--- decoder.c 2002/12/08 05:38:56 1.37.2.15 +++ decoder.c 2003/08/02 15:08:12 1.49.2.10 @@ -1,64 +1,30 @@ -/************************************************************************** +/***************************************************************************** * * XVID MPEG-4 VIDEO CODEC - * - Decoder main module - + * - Decoder Module - * - * This program is an implementation of a part of one or more MPEG-4 - * Video tools as specified in ISO/IEC 14496-2 standard. Those intending - * to use this software module in hardware or software products are - * advised that its use may infringe existing patents or copyrights, and - * any such use would be at such party's own risk. The original - * developer of this software module and his/her company, and subsequent - * editors and their companies, will have no liability for use of this - * software or modifications or derivatives thereof. + * Copyright(C) 2002 MinChen + * 2002-2003 Peter Ross * - * This program is free software; you can redistribute it and/or modify + * This program is free software ; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or + * the Free Software Foundation ; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of + * but WITHOUT ANY WARRANTY ; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software + * along with this program ; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - *************************************************************************/ - -/************************************************************************** - * - * History: - * - * 15.07.2002 fix a bug in B-frame decode at DIRECT mode - * MinChen - * 10.07.2002 added BFRAMES_DEC_DEBUG support - * Fix a little bug for low_delay flage - * MinChen - * 28.06.2002 added basic resync support to iframe/pframe_decode() - * 22.06.2002 added primative N_VOP support - * #define BFRAMES_DEC now enables Minchen's bframe decoder - * 08.05.2002 add low_delay support for B_VOP decode - * MinChen - * 05.05.2002 fix some B-frame decode problem - * 02.05.2002 add B-frame decode support(have some problem); - * MinChen - * 22.04.2002 add some B-frame decode support; chenm001 - * 29.03.2002 interlacing fix - compensated block wasn't being used when - * reconstructing blocks, thus artifacts - * interlacing speedup - used transfers to re-interlace - * interlaced decoding should be as fast as progressive now - * 26.03.2002 interlacing support - moved transfers outside decode loop - * 26.12.2001 decoder_mbinter: dequant/idct moved within if(coded) block - * 22.12.2001 lock based interpolation - * 01.12.2001 inital version; (c)2001 peter ross - * - * $Id: decoder.c,v 1.37.2.15 2002/12/08 05:38:56 suxen_drol Exp $ + * $Id: decoder.c,v 1.49.2.10 2003/08/02 15:08:12 edgomez Exp $ * - *************************************************************************/ + ****************************************************************************/ +#include #include #include @@ -68,6 +34,7 @@ #include "xvid.h" #include "portab.h" +#include "global.h" #include "decoder.h" #include "bitstream/bitstream.h" @@ -80,12 +47,14 @@ #include "utils/mem_transfer.h" #include "image/interpolate8x8.h" #include "image/reduced.h" +#include "image/font.h" #include "bitstream/mbcoding.h" #include "prediction/mbprediction.h" #include "utils/timer.h" #include "utils/emms.h" #include "motion/motion.h" +#include "motion/gmc.h" #include "image/image.h" #include "image/colorspace.h" @@ -95,12 +64,13 @@ decoder_resize(DECODER * dec) { /* free existing */ - + image_destroy(&dec->cur, dec->edged_width, dec->edged_height); image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height); - image_destroy(&dec->refn[2], dec->edged_width, dec->edged_height); - image_destroy(&dec->refh, dec->edged_width, dec->edged_height); - image_destroy(&dec->cur, dec->edged_width, dec->edged_height); + image_destroy(&dec->tmp, dec->edged_width, dec->edged_height); + image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height); + + image_destroy(&dec->gmc, dec->edged_width, dec->edged_height); if (dec->last_mbs) xvid_free(dec->last_mbs); @@ -108,7 +78,6 @@ xvid_free(dec->mbs); /* realloc */ - dec->mb_width = (dec->width + 15) / 16; dec->mb_height = (dec->height + 15) / 16; @@ -126,27 +95,36 @@ return XVID_ERR_MEMORY; } - // add by chenm001 - // for support B-frame to reference last 2 frame + /* Support B-frame to reference last 2 frame */ if (image_create(&dec->refn[1], dec->edged_width, dec->edged_height)) { image_destroy(&dec->cur, dec->edged_width, dec->edged_height); image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); xvid_free(dec); return XVID_ERR_MEMORY; } - if (image_create(&dec->refn[2], dec->edged_width, dec->edged_height)) { + if (image_create(&dec->tmp, dec->edged_width, dec->edged_height)) { + image_destroy(&dec->cur, dec->edged_width, dec->edged_height); + image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); + image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height); + xvid_free(dec); + return XVID_ERR_MEMORY; + } + + if (image_create(&dec->qtmp, dec->edged_width, dec->edged_height)) { image_destroy(&dec->cur, dec->edged_width, dec->edged_height); image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height); + image_destroy(&dec->tmp, dec->edged_width, dec->edged_height); xvid_free(dec); return XVID_ERR_MEMORY; } - if (image_create(&dec->refh, dec->edged_width, dec->edged_height)) { + if (image_create(&dec->gmc, dec->edged_width, dec->edged_height)) { + image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height); image_destroy(&dec->cur, dec->edged_width, dec->edged_height); image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height); - image_destroy(&dec->refn[2], dec->edged_width, dec->edged_height); + image_destroy(&dec->tmp, dec->edged_width, dec->edged_height); xvid_free(dec); return XVID_ERR_MEMORY; } @@ -158,15 +136,14 @@ image_destroy(&dec->cur, dec->edged_width, dec->edged_height); image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height); - image_destroy(&dec->refn[2], dec->edged_width, dec->edged_height); - image_destroy(&dec->refh, dec->edged_width, dec->edged_height); + image_destroy(&dec->tmp, dec->edged_width, dec->edged_height); + image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height); xvid_free(dec); return XVID_ERR_MEMORY; } memset(dec->mbs, 0, sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height); - // add by chenm001 - // for skip MB flag + /* For skip MB flag */ dec->last_mbs = xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height, CACHE_LINE); @@ -175,48 +152,54 @@ image_destroy(&dec->cur, dec->edged_width, dec->edged_height); image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height); - image_destroy(&dec->refn[2], dec->edged_width, dec->edged_height); - image_destroy(&dec->refh, dec->edged_width, dec->edged_height); + image_destroy(&dec->tmp, dec->edged_width, dec->edged_height); + image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height); xvid_free(dec); return XVID_ERR_MEMORY; } memset(dec->last_mbs, 0, sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height); - return XVID_ERR_OK; + return 0; } int -decoder_create(XVID_DEC_PARAM * param) +decoder_create(xvid_dec_create_t * create) { DECODER *dec; + if (XVID_VERSION_MAJOR(create->version) != 1) /* v1.x.x */ + return XVID_ERR_VERSION; + dec = xvid_malloc(sizeof(DECODER), CACHE_LINE); if (dec == NULL) { return XVID_ERR_MEMORY; } memset(dec, 0, sizeof(DECODER)); - param->handle = dec; + create->handle = dec; - dec->width = param->width; - dec->height = param->height; + dec->width = create->width; + dec->height = create->height; image_null(&dec->cur); image_null(&dec->refn[0]); image_null(&dec->refn[1]); - image_null(&dec->refn[2]); - image_null(&dec->refh); + image_null(&dec->tmp); + image_null(&dec->qtmp); + + /* image based GMC */ + image_null(&dec->gmc); + dec->mbs = NULL; dec->last_mbs = NULL; init_timer(); - // add by chenm001 - // for support B-frame to save reference frame's time - dec->frames = -1; + /* For B-frame support (used to save reference frame's time */ + dec->frames = 0; dec->time = dec->time_base = dec->last_time_base = 0; dec->low_delay = 0; dec->packed_mode = 0; @@ -226,7 +209,7 @@ if (dec->fixed_dimensions) return decoder_resize(dec); else - return XVID_ERR_OK; + return 0; } @@ -235,15 +218,19 @@ { xvid_free(dec->last_mbs); xvid_free(dec->mbs); + + /* image based GMC */ + image_destroy(&dec->gmc, dec->edged_width, dec->edged_height); + image_destroy(&dec->refn[0], dec->edged_width, dec->edged_height); image_destroy(&dec->refn[1], dec->edged_width, dec->edged_height); - image_destroy(&dec->refn[2], dec->edged_width, dec->edged_height); - image_destroy(&dec->refh, dec->edged_width, dec->edged_height); + image_destroy(&dec->tmp, dec->edged_width, dec->edged_height); + image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height); image_destroy(&dec->cur, dec->edged_width, dec->edged_height); xvid_free(dec); write_timer(); - return XVID_ERR_OK; + return 0; } @@ -255,8 +242,7 @@ -// decode an intra macroblock - +/* decode an intra macroblock */ void decoder_mbintra(DECODER * dec, MACROBLOCK * pMB, @@ -291,7 +277,7 @@ pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3); } - memset(block, 0, 6 * 64 * sizeof(int16_t)); // clear + memset(block, 0, 6 * 64 * sizeof(int16_t)); /* clear */ for (i = 0; i < 6; i++) { uint32_t iDcScaler = get_dc_scaler(iQuant, i < 4); @@ -314,19 +300,19 @@ dc_dif = dc_size ? get_dc_dif(bs, dc_size) : 0; if (dc_size > 8) { - BitstreamSkip(bs, 1); // marker + BitstreamSkip(bs, 1); /* marker */ } block[i * 64 + 0] = dc_dif; start_coeff = 1; - DPRINTF(DPRINTF_COEFF,"block[0] %i", dc_dif); + DPRINTF(XVID_DEBUG_COEFF,"block[0] %i\n", dc_dif); } else { start_coeff = 0; } start_timer(); - if (cbp & (1 << (5 - i))) // coded + if (cbp & (1 << (5 - i))) /* coded */ { int direction = dec->alternate_vertical_scan ? 2 : pMB->acpred_directions[i]; @@ -383,47 +369,13 @@ - -#define SIGN(X) (((X)>0)?1:-1) -#define ABS(X) (((X)>0)?(X):-(X)) - -// decode an inter macroblock - -static void -set_block(uint8_t * dst, int stride, int width, int height, int color) -{ - int i; - for (i = 0; i < height; i++) - { - memset(dst, color, width); - dst += stride; - } -} - -static void -rrv_mv_scaleup(VECTOR * mv) -{ - if (mv->x > 0) { - mv->x = 2*mv->x - 1; - } else if (mv->x < 0) { - mv->x = 2*mv->x + 1; - } - - if (mv->y > 0) { - mv->y = 2*mv->y - 1; - } else if (mv->y < 0) { - mv->y = 2*mv->y + 1; - } -} - - - +/* decode an inter macroblock */ void decoder_mbinter(DECODER * dec, const MACROBLOCK * pMB, const uint32_t x_pos, const uint32_t y_pos, - const uint32_t acpred_flag, + const uint32_t fcode, const uint32_t cbp, Bitstream * bs, const uint32_t quant, @@ -442,51 +394,29 @@ uint8_t *pY_Cur, *pU_Cur, *pV_Cur; int uv_dx, uv_dy; - VECTOR mv[4]; - - for (i = 0; i < 4; i++) - { - mv[i] = pMB->mvs[i]; - //DPRINTF(DPRINTF_MB, "mv[%i] orig=%i,%i local=%i", i, pMB->mvs[i].x, pMB->mvs[i].y, mv[i].x, mv[i].y); - } + VECTOR mv[4]; /* local copy of mvs */ if (reduced_resolution) { pY_Cur = dec->cur.y + (y_pos << 5) * stride + (x_pos << 5); pU_Cur = dec->cur.u + (y_pos << 4) * stride2 + (x_pos << 4); pV_Cur = dec->cur.v + (y_pos << 4) * stride2 + (x_pos << 4); - DPRINTF(DPRINTF_MB,"[%i,%i] %i,%i %i,%i %i,%i %i,%i", - x_pos, y_pos, - mv[0].x, mv[0].y, - mv[1].x, mv[1].y, - mv[2].x, mv[2].y, - mv[3].x, mv[3].y); - - rrv_mv_scaleup(&mv[0]); - rrv_mv_scaleup(&mv[1]); - rrv_mv_scaleup(&mv[2]); - rrv_mv_scaleup(&mv[3]); - - DPRINTF(DPRINTF_MB," %i,%i %i,%i %i,%i %i,%i", - mv[0].x, mv[0].y, - mv[1].x, mv[1].y, - mv[2].x, mv[2].y, - mv[3].x, mv[3].y); - }else{ + for (i = 0; i < 4; i++) { + mv[i].x = RRV_MV_SCALEUP(pMB->mvs[i].x); + mv[i].y = RRV_MV_SCALEUP(pMB->mvs[i].y); + } + } else { pY_Cur = dec->cur.y + (y_pos << 4) * stride + (x_pos << 4); pU_Cur = dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3); pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3); + for (i = 0; i < 4; i++) + mv[i] = pMB->mvs[i]; } - + if (pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q) { - uv_dx = mv[0].x; - uv_dy = mv[0].y; - - if (dec->quarterpel) - { - uv_dx /= 2; - uv_dy /= 2; - } + uv_dx = mv[0].x / (1 + dec->quarterpel); + uv_dy = mv[0].y / (1 + dec->quarterpel); + uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3]; uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3]; @@ -504,13 +434,13 @@ else { if(dec->quarterpel) { - interpolate16x16_quarterpel(dec->cur.y, dec->refn[0].y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos, - mv[0].x, mv[0].y, stride, rounding); + interpolate16x16_quarterpel(dec->cur.y, dec->refn[0].y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos, + mv[0].x, mv[0].y, stride, rounding); } else { interpolate16x16_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos, - mv[0].x, mv[0].y, stride, rounding); + mv[0].x, mv[0].y, stride, rounding); } interpolate8x8_switch(dec->cur.u, dec->refn[0].u, 8 * x_pos, 8 * y_pos, @@ -542,49 +472,49 @@ { interpolate16x16_switch(dec->cur.y, dec->refn[0].y, 32*x_pos, 32*y_pos, mv[0].x, mv[0].y, stride, rounding); - interpolate16x16_switch(dec->cur.y, dec->refn[0].y, 32*x_pos + 16, 32*y_pos, + interpolate16x16_switch(dec->cur.y, dec->refn[0].y , 32*x_pos + 16, 32*y_pos, mv[1].x, mv[1].y, stride, rounding); - interpolate16x16_switch(dec->cur.y, dec->refn[0].y, 32*x_pos, 32*y_pos + 16, + interpolate16x16_switch(dec->cur.y, dec->refn[0].y , 32*x_pos, 32*y_pos + 16, mv[2].x, mv[2].y, stride, rounding); - interpolate16x16_switch(dec->cur.y, dec->refn[0].y, 32*x_pos + 16, 32*y_pos + 16, + interpolate16x16_switch(dec->cur.y, dec->refn[0].y , 32*x_pos + 16, 32*y_pos + 16, mv[3].x, mv[3].y, stride, rounding); - interpolate16x16_switch(dec->cur.u, dec->refn[0].u, 16 * x_pos, 16 * y_pos, + interpolate16x16_switch(dec->cur.u, dec->refn[0].u , 16 * x_pos, 16 * y_pos, uv_dx, uv_dy, stride2, rounding); - interpolate16x16_switch(dec->cur.v, dec->refn[0].v, 16 * x_pos, 16 * y_pos, + interpolate16x16_switch(dec->cur.v, dec->refn[0].v , 16 * x_pos, 16 * y_pos, uv_dx, uv_dy, stride2, rounding); - // set_block(pY_Cur, stride, 32, 32, 127); + /* set_block(pY_Cur, stride, 32, 32, 127); */ } else { if(dec->quarterpel) { - interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos, + interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y , dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos, mv[0].x, mv[0].y, stride, rounding); - interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos + 8, 16*y_pos, + interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y , dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos, mv[1].x, mv[1].y, stride, rounding); - interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos + 8, + interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y , dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos + 8, mv[2].x, mv[2].y, stride, rounding); - interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos + 8, 16*y_pos + 8, + interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y , dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos + 8, mv[3].x, mv[3].y, stride, rounding); } else { - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos, + interpolate8x8_switch(dec->cur.y, dec->refn[0].y , 16*x_pos, 16*y_pos, mv[0].x, mv[0].y, stride, rounding); - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos + 8, 16*y_pos, + interpolate8x8_switch(dec->cur.y, dec->refn[0].y , 16*x_pos + 8, 16*y_pos, mv[1].x, mv[1].y, stride, rounding); - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos + 8, + interpolate8x8_switch(dec->cur.y, dec->refn[0].y , 16*x_pos, 16*y_pos + 8, mv[2].x, mv[2].y, stride, rounding); - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos + 8, 16*y_pos + 8, + interpolate8x8_switch(dec->cur.y, dec->refn[0].y , 16*x_pos + 8, 16*y_pos + 8, mv[3].x, mv[3].y, stride, rounding); } - interpolate8x8_switch(dec->cur.u, dec->refn[0].u, 8 * x_pos, 8 * y_pos, + interpolate8x8_switch(dec->cur.u, dec->refn[0].u , 8 * x_pos, 8 * y_pos, uv_dx, uv_dy, stride2, rounding); - interpolate8x8_switch(dec->cur.v, dec->refn[0].v, 8 * x_pos, 8 * y_pos, + interpolate8x8_switch(dec->cur.v, dec->refn[0].v , 8 * x_pos, 8 * y_pos, uv_dx, uv_dy, stride2, rounding); } stop_comp_timer(); @@ -593,9 +523,9 @@ for (i = 0; i < 6; i++) { int direction = dec->alternate_vertical_scan ? 2 : 0; - if (cbp & (1 << (5 - i))) // coded + if (cbp & (1 << (5 - i))) /* coded */ { - memset(&block[i * 64], 0, 64 * sizeof(int16_t)); // clear + memset(&block[i * 64], 0, 64 * sizeof(int16_t)); /* clear */ start_timer(); get_inter_block(bs, &block[i * 64], direction); @@ -655,6 +585,115 @@ } +static void +decoder_mbgmc(DECODER * dec, + MACROBLOCK * const pMB, + const uint32_t x_pos, + const uint32_t y_pos, + const uint32_t fcode, + const uint32_t cbp, + Bitstream * bs, + const uint32_t quant, + const uint32_t rounding, + const int reduced_resolution) /* no reduced res support */ +{ + + DECLARE_ALIGNED_MATRIX(block, 6, 64, int16_t, CACHE_LINE); + DECLARE_ALIGNED_MATRIX(data, 6, 64, int16_t, CACHE_LINE); + + const uint32_t stride = dec->edged_width; + const uint32_t stride2 = stride / 2; + const uint32_t next_block = stride * (reduced_resolution ? 16 : 8); + uint32_t i; + const uint32_t iQuant = pMB->quant; + uint8_t *const pY_Cur=dec->cur.y + (y_pos << 4) * stride + (x_pos << 4); + uint8_t *const pU_Cur=dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3); + uint8_t *const pV_Cur=dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3); + + pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; + + start_timer(); + +/* this is where the calculations are done */ + + { NEW_GMC_DATA * gmc_data = &dec->new_gmc_data; + + gmc_data->predict_16x16(gmc_data, + dec->cur.y + y_pos*16*stride + x_pos*16, dec->refn[0].y, + stride, stride, x_pos, y_pos, rounding); + + gmc_data->predict_8x8(gmc_data, + dec->cur.u + y_pos*8*stride2 + x_pos*8, dec->refn[0].u, + dec->cur.v + y_pos*8*stride2 + x_pos*8, dec->refn[0].v, + stride2, stride2, x_pos, y_pos, rounding); + + gmc_data->get_average_mv(gmc_data, &pMB->amv, x_pos, y_pos, dec->quarterpel); + + pMB->amv.x = gmc_sanitize(pMB->amv.x, dec->quarterpel, fcode); + pMB->amv.y = gmc_sanitize(pMB->amv.y, dec->quarterpel, fcode); + } + pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->amv; + +/* + transfer16x16_copy(pY_Cur, dec->gmc.y + (y_pos << 4)*stride + (x_pos << 4), stride); + transfer8x8_copy(pU_Cur, dec->gmc.u + (y_pos << 3)*stride2 + (x_pos << 3), stride2); + transfer8x8_copy(pV_Cur, dec->gmc.v + (y_pos << 3)*stride2 + (x_pos << 3), stride2); +*/ + + + stop_transfer_timer(); + + if (!cbp) return; + + for (i = 0; i < 6; i++) { + int direction = dec->alternate_vertical_scan ? 