + tmp10 = tmp0 + tmp12;
1.868 + tmp2 = tmp0 - tmp12 - tmp12;
1.869 +
1.870 + /* Odd part */
1.871 +
1.872 + tmp12 = (INT32) wsptr[1];
1.873 + tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */
1.874 +
1.875 + /* Final output stage */
1.876 +
1.877 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1.878 + CONST_BITS+PASS1_BITS+3)
1.879 + & RANGE_MASK];
1.880 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1.881 + CONST_BITS+PASS1_BITS+3)
1.882 + & RANGE_MASK];
1.883 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
1.884 + CONST_BITS+PASS1_BITS+3)
1.885 + & RANGE_MASK];
1.886 +
1.887 + wsptr += 3; /* advance pointer to next row */
1.888 + }
1.889 +}
1.890 +
1.891 +
1.892 +/*
1.893 + * Perform dequantization and inverse DCT on one block of coefficients,
1.894 + * producing a 9x9 output block.
1.895 + *
1.896 + * Optimized algorithm with 10 multiplications in the 1-D kernel.
1.897 + * cK represents sqrt(2) * cos(K*pi/18).
1.898 + */
1.899 +
1.900 +GLOBAL(void)
1.901 +jpeg_idct_9x9 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.902 + JCOEFPTR coef_block,
1.903 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.904 +{
1.905 + INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14;
1.906 + INT32 z1, z2, z3, z4;
1.907 + JCOEFPTR inptr;
1.908 + ISLOW_MULT_TYPE * quantptr;
1.909 + int * wsptr;
1.910 + JSAMPROW outptr;
1.911 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.912 + int ctr;
1.913 + int workspace[8*9]; /* buffers data between passes */
1.914 + SHIFT_TEMPS
1.915 +
1.916 + /* Pass 1: process columns from input, store into work array. */
1.917 +
1.918 + inptr = coef_block;
1.919 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.920 + wsptr = workspace;
1.921 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.922 + /* Even part */
1.923 +
1.924 + tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.925 + tmp0 <<= CONST_BITS;
1.926 + /* Add fudge factor here for final descale. */
1.927 + tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
1.928 +
1.929 + z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.930 + z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.931 + z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.932 +
1.933 + tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
1.934 + tmp1 = tmp0 + tmp3;
1.935 + tmp2 = tmp0 - tmp3 - tmp3;
1.936 +
1.937 + tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1.938 + tmp11 = tmp2 + tmp0;
1.939 + tmp14 = tmp2 - tmp0 - tmp0;
1.940 +
1.941 + tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1.942 + tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1.943 + tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1.944 +
1.945 + tmp10 = tmp1 + tmp0 - tmp3;
1.946 + tmp12 = tmp1 - tmp0 + tmp2;
1.947 + tmp13 = tmp1 - tmp2 + tmp3;
1.948 +
1.949 + /* Odd part */
1.950 +
1.951 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.952 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.953 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.954 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.955 +
1.956 + z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
1.957 +
1.958 + tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1.959 + tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1.960 + tmp0 = tmp2 + tmp3 - z2;
1.961 + tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1.962 + tmp2 += z2 - tmp1;
1.963 + tmp3 += z2 + tmp1;
1.964 + tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1.965 +
1.966 + /* Final output stage */
1.967 +
1.968 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
1.969 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
1.970 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
1.971 + wsptr[8*7] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
1.972 + wsptr[8*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
1.973 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
1.974 + wsptr[8*3] = (int) RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS-PASS1_BITS);
1.975 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS-PASS1_BITS);
1.976 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp14, CONST_BITS-PASS1_BITS);
1.977 + }
1.978 +
1.979 + /* Pass 2: process 9 rows from work array, store into output array. */
1.980 +
1.981 + wsptr = workspace;
1.982 + for (ctr = 0; ctr < 9; ctr++) {
1.983 + outptr = output_buf[ctr] + output_col;
1.984 +
1.985 + /* Even part */
1.986 +
1.987 + /* Add fudge factor here for final descale. */
1.988 + tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.989 + tmp0 <<= CONST_BITS;
1.990 +
1.991 + z1 = (INT32) wsptr[2];
1.992 + z2 = (INT32) wsptr[4];
1.993 + z3 = (INT32) wsptr[6];
1.994 +
1.995 + tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */
1.996 + tmp1 = tmp0 + tmp3;
1.997 + tmp2 = tmp0 - tmp3 - tmp3;
1.998 +
1.999 + tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
1.1000 + tmp11 = tmp2 + tmp0;
1.1001 + tmp14 = tmp2 - tmp0 - tmp0;
1.1002 +
1.1003 + tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
1.1004 + tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */
1.1005 + tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */
1.1006 +
1.1007 + tmp10 = tmp1 + tmp0 - tmp3;
1.1008 + tmp12 = tmp1 - tmp0 + tmp2;
1.1009 + tmp13 = tmp1 - tmp2 + tmp3;
1.1010 +
1.1011 + /* Odd part */
1.1012 +
1.1013 + z1 = (INT32) wsptr[1];
1.1014 + z2 = (INT32) wsptr[3];
1.1015 + z3 = (INT32) wsptr[5];
1.1016 + z4 = (INT32) wsptr[7];
1.1017 +
1.1018 + z2 = MULTIPLY(z2, - FIX(1.224744871)); /* -c3 */
1.1019 +
1.1020 + tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */
1.1021 + tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */
1.1022 + tmp0 = tmp2 + tmp3 - z2;
1.1023 + tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */
1.1024 + tmp2 += z2 - tmp1;
1.1025 + tmp3 += z2 + tmp1;
1.1026 + tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */
1.1027 +
1.1028 + /* Final output stage */
1.1029 +
1.1030 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
1.1031 + CONST_BITS+PASS1_BITS+3)
1.1032 + & RANGE_MASK];
1.1033 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
1.1034 + CONST_BITS+PASS1_BITS+3)
1.1035 + & RANGE_MASK];
1.1036 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
1.1037 + CONST_BITS+PASS1_BITS+3)
1.1038 + & RANGE_MASK];
1.1039 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
1.1040 + CONST_BITS+PASS1_BITS+3)
1.1041 + & RANGE_MASK];
1.1042 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
1.1043 + CONST_BITS+PASS1_BITS+3)
1.1044 + & RANGE_MASK];
1.1045 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
1.1046 + CONST_BITS+PASS1_BITS+3)
1.1047 + & RANGE_MASK];
1.1048 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp3,
1.1049 + CONST_BITS+PASS1_BITS+3)
1.1050 + & RANGE_MASK];
1.1051 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp3,
1.1052 + CONST_BITS+PASS1_BITS+3)
1.1053 + & RANGE_MASK];
1.1054 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp14,
1.1055 + CONST_BITS+PASS1_BITS+3)
1.1056 + & RANGE_MASK];
1.1057 +
1.1058 + wsptr += 8; /* advance pointer to next row */
1.1059 + }
1.1060 +}
1.1061 +
1.1062 +
1.1063 +/*
1.1064 + * Perform dequantization and inverse DCT on one block of coefficients,
1.1065 + * producing a 10x10 output block.
1.1066 + *
1.1067 + * Optimized algorithm with 12 multiplications in the 1-D kernel.
1.1068 + * cK represents sqrt(2) * cos(K*pi/20).
