2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
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.
11 #include "./vpx_config.h"
13 #include "vpx_mem/vpx_mem.h"
14 #include "vpx_ports/vpx_once.h"
16 #include "./vp9_rtcd.h"
18 #include "vp9/common/vp9_reconintra.h"
19 #include "vp9/common/vp9_onyxc_int.h"
21 const TX_TYPE intra_mode_to_tx_type_lookup[INTRA_MODES] = {
34 #define intra_pred_sized(type, size) \
35 void vp9_##type##_predictor_##size##x##size##_c(uint8_t *dst, \
37 const uint8_t *above, \
38 const uint8_t *left) { \
39 type##_predictor(dst, stride, size, above, left); \
42 #define intra_pred_allsizes(type) \
43 intra_pred_sized(type, 4) \
44 intra_pred_sized(type, 8) \
45 intra_pred_sized(type, 16) \
46 intra_pred_sized(type, 32)
48 static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
49 const uint8_t *above, const uint8_t *left) {
53 for (r = 0; r < bs - 1; ++r)
54 dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
55 dst[(bs - 1) * stride] = left[bs - 1];
59 for (r = 0; r < bs - 2; ++r)
60 dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 +
62 dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
64 dst[(bs - 1) * stride] = left[bs - 1];
68 for (c = 0; c < bs - 2; ++c)
69 dst[(bs - 1) * stride + c] = left[bs - 1];
71 for (r = bs - 2; r >= 0; --r)
72 for (c = 0; c < bs - 2; ++c)
73 dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
75 intra_pred_allsizes(d207)
77 static INLINE void d63_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
78 const uint8_t *above, const uint8_t *left) {
80 for (r = 0; r < bs; ++r) {
81 for (c = 0; c < bs; ++c)
82 dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] +
83 above[r/2 + c + 1] * 2 +
84 above[r/2 + c + 2], 2)
85 : ROUND_POWER_OF_TWO(above[r/2 + c] +
86 above[r/2 + c + 1], 1);
90 intra_pred_allsizes(d63)
92 static INLINE void d45_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
93 const uint8_t *above, const uint8_t *left) {
95 for (r = 0; r < bs; ++r) {
96 for (c = 0; c < bs; ++c)
97 dst[c] = r + c + 2 < bs * 2 ? ROUND_POWER_OF_TWO(above[r + c] +
98 above[r + c + 1] * 2 +
104 intra_pred_allsizes(d45)
106 static INLINE void d117_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
107 const uint8_t *above, const uint8_t *left) {
111 for (c = 0; c < bs; c++)
112 dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
116 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
117 for (c = 1; c < bs; c++)
118 dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
121 // the rest of first col
122 dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
123 for (r = 3; r < bs; ++r)
124 dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 +
127 // the rest of the block
128 for (r = 2; r < bs; ++r) {
129 for (c = 1; c < bs; c++)
130 dst[c] = dst[-2 * stride + c - 1];
134 intra_pred_allsizes(d117)
136 static INLINE void d135_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
137 const uint8_t *above, const uint8_t *left) {
139 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
140 for (c = 1; c < bs; c++)
141 dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
143 dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
144 for (r = 2; r < bs; ++r)
145 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
149 for (r = 1; r < bs; ++r) {
150 for (c = 1; c < bs; c++)
151 dst[c] = dst[-stride + c - 1];
155 intra_pred_allsizes(d135)
157 static INLINE void d153_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
158 const uint8_t *above, const uint8_t *left) {
160 dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
161 for (r = 1; r < bs; r++)
162 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
165 dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
166 dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
167 for (r = 2; r < bs; r++)
168 dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
172 for (c = 0; c < bs - 2; c++)
173 dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2);
176 for (r = 1; r < bs; ++r) {
