2 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/log.h"
22 #include "libavutil/thread.h"
29 void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
30 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
34 //FIXME ugly misuse of obmc_stride
35 const uint8_t *obmc1= obmc + y*obmc_stride;
36 const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
37 const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
38 const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
39 dst = slice_buffer_get_line(sb, src_y + y);
41 int v= obmc1[x] * block[3][x + y*src_stride]
42 +obmc2[x] * block[2][x + y*src_stride]
43 +obmc3[x] * block[1][x + y*src_stride]
44 +obmc4[x] * block[0][x + y*src_stride];
46 v <<= 8 - LOG2_OBMC_MAX;
52 v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
53 if(v&(~255)) v= ~(v>>31);
54 dst8[x + y*src_stride] = v;
62 void ff_snow_reset_contexts(SnowContext *s){ //FIXME better initial contexts
63 int plane_index, level, orientation;
65 for(plane_index=0; plane_index<3; plane_index++){
66 for(level=0; level<MAX_DECOMPOSITIONS; level++){
67 for(orientation=level ? 1:0; orientation<4; orientation++){
68 memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state));
72 memset(s->header_state, MID_STATE, sizeof(s->header_state));
73 memset(s->block_state, MID_STATE, sizeof(s->block_state));
76 int ff_snow_alloc_blocks(SnowContext *s){
77 int w= AV_CEIL_RSHIFT(s->avctx->width, LOG2_MB_SIZE);
78 int h= AV_CEIL_RSHIFT(s->avctx->height, LOG2_MB_SIZE);
84 s->block = av_calloc(w * h, sizeof(*s->block) << (s->block_max_depth*2));
86 return AVERROR(ENOMEM);
91 static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, int stride, int b_w, int b_h, int dx, int dy){
92 static const uint8_t weight[64]={
103 static const uint8_t brane[256]={
104 0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x11,0x12,0x12,0x12,0x12,0x12,0x12,0x12,
105 0x04,0x05,0xcc,0xcc,0xcc,0xcc,0xcc,0x41,0x15,0x16,0xcc,0xcc,0xcc,0xcc,0xcc,0x52,
106 0x04,0xcc,0x05,0xcc,0xcc,0xcc,0x41,0xcc,0x15,0xcc,0x16,0xcc,0xcc,0xcc,0x52,0xcc,
107 0x04,0xcc,0xcc,0x05,0xcc,0x41,0xcc,0xcc,0x15,0xcc,0xcc,0x16,0xcc,0x52,0xcc,0xcc,
108 0x04,0xcc,0xcc,0xcc,0x41,0xcc,0xcc,0xcc,0x15,0xcc,0xcc,0xcc,0x16,0xcc,0xcc,0xcc,
109 0x04,0xcc,0xcc,0x41,0xcc,0x05,0xcc,0xcc,0x15,0xcc,0xcc,0x52,0xcc,0x16,0xcc,0xcc,
110 0x04,0xcc,0x41,0xcc,0xcc,0xcc,0x05,0xcc,0x15,0xcc,0x52,0xcc,0xcc,0xcc,0x16,0xcc,
111 0x04,0x41,0xcc,0xcc,0xcc,0xcc,0xcc,0x05,0x15,0x52,0xcc,0xcc,0xcc,0xcc,0xcc,0x16,
112 0x44,0x45,0x45,0x45,0x45,0x45,0x45,0x45,0x55,0x56,0x56,0x56,0x56,0x56,0x56,0x56,
113 0x48,0x49,0xcc,0xcc,0xcc,0xcc,0xcc,0x85,0x59,0x5A,0xcc,0xcc,0xcc,0xcc,0xcc,0x96,
114 0x48,0xcc,0x49,0xcc,0xcc,0xcc,0x85,0xcc,0x59,0xcc,0x5A,0xcc,0xcc,0xcc,0x96,0xcc,
115 0x48,0xcc,0xcc,0x49,0xcc,0x85,0xcc,0xcc,0x59,0xcc,0xcc,0x5A,0xcc,0x96,0xcc,0xcc,
116 0x48,0xcc,0xcc,0xcc,0x49,0xcc,0xcc,0xcc,0x59,0xcc,0xcc,0xcc,0x96,0xcc,0xcc,0xcc,