1.1069 + */
1.1070 +
1.1071 +GLOBAL(void)
1.1072 +jpeg_idct_10x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.1073 + JCOEFPTR coef_block,
1.1074 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.1075 +{
1.1076 + INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1.1077 + INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
1.1078 + INT32 z1, z2, z3, z4, z5;
1.1079 + JCOEFPTR inptr;
1.1080 + ISLOW_MULT_TYPE * quantptr;
1.1081 + int * wsptr;
1.1082 + JSAMPROW outptr;
1.1083 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.1084 + int ctr;
1.1085 + int workspace[8*10]; /* buffers data between passes */
1.1086 + SHIFT_TEMPS
1.1087 +
1.1088 + /* Pass 1: process columns from input, store into work array. */
1.1089 +
1.1090 + inptr = coef_block;
1.1091 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.1092 + wsptr = workspace;
1.1093 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.1094 + /* Even part */
1.1095 +
1.1096 + z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.1097 + z3 <<= CONST_BITS;
1.1098 + /* Add fudge factor here for final descale. */
1.1099 + z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1.1100 + z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.1101 + z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1.1102 + z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1.1103 + tmp10 = z3 + z1;
1.1104 + tmp11 = z3 - z2;
1.1105 +
1.1106 + tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1), /* c0 = (c4-c8)*2 */
1.1107 + CONST_BITS-PASS1_BITS);
1.1108 +
1.1109 + z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.1110 + z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.1111 +
1.1112 + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1.1113 + tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1.1114 + tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1.1115 +
1.1116 + tmp20 = tmp10 + tmp12;
1.1117 + tmp24 = tmp10 - tmp12;
1.1118 + tmp21 = tmp11 + tmp13;
1.1119 + tmp23 = tmp11 - tmp13;
1.1120 +
1.1121 + /* Odd part */
1.1122 +
1.1123 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.1124 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.1125 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.1126 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.1127 +
1.1128 + tmp11 = z2 + z4;
1.1129 + tmp13 = z2 - z4;
1.1130 +
1.1131 + tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1.1132 + z5 = z3 << CONST_BITS;
1.1133 +
1.1134 + z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1.1135 + z4 = z5 + tmp12;
1.1136 +
1.1137 + tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1.1138 + tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1.1139 +
1.1140 + z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1.1141 + z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));
1.1142 +
1.1143 + tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;
1.1144 +
1.1145 + tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1.1146 + tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1.1147 +
1.1148 + /* Final output stage */
1.1149 +
1.1150 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1.1151 + wsptr[8*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1.1152 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1.1153 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1.1154 + wsptr[8*2] = (int) (tmp22 + tmp12);
1.1155 + wsptr[8*7] = (int) (tmp22 - tmp12);
1.1156 + wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1.1157 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1.1158 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1.1159 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1.1160 + }
1.1161 +
1.1162 + /* Pass 2: process 10 rows from work array, store into output array. */
1.1163 +
1.1164 + wsptr = workspace;
1.1165 + for (ctr = 0; ctr < 10; ctr++) {
1.1166 + outptr = output_buf[ctr] + output_col;
1.1167 +
1.1168 + /* Even part */
1.1169 +
1.1170 + /* Add fudge factor here for final descale. */
1.1171 + z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.1172 + z3 <<= CONST_BITS;
1.1173 + z4 = (INT32) wsptr[4];
1.1174 + z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */
1.1175 + z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */
1.1176 + tmp10 = z3 + z1;
1.1177 + tmp11 = z3 - z2;
1.1178 +
1.1179 + tmp22 = z3 - ((z1 - z2) << 1); /* c0 = (c4-c8)*2 */
1.1180 +
1.1181 + z2 = (INT32) wsptr[2];
1.1182 + z3 = (INT32) wsptr[6];
1.1183 +
1.1184 + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */
1.1185 + tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
1.1186 + tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */
1.1187 +
1.1188 + tmp20 = tmp10 + tmp12;
1.1189 + tmp24 = tmp10 - tmp12;
1.1190 + tmp21 = tmp11 + tmp13;
1.1191 + tmp23 = tmp11 - tmp13;
1.1192 +
1.1193 + /* Odd part */
1.1194 +
1.1195 + z1 = (INT32) wsptr[1];
1.1196 + z2 = (INT32) wsptr[3];
1.1197 + z3 = (INT32) wsptr[5];
1.1198 + z3 <<= CONST_BITS;
1.1199 + z4 = (INT32) wsptr[7];
1.1200 +
1.1201 + tmp11 = z2 + z4;
1.1202 + tmp13 = z2 - z4;
1.1203 +
1.1204 + tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */
1.1205 +
1.1206 + z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */
1.1207 + z4 = z3 + tmp12;
1.1208 +
1.1209 + tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
1.1210 + tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */
1.1211 +
1.1212 + z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */
1.1213 + z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));
1.1214 +
1.1215 + tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;
1.1216 +
1.1217 + tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
1.1218 + tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */
1.1219 +
1.1220 + /* Final output stage */
1.1221 +
1.1222 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1.1223 + CONST_BITS+PASS1_BITS+3)
1.1224 + & RANGE_MASK];
1.1225 + outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1.1226 + CONST_BITS+PASS1_BITS+3)
1.1227 + & RANGE_MASK];
1.1228 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1.1229 + CONST_BITS+PASS1_BITS+3)
1.1230 + & RANGE_MASK];
1.1231 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1.1232 + CONST_BITS+PASS1_BITS+3)
1.1233 + & RANGE_MASK];
1.1234 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1.1235 + CONST_BITS+PASS1_BITS+3)
1.1236 + & RANGE_MASK];
1.1237 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1.1238 + CONST_BITS+PASS1_BITS+3)
1.1239 + & RANGE_MASK];
1.1240 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1.1241 + CONST_BITS+PASS1_BITS+3)
1.1242 + & RANGE_MASK];
1.1243 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1.1244 + CONST_BITS+PASS1_BITS+3)
1.1245 + & RANGE_MASK];
1.1246 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1.1247 + CONST_BITS+PASS1_BITS+3)
1.1248 + & RANGE_MASK];
1.1249 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1.1250 + CONST_BITS+PASS1_BITS+3)
1.1251 + & RANGE_MASK];
1.1252 +
1.1253 + wsptr += 8; /* advance pointer to next row */
1.1254 + }
1.1255 +}
1.1256 +
1.1257 +
1.1258 +/*
1.1259 + * Perform dequantization and inverse DCT on one block of coefficients,
1.1260 + * producing a 11x11 output block.
1.1261 + *
1.1262 + * Optimized algorithm with 24 multiplications in the 1-D kernel.
1.1263 + * cK represents sqrt(2) * cos(K*pi/22).
