annotate jfdctfst.c @ 5134:879b809b00cb libavcodec

kill one vector constant value load by the right combination of vec_splatX/vec_sl patch by Sigbjorn Skjaeret %sskjer-1 A broadpark P no%
author gpoirier
date Tue, 12 Jun 2007 21:24:25 +0000
parents d6f83e2f8804
children f7cbb7733146
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
1 /*
986e461dc072 Initial revision
glantau
parents:
diff changeset
2 * jfdctfst.c
986e461dc072 Initial revision
glantau
parents:
diff changeset
3 *
986e461dc072 Initial revision
glantau
parents:
diff changeset
4 * This file is part of the Independent JPEG Group's software.
3669
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
5 *
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
6 * The authors make NO WARRANTY or representation, either express or implied,
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
7 * with respect to this software, its quality, accuracy, merchantability, or
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
8 * fitness for a particular purpose. This software is provided "AS IS", and
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
9 * you, its user, assume the entire risk as to its quality and accuracy.
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
10 *
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
11 * This software is copyright (C) 1994-1996, Thomas G. Lane.
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
12 * All Rights Reserved except as specified below.
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
13 *
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
14 * Permission is hereby granted to use, copy, modify, and distribute this
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
15 * software (or portions thereof) for any purpose, without fee, subject to
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
16 * these conditions:
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
17 * (1) If any part of the source code for this software is distributed, then
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
18 * this README file must be included, with this copyright and no-warranty
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
19 * notice unaltered; and any additions, deletions, or changes to the original
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
20 * files must be clearly indicated in accompanying documentation.
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
21 * (2) If only executable code is distributed, then the accompanying
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
22 * documentation must state that "this software is based in part on the work
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
23 * of the Independent JPEG Group".
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
24 * (3) Permission for use of this software is granted only if the user accepts
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
25 * full responsibility for any undesirable consequences; the authors accept
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
26 * NO LIABILITY for damages of any kind.
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
27 *
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
28 * These conditions apply to any software derived from or based on the IJG
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
29 * code, not just to the unmodified library. If you use our work, you ought
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
30 * to acknowledge us.
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
31 *
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
32 * Permission is NOT granted for the use of any IJG author's name or company
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
33 * name in advertising or publicity relating to this software or products
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
34 * derived from it. This software may be referred to only as "the Independent
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
35 * JPEG Group's software".
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
36 *
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
37 * We specifically permit and encourage the use of this software as the basis
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
38 * of commercial products, provided that all warranty or liability claims are
9b98e18a1b1c Add copyright notice from the Independent JPEG Group instead of referring
diego
parents: 2979
diff changeset
39 * assumed by the product vendor.
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
40 *
986e461dc072 Initial revision
glantau
parents:
diff changeset
41 * This file contains a fast, not so accurate integer implementation of the
986e461dc072 Initial revision
glantau
parents:
diff changeset
42 * forward DCT (Discrete Cosine Transform).
986e461dc072 Initial revision
glantau
parents:
diff changeset
43 *
986e461dc072 Initial revision
glantau
parents:
diff changeset
44 * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT
986e461dc072 Initial revision
glantau
parents:
diff changeset
45 * on each column. Direct algorithms are also available, but they are
986e461dc072 Initial revision
glantau
parents:
diff changeset
46 * much more complex and seem not to be any faster when reduced to code.
986e461dc072 Initial revision
glantau
parents:
diff changeset
47 *
986e461dc072 Initial revision
glantau
parents:
diff changeset
48 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
986e461dc072 Initial revision
glantau
parents:
diff changeset
49 * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in
986e461dc072 Initial revision
glantau
parents:
diff changeset
50 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
986e461dc072 Initial revision
glantau
parents:
diff changeset
51 * JPEG textbook (see REFERENCES section in file README). The following code
986e461dc072 Initial revision
glantau
parents:
diff changeset
52 * is based directly on figure 4-8 in P&M.
