nengel@2: #include "h264_types.h"
nengel@2: #include "h264_data.h"
nengel@2: 
nengel@2: static inline void mc_dir_part(MBRecContext *d, MBRecState *mrs, H264Mb *m, DecodedPicture *pic, int n, int square,
nengel@2: 							   int chroma_height, int delta, int list,uint8_t *dest_y,
nengel@2: 							   uint8_t *dest_cb, uint8_t *dest_cr, int src_x_offset, int src_y_offset,
nengel@2: 							   qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
nengel@2: 	const int mx= mrs->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
nengel@2: 	const int my= mrs->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
nengel@2: 	const int luma_xy= (mx&3) + ((my&3)<<2);
nengel@2: 	const int pic_width  = 16*d->mb_width;
nengel@2: 	const int pic_height = 16*d->mb_height;
nengel@2: 
nengel@2: 	uint8_t *src_y, *src_cb, *src_cr;
nengel@2: 	int ymx= mx>>2;
nengel@2: 	int ymy= my>>2;
nengel@2: 	int cmy= my>>3;
nengel@2: 	int cmx= mx>>3;
nengel@2: 
nengel@2: 	//truncate the motion vectors references
nengel@2: 	if(ymy>= pic_height+2){
nengel@2: 		ymy=pic_height+1;
nengel@2: 	}else if(ymy <=-19){
nengel@2: 		ymy=-18;
nengel@2: 	}
nengel@2: 	if(ymx>= pic_width+2){
nengel@2: 		ymx= pic_width+1;
nengel@2: 	}else if(ymx<=-19){
nengel@2: 		ymx=-19;
nengel@2: 	}
nengel@2: 
nengel@2: 	src_y = pic->data[0] + ymx + ymy*d->linesize;
nengel@2: 	qpix_op[luma_xy](dest_y, src_y, d->linesize); //FIXME try variable height perhaps?
nengel@2: 	if(!square){
nengel@2: 		qpix_op[luma_xy](dest_y + delta, src_y + delta, d->linesize);
nengel@2: 	}
nengel@2: 
nengel@2: 	if(cmy >= pic_height>>1){
nengel@2: 		cmy = (pic_height>>1) -1;
nengel@2: 	}else if(cmy<=-9){
nengel@2: 		cmy=-8;
nengel@2: 	}
nengel@2: 	if(cmx >= pic_width>>1){
nengel@2: 		cmx = (pic_width>>1) -1;
nengel@2: 	}else if(cmx<=-9){
nengel@2: 		cmx=-8;
nengel@2: 	}
nengel@2: 
nengel@2: 	src_cb= pic->data[1] + cmx + cmy*d->uvlinesize;
nengel@2: 	src_cr= pic->data[2] + cmx + cmy*d->uvlinesize;
nengel@2: 
nengel@2: 	chroma_op(dest_cb, src_cb, d->uvlinesize, chroma_height, mx&7, my&7);
nengel@2: 	chroma_op(dest_cr, src_cr, d->uvlinesize, chroma_height, mx&7, my&7);
nengel@2: }
nengel@2: 
nengel@2: static inline void mc_part_std(MBRecContext *d, MBRecState *mrs, H264Slice *s, H264Mb *m, int n, int square, int chroma_height, int delta,
nengel@2: 								uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
nengel@2: 								int x_offset, int y_offset,
nengel@2: 								qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
nengel@2: 								qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
nengel@2: 								int list0, int list1){
nengel@2: 	qpel_mc_func *qpix_op=  qpix_put;
nengel@2: 	h264_chroma_mc_func chroma_op= chroma_put;
nengel@2: 
nengel@2: 	dest_y  += 2*x_offset + 2*y_offset*d->  linesize;
nengel@2: 	dest_cb +=   x_offset +   y_offset*d->uvlinesize;
nengel@2: 	dest_cr +=   x_offset +   y_offset*d->uvlinesize;
nengel@2: 	x_offset += 8*m->mb_x;
nengel@2: 	y_offset += 8*m->mb_y;
nengel@2: 
nengel@2: 	if(list0){
nengel@2: 		DecodedPicture *ref= s->dp_ref_list[0][ mrs->ref_cache[0][ scan8[n] ] ];
nengel@2: 		mc_dir_part(d, mrs, m, ref, n, square, chroma_height, delta, 0,
nengel@2: 					dest_y, dest_cb, dest_cr, x_offset, y_offset, qpix_op, chroma_op);
nengel@2: 
nengel@2: 		qpix_op=  qpix_avg;
nengel@2: 		chroma_op= chroma_avg;
nengel@2: 	}
nengel@2: 
nengel@2: 	if(list1){
nengel@2: 		DecodedPicture *ref= s->dp_ref_list[1][ mrs->ref_cache[1][ scan8[n] ] ];
nengel@2: 		mc_dir_part(d, mrs, m, ref, n, square, chroma_height, delta, 1,