1.1264 + */
1.1265 +
1.1266 +GLOBAL(void)
1.1267 +jpeg_idct_11x11 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.1268 + JCOEFPTR coef_block,
1.1269 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.1270 +{
1.1271 + INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
1.1272 + INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1.1273 + INT32 z1, z2, z3, z4;
1.1274 + JCOEFPTR inptr;
1.1275 + ISLOW_MULT_TYPE * quantptr;
1.1276 + int * wsptr;
1.1277 + JSAMPROW outptr;
1.1278 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.1279 + int ctr;
1.1280 + int workspace[8*11]; /* buffers data between passes */
1.1281 + SHIFT_TEMPS
1.1282 +
1.1283 + /* Pass 1: process columns from input, store into work array. */
1.1284 +
1.1285 + inptr = coef_block;
1.1286 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.1287 + wsptr = workspace;
1.1288 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.1289 + /* Even part */
1.1290 +
1.1291 + tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.1292 + tmp10 <<= CONST_BITS;
1.1293 + /* Add fudge factor here for final descale. */
1.1294 + tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
1.1295 +
1.1296 + z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.1297 + z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.1298 + z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.1299 +
1.1300 + tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1.1301 + tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1.1302 + z4 = z1 + z3;
1.1303 + tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1.1304 + z4 -= z2;
1.1305 + tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1.1306 + tmp21 = tmp20 + tmp23 + tmp25 -
1.1307 + MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1.1308 + tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1.1309 + tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1.1310 + tmp24 += tmp25;
1.1311 + tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1.1312 + tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1.1313 + MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1.1314 + tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1.1315 +
1.1316 + /* Odd part */
1.1317 +
1.1318 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.1319 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.1320 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.1321 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.1322 +
1.1323 + tmp11 = z1 + z2;
1.1324 + tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1.1325 + tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1.1326 + tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1.1327 + tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1.1328 + tmp10 = tmp11 + tmp12 + tmp13 -
1.1329 + MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1.1330 + z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1.1331 + tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1.1332 + tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1.1333 + z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1.1334 + tmp11 += z1;
1.1335 + tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1.1336 + tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1.1337 + MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1.1338 + MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1.1339 +
1.1340 + /* Final output stage */
1.1341 +
1.1342 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1.1343 + wsptr[8*10] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1.1344 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1.1345 + wsptr[8*9] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1.1346 + wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1.1347 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1.1348 + wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1.1349 + wsptr[8*7] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1.1350 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1.1351 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1.1352 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp25, CONST_BITS-PASS1_BITS);
1.1353 + }
1.1354 +
1.1355 + /* Pass 2: process 11 rows from work array, store into output array. */
1.1356 +
1.1357 + wsptr = workspace;
1.1358 + for (ctr = 0; ctr < 11; ctr++) {
1.1359 + outptr = output_buf[ctr] + output_col;
1.1360 +
1.1361 + /* Even part */
1.1362 +
1.1363 + /* Add fudge factor here for final descale. */
1.1364 + tmp10 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.1365 + tmp10 <<= CONST_BITS;
1.1366 +
1.1367 + z1 = (INT32) wsptr[2];
1.1368 + z2 = (INT32) wsptr[4];
1.1369 + z3 = (INT32) wsptr[6];
1.1370 +
1.1371 + tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */
1.1372 + tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */
1.1373 + z4 = z1 + z3;
1.1374 + tmp24 = MULTIPLY(z4, - FIX(1.155664402)); /* -(c2-c10) */
1.1375 + z4 -= z2;
1.1376 + tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */
1.1377 + tmp21 = tmp20 + tmp23 + tmp25 -
1.1378 + MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */
1.1379 + tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
1.1380 + tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
1.1381 + tmp24 += tmp25;
1.1382 + tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */
1.1383 + tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */
1.1384 + MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */
1.1385 + tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */
1.1386 +
1.1387 + /* Odd part */
1.1388 +
1.1389 + z1 = (INT32) wsptr[1];
1.1390 + z2 = (INT32) wsptr[3];
1.1391 + z3 = (INT32) wsptr[5];
1.1392 + z4 = (INT32) wsptr[7];
1.1393 +
1.1394 + tmp11 = z1 + z2;
1.1395 + tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
1.1396 + tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */
1.1397 + tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */
1.1398 + tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
1.1399 + tmp10 = tmp11 + tmp12 + tmp13 -
1.1400 + MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */
1.1401 + z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
1.1402 + tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */
1.1403 + tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */
1.1404 + z1 = MULTIPLY(z2 + z4, - FIX(1.798248910)); /* -(c1+c9) */
1.1405 + tmp11 += z1;
1.1406 + tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */
1.1407 + tmp14 += MULTIPLY(z2, - FIX(1.467221301)) + /* -(c5+c9) */
1.1408 + MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */
1.1409 + MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */
1.1410 +
1.1411 + /* Final output stage */
1.1412 +
1.1413 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1.1414 + CONST_BITS+PASS1_BITS+3)
1.1415 + & RANGE_MASK];
1.1416 + outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1.1417 + CONST_BITS+PASS1_BITS+3)
1.1418 + & RANGE_MASK];
1.1419 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1.1420 + CONST_BITS+PASS1_BITS+3)
1.1421 + & RANGE_MASK];
1.1422 + outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1.1423 + CONST_BITS+PASS1_BITS+3)
1.1424 + & RANGE_MASK];
1.1425 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1.1426 + CONST_BITS+PASS1_BITS+3)
1.1427 + & RANGE_MASK];
1.1428 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1.1429 + CONST_BITS+PASS1_BITS+3)
1.1430 + & RANGE_MASK];
1.1431 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1.1432 + CONST_BITS+PASS1_BITS+3)
1.1433 + & RANGE_MASK];
1.1434 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1.1435 + CONST_BITS+PASS1_BITS+3)
1.1436 + & RANGE_MASK];
1.1437 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1.1438 + CONST_BITS+PASS1_BITS+3)
1.1439 + & RANGE_MASK];
1.1440 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1.1441 + CONST_BITS+PASS1_BITS+3)
1.1442 + & RANGE_MASK];
1.1443 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25,
1.1444 + CONST_BITS+PASS1_BITS+3)
1.1445 + & RANGE_MASK];
1.1446 +
1.1447 + wsptr += 8; /* advance pointer to next row */
1.1448 + }
1.1449 +}
1.1450 +
1.1451 +
1.1452 +/*
1.1453 + * Perform dequantization and inverse DCT on one block of coefficients,
1.1454 + * producing a 12x12 output block.
1.1455 + *
1.1456 + * Optimized algorithm with 15 multiplications in the 1-D kernel.
1.1457 + * cK represents sqrt(2) * cos(K*pi/24).
1.1458 + */
1.1459 +
1.1460 +GLOBAL(void)
1.1461 +jpeg_idct_12x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.1462 + JCOEFPTR coef_block,
1.1463 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.1464 +{
1.1465 + INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1.1466 + INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
1.1467 + INT32 z1, z2, z3, z4;
1.1468 + JCOEFPTR inptr;
1.1469 + ISLOW_MULT_TYPE * quantptr;
1.1470 + int * wsptr;
1.1471 + JSAMPROW outptr;
1.1472 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.1473 + int ctr;
1.1474 + int workspace[8*12]; /* buffers data between passes */
