X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fsnow.c;h=6bc9a8f1aece6e215c410513fff8da2e23ee7419;hb=486497e07b4617d9548a5b7347076f1b329bd417;hp=c9d47a8aa14dba13aced04d36b7b737469669e5c;hpb=8ff53f5b073d14528fd12f3a99d5beed4d8e2839;p=ffmpeg diff --git a/libavcodec/snow.c b/libavcodec/snow.c index c9d47a8aa14..6bc9a8f1aec 100644 --- a/libavcodec/snow.c +++ b/libavcodec/snow.c @@ -1,40 +1,35 @@ /* * Copyright (C) 2004 Michael Niedermayer * - * This library is free software; you can redistribute it and/or + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. + * version 2.1 of the License, or (at your option) any later version. * - * This library is distributed in the hope that it will be useful, + * FFmpeg is distributed in the hope that it will be useful, * 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 - * License along with this library; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ #include "avcodec.h" #include "common.h" #include "dsputil.h" +#include "snow.h" #include "rangecoder.h" -#define MID_STATE 128 #include "mpegvideo.h" #undef NDEBUG #include -#define MAX_DECOMPOSITIONS 8 -#define MAX_PLANES 4 -#define DWTELEM int -#define QROOT 8 -#define LOSSLESS_QLOG -128 -#define FRAC_BITS 8 - static const int8_t quant3[256]={ 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, @@ -71,6 +66,24 @@ static const int8_t quant3b[256]={ -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, -1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1,-1, }; +static const int8_t quant3bA[256]={ + 0, 0, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, + 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, 1,-1, +}; static const int8_t quant5[256]={ 0, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, @@ -162,8 +175,6 @@ static const int8_t quant13[256]={ -4,-4,-4,-4,-4,-4,-4,-4,-4,-3,-3,-3,-3,-2,-2,-1, }; -#define LOG2_OBMC_MAX 6 -#define OBMC_MAX (1<<(LOG2_OBMC_MAX)) #if 0 //64*cubic static const uint8_t obmc32[1024]={ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, @@ -221,57 +232,57 @@ static const uint8_t obmc16[256]={ }; #elif 1 // 64*linear static const uint8_t obmc32[1024]={ - 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, - 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 5, 5, 5, 4, 4, 4, 3, 3, 2, 2, 2, 1, 1, 1, 0, - 0, 1, 2, 2, 3, 3, 4, 5, 5, 6, 7, 7, 8, 8, 9,10,10, 9, 8, 8, 7, 7, 6, 5, 5, 4, 3, 3, 2, 2, 1, 0, - 0, 1, 2, 3, 4, 5, 6, 7, 7, 8, 9,10,11,12,13,14,14,13,12,11,10, 9, 8, 7, 7, 6, 5, 4, 3, 2, 1, 0, - 1, 2, 3, 4, 5, 6, 7, 8,10,11,12,13,14,15,16,17,17,16,15,14,13,12,11,10, 8, 7, 6, 5, 4, 3, 2, 1, - 1, 2, 3, 5, 6, 8, 9,10,12,13,14,16,17,19,20,21,21,20,19,17,16,14,13,12,10, 9, 8, 6, 5, 3, 2, 1, - 1, 2, 4, 6, 7, 9,11,12,14,15,17,19,20,22,24,25,25,24,22,20,19,17,15,14,12,11, 9, 7, 6, 4, 2, 1, - 1, 3, 5, 7, 8,10,12,14,16,18,20,22,23,25,27,29,29,27,25,23,22,20,18,16,14,12,10, 8, 7, 5, 3, 1, - 1, 3, 5, 7,10,12,14,16,18,20,22,24,27,29,31,33,33,31,29,27,24,22,20,18,16,14,12,10, 7, 5, 3, 1, - 1, 4, 6, 8,11,13,15,18,20,23,25,27,30,32,34,37,37,34,32,30,27,25,23,20,18,15,13,11, 8, 6, 4, 1, - 1, 4, 7, 9,12,14,17,20,22,25,28,30,33,35,38,41,41,38,35,33,30,28,25,22,20,17,14,12, 9, 7, 4, 1, - 1, 4, 7,10,13,16,19,22,24,27,30,33,36,39,42,45,45,42,39,36,33,30,27,24,22,19,16,13,10, 7, 4, 1, - 2, 5, 8,11,14,17,20,23,27,30,33,36,39,42,45,48,48,45,42,39,36,33,30,27,23,20,17,14,11, 8, 5, 2, - 2, 5, 8,12,15,19,22,25,29,32,35,39,42,46,49,52,52,49,46,42,39,35,32,29,25,22,19,15,12, 8, 5, 2, - 2, 5, 9,13,16,20,24,27,31,34,38,42,45,49,53,56,56,53,49,45,42,38,34,31,27,24,20,16,13, 9, 5, 2, - 2, 6,10,14,17,21,25,29,33,37,41,45,48,52,56,60,60,56,52,48,45,41,37,33,29,25,21,17,14,10, 6, 2, - 2, 6,10,14,17,21,25,29,33,37,41,45,48,52,56,60,60,56,52,48,45,41,37,33,29,25,21,17,14,10, 6, 2, - 2, 5, 9,13,16,20,24,27,31,34,38,42,45,49,53,56,56,53,49,45,42,38,34,31,27,24,20,16,13, 9, 5, 2, - 2, 5, 8,12,15,19,22,25,29,32,35,39,42,46,49,52,52,49,46,42,39,35,32,29,25,22,19,15,12, 8, 5, 2, - 2, 5, 8,11,14,17,20,23,27,30,33,36,39,42,45,48,48,45,42,39,36,33,30,27,23,20,17,14,11, 8, 5, 2, - 1, 4, 7,10,13,16,19,22,24,27,30,33,36,39,42,45,45,42,39,36,33,30,27,24,22,19,16,13,10, 7, 4, 1, - 1, 4, 7, 9,12,14,17,20,22,25,28,30,33,35,38,41,41,38,35,33,30,28,25,22,20,17,14,12, 9, 7, 4, 1, - 1, 4, 6, 8,11,13,15,18,20,23,25,27,30,32,34,37,37,34,32,30,27,25,23,20,18,15,13,11, 8, 6, 4, 1, - 1, 3, 5, 7,10,12,14,16,18,20,22,24,27,29,31,33,33,31,29,27,24,22,20,18,16,14,12,10, 7, 5, 3, 1, - 1, 3, 5, 7, 8,10,12,14,16,18,20,22,23,25,27,29,29,27,25,23,22,20,18,16,14,12,10, 8, 7, 5, 3, 1, - 1, 2, 4, 6, 7, 9,11,12,14,15,17,19,20,22,24,25,25,24,22,20,19,17,15,14,12,11, 9, 7, 6, 4, 2, 1, - 1, 2, 3, 5, 6, 8, 9,10,12,13,14,16,17,19,20,21,21,20,19,17,16,14,13,12,10, 9, 8, 6, 5, 3, 2, 1, - 1, 2, 3, 4, 5, 6, 7, 8,10,11,12,13,14,15,16,17,17,16,15,14,13,12,11,10, 8, 7, 6, 5, 4, 3, 2, 1, - 0, 1, 2, 3, 4, 5, 6, 7, 7, 8, 9,10,11,12,13,14,14,13,12,11,10, 9, 8, 7, 7, 6, 5, 4, 3, 2, 1, 0, - 0, 1, 2, 2, 3, 3, 4, 5, 5, 6, 7, 7, 8, 8, 9,10,10, 9, 8, 8, 7, 7, 6, 5, 5, 4, 3, 3, 2, 2, 1, 0, - 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 5, 5, 5, 4, 4, 4, 3, 3, 2, 2, 2, 1, 1, 1, 0, - 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, + 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, + 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0, + 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0, + 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0, + 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4, + 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4, + 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4, + 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4, + 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4, + 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4, + 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4, + 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4, + 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8, + 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8, + 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8, + 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8, + 8, 24, 40, 56, 68, 84,100,116,132,148,164,180,192,208,224,240,240,224,208,192,180,164,148,132,116,100, 84, 68, 56, 40, 24, 8, + 8, 20, 36, 52, 64, 80, 96,108,124,136,152,168,180,196,212,224,224,212,196,180,168,152,136,124,108, 96, 80, 64, 52, 36, 20, 8, + 8, 20, 32, 48, 60, 76, 88,100,116,128,140,156,168,184,196,208,208,196,184,168,156,140,128,116,100, 88, 76, 60, 48, 32, 20, 8, + 8, 20, 32, 44, 56, 68, 80, 92,108,120,132,144,156,168,180,192,192,180,168,156,144,132,120,108, 92, 80, 68, 56, 44, 32, 20, 8, + 4, 16, 28, 40, 52, 64, 76, 88, 96,108,120,132,144,156,168,180,180,168,156,144,132,120,108, 96, 88, 76, 64, 52, 40, 28, 16, 4, + 4, 16, 28, 36, 48, 56, 68, 80, 88,100,112,120,132,140,152,164,164,152,140,132,120,112,100, 88, 80, 68, 56, 48, 36, 28, 16, 4, + 4, 16, 24, 32, 44, 52, 60, 72, 80, 92,100,108,120,128,136,148,148,136,128,120,108,100, 92, 80, 72, 60, 52, 44, 32, 24, 16, 4, + 4, 12, 20, 28, 40, 48, 56, 64, 72, 80, 88, 96,108,116,124,132,132,124,116,108, 96, 88, 80, 72, 64, 56, 48, 40, 28, 20, 12, 4, + 4, 12, 20, 28, 32, 40, 48, 56, 64, 72, 80, 88, 92,100,108,116,116,108,100, 92, 88, 80, 72, 64, 56, 48, 40, 32, 28, 20, 12, 4, + 4, 8, 16, 24, 28, 36, 44, 48, 56, 60, 68, 76, 80, 88, 96,100,100, 96, 88, 80, 76, 68, 60, 56, 48, 44, 36, 28, 24, 16, 8, 4, + 4, 8, 12, 20, 24, 32, 36, 40, 48, 52, 56, 64, 68, 76, 80, 84, 84, 80, 76, 68, 64, 56, 52, 48, 40, 36, 32, 24, 20, 12, 8, 4, + 4, 8, 12, 16, 20, 24, 28, 32, 40, 44, 48, 52, 56, 60, 64, 68, 68, 64, 60, 56, 52, 48, 44, 40, 32, 28, 24, 20, 16, 12, 8, 4, + 0, 4, 8, 12, 16, 20, 24, 28, 28, 32, 36, 40, 44, 48, 52, 56, 56, 52, 48, 44, 40, 36, 32, 28, 28, 24, 20, 16, 12, 8, 4, 0, + 0, 4, 8, 8, 12, 12, 16, 20, 20, 24, 28, 28, 32, 32, 36, 40, 40, 36, 32, 32, 28, 28, 24, 20, 20, 16, 12, 12, 8, 8, 4, 0, + 0, 4, 4, 4, 8, 8, 8, 12, 12, 16, 16, 16, 20, 20, 20, 24, 24, 20, 20, 20, 16, 16, 16, 12, 12, 8, 8, 8, 4, 4, 4, 0, + 0, 0, 0, 0, 4, 4, 4, 4, 4, 4, 4, 4, 8, 8, 8, 8, 8, 8, 8, 8, 4, 4, 4, 4, 4, 4, 4, 4, 0, 0, 0, 0, //error:0.000020 }; static const uint8_t obmc16[256]={ - 0, 1, 1, 2, 2, 3, 3, 4, 4, 3, 3, 2, 2, 1, 1, 0, - 1, 2, 4, 5, 7, 8,10,11,11,10, 8, 7, 5, 4, 2, 1, - 1, 4, 6, 9,11,14,16,19,19,16,14,11, 9, 6, 4, 1, - 2, 5, 9,12,16,19,23,26,26,23,19,16,12, 9, 5, 2, - 2, 7,11,16,20,25,29,34,34,29,25,20,16,11, 7, 2, - 3, 8,14,19,25,30,36,41,41,36,30,25,19,14, 8, 3, - 3,10,16,23,29,36,42,49,49,42,36,29,23,16,10, 3, - 4,11,19,26,34,41,49,56,56,49,41,34,26,19,11, 4, - 4,11,19,26,34,41,49,56,56,49,41,34,26,19,11, 4, - 3,10,16,23,29,36,42,49,49,42,36,29,23,16,10, 3, - 3, 8,14,19,25,30,36,41,41,36,30,25,19,14, 8, 3, - 2, 7,11,16,20,25,29,34,34,29,25,20,16,11, 7, 2, - 2, 5, 9,12,16,19,23,26,26,23,19,16,12, 9, 5, 2, - 1, 4, 6, 9,11,14,16,19,19,16,14,11, 9, 6, 4, 1, - 1, 2, 4, 5, 7, 8,10,11,11,10, 8, 7, 5, 4, 2, 1, - 0, 1, 1, 2, 2, 3, 3, 4, 4, 3, 3, 2, 2, 1, 1, 0, + 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0, + 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4, + 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4, + 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8, + 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8, + 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12, + 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12, + 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16, + 16, 44, 76,104,136,164,196,224,224,196,164,136,104, 76, 44, 16, + 12, 40, 64, 92,116,144,168,196,196,168,144,116, 92, 64, 40, 12, + 12, 32, 56, 76,100,120,144,164,164,144,120,100, 76, 56, 32, 12, + 8, 28, 44, 64, 80,100,116,136,136,116,100, 80, 64, 44, 28, 8, + 8, 20, 36, 48, 64, 76, 92,104,104, 92, 76, 64, 48, 36, 20, 8, + 4, 16, 24, 36, 44, 56, 64, 76, 76, 64, 56, 44, 36, 24, 16, 4, + 4, 8, 16, 20, 28, 32, 40, 44, 44, 40, 32, 28, 20, 16, 8, 4, + 0, 4, 4, 8, 8, 12, 12, 16, 16, 12, 12, 8, 8, 4, 4, 0, //error:0.000015 }; #else //64*cos @@ -333,23 +344,23 @@ static const uint8_t obmc16[256]={ //linear *64 static const uint8_t obmc8[64]={ - 1, 3, 5, 7, 7, 5, 3, 1, - 3, 9,15,21,21,15, 9, 3, - 5,15,25,35,35,25,15, 5, - 7,21,35,49,49,35,21, 7, - 7,21,35,49,49,35,21, 7, - 5,15,25,35,35,25,15, 5, - 3, 9,15,21,21,15, 9, 3, - 1, 3, 5, 7, 7, 5, 3, 1, + 4, 12, 20, 28, 28, 20, 12, 4, + 12, 36, 60, 84, 84, 60, 36, 12, + 20, 60,100,140,140,100, 60, 20, + 28, 84,140,196,196,140, 84, 28, + 28, 84,140,196,196,140, 84, 28, + 20, 60,100,140,140,100, 60, 20, + 12, 36, 60, 84, 84, 60, 36, 12, + 4, 12, 20, 28, 28, 20, 12, 4, //error:0.000000 }; //linear *64 static const uint8_t obmc4[16]={ - 4,12,12, 4, -12,36,36,12, -12,36,36,12, - 4,12,12, 4, + 16, 48, 48, 16, + 48,144,144, 48, + 48,144,144, 48, + 16, 48, 48, 16, //error:0.000000 }; @@ -357,20 +368,38 @@ static const uint8_t *obmc_tab[4]={ obmc32, obmc16, obmc8, obmc4 }; +static int scale_mv_ref[MAX_REF_FRAMES][MAX_REF_FRAMES]; + typedef struct BlockNode{ int16_t mx; int16_t my; + uint8_t ref; uint8_t color[3]; uint8_t type; //#define TYPE_SPLIT 1 #define BLOCK_INTRA 1 +#define BLOCK_OPT 2 //#define TYPE_NOCOLOR 4 uint8_t level; //FIXME merge into type? }BlockNode; +static const BlockNode null_block= { //FIXME add border maybe + .color= {128,128,128}, + .mx= 0, + .my= 0, + .ref= 0, + .type= 0, + .level= 0, +}; + #define LOG2_MB_SIZE 4 #define MB_SIZE (1<line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num))) +//#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num))) + +static void iterative_me(SnowContext *s); + +static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, DWTELEM * base_buffer) +{ + int i; + + buf->base_buffer = base_buffer; + buf->line_count = line_count; + buf->line_width = line_width; + buf->data_count = max_allocated_lines; + buf->line = (DWTELEM * *) av_mallocz (sizeof(DWTELEM *) * line_count); + buf->data_stack = (DWTELEM * *) av_malloc (sizeof(DWTELEM *) * max_allocated_lines); + + for (i = 0; i < max_allocated_lines; i++) + { + buf->data_stack[i] = (DWTELEM *) av_malloc (sizeof(DWTELEM) * line_width); + } + + buf->data_stack_top = max_allocated_lines - 1; +} + +static DWTELEM * slice_buffer_load_line(slice_buffer * buf, int line) +{ + int offset; + DWTELEM * buffer; + +// av_log(NULL, AV_LOG_DEBUG, "Cache hit: %d\n", line); + + assert(buf->data_stack_top >= 0); +// assert(!buf->line[line]); + if (buf->line[line]) + return buf->line[line]; + + offset = buf->line_width * line; + buffer = buf->data_stack[buf->data_stack_top]; + buf->data_stack_top--; + buf->line[line] = buffer; + +// av_log(NULL, AV_LOG_DEBUG, "slice_buffer_load_line: line: %d remaining: %d\n", line, buf->data_stack_top + 1); + + return buffer; +} + +static void slice_buffer_release(slice_buffer * buf, int line) +{ + int offset; + DWTELEM * buffer; + + assert(line >= 0 && line < buf->line_count); + assert(buf->line[line]); + + offset = buf->line_width * line; + buffer = buf->line[line]; + buf->data_stack_top++; + buf->data_stack[buf->data_stack_top] = buffer; + buf->line[line] = NULL; + +// av_log(NULL, AV_LOG_DEBUG, "slice_buffer_release: