2
* Sun mediaLib optimized DSP utils
3
* Copyright (c) 2001 Fabrice Bellard.
5
* This library is free software; you can redistribute it and/or
6
* modify it under the terms of the GNU Lesser General Public
7
* License as published by the Free Software Foundation; either
8
* version 2 of the License, or (at your option) any later version.
10
* This library is distributed in the hope that it will be useful,
11
* but WITHOUT ANY WARRANTY; without even the implied warranty of
12
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13
* Lesser General Public License for more details.
15
* You should have received a copy of the GNU Lesser General Public
16
* License along with this library; if not, write to the Free Software
17
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20
#include "../dsputil.h"
21
#include "../mpegvideo.h"
23
#include <mlib_types.h>
24
#include <mlib_status.h>
26
#include <mlib_algebra.h>
27
#include <mlib_video.h>
31
static void get_pixels_mlib(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
36
mlib_VectorConvert_S16_U8_Mod((mlib_s16 *)block, (mlib_u8 *)pixels, 8);
43
static void diff_pixels_mlib(DCTELEM *restrict block, const uint8_t *s1, const uint8_t *s2, int line_size)
48
mlib_VectorSub_S16_U8_Mod((mlib_s16 *)block, (mlib_u8 *)s1, (mlib_u8 *)s2, 8);
56
static void add_pixels_clamped_mlib(const DCTELEM *block, uint8_t *pixels, int line_size)
58
mlib_VideoAddBlock_U8_S16(pixels, (mlib_s16 *)block, line_size);
61
/* put block, width 16 pixel, height 8/16 */
63
static void put_pixels16_mlib (uint8_t * dest, const uint8_t * ref,
64
int stride, int height)
68
mlib_VideoCopyRef_U8_U8_16x8(dest, (uint8_t *)ref, stride);
72
mlib_VideoCopyRef_U8_U8_16x16(dest, (uint8_t *)ref, stride);
80
static void put_pixels16_x2_mlib (uint8_t * dest, const uint8_t * ref,
81
int stride, int height)
85
mlib_VideoInterpX_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
89
mlib_VideoInterpX_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
97
static void put_pixels16_y2_mlib (uint8_t * dest, const uint8_t * ref,
98
int stride, int height)
102
mlib_VideoInterpY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
106
mlib_VideoInterpY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
114
static void put_pixels16_xy2_mlib(uint8_t * dest, const uint8_t * ref,
115
int stride, int height)
119
mlib_VideoInterpXY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
123
mlib_VideoInterpXY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
131
/* put block, width 8 pixel, height 4/8/16 */
133
static void put_pixels8_mlib (uint8_t * dest, const uint8_t * ref,
134
int stride, int height)
138
mlib_VideoCopyRef_U8_U8_8x4(dest, (uint8_t *)ref, stride);
142
mlib_VideoCopyRef_U8_U8_8x8(dest, (uint8_t *)ref, stride);
146
mlib_VideoCopyRef_U8_U8_8x16(dest, (uint8_t *)ref, stride);
154
static void put_pixels8_x2_mlib (uint8_t * dest, const uint8_t * ref,
155
int stride, int height)
159
mlib_VideoInterpX_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
163
mlib_VideoInterpX_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
167
mlib_VideoInterpX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
175
static void put_pixels8_y2_mlib (uint8_t * dest, const uint8_t * ref,
176
int stride, int height)
180
mlib_VideoInterpY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
184
mlib_VideoInterpY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
188
mlib_VideoInterpY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
196
static void put_pixels8_xy2_mlib(uint8_t * dest, const uint8_t * ref,
197
int stride, int height)
201
mlib_VideoInterpXY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
205
mlib_VideoInterpXY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
209
mlib_VideoInterpXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
217
/* average block, width 16 pixel, height 8/16 */
219
static void avg_pixels16_mlib (uint8_t * dest, const uint8_t * ref,
220
int stride, int height)
224
mlib_VideoCopyRefAve_U8_U8_16x8(dest, (uint8_t *)ref, stride);
228
mlib_VideoCopyRefAve_U8_U8_16x16(dest, (uint8_t *)ref, stride);
236
static void avg_pixels16_x2_mlib (uint8_t * dest, const uint8_t * ref,
237
int stride, int height)
241
mlib_VideoInterpAveX_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
245
mlib_VideoInterpAveX_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
253
