/*!
**************************************************************************************
* \brief
* Allocate memory for error resilient RDO.
**************************************************************************************
*/
int allocate_errdo_mem(VideoParameters *p_Vid, InputParameters *p_Inp)
{
int memory_size = 0;
//allocate shared memory for all algorithms
p_Vid->p_decs = (Decoders *) malloc(sizeof(Decoders));
memory_size += get_mem3Dint(&p_Vid->p_decs->res_img, MAX_PLANE, MB_BLOCK_SIZE, MB_BLOCK_SIZE);
memory_size += get_mem3Dint(&p_Vid->p_decs->res_mb_best8x8, MAX_PLANE, MB_BLOCK_SIZE, MB_BLOCK_SIZE);
p_Vid->p_decs->RCD_bestY_mb = NULL;
p_Vid->p_decs->RCD_bestY_b8x8 = NULL;
p_Vid->p_decs->MVCD_bestY_mb = NULL;
p_Vid->p_decs->MVCD_bestY_b8x8 = NULL;
p_Vid->p_decs->flag_bestY_mb = NULL;
p_Vid->p_decs->flag_bestY_b8x8 = NULL;
p_Vid->p_decs->flag_wo_res = NULL;
p_Vid->p_decs->flag_wo_res_bestY_b8x8 = NULL;
p_Vid->p_decs->trans_dist_bestY_mb = NULL;
p_Vid->p_decs->trans_dist_bestY_b8x8 = NULL;
p_Vid->p_decs->trans_dist_wo_res = NULL; //it is used for P8x8, where residual may be set to 0
p_Vid->p_decs->trans_dist_wo_res_bestY_b8x8 = NULL; //it is used for P8x8, where residual may be set to 0
p_Vid->p_decs->trans_err_bestY_mb = NULL;
p_Vid->p_decs->trans_err_bestY_b8x8 = NULL;
p_Vid->p_decs->trans_err_wo_res = NULL; //it is used for P8x8, where residual may be set to 0
p_Vid->p_decs->trans_err_wo_res_bestY_b8x8 = NULL; //it is used for P8x8, where residual may be set to 0
p_Vid->p_decs->dec_mb_pred = NULL;
p_Vid->p_decs->dec_mbY_best = NULL;
p_Vid->p_decs->dec_mb_pred_best8x8 = NULL;
p_Vid->p_decs->dec_mbY_best8x8 = NULL;
p_Vid->p_decs->first_moment_bestY_mb = NULL;
p_Vid->p_decs->first_moment_bestY_b8x8 = NULL;
p_Vid->p_decs->first_moment_pred_bestY_b8x8 = NULL;
p_Vid->p_decs->first_moment_pred = NULL;
p_Vid->p_decs->second_moment_bestY_mb = NULL;
p_Vid->p_decs->second_moment_bestY_b8x8 = NULL;
p_Vid->p_decs->second_moment_pred_bestY_b8x8 = NULL;
p_Vid->p_decs->second_moment_pred = NULL;
//Zhifeng 090630
switch (p_Inp->de)
{
case LLN:
//allocate memory for lln algorithm
memory_size += get_mem3Dpel(&p_Vid->p_decs->dec_mb_pred, p_Inp->NoOfDecoders, MB_BLOCK_SIZE, MB_BLOCK_SIZE);
memory_size += get_mem3Dpel(&p_Vid->p_decs->dec_mbY_best, p_Inp->NoOfDecoders, MB_BLOCK_SIZE, MB_BLOCK_SIZE);
memory_size += get_mem4Dpel(&p_Vid->p_decs->dec_mbY_best8x8, 2, p_Inp->NoOfDecoders, MB_BLOCK_SIZE, MB_BLOCK_SIZE);
memory_size += get_mem4Dpel(&p_Vid->p_decs->dec_mb_pred_best8x8, 2, p_Inp->NoOfDecoders, MB_BLOCK_SIZE, MB_BLOCK_SIZE);
break;
default:
;
}
return memory_size;
}
/*!
