1 /*
2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <limits.h>
12 #include <string.h>
13
14 #include "vpx_config.h"
15 #include "vp8_rtcd.h"
16 #include "vpx/vpx_integer.h"
17 #include "blockd.h"
18 #include "reconinter.h"
19 #if CONFIG_RUNTIME_CPU_DETECT
20 #include "onyxc_int.h"
21 #endif
22
vp8_copy_mem16x16_c(unsigned char * src,int src_stride,unsigned char * dst,int dst_stride)23 void vp8_copy_mem16x16_c(unsigned char *src, int src_stride, unsigned char *dst,
24 int dst_stride) {
25 int r;
26
27 for (r = 0; r < 16; ++r) {
28 memcpy(dst, src, 16);
29
30 src += src_stride;
31 dst += dst_stride;
32 }
33 }
34
vp8_copy_mem8x8_c(unsigned char * src,int src_stride,unsigned char * dst,int dst_stride)35 void vp8_copy_mem8x8_c(unsigned char *src, int src_stride, unsigned char *dst,
36 int dst_stride) {
37 int r;
38
39 for (r = 0; r < 8; ++r) {
40 memcpy(dst, src, 8);
41
42 src += src_stride;
43 dst += dst_stride;
44 }
45 }
46
vp8_copy_mem8x4_c(unsigned char * src,int src_stride,unsigned char * dst,int dst_stride)47 void vp8_copy_mem8x4_c(unsigned char *src, int src_stride, unsigned char *dst,
48 int dst_stride) {
49 int r;
50
51 for (r = 0; r < 4; ++r) {
52 memcpy(dst, src, 8);
53
54 src += src_stride;
55 dst += dst_stride;
56 }
57 }
58
vp8_build_inter_predictors_b(BLOCKD * d,int pitch,unsigned char * base_pre,int pre_stride,vp8_subpix_fn_t sppf)59 void vp8_build_inter_predictors_b(BLOCKD *d, int pitch, unsigned char *base_pre,
60 int pre_stride, vp8_subpix_fn_t sppf) {
61 int r;
62 unsigned char *pred_ptr = d->predictor;
63 unsigned char *ptr;
64 ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
65 (d->bmi.mv.as_mv.col >> 3);
66
67 if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
68 sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7,
69 pred_ptr, pitch);
70 } else {
71 for (r = 0; r < 4; ++r) {
72 pred_ptr[0] = ptr[0];
73 pred_ptr[1] = ptr[1];
74 pred_ptr[2] = ptr[2];
75 pred_ptr[3] = ptr[3];
76 pred_ptr += pitch;
77 ptr += pre_stride;
78 }
79 }
80 }
81
build_inter_predictors4b(MACROBLOCKD * x,BLOCKD * d,unsigned char * dst,int dst_stride,unsigned char * base_pre,int pre_stride)82 static void build_inter_predictors4b(MACROBLOCKD *x, BLOCKD *d,
83 unsigned char *dst, int dst_stride,
84 unsigned char *base_pre, int pre_stride) {
85 unsigned char *ptr;
86 ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
87 (d->bmi.mv.as_mv.col >> 3);
88
89 if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
90 x->subpixel_predict8x8(ptr, pre_stride, d->bmi.mv.as_mv.col & 7,
91 d->bmi.mv.as_mv.row & 7, dst, dst_stride);
92 } else {
93 vp8_copy_mem8x8(ptr, pre_stride, dst, dst_stride);
94 }
95 }
96
build_inter_predictors2b(MACROBLOCKD * x,BLOCKD * d,unsigned char * dst,int dst_stride,unsigned char * base_pre,int pre_stride)97 static void build_inter_predictors2b(MACROBLOCKD *x, BLOCKD *d,
98 unsigned char *dst, int dst_stride,
99 unsigned char *base_pre, int pre_stride) {
100 unsigned char *ptr;
101 ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
102 (d->bmi.mv.as_mv.col >> 3);
103
104 if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
105 x->subpixel_predict8x4(ptr, pre_stride, d->bmi.mv.as_mv.col & 7,
106 d->bmi.mv.as_mv.row & 7, dst, dst_stride);
107 } else {
108 vp8_copy_mem8x4(ptr, pre_stride, dst, dst_stride);
109 }
110 }
111
build_inter_predictors_b(BLOCKD * d,unsigned char * dst,int dst_stride,unsigned char * base_pre,int pre_stride,vp8_subpix_fn_t sppf)112 static void build_inter_predictors_b(BLOCKD *d, unsigned char *dst,