117 0x48,0xcc,0xcc,0x85,0xcc,0x49,0xcc,0xcc,0x59,0xcc,0xcc,0x96,0xcc,0x5A,0xcc,0xcc,
118 0x48,0xcc,0x85,0xcc,0xcc,0xcc,0x49,0xcc,0x59,0xcc,0x96,0xcc,0xcc,0xcc,0x5A,0xcc,
119 0x48,0x85,0xcc,0xcc,0xcc,0xcc,0xcc,0x49,0x59,0x96,0xcc,0xcc,0xcc,0xcc,0xcc,0x5A,
122 static const uint8_t needs[16]={
130 int16_t tmpIt [64*(32+HTAPS_MAX)];
131 uint8_t tmp2t[3][64*(32+HTAPS_MAX)];
132 int16_t *tmpI= tmpIt;
133 uint8_t *tmp2= tmp2t[0];
134 const uint8_t *hpel[11];
135 av_assert2(dx<16 && dy<16);
136 r= brane[dx + 16*dy]&15;
137 l= brane[dx + 16*dy]>>4;
139 b= needs[l] | needs[r];
144 for(y=0; y < b_h+HTAPS_MAX-1; y++){
145 for(x=0; x < b_w; x++){
146 int a_1=src[x + HTAPS_MAX/2-4];
147 int a0= src[x + HTAPS_MAX/2-3];
148 int a1= src[x + HTAPS_MAX/2-2];
149 int a2= src[x + HTAPS_MAX/2-1];
150 int a3= src[x + HTAPS_MAX/2+0];
151 int a4= src[x + HTAPS_MAX/2+1];
152 int a5= src[x + HTAPS_MAX/2+2];
153 int a6= src[x + HTAPS_MAX/2+3];
155 if(!p || p->fast_mc){
156 am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
160 am= p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6);
165 if(am&(~255)) am= ~(am>>31);
174 src += HTAPS_MAX/2 - 1;
178 for(y=0; y < b_h; y++){
179 for(x=0; x < b_w+1; x++){
180 int a_1=src[x + (HTAPS_MAX/2-4)*stride];
181 int a0= src[x + (HTAPS_MAX/2-3)*stride];
182 int a1= src[x + (HTAPS_MAX/2-2)*stride];
183 int a2= src[x + (HTAPS_MAX/2-1)*stride];
184 int a3= src[x + (HTAPS_MAX/2+0)*stride];
185 int a4= src[x + (HTAPS_MAX/2+1)*stride];
186 int a5= src[x + (HTAPS_MAX/2+2)*stride];
187 int a6= src[x + (HTAPS_MAX/2+3)*stride];
190 am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
192 am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;
194 if(am&(~255)) am= ~(am>>31);
202 src += stride*(HTAPS_MAX/2 - 1);
206 for(y=0; y < b_h; y++){
207 for(x=0; x < b_w; x++){
208 int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
209 int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
210 int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
211 int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
212 int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
213 int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
214 int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
215 int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
218 am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
220 am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;
221 if(am&(~255)) am= ~(am>>31);
230 hpel[ 1]= tmp2t[0] + 64*(HTAPS_MAX/2-1);
235 hpel[ 6]= tmp2t[1] + 1;
237 hpel[ 8]= src + stride;
238 hpel[ 9]= hpel[1] + 64;
239 hpel[10]= hpel[8] + 1;
241 #define MC_STRIDE(x) (needs[x] ? 64 : stride)
244 int dxy = dx / 8 + dy / 8 * 4;
245 const uint8_t *src1 = hpel[dxy ];
246 const uint8_t *src2 = hpel[dxy + 1];
247 const uint8_t *src3 = hpel[dxy + 4];
248 const uint8_t *src4 = hpel[dxy + 5];
249 int stride1 = MC_STRIDE(dxy);
