--- decoder.c 2002/06/14 13:21:13 1.20 +++ decoder.c 2002/10/30 23:11:48 1.37.2.7 @@ -32,6 +32,14 @@ * * 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 @@ -47,13 +55,17 @@ * 22.12.2001 lock based interpolation * 01.12.2001 inital version; (c)2001 peter ross * - * $Id: decoder.c,v 1.20 2002/06/14 13:21:13 Isibaar Exp $ + * $Id: decoder.c,v 1.37.2.7 2002/10/30 23:11:48 Isibaar Exp $ * *************************************************************************/ #include #include +#ifdef BFRAMES_DEC_DEBUG + #define BFRAMES_DEC +#endif + #include "xvid.h" #include "portab.h" @@ -72,6 +84,7 @@ #include "prediction/mbprediction.h" #include "utils/timer.h" #include "utils/emms.h" +#include "motion/motion.h" #include "image/image.h" #include "image/colorspace.h" @@ -124,6 +137,15 @@ return XVID_ERR_MEMORY; } + if (image_create(&dec->refh, 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); + xvid_free(dec); + return XVID_ERR_MEMORY; + } + dec->mbs = xvid_malloc(sizeof(MACROBLOCK) * dec->mb_width * dec->mb_height, CACHE_LINE); @@ -132,6 +154,7 @@ 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); xvid_free(dec); return XVID_ERR_MEMORY; } @@ -149,6 +172,7 @@ 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); xvid_free(dec); return XVID_ERR_MEMORY; } @@ -174,6 +198,7 @@ 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); xvid_free(dec); @@ -201,7 +226,8 @@ const uint32_t cbp, Bitstream * bs, const uint32_t quant, - const uint32_t intra_dc_threshold) + const uint32_t intra_dc_threshold, + const unsigned int bound) { DECLARE_ALIGNED_MATRIX(block, 6, 64, int16_t, CACHE_LINE); @@ -227,7 +253,7 @@ start_timer(); predict_acdc(dec->mbs, x_pos, y_pos, dec->mb_width, i, &block[i * 64], - iQuant, iDcScaler, predictors); + iQuant, iDcScaler, predictors, bound); if (!acpred_flag) { pMB->acpred_directions[i] = 0; } @@ -246,6 +272,8 @@ block[i * 64 + 0] = dc_dif; start_coeff = 1; + + DPRINTF(DPRINTF_COEFF,"block[0] %i", dc_dif); } else { start_coeff = 0; } @@ -253,8 +281,10 @@ start_timer(); if (cbp & (1 << (5 - i))) // coded { - get_intra_block(bs, &block[i * 64], pMB->acpred_directions[i], - start_coeff); + int direction = dec->alternate_vertical_scan ? + 2 : pMB->acpred_directions[i]; + + get_intra_block(bs, &block[i * 64], direction, start_coeff); } stop_coding_timer(); @@ -296,9 +326,6 @@ #define SIGN(X) (((X)>0)?1:-1) #define ABS(X) (((X)>0)?(X):-(X)) -static const uint32_t roundtab[16] = - { 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 }; - // decode an inter macroblock @@ -333,49 +360,97 @@ uv_dx = pMB->mvs[0].x; uv_dy = pMB->mvs[0].y; - uv_dx = (uv_dx & 3) ? (uv_dx >> 1) | 1 : uv_dx / 2; - uv_dy = (uv_dy & 3) ? (uv_dy >> 1) | 1 : uv_dy / 2; + if (dec->quarterpel) + { + uv_dx /= 2; + uv_dy /= 2; + } + + uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3]; + uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3]; + + start_timer(); + 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, + pMB->mvs[0].x, pMB->mvs[0].y, stride, rounding); + } + else { + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos, + pMB->mvs[0].x, pMB->mvs[0].