3dphotoshoot

annotate libs/libjpeg/jfdctflt.c @ 14:06dc8b9b4f89

find changesets by author, revision, files, or words in the commit message
added libimago, libjpeg and libpng
author John Tsiombikas <nuclear@member.fsf.org>
date Sun, 07 Jun 2015 17:25:49 +0300
parents
children
rev   line source
nuclear@14 1 /*
nuclear@14 2 * jfdctflt.c
nuclear@14 3 *
nuclear@14 4 * Copyright (C) 1994-1996, Thomas G. Lane.
nuclear@14 5 * This file is part of the Independent JPEG Group's software.
nuclear@14 6 * For conditions of distribution and use, see the accompanying README file.
nuclear@14 7 *
nuclear@14 8 * This file contains a floating-point implementation of the
nuclear@14 9 * forward DCT (Discrete Cosine Transform).
nuclear@14 10 *
nuclear@14 11 * This implementation should be more accurate than either of the integer
nuclear@14 12 * DCT implementations. However, it may not give the same results on all
nuclear@14 13 * machines because of differences in roundoff behavior. Speed will depend
nuclear@14 14 * on the hardware's floating point capacity.
nuclear@14 15 *
nuclear@14 16 * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
nuclear@14 17 * on each column. Direct algorithms are also available, but they are
nuclear@14 18 * much more complex and seem not to be any faster when reduced to code.
nuclear@14 19 *
nuclear@14 20 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
nuclear@14 21 * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in
nuclear@14 22 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
nuclear@14 23 * JPEG textbook (see REFERENCES section in file README). The following code
nuclear@14 24 * is based directly on figure 4-8 in P&M.
nuclear@14 25 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
nuclear@14 26 * possible to arrange the computation so that many of the multiplies are
nuclear@14 27 * simple scalings of the final outputs. These multiplies can then be
nuclear@14 28 * folded into the multiplications or divisions by the JPEG quantization
nuclear@14 29 * table entries. The AA&N method leaves only 5 multiplies and 29 adds
nuclear@14 30 * to be done in the DCT itself.
nuclear@14 31 * The primary disadvantage of this method is that with a fixed-point
nuclear@14 32 * implementation, accuracy is lost due to imprecise representation of the
nuclear@14 33 * scaled quantization values. However, that problem does not arise if
nuclear@14 34 * we use floating point arithmetic.
nuclear@14 35 */
nuclear@14 36
nuclear@14 37 #define JPEG_INTERNALS
nuclear@14 38 #include "jinclude.h"
nuclear@14 39 #include "jpeglib.h"
nuclear@14 40 #include "jdct.h" /* Private declarations for DCT subsystem */
nuclear@14 41
nuclear@14 42 #ifdef DCT_FLOAT_SUPPORTED
nuclear@14 43
nuclear@14 44
nuclear@14 45 /*
nuclear@14 46 * This module is specialized to the case DCTSIZE = 8.
nuclear@14 47 */
nuclear@14 48
nuclear@14 49 #if DCTSIZE != 8
nuclear@14 50 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
nuclear@14 51 #endif
nuclear@14 52
nuclear@14 53
nuclear@14 54 /*
nuclear@14 55 * Perform the forward DCT on one block of samples.
nuclear@14 56 */
nuclear@14 57
nuclear@14 58 GLOBAL(void)
nuclear@14 59 jpeg_fdct_float (FAST_FLOAT * data)
nuclear@14 60 {
nuclear@14 61 FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
nuclear@14 62 FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
nuclear@14 63 FAST_FLOAT z1, z2, z3, z4, z5, z11, z13;
nuclear@14 64 FAST_FLOAT *dataptr;
nuclear@14 65 int ctr;
nuclear@14 66
nuclear@14 67 /* Pass 1: process rows. */
nuclear@14 68
nuclear@14 69 dataptr = data;
nuclear@14 70 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
nuclear@14 71 tmp0 = dataptr[0] + dataptr[7];
nuclear@14 72 tmp7 = dataptr[0] - dataptr[7];
nuclear@14 73 tmp1 = dataptr[1] + dataptr[6];
nuclear@14 74 tmp6 = dataptr[1] - dataptr[6];
nuclear@14 75 tmp2 = dataptr[2] + dataptr[5];
nuclear@14 76 tmp5 = dataptr[2] - dataptr[5];