*************************************************************************************
* \brief
* allocate storable picture memory for errdo
*
*************************************************************************************
*/
void errdo_alloc_storable_picture(StorablePicture *p, VideoParameters *p_Vid, InputParameters *p_Inp, int size_x, int size_y, int size_x_cr, int size_y_cr)
{
Dist_Estm *s;
int dec, ndec, nplane;
p->de_mem = (Dist_Estm *)malloc( sizeof(Dist_Estm) );
s = p->de_mem;
s->res_con_diff_Y = NULL;
s->res_con_diff_UV = NULL;
s->MV_con_diff_Y = NULL;
s->MV_con_diff_UV = NULL;
s->error_sign_flag_Y = NULL;
s->error_sign_flag_UV = NULL;
s->transmission_dist_Y = NULL;
s->transmission_dist_UV = NULL;
s->transmission_err_Y = NULL;
s->transmission_err_UV = NULL;
s->dec_imgY = NULL;
s->dec_imgUV = NULL;
s->mb_error_map = NULL;
s->first_moment_Y = NULL;
s->first_moment_UV = NULL;
s->second_moment_Y = NULL;
s->second_moment_UV = NULL;
for (nplane = 0; nplane < 3; nplane++)
{
s->p_res_con_diff[nplane] = NULL;
s->p_MV_con_diff[nplane] = NULL;
s->p_error_sign_flag[nplane] = NULL;
s->p_transmission_dist[nplane] = NULL;
s->p_transmission_err[nplane] = NULL;
s->p_dec_img[nplane] = NULL;
s->p_first_moment[nplane] = NULL;
s->p_second_moment[nplane] = NULL;
}
switch (p_Inp->de)
{
case LLN:
ndec = p_Inp->NoOfDecoders;
//check the consistent
if (ndec == 0)
{
printf("Number of decoders cannot be zero for LLN and fast LLN algorithms, resetting to 30");
ndec = 30;
}
get_mem3D(&(s->mb_error_map), ndec, size_y/MB_BLOCK_SIZE, size_x/MB_BLOCK_SIZE);
get_mem3Dpel(&(s->dec_imgY), ndec, size_y, size_x);
// This seems somewhat inefficient. Why not allocate array as [ndec][x] where x goes from 0 to 2?
if ((s->p_dec_img[0] = (imgpel***)calloc(ndec,sizeof(imgpel**))) == NULL)
{
no_mem_exit("errdo.c: p_dec_img[0]");
}
if (p_Vid->yuv_format != YUV400)
{
get_mem4Dpel(&(s->dec_imgUV), ndec, 2, size_y_cr, size_x_cr);
if ((s->p_dec_img[1] = (imgpel***)calloc(ndec,sizeof(imgpel**))) == NULL)
{
no_mem_exit("errdo.c: p_dec_img[1]");
}
if ((s->p_dec_img[2] = (imgpel***)calloc(ndec,sizeof(imgpel**))) == NULL)
{
no_mem_exit("errdo.c: p_dec_img[2]");
}
}
for (dec = 0; dec < ndec; dec++)
{
s->p_dec_img[0][dec] = s->dec_imgY[dec];
}
if (p_Vid->yuv_format != YUV400)
{
for (dec = 0; dec < ndec; dec++)
{
s->p_dec_img[1][dec] = s->dec_imgUV[dec][0];
s->p_dec_img[2][dec] = s->dec_imgUV[dec][1];
}
}
break;
default:
;
}
}
/*!
