snow.c 24.9 KB
Newer Older
1 2 3
/*
 * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
 *
4 5 6
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
7 8
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12 13 14 15 16
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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
19 20
 */

21
#include "libavutil/intmath.h"
22 23
#include "libavutil/log.h"
#include "libavutil/opt.h"
24
#include "avcodec.h"
25
#include "me_cmp.h"
26
#include "snow_dwt.h"
27
#include "internal.h"
28
#include "snow.h"
29
#include "snowdata.h"
30 31

#include "rangecoder.h"
32
#include "mathops.h"
33
#include "h263.h"
34

35

36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51
void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
                              int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
    int y, x;
    IDWTELEM * dst;
    for(y=0; y<b_h; y++){
        //FIXME ugly misuse of obmc_stride
        const uint8_t *obmc1= obmc + y*obmc_stride;
        const uint8_t *obmc2= obmc1+ (obmc_stride>>1);
        const uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1);
        const uint8_t *obmc4= obmc3+ (obmc_stride>>1);
        dst = slice_buffer_get_line(sb, src_y + y);
        for(x=0; x<b_w; x++){
            int v=   obmc1[x] * block[3][x + y*src_stride]
                    +obmc2[x] * block[2][x + y*src_stride]
                    +obmc3[x] * block[1][x + y*src_stride]
                    +obmc4[x] * block[0][x + y*src_stride];
52

53 54 55
            v <<= 8 - LOG2_OBMC_MAX;
            if(FRAC_BITS != 8){
                v >>= 8 - FRAC_BITS;
56
            }
57 58 59 60 61
            if(add){
                v += dst[x + src_x];
                v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS;
                if(v&(~255)) v= ~(v>>31);
                dst8[x + y*src_stride] = v;
62
            }else{
63
                dst[x + src_x] -= v;
64 65
            }
        }
66
    }
67 68
}

69 70 71
int ff_snow_get_buffer(SnowContext *s, AVFrame *frame)
{
    int ret, i;
72
    int edges_needed = av_codec_is_encoder(s->avctx->codec);
73

74 75 76 77 78 79
    frame->width  = s->avctx->width ;
    frame->height = s->avctx->height;
    if (edges_needed) {
        frame->width  += 2 * EDGE_WIDTH;
        frame->height += 2 * EDGE_WIDTH;
    }
80 81
    if ((ret = ff_get_buffer(s->avctx, frame, AV_GET_BUFFER_FLAG_REF)) < 0)
        return ret;
82 83 84 85 86 87 88 89 90
    if (edges_needed) {
        for (i = 0; frame->data[i]; i++) {
            int offset = (EDGE_WIDTH >> (i ? s->chroma_v_shift : 0)) *
                            frame->linesize[i] +
                            (EDGE_WIDTH >> (i ? s->chroma_h_shift : 0));
            frame->data[i] += offset;
        }
        frame->width  = s->avctx->width;
        frame->height = s->avctx->height;
91 92 93 94 95
    }

    return 0;
}

96
void ff_snow_reset_contexts(SnowContext *s){ //FIXME better initial contexts
97 98
    int plane_index, level, orientation;

99
    for(plane_index=0; plane_index<3; plane_index++){
100
        for(level=0; level<MAX_DECOMPOSITIONS; level++){
101
            for(orientation=level ? 1:0; orientation<4; orientation++){
102
                memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state));
103 104 105
            }
        }
    }
106 107
    memset(s->header_state, MID_STATE, sizeof(s->header_state));
    memset(s->block_state, MID_STATE, sizeof(s->block_state));
108 109
}

110
int ff_snow_alloc_blocks(SnowContext *s){
111 112
    int w= FF_CEIL_RSHIFT(s->avctx->width,  LOG2_MB_SIZE);
    int h= FF_CEIL_RSHIFT(s->avctx->height, LOG2_MB_SIZE);
113

114 115
    s->b_width = w;
    s->b_height= h;
116

117
    av_free(s->block);
118
    s->block= av_mallocz_array(w * h,  sizeof(BlockNode) << (s->block_max_depth*2));
119 120 121
    if (!s->block)
        return AVERROR(ENOMEM);

122 123 124
    return 0;
}

125
static av_cold void init_qexp(void){
126
    int i;
127
    double v=128;
128