986e461dc072 Initial revision
glantau
parents:
diff changeset
53 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
986e461dc072 Initial revision
glantau
parents:
diff changeset
54 * possible to arrange the computation so that many of the multiplies are
986e461dc072 Initial revision
glantau
parents:
diff changeset
55 * simple scalings of the final outputs. These multiplies can then be
986e461dc072 Initial revision
glantau
parents:
diff changeset
56 * folded into the multiplications or divisions by the JPEG quantization
986e461dc072 Initial revision
glantau
parents:
diff changeset
57 * table entries. The AA&N method leaves only 5 multiplies and 29 adds
986e461dc072 Initial revision
glantau
parents:
diff changeset
58 * to be done in the DCT itself.
986e461dc072 Initial revision
glantau
parents:
diff changeset
59 * The primary disadvantage of this method is that with fixed-point math,
986e461dc072 Initial revision
glantau
parents:
diff changeset
60 * accuracy is lost due to imprecise representation of the scaled
986e461dc072 Initial revision
glantau
parents:
diff changeset
61 * quantization values. The smaller the quantization table entry, the less
986e461dc072 Initial revision
glantau
parents:
diff changeset
62 * precise the scaled value, so this implementation does worse with high-
986e461dc072 Initial revision
glantau
parents:
diff changeset
63 * quality-setting files than with low-quality ones.
986e461dc072 Initial revision
glantau
parents:
diff changeset
64 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
65
1106
1e39f273ecd6 per file doxy
michaelni
parents: 1064
diff changeset
66 /**
1e39f273ecd6 per file doxy
michaelni
parents: 1064
diff changeset
67 * @file jfdctfst.c
1e39f273ecd6 per file doxy
michaelni
parents: 1064
diff changeset
68 * Independent JPEG Group's fast AAN dct.
1e39f273ecd6 per file doxy
michaelni
parents: 1064
diff changeset
69 */
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
70
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
71 #include <stdlib.h>
986e461dc072 Initial revision
glantau
parents:
diff changeset
72 #include <stdio.h>
986e461dc072 Initial revision
glantau
parents:
diff changeset
73 #include "common.h"
986e461dc072 Initial revision
glantau
parents:
diff changeset
74 #include "dsputil.h"
986e461dc072 Initial revision
glantau
parents:
diff changeset
75
986e461dc072 Initial revision
glantau
parents:
diff changeset
76 #define DCTSIZE 8
986e461dc072 Initial revision
glantau
parents:
diff changeset
77 #define GLOBAL(x) x
986e461dc072 Initial revision
glantau
parents:
diff changeset
78 #define RIGHT_SHIFT(x, n) ((x) >> (n))
986e461dc072 Initial revision
glantau
parents:
diff changeset
79 #define SHIFT_TEMPS
986e461dc072 Initial revision
glantau
parents:
diff changeset
80
986e461dc072 Initial revision
glantau
parents:
diff changeset
81 /*
986e461dc072 Initial revision
glantau
parents:
diff changeset
82 * This module is specialized to the case DCTSIZE = 8.
986e461dc072 Initial revision
glantau
parents:
diff changeset
83 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
84
986e461dc072 Initial revision
glantau
parents:
diff changeset
85 #if DCTSIZE != 8
986e461dc072 Initial revision
glantau
parents:
diff changeset
86 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
986e461dc072 Initial revision
glantau
parents:
diff changeset
87 #endif
986e461dc072 Initial revision
glantau
parents:
diff changeset
88
986e461dc072 Initial revision
glantau
parents:
diff changeset
89
986e461dc072 Initial revision
glantau
parents:
diff changeset
90 /* Scaling decisions are generally the same as in the LL&M algorithm;
986e461dc072 Initial revision
glantau
parents:
diff changeset
91 * see jfdctint.c for more details. However, we choose to descale
986e461dc072 Initial revision
glantau
parents:
diff changeset
92 * (right shift) multiplication products as soon as they are formed,
986e461dc072 Initial revision
glantau
parents:
diff changeset
93 * rather than carrying additional fractional bits into subsequent additions.
986e461dc072 Initial revision
glantau
parents:
diff changeset
94 * This compromises accuracy slightly, but it lets us save a few shifts.