line: %d remaining: %d\n", line, buf->data_stack_top + 1); +} + +static void slice_buffer_flush(slice_buffer * buf) +{ + int i; + for (i = 0; i < buf->line_count; i++) + { + if (buf->line[i]) + { +// av_log(NULL, AV_LOG_DEBUG, "slice_buffer_flush: line: %d \n", i); + slice_buffer_release(buf, i); + } + } +} + +static void slice_buffer_destroy(slice_buffer * buf) +{ + int i; + slice_buffer_flush(buf); + + for (i = buf->data_count - 1; i >= 0; i--) + { + assert(buf->data_stack[i]); + av_freep(&buf->data_stack[i]); + } + assert(buf->data_stack); + av_freep(&buf->data_stack); + assert(buf->line); + av_freep(&buf->line); +} + +#ifdef __sgi // Avoid a name clash on SGI IRIX -#undef qexp +#undef qexp #endif #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0 -static const uint8_t qexp[8]={ - 128, 140, 152, 166, 181, 197, 215, 235 -// 64, 70, 76, 83, 91, 99, 108, 117 -// 32, 35, 38, 41, 45, 49, 54, 59 -// 16, 17, 19, 21, 23, 25, 27, 29 -// 8, 9, 10, 10, 11, 12, 13, 15 -}; +static uint8_t qexp[QROOT]; static inline int mirror(int v, int m){ - if (v<0) return -v; - else if(v>m) return 2*m-v; - else return v; + while((unsigned)v > (unsigned)m){ + v=-v; + if(v<0) v+= 2*m; + } + return v; } static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){ int i; if(v){ - const int a= ABS(v); + const int a= FFABS(v); const int e= av_log2(a); #if 1 - const int el= FFMIN(e, 10); + const int el= FFMIN(e, 10); put_rac(c, state+0, 0); for(i=0; i0) r+=r; } put_rac(c, state+4+log2, 0); - + for(i=log2-1; i>=0; i--){ put_rac(c, state+31-i, (v>>i)&1); } @@ -564,7 +701,7 @@ static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){ log2++; if(log2>0) r+=r; } - + for(i=log2-1; i>=0; i--){ v+= get_rac(c, state+31-i)<>shift), inverse); } - + if(mirror_right){ dst[w*dst_step] = LIFT(src[w*src_step], ((mul*2*ref[w*ref_step]+add)>>shift), inverse); } } +#ifndef lift5 static always_inline void lift5(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){ const int mirror_left= !highpass; const int mirror_right= (width&1) ^ highpass; @@ -608,14 +746,14 @@ static always_inline void lift5(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int ds dst += dst_step; src += src_step; } - + for(i=0; i>4; r += r>>8; dst[i*dst_step] = LIFT(src[i*src_step], ((r+add)>>shift), inverse); } - + if(mirror_right){ int r= 3*2*ref[w*ref_step]; r += r>>4; @@ -623,11 +761,37 @@ static always_inline void lift5(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int ds dst[w*dst_step] = LIFT(src[w*src_step], ((r+add)>>shift), inverse); } } +#endif + +#ifndef liftS +static always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){ + const int mirror_left= !highpass; + const int mirror_right= (width&1) ^ highpass; + const int w= (width>>1) - 1 + (highpass & width); + int i; + + assert(shift == 4); +#define LIFTS(src, ref, inv) ((inv) ? (src) - (((ref) - 4*(src))>>shift): (16*4*(src) + 4*(ref) + 8 + (5<<27))/(5*16) - (1<<23)) + if(mirror_left){ + dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse); + dst += dst_step; + src += src_step; + } + + for(i=0; i>1; const int w2= (width+1)>>1; - int A1,A2,A3,A4, x; + int x; inplace_lift(b, width, COEFFS1, N1, SHIFT1, LX1, 0); inplace_lift(b, width, COEFFS2, N2, SHIFT2, LX0, 0); inplace_lift(b, width, COEFFS3, N3, SHIFT3, LX1, 0); inplace_lift(b, width, COEFFS4, N4, SHIFT4, LX0, 0); - + for(x=0; x>1; - int A1,A2,A3,A4, x; + int x; const int w2= (width+1)>>1; memcpy(temp, b, width*sizeof(int)); @@ -816,7 +980,7 @@ static void horizontal_composeX(DWTELEM *b, int width){ static void spatial_decomposeX(DWTELEM *buffer, int width, int height, int stride){ int x, y; - + for(y=0; y>1; - int A1,A2,A3,A4, x; + int x; const int w2= (width+1)>>1; for(x=0; x>2; b[width -1] = A3; b[width2-1] = A2; -#else + } +#else lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0); lift(b , temp , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 0); #endif @@ -900,7 +1067,7 @@ static void horizontal_decompose53i(DWTELEM *b, int width){ static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>1; } @@ -908,7 +1075,7 @@ static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int w static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>2; } @@ -918,101 +1085,32 @@ static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int str int y; DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride; DWTELEM *b1= buffer + mirror(-2 , height-1)*stride; - + for(y=-2; y>1; lift (temp+w2, b +1, b , 1, 2, 2, width, -W_AM, W_AO, W_AS, 1, 0); - lift (temp , b , temp+w2, 1, 2, 1, width, -W_BM, W_BO, W_BS, 0, 0); + liftS(temp , b , temp+w2, 1, 2, 1, width, -W_BM, W_BO, W_BS, 0, 0); lift5(b +w2, temp+w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 0); lift (b , temp , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 0); } @@ -1020,7 +1118,7 @@ static void horizontal_decompose97i(DWTELEM *b, int width){ static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_AS; } @@ -1028,7 +1126,7 @@ static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int w static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_CS; @@ -1043,15 +1141,19 @@ static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int w static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_BS; +#else + b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + 8*5 + (5<<27)) / (5*16) - (1<<23); +#endif } } static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_DS; } @@ -1063,28 +1165,28 @@ static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int str DWTELEM *b1= buffer + mirror(-4 , height-1)*stride; DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride; DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride; - + for(y=-4; y400){ STOP_TIMER("horizontal_decompose97i") }} - + {START_TIMER - if(b3 <= b5) vertical_decompose97iH0(b3, b4, b5, width); - if(b2 <= b4) vertical_decompose97iL0(b2, b3, b4, width); - if(b1 <= b3) vertical_decompose97iH1(b1, b2, b3, width); - if(b0 <= b2) vertical_decompose97iL1(b0, b1, b2, width); + if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width); + if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width); + if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width); + if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width); if(width>400){ STOP_TIMER("vertical_decompose97i") }} - + b0=b2; b1=b3; b2=b4; @@ -1094,12 +1196,12 @@ STOP_TIMER("vertical_decompose97i") void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ int level; - + for(level=0; level>level, height>>level, stride<>level, height>>level, stride<>level, height>>level, stride<>level, height>>level, stride<>level, height>>level, stride<>level, height>>level, stride<>1; const int w2= (width+1)>>1; - int A1,A2,A3,A4, x; + int x; #if 0 + int A1,A2,A3,A4; A2= temp[1 ]; A4= temp[0 ]; A1= temp[0+width2]; @@ -1138,7 +1241,7 @@ static void horizontal_compose53i(DWTELEM *b, int width){ A2 += (A1 + A3 + 2)>>2; b[width -1] = A3; b[width2-1] = A2; -#else +#else lift(temp , b , b+w2, 1, 1, 1, width, 1, 2, 2, 0, 1); lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1); #endif @@ -1152,7 +1255,7 @@ static void horizontal_compose53i(DWTELEM *b, int width){ static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>1; } @@ -1160,50 +1263,90 @@ static void vertical_compose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int wid static void vertical_compose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>2; } } -static void spatial_compose53i(DWTELEM *buffer, int width, int height, int stride){ - int y; - DWTELEM *b0= buffer + mirror(-1-1, height-1)*stride; - DWTELEM *b1= buffer + mirror(-1 , height-1)*stride; - - for(y=-1; y<=height; y+=2){ - DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; - DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; +static void spatial_compose53i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){ + cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line); + cs->b1 = slice_buffer_get_line(sb, mirror(-1 , height-1) * stride_line); + cs->y = -1; +} + +static void spatial_compose53i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){ + cs->b0 = buffer + mirror(-1-1, height-1)*stride; + cs->b1 = buffer + mirror(-1 , height-1)*stride; + cs->y = -1; +} + +static void spatial_compose53i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ + int y= cs->y; + + DWTELEM *b0= cs->b0; + DWTELEM *b1= cs->b1; + DWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line); + DWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line); {START_TIMER - if(b1 <= b3) vertical_compose53iL0(b1, b2, b3, width); - if(b0 <= b2) vertical_compose53iH0(b0, b1, b2, width); + if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); + if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); STOP_TIMER("vertical_compose53i*")} {START_TIMER - if(y-1 >= 0) horizontal_compose53i(b0, width); - if(b0 <= b2) horizontal_compose53i(b1, width); + if(y-1<(unsigned)height) horizontal_compose53i(b0, width); + if(y+0<(unsigned)height) horizontal_compose53i(b1, width); STOP_TIMER("horizontal_compose53i")} - b0=b2; - b1=b3; - } -} + cs->b0 = b2; + cs->b1 = b3; + cs->y += 2; +} + +static void spatial_compose53i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){ + int y= cs->y; + DWTELEM *b0= cs->b0; + DWTELEM *b1= cs->b1; + DWTELEM *b2= buffer + mirror(y+1, height-1)*stride; + DWTELEM *b3= buffer + mirror(y+2, height-1)*stride; + +{START_TIMER + if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width); + if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width); +STOP_TIMER("vertical_compose53i*")} + +{START_TIMER + if(y-1<(unsigned)height) horizontal_compose53i(b0, width); + if(y+0<(unsigned)height) horizontal_compose53i(b1, width); +STOP_TIMER("horizontal_compose53i")} + + cs->b0 = b2; + cs->b1 = b3; + cs->y += 2; +} + +static void spatial_compose53i(DWTELEM *buffer, int width, int height, int stride){ + dwt_compose_t cs; + spatial_compose53i_init(&cs, buffer, height, stride); + while(cs.y <= height) + spatial_compose53i_dy(&cs, buffer, width, height, stride); +} + - -static void horizontal_compose97i(DWTELEM *b, int width){ +void ff_snow_horizontal_compose97i(DWTELEM *b, int width){ DWTELEM temp[width]; const int w2= (width+1)>>1; lift (temp , b , b +w2, 1, 1, 1, width, W_DM, W_DO, W_DS, 0, 1); lift5(temp+w2, b +w2, temp , 1, 1, 1, width, W_CM, W_CO, W_CS, 1, 1); - lift (b , temp , temp+w2, 2, 1, 1, width, -W_BM, W_BO, W_BS, 0, 1); + liftS(b , temp , temp+w2, 2, 1, 1, width, -W_BM, W_BO, W_BS, 0, 1); lift (b+1 , temp+w2, b , 2, 1, 2, width, -W_AM, W_AO, W_AS, 1, 1); } static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_AS; } @@ -1211,7 +1354,7 @@ static void vertical_compose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int wid static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_CS; @@ -1226,73 +1369,213 @@ static void vertical_compose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int wid static void vertical_compose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_BS; +#else + b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS; +#endif } } static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){ int i; - + for(i=0; i>W_DS; } } -static void spatial_compose97i(DWTELEM *buffer, int width, int height, int stride){ - int y; - DWTELEM *b0= buffer + mirror(-3-1, height-1)*stride; - DWTELEM *b1= buffer + mirror(-3 , height-1)*stride; - DWTELEM *b2= buffer + mirror(-3+1, height-1)*stride; - DWTELEM *b3= buffer + mirror(-3+2, height-1)*stride; +void ff_snow_vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){ + int i; - for(y=-3; y<=height; y+=2){ - DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; - DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; + for(i=0; i>W_DS; +#ifdef lift5 + b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS; +#else + r= 3*(b2[i] + b4[i]); + r+= r>>4; + r+= r>>8; + b3[i] -= (r+W_CO)>>W_CS; +#endif +#ifdef liftS + b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS; +#else + b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS; +#endif + b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS; + } +} + +static void spatial_compose97i_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int height, int stride_line){ + cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line); + cs->b1 = slice_buffer_get_line(sb, mirror(-3 , height-1) * stride_line); + cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line); + cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line); + cs->y = -3; +} + +static void spatial_compose97i_init(dwt_compose_t *cs, DWTELEM *buffer, int height, int stride){ + cs->b0 = buffer + mirror(-3-1, height-1)*stride; + cs->b1 = buffer + mirror(-3 , height-1)*stride; + cs->b2 = buffer + mirror(-3+1, height-1)*stride; + cs->b3 = buffer + mirror(-3+2, height-1)*stride; + cs->y = -3; +} + +static void spatial_compose97i_dy_buffered(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ + int y = cs->y; + + DWTELEM *b0= cs->b0; + DWTELEM *b1= cs->b1; + DWTELEM *b2= cs->b2; + DWTELEM *b3= cs->b3; + DWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line); + DWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line); + +{START_TIMER + if(y>0 && y+4vertical_compose97i(b0, b1, b2, b3, b4, b5, width); + }else{ + if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); + if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); + if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width); + if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width); + } +if(width>400){ +STOP_TIMER("vertical_compose97i")}} + +{START_TIMER + if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width); + if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width); +if(width>400 && y+0<(unsigned)height){ +STOP_TIMER("horizontal_compose97i")}} + + cs->b0=b2; + cs->b1=b3; + cs->b2=b4; + cs->b3=b5; + cs->y += 2; +} + +static void spatial_compose97i_dy(dwt_compose_t *cs, DWTELEM *buffer, int width, int height, int stride){ + int y = cs->y; + DWTELEM *b0= cs->b0; + DWTELEM *b1= cs->b1; + DWTELEM *b2= cs->b2; + DWTELEM *b3= cs->b3; + DWTELEM *b4= buffer + mirror(y+3, height-1)*stride; + DWTELEM *b5= buffer + mirror(y+4, height-1)*stride; - if(stride == width && y+4 < height && 0){ - int x; - for(x=0; x400){ STOP_TIMER("vertical_compose97i")}} {START_TIMER - if(y-1>= 0) horizontal_compose97i(b0, width); - if(b0 <= b2) horizontal_compose97i(b1, width); + if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width); + if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width); if(width>400 && b0 <= b2){ STOP_TIMER("horizontal_compose97i")}} - - b0=b2; - b1=b3; - b2=b4; - b3=b5; - } + + cs->b0=b2; + cs->b1=b3; + cs->b2=b4; + cs->b3=b5; + cs->y += 2; +} + +static void spatial_compose97i(DWTELEM *buffer, int width, int height, int stride){ + dwt_compose_t cs; + spatial_compose97i_init(&cs, buffer, height, stride); + while(cs.y <= height) + spatial_compose97i_dy(&cs, buffer, width, height, stride); } -void ff_spatial_idwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){ +static void ff_spatial_idwt_buffered_init(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){ int level; + for(level=decomposition_count-1; level>=0; level--){ + switch(type){ + case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<>level, stride_line<>level, height>>level, stride<=0; level--){ switch(type){ - case 0: spatial_compose97i(buffer, width>>level, height>>level, stride<>level, height>>level, stride<>level, height>>level, stride<>level, stride<>level, stride<>level, height>>level, stride<=0; level--){ + while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ + switch(type){ + case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<>level, height>>level, stride<=0; level--){ + while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ + switch(type){ + case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<>level, height>>level, stride_line<=0; level--) + spatial_composeX (buffer, width>>level, height>>level, stride<width; const int h= b->height; int x, y; @@ -1301,7 +1584,8 @@ static void encode_subband_c0run(SnowContext *s, SubBand *b, DWTELEM *src, DWTEL int run=0; int runs[w*h]; int run_index=0; - + int max_index; + for(y=0; y>1; int py= y>>1; - if(pxparent->width && pyparent->height) + if(pxparent->width && pyparent->height) p= parent[px + py*2*stride]; } if(!(/*ll|*/l|lt|t|rt|p)){ @@ -1340,13 +1624,20 @@ static void encode_subband_c0run(SnowContext *s, SubBand *b, DWTELEM *src, DWTEL } } } + max_index= run_index; runs[run_index++]= run; run_index=0; run= runs[run_index++]; - put_symbol2(&s->c, b->state[1], run, 3); - + put_symbol2(&s->c, b->state[30], max_index, 0); + if(run_index <= max_index) + put_symbol2(&s->c, b->state[1], run, 3); + for(y=0; yc.bytestream_end - s->c.bytestream < w*40){ + av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); + return -1; + } for(x=0; x>1; int py= y>>1; - if(pxparent->width && pyparent->height) + if(pxparent->width && pyparent->height) p= parent[px + py*2*stride]; } if(/*ll|*/l|lt|t|rt|p){ - int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); + int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p)); put_rac(&s->c, &b->state[0][context], !!v); }else{ if(!run){ run= runs[run_index++]; - put_symbol2(&s->c, b->state[1], run, 3); + if(run_index <= max_index) + put_symbol2(&s->c, b->state[1], run, 3); assert(v); }else{ run--; @@ -1390,159 +1682,194 @@ static void encode_subband_c0run(SnowContext *s, SubBand *b, DWTELEM *src, DWTEL } } if(v){ - int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); + int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p)); + int l2= 2*FFABS(l) + (l<0); + int t2= 2*FFABS(t) + (t<0); - put_symbol2(&s->c, b->state[context + 2], ABS(v)-1, context-4); - put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3b[l&0xFF] + 3*quant3b[t&0xFF]], v<0); + put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4); + put_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l2&0xFF] + 3*quant3bA[t2&0xFF]], v<0); } } } } + return 0; } -static void encode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){ +static int encode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){ // encode_subband_qtree(s, b, src, parent, stride, orientation); // encode_subband_z0run(s, b, src, parent, stride, orientation); - encode_subband_c0run(s, b, src, parent, stride, orientation); + return encode_subband_c0run(s, b, src, parent, stride, orientation); // encode_subband_dzr(s, b, src, parent, stride, orientation); } -static inline void decode_subband(SnowContext *s, SubBand *b, DWTELEM *src, DWTELEM *parent, int stride, int orientation){ +static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){ const int w= b->width; const int h= b->height; int x,y; - const int qlog= clip(s->qlog + b->qlog, 0, 128); - int qmul= qexp[qlog&7]<<(qlog>>3); - int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; - - START_TIMER - - if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){ - qadd= 0; - qmul= 1<height; y++) - memset(&src[y*stride], 0, b->width*sizeof(DWTELEM)); - - run= get_symbol2(&s->c, b->state[1], 3); + int run, runs; + x_and_coeff *xc= b->x_coeff; + x_and_coeff *prev_xc= NULL; + x_and_coeff *prev2_xc= xc; + x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL; + x_and_coeff *prev_parent_xc= parent_xc; + + runs= get_symbol2(&s->c, b->state[30], 0); + if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); + else run= INT_MAX; + for(y=0; yx[prev_index] == 0){ - rt= b->coeff[prev_index]; + if(y && prev_xc->x == 0){ + rt= prev_xc->coeff; } for(x=0; xx[prev_index] <= x) - prev_index++; - if(b->x[prev_index] == x + 1) - rt= b->coeff[prev_index]; + if(prev_xc->x <= x) + prev_xc++; + if(prev_xc->x == x + 1) + rt= prev_xc->coeff; else rt=0; } - if(parent){ - if(x>>1 > b->parent->x[parent_index]){ - parent_index++; + if(parent_xc){ + if(x>>1 > parent_xc->x){ + parent_xc++; } - if(x>>1 == b->parent->x[parent_index]){ - p= b->parent->coeff[parent_index]; + if(x>>1 == parent_xc->x){ + p= parent_xc->coeff; } } if(/*ll|*/l|lt|t|rt|p){ - int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); + int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1)); v=get_rac(&s->c, &b->state[0][context]); + if(v){ + v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1); + v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]); + + xc->x=x; + (xc++)->coeff= v; + } }else{ if(!run){ - run= get_symbol2(&s->c, b->state[1], 3); - v=1; + if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3); + else run= INT_MAX; + v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1); + v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]); + + xc->x=x; + (xc++)->coeff= v; }else{ + int max_run; run--; v=0; - if(y && parent){ - int max_run; - - max_run= FFMIN(run, b->x[prev_index] - x - 2); - max_run= FFMIN(max_run, 2*b->parent->x[parent_index] - x - 1); - x+= max_run; - run-= max_run; - } - } - } - if(v){ - int context= av_log2(/*ABS(ll) + */3*ABS(l) + ABS(lt) + 2*ABS(t) + ABS(rt) + ABS(p)); - v= get_symbol2(&s->c, b->state[context + 2], context-4) + 1; - if(get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3b[l&0xFF] + 3*quant3b[t&0xFF]])){ - src[x + y*stride]=-(( v*qmul + qadd)>>(QEXPSHIFT)); - v= -v; - }else{ - src[x + y*stride]= (( v*qmul + qadd)>>(QEXPSHIFT)); + if(y) max_run= FFMIN(run, prev_xc->x - x - 2); + else max_run= FFMIN(run, w-x-1); + if(parent_xc) + max_run= FFMIN(max_run, 2*parent_xc->x - x - 1); + x+= max_run; + run-= max_run; } - b->x[index]=x; //FIXME interleave x/coeff - b->coeff[index++]= v; } } - b->x[index++]= w+1; //end marker - prev_index= prev2_index; - prev2_index= index; - - if(parent){ - while(b->parent->x[parent_index] != b->parent->width+1) - parent_index++; - parent_index++; + (xc++)->x= w+1; //end marker + prev_xc= prev2_xc; + prev2_xc= xc; + + if(parent_xc){ if(y&1){ - prev_parent_index= parent_index; + while(parent_xc->x != parent->width+1) + parent_xc++; + parent_xc++; + prev_parent_xc= parent_xc; }else{ - parent_index= prev_parent_index; + parent_xc= prev_parent_xc; } } } - b->x[index++]= w+1; //end marker - if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ - STOP_TIMER("decode_subband") - } - - return; - } -} - -static void reset_contexts(SnowContext *s){ - int plane_index, level, orientation; - for(plane_index=0; plane_index<3; plane_index++){ - for(level=0; levelspatial_decomposition_count; level++){ - for(orientation=level ? 1:0; orientation<4; orientation++){ - memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state)); - } - } + (xc++)->x= w+1; //end marker } - memset(s->header_state, MID_STATE, sizeof(s->header_state)); - memset(s->block_state, MID_STATE, sizeof(s->block_state)); +} + +static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){ + const int w= b->width; + int y; + const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); + int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); + int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; + int new_index = 0; + + START_TIMER + + if(b->buf == s->spatial_dwt_buffer || s->qlog == LOSSLESS_QLOG){ + qadd= 0; + qmul= 1<stride_line + b->buf_y_offset) + b->buf_x_offset; + memset(line, 0, b->width*sizeof(DWTELEM)); + v = b->x_coeff[new_index].coeff; + x = b->x_coeff[new_index++].x; + while(x < w) + { + register int t= ( (v>>1)*qmul + qadd)>>QEXPSHIFT; + register int u= -(v&1); + line[x] = (t^u) - u; + + v = b->x_coeff[new_index].coeff; + x = b->x_coeff[new_index++].x; + } + } + if(w > 200 && start_y != 0/*level+1 == s->spatial_decomposition_count*/){ + STOP_TIMER("decode_subband") + } + + /* Save our variables for the next slice. */ + save_state[0] = new_index; + + return; +} + +static void reset_contexts(SnowContext *s){ + int plane_index, level, orientation; + + for(plane_index=0; plane_index<3; plane_index++){ + for(level=0; levelspatial_decomposition_count; level++){ + for(orientation=level ? 1:0; orientation<4; orientation++){ + memset(s->plane[plane_index].band[level][orientation].state, MID_STATE, sizeof(s->plane[plane_index].band[level][orientation].state)); + } + } + } + memset(s->header_state, MID_STATE, sizeof(s->header_state)); + memset(s->block_state, MID_STATE, sizeof(s->block_state)); } static int alloc_blocks(SnowContext *s){ int w= -((-s->avctx->width )>>LOG2_MB_SIZE); int h= -((-s->avctx->height)>>LOG2_MB_SIZE); - + s->b_width = w; s->b_height= h; - + s->block= av_mallocz(w * h * sizeof(BlockNode) << (s->block_max_depth*2)); return 0; } @@ -1575,7 +1902,7 @@ static int pix_sum(uint8_t * pix, int line_size, int w) static int pix_norm1(uint8_t * pix, int line_size, int w) { int s, i, j; - uint32_t *sq = squareTbl + 256; + uint32_t *sq = ff_squareTbl + 256; s = 0; for (i = 0; i < w; i++) { @@ -1588,19 +1915,20 @@ static int pix_norm1(uint8_t * pix, int line_size, int w) return s; } -static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int type){ +static inline void set_blocks(SnowContext *s, int level, int x, int y, int l, int cb, int cr, int mx, int my, int ref, int type){ const int w= s->b_width << s->block_max_depth; const int rem_depth= s->block_max_depth - level; const int index= (x + y*w) << rem_depth; const int block_w= 1<ref_frames == 1){ + *mx = mid_pred(left->mx, top->mx, tr->mx); + *my = mid_pred(left->my, top->my, tr->my); + }else{ + const int *scale = scale_mv_ref[ref]; + *mx = mid_pred(left->mx * scale[left->ref] + 128 >>8, + top ->mx * scale[top ->ref] + 128 >>8, + tr ->mx * scale[tr ->ref] + 128 >>8); + *my = mid_pred(left->my * scale[left->ref] + 128 >>8, + top ->my * scale[top ->ref] + 128 >>8, + tr ->my * scale[tr ->ref] + 128 >>8); + } +} + //FIXME copy&paste #define P_LEFT P[1] #define P_TOP P[2] @@ -1647,13 +1991,6 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ const int rem_depth= s->block_max_depth - level; const int index= (x + y*w) << rem_depth; const int block_w= 1<<(LOG2_MB_SIZE - level); - static BlockNode null_block= { //FIXME add border maybe - .color= {128,128,128}, - .mx= 0, - .my= 0, - .type= 0, - .level= 0, - }; int trx= (x+1)<block[index-1] : &null_block; @@ -1665,42 +2002,31 