2 : 0; + + if (cbp & (1 << (5 - i))) /* coded */ + { + memset(&block[i * 64], 0, 64 * sizeof(int16_t)); /* clear */ + + start_timer(); + get_inter_block(bs, &block[i * 64], direction); + stop_coding_timer(); + + start_timer(); + if (dec->quant_type == 0) { + dequant_inter(&data[i * 64], &block[i * 64], iQuant); + } else { + dequant4_inter(&data[i * 64], &block[i * 64], iQuant); + } + stop_iquant_timer(); + + start_timer(); + idct(&data[i * 64]); + stop_idct_timer(); + } + } + +/* interlace + GMC is this possible ??? */ +/* + if (dec->interlacing && pMB->field_dct) { + next_block = stride; + stride *= 2; + } +*/ + start_timer(); + if (cbp & 32) + transfer_16to8add(pY_Cur, &data[0 * 64], stride); + if (cbp & 16) + transfer_16to8add(pY_Cur + 8, &data[1 * 64], stride); + if (cbp & 8) + transfer_16to8add(pY_Cur + next_block, &data[2 * 64], stride); + if (cbp & 4) + transfer_16to8add(pY_Cur + 8 + next_block, &data[3 * 64], stride); + if (cbp & 2) + transfer_16to8add(pU_Cur, &data[4 * 64], stride2); + if (cbp & 1) + transfer_16to8add(pV_Cur, &data[5 * 64], stride2); + stop_transfer_timer(); +} + + void decoder_iframe(DECODER * dec, Bitstream * bs, @@ -664,13 +703,13 @@ { uint32_t bound; uint32_t x, y; - int mb_width = dec->mb_width; - int mb_height = dec->mb_height; + uint32_t mb_width = dec->mb_width; + uint32_t mb_height = dec->mb_height; if (reduced_resolution) { - mb_width /= 2; - mb_height /= 2; + mb_width = (dec->width + 31) / 32; + mb_height = (dec->height + 31) / 32; } bound = 0; @@ -696,7 +735,7 @@ } mb = &dec->mbs[y * dec->mb_width + x]; - DPRINTF(DPRINTF_MB, "macroblock (%i,%i) %08x", x, y, BitstreamShowBits(bs, 32)); + DPRINTF(XVID_DEBUG_MB, "macroblock (%i,%i) %08x\n", x, y, BitstreamShowBits(bs, 32)); mcbpc = get_mcbpc_intra(bs); mb->mode = mcbpc & 7; @@ -723,7 +762,7 @@ if (dec->interlacing) { mb->field_dct = BitstreamGetBit(bs); - DEBUG1("deci: field_dct: ", mb->field_dct); + DPRINTF(XVID_DEBUG_MB,"deci: field_dct: %i\n", mb->field_dct); } decoder_mbintra(dec, mb, x, y, acpred_flag, cbp, bs, quant, @@ -761,7 +800,7 @@ mv.x = get_mv(bs, fcode); mv.y = get_mv(bs, fcode); - DPRINTF(DPRINTF_MV,"mv_diff (%i,%i) pred (%i,%i)", mv.x, mv.y, pmv.x, pmv.y); + DPRINTF(XVID_DEBUG_MV,"mv_diff (%i,%i) pred (%i,%i) result (%i,%i)\n", mv.x, mv.y, pmv.x, pmv.y, mv.x+pmv.x, mv.y+pmv.y); mv.x += pmv.x; mv.y += pmv.y; @@ -784,21 +823,9 @@ -static __inline int gmc_sanitize(int value, int quarterpel, int fcode) -{ - int length = 1 << (fcode+4); - - if (quarterpel) value *= 2; - - if (value < -length) - return -length; - else if (value >= length) - return length-1; - else return value; -} -/* for P_VOP set gmc_mv to NULL */ +/* for P_VOP set gmc_warp to NULL */ void decoder_pframe(DECODER * dec, Bitstream * bs, @@ -807,19 +834,19 @@ int quant, int fcode, int intra_dc_threshold, - VECTOR * gmc_mv) + const WARPPOINTS *const gmc_warp) { uint32_t x, y; uint32_t bound; int cp_mb, st_mb; - int mb_width = dec->mb_width; - int mb_height = dec->mb_height; + uint32_t mb_width = dec->mb_width; + uint32_t mb_height = dec->mb_height; if (reduced_resolution) { - mb_width /= 2; - mb_height /= 2; + mb_width = (dec->width + 31) / 32; + mb_height = (dec->height + 31) / 32; } start_timer(); @@ -827,6 +854,23 @@ dec->width, dec->height); stop_edges_timer(); + if (gmc_warp) + { + + /* accuracy: 0==1/2, 1=1/4, 2=1/8, 3=1/16 */ +/* { + fprintf(stderr,"GMC parameters acc=%d(-> 1/%d), %d pts!!!\n", + dec->sprite_warping_accuracy,(2<sprite_warping_accuracy), + dec->sprite_warping_points); + }*/ + + generate_GMCparameters( dec->sprite_warping_points, + dec->sprite_warping_accuracy, gmc_warp, + dec->width, dec->height, &dec->new_gmc_data); + +/* image warping is done block-based in decoder_mbgmc(), now */ + } + bound = 0; for (y = 0; y < mb_height; y++) { @@ -834,7 +878,7 @@ for (x = 0; x < mb_width; x++) { MACROBLOCK *mb; - // skip stuffing + /* skip stuffing */ while (BitstreamShowBits(bs, 10) == 1) BitstreamSkip(bs, 10); @@ -847,10 +891,10 @@ } mb = &dec->mbs[y * dec->mb_width + x]; - DPRINTF(DPRINTF_MB, "macroblock (%i,%i) %08x", x, y, BitstreamShowBits(bs, 32)); + DPRINTF(XVID_DEBUG_MB, "macroblock (%i,%i) %08x\n", x, y, BitstreamShowBits(bs, 32)); - //if (!(dec->mb_skip[y*dec->mb_width + x]=BitstreamGetBit(bs))) // not_coded - if (!(BitstreamGetBit(bs))) // not_coded + /* if (!(dec->mb_skip[y*dec->mb_width + x]=BitstreamGetBit(bs))) */ /* not_coded */ + if (!(BitstreamGetBit(bs))) /* block _is_ coded */ { uint32_t mcbpc; uint32_t cbpc; @@ -858,15 +902,15 @@ uint32_t cbpy; uint32_t cbp; uint32_t intra; - int mcsel = 0; // mcsel: '0'=local motion, '1'=GMC + int mcsel = 0; /* mcsel: '0'=local motion, '1'=GMC */ cp_mb++; mcbpc = get_mcbpc_inter(bs); mb->mode = mcbpc & 7; cbpc = (mcbpc >> 4); - DPRINTF(DPRINTF_MB, "mode %i", mb->mode); - DPRINTF(DPRINTF_MB, "cbpc %i", cbpc); + DPRINTF(XVID_DEBUG_MB, "mode %i\n", mb->mode); + DPRINTF(XVID_DEBUG_MB, "cbpc %i\n", cbpc); acpred_flag = 0; intra = (mb->mode == MODE_INTRA || mb->mode == MODE_INTRA_Q); @@ -875,56 +919,56 @@ acpred_flag = BitstreamGetBit(bs); } - if (gmc_mv && (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q)) + if (gmc_warp && (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q)) { mcsel = BitstreamGetBit(bs); } cbpy = get_cbpy(bs, intra); - DPRINTF(DPRINTF_MB, "cbpy %i", cbpy); + DPRINTF(XVID_DEBUG_MB, "cbpy %i mcsel %i \n", cbpy,mcsel); cbp = (cbpy << 2) | cbpc; if (mb->mode == MODE_INTER_Q || mb->mode == MODE_INTRA_Q) { int dquant = dquant_table[BitstreamGetBits(bs, 2)]; - DPRINTF(DPRINTF_MB, "dquant %i", dquant); + DPRINTF(XVID_DEBUG_MB, "dquant %i\n", dquant); quant += dquant; if (quant > 31) { quant = 31; } else if (quant < 1) { quant = 1; } - DPRINTF(DPRINTF_MB, "quant %i", quant); + DPRINTF(XVID_DEBUG_MB, "quant %i\n", quant); } mb->quant = quant; if (dec->interlacing) { if (cbp || intra) { mb->field_dct = BitstreamGetBit(bs); - DEBUG1("decp: field_dct: ", mb->field_dct); + DPRINTF(XVID_DEBUG_MB,"decp: field_dct: %i\n", mb->field_dct); } if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { mb->field_pred = BitstreamGetBit(bs); - DEBUG1("decp: field_pred: ", mb->field_pred); + DPRINTF(XVID_DEBUG_MB, "decp: field_pred: %i\n", mb->field_pred); if (mb->field_pred) { mb->field_for_top = BitstreamGetBit(bs); - DEBUG1("decp: field_for_top: ", mb->field_for_top); + DPRINTF(XVID_DEBUG_MB,"decp: field_for_top: %i\n", mb->field_for_top); mb->field_for_bot = BitstreamGetBit(bs); - DEBUG1("decp: field_for_bot: ", mb->field_for_bot); + DPRINTF(XVID_DEBUG_MB,"decp: field_for_bot: %i\n", mb->field_for_bot); } } } + + if (mcsel) { + decoder_mbgmc(dec, mb, x, y, fcode, cbp, bs, quant, + rounding, reduced_resolution); + continue; - if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { - - if (mcsel) - { - mb->mvs[0].x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = gmc_sanitize(gmc_mv[0].x, dec->quarterpel, fcode); - mb->mvs[0].y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = gmc_sanitize(gmc_mv[0].y, dec->quarterpel, fcode); + } else if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { - } else if (dec->interlacing && mb->field_pred) { + if (dec->interlacing && mb->field_pred) { get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); get_motion_vector(dec, bs, x, y, 