986e461dc072 Initial revision
glantau
parents:
diff changeset
95 * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
986e461dc072 Initial revision
glantau
parents:
diff changeset
96 * everywhere except in the multiplications proper; this saves a good deal
986e461dc072 Initial revision
glantau
parents:
diff changeset
97 * of work on 16-bit-int machines.
986e461dc072 Initial revision
glantau
parents:
diff changeset
98 *
986e461dc072 Initial revision
glantau
parents:
diff changeset
99 * Again to save a few shifts, the intermediate results between pass 1 and
986e461dc072 Initial revision
glantau
parents:
diff changeset
100 * pass 2 are not upscaled, but are represented only to integral precision.
986e461dc072 Initial revision
glantau
parents:
diff changeset
101 *
986e461dc072 Initial revision
glantau
parents:
diff changeset
102 * A final compromise is to represent the multiplicative constants to only
986e461dc072 Initial revision
glantau
parents:
diff changeset
103 * 8 fractional bits, rather than 13. This saves some shifting work on some
986e461dc072 Initial revision
glantau
parents:
diff changeset
104 * machines, and may also reduce the cost of multiplication (since there
986e461dc072 Initial revision
glantau
parents:
diff changeset
105 * are fewer one-bits in the constants).
986e461dc072 Initial revision
glantau
parents:
diff changeset
106 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
107
986e461dc072 Initial revision
glantau
parents:
diff changeset
108 #define CONST_BITS 8
986e461dc072 Initial revision
glantau
parents:
diff changeset
109
986e461dc072 Initial revision
glantau
parents:
diff changeset
110
986e461dc072 Initial revision
glantau
parents:
diff changeset
111 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
986e461dc072 Initial revision
glantau
parents:
diff changeset
112 * causing a lot of useless floating-point operations at run time.
986e461dc072 Initial revision
glantau
parents:
diff changeset
113 * To get around this we use the following pre-calculated constants.
986e461dc072 Initial revision
glantau
parents:
diff changeset
114 * If you change CONST_BITS you may want to add appropriate values.
986e461dc072 Initial revision
glantau
parents:
diff changeset
115 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
986e461dc072 Initial revision
glantau
parents:
diff changeset
116 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
117
986e461dc072 Initial revision
glantau
parents:
diff changeset
118 #if CONST_BITS == 8
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
119 #define FIX_0_382683433 ((int32_t) 98) /* FIX(0.382683433) */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
120 #define FIX_0_541196100 ((int32_t) 139) /* FIX(0.541196100) */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
121 #define FIX_0_707106781 ((int32_t) 181) /* FIX(0.707106781) */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
122 #define FIX_1_306562965 ((int32_t) 334) /* FIX(1.306562965) */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
123 #else
986e461dc072 Initial revision
glantau
parents:
diff changeset
124 #define FIX_0_382683433 FIX(0.382683433)
986e461dc072 Initial revision
glantau
parents:
diff changeset
125 #define FIX_0_541196100 FIX(0.541196100)
986e461dc072 Initial revision
glantau
parents:
diff changeset
126 #define FIX_0_707106781 FIX(0.707106781)
986e461dc072 Initial revision
glantau
parents:
diff changeset
127 #define FIX_1_306562965 FIX(1.306562965)
986e461dc072 Initial revision
glantau
parents:
diff changeset
128 #endif
986e461dc072 Initial revision
glantau
parents:
diff changeset
129
986e461dc072 Initial revision
glantau
parents:
diff changeset
130
986e461dc072 Initial revision
glantau
parents:
diff changeset
131 /* We can gain a little more speed, with a further compromise in accuracy,
986e461dc072 Initial revision
glantau
parents:
diff changeset
132 * by omitting the addition in a descaling shift. This yields an incorrectly
986e461dc072 Initial revision
glantau
parents:
diff changeset
133 * rounded result half the time...