nengel@2: 					dest_y, dest_cb, dest_cr, x_offset, y_offset, qpix_op, chroma_op);
nengel@2: 	}
nengel@2: }
nengel@2: 
nengel@2: static inline void mc_part_weighted(MBRecContext *d, MBRecState *mrs, H264Slice *s, H264Mb *m, int n, int square, int chroma_height, int delta,
nengel@2: 									uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
nengel@2: 									int x_offset, int y_offset,
nengel@2: 									qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
nengel@2: 									h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
nengel@2: 									h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
nengel@2: 									int list0, int list1){
nengel@2: 	dest_y  += 2*x_offset + 2*y_offset*d->  linesize;
nengel@2: 	dest_cb +=   x_offset +   y_offset*d->uvlinesize;
nengel@2: 	dest_cr +=   x_offset +   y_offset*d->uvlinesize;
nengel@2: 	x_offset += 8*m->mb_x;
nengel@2: 	y_offset += 8*m->mb_y;
nengel@2: 
nengel@2: 	if(list0 && list1){
nengel@2: 		/* don't optimize for luma-only case, since B-frames usually
nengel@2: 		* use implicit weights => chroma too. */
nengel@2: 		uint8_t *tmp_y  = d->scratchpad_y  + 2*x_offset +16 ;
nengel@2: 		uint8_t *tmp_cb = d->scratchpad_cb + x_offset + 8;
nengel@2: 		uint8_t *tmp_cr = d->scratchpad_cr + x_offset + 8;
nengel@2: 
nengel@2: /*
nengel@2: 		uint8_t *tmp_cb = d->scratchpad;
nengel@2: 		uint8_t *tmp_cr = d->scratchpad + 8;
nengel@2: 		uint8_t *tmp_y  = d->scratchpad + 8*d->uvlinesize;*/
nengel@2: 		int refn0 = mrs->ref_cache[0][ scan8[n] ];
nengel@2: 		int refn1 = mrs->ref_cache[1][ scan8[n] ];
nengel@2: 
nengel@2: 		mc_dir_part(d, mrs, m, s->dp_ref_list[0][refn0], n, square, chroma_height, delta, 0,
nengel@2: 					dest_y, dest_cb, dest_cr, x_offset, y_offset, qpix_put, chroma_put);
nengel@2: 		mc_dir_part(d, mrs, m, s->dp_ref_list[1][refn1], n, square, chroma_height, delta, 1,
nengel@2: 					tmp_y, tmp_cb, tmp_cr, x_offset, y_offset, qpix_put, chroma_put);
nengel@2: 
nengel@2: 		if(s->use_weight == 2){
nengel@2: 			int weight0 = s->implicit_weight[refn0][refn1][m->mb_y&1];
nengel@2: 			int weight1 = 64 - weight0;
nengel@2: 			luma_weight_avg(  dest_y,  tmp_y,  d->  linesize, 5, weight0, weight1, 0);
nengel@2: 			chroma_weight_avg(dest_cb, tmp_cb, d->uvlinesize, 5, weight0, weight1, 0);
nengel@2: 			chroma_weight_avg(dest_cr, tmp_cr, d->uvlinesize, 5, weight0, weight1, 0);
nengel@2: 		}else{
nengel@2: 			luma_weight_avg(dest_y, tmp_y, d->linesize, s->luma_log2_weight_denom,
nengel@2: 							s->luma_weight[refn0][0][0] , s->luma_weight[refn1][1][0],
nengel@2: 							s->luma_weight[refn0][0][1] + s->luma_weight[refn1][1][1]);
nengel@2: 			chroma_weight_avg(dest_cb, tmp_cb, d->uvlinesize, s->chroma_log2_weight_denom,
nengel@2: 							s->chroma_weight[refn0][0][0][0] , s->chroma_weight[refn1][1][0][0],
nengel@2: 							s->chroma_weight[refn0][0][0][1] + s->chroma_weight[refn1][1][0][1]);
nengel@2: 			chroma_weight_avg(dest_cr, tmp_cr, d->uvlinesize, s->chroma_log2_weight_denom,
nengel@2: 							s->chroma_weight[refn0][0][1][0] , s->chroma_weight[refn1][1][1][0],
nengel@2: 							s->chroma_weight[refn0][0][1][1] + s->chroma_weight[refn1][1][1][1]);
nengel@2: 		}
nengel@2: 	}else{
nengel@2: 		int list = list1 ? 1 : 0;
nengel@2: 		int refn = mrs->ref_cache[list][ scan8[n] ];
nengel@2: 		DecodedPicture *ref= s->dp_ref_list[list][refn];
nengel@2: 		mc_dir_part(d, mrs, m, ref, n, square, chroma_height, delta, list,
nengel@2: 					dest_y, dest_cb, dest_cr, x_offset, y_offset, qpix_put, chroma_put);
nengel@2: 
nengel@2: 		luma_weight_op(dest_y, d->linesize, s->luma_log2_weight_denom,