129
    for(i=0; i<QROOT; i++){
130
        ff_qexp[i]= lrintf(v);
131
        v *= pow(2, 1.0 / QROOT);
132
    }
133
}
134
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){
135
    static const uint8_t weight[64]={
Michael Niedermayer's avatar
Michael Niedermayer committed
136 137 138 139 140 141 142 143 144 145
    8,7,6,5,4,3,2,1,
    7,7,0,0,0,0,0,1,
    6,0,6,0,0,0,2,0,
    5,0,0,5,0,3,0,0,
    4,0,0,0,4,0,0,0,
    3,0,0,5,0,3,0,0,
    2,0,6,0,0,0,2,0,
    1,7,0,0,0,0,0,1,
    };

146
    static const uint8_t brane[256]={
Michael Niedermayer's avatar
Michael Niedermayer committed
147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
    0x00,0x01,0x01,0x01,0x01,0x01,0x01,0x01,0x11,0x12,0x12,0x12,0x12,0x12,0x12,0x12,
    0x04,0x05,0xcc,0xcc,0xcc,0xcc,0xcc,0x41,0x15,0x16,0xcc,0xcc,0xcc,0xcc,0xcc,0x52,
    0x04,0xcc,0x05,0xcc,0xcc,0xcc,0x41,0xcc,0x15,0xcc,0x16,0xcc,0xcc,0xcc,0x52,0xcc,
    0x04,0xcc,0xcc,0x05,0xcc,0x41,0xcc,0xcc,0x15,0xcc,0xcc,0x16,0xcc,0x52,0xcc,0xcc,
    0x04,0xcc,0xcc,0xcc,0x41,0xcc,0xcc,0xcc,0x15,0xcc,0xcc,0xcc,0x16,0xcc,0xcc,0xcc,
    0x04,0xcc,0xcc,0x41,0xcc,0x05,0xcc,0xcc,0x15,0xcc,0xcc,0x52,0xcc,0x16,0xcc,0xcc,
    0x04,0xcc,0x41,0xcc,0xcc,0xcc,0x05,0xcc,0x15,0xcc,0x52,0xcc,0xcc,0xcc,0x16,0xcc,
    0x04,0x41,0xcc,0xcc,0xcc,0xcc,0xcc,0x05,0x15,0x52,0xcc,0xcc,0xcc,0xcc,0xcc,0x16,
    0x44,0x45,0x45,0x45,0x45,0x45,0x45,0x45,0x55,0x56,0x56,0x56,0x56,0x56,0x56,0x56,
    0x48,0x49,0xcc,0xcc,0xcc,0xcc,0xcc,0x85,0x59,0x5A,0xcc,0xcc,0xcc,0xcc,0xcc,0x96,
    0x48,0xcc,0x49,0xcc,0xcc,0xcc,0x85,0xcc,0x59,0xcc,0x5A,0xcc,0xcc,0xcc,0x96,0xcc,
    0x48,0xcc,0xcc,0x49,0xcc,0x85,0xcc,0xcc,0x59,0xcc,0xcc,0x5A,0xcc,0x96,0xcc,0xcc,
    0x48,0xcc,0xcc,0xcc,0x49,0xcc,0xcc,0xcc,0x59,0xcc,0xcc,0xcc,0x96,0xcc,0xcc,0xcc,
    0x48,0xcc,0xcc,0x85,0xcc,0x49,0xcc,0xcc,0x59,0xcc,0xcc,0x96,0xcc,0x5A,0xcc,0xcc,
    0x48,0xcc,0x85,0xcc,0xcc,0xcc,0x49,0xcc,0x59,0xcc,0x96,0xcc,0xcc,0xcc,0x5A,0xcc,
    0x48,0x85,0xcc,0xcc,0xcc,0xcc,0xcc,0x49,0x59,0x96,0xcc,0xcc,0xcc,0xcc,0xcc,0x5A,
    };

