2 * Sun mediaLib optimized DSP utils
3 * Copyright (c) 2001 Fabrice Bellard
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavcodec/dsputil.h"
23 #include "libavcodec/mpegvideo.h"
25 #include <mlib_types.h>
26 #include <mlib_status.h>
28 #include <mlib_algebra.h>
29 #include <mlib_video.h>
33 static void get_pixels_mlib(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
38 mlib_VectorConvert_S16_U8_Mod((mlib_s16 *)block, (mlib_u8 *)pixels, 8);
45 static void diff_pixels_mlib(DCTELEM *restrict block, const uint8_t *s1, const uint8_t *s2, int line_size)
50 mlib_VectorSub_S16_U8_Mod((mlib_s16 *)block, (mlib_u8 *)s1, (mlib_u8 *)s2, 8);
58 static void add_pixels_clamped_mlib(const DCTELEM *block, uint8_t *pixels, int line_size)
60 mlib_VideoAddBlock_U8_S16(pixels, (mlib_s16 *)block, line_size);
63 /* put block, width 16 pixel, height 8/16 */
65 static void put_pixels16_mlib (uint8_t * dest, const uint8_t * ref,
66 int stride, int height)
70 mlib_VideoCopyRef_U8_U8_16x8(dest, (uint8_t *)ref, stride);
74 mlib_VideoCopyRef_U8_U8_16x16(dest, (uint8_t *)ref, stride);
82 static void put_pixels16_x2_mlib (uint8_t * dest, const uint8_t * ref,
83 int stride, int height)
87 mlib_VideoInterpX_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
91 mlib_VideoInterpX_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
99 static void put_pixels16_y2_mlib (uint8_t * dest, const uint8_t * ref,
100 int stride, int height)
104 mlib_VideoInterpY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
108 mlib_VideoInterpY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
116 static void put_pixels16_xy2_mlib(uint8_t * dest, const uint8_t * ref,
117 int stride, int height)
121 mlib_VideoInterpXY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
125 mlib_VideoInterpXY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
133 /* put block, width 8 pixel, height 4/8/16 */
135 static void put_pixels8_mlib (uint8_t * dest, const uint8_t * ref,
136 int stride, int height)
140 mlib_VideoCopyRef_U8_U8_8x4(dest, (uint8_t *)ref, stride);
144 mlib_VideoCopyRef_U8_U8_8x8(dest, (uint8_t *)ref, stride);
148 mlib_VideoCopyRef_U8_U8_8x16(dest, (uint8_t *)ref, stride);
156 static void put_pixels8_x2_mlib (uint8_t * dest, const uint8_t * ref,
157 int stride, int height)
161 mlib_VideoInterpX_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
165 mlib_VideoInterpX_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
169 mlib_VideoInterpX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
177 static void put_pixels8_y2_mlib (uint8_t * dest, const uint8_t * ref,
178 int stride, int height)
182 mlib_VideoInterpY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
186 mlib_VideoInterpY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
190 mlib_VideoInterpY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
198 static void put_pixels8_xy2_mlib(uint8_t * dest, const uint8_t * ref,
199 int stride, int height)
203 mlib_VideoInterpXY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
207 mlib_VideoInterpXY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
211 mlib_VideoInterpXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
219 /* average block, width 16 pixel, height 8/16 */
221 static void avg_pixels16_mlib (uint8_t * dest, const uint8_t * ref,
222 int stride, int height)
226 mlib_VideoCopyRefAve_U8_U8_16x8(dest, (uint8_t *)ref, stride);
230 mlib_VideoCopyRefAve_U8_U8_16x16(dest, (uint8_t *)ref, stride);
238 static void avg_pixels16_x2_mlib (uint8_t * dest, const uint8_t * ref,
239 int stride, int height)
243 mlib_VideoInterpAveX_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
247 mlib_VideoInterpAveX_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
255 static void avg_pixels16_y2_mlib (uint8_t * dest, const uint8_t * ref,
256 int stride, int height)
260 mlib_VideoInterpAveY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
264 mlib_VideoInterpAveY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
272 static void avg_pixels16_xy2_mlib(uint8_t * dest, const uint8_t * ref,
273 int stride, int height)
277 mlib_VideoInterpAveXY_U8_U8_16x8(dest, (uint8_t *)ref, stride, stride);
