PageRenderTime 43ms CodeModel.GetById 15ms app.highlight 23ms RepoModel.GetById 1ms app.codeStats 0ms

/libavcodec/bfin/dsputil_bfin.c

http://github.com/FFmpeg/FFmpeg
C | 272 lines | 182 code | 51 blank | 39 comment | 10 complexity | 832ce531d3acceecff8e009faebcd5d2 MD5 | raw file
  1/*
  2 * BlackFin DSPUTILS
  3 *
  4 * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
  5 * Copyright (c) 2006 Michael Benjamin <michael.benjamin@analog.com>
  6 *
  7 * This file is part of FFmpeg.
  8 *
  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.
 13 *
 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.
 18 *
 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
 22 */
 23
 24#include "libavcodec/avcodec.h"
 25#include "libavcodec/dsputil.h"
 26#include "dsputil_bfin.h"
 27
 28int off;
 29
 30static void bfin_idct_add (uint8_t *dest, int line_size, DCTELEM *block)
 31{
 32    ff_bfin_idct (block);
 33    ff_bfin_add_pixels_clamped (block, dest, line_size);
 34}
 35
 36static void bfin_idct_put (uint8_t *dest, int line_size, DCTELEM *block)
 37{
 38    ff_bfin_idct (block);
 39    ff_bfin_put_pixels_clamped (block, dest, line_size);
 40}
 41
 42
 43static void bfin_clear_blocks (DCTELEM *blocks)
 44{
 45    // This is just a simple memset.
 46    //
 47    __asm__("P0=192; "
 48        "I0=%0;  "
 49        "R0=0;   "
 50        "LSETUP(clear_blocks_blkfn_lab,clear_blocks_blkfn_lab)LC0=P0;"
 51        "clear_blocks_blkfn_lab:"
 52        "[I0++]=R0;"
 53        ::"a" (blocks):"P0","I0","R0");
 54}
 55
 56
 57
 58static void bfin_put_pixels8 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
 59{
 60    ff_bfin_put_pixels8uc (block, pixels, pixels, line_size, line_size, h);
 61}
 62
 63static void bfin_put_pixels8_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)
 64{
 65    ff_bfin_put_pixels8uc (block, pixels, pixels+1, line_size, line_size, h);
 66}
 67
 68static void bfin_put_pixels8_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
 69{
 70    ff_bfin_put_pixels8uc (block, pixels, pixels+line_size, line_size, line_size, h);
 71}
 72
 73static void bfin_put_pixels8_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)
 74{
 75    ff_bfin_z_put_pixels8_xy2 (block,s0,line_size, line_size, h);
 76}
 77
 78static void bfin_put_pixels16 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
 79{
 80    ff_bfin_put_pixels16uc (block, pixels, pixels, line_size, line_size, h);
 81}
 82
 83static void bfin_put_pixels16_x2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
 84{
 85    ff_bfin_put_pixels16uc (block, pixels, pixels+1, line_size, line_size, h);
 86}
 87
 88static void bfin_put_pixels16_y2 (uint8_t *block, const uint8_t *pixels, int line_size, int h)
 89{
 90    ff_bfin_put_pixels16uc (block, pixels, pixels+line_size, line_size, line_size, h);
 91}
 92
 93static void bfin_put_pixels16_xy2 (uint8_t *block, const uint8_t *s0, int line_size, int h)
 94{
 95    ff_bfin_z_put_pixels16_xy2 (block,s0,line_size, line_size, h);
 96}
 97
 98static void bfin_put_pixels8_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
 99{
100    ff_bfin_put_pixels8uc_nornd (block, pixels, pixels, line_size, h);
101}
102
103static void bfin_put_pixels8_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
104{
105    ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+1, line_size, h);
106}
107
108static void bfin_put_pixels8_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
109{
110    ff_bfin_put_pixels8uc_nornd (block, pixels, pixels+line_size, line_size, h);
111}
112
113
114static void bfin_put_pixels16_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
115{
116    ff_bfin_put_pixels16uc_nornd (block, pixels, pixels, line_size, h);
117}
118
119static void bfin_put_pixels16_x2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
120{
121    ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+1, line_size, h);
122}
123
124static void bfin_put_pixels16_y2_nornd (uint8_t *block, const uint8_t *pixels, int line_size, int h)
125{
126    ff_bfin_put_pixels16uc_nornd (block, pixels, pixels+line_size, line_size, h);
127}
128
129static int bfin_pix_abs16 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
130{
131    return ff_bfin_z_sad16x16 (blk1,blk2,line_size,line_size,h);
132}
133
134static int bfin_vsad_intra16 (void *c, uint8_t *blk1, uint8_t *dummy, int stride, int h) {
135    return ff_bfin_z_sad16x16 (blk1,blk1+stride,stride<<1,stride<<1,h);
136}
137
138static int bfin_vsad (void *c, uint8_t *blk1, uint8_t *blk2, int stride, int h) {
139    return ff_bfin_z_sad16x16 (blk1,blk1+stride,stride<<1,stride<<1,h)
140        + ff_bfin_z_sad16x16 (blk2,blk2+stride,stride<<1,stride<<1,h);
141}
142
143static uint8_t vtmp_blk[256] attribute_l1_data_b;
144
145static int bfin_pix_abs16_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
146{