250 int stride2 = MC_STRIDE(dxy + 1);
251 int stride3 = MC_STRIDE(dxy + 4);
252 int stride4 = MC_STRIDE(dxy + 5);
255 for(y=0; y < b_h; y++){
256 for(x=0; x < b_w; x++){
257 dst[x]= ((8-dx)*(8-dy)*src1[x] + dx*(8-dy)*src2[x]+
258 (8-dx)* dy *src3[x] + dx* dy *src4[x]+32)>>6;
267 const uint8_t *src1= hpel[l];
268 const uint8_t *src2= hpel[r];
269 int stride1 = MC_STRIDE(l);
270 int stride2 = MC_STRIDE(r);
271 int a= weight[((dx&7) + (8*(dy&7)))];
273 for(y=0; y < b_h; y++){
274 for(x=0; x < b_w; x++){
275 dst[x]= (a*src1[x] + b*src2[x] + 4)>>3;
284 void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t stride, int sx, int sy, int b_w, int b_h, const BlockNode *block, int plane_index, int w, int h){
285 if(block->type & BLOCK_INTRA){
287 const unsigned color = block->color[plane_index];
288 const unsigned color4 = color*0x01010101;
290 for(y=0; y < b_h; y++){
291 *(uint32_t*)&dst[0 + y*stride]= color4;
292 *(uint32_t*)&dst[4 + y*stride]= color4;
293 *(uint32_t*)&dst[8 + y*stride]= color4;
294 *(uint32_t*)&dst[12+ y*stride]= color4;
295 *(uint32_t*)&dst[16+ y*stride]= color4;
296 *(uint32_t*)&dst[20+ y*stride]= color4;
297 *(uint32_t*)&dst[24+ y*stride]= color4;
298 *(uint32_t*)&dst[28+ y*stride]= color4;
301 for(y=0; y < b_h; y++){
302 *(uint32_t*)&dst[0 + y*stride]= color4;
303 *(uint32_t*)&dst[4 + y*stride]= color4;
304 *(uint32_t*)&dst[8 + y*stride]= color4;
305 *(uint32_t*)&dst[12+ y*stride]= color4;
308 for(y=0; y < b_h; y++){
309 *(uint32_t*)&dst[0 + y*stride]= color4;
310 *(uint32_t*)&dst[4 + y*stride]= color4;
313 for(y=0; y < b_h; y++){
314 *(uint32_t*)&dst[0 + y*stride]= color4;
317 for(y=0; y < b_h; y++){
318 for(x=0; x < b_w; x++){
319 dst[x + y*stride]= color;
324 const uint8_t *src = s->last_picture[block->ref]->data[plane_index];
325 const int scale= plane_index ? (2*s->mv_scale)>>s->chroma_h_shift : 2*s->mv_scale;
326 int mx= block->mx*scale;
327 int my= block->my*scale;
330 const int tab_index= 3 - (b_w>>2) + (b_w>>4);
331 sx += (mx>>4) - (HTAPS_MAX/2-1);
332 sy += (my>>4) - (HTAPS_MAX/2-1);
333 src += sx + sy*stride;
334 if( (unsigned)sx >= FFMAX(w - b_w - (HTAPS_MAX-2), 0)
335 || (unsigned)sy >= FFMAX(h - b_h - (HTAPS_MAX-2), 0)){
336 s->vdsp.emulated_edge_mc(tmp + MB_SIZE, src,
338 b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1,
343 av_assert2(s->chroma_h_shift == s->chroma_v_shift); // only one mv_scale
345 av_assert2((tab_index>=0 && tab_index<4) || b_w==32);
347 || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h)
351 || !s->plane[plane_index].fast_mc )
352 mc_block(&s->plane[plane_index], dst, src, stride, b_w, b_h, dx, dy);
355 for(y=0; y<b_h; y+=16){
356 s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 3 + (y+3)*stride,stride);
357 s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 19 + (y+3)*stride,stride);
360 s->h264qpel.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst,src + 3 + 3*stride,stride);