1.1475 + SHIFT_TEMPS
1.1476 +
1.1477 + /* Pass 1: process columns from input, store into work array. */
1.1478 +
1.1479 + inptr = coef_block;
1.1480 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.1481 + wsptr = workspace;
1.1482 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.1483 + /* Even part */
1.1484 +
1.1485 + z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.1486 + z3 <<= CONST_BITS;
1.1487 + /* Add fudge factor here for final descale. */
1.1488 + z3 += ONE << (CONST_BITS-PASS1_BITS-1);
1.1489 +
1.1490 + z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.1491 + z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1.1492 +
1.1493 + tmp10 = z3 + z4;
1.1494 + tmp11 = z3 - z4;
1.1495 +
1.1496 + z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.1497 + z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1.1498 + z1 <<= CONST_BITS;
1.1499 + z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.1500 + z2 <<= CONST_BITS;
1.1501 +
1.1502 + tmp12 = z1 - z2;
1.1503 +
1.1504 + tmp21 = z3 + tmp12;
1.1505 + tmp24 = z3 - tmp12;
1.1506 +
1.1507 + tmp12 = z4 + z2;
1.1508 +
1.1509 + tmp20 = tmp10 + tmp12;
1.1510 + tmp25 = tmp10 - tmp12;
1.1511 +
1.1512 + tmp12 = z4 - z1 - z2;
1.1513 +
1.1514 + tmp22 = tmp11 + tmp12;
1.1515 + tmp23 = tmp11 - tmp12;
1.1516 +
1.1517 + /* Odd part */
1.1518 +
1.1519 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.1520 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.1521 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.1522 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.1523 +
1.1524 + tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1.1525 + tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1.1526 +
1.1527 + tmp10 = z1 + z3;
1.1528 + tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1.1529 + tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1.1530 + tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1.1531 + tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1.1532 + tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1.1533 + tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1.1534 + tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1.1535 + MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1.1536 +
1.1537 + z1 -= z4;
1.1538 + z2 -= z3;
1.1539 + z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1.1540 + tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1.1541 + tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1.1542 +
1.1543 + /* Final output stage */
1.1544 +
1.1545 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1.1546 + wsptr[8*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1.1547 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1.1548 + wsptr[8*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1.1549 + wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1.1550 + wsptr[8*9] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1.1551 + wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1.1552 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1.1553 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1.1554 + wsptr[8*7] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1.1555 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1.1556 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1.1557 + }
1.1558 +
1.1559 + /* Pass 2: process 12 rows from work array, store into output array. */
1.1560 +
1.1561 + wsptr = workspace;
1.1562 + for (ctr = 0; ctr < 12; ctr++) {
1.1563 + outptr = output_buf[ctr] + output_col;
1.1564 +
1.1565 + /* Even part */
1.1566 +
1.1567 + /* Add fudge factor here for final descale. */
1.1568 + z3 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.1569 + z3 <<= CONST_BITS;
1.1570 +
1.1571 + z4 = (INT32) wsptr[4];
1.1572 + z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */
1.1573 +
1.1574 + tmp10 = z3 + z4;
1.1575 + tmp11 = z3 - z4;
1.1576 +
1.1577 + z1 = (INT32) wsptr[2];
1.1578 + z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
1.1579 + z1 <<= CONST_BITS;
1.1580 + z2 = (INT32) wsptr[6];
1.1581 + z2 <<= CONST_BITS;
1.1582 +
1.1583 + tmp12 = z1 - z2;
1.1584 +
1.1585 + tmp21 = z3 + tmp12;
1.1586 + tmp24 = z3 - tmp12;
1.1587 +
1.1588 + tmp12 = z4 + z2;
1.1589 +
1.1590 + tmp20 = tmp10 + tmp12;
1.1591 + tmp25 = tmp10 - tmp12;
1.1592 +
1.1593 + tmp12 = z4 - z1 - z2;
1.1594 +
1.1595 + tmp22 = tmp11 + tmp12;
1.1596 + tmp23 = tmp11 - tmp12;
1.1597 +
1.1598 + /* Odd part */
1.1599 +
1.1600 + z1 = (INT32) wsptr[1];
1.1601 + z2 = (INT32) wsptr[3];
1.1602 + z3 = (INT32) wsptr[5];
1.1603 + z4 = (INT32) wsptr[7];
1.1604 +
1.1605 + tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */
1.1606 + tmp14 = MULTIPLY(z2, - FIX_0_541196100); /* -c9 */
1.1607 +
1.1608 + tmp10 = z1 + z3;
1.1609 + tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */
1.1610 + tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */
1.1611 + tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */
1.1612 + tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580)); /* -(c7+c11) */
1.1613 + tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
1.1614 + tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
1.1615 + tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */
1.1616 + MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */
1.1617 +
1.1618 + z1 -= z4;
1.1619 + z2 -= z3;
1.1620 + z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */
1.1621 + tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */
1.1622 + tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */
1.1623 +
1.1624 + /* Final output stage */
1.1625 +
1.1626 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1.1627 + CONST_BITS+PASS1_BITS+3)
1.1628 + & RANGE_MASK];
1.1629 + outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1.1630 + CONST_BITS+PASS1_BITS+3)
1.1631 + & RANGE_MASK];
1.1632 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1.1633 + CONST_BITS+PASS1_BITS+3)
1.1634 + & RANGE_MASK];
1.1635 + outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1.1636 + CONST_BITS+PASS1_BITS+3)
1.1637 + & RANGE_MASK];
1.1638 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1.1639 + CONST_BITS+PASS1_BITS+3)
1.1640 + & RANGE_MASK];
1.1641 + outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1.1642 + CONST_BITS+PASS1_BITS+3)
1.1643 + & RANGE_MASK];
1.1644 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1.1645 + CONST_BITS+PASS1_BITS+3)
1.1646 + & RANGE_MASK];
1.1647 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1.1648 + CONST_BITS+PASS1_BITS+3)
1.1649 + & RANGE_MASK];
1.1650 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1.1651 + CONST_BITS+PASS1_BITS+3)
1.1652 + & RANGE_MASK];
1.1653 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1.1654 + CONST_BITS+PASS1_BITS+3)
1.1655 + & RANGE_MASK];
1.1656 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1.1657 + CONST_BITS+PASS1_BITS+3)
1.1658 + & RANGE_MASK];
1.1659 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1.1660 + CONST_BITS+PASS1_BITS+3)
1.1661 + & RANGE_MASK];
1.1662 +
1.1663 + wsptr += 8; /* advance pointer to next row */
1.1664 + }
1.1665 +}
1.1666 +
1.1667 +
1.1668 +/*
1.1669 + * Perform dequantization and inverse DCT on one block of coefficients,
1.1670 + * producing a 13x13 output block.
1.1671 + *
1.1672 + * Optimized algorithm with 29 multiplications in the 1-D kernel.
1.1673 + * cK represents sqrt(2) * cos(K*pi/26).
1.1674 + */
1.1675 +
1.1676 +GLOBAL(void)
1.1677 +jpeg_idct_13x13 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.1678 + JCOEFPTR coef_block,
1.1679 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.1680 +{
1.1681 + INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
1.1682 + INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1.1683 + INT32 z1, z2, z3, z4;
1.1684 + JCOEFPTR inptr;
1.1685 + ISLOW_MULT_TYPE * quantptr;
1.1686 + int * wsptr;
1.1687 + JSAMPROW outptr;
1.1688 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.1689 + int ctr;
1.1690 + int workspace[8*13]; /* buffers data between passes */
1.1691 + SHIFT_TEMPS
1.1692 +
1.1693 + /* Pass 1: process columns from input, store into work array. */
1.1694 +
1.1695 + inptr = coef_block;
1.1696 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.1697 + wsptr = workspace;
1.1698 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.1699 + /* Even part */
1.1700 +
1.1701 + z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.1702 + z1 <<= CONST_BITS;
1.1703 + /* Add fudge factor here for final descale. */
1.1704 + z1 += ONE << (CONST_BITS-PASS1_BITS-1);
1.1705 +
1.1706 + z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.1707 + z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.1708 + z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.1709 +
1.1710 + tmp10 = z3 + z4;
1.1711 + tmp11 = z3 - z4;
1.1712 +
1.1713 + tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1.1714 + tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1.1715 +
