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.
15 #include "./vpx_config.h"
17 #include "vp10/common/alloccommon.h"
18 #include "vp10/common/filter.h"
19 #include "vp10/common/idct.h"
20 #if CONFIG_VP9_POSTPROC
21 #include "vp10/common/postproc.h"
23 #include "vp10/common/reconinter.h"
24 #include "vp10/common/reconintra.h"
25 #include "vp10/common/tile_common.h"
27 #include "vp10/encoder/aq_complexity.h"
28 #include "vp10/encoder/aq_cyclicrefresh.h"
29 #include "vp10/encoder/aq_variance.h"
30 #include "vp10/encoder/bitstream.h"
31 #include "vp10/encoder/context_tree.h"
32 #include "vp10/encoder/encodeframe.h"
33 #include "vp10/encoder/encodemv.h"
34 #include "vp10/encoder/encoder.h"
35 #include "vp10/encoder/ethread.h"
36 #include "vp10/encoder/firstpass.h"
37 #include "vp10/encoder/mbgraph.h"
38 #include "vp10/encoder/picklpf.h"
39 #include "vp10/encoder/ratectrl.h"
40 #include "vp10/encoder/rd.h"
41 #include "vp10/encoder/resize.h"
42 #include "vp10/encoder/segmentation.h"
43 #include "vp10/encoder/skin_detection.h"
44 #include "vp10/encoder/speed_features.h"
45 #include "vp10/encoder/temporal_filter.h"
47 #include "./vp10_rtcd.h"
48 #include "./vpx_dsp_rtcd.h"
49 #include "./vpx_scale_rtcd.h"
50 #include "vpx/internal/vpx_psnr.h"
51 #if CONFIG_INTERNAL_STATS
52 #include "vpx_dsp/ssim.h"
54 #include "vpx_dsp/vpx_dsp_common.h"
55 #include "vpx_dsp/vpx_filter.h"
56 #include "vpx_ports/mem.h"
57 #include "vpx_ports/system_state.h"
58 #include "vpx_ports/vpx_timer.h"
59 #include "vpx_scale/vpx_scale.h"
61 #define AM_SEGMENT_ID_INACTIVE 7
62 #define AM_SEGMENT_ID_ACTIVE 0
64 #define SHARP_FILTER_QTHRESH 0 /* Q threshold for 8-tap sharp filter */
66 #define ALTREF_HIGH_PRECISION_MV 1 // Whether to use high precision mv
67 // for altref computation.
68 #define HIGH_PRECISION_MV_QTHRESH 200 // Q threshold for high precision
69 // mv. Choose a very high value for
70 // now so that HIGH_PRECISION is always
72 // #define OUTPUT_YUV_REC
74 #ifdef OUTPUT_YUV_DENOISED
75 FILE *yuv_denoised_file = NULL;
77 #ifdef OUTPUT_YUV_SKINMAP
78 FILE *yuv_skinmap_file = NULL;
90 static INLINE void Scale2Ratio(VPX_SCALING mode, int *hr, int *hs) {
116 // Mark all inactive blocks as active. Other segmentation features may be set
117 // so memset cannot be used, instead only inactive blocks should be reset.
118 static void suppress_active_map(VP10_COMP *cpi) {
119 unsigned char *const seg_map = cpi->segmentation_map;
121 if (cpi->active_map.enabled || cpi->active_map.update)
122 for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i)
123 if (seg_map[i] == AM_SEGMENT_ID_INACTIVE)
124 seg_map[i] = AM_SEGMENT_ID_ACTIVE;
127 static void apply_active_map(VP10_COMP *cpi) {
128 struct segmentation *const seg = &cpi->common.seg;
129 unsigned char *const seg_map = cpi->segmentation_map;
130 const unsigned char *const active_map = cpi->active_map.map;
133 assert(AM_SEGMENT_ID_ACTIVE == CR_SEGMENT_ID_BASE);
135 if (frame_is_intra_only(&cpi->common)) {
136 cpi->active_map.enabled = 0;
137 cpi->active_map.update = 1;
140 if (cpi->active_map.update) {
141 if (cpi->active_map.enabled) {
142 for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i)
143 if (seg_map[i] == AM_SEGMENT_ID_ACTIVE) seg_map[i] = active_map[i];
144 vp10_enable_segmentation(seg);
145 vp10_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP);
146 vp10_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF);
147 // Setting the data to -MAX_LOOP_FILTER will result in the computed loop
148 // filter level being zero regardless of the value of seg->abs_delta.
149 vp10_set_segdata(seg, AM_SEGMENT_ID_INACTIVE,
150 SEG_LVL_ALT_LF, -MAX_LOOP_FILTER);
152 vp10_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP);
153 vp10_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF);
155 seg->update_data = 1;
159 cpi->active_map.update = 0;
163 int vp10_set_active_map(VP10_COMP* cpi,
164 unsigned char* new_map_16x16,
167 if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols) {
168 unsigned char *const active_map_8x8 = cpi->active_map.map;
169 const int mi_rows = cpi->common.mi_rows;
170 const int mi_cols = cpi->common.mi_cols;
171 cpi->active_map.update = 1;
174 for (r = 0; r < mi_rows; ++r) {
175 for (c = 0; c < mi_cols; ++c) {
176 active_map_8x8[r * mi_cols + c] =
177 new_map_16x16[(r >> 1) * cols + (c >> 1)]
178 ? AM_SEGMENT_ID_ACTIVE
179 : AM_SEGMENT_ID_INACTIVE;
182 cpi->active_map.enabled = 1;
184 cpi->active_map.enabled = 0;
192 int vp10_get_active_map(VP10_COMP* cpi,
193 unsigned char* new_map_16x16,
196 if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols &&
198 unsigned char* const seg_map_8x8 = cpi->segmentation_map;
199 const int mi_rows = cpi->common.mi_rows;
200 const int mi_cols = cpi->common.mi_cols;
201 memset(new_map_16x16, !cpi->active_map.enabled, rows * cols);
202 if (cpi->active_map.enabled) {
204 for (r = 0; r < mi_rows; ++r) {
205 for (c = 0; c < mi_cols; ++c) {
206 // Cyclic refresh segments are considered active despite not having
207 // AM_SEGMENT_ID_ACTIVE
208 new_map_16x16[(r >> 1) * cols + (c >> 1)] |=
209 seg_map_8x8[r * mi_cols + c] != AM_SEGMENT_ID_INACTIVE;
219 void vp10_set_high_precision_mv(VP10_COMP *cpi, int allow_high_precision_mv) {
220 MACROBLOCK *const mb = &cpi->td.mb;
221 cpi->common.allow_high_precision_mv = allow_high_precision_mv;
222 if (cpi->common.allow_high_precision_mv) {
223 mb->mvcost = mb->nmvcost_hp;
224 mb->mvsadcost = mb->nmvsadcost_hp;
226 mb->mvcost = mb->nmvcost;
227 mb->mvsadcost = mb->nmvsadcost;
231 static void setup_frame(VP10_COMP *cpi) {
232 VP10_COMMON *const cm = &cpi->common;
233 // Set up entropy context depending on frame type. The decoder mandates
234 // the use of the default context, index 0, for keyframes and inter
235 // frames where the error_resilient_mode or intra_only flag is set. For
236 // other inter-frames the encoder currently uses only two contexts;
237 // context 1 for ALTREF frames and context 0 for the others.
238 if (frame_is_intra_only(cm) || cm->error_resilient_mode) {
239 vp10_setup_past_independence(cm);
241 cm->frame_context_idx = cpi->refresh_alt_ref_frame;
244 if (cm->frame_type == KEY_FRAME) {
245 cpi->refresh_golden_frame = 1;
246 cpi->refresh_alt_ref_frame = 1;
247 vp10_zero(cpi->interp_filter_selected);
249 *cm->fc = cm->frame_contexts[cm->frame_context_idx];
250 vp10_zero(cpi->interp_filter_selected[0]);
254 static void vp10_enc_setup_mi(VP10_COMMON *cm) {
256 cm->mi = cm->mip + cm->mi_stride + 1;
257 memset(cm->mip, 0, cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mip));
258 cm->prev_mi = cm->prev_mip + cm->mi_stride + 1;
259 // Clear top border row
260 memset(cm->prev_mip, 0, sizeof(*cm->prev_mip) * cm->mi_stride);
261 // Clear left border column
262 for (i = 1; i < cm->mi_rows + 1; ++i)
263 memset(&cm->prev_mip[i * cm->mi_stride], 0, sizeof(*cm->prev_mip));
265 cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
266 cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mi_stride + 1;
268 memset(cm->mi_grid_base, 0,
269 cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mi_grid_base));
272 static int vp10_enc_alloc_mi(VP10_COMMON *cm, int mi_size) {
273 cm->mip = vpx_calloc(mi_size, sizeof(*cm->mip));
276 cm->prev_mip = vpx_calloc(mi_size, sizeof(*cm->prev_mip));
279 cm->mi_alloc_size = mi_size;
281 cm->mi_grid_base = (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO*));
282 if (!cm->mi_grid_base)
284 cm->prev_mi_grid_base = (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO*));
285 if (!cm->prev_mi_grid_base)
291 static void vp10_enc_free_mi(VP10_COMMON *cm) {
294 vpx_free(cm->prev_mip);
296 vpx_free(cm->mi_grid_base);
297 cm->mi_grid_base = NULL;
298 vpx_free(cm->prev_mi_grid_base);
299 cm->prev_mi_grid_base = NULL;
302 static void vp10_swap_mi_and_prev_mi(VP10_COMMON *cm) {
303 // Current mip will be the prev_mip for the next frame.
