4 * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 based upon some outcommented c code from mpeg2dec (idct_mmx.c
30 written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
33 #include "bit_depth_template.c"
50 #define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
51 #define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
52 #define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
53 #define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
54 #define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
55 #define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
56 #define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
62 #define MUL(a, b) MUL16(a, b)
63 #define MAC(a, b, c) MAC16(a, b, c)
79 #define MUL(a, b) ((a) * (b))
80 #define MAC(a, b, c) ((a) += (b) * (c))
84 #error "Unsupported bitdepth"
88 static inline void FUNC(idctRowCondDC)(DCTELEM *row, int extra_shift)
90 int a0, a1, a2, a3, b0, b1, b2, b3;
93 #define ROW0_MASK (0xffffLL << 48 * HAVE_BIGENDIAN)
94 if (((((uint64_t *)row)[0] & ~ROW0_MASK) | ((uint64_t *)row)[1]) == 0) {
96 if (DC_SHIFT - extra_shift > 0) {
97 temp = (row[0] << (DC_SHIFT - extra_shift)) & 0xffff;
99 temp = (row[0] >> (extra_shift - DC_SHIFT)) & 0xffff;
103 ((uint64_t *)row)[0] = temp;
104 ((uint64_t *)row)[1] = temp;
108 if (!(((uint32_t*)row)[1] |
109 ((uint32_t*)row)[2] |
110 ((uint32_t*)row)[3] |
113 if (DC_SHIFT - extra_shift > 0) {
114 temp = (row[0] << (DC_SHIFT - extra_shift)) & 0xffff;
116 temp = (row[0] >> (extra_shift - DC_SHIFT)) & 0xffff;
119 ((uint32_t*)row)[0]=((uint32_t*)row)[1] =
120 ((uint32_t*)row)[2]=((uint32_t*)row)[3] = temp;
125 a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1));
135 b0 = MUL(W1, row[1]);
137 b1 = MUL(W3, row[1]);
138 MAC(b1, -W7, row[3]);
139 b2 = MUL(W5, row[1]);
140 MAC(b2, -W1, row[3]);
141 b3 = MUL(W7, row[1]);
142 MAC(b3, -W5, row[3]);
144 if (AV_RN64A(row + 4)) {
145 a0 += W4*row[4] + W6*row[6];
146 a1 += - W4*row[4] - W2*row[6];
147 a2 += - W4*row[4] + W2*row[6];
148 a3 += W4*row[4] - W6*row[6];
153 MAC(b1, -W1, row[5]);
154 MAC(b1, -W5, row[7]);
160 MAC(b3, -W1, row[7]);
163 row[0] = (a0 + b0) >> (ROW_SHIFT + extra_shift);
164 row[7] = (a0 - b0) >> (ROW_SHIFT + extra_shift);
165 row[1] = (a1 + b1) >> (ROW_SHIFT + extra_shift);
166 row[6] = (a1 - b1) >> (ROW_SHIFT + extra_shift);
167 row[2] = (a2 + b2) >> (ROW_SHIFT + extra_shift);
168 row[5] = (a2 - b2) >> (ROW_SHIFT + extra_shift);
169 row[3] = (a3 + b3) >> (ROW_SHIFT + extra_shift);
170 row[4] = (a3 - b3) >> (ROW_SHIFT + extra_shift);
173 #define IDCT_COLS do { \
174 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4)); \
181 a2 += -W6*col[8*2]; \
182 a3 += -W2*col[8*2]; \
184 b0 = MUL(W1, col[8*1]); \
185 b1 = MUL(W3, col[8*1]); \
186 b2 = MUL(W5, col[8*1]); \
187 b3 = MUL(W7, col[8*1]); \