0, &mb->mvs[1], @@ -932,10 +976,7 @@ } else { get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); - mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = - mb->mvs[0].x; - mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = - mb->mvs[0].y; + mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; } } else if (mb->mode == MODE_INTER4V ) { @@ -943,7 +984,7 @@ get_motion_vector(dec, bs, x, y, 1, &mb->mvs[1], fcode, bound); get_motion_vector(dec, bs, x, y, 2, &mb->mvs[2], fcode, bound); get_motion_vector(dec, bs, x, y, 3, &mb->mvs[3], fcode, bound); - } else // MODE_INTRA, MODE_INTRA_Q + } else /* MODE_INTRA, MODE_INTRA_Q */ { mb->mvs[0].x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = 0; @@ -954,16 +995,26 @@ continue; } - decoder_mbinter(dec, mb, x, y, acpred_flag, cbp, bs, quant, + decoder_mbinter(dec, mb, x, y, fcode, cbp, bs, quant, rounding, reduced_resolution); - } - else if (gmc_mv) /* not coded S_VOP macroblock */ + } + else if (gmc_warp) /* a not coded S(GMC)-VOP macroblock */ { - mb->mode = MODE_NOT_CODED; - mb->mvs[0].x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = gmc_sanitize(gmc_mv[0].x, dec->quarterpel, fcode); - mb->mvs[0].y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = gmc_sanitize(gmc_mv[0].y, dec->quarterpel, fcode); - decoder_mbinter(dec, mb, x, y, 0, 0, bs, quant, rounding, reduced_resolution); + mb->mode = MODE_NOT_CODED_GMC; + + start_timer(); + + decoder_mbgmc(dec, mb, x, y, fcode, 0x00, bs, quant, + rounding, reduced_resolution); + + stop_transfer_timer(); + + if(dec->out_frm && cp_mb > 0) { + output_slice(&dec->cur, dec->edged_width,dec->width,dec->out_frm,st_mb,y,cp_mb); + cp_mb = 0; + } + st_mb = x+1; } else /* not coded P_VOP macroblock */ { @@ -971,7 +1022,7 @@ mb->mvs[0].x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = 0; mb->mvs[0].y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = 0; - // copy macroblock directly from ref to cur + /* copy macroblock directly from ref to cur */ start_timer(); @@ -1019,8 +1070,7 @@ } -// add by MinChen -// decode B-frame motion vector +/* decode B-frame motion vector */ void get_b_motion_vector(DECODER * dec, Bitstream * bs, @@ -1064,8 +1114,7 @@ } -// add by MinChen -// decode an B-frame forward & backward inter macroblock +/* decode an B-frame forward & backward inter macroblock */ void decoder_bf_mbinter(DECODER * dec, const MACROBLOCK * pMB, @@ -1125,8 +1174,8 @@ start_timer(); if(dec->quarterpel) { - interpolate16x16_quarterpel(dec->cur.y, dec->refn[ref].y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos, + interpolate16x16_quarterpel(dec->cur.y, dec->refn[ref].y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos, pMB->mvs[0].x, pMB->mvs[0].y, stride, 0); } else { @@ -1149,9 +1198,9 @@ for (i = 0; i < 6; i++) { int direction = dec->alternate_vertical_scan ? 2 : 0; - if (cbp & (1 << (5 - i))) // coded + if (cbp & (1 << (5 - i))) /* coded */ { - memset(&block[i * 64], 0, 64 * sizeof(int16_t)); // clear + memset(&block[i * 64], 0, 64 * sizeof(int16_t)); /* clear */ start_timer(); get_inter_block(bs, &block[i * 64], direction); @@ -1192,8 +1241,7 @@ stop_transfer_timer(); } -// add by MinChen -// decode an B-frame direct & inter macroblock +/* decode an B-frame direct & inter macroblock */ void decoder_bf_interpolate_mbinter(DECODER * dec, IMAGE forward, @@ -1280,21 +1328,21 @@ start_timer(); if(dec->quarterpel) { if((pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q)) - interpolate16x16_quarterpel(dec->cur.y, forward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos, + interpolate16x16_quarterpel(dec->cur.y, forward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos, pMB->mvs[0].x, pMB->mvs[0].y, stride, 0); else { - interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos, + interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos, pMB->mvs[0].x, pMB->mvs[0].y, stride, 0); - interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos + 8, 16*y_pos, + interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos, pMB->mvs[1].x, pMB->mvs[1].y, stride, 0); - interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos + 8, + interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos + 8, pMB->mvs[2].x, pMB->mvs[2].y, stride, 0); - interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos + 8, 16*y_pos + 8, + interpolate8x8_quarterpel(dec->cur.y, forward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos + 8, pMB->mvs[3].x, pMB->mvs[3].y, stride, 0); } } @@ -1318,71 +1366,71 @@ if(dec->quarterpel) { if((pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q)) - interpolate16x16_quarterpel(dec->refn[2].y, backward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos, + interpolate16x16_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos, pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0); else { - interpolate8x8_quarterpel(dec->refn[2].y, backward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos, + interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos, pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0); - interpolate8x8_quarterpel(dec->refn[2].y, backward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos + 8, 16*y_pos, + interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos, pMB->b_mvs[1].x, pMB->b_mvs[1].y, stride, 0); - interpolate8x8_quarterpel(dec->refn[2].y, backward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos, 16*y_pos + 8, + interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos, 16*y_pos + 8, pMB->b_mvs[2].x, pMB->b_mvs[2].y, stride, 0); - interpolate8x8_quarterpel(dec->refn[2].y, backward.y, dec->refh.y, dec->refh.y + 64, - dec->refh.y + 128, 16*x_pos + 8, 16*y_pos + 8, + interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64, + dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos + 8, pMB->b_mvs[3].x, pMB->b_mvs[3].y, stride, 0); } } else { - interpolate8x8_switch(dec->refn[2].y, backward.y, 16 * x_pos, 16 * y_pos, + interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos, 16 * y_pos, pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0); - interpolate8x8_switch(dec->refn[2].y, backward.y, 16 * x_pos + 8, + interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos + 8, 16 * y_pos, pMB->b_mvs[1].x, pMB->b_mvs[1].y, stride, 0); - interpolate8x8_switch(dec->refn[2].y, backward.y, 16 * x_pos, + interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos, 16 * y_pos + 8, pMB->b_mvs[2].x, pMB->b_mvs[2].y, stride, 