304 MODE_INFO **temp_base = cm->prev_mi_grid_base;
305 MODE_INFO *temp = cm->prev_mip;
306 cm->prev_mip = cm->mip;
309 // Update the upper left visible macroblock ptrs.
310 cm->mi = cm->mip + cm->mi_stride + 1;
311 cm->prev_mi = cm->prev_mip + cm->mi_stride + 1;
313 cm->prev_mi_grid_base = cm->mi_grid_base;
314 cm->mi_grid_base = temp_base;
315 cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
316 cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mi_stride + 1;
319 void vp10_initialize_enc(void) {
320 static volatile int init_done = 0;
326 vp10_init_intra_predictors();
328 vp10_rc_init_minq_luts();
329 vp10_entropy_mv_init();
330 vp10_temporal_filter_init();
335 static void dealloc_compressor_data(VP10_COMP *cpi) {
336 VP10_COMMON *const cm = &cpi->common;
338 vpx_free(cpi->mbmi_ext_base);
339 cpi->mbmi_ext_base = NULL;
341 vpx_free(cpi->tile_data);
342 cpi->tile_data = NULL;
344 // Delete sementation map
345 vpx_free(cpi->segmentation_map);
346 cpi->segmentation_map = NULL;
347 vpx_free(cpi->coding_context.last_frame_seg_map_copy);
348 cpi->coding_context.last_frame_seg_map_copy = NULL;
350 vpx_free(cpi->nmvcosts[0]);
351 vpx_free(cpi->nmvcosts[1]);
352 cpi->nmvcosts[0] = NULL;
353 cpi->nmvcosts[1] = NULL;
355 vpx_free(cpi->nmvcosts_hp[0]);
356 vpx_free(cpi->nmvcosts_hp[1]);
357 cpi->nmvcosts_hp[0] = NULL;
358 cpi->nmvcosts_hp[1] = NULL;
360 vpx_free(cpi->nmvsadcosts[0]);
361 vpx_free(cpi->nmvsadcosts[1]);
362 cpi->nmvsadcosts[0] = NULL;
363 cpi->nmvsadcosts[1] = NULL;
365 vpx_free(cpi->nmvsadcosts_hp[0]);
366 vpx_free(cpi->nmvsadcosts_hp[1]);
367 cpi->nmvsadcosts_hp[0] = NULL;
368 cpi->nmvsadcosts_hp[1] = NULL;
370 vp10_cyclic_refresh_free(cpi->cyclic_refresh);
371 cpi->cyclic_refresh = NULL;
373 vpx_free(cpi->active_map.map);
374 cpi->active_map.map = NULL;
376 vp10_free_ref_frame_buffers(cm->buffer_pool);
377 #if CONFIG_VP9_POSTPROC
378 vp10_free_postproc_buffers(cm);
380 vp10_free_context_buffers(cm);
382 vpx_free_frame_buffer(&cpi->last_frame_uf);
383 vpx_free_frame_buffer(&cpi->scaled_source);
384 vpx_free_frame_buffer(&cpi->scaled_last_source);
385 vpx_free_frame_buffer(&cpi->alt_ref_buffer);
386 vp10_lookahead_destroy(cpi->lookahead);
388 vpx_free(cpi->tile_tok[0][0]);
389 cpi->tile_tok[0][0] = 0;
391 vp10_free_pc_tree(&cpi->td);
393 if (cpi->source_diff_var != NULL) {
394 vpx_free(cpi->source_diff_var);
395 cpi->source_diff_var = NULL;
399 static void save_coding_context(VP10_COMP *cpi) {
400 CODING_CONTEXT *const cc = &cpi->coding_context;
401 VP10_COMMON *cm = &cpi->common;
403 // Stores a snapshot of key state variables which can subsequently be
404 // restored with a call to vp10_restore_coding_context. These functions are
405 // intended for use in a re-code loop in vp10_compress_frame where the
406 // quantizer value is adjusted between loop iterations.
407 vp10_copy(cc->nmvjointcost, cpi->td.mb.nmvjointcost);
409 memcpy(cc->nmvcosts[0], cpi->nmvcosts[0],
410 MV_VALS * sizeof(*cpi->nmvcosts[0]));
411 memcpy(cc->nmvcosts[1], cpi->nmvcosts[1],
412 MV_VALS * sizeof(*cpi->nmvcosts[1]));
413 memcpy(cc->nmvcosts_hp[0], cpi->nmvcosts_hp[0],
414 MV_VALS * sizeof(*cpi->nmvcosts_hp[0]));
415 memcpy(cc->nmvcosts_hp[1], cpi->nmvcosts_hp[1],
416 MV_VALS * sizeof(*cpi->nmvcosts_hp[1]));
418 vp10_copy(cc->segment_pred_probs, cm->seg.pred_probs);
420 memcpy(cpi->coding_context.last_frame_seg_map_copy,
421 cm->last_frame_seg_map, (cm->mi_rows * cm->mi_cols));
423 vp10_copy(cc->last_ref_lf_deltas, cm->lf.last_ref_deltas);
424 vp10_copy(cc->last_mode_lf_deltas, cm->lf.last_mode_deltas);
429 static void restore_coding_context(VP10_COMP *cpi) {
430 CODING_CONTEXT *const cc = &cpi->coding_context;
431 VP10_COMMON *cm = &cpi->common;
433 // Restore key state variables to the snapshot state stored in the
434 // previous call to vp10_save_coding_context.
435 vp10_copy(cpi->td.mb.nmvjointcost, cc->nmvjointcost);
437 memcpy(cpi->nmvcosts[0], cc->nmvcosts[0], MV_VALS * sizeof(*cc->nmvcosts[0]));
438 memcpy(cpi->nmvcosts[1], cc->nmvcosts[1], MV_VALS * sizeof(*cc->nmvcosts[1]));
439 memcpy(cpi->nmvcosts_hp[0], cc->nmvcosts_hp[0],
440 MV_VALS * sizeof(*cc->nmvcosts_hp[0]));
441 memcpy(cpi->nmvcosts_hp[1], cc->nmvcosts_hp[1],
442 MV_VALS * sizeof(*cc->nmvcosts_hp[1]));
444 vp10_copy(cm->seg.pred_probs, cc->segment_pred_probs);
446 memcpy(cm->last_frame_seg_map,
447 cpi->coding_context.last_frame_seg_map_copy,
448 (cm->mi_rows * cm->mi_cols));
450 vp10_copy(cm->lf.last_ref_deltas, cc->last_ref_lf_deltas);
451 vp10_copy(cm->lf.last_mode_deltas, cc->last_mode_lf_deltas);
456 static void configure_static_seg_features(VP10_COMP *cpi) {
457 VP10_COMMON *const cm = &cpi->common;
458 const RATE_CONTROL *const rc = &cpi->rc;
459 struct segmentation *const seg = &cm->seg;
461 int high_q = (int)(rc->avg_q > 48.0);
464 // Disable and clear down for KF
465 if (cm->frame_type == KEY_FRAME) {
466 // Clear down the global segmentation map
467 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
469 seg->update_data = 0;
470 cpi->static_mb_pct = 0;
472 // Disable segmentation
473 vp10_disable_segmentation(seg);
475 // Clear down the segment features.
476 vp10_clearall_segfeatures(seg);
477 } else if (cpi->refresh_alt_ref_frame) {
478 // If this is an alt ref frame
479 // Clear down the global segmentation map
480 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
482 seg->update_data = 0;
483 cpi->static_mb_pct = 0;
485 // Disable segmentation and individual segment features by default
486 vp10_disable_segmentation(seg);
487 vp10_clearall_segfeatures(seg);
489 // Scan frames from current to arf frame.
490 // This function re-enables segmentation if appropriate.