986e461dc072 Initial revision
glantau
parents:
diff changeset
134 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
135
986e461dc072 Initial revision
glantau
parents:
diff changeset
136 #ifndef USE_ACCURATE_ROUNDING
986e461dc072 Initial revision
glantau
parents:
diff changeset
137 #undef DESCALE
986e461dc072 Initial revision
glantau
parents:
diff changeset
138 #define DESCALE(x,n) RIGHT_SHIFT(x, n)
986e461dc072 Initial revision
glantau
parents:
diff changeset
139 #endif
986e461dc072 Initial revision
glantau
parents:
diff changeset
140
986e461dc072 Initial revision
glantau
parents:
diff changeset
141
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 474
diff changeset
142 /* Multiply a DCTELEM variable by an int32_t constant, and immediately
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
143 * descale to yield a DCTELEM result.
986e461dc072 Initial revision
glantau
parents:
diff changeset
144 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
145
986e461dc072 Initial revision
glantau
parents:
diff changeset
146 #define MULTIPLY(var,const) ((DCTELEM) DESCALE((var) * (const), CONST_BITS))
986e461dc072 Initial revision
glantau
parents:
diff changeset
147
4283
d6f83e2f8804 rename always_inline to av_always_inline and move to common.h
mru
parents: 3669
diff changeset
148 static av_always_inline void row_fdct(DCTELEM * data){
1589
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
149 int_fast16_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
150 int_fast16_t tmp10, tmp11, tmp12, tmp13;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
151 int_fast16_t z1, z2, z3, z4, z5, z11, z13;
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
152 DCTELEM *dataptr;
986e461dc072 Initial revision
glantau
parents:
diff changeset
153 int ctr;
986e461dc072 Initial revision
glantau
parents:
diff changeset
154 SHIFT_TEMPS
986e461dc072 Initial revision
glantau
parents:
diff changeset
155
986e461dc072 Initial revision
glantau
parents:
diff changeset
156 /* Pass 1: process rows. */
986e461dc072 Initial revision
glantau
parents:
diff changeset
157
986e461dc072 Initial revision
glantau
parents:
diff changeset
158 dataptr = data;
986e461dc072 Initial revision
glantau
parents:
diff changeset
159 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
986e461dc072 Initial revision
glantau
parents:
diff changeset
160 tmp0 = dataptr[0] + dataptr[7];
986e461dc072 Initial revision
glantau
parents:
diff changeset
161 tmp7 = dataptr[0] - dataptr[7];
986e461dc072 Initial revision
glantau
parents:
diff changeset
162 tmp1 = dataptr[1] + dataptr[6];
986e461dc072 Initial revision
glantau
parents:
diff changeset
163 tmp6 = dataptr[1] - dataptr[6];
986e461dc072 Initial revision
glantau
parents:
diff changeset
164 tmp2 = dataptr[2] + dataptr[5];
986e461dc072 Initial revision
glantau
parents:
diff changeset
165 tmp5 = dataptr[2] - dataptr[5];
986e461dc072 Initial revision
glantau
parents:
diff changeset
166 tmp3 = dataptr[3] + dataptr[4];
986e461dc072 Initial revision
glantau
parents:
diff changeset
167 tmp4 = dataptr[3] - dataptr[4];
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
168
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
169 /* Even part */
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
170
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
171 tmp10 = tmp0 + tmp3; /* phase 2 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
172 tmp13 = tmp0 - tmp3;
986e461dc072 Initial revision
glantau
parents:
diff changeset
173 tmp11 = tmp1 + tmp2;
986e461dc072 Initial revision
glantau
parents:
diff changeset
174 tmp12 = tmp1 - tmp2;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
175
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
176 dataptr[0] = tmp10 + tmp11; /* phase 3 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
177 dataptr[4] = tmp10 - tmp11;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
178
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
179 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
180 dataptr[2] = tmp13 + z1; /* phase 5 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
181 dataptr[6] = tmp13 - z1;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
182
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
183 /* Odd part */
986e461dc072 Initial revision
glantau
parents:
diff changeset
184
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
185 tmp10 = tmp4 + tmp5; /* phase 2 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
186 tmp11 = tmp5 + tmp6;
986e461dc072 Initial revision
glantau
parents:
diff changeset
187 tmp12 = tmp6 + tmp7;
986e461dc072 Initial revision
glantau
parents:
diff changeset
188
986e461dc072 Initial revision
glantau
parents:
diff changeset
189 /* The rotator is modified from fig 4-8 to avoid extra negations. */
986e461dc072 Initial revision
glantau
parents:
diff changeset
190 z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
191 z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
192 z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
193 z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
194
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
195 z11 = tmp7 + z3; /* phase 5 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
196 z13 = tmp7 - z3;
986e461dc072 Initial revision
glantau
parents:
diff changeset
197
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
198 dataptr[5] = z13 + z2; /* phase 6 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
199 dataptr[3] = z13 - z2;
986e461dc072 Initial revision
glantau
parents:
diff changeset
200 dataptr[1] = z11 + z4;
986e461dc072 Initial revision
glantau
parents:
diff changeset
201 dataptr[7] = z11 - z4;
986e461dc072 Initial revision
glantau
parents:
diff changeset
202
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
203 dataptr += DCTSIZE; /* advance pointer to next row */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
204 }
1589
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
205 }
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
206
1589
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
207 /*
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
208 * Perform the forward DCT on one block of samples.