nengel@2: 						s->luma_weight[refn][list][0], s->luma_weight[refn][list][1]);
nengel@2: 		if(s->use_weight_chroma){
nengel@2: 			chroma_weight_op(dest_cb, d->uvlinesize, s->chroma_log2_weight_denom,
nengel@2: 							s->chroma_weight[refn][list][0][0], s->chroma_weight[refn][list][0][1]);
nengel@2: 			chroma_weight_op(dest_cr, d->uvlinesize, s->chroma_log2_weight_denom,
nengel@2: 							s->chroma_weight[refn][list][1][0], s->chroma_weight[refn][list][1][1]);
nengel@2: 		}
nengel@2: 	}
nengel@2: }
nengel@2: 
nengel@2: static inline void mc_part(MBRecContext *d, MBRecState *mrs, H264Slice *s, H264Mb *m, int n, int square, int chroma_height, int delta,
nengel@2: 							uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
nengel@2: 							int x_offset, int y_offset,
nengel@2: 							qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
nengel@2: 							qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
nengel@2: 							h264_weight_func *weight_op, h264_biweight_func *weight_avg,
nengel@2: 							int list0, int list1){
nengel@2: 	if((s->use_weight==2 && list0 && list1
nengel@2: 		&& (s->implicit_weight[ mrs->ref_cache[0][scan8[n]] ][ mrs->ref_cache[1][scan8[n]] ][m->mb_y&1] != 32))
nengel@2: 		|| s->use_weight==1)
nengel@2: 		mc_part_weighted(d, mrs, s, m, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
nengel@2: 						x_offset, y_offset, qpix_put, chroma_put,
nengel@2: 						weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
nengel@2: 	else
nengel@2: 		mc_part_std(d, mrs, s, m, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
nengel@2: 					x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
nengel@2: }
nengel@2: 
nengel@2: static inline void prefetch_motion(MBRecContext *d, MBRecState *mrs, H264Slice *s, H264Mb *m, int list){
nengel@2: 	/* fetch pixels for estimated mv 4 macroblocks ahead
nengel@2: 	* optimized for 64byte cache lines */
nengel@2: 	const int refn = mrs->ref_cache[list][scan8[0]];
nengel@2: 
nengel@2: 	if(refn >= 0){
nengel@2: 		const int mx= (mrs->mv_cache[list][scan8[0]][0]>>2) + 16*m->mb_x + 8;
nengel@2: 		const int my= (mrs->mv_cache[list][scan8[0]][1]>>2) + 16*m->mb_y;
nengel@2: 		uint8_t **src= s->dp_ref_list[list][refn]->data;
nengel@2: 		int off= mx + (my + (m->mb_x&3)*4)*d->linesize + 64;
nengel@2: 
nengel@2: 		d->dsp.prefetch(src[0]+off, d->linesize, 4);
nengel@2: 		off= (mx>>1) + ((my>>1) + (m->mb_x&7))*d->uvlinesize + 64;
nengel@2: 		d->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
nengel@2: 	}
nengel@2: }
nengel@2: 
nengel@2: void hl_motion(MBRecContext *d, MBRecState *mrs, H264Slice *s, H264Mb *m, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
nengel@2: 					qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
nengel@2: 					qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
nengel@2: 					h264_weight_func *weight_op, h264_biweight_func *weight_avg){
nengel@2: 	const int mb_type= m->mb_type;
nengel@2: 	assert(IS_INTER(mb_type));
nengel@2: 
nengel@2: 	if (mb_type & MB_TYPE_L0)
nengel@2: 		prefetch_motion(d, mrs, s, m, 0);
nengel@2: 	if (mb_type & MB_TYPE_L1)
nengel@2: 		prefetch_motion(d, mrs, s, m, 1);
nengel@2: 
nengel@2: 	if(IS_16X16(mb_type)){
nengel@2: 		mc_part(d, mrs, s, m, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
nengel@2: 				qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
nengel@2: 				weight_op, weight_avg,
nengel@2: 				IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
nengel@2: 	}else if(IS_16X8(mb_type)){