0); - interpolate8x8_switch(dec->refn[2].y, backward.y, 16 * x_pos + 8, + interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos + 8, 16 * y_pos + 8, pMB->b_mvs[3].x, pMB->b_mvs[3].y, stride, 0); } - interpolate8x8_switch(dec->refn[2].u, backward.u, 8 * x_pos, 8 * y_pos, + interpolate8x8_switch(dec->tmp.u, backward.u, 8 * x_pos, 8 * y_pos, b_uv_dx, b_uv_dy, stride2, 0); - interpolate8x8_switch(dec->refn[2].v, backward.v, 8 * x_pos, 8 * y_pos, + interpolate8x8_switch(dec->tmp.v, backward.v, 8 * x_pos, 8 * y_pos, b_uv_dx, b_uv_dy, stride2, 0); interpolate8x8_avg2(dec->cur.y + (16 * y_pos * stride) + 16 * x_pos, dec->cur.y + (16 * y_pos * stride) + 16 * x_pos, - dec->refn[2].y + (16 * y_pos * stride) + 16 * x_pos, + dec->tmp.y + (16 * y_pos * stride) + 16 * x_pos, stride, 1, 8); interpolate8x8_avg2(dec->cur.y + (16 * y_pos * stride) + 16 * x_pos + 8, dec->cur.y + (16 * y_pos * stride) + 16 * x_pos + 8, - dec->refn[2].y + (16 * y_pos * stride) + 16 * x_pos + 8, + dec->tmp.y + (16 * y_pos * stride) + 16 * x_pos + 8, stride, 1, 8); interpolate8x8_avg2(dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos, dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos, - dec->refn[2].y + ((16 * y_pos + 8) * stride) + 16 * x_pos, + dec->tmp.y + ((16 * y_pos + 8) * stride) + 16 * x_pos, stride, 1, 8); interpolate8x8_avg2(dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos + 8, dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos + 8, - dec->refn[2].y + ((16 * y_pos + 8) * stride) + 16 * x_pos + 8, + dec->tmp.y + ((16 * y_pos + 8) * stride) + 16 * x_pos + 8, stride, 1, 8); interpolate8x8_avg2(dec->cur.u + (8 * y_pos * stride2) + 8 * x_pos, dec->cur.u + (8 * y_pos * stride2) + 8 * x_pos, - dec->refn[2].u + (8 * y_pos * stride2) + 8 * x_pos, + dec->tmp.u + (8 * y_pos * stride2) + 8 * x_pos, stride2, 1, 8); interpolate8x8_avg2(dec->cur.v + (8 * y_pos * stride2) + 8 * x_pos, dec->cur.v + (8 * y_pos * stride2) + 8 * x_pos, - dec->refn[2].v + (8 * y_pos * stride2) + 8 * x_pos, + dec->tmp.v + (8 * y_pos * stride2) + 8 * x_pos, stride2, 1, 8); stop_comp_timer(); @@ -1390,9 +1438,9 @@ for (i = 0; i < 6; i++) { int direction = dec->alternate_vertical_scan ? 2 : 0; - if (cbp & (1 << (5 - i))) // coded + if (cbp & (1 << (5 - i))) /* coded */ { - memset(&block[i * 64], 0, 64 * sizeof(int16_t)); // clear + memset(&block[i * 64], 0, 64 * sizeof(int16_t)); /* clear */ start_timer(); get_inter_block(bs, &block[i * 64], direction); @@ -1434,26 +1482,26 @@ } -// add by MinChen -// for decode B-frame dbquant +/* for decode B-frame dbquant */ int32_t __inline get_dbquant(Bitstream * bs) { - if (!BitstreamGetBit(bs)) // '0' + if (!BitstreamGetBit(bs)) /* '0' */ return (0); - else if (!BitstreamGetBit(bs)) // '10' + else if (!BitstreamGetBit(bs)) /* '10' */ return (-2); - else - return (2); // '11' + else /* '11' */ + return (2); } -// add by MinChen -// for decode B-frame mb_type -// bit ret_value -// 1 0 -// 01 1 -// 001 2 -// 0001 3 +/* + * For decode B-frame mb_type + * bit ret_value + * 1 0 + * 01 1 + * 001 2 + * 0001 3 + */ int32_t __inline get_mbtype(Bitstream * bs) { @@ -1502,7 +1550,7 @@ #endif for (y = 0; y < dec->mb_height; y++) { - // Initialize Pred Motion Vector + /* Initialize Pred Motion Vector */ dec->p_fmv = dec->p_bmv = zeromv; for (x = 0; x < dec->mb_width; x++) { MACROBLOCK *mb = &dec->mbs[y * dec->mb_width + x]; @@ -1512,9 +1560,14 @@ mb->b_mvs[0] = mb->b_mvs[1] = mb->b_mvs[2] = mb->b_mvs[3] = mb->mvs[0] = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = zeromv; - // the last P_VOP is skip macroblock ? + /* + * skip if the co-located P_VOP macroblock is not coded + * if not codec in co-located S_VOP macroblock is _not_ + * automatically skipped + */ + if (last_mb->mode == MODE_NOT_CODED) { - //DEBUG2("Skip MB in B-frame at (X,Y)=!",x,y); + /* DEBUG2("Skip MB in B-frame at (X,Y)=!",x,y); */ mb->cbp = 0; #ifdef BFRAMES_DEC_DEBUG mb->mb_type = MODE_NOT_CODED; @@ -1522,19 +1575,21 @@ #endif mb->mb_type = MODE_FORWARD; mb->quant = last_mb->quant; - //mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = mb->mvs[0].x; - //mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = mb->mvs[0].y; + /* + mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = mb->mvs[0].x; + mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = mb->mvs[0].y; + */ decoder_bf_mbinter(dec, mb, x, y, mb->cbp, bs, mb->quant, 1); continue; } - if (!BitstreamGetBit(bs)) { // modb=='0' + if (!BitstreamGetBit(bs)) { /* modb=='0' */ const uint8_t modb2 = BitstreamGetBit(bs); mb->mb_type = get_mbtype(bs); - if (!modb2) { // modb=='00' + if (!modb2) { /* modb=='00' */ mb->cbp = BitstreamGetBits(bs, 6); } else { mb->cbp = 0; @@ -1555,7 +1610,7 @@ mb->quant = quant; mb->mode = MODE_INTER4V; - //DEBUG1("Switch bm_type=",mb->mb_type); + /* DEBUG1("Switch bm_type=",mb->mb_type); */ #ifdef BFRAMES_DEC_DEBUG BFRAME_DEBUG @@ -1584,7 +1639,7 @@ / TRD : mb->mvs[i].y - last_mb->mvs[i].y); } - //DEBUG("B-frame Direct!\n"); + /* DEBUG("B-frame Direct!\n"); */ } decoder_bf_interpolate_mbinter(dec, dec->refn[1], dec->refn[0], mb, x, y, bs); @@ -1602,7 +1657,7 @@ decoder_bf_interpolate_mbinter(dec, dec->refn[1], dec->refn[0], mb, x, y, bs); - //DEBUG("B-frame Bidir!\n"); + /* DEBUG("B-frame Bidir!\n"); */ break; case MODE_BACKWARD: @@ -1612,7 +1667,7 @@ mb->mode = MODE_INTER; decoder_bf_mbinter(dec, mb, x, y, mb->cbp, bs, quant, 0); - //DEBUG("B-frame Backward!\n"); + /* DEBUG("B-frame Backward!\n"); */ break; case MODE_FORWARD: @@ -1622,15 +1677,15 @@ mb->mode = MODE_INTER; decoder_bf_mbinter(dec, mb, x, y, mb->cbp, bs, quant, 1); - //DEBUG("B-frame Forward!\n"); + /* DEBUG("B-frame Forward!