491 vp10_update_mbgraph_stats(cpi);
493 // If segmentation was enabled set those features needed for the
497 seg->update_data = 1;
499 qi_delta = vp10_compute_qdelta(rc, rc->avg_q, rc->avg_q * 0.875,
501 vp10_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2);
502 vp10_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
504 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
505 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
507 // Where relevant assume segment data is delta data
508 seg->abs_delta = SEGMENT_DELTADATA;
510 } else if (seg->enabled) {
511 // All other frames if segmentation has been enabled
513 // First normal frame in a valid gf or alt ref group
514 if (rc->frames_since_golden == 0) {
515 // Set up segment features for normal frames in an arf group
516 if (rc->source_alt_ref_active) {
518 seg->update_data = 1;
519 seg->abs_delta = SEGMENT_DELTADATA;
521 qi_delta = vp10_compute_qdelta(rc, rc->avg_q, rc->avg_q * 1.125,
523 vp10_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta + 2);
524 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
526 vp10_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
527 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
529 // Segment coding disabled for compred testing
530 if (high_q || (cpi->static_mb_pct == 100)) {
531 vp10_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
532 vp10_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
533 vp10_enable_segfeature(seg, 1, SEG_LVL_SKIP);
536 // Disable segmentation and clear down features if alt ref
537 // is not active for this group
539 vp10_disable_segmentation(seg);
541 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
544 seg->update_data = 0;
546 vp10_clearall_segfeatures(seg);
548 } else if (rc->is_src_frame_alt_ref) {
549 // Special case where we are coding over the top of a previous
551 // Segment coding disabled for compred testing
553 // Enable ref frame features for segment 0 as well
554 vp10_enable_segfeature(seg, 0, SEG_LVL_REF_FRAME);
555 vp10_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
557 // All mbs should use ALTREF_FRAME
558 vp10_clear_segdata(seg, 0, SEG_LVL_REF_FRAME);
559 vp10_set_segdata(seg, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME);
560 vp10_clear_segdata(seg, 1, SEG_LVL_REF_FRAME);
561 vp10_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
563 // Skip all MBs if high Q (0,0 mv and skip coeffs)
565 vp10_enable_segfeature(seg, 0, SEG_LVL_SKIP);
566 vp10_enable_segfeature(seg, 1, SEG_LVL_SKIP);
568 // Enable data update
569 seg->update_data = 1;
573 // No updates.. leave things as they are.
575 seg->update_data = 0;
580 static void update_reference_segmentation_map(VP10_COMP *cpi) {
581 VP10_COMMON *const cm = &cpi->common;
582 MODE_INFO **mi_8x8_ptr = cm->mi_grid_visible;
583 uint8_t *cache_ptr = cm->last_frame_seg_map;
586 for (row = 0; row < cm->mi_rows; row++) {
587 MODE_INFO **mi_8x8 = mi_8x8_ptr;
588 uint8_t *cache = cache_ptr;
589 for (col = 0; col < cm->mi_cols; col++, mi_8x8++, cache++)
590 cache[0] = mi_8x8[0]->mbmi.segment_id;
591 mi_8x8_ptr += cm->mi_stride;
592 cache_ptr += cm->mi_cols;
596 static void alloc_raw_frame_buffers(VP10_COMP *cpi) {
597 VP10_COMMON *cm = &cpi->common;
598 const VP10EncoderConfig *oxcf = &cpi->oxcf;
601 cpi->lookahead = vp10_lookahead_init(oxcf->width, oxcf->height,
602 cm->subsampling_x, cm->subsampling_y,
603 #if CONFIG_VP9_HIGHBITDEPTH
604 cm->use_highbitdepth,
606 oxcf->lag_in_frames);
608 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
609 "Failed to allocate lag buffers");
611 // TODO(agrange) Check if ARF is enabled and skip allocation if not.
612 if (vpx_realloc_frame_buffer(&cpi->alt_ref_buffer,
613 oxcf->width, oxcf->height,
614 cm->subsampling_x, cm->subsampling_y,
615 #if CONFIG_VP9_HIGHBITDEPTH
616 cm->use_highbitdepth,
618 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
620 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
621 "Failed to allocate altref buffer");
624 static void alloc_util_frame_buffers(VP10_COMP *cpi) {
625 VP10_COMMON *const cm = &cpi->common;
626 if (vpx_realloc_frame_buffer(&cpi->last_frame_uf,
627 cm->width, cm->height,
628 cm->subsampling_x, cm->subsampling_y,
629 #if CONFIG_VP9_HIGHBITDEPTH
630 cm->use_highbitdepth,
632 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
634 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
635 "Failed to allocate last frame buffer");
637 if (vpx_realloc_frame_buffer(&cpi->scaled_source,
638 cm->width, cm->height,
639 cm->subsampling_x, cm->subsampling_y,
640 #if CONFIG_VP9_HIGHBITDEPTH
641 cm->use_highbitdepth,
643 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
645 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
646 "Failed to allocate scaled source buffer");
648 if (vpx_realloc_frame_buffer(&cpi->scaled_last_source,
649 cm->width, cm->height,
650 cm->subsampling_x, cm->subsampling_y,
651 #if CONFIG_VP9_HIGHBITDEPTH
652 cm->use_highbitdepth,
654 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
656 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
657 "Failed to allocate scaled last source buffer");
661 static int alloc_context_buffers_ext(VP10_COMP *cpi) {
662 VP10_COMMON *cm = &cpi->common;
663 int mi_size = cm->mi_cols * cm->mi_rows;
665 cpi->mbmi_ext_base = vpx_calloc(mi_size, sizeof(*cpi->mbmi_ext_base));
666 if (!cpi->mbmi_ext_base)
672 void vp10_alloc_compressor_data(VP10_COMP *cpi) {
673 VP10_COMMON *cm = &cpi->common;
675 vp10_alloc_context_buffers(cm, cm->width, cm->height);
677 alloc_context_buffers_ext(cpi);
679 vpx_free(cpi->tile_tok[0][0]);
682 unsigned int tokens = get_token_alloc(cm->mb_rows, cm->mb_cols);
683 CHECK_MEM_ERROR(cm, cpi->tile_tok[0][0],
684 vpx_calloc(tokens, sizeof(*cpi->tile_tok[0][0])));
687 vp10_setup_pc_tree(&cpi->common, &cpi->td);
690 void vp10_new_framerate(VP10_COMP *cpi, double framerate) {
691 cpi->framerate = framerate < 0.1 ? 30 : framerate;
692 vp10_rc_update_framerate(cpi);
695 static void set_tile_limits(VP10_COMP *cpi) {
696 VP10_COMMON *const cm = &cpi->common;
698 int min_log2_tile_cols, max_log2_tile_cols;
699 vp10_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
701 cm->log2_tile_cols = clamp(cpi->oxcf.tile_columns,
702 min_log2_tile_cols, max_log2_tile_cols);
703 cm->log2_tile_rows = cpi->oxcf.tile_rows;
706 static void update_frame_size(VP10_COMP *cpi) {
707 VP10_COMMON *const cm = &cpi->common;
708 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
710 vp10_set_mb_mi(cm, cm->width, cm->height);
711 vp10_init_context_buffers(cm);
712 vp10_init_macroblockd(cm, xd, NULL);
713 memset(cpi->mbmi_ext_base, 0,
714 cm->mi_rows * cm->mi_cols * sizeof(*cpi->mbmi_ext_base));
716 set_tile_limits(cpi);
719 static void init_buffer_indices(VP10_COMP *cpi) {
725 static void init_config(struct VP10_COMP *cpi, VP10EncoderConfig *oxcf) {
726 VP10_COMMON *const cm = &cpi->common;
729 cpi->framerate = oxcf->init_framerate;
731 cm->profile = oxcf->profile;
732 cm->bit_depth = oxcf->bit_depth;
733 #if CONFIG_VP9_HIGHBITDEPTH
734 cm->use_highbitdepth = oxcf->use_highbitdepth;
736 cm->color_space = oxcf->color_space;
737 cm->color_range = oxcf->color_range;
739 cm->width = oxcf->width;
740 cm->height = oxcf->height;
741 vp10_alloc_compressor_data(cpi);
743 // Single thread case: use counts in common.