113 int dst_stride, unsigned char *base_pre,
114 int pre_stride, vp8_subpix_fn_t sppf) {
115 int r;
116 unsigned char *ptr;
117 ptr = base_pre + d->offset + (d->bmi.mv.as_mv.row >> 3) * pre_stride +
118 (d->bmi.mv.as_mv.col >> 3);
119
120 if (d->bmi.mv.as_mv.row & 7 || d->bmi.mv.as_mv.col & 7) {
121 sppf(ptr, pre_stride, d->bmi.mv.as_mv.col & 7, d->bmi.mv.as_mv.row & 7, dst,
122 dst_stride);
123 } else {
124 for (r = 0; r < 4; ++r) {
125 dst[0] = ptr[0];
126 dst[1] = ptr[1];
127 dst[2] = ptr[2];
128 dst[3] = ptr[3];
129 dst += dst_stride;
130 ptr += pre_stride;
131 }
132 }
133 }
134
135 /*encoder only*/
vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD * x)136 void vp8_build_inter16x16_predictors_mbuv(MACROBLOCKD *x) {
137 unsigned char *uptr, *vptr;
138 unsigned char *upred_ptr = &x->predictor[256];
139 unsigned char *vpred_ptr = &x->predictor[320];
140
141 int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
142 int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
143 int offset;
144 int pre_stride = x->pre.uv_stride;
145
146 /* calc uv motion vectors */
147 mv_row += 1 | (mv_row >> (sizeof(int) * CHAR_BIT - 1));
148 mv_col += 1 | (mv_col >> (sizeof(int) * CHAR_BIT - 1));
149 mv_row /= 2;
150 mv_col /= 2;
151 mv_row &= x->fullpixel_mask;
152 mv_col &= x->fullpixel_mask;
153
154 offset = (mv_row >> 3) * pre_stride + (mv_col >> 3);
155 uptr = x->pre.u_buffer + offset;
156 vptr = x->pre.v_buffer + offset;
157
158 if ((mv_row | mv_col) & 7) {
159 x->subpixel_predict8x8(uptr, pre_stride, mv_col & 7, mv_row & 7, upred_ptr,
160 8);
161 x->subpixel_predict8x8(vptr, pre_stride, mv_col & 7, mv_row & 7, vpred_ptr,
162 8);
163 } else {
164 vp8_copy_mem8x8(uptr, pre_stride, upred_ptr, 8);
165 vp8_copy_mem8x8(vptr, pre_stride, vpred_ptr, 8);
166 }
167 }
168
169 /*encoder only*/
vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD * x)170 void vp8_build_inter4x4_predictors_mbuv(MACROBLOCKD *x) {
171 int i, j;
172 int pre_stride = x->pre.uv_stride;
173 unsigned char *base_pre;
174
175 /* build uv mvs */
176 for (i = 0; i < 2; ++i) {
177 for (j = 0; j < 2; ++j) {
178 int yoffset = i * 8 + j * 2;
179 int uoffset = 16 + i * 2 + j;
180 int voffset = 20 + i * 2 + j;
181
182 int temp;
183
184 temp = x->block[yoffset].bmi.mv.as_mv.row +
185 x->block[yoffset + 1].bmi.mv.as_mv.row +
186 x->block[yoffset + 4].bmi.mv.as_mv.row +
187 x->block[yoffset + 5].bmi.mv.as_mv.row;
188
189 temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
190
191 x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
192
193 temp = x->block[yoffset].bmi.mv.as_mv.col +
194 x->block[yoffset + 1].bmi.mv.as_mv.col +
195 x->block[yoffset + 4].bmi.mv.as_mv.col +
196 x->block[yoffset + 5].bmi.mv.as_mv.col;
197
198 temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
199
200 x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
201
202 x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
203 }
204 }
205
206 base_pre = x->pre.u_buffer;
207 for (i = 16; i < 20; i += 2) {
208 BLOCKD *d0 = &x->block[i];
209 BLOCKD *d1 = &x->block[i + 1];
210
211 if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
212 build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
213 } else {
214 vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride,
215 x->subpixel_predict);