165
    static const uint8_t needs[16]={
Michael Niedermayer's avatar
Michael Niedermayer committed
166 167 168 169 170 171 172
    0,1,0,0,
    2,4,2,0,
    0,1,0,0,
    15
    };

    int x, y, b, r, l;
173
    int16_t tmpIt   [64*(32+HTAPS_MAX)];
174
    uint8_t tmp2t[3][64*(32+HTAPS_MAX)];
Michael Niedermayer's avatar
Michael Niedermayer committed
175 176
    int16_t *tmpI= tmpIt;
    uint8_t *tmp2= tmp2t[0];
177
    const uint8_t *hpel[11];
178
    av_assert2(dx<16 && dy<16);
Michael Niedermayer's avatar
Michael Niedermayer committed
179 180 181 182
    r= brane[dx + 16*dy]&15;
    l= brane[dx + 16*dy]>>4;

    b= needs[l] | needs[r];
183 184
    if(p && !p->diag_mc)
        b= 15;
Michael Niedermayer's avatar
Michael Niedermayer committed
185 186

    if(b&5){
187
        for(y=0; y < b_h+HTAPS_MAX-1; y++){
Michael Niedermayer's avatar
Michael Niedermayer committed
188
            for(x=0; x < b_w; x++){
189 190 191 192 193 194 195 196
                int a_1=src[x + HTAPS_MAX/2-4];
                int a0= src[x + HTAPS_MAX/2-3];
                int a1= src[x + HTAPS_MAX/2-2];
                int a2= src[x + HTAPS_MAX/2-1];
                int a3= src[x + HTAPS_MAX/2+0];
                int a4= src[x + HTAPS_MAX/2+1];
                int a5= src[x + HTAPS_MAX/2+2];
                int a6= src[x + HTAPS_MAX/2+3];
197 198 199 200 201 202 203 204 205 206
                int am=0;
                if(!p || p->fast_mc){
                    am= 20*(a2+a3) - 5*(a1+a4) + (a0+a5);
                    tmpI[x]= am;
                    am= (am+16)>>5;
                }else{
                    am= p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6);
                    tmpI[x]= am;
                    am= (am+32)>>6;
                }
207

Michael Niedermayer's avatar
Michael Niedermayer committed
208 209 210 211
                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            tmpI+= 64;
212
            tmp2+= 64;
Michael Niedermayer's avatar
Michael Niedermayer committed
213
            src += stride;
214
        }
Michael Niedermayer's avatar
Michael Niedermayer committed
215
        src -= stride*y;
Michael Niedermayer's avatar
Michael Niedermayer committed
216
    }
217
    src += HTAPS_MAX/2 - 1;
Michael Niedermayer's avatar
Michael Niedermayer committed
218
    tmp2= tmp2t[1];
219

Michael Niedermayer's avatar
Michael Niedermayer committed
220
    if(b&2){
Michael Niedermayer's avatar
Michael Niedermayer committed
221 222
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w+1; x++){
223 224 225 226 227 228 229 230
                int a_1=src[x + (HTAPS_MAX/2-4)*stride];
                int a0= src[x + (HTAPS_MAX/2-3)*stride];
                int a1= src[x + (HTAPS_MAX/2-2)*stride];
                int a2= src[x + (HTAPS_MAX/2-1)*stride];
                int a3= src[x + (HTAPS_MAX/2+0)*stride];
                int a4= src[x + (HTAPS_MAX/2+1)*stride];
                int a5= src[x + (HTAPS_MAX/2+2)*stride];
                int a6= src[x + (HTAPS_MAX/2+3)*stride];
231 232 233 234 235
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 16)>>5;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 32)>>6;
236