744 cpi->td.counts = &cm->counts;
746 // change includes all joint functionality
747 vp10_change_config(cpi, oxcf);
749 cpi->static_mb_pct = 0;
750 cpi->ref_frame_flags = 0;
752 init_buffer_indices(cpi);
755 static void set_rc_buffer_sizes(RATE_CONTROL *rc,
756 const VP10EncoderConfig *oxcf) {
757 const int64_t bandwidth = oxcf->target_bandwidth;
758 const int64_t starting = oxcf->starting_buffer_level_ms;
759 const int64_t optimal = oxcf->optimal_buffer_level_ms;
760 const int64_t maximum = oxcf->maximum_buffer_size_ms;
762 rc->starting_buffer_level = starting * bandwidth / 1000;
763 rc->optimal_buffer_level = (optimal == 0) ? bandwidth / 8
764 : optimal * bandwidth / 1000;
765 rc->maximum_buffer_size = (maximum == 0) ? bandwidth / 8
766 : maximum * bandwidth / 1000;
769 #if CONFIG_VP9_HIGHBITDEPTH
770 #define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF) \
771 cpi->fn_ptr[BT].sdf = SDF; \
772 cpi->fn_ptr[BT].sdaf = SDAF; \
773 cpi->fn_ptr[BT].vf = VF; \
774 cpi->fn_ptr[BT].svf = SVF; \
775 cpi->fn_ptr[BT].svaf = SVAF; \
776 cpi->fn_ptr[BT].sdx3f = SDX3F; \
777 cpi->fn_ptr[BT].sdx8f = SDX8F; \
778 cpi->fn_ptr[BT].sdx4df = SDX4DF;
780 #define MAKE_BFP_SAD_WRAPPER(fnname) \
781 static unsigned int fnname##_bits8(const uint8_t *src_ptr, \
783 const uint8_t *ref_ptr, \
785 return fnname(src_ptr, source_stride, ref_ptr, ref_stride); \
787 static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
789 const uint8_t *ref_ptr, \
791 return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 2; \
793 static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
795 const uint8_t *ref_ptr, \
797 return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 4; \
800 #define MAKE_BFP_SADAVG_WRAPPER(fnname) static unsigned int \
801 fnname##_bits8(const uint8_t *src_ptr, \
803 const uint8_t *ref_ptr, \
805 const uint8_t *second_pred) { \
806 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred); \
808 static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
810 const uint8_t *ref_ptr, \
812 const uint8_t *second_pred) { \
813 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
816 static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
818 const uint8_t *ref_ptr, \
820 const uint8_t *second_pred) { \
821 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
825 #define MAKE_BFP_SAD3_WRAPPER(fnname) \
826 static void fnname##_bits8(const uint8_t *src_ptr, \
828 const uint8_t *ref_ptr, \
830 unsigned int *sad_array) { \
831 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
833 static void fnname##_bits10(const uint8_t *src_ptr, \
835 const uint8_t *ref_ptr, \
837 unsigned int *sad_array) { \
839 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
840 for (i = 0; i < 3; i++) \
841 sad_array[i] >>= 2; \
843 static void fnname##_bits12(const uint8_t *src_ptr, \
845 const uint8_t *ref_ptr, \
847 unsigned int *sad_array) { \
849 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
850 for (i = 0; i < 3; i++) \
851 sad_array[i] >>= 4; \
854 #define MAKE_BFP_SAD8_WRAPPER(fnname) \
855 static void fnname##_bits8(const uint8_t *src_ptr, \
857 const uint8_t *ref_ptr, \
859 unsigned int *sad_array) { \
860 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
862 static void fnname##_bits10(const uint8_t *src_ptr, \
864 const uint8_t *ref_ptr, \
866 unsigned int *sad_array) { \
868 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
869 for (i = 0; i < 8; i++) \
870 sad_array[i] >>= 2; \
872 static void fnname##_bits12(const uint8_t *src_ptr, \
874 const uint8_t *ref_ptr, \
876 unsigned int *sad_array) { \
878 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
879 for (i = 0; i < 8; i++) \
880 sad_array[i] >>= 4; \
882 #define MAKE_BFP_SAD4D_WRAPPER(fnname) \
883 static void fnname##_bits8(const uint8_t *src_ptr, \
885 const uint8_t* const ref_ptr[], \
887 unsigned int *sad_array) { \
888 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
890 static void fnname##_bits10(const uint8_t *src_ptr, \
892 const uint8_t* const ref_ptr[], \
894 unsigned int *sad_array) { \
896 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
897 for (i = 0; i < 4; i++) \
898 sad_array[i] >>= 2; \
900 static void fnname##_bits12(const uint8_t *src_ptr, \
902 const uint8_t* const ref_ptr[], \
904 unsigned int *sad_array) { \
906 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
907 for (i = 0; i < 4; i++) \
908 sad_array[i] >>= 4; \
911 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x16)
912 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x16_avg)
913 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x16x4d)
914 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x32)
915 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x32_avg)
916 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x32x4d)
917 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad64x32)
918 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad64x32_avg)
919 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad64x32x4d)
920 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x64)
921 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x64_avg)
922 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x64x4d)
923 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x32)
924 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x32_avg)
925 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad32x32x3)
926 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad32x32x8)
927 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x32x4d)
928 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad64x64)
929 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad64x64_avg)
930 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad64x64x3)
931 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad64x64x8)
932 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad64x64x4d)
933 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x16)
934 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x16_avg)
935 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x16x3)
936 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x16x8)
937 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x16x4d)
938 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x8)
939 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x8_avg)
940 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x8x3)
941 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x8x8)
942 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x8x4d)
943 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x16)
944 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x16_avg)
945 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x16x3)
946 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x16x8)