216 vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride,
217 x->subpixel_predict);
218 }
219 }
220
221 base_pre = x->pre.v_buffer;
222 for (i = 20; i < 24; i += 2) {
223 BLOCKD *d0 = &x->block[i];
224 BLOCKD *d1 = &x->block[i + 1];
225
226 if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
227 build_inter_predictors2b(x, d0, d0->predictor, 8, base_pre, pre_stride);
228 } else {
229 vp8_build_inter_predictors_b(d0, 8, base_pre, pre_stride,
230 x->subpixel_predict);
231 vp8_build_inter_predictors_b(d1, 8, base_pre, pre_stride,
232 x->subpixel_predict);
233 }
234 }
235 }
236
237 /*encoder only*/
vp8_build_inter16x16_predictors_mby(MACROBLOCKD * x,unsigned char * dst_y,int dst_ystride)238 void vp8_build_inter16x16_predictors_mby(MACROBLOCKD *x, unsigned char *dst_y,
239 int dst_ystride) {
240 unsigned char *ptr_base;
241 unsigned char *ptr;
242 int mv_row = x->mode_info_context->mbmi.mv.as_mv.row;
243 int mv_col = x->mode_info_context->mbmi.mv.as_mv.col;
244 int pre_stride = x->pre.y_stride;
245
246 ptr_base = x->pre.y_buffer;
247 ptr = ptr_base + (mv_row >> 3) * pre_stride + (mv_col >> 3);
248
249 if ((mv_row | mv_col) & 7) {
250 x->subpixel_predict16x16(ptr, pre_stride, mv_col & 7, mv_row & 7, dst_y,
251 dst_ystride);
252 } else {
253 vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
254 }
255 }
256
clamp_mv_to_umv_border(MV * mv,const MACROBLOCKD * xd)257 static void clamp_mv_to_umv_border(MV *mv, const MACROBLOCKD *xd) {
258 /* If the MV points so far into the UMV border that no visible pixels
259 * are used for reconstruction, the subpel part of the MV can be
260 * discarded and the MV limited to 16 pixels with equivalent results.
261 *
262 * This limit kicks in at 19 pixels for the top and left edges, for
263 * the 16 pixels plus 3 taps right of the central pixel when subpel
264 * filtering. The bottom and right edges use 16 pixels plus 2 pixels
265 * left of the central pixel when filtering.
266 */
267 if (mv->col < (xd->mb_to_left_edge - (19 << 3))) {
268 mv->col = xd->mb_to_left_edge - (16 << 3);
269 } else if (mv->col > xd->mb_to_right_edge + (18 << 3)) {
270 mv->col = xd->mb_to_right_edge + (16 << 3);
271 }
272
273 if (mv->row < (xd->mb_to_top_edge - (19 << 3))) {
274 mv->row = xd->mb_to_top_edge - (16 << 3);
275 } else if (mv->row > xd->mb_to_bottom_edge + (18 << 3)) {
276 mv->row = xd->mb_to_bottom_edge + (16 << 3);
277 }
278 }
279
280 /* A version of the above function for chroma block MVs.*/
clamp_uvmv_to_umv_border(MV * mv,const MACROBLOCKD * xd)281 static void clamp_uvmv_to_umv_border(MV *mv, const MACROBLOCKD *xd) {
282 mv->col = (2 * mv->col < (xd->mb_to_left_edge - (19 << 3)))
283 ? (xd->mb_to_left_edge - (16 << 3)) >> 1
284 : mv->col;
285 mv->col = (2 * mv->col > xd->mb_to_right_edge + (18 << 3))
286 ? (xd->mb_to_right_edge + (16 << 3)) >> 1
287 : mv->col;
288
289 mv->row = (2 * mv->row < (xd->mb_to_top_edge - (19 << 3)))
290 ? (xd->mb_to_top_edge - (16 << 3)) >> 1
291 : mv->row;
292 mv->row = (2 * mv->row > xd->mb_to_bottom_edge + (18 << 3))
293 ? (xd->mb_to_bottom_edge + (16 << 3)) >> 1
294 : mv->row;
295 }
296
vp8_build_inter16x16_predictors_mb(MACROBLOCKD * x,unsigned char * dst_y,unsigned char * dst_u,unsigned char * dst_v,int dst_ystride,int dst_uvstride)297 void vp8_build_inter16x16_predictors_mb(MACROBLOCKD *x, unsigned char *dst_y,
298 unsigned char *dst_u,
299 unsigned char *dst_v, int dst_ystride,
300 int dst_uvstride) {
301 int offset;
