2 static inline void FUNC(mc_dir_part)(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
3 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
4 int src_x_offset, int src_y_offset,
5 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
6 MpegEncContext * const s = &h->s;
7 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
8 int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
9 const int luma_xy= (mx&3) + ((my&3)<<2);
10 uint8_t * src_y = pic->data[0] + ((mx>>2)<<PIXEL_SHIFT) + (my>>2)*h->mb_linesize;
11 uint8_t * src_cb, * src_cr;
12 int extra_width= h->emu_edge_width;
13 int extra_height= h->emu_edge_height;
15 const int full_mx= mx>>2;
16 const int full_my= my>>2;
17 const int pic_width = 16*s->mb_width;
18 const int pic_height = 16*s->mb_height >> MB_FIELD;
20 if(mx&7) extra_width -= 3;
21 if(my&7) extra_height -= 3;
23 if( full_mx < 0-extra_width
24 || full_my < 0-extra_height
25 || full_mx + 16/*FIXME*/ > pic_width + extra_width
26 || full_my + 16/*FIXME*/ > pic_height + extra_height){
27 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2<<PIXEL_SHIFT) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
28 src_y= s->edge_emu_buffer + (2<<PIXEL_SHIFT) + 2*h->mb_linesize;
32 qpix_op[luma_xy](dest_y, src_y, h->mb_linesize); //FIXME try variable height perhaps?
34 qpix_op[luma_xy](dest_y + delta, src_y + delta, h->mb_linesize);
37 if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
40 // chroma offset when predicting from a field of opposite parity
41 my += 2 * ((s->mb_y & 1) - (pic->reference - 1));
42 emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
44 src_cb= pic->data[1] + ((mx>>3)<<PIXEL_SHIFT) + (my>>3)*h->mb_uvlinesize;
45 src_cr= pic->data[2] + ((mx>>3)<<PIXEL_SHIFT) + (my>>3)*h->mb_uvlinesize;
48 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
49 src_cb= s->edge_emu_buffer;
51 chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
54 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
55 src_cr= s->edge_emu_buffer;
57 chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
60 static inline void FUNC(mc_part_std)(H264Context *h, int n, int square, int chroma_height, int delta,
61 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
62 int x_offset, int y_offset,
63 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
64 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
65 int list0, int list1){
66 MpegEncContext * const s = &h->s;
67 qpel_mc_func *qpix_op= qpix_put;
68 h264_chroma_mc_func chroma_op= chroma_put;
70 dest_y += (2*x_offset<<PIXEL_SHIFT) + 2*y_offset*h-> mb_linesize;
71 dest_cb += ( x_offset<<PIXEL_SHIFT) + y_offset*h->mb_uvlinesize;
72 dest_cr += ( x_offset<<PIXEL_SHIFT) + y_offset*h->mb_uvlinesize;
73 x_offset += 8*s->mb_x;
74 y_offset += 8*(s->mb_y >> MB_FIELD);
77 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
78 FUNC(mc_dir_part)(h, ref, n, square, chroma_height, delta, 0,
79 dest_y, dest_cb, dest_cr, x_offset, y_offset,
83 chroma_op= chroma_avg;
87 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
88 FUNC(mc_dir_part)(h, ref, n, square, chroma_height, delta, 1,
89 dest_y, dest_cb, dest_cr, x_offset, y_offset,
94 static inline void FUNC(mc_part_weighted)(H264Context *h, int n, int square, int chroma_height, int delta,
95 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
96 int x_offset, int y_offset,
97 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
98 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
99 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
100 int list0, int list1){
101 MpegEncContext * const s = &h->s;
103 dest_y += (2*x_offset<<PIXEL_SHIFT) + 2*y_offset*h-> mb_linesize;
104 dest_cb += ( x_offset<<PIXEL_SHIFT) + y_offset*h->mb_uvlinesize;
105 dest_cr += ( x_offset<<PIXEL_SHIFT) + y_offset*h->mb_uvlinesize;
106 x_offset += 8*s->mb_x;
107 y_offset += 8*(s->mb_y >> MB_FIELD);
110 /* don't optimize for luma-only case, since B-frames usually
111 * use implicit weights => chroma too. */
112 uint8_t *tmp_cb = s->obmc_scratchpad;
113 uint8_t *tmp_cr = s->obmc_scratchpad + (8<<PIXEL_SHIFT);
114 uint8_t *tmp_y = s->obmc_scratchpad + 8*h->mb_uvlinesize;
115 int refn0 = h->ref_cache[0][ scan8[n] ];
116 int refn1 = h->ref_cache[1][ scan8[n] ];
118 FUNC(mc_dir_part)(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
119 dest_y, dest_cb, dest_cr,
120 x_offset, y_offset, qpix_put, chroma_put);
121 FUNC(mc_dir_part)(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
122 tmp_y, tmp_cb, tmp_cr,
123 x_offset, y_offset, qpix_put, chroma_put);
125 if(h->use_weight == 2){
126 int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
127 int weight1 = 64 - weight0;
128 luma_weight_avg( dest_y, tmp_y, h-> mb_linesize, 5, weight0, weight1, 0);
129 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
130 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
132 luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
133 h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
134 h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
135 chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
136 h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
137 h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
138 chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
139 h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
140 h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
143 int list = list1 ? 1 : 0;
144 int refn = h->ref_cache[list][ scan8[n] ];
145 Picture *ref= &h->ref_list[list][refn];
146 FUNC(mc_dir_part)(h, ref, n, square, chroma_height, delta, list,