1.1716 + tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1.1717 + tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1.1718 +
1.1719 + tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1.1720 + tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1.1721 +
1.1722 + tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1.1723 + tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1.1724 +
1.1725 + tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1.1726 + tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1.1727 +
1.1728 + tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1.1729 + tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1.1730 +
1.1731 + tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1.1732 +
1.1733 + /* Odd part */
1.1734 +
1.1735 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.1736 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.1737 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.1738 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.1739 +
1.1740 + tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1.1741 + tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1.1742 + tmp15 = z1 + z4;
1.1743 + tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1.1744 + tmp10 = tmp11 + tmp12 + tmp13 -
1.1745 + MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1.1746 + tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
1.1747 + tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1.1748 + tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1.1749 + tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
1.1750 + tmp11 += tmp14;
1.1751 + tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1.1752 + tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
1.1753 + tmp12 += tmp14;
1.1754 + tmp13 += tmp14;
1.1755 + tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1.1756 + tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1.1757 + MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1.1758 + z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1.1759 + tmp14 += z1;
1.1760 + tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1.1761 + MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1.1762 +
1.1763 + /* Final output stage */
1.1764 +
1.1765 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1.1766 + wsptr[8*12] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1.1767 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1.1768 + wsptr[8*11] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1.1769 + wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1.1770 + wsptr[8*10] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1.1771 + wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1.1772 + wsptr[8*9] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1.1773 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1.1774 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1.1775 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1.1776 + wsptr[8*7] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1.1777 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp26, CONST_BITS-PASS1_BITS);
1.1778 + }
1.1779 +
1.1780 + /* Pass 2: process 13 rows from work array, store into output array. */
1.1781 +
1.1782 + wsptr = workspace;
1.1783 + for (ctr = 0; ctr < 13; ctr++) {
1.1784 + outptr = output_buf[ctr] + output_col;
1.1785 +
1.1786 + /* Even part */
1.1787 +
1.1788 + /* Add fudge factor here for final descale. */
1.1789 + z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.1790 + z1 <<= CONST_BITS;
1.1791 +
1.1792 + z2 = (INT32) wsptr[2];
1.1793 + z3 = (INT32) wsptr[4];
1.1794 + z4 = (INT32) wsptr[6];
1.1795 +
1.1796 + tmp10 = z3 + z4;
1.1797 + tmp11 = z3 - z4;
1.1798 +
1.1799 + tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */
1.1800 + tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */
1.1801 +
1.1802 + tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */
1.1803 + tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */
1.1804 +
1.1805 + tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */
1.1806 + tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */
1.1807 +
1.1808 + tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */
1.1809 + tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */
1.1810 +
1.1811 + tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */
1.1812 + tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */
1.1813 +
1.1814 + tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
1.1815 + tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */
1.1816 +
1.1817 + tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */
1.1818 +
1.1819 + /* Odd part */
1.1820 +
1.1821 + z1 = (INT32) wsptr[1];
1.1822 + z2 = (INT32) wsptr[3];
1.1823 + z3 = (INT32) wsptr[5];
1.1824 + z4 = (INT32) wsptr[7];
1.1825 +
1.1826 + tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */
1.1827 + tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */
1.1828 + tmp15 = z1 + z4;
1.1829 + tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */
1.1830 + tmp10 = tmp11 + tmp12 + tmp13 -
1.1831 + MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */
1.1832 + tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458)); /* -c11 */
1.1833 + tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
1.1834 + tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
1.1835 + tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945)); /* -c5 */
1.1836 + tmp11 += tmp14;
1.1837 + tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
1.1838 + tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813)); /* -c9 */
1.1839 + tmp12 += tmp14;
1.1840 + tmp13 += tmp14;
1.1841 + tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */
1.1842 + tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
1.1843 + MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */
1.1844 + z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */
1.1845 + tmp14 += z1;
1.1846 + tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */
1.1847 + MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */
1.1848 +
1.1849 + /* Final output stage */
1.1850 +
1.1851 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1.1852 + CONST_BITS+PASS1_BITS+3)
1.1853 + & RANGE_MASK];
1.1854 + outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1.1855 + CONST_BITS+PASS1_BITS+3)
1.1856 + & RANGE_MASK];
1.1857 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1.1858 + CONST_BITS+PASS1_BITS+3)
1.1859 + & RANGE_MASK];
1.1860 + outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1.1861 + CONST_BITS+PASS1_BITS+3)
1.1862 + & RANGE_MASK];
1.1863 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1.1864 + CONST_BITS+PASS1_BITS+3)
1.1865 + & RANGE_MASK];
1.1866 + outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1.1867 + CONST_BITS+PASS1_BITS+3)
1.1868 + & RANGE_MASK];
1.1869 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1.1870 + CONST_BITS+PASS1_BITS+3)
1.1871 + & RANGE_MASK];
1.1872 + outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1.1873 + CONST_BITS+PASS1_BITS+3)
1.1874 + & RANGE_MASK];
1.1875 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1.1876 + CONST_BITS+PASS1_BITS+3)
1.1877 + & RANGE_MASK];
1.1878 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1.1879 + CONST_BITS+PASS1_BITS+3)
1.1880 + & RANGE_MASK];
1.1881 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1.1882 + CONST_BITS+PASS1_BITS+3)
1.1883 + & RANGE_MASK];
1.1884 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1.1885 + CONST_BITS+PASS1_BITS+3)
1.1886 + & RANGE_MASK];
1.1887 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26,
1.1888 + CONST_BITS+PASS1_BITS+3)
1.1889 + & RANGE_MASK];
1.1890 +
1.1891 + wsptr += 8; /* advance pointer to next row */
1.1892 + }
1.1893 +}
1.1894 +
1.1895 +
1.1896 +/*
1.1897 + * Perform dequantization and inverse DCT on one block of coefficients,
1.1898 + * producing a 14x14 output block.
1.1899 + *
1.1900 + * Optimized algorithm with 20 multiplications in the 1-D kernel.
1.1901 + * cK represents sqrt(2) * cos(K*pi/28).
1.1902 + */
1.1903 +
1.1904 +GLOBAL(void)
1.1905 +jpeg_idct_14x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.1906 + JCOEFPTR coef_block,
1.1907 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.1908 +{
1.1909 + INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
1.1910 + INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
1.1911 + INT32 z1, z2, z3, z4;
1.1912 + JCOEFPTR inptr;
1.1913 + ISLOW_MULT_TYPE * quantptr;
1.1914 + int * wsptr;
1.1915 + JSAMPROW outptr;
1.1916 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.1917 + int ctr;
1.1918 + int workspace[8*14]; /* buffers data between passes */
1.1919 + SHIFT_TEMPS
1.1920 +
1.1921 + /* Pass 1: process columns from input, store into work array. */
1.1922 +
1.1923 + inptr = coef_block;
1.1924 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.1925 + wsptr = workspace;
1.1926 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.1927 + /* Even part */
1.1928 +