static void avg_pixels16_y2_mlib (uint8_t * dest, const uint8_t * ref,
254
int stride, int height)
258
mlib_VideoInterpAveY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
262
mlib_VideoInterpAveY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
270
static void avg_pixels16_xy2_mlib(uint8_t * dest, const uint8_t * ref,
271
int stride, int height)
275
mlib_VideoInterpAveXY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
279
mlib_VideoInterpAveXY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
287
/* average block, width 8 pixel, height 4/8/16 */
289
static void avg_pixels8_mlib (uint8_t * dest, const uint8_t * ref,
290
int stride, int height)
294
mlib_VideoCopyRefAve_U8_U8_8x4(dest, (uint8_t *)ref, stride);
298
mlib_VideoCopyRefAve_U8_U8_8x8(dest, (uint8_t *)ref, stride);
302
mlib_VideoCopyRefAve_U8_U8_8x16(dest, (uint8_t *)ref, stride);
310
static void avg_pixels8_x2_mlib (uint8_t * dest, const uint8_t * ref,
311
int stride, int height)
315
mlib_VideoInterpAveX_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
319
mlib_VideoInterpAveX_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
323
mlib_VideoInterpAveX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
331
static void avg_pixels8_y2_mlib (uint8_t * dest, const uint8_t * ref,
332
int stride, int height)
336
mlib_VideoInterpAveY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
340
mlib_VideoInterpAveY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
344
mlib_VideoInterpAveY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
352
static void avg_pixels8_xy2_mlib(uint8_t * dest, const uint8_t * ref,
353
int stride, int height)
357
mlib_VideoInterpAveXY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
361
mlib_VideoInterpAveXY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
365
mlib_VideoInterpAveXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
373
/* swap byte order of a buffer */
375
static void bswap_buf_mlib(uint32_t *dst, uint32_t *src, int w)
377
mlib_VectorReverseByteOrder_U32_U32(dst, src, w);
380
/* transformations */
382
static void ff_idct_put_mlib(uint8_t *dest, int line_size, DCTELEM *data)
385
uint8_t *cm = cropTbl + MAX_NEG_CROP;
387
mlib_VideoIDCT8x8_S16_S16 (data, data);
390
dest[0] = cm[data[0]];
391
dest[1] = cm[data[1]];
392
dest[2] = cm[data[2]];
393
dest[3] = cm[data[3]];
394
dest[4] = cm[data[4]];
395
dest[5] = cm[data[5]];
396
dest[6] = cm[data[6]];
397
dest[7] = cm[data[7]];
404
static void ff_idct_add_mlib(uint8_t *dest, int line_size, DCTELEM *data)
406
mlib_VideoIDCT8x8_S16_S16 (data, data);
407
mlib_VideoAddBlock_U8_S16(dest, (mlib_s16 *)data, line_size);
410
static void ff_idct_mlib(DCTELEM *data)
412
mlib_VideoIDCT8x8_S16_S16 (data, data);
415
static void ff_fdct_mlib(DCTELEM *data)
417
mlib_VideoDCT8x8_S16_S16 (data, data);
420
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx)
422
c->get_pixels = get_pixels_mlib;
423
c->diff_pixels = diff_pixels_mlib;
424
c->add_pixels_clamped = add_pixels_clamped_mlib;
426
c->put_pixels_tab[0][0] = put_pixels16_mlib;
427
c->put_pixels_tab[0][1] = put_pixels16_x2_mlib;
428
c->put_pixels_tab[0][2] = put_pixels16_y2_mlib;
429
c->put_pixels_tab[0][3] = put_pixels16_xy2_mlib;
430
c->put_pixels_tab[1][0] = put_pixels8_mlib;
431
c->put_pixels_tab[1][1] = put_pixels8_x2_mlib;
432
c->put_pixels_tab[1][2] = put_pixels8_y2_mlib;
433
c->put_pixels_tab[1][3] = put_pixels8_xy2_mlib;
435
c->avg_pixels_tab[0][0] = avg_pixels16_mlib;
436
c->avg_pixels_tab[0][1] = avg_pixels16_x2_mlib;
437
c->avg_pixels_tab[0][2] = avg_pixels16_y2_mlib;
438
c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mlib;
439
c->avg_pixels_tab[1][0] = avg_pixels8_mlib;
440
c->avg_pixels_tab[1][1] = avg_pixels8_x2_mlib;
441
c->avg_pixels_tab[1][2] = avg_pixels8_y2_mlib;
442
c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mlib;
444
c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mlib;
445
c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mlib;
447
c->bswap_buf = bswap_buf_mlib;
450
void MPV_common_init_mlib(MpegEncContext *s)
452
if(s->avctx->dct_algo==FF_DCT_AUTO || s->avctx->dct_algo==FF_DCT_MLIB){
453
s->dsp.fdct = ff_fdct_mlib;
456
if(s->avctx->idct_algo==FF_IDCT_AUTO || s->avctx->idct_algo==FF_IDCT_MLIB){
457
s->dsp.idct_put= ff_idct_put_mlib;
458
s->dsp.idct_add= ff_idct_add_mlib;
459
s->dsp.idct = ff_idct_mlib;
460
s->dsp.idct_permutation_type= FF_NO_IDCT_PERM;