947 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x16x4d)
948 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x8)
949 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x8_avg)
950 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x8x3)
951 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x8x8)
952 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x8x4d)
953 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x4)
954 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x4_avg)
955 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x4x8)
956 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x4x4d)
957 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x8)
958 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x8_avg)
959 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x8x8)
960 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x8x4d)
961 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x4)
962 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x4_avg)
963 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad4x4x3)
964 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x4x8)
965 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x4x4d)
967 static void highbd_set_var_fns(VP10_COMP *const cpi) {
968 VP10_COMMON *const cm = &cpi->common;
969 if (cm->use_highbitdepth) {
970 switch (cm->bit_depth) {
972 HIGHBD_BFP(BLOCK_32X16,
973 vpx_highbd_sad32x16_bits8,
974 vpx_highbd_sad32x16_avg_bits8,
975 vpx_highbd_8_variance32x16,
976 vpx_highbd_8_sub_pixel_variance32x16,
977 vpx_highbd_8_sub_pixel_avg_variance32x16,
980 vpx_highbd_sad32x16x4d_bits8)
982 HIGHBD_BFP(BLOCK_16X32,
983 vpx_highbd_sad16x32_bits8,
984 vpx_highbd_sad16x32_avg_bits8,
985 vpx_highbd_8_variance16x32,
986 vpx_highbd_8_sub_pixel_variance16x32,
987 vpx_highbd_8_sub_pixel_avg_variance16x32,
990 vpx_highbd_sad16x32x4d_bits8)
992 HIGHBD_BFP(BLOCK_64X32,
993 vpx_highbd_sad64x32_bits8,
994 vpx_highbd_sad64x32_avg_bits8,
995 vpx_highbd_8_variance64x32,
996 vpx_highbd_8_sub_pixel_variance64x32,
997 vpx_highbd_8_sub_pixel_avg_variance64x32,
1000 vpx_highbd_sad64x32x4d_bits8)
1002 HIGHBD_BFP(BLOCK_32X64,
1003 vpx_highbd_sad32x64_bits8,
1004 vpx_highbd_sad32x64_avg_bits8,
1005 vpx_highbd_8_variance32x64,
1006 vpx_highbd_8_sub_pixel_variance32x64,
1007 vpx_highbd_8_sub_pixel_avg_variance32x64,
1010 vpx_highbd_sad32x64x4d_bits8)
1012 HIGHBD_BFP(BLOCK_32X32,
1013 vpx_highbd_sad32x32_bits8,
1014 vpx_highbd_sad32x32_avg_bits8,
1015 vpx_highbd_8_variance32x32,
1016 vpx_highbd_8_sub_pixel_variance32x32,
1017 vpx_highbd_8_sub_pixel_avg_variance32x32,
1018 vpx_highbd_sad32x32x3_bits8,
1019 vpx_highbd_sad32x32x8_bits8,
1020 vpx_highbd_sad32x32x4d_bits8)
1022 HIGHBD_BFP(BLOCK_64X64,
1023 vpx_highbd_sad64x64_bits8,
1024 vpx_highbd_sad64x64_avg_bits8,
1025 vpx_highbd_8_variance64x64,
1026 vpx_highbd_8_sub_pixel_variance64x64,
1027 vpx_highbd_8_sub_pixel_avg_variance64x64,
1028 vpx_highbd_sad64x64x3_bits8,
1029 vpx_highbd_sad64x64x8_bits8,
1030 vpx_highbd_sad64x64x4d_bits8)
1032 HIGHBD_BFP(BLOCK_16X16,
1033 vpx_highbd_sad16x16_bits8,
1034 vpx_highbd_sad16x16_avg_bits8,
1035 vpx_highbd_8_variance16x16,
1036 vpx_highbd_8_sub_pixel_variance16x16,
1037 vpx_highbd_8_sub_pixel_avg_variance16x16,
1038 vpx_highbd_sad16x16x3_bits8,
1039 vpx_highbd_sad16x16x8_bits8,
1040 vpx_highbd_sad16x16x4d_bits8)
1042 HIGHBD_BFP(BLOCK_16X8,
1043 vpx_highbd_sad16x8_bits8,
1044 vpx_highbd_sad16x8_avg_bits8,
1045 vpx_highbd_8_variance16x8,
1046 vpx_highbd_8_sub_pixel_variance16x8,
1047 vpx_highbd_8_sub_pixel_avg_variance16x8,
1048 vpx_highbd_sad16x8x3_bits8,
1049 vpx_highbd_sad16x8x8_bits8,
1050 vpx_highbd_sad16x8x4d_bits8)
1052 HIGHBD_BFP(BLOCK_8X16,
1053 vpx_highbd_sad8x16_bits8,
1054 vpx_highbd_sad8x16_avg_bits8,
1055 vpx_highbd_8_variance8x16,
1056 vpx_highbd_8_sub_pixel_variance8x16,
1057 vpx_highbd_8_sub_pixel_avg_variance8x16,
1058 vpx_highbd_sad8x16x3_bits8,
1059 vpx_highbd_sad8x16x8_bits8,
1060 vpx_highbd_sad8x16x4d_bits8)
1062 HIGHBD_BFP(BLOCK_8X8,
1063 vpx_highbd_sad8x8_bits8,
1064 vpx_highbd_sad8x8_avg_bits8,
1065 vpx_highbd_8_variance8x8,
1066 vpx_highbd_8_sub_pixel_variance8x8,
1067 vpx_highbd_8_sub_pixel_avg_variance8x8,
1068 vpx_highbd_sad8x8x3_bits8,
1069 vpx_highbd_sad8x8x8_bits8,
1070 vpx_highbd_sad8x8x4d_bits8)
1072 HIGHBD_BFP(BLOCK_8X4,
1073 vpx_highbd_sad8x4_bits8,
1074 vpx_highbd_sad8x4_avg_bits8,
1075 vpx_highbd_8_variance8x4,
1076 vpx_highbd_8_sub_pixel_variance8x4,
1077 vpx_highbd_8_sub_pixel_avg_variance8x4,
1079 vpx_highbd_sad8x4x8_bits8,
1080 vpx_highbd_sad8x4x4d_bits8)
1082 HIGHBD_BFP(BLOCK_4X8,
1083 vpx_highbd_sad4x8_bits8,
1084 vpx_highbd_sad4x8_avg_bits8,
1085 vpx_highbd_8_variance4x8,
1086 vpx_highbd_8_sub_pixel_variance4x8,
1087 vpx_highbd_8_sub_pixel_avg_variance4x8,
1089 vpx_highbd_sad4x8x8_bits8,
1090 vpx_highbd_sad4x8x4d_bits8)
1092 HIGHBD_BFP(BLOCK_4X4,
1093 vpx_highbd_sad4x4_bits8,
1094 vpx_highbd_sad4x4_avg_bits8,
1095 vpx_highbd_8_variance4x4,
1096 vpx_highbd_8_sub_pixel_variance4x4,
1097 vpx_highbd_8_sub_pixel_avg_variance4x4,
1098 vpx_highbd_sad4x4x3_bits8,
1099 vpx_highbd_sad4x4x8_bits8,
1100 vpx_highbd_sad4x4x4d_bits8)
1104 HIGHBD_BFP(BLOCK_32X16,
1105 vpx_highbd_sad32x16_bits10,
1106 vpx_highbd_sad32x16_avg_bits10,
1107 vpx_highbd_10_variance32x16,
1108 vpx_highbd_10_sub_pixel_variance32x16,
1109 vpx_highbd_10_sub_pixel_avg_variance32x16,
1112 vpx_highbd_sad32x16x4d_bits10)
1114 HIGHBD_BFP(BLOCK_16X32,
1115 vpx_highbd_sad16x32_bits10,
1116 vpx_highbd_sad16x32_avg_bits10,
1117 vpx_highbd_10_variance16x32,
1118 vpx_highbd_10_sub_pixel_variance16x32,
1119 vpx_highbd_10_sub_pixel_avg_variance16x32,
1122 vpx_highbd_sad16x32x4d_bits10)
1124 HIGHBD_BFP(BLOCK_64X32,
1125 vpx_highbd_sad64x32_bits10,
1126 vpx_highbd_sad64x32_avg_bits10,
1127 vpx_highbd_10_variance64x32,
1128 vpx_highbd_10_sub_pixel_variance64x32,
1129 vpx_highbd_10_sub_pixel_avg_variance64x32,
1132 vpx_highbd_sad64x32x4d_bits10)
1134 HIGHBD_BFP(BLOCK_32X64,
1135 vpx_highbd_sad32x64_bits10,
1136 vpx_highbd_sad32x64_avg_bits10,
1137 vpx_highbd_10_variance32x64,
1138 vpx_highbd_10_sub_pixel_variance32x64,
1139 vpx_highbd_10_sub_pixel_avg_variance32x64,
1142 vpx_highbd_sad32x64x4d_bits10)
1144 HIGHBD_BFP(BLOCK_32X32,
1145 vpx_highbd_sad32x32_bits10,
1146 vpx_highbd_sad32x32_avg_bits10,
1147 vpx_highbd_10_variance32x32,
1148 vpx_highbd_10_sub_pixel_variance32x32,
1149 vpx_highbd_10_sub_pixel_avg_variance32x32,
1150 vpx_highbd_sad32x32x3_bits10,
1151 vpx_highbd_sad32x32x8_bits10,
1152 vpx_highbd_sad32x32x4d_bits10)
1154 HIGHBD_BFP(BLOCK_64X64,
1155 vpx_highbd_sad64x64_bits10,
1156 vpx_highbd_sad64x64_avg_bits10,
1157 vpx_highbd_10_variance64x64,
1158 vpx_highbd_10_sub_pixel_variance64x64,
1159 vpx_highbd_10_sub_pixel_avg_variance64x64,
1160 vpx_highbd_sad64x64x3_bits10,
1161 vpx_highbd_sad64x64x8_bits10,
1162 vpx_highbd_sad64x64x4d_bits10)
