--- decoder.c 2002/07/24 00:49:04 1.34 +++ decoder.c 2003/06/09 19:41:53 1.49.2.8 @@ -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 1.49.2.8 2003/06/09 19:41:53 edgomez Exp $ * - * $Id: decoder.c,v 1.34 2002/07/24 00:49:04 Isibaar Exp $ - * - *************************************************************************/ + ****************************************************************************/ +#include #include #include @@ -68,6 +34,7 @@ #include "xvid.h" #include "portab.h" +#include "global.h" #include "decoder.h" #include "bitstream/bitstream.h" @@ -79,36 +46,42 @@ #include "dct/fdct.h" #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 "image/image.h" #include "image/colorspace.h" #include "utils/mem_align.h" int -decoder_create(XVID_DEC_PARAM * param) +decoder_resize(DECODER * dec) { - 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->tmp, dec->edged_width, dec->edged_height); + image_destroy(&dec->qtmp, dec->edged_width, dec->edged_height); - dec = xvid_malloc(sizeof(DECODER), CACHE_LINE); - if (dec == NULL) { - return XVID_ERR_MEMORY; - } - param->handle = dec; + image_destroy(&dec->gmc, dec->edged_width, dec->edged_height); - dec->width = param->width; - dec->height = param->height; + if (dec->last_mbs) + xvid_free(dec->last_mbs); + if (dec->mbs) + xvid_free(dec->mbs); + /* realloc */ dec->mb_width = (dec->width + 15) / 16; dec->mb_height = (dec->height + 15) / 16; dec->edged_width = 16 * dec->mb_width + 2 * EDGE_SIZE; dec->edged_height = 16 * dec->mb_height + 2 * EDGE_SIZE; - dec->low_delay = 0; if (image_create(&dec->cur, dec->edged_width, dec->edged_height)) { xvid_free(dec); @@ -120,15 +93,15 @@ xvid_free(dec); 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); @@ -136,6 +109,25 @@ 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->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); @@ -143,15 +135,14 @@ 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->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); @@ -160,21 +151,64 @@ 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); + 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 0; +} + + +int +decoder_create(xvid_dec_create_t * create) +{ + DECODER *dec; + + if (XVID_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)); + + create->handle = dec; + + 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->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; - return XVID_ERR_OK; + dec->fixed_dimensions = (dec->width > 0 && dec->height > 0); + + if (dec->fixed_dimensions) + return decoder_resize(dec); + else + return 0; } @@ -183,14 +217,19 @@ { 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->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; } @@ -202,8 +241,7 @@ -// decode an intra macroblock - +/* decode an intra macroblock */ void decoder_mbintra(DECODER * dec, MACROBLOCK * pMB, @@ -214,7 +252,8 @@ Bitstream * bs, const uint32_t quant, const uint32_t intra_dc_threshold, - const unsigned int bound) + const unsigned int bound, + const int reduced_resolution) { DECLARE_ALIGNED_MATRIX(block, 6, 64, int16_t, CACHE_LINE); @@ -227,11 +266,17 @@ uint32_t iQuant = pMB->quant; uint8_t *pY_Cur, *pU_Cur, *pV_Cur; - 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 (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); + }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); + } - 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); @@ -254,22 +299,24 @@ 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 */ { - 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(); @@ -288,6 +335,7 @@ start_timer(); idct(&data[i * 64]); stop_idct_timer(); + } if (dec->interlacing && pMB->field_dct) { @@ -296,37 +344,42 @@ } start_timer(); - transfer_16to8copy(pY_Cur, &data[0 * 64], stride); - transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); - transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); - transfer_16to8copy(pY_Cur + 8 + next_block, &data[3 * 64], stride); - transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); - transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); + + if (reduced_resolution) + { + next_block*=2; + copy_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); + copy_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); + copy_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); + copy_upsampled_8x8_16to8(pY_Cur + 16 + next_block, &data[3 * 64], stride); + copy_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); + copy_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); + }else{ + transfer_16to8copy(pY_Cur, &data[0 * 64], stride); + transfer_16to8copy(pY_Cur + 8, &data[1 * 64], stride); + transfer_16to8copy(pY_Cur + next_block, &data[2 * 64], stride); + transfer_16to8copy(pY_Cur + 8 + next_block, &data[3 * 64], stride); + transfer_16to8copy(pU_Cur, &data[4 * 64], stride2); + transfer_16to8copy(pV_Cur, &data[5 * 64], stride2); + } stop_transfer_timer(); } - -#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 - +/* 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, - const uint32_t rounding) + const uint32_t rounding, + const int reduced_resolution) { DECLARE_ALIGNED_MATRIX(block, 6, 64, int16_t, CACHE_LINE); @@ -334,85 +387,276 @@ uint32_t stride = dec->edged_width; uint32_t stride2 = stride / 2; - uint32_t next_block = stride * 8; + uint32_t next_block = stride * (reduced_resolution ? 16 : 8); uint32_t i; uint32_t iQuant = pMB->quant; uint8_t *pY_Cur, *pU_Cur, *pV_Cur; - int uv_dx, uv_dy; - 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); + int uv_dx, uv_dy; + 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); + 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 = pMB->mvs[0].x; - uv_dy = pMB->mvs[0].y; - if (dec->quarterpel) + 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]; + + start_timer(); + if (reduced_resolution) { - uv_dx = (uv_dx >> 1) | (uv_dx & 1); - uv_dy = (uv_dy >> 1) | (uv_dy & 1); + interpolate32x32_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.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, + uv_dx, uv_dy, stride2, rounding); + } + else + { + 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); + } + else { + interpolate16x16_switch(dec->cur.y, dec->refn[0].y, 16*x_pos, 16*y_pos, + mv[0].x, mv[0].y, stride, rounding); + } - uv_dx = (uv_dx & 3) ? (uv_dx >> 1) | 1 : uv_dx / 2; - uv_dy = (uv_dy & 3) ? (uv_dy >> 1) | 1 : uv_dy / 2; - } else { + 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 { /* MODE_INTER4V */ int sum; - sum = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x; + + if(dec->quarterpel) + sum = (mv[0].x / 2) + (mv[1].x / 2) + (mv[2].x / 2) + (mv[3].x / 2); + else + sum = mv[0].x + mv[1].x + mv[2].x + mv[3].x; + + uv_dx = (sum >> 3) + roundtab_76[sum & 0xf]; + + if(dec->quarterpel) + sum = (mv[0].y / 2) + (mv[1].y / 2) + (mv[2].y / 2) + (mv[3].y / 2); + else + sum = mv[0].y + mv[1].y + mv[2].y + mv[3].y; - if (dec->quarterpel) + uv_dy = (sum >> 3) + roundtab_76[sum & 0xf]; + + start_timer(); + if (reduced_resolution) { - sum /= 2; - } + 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, + mv[1].x, mv[1].y, stride, rounding); + 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, + mv[3].x, mv[3].y, stride, rounding); + 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, + uv_dx, uv_dy, stride2, rounding); + + /* set_block(pY_Cur, stride, 32, 32, 127); */ + } + else + { + if(dec->quarterpel) { + 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, + 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, + 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, + 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, + mv[0].x, mv[0].y, stride, rounding); + 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, + 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, + mv[3].x, mv[3].y, stride, rounding); + } - uv_dx = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); + 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(); + } - sum = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y; + for (i = 0; i < 6; i++) { + int direction = dec->alternate_vertical_scan ? 