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
209 */
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
210
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
211 GLOBAL(void)
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
212 fdct_ifast (DCTELEM * data)
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
213 {
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
214 int_fast16_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
215 int_fast16_t tmp10, tmp11, tmp12, tmp13;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
216 int_fast16_t z1, z2, z3, z4, z5, z11, z13;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
217 DCTELEM *dataptr;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
218 int ctr;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
219 SHIFT_TEMPS
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
220
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
221 row_fdct(data);
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
222
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
223 /* Pass 2: process columns. */
986e461dc072 Initial revision
glantau
parents:
diff changeset
224
986e461dc072 Initial revision
glantau
parents:
diff changeset
225 dataptr = data;
986e461dc072 Initial revision
glantau
parents:
diff changeset
226 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
986e461dc072 Initial revision
glantau
parents:
diff changeset
227 tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
986e461dc072 Initial revision
glantau
parents:
diff changeset
228 tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
986e461dc072 Initial revision
glantau
parents:
diff changeset
229 tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
986e461dc072 Initial revision
glantau
parents:
diff changeset
230 tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
986e461dc072 Initial revision
glantau
parents:
diff changeset
231 tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
986e461dc072 Initial revision
glantau
parents:
diff changeset
232 tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
986e461dc072 Initial revision
glantau
parents:
diff changeset
233 tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
986e461dc072 Initial revision
glantau
parents:
diff changeset
234 tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
235
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
236 /* Even part */
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
237
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
238 tmp10 = tmp0 + tmp3; /* phase 2 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
239 tmp13 = tmp0 - tmp3;
986e461dc072 Initial revision
glantau
parents:
diff changeset
240 tmp11 = tmp1 + tmp2;
986e461dc072 Initial revision
glantau
parents:
diff changeset
241 tmp12 = tmp1 - tmp2;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
242
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
243 dataptr[DCTSIZE*0] = tmp10 + tmp11; /* phase 3 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
244 dataptr[DCTSIZE*4] = tmp10 - tmp11;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
245
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
246 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
247 dataptr[DCTSIZE*2] = tmp13 + z1; /* phase 5 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
248 dataptr[DCTSIZE*6] = tmp13 - z1;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
249
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
250 /* Odd part */
986e461dc072 Initial revision
glantau
parents:
diff changeset
251
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
252 tmp10 = tmp4 + tmp5; /* phase 2 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
253 tmp11 = tmp5 + tmp6;
986e461dc072 Initial revision
glantau
parents:
diff changeset
254 tmp12 = tmp6 + tmp7;
986e461dc072 Initial revision
glantau
parents:
diff changeset
255
986e461dc072 Initial revision
glantau
parents:
diff changeset
256 /* The rotator is modified from fig 4-8 to avoid extra negations. */
986e461dc072 Initial revision
glantau
parents:
diff changeset
257 z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
258 z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
259 z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
260 z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
261
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
262 z11 = tmp7 + z3; /* phase 5 */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
263 z13 = tmp7 - z3;
986e461dc072 Initial revision
glantau
parents:
diff changeset
264
986e461dc072 Initial revision
glantau
parents:
diff changeset
265 dataptr[DCTSIZE*5] = z13 + z2; /* phase 6 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
266 dataptr[DCTSIZE*3] = z13 - z2;
986e461dc072 Initial revision
glantau
parents:
diff changeset
267 dataptr[DCTSIZE*1] = z11 + z4;
986e461dc072 Initial revision
glantau
parents:
diff changeset
268 dataptr[DCTSIZE*7] = z11 - z4;
986e461dc072 Initial revision
glantau
parents:
diff changeset
269
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
270 dataptr++; /* advance pointer to next column */
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
271 }
986e461dc072 Initial revision
glantau
parents:
diff changeset
272 }
440
000aeeac27a2 * started to cleanup name clashes for onetime compilation
kabi
parents: 0
diff changeset
273
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
274 /*
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
275 * Perform the forward 2-4-8 DCT on one block of samples.