177 for (c = 0; c < bs - 2; c++)
178 dst[c] = dst[-stride + c - 2];
182 intra_pred_allsizes(d153)
184 static INLINE void v_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
185 const uint8_t *above, const uint8_t *left) {
188 for (r = 0; r < bs; r++) {
189 vpx_memcpy(dst, above, bs);
193 intra_pred_allsizes(v)
195 static INLINE void h_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
196 const uint8_t *above, const uint8_t *left) {
199 for (r = 0; r < bs; r++) {
200 vpx_memset(dst, left[r], bs);
204 intra_pred_allsizes(h)
206 static INLINE void tm_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
207 const uint8_t *above, const uint8_t *left) {
209 int ytop_left = above[-1];
211 for (r = 0; r < bs; r++) {
212 for (c = 0; c < bs; c++)
213 dst[c] = clip_pixel(left[r] + above[c] - ytop_left);
217 intra_pred_allsizes(tm)
219 static INLINE void dc_128_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
220 const uint8_t *above, const uint8_t *left) {
223 for (r = 0; r < bs; r++) {
224 vpx_memset(dst, 128, bs);
228 intra_pred_allsizes(dc_128)
230 static INLINE void dc_left_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
231 const uint8_t *above,
232 const uint8_t *left) {
233 int i, r, expected_dc, sum = 0;
235 for (i = 0; i < bs; i++)
237 expected_dc = (sum + (bs >> 1)) / bs;
239 for (r = 0; r < bs; r++) {
240 vpx_memset(dst, expected_dc, bs);
244 intra_pred_allsizes(dc_left)
246 static INLINE void dc_top_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
247 const uint8_t *above, const uint8_t *left) {
248 int i, r, expected_dc, sum = 0;
250 for (i = 0; i < bs; i++)
252 expected_dc = (sum + (bs >> 1)) / bs;
254 for (r = 0; r < bs; r++) {
255 vpx_memset(dst, expected_dc, bs);
259 intra_pred_allsizes(dc_top)
261 static INLINE void dc_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
262 const uint8_t *above, const uint8_t *left) {
263 int i, r, expected_dc, sum = 0;
264 const int count = 2 * bs;
266 for (i = 0; i < bs; i++) {
271 expected_dc = (sum + (count >> 1)) / count;
273 for (r = 0; r < bs; r++) {
274 vpx_memset(dst, expected_dc, bs);
278 intra_pred_allsizes(dc)
279 #undef intra_pred_allsizes
281 typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride,
282 const uint8_t *above, const uint8_t *left);
284 static intra_pred_fn pred[INTRA_MODES][4];
285 static intra_pred_fn dc_pred[2][2][4];
287 static void init_intra_pred_fn_ptrs(void) {
288 #define intra_pred_allsizes(l, type) \
289 l[0] = vp9_##type##_predictor_4x4; \
290 l[1] = vp9_##type##_predictor_8x8; \
291 l[2] = vp9_##type##_predictor_16x16; \
292 l[3] = vp9_##type##_predictor_32x32
294 intra_pred_allsizes(pred[V_PRED], v);
295 intra_pred_allsizes(pred[H_PRED], h);
296 intra_pred_allsizes(pred[D207_PRED], d207);
297 intra_pred_allsizes(pred[D45_PRED], d45);
298 intra_pred_allsizes(pred[D63_PRED], d63);
299 intra_pred_allsizes(pred[D117_PRED], d117);
300 intra_pred_allsizes(pred[D135_PRED], d135);
301 intra_pred_allsizes(pred[D153_PRED], d153);
302 intra_pred_allsizes(pred[TM_PRED], tm);
304 intra_pred_allsizes(dc_pred[0][0], dc_128);
305 intra_pred_allsizes(dc_pred[0][1], dc_top);
306 intra_pred_allsizes(dc_pred[1][0], dc_left);
307 intra_pred_allsizes(dc_pred[1][1], dc);
309 #undef intra_pred_allsizes
312 static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
313 int ref_stride, uint8_t *dst, int dst_stride,
314 PREDICTION_MODE mode, TX_SIZE tx_size,
315 int up_available, int left_available,
316 int right_available, int x, int y,
319 DECLARE_ALIGNED_ARRAY(16, uint8_t, left_col, 64);
320 DECLARE_ALIGNED_ARRAY(16, uint8_t, above_data, 128 + 16);
321 uint8_t *above_row = above_data + 16;
322 const uint8_t *const_above_row = above_row;
323 const int bs = 4 << tx_size;
324 int frame_width, frame_height;
326 const struct macroblockd_plane *const pd = &xd->plane[plane];
328 // 127 127 127 .. 127 127 127 127 127 127
332 // 129 G H .. S T T T T T
335 once(init_intra_pred_fn_ptrs);
337 // Get current frame pointer, width and height.