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ int pl = left->color[0]; int pcb= left->color[1]; int pcr= left->color[2]; - int pmx= mid_pred(left->mx, top->mx, tr->mx); - int pmy= mid_pred(left->my, top->my, tr->my); + int pmx, pmy; int mx=0, my=0; - int l,cr,cb, i; + int l,cr,cb; const int stride= s->current_picture.linesize[0]; const int uvstride= s->current_picture.linesize[1]; - const int instride= s->input_picture.linesize[0]; - const int uvinstride= s->input_picture.linesize[1]; - uint8_t *new_l = s->input_picture.data[0] + (x + y* instride)*block_w; - uint8_t *new_cb= s->input_picture.data[1] + (x + y*uvinstride)*block_w/2; - uint8_t *new_cr= s->input_picture.data[2] + (x + y*uvinstride)*block_w/2; - uint8_t current_mb[3][stride*block_w]; - uint8_t *current_data[3]= {¤t_mb[0][0], ¤t_mb[1][0], ¤t_mb[2][0]}; + uint8_t *current_data[3]= { s->input_picture.data[0] + (x + y* stride)*block_w, + s->input_picture.data[1] + (x + y*uvstride)*block_w/2, + s->input_picture.data[2] + (x + y*uvstride)*block_w/2}; int P[10][2]; int16_t last_mv[3][2]; int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused const int shift= 1+qpel; MotionEstContext *c= &s->m.me; - int mx_context= av_log2(2*ABS(left->mx - top->mx)); - int my_context= av_log2(2*ABS(left->my - top->my)); + int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); + int mx_context= av_log2(2*FFABS(left->mx - top->mx)); + int my_context= av_log2(2*FFABS(left->my - top->my)); int s_context= 2*left->level + 2*top->level + tl->level + tr->level; + int ref, best_ref, ref_score, ref_mx, ref_my; assert(sizeof(s->block_state) >= 256); if(s->keyframe){ - set_blocks(s, level, x, y, pl, pcb, pcr, pmx, pmy, BLOCK_INTRA); + set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); return 0; } - //FIXME optimize - for(i=0; i>1; i++) - memcpy(¤t_mb[1][0] + uvstride*i, new_cb + uvinstride*i, block_w>>1); - for(i=0; i>1; i++) - memcpy(¤t_mb[2][0] + uvstride*i, new_cr + uvinstride*i, block_w>>1); - // clip predictors / edge ? P_LEFT[0]= left->mx; @@ -1709,36 +2035,34 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ P_TOP [1]= top->my; P_TOPRIGHT[0]= tr->mx; P_TOPRIGHT[1]= tr->my; - + last_mv[0][0]= s->block[index].mx; last_mv[0][1]= s->block[index].my; last_mv[1][0]= right->mx; last_mv[1][1]= right->my; last_mv[2][0]= bottom->mx; last_mv[2][1]= bottom->my; - + s->m.mb_stride=2; - s->m.mb_x= + s->m.mb_x= s->m.mb_y= 0; s->m.me.skip= 0; - init_ref(c, current_data, s->last_picture.data, NULL, block_w*x, block_w*y, 0); - assert(s->m.me. stride == stride); assert(s->m.me.uvstride == uvstride); - + c->penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp); c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp); c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp); c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV; - + c->xmin = - x*block_w - 16+2; c->ymin = - y*block_w - 16+2; c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2; c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-2; if(P_LEFT[0] > (c->xmax<xmax< (c->ymax<ymax< (c->ymax<ymax< (c->xmax<xmax< (c->ymax<ymax<xmin<xmin<pred_y = P_MEDIAN[1]; } - score= ff_epzs_motion_search(&s->m, &mx, &my, P, 0, /*ref_index*/ 0, last_mv, - (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w); - - assert(mx >= c->xmin); - assert(mx <= c->xmax); - assert(my >= c->ymin); - assert(my <= c->ymax); - - score= s->m.me.sub_motion_search(&s->m, &mx, &my, score, 0, 0, level-LOG2_MB_SIZE+4, block_w); - score= ff_get_mb_score(&s->m, mx, my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0); + score= INT_MAX; + best_ref= 0; + for(ref=0; refref_frames; ref++){ + init_ref(c, current_data, s->last_picture[ref].data, NULL, block_w*x, block_w*y, 0); + + ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv, + (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w); + + assert(ref_mx >= c->xmin); + assert(ref_mx <= c->xmax); + assert(ref_my >= c->ymin); + assert(ref_my <= c->ymax); + + ref_score= s->m.me.sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w); + ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0); + ref_score+= 2*av_log2(2*ref)*c->penalty_factor; + if(s->ref_mvs[ref]){ + s->ref_mvs[ref][index][0]= ref_mx; + s->ref_mvs[ref][index][1]= ref_my; + s->ref_scores[ref][index]= ref_score; + } + if(score > ref_score){ + score= ref_score; + best_ref= ref; + mx= ref_mx; + my= ref_my; + } + } //FIXME if mb_cmp != SSE then intra cant be compared currently and mb_penalty vs. lambda2 - + // subpel search pc= s->c; pc.bytestream_start= @@ -1777,8 +2119,11 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ if(level!=s->block_max_depth) put_rac(&pc, &p_state[4 + s_context], 1); put_rac(&pc, &p_state[1 + left->type + top->type], 0); - put_symbol(&pc, &p_state[128 + 32*mx_context], mx - pmx, 1); - put_symbol(&pc, &p_state[128 + 32*my_context], my - pmy, 1); + if(s->ref_frames > 1) + put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0); + pred_mv(s, &pmx, &pmy, best_ref, left, top, tr); + put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1); + put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1); p_len= pc.bytestream - pc.bytestream_start; score += (s->lambda2*(p_len*8 + (pc.outstanding_count - s->c.outstanding_count)*8 @@ -1786,15 +2131,15 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ ))>>FF_LAMBDA_SHIFT; block_s= block_w*block_w; - sum = pix_sum(¤t_mb[0][0], stride, block_w); + sum = pix_sum(current_data[0], stride, block_w); l= (sum + block_s/2)/block_s; - iscore = pix_norm1(¤t_mb[0][0], stride, block_w) - 2*l*sum + l*l*block_s; - + iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s; + block_s= block_w*block_w>>2; - sum = pix_sum(¤t_mb[1][0], uvstride, block_w>>1); + sum = pix_sum(current_data[1], uvstride, block_w>>1); cb= (sum + block_s/2)/block_s; // iscore += pix_norm1(¤t_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s; - sum = pix_sum(¤t_mb[2][0], uvstride, block_w>>1); + sum = pix_sum(current_data[2], uvstride, block_w>>1); cr= (sum + block_s/2)/block_s; // iscore += pix_norm1(¤t_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s; @@ -1828,7 +2173,7 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ else c->scene_change_score+= s->m.qscale; } - + if(level!=s->block_max_depth){ put_rac(&s->c, &s->block_state[4 + s_context], 0); score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0); @@ -1836,17 +2181,18 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1); score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1); score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead - + if(score2 < score && score2 < iscore) return score2; } - + if(iscore < score){ + pred_mv(s, &pmx, &pmy, 0, left, top, tr); memcpy(pbbak, i_buffer, i_len); s->c= ic; s->c.bytestream_start= pbbak_start; s->c.bytestream= pbbak + i_len; - set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, BLOCK_INTRA); + set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA); memcpy(s->block_state, i_state, sizeof(s->block_state)); return iscore; }else{ @@ -1854,32 +2200,87 @@ static int encode_q_branch(SnowContext *s, int level, int x, int y){ s->c= pc; s->c.bytestream_start= pbbak_start; s->c.bytestream= pbbak + p_len; - set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, 0); + set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0); memcpy(s->block_state, p_state, sizeof(s->block_state)); return score; } } +static always_inline int same_block(BlockNode *a, BlockNode *b){ + if((a->type&BLOCK_INTRA) && (b->type&BLOCK_INTRA)){ + return !((a->color[0] - b->color[0]) | (a->color[1] - b->color[1]) | (a->color[2] - b->color[2])); + }else{ + return !((a->mx - b->mx) | (a->my - b->my) | (a->ref - b->ref) | ((a->type ^ b->type)&BLOCK_INTRA)); + } +} + +static void encode_q_branch2(SnowContext *s, int level, int x, int y){ + const int w= s->b_width << s->block_max_depth; + const int rem_depth= s->block_max_depth - level; + const int index= (x + y*w) << rem_depth; + int trx= (x+1)<block[index]; + BlockNode *left = x ? &s->block[index-1] : &null_block; + BlockNode *top = y ? &s->block[index-w] : &null_block; + BlockNode *tl = y && x ? &s->block[index-w-1] : left; + BlockNode *tr = y && trxblock[index-w+(1<color[0]; + int pcb= left->color[1]; + int pcr= left->color[2]; + int pmx, pmy; + int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); + int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref; + int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref; + int s_context= 2*left->level + 2*top->level + tl->level + tr->level; + + if(s->keyframe){ + set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA); + return; + } + + if(level!=s->block_max_depth){ + if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){ + put_rac(&s->c, &s->block_state[4 + s_context], 1); + }else{ + put_rac(&s->c, &s->block_state[4 + s_context], 0); + encode_q_branch2(s, level+1, 2*x+0, 2*y+0); + encode_q_branch2(s, level+1, 2*x+1, 2*y+0); + encode_q_branch2(s, level+1, 2*x+0, 2*y+1); + encode_q_branch2(s, level+1, 2*x+1, 2*y+1); + return; + } + } + if(b->type & BLOCK_INTRA){ + pred_mv(s, &pmx, &pmy, 0, left, top, tr); + put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1); + put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1); + put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1); + put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1); + set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA); + }else{ + pred_mv(s, &pmx, &pmy, b->ref, left, top, tr); + put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0); + if(s->ref_frames > 1) + put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0); + put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1); + put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1); + set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0); + } +} + static void decode_q_branch(SnowContext *s, int level, int x, int y){ const int w= s->b_width << s->block_max_depth; const int rem_depth= s->block_max_depth - level; const int index= (x + y*w) << rem_depth; - static BlockNode null_block= { //FIXME add border maybe - .color= {128,128,128}, - .mx= 0, - .my= 0, - .type= 0, - .level= 0, - }; int trx= (x+1)<block[index-1] : &null_block; BlockNode *top = y ? &s->block[index-w] : &null_block; BlockNode *tl = y && x ? &s->block[index-w-1] : left; BlockNode *tr = y && trxblock[index-w+(1<level + 2*top->level + tl->level + tr->level; - + if(s->keyframe){ - set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, BLOCK_INTRA); + set_blocks(s, level, x, y, null_block.color[0], null_block.color[1], null_block.color[2], null_block.mx, null_block.my, null_block.ref, BLOCK_INTRA); return; } @@ -1890,20 +2291,26 @@ static void decode_q_branch(SnowContext *s, int level, int x, int y){ int cr= left->color[2]; int mx= mid_pred(left->mx, top->mx, tr->mx); int my= mid_pred(left->my, top->my, tr->my); - int mx_context= av_log2(2*ABS(left->mx - top->mx)) + 0*av_log2(2*ABS(tr->mx - top->mx)); - int my_context= av_log2(2*ABS(left->my - top->my)) + 0*av_log2(2*ABS(tr->my - top->my)); - + int ref = 0; + int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref); + int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 0*av_log2(2*FFABS(tr->mx - top->mx)); + int my_context= av_log2(2*FFABS(left->my - top->my)) + 0*av_log2(2*FFABS(tr->my - top->my)); + type= get_rac(&s->c, &s->block_state[1 + left->type + top->type]) ? BLOCK_INTRA : 0; if(type){ + pred_mv(s, &mx, &my, 0, left, top, tr); l += get_symbol(&s->c, &s->block_state[32], 1); cb+= get_symbol(&s->c, &s->block_state[64], 1); cr+= get_symbol(&s->c, &s->block_state[96], 1); }else{ - mx+= get_symbol(&s->c, &s->block_state[128 + 32*mx_context], 1); - my+= get_symbol(&s->c, &s->block_state[128 + 32*my_context], 1); + if(s->ref_frames > 1) + ref= get_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], 0); + pred_mv(s, &mx, &my, ref, left, top, tr); + mx+= get_symbol(&s->c, &s->block_state[128 + 32*(mx_context + 16*!!ref)], 1); + my+= get_symbol(&s->c, &s->block_state[128 + 32*(my_context + 16*!!ref)], 1); } - set_blocks(s, level, x, y, l, cb, cr, mx, my, type); + set_blocks(s, level, x, y, l, cb, cr, mx, my, ref, type); }else{ decode_q_branch(s, level+1, 2*x+0, 2*y+0); decode_q_branch(s, level+1, 2*x+1, 2*y+0); @@ -1912,14 +2319,24 @@ static void decode_q_branch(SnowContext *s, int level, int x, int y){ } } -static void encode_blocks(SnowContext *s){ +static void encode_blocks(SnowContext *s, int search){ int x, y; int w= s->b_width; int h= s->b_height; + if(s->avctx->me_method == ME_ITER && !s->keyframe && search) + iterative_me(s); + for(y=0; yc.