r125: This commit was manufactured by cvs2svn to create tag 'r1_1_7-last'.
[cinelerra_cv/mob.git] / hvirtual / quicktime / libavcodec / alpha / simple_idct_alpha.c
blob5ce017740868f5a34519fbbb5df273022fe05d05
1 /*
2 * Simple IDCT (Alpha optimized)
4 * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 * based upon some outcommented c code from mpeg2dec (idct_mmx.c
21 * written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
23 * Alpha optimiziations by Måns Rullgård <mru@users.sourceforge.net>
24 * and Falk Hueffner <falk@debian.org>
27 #include "asm.h"
28 #include "../dsputil.h"
30 extern void (*put_pixels_clamped_axp_p)(const DCTELEM *block, uint8_t *pixels,
31 int line_size);
32 extern void (*add_pixels_clamped_axp_p)(const DCTELEM *block, uint8_t *pixels,
33 int line_size);
35 // cos(i * M_PI / 16) * sqrt(2) * (1 << 14)
36 // W4 is actually exactly 16384, but using 16383 works around
37 // accumulating rounding errors for some encoders
38 #define W1 ((int_fast32_t) 22725)
39 #define W2 ((int_fast32_t) 21407)
40 #define W3 ((int_fast32_t) 19266)
41 #define W4 ((int_fast32_t) 16383)
42 #define W5 ((int_fast32_t) 12873)
43 #define W6 ((int_fast32_t) 8867)
44 #define W7 ((int_fast32_t) 4520)
45 #define ROW_SHIFT 11
46 #define COL_SHIFT 20
48 /* 0: all entries 0, 1: only first entry nonzero, 2: otherwise */
49 static inline int idct_row(DCTELEM *row)
51 int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3, t;
52 uint64_t l, r;
53 l = ldq(row);
54 r = ldq(row + 4);
56 if (l == 0 && r == 0)
57 return 0;
59 a0 = W4 * sextw(l) + (1 << (ROW_SHIFT - 1));
61 if (((l & ~0xffffUL) | r) == 0) {
62 a0 >>= ROW_SHIFT;
63 a0 = (uint16_t) a0;
64 a0 |= a0 << 16;
65 a0 |= a0 << 32;
67 stq(a0, row);
68 stq(a0, row + 4);
69 return 1;
72 a1 = a0;
73 a2 = a0;
74 a3 = a0;
76 t = extwl(l, 4); /* row[2] */
77 if (t != 0) {
78 t = sextw(t);
79 a0 += W2 * t;
80 a1 += W6 * t;
81 a2 -= W6 * t;
82 a3 -= W2 * t;
85 t = extwl(r, 0); /* row[4] */
86 if (t != 0) {
87 t = sextw(t);
88 a0 += W4 * t;
89 a1 -= W4 * t;
90 a2 -= W4 * t;
91 a3 += W4 * t;
94 t = extwl(r, 4); /* row[6] */
95 if (t != 0) {
96 t = sextw(t);
97 a0 += W6 * t;
98 a1 -= W2 * t;
99 a2 += W2 * t;
100 a3 -= W6 * t;
103 t = extwl(l, 2); /* row[1] */
104 if (t != 0) {
105 t = sextw(t);
106 b0 = W1 * t;
107 b1 = W3 * t;
108 b2 = W5 * t;
109 b3 = W7 * t;
110 } else {
111 b0 = 0;
112 b1 = 0;
113 b2 = 0;
114 b3 = 0;
117 t = extwl(l, 6); /* row[3] */
118 if (t) {
119 t = sextw(t);
120 b0 += W3 * t;
121 b1 -= W7 * t;
122 b2 -= W1 * t;
123 b3 -= W5 * t;
127 t = extwl(r, 2); /* row[5] */
128 if (t) {
129 t = sextw(t);
130 b0 += W5 * t;
131 b1 -= W1 * t;
132 b2 += W7 * t;
133 b3 += W3 * t;
136 t = extwl(r, 6); /* row[7] */
137 if (t) {
138 t = sextw(t);
139 b0 += W7 * t;
140 b1 -= W5 * t;
141 b2 += W3 * t;
142 b3 -= W1 * t;
145 row[0] = (a0 + b0) >> ROW_SHIFT;
146 row[1] = (a1 + b1) >> ROW_SHIFT;
147 row[2] = (a2 + b2) >> ROW_SHIFT;
148 row[3] = (a3 + b3) >> ROW_SHIFT;
149 row[4] = (a3 - b3) >> ROW_SHIFT;
150 row[5] = (a2 - b2) >> ROW_SHIFT;
151 row[6] = (a1 - b1) >> ROW_SHIFT;
152 row[7] = (a0 - b0) >> ROW_SHIFT;
154 return 2;
157 static inline void idct_col(DCTELEM *col)
159 int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3;