302 unsigned char *ptr;
303 unsigned char *uptr, *vptr;
304
305 int_mv _16x16mv;
306
307 unsigned char *ptr_base = x->pre.y_buffer;
308 int pre_stride = x->pre.y_stride;
309
310 _16x16mv.as_int = x->mode_info_context->mbmi.mv.as_int;
311
312 if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
313 clamp_mv_to_umv_border(&_16x16mv.as_mv, x);
314 }
315
316 ptr = ptr_base + (_16x16mv.as_mv.row >> 3) * pre_stride +
317 (_16x16mv.as_mv.col >> 3);
318
319 if (_16x16mv.as_int & 0x00070007) {
320 x->subpixel_predict16x16(ptr, pre_stride, _16x16mv.as_mv.col & 7,
321 _16x16mv.as_mv.row & 7, dst_y, dst_ystride);
322 } else {
323 vp8_copy_mem16x16(ptr, pre_stride, dst_y, dst_ystride);
324 }
325
326 /* calc uv motion vectors */
327 _16x16mv.as_mv.row +=
328 1 | (_16x16mv.as_mv.row >> (sizeof(int) * CHAR_BIT - 1));
329 _16x16mv.as_mv.col +=
330 1 | (_16x16mv.as_mv.col >> (sizeof(int) * CHAR_BIT - 1));
331 _16x16mv.as_mv.row /= 2;
332 _16x16mv.as_mv.col /= 2;
333 _16x16mv.as_mv.row &= x->fullpixel_mask;
334 _16x16mv.as_mv.col &= x->fullpixel_mask;
335
336 pre_stride >>= 1;
337 offset = (_16x16mv.as_mv.row >> 3) * pre_stride + (_16x16mv.as_mv.col >> 3);
338 uptr = x->pre.u_buffer + offset;
339 vptr = x->pre.v_buffer + offset;
340
341 if (_16x16mv.as_int & 0x00070007) {
342 x->subpixel_predict8x8(uptr, pre_stride, _16x16mv.as_mv.col & 7,
343 _16x16mv.as_mv.row & 7, dst_u, dst_uvstride);
344 x->subpixel_predict8x8(vptr, pre_stride, _16x16mv.as_mv.col & 7,
345 _16x16mv.as_mv.row & 7, dst_v, dst_uvstride);
346 } else {
347 vp8_copy_mem8x8(uptr, pre_stride, dst_u, dst_uvstride);
348 vp8_copy_mem8x8(vptr, pre_stride, dst_v, dst_uvstride);
349 }
350 }
351
build_inter4x4_predictors_mb(MACROBLOCKD * x)352 static void build_inter4x4_predictors_mb(MACROBLOCKD *x) {
353 int i;
354 unsigned char *base_dst = x->dst.y_buffer;
355 unsigned char *base_pre = x->pre.y_buffer;
356
357 if (x->mode_info_context->mbmi.partitioning < 3) {
358 BLOCKD *b;
359 int dst_stride = x->dst.y_stride;
360
361 x->block[0].bmi = x->mode_info_context->bmi[0];
362 x->block[2].bmi = x->mode_info_context->bmi[2];
363 x->block[8].bmi = x->mode_info_context->bmi[8];
364 x->block[10].bmi = x->mode_info_context->bmi[10];
365 if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
366 clamp_mv_to_umv_border(&x->block[0].bmi.mv.as_mv, x);
367 clamp_mv_to_umv_border(&x->block[2].bmi.mv.as_mv, x);
368 clamp_mv_to_umv_border(&x->block[8].bmi.mv.as_mv, x);
369 clamp_mv_to_umv_border(&x->block[10].bmi.mv.as_mv, x);
370 }
371
372 b = &x->block[0];
373 build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
374 dst_stride);
375 b = &x->block[2];
376 build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
377 dst_stride);
378 b = &x->block[8];
379 build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
380 dst_stride);
381 b = &x->block[10];
382 build_inter_predictors4b(x, b, base_dst + b->offset, dst_stride, base_pre,
383 dst_stride);
384 } else {
385 for (i = 0; i < 16; i += 2) {
386 BLOCKD *d0 = &x->block[i];
387 BLOCKD *d1 = &x->block[i + 1];
388 int dst_stride = x->dst.y_stride;
389
390 x->block[i + 0].bmi = x->mode_info_context->bmi[i + 0];
391 x->block[i + 1].bmi = x->mode_info_context->bmi[i + 1];
392 if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
393 clamp_mv_to_umv_border(&x->block[i + 0].bmi.mv.as_mv, x);
394 clamp_mv_to_umv_border(&x->block[i + 1].bmi.mv.as_mv, x);
395 }
396