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos + 8, 16*y_pos, + pMB->mvs[1].x, pMB->mvs[1].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos + 8, + pMB->mvs[2].x, pMB->mvs[2].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos + 8, 16*y_pos + 8, + pMB->mvs[3].x, pMB->mvs[3].y, stride, rounding); + } + + 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, + uv_dx, uv_dy, stride2, rounding); + stop_comp_timer(); + } else { int sum; + + if(dec->quarterpel) + sum = (pMB->mvs[0].x / 2) + (pMB->mvs[1].x / 2) + (pMB->mvs[2].x / 2) + (pMB->mvs[3].x / 2); + else + sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; + + uv_dx = (sum >> 3) + roundtab_76[sum & 0xf]; + + if(dec->quarterpel) + sum = (pMB->mvs[0].y / 2) + (pMB->mvs[1].y / 2) + (pMB->mvs[2].y / 2) + (pMB->mvs[3].y / 2); + else + sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; - sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; - uv_dx = - (sum == - 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + - (ABS(sum) / 16) * 2)); + uv_dy = (sum >> 3) + roundtab_76[sum & 0xf]; - sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; - uv_dy = - (sum == - 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + - (ABS(sum) / 16) * 2)); - } + start_timer(); + 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, + pMB->mvs[0].x, pMB->mvs[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, + pMB->mvs[1].x, pMB->mvs[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, + pMB->mvs[2].x, pMB->mvs[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, + pMB->mvs[3].x, pMB->mvs[3].y, stride, rounding); + } + else { + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos, + pMB->mvs[0].x, pMB->mvs[0].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos + 8, 16*y_pos, + pMB->mvs[1].x, pMB->mvs[1].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos + 8, + pMB->mvs[2].x, pMB->mvs[2].y, stride, rounding); + interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16*x_pos + 8, 16*y_pos + 8, + pMB->mvs[3].x, pMB->mvs[3].y, stride, rounding); + } - start_timer(); - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16 * x_pos, 16 * y_pos, - pMB->mvs[0].x, pMB->mvs[0].y, stride, rounding); - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16 * x_pos + 8, - 16 * y_pos, pMB->mvs[1].x, pMB->mvs[1].y, stride, - rounding); - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16 * x_pos, - 16 * y_pos + 8, pMB->mvs[2].x, pMB->mvs[2].y, stride, - rounding); - interpolate8x8_switch(dec->cur.y, dec->refn[0].y, 16 * x_pos + 8, - 16 * y_pos + 8, pMB->mvs[3].x, pMB->mvs[3].y, stride, - rounding); - 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, - uv_dx, uv_dy, stride2, rounding); - stop_comp_timer(); + 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, + uv_dx, uv_dy, stride2, rounding); + stop_comp_timer(); + } 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]); + get_inter_block(bs, &block[i * 64], direction); stop_coding_timer(); start_timer(); @@ -420,30 +495,39 @@ int quant, int intra_dc_threshold) { - + uint32_t