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit + av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n"); + return; + } for(x=0; xavctx->me_method == ME_ITER || !search) + encode_q_branch2(s, 0, x, y); + else + encode_q_branch (s, 0, x, y); } } } @@ -1955,8 +2372,13 @@ START_TIMER // if(b_w==16) am= 8*(a1+a2); - if(dx<8) tmp[x]= (32*a2*( 8-dx) + am* dx + 128)>>8; - else tmp[x]= ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8; + if(dx<8) am = (32*a2*( 8-dx) + am* dx + 128)>>8; + else am = ( am*(16-dx) + 32*a3*(dx-8) + 128)>>8; + + /* FIXME Try increasing tmp buffer to 16 bits and not clipping here. Should give marginally better results. - Robert*/ + if(am&(~255)) am= ~(am>>31); + + tmp[x] = am; /* if (dx< 4) tmp[x + y*stride]= (16*a1*( 4-dx) + aL* dx + 32)>>6; else if(dx< 8) tmp[x + y*stride]= ( aL*( 8-dx) + am*(dx- 4) + 32)>>6; @@ -1967,7 +2389,7 @@ START_TIMER src += stride; } tmp -= (b_h+5)*stride; - + for(y=0; y < b_h; y++){ for(x=0; x < b_w; x++){ int a0= tmp[x + 0*stride]; @@ -1980,12 +2402,15 @@ START_TIMER // int am= 18*(a2+a3) - 2*(a1+a4); /* int aL= (-7*a0 + 105*a1 + 35*a2 - 5*a3)>>3; int aR= (-7*a3 + 105*a2 + 35*a1 - 5*a0)>>3;*/ - + // if(b_w==16) am= 8*(a1+a2); - if(dy<8) dst[x]= (32*a2*( 8-dy) + am* dy + 128)>>8; - else dst[x]= ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8; + if(dy<8) am = (32*a2*( 8-dy) + am* dy + 128)>>8; + else am = ( am*(16-dy) + 32*a3*(dy-8) + 128)>>8; + if(am&(~255)) am= ~(am>>31); + + dst[x] = am; /* if (dy< 4) tmp[x + y*stride]= (16*a1*( 4-dy) + aL* dy + 32)>>6; else if(dy< 8) tmp[x + y*stride]= ( aL*( 8-dy) + am*(dy- 4) + 32)>>6; else if(dy<12) tmp[x + y*stride]= ( am*(12-dy) + aR*(dy- 8) + 32)>>6; @@ -2013,21 +2438,53 @@ mca( 8, 0,8) mca( 0, 8,8) mca( 8, 8,8) -static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *src, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){ - if(block->type){ +static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, int sx, int sy, int b_w, int b_h, BlockNode *block, int plane_index, int w, int h){ + if(block->type & BLOCK_INTRA){ int x, y; - const int color= block->color[plane_index]; - for(y=0; y < b_h; y++){ - for(x=0; x < b_w; x++){ - dst[x + y*stride]= color; + const int color = block->color[plane_index]; + const int color4= color*0x01010101; + 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){ + 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; + } } } }else{ + uint8_t *src= s->last_picture[block->ref].data[plane_index]; const int scale= plane_index ? s->mv_scale : 2*s->mv_scale; int mx= block->mx*scale; int my= block->my*scale; const int dx= mx&15; const int dy= my&15; + const int tab_index= 3 - (b_w>>2) + (b_w>>4); sx += (mx>>4) - 2; sy += (my>>4) - 2; src += sx + sy*stride; @@ -2036,19 +2493,67 @@ static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *src, uint8_t *tmp, ff_emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+5, b_h+5, sx, sy, w, h); src= tmp + MB_SIZE; } - if((dx&3) || (dy&3) || b_w!=b_h || (b_w!=4 && b_w!=8 && b_w!=16)) +// assert(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h); +// assert(!(b_w&(b_w-1))); + assert(b_w>1 && b_h>1); + assert(tab_index>=0 && tab_index<4 || b_w==32); + if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1))) mc_block(dst, src, tmp, stride, b_w, b_h, dx, dy); - else - s->dsp.put_h264_qpel_pixels_tab[2-(b_w>>3)][dy+(dx>>2)](dst,src + 2 + 2*stride,stride); + else if(b_w==32){ + int y; + for(y=0; ydsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + y*stride, src + 2 + (y+2)*stride,stride); + s->dsp.put_h264_qpel_pixels_tab[0][dy+(dx>>2)](dst + 16 + y*stride, src + 18 + (y+2)*stride,stride); + } + }else if(b_w==b_h) + s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst,src + 2 + 2*stride,stride); + else if(b_w==2*b_h){ + s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst ,src + 2 + 2*stride,stride); + s->dsp.put_h264_qpel_pixels_tab[tab_index+1][dy+(dx>>2)](dst+b_h,src + 2 + b_h + 2*stride,stride); + }else{ + assert(2*b_w==b_h); + s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst ,src + 2 + 2*stride ,stride); + s->dsp.put_h264_qpel_pixels_tab[tab_index ][dy+(dx>>2)](dst+b_w*stride,src + 2 + 2*stride+b_w*stride,stride); + } } } -static always_inline int same_block(BlockNode *a, BlockNode *b){ - return !((a->mx - b->mx) | (a->my - b->my) | a->type | b->type); +void ff_snow_inner_add_yblock(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; + DWTELEM * dst; + for(y=0; y>1); + uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); + uint8_t *obmc4= obmc3+ (obmc_stride>>1); + dst = slice_buffer_get_line(sb, src_y + y); + for(x=0; x>= 8 - FRAC_BITS; + } + 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; + }else{ + dst[x + src_x] -= v; + } + } + } } //FIXME name clenup (b_w, block_w, b_width stuff) -static always_inline void add_yblock(SnowContext *s, DWTELEM *dst, uint8_t *dst8, uint8_t *src, uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int plane_index){ +static always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer *sb, DWTELEM *dst, uint8_t *dst8, const uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int offset_dst, int plane_index){ const int b_width = s->b_width << s->block_max_depth; const int b_height= s->b_height << s->block_max_depth; const int b_stride= b_width; @@ -2056,8 +2561,10 @@ static always_inline void add_yblock(SnowContext *s, DWTELEM *dst, uint8_t *dst8 BlockNode *rt= lt+1; BlockNode *lb= lt+b_stride; BlockNode *rb= lb+1; - uint8_t *block[4]; - uint8_t tmp[src_stride*(b_h+5)]; //FIXME align + uint8_t *block[4]; + int tmp_step= src_stride >= 7*MB_SIZE ? MB_SIZE : MB_SIZE*src_stride; + uint8_t tmp[src_stride*7*MB_SIZE]; //FIXME align + uint8_t *ptmp; int x,y; if(b_x<0){ @@ -2074,10 +2581,12 @@ static always_inline void add_yblock(SnowContext *s, DWTELEM *dst, uint8_t *dst8 lb= lt; rb= rt; } - + if(src_x<0){ //FIXME merge with prev & always round internal width upto *16 obmc -= src_x; b_w += src_x; + if(!sliced && !offset_dst) + dst -= src_x; src_x=0; }else if(src_x + b_w > w){ b_w = w - src_x; @@ -2085,35 +2594,42 @@ static always_inline void add_yblock(SnowContext *s, DWTELEM *dst, uint8_t *dst8 if(src_y<0){ obmc -= src_y*obmc_stride; b_h += src_y; + if(!sliced && !offset_dst) + dst -= src_y*dst_stride; src_y=0; }else if(src_y + b_h> h){ b_h = h - src_y; } - + if(b_w<=0 || b_h<=0) return; -assert(src_stride > 7*MB_SIZE); - dst += src_x + src_y*dst_stride; +assert(src_stride > 2*MB_SIZE + 5); + if(!sliced && offset_dst) + dst += src_x + src_y*dst_stride; dst8+= src_x + src_y*src_stride; // src += src_x + src_y*src_stride; - block[0]= tmp+3*MB_SIZE; - pred_block(s, block[0], src, tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); + ptmp= tmp + 3*tmp_step; + block[0]= ptmp; + ptmp+=tmp_step; + pred_block(s, block[0], tmp, src_stride, src_x, src_y, b_w, b_h, lt, plane_index, w, h); if(same_block(lt, rt)){ block[1]= block[0]; }else{ - block[1]= tmp + 4*MB_SIZE; - pred_block(s, block[1], src, tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); + block[1]= ptmp; + ptmp+=tmp_step; + pred_block(s, block[1], tmp, src_stride, src_x, src_y, b_w, b_h, rt, plane_index, w, h); } - + if(same_block(lt, lb)){ block[2]= block[0]; }else if(same_block(rt, lb)){ block[2]= block[1]; }else{ - block[2]= tmp+5*MB_SIZE; - pred_block(s, block[2], src, tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); + block[2]= ptmp; + ptmp+=tmp_step; + pred_block(s, block[2], tmp, src_stride, src_x, src_y, b_w, b_h, lb, plane_index, w, h); } if(same_block(lt, rb) ){ @@ -2123,8 +2639,8 @@ assert(src_stride > 7*MB_SIZE); }else if(same_block(lb, rb)){ block[3]= block[2]; }else{ - block[3]= tmp+6*MB_SIZE; - pred_block(s, block[3], src, tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); + block[3]= ptmp; + pred_block(s, block[3], tmp, src_stride, src_x, src_y, b_w, b_h, rb, plane_index, w, h); } #if 0 for(y=0; y 7*MB_SIZE); } } #else + if(sliced){ + START_TIMER + + s->dsp.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); + STOP_TIMER("inner_add_yblock") + }else for(y=0; y 7*MB_SIZE); +obmc2[x] * block[2][x + y*src_stride] +obmc3[x] * block[1][x + y*src_stride] +obmc4[x] * block[0][x + y*src_stride]; - + v <<= 8 - LOG2_OBMC_MAX; if(FRAC_BITS != 8){ v += 1<<(7 - FRAC_BITS); @@ -2190,85 +2712,714 @@ assert(src_stride > 7*MB_SIZE); #endif } -static always_inline void predict_plane(SnowContext *s, DWTELEM *buf, int plane_index, int add){ +static always_inline void predict_slice_buffered(SnowContext *s, slice_buffer * sb, DWTELEM * old_buffer, int plane_index, int add, int mb_y){ Plane *p= &s->plane[plane_index]; const int mb_w= s->b_width << s->block_max_depth; const int mb_h= s->b_height << s->block_max_depth; - int x, y, mb_x, mb_y; + int x, y, mb_x; int block_size = MB_SIZE >> s->block_max_depth; int block_w = plane_index ? block_size/2 : block_size; const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; int obmc_stride= plane_index ? block_size : 2*block_size; int ref_stride= s->current_picture.linesize[plane_index]; - uint8_t *ref = s->last_picture.data[plane_index]; uint8_t *dst8= s->current_picture.data[plane_index]; int w= p->width; int h= p->height; START_TIMER - + if(s->keyframe || (s->avctx->debug&512)){ + if(mb_y==mb_h) + return; + if(add){ - for(y=0; yline[y]; + for(x=0; x>= FRAC_BITS; + if(v&(~255)) v= ~(v>>31); + dst8[x + y*ref_stride]= v; + } + } + }else{ + for(y=block_w*mb_y; yline[y]; + for(x=0; xplane[plane_index]; + const int mb_w= s->b_width << s->block_max_depth; + const int mb_h= s->b_height << s->block_max_depth; + int x, y, mb_x; + int block_size = MB_SIZE >> s->block_max_depth; + int block_w = plane_index ? block_size/2 : block_size; + const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; + const int obmc_stride= plane_index ? block_size : 2*block_size; + int ref_stride= s->current_picture.linesize[plane_index]; + uint8_t *dst8= s->current_picture.data[plane_index]; + int w= p->width; + int h= p->height; + START_TIMER + + if(s->keyframe || (s->avctx->debug&512)){ + if(mb_y==mb_h) + return; + + if(add){ + for(y=block_w*mb_y; y>= FRAC_BITS; if(v&(~255)) v= ~(v>>31); dst8[x + y*ref_stride]= v; } - } - }else{ - for(y=0; yb_height << s->block_max_depth; + int mb_y; + for(mb_y=0; mb_y<=mb_h; mb_y++) + predict_slice(s, buf, plane_index, add, mb_y); +} + +static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){ + int i, x2, y2; + Plane *p= &s->plane[plane_index]; + const int block_size = MB_SIZE >> s->block_max_depth; + const int block_w = plane_index ? block_size/2 : block_size; + const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; + const int obmc_stride= plane_index ? block_size : 2*block_size; + const int ref_stride= s->current_picture.linesize[plane_index]; + uint8_t *src= s-> input_picture.data[plane_index]; + DWTELEM *dst= (DWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; + const int b_stride = s->b_width << s->block_max_depth; + const int w= p->width; + const int h= p->height; + int index= mb_x + mb_y*b_stride; + BlockNode *b= &s->block[index]; + BlockNode backup= *b; + int ab=0; + int aa=0; + + b->type|= BLOCK_INTRA; + b->color[plane_index]= 0; + memset(dst, 0, obmc_stride*obmc_stride*sizeof(DWTELEM)); + + for(i=0; i<4; i++){ + int mb_x2= mb_x + (i &1) - 1; + int mb_y2= mb_y + (i>>1) - 1; + int x= block_w*mb_x2 + block_w/2; + int y= block_w*mb_y2 + block_w/2; + + add_yblock(s, 0, NULL, dst + ((i&1)+(i>>1)*obmc_stride)*block_w, NULL, obmc, + x, y, block_w, block_w, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index); + + for(y2= FFMAX(y, 0); y2h) obmc_v += obmc[index - block_w*obmc_stride]; + if(x+block_w>w) obmc_v += obmc[index - block_w]; + //FIXME precalc this or simplify it somehow else + + d = -dst[index] + (1<<(FRAC_BITS-1)); + dst[index] = d; + ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v; + aa += obmc_v * obmc_v; //FIXME precalclate this + } + } + } + *b= backup; + + return clip(((ab<b_width << s->block_max_depth; + const int b_height = s->b_height<< s->block_max_depth; + int index= x + y*b_stride; + BlockNode *b = &s->block[index]; + BlockNode *left = x ? &s->block[index-1] : &null_block; + BlockNode *top = y ? &s->block[index-b_stride] : &null_block; + BlockNode *tl = y && x ? &s->block[index-b_stride-1] : left; + BlockNode *tr = y && x+wblock[index-b_stride+w] : tl; + int dmx, dmy; +// int mx_context= av_log2(2*FFABS(left->mx - top->mx)); +// int my_context= av_log2(2*FFABS(left->my - top->my)); + + if(x<0 || x>=b_stride || y>=b_height) + return 0; +/* +1 0 0 +01X 1-2 1 +001XX 3-6 2-3 +0001XXX 7-14 4-7 +00001XXXX 15-30 8-15 +*/ +//FIXME try accurate rate +//FIXME intra and inter predictors if surrounding blocks arent the same type + if(b->type & BLOCK_INTRA){ + return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0])) + + av_log2(2*FFABS(left->color[1] - b->color[1])) + + av_log2(2*FFABS(left->color[2] - b->color[2]))); + }else{ + pred_mv(s, &dmx, &dmy, b->ref, left, top, tr); + dmx-= b->mx; + dmy-= b->my; + return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda + + av_log2(2*FFABS(dmy)) + + av_log2(2*b->ref)); + } +} + +static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, const uint8_t *obmc_edged){ + Plane *p= &s->plane[plane_index]; + const int block_size = MB_SIZE >> s->block_max_depth; + const int block_w = plane_index ? block_size/2 : block_size; + const int obmc_stride= plane_index ? block_size : 2*block_size; + const int ref_stride= s->current_picture.linesize[plane_index]; + uint8_t *dst= s->current_picture.data[plane_index]; + uint8_t *src= s-> input_picture.data[plane_index]; + DWTELEM *pred= (DWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; + uint8_t cur[ref_stride*2*MB_SIZE]; //FIXME alignment + uint8_t tmp[ref_stride*(2*MB_SIZE+5)]; + const int b_stride = s->b_width << s->block_max_depth; + const int b_height = s->b_height<< s->block_max_depth; + const int w= p->width; + const int h= p->height; + int distortion; + int rate= 0; + const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); + int sx= block_w*mb_x - block_w/2; + int sy= block_w*mb_y - block_w/2; + int x0= FFMAX(0,-sx); + int y0= FFMAX(0,-sy); + int x1= FFMIN(block_w*2, w-sx); + int y1= FFMIN(block_w*2, h-sy); + int i,x,y; + + pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_w*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h); + + for(y=y0; y> FRAC_BITS; + if(v&(~255)) v= ~(v>>31); + dst1[x] = v; + } + } + + /* copy the regions where obmc[] = (uint8_t)256 */ + if(LOG2_OBMC_MAX == 8 + && (mb_x == 0 || mb_x == b_stride-1) + && (mb_y == 0 || mb_y == b_height-1)){ + if(mb_x == 0) + x1 = block_w; + else + x0 = block_w; + if(mb_y == 0) + y1 = block_w; + else + y0 = block_w; + for(y=y0; yavctx->me_cmp == FF_CMP_W97) + distortion = w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); + else if(s->avctx->me_cmp == FF_CMP_W53) + distortion = w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32); + else{ + distortion = 0; + for(i=0; i<4; i++){ + int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride; + distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16); + } + } + }else{ + assert(block_w==8); + distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2); + } + + if(plane_index==0){ + for(i=0; i<4; i++){ +/* ..RRr + * .RXx. + * rxx.. + */ + rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1); + } + if(mb_x == b_stride-2) + rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1); + } + return distortion + rate*penalty_factor; +} + +static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){ + int i, y2; + Plane *p= &s->plane[plane_index]; + const int block_size = MB_SIZE >> s->block_max_depth; + const int block_w = plane_index ? block_size/2 : block_size; + const uint8_t *obmc = plane_index ? obmc_tab[s->block_max_depth+1] : obmc_tab[s->block_max_depth]; + const int obmc_stride= plane_index ? block_size : 2*block_size; + const int ref_stride= s->current_picture.linesize[plane_index]; + uint8_t *dst= s->current_picture.data[plane_index]; + uint8_t *src= s-> input_picture.data[plane_index]; + static const DWTELEM zero_dst[4096]; //FIXME + const int b_stride = s->b_width << s->block_max_depth; + const int w= p->width; + const int h= p->height; + int distortion= 0; + int rate= 0; + const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp); + + for(i=0; i<9; i++){ + int mb_x2= mb_x + (i%3) - 1; + int mb_y2= mb_y + (i/3) - 1; + int x= block_w*mb_x2 + block_w/2; + int y= block_w*mb_y2 + block_w/2; + + add_yblock(s, 0, NULL, zero_dst, dst, obmc, + x, y, block_w, block_w, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index); + + //FIXME find a cleaner/simpler way to skip the outside stuff + for(y2= y; y2<0; y2++) + memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w); + for(y2= h; y2 w){ + for(y2= y; y2dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_w); + } + + if(plane_index==0){ + BlockNode *b= &s->block[mb_x+mb_y*b_stride]; + int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1); + +/* ..RRRr + * .RXXx. + * .RXXx. + * rxxx. + */ + if(merged) + rate = get_block_bits(s, mb_x, mb_y, 2); + for(i=merged?4:0; i<9; i++){ + static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}}; + rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1); + } + } + return distortion + rate*penalty_factor; +} + +static always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, const uint8_t *obmc_edged, int *best_rd){ + const int b_stride= s->b_width << s->block_max_depth; + BlockNode *block= &s->block[mb_x + mb_y * b_stride]; + BlockNode backup= *block; + int rd, index, value; + + assert(mb_x>=0 && mb_y>=0); + assert(mb_xcolor[0] = p[0]; + block->color[1] = p[1]; + block->color[2] = p[2]; + block->type |= BLOCK_INTRA; + }else{ + index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1); + value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12); + if(s->me_cache[index] == value) + return 0; + s->me_cache[index]= value; + + block->mx= p[0]; + block->my= p[1]; + block->type &= ~BLOCK_INTRA; + } + + rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged); + +//FIXME chroma + if(rd < *best_rd){ + *best_rd= rd; + return 1; + }else{ + *block= backup; + return 0; + } +} + +/* special case for int[2] args we discard afterward, fixes compilation prob with gcc 2.95 */ +static always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, const uint8_t *obmc_edged, int *best_rd){ + int p[2] = {p0, p1}; + return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd); +} + +static always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){ + const int b_stride= s->b_width << s->block_max_depth; + BlockNode *block= &s->block[mb_x + mb_y * b_stride]; + BlockNode backup[4]= {block[0], block[1], block[b_stride], block[b_stride+1]}; + int rd, index, value; + + assert(mb_x>=0 && mb_y>=0); + assert(mb_xme_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12); + if(s->me_cache[index] == value) + return 0; + s->me_cache[index]= value; + + block->mx= p0; + block->my= p1; + block->ref= ref; + block->type &= ~BLOCK_INTRA; + block[1]= block[b_stride]= block[b_stride+1]= *block; + + rd= get_4block_rd(s, mb_x, mb_y, 0); + +//FIXME chroma + if(rd < *best_rd){ + *best_rd= rd; + return 1; + }else{ + block[0]= backup[0]; + block[1]= backup[1]; + block[b_stride]= backup[2]; + block[b_stride+1]= backup[3]; + return 0; + } +} + +static void iterative_me(SnowContext *s){ + int pass, mb_x, mb_y; + const int b_width = s->b_width << s->block_max_depth; + const int b_height= s->b_height << s->block_max_depth; + const int b_stride= b_width; + int color[3]; + + { + RangeCoder r = s->c; + uint8_t state[sizeof(s->block_state)]; + memcpy(state, s->block_state, sizeof(s->block_state)); + for(mb_y= 0; mb_yb_height; mb_y++) + for(mb_x= 0; mb_xb_width; mb_x++) + encode_q_branch(s, 0, mb_x, mb_y); + s->c = r; + memcpy(s->block_state, state, sizeof(s->block_state)); + } + + for(pass=0; pass<25; pass++){ + int change= 0; + + for(mb_y= 0; mb_yblock[index]; + BlockNode *tb = mb_y ? &s->block[index-b_stride ] : NULL; + BlockNode *lb = mb_x ? &s->block[index -1] : NULL; + BlockNode *rb = mb_x+1block[index +1] : NULL; + BlockNode *bb = mb_y+1block[index+b_stride ] : NULL; + BlockNode *tlb= mb_x && mb_y ? &s->block[index-b_stride-1] : NULL; + BlockNode *trb= mb_x+1block[index-b_stride+1] : NULL; + BlockNode *blb= mb_x && mb_y+1block[index+b_stride-1] : NULL; + BlockNode *brb= mb_x+1block[index+b_stride+1] : NULL; + const int b_w= (MB_SIZE >> s->block_max_depth); + uint8_t obmc_edged[b_w*2][b_w*2]; + + if(pass && (block->type & BLOCK_OPT)) + continue; + block->type |= BLOCK_OPT; + + backup= *block; + + if(!s->me_cache_generation) + memset(s->me_cache, 0, sizeof(s->me_cache)); + s->me_cache_generation += 1<<22; + + //FIXME precalc + { + int x, y; + memcpy(obmc_edged, obmc_tab[s->block_max_depth], b_w*b_w*4); + if(mb_x==0) + for(y=0; y input_picture.data[0]; + uint8_t *dst= s->current_picture.data[0]; + const int stride= s->current_picture.linesize[0]; + const int block_w= MB_SIZE >> s->block_max_depth; + const int sx= block_w*mb_x - block_w/2; + const int sy= block_w*mb_y - block_w/2; + const int w= s->plane[0].width; + const int h= s->plane[0].height; + int y; + + for(y=sy; y<0; y++) + memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2); + for(y=h; y w){ + for(y=sy; y 0 && (block->type&BLOCK_INTRA)){ + int color0[3]= {block->color[0], block->color[1], block->color[2]}; + check_block(s, mb_x, mb_y, color0, 1, *obmc_edged, &best_rd); + }else + check_block_inter(s, mb_x, mb_y, block->mx, block->my, *obmc_edged, &best_rd); + + ref_b= *block; + ref_rd= best_rd; + for(ref=0; ref < s->ref_frames; ref++){ + int16_t (*mvr)[2]= &s->ref_mvs[ref][index]; + if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold + continue; + block->ref= ref; + best_rd= INT_MAX; + + check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], *obmc_edged, &best_rd); + check_block_inter(s, mb_x, mb_y, 0, 0, *obmc_edged, &best_rd); + if(tb) + check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], *obmc_edged, &best_rd); + if(lb) + check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], *obmc_edged, &best_rd); + if(rb) + check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], *obmc_edged, &best_rd); + if(bb) + check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], *obmc_edged, &best_rd); + + /* fullpel ME */ + //FIXME avoid subpel interpol / round to nearest integer + do{ + dia_change=0; + for(i=0; iavctx->dia_size, 1); i++){ + for(j=0; jmx+4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); + dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); + dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), *obmc_edged, &best_rd); + dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), *obmc_edged, &best_rd); + } + } + }while(dia_change); + /* subpel ME */ + do{ + static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},}; + dia_change=0; + for(i=0; i<8; i++) + dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], *obmc_edged, &best_rd); + }while(dia_change); + //FIXME or try the standard 2 pass qpel or similar + + mvr[0][0]= block->mx; + mvr[0][1]= block->my; + if(ref_rd > best_rd){ + ref_rd= best_rd; + ref_b= *block; + } + } + best_rd= ref_rd; + *block= ref_b; +#if 1 + check_block(s, mb_x, mb_y, color, 1, *obmc_edged, &best_rd); + //FIXME RD style color selection +#endif + if(!same_block(block, &backup)){ + if(tb ) tb ->type &= ~BLOCK_OPT; + if(lb ) lb ->type &= ~BLOCK_OPT; + if(rb ) rb ->type &= ~BLOCK_OPT; + if(bb ) bb ->type &= ~BLOCK_OPT; + if(tlb) tlb->type &= ~BLOCK_OPT; + if(trb) trb->type &= ~BLOCK_OPT; + if(blb) blb->type &= ~BLOCK_OPT; + if(brb) brb->type &= ~BLOCK_OPT; + change ++; } } } - - return; + av_log(NULL, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change); + if(!change) + break; } - - for(mb_y=0; mb_y<=mb_h; mb_y++){ - for(mb_x=0; mb_x<=mb_w; mb_x++){ - START_TIMER - add_yblock(s, buf, dst8, ref, obmc, - block_w*mb_x - block_w/2, - block_w*mb_y - block_w/2, - block_w, block_w, - w, h, - w, ref_stride, obmc_stride, - mb_x - 1, mb_y - 1, - add, plane_index); - - STOP_TIMER("add_yblock") + if(s->block_max_depth == 1){ + int change= 0; + for(mb_y= 0; mb_yblock[index]; + b[1]= b[0]+1; + b[2]= b[0]+b_stride; + b[3]= b[2]+1; + if(same_block(b[0], b[1]) && + same_block(b[0], b[2]) && + same_block(b[0], b[3])) + continue; + + if(!s->me_cache_generation) + memset(s->me_cache, 0, sizeof(s->me_cache)); + s->me_cache_generation += 1<<22; + + init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0); + + //FIXME more multiref search? + check_4block_inter(s, mb_x, mb_y, + (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2, + (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd); + + for(i=0; i<4; i++) + if(!(b[i]->type&BLOCK_INTRA)) + check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd); + + if(init_rd != best_rd) + change++; + } } + av_log(NULL, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4); } - - STOP_TIMER("predict_plane") } static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int bias){ const int level= b->level; const int w= b->width; const int h= b->height; - const int qlog= clip(s->qlog + b->qlog, 0, 128); - const int qmul= qexp[qlog&7]<<(qlog>>3); + const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); + const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); int x,y, thres1, thres2; - START_TIMER - - assert(QROOT==8); +// START_TIMER if(s->qlog == LOSSLESS_QLOG) return; - + bias= bias ? 