161 col[0] += (1 << (COL_SHIFT - 1)) / W4;
163 a0 = W4 * col[8 * 0];
164 a1 = W4 * col[8 * 0];
165 a2 = W4 * col[8 * 0];
166 a3 = W4 * col[8 * 0];
168 if (col[8 * 2]) {
169 a0 += W2 * col[8 * 2];
170 a1 += W6 * col[8 * 2];
171 a2 -= W6 * col[8 * 2];
172 a3 -= W2 * col[8 * 2];
175 if (col[8 * 4]) {
176 a0 += W4 * col[8 * 4];
177 a1 -= W4 * col[8 * 4];
178 a2 -= W4 * col[8 * 4];
179 a3 += W4 * col[8 * 4];
182 if (col[8 * 6]) {
183 a0 += W6 * col[8 * 6];
184 a1 -= W2 * col[8 * 6];
185 a2 += W2 * col[8 * 6];
186 a3 -= W6 * col[8 * 6];
189 if (col[8 * 1]) {
190 b0 = W1 * col[8 * 1];
191 b1 = W3 * col[8 * 1];
192 b2 = W5 * col[8 * 1];
193 b3 = W7 * col[8 * 1];
194 } else {
195 b0 = 0;
196 b1 = 0;
197 b2 = 0;
198 b3 = 0;
201 if (col[8 * 3]) {
202 b0 += W3 * col[8 * 3];
203 b1 -= W7 * col[8 * 3];
204 b2 -= W1 * col[8 * 3];
205 b3 -= W5 * col[8 * 3];
208 if (col[8 * 5]) {
209 b0 += W5 * col[8 * 5];
210 b1 -= W1 * col[8 * 5];
211 b2 += W7 * col[8 * 5];
212 b3 += W3 * col[8 * 5];
215 if (col[8 * 7]) {
216 b0 += W7 * col[8 * 7];
217 b1 -= W5 * col[8 * 7];
218 b2 += W3 * col[8 * 7];
219 b3 -= W1 * col[8 * 7];
222 col[8 * 0] = (a0 + b0) >> COL_SHIFT;
223 col[8 * 7] = (a0 - b0) >> COL_SHIFT;
224 col[8 * 1] = (a1 + b1) >> COL_SHIFT;
225 col[8 * 6] = (a1 - b1) >> COL_SHIFT;
226 col[8 * 2] = (a2 + b2) >> COL_SHIFT;
227 col[8 * 5] = (a2 - b2) >> COL_SHIFT;
228 col[8 * 3] = (a3 + b3) >> COL_SHIFT;
229 col[8 * 4] = (a3 - b3) >> COL_SHIFT;
232 /* If all rows but the first one are zero after row transformation,
233 all rows will be identical after column transformation. */
234 static inline void idct_col2(DCTELEM *col)
236 int i;
237 uint64_t l, r;
238 uint64_t *lcol = (uint64_t *) col;
240 for (i = 0; i < 8; ++i) {
241 int_fast32_t a0 = col[0] + (1 << (COL_SHIFT - 1)) / W4;
243 a0 *= W4;
244 col[0] = a0 >> COL_SHIFT;
245 ++col;
248 l = lcol[0];
249 r = lcol[1];
250 lcol[ 2] = l; lcol[ 3] = r;
251 lcol[ 4] = l; lcol[ 5] = r;
252 lcol[ 6] = l; lcol[ 7] = r;
253 lcol[ 8] = l; lcol[ 9] = r;
254 lcol[10] = l; lcol[11] = r;
255 lcol[12] = l; lcol[13] = r;
256 lcol[14] = l; lcol[15] = r;
259 void simple_idct_axp(DCTELEM *block)
262 int i;
263 int rowsZero = 1; /* all rows except row 0 zero */
264 int rowsConstant = 1; /* all rows consist of a constant value */
266 for (i = 0; i < 8; i++) {
267 int sparseness = idct_row(block + 8 * i);
269 if (i > 0 && sparseness > 0)
270 rowsZero = 0;
271 if (sparseness == 2)
272 rowsConstant = 0;
275 if (rowsZero) {
276 idct_col2(block);
277 } else if (rowsConstant) {
278 uint64_t *lblock = (uint64_t *) block;
280 idct_col(block);
281 for (i = 0; i < 8; i += 2) {
282 uint64_t v = (uint16_t) block[i * 8];
283 uint64_t w = (uint16_t) block[i * 8 + 8];
285 v |= v << 16;
286 w |= w << 16;
287 v |= v << 32;
288 w |= w << 32;
289 lblock[0] = v;
290 lblock[1] = v;
291 lblock[2] = w;
292 lblock[3] = w;
293 lblock += 4;
295 } else {
296 for (i = 0; i < 8; i++)
297 idct_col(block + i);
301 void simple_idct_put_axp(uint8_t *dest, int line_size, DCTELEM *block)
303 simple_idct_axp(block);
304 put_pixels_clamped_axp_p(block, dest, line_size);
307 void simple_idct_add_axp(uint8_t *dest, int line_size, DCTELEM *block)
309 simple_idct_axp(block);
310 add_pixels_clamped_axp_p(block, dest, line_size);