1.1929 + z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.1930 + z1 <<= CONST_BITS;
1.1931 + /* Add fudge factor here for final descale. */
1.1932 + z1 += ONE << (CONST_BITS-PASS1_BITS-1);
1.1933 + z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.1934 + z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
1.1935 + z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
1.1936 + z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
1.1937 +
1.1938 + tmp10 = z1 + z2;
1.1939 + tmp11 = z1 + z3;
1.1940 + tmp12 = z1 - z4;
1.1941 +
1.1942 + tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)*2 */
1.1943 + CONST_BITS-PASS1_BITS);
1.1944 +
1.1945 + z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.1946 + z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.1947 +
1.1948 + z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
1.1949 +
1.1950 + tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
1.1951 + tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
1.1952 + tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
1.1953 + MULTIPLY(z2, FIX(1.378756276)); /* c2 */
1.1954 +
1.1955 + tmp20 = tmp10 + tmp13;
1.1956 + tmp26 = tmp10 - tmp13;
1.1957 + tmp21 = tmp11 + tmp14;
1.1958 + tmp25 = tmp11 - tmp14;
1.1959 + tmp22 = tmp12 + tmp15;
1.1960 + tmp24 = tmp12 - tmp15;
1.1961 +
1.1962 + /* Odd part */
1.1963 +
1.1964 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.1965 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.1966 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.1967 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.1968 + tmp13 = z4 << CONST_BITS;
1.1969 +
1.1970 + tmp14 = z1 + z3;
1.1971 + tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
1.1972 + tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
1.1973 + tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
1.1974 + tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
1.1975 + tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
1.1976 + z1 -= z2;
1.1977 + tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */
1.1978 + tmp16 += tmp15;
1.1979 + z1 += z4;
1.1980 + z4 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
1.1981 + tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
1.1982 + tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
1.1983 + z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
1.1984 + tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
1.1985 + tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
1.1986 +
1.1987 + tmp13 = (z1 - z3) << PASS1_BITS;
1.1988 +
1.1989 + /* Final output stage */
1.1990 +
1.1991 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1.1992 + wsptr[8*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1.1993 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1.1994 + wsptr[8*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1.1995 + wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1.1996 + wsptr[8*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1.1997 + wsptr[8*3] = (int) (tmp23 + tmp13);
1.1998 + wsptr[8*10] = (int) (tmp23 - tmp13);
1.1999 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1.2000 + wsptr[8*9] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1.2001 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1.2002 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1.2003 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
1.2004 + wsptr[8*7] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
1.2005 + }
1.2006 +
1.2007 + /* Pass 2: process 14 rows from work array, store into output array. */
1.2008 +
1.2009 + wsptr = workspace;
1.2010 + for (ctr = 0; ctr < 14; ctr++) {
1.2011 + outptr = output_buf[ctr] + output_col;
1.2012 +
1.2013 + /* Even part */
1.2014 +
1.2015 + /* Add fudge factor here for final descale. */
1.2016 + z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.2017 + z1 <<= CONST_BITS;
1.2018 + z4 = (INT32) wsptr[4];
1.2019 + z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */
1.2020 + z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */
1.2021 + z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */
1.2022 +
1.2023 + tmp10 = z1 + z2;
1.2024 + tmp11 = z1 + z3;
1.2025 + tmp12 = z1 - z4;
1.2026 +
1.2027 + tmp23 = z1 - ((z2 + z3 - z4) << 1); /* c0 = (c4+c12-c8)*2 */
1.2028 +
1.2029 + z1 = (INT32) wsptr[2];
1.2030 + z2 = (INT32) wsptr[6];
1.2031 +
1.2032 + z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */
1.2033 +
1.2034 + tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
1.2035 + tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
1.2036 + tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */
1.2037 + MULTIPLY(z2, FIX(1.378756276)); /* c2 */
1.2038 +
1.2039 + tmp20 = tmp10 + tmp13;
1.2040 + tmp26 = tmp10 - tmp13;
1.2041 + tmp21 = tmp11 + tmp14;
1.2042 + tmp25 = tmp11 - tmp14;
1.2043 + tmp22 = tmp12 + tmp15;
1.2044 + tmp24 = tmp12 - tmp15;
1.2045 +
1.2046 + /* Odd part */
1.2047 +
1.2048 + z1 = (INT32) wsptr[1];
1.2049 + z2 = (INT32) wsptr[3];
1.2050 + z3 = (INT32) wsptr[5];
1.2051 + z4 = (INT32) wsptr[7];
1.2052 + z4 <<= CONST_BITS;
1.2053 +
1.2054 + tmp14 = z1 + z3;
1.2055 + tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */
1.2056 + tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */
1.2057 + tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
1.2058 + tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */
1.2059 + tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */
1.2060 + z1 -= z2;
1.2061 + tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */
1.2062 + tmp16 += tmp15;
1.2063 + tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4; /* -c13 */
1.2064 + tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */
1.2065 + tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */
1.2066 + tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */
1.2067 + tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
1.2068 + tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */
1.2069 +
1.2070 + tmp13 = ((z1 - z3) << CONST_BITS) + z4;
1.2071 +
1.2072 + /* Final output stage */
1.2073 +
1.2074 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1.2075 + CONST_BITS+PASS1_BITS+3)
1.2076 + & RANGE_MASK];
1.2077 + outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1.2078 + CONST_BITS+PASS1_BITS+3)
1.2079 + & RANGE_MASK];
1.2080 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1.2081 + CONST_BITS+PASS1_BITS+3)
1.2082 + & RANGE_MASK];
1.2083 + outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1.2084 + CONST_BITS+PASS1_BITS+3)
1.2085 + & RANGE_MASK];
1.2086 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1.2087 + CONST_BITS+PASS1_BITS+3)
1.2088 + & RANGE_MASK];
1.2089 + outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1.2090 + CONST_BITS+PASS1_BITS+3)
1.2091 + & RANGE_MASK];
1.2092 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1.2093 + CONST_BITS+PASS1_BITS+3)
1.2094 + & RANGE_MASK];
1.2095 + outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1.2096 + CONST_BITS+PASS1_BITS+3)
1.2097 + & RANGE_MASK];
1.2098 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1.2099 + CONST_BITS+PASS1_BITS+3)
1.2100 + & RANGE_MASK];
1.2101 + outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1.2102 + CONST_BITS+PASS1_BITS+3)
1.2103 + & RANGE_MASK];
1.2104 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1.2105 + CONST_BITS+PASS1_BITS+3)
1.2106 + & RANGE_MASK];
1.2107 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1.2108 + CONST_BITS+PASS1_BITS+3)
1.2109 + & RANGE_MASK];
1.2110 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
1.2111 + CONST_BITS+PASS1_BITS+3)
1.2112 + & RANGE_MASK];
1.2113 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
1.2114 + CONST_BITS+PASS1_BITS+3)
1.2115 + & RANGE_MASK];
1.2116 +
1.2117 + wsptr += 8; /* advance pointer to next row */
1.2118 + }
1.2119 +}
1.2120 +
1.2121 +
1.2122 +/*
1.2123 + * Perform dequantization and inverse DCT on one block of coefficients,
1.2124 + * producing a 15x15 output block.
1.2125 + *
1.2126 + * Optimized algorithm with 22 multiplications in the 1-D kernel.
1.2127 + * cK represents sqrt(2) * cos(K*pi/30).
1.2128 + */
1.2129 +
1.2130 +GLOBAL(void)
1.2131 +jpeg_idct_15x15 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.2132 + JCOEFPTR coef_block,
1.2133 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.2134 +{
1.2135 + INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
1.2136 + INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
1.2137 + INT32 z1, z2, z3, z4;
1.2138 + JCOEFPTR inptr;
1.2139 + ISLOW_MULT_TYPE * quantptr;
1.2140 + int * wsptr;
1.2141 + JSAMPROW outptr;
1.2142 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.2143 + int ctr;
1.2144 + int workspace[8*15]; /* buffers data between passes */
1.2145 + SHIFT_TEMPS
1.2146 +
1.2147 + /* Pass 1: process columns from input, store into work array. */
1.2148 +
1.2149 + inptr = coef_block;
1.2150 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.2151 + wsptr = workspace;