0 : (3*qmul)>>3; thres1= ((qmul - bias)>>QEXPSHIFT) - 1; thres2= 2*thres1; - + if(!bias){ for(y=0; y thres2){ if(i>=0){ i<<= QEXPSHIFT; @@ -2287,8 +3438,8 @@ static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int b }else{ for(y=0; y thres2){ if(i>=0){ i<<= QEXPSHIFT; @@ -2310,18 +3461,43 @@ static void quantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride, int b } } +static void dequantize_slice_buffered(SnowContext *s, slice_buffer * sb, SubBand *b, DWTELEM *src, int stride, int start_y, int end_y){ + const int w= b->width; + const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); + const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); + const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; + int x,y; + START_TIMER + + if(s->qlog == LOSSLESS_QLOG) return; + + for(y=start_y; ystride_line) + b->buf_y_offset) + b->buf_x_offset; + for(x=0; x>(QEXPSHIFT)); //FIXME try different bias + }else if(i>0){ + line[x]= (( i*qmul + qadd)>>(QEXPSHIFT)); + } + } + } + if(w > 200 /*level+1 == s->spatial_decomposition_count*/){ + STOP_TIMER("dquant") + } +} + static void dequantize(SnowContext *s, SubBand *b, DWTELEM *src, int stride){ const int w= b->width; const int h= b->height; - const int qlog= clip(s->qlog + b->qlog, 0, 128); - const int qmul= qexp[qlog&7]<<(qlog>>3); + const int qlog= clip(s->qlog + b->qlog, 0, QROOT*16); + const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT; int x,y; START_TIMER - + if(s->qlog == LOSSLESS_QLOG) return; - - assert(QROOT==8); for(y=0; ywidth; const int h= b->height; int x,y; - + for(y=h-1; y>=0; y--){ for(x=w-1; x>=0; x--){ int i= x + y*stride; - + if(x){ if(use_median){ if(y && x+1width; + int x,y; + +// START_TIMER + + DWTELEM * line; + DWTELEM * prev; + + if (start_y != 0) + line = slice_buffer_get_line(sb, ((start_y - 1) * b->stride_line) + b->buf_y_offset) + b->buf_x_offset; + + for(y=start_y; ystride_line) + b->buf_y_offset) + b->buf_x_offset; + for(x=0; xwidth; const int h= b->height; int x,y; - + for(y=0; yc, kstate, s->keyframe); if(s->keyframe || s->always_reset) @@ -2406,6 +3616,7 @@ static void encode_header(SnowContext *s){ put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0); put_rac(&s->c, s->header_state, s->spatial_scalability); // put_rac(&s->c, s->header_state, s->rate_scalability); + put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0); for(plane_index=0; plane_index<2; plane_index++){ for(level=0; levelspatial_decomposition_count; level++){ @@ -2417,8 +3628,8 @@ static void encode_header(SnowContext *s){ } } put_symbol(&s->c, s->header_state, s->spatial_decomposition_type, 0); - put_symbol(&s->c, s->header_state, s->qlog, 1); - put_symbol(&s->c, s->header_state, s->mv_scale, 0); + put_symbol(&s->c, s->header_state, s->qlog, 1); + put_symbol(&s->c, s->header_state, s->mv_scale, 0); put_symbol(&s->c, s->header_state, s->qbias, 1); put_symbol(&s->c, s->header_state, s->block_max_depth, 0); } @@ -2427,7 +3638,7 @@ static int decode_header(SnowContext *s){ int plane_index, level, orientation; uint8_t kstate[32]; - memset(kstate, MID_STATE, sizeof(kstate)); + memset(kstate, MID_STATE, sizeof(kstate)); s->keyframe= get_rac(&s->c, kstate); if(s->keyframe || s->always_reset) @@ -2447,6 +3658,7 @@ static int decode_header(SnowContext *s){ s->chroma_v_shift= get_symbol(&s->c, s->header_state, 0); s->spatial_scalability= get_rac(&s->c, s->header_state); // s->rate_scalability= get_rac(&s->c, s->header_state); + s->max_ref_frames= get_symbol(&s->c, s->header_state, 0)+1; for(plane_index=0; plane_index<3; plane_index++){ for(level=0; levelspatial_decomposition_count; level++){ @@ -2460,28 +3672,44 @@ static int decode_header(SnowContext *s){ } } } - + s->spatial_decomposition_type= get_symbol(&s->c, s->header_state, 0); if(s->spatial_decomposition_type > 2){ av_log(s->avctx, AV_LOG_ERROR, "spatial_decomposition_type %d not supported", s->spatial_decomposition_type); return -1; } - + s->qlog= get_symbol(&s->c, s->header_state, 1); s->mv_scale= get_symbol(&s->c, s->header_state, 0); s->qbias= get_symbol(&s->c, s->header_state, 1); s->block_max_depth= get_symbol(&s->c, s->header_state, 0); + if(s->block_max_depth > 1 || s->block_max_depth < 0){ + av_log(s->avctx, AV_LOG_ERROR, "block_max_depth= %d is too large", s->block_max_depth); + s->block_max_depth= 0; + return -1; + } return 0; } +static void init_qexp(void){ + int i; + double v=128; + + for(i=0; ipriv_data; int width, height; int level, orientation, plane_index, dec; + int i, j; s->avctx= avctx; - + dsputil_init(&s->dsp, avctx); #define mcf(dx,dy)\ @@ -2521,24 +3749,27 @@ static int common_init(AVCodecContext *avctx){ mcfh(8, 0) mcfh(0, 8) mcfh(8, 8) - + + if(!qexp[0]) + init_qexp(); + dec= s->spatial_decomposition_count= 5; s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type - + s->chroma_h_shift= 1; //FIXME XXX s->chroma_v_shift= 1; - + // dec += FFMAX(s->chroma_h_shift, s->chroma_v_shift); - + width= s->avctx->width; height= s->avctx->height; s->spatial_dwt_buffer= av_mallocz(width*height*sizeof(DWTELEM)); - + s->mv_scale= (s->avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4; s->block_max_depth= (s->avctx->flags & CODEC_FLAG_4MV) ? 1 : 0; - - for(plane_index=0; plane_index<3; plane_index++){ + + for(plane_index=0; plane_index<3; plane_index++){ int w= s->avctx->width; int h= s->avctx->height; @@ -2552,38 +3783,105 @@ static int common_init(AVCodecContext *avctx){ 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; - - if(orientation&1) b->buf += (w+1)>>1; - if(orientation>1) b->buf += b->stride>>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; + } + if(level) b->parent= &s->plane[plane_index].band[level-1][orientation]; - b->x = av_mallocz(((b->width+1) * b->height+1)*sizeof(int16_t)); - b->coeff= av_mallocz(((b->width+1) * b->height+1)*sizeof(DWTELEM)); + b->x_coeff=av_mallocz(((b->width+1) * b->height+1)*sizeof(x_and_coeff)); } w= (w+1)>>1; h= (h+1)>>1; } } - + + for(i=0; iwidth= avctx->width; height= s->height= avctx->height; - + assert(width && height); */ s->avctx->get_buffer(s->avctx, &s->mconly_picture); - + return 0; } +static int qscale2qlog(int qscale){ + return rint(QROOT*log(qscale / (float)FF_QP2LAMBDA)/log(2)) + + 61*QROOT/8; //<64 >60 +} + +static int ratecontrol_1pass(SnowContext *s, AVFrame *pict) +{ + /* estimate the frame's complexity as a sum of weighted dwt coefs. + * FIXME we know exact mv bits at this point, + * but ratecontrol isn't set up to include them. */ + uint32_t coef_sum= 0; + int level, orientation, delta_qlog; + + for(level=0; levelspatial_decomposition_count; level++){ + for(orientation=level ? 1 : 0; orientation<4; orientation++){ + SubBand *b= &s->plane[0].band[level][orientation]; + DWTELEM *buf= b->buf; + const int w= b->width; + const int h= b->height; + const int stride= b->stride; + const int qlog= clip(2*QROOT + b->qlog, 0, QROOT*16); + const int qmul= qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT); + const int qdiv= (1<<16)/qmul; + int x, y; + if(orientation==0) + decorrelate(s, b, buf, stride, 1, 0); + for(y=0; y> 16; + if(orientation==0) + correlate(s, b, buf, stride, 1, 0); + } + } + + /* ugly, ratecontrol just takes a sqrt again */ + coef_sum = (uint64_t)coef_sum * coef_sum >> 16; + assert(coef_sum < INT_MAX); + + if(pict->pict_type == I_TYPE){ + s->m.current_picture.mb_var_sum= coef_sum; + s->m.current_picture.mc_mb_var_sum= 0; + }else{ + s->m.current_picture.mc_mb_var_sum= coef_sum; + s->m.current_picture.mb_var_sum= 0; + } + + pict->quality= ff_rate_estimate_qscale(&s->m, 1); + if (pict->quality < 0) + return INT_MIN; + s->lambda= pict->quality * 3/2; + delta_qlog= qscale2qlog(pict->quality) - s->qlog; + s->qlog+= delta_qlog; + return delta_qlog; +} static void calculate_vissual_weight(SnowContext *s, Plane *p){ int width = p->width; @@ -2595,7 +3893,7 @@ static void calculate_vissual_weight(SnowContext *s, Plane *p){ SubBand *b= &p->band[level][orientation]; DWTELEM *buf= b->buf; int64_t error=0; - + memset(s->spatial_dwt_buffer, 0, sizeof(int)*width*height); buf[b->width/2 + b->height/2*b->stride]= 256*256; ff_spatial_idwt(s->spatial_dwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); @@ -2617,27 +3915,51 @@ static int encode_init(AVCodecContext *avctx) SnowContext *s = avctx->priv_data; int plane_index; - if(avctx->strict_std_compliance >= 0){ - av_log(avctx, AV_LOG_ERROR, "this codec is under development, files encoded with it wont be decodeable with future versions!!!\n" - "use vstrict=-1 to use it anyway\n"); + if(avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL){ + av_log(avctx, AV_LOG_ERROR, "this codec is under development, files encoded with it may not be decodable with future versions!!!\n" + "use vstrict=-2 / -strict -2 to use it anyway\n"); + return -1; + } + + if(avctx->prediction_method == DWT_97 + && (avctx->flags & CODEC_FLAG_QSCALE) + && avctx->global_quality == 0){ + av_log(avctx, AV_LOG_ERROR, "the 9/7 wavelet is incompatible with lossless mode\n"); return -1; } - + common_init(avctx); alloc_blocks(s); - + s->version=0; - + + s->m.avctx = avctx; + s->m.flags = avctx->flags; + s->m.bit_rate= avctx->bit_rate; + s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t)); s->m.me.map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t)); + s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t)); h263_encode_init(&s->m); //mv_penalty + s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1); + + if(avctx->flags&CODEC_FLAG_PASS1){ + if(!avctx->stats_out) + avctx->stats_out = av_mallocz(256); + } + if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){ + if(ff_rate_control_init(&s->m) < 0) + return -1; + } + s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2)); + for(plane_index=0; plane_index<3; plane_index++){ calculate_vissual_weight(s, &s->plane[plane_index]); } - - + + avctx->coded_frame= &s->current_picture; switch(avctx->pix_fmt){ // case PIX_FMT_YUV444P: @@ -2658,6 +3980,21 @@ static int encode_init(AVCodecContext *avctx) // avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift); s->chroma_h_shift= 1; s->chroma_v_shift= 1; + + ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp); + ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp); + + s->avctx->get_buffer(s->avctx, &s->input_picture); + + if(s->avctx->me_method == ME_ITER){ + int i; + int size= s->b_width * s->b_height << 2*s->block_max_depth; + for(i=0; imax_ref_frames; i++){ + s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2])); + s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t)); + } + } + return 0; } @@ -2672,16 +4009,29 @@ static int frame_start(SnowContext *s){ draw_edges(s->current_picture.data[2], s->current_picture.linesize[2], w>>1, h>>1, EDGE_WIDTH/2); } - tmp= s->last_picture; - s->last_picture= s->current_picture; + tmp= s->last_picture[s->max_ref_frames-1]; + memmove(s->last_picture+1, s->last_picture, (s->max_ref_frames-1)*sizeof(AVFrame)); + s->last_picture[0]= s->current_picture; s->current_picture= tmp; - + + if(s->keyframe){ + s->ref_frames= 0; + }else{ + int i; + for(i=0; imax_ref_frames && s->last_picture[i].data[0]; i++) + if(i && s->last_picture[i-1].key_frame) + break; + s->ref_frames= i; + } + s->current_picture.reference= 1; if(s->avctx->get_buffer(s->avctx, &s->current_picture) < 0){ av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed\n"); return -1; } - + + s->current_picture.key_frame= s->keyframe; + return 0; } @@ -2691,40 +4041,64 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, AVFrame *pict = data; const int width= s->avctx->width; const int height= s->avctx->height; - int level, orientation, plane_index; + int level, orientation, plane_index, i, y; + uint8_t rc_header_bak[sizeof(s->header_state)]; + uint8_t rc_block_bak[sizeof(s->block_state)]; ff_init_range_encoder(c, buf, buf_size); ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); - - s->input_picture = *pict; - s->keyframe=avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0; - pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE; - - if(pict->quality){ - s->qlog= rint(QROOT*log(pict->quality / (float)FF_QP2LAMBDA)/log(2)); - //<64 >60 - s->qlog += 61; + for(i=0; i<3; i++){ + int shift= !!i; + for(y=0; y<(height>>shift); y++) + memcpy(&s->input_picture.data[i][y * s->input_picture.linesize[i]], + &pict->data[i][y * pict->linesize[i]], + width>>shift); + } + s->new_picture = *pict; + + s->m.picture_number= avctx->frame_number; + if(avctx->flags&CODEC_FLAG_PASS2){ + s->m.pict_type = + pict->pict_type= s->m.rc_context.entry[avctx->frame_number].new_pict_type; + s->keyframe= pict->pict_type==FF_I_TYPE; + if(!(avctx->flags&CODEC_FLAG_QSCALE)) { + pict->quality= ff_rate_estimate_qscale(&s->m, 0); + if (pict->quality < 0) + return -1; + } }else{ - s->qlog= LOSSLESS_QLOG; + s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0; + s->m.pict_type= + pict->pict_type= s->keyframe ? FF_I_TYPE : FF_P_TYPE; } + if(s->pass1_rc && avctx->frame_number == 0) + pict->quality= 2*FF_QP2LAMBDA; + if(pict->quality){ + s->qlog= qscale2qlog(pict->quality); + s->lambda = pict->quality * 3/2; + } + if(s->qlog < 0 || (!pict->quality && (avctx->flags & CODEC_FLAG_QSCALE))){ + s->qlog= LOSSLESS_QLOG; + s->lambda = 0; + }//else keep previous frame's qlog until after motion est + frame_start(s); - s->current_picture.key_frame= s->keyframe; + s->m.current_picture_ptr= &s->m.current_picture; if(pict->pict_type == P_TYPE){ int block_width = (width +15)>>4; int block_height= (height+15)>>4; int stride= s->current_picture.linesize[0]; - + assert(s->current_picture.data[0]); - assert(s->last_picture.data[0]); - + assert(s->last_picture[0].data[0]); + s->m.avctx= s->avctx; s->m.current_picture.data[0]= s->current_picture.data[0]; - s->m. last_picture.data[0]= s-> last_picture.data[0]; + s->m. last_picture.data[0]= s->last_picture[0].data[0]; s->m. new_picture.data[0]= s-> input_picture.data[0]; - s->m.current_picture_ptr= &s->m.current_picture; s->m. last_picture_ptr= &s->m. last_picture; s->m.linesize= s->m. last_picture.linesize[0]= @@ -2746,21 +4120,30 @@ static int encode_frame(AVCodecContext *avctx, unsigned char *buf, int buf_size, s->m.out_format= FMT_H263; s->m.unrestricted_mv= 1; - s->lambda = s->m.lambda= pict->quality * 3/2; //FIXME bug somewhere else + s->m.lambda = s->lambda; s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7); s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT; s->m.dsp= s->dsp; //move ff_init_me(&s->m); + s->dsp= s->m.dsp; + } + + if(s->pass1_rc){ + memcpy(rc_header_bak, s->header_state, sizeof(s->header_state)); + memcpy(rc_block_bak, s->block_state, sizeof(s->block_state)); } - + redo_frame: - + + s->m.pict_type = pict->pict_type; s->qbias= pict->pict_type == P_TYPE ? 