147 dest_y, dest_cb, dest_cr, x_offset, y_offset,
148 qpix_put, chroma_put);
150 luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
151 h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
152 if(h->use_weight_chroma){
153 chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
154 h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
155 chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
156 h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
161 static inline void FUNC(mc_part)(H264Context *h, int n, int square, int chroma_height, int delta,
162 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
163 int x_offset, int y_offset,
164 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
165 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
166 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
167 int list0, int list1){
168 if((h->use_weight==2 && list0 && list1
169 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
171 FUNC(mc_part_weighted)(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
172 x_offset, y_offset, qpix_put, chroma_put,
173 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
175 FUNC(mc_part_std)(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
176 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
179 static inline void FUNC(prefetch_motion)(H264Context *h, int list){
180 /* fetch pixels for estimated mv 4 macroblocks ahead
181 * optimized for 64byte cache lines */
182 MpegEncContext * const s = &h->s;
183 const int refn = h->ref_cache[list][scan8[0]];
185 const int mx= (h->mv_cache[list][scan8[0]][0]>>2) + 16*s->mb_x + 8;
186 const int my= (h->mv_cache[list][scan8[0]][1]>>2) + 16*s->mb_y;
187 uint8_t **src= h->ref_list[list][refn].data;
188 int off= ((mx+64)<<PIXEL_SHIFT) + (my + (s->mb_x&3)*4)*h->mb_linesize;
189 s->dsp.prefetch(src[0]+off, s->linesize, 4);
190 off= (((mx>>1)+64)<<PIXEL_SHIFT) + ((my>>1) + (s->mb_x&7))*s->uvlinesize;
191 s->dsp.prefetch(src[1]+off, src[2]-src[1], 2);
195 static void FUNC(hl_motion)(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
196 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
197 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
198 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
199 MpegEncContext * const s = &h->s;
200 const int mb_xy= h->mb_xy;
201 const int mb_type= s->current_picture.mb_type[mb_xy];
203 assert(IS_INTER(mb_type));
205 if(HAVE_PTHREADS && s->avctx->active_thread_type&FF_THREAD_FRAME)
207 FUNC(prefetch_motion)(h, 0);
209 if(IS_16X16(mb_type)){
210 FUNC(mc_part)(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
211 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
212 weight_op, weight_avg,
213 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
214 }else if(IS_16X8(mb_type)){
215 FUNC(mc_part)(h, 0, 0, 4, (8<<PIXEL_SHIFT), dest_y, dest_cb, dest_cr, 0, 0,
216 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
217 &weight_op[1], &weight_avg[1],
218 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
219 FUNC(mc_part)(h, 8, 0, 4, (8<<PIXEL_SHIFT), dest_y, dest_cb, dest_cr, 0, 4,
220 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
221 &weight_op[1], &weight_avg[1],
222 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
223 }else if(IS_8X16(mb_type)){
224 FUNC(mc_part)(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
225 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
226 &weight_op[2], &weight_avg[2],
227 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
228 FUNC(mc_part)(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
229 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
230 &weight_op[2], &weight_avg[2],
231 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
235 assert(IS_8X8(mb_type));
238 const int sub_mb_type= h->sub_mb_type[i];
240 int x_offset= (i&1)<<2;
241 int y_offset= (i&2)<<1;
243 if(IS_SUB_8X8(sub_mb_type)){
244 FUNC(mc_part)(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
245 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
246 &weight_op[3], &weight_avg[3],
247 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
248 }else if(IS_SUB_8X4(sub_mb_type)){
249 FUNC(mc_part)(h, n , 0, 2, (4<<PIXEL_SHIFT), dest_y, dest_cb, dest_cr, x_offset, y_offset,
250 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
251 &weight_op[4], &weight_avg[4],
252 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
253 FUNC(mc_part)(h, n+2, 0, 2, (4<<PIXEL_SHIFT), dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
254 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
255 &weight_op[4], &weight_avg[4],
256 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
257 }else if(IS_SUB_4X8(sub_mb_type)){
258 FUNC(mc_part)(h, n , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
259 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
260 &weight_op[5], &weight_avg[5],
261 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
262 FUNC(mc_part)(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
263 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
264 &weight_op[5], &weight_avg[5],
265 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
268 assert(IS_SUB_4X4(sub_mb_type));
270 int sub_x_offset= x_offset + 2*(j&1);
271 int sub_y_offset= y_offset + (j&2);
272 FUNC(mc_part)(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
273 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
274 &weight_op[6], &weight_avg[6],
275 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
281 FUNC(prefetch_motion)(h, 1);