nengel@2: 		mc_part(d, mrs, s, m, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
nengel@2: 				qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
nengel@2: 				&weight_op[1], &weight_avg[1],
nengel@2: 				IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
nengel@2: 		mc_part(d, mrs, s, m, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
nengel@2: 				qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
nengel@2: 				&weight_op[1], &weight_avg[1],
nengel@2: 				IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
nengel@2: 	}else if(IS_8X16(mb_type)){
nengel@2: 		mc_part(d, mrs, s, m, 0, 0, 8, 8*d->linesize, dest_y, dest_cb, dest_cr, 0, 0,
nengel@2: 				qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
nengel@2: 				&weight_op[2], &weight_avg[2],
nengel@2: 				IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
nengel@2: 		mc_part(d, mrs, s, m, 4, 0, 8, 8*d->linesize, dest_y, dest_cb, dest_cr, 4, 0,
nengel@2: 				qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
nengel@2: 				&weight_op[2], &weight_avg[2],
nengel@2: 				IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
nengel@2: 	}else{
nengel@2: 		int i;
nengel@2: 
nengel@2: 		assert(IS_8X8(mb_type));
nengel@2: 
nengel@2: 		for(i=0; i<4; i++){
nengel@2: 			const int sub_mb_type= m->sub_mb_type[i];
nengel@2: 			const int n= 4*i;
nengel@2: 			int x_offset= (i&1)<<2;
nengel@2: 			int y_offset= (i&2)<<1;
nengel@2: 
nengel@2: 			if(IS_SUB_8X8(sub_mb_type)){
nengel@2: 				mc_part(d, mrs, s, m, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
nengel@2: 						qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
nengel@2: 						&weight_op[3], &weight_avg[3],
nengel@2: 						IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
nengel@2: 			}else if(IS_SUB_8X4(sub_mb_type)){
nengel@2: 				mc_part(d, mrs, s, m, n, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
nengel@2: 						qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
nengel@2: 						&weight_op[4], &weight_avg[4],
nengel@2: 						IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
nengel@2: 				mc_part(d, mrs, s, m, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
nengel@2: 						qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
nengel@2: 						&weight_op[4], &weight_avg[4],
nengel@2: 						IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
nengel@2: 			}else if(IS_SUB_4X8(sub_mb_type)){
nengel@2: 				mc_part(d, mrs, s, m, n, 0, 4, 4*d->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
nengel@2: 						qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
nengel@2: 						&weight_op[5], &weight_avg[5],
nengel@2: 						IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
nengel@2: 				mc_part(d, mrs, s, m, n+1, 0, 4, 4*d->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
nengel@2: 						qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
nengel@2: 						&weight_op[5], &weight_avg[5],
nengel@2: 						IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
nengel@2: 			}else{
nengel@2: 				int j;
nengel@2: 				assert(IS_SUB_4X4(sub_mb_type));
nengel@2: 				for(j=0; j<4; j++){
nengel@2: 					int sub_x_offset= x_offset + 2*(j&1);
nengel@2: 					int sub_y_offset= y_offset +   (j&2);
nengel@2: 					mc_part(d, mrs, s, m, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
nengel@2: 							qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
nengel@2: 							&weight_op[6], &weight_avg[6],
nengel@2: 							IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
nengel@2: 				}
nengel@2: 			}
nengel@2: 		}
nengel@2: 	}
nengel@2: }