1164 HIGHBD_BFP(BLOCK_16X16,
1165 vpx_highbd_sad16x16_bits10,
1166 vpx_highbd_sad16x16_avg_bits10,
1167 vpx_highbd_10_variance16x16,
1168 vpx_highbd_10_sub_pixel_variance16x16,
1169 vpx_highbd_10_sub_pixel_avg_variance16x16,
1170 vpx_highbd_sad16x16x3_bits10,
1171 vpx_highbd_sad16x16x8_bits10,
1172 vpx_highbd_sad16x16x4d_bits10)
1174 HIGHBD_BFP(BLOCK_16X8,
1175 vpx_highbd_sad16x8_bits10,
1176 vpx_highbd_sad16x8_avg_bits10,
1177 vpx_highbd_10_variance16x8,
1178 vpx_highbd_10_sub_pixel_variance16x8,
1179 vpx_highbd_10_sub_pixel_avg_variance16x8,
1180 vpx_highbd_sad16x8x3_bits10,
1181 vpx_highbd_sad16x8x8_bits10,
1182 vpx_highbd_sad16x8x4d_bits10)
1184 HIGHBD_BFP(BLOCK_8X16,
1185 vpx_highbd_sad8x16_bits10,
1186 vpx_highbd_sad8x16_avg_bits10,
1187 vpx_highbd_10_variance8x16,
1188 vpx_highbd_10_sub_pixel_variance8x16,
1189 vpx_highbd_10_sub_pixel_avg_variance8x16,
1190 vpx_highbd_sad8x16x3_bits10,
1191 vpx_highbd_sad8x16x8_bits10,
1192 vpx_highbd_sad8x16x4d_bits10)
1194 HIGHBD_BFP(BLOCK_8X8,
1195 vpx_highbd_sad8x8_bits10,
1196 vpx_highbd_sad8x8_avg_bits10,
1197 vpx_highbd_10_variance8x8,
1198 vpx_highbd_10_sub_pixel_variance8x8,
1199 vpx_highbd_10_sub_pixel_avg_variance8x8,
1200 vpx_highbd_sad8x8x3_bits10,
1201 vpx_highbd_sad8x8x8_bits10,
1202 vpx_highbd_sad8x8x4d_bits10)
1204 HIGHBD_BFP(BLOCK_8X4,
1205 vpx_highbd_sad8x4_bits10,
1206 vpx_highbd_sad8x4_avg_bits10,
1207 vpx_highbd_10_variance8x4,
1208 vpx_highbd_10_sub_pixel_variance8x4,
1209 vpx_highbd_10_sub_pixel_avg_variance8x4,
1211 vpx_highbd_sad8x4x8_bits10,
1212 vpx_highbd_sad8x4x4d_bits10)
1214 HIGHBD_BFP(BLOCK_4X8,
1215 vpx_highbd_sad4x8_bits10,
1216 vpx_highbd_sad4x8_avg_bits10,
1217 vpx_highbd_10_variance4x8,
1218 vpx_highbd_10_sub_pixel_variance4x8,
1219 vpx_highbd_10_sub_pixel_avg_variance4x8,
1221 vpx_highbd_sad4x8x8_bits10,
1222 vpx_highbd_sad4x8x4d_bits10)
1224 HIGHBD_BFP(BLOCK_4X4,
1225 vpx_highbd_sad4x4_bits10,
1226 vpx_highbd_sad4x4_avg_bits10,
1227 vpx_highbd_10_variance4x4,
1228 vpx_highbd_10_sub_pixel_variance4x4,
1229 vpx_highbd_10_sub_pixel_avg_variance4x4,
1230 vpx_highbd_sad4x4x3_bits10,
1231 vpx_highbd_sad4x4x8_bits10,
1232 vpx_highbd_sad4x4x4d_bits10)
1236 HIGHBD_BFP(BLOCK_32X16,
1237 vpx_highbd_sad32x16_bits12,
1238 vpx_highbd_sad32x16_avg_bits12,
1239 vpx_highbd_12_variance32x16,
1240 vpx_highbd_12_sub_pixel_variance32x16,
1241 vpx_highbd_12_sub_pixel_avg_variance32x16,
1244 vpx_highbd_sad32x16x4d_bits12)
1246 HIGHBD_BFP(BLOCK_16X32,
1247 vpx_highbd_sad16x32_bits12,
1248 vpx_highbd_sad16x32_avg_bits12,
1249 vpx_highbd_12_variance16x32,
1250 vpx_highbd_12_sub_pixel_variance16x32,
1251 vpx_highbd_12_sub_pixel_avg_variance16x32,
1254 vpx_highbd_sad16x32x4d_bits12)
1256 HIGHBD_BFP(BLOCK_64X32,
1257 vpx_highbd_sad64x32_bits12,
1258 vpx_highbd_sad64x32_avg_bits12,
1259 vpx_highbd_12_variance64x32,
1260 vpx_highbd_12_sub_pixel_variance64x32,
1261 vpx_highbd_12_sub_pixel_avg_variance64x32,
1264 vpx_highbd_sad64x32x4d_bits12)
1266 HIGHBD_BFP(BLOCK_32X64,
1267 vpx_highbd_sad32x64_bits12,
1268 vpx_highbd_sad32x64_avg_bits12,
1269 vpx_highbd_12_variance32x64,
1270 vpx_highbd_12_sub_pixel_variance32x64,
1271 vpx_highbd_12_sub_pixel_avg_variance32x64,
1274 vpx_highbd_sad32x64x4d_bits12)
1276 HIGHBD_BFP(BLOCK_32X32,
1277 vpx_highbd_sad32x32_bits12,
1278 vpx_highbd_sad32x32_avg_bits12,
1279 vpx_highbd_12_variance32x32,
1280 vpx_highbd_12_sub_pixel_variance32x32,
1281 vpx_highbd_12_sub_pixel_avg_variance32x32,
1282 vpx_highbd_sad32x32x3_bits12,
1283 vpx_highbd_sad32x32x8_bits12,
1284 vpx_highbd_sad32x32x4d_bits12)
1286 HIGHBD_BFP(BLOCK_64X64,
1287 vpx_highbd_sad64x64_bits12,
1288 vpx_highbd_sad64x64_avg_bits12,
1289 vpx_highbd_12_variance64x64,
1290 vpx_highbd_12_sub_pixel_variance64x64,
1291 vpx_highbd_12_sub_pixel_avg_variance64x64,
1292 vpx_highbd_sad64x64x3_bits12,
1293 vpx_highbd_sad64x64x8_bits12,
1294 vpx_highbd_sad64x64x4d_bits12)
1296 HIGHBD_BFP(BLOCK_16X16,
1297 vpx_highbd_sad16x16_bits12,
1298 vpx_highbd_sad16x16_avg_bits12,
1299 vpx_highbd_12_variance16x16,
1300 vpx_highbd_12_sub_pixel_variance16x16,
1301 vpx_highbd_12_sub_pixel_avg_variance16x16,
1302 vpx_highbd_sad16x16x3_bits12,
1303 vpx_highbd_sad16x16x8_bits12,
1304 vpx_highbd_sad16x16x4d_bits12)
1306 HIGHBD_BFP(BLOCK_16X8,
1307 vpx_highbd_sad16x8_bits12,
1308 vpx_highbd_sad16x8_avg_bits12,
1309 vpx_highbd_12_variance16x8,
1310 vpx_highbd_12_sub_pixel_variance16x8,
1311 vpx_highbd_12_sub_pixel_avg_variance16x8,
1312 vpx_highbd_sad16x8x3_bits12,
1313 vpx_highbd_sad16x8x8_bits12,
1314 vpx_highbd_sad16x8x4d_bits12)
1316 HIGHBD_BFP(BLOCK_8X16,
1317 vpx_highbd_sad8x16_bits12,
1318 vpx_highbd_sad8x16_avg_bits12,
1319 vpx_highbd_12_variance8x16,
1320 vpx_highbd_12_sub_pixel_variance8x16,
1321 vpx_highbd_12_sub_pixel_avg_variance8x16,
1322 vpx_highbd_sad8x16x3_bits12,
1323 vpx_highbd_sad8x16x8_bits12,
1324 vpx_highbd_sad8x16x4d_bits12)
1326 HIGHBD_BFP(BLOCK_8X8,
1327 vpx_highbd_sad8x8_bits12,
1328 vpx_highbd_sad8x8_avg_bits12,
1329 vpx_highbd_12_variance8x8,
1330 vpx_highbd_12_sub_pixel_variance8x8,
1331 vpx_highbd_12_sub_pixel_avg_variance8x8,
1332 vpx_highbd_sad8x8x3_bits12,
1333 vpx_highbd_sad8x8x8_bits12,
1334 vpx_highbd_sad8x8x4d_bits12)
1336 HIGHBD_BFP(BLOCK_8X4,
1337 vpx_highbd_sad8x4_bits12,
1338 vpx_highbd_sad8x4_avg_bits12,
1339 vpx_highbd_12_variance8x4,
1340 vpx_highbd_12_sub_pixel_variance8x4,
1341 vpx_highbd_12_sub_pixel_avg_variance8x4,
1343 vpx_highbd_sad8x4x8_bits12,
1344 vpx_highbd_sad8x4x4d_bits12)
1346 HIGHBD_BFP(BLOCK_4X8,
1347 vpx_highbd_sad4x8_bits12,
1348 vpx_highbd_sad4x8_avg_bits12,
1349 vpx_highbd_12_variance4x8,
1350 vpx_highbd_12_sub_pixel_variance4x8,
1351 vpx_highbd_12_sub_pixel_avg_variance4x8,
1353 vpx_highbd_sad4x8x8_bits12,
1354 vpx_highbd_sad4x8x4d_bits12)
1356 HIGHBD_BFP(BLOCK_4X4,
1357 vpx_highbd_sad4x4_bits12,
1358 vpx_highbd_sad4x4_avg_bits12,
1359 vpx_highbd_12_variance4x4,
1360 vpx_highbd_12_sub_pixel_variance4x4,
1361 vpx_highbd_12_sub_pixel_avg_variance4x4,
1362 vpx_highbd_sad4x4x3_bits12,
1363 vpx_highbd_sad4x4x8_bits12,
1364 vpx_highbd_sad4x4x4d_bits12)
1368 assert(0 && "cm->bit_depth should be VPX_BITS_8, "
1369 "VPX_BITS_10 or VPX_BITS_12");
1373 #endif // CONFIG_VP9_HIGHBITDEPTH
1375 static void realloc_segmentation_maps(VP10_COMP *cpi) {
1376 VP10_COMMON *const cm = &cpi->common;
1378 // Create the encoder segmentation map and set all entries to 0
1379 vpx_free(cpi->segmentation_map);
1380 CHECK_MEM_ERROR(cm, cpi->segmentation_map,
1381 vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
1383 // Create a map used for cyclic background refresh.
1384 if (cpi->cyclic_refresh)
1385 vp10_cyclic_refresh_free(cpi->cyclic_refresh);
1386 CHECK_MEM_ERROR(cm, cpi->cyclic_refresh,
1387 vp10_cyclic_refresh_alloc(cm->mi_rows, cm->mi_cols));
1389 // Create a map used to mark inactive areas.