Michael Niedermayer's avatar
Michael Niedermayer committed
237 238 239 240
                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            src += stride;
241
            tmp2+= 64;
Michael Niedermayer's avatar
Michael Niedermayer committed
242
        }
Michael Niedermayer's avatar
Michael Niedermayer committed
243
        src -= stride*y;
Michael Niedermayer's avatar
Michael Niedermayer committed
244
    }
245
    src += stride*(HTAPS_MAX/2 - 1);
Michael Niedermayer's avatar
Michael Niedermayer committed
246 247 248 249 250
    tmp2= tmp2t[2];
    tmpI= tmpIt;
    if(b&4){
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
251 252 253 254 255 256 257 258
                int a_1=tmpI[x + (HTAPS_MAX/2-4)*64];
                int a0= tmpI[x + (HTAPS_MAX/2-3)*64];
                int a1= tmpI[x + (HTAPS_MAX/2-2)*64];
                int a2= tmpI[x + (HTAPS_MAX/2-1)*64];
                int a3= tmpI[x + (HTAPS_MAX/2+0)*64];
                int a4= tmpI[x + (HTAPS_MAX/2+1)*64];
                int a5= tmpI[x + (HTAPS_MAX/2+2)*64];
                int a6= tmpI[x + (HTAPS_MAX/2+3)*64];
259 260 261 262 263
                int am=0;
                if(!p || p->fast_mc)
                    am= (20*(a2+a3) - 5*(a1+a4) + (a0+a5) + 512)>>10;
                else
                    am= (p->hcoeff[0]*(a2+a3) + p->hcoeff[1]*(a1+a4) + p->hcoeff[2]*(a0+a5) + p->hcoeff[3]*(a_1+a6) + 2048)>>12;
Michael Niedermayer's avatar
Michael Niedermayer committed
264 265 266 267
                if(am&(~255)) am= ~(am>>31);
                tmp2[x]= am;
            }
            tmpI+= 64;
268
            tmp2+= 64;
Michael Niedermayer's avatar
Michael Niedermayer committed
269 270
        }
    }
271

Michael Niedermayer's avatar
Michael Niedermayer committed
272
    hpel[ 0]= src;
273
    hpel[ 1]= tmp2t[0] + 64*(HTAPS_MAX/2-1);
Michael Niedermayer's avatar
Michael Niedermayer committed
274 275 276 277 278 279 280
    hpel[ 2]= src + 1;

    hpel[ 4]= tmp2t[1];
    hpel[ 5]= tmp2t[2];
    hpel[ 6]= tmp2t[1] + 1;

    hpel[ 8]= src + stride;
281
    hpel[ 9]= hpel[1] + 64;
Michael Niedermayer's avatar
Michael Niedermayer committed
282 283
    hpel[10]= hpel[8] + 1;

284 285
#define MC_STRIDE(x) (needs[x] ? 64 : stride)

Michael Niedermayer's avatar
Michael Niedermayer committed
286
    if(b==15){
287 288 289 290 291 292 293 294 295
        int dxy = dx / 8 + dy / 8 * 4;
        const uint8_t *src1 = hpel[dxy    ];
        const uint8_t *src2 = hpel[dxy + 1];
        const uint8_t *src3 = hpel[dxy + 4];
        const uint8_t *src4 = hpel[dxy + 5];
        int stride1 = MC_STRIDE(dxy);
        int stride2 = MC_STRIDE(dxy + 1);
        int stride3 = MC_STRIDE(dxy + 4);
        int stride4 = MC_STRIDE(dxy + 5);
Michael Niedermayer's avatar
Michael Niedermayer committed
296 297 298 299 300 301 302
        dx&=7;
        dy&=7;
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                dst[x]= ((8-dx)*(8-dy)*src1[x] + dx*(8-dy)*src2[x]+
                         (8-dx)*   dy *src3[x] + dx*   dy *src4[x]+32)>>6;
            }
303 304 305 306
            src1+=stride1;
            src2+=stride2;
            src3+=stride3;
            src4+=stride4;
Michael Niedermayer's avatar
Michael Niedermayer committed
307 308 309
            dst +=stride;
        }
    }else{
310 311
        const uint8_t *src1= hpel[l];
        const uint8_t *src2= hpel[r];
312 313
        int stride1 = MC_STRIDE(l);
        int stride2 = MC_STRIDE(r);
Michael Niedermayer's avatar
Michael Niedermayer committed
314 315 316 317 318 319
        int a= weight[((dx&7) + (8*(dy&7)))];
        int b= 8-a;
        for(y=0; y < b_h; y++){
            for(x=0; x < b_w; x++){
                dst[x]= (a*src1[x] + b*src2[x] + 4)>>3;
            }
320 321
            src1+=stride1;
            src2+=stride2;
Michael Niedermayer's avatar
Michael Niedermayer committed
322
            dst +=stride;
323 324 325 326
        }
    }
}