************************************************************************
* \brief
* Dynamic memory allocation of frame size related global buffers
* buffers are defined in global.h, allocated memory must be freed in
* void free_global_buffers()
*
* \par Input:
* Input Parameters struct inp_par *inp, Image Parameters struct img_par *img
*
* \par Output:
* Number of allocated bytes
***********************************************************************
*/
int init_global_buffers()
{
int memory_size=0;
int quad_range, i;
if (global_init_done)
{
free_global_buffers();
}
// allocate memory for reference frame in find_snr
memory_size += get_mem2Dpel(&imgY_ref, img->height, img->width);
if (active_sps->chroma_format_idc != YUV400)
memory_size += get_mem3Dpel(&imgUV_ref, 2, img->height_cr, img->width_cr);
else
imgUV_ref=NULL;
// allocate memory in structure img
if(((img->mb_data) = (Macroblock *) calloc(img->FrameSizeInMbs, sizeof(Macroblock))) == NULL)
no_mem_exit("init_global_buffers: img->mb_data");
if(((img->intra_block) = (int*)calloc(img->FrameSizeInMbs, sizeof(int))) == NULL)
no_mem_exit("init_global_buffers: img->intra_block");
memory_size += get_mem2Dint(&(img->ipredmode), 4*img->PicWidthInMbs , 4*img->FrameHeightInMbs);
memory_size += get_mem2Dint(&(img->field_anchor),4*img->FrameHeightInMbs, 4*img->PicWidthInMbs);
memory_size += get_mem3Dint(&(img->wp_weight), 2, MAX_REFERENCE_PICTURES, 3);
memory_size += get_mem3Dint(&(img->wp_offset), 6, MAX_REFERENCE_PICTURES, 3);
memory_size += get_mem4Dint(&(img->wbp_weight), 6, MAX_REFERENCE_PICTURES, MAX_REFERENCE_PICTURES, 3);
// CAVLC mem
memory_size += get_mem3Dint(&(img->nz_coeff), img->FrameSizeInMbs, 4, 4 + img->num_blk8x8_uv);
memory_size += get_mem2Dint(&(img->siblock),img->PicWidthInMbs , img->FrameHeightInMbs);
if(img->max_imgpel_value > img->max_imgpel_value_uv || active_sps->chroma_format_idc == YUV400)
quad_range = (img->max_imgpel_value + 1) * 2;
else
quad_range = (img->max_imgpel_value_uv + 1) * 2;
if ((img->quad = (int*)calloc (quad_range, sizeof(int))) == NULL)
no_mem_exit ("init_img: img->quad");
for (i=0; i < quad_range/2; ++i)
{
img->quad[i]=i*i;
}
#ifdef ADAPTIVE_FILTER
memory_size += get_mem2Ddouble (&tmp_coef, 21, 16);
#endif
#ifdef ADAPTIVE_LOOP_FILTER
memory_size += InitALFGlobalBuffers();
#endif
global_init_done = 1;
img->oldFrameSizeInMbs = img->FrameSizeInMbs;
return (memory_size);
}
/*!
*************************************************************************************
* \brief
* Mode Decision for a macroblock
*************************************************************************************
*/
void encode_one_macroblock_low (Macroblock *currMB)
{
Slice *currSlice = currMB->p_slice;
RDOPTStructure *p_RDO = currSlice->p_RDO;
ImageParameters *p_Img = currMB->p_Img;
InputParameters *p_Inp = currMB->p_Inp;
PicMotionParams *motion = &p_Img->enc_picture->motion;
imgpel ***temp_img; // to temp store the Y data for 8x8 transform
int block, mode, i=0, j;
RD_PARAMS enc_mb;
int bmcost[5] = {INT_MAX};
double rd_cost = 0;
int cost = 0;
int min_cost = INT_MAX, cost_direct=0, have_direct=0;
int intra1 = 0;
int temp_cpb = 0;
Boolean best_transform_flag = FALSE;
int cost8x8_direct = 0;
short islice = (short) (currSlice->slice_type == I_SLICE);
short bslice = (short) (currSlice->slice_type == B_SLICE);