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
276 */
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
277
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
278 GLOBAL(void)
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
279 fdct_ifast248 (DCTELEM * data)
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
280 {
1589
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
281 int_fast16_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
282 int_fast16_t tmp10, tmp11, tmp12, tmp13;
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
283 int_fast16_t z1;
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
284 DCTELEM *dataptr;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
285 int ctr;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
286 SHIFT_TEMPS
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
287
1589
eb26d190cf5a move identical code into its own function
michael
parents: 1571
diff changeset
288 row_fdct(data);
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
289
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
290 /* Pass 2: process columns. */
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
291
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
292 dataptr = data;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
293 for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
294 tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*1];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
295 tmp1 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
296 tmp2 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
297 tmp3 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
298 tmp4 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*1];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
299 tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
300 tmp6 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
301 tmp7 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7];
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
302
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
303 /* Even part */
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
304
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
305 tmp10 = tmp0 + tmp3;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
306 tmp11 = tmp1 + tmp2;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
307 tmp12 = tmp1 - tmp2;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
308 tmp13 = tmp0 - tmp3;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
309
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
310 dataptr[DCTSIZE*0] = tmp10 + tmp11;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
311 dataptr[DCTSIZE*4] = tmp10 - tmp11;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
312
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
313 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781);
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
314 dataptr[DCTSIZE*2] = tmp13 + z1;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
315 dataptr[DCTSIZE*6] = tmp13 - z1;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
316
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
317 tmp10 = tmp4 + tmp7;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
318 tmp11 = tmp5 + tmp6;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
319 tmp12 = tmp5 - tmp6;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
320 tmp13 = tmp4 - tmp7;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
321
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
322 dataptr[DCTSIZE*1] = tmp10 + tmp11;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
323 dataptr[DCTSIZE*5] = tmp10 - tmp11;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
324
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
325 z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781);
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
326 dataptr[DCTSIZE*3] = tmp13 + z1;
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
327 dataptr[DCTSIZE*7] = tmp13 - z1;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 1589
diff changeset
328
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
329 dataptr++; /* advance pointer to next column */
1571
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
330 }
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
331 }
aa4dc16c0f18 * adding integer/floating point AAN implementations for DCT 2-4-8
romansh
parents: 1106
diff changeset
332
440
000aeeac27a2 * started to cleanup name clashes for onetime compilation
kabi
parents: 0
diff changeset
333
000aeeac27a2 * started to cleanup name clashes for onetime compilation
kabi
parents: 0
diff changeset
334 #undef GLOBAL
000aeeac27a2 * started to cleanup name clashes for onetime compilation
kabi
parents: 0
diff changeset
335 #undef CONST_BITS
000aeeac27a2 * started to cleanup name clashes for onetime compilation
kabi
parents: 0
diff changeset
336 #undef DESCALE
000aeeac27a2 * started to cleanup name clashes for onetime compilation
kabi
parents: 0
diff changeset
337 #undef FIX_0_541196100
000aeeac27a2 * started to cleanup name clashes for onetime compilation
kabi
parents: 0
diff changeset
338 #undef FIX_1_306562965