362 s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst ,src + 3 + 3*stride,stride);
363 s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 3 + b_h + 3*stride,stride);
365 av_assert2(2*b_w==b_h);
366 s->h264qpel.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst ,src + 3 + 3*stride ,stride);
367 s->h264qpel.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst+b_w*stride,src + 3 + 3*stride+b_w*stride,stride);
372 #define mca(dx,dy,b_w)\
373 static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, ptrdiff_t stride, int h){\
375 mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, stride, b_w, b_w, dx, dy);\
387 static av_cold void snow_static_init(void)
389 for (int i = 0; i < MAX_REF_FRAMES; i++)
390 for (int j = 0; j < MAX_REF_FRAMES; j++)
391 ff_scale_mv_ref[i][j] = 256 * (i + 1) / (j + 1);
394 av_cold int ff_snow_common_init(AVCodecContext *avctx){
395 static AVOnce init_static_once = AV_ONCE_INIT;
396 SnowContext *s = avctx->priv_data;
401 s->max_ref_frames=1; //just make sure it's not an invalid value in case of no initial keyframe
402 s->spatial_decomposition_count = 1;
404 ff_videodsp_init(&s->vdsp, 8);
405 ff_dwt_init(&s->dwt);
406 ff_h264qpel_init(&s->h264qpel, 8);
409 s->hdsp.put_pixels_tab [0][dy/4+dx/8]=\
410 s->hdsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\
411 mc_block_hpel ## dx ## dy ## 16;\
412 s->hdsp.put_pixels_tab [1][dy/4+dx/8]=\
413 s->hdsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\
414 mc_block_hpel ## dx ## dy ## 8;
421 // dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);
423 width= s->avctx->width;
424 height= s->avctx->height;
426 if (!FF_ALLOCZ_TYPED_ARRAY(s->spatial_idwt_buffer, width * height) ||
427 !FF_ALLOCZ_TYPED_ARRAY(s->spatial_dwt_buffer, width * height) || //FIXME this does not belong here
428 !FF_ALLOCZ_TYPED_ARRAY(s->temp_dwt_buffer, width) ||
429 !FF_ALLOCZ_TYPED_ARRAY(s->temp_idwt_buffer, width) ||
430 !FF_ALLOCZ_TYPED_ARRAY(s->run_buffer, ((width + 1) >> 1) * ((height + 1) >> 1)))
431 return AVERROR(ENOMEM);
433 for(i=0; i<MAX_REF_FRAMES; i++) {
434 s->last_picture[i] = av_frame_alloc();
435 if (!s->last_picture[i])
436 return AVERROR(ENOMEM);
439 s->mconly_picture = av_frame_alloc();
440 s->current_picture = av_frame_alloc();
441 if (!s->mconly_picture || !s->current_picture)
442 return AVERROR(ENOMEM);
444 ff_thread_once(&init_static_once, snow_static_init);
449 int ff_snow_common_init_after_header(AVCodecContext *avctx) {
450 SnowContext *s = avctx->priv_data;
451 int plane_index, level, orientation;
455 emu_buf_size = FFMAX(s->mconly_picture->linesize[0], 2*avctx->width+256) * (2 * MB_SIZE + HTAPS_MAX - 1);
456 if (!FF_ALLOCZ_TYPED_ARRAY(s->scratchbuf, FFMAX(s->mconly_picture->linesize[0], 2*avctx->width+256) * 7 * MB_SIZE) ||
457 !FF_ALLOCZ_TYPED_ARRAY(s->emu_edge_buffer, emu_buf_size))
458 return AVERROR(ENOMEM);
