4 * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
5 * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
27 #include "dsputil_bfin.h"
32 extern void ff_bfin_idct (DCTELEM *block) attribute_l1_text;
33 extern void ff_bfin_fdct (DCTELEM *block) attribute_l1_text;
34 extern void ff_bfin_add_pixels_clamped (DCTELEM *block, uint8_t *dest, int line_size) attribute_l1_text;
35 extern void ff_bfin_put_pixels_clamped (DCTELEM *block, uint8_t *dest, int line_size) attribute_l1_text;
36 extern void ff_bfin_diff_pixels (DCTELEM *block, uint8_t *s1, uint8_t *s2, int stride) attribute_l1_text;
37 extern void ff_bfin_get_pixels (DCTELEM *restrict block, const uint8_t *pixels, int line_size) attribute_l1_text;
38 extern int ff_bfin_pix_norm1 (uint8_t * pix, int line_size) attribute_l1_text;
39 extern int ff_bfin_z_sad8x8 (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h) attribute_l1_text;
40 extern int ff_bfin_z_sad16x16 (uint8_t *blk1, uint8_t *blk2, int dsz, int line_size, int h) attribute_l1_text;
42 extern void ff_bfin_z_put_pixels16_xy2 (uint8_t *block, const uint8_t *s0, int dest_size, int line_size, int h) attribute_l1_text;
43 extern void ff_bfin_z_put_pixels8_xy2 (uint8_t *block, const uint8_t *s0, int dest_size, int line_size, int h) attribute_l1_text;
44 extern void ff_bfin_put_pixels16_xy2_nornd (uint8_t *block, const uint8_t *s0, int line_size, int h) attribute_l1_text;
45 extern void ff_bfin_put_pixels8_xy2_nornd (uint8_t *block, const uint8_t *s0, int line_size, int h) attribute_l1_text;
48 extern int ff_bfin_pix_sum (uint8_t *p, int stride) attribute_l1_text;
50 extern void ff_bfin_put_pixels8uc (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int dest_size, int line_size, int h) attribute_l1_text;
51 extern void ff_bfin_put_pixels16uc (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int dest_size, int line_size, int h) attribute_l1_text;
52 extern void ff_bfin_put_pixels8uc_nornd (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int line_size, int h) attribute_l1_text;
53 extern void ff_bfin_put_pixels16uc_nornd (uint8_t *block, const uint8_t *s0, const uint8_t *s1, int line_size, int h) attribute_l1_text;
55 extern int ff_bfin_sse4 (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) attribute_l1_text;
56 extern int ff_bfin_sse8 (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) attribute_l1_text;
57 extern int ff_bfin_sse16 (void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) attribute_l1_text;
60 static void bfin_idct_add (uint8_t *dest, int line_size, DCTELEM *block)
63 ff_bfin_add_pixels_clamped (block, dest, line_size);
66 static void bfin_idct_put (uint8_t *dest, int line_size, DCTELEM *block)
69 ff_bfin_put_pixels_clamped (block, dest, line_size);
73 static void bfin_clear_blocks (DCTELEM *blocks)
75 // This is just a simple memset.
80 "LSETUP(clear_blocks_blkfn_lab,clear_blocks_blkfn_lab)LC0=P0;"
81 "clear_blocks_blkfn_lab:"
83 ::"a" (blocks):"P0","I0","R0");
88 static void bfin_put_pixels8 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
90 ff_bfin_put_pixels8uc (block, pixels, pixels, line_size, line_size, h);
93 static void bfin_put_pixels8_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
95 ff_bfin_put_pixels8uc (block, pixels, pixels+1, line_size, line_size, h);
98 static void bfin_put_pixels8_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
100 ff_bfin_put_pixels8uc (block, pixels, pixels+line_size, line_size, line_size, h);
103 static void bfin_put_pixels8_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)
105 ff_bfin_z_put_pixels8_xy2 (block,s0,line_size, line_size, h);
108 static void bfin_put_pixels16 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
110 ff_bfin_put_pixels16uc (block, pixels, pixels, line_size, line_size, h);
113 static void bfin_put_pixels16_x2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
115 ff_bfin_put_pixels16uc (block, pixels, pixels+1, line_size, line_size, h);
118 static void bfin_put_pixels16_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
120 ff_bfin_put_pixels16uc (block, pixels, pixels+line_size, line_size, line_size, h);
123 static void bfin_put_pixels16_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)
125 ff_bfin_z_put_pixels16_xy2 (block,s0,line_size, line_size, h);
128 void bfin_put_pixels8_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
130 ff_bfin_put_pixels8uc_nornd (block, pixels, pixels, line_size, h);
133 static void bfin_put_pixels8_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
135 ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+1, line_size, h);
138 static void bfin_put_pixels8_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
140 ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+line_size, line_size, h);