1390 vpx_free(cpi->active_map.map);
1391 CHECK_MEM_ERROR(cm, cpi->active_map.map,
1392 vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
1394 // And a place holder structure is the coding context
1395 // for use if we want to save and restore it
1396 vpx_free(cpi->coding_context.last_frame_seg_map_copy);
1397 CHECK_MEM_ERROR(cm, cpi->coding_context.last_frame_seg_map_copy,
1398 vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
1401 void vp10_change_config(struct VP10_COMP *cpi, const VP10EncoderConfig *oxcf) {
1402 VP10_COMMON *const cm = &cpi->common;
1403 RATE_CONTROL *const rc = &cpi->rc;
1405 if (cm->profile != oxcf->profile)
1406 cm->profile = oxcf->profile;
1407 cm->bit_depth = oxcf->bit_depth;
1408 cm->color_space = oxcf->color_space;
1409 cm->color_range = oxcf->color_range;
1411 if (cm->profile <= PROFILE_1)
1412 assert(cm->bit_depth == VPX_BITS_8);
1414 assert(cm->bit_depth > VPX_BITS_8);
1417 #if CONFIG_VP9_HIGHBITDEPTH
1418 cpi->td.mb.e_mbd.bd = (int)cm->bit_depth;
1419 #endif // CONFIG_VP9_HIGHBITDEPTH
1421 rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
1423 cpi->refresh_golden_frame = 0;
1424 cpi->refresh_last_frame = 1;
1425 cm->refresh_frame_context =
1426 oxcf->error_resilient_mode ? REFRESH_FRAME_CONTEXT_OFF :
1427 oxcf->frame_parallel_decoding_mode ? REFRESH_FRAME_CONTEXT_FORWARD
1428 : REFRESH_FRAME_CONTEXT_BACKWARD;
1429 cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
1431 vp10_reset_segment_features(&cm->seg);
1432 vp10_set_high_precision_mv(cpi, 0);
1437 for (i = 0; i < MAX_SEGMENTS; i++)
1438 cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout;
1440 cpi->encode_breakout = cpi->oxcf.encode_breakout;
1442 set_rc_buffer_sizes(rc, &cpi->oxcf);
1444 // Under a configuration change, where maximum_buffer_size may change,
1445 // keep buffer level clipped to the maximum allowed buffer size.
1446 rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
1447 rc->buffer_level = VPXMIN(rc->buffer_level, rc->maximum_buffer_size);
1449 // Set up frame rate and related parameters rate control values.
1450 vp10_new_framerate(cpi, cpi->framerate);
1452 // Set absolute upper and lower quality limits
1453 rc->worst_quality = cpi->oxcf.worst_allowed_q;
1454 rc->best_quality = cpi->oxcf.best_allowed_q;
1456 cm->interp_filter = cpi->sf.default_interp_filter;
1458 cm->display_width = cpi->oxcf.width;
1459 cm->display_height = cpi->oxcf.height;
1460 cm->width = cpi->oxcf.width;
1461 cm->height = cpi->oxcf.height;
1463 if (cpi->initial_width) {
1464 if (cm->width > cpi->initial_width || cm->height > cpi->initial_height) {
1465 vp10_free_context_buffers(cm);
1466 vp10_alloc_compressor_data(cpi);
1467 realloc_segmentation_maps(cpi);
1468 cpi->initial_width = cpi->initial_height = 0;
1471 update_frame_size(cpi);
1473 cpi->alt_ref_source = NULL;
1474 rc->is_src_frame_alt_ref = 0;
1477 // Experimental RD Code
1478 cpi->frame_distortion = 0;
1479 cpi->last_frame_distortion = 0;
1482 set_tile_limits(cpi);
1484 cpi->ext_refresh_frame_flags_pending = 0;
1485 cpi->ext_refresh_frame_context_pending = 0;
1487 #if CONFIG_VP9_HIGHBITDEPTH
1488 highbd_set_var_fns(cpi);
1493 #define M_LOG2_E 0.693147180559945309417
1495 #define log2f(x) (log (x) / (float) M_LOG2_E)
1497 static void cal_nmvjointsadcost(int *mvjointsadcost) {
1498 mvjointsadcost[0] = 600;
1499 mvjointsadcost[1] = 300;
1500 mvjointsadcost[2] = 300;
1501 mvjointsadcost[3] = 300;
1504 static void cal_nmvsadcosts(int *mvsadcost[2]) {
1507 mvsadcost[0][0] = 0;
1508 mvsadcost[1][0] = 0;
1511 double z = 256 * (2 * (log2f(8 * i) + .6));
1512 mvsadcost[0][i] = (int)z;
1513 mvsadcost[1][i] = (int)z;
1514 mvsadcost[0][-i] = (int)z;
1515 mvsadcost[1][-i] = (int)z;
1516 } while (++i <= MV_MAX);
1519 static void cal_nmvsadcosts_hp(int *mvsadcost[2]) {
1522 mvsadcost[0][0] = 0;
1523 mvsadcost[1][0] = 0;
1526 double z = 256 * (2 * (log2f(8 * i) + .6));
1527 mvsadcost[0][i] = (int)z;
1528 mvsadcost[1][i] = (int)z;
1529 mvsadcost[0][-i] = (int)z;
1530 mvsadcost[1][-i] = (int)z;
1531 } while (++i <= MV_MAX);
1535 VP10_COMP *vp10_create_compressor(VP10EncoderConfig *oxcf,
1536 BufferPool *const pool) {
1538 VP10_COMP *volatile const cpi = vpx_memalign(32, sizeof(VP10_COMP));
1539 VP10_COMMON *volatile const cm = cpi != NULL ? &cpi->common : NULL;
1546 if (setjmp(cm->error.jmp)) {
1547 cm->error.setjmp = 0;
1548 vp10_remove_compressor(cpi);
1552 cm->error.setjmp = 1;
1553 cm->alloc_mi = vp10_enc_alloc_mi;
1554 cm->free_mi = vp10_enc_free_mi;
1555 cm->setup_mi = vp10_enc_setup_mi;
1557 CHECK_MEM_ERROR(cm, cm->fc,
1558 (FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
1559 CHECK_MEM_ERROR(cm, cm->frame_contexts,
1560 (FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
1561 sizeof(*cm->frame_contexts)));
1563 cpi->resize_state = 0;
1564 cpi->resize_avg_qp = 0;
1565 cpi->resize_buffer_underflow = 0;
1566 cpi->common.buffer_pool = pool;
1568 init_config(cpi, oxcf);
1569 vp10_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc);
1571 cm->current_video_frame = 0;
1572 cpi->partition_search_skippable_frame = 0;
1573 cpi->tile_data = NULL;
1575 realloc_segmentation_maps(cpi);
1577 CHECK_MEM_ERROR(cm, cpi->nmvcosts[0],
1578 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[0])));
1579 CHECK_MEM_ERROR(cm, cpi->nmvcosts[1],
1580 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[1])));
1581 CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[0],
1582 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[0])));
1583 CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[1],
1584 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[1])));
1585 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[0],
1586 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[0])));
1587 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[1],
1588 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[1])));
1589 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[0],
1590 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[0])));
1591 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[1],
1592 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[1])));
1594 for (i = 0; i < (sizeof(cpi->mbgraph_stats) /
1595 sizeof(cpi->mbgraph_stats[0])); i++) {
1596 CHECK_MEM_ERROR(cm, cpi->mbgraph_stats[i].mb_stats,
