--- decoder.c 2003/01/03 16:25:14 1.37.2.25 +++ decoder.c 2003/01/11 21:22:24 1.37.2.29 @@ -55,10 +55,11 @@ * 22.12.2001 lock based interpolation * 01.12.2001 inital version; (c)2001 peter ross * - * $Id: decoder.c,v 1.37.2.25 2003/01/03 16:25:14 suxen_drol Exp $ + * $Id: decoder.c,v 1.37.2.29 2003/01/11 21:22:24 chl Exp $ * *************************************************************************/ +#include #include #include @@ -68,6 +69,7 @@ #include "xvid.h" #include "portab.h" +#include "global.h" #include "decoder.h" #include "bitstream/bitstream.h" @@ -103,6 +105,8 @@ 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); if (dec->mbs) @@ -152,6 +156,16 @@ return XVID_ERR_MEMORY; } + 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->tmp, 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); @@ -210,6 +224,10 @@ image_null(&dec->tmp); image_null(&dec->qtmp); +/* image based GMC */ + image_null(&dec->gmc); + + dec->mbs = NULL; dec->last_mbs = NULL; @@ -236,6 +254,9 @@ { xvid_free(dec->last_mbs); xvid_free(dec->mbs); + + image_destroy(&dec->gmc, dec->edged_width, dec->edged_height); /* image based GMC */ + 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); @@ -384,10 +405,6 @@ - -#define SIGN(X) (((X)>0)?1:-1) -#define ABS(X) (((X)>0)?(X):-(X)) - // decode an inter macroblock void @@ -395,7 +412,7 @@ 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, @@ -424,24 +441,19 @@ mv[i].x = RRV_MV_SCALEUP(pMB->mvs[i].x); mv[i].y = RRV_MV_SCALEUP(pMB->mvs[i].y); } - }else{ + } 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]; @@ -461,11 +473,11 @@ if(dec->quarterpel) { 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); + 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, @@ -497,15 +509,15 @@ { 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); @@ -513,33 +525,33 @@ else { if(dec->quarterpel) { - interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y, dec->qtmp.y, dec->qtmp.y + 64, + 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->qtmp.y, dec->qtmp.y + 64, + 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->qtmp.y, dec->qtmp.y + 64, + 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->qtmp.y, dec->qtmp.y + 64, + 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(); @@ -609,6 +621,118 @@ stop_transfer_timer(); } +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; +} + + +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 */ + + { + pMB->amv = generate_GMCimageMB(&dec->gmc_data, &dec->refn[0], x_pos, y_pos, + stride, stride2, dec->quarterpel, rounding, &dec->cur); + + 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, @@ -716,7 +840,7 @@ 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(DPRINTF_MV,"mv_diff (%i,%i) pred (%i,%i) result (%i,%i)", mv.x, mv.y, pmv.x, pmv.y, mv.x+pmv.x, mv.y+pmv.y); mv.x += pmv.x; mv.y += pmv.y; @@ -739,21 +863,9 @@ -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, @@ -762,7 +874,7 @@ int quant, int fcode, int intra_dc_threshold, - VECTOR * gmc_mv) + const WARPPOINTS *const gmc_warp) { uint32_t x, y; @@ -782,6 +894,31 @@ dec->width, dec->height); stop_edges_timer(); + if (gmc_warp) + { + + // accuracy: 0==1/2, 1=1/4, 2=1/8, 3=1/16 + if ( (dec->sprite_warping_accuracy != 3) || (dec->sprite_warping_points != 2) ) + { + fprintf(stderr,"Wrong GMC parameters acc=%d(-> 1/%d), %d!!!\n", + dec->sprite_warping_accuracy,(2<sprite_warping_accuracy), + dec->sprite_warping_points); + } + + generate_GMCparameters( dec->sprite_warping_points, + (2 << dec->sprite_warping_accuracy), gmc_warp, + dec->width, dec->height, &dec->gmc_data); + +/* image warping is done block-based in decoder_mbgmc(), now */ +/* + generate_GMCimage(&dec->gmc_data, &dec->refn[0], + mb_width, mb_height, + dec->edged_width, dec->edged_width/2, + fcode, dec->quarterpel, 0, + rounding, dec->mbs, &dec->gmc); +*/ + } + bound = 0; for (y = 0; y < mb_height; y++) { @@ -805,7 +942,7 @@ 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 + if (!(BitstreamGetBit(bs))) // block _is_ coded { uint32_t mcbpc; uint32_t cbpc; @@ -830,13 +967,13 @@ 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(DPRINTF_MB, "cbpy %i mcsel %i ", cbpy,mcsel); cbp = (cbpy << 2) | cbpc; @@ -871,15 +1008,15 @@ } } } + + 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], @@ -887,10 +1024,7 @@ } 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 ) { @@ -909,16 +1043,26 @@ 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_GMC; - 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); + + 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 */ { @@ -1468,7 +1612,7 @@ mb->mvs[0] = mb->mvs[1] = mb->mvs[2] = mb->mvs[3] = zeromv; // skip if the co-located P_VOP macroblock is not coded - // note: gmc+not_coded isn't skipped + // 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); @@ -1642,7 +1786,7 @@ uint32_t fcode_forward; uint32_t fcode_backward; uint32_t intra_dc_threshold; - VECTOR gmc_mv[5]; + WARPPOINTS gmc_warp; uint32_t vop_type; int success = 0; int output = 0; @@ -1697,9 +1841,9 @@ repeat: vop_type = BitstreamReadHeaders(&bs, dec, &rounding, &reduced_resolution, - &quant, &fcode_forward, &fcode_backward, &intra_dc_threshold, gmc_mv); + &quant, &fcode_forward, &fcode_backward, &intra_dc_threshold, &gmc_warp); - DPRINTF(DPRINTF_HEADER, "vop_type=%i, packed=%i, time=%i, time_pp=%i, time_bp=%i", + DPRINTF(DPRINTF_HEADER, "vop_type=%i, packed=%i, time=%lli, time_pp=%i, time_bp=%i", vop_type, dec->packed_mode, dec->time, dec->time_pp, dec->time_bp); if (vop_type == - 1) @@ -1758,7 +1902,7 @@ break; case S_VOP : decoder_pframe(dec, &bs, rounding, reduced_resolution, quant, - fcode_forward, intra_dc_threshold, gmc_mv); + fcode_forward, intra_dc_threshold, &gmc_warp); break; case N_VOP : image_copy(&dec->cur, &dec->refn[0], dec->edged_width, dec->height);