bound; uint32_t x, y; + bound = 0; + for (y = 0; y < dec->mb_height; y++) { for (x = 0; x < dec->mb_width; x++) { - MACROBLOCK *mb = &dec->mbs[y * dec->mb_width + x]; - + MACROBLOCK *mb; uint32_t mcbpc; uint32_t cbpc; uint32_t acpred_flag; uint32_t cbpy; uint32_t cbp; + while (BitstreamShowBits(bs, 9) == 1) + BitstreamSkip(bs, 9); + + if (check_resync_marker(bs, 0)) + { + bound = read_video_packet_header(bs, 0, &quant); + x = bound % dec->mb_width; + y = bound / dec->mb_width; + } + mb = &dec->mbs[y * dec->mb_width + x]; + + DPRINTF(DPRINTF_MB, "macroblock (%i,%i) %08x", x, y, BitstreamShowBits(bs, 32)); + mcbpc = get_mcbpc_intra(bs); mb->mode = mcbpc & 7; cbpc = (mcbpc >> 4); acpred_flag = BitstreamGetBit(bs); - if (mb->mode == MODE_STUFFING) { - DEBUG("-- STUFFING ?"); - continue; - } - cbpy = get_cbpy(bs, 1); cbp = (cbpy << 2) | cbpc; @@ -456,6 +540,10 @@ } } mb->quant = quant; + mb->mvs[0].x = mb->mvs[0].y = + mb->mvs[1].x = mb->mvs[1].y = + mb->mvs[2].x = mb->mvs[2].y = + mb->mvs[3].x = mb->mvs[3].y =0; if (dec->interlacing) { mb->field_dct = BitstreamGetBit(bs); @@ -463,8 +551,11 @@ } decoder_mbintra(dec, mb, x, y, acpred_flag, cbp, bs, quant, - intra_dc_threshold); + intra_dc_threshold, bound); } + if(dec->out_frm) + output_slice(&dec->cur, dec->edged_width,dec->width,dec->out_frm,0,y,dec->mb_width); + } } @@ -477,7 +568,8 @@ int y, int k, VECTOR * mv, - int fcode) + int fcode, + const int bound) { int scale_fac = 1 << (fcode - 1); @@ -485,23 +577,18 @@ int low = ((-32) * scale_fac); int range = (64 * scale_fac); - VECTOR pmv[4]; - int32_t psad[4]; - + VECTOR pmv; int mv_x, mv_y; - int pmv_x, pmv_y; - - - get_pmvdata(dec->mbs, x, y, dec->mb_width, k, pmv, psad); - pmv_x = pmv[0].x; - pmv_y = pmv[0].y; + pmv = get_pmv2(dec->mbs, dec->mb_width, bound, x, y, k); mv_x = get_mv(bs, fcode); mv_y = get_mv(bs, fcode); - mv_x += pmv_x; - mv_y += pmv_y; + DPRINTF(DPRINTF_MV,"mv_diff (%i,%i) pred (%i,%i)", mv_x, mv_y, pmv.x, pmv.y); + + mv_x += pmv.x; + mv_y += pmv.y; if (mv_x < low) { mv_x += range; @@ -531,15 +618,34 @@ { uint32_t x, y; + uint32_t bound; + int cp_mb, st_mb; start_timer(); image_setedges(&dec->refn[0], dec->edged_width, dec->edged_height, - dec->width, dec->height, dec->interlacing); + dec->width, dec->height); stop_edges_timer(); + bound = 0; + for (y = 0; y < dec->mb_height; y++) { + cp_mb = st_mb = 0; for (x = 0; x < dec->mb_width; x++) { - MACROBLOCK *mb = &dec->mbs[y * dec->mb_width + x]; + MACROBLOCK *mb; + + // skip stuffing + while (BitstreamShowBits(bs, 10) == 1) + BitstreamSkip(bs, 10); + + if (check_resync_marker(bs, fcode - 1)) + { + bound = read_video_packet_header(bs, fcode - 1, &quant); + x = bound % dec->mb_width; + y = bound / dec->mb_width; + } + mb = &dec->mbs[y * dec->mb_width + x]; + + DPRINTF(DPRINTF_MB, "macroblock (%i,%i) %08x", x, y, BitstreamShowBits(bs, 32)); //if (!(dec->mb_skip[y*dec->mb_width + x]=BitstreamGetBit(bs))) // not_coded if (!