1.2152 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.2153 + /* Even part */
1.2154 +
1.2155 + z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.2156 + z1 <<= CONST_BITS;
1.2157 + /* Add fudge factor here for final descale. */
1.2158 + z1 += ONE << (CONST_BITS-PASS1_BITS-1);
1.2159 +
1.2160 + z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.2161 + z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.2162 + z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.2163 +
1.2164 + tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
1.2165 + tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
1.2166 +
1.2167 + tmp12 = z1 - tmp10;
1.2168 + tmp13 = z1 + tmp11;
1.2169 + z1 -= (tmp11 - tmp10) << 1; /* c0 = (c6-c12)*2 */
1.2170 +
1.2171 + z4 = z2 - z3;
1.2172 + z3 += z2;
1.2173 + tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
1.2174 + tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
1.2175 + z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
1.2176 +
1.2177 + tmp20 = tmp13 + tmp10 + tmp11;
1.2178 + tmp23 = tmp12 - tmp10 + tmp11 + z2;
1.2179 +
1.2180 + tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
1.2181 + tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
1.2182 +
1.2183 + tmp25 = tmp13 - tmp10 - tmp11;
1.2184 + tmp26 = tmp12 + tmp10 - tmp11 - z2;
1.2185 +
1.2186 + tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
1.2187 + tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
1.2188 +
1.2189 + tmp21 = tmp12 + tmp10 + tmp11;
1.2190 + tmp24 = tmp13 - tmp10 + tmp11;
1.2191 + tmp11 += tmp11;
1.2192 + tmp22 = z1 + tmp11; /* c10 = c6-c12 */
1.2193 + tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
1.2194 +
1.2195 + /* Odd part */
1.2196 +
1.2197 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.2198 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.2199 + z4 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.2200 + z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
1.2201 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.2202 +
1.2203 + tmp13 = z2 - z4;
1.2204 + tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
1.2205 + tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
1.2206 + tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
1.2207 +
1.2208 + tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
1.2209 + tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
1.2210 + z2 = z1 - z4;
1.2211 + tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
1.2212 +
1.2213 + tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
1.2214 + tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
1.2215 + tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
1.2216 + z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
1.2217 + tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
1.2218 + tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
1.2219 +
1.2220 + /* Final output stage */
1.2221 +
1.2222 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
1.2223 + wsptr[8*14] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
1.2224 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
1.2225 + wsptr[8*13] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
1.2226 + wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
1.2227 + wsptr[8*12] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
1.2228 + wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
1.2229 + wsptr[8*11] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
1.2230 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
1.2231 + wsptr[8*10] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
1.2232 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
1.2233 + wsptr[8*9] = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
1.2234 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
1.2235 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
1.2236 + wsptr[8*7] = (int) RIGHT_SHIFT(tmp27, CONST_BITS-PASS1_BITS);
1.2237 + }
1.2238 +
1.2239 + /* Pass 2: process 15 rows from work array, store into output array. */
1.2240 +
1.2241 + wsptr = workspace;
1.2242 + for (ctr = 0; ctr < 15; ctr++) {
1.2243 + outptr = output_buf[ctr] + output_col;
1.2244 +
1.2245 + /* Even part */
1.2246 +
1.2247 + /* Add fudge factor here for final descale. */
1.2248 + z1 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.2249 + z1 <<= CONST_BITS;
1.2250 +
1.2251 + z2 = (INT32) wsptr[2];
1.2252 + z3 = (INT32) wsptr[4];
1.2253 + z4 = (INT32) wsptr[6];
1.2254 +
1.2255 + tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
1.2256 + tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */
1.2257 +
1.2258 + tmp12 = z1 - tmp10;
1.2259 + tmp13 = z1 + tmp11;
1.2260 + z1 -= (tmp11 - tmp10) << 1; /* c0 = (c6-c12)*2 */
1.2261 +
1.2262 + z4 = z2 - z3;
1.2263 + z3 += z2;
1.2264 + tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
1.2265 + tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
1.2266 + z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */
1.2267 +
1.2268 + tmp20 = tmp13 + tmp10 + tmp11;
1.2269 + tmp23 = tmp12 - tmp10 + tmp11 + z2;
1.2270 +
1.2271 + tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
1.2272 + tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */
1.2273 +
1.2274 + tmp25 = tmp13 - tmp10 - tmp11;
1.2275 + tmp26 = tmp12 + tmp10 - tmp11 - z2;
1.2276 +
1.2277 + tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
1.2278 + tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */
1.2279 +
1.2280 + tmp21 = tmp12 + tmp10 + tmp11;
1.2281 + tmp24 = tmp13 - tmp10 + tmp11;
1.2282 + tmp11 += tmp11;
1.2283 + tmp22 = z1 + tmp11; /* c10 = c6-c12 */
1.2284 + tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */
1.2285 +
1.2286 + /* Odd part */
1.2287 +
1.2288 + z1 = (INT32) wsptr[1];
1.2289 + z2 = (INT32) wsptr[3];
1.2290 + z4 = (INT32) wsptr[5];
1.2291 + z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */
1.2292 + z4 = (INT32) wsptr[7];
1.2293 +
1.2294 + tmp13 = z2 - z4;
1.2295 + tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */
1.2296 + tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */
1.2297 + tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */
1.2298 +
1.2299 + tmp13 = MULTIPLY(z2, - FIX(0.831253876)); /* -c9 */
1.2300 + tmp15 = MULTIPLY(z2, - FIX(1.344997024)); /* -c3 */
1.2301 + z2 = z1 - z4;
1.2302 + tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */
1.2303 +
1.2304 + tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
1.2305 + tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
1.2306 + tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */
1.2307 + z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */
1.2308 + tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */
1.2309 + tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */
1.2310 +
1.2311 + /* Final output stage */
1.2312 +
1.2313 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
1.2314 + CONST_BITS+PASS1_BITS+3)
1.2315 + & RANGE_MASK];
1.2316 + outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
1.2317 + CONST_BITS+PASS1_BITS+3)
1.2318 + & RANGE_MASK];
1.2319 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
1.2320 + CONST_BITS+PASS1_BITS+3)
1.2321 + & RANGE_MASK];
1.2322 + outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
1.2323 + CONST_BITS+PASS1_BITS+3)
1.2324 + & RANGE_MASK];
1.2325 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
1.2326 + CONST_BITS+PASS1_BITS+3)
1.2327 + & RANGE_MASK];
1.2328 + outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
1.2329 + CONST_BITS+PASS1_BITS+3)
1.2330 + & RANGE_MASK];
1.2331 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
1.2332 + CONST_BITS+PASS1_BITS+3)
1.2333 + & RANGE_MASK];
1.2334 + outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
1.2335 + CONST_BITS+PASS1_BITS+3)
1.2336 + & RANGE_MASK];
1.2337 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
1.2338 + CONST_BITS+PASS1_BITS+3)
1.2339 + & RANGE_MASK];
1.2340 + outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
1.2341 + CONST_BITS+PASS1_BITS+3)
1.2342 + & RANGE_MASK];
1.2343 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
1.2344 + CONST_BITS+PASS1_BITS+3)
1.2345 + & RANGE_MASK];
1.2346 + outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
1.2347 + CONST_BITS+PASS1_BITS+3)
1.2348 + & RANGE_MASK];
1.2349 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
1.2350 + CONST_BITS+PASS1_BITS+3)
1.2351 + & RANGE_MASK];
1.2352 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
1.2353 + CONST_BITS+PASS1_BITS+3)
1.2354 + & RANGE_MASK];
1.2355 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27,
1.2356 + CONST_BITS+PASS1_BITS+3)
1.2357 + & RANGE_MASK];
1.2358 +
1.2359 + wsptr += 8; /* advance pointer to next row */
1.2360 + }
1.2361 +}
1.2362 +
1.2363 +
1.2364 +/*
1.2365 + * Perform dequantization and inverse DCT on one block of coefficients,
1.2366 + * producing a 16x16 output block.
1.2367 + *
1.2368 + * Optimized algorithm with 28 multiplications in the 1-D kernel.
1.2369 + * cK represents sqrt(2) * cos(K*pi/32).