short pslice = (short) ((currSlice->slice_type == P_SLICE) || (currSlice->slice_type == SP_SLICE));
short intra = (short) (islice || (pslice && currMB->mb_y == p_Img->mb_y_upd && p_Img->mb_y_upd!=p_Img->mb_y_intra));
int lambda_mf[3];
Block8x8Info *b8x8info = p_Img->b8x8info;
//int mb_available[3] = { 1, 1, 1};
char **ipredmodes = p_Img->ipredmode;
short *allmvs = (currSlice->slice_type == I_SLICE) ? NULL: currSlice->all_mv[0][0][0][0][0];
int ****i4p; //for non-RD-opt. mode
imgpel **mb_pred = currSlice->mb_pred[0];
Boolean tmp_8x8_flag, tmp_no_mbpart;
BestMode md_best;
Info8x8 best;
init_md_best(&md_best);
// Init best (need to create simple function)
best.pdir = 0;
best.bipred = 0;
best.ref[LIST_0] = 0;
best.ref[LIST_1] = -1;
get_mem3Dpel(&temp_img, 3, MB_BLOCK_SIZE, MB_BLOCK_SIZE);
intra |= RandomIntra (p_Img, currMB->mbAddrX); // Forced Pseudo-Random Intra
//===== Setup Macroblock encoding parameters =====
init_enc_mb_params(currMB, &enc_mb, intra);
if (p_Inp->AdaptiveRounding)
{
reset_adaptive_rounding(p_Img);
}
if (currSlice->MbaffFrameFlag)
{
reset_mb_nz_coeff(p_Img, currMB->mbAddrX);
}
//===== S T O R E C O D I N G S T A T E =====
//---------------------------------------------------
currSlice->store_coding_state (currMB, currSlice->p_RDO->cs_cm);
if (!intra)
{
//===== set direct motion vectors =====
currMB->best_mode = 10; // let us set best_mode to an intra mode to avoid possible bug with RDOQ
if (bslice && enc_mb.valid[0])
{
currSlice->Get_Direct_Motion_Vectors (currMB);
}
if (p_Inp->CtxAdptLagrangeMult == 1)
{
get_initial_mb16x16_cost(currMB);
}
//===== MOTION ESTIMATION FOR 16x16, 16x8, 8x16 BLOCKS =====
for (mode = 1; mode < 4; mode++)
{
best.bipred = 0;
best.mode = (char) mode;
b8x8info->best[mode][0].bipred = 0;
if (enc_mb.valid[mode])
{
for (cost=0, block=0; block<(mode==1?1:2); block++)
{
update_lambda_costs(currMB, &enc_mb, lambda_mf);
PartitionMotionSearch (currMB, mode, block, lambda_mf);
//--- set 4x4 block indizes (for getting MV) ---
j = (block==1 && mode==2 ? 2 : 0);
i = (block==1 && mode==3 ? 2 : 0);
//--- get cost and reference frame for List 0 prediction ---
bmcost[LIST_0] = INT_MAX;
list_prediction_cost(currMB, LIST_0, block, mode, &enc_mb, bmcost, best.ref);
if (bslice)
{
//--- get cost and reference frame for List 1 prediction ---
bmcost[LIST_1] = INT_MAX;
list_prediction_cost(currMB, LIST_1, block, mode, &enc_mb, bmcost, best.ref);
// Compute bipredictive cost between best list 0 and best list 1 references
list_prediction_cost(currMB, BI_PRED, block, mode, &enc_mb, bmcost, best.ref);
// currently Bi predictive ME is only supported for modes 1, 2, 3 and ref 0
if (is_bipred_enabled(p_Inp, mode))
{
list_prediction_cost(currMB, BI_PRED_L0, block, mode, &enc_mb, bmcost, 0);
list_prediction_cost(currMB, BI_PRED_L1, block, mode, &enc_mb, bmcost, 0);
}
else
{
bmcost[BI_PRED_L0] = INT_MAX;
bmcost[BI_PRED_L1] = INT_MAX;
}
// Determine prediction list based on mode cost
determine_prediction_list(bmcost, &best, &cost);
}
else // if (bslice)
{
best.pdir = 0;
cost += bmcost[LIST_0];
}
assign_enc_picture_params(currMB, mode, &best, 2 * block);
//----- set reference frame and direction parameters -----
set_block8x8_info(b8x8info, mode, block, &best);
//--- set reference frames and motion vectors ---
if (mode>1 && block==0)
currSlice->set_ref_and_motion_vectors (currMB, motion, &best, block);