2 : 0; - if (dec->quarterpel) + if (cbp & (1 << (5 - i))) /* coded */ { - sum /= 2; + 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(); } + } - uv_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2)); + if (dec->interlacing && pMB->field_dct) { + next_block = stride; + stride *= 2; } start_timer(); - if(dec->quarterpel) { - DEBUG("QUARTERPEL"); - interpolate8x8_quarterpel(dec->cur.y, dec->refn[0].y, 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, 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, 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, 16*x_pos + 8, 16*y_pos + 8, - pMB->mvs[3].x, pMB->mvs[3].y, stride, rounding); + if (reduced_resolution) + { + if (cbp & 32) + add_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride); + if (cbp & 16) + add_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride); + if (cbp & 8) + add_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride); + if (cbp & 4) + add_upsampled_8x8_16to8(pY_Cur + 16 + next_block, &data[3 * 64], stride); + if (cbp & 2) + add_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2); + if (cbp & 1) + add_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2); } - 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); + else + { + 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(); +} - 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(); +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++) { - if (cbp & (1 << (5 - i))) // coded + int direction = dec->alternate_vertical_scan ? 2 : 0; + + 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]); + get_inter_block(bs, &block[i * 64], direction); stop_coding_timer(); start_timer(); @@ -429,11 +673,13 @@ } } - if (dec->interlacing && pMB->field_dct) { - next_block = stride; - stride *= 2; - } - +/* 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); @@ -454,16 +700,25 @@ void decoder_iframe(DECODER * dec, Bitstream * bs, + int reduced_resolution, int quant, int intra_dc_threshold) { uint32_t bound; uint32_t x, y; + uint32_t mb_width = dec->mb_width; + uint32_t mb_height = dec->mb_height; + + if (reduced_resolution) + { + mb_width = (dec->width + 31) / 32; + mb_height = (dec->height + 31) / 32; + } bound = 0; - for (y = 0; y < dec->mb_height; y++) { - for (x = 0; x < dec->mb_width; x++) { + for (y = 0; y < mb_height; y++) { + for (x = 0; x < mb_width; x++) { MACROBLOCK *mb; uint32_t mcbpc; uint32_t cbpc; @@ -476,13 +731,14 @@ if (check_resync_marker(bs, 0)) { - bound = read_video_packet_header(bs, 0, &quant); - x = bound % dec->mb_width; - y = bound / dec->mb_width; + bound = read_video_packet_header(bs, dec, 0, + &quant, NULL, NULL, &intra_dc_threshold); + x = bound % mb_width; + y = bound / mb_width; } 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; @@ -509,15 +765,15 @@ 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, - intra_dc_threshold, bound); + intra_dc_threshold, bound, reduced_resolution); + } if(dec->out_frm) - output_slice(&dec->cur, dec->edged_width,dec->width,dec->out_frm,0,y,dec->mb_width); - + output_slice(&dec->cur, dec->edged_width,dec->width,dec->out_frm,0,y,mb_width); } } @@ -529,7 +785,7 @@ int x, int y, int k, - VECTOR * mv, + VECTOR * ret_mv, int fcode, const int bound) { @@ -540,77 +796,116 @@ int range = (64 * scale_fac); VECTOR pmv; - int mv_x, mv_y; + VECTOR mv; 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 = 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; + mv.x += pmv.x; + mv.y += pmv.y; - if (mv_x < low) { - mv_x += range; - } else if (mv_x > high) { - mv_x -= range; + if (mv.x < low) { + mv.x += range; + } else if (mv.x > high) { + mv.x -= range; } - if (mv_y < low) { - mv_y += range; - } else if (mv_y > high) { - mv_y -= range; + if (mv.y < low) { + mv.y += range; + } else if (mv.y > high) { + mv.y -= range; } - mv->x = mv_x; - mv->y = mv_y; - + ret_mv->x = mv.x; + ret_mv->y = mv.y; } + + + +/* for P_VOP set gmc_warp to NULL */ void decoder_pframe(DECODER * dec, Bitstream * bs, int rounding, + int reduced_resolution, int quant, int fcode, - int intra_dc_threshold) + int intra_dc_threshold, + const WARPPOINTS *const gmc_warp) { uint32_t x, y; uint32_t bound; int cp_mb, st_mb; + uint32_t mb_width = dec->mb_width; + uint32_t mb_height = dec->mb_height; + + if (reduced_resolution) + { + mb_width = (dec->width + 31) / 32; + mb_height = (dec->height + 31) / 32; + } 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(); + 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 < dec->mb_height; y++) { + for (y = 0; y < mb_height; y++) { cp_mb = st_mb = 0; - for (x = 0; x < dec->mb_width; x++) { + for (x = 0; x < mb_width; x++) { MACROBLOCK *mb; - // skip stuffing + /* 