339 frame_width = xd->cur_buf->y_width;
340 frame_height = xd->cur_buf->y_height;
342 frame_width = xd->cur_buf->uv_width;
343 frame_height = xd->cur_buf->uv_height;
346 // Get block position in current frame.
347 x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
348 y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
350 vpx_memset(left_col, 129, 64);
353 if (left_available) {
354 if (xd->mb_to_bottom_edge < 0) {
355 /* slower path if the block needs border extension */
356 if (y0 + bs <= frame_height) {
357 for (i = 0; i < bs; ++i)
358 left_col[i] = ref[i * ref_stride - 1];
360 const int extend_bottom = frame_height - y0;
361 for (i = 0; i < extend_bottom; ++i)
362 left_col[i] = ref[i * ref_stride - 1];
364 left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
367 /* faster path if the block does not need extension */
368 for (i = 0; i < bs; ++i)
369 left_col[i] = ref[i * ref_stride - 1];
373 // TODO(hkuang) do not extend 2*bs pixels for all modes.
376 const uint8_t *above_ref = ref - ref_stride;
377 if (xd->mb_to_right_edge < 0) {
378 /* slower path if the block needs border extension */
379 if (x0 + 2 * bs <= frame_width) {
380 if (right_available && bs == 4) {
381 vpx_memcpy(above_row, above_ref, 2 * bs);
383 vpx_memcpy(above_row, above_ref, bs);
384 vpx_memset(above_row + bs, above_row[bs - 1], bs);
386 } else if (x0 + bs <= frame_width) {
387 const int r = frame_width - x0;
388 if (right_available && bs == 4) {
389 vpx_memcpy(above_row, above_ref, r);
390 vpx_memset(above_row + r, above_row[r - 1],
391 x0 + 2 * bs - frame_width);
393 vpx_memcpy(above_row, above_ref, bs);
394 vpx_memset(above_row + bs, above_row[bs - 1], bs);
396 } else if (x0 <= frame_width) {
397 const int r = frame_width - x0;
398 if (right_available && bs == 4) {
399 vpx_memcpy(above_row, above_ref, r);
400 vpx_memset(above_row + r, above_row[r - 1],
401 x0 + 2 * bs - frame_width);
403 vpx_memcpy(above_row, above_ref, r);
404 vpx_memset(above_row + r, above_row[r - 1],
405 x0 + 2 * bs - frame_width);
408 above_row[-1] = left_available ? above_ref[-1] : 129;
410 /* faster path if the block does not need extension */
411 if (bs == 4 && right_available && left_available) {
412 const_above_row = above_ref;
414 vpx_memcpy(above_row, above_ref, bs);
415 if (bs == 4 && right_available)
416 vpx_memcpy(above_row + bs, above_ref + bs, bs);
418 vpx_memset(above_row + bs, above_row[bs - 1], bs);
419 above_row[-1] = left_available ? above_ref[-1] : 129;
423 vpx_memset(above_row, 127, bs * 2);
428 if (mode == DC_PRED) {
429 dc_pred[left_available][up_available][tx_size](dst, dst_stride,
430 const_above_row, left_col);
432 pred[mode][tx_size](dst, dst_stride, const_above_row, left_col);
436 void vp9_predict_intra_block(const MACROBLOCKD *xd, int block_idx, int bwl_in,
437 TX_SIZE tx_size, PREDICTION_MODE mode,
438 const uint8_t *ref, int ref_stride,
439 uint8_t *dst, int dst_stride,
440 int aoff, int loff, int plane) {
441 const int bwl = bwl_in - tx_size;
442 const int wmask = (1 << bwl) - 1;
443 const int have_top = (block_idx >> bwl) || xd->up_available;
444 const int have_left = (block_idx & wmask) || xd->left_available;
445 const int have_right = ((block_idx & wmask) != wmask);
446 const int x = aoff * 4;
447 const int y = loff * 4;
450 build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
451 have_top, have_left, have_right, x, y, plane);