(BitstreamGetBit(bs))) // not_coded @@ -551,9 +657,13 @@ uint32_t cbp; uint32_t intra; + 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); acpred_flag = 0; intra = (mb->mode == MODE_INTRA || mb->mode == MODE_INTRA_Q); @@ -562,27 +672,29 @@ acpred_flag = BitstreamGetBit(bs); } - if (mb->mode == MODE_STUFFING) { - DEBUG("-- STUFFING ?"); - continue; - } - cbpy = get_cbpy(bs, intra); + DPRINTF(DPRINTF_MB, "cbpy %i", cbpy); + cbp = (cbpy << 2) | cbpc; if (mb->mode == MODE_INTER_Q || mb->mode == MODE_INTRA_Q) { - quant += dquant_table[BitstreamGetBits(bs, 2)]; + int dquant = dquant_table[BitstreamGetBits(bs, 2)]; + DPRINTF(DPRINTF_MB, "dquant %i", dquant); + quant += dquant; if (quant > 31) { quant = 31; - } else if (mb->quant < 1) { + } else if (quant < 1) { quant = 1; } + DPRINTF(DPRINTF_MB, "quant %i", quant); } mb->quant = quant; if (dec->interlacing) { - mb->field_dct = BitstreamGetBit(bs); - DEBUG1("decp: field_dct: ", mb->field_dct); + if (cbp || intra) { + mb->field_dct = BitstreamGetBit(bs); + DEBUG1("decp: field_dct: ", mb->field_dct); + } if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { mb->field_pred = BitstreamGetBit(bs); @@ -600,23 +712,23 @@ if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { if (dec->interlacing && mb->field_pred) { get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], - fcode); + fcode, bound); get_motion_vector(dec, bs, x, y, 0, &mb->mvs[1], - fcode); + fcode, bound); } else { get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], - fcode); + 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; } - } else if (mb->mode == - MODE_INTER4V /* || mb->mode == MODE_INTER4V_Q */ ) { - get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode); - get_motion_vector(dec, bs, x, y, 1, &mb->mvs[1], fcode); - get_motion_vector(dec, bs, x, y, 2, &mb->mvs[2], fcode); - get_motion_vector(dec, bs, x, y, 3, &mb->mvs[3], fcode); + } else if (mb->mode == MODE_INTER4V ) { + + get_motion_vector(dec, bs, x, y, 0, &mb->mvs[0], fcode, bound); + 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 { mb->mvs[0].x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = @@ -624,7 +736,7 @@ mb->mvs[0].y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = 0; decoder_mbintra(dec, mb, x, y, acpred_flag, cbp, bs, quant, - intra_dc_threshold); + intra_dc_threshold, bound); continue; } @@ -632,7 +744,6 @@ rounding); } else // not coded { - //DEBUG2("P-frame MB at (X,Y)=",x,y); mb->mode = MODE_NOT_CODED; 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; @@ -674,10 +785,16 @@ dec->refn[0].v + (8 * y) * dec->edged_width / 2 + (8 * x), dec->edged_width / 2); - 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; } } + if(dec->out_frm && cp_mb > 0) + output_slice(&dec->cur, dec->edged_width,dec->width,dec->out_frm,st_mb,y,cp_mb); } } @@ -755,6 +872,7 @@ pU_Cur = dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3); pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3); + if (!(pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q)) { uv_dx = pMB->mvs[0].x; uv_dy = pMB->mvs[0].y; @@ -795,12 +913,14 @@ stop_comp_timer(); 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]); + get_inter_block(bs, &block[i * 64], direction); stop_coding_timer(); start_timer(); @@ -838,7 +958,6 @@ stop_transfer_timer(); } - // add by MinChen // decode an B-frame direct & inter macroblock void @@ -848,7 +967,6 @@ const MACROBLOCK * pMB, const uint32_t x_pos, const uint32_t y_pos, - const uint32_t cbp, Bitstream * bs) { @@ -863,11 +981,13 @@ int b_uv_dx, b_uv_dy; uint32_t i; uint8_t *pY_Cur, *pU_Cur, *pV_Cur; + const uint32_t cbp = pMB->cbp; 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); + if ((pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q)) { uv_dx = pMB->mvs[0].x; uv_dy = pMB->mvs[0].y; @@ -945,28 +1065,47 @@ interpolate8x8_switch(dec->refn[2].v, backward.v, 8 * x_pos, 8 * y_pos, b_uv_dx, b_uv_dy, stride2, 0); - interpolate8x8_c(dec->cur.y, dec->refn[2].y, 16 * x_pos, 16 * y_pos, - stride); - interpolate8x8_c(dec->cur.y, dec->refn[2].y, 16 * x_pos + 8, 16 * y_pos, - stride); - interpolate8x8_c(dec->cur.y, dec->refn[2].y, 16 * x_pos, 16 * y_pos + 8, - stride); - interpolate8x8_c(dec->cur.y, dec->refn[2].y, 16 * x_pos + 8, - 16 * y_pos + 8, stride); - interpolate8x8_c(dec->cur.u, dec->refn[2].u, 8 * x_pos, 8 * y_pos, - stride2); - interpolate8x8_c(dec->cur.v, dec->refn[2].v, 8 * x_pos, 8 * y_pos, - stride2); + 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, + stride, 0); + + 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, + stride, 0); + + 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, + stride, 0); + + 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, + stride, 0); + + 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, + stride2, 0); + + 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, + stride2, 0); stop_comp_timer(); 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]); + get_inter_block(bs, &block[i * 64], direction); stop_coding_timer(); start_timer(); @@ -1048,40 +1187,57 @@ int fcode_forward, int fcode_backward) { - uint32_t x, y; - VECTOR mv, zeromv; + VECTOR mv; + const VECTOR zeromv = {0,0}; +#ifdef BFRAMES_DEC_DEBUG + FILE *fp; + static char first=0; +#define BFRAME_DEBUG if (!first && fp){ \ + fprintf(fp,"Y=%3d X=%3d MB=%2d CBP=%02X\n",y,x,mb->mb_type,mb->cbp); \ + } +#endif start_timer(); image_setedges(&dec->refn[0], dec->edged_width, dec->edged_height, - dec->width, dec->height, dec->interlacing); - //image_setedges(&dec->refn[1], dec->edged_width, dec->edged_height, dec->width, dec->height, dec->interlacing); + dec->width, dec->height); + image_setedges(&dec->refn[1], dec->edged_width, dec->edged_height, + dec->width, dec->height); stop_edges_timer(); +#ifdef BFRAMES_DEC_DEBUG + if (!first){ + fp=fopen("C:\\XVIDDBG.TXT","w"); + } +#endif for (y = 0; y < dec->mb_height; y++) { // Initialize Pred Motion Vector - dec->p_fmv.x = dec->p_fmv.y = dec->p_bmv.x = dec->p_bmv.y = 0; + dec->p_fmv = dec->p_bmv = zeromv; for (x = 0; x < dec->mb_width; x++) { MACROBLOCK *mb = &dec->mbs[y * dec->mb_width + x]; MACROBLOCK *last_mb = &dec->last_mbs[y * dec->mb_width + x]; - mb->mvs[0].x = mb->mvs[0].y = zeromv.x = zeromv.y = mv.x = mv.y = - 0; + mv = + 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 ? if (last_mb->mode == MODE_NOT_CODED) { //DEBUG2("Skip MB in B-frame at (X,Y)=!",x,y); - mb->mb_type = MODE_FORWARD; mb->cbp = 0; - 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->quant = 8; +#ifdef BFRAMES_DEC_DEBUG + mb->mb_type = MODE_NOT_CODED; + BFRAME_DEBUG +#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; - decoder_bf_mbinter(dec, mb, x, y, mb->cbp, bs, quant, 1); + decoder_bf_mbinter(dec, mb, x, y, mb->cbp, bs, mb->quant, 1); continue; } - //t=BitstreamShowBits(bs,32); if (!BitstreamGetBit(bs)) { // modb=='0' const uint8_t modb2 = BitstreamGetBit(bs); @@ -1098,84 +1254,73 @@ if (quant > 31) { quant = 31; - } else if (mb->quant < 1) { + } else if (quant < 1) { quant = 1; } - } else { - quant = 8; } - mb->quant = quant; } else { mb->mb_type = MODE_DIRECT_NONE_MV; mb->cbp = 0; } - mb->mode = MODE_INTER; + mb->quant = quant; + mb->mode = MODE_INTER4V; //DEBUG1("Switch bm_type=",mb->mb_type); +#ifdef BFRAMES_DEC_DEBUG + BFRAME_DEBUG +#endif + switch (mb->mb_type) { case MODE_DIRECT: - get_b_motion_vector(dec, bs, x, y, &mb->mvs[0], 1, zeromv); + get_b_motion_vector(dec, bs, x, y, &mv, 1, zeromv); case MODE_DIRECT_NONE_MV: - { // Because this file is a C file not C++ so I use '{' to define var - const int64_t TRB = dec->time_pp - dec->time_bp, TRD = - dec->time_pp; + { + const int64_t TRB = dec->time_pp - dec->time_bp, TRD = dec->time_pp; int i; for (i = 0; i < 4; i++) { - mb->mvs[i].x = - (int32_t) ((TRB * last_mb->mvs[i].x) / TRD + - mb->mvs[0].x); - mb->b_mvs[i].x = - (int32_t) ((mb->mvs[0].x == - 0) ? ((TRB - - TRD) * last_mb->mvs[i].x) / - TRD : mb->mvs[i].x - last_mb->mvs[i].x); - mb->mvs[i].y = - (int32_t) ((TRB * last_mb->mvs[i].y) / TRD + - mb->mvs[0].y); - mb->b_mvs[i].y = - (int32_t) ((mb->mvs[0].y == - 0) ? ((TRB - - TRD) * last_mb->mvs[i].y) / - TRD : mb->mvs[i].y - last_mb->mvs[i].y); + mb->mvs[i].x = (int32_t) ((TRB * last_mb->mvs[i].x) + / TRD + mv.x); + mb->b_mvs[i].x = (int32_t) ((mv.x == 0) + ? ((TRB - TRD) * last_mb->mvs[i].x) + / TRD + : mb->mvs[i].x - last_mb->mvs[i].x); + mb->mvs[i].y = (int32_t) ((TRB * last_mb->mvs[i].y) + / TRD + mv.y); + mb->b_mvs[i].y = (int32_t) ((mv.y == 0) + ? ((TRB - TRD) * last_mb->mvs[i].y) + / TRD + : mb->mvs[i].y - last_mb->mvs[i].y); } //DEBUG("B-frame Direct!\n"); } - mb->mode = MODE_INTER4V; decoder_bf_interpolate_mbinter(dec, dec->refn[1], dec->refn[0], - mb, x, y, mb->cbp, bs); + mb, x, y, bs); break; case MODE_INTERPOLATE: get_b_motion_vector(dec, bs, x, y, &mb->mvs[0], fcode_forward, dec->p_fmv); - dec->p_fmv.x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = - mb->mvs[0].x; - dec->p_fmv.y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = - mb->mvs[0].y; + dec->p_fmv = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; get_b_motion_vector(dec, bs, x, y, &mb->b_mvs[0], fcode_backward, dec->p_bmv); - dec->p_bmv.x = mb->b_mvs[1].x = mb->b_mvs[2].x = - mb->b_mvs[3].x = mb->b_mvs[0].x; - dec->p_bmv.y = mb->b_mvs[1].y = mb->b_mvs[2].y = - mb->b_mvs[3].y = mb->b_mvs[0].y; + dec->p_bmv = mb->b_mvs[1] = mb->b_mvs[2] = + mb->b_mvs[3] = mb->b_mvs[0]; decoder_bf_interpolate_mbinter(dec, dec->refn[1], dec->refn[0], - mb, x, y, mb->cbp, bs); + mb, x, y, bs); //DEBUG("B-frame Bidir!