327
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){
328
    if(block->type & BLOCK_INTRA){
329
        int x, y;
330 331
        const unsigned color  = block->color[plane_index];
        const unsigned color4 = color*0x01010101;
332 333 334 335 336 337 338 339 340 341 342 343
        if(b_w==32){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
                *(uint32_t*)&dst[8 + y*stride]= color4;
                *(uint32_t*)&dst[12+ y*stride]= color4;
                *(uint32_t*)&dst[16+ y*stride]= color4;
                *(uint32_t*)&dst[20+ y*stride]= color4;
                *(uint32_t*)&dst[24+ y*stride]= color4;
                *(uint32_t*)&dst[28+ y*stride]= color4;
            }
        }else if(b_w==16){
344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
                *(uint32_t*)&dst[8 + y*stride]= color4;
                *(uint32_t*)&dst[12+ y*stride]= color4;
            }
        }else if(b_w==8){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
                *(uint32_t*)&dst[4 + y*stride]= color4;
            }
        }else if(b_w==4){
            for(y=0; y < b_h; y++){
                *(uint32_t*)&dst[0 + y*stride]= color4;
            }
        }else{
            for(y=0; y < b_h; y++){
                for(x=0; x < b_w; x++){
                    dst[x + y*stride]= color;
                }
364 365 366
            }
        }
    }else{
367
        uint8_t *src= s->last_picture[block->ref]->data[plane_index];
368
        const int scale= plane_index ?  (2*s->mv_scale)>>s->chroma_h_shift : 2*s->mv_scale;
369 370
        int mx= block->mx*scale;
        int my= block->my*scale;
371 372
        const int dx= mx&15;
        const int dy= my&15;
373
        const int tab_index= 3 - (b_w>>2) + (b_w>>4);
374 375
        sx += (mx>>4) - (HTAPS_MAX/2-1);
        sy += (my>>4) - (HTAPS_MAX/2-1);
376
        src += sx + sy*stride;
377 378
        if(   (unsigned)sx >= FFMAX(w - b_w - (HTAPS_MAX-2), 0)
           || (unsigned)sy >= FFMAX(h - b_h - (HTAPS_MAX-2), 0)){
379 380
            s->vdsp.emulated_edge_mc(tmp + MB_SIZE, src,
                                     stride, stride,
381 382
                                     b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1,
                                     sx, sy, w, h);
383 384
            src= tmp + MB_SIZE;
        }
385 386 387

        av_assert2(s->chroma_h_shift == s->chroma_v_shift); // only one mv_scale

388
        av_assert2((tab_index>=0 && tab_index<4) || b_w==32);
389 390 391 392 393 394
        if(    (dx&3) || (dy&3)
            || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h)
            || (b_w&(b_w-1))
            || b_w == 1
            || b_h == 1
            || !s->plane[plane_index].fast_mc )
395
            mc_block(&s->plane[plane_index], dst, src, stride, b_w, b_h, dx, dy);
396 397 398
        else if(b_w==32){
            int y;
            for(y=0; y<b_h; y+=16){
399 400
                s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 3 + (y+3)*stride,stride);
                s->h264qpel.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 19 + (y+3)*stride,stride);
401 402
            }
        }else if(b_w==b_h)
403
            s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst,src + 3 + 3*stride,stride);
404
        else if(b_w==2*b_h){
405 406
            s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst    ,src + 3       + 3*stride,stride);
            s->h264qpel.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 3 + b_h + 3*stride,stride);
407
        }else{
408
            av_assert2(2*b_w==b_h);
409 410
            s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst           ,src + 3 + 3*stride           ,stride);
            s->h264qpel.put_h264_qpel_pixels_tab[tab_index  ][dy+(dx>>2)](dst+b_w*stride,src + 3 + 3*stride+b_w*stride,stride);
411
        }
412 413 414
    }
}

415
#define mca(dx,dy,b_w)\
416
static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, ptrdiff_t stride, int h){\
417
    av_assert2(h==b_w);\
418
    mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, stride, b_w, b_w, dx, dy);\
419 420
}

421 422 423 424 425 426 427 428
mca( 0, 0,16)
mca( 8, 0,16)
mca( 0, 8,16)
mca( 8, 8,16)
mca( 0, 0,8)
mca( 8, 0,8)
mca( 0, 8,8)
mca( 8, 8,8)
429