461 for(plane_index=0; plane_index < s->nb_planes; plane_index++){
462 int w= s->avctx->width;
463 int h= s->avctx->height;
466 w = AV_CEIL_RSHIFT(w, s->chroma_h_shift);
467 h = AV_CEIL_RSHIFT(h, s->chroma_v_shift);
469 s->plane[plane_index].width = w;
470 s->plane[plane_index].height= h;
472 for(level=s->spatial_decomposition_count-1; level>=0; level--){
473 for(orientation=level ? 1 : 0; orientation<4; orientation++){
474 SubBand *b= &s->plane[plane_index].band[level][orientation];
476 b->buf= s->spatial_dwt_buffer;
478 b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
479 b->width = (w + !(orientation&1))>>1;
480 b->height= (h + !(orientation>1))>>1;
482 b->stride_line = 1 << (s->spatial_decomposition_count - level);
488 b->buf_x_offset = (w+1)>>1;
491 b->buf += b->stride>>1;
492 b->buf_y_offset = b->stride_line >> 1;
494 b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);
497 b->parent= &s->plane[plane_index].band[level-1][orientation];
498 //FIXME avoid this realloc
499 av_freep(&b->x_coeff);
500 b->x_coeff = av_calloc((b->width + 1) * b->height + 1,
501 sizeof(*b->x_coeff));
503 return AVERROR(ENOMEM);
513 void ff_snow_release_buffer(AVCodecContext *avctx)
515 SnowContext *s = avctx->priv_data;
517 if(s->last_picture[s->max_ref_frames-1]->data[0]){
518 av_frame_unref(s->last_picture[s->max_ref_frames-1]);
522 int ff_snow_frames_prepare(SnowContext *s)
526 ff_snow_release_buffer(s->avctx);
528 tmp= s->last_picture[s->max_ref_frames-1];
529 for (int i = s->max_ref_frames - 1; i > 0; i--)
530 s->last_picture[i] = s->last_picture[i-1];
531 s->last_picture[0] = s->current_picture;
532 s->current_picture = tmp;
536 s->current_picture->flags |= AV_FRAME_FLAG_KEY;
539 for(i=0; i<s->max_ref_frames && s->last_picture[i]->data[0]; i++)
540 if(i && (s->last_picture[i-1]->flags & AV_FRAME_FLAG_KEY))
543 if(s->ref_frames==0){
544 av_log(s->avctx,AV_LOG_ERROR, "No reference frames\n");
545 return AVERROR_INVALIDDATA;
547 s->current_picture->flags &= ~AV_FRAME_FLAG_KEY;
553 av_cold void ff_snow_common_end(SnowContext *s)
555 int plane_index, level, orientation, i;
557 av_freep(&s->spatial_dwt_buffer);
558 av_freep(&s->temp_dwt_buffer);
559 av_freep(&s->spatial_idwt_buffer);
560 av_freep(&s->temp_idwt_buffer);
561 av_freep(&s->run_buffer);
564 av_freep(&s->scratchbuf);
565 av_freep(&s->emu_edge_buffer);
567 for(i=0; i<MAX_REF_FRAMES; i++){
568 if(s->last_picture[i] && s->last_picture[i]->data[0]) {
569 av_assert0(s->last_picture[i]->data[0] != s->current_picture->data[0]);
571 av_frame_free(&s->last_picture[i]);
574 for(plane_index=0; plane_index < MAX_PLANES; plane_index++){
575 for(level=MAX_DECOMPOSITIONS-1; level>=0; level--){
576 for(orientation=level ? 1 : 0; orientation<4; orientation++){
577 SubBand *b= &s->plane[plane_index].band[level][orientation];
579 av_freep(&b->x_coeff);
583 av_frame_free(&s->mconly_picture);
584 av_frame_free(&s->current_picture);