} // for (block=0; block<(mode==1?1:2); block++)
currMB->luma_transform_size_8x8_flag = FALSE;
if (p_Inp->Transform8x8Mode) //for inter rd-off, set 8x8 to do 8x8 transform
{
SetModesAndRefframeForBlocks(currMB, (short) mode);
currMB->luma_transform_size_8x8_flag = (byte) TransformDecision(currMB, -1, &cost);
}
if (cost < min_cost)
{
currMB->best_mode = (short) mode;
min_cost = cost;
best_transform_flag = currMB->luma_transform_size_8x8_flag;
if (p_Inp->CtxAdptLagrangeMult == 1)
{
adjust_mb16x16_cost(currMB, cost);
}
}
} // if (enc_mb.valid[mode])
} // for (mode=1; mode<4; mode++)
if (enc_mb.valid[P8x8])
{
//===== store coding state of macroblock =====
currSlice->store_coding_state (currMB, currSlice->p_RDO->cs_mb);
memset( currSlice->cofAC[0][0][0], 0, 2080 * sizeof(int)); // 4 * 4 * 2 * 65
currMB->valid_8x8 = FALSE;
if (p_Inp->Transform8x8Mode)
{
ResetRD8x8Data(p_Img, p_RDO->tr8x8);
//===========================================================
// Check 8x8 partition with transform size 8x8
//===========================================================
//===== LOOP OVER 8x8 SUB-PARTITIONS (Motion Estimation & Mode Decision) =====
for (cost_direct = 0, block = 0; block < 4; block++)
{
submacroblock_mode_decision_low(currMB, &enc_mb, p_RDO->tr8x8, p_RDO->cofAC8x8ts[block],
&have_direct, block, &cost_direct, &cost, &cost8x8_direct, 1);
set_subblock8x8_info(b8x8info, P8x8, block, p_RDO->tr8x8);
}
currMB->luma_transform_size_8x8_flag = FALSE; //switch to 4x4 transform size
}// if (p_Inp->Transform8x8Mode)
if (p_Inp->Transform8x8Mode != 2)
{
ResetRD8x8Data(p_Img, p_RDO->tr4x4);
//=================================================================
// Check 8x8, 8x4, 4x8 and 4x4 partitions with transform size 4x4
//=================================================================
//===== LOOP OVER 8x8 SUB-PARTITIONS (Motion Estimation & Mode Decision) =====
for (cost_direct = 0, block=0; block<4; block++)
{
submacroblock_mode_decision_low(currMB, &enc_mb, p_RDO->tr4x4, p_RDO->coefAC8x8[block],
&have_direct, block, &cost_direct, &cost, &cost8x8_direct, 0);
set_subblock8x8_info(b8x8info, P8x8, block, p_RDO->tr4x4);
}
}// if (p_Inp->Transform8x8Mode != 2)
if (p_Inp->RCEnable)
rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);
//check cost for P8x8 for non-rdopt mode
if (((p_Inp->Transform8x8Mode < 2) && (p_RDO->tr4x4->mb_p8x8_cost < min_cost)) ||
((p_Inp->Transform8x8Mode > 0) && (p_RDO->tr8x8->mb_p8x8_cost < min_cost)))
{
currMB->best_mode = P8x8;
if (p_Inp->Transform8x8Mode == 2)
{
min_cost = p_RDO->tr8x8->mb_p8x8_cost;
currMB->luma_transform_size_8x8_flag = TRUE;
}
else if (p_Inp->Transform8x8Mode)
{
if (p_RDO->tr8x8->mb_p8x8_cost < p_RDO->tr4x4->mb_p8x8_cost)
{
min_cost = p_RDO->tr8x8->mb_p8x8_cost;
currMB->luma_transform_size_8x8_flag = TRUE;
}
else if(p_RDO->tr4x4->mb_p8x8_cost < p_RDO->tr8x8->mb_p8x8_cost)
{
min_cost = p_RDO->tr4x4->mb_p8x8_cost;
currMB->luma_transform_size_8x8_flag = FALSE;
}
else
{
if (GetBestTransformP8x8(currMB) == 0)
{
min_cost = p_RDO->tr4x4->mb_p8x8_cost;
currMB->luma_transform_size_8x8_flag = FALSE;
}
else
{
min_cost = p_RDO->tr8x8->mb_p8x8_cost;
currMB->luma_transform_size_8x8_flag = TRUE;
}
}
}
else
{
min_cost = p_RDO->tr4x4->mb_p8x8_cost;
currMB->luma_transform_size_8x8_flag = FALSE;
}