2 : 0; encode_header(s); - encode_blocks(s); - + s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start); + encode_blocks(s, 1); + s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits; + for(plane_index=0; plane_index<3; plane_index++){ Plane *p= &s->plane[plane_index]; int w= p->width; @@ -2768,6 +4151,7 @@ redo_frame: int x, y; // int bits= put_bits_count(&s->c.pb); + if(!(avctx->flags2 & CODEC_FLAG2_MEMC_ONLY)){ //FIXME optimize if(pict->data[plane_index]) //FIXME gray hack for(y=0; yspatial_dwt_buffer, plane_index, 0); - - if( plane_index==0 - && pict->pict_type == P_TYPE + + if( plane_index==0 + && pict->pict_type == P_TYPE + && !(avctx->flags&CODEC_FLAG_PASS2) && s->m.me.scene_change_score > s->avctx->scenechange_threshold){ ff_init_range_encoder(c, buf, buf_size); ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); pict->pict_type= FF_I_TYPE; s->keyframe=1; + s->current_picture.key_frame=1; reset_contexts(s); goto redo_frame; } - + if(s->qlog == LOSSLESS_QLOG){ for(y=0; yspatial_dwt_buffer[y*w + x]= (s->spatial_dwt_buffer[y*w + x] + (1<<(FRAC_BITS-1)))>>FRAC_BITS; + s->spatial_dwt_buffer[y*w + x]= (s->spatial_dwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS; } } } - + ff_spatial_dwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); + if(s->pass1_rc && plane_index==0){ + int delta_qlog = ratecontrol_1pass(s, pict); + if (delta_qlog <= INT_MIN) + return -1; + if(delta_qlog){ + //reordering qlog in the bitstream would eliminate this reset + ff_init_range_encoder(c, buf, buf_size); + memcpy(s->header_state, rc_header_bak, sizeof(s->header_state)); + memcpy(s->block_state, rc_block_bak, sizeof(s->block_state)); + encode_header(s); + encode_blocks(s, 0); + } + } + for(level=0; levelspatial_decomposition_count; level++){ for(orientation=level ? 1 : 0; orientation<4; orientation++){ SubBand *b= &p->band[level][orientation]; - + quantize(s, b, b->buf, b->stride, s->qbias); if(orientation==0) decorrelate(s, b, b->buf, b->stride, pict->pict_type == P_TYPE, 0); @@ -2832,9 +4232,23 @@ redo_frame: {START_TIMER predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); STOP_TIMER("pred-conv")} + }else{ + //ME/MC only + if(pict->pict_type == I_TYPE){ + for(y=0; ycurrent_picture.data[plane_index][y*s->current_picture.linesize[plane_index] + x]= + pict->data[plane_index][y*pict->linesize[plane_index] + x]; + } + } + }else{ + memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h); + predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); + } + } if(s->avctx->flags&CODEC_FLAG_PSNR){ int64_t error= 0; - + if(pict->data[plane_index]) //FIXME gray hack for(y=0; ylast_picture.data[0]) - avctx->release_buffer(avctx, &s->last_picture); + if(s->last_picture[s->max_ref_frames-1].data[0]) + avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]); + + s->current_picture.coded_picture_number = avctx->frame_number; + s->current_picture.pict_type = pict->pict_type; + s->current_picture.quality = pict->quality; + s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start); + s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits; + s->m.current_picture.display_picture_number = + s->m.current_picture.coded_picture_number = avctx->frame_number; + s->m.current_picture.quality = pict->quality; + s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start); + if(s->pass1_rc) + if (ff_rate_estimate_qscale(&s->m, 0) < 0) + return -1; + if(avctx->flags&CODEC_FLAG_PASS1) + ff_write_pass1_stats(&s->m); + s->m.last_pict_type = s->m.pict_type; + avctx->frame_bits = s->m.frame_bits; + avctx->mv_bits = s->m.mv_bits; + avctx->misc_bits = s->m.misc_bits; + avctx->p_tex_bits = s->m.p_tex_bits; emms_c(); - + return ff_rac_terminate(c); } static void common_end(SnowContext *s){ - int plane_index, level, orientation; + int plane_index, level, orientation, i; av_freep(&s->spatial_dwt_buffer); - av_freep(&s->m.me.scratchpad); + 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); - for(plane_index=0; plane_index<3; plane_index++){ + for(i=0; iref_mvs[i]); + av_freep(&s->ref_scores[i]); + if(s->last_picture[i].data[0]) + s->avctx->release_buffer(s->avctx, &s->last_picture[i]); + } + + for(plane_index=0; plane_index<3; plane_index++){ 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); - av_freep(&b->coeff); + + av_freep(&b->x_coeff); } } } @@ -2883,16 +4324,23 @@ static int encode_end(AVCodecContext *avctx) SnowContext *s = avctx->priv_data; common_end(s); + av_free(avctx->stats_out); return 0; } static int decode_init(AVCodecContext *avctx) { -// SnowContext *s = avctx->priv_data; + SnowContext *s = avctx->priv_data; + int block_size; + + avctx->pix_fmt= PIX_FMT_YUV420P; common_init(avctx); - + + block_size = MB_SIZE >> s->block_max_depth; + slice_buffer_init(&s->sb, s->plane[0].height, (block_size) + (s->spatial_decomposition_count * (s->spatial_decomposition_count + 3)) + 1, s->plane[0].width, s->spatial_dwt_buffer); + return 0; } @@ -2902,11 +4350,6 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8 int bytes_read; AVFrame *picture = data; int level, orientation, plane_index; - - - /* no supplementary picture */ - if (buf_size == 0) - return 0; ff_init_range_decoder(c, buf, buf_size); ff_build_rac_states(c, 0.05*(1LL<<32), 256-8); @@ -2919,7 +4362,7 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8 //keyframe flag dupliaction mess FIXME if(avctx->debug&FF_DEBUG_PICT_INFO) av_log(avctx, AV_LOG_ERROR, "keyframe:%d qlog:%d\n", s->keyframe, s->qlog); - + decode_blocks(s); for(plane_index=0; plane_index<3; plane_index++){ @@ -2927,7 +4370,8 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, uint8 int w= p->width; int h= p->height; int x, y; - + int decode_state[MAX_DECOMPOSITIONS][4][1]; /* Stored state info for unpack_coeffs. 1 variable per instance. */ + if(s->avctx->debug&2048){ memset(s->spatial_dwt_buffer, 0, sizeof(DWTELEM)*w*h); predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); @@ -2939,44 +4383,114 @@ if(s->avctx->debug&2048){ } } } + +{ START_TIMER + for(level=0; levelspatial_decomposition_count; level++){ + for(orientation=level ? 1 : 0; orientation<4; orientation++){ + SubBand *b= &p->band[level][orientation]; + unpack_coeffs(s, b, b->parent, orientation); + } + } + STOP_TIMER("unpack coeffs"); +} + +{START_TIMER + const int mb_h= s->b_height << s->block_max_depth; + const int block_size = MB_SIZE >> s->block_max_depth; + const int block_w = plane_index ? block_size/2 : block_size; + int mb_y; + dwt_compose_t cs[MAX_DECOMPOSITIONS]; + int yd=0, yq=0; + int y; + int end_y; + + ff_spatial_idwt_buffered_init(cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count); + for(mb_y=0; mb_y<=mb_h; mb_y++){ + + int slice_starty = block_w*mb_y; + int slice_h = block_w*(mb_y+1); + if (!(s->keyframe || s->avctx->debug&512)){ + slice_starty = FFMAX(0, slice_starty - (block_w >> 1)); + slice_h -= (block_w >> 1); + } + + { + START_TIMER for(level=0; levelspatial_decomposition_count; level++){ for(orientation=level ? 1 : 0; orientation<4; orientation++){ SubBand *b= &p->band[level][orientation]; - - decode_subband(s, b, b->buf, b->parent ? b->parent->buf : NULL, b->stride, orientation); - if(orientation==0){ - correlate(s, b, b->buf, b->stride, 1, 0); - dequantize(s, b, b->buf, b->stride); - assert(b->buf == s->spatial_dwt_buffer); + int start_y; + int end_y; + int our_mb_start = mb_y; + int our_mb_end = (mb_y + 1); + const int extra= 3; + start_y = (mb_y ? ((block_w * our_mb_start) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra: 0); + end_y = (((block_w * our_mb_end) >> (s->spatial_decomposition_count - level)) + s->spatial_decomposition_count - level + extra); + if (!(s->keyframe || s->avctx->debug&512)){ + start_y = FFMAX(0, start_y - (block_w >> (1+s->spatial_decomposition_count - level))); + end_y = FFMAX(0, end_y - (block_w >> (1+s->spatial_decomposition_count - level))); + } + start_y = FFMIN(b->height, start_y); + end_y = FFMIN(b->height, end_y); + + if (start_y != end_y){ + if (orientation == 0){ + SubBand * correlate_band = &p->band[0][0]; + int correlate_end_y = FFMIN(b->height, end_y + 1); + int correlate_start_y = FFMIN(b->height, (start_y ? start_y + 1 : 0)); + decode_subband_slice_buffered(s, correlate_band, &s->sb, correlate_start_y, correlate_end_y, decode_state[0][0]); + correlate_slice_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride, 1, 0, correlate_start_y, correlate_end_y); + dequantize_slice_buffered(s, &s->sb, correlate_band, correlate_band->buf, correlate_band->stride, start_y, end_y); + } + else + decode_subband_slice_buffered(s, b, &s->sb, start_y, end_y, decode_state[level][orientation]); } } } + STOP_TIMER("decode_subband_slice"); + } + +{ START_TIMER + for(; yddsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd); + } + STOP_TIMER("idwt slice");} + - ff_spatial_idwt(s->spatial_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count); if(s->qlog == LOSSLESS_QLOG){ - for(y=0; ysb, yq); for(x=0; xspatial_dwt_buffer[y*w + x]<<=FRAC_BITS; + line[x] <<= FRAC_BITS; } } } -{START_TIMER - predict_plane(s, s->spatial_dwt_buffer, plane_index, 1); -STOP_TIMER("predict_plane conv2")} + + predict_slice_buffered(s, &s->sb, s->spatial_dwt_buffer, plane_index, 1, mb_y); + + y = FFMIN(p->height, slice_starty); + end_y = FFMIN(p->height, slice_h); + while(y < end_y) + slice_buffer_release(&s->sb, y++); } - + + slice_buffer_flush(&s->sb); + +STOP_TIMER("idwt + predict_slices")} + } + emms_c(); - if(s->last_picture.data[0]) - avctx->release_buffer(avctx, &s->last_picture); + if(s->last_picture[s->max_ref_frames-1].data[0]) + avctx->release_buffer(avctx, &s->last_picture[s->max_ref_frames-1]); -if(!(s->avctx->debug&2048)) +if(!(s->avctx->debug&2048)) *picture= s->current_picture; else *picture= s->mconly_picture; - + *data_size = sizeof(AVFrame); - + bytes_read= c->bytestream - c->bytestream_start; if(bytes_read ==0) av_log(s->avctx, AV_LOG_ERROR, "error at end of frame\n"); //FIXME @@ -2987,6 +4501,8 @@ static int decode_end(AVCodecContext *avctx) { SnowContext *s = avctx->priv_data; + slice_buffer_destroy(&s->sb); + common_end(s); return 0; @@ -3031,14 +4547,14 @@ int main(){ int i; s.spatial_decomposition_count=6; s.spatial_decomposition_type=1; - + printf("testing 5/3 DWT\n"); for(i=0; ispatial_decomposition_type, s->spatial_decomposition_count); - ff_spatial_idwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); - + + ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); + ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); + for(i=0; ispatial_decomposition_type, s->spatial_decomposition_count); - ff_spatial_idwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); - + + ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); + ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); + for(i=0; i20) printf("fsck: %d %d %d\n",i, buffer[0][i], buffer[1][i]); + +#if 0 printf("testing AC coder\n"); memset(s.header_state, 0, sizeof(s.header_state)); ff_init_range_encoder(&s.c, buffer[0], 256*256); ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); - + for(i=-256; i<256; i++){ START_TIMER - put_symbol(&s.c, s.header_state, i*i*i/3*ABS(i), 1); + put_symbol(&s.c, s.header_state, i*i*i/3*FFABS(i), 1); STOP_TIMER("put_symbol") } ff_rac_terminate(&s.c); @@ -3068,14 +4585,15 @@ STOP_TIMER("put_symbol") memset(s.header_state, 0, sizeof(s.header_state)); ff_init_range_decoder(&s.c, buffer[0], 256*256); ff_init_cabac_states(&s.c, ff_h264_lps_range, ff_h264_mps_state, ff_h264_lps_state, 64); - + for(i=-256; i<256; i++){ int j; START_TIMER j= get_symbol(&s.c, s.header_state, 1); STOP_TIMER("get_symbol") - if(j!=i*i*i/3*ABS(i)) printf("fsck: %d != %d\n", i, j); + if(j!=i*i*i/3*FFABS(i)) printf("fsck: %d != %d\n", i, j); } +#endif { int level, orientation, x, y; int64_t errors[8][4]; @@ -3094,17 +4612,17 @@ int64_t g=0; if(orientation&1) buf+=w; if(orientation>1) buf+=stride>>1; - + memset(buffer[0], 0, sizeof(int)*width*height); buf[w/2 + h/2*stride]= 256*256; - ff_spatial_idwt(buffer[0], width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count); + ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); for(y=0; y>1; - + memset(buffer[0], 0, sizeof(int)*width*height); #if 1 for(y=0; yspatial_decomposition_type, s->spatial_decomposition_count); + ff_spatial_dwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); #else for(y=0; yspatial_decomposition_type, s->spatial_decomposition_count); + ff_spatial_idwt(buffer[0], width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count); #endif for(y=0; y