yading@10: /* yading@10: * This file is part of the Independent JPEG Group's software. yading@10: * yading@10: * The authors make NO WARRANTY or representation, either express or implied, yading@10: * with respect to this software, its quality, accuracy, merchantability, or yading@10: * fitness for a particular purpose. This software is provided "AS IS", and yading@10: * you, its user, assume the entire risk as to its quality and accuracy. yading@10: * yading@10: * This software is copyright (C) 1991, 1992, Thomas G. Lane. yading@10: * All Rights Reserved except as specified below. yading@10: * yading@10: * Permission is hereby granted to use, copy, modify, and distribute this yading@10: * software (or portions thereof) for any purpose, without fee, subject to yading@10: * these conditions: yading@10: * (1) If any part of the source code for this software is distributed, then yading@10: * this README file must be included, with this copyright and no-warranty yading@10: * notice unaltered; and any additions, deletions, or changes to the original yading@10: * files must be clearly indicated in accompanying documentation. yading@10: * (2) If only executable code is distributed, then the accompanying yading@10: * documentation must state that "this software is based in part on the work yading@10: * of the Independent JPEG Group". yading@10: * (3) Permission for use of this software is granted only if the user accepts yading@10: * full responsibility for any undesirable consequences; the authors accept yading@10: * NO LIABILITY for damages of any kind. yading@10: * yading@10: * These conditions apply to any software derived from or based on the IJG yading@10: * code, not just to the unmodified library. If you use our work, you ought yading@10: * to acknowledge us. yading@10: * yading@10: * Permission is NOT granted for the use of any IJG author's name or company yading@10: * name in advertising or publicity relating to this software or products yading@10: * derived from it. This software may be referred to only as "the Independent yading@10: * JPEG Group's software". yading@10: * yading@10: * We specifically permit and encourage the use of this software as the basis yading@10: * of commercial products, provided that all warranty or liability claims are yading@10: * assumed by the product vendor. yading@10: * yading@10: * This file contains the basic inverse-DCT transformation subroutine. yading@10: * yading@10: * This implementation is based on an algorithm described in yading@10: * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT yading@10: * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics, yading@10: * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991. yading@10: * The primary algorithm described there uses 11 multiplies and 29 adds. yading@10: * We use their alternate method with 12 multiplies and 32 adds. yading@10: * The advantage of this method is that no data path contains more than one yading@10: * multiplication; this allows a very simple and accurate implementation in yading@10: * scaled fixed-point arithmetic, with a minimal number of shifts. yading@10: * yading@10: * I've made lots of modifications to attempt to take advantage of the yading@10: * sparse nature of the DCT matrices we're getting. Although the logic yading@10: * is cumbersome, it's straightforward and the resulting code is much yading@10: * faster. yading@10: * yading@10: * A better way to do this would be to pass in the DCT block as a sparse yading@10: * matrix, perhaps with the difference cases encoded. yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * Independent JPEG Group's LLM idct. yading@10: */ yading@10: yading@10: #include "libavutil/common.h" yading@10: #include "dct.h" yading@10: yading@10: #define EIGHT_BIT_SAMPLES yading@10: yading@10: #define DCTSIZE 8 yading@10: #define DCTSIZE2 64 yading@10: yading@10: #define GLOBAL yading@10: yading@10: #define RIGHT_SHIFT(x, n) ((x) >> (n)) yading@10: yading@10: typedef int16_t DCTBLOCK[DCTSIZE2]; yading@10: yading@10: #define CONST_BITS 13 yading@10: yading@10: /* yading@10: * This routine is specialized to the case DCTSIZE = 8. yading@10: */ yading@10: yading@10: #if DCTSIZE != 8 yading@10: Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ yading@10: #endif yading@10: yading@10: yading@10: /* yading@10: * A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT yading@10: * on each column. Direct algorithms are also available, but they are yading@10: * much more complex and seem not to be any faster when reduced to code. yading@10: * yading@10: * The poop on this scaling stuff is as follows: yading@10: * yading@10: * Each 1-D IDCT step produces outputs which are a factor of sqrt(N) yading@10: * larger than the true IDCT outputs. The final outputs are therefore yading@10: * a factor of N larger than desired; since N=8 this can be cured by yading@10: * a simple right shift at the end of the algorithm. The advantage of yading@10: * this arrangement is that we save two multiplications per 1-D IDCT, yading@10: * because the y0 and y4 inputs need not be divided by sqrt(N). yading@10: * yading@10: * We have to do addition and subtraction of the integer inputs, which yading@10: * is no problem, and multiplication by fractional constants, which is yading@10: * a problem to do in integer arithmetic. We multiply all the constants yading@10: * by CONST_SCALE and convert them to integer constants (thus retaining yading@10: * CONST_BITS bits of precision in the constants). After doing a yading@10: * multiplication we have to divide the product by CONST_SCALE, with proper yading@10: * rounding, to produce the correct output. This division can be done yading@10: * cheaply as a right shift of CONST_BITS bits. We postpone shifting yading@10: * as long as possible so that partial sums can be added together with yading@10: * full fractional precision. yading@10: * yading@10: * The outputs of the first pass are scaled up by PASS1_BITS bits so that yading@10: * they are represented to better-than-integral precision. These outputs yading@10: * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word yading@10: * with the recommended scaling. (To scale up 12-bit sample data further, an yading@10: * intermediate int32 array would be needed.) yading@10: * yading@10: * To avoid overflow of the 32-bit intermediate results in pass 2, we must yading@10: * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis yading@10: * shows that the values given below are the most effective. yading@10: */ yading@10: yading@10: #ifdef EIGHT_BIT_SAMPLES yading@10: #define PASS1_BITS 2 yading@10: #else yading@10: #define PASS1_BITS 1 /* lose a little precision to avoid overflow */ yading@10: #endif yading@10: yading@10: #define ONE ((int32_t) 1) yading@10: yading@10: #define CONST_SCALE (ONE << CONST_BITS) yading@10: yading@10: /* Convert a positive real constant to an integer scaled by CONST_SCALE. yading@10: * IMPORTANT: if your compiler doesn't do this arithmetic at compile time, yading@10: * you will pay a significant penalty in run time. In that case, figure yading@10: * the correct integer constant values and insert them by hand. yading@10: */ yading@10: yading@10: /* Actually FIX is no longer used, we precomputed them all */ yading@10: #define FIX(x) ((int32_t) ((x) * CONST_SCALE + 0.5)) yading@10: yading@10: /* Descale and correctly round an int32_t value that's scaled by N bits. yading@10: * We assume RIGHT_SHIFT rounds towards minus infinity, so adding yading@10: * the fudge factor is correct for either sign of X. yading@10: */ yading@10: yading@10: #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n) yading@10: yading@10: /* Multiply an int32_t variable by an int32_t constant to yield an int32_t result. yading@10: * For 8-bit samples with the recommended scaling, all the variable yading@10: * and constant values involved are no more than 16 bits wide, so a yading@10: * 16x16->32 bit multiply can be used instead of a full 32x32 multiply; yading@10: * this provides a useful speedup on many machines. yading@10: * There is no way to specify a 16x16->32 multiply in portable C, but yading@10: * some C compilers will do the right thing if you provide the correct yading@10: * combination of casts. yading@10: * NB: for 12-bit samples, a full 32-bit multiplication will be needed. yading@10: */ yading@10: yading@10: #ifdef EIGHT_BIT_SAMPLES yading@10: #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ yading@10: #define MULTIPLY(var,const) (((int16_t) (var)) * ((int16_t) (const))) yading@10: #endif yading@10: #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */ yading@10: #define MULTIPLY(var,const) (((int16_t) (var)) * ((int32_t) (const))) yading@10: #endif yading@10: #endif yading@10: yading@10: #ifndef MULTIPLY /* default definition */ yading@10: #define MULTIPLY(var,const) ((var) * (const)) yading@10: #endif yading@10: yading@10: yading@10: /* yading@10: Unlike our decoder where we approximate the FIXes, we need to use exact yading@10: ones here or successive P-frames will drift too much with Reference frame coding yading@10: */ yading@10: #define FIX_0_211164243 1730 yading@10: #define FIX_0_275899380 2260 yading@10: #define FIX_0_298631336 2446 yading@10: #define FIX_0_390180644 3196 yading@10: #define FIX_0_509795579 4176 yading@10: #define FIX_0_541196100 4433 yading@10: #define FIX_0_601344887 4926 yading@10: #define FIX_0_765366865 6270 yading@10: #define