397 if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
398 build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
399 base_pre, dst_stride);
400 } else {
401 build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride,
402 base_pre, dst_stride, x->subpixel_predict);
403 build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride,
404 base_pre, dst_stride, x->subpixel_predict);
405 }
406 }
407 }
408 base_dst = x->dst.u_buffer;
409 base_pre = x->pre.u_buffer;
410 for (i = 16; i < 20; i += 2) {
411 BLOCKD *d0 = &x->block[i];
412 BLOCKD *d1 = &x->block[i + 1];
413 int dst_stride = x->dst.uv_stride;
414
415 /* Note: uv mvs already clamped in build_4x4uvmvs() */
416
417 if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
418 build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
419 base_pre, dst_stride);
420 } else {
421 build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre,
422 dst_stride, x->subpixel_predict);
423 build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre,
424 dst_stride, x->subpixel_predict);
425 }
426 }
427
428 base_dst = x->dst.v_buffer;
429 base_pre = x->pre.v_buffer;
430 for (i = 20; i < 24; i += 2) {
431 BLOCKD *d0 = &x->block[i];
432 BLOCKD *d1 = &x->block[i + 1];
433 int dst_stride = x->dst.uv_stride;
434
435 /* Note: uv mvs already clamped in build_4x4uvmvs() */
436
437 if (d0->bmi.mv.as_int == d1->bmi.mv.as_int) {
438 build_inter_predictors2b(x, d0, base_dst + d0->offset, dst_stride,
439 base_pre, dst_stride);
440 } else {
441 build_inter_predictors_b(d0, base_dst + d0->offset, dst_stride, base_pre,
442 dst_stride, x->subpixel_predict);
443 build_inter_predictors_b(d1, base_dst + d1->offset, dst_stride, base_pre,
444 dst_stride, x->subpixel_predict);
445 }
446 }
447 }
448
build_4x4uvmvs(MACROBLOCKD * x)449 static void build_4x4uvmvs(MACROBLOCKD *x) {
450 int i, j;
451
452 for (i = 0; i < 2; ++i) {
453 for (j = 0; j < 2; ++j) {
454 int yoffset = i * 8 + j * 2;
455 int uoffset = 16 + i * 2 + j;
456 int voffset = 20 + i * 2 + j;
457
458 int temp;
459
460 temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.row +
461 x->mode_info_context->bmi[yoffset + 1].mv.as_mv.row +
462 x->mode_info_context->bmi[yoffset + 4].mv.as_mv.row +
463 x->mode_info_context->bmi[yoffset + 5].mv.as_mv.row;
464
465 temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
466
467 x->block[uoffset].bmi.mv.as_mv.row = (temp / 8) & x->fullpixel_mask;
468
469 temp = x->mode_info_context->bmi[yoffset + 0].mv.as_mv.col +
470 x->mode_info_context->bmi[yoffset + 1].mv.as_mv.col +
471 x->mode_info_context->bmi[yoffset + 4].mv.as_mv.col +
472 x->mode_info_context->bmi[yoffset + 5].mv.as_mv.col;
473
474 temp += 4 + ((temp >> (sizeof(temp) * CHAR_BIT - 1)) * 8);
475
476 x->block[uoffset].bmi.mv.as_mv.col = (temp / 8) & x->fullpixel_mask;
477
478 if (x->mode_info_context->mbmi.need_to_clamp_mvs) {
479 clamp_uvmv_to_umv_border(&x->block[uoffset].bmi.mv.as_mv, x);
480 }
481
482 x->block[voffset].bmi.mv.as_int = x->block[uoffset].bmi.mv.as_int;
483 }
484 }
485 }
486
vp8_build_inter_predictors_mb(MACROBLOCKD * xd)487 void vp8_build_inter_predictors_mb(MACROBLOCKD *xd) {
488 if (xd->mode_info_context->mbmi.mode != SPLITMV) {
489 vp8_build_inter16x16_predictors_mb(xd, xd->dst.y_buffer, xd->dst.u_buffer,
490 xd->dst.v_buffer, xd->dst.y_stride,
491 xd->dst.uv_stride);
492 } else {
493 build_4x4uvmvs(xd);
494 build_inter4x4_predictors_mb(xd);
495 }
496 }
497