144 void bfin_put_pixels16_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
146 ff_bfin_put_pixels16uc_nornd (block, pixels, pixels, line_size, h);
149 static void bfin_put_pixels16_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
151 ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+1, line_size, h);
154 static void bfin_put_pixels16_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
156 ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+line_size, line_size, h);
159 static int bfin_pix_abs16 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
161 return ff_bfin_z_sad16x16 (blk1,blk2,line_size,line_size,h);
164 static uint8_t vtmp_blk[256] __attribute__((l1_data_B));
166 static int bfin_pix_abs16_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
168 ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+1, 16, line_size, h);
169 return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
172 static int bfin_pix_abs16_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
174 ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+line_size, 16, line_size, h);
175 return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
178 static int bfin_pix_abs16_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
180 ff_bfin_z_put_pixels16_xy2 (vtmp_blk, blk2, 16, line_size, h);
181 return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
184 static int bfin_pix_abs8 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
186 return ff_bfin_z_sad8x8 (blk1,blk2,line_size,line_size, h);
189 static int bfin_pix_abs8_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
191 ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+1, 8, line_size, h);
192 return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
195 static int bfin_pix_abs8_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
197 ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+line_size, 8, line_size, h);
198 return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
201 static int bfin_pix_abs8_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
203 ff_bfin_z_put_pixels8_xy2 (vtmp_blk, blk2, 8, line_size, h);
204 return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
210 start on 2/11 100 frames of 352x240@25 compiled with no optimization -g debugging
212 6.360s ~ 1.58x off with -O2
213 5.740s ~ 1.43x off with idcts
215 2.64s 2/20 same sman.mp4 decode only
219 void dsputil_init_bfin( DSPContext* c, AVCodecContext *avctx )
221 c->get_pixels = ff_bfin_get_pixels;
222 c->diff_pixels = ff_bfin_diff_pixels;
223 c->put_pixels_clamped = ff_bfin_put_pixels_clamped;
224 c->add_pixels_clamped = ff_bfin_add_pixels_clamped;
226 c->clear_blocks = bfin_clear_blocks;
227 c->pix_sum = ff_bfin_pix_sum;
228 c->pix_norm1 = ff_bfin_pix_norm1;
230 c->sad[0] = bfin_pix_abs16;
231 c->sad[1] = bfin_pix_abs8;
233 /* TODO [0] 16 [1] 8 */
234 c->pix_abs[0][0] = bfin_pix_abs16;
235 c->pix_abs[0][1] = bfin_pix_abs16_x2;
236 c->pix_abs[0][2] = bfin_pix_abs16_y2;
237 c->pix_abs[0][3] = bfin_pix_abs16_xy2;
239 c->pix_abs[1][0] = bfin_pix_abs8;
240 c->pix_abs[1][1] = bfin_pix_abs8_x2;
241 c->pix_abs[1][2] = bfin_pix_abs8_y2;
242 c->pix_abs[1][3] = bfin_pix_abs8_xy2;
245 c->sse[0] = ff_bfin_sse16;
246 c->sse[1] = ff_bfin_sse8;
247 c->sse[2] = ff_bfin_sse4;
251 * Halfpel motion compensation with rounding (a+b+1)>>1.
252 * This is an array[4][4] of motion compensation functions for 4
253 * horizontal blocksizes (8,16) and the 4 halfpel positions
254 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
255 * @param block destination where the result is stored
256 * @param pixels source
257 * @param line_size number of bytes in a horizontal line of block
261 c->put_pixels_tab[0][0] = bfin_put_pixels16;
262 c->put_pixels_tab[0][1] = bfin_put_pixels16_x2;
263 c->put_pixels_tab[0][2] = bfin_put_pixels16_y2;
264 c->put_pixels_tab[0][3] = bfin_put_pixels16_xy2;
266 c->put_pixels_tab[1][0] = bfin_put_pixels8;
267 c->put_pixels_tab[1][1] = bfin_put_pixels8_x2;
268 c->put_pixels_tab[1][2] = bfin_put_pixels8_y2;
269 c->put_pixels_tab[1][3] = bfin_put_pixels8_xy2;
271 c->put_no_rnd_pixels_tab[1][0] = bfin_put_pixels8_nornd;
272 c->put_no_rnd_pixels_tab[1][1] = bfin_put_pixels8_x2_nornd;
273 c->put_no_rnd_pixels_tab[1][2] = bfin_put_pixels8_y2_nornd;
274 c->put_no_rnd_pixels_tab[1][3] = ff_bfin_put_pixels8_xy2_nornd;
276 c->put_no_rnd_pixels_tab[0][0] = bfin_put_pixels16_nornd;
277 c->put_no_rnd_pixels_tab[0][1] = bfin_put_pixels16_x2_nornd;
278 c->put_no_rnd_pixels_tab[0][2] = bfin_put_pixels16_y2_nornd;
279 c->put_no_rnd_pixels_tab[0][3] = ff_bfin_put_pixels16_xy2_nornd;
281 c->idct_permutation_type = FF_NO_IDCT_PERM;
282 c->fdct = ff_bfin_fdct;
283 if (avctx->idct_algo!=FF_IDCT_VP3) {
284 c->idct = ff_bfin_idct;
285 c->idct_add = bfin_idct_add;
286 c->idct_put = bfin_idct_put;