}// if ((p_RDO->tr4x4->mb_p8x8_cost < min_cost || p_RDO->tr8x8->mb_p8x8_cost < min_cost))
p_Img->giRDOpt_B8OnlyFlag = FALSE;
}
// Find a motion vector for the Skip mode
if(pslice)
FindSkipModeMotionVector (currMB);
}
else // if (!intra)
{
min_cost = INT_MAX;
}
//========= C H O O S E B E S T M A C R O B L O C K M O D E =========
//-------------------------------------------------------------------------
tmp_8x8_flag = currMB->luma_transform_size_8x8_flag; //save 8x8_flag
tmp_no_mbpart = currMB->NoMbPartLessThan8x8Flag; //save no-part-less
if ((p_Img->yuv_format != YUV400) && (p_Img->yuv_format != YUV444))
// precompute all chroma intra prediction modes
intra_chroma_prediction(currMB, NULL, NULL, NULL);
if (enc_mb.valid[0] && bslice) // check DIRECT MODE
{
if(have_direct)
{
switch(p_Inp->Transform8x8Mode)
{
case 1: // Mixture of 8x8 & 4x4 transform
cost = ((cost8x8_direct < cost_direct) || !(enc_mb.valid[5] && enc_mb.valid[6] && enc_mb.valid[7]))
? cost8x8_direct : cost_direct;
break;
case 2: // 8x8 Transform only
cost = cost8x8_direct;
break;
default: // 4x4 Transform only
cost = cost_direct;
break;
}
}
else
{ //!have_direct
cost = GetDirectCostMB (currMB);
}
if (cost!=INT_MAX)
{
cost -= (int)floor(16 * enc_mb.lambda_md + 0.4999);
}
if (cost <= min_cost)
{
if(p_Img->active_sps->direct_8x8_inference_flag && p_Inp->Transform8x8Mode)
{
if(p_Inp->Transform8x8Mode==2)
currMB->luma_transform_size_8x8_flag = TRUE;
else
{
if(cost8x8_direct < cost_direct)
currMB->luma_transform_size_8x8_flag = TRUE;
else
currMB->luma_transform_size_8x8_flag = FALSE;
}
}
else
currMB->luma_transform_size_8x8_flag = FALSE;
//Rate control
if (p_Inp->RCEnable)
rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);
min_cost = cost;
currMB->best_mode = 0;
tmp_8x8_flag = currMB->luma_transform_size_8x8_flag;
}
else
{
currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore if not best
currMB->NoMbPartLessThan8x8Flag = (byte) tmp_no_mbpart; // restore if not best
}
}
currMB->min_rdcost = (double) min_cost;
if (enc_mb.valid[I8MB]) // check INTRA8x8
{
currMB->luma_transform_size_8x8_flag = TRUE; // at this point cost will ALWAYS be less than min_cost
currMB->mb_type = currMB->ar_mode = I8MB;
temp_cpb = Mode_Decision_for_Intra8x8Macroblock (currMB, enc_mb.lambda_md, &rd_cost);
if (rd_cost <= currMB->min_rdcost) //HYU_NOTE. bug fix. 08/15/07
{
currMB->cbp = temp_cpb;
if (p_Img->P444_joined)
{
currSlice->curr_cbp[0] = currSlice->cmp_cbp[1];
currSlice->curr_cbp[1] = currSlice->cmp_cbp[2];
}
if(enc_mb.valid[I4MB])
{
//coeffs
if (p_Inp->Transform8x8Mode != 2)
{
i4p = p_RDO->cofAC;
p_RDO->cofAC = currSlice->cofAC;
currSlice->cofAC = i4p;
}
}
copy_image_data_16x16(temp_img[0], &p_Img->enc_picture->imgY[currMB->pix_y], 0, currMB->pix_x);
if (p_Img->P444_joined)
{
copy_image_data_16x16(temp_img[1], &p_Img->enc_picture->imgUV[0][currMB->pix_y], 0, currMB->pix_x);
copy_image_data_16x16(temp_img[2], &p_Img->enc_picture->imgUV[1][currMB->pix_y], 0, currMB->pix_x);
}
//Rate control
if (p_Inp->RCEnable)
rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);
currMB->min_rdcost = rd_cost;
currMB->best_mode = I8MB;
tmp_8x8_flag = currMB->luma_transform_size_8x8_flag;
}
else
{
currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore if not best
if (p_Img->P444_joined)
{
currMB->cbp |= currSlice->curr_cbp[0];
currMB->cbp |= currSlice->curr_cbp[1];