1.2370 + */
1.2371 +
1.2372 +GLOBAL(void)
1.2373 +jpeg_idct_16x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
1.2374 + JCOEFPTR coef_block,
1.2375 + JSAMPARRAY output_buf, JDIMENSION output_col)
1.2376 +{
1.2377 + INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
1.2378 + INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
1.2379 + INT32 z1, z2, z3, z4;
1.2380 + JCOEFPTR inptr;
1.2381 + ISLOW_MULT_TYPE * quantptr;
1.2382 + int * wsptr;
1.2383 + JSAMPROW outptr;
1.2384 + JSAMPLE *range_limit = IDCT_range_limit(cinfo);
1.2385 + int ctr;
1.2386 + int workspace[8*16]; /* buffers data between passes */
1.2387 + SHIFT_TEMPS
1.2388 +
1.2389 + /* Pass 1: process columns from input, store into work array. */
1.2390 +
1.2391 + inptr = coef_block;
1.2392 + quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
1.2393 + wsptr = workspace;
1.2394 + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
1.2395 + /* Even part */
1.2396 +
1.2397 + tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
1.2398 + tmp0 <<= CONST_BITS;
1.2399 + /* Add fudge factor here for final descale. */
1.2400 + tmp0 += 1 << (CONST_BITS-PASS1_BITS-1);
1.2401 +
1.2402 + z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
1.2403 + tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
1.2404 + tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
1.2405 +
1.2406 + tmp10 = tmp0 + tmp1;
1.2407 + tmp11 = tmp0 - tmp1;
1.2408 + tmp12 = tmp0 + tmp2;
1.2409 + tmp13 = tmp0 - tmp2;
1.2410 +
1.2411 + z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
1.2412 + z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
1.2413 + z3 = z1 - z2;
1.2414 + z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
1.2415 + z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
1.2416 +
1.2417 + tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
1.2418 + tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
1.2419 + tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
1.2420 + tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
1.2421 +
1.2422 + tmp20 = tmp10 + tmp0;
1.2423 + tmp27 = tmp10 - tmp0;
1.2424 + tmp21 = tmp12 + tmp1;
1.2425 + tmp26 = tmp12 - tmp1;
1.2426 + tmp22 = tmp13 + tmp2;
1.2427 + tmp25 = tmp13 - tmp2;
1.2428 + tmp23 = tmp11 + tmp3;
1.2429 + tmp24 = tmp11 - tmp3;
1.2430 +
1.2431 + /* Odd part */
1.2432 +
1.2433 + z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
1.2434 + z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
1.2435 + z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
1.2436 + z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
1.2437 +
1.2438 + tmp11 = z1 + z3;
1.2439 +
1.2440 + tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
1.2441 + tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
1.2442 + tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
1.2443 + tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
1.2444 + tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
1.2445 + tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
1.2446 + tmp0 = tmp1 + tmp2 + tmp3 -
1.2447 + MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
1.2448 + tmp13 = tmp10 + tmp11 + tmp12 -
1.2449 + MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
1.2450 + z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
1.2451 + tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
1.2452 + tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
1.2453 + z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
1.2454 + tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
1.2455 + tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
1.2456 + z2 += z4;
1.2457 + z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
1.2458 + tmp1 += z1;
1.2459 + tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
1.2460 + z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
1.2461 + tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
1.2462 + tmp12 += z2;
1.2463 + z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
1.2464 + tmp2 += z2;
1.2465 + tmp3 += z2;
1.2466 + z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
1.2467 + tmp10 += z2;
1.2468 + tmp11 += z2;
1.2469 +
1.2470 + /* Final output stage */
1.2471 +
1.2472 + wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS-PASS1_BITS);
1.2473 + wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS-PASS1_BITS);
1.2474 + wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS-PASS1_BITS);
1.2475 + wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS-PASS1_BITS);
1.2476 + wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS-PASS1_BITS);
1.2477 + wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS-PASS1_BITS);
1.2478 + wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS-PASS1_BITS);
1.2479 + wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS-PASS1_BITS);
1.2480 + wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
1.2481 + wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
1.2482 + wsptr[8*5] = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
1.2483 + wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
1.2484 + wsptr[8*6] = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
1.2485 + wsptr[8*9] = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
1.2486 + wsptr[8*7] = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
1.2487 + wsptr[8*8] = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
1.2488 + }
1.2489 +
1.2490 + /* Pass 2: process 16 rows from work array, store into output array. */
1.2491 +
1.2492 + wsptr = workspace;
1.2493 + for (ctr = 0; ctr < 16; ctr++) {
1.2494 + outptr = output_buf[ctr] + output_col;
1.2495 +
1.2496 + /* Even part */
1.2497 +
1.2498 + /* Add fudge factor here for final descale. */
1.2499 + tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
1.2500 + tmp0 <<= CONST_BITS;
1.2501 +
1.2502 + z1 = (INT32) wsptr[4];
1.2503 + tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */
1.2504 + tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */
1.2505 +
1.2506 + tmp10 = tmp0 + tmp1;
1.2507 + tmp11 = tmp0 - tmp1;
1.2508 + tmp12 = tmp0 + tmp2;
1.2509 + tmp13 = tmp0 - tmp2;
1.2510 +
1.2511 + z1 = (INT32) wsptr[2];
1.2512 + z2 = (INT32) wsptr[6];
1.2513 + z3 = z1 - z2;
1.2514 + z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */
1.2515 + z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */
1.2516 +
1.2517 + tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */
1.2518 + tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */
1.2519 + tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
1.2520 + tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */
1.2521 +
1.2522 + tmp20 = tmp10 + tmp0;
1.2523 + tmp27 = tmp10 - tmp0;
1.2524 + tmp21 = tmp12 + tmp1;
1.2525 + tmp26 = tmp12 - tmp1;
1.2526 + tmp22 = tmp13 + tmp2;
1.2527 + tmp25 = tmp13 - tmp2;
1.2528 + tmp23 = tmp11 + tmp3;
1.2529 + tmp24 = tmp11 - tmp3;
1.2530 +
1.2531 + /* Odd part */
1.2532 +
1.2533 + z1 = (INT32) wsptr[1];
1.2534 + z2 = (INT32) wsptr[3];
1.2535 + z3 = (INT32) wsptr[5];
1.2536 + z4 = (INT32) wsptr[7];
1.2537 +
1.2538 + tmp11 = z1 + z3;
1.2539 +
1.2540 + tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */
1.2541 + tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */
1.2542 + tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */
1.2543 + tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */
1.2544 + tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */
1.2545 + tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */
1.2546 + tmp0 = tmp1 + tmp2 + tmp3 -
1.2547 + MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */
1.2548 + tmp13 = tmp10 + tmp11 + tmp12 -
1.2549 + MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */
1.2550 + z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */
1.2551 + tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */
1.2552 + tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */
1.2553 + z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */
1.2554 + tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */
1.2555 + tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */
1.2556 + z2 += z4;
1.2557 + z1 = MULTIPLY(z2, - FIX(0.666655658)); /* -c11 */
1.2558 + tmp1 += z1;
1.2559 + tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */
1.2560 + z2 = MULTIPLY(z2, - FIX(1.247225013)); /* -c5 */
1.2561 + tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */
1.2562 + tmp12 += z2;
1.2563 + z2 = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
1.2564 + tmp2 += z2;
1.2565 + tmp3 += z2;
1.2566 + z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */
1.2567 + tmp10 += z2;
1.2568 + tmp11 += z2;
1.2569 +
1.2570 + /* Final output stage */
1.2571 +
1.2572 + outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
1.2573 + CONST_BITS+PASS1_BITS+3)
1.2574 + & RANGE_MASK];
1.2575 + outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
1.2576 + CONST_BITS+PASS1_BITS+3)
1.2577 + & RANGE_MASK];
1.2578 + outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
1.2579 + CONST_BITS+PASS1_BITS+3)
1.2580 + & RANGE_MASK];
1.2581 + outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
1.2582 + CONST_BITS+PASS1_BITS+3)
1.2583 + & RANGE_MASK];
1.2584 + outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
1.2585 + CONST_BITS+PASS1_BITS+3)
1.2586 + & RANGE_MASK];
1.2587 + outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
1.2588 + CONST_BITS+PASS1_BITS+3)
1.2589 + & RANGE_MASK];
1.2590 + outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
1.2591 + CONST_BITS+PASS1_BITS+3)
1.2592 + & RANGE_MASK];
1.2593 + outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
1.2594 + CONST_BITS+PASS1_BITS+3)
1.2595 + & RANGE_MASK];
1.2596 + outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
1.2597 + CONST_BITS+PASS1_BITS+3)
1.2598 + & RANGE_MASK];
1.2599 + outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
1.2600 + CONST_BITS+PASS1_BITS+3)
1.2601 + & RANGE_MASK];
1.2602 + outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
1.2603 + CONST_BITS+PASS1_BITS+3)
1.2604 + & RANGE_MASK];
1.2605 + outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
1.2606 + CONST_BITS+PASS1_BITS+3)
1.2607 + & RANGE_MASK];
1.2608 + outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
1.2609 + CONST_BITS+PASS1_BITS+3)
1.2610 + & RANGE_MASK];
1.2611 + outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
1.2612 + CONST_BITS+PASS1_BITS+3)
1.2613 + & RANGE_MASK];
1.2614 + outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
1.2615 + CONST_BITS+PASS1_BITS+3)
1.2616 + & RANGE_MASK];
1.2617 + outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
1.2618 + CONST_BITS+PASS1_BITS+3)
1.2619 + & RANGE_MASK];
1.2620 +
1.2621 + wsptr += 8; /* advance pointer to next row */
1.2622 + }
1.2623 +}
1.2624 +
1.2625 +#endif /* IDCT_SCALING_SUPPORTED */
1.2626 +#endif /* DCT_ISLOW_SUPPORTED */