\n"); break; case MODE_BACKWARD: get_b_motion_vector(dec, bs, x, y, &mb->mvs[0], fcode_backward, dec->p_bmv); - dec->p_bmv.x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = - mb->mvs[0].x; - dec->p_bmv.y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = - mb->mvs[0].y; + dec->p_bmv = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; + mb->mode = MODE_INTER; decoder_bf_mbinter(dec, mb, x, y, mb->cbp, bs, quant, 0); //DEBUG("B-frame Backward!\n"); break; @@ -1183,11 +1328,9 @@ case MODE_FORWARD: get_b_motion_vector(dec, bs, x, y, &mb->mvs[0], fcode_forward, dec->p_fmv); - dec->p_fmv.x = mb->mvs[1].x = mb->mvs[2].x = mb->mvs[3].x = - mb->mvs[0].x; - dec->p_fmv.y = mb->mvs[1].y = mb->mvs[2].y = mb->mvs[3].y = - mb->mvs[0].y; + dec->p_fmv = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = mb->mvs[0]; + mb->mode = MODE_INTER; decoder_bf_mbinter(dec, mb, x, y, mb->cbp, bs, quant, 1); //DEBUG("B-frame Forward!\n"); break; @@ -1198,6 +1341,13 @@ } // end of FOR } +#ifdef BFRAMES_DEC_DEBUG + if (!first){ + first=1; + if (fp) + fclose(fp); + } +#endif } // swap two MACROBLOCK array @@ -1226,8 +1376,13 @@ start_global_timer(); + dec->out_frm = (frame->colorspace == XVID_CSP_EXTERN) ? frame->image : NULL; + BitstreamInit(&bs, frame->bitstream, frame->length); + if(BitstreamShowBits(&bs, 8) == 0x7f) + return XVID_ERR_OK; + // add by chenm001 // for support B-frame to reference last 2 frame dec->frames++; @@ -1241,43 +1396,61 @@ case P_VOP: decoder_pframe(dec, &bs, rounding, quant, fcode_forward, intra_dc_threshold); +#ifdef BFRAMES_DEC DEBUG1("P_VOP Time=", dec->time); +#endif break; case I_VOP: decoder_iframe(dec, &bs, quant, intra_dc_threshold); +#ifdef BFRAMES_DEC DEBUG1("I_VOP Time=", dec->time); +#endif break; case B_VOP: -#ifdef BFRAMES +#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!"); } +#else + image_copy(&dec->cur, &dec->refn[0], dec->edged_width, dec->height); #endif 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: return XVID_ERR_FAIL; } +#ifdef BFRAMES_DEC_DEBUG + if (frame->length != BitstreamPos(&bs) / 8){ + DEBUG2("InLen/UseLen",frame->length, BitstreamPos(&bs) / 8); + } +#endif frame->length = BitstreamPos(&bs) / 8; -#ifdef BFRAMES + +#ifdef BFRAMES_DEC // test if no B_VOP - if (dec->low_delay) { + if (dec->low_delay || dec->frames == 0) { #endif - image_output(&dec->cur, dec->width, dec->height, dec->edged_width, + image_output(&dec->cur, dec->width, dec->height, dec->edged_width, frame->image, frame->stride, frame->colorspace); -#ifdef BFRAMES + +#ifdef BFRAMES_DEC } else { - if (dec->frames >= 1) { + if (dec->frames >= 1) { start_timer(); if ((vop_type == I_VOP || vop_type == P_VOP)) { image_output(&dec->refn[0], dec->width, dec->height, @@ -1296,8 +1469,16 @@ if (vop_type == I_VOP || vop_type == P_VOP) { image_swap(&dec->refn[0], &dec->refn[1]); image_swap(&dec->cur, &dec->refn[0]); + // swap MACROBLOCK - if (dec->low_delay && vop_type == P_VOP) + // 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); }