FIX_0_785694958 6436 yading@10: #define FIX_0_899976223 7373 yading@10: #define FIX_1_061594337 8697 yading@10: #define FIX_1_111140466 9102 yading@10: #define FIX_1_175875602 9633 yading@10: #define FIX_1_306562965 10703 yading@10: #define FIX_1_387039845 11363 yading@10: #define FIX_1_451774981 11893 yading@10: #define FIX_1_501321110 12299 yading@10: #define FIX_1_662939225 13623 yading@10: #define FIX_1_847759065 15137 yading@10: #define FIX_1_961570560 16069 yading@10: #define FIX_2_053119869 16819 yading@10: #define FIX_2_172734803 17799 yading@10: #define FIX_2_562915447 20995 yading@10: #define FIX_3_072711026 25172 yading@10: yading@10: /* yading@10: * Perform the inverse DCT on one block of coefficients. yading@10: */ yading@10: yading@10: void ff_j_rev_dct(DCTBLOCK data) yading@10: { yading@10: int32_t tmp0, tmp1, tmp2, tmp3; yading@10: int32_t tmp10, tmp11, tmp12, tmp13; yading@10: int32_t z1, z2, z3, z4, z5; yading@10: int32_t d0, d1, d2, d3, d4, d5, d6, d7; yading@10: register int16_t *dataptr; yading@10: int rowctr; yading@10: yading@10: /* Pass 1: process rows. */ yading@10: /* Note results are scaled up by sqrt(8) compared to a true IDCT; */ yading@10: /* furthermore, we scale the results by 2**PASS1_BITS. */ yading@10: yading@10: dataptr = data; yading@10: yading@10: for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) { yading@10: /* Due to quantization, we will usually find that many of the input yading@10: * coefficients are zero, especially the AC terms. We can exploit this yading@10: * by short-circuiting the IDCT calculation for any row in which all yading@10: * the AC terms are zero. In that case each output is equal to the yading@10: * DC coefficient (with scale factor as needed). yading@10: * With typical images and quantization tables, half or more of the yading@10: * row DCT calculations can be simplified this way. yading@10: */ yading@10: yading@10: register int *idataptr = (int*)dataptr; yading@10: yading@10: /* WARNING: we do the same permutation as MMX idct to simplify the yading@10: video core */ yading@10: d0 = dataptr[0]; yading@10: d2 = dataptr[1]; yading@10: d4 = dataptr[2]; yading@10: d6 = dataptr[3]; yading@10: d1 = dataptr[4]; yading@10: d3 = dataptr[5]; yading@10: d5 = dataptr[6]; yading@10: d7 = dataptr[7]; yading@10: yading@10: if ((d1 | d2 | d3 | d4 | d5 | d6 | d7) == 0) { yading@10: /* AC terms all zero */ yading@10: if (d0) { yading@10: /* Compute a 32 bit value to assign. */ yading@10: int16_t dcval = (int16_t) (d0 << PASS1_BITS); yading@10: register int v = (dcval & 0xffff) | ((dcval << 16) & 0xffff0000); yading@10: yading@10: idataptr[0] = v; yading@10: idataptr[1] = v; yading@10: idataptr[2] = v; yading@10: idataptr[3] = v; yading@10: } yading@10: yading@10: dataptr += DCTSIZE; /* advance pointer to next row */ yading@10: continue; yading@10: } yading@10: yading@10: /* Even part: reverse the even part of the forward DCT. */ yading@10: /* The rotator is sqrt(2)*c(-6). */ yading@10: { yading@10: if (d6) { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */ yading@10: z1 = MULTIPLY(d2 + d6, FIX_0_541196100); yading@10: tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065); yading@10: tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */ yading@10: tmp2 = MULTIPLY(-d6, FIX_1_306562965); yading@10: tmp3 = MULTIPLY(d6, FIX_0_541196100); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } yading@10: } else { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */ yading@10: tmp2 = MULTIPLY(d2, FIX_0_541196100); yading@10: tmp3 = MULTIPLY(d2, FIX_1_306562965); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */ yading@10: tmp10 = tmp13 = (d0 + d4) << CONST_BITS; yading@10: tmp11 = tmp12 = (d0 - d4) << CONST_BITS; yading@10: } yading@10: } yading@10: yading@10: /* Odd part per figure 8; the matrix is unitary and hence its yading@10: * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively. yading@10: */ yading@10: yading@10: if (d7) { yading@10: if (d5) { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 != 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z2 = d5 + d3; yading@10: z3 = d7 + d3; yading@10: z4 = d5 + d1; yading@10: z5 = MULTIPLY(z3 + z4, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-z1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-z2, FIX_2_562915447); yading@10: z3 = MULTIPLY(-z3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-z4, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 != 0, d7 != 0 */ yading@10: z2 = d5 + d3; yading@10: z3 = d7 + d3; yading@10: z5 = MULTIPLY(z3 + d5, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: z1 = MULTIPLY(-d7, FIX_0_899976223); yading@10: z2 = MULTIPLY(-z2, FIX_2_562915447); yading@10: z3 = MULTIPLY(-z3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-d5, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 = z1 + z4; yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 != 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z4 = d5 + d1; yading@10: z5 = MULTIPLY(d7 + z4, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-z1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-d5, FIX_2_562915447); yading@10: z3 = MULTIPLY(-d7, FIX_1_961570560); yading@10: z4 = MULTIPLY(-z4, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 = z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 != 0, d7 != 0 */ yading@10: tmp0 = MULTIPLY(-d7, FIX_0_601344887); yading@10: z1 = MULTIPLY(-d7, FIX_0_899976223); yading@10: z3 = MULTIPLY(-d7, FIX_1_961570560); yading@10: tmp1 = MULTIPLY(-d5, FIX_0_509795579); yading@10: z2 = MULTIPLY(-d5, FIX_2_562915447); yading@10: z4 = MULTIPLY(-d5, FIX_0_390180644); yading@10: z5 = MULTIPLY(d5 + d7, FIX_1_175875602); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z3; yading@10: tmp1 += z4; yading@10: tmp2 = z2 + z3; yading@10: tmp3 = z1 + z4; yading@10: } yading@10: } yading@10: } else { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 == 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z3 = d7 + d3; yading@10: z5 = MULTIPLY(z3 + d1, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-z1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-d3, FIX_2_562915447); yading@10: z3 = MULTIPLY(-z3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-d1, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 = z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 == 0, d7 != 0 */ yading@10: z3 = d7 + d3; yading@10: yading@10: tmp0 = MULTIPLY(-d7, FIX_0_601344887); yading@10: z1 = MULTIPLY(-d7, FIX_0_899976223); yading@10: tmp2 = MULTIPLY(d3, FIX_0_509795579); yading@10: z2 = MULTIPLY(-d3, FIX_2_562915447); yading@10: z5 = MULTIPLY(z3, FIX_1_175875602); yading@10: z3 = MULTIPLY(-z3, FIX_0_785694958); yading@10: yading@10: tmp0 += z3; yading@10: tmp1 = z2 + z5; yading@10: tmp2 += z3; yading@10: tmp3 = z1 + z5; yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 == 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z5 = MULTIPLY(z1, FIX_1_175875602); yading@10: yading@10: z1 = MULTIPLY(z1, FIX_0_275899380); yading@10: z3 = MULTIPLY(-d7, FIX_1_961570560); yading@10: tmp0 = MULTIPLY(-d7, FIX_1_662939225); yading@10: z4 = MULTIPLY(-d1, FIX_0_390180644); yading@10: tmp3 = MULTIPLY(d1, FIX_1_111140466); yading@10: yading@10: tmp0 += z1; yading@10: tmp1 = z4 + z5; yading@10: tmp2 = z3 + z5; yading@10: tmp3 += z1; yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 == 0, d7 != 0 */ yading@10: tmp0 = MULTIPLY(-d7, FIX_1_387039845); yading@10: tmp1 = MULTIPLY(d7, FIX_1_175875602); yading@10: tmp2 = MULTIPLY(-d7, FIX_0_785694958); yading@10: tmp3 = MULTIPLY(d7, FIX_0_275899380); yading@10: } yading@10: } yading@10: } yading@10: } else { yading@10: if (d5) { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 != 0, d7 == 0 */ yading@10: z2 = d5 + d3; yading@10: z4 = d5 + d1; yading@10: z5 = MULTIPLY(d3 + z4, FIX_1_175875602); yading@10: yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-d1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-z2, FIX_2_562915447); yading@10: z3 = MULTIPLY(-d3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-z4, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 = z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 != 0, d7 == 0 */ yading@10: z2 = d5 + d3; yading@10: yading@10: z5 = MULTIPLY(z2, FIX_1_175875602); yading@10: tmp1 = MULTIPLY(d5, FIX_1_662939225); yading@10: z4 = MULTIPLY(-d5, FIX_0_390180644); yading@10: z2 = MULTIPLY(-z2, FIX_1_387039845); yading@10: tmp2 = MULTIPLY(d3, FIX_1_111140466); yading@10: z3 = MULTIPLY(-d3, FIX_1_961570560); yading@10: yading@10: tmp0 = z3 + z5; yading@10: tmp1 += z2; yading@10: tmp2 += z2; yading@10: tmp3 = z4 + z5; yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 != 0, d7 == 0 */ yading@10: z4 = d5 + d1; yading@10: yading@10: z5 = MULTIPLY(z4, FIX_1_175875602); yading@10: z1 = MULTIPLY(-d1, FIX_0_899976223); yading@10: tmp3 = MULTIPLY(d1, FIX_0_601344887); yading@10: tmp1 = MULTIPLY(-d5, FIX_0_509795579); yading@10: z2 = MULTIPLY(-d5, FIX_2_562915447); yading@10: z4 = MULTIPLY(z4, FIX_0_785694958); yading@10: yading@10: tmp0 = z1 + z5; yading@10: tmp1 += z4; yading@10: tmp2 = z2 + z5; yading@10: tmp3 += z4; yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 != 0, d7 == 0 */ yading@10: tmp0 = MULTIPLY(d5, FIX_1_175875602); yading@10: tmp1 = MULTIPLY(d5, FIX_0_275899380); yading@10: tmp2 = MULTIPLY(-d5, FIX_1_387039845); yading@10: tmp3 = MULTIPLY(d5, FIX_0_785694958); yading@10: } yading@10: } yading@10: } else { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 == 0, d7 == 0 */ yading@10: z5 = d1 + d3; yading@10: tmp3 = MULTIPLY(d1, FIX_0_211164243); yading@10: tmp2 = MULTIPLY(-d3, FIX_1_451774981); yading@10: z1 = MULTIPLY(d1, FIX_1_061594337); yading@10: z2 = MULTIPLY(-d3, FIX_2_172734803); yading@10: z4 = MULTIPLY(z5, FIX_0_785694958); yading@10: z5 = MULTIPLY(z5, FIX_1_175875602); yading@10: yading@10: tmp0 = z1 - z4; yading@10: tmp1 = z2 + z4; yading@10: tmp2 += z5; yading@10: tmp3 += z5; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 == 0, d7 == 0 */ yading@10: tmp0 = MULTIPLY(-d3, FIX_0_785694958); yading@10: tmp1 = MULTIPLY(-d3, FIX_1_387039845); yading@10: tmp2 = MULTIPLY(-d3, FIX_0_275899380); yading@10: tmp3 = MULTIPLY(d3, FIX_1_175875602); yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 == 0, d7 == 0 */ yading@10: tmp0 = MULTIPLY(d1, FIX_0_275899380); yading@10: tmp1 = MULTIPLY(d1, FIX_0_785694958); yading@10: tmp2 = MULTIPLY(d1, FIX_1_175875602); yading@10: tmp3 = MULTIPLY(d1, FIX_1_387039845); yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 == 0, d7 == 0 */ yading@10: tmp0 = tmp1 = tmp2 = tmp3 = 0; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: } yading@10: /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */ yading@10: yading@10: dataptr[0] = (int16_t) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS); yading@10: dataptr[7] = (int16_t) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS); yading@10: dataptr[1] = (int16_t) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS); yading@10: dataptr[6] = (int16_t) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS); yading@10: dataptr[2] = (int16_t) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS); yading@10: dataptr[5] = (int16_t) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS); yading@10: dataptr[3] = (int16_t) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS); yading@10: dataptr[4] = (int16_t) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS); yading@10: yading@10: dataptr += DCTSIZE; /* advance pointer to next row */ yading@10: } yading@10: yading@10: /* Pass 2: process columns. */ yading@10: /* Note that we must descale the results by a factor of 8 == 2**3, */ yading@10: /* and also undo the PASS1_BITS scaling. */ yading@10: yading@10: dataptr = data; yading@10: for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) { yading@10: /* Columns of zeroes can be exploited in the same way as we did with rows. yading@10: * However, the row calculation has created many nonzero AC terms, so the yading@10: * simplification applies less often (typically 5% to 10% of the time). yading@10: * On machines with very fast multiplication, it's possible that the yading@10: * test takes more time than it's worth. In that case this section yading@10: * may be commented out. yading@10: */ yading@10: yading@10: d0 = dataptr[DCTSIZE*0]; yading@10: d1 = dataptr[DCTSIZE*1]; yading@10: d2 = dataptr[DCTSIZE*2]; yading@10: d3 = dataptr[DCTSIZE*3]; yading@10: d4 = dataptr[DCTSIZE*4]; yading@10: d5 = dataptr[DCTSIZE*5]; yading@10: d6 = dataptr[DCTSIZE*6]; yading@10: d7 = dataptr[DCTSIZE*7]; yading@10: yading@10: /* Even part: reverse the even part of the forward DCT. */ yading@10: /* The rotator is sqrt(2)*c(-6). */ yading@10: if (d6) { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */ yading@10: z1 = MULTIPLY(d2 + d6, FIX_0_541196100); yading@10: tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065); yading@10: tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */ yading@10: tmp2 = MULTIPLY(-d6, FIX_1_306562965); yading@10: tmp3 = MULTIPLY(d6, FIX_0_541196100); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } yading@10: } else { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */ yading@10: tmp2 = MULTIPLY(d2, FIX_0_541196100); yading@10: tmp3 = MULTIPLY(d2, FIX_1_306562965); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */ yading@10: tmp10 = tmp13 = (d0 + d4) << CONST_BITS; yading@10: tmp11 = tmp12 = (d0 - d4) << CONST_BITS; yading@10: } yading@10: } yading@10: yading@10: /* Odd part per figure 8; the matrix is unitary and hence its yading@10: * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively. yading@10: */ yading@10: if (d7) { yading@10: if (d5) { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 != 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z2 = d5 + d3; yading@10: z3 = d7 + d3; yading@10: z4 = d5 + d1; yading@10: z5 = MULTIPLY(z3 + z4, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-z1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-z2, FIX_2_562915447); yading@10: z3 = MULTIPLY(-z3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-z4, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 != 0, d7 != 0 */ yading@10: z2 = d5 + d3; yading@10: z3 = d7 + d3; yading@10: z5 = MULTIPLY(z3 + d5, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: z1 = MULTIPLY(-d7, FIX_0_899976223); yading@10: z2 = MULTIPLY(-z2, FIX_2_562915447); yading@10: z3 = MULTIPLY(-z3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-d5, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 = z1 + z4; yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 != 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z3 = d7; yading@10: z4 = d5 + d1; yading@10: z5 = MULTIPLY(z3 + z4, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-z1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-d5, FIX_2_562915447); yading@10: z3 = MULTIPLY(-d7, FIX_1_961570560); yading@10: z4 = MULTIPLY(-z4, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 = z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 != 0, d7 != 0 */ yading@10: tmp0 = MULTIPLY(-d7, FIX_0_601344887); yading@10: z1 = MULTIPLY(-d7, FIX_0_899976223); yading@10: z3 = MULTIPLY(-d7, FIX_1_961570560); yading@10: tmp1 = MULTIPLY(-d5, FIX_0_509795579); yading@10: z2 = MULTIPLY(-d5, FIX_2_562915447); yading@10: z4 = MULTIPLY(-d5, FIX_0_390180644); yading@10: z5 = MULTIPLY(d5 + d7, FIX_1_175875602); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z3; yading@10: tmp1 += z4; yading@10: tmp2 = z2 + z3; yading@10: tmp3 = z1 + z4; yading@10: } yading@10: } yading@10: } else { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 == 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z3 = d7 + d3; yading@10: z5 = MULTIPLY(z3 + d1, FIX_1_175875602); yading@10: yading@10: tmp0 = MULTIPLY(d7, FIX_0_298631336); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-z1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-d3, FIX_2_562915447); yading@10: z3 = MULTIPLY(-z3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-d1, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 += z1 + z3; yading@10: tmp1 = z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 == 0, d7 != 0 */ yading@10: z3 = d7 + d3; yading@10: yading@10: tmp0 = MULTIPLY(-d7, FIX_0_601344887); yading@10: z1 = MULTIPLY(-d7, FIX_0_899976223); yading@10: tmp2 = MULTIPLY(d3, FIX_0_509795579); yading@10: z2 = MULTIPLY(-d3, FIX_2_562915447); yading@10: z5 = MULTIPLY(z3, FIX_1_175875602); yading@10: z3 = MULTIPLY(-z3, FIX_0_785694958); yading@10: yading@10: tmp0 += z3; yading@10: tmp1 = z2 + z5; yading@10: tmp2 += z3; yading@10: tmp3 = z1 + z5; yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 == 0, d7 != 0 */ yading@10: z1 = d7 + d1; yading@10: z5 = MULTIPLY(z1, FIX_1_175875602); yading@10: yading@10: z1 = MULTIPLY(z1, FIX_0_275899380); yading@10: z3 = MULTIPLY(-d7, FIX_1_961570560); yading@10: tmp0 = MULTIPLY(-d7, FIX_1_662939225); yading@10: z4 = MULTIPLY(-d1, FIX_0_390180644); yading@10: tmp3 = MULTIPLY(d1, FIX_1_111140466); yading@10: yading@10: tmp0 += z1; yading@10: tmp1 = z4 + z5; yading@10: tmp2 = z3 + z5; yading@10: tmp3 += z1; yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 == 0, d7 != 0 */ yading@10: tmp0 = MULTIPLY(-d7, FIX_1_387039845); yading@10: tmp1 = MULTIPLY(d7, FIX_1_175875602); yading@10: tmp2 = MULTIPLY(-d7, FIX_0_785694958); yading@10: tmp3 = MULTIPLY(d7, FIX_0_275899380); yading@10: } yading@10: } yading@10: } yading@10: } else { yading@10: if (d5) { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 != 0, d7 == 0 */ yading@10: z2 = d5 + d3; yading@10: z4 = d5 + d1; yading@10: z5 = MULTIPLY(d3 + z4, FIX_1_175875602); yading@10: yading@10: tmp1 = MULTIPLY(d5, FIX_2_053119869); yading@10: tmp2 = MULTIPLY(d3, FIX_3_072711026); yading@10: tmp3 = MULTIPLY(d1, FIX_1_501321110); yading@10: z1 = MULTIPLY(-d1, FIX_0_899976223); yading@10: z2 = MULTIPLY(-z2, FIX_2_562915447); yading@10: z3 = MULTIPLY(-d3, FIX_1_961570560); yading@10: z4 = MULTIPLY(-z4, FIX_0_390180644); yading@10: yading@10: z3 += z5; yading@10: z4 += z5; yading@10: yading@10: tmp0 = z1 + z3; yading@10: tmp1 += z2 + z4; yading@10: tmp2 += z2 + z3; yading@10: tmp3 += z1 + z4; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 != 0, d7 == 0 */ yading@10: z2 = d5 + d3; yading@10: yading@10: z5 = MULTIPLY(z2, FIX_1_175875602); yading@10: tmp1 = MULTIPLY(d5, FIX_1_662939225); yading@10: z4 = MULTIPLY(-d5, FIX_0_390180644); yading@10: z2 = MULTIPLY(-z2, FIX_1_387039845); yading@10: tmp2 = MULTIPLY(d3, FIX_1_111140466); yading@10: z3 = MULTIPLY(-d3, FIX_1_961570560); yading@10: yading@10: tmp0 = z3 + z5; yading@10: tmp1 += z2; yading@10: tmp2 += z2; yading@10: tmp3 = z4 + z5; yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 != 0, d7 == 0 */ yading@10: z4 = d5 + d1; yading@10: yading@10: z5 = MULTIPLY(z4, FIX_1_175875602); yading@10: z1 = MULTIPLY(-d1, FIX_0_899976223); yading@10: tmp3 = MULTIPLY(d1, FIX_0_601344887); yading@10: tmp1 = MULTIPLY(-d5, FIX_0_509795579); yading@10: z2 = MULTIPLY(-d5, FIX_2_562915447); yading@10: z4 = MULTIPLY(z4, FIX_0_785694958); yading@10: yading@10: tmp0 = z1 + z5; yading@10: tmp1 += z4; yading@10: tmp2 = z2 + z5; yading@10: tmp3 += z4; yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 != 0, d7 == 0 */ yading@10: tmp0 = MULTIPLY(d5, FIX_1_175875602); yading@10: tmp1 = MULTIPLY(d5, FIX_0_275899380); yading@10: tmp2 = MULTIPLY(-d5, FIX_1_387039845); yading@10: tmp3 = MULTIPLY(d5, FIX_0_785694958); yading@10: } yading@10: } yading@10: } else { yading@10: if (d3) { yading@10: if (d1) { yading@10: /* d1 != 0, d3 != 0, d5 == 0, d7 == 0 */ yading@10: z5 = d1 + d3; yading@10: tmp3 = MULTIPLY(d1, FIX_0_211164243); yading@10: tmp2 = MULTIPLY(-d3, FIX_1_451774981); yading@10: z1 = MULTIPLY(d1, FIX_1_061594337); yading@10: z2 = MULTIPLY(-d3, FIX_2_172734803); yading@10: z4 = MULTIPLY(z5, FIX_0_785694958); yading@10: z5 = MULTIPLY(z5, FIX_1_175875602); yading@10: yading@10: tmp0 = z1 - z4; yading@10: tmp1 = z2 + z4; yading@10: tmp2 += z5; yading@10: tmp3 += z5; yading@10: } else { yading@10: /* d1 == 0, d3 != 0, d5 == 0, d7 == 0 */ yading@10: tmp0 = MULTIPLY(-d3, FIX_0_785694958); yading@10: tmp1 = MULTIPLY(-d3, FIX_1_387039845); yading@10: tmp2 = MULTIPLY(-d3, FIX_0_275899380); yading@10: tmp3 = MULTIPLY(d3, FIX_1_175875602); yading@10: } yading@10: } else { yading@10: if (d1) { yading@10: /* d1 != 0, d3 == 0, d5 == 0, d7 == 0 */ yading@10: tmp0 = MULTIPLY(d1, FIX_0_275899380); yading@10: tmp1 = MULTIPLY(d1, FIX_0_785694958); yading@10: tmp2 = MULTIPLY(d1, FIX_1_175875602); yading@10: tmp3 = MULTIPLY(d1, FIX_1_387039845); yading@10: } else { yading@10: /* d1 == 0, d3 == 0, d5 == 0, d7 == 0 */ yading@10: tmp0 = tmp1 = tmp2 = tmp3 = 0; yading@10: } yading@10: } yading@10: } yading@10: } yading@10: yading@10: /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */ yading@10: yading@10: dataptr[DCTSIZE*0] = (int16_t) DESCALE(tmp10 + tmp3, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSIZE*7] = (int16_t) DESCALE(tmp10 - tmp3, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSIZE*1] = (int16_t) DESCALE(tmp11 + tmp2, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSIZE*6] = (int16_t) DESCALE(tmp11 - tmp2, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSIZE*2] = (int16_t) DESCALE(tmp12 + tmp1, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSIZE*5] = (int16_t) DESCALE(tmp12 - tmp1, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSIZE*3] = (int16_t) DESCALE(tmp13 + tmp0, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSIZE*4] = (int16_t) DESCALE(tmp13 - tmp0, yading@10: CONST_BITS+PASS1_BITS+3); yading@10: yading@10: dataptr++; /* advance pointer to next column */ yading@10: } yading@10: } yading@10: yading@10: #undef DCTSIZE yading@10: #define DCTSIZE 4 yading@10: #define DCTSTRIDE 8 yading@10: yading@10: void ff_j_rev_dct4(DCTBLOCK data) yading@10: { yading@10: int32_t tmp0, tmp1, tmp2, tmp3; yading@10: int32_t tmp10, tmp11, tmp12, tmp13; yading@10: int32_t z1; yading@10: int32_t d0, d2, d4, d6; yading@10: register int16_t *dataptr; yading@10: int rowctr; yading@10: yading@10: /* Pass 1: process rows. */ yading@10: /* Note results are scaled up by sqrt(8) compared to a true IDCT; */ yading@10: /* furthermore, we scale the results by 2**PASS1_BITS. */ yading@10: yading@10: data[0] += 4; yading@10: yading@10: dataptr = data; yading@10: yading@10: for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) { yading@10: /* Due to quantization, we will usually find that many of the input yading@10: * coefficients are zero, especially the AC terms. We can exploit this yading@10: * by short-circuiting the IDCT calculation for any row in which all yading@10: * the AC terms are zero. In that case each output is equal to the yading@10: * DC coefficient (with scale factor as needed). yading@10: * With typical images and quantization tables, half or more of the yading@10: * row DCT calculations can be simplified this way. yading@10: */ yading@10: yading@10: register int *idataptr = (int*)dataptr; yading@10: yading@10: d0 = dataptr[0]; yading@10: d2 = dataptr[1]; yading@10: d4 = dataptr[2]; yading@10: d6 = dataptr[3]; yading@10: yading@10: if ((d2 | d4 | d6) == 0) { yading@10: /* AC terms all zero */ yading@10: if (d0) { yading@10: /* Compute a 32 bit value to assign. */ yading@10: int16_t dcval = (int16_t) (d0 << PASS1_BITS); yading@10: register int v = (dcval & 0xffff) | ((dcval << 16) & 0xffff0000); yading@10: yading@10: idataptr[0] = v; yading@10: idataptr[1] = v; yading@10: } yading@10: yading@10: dataptr += DCTSTRIDE; /* advance pointer to next row */ yading@10: continue; yading@10: } yading@10: yading@10: /* Even part: reverse the even part of the forward DCT. */ yading@10: /* The rotator is sqrt(2)*c(-6). */ yading@10: if (d6) { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */ yading@10: z1 = MULTIPLY(d2 + d6, FIX_0_541196100); yading@10: tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065); yading@10: tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */ yading@10: tmp2 = MULTIPLY(-d6, FIX_1_306562965); yading@10: tmp3 = MULTIPLY(d6, FIX_0_541196100); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } yading@10: } else { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */ yading@10: tmp2 = MULTIPLY(d2, FIX_0_541196100); yading@10: tmp3 = MULTIPLY(d2, FIX_1_306562965); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */ yading@10: tmp10 = tmp13 = (d0 + d4) << CONST_BITS; yading@10: tmp11 = tmp12 = (d0 - d4) << CONST_BITS; yading@10: } yading@10: } yading@10: yading@10: /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */ yading@10: yading@10: dataptr[0] = (int16_t) DESCALE(tmp10, CONST_BITS-PASS1_BITS); yading@10: dataptr[1] = (int16_t) DESCALE(tmp11, CONST_BITS-PASS1_BITS); yading@10: dataptr[2] = (int16_t) DESCALE(tmp12, CONST_BITS-PASS1_BITS); yading@10: dataptr[3] = (int16_t) DESCALE(tmp13, CONST_BITS-PASS1_BITS); yading@10: yading@10: dataptr += DCTSTRIDE; /* advance pointer to next row */ yading@10: } yading@10: yading@10: /* Pass 2: process columns. */ yading@10: /* Note that we must descale the results by a factor of 8 == 2**3, */ yading@10: /* and also undo the PASS1_BITS scaling. */ yading@10: yading@10: dataptr = data; yading@10: for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) { yading@10: /* Columns of zeroes can be exploited in the same way as we did with rows. yading@10: * However, the row calculation has created many nonzero AC terms, so the yading@10: * simplification applies less often (typically 5% to 10% of the time). yading@10: * On machines with very fast multiplication, it's possible that the yading@10: * test takes more time than it's worth. In that case this section yading@10: * may be commented out. yading@10: */ yading@10: yading@10: d0 = dataptr[DCTSTRIDE*0]; yading@10: d2 = dataptr[DCTSTRIDE*1]; yading@10: d4 = dataptr[DCTSTRIDE*2]; yading@10: d6 = dataptr[DCTSTRIDE*3]; yading@10: yading@10: /* Even part: reverse the even part of the forward DCT. */ yading@10: /* The rotator is sqrt(2)*c(-6). */ yading@10: if (d6) { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */ yading@10: z1 = MULTIPLY(d2 + d6, FIX_0_541196100); yading@10: tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065); yading@10: tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */ yading@10: tmp2 = MULTIPLY(-d6, FIX_1_306562965); yading@10: tmp3 = MULTIPLY(d6, FIX_0_541196100); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } yading@10: } else { yading@10: if (d2) { yading@10: /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */ yading@10: tmp2 = MULTIPLY(d2, FIX_0_541196100); yading@10: tmp3 = MULTIPLY(d2, FIX_1_306562965); yading@10: yading@10: tmp0 = (d0 + d4) << CONST_BITS; yading@10: tmp1 = (d0 - d4) << CONST_BITS; yading@10: yading@10: tmp10 = tmp0 + tmp3; yading@10: tmp13 = tmp0 - tmp3; yading@10: tmp11 = tmp1 + tmp2; yading@10: tmp12 = tmp1 - tmp2; yading@10: } else { yading@10: /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */ yading@10: tmp10 = tmp13 = (d0 + d4) << CONST_BITS; yading@10: tmp11 = tmp12 = (d0 - d4) << CONST_BITS; yading@10: } yading@10: } yading@10: yading@10: /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */ yading@10: yading@10: dataptr[DCTSTRIDE*0] = tmp10 >> (CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSTRIDE*1] = tmp11 >> (CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSTRIDE*2] = tmp12 >> (CONST_BITS+PASS1_BITS+3); yading@10: dataptr[DCTSTRIDE*3] = tmp13 >> (CONST_BITS+PASS1_BITS+3); yading@10: yading@10: dataptr++; /* advance pointer to next column */ yading@10: } yading@10: } yading@10: yading@10: void ff_j_rev_dct2(DCTBLOCK data){ yading@10: int d00, d01, d10, d11; yading@10: yading@10: data[0] += 4; yading@10: d00 = data[0+0*DCTSTRIDE] + data[1+0*DCTSTRIDE]; yading@10: d01 = data[0+0*DCTSTRIDE] - data[1+0*DCTSTRIDE]; yading@10: d10 = data[0+1*DCTSTRIDE] + data[1+1*DCTSTRIDE]; yading@10: d11 = data[0+1*DCTSTRIDE] - data[1+1*DCTSTRIDE]; yading@10: yading@10: data[0+0*DCTSTRIDE]= (d00 + d10)>>3; yading@10: data[1+0*DCTSTRIDE]= (d01 + d11)>>3; yading@10: data[0+1*DCTSTRIDE]= (d00 - d10)>>3; yading@10: data[1+1*DCTSTRIDE]= (d01 - d11)>>3; yading@10: } yading@10: yading@10: void ff_j_rev_dct1(DCTBLOCK data){ yading@10: data[0] = (data[0] + 4)>>3; yading@10: } yading@10: yading@10: #undef FIX yading@10: #undef CONST_BITS