currSlice->cmp_cbp[1] = currMB->cbp;
currSlice->cmp_cbp[2] = currMB->cbp;
}
}
}
if (enc_mb.valid[I4MB]) // check INTRA4x4
{
currMB->luma_transform_size_8x8_flag = FALSE;
currMB->mb_type = currMB->ar_mode = I4MB;
temp_cpb = Mode_Decision_for_Intra4x4Macroblock (currMB, enc_mb.lambda_md, &rd_cost);
if (rd_cost <= currMB->min_rdcost)
{
currMB->cbp = temp_cpb;
//Rate control
if (p_Inp->RCEnable)
rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);
currMB->min_rdcost = rd_cost;
currMB->best_mode = I4MB;
tmp_8x8_flag = currMB->luma_transform_size_8x8_flag;
}
else
{
currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore if not best
if (p_Img->P444_joined)
{
currMB->cbp |= currSlice->curr_cbp[0];
currMB->cbp |= currSlice->curr_cbp[1];
currSlice->cmp_cbp[1] = currMB->cbp;
currSlice->cmp_cbp[2] = currMB->cbp;
}
//coeffs
i4p = p_RDO->cofAC;
p_RDO->cofAC = currSlice->cofAC;
currSlice->cofAC=i4p;
}
}
if (enc_mb.valid[I16MB]) // check INTRA16x16
{
currMB->luma_transform_size_8x8_flag = FALSE;
intrapred_16x16 (currMB, PLANE_Y);
if (p_Img->P444_joined)
{
select_plane(p_Img, PLANE_U);
intrapred_16x16 (currMB, PLANE_U);
select_plane(p_Img, PLANE_V);
intrapred_16x16 (currMB, PLANE_V);
select_plane(p_Img, PLANE_Y);
}
rd_cost = currSlice->find_sad_16x16 (currMB);
if (rd_cost < currMB->min_rdcost)
{
//Rate control
if (p_Inp->RCEnable)
rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, currSlice->mpr_16x16[0][(short) currMB->i16mode]);
currMB->best_mode = I16MB;
currMB->min_rdcost = rd_cost;
currMB->cbp = currMB->trans_16x16 (currMB, PLANE_Y);
if (p_Img->P444_joined)
{
select_plane(p_Img, PLANE_U);
currSlice->cmp_cbp[1] = currMB->trans_16x16(currMB, PLANE_U);
select_plane(p_Img, PLANE_V);
currSlice->cmp_cbp[2] = currMB->trans_16x16(currMB, PLANE_V);
select_plane(p_Img, PLANE_Y);
currMB->cbp |= currSlice->cmp_cbp[1];
currMB->cbp |= currSlice->cmp_cbp[2];
currSlice->cmp_cbp[1] = currMB->cbp;
currSlice->cmp_cbp[2] = currMB->cbp;
}
}
else
{
currMB->luma_transform_size_8x8_flag = (byte) tmp_8x8_flag; // restore
currMB->NoMbPartLessThan8x8Flag = (byte) tmp_no_mbpart; // restore
}
}
intra1 = IS_INTRA(currMB);
//===== S E T F I N A L M A C R O B L O C K P A R A M E T E R S ======
//---------------------------------------------------------------------------
{
//===== set parameters for chosen mode =====
SetModesAndRefframeForBlocks (currMB, currMB->best_mode);
if (currMB->best_mode == P8x8)
{
if (currMB->luma_transform_size_8x8_flag && (p_RDO->tr8x8->cbp8x8 == 0) && p_Inp->Transform8x8Mode != 2)
currMB->luma_transform_size_8x8_flag = FALSE;
SetCoeffAndReconstruction8x8 (currMB);
memset(currMB->intra_pred_modes, DC_PRED, MB_BLOCK_PARTITIONS * sizeof(char));
for (j = currMB->block_y; j < currMB->block_y + BLOCK_MULTIPLE; j++)
memset(&ipredmodes[j][currMB->block_x], DC_PRED, BLOCK_MULTIPLE * sizeof(char));
}
else
{
//===== set parameters for chosen mode =====
if (currMB->best_mode == I8MB)
{
memcpy(currMB->intra_pred_modes,currMB->intra_pred_modes8x8, MB_BLOCK_PARTITIONS * sizeof(char));
for(j = currMB->block_y; j < currMB->block_y + BLOCK_MULTIPLE; j++)
memcpy(&p_Img->ipredmode[j][currMB->block_x],&p_Img->ipredmode8x8[j][currMB->block_x], BLOCK_MULTIPLE * sizeof(char));
//--- restore reconstruction for 8x8 transform ---
copy_image_data_16x16(&p_Img->enc_picture->imgY[currMB->pix_y], temp_img[0], currMB->pix_x, 0);