430 431 432
av_cold int ff_snow_common_init(AVCodecContext *avctx){
    SnowContext *s = avctx->priv_data;
    int width, height;
433
    int i, j;
434

435
    s->avctx= avctx;
436
    s->max_ref_frames=1; //just make sure it's not an invalid value in case of no initial keyframe
437

438
    ff_me_cmp_init(&s->mecc, avctx);
439
    ff_hpeldsp_init(&s->hdsp, avctx->flags);
440
    ff_videodsp_init(&s->vdsp, 8);
441
    ff_dwt_init(&s->dwt);
442
    ff_h264qpel_init(&s->h264qpel, 8);
443

444
#define mcf(dx,dy)\
445 446
    s->qdsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
    s->qdsp.put_no_rnd_qpel_pixels_tab[0][dy+dx/4]=\
447
        s->h264qpel.put_h264_qpel_pixels_tab[0][dy+dx/4];\
448 449
    s->qdsp.put_qpel_pixels_tab       [1][dy+dx/4]=\
    s->qdsp.put_no_rnd_qpel_pixels_tab[1][dy+dx/4]=\
450
        s->h264qpel.put_h264_qpel_pixels_tab[1][dy+dx/4];
451

452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467
    mcf( 0, 0)
    mcf( 4, 0)
    mcf( 8, 0)
    mcf(12, 0)
    mcf( 0, 4)
    mcf( 4, 4)
    mcf( 8, 4)
    mcf(12, 4)
    mcf( 0, 8)
    mcf( 4, 8)
    mcf( 8, 8)
    mcf(12, 8)
    mcf( 0,12)
    mcf( 4,12)
    mcf( 8,12)
    mcf(12,12)
468

469
#define mcfh(dx,dy)\
470 471
    s->hdsp.put_pixels_tab       [0][dy/4+dx/8]=\
    s->hdsp.put_no_rnd_pixels_tab[0][dy/4+dx/8]=\
472
        mc_block_hpel ## dx ## dy ## 16;\
473 474
    s->hdsp.put_pixels_tab       [1][dy/4+dx/8]=\
    s->hdsp.put_no_rnd_pixels_tab[1][dy/4+dx/8]=\
475
        mc_block_hpel ## dx ## dy ## 8;
476

477 478 479 480
    mcfh(0, 0)
    mcfh(8, 0)
    mcfh(0, 8)
    mcfh(8, 8)
481

482
    init_qexp();
483

484
//    dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift);
485

486 487
    width= s->avctx->width;
    height= s->avctx->height;
488

489 490 491 492 493
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->spatial_idwt_buffer, width, height * sizeof(IDWTELEM), fail);
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->spatial_dwt_buffer,  width, height * sizeof(DWTELEM),  fail); //FIXME this does not belong here
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->temp_dwt_buffer,     width, sizeof(DWTELEM),  fail);
    FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->temp_idwt_buffer,    width, sizeof(IDWTELEM), fail);
    FF_ALLOC_ARRAY_OR_GOTO(avctx,  s->run_buffer,          ((width + 1) >> 1), ((height + 1) >> 1) * sizeof(*s->run_buffer), fail);
Loren Merritt's avatar
Loren Merritt committed
494

495
    for(i=0; i<MAX_REF_FRAMES; i++) {
496
        for(j=0; j<MAX_REF_FRAMES; j++)
497
            ff_scale_mv_ref[i][j] = 256*(i+1)/(j+1);
498
        s->last_picture[i] = av_frame_alloc();
499 500
        if (!s->last_picture[i])
            goto fail;
501 502
    }

503 504
    s->mconly_picture = av_frame_alloc();
    s->current_picture = av_frame_alloc();
505 506
    if (!s->mconly_picture || !s->current_picture)
        goto fail;
Loren Merritt's avatar
Loren Merritt committed
507

508
    return 0;
509 510
fail:
    return AVERROR(ENOMEM);
Loren Merritt's avatar
Loren Merritt committed
511 512
}