nuclear@14 77 tmp3 = dataptr[3] + dataptr[4];
nuclear@14 78 tmp4 = dataptr[3] - dataptr[4];
nuclear@14 79
nuclear@14 80 /* Even part */
nuclear@14 81
nuclear@14 82 tmp10 = tmp0 + tmp3; /* phase 2 */
nuclear@14 83 tmp13 = tmp0 - tmp3;
nuclear@14 84 tmp11 = tmp1 + tmp2;
nuclear@14 85 tmp12 = tmp1 - tmp2;
nuclear@14 86
nuclear@14 87 dataptr[0] = tmp10 + tmp11; /* phase 3 */
nuclear@14 88 dataptr[4] = tmp10 - tmp11;
nuclear@14 89
nuclear@14 90 z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
nuclear@14 91 dataptr[2] = tmp13 + z1; /* phase 5 */
nuclear@14 92 dataptr[6] = tmp13 - z1;
nuclear@14 93
nuclear@14 94 /* Odd part */
nuclear@14 95
nuclear@14 96 tmp10 = tmp4 + tmp5; /* phase 2 */
nuclear@14 97 tmp11 = tmp5 + tmp6;
nuclear@14 98 tmp12 = tmp6 + tmp7;
nuclear@14 99
nuclear@14 100 /* The rotator is modified from fig 4-8 to avoid extra negations. */
nuclear@14 101 z5 = (tmp10 - tmp12) * ((FAST_FLOAT) 0.382683433); /* c6 */
nuclear@14 102 z2 = ((FAST_FLOAT) 0.541196100) * tmp10 + z5; /* c2-c6 */
nuclear@14 103 z4 = ((FAST_FLOAT) 1.306562965) * tmp12 + z5; /* c2+c6 */
nuclear@14 104 z3 = tmp11 * ((FAST_FLOAT) 0.707106781); /* c4 */
nuclear@14 105
nuclear@14 106 z11 = tmp7 + z3; /* phase 5 */
nuclear@14 107 z13 = tmp7 - z3;
nuclear@14 108
nuclear@14 109 dataptr[5] = z13 + z2; /* phase 6 */
nuclear@14 110 dataptr[3] = z13 - z2;
nuclear@14 111 dataptr[1] = z11 + z4;
nuclear@14 112 dataptr[7] = z11 - z4;
nuclear@14 113
nuclear@14 114 dataptr += DCTSIZE; /* advance pointer to next row */
nuclear@14 115 }
nuclear@14 116
nuclear@14 117 /* Pass 2: process columns. */
nuclear@14 118
nuclear@14 119 dataptr = data;
nuclear@14 120 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
nuclear@14 121 tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
nuclear@14 122 tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
nuclear@14 123 tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
nuclear@14 124 tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
nuclear@14 125 tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
nuclear@14 126 tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
nuclear@14 127 tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
nuclear@14 128 tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
nuclear@14 129
nuclear@14 130 /* Even part */
nuclear@14 131
nuclear@14 132 tmp10 = tmp0 + tmp3; /* phase 2 */
nuclear@14 133 tmp13 = tmp0 - tmp3;
nuclear@14 134 tmp11 = tmp1 + tmp2;
nuclear@14 135 tmp12 = tmp1 - tmp2;
nuclear@14 136
nuclear@14 137 dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
nuclear@14 138 dataptr[DCTSIZE*4] = tmp10 - tmp11;
nuclear@14 139
nuclear@14 140 z1 = (tmp12 + tmp13) * ((FAST_FLOAT) 0.707106781); /* c4 */
nuclear@14 141 dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
nuclear@14 142 dataptr[DCTSIZE*6] = tmp13 - z1;
nuclear@14 143
nuclear@14 144 /* Odd part */
nuclear@14 145
nuclear@14 146 tmp10 = tmp4 + tmp5; /* phase 2 */
nuclear@14 147 tmp11 = tmp5 + tmp6;
nuclear@14 148 tmp12 = tmp6 + tmp7;
nuclear@14 149
nuclear@14 150 /* The rotator is modified from fig 4-8 to avoid extra negations. */
nuclear@14 151 z5 = (tmp10 - tmp12) * ((FAST_FLOAT) 0.382683433); /* c6 */
nuclear@14 152 z2 = ((FAST_FLOAT) 0.541196100) * tmp10 + z5; /* c2-c6 */
nuclear@14 153 z4 = ((FAST_FLOAT) 1.306562965) * tmp12 + z5; /* c2+c6 */
nuclear@14 154 z3 = tmp11 * ((FAST_FLOAT) 0.707106781); /* c4 */
nuclear@14 155
nuclear@14 156 z11 = tmp7 + z3; /* phase 5 */
nuclear@14 157 z13 = tmp7 - z3;
nuclear@14 158
nuclear@14 159 dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */
nuclear@14 160 dataptr[DCTSIZE*3] = z13 - z2;
nuclear@14 161 dataptr[DCTSIZE*1] = z11 + z4;
nuclear@14 162 dataptr[DCTSIZE*7] = z11 - z4;
nuclear@14 163
nuclear@14 164 dataptr++; /* advance pointer to next column */
nuclear@14 165 }
nuclear@14 166 }
nuclear@14 167
nuclear@14 168 #endif /* DCT_FLOAT_SUPPORTED */