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; + bound = read_video_packet_header(bs, dec, fcode - 1, + &quant, &fcode, NULL, &intra_dc_threshold); + x = bound % mb_width; + y = bound / mb_width; } 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; @@ -618,14 +913,15 @@ uint32_t cbpy; uint32_t cbp; uint32_t intra; + 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); @@ -634,42 +930,55 @@ acpred_flag = BitstreamGetBit(bs); } + 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) { - mb->field_dct = BitstreamGetBit(bs); - DEBUG1("decp: field_dct: ", mb->field_dct); + if (cbp || intra) { + mb->field_dct = BitstreamGetBit(bs); + 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; + + } else if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) { - 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, bound); @@ -678,75 +987,87 @@ } 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 /* || mb->mode == MODE_INTER4V_Q */ ) { + } 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 + } else /* MODE_INTRA, MODE_INTRA_Q */ { 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; decoder_mbintra(dec, mb, x, y, acpred_flag, cbp, bs, quant, - intra_dc_threshold, bound); + intra_dc_threshold, bound, reduced_resolution); continue; } - decoder_mbinter(dec, mb, x, y, acpred_flag, cbp, bs, quant, - rounding); - } else // not coded + decoder_mbinter(dec, mb, x, y, fcode, cbp, bs, quant, + rounding, reduced_resolution); + + } + else if (gmc_warp) /* a not coded S(GMC)-VOP macroblock */ + { + 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 */ { - //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; - - // copy macroblock directly from ref to cur + /* copy macroblock directly from ref to cur */ start_timer(); - transfer8x8_copy(dec->cur.y + (16 * y) * dec->edged_width + - (16 * x), - dec->refn[0].y + (16 * y) * dec->edged_width + - (16 * x), dec->edged_width); - - transfer8x8_copy(dec->cur.y + (16 * y) * dec->edged_width + - (16 * x + 8), - dec->refn[0].y + (16 * y) * dec->edged_width + - (16 * x + 8), dec->edged_width); - - transfer8x8_copy(dec->cur.y + (16 * y + 8) * dec->edged_width + - (16 * x), - dec->refn[0].y + (16 * y + - 8) * dec->edged_width + - (16 * x), dec->edged_width); - - transfer8x8_copy(dec->cur.y + (16 * y + 8) * dec->edged_width + - (16 * x + 8), - dec->refn[0].y + (16 * y + - 8) * dec->edged_width + - (16 * x + 8), dec->edged_width); - - transfer8x8_copy(dec->cur.u + (8 * y) * dec->edged_width / 2 + - (8 * x), - dec->refn[0].u + - (8 * y) * dec->edged_width / 2 + (8 * x), - dec->edged_width / 2); - - transfer8x8_copy(dec->cur.v + (8 * y) * dec->edged_width / 2 + - (8 * x), - dec->refn[0].v + - (8 * y) * dec->edged_width / 2 + (8 * x), - dec->edged_width / 2); + if (reduced_resolution) + { + transfer32x32_copy(dec->cur.y + (32*y)*dec->edged_width + (32*x), + dec->refn[0].y + (32*y)*dec->edged_width + (32*x), + dec->edged_width); + + transfer16x16_copy(dec->cur.u + (16*y)*dec->edged_width/2 + (16*x), + dec->refn[0].u + (16*y)*dec->edged_width/2 + (16*x), + dec->edged_width/2); + + transfer16x16_copy(dec->cur.v + (16*y)*dec->edged_width/2 + (16*x), + dec->refn[0].v + (16*y)*dec->edged_width/2 + (16*x), + dec->edged_width/2); + } + else + { + transfer16x16_copy(dec->cur.y + (16*y)*dec->edged_width + (16*x), + dec->refn[0].y + (16*y)*dec->edged_width + (16*x), + dec->edged_width); + + transfer8x8_copy(dec->cur.u + (8*y)*dec->edged_width/2 + (8*x), + dec->refn[0].u + (8*y)*dec->edged_width/2 + (8*x), + dec->edged_width/2); + + transfer8x8_copy(dec->cur.v + (8*y)*dec->edged_width/2 + (8*x), + 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; @@ -760,8 +1081,7 @@ } -// add by MinChen -// decode B-frame motion vector +/* decode B-frame motion vector */ void get_b_motion_vector(DECODER * dec, Bitstream * bs, @@ -805,8 +1125,7 @@ } -// 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, @@ -838,35 +1157,49 @@ 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]; } else { int sum; - 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)); - - 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)); + 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; + + uv_dy = (sum >> 3) + roundtab_76[sum & 0xf]; } start_timer(); - interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16 * x_pos, 16 * y_pos, - pMB->mvs[0].x, pMB->mvs[0].y, stride, 0); - interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16 * x_pos + 8, - 16 * y_pos, pMB->mvs[1].x, pMB->mvs[1].y, stride, 0); - interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16 * x_pos, - 16 * y_pos + 8, pMB->mvs[2].x, pMB->mvs[2].y, stride, - 0); - interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16 * x_pos + 8, - 16 * y_pos + 8, pMB->mvs[3].x, pMB->mvs[3].y, stride, - 0); + if(dec->quarterpel) { + 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 { + interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16*x_pos, 16*y_pos, + pMB->mvs[0].x, pMB->mvs[0].y, stride, 0); + interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16*x_pos + 8, 16*y_pos, + pMB->mvs[1].x, pMB->mvs[1].y, stride, 0); + interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16*x_pos, 16*y_pos + 8, + pMB->mvs[2].x, pMB->mvs[2].y, stride, 0); + interpolate8x8_switch(dec->cur.y, dec->refn[ref].y, 16*x_pos + 8, 16*y_pos + 8, + pMB->mvs[3].x, pMB->mvs[3].y, stride, 0); + } + interpolate8x8_switch(dec->cur.u, dec->refn[ref].u, 8 * x_pos, 8 * y_pos, uv_dx, uv_dy, stride2, 0); interpolate8x8_switch(dec->cur.v, dec->refn[ref].v, 8 * x_pos, 8 * y_pos, @@ -874,12 +1207,14 @@ stop_comp_timer(); for (i = 0; i < 6; i++) { - if (cbp & (1 << (5 - i))) // coded + int direction = dec->alternate_vertical_scan ? 2 : 0; + + 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]); + get_inter_block(bs, &block[i * 64], direction); stop_coding_timer(); start_timer(); @@ -917,9 +1252,7 @@ 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, @@ -952,100 +1285,176 @@ 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; - b_uv_dx = pMB->b_mvs[0].x; b_uv_dy = pMB->b_mvs[0].y; - b_uv_dx = (uv_dx & 3) ? (uv_dx >> 1) | 1 : uv_dx / 2; - b_uv_dy = (uv_dy & 3) ? (uv_dy >> 1) | 1 : uv_dy / 2; + if (dec->quarterpel) + { + uv_dx /= 2; + uv_dy /= 2; + + b_uv_dx /= 2; + b_uv_dy /= 2; + } + + uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3]; + uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3]; + + b_uv_dx = (b_uv_dx >> 1) + roundtab_79[b_uv_dx & 0x3]; + b_uv_dy = (b_uv_dy >> 1) + roundtab_79[b_uv_dy & 0x3]; } else { int sum; - 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)); - - 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)); - - sum = - pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + - pMB->b_mvs[3].x; - b_uv_dx = - (sum == - 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + - (ABS(sum) / 16) * 2)); - - sum = - pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + - pMB->b_mvs[3].y; - b_uv_dy = - (sum == - 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + - (ABS(sum) / 16) * 2)); + 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; + + uv_dy = (sum >> 3) + roundtab_76[sum & 0xf]; + + + if(dec->quarterpel) + sum = (pMB->b_mvs[0].x / 2) + (pMB->b_mvs[1].x / 2) + (pMB->b_mvs[2].x / 2) + (pMB->b_mvs[3].x / 2); + else + sum = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x; + + b_uv_dx = (sum >> 3) + roundtab_76[sum & 0xf]; + + if(dec->quarterpel) + sum = (pMB->b_mvs[0].y / 2) + (pMB->b_mvs[1].y / 2) + (pMB->b_mvs[2].y / 2) + (pMB->b_mvs[3].y / 2); + else + sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y; + + b_uv_dy = (sum >> 3) + roundtab_76[sum & 0xf]; } start_timer(); - interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos, 16 * y_pos, - pMB->mvs[0].x, pMB->mvs[0].y, stride, 0); - interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos + 8, 16 * y_pos, - pMB->mvs[1].x, pMB->mvs[1].y, stride, 0); - interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos, 16 * y_pos + 8, - pMB->mvs[2].x, pMB->mvs[2].y, stride, 0); - interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos + 8, - 16 * y_pos + 8, pMB->mvs[3].x, pMB->mvs[3].y, stride, - 0); + if(dec->quarterpel) { + if((pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q)) + 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->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->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->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->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); + } + } + else { + interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos, 16 * y_pos, + pMB->mvs[0].x, pMB->mvs[0].y, stride, 0); + interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos + 8, 16 * y_pos, + pMB->mvs[1].x, pMB->mvs[1].y, stride, 0); + interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos, 16 * y_pos + 8, + pMB->mvs[2].x, pMB->mvs[2].y, stride, 0); + interpolate8x8_switch(dec->cur.y, forward.y, 16 * x_pos + 8, + 16 * y_pos + 8, pMB->mvs[3].x, pMB->mvs[3].y, stride, + 0); + } + interpolate8x8_switch(dec->cur.u, forward.u, 8 * x_pos, 8 * y_pos, uv_dx, uv_dy, stride2, 0); interpolate8x8_switch(dec->cur.v, forward.v, 8 * x_pos, 8 * y_pos, uv_dx, uv_dy, stride2, 0); - interpolate8x8_switch(dec->refn[2].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, - 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, - 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, - 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, + if(dec->quarterpel) { + if((pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q)) + 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->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->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->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->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->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->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->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->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->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_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->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->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->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->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->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->tmp.v + (8 * y_pos * stride2) + 8 * x_pos, + stride2, 1, 8); + stop_comp_timer(); for (i = 0; i < 6; i++) { - if (cbp & (1 << (5 - i))) // coded + int direction = dec->alternate_vertical_scan ? 2 : 0; + + 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]); + get_inter_block(bs, &block[i * 64], direction); stop_coding_timer(); start_timer(); @@ -1084,26 +1493,26 @@ } -// 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) { @@ -1140,9 +1549,9 @@ start_timer(); image_setedges(&dec->refn[0], 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, dec->interlacing); + dec->width, dec->height); stop_edges_timer(); #ifdef BFRAMES_DEC_DEBUG @@ -1152,7 +1561,7 @@ #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]; @@ -1162,9 +1571,14 @@ 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; @@ -1172,19 +1586,21 @@ #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; @@ -1205,7 +1621,7 @@ 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 @@ -1234,7 +1650,7 @@ / 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); @@ -1252,7 +1668,7 @@ 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: @@ -1262,7 +1678,7 @@ 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: @@ -1272,24 +1688,15 @@ 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); - ; - } - - if ( (x==19) && (y==8) ) - { - fprintf(stderr,"D %d %d %d %d %d %d \n",0, mb->mb_type, - mb->mvs[0].x, mb->mvs[0].y,mb->b_mvs[0].x, mb->b_mvs[0].y ); + 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; @@ -1299,135 +1706,268 @@ #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_frame_t * frame, xvid_dec_stats_t * stats) { Bitstream bs; uint32_t rounding; + uint32_t reduced_resolution; uint32_t quant; uint32_t fcode_forward; uint32_t fcode_backward; uint32_t intra_dc_threshold; - uint32_t vop_type; + WARPPOINTS gmc_warp; + int coding_type; + int success, output, seen_something; + idctFuncPtr idct_save; + + if (XVID_MAJOR(frame->version) != 1 || (stats && XVID_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); - // add by chenm001 - // for support B-frame to reference last 2 frame - dec->frames++; - vop_type = - BitstreamReadHeaders(&bs, dec, &rounding, &quant, &fcode_forward, - &fcode_backward, &intra_dc_threshold); - - 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, quant, fcode_forward, - intra_dc_threshold); -#ifdef BFRAMES_DEC - DEBUG1("P_VOP Time=", dec->time); -#endif - break; + /* XXX: 0x7f is only valid whilst decoding vfw xvid/divx5 avi's */ + if(dec->low_delay_default && frame->length == 1 && BitstreamShowBits(&bs, 8) == 0x7f) + { + image_output(&dec->refn[0], dec->width, dec->height, dec->edged_width, + (uint8_t**)frame->output.plane, frame->output.stride, frame->output.csp, dec->interlacing); + if (stats) stats->type = XVID_TYPE_NOTHING; + emms(); + return 1; /* one byte consumed */ + } + + success = 0; + output = 0; + seen_something = 0; + idct_save = idct; + +repeat: + + coding_type = BitstreamReadHeaders(&bs, dec, &rounding, &reduced_resolution, + &quant, &fcode_forward, &fcode_backward, &intra_dc_threshold, &gmc_warp); + + 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 (coding_type == -1) /* nothing */ + { + 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->type = XVID_TYPE_VOL; + stats->data.vol.general = 0; + /*XXX: if (dec->interlacing) + stats->data.vol.general |= ++INTERLACING; */ + stats->data.vol.width = dec->width; + stats->data.vol.height = dec->height; + stats->data.vol.par = dec->aspect_ratio; + stats->data.vol.par_width = dec->par_width; + stats->data.vol.par_height = dec->par_height; + emms(); + return BitstreamPos(&bs)/8; /* number of bytes consumed */ + } + goto repeat; + } - case I_VOP: - decoder_iframe(dec, &bs, quant, intra_dc_threshold); -#ifdef BFRAMES_DEC - DEBUG1("I_VOP Time=", dec->time); + dec->p_bmv.x = dec->p_bmv.y = dec->p_fmv.y = dec->p_fmv.y = 0; /* init pred vector to 0 */ + +#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; + } + /* ignore otherwise */ + } + else if (coding_type != B_VOP) + { + switch(coding_type) + { + 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; } -#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); - break; + if (reduced_resolution) + { + image_deblock_rrv(&dec->cur, dec->edged_width, dec->mbs, + (dec->width + 31) / 32, (dec->height + 31) / 32, dec->mb_width, + 16, 0); + } - default: - return XVID_ERR_FAIL; - } + /* note: for packed_mode, output is performed when the special-N_VOP is decoded */ + if (!(dec->low_delay_default && dec->packed_mode)) + { + if (dec->low_delay) + { + decoder_output(dec, &dec->cur, dec->mbs, frame, stats, coding_type); + output = 1; + } + else if (dec->frames > 0) /* is the reference frame valid? */ + { + /* output the reference frame */ + decoder_output(dec, &dec->refn[0], dec->last_mbs, frame, stats, dec->last_coding_type); + output = 1; + } + } -#ifdef BFRAMES_DEC_DEBUG - if (frame->length != BitstreamPos(&bs) / 8){ - DEBUG2("InLen/UseLen",frame->length, BitstreamPos(&bs) / 8); - } -#endif - frame->length = BitstreamPos(&bs) / 8; + 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; + dec->frames++; + seen_something = 1; -#ifdef BFRAMES_DEC - // test if no B_VOP - if (dec->low_delay) { -#endif - image_output(&dec->cur, dec->width, dec->height, dec->edged_width, - frame->image, frame->stride, frame->colorspace); + }else{ /* B_VOP */ -#ifdef BFRAMES_DEC - } else { - if (dec->frames >= 0) { - start_timer(); - if ((vop_type == I_VOP || vop_type == P_VOP)) { - image_output(&dec->refn[0], dec->width, dec->height, - dec->edged_width, frame->image, frame->stride, - frame->colorspace); - } else if (vop_type == B_VOP) { - image_output(&dec->cur, dec->width, dec->height, - dec->edged_width, frame->image, frame->stride, - frame->colorspace); - } - stop_conv_timer(); + if (dec->low_delay) + { + DPRINTF(XVID_DEBUG_ERROR, "warning: bvop found in low_delay==1 stream\n"); + dec->low_delay = 1; + } + + 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); } + + decoder_output(dec, &dec->cur, dec->mbs, frame, stats, coding_type); + output = 1; + dec->frames++; } -#endif - if (vop_type == I_VOP || vop_type == P_VOP) { - image_swap(&dec->refn[0], &dec->refn[1]); - image_swap(&dec->cur, &dec->refn[0]); + BitstreamByteAlign(&bs); - // 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); + /* low_delay_default mode: repeat in packed_mode */ + if (dec->low_delay_default && dec->packed_mode && output == 0 && success == 0) + { + success = 1; + goto repeat; } - emms(); +done : + /* 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) + { + 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 */ }