513
int ff_snow_common_init_after_header(AVCodecContext *avctx) {
514 515
    SnowContext *s = avctx->priv_data;
    int plane_index, level, orientation;
516 517 518
    int ret, emu_buf_size;

    if(!s->scratchbuf) {
519
        if ((ret = ff_get_buffer(s->avctx, s->mconly_picture,
520
                                 AV_GET_BUFFER_FLAG_REF)) < 0)
521
            return ret;
522
        FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->scratchbuf, FFMAX(s->mconly_picture->linesize[0], 2*avctx->width+256), 7*MB_SIZE, fail);
523
        emu_buf_size = FFMAX(s->mconly_picture->linesize[0], 2*avctx->width+256) * (2 * MB_SIZE + HTAPS_MAX - 1);
524 525 526
        FF_ALLOC_OR_GOTO(avctx, s->emu_edge_buffer, emu_buf_size, fail);
    }

527
    if(s->mconly_picture->format != avctx->pix_fmt) {
528 529 530
        av_log(avctx, AV_LOG_ERROR, "pixel format changed\n");
        return AVERROR_INVALIDDATA;
    }
531

532
    for(plane_index=0; plane_index < s->nb_planes; plane_index++){
533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570
        int w= s->avctx->width;
        int h= s->avctx->height;

        if(plane_index){
            w>>= s->chroma_h_shift;
            h>>= s->chroma_v_shift;
        }
        s->plane[plane_index].width = w;
        s->plane[plane_index].height= h;

        for(level=s->spatial_decomposition_count-1; level>=0; level--){
            for(orientation=level ? 1 : 0; orientation<4; orientation++){
                SubBand *b= &s->plane[plane_index].band[level][orientation];

                b->buf= s->spatial_dwt_buffer;
                b->level= level;
                b->stride= s->plane[plane_index].width << (s->spatial_decomposition_count - level);
                b->width = (w + !(orientation&1))>>1;
                b->height= (h + !(orientation>1))>>1;

                b->stride_line = 1 << (s->spatial_decomposition_count - level);
                b->buf_x_offset = 0;
                b->buf_y_offset = 0;

                if(orientation&1){
                    b->buf += (w+1)>>1;
                    b->buf_x_offset = (w+1)>>1;
                }
                if(orientation>1){
                    b->buf += b->stride>>1;
                    b->buf_y_offset = b->stride_line >> 1;
                }
                b->ibuf= s->spatial_idwt_buffer + (b->buf - s->spatial_dwt_buffer);

                if(level)
                    b->parent= &s->plane[plane_index].band[level-1][orientation];
                //FIXME avoid this realloc
                av_freep(&b->x_coeff);
571
                b->x_coeff=av_mallocz_array(((b->width+1) * b->height+1), sizeof(x_and_coeff));
572 573
                if (!b->x_coeff)
                    goto fail;
574 575 576 577 578 579 580
            }
            w= (w+1)>>1;
            h= (h+1)>>1;
        }
    }

    return 0;
581 582
fail:
    return AVERROR(ENOMEM);
583 584 585 586
}

#define USE_HALFPEL_PLANE 0

587
static int halfpel_interpol(SnowContext *s, uint8_t *halfpel[4][4], AVFrame *frame){
588 589
    int p,x,y;

590
    for(p=0; p < s->nb_planes; p++){
591
        int is_chroma= !!p;
592 593
        int w= is_chroma ? s->avctx->width >>s->chroma_h_shift : s->avctx->width;
        int h= is_chroma ? s->avctx->height>>s->chroma_v_shift : s->avctx->height;
594 595 596
        int ls= frame->linesize[p];
        uint8_t *src= frame->data[p];

597 598 599
        halfpel[1][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
        halfpel[2][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
        halfpel[3][p] = (uint8_t*) av_malloc(ls * (h + 2 * EDGE_WIDTH)) + EDGE_WIDTH * (1 + ls);
600 601
        if (!halfpel[1][p] || !halfpel[2][p] || !halfpel[3][p])
            return AVERROR(ENOMEM);
602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628

        halfpel[0][p]= src;
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[1][p][i]= (20*(src[i] + src[i+1]) - 5*(src[i-1] + src[i+2]) + (src[i-2] + src[i+3]) + 16 )>>5;
            }
        }
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[2][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
            }
        }
        src= halfpel[1][p];
        for(y=0; y<h; y++){
            for(x=0; x<w; x++){
                int i= y*ls + x;

                halfpel[3][p][i]= (20*(src[i] + src[i+ls]) - 5*(src[i-ls] + src[i+2*ls]) + (src[i-2*ls] + src[i+3*ls]) + 16 )>>5;
            }
        }