281 mlib_VideoInterpAveXY_U8_U8_16x16(dest, (uint8_t *)ref, stride, stride);
289 /* average block, width 8 pixel, height 4/8/16 */
291 static void avg_pixels8_mlib (uint8_t * dest, const uint8_t * ref,
292 int stride, int height)
296 mlib_VideoCopyRefAve_U8_U8_8x4(dest, (uint8_t *)ref, stride);
300 mlib_VideoCopyRefAve_U8_U8_8x8(dest, (uint8_t *)ref, stride);
304 mlib_VideoCopyRefAve_U8_U8_8x16(dest, (uint8_t *)ref, stride);
312 static void avg_pixels8_x2_mlib (uint8_t * dest, const uint8_t * ref,
313 int stride, int height)
317 mlib_VideoInterpAveX_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
321 mlib_VideoInterpAveX_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
325 mlib_VideoInterpAveX_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
333 static void avg_pixels8_y2_mlib (uint8_t * dest, const uint8_t * ref,
334 int stride, int height)
338 mlib_VideoInterpAveY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
342 mlib_VideoInterpAveY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
346 mlib_VideoInterpAveY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
354 static void avg_pixels8_xy2_mlib(uint8_t * dest, const uint8_t * ref,
355 int stride, int height)
359 mlib_VideoInterpAveXY_U8_U8_8x4(dest, (uint8_t *)ref, stride, stride);
363 mlib_VideoInterpAveXY_U8_U8_8x8(dest, (uint8_t *)ref, stride, stride);
367 mlib_VideoInterpAveXY_U8_U8_8x16(dest, (uint8_t *)ref, stride, stride);
375 /* swap byte order of a buffer */
377 static void bswap_buf_mlib(uint32_t *dst, const uint32_t *src, int w)
379 mlib_VectorReverseByteOrder_U32_U32(dst, src, w);
382 /* transformations */
384 static void ff_idct_put_mlib(uint8_t *dest, int line_size, DCTELEM *data)
387 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
389 mlib_VideoIDCT8x8_S16_S16 (data, data);
392 dest[0] = cm[data[0]];
393 dest[1] = cm[data[1]];
394 dest[2] = cm[data[2]];
395 dest[3] = cm[data[3]];
396 dest[4] = cm[data[4]];
397 dest[5] = cm[data[5]];
398 dest[6] = cm[data[6]];
399 dest[7] = cm[data[7]];
406 static void ff_idct_add_mlib(uint8_t *dest, int line_size, DCTELEM *data)
408 mlib_VideoIDCT8x8_S16_S16 (data, data);
409 mlib_VideoAddBlock_U8_S16(dest, (mlib_s16 *)data, line_size);
412 static void ff_idct_mlib(DCTELEM *data)
414 mlib_VideoIDCT8x8_S16_S16 (data, data);
417 static void ff_fdct_mlib(DCTELEM *data)
419 mlib_VideoDCT8x8_S16_S16 (data, data);
422 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx)
424 const int high_bit_depth = avctx->codec_id == CODEC_ID_H264 && avctx->bits_per_raw_sample > 8;
426 c->get_pixels = get_pixels_mlib;
427 c->diff_pixels = diff_pixels_mlib;
428 c->add_pixels_clamped = add_pixels_clamped_mlib;
430 if (!high_bit_depth) {
431 c->put_pixels_tab[0][0] = put_pixels16_mlib;
432 c->put_pixels_tab[0][1] = put_pixels16_x2_mlib;
433 c->put_pixels_tab[0][2] = put_pixels16_y2_mlib;
434 c->put_pixels_tab[0][3] = put_pixels16_xy2_mlib;
435 c->put_pixels_tab[1][0] = put_pixels8_mlib;
436 c->put_pixels_tab[1][1] = put_pixels8_x2_mlib;
437 c->put_pixels_tab[1][2] = put_pixels8_y2_mlib;
438 c->put_pixels_tab[1][3] = put_pixels8_xy2_mlib;
440 c->avg_pixels_tab[0][0] = avg_pixels16_mlib;
441 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mlib;
442 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mlib;
443 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mlib;
444 c->avg_pixels_tab[1][0] = avg_pixels8_mlib;
445 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mlib;
446 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mlib;
447 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mlib;
449 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mlib;
450 c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mlib;
453 c->bswap_buf = bswap_buf_mlib;
456 void MPV_common_init_mlib(MpegEncContext *s)
458 if(s->avctx->dct_algo==FF_DCT_AUTO || s->avctx->dct_algo==FF_DCT_MLIB){
459 s->dsp.fdct = ff_fdct_mlib;
462 if(s->avctx->idct_algo==FF_IDCT_MLIB){
463 s->dsp.idct_put= ff_idct_put_mlib;
464 s->dsp.idct_add= ff_idct_add_mlib;
465 s->dsp.idct = ff_idct_mlib;
466 s->dsp.idct_permutation_type= FF_NO_IDCT_PERM;