147    ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+1, 16, line_size, h);
148    return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
149}
150
151static int bfin_pix_abs16_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
152{
153    ff_bfin_put_pixels16uc (vtmp_blk, blk2, blk2+line_size, 16, line_size, h);
154    return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
155}
156
157static int bfin_pix_abs16_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
158{
159    ff_bfin_z_put_pixels16_xy2 (vtmp_blk, blk2, 16, line_size, h);
160    return ff_bfin_z_sad16x16 (blk1, vtmp_blk, line_size, 16, h);
161}
162
163static int bfin_pix_abs8 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
164{
165    return ff_bfin_z_sad8x8 (blk1,blk2,line_size,line_size, h);
166}
167
168static int bfin_pix_abs8_x2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
169{
170    ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+1, 8, line_size, h);
171    return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
172}
173
174static int bfin_pix_abs8_y2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
175{
176    ff_bfin_put_pixels8uc (vtmp_blk, blk2, blk2+line_size, 8, line_size, h);
177    return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
178}
179
180static int bfin_pix_abs8_xy2 (void *c, uint8_t *blk1, uint8_t *blk2, int line_size, int h)
181{
182    ff_bfin_z_put_pixels8_xy2 (vtmp_blk, blk2, 8, line_size, h);
183    return ff_bfin_z_sad8x8 (blk1, vtmp_blk, line_size, 8, h);
184}
185
186
187/*
188  decoder optimization
189  start on 2/11 100 frames of 352x240@25 compiled with no optimization -g debugging
190  9.824s ~ 2.44x off
191  6.360s ~ 1.58x off with -O2
192  5.740s ~ 1.43x off with idcts
193
194  2.64s    2/20 same sman.mp4 decode only
195
196*/
197
198void ff_dsputil_init_bfin( DSPContext* c, AVCodecContext *avctx )
199{
200    const int high_bit_depth = avctx->bits_per_raw_sample > 8;
201
202    c->diff_pixels        = ff_bfin_diff_pixels;
203    c->put_pixels_clamped = ff_bfin_put_pixels_clamped;
204    c->add_pixels_clamped = ff_bfin_add_pixels_clamped;
205
206    if (!high_bit_depth)
207    c->get_pixels         = ff_bfin_get_pixels;
208    c->clear_blocks       = bfin_clear_blocks;
209    c->pix_sum            = ff_bfin_pix_sum;
210    c->pix_norm1          = ff_bfin_pix_norm1;
211
212    c->sad[0]             = bfin_pix_abs16;
213    c->sad[1]             = bfin_pix_abs8;
214
215/*     c->vsad[0]            = bfin_vsad; */
216/*     c->vsad[4]            = bfin_vsad_intra16; */
217
218    /* TODO [0] 16  [1] 8 */
219    c->pix_abs[0][0] = bfin_pix_abs16;
220    c->pix_abs[0][1] = bfin_pix_abs16_x2;
221    c->pix_abs[0][2] = bfin_pix_abs16_y2;
222    c->pix_abs[0][3] = bfin_pix_abs16_xy2;
223
224    c->pix_abs[1][0] = bfin_pix_abs8;
225    c->pix_abs[1][1] = bfin_pix_abs8_x2;
226    c->pix_abs[1][2] = bfin_pix_abs8_y2;
227    c->pix_abs[1][3] = bfin_pix_abs8_xy2;
228
229
230    c->sse[0] = ff_bfin_sse16;
231    c->sse[1] = ff_bfin_sse8;
232    c->sse[2] = ff_bfin_sse4;
233
234    if (!high_bit_depth) {
235    c->put_pixels_tab[0][0] = bfin_put_pixels16;
236    c->put_pixels_tab[0][1] = bfin_put_pixels16_x2;
237    c->put_pixels_tab[0][2] = bfin_put_pixels16_y2;
238    c->put_pixels_tab[0][3] = bfin_put_pixels16_xy2;
239
240    c->put_pixels_tab[1][0] = bfin_put_pixels8;
241    c->put_pixels_tab[1][1] = bfin_put_pixels8_x2;
242    c->put_pixels_tab[1][2] = bfin_put_pixels8_y2;
243    c->put_pixels_tab[1][3] = bfin_put_pixels8_xy2;
244
245    c->put_no_rnd_pixels_tab[1][0] = bfin_put_pixels8_nornd;
246    c->put_no_rnd_pixels_tab[1][1] = bfin_put_pixels8_x2_nornd;
247    c->put_no_rnd_pixels_tab[1][2] = bfin_put_pixels8_y2_nornd;
248/*     c->put_no_rnd_pixels_tab[1][3] = ff_bfin_put_pixels8_xy2_nornd; */
249
250    c->put_no_rnd_pixels_tab[0][0] = bfin_put_pixels16_nornd;
251    c->put_no_rnd_pixels_tab[0][1] = bfin_put_pixels16_x2_nornd;
252    c->put_no_rnd_pixels_tab[0][2] = bfin_put_pixels16_y2_nornd;
253/*     c->put_no_rnd_pixels_tab[0][3] = ff_bfin_put_pixels16_xy2_nornd; */
254    }
255
256    if (avctx->bits_per_raw_sample <= 8) {
257        if (avctx->dct_algo == FF_DCT_AUTO)
258            c->fdct                  = ff_bfin_fdct;
259
260        if (avctx->idct_algo == FF_IDCT_VP3) {
261            c->idct_permutation_type = FF_NO_IDCT_PERM;
262            c->idct                  = ff_bfin_vp3_idct;
263            c->idct_add              = ff_bfin_vp3_idct_add;
264            c->idct_put              = ff_bfin_vp3_idct_put;
265        } else if (avctx->idct_algo == FF_IDCT_AUTO) {
266            c->idct_permutation_type = FF_NO_IDCT_PERM;
267            c->idct                  = ff_bfin_idct;
268            c->idct_add              = bfin_idct_add;
269            c->idct_put              = bfin_idct_put;
270        }
271    }
272}