189 MAC(b0, W3, col[8*3]); \
190 MAC(b1, -W7, col[8*3]); \
191 MAC(b2, -W1, col[8*3]); \
192 MAC(b3, -W5, col[8*3]); \
196 a1 += -W4*col[8*4]; \
197 a2 += -W4*col[8*4]; \
202 MAC(b0, W5, col[8*5]); \
203 MAC(b1, -W1, col[8*5]); \
204 MAC(b2, W7, col[8*5]); \
205 MAC(b3, W3, col[8*5]); \
210 a1 += -W2*col[8*6]; \
212 a3 += -W6*col[8*6]; \
216 MAC(b0, W7, col[8*7]); \
217 MAC(b1, -W5, col[8*7]); \
218 MAC(b2, W3, col[8*7]); \
219 MAC(b3, -W1, col[8*7]); \
223 static inline void FUNC(idctSparseColPut)(pixel *dest, int line_size,
226 int a0, a1, a2, a3, b0, b1, b2, b3;
230 dest[0] = av_clip_pixel((a0 + b0) >> COL_SHIFT);
232 dest[0] = av_clip_pixel((a1 + b1) >> COL_SHIFT);
234 dest[0] = av_clip_pixel((a2 + b2) >> COL_SHIFT);
236 dest[0] = av_clip_pixel((a3 + b3) >> COL_SHIFT);
238 dest[0] = av_clip_pixel((a3 - b3) >> COL_SHIFT);
240 dest[0] = av_clip_pixel((a2 - b2) >> COL_SHIFT);
242 dest[0] = av_clip_pixel((a1 - b1) >> COL_SHIFT);
244 dest[0] = av_clip_pixel((a0 - b0) >> COL_SHIFT);
247 static inline void FUNC(idctSparseColAdd)(pixel *dest, int line_size,
250 int a0, a1, a2, a3, b0, b1, b2, b3;
254 dest[0] = av_clip_pixel(dest[0] + ((a0 + b0) >> COL_SHIFT));
256 dest[0] = av_clip_pixel(dest[0] + ((a1 + b1) >> COL_SHIFT));
258 dest[0] = av_clip_pixel(dest[0] + ((a2 + b2) >> COL_SHIFT));
260 dest[0] = av_clip_pixel(dest[0] + ((a3 + b3) >> COL_SHIFT));
262 dest[0] = av_clip_pixel(dest[0] + ((a3 - b3) >> COL_SHIFT));
264 dest[0] = av_clip_pixel(dest[0] + ((a2 - b2) >> COL_SHIFT));
266 dest[0] = av_clip_pixel(dest[0] + ((a1 - b1) >> COL_SHIFT));
268 dest[0] = av_clip_pixel(dest[0] + ((a0 - b0) >> COL_SHIFT));
271 static inline void FUNC(idctSparseCol)(DCTELEM *col)
273 int a0, a1, a2, a3, b0, b1, b2, b3;
277 col[0 ] = ((a0 + b0) >> COL_SHIFT);
278 col[8 ] = ((a1 + b1) >> COL_SHIFT);
279 col[16] = ((a2 + b2) >> COL_SHIFT);
280 col[24] = ((a3 + b3) >> COL_SHIFT);
281 col[32] = ((a3 - b3) >> COL_SHIFT);
282 col[40] = ((a2 - b2) >> COL_SHIFT);
283 col[48] = ((a1 - b1) >> COL_SHIFT);
284 col[56] = ((a0 - b0) >> COL_SHIFT);
287 void FUNC(ff_simple_idct_put)(uint8_t *dest_, int line_size, DCTELEM *block)
289 pixel *dest = (pixel *)dest_;
292 line_size /= sizeof(pixel);
294 for (i = 0; i < 8; i++)
295 FUNC(idctRowCondDC)(block + i*8, 0);
297 for (i = 0; i < 8; i++)
298 FUNC(idctSparseColPut)(dest + i, line_size, block + i);
301 void FUNC(ff_simple_idct_add)(uint8_t *dest_, int line_size, DCTELEM *block)
303 pixel *dest = (pixel *)dest_;
306 line_size /= sizeof(pixel);
308 for (i = 0; i < 8; i++)
309 FUNC(idctRowCondDC)(block + i*8, 0);
311 for (i = 0; i < 8; i++)
312 FUNC(idctSparseColAdd)(dest + i, line_size, block + i);
315 void FUNC(ff_simple_idct)(DCTELEM *block)
319 for (i = 0; i < 8; i++)
320 FUNC(idctRowCondDC)(block + i*8, 0);
322 for (i = 0; i < 8; i++)
323 FUNC(idctSparseCol)(block + i);