if (p_Img->P444_joined)
{
copy_image_data_16x16(&p_Img->enc_picture->imgUV[0][currMB->pix_y], temp_img[1], currMB->pix_x, 0);
copy_image_data_16x16(&p_Img->enc_picture->imgUV[1][currMB->pix_y], temp_img[2], currMB->pix_x, 0);
}
}
if ((currMB->best_mode!=I4MB)&&(currMB->best_mode != I8MB))
{
memset(currMB->intra_pred_modes,DC_PRED, MB_BLOCK_PARTITIONS * sizeof(char));
for(j = currMB->block_y; j < currMB->block_y + BLOCK_MULTIPLE; j++)
memset(&ipredmodes[j][currMB->block_x],DC_PRED, BLOCK_MULTIPLE * sizeof(char));
currMB->ar_mode = currMB->best_mode;
if (currMB->best_mode!=I16MB)
{
if((currMB->best_mode >= 1) && (currMB->best_mode <= 3))
currMB->luma_transform_size_8x8_flag = (byte) best_transform_flag;
if (currSlice->P444_joined)
{
luma_residual_coding_p444(currMB);
if((currMB->cbp==0 && currSlice->cmp_cbp[1] == 0 && currSlice->cmp_cbp[2] == 0) &&(currMB->best_mode == 0))
currMB->luma_transform_size_8x8_flag = FALSE;
}
else
{
luma_residual_coding(currMB);
if((currMB->cbp==0)&&(currMB->best_mode == 0))
currMB->luma_transform_size_8x8_flag = FALSE;
}
//Rate control
if (p_Inp->RCEnable)
rc_store_diff(currSlice->diffy, &p_Img->pCurImg[currMB->opix_y], currMB->pix_x, mb_pred);
}
}
}
//check luma cbp for transform size flag
if (((currMB->cbp&15) == 0) && currMB->mb_type != I4MB && currMB->mb_type != I8MB)
currMB->luma_transform_size_8x8_flag = FALSE;
// precompute all chroma intra prediction modes
if ((p_Img->yuv_format != YUV400) && (p_Img->yuv_format != YUV444))
intra_chroma_prediction(currMB, NULL, NULL, NULL);
currMB->i16offset = 0;
if ((p_Img->yuv_format != YUV400) && (p_Img->yuv_format != YUV444))
chroma_residual_coding (currMB);
if (currMB->best_mode == I16MB)
{
currMB->i16offset = I16Offset (currMB->cbp, currMB->i16mode);
}
currSlice->SetMotionVectorsMB (currMB, motion);
//===== check for SKIP mode =====
if(p_Img->P444_joined)
{
if ((pslice) && currMB->best_mode == 1 && currMB->cbp==0 && currSlice->cmp_cbp[1] == 0 && currSlice->cmp_cbp[2] == 0 &&
motion->ref_idx[LIST_0][currMB->block_y][currMB->block_x] == 0 &&
motion->mv [LIST_0][currMB->block_y][currMB->block_x][0] == allmvs[0] &&
motion->mv [LIST_0][currMB->block_y][currMB->block_x][1] == allmvs[1])
{
currMB->mb_type = currMB->b8x8[0].mode = currMB->b8x8[1].mode = currMB->b8x8[2].mode = currMB->b8x8[3].mode = 0;
currMB->luma_transform_size_8x8_flag = FALSE;
}
}
else if ((pslice) && currMB->best_mode == 1 && currMB->cbp==0 &&
motion->ref_idx[LIST_0][currMB->block_y][currMB->block_x] == 0 &&
motion->mv [LIST_0][currMB->block_y][currMB->block_x][0] == allmvs[0] &&
motion->mv [LIST_0][currMB->block_y][currMB->block_x][1] == allmvs[1])
{
currMB->mb_type = currMB->b8x8[0].mode = currMB->b8x8[1].mode = currMB->b8x8[2].mode = currMB->b8x8[3].mode = 0;
currMB->luma_transform_size_8x8_flag = FALSE;
}
if (currSlice->MbaffFrameFlag || (p_Inp->UseRDOQuant && currSlice->RDOQ_QP_Num > 1))
set_mbaff_parameters(currMB);
}
// Rate control
if(p_Inp->RCEnable && p_Inp->RCUpdateMode <= MAX_RC_MODE)
rc_store_mad(currMB);
//===== Decide if this MB will restrict the reference frames =====
if (p_Inp->RestrictRef)
update_refresh_map(currMB, intra, intra1);
/*update adaptive rounding offset p_Inp*/
if (p_Img->AdaptiveRounding)
{
update_offset_params(currMB, currMB->best_mode, currMB->luma_transform_size_8x8_flag);
}
free_mem3Dpel(temp_img);
}