//FIXME border!
    }
629
    return 0;
630 631
}

632 633
void ff_snow_release_buffer(AVCodecContext *avctx)
{
634 635 636
    SnowContext *s = avctx->priv_data;
    int i;

637 638
    if(s->last_picture[s->max_ref_frames-1]->data[0]){
        av_frame_unref(s->last_picture[s->max_ref_frames-1]);
639
        for(i=0; i<9; i++)
640
            if(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3]) {
641
                av_free(s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] - EDGE_WIDTH*(1+s->current_picture->linesize[i%3]));
642 643
                s->halfpel_plane[s->max_ref_frames-1][1+i/3][i%3] = NULL;
            }
644 645 646
    }
}

647
int ff_snow_frame_start(SnowContext *s){
648
   AVFrame *tmp;
649
   int i, ret;
650

651
    ff_snow_release_buffer(s->avctx);
652

653
    tmp= s->last_picture[s->max_ref_frames-1];
654
    for(i=s->max_ref_frames-1; i>0; i--)
655
        s->last_picture[i] = s->last_picture[i-1];
656
    memmove(s->halfpel_plane+1, s->halfpel_plane, (s->max_ref_frames-1)*sizeof(void*)*4*4);
657 658
    if(USE_HALFPEL_PLANE && s->current_picture->data[0]) {
        if((ret = halfpel_interpol(s, s->halfpel_plane[0], s->current_picture)) < 0)
659 660
            return ret;
    }
661 662
    s->last_picture[0] = s->current_picture;
    s->current_picture = tmp;
663 664 665 666 667

    if(s->keyframe){
        s->ref_frames= 0;
    }else{
        int i;
668 669
        for(i=0; i<s->max_ref_frames && s->last_picture[i]->data[0]; i++)
            if(i && s->last_picture[i-1]->key_frame)
670 671 672 673 674 675 676
                break;
        s->ref_frames= i;
        if(s->ref_frames==0){
            av_log(s->avctx,AV_LOG_ERROR, "No reference frames\n");
            return -1;
        }
    }
677
    if ((ret = ff_snow_get_buffer(s, s->current_picture)) < 0)
678
        return ret;
679

680
    s->current_picture->key_frame= s->keyframe;
681 682 683 684

    return 0;
}

685 686
av_cold void ff_snow_common_end(SnowContext *s)
{
687 688 689
    int plane_index, level, orientation, i;

    av_freep(&s->spatial_dwt_buffer);
690
    av_freep(&s->temp_dwt_buffer);
691
    av_freep(&s->spatial_idwt_buffer);
692
    av_freep(&s->temp_idwt_buffer);
693
    av_freep(&s->run_buffer);
694 695 696 697 698 699 700 701 702

    s->m.me.temp= NULL;
    av_freep(&s->m.me.scratchpad);
    av_freep(&s->m.me.map);
    av_freep(&s->m.me.score_map);
    av_freep(&s->m.obmc_scratchpad);

    av_freep(&s->block);
    av_freep(&s->scratchbuf);
703
    av_freep(&s->emu_edge_buffer);
704 705 706 707

    for(i=0; i<MAX_REF_FRAMES; i++){
        av_freep(&s->ref_mvs[i]);
        av_freep(&s->ref_scores[i]);
708
        if(s->last_picture[i] && s->last_picture[i]->data[0]) {
709
            av_assert0(s->last_picture[i]->data[0] != s->current_picture->data[0]);
710
        }
711
        av_frame_free(&s->last_picture[i]);
712 713
    }

714
    for(plane_index=0; plane_index < s->nb_planes; plane_index++){
715 716 717 718 719 720 721 722
        for(level=s->spatial_decomposition_count-1; level>=0; level--){
            for(orientation=level ? 1 : 0; orientation<4; orientation++){
                SubBand *b= &s->plane[plane_index].band[level][orientation];

                av_freep(&b->x_coeff);
            }
        }
    }
723 724
    av_frame_free(&s->mconly_picture);
    av_frame_free(&s->current_picture);
725
}