\n"); */ break; default: - DEBUG1("Not support B-frame mb_type =", mb->mb_type); + DPRINTF(XVID_DEBUG_ERROR,"Not support B-frame mb_type = %i\n", mb->mb_type); } - - } // end of FOR + } /* End of for */ } + #ifdef BFRAMES_DEC_DEBUG if (!first){ first=1; @@ -1640,20 +1695,30 @@ #endif } -// swap two MACROBLOCK array -void -mb_swap(MACROBLOCK ** mb1, - MACROBLOCK ** mb2) + + +/* perform post processing if necessary, and output the image */ +void decoder_output(DECODER * dec, IMAGE * img, MACROBLOCK * mbs, + xvid_dec_frame_t * frame, xvid_dec_stats_t * stats, int coding_type) { - MACROBLOCK *temp = *mb1; - *mb1 = *mb2; - *mb2 = temp; + + image_output(img, dec->width, dec->height, + dec->edged_width, (uint8_t**)frame->output.plane, frame->output.stride, + frame->output.csp, dec->interlacing); + + if (stats) + { + stats->type = coding2type(coding_type); + stats->data.vop.time_base = (int)dec->time_base; + stats->data.vop.time_increment = 0; /* XXX: todo */ + } } + int decoder_decode(DECODER * dec, - XVID_DEC_FRAME * frame, XVID_DEC_STATS * stats) + xvid_dec_frame_t * frame, xvid_dec_stats_t * stats) { Bitstream bs; @@ -1663,248 +1728,235 @@ uint32_t fcode_forward; uint32_t fcode_backward; uint32_t intra_dc_threshold; - VECTOR gmc_mv[5]; - uint32_t vop_type; - int success = 0; + WARPPOINTS gmc_warp; + int coding_type; + int success, output, seen_something; + idctFuncPtr idct_save; + + if (XVID_VERSION_MAJOR(frame->version) != 1 || (stats && XVID_VERSION_MAJOR(stats->version) != 1)) /* v1.x.x */ + return XVID_ERR_VERSION; start_global_timer(); - dec->out_frm = (frame->colorspace == XVID_CSP_EXTERN) ? frame->image : NULL; + dec->low_delay_default = (frame->general & XVID_LOWDELAY); + if ((frame->general & XVID_DISCONTINUITY)) + dec->frames = 0; + dec->out_frm = (frame->output.csp == XVID_CSP_SLICE) ? &frame->output : NULL; + + if (frame->length < 0) /* decoder flush */ + { + int ret; + /* if not decoding "low_delay/packed", and this isn't low_delay and + we have a reference frame, then outout the reference frame */ + if (!(dec->low_delay_default && dec->packed_mode) && !dec->low_delay && dec->frames>0) { + decoder_output(dec, &dec->refn[0], dec->last_mbs, frame, stats, dec->last_coding_type); + dec->frames = 0; + ret = 0; + }else{ + if (stats) stats->type = XVID_TYPE_NOTHING; + ret = XVID_ERR_END; + } + + emms(); + stop_global_timer(); + return ret; + } BitstreamInit(&bs, frame->bitstream, frame->length); - // XXX: 0x7f is only valid whilst decoding vfw xvid/divx5 avi's - if(frame->length == 1 && BitstreamShowBits(&bs, 8) == 0x7f) + /* XXX: 0x7f is only valid whilst decoding vfw xvid/divx5 avi's */ + if(dec->low_delay_default && frame->length == 1 && BitstreamShowBits(&bs, 8) == 0x7f) { - if (stats) - stats->notify = XVID_DEC_VOP; - frame->length = 1; image_output(&dec->refn[0], dec->width, dec->height, dec->edged_width, - frame->image, frame->stride, frame->colorspace, dec->interlacing); + (uint8_t**)frame->output.plane, frame->output.stride, frame->output.csp, dec->interlacing); + if (stats) stats->type = XVID_TYPE_NOTHING; emms(); - return XVID_ERR_OK; + return 1; /* one byte consumed */ } -start: - // add by chenm001 - // for support B-frame to reference last 2 frame - dec->frames++; + success = 0; + output = 0; + seen_something = 0; + idct_save = idct; -xxx: - vop_type = - BitstreamReadHeaders(&bs, dec, &rounding, &reduced_resolution, - &quant, &fcode_forward, &fcode_backward, &intra_dc_threshold, gmc_mv); +repeat: - //DPRINTF(DPRINTF_HEADER, "vop_type=%i", vop_type); + coding_type = BitstreamReadHeaders(&bs, dec, &rounding, &reduced_resolution, + &quant, &fcode_forward, &fcode_backward, &intra_dc_threshold, &gmc_warp); - if (vop_type == -1 && success) - goto done; + DPRINTF(XVID_DEBUG_HEADER, "coding_type=%i, packed=%i, time=%lli, time_pp=%i, time_bp=%i\n", + coding_type, dec->packed_mode, dec->time, dec->time_pp, dec->time_bp); - if (vop_type == -2 || vop_type == -3) + if (coding_type == -1) /* nothing */ { - if (vop_type == -3) + if (success) goto done; + if (stats) stats->type = XVID_TYPE_NOTHING; + emms(); + return BitstreamPos(&bs)/8; + } + + if (coding_type == -2 || coding_type == -3) /* vol and/or resize */ + { + if (coding_type == -3) decoder_resize(dec); if (stats) { - stats->notify = XVID_DEC_VOL; + stats->type = XVID_TYPE_VOL; stats->data.vol.general = 0; - if (dec->interlacing) - stats->data.vol.general |= XVID_INTERLACING; + /*XXX: if (dec->interlacing) + stats->data.vol.general |= ++INTERLACING; */ stats->data.vol.width = dec->width; stats->data.vol.height = dec->height; - stats->data.vol.aspect_ratio = dec->aspect_ratio; + stats->data.vol.par = dec->aspect_ratio; stats->data.vol.par_width = dec->par_width; stats->data.vol.par_height = dec->par_height; - frame->length = BitstreamPos(&bs) / 8; - return XVID_ERR_OK; + emms(); + return BitstreamPos(&bs)/8; /* number of bytes consumed */ } - goto xxx; + goto repeat; } - dec->p_bmv.x = dec->p_bmv.y = dec->p_fmv.y = dec->p_fmv.y = 0; // init pred vector to 0 - - switch (vop_type) { - case P_VOP: - decoder_pframe(dec, &bs, rounding, reduced_resolution, quant, - fcode_forward, intra_dc_threshold, NULL); -#ifdef BFRAMES_DEC - DEBUG1("P_VOP Time=", dec->time); -#endif - break; + dec->p_bmv.x = dec->p_bmv.y = dec->p_fmv.y = dec->p_fmv.y = 0; /* init pred vector to 0 */ - case I_VOP: - decoder_iframe(dec, &bs, reduced_resolution, quant, intra_dc_threshold); -#ifdef BFRAMES_DEC - DEBUG1("I_VOP Time=", dec->time); +#if defined(ARCH_IS_IA32) + /* + * /!\ Ugly hack /!\ + * IA32: Prior to xvid bitstream 10, we were using Walten's mmx/xmm idct + */ + if((idct == simple_idct_mmx) && (dec->bs_version < 10)) + idct = idct_mmx; #endif - break; - case B_VOP: -#ifdef BFRAMES_DEC - if (dec->time_pp > dec->time_bp) { - DEBUG1("B_VOP Time=", dec->time); - decoder_bframe(dec, &bs, quant, fcode_forward, fcode_backward); - } else { - DEBUG("broken B-frame!"); + /* packed_mode: special-N_VOP treament */ + if (dec->packed_mode && coding_type == N_VOP) + { + if (dec->low_delay_default && dec->frames > 0) + { + decoder_output(dec, &dec->refn[0], dec->last_mbs, frame, stats, dec->last_coding_type); + output = 1; } -#else - image_copy(&dec->cur, &dec->refn[0], dec->edged_width, dec->height); -#endif - break; - - case S_VOP : - decoder_pframe(dec, &bs, rounding, reduced_resolution, quant, - fcode_forward, intra_dc_threshold, gmc_mv); - break; - - case N_VOP: // vop not coded - // when low_delay==0, N_VOP's should interpolate between the past and future frames - image_copy(&dec->cur, &dec->refn[0], dec->edged_width, dec->height); -#ifdef BFRAMES_DEC - DEBUG1("N_VOP Time=", dec->time); -#endif - break; - - default: - if (stats) - stats->notify = 0; - - emms(); - return XVID_ERR_FAIL; + /* ignore otherwise */ } - - - /* reduced resolution deblocking filter */ - - if (reduced_resolution) + else if (coding_type != B_VOP) { - const int rmb_height = dec->mb_height / 2; - const int rmb_width = dec->mb_width / 2; - const int edged_width2 = dec->edged_width /2; - int i,j; - - /* horizontal deblocking */ - - for (j = 1; j < rmb_height*2; j++) // luma: j,i in block units - for (i = 0; i < rmb_width*2; i++) + switch(coding_type) { - if (dec->mbs[(j-1)/2*dec->mb_width + (i/2)].mode != MODE_NOT_CODED || - dec->mbs[(j+0)/2*dec->mb_width + (i/2)].mode != MODE_NOT_CODED) - { - xvid_HFilter_31_C(dec->cur.y + (j*16 - 1)*dec->edged_width + i*16, - dec->cur.y + (j*16 + 0)*dec->edged_width + i*16, 2); - } + case I_VOP : + decoder_iframe(dec, &bs, reduced_resolution, quant, intra_dc_threshold); + break; + case P_VOP : + decoder_pframe(dec, &bs, rounding, reduced_resolution, quant, + fcode_forward, intra_dc_threshold, NULL); + break; + case S_VOP : + decoder_pframe(dec, &bs, rounding, reduced_resolution, quant, + fcode_forward, intra_dc_threshold, &gmc_warp); + break; + case N_VOP : + /* XXX: not_coded vops are not used for forward prediction */ + /* we should not swap(last_mbs,mbs) */ + image_copy(&dec->cur, &dec->refn[0], dec->edged_width, dec->height); + break; } - for (j = 1; j < rmb_height; j++) // chroma - for (i = 0; i < rmb_width; i++) + if (reduced_resolution) { - if (dec->mbs[(j-1)*dec->mb_width + i].mode != MODE_NOT_CODED || - dec->mbs[(j+0)*dec->mb_width + i].mode != MODE_NOT_CODED) - { - hfilter_31(dec->cur.u + (j*16 - 1)*edged_width2 + i*16, - dec->cur.u + (j*16 + 0)*edged_width2 + i*16, 2); - hfilter_31(dec->cur.v + (j*16 - 1)*edged_width2 + i*16, - dec->cur.v + (j*16 + 0)*edged_width2 + i*16, 2); - } + image_deblock_rrv(&dec->cur, dec->edged_width, dec->mbs, + (dec->width + 31) / 32, (dec->height + 31) / 32, dec->mb_width, + 16, 0); } - /* vertical deblocking */ - - for (j = 0; j < rmb_height*2; j++) // luma: i,j in block units - for (i = 1; i < rmb_width*2; i++) + /* note: for packed_mode, output is performed when the special-N_VOP is decoded */ + if (!(dec->low_delay_default && dec->packed_mode)) { - if (dec->mbs[(j/2)*dec->mb_width + (i-1)/2].mode != MODE_NOT_CODED || - dec->mbs[(j/2)*dec->mb_width + (i+0)/2].mode != MODE_NOT_CODED) + if (dec->low_delay) { - vfilter_31(dec->cur.y + (j*16)*dec->edged_width + i*16 - 1, - dec->cur.y + (j*16)*dec->edged_width + i*16 + 0, - dec->edged_width, 2); + decoder_output(dec, &dec->cur, dec->mbs, frame, stats, coding_type); + output = 1; } - } - - for (j = 0; j < rmb_height; j++) // chroma - for (i = 1; i < rmb_width; i++) - { - if (dec->mbs[j*dec->mb_width + i - 1].mode != MODE_NOT_CODED || - dec->mbs[j*dec->mb_width + i + 0].mode != MODE_NOT_CODED) + else if (dec->frames > 0) /* is the reference frame valid? */ { - vfilter_31(dec->cur.u + (j*16)*edged_width2 + i*16 - 1, - dec->cur.u + (j*16)*edged_width2 + i*16 + 0, - edged_width2, 2); - vfilter_31(dec->cur.v + (j*16)*edged_width2 + i*16 - 1, - dec->cur.v + (j*16)*edged_width2 + i*16 + 0, - edged_width2, 2); + /* output the reference frame */ + decoder_output(dec, &dec->refn[0], dec->last_mbs, frame, stats, dec->last_coding_type); + output = 1; } } - } - BitstreamByteAlign(&bs); + image_swap(&dec->refn[0], &dec->refn[1]); + image_swap(&dec->cur, &dec->refn[0]); + SWAP(MACROBLOCK *, dec->mbs, dec->last_mbs); + dec->last_reduced_resolution = reduced_resolution; + dec->last_coding_type = coding_type; -#ifdef BFRAMES_DEC - // test if no B_VOP - if (dec->low_delay || dec->frames == 0 || ((dec->packed_mode) && !(frame->length > BitstreamPos(&bs) / 8))) { -#endif - image_output(&dec->cur, dec->width, dec->height, dec->edged_width, - frame->image, frame->stride, frame->colorspace, dec->interlacing); + dec->frames++; + seen_something = 1; -#ifdef BFRAMES_DEC - } else { - if (dec->frames >= 1 && !(dec->packed_mode)) { - start_timer(); - if ((vop_type == I_VOP || vop_type == P_VOP || vop_type == S_VOP)) { - image_output(&dec->refn[0], dec->width, dec->height, - dec->edged_width, frame->image, frame->stride, - frame->colorspace, dec->interlacing); - } else if (vop_type == B_VOP) { - image_output(&dec->cur, dec->width, dec->height, - dec->edged_width, frame->image, frame->stride, - frame->colorspace, dec->interlacing); - } - stop_conv_timer(); + }else{ /* B_VOP */ + + if (dec->low_delay) + { + DPRINTF(XVID_DEBUG_ERROR, "warning: bvop found in low_delay==1 stream\n"); + dec->low_delay = 1; } - } -#endif - if (vop_type == I_VOP || vop_type == P_VOP || vop_type == S_VOP) { - image_swap(&dec->refn[0], &dec->refn[1]); - image_swap(&dec->cur, &dec->refn[0]); + if (dec->frames < 2) + { + /* attemping to decode a bvop without atleast 2 reference frames */ + image_printf(&dec->cur, dec->edged_width, dec->height, 16, 16, + "broken b-frame, mising ref frames"); + }else if (dec->time_pp <= dec->time_bp) { + /* this occurs when dx50_bvop_compatibility==0 sequences are + decoded in vfw. */ + image_printf(&dec->cur, dec->edged_width, dec->height, 16, 16, + "broken b-frame, tpp=%i tbp=%i", dec->time_pp, dec->time_bp); + }else{ + decoder_bframe(dec, &bs, quant, fcode_forward, fcode_backward); + } - // swap MACROBLOCK - // the Divx will not set the low_delay flage some times - // so follow code will wrong to not swap at that time - // this will broken bitstream! so I'm change it, - // But that is not the best way! can anyone tell me how - // to do another way? - // 18-07-2002 MinChen - //if (!dec->low_delay && vop_type == P_VOP) - if (vop_type == P_VOP) - mb_swap(&dec->mbs, &dec->last_mbs); + decoder_output(dec, &dec->cur, dec->mbs, frame, stats, coding_type); + output = 1; + dec->frames++; } + BitstreamByteAlign(&bs); - if (success == 0 && dec->packed_mode) + /* low_delay_default mode: repeat in packed_mode */ + if (dec->low_delay_default && dec->packed_mode && output == 0 && success == 0) { success = 1; - // if (frame->length > BitstreamPos(&bs) / 8) // multiple vops packed together - goto start; + goto repeat; } done : - frame->length = BitstreamPos(&bs) / 8; - - if (stats) + /* low_delay_default mode: if we've gotten here without outputting anything, + then output the recently decoded frame, or print an error message */ + if (dec->low_delay_default && output == 0) { - stats->notify = XVID_DEC_VOP; - stats->data.vop.time_base = (int)dec->time_base; - stats->data.vop.time_increment = 0; //XXX: todo + if (dec->packed_mode && seen_something) + { + /* output the recently decoded frame */ + decoder_output(dec, &dec->refn[0], dec->last_mbs, frame, stats, dec->last_coding_type); + } + else + { + image_clear(&dec->cur, dec->width, dec->height, dec->edged_width, 0, 128, 128); + image_printf(&dec->cur, dec->edged_width, dec->height, 16, 16, + "warning: nothing to output"); + image_printf(&dec->cur, dec->edged_width, dec->height, 16, 64, + "bframe decoder lag"); + + decoder_output(dec, &dec->cur, NULL, frame, stats, P_VOP); + if (stats) stats->type = XVID_TYPE_NOTHING; + + } } emms(); - stop_global_timer(); - return XVID_ERR_OK; + idct = idct_save; + + return BitstreamPos(&bs) / 8; /* number of bytes consumed */ }