1597 vpx_calloc(cm->MBs *
1598 sizeof(*cpi->mbgraph_stats[i].mb_stats), 1));
1601 #if CONFIG_FP_MB_STATS
1602 cpi->use_fp_mb_stats = 0;
1603 if (cpi->use_fp_mb_stats) {
1604 // a place holder used to store the first pass mb stats in the first pass
1605 CHECK_MEM_ERROR(cm, cpi->twopass.frame_mb_stats_buf,
1606 vpx_calloc(cm->MBs * sizeof(uint8_t), 1));
1608 cpi->twopass.frame_mb_stats_buf = NULL;
1612 cpi->refresh_alt_ref_frame = 0;
1613 cpi->multi_arf_last_grp_enabled = 0;
1615 cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
1616 #if CONFIG_INTERNAL_STATS
1617 cpi->b_calculate_ssimg = 0;
1618 cpi->b_calculate_blockiness = 1;
1619 cpi->b_calculate_consistency = 1;
1620 cpi->total_inconsistency = 0;
1621 cpi->psnr.worst = 100.0;
1622 cpi->worst_ssim = 100.0;
1627 if (cpi->b_calculate_psnr) {
1628 cpi->total_sq_error = 0;
1629 cpi->total_samples = 0;
1631 cpi->totalp_sq_error = 0;
1632 cpi->totalp_samples = 0;
1634 cpi->tot_recode_hits = 0;
1635 cpi->summed_quality = 0;
1636 cpi->summed_weights = 0;
1637 cpi->summedp_quality = 0;
1638 cpi->summedp_weights = 0;
1641 if (cpi->b_calculate_ssimg) {
1642 cpi->ssimg.worst= 100.0;
1644 cpi->fastssim.worst = 100.0;
1646 cpi->psnrhvs.worst = 100.0;
1648 if (cpi->b_calculate_blockiness) {
1649 cpi->total_blockiness = 0;
1650 cpi->worst_blockiness = 0.0;
1653 if (cpi->b_calculate_consistency) {
1654 cpi->ssim_vars = vpx_malloc(sizeof(*cpi->ssim_vars) *
1655 4 * cpi->common.mi_rows * cpi->common.mi_cols);
1656 cpi->worst_consistency = 100.0;
1661 cpi->first_time_stamp_ever = INT64_MAX;
1663 cal_nmvjointsadcost(cpi->td.mb.nmvjointsadcost);
1664 cpi->td.mb.nmvcost[0] = &cpi->nmvcosts[0][MV_MAX];
1665 cpi->td.mb.nmvcost[1] = &cpi->nmvcosts[1][MV_MAX];
1666 cpi->td.mb.nmvsadcost[0] = &cpi->nmvsadcosts[0][MV_MAX];
1667 cpi->td.mb.nmvsadcost[1] = &cpi->nmvsadcosts[1][MV_MAX];
1668 cal_nmvsadcosts(cpi->td.mb.nmvsadcost);
1670 cpi->td.mb.nmvcost_hp[0] = &cpi->nmvcosts_hp[0][MV_MAX];
1671 cpi->td.mb.nmvcost_hp[1] = &cpi->nmvcosts_hp[1][MV_MAX];
1672 cpi->td.mb.nmvsadcost_hp[0] = &cpi->nmvsadcosts_hp[0][MV_MAX];
1673 cpi->td.mb.nmvsadcost_hp[1] = &cpi->nmvsadcosts_hp[1][MV_MAX];
1674 cal_nmvsadcosts_hp(cpi->td.mb.nmvsadcost_hp);
1676 #if CONFIG_VP9_TEMPORAL_DENOISING
1677 #ifdef OUTPUT_YUV_DENOISED
1678 yuv_denoised_file = fopen("denoised.yuv", "ab");
1681 #ifdef OUTPUT_YUV_SKINMAP
1682 yuv_skinmap_file = fopen("skinmap.yuv", "ab");
1684 #ifdef OUTPUT_YUV_REC
1685 yuv_rec_file = fopen("rec.yuv", "wb");
1689 framepsnr = fopen("framepsnr.stt", "a");
1690 kf_list = fopen("kf_list.stt", "w");
1693 cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
1695 if (oxcf->pass == 1) {
1696 vp10_init_first_pass(cpi);
1697 } else if (oxcf->pass == 2) {
1698 const size_t packet_sz = sizeof(FIRSTPASS_STATS);
1699 const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz);
1701 #if CONFIG_FP_MB_STATS
1702 if (cpi->use_fp_mb_stats) {
1703 const size_t psz = cpi->common.MBs * sizeof(uint8_t);
1704 const int ps = (int)(oxcf->firstpass_mb_stats_in.sz / psz);
1706 cpi->twopass.firstpass_mb_stats.mb_stats_start =
1707 oxcf->firstpass_mb_stats_in.buf;
1708 cpi->twopass.firstpass_mb_stats.mb_stats_end =
1709 cpi->twopass.firstpass_mb_stats.mb_stats_start +
1710 (ps - 1) * cpi->common.MBs * sizeof(uint8_t);
1714 cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf;
1715 cpi->twopass.stats_in = cpi->twopass.stats_in_start;
1716 cpi->twopass.stats_in_end = &cpi->twopass.stats_in[packets - 1];
1718 vp10_init_second_pass(cpi);
1721 vp10_set_speed_features_framesize_independent(cpi);
1722 vp10_set_speed_features_framesize_dependent(cpi);
1724 // Allocate memory to store variances for a frame.
1725 CHECK_MEM_ERROR(cm, cpi->source_diff_var,
1726 vpx_calloc(cm->MBs, sizeof(diff)));
1727 cpi->source_var_thresh = 0;
1728 cpi->frames_till_next_var_check = 0;
1730 #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF)\
1731 cpi->fn_ptr[BT].sdf = SDF; \
1732 cpi->fn_ptr[BT].sdaf = SDAF; \
1733 cpi->fn_ptr[BT].vf = VF; \
1734 cpi->fn_ptr[BT].svf = SVF; \
1735 cpi->fn_ptr[BT].svaf = SVAF; \
1736 cpi->fn_ptr[BT].sdx3f = SDX3F; \
1737 cpi->fn_ptr[BT].sdx8f = SDX8F; \
1738 cpi->fn_ptr[BT].sdx4df = SDX4DF;
1740 BFP(BLOCK_32X16, vpx_sad32x16, vpx_sad32x16_avg,
1741 vpx_variance32x16, vpx_sub_pixel_variance32x16,
1742 vpx_sub_pixel_avg_variance32x16, NULL, NULL, vpx_sad32x16x4d)
1744 BFP(BLOCK_16X32, vpx_sad16x32, vpx_sad16x32_avg,
1745 vpx_variance16x32, vpx_sub_pixel_variance16x32,
1746 vpx_sub_pixel_avg_variance16x32, NULL, NULL, vpx_sad16x32x4d)
1748 BFP(BLOCK_64X32, vpx_sad64x32, vpx_sad64x32_avg,
1749 vpx_variance64x32, vpx_sub_pixel_variance64x32,
1750 vpx_sub_pixel_avg_variance64x32, NULL, NULL, vpx_sad64x32x4d)
1752 BFP(BLOCK_32X64, vpx_sad32x64, vpx_sad32x64_avg,
1753 vpx_variance32x64, vpx_sub_pixel_variance32x64,
1754 vpx_sub_pixel_avg_variance32x64, NULL, NULL, vpx_sad32x64x4d)
1756 BFP(BLOCK_32X32, vpx_sad32x32, vpx_sad32x32_avg,
1757 vpx_variance32x32, vpx_sub_pixel_variance32x32,
1758 vpx_sub_pixel_avg_variance32x32, vpx_sad32x32x3, vpx_sad32x32x8,
1761 BFP(BLOCK_64X64, vpx_sad64x64, vpx_sad64x64_avg,
1762 vpx_variance64x64, vpx_sub_pixel_variance64x64,
1763 vpx_sub_pixel_avg_variance64x64, vpx_sad64x64x3, vpx_sad64x64x8,
1766 BFP(BLOCK_16X16, vpx_sad16x16, vpx_sad16x16_avg,
1767 vpx_variance16x16, vpx_sub_pixel_variance16x16,
1768 vpx_sub_pixel_avg_variance16x16, vpx_sad16x16x3, vpx_sad16x16x8,
1771 BFP(BLOCK_16X8, vpx_sad16x8, vpx_sad16x8_avg,
1772 vpx_variance16x8, vpx_sub_pixel_variance16x8,
1773 vpx_sub_pixel_avg_variance16x8,
1774 vpx_sad16x8x3, vpx_sad16x8x8, vpx_sad16x8x4d)
1776 BFP(BLOCK_8X16, vpx_sad8x16, vpx_sad8x16_avg,
1777 vpx_variance8x16, vpx_sub_pixel_variance8x16,
1778 vpx_sub_pixel_avg_variance8x16,
1779 vpx_sad8x16x3, vpx_sad8x16x8, vpx_sad8x16x4d)
1781 BFP(BLOCK_8X8, vpx_sad8x8, vpx_sad8x8_avg,
1782 vpx_variance8x8, vpx_sub_pixel_variance8x8,
1783 vpx_sub_pixel_avg_variance8x8,
1784 vpx_sad8x8x3, vpx_sad8x8x8, vpx_sad8x8x4d)
1786 BFP(BLOCK_8X4, vpx_sad8x4, vpx_sad8x4_avg,
1787 vpx_variance8x4, vpx_sub_pixel_variance8x4,
1788 vpx_sub_pixel_avg_variance8x4, NULL, vpx_sad8x4x8, vpx_sad8x4x4d)
1790 BFP(BLOCK_4X8, vpx_sad4x8, vpx_sad4x8_avg,
1791 vpx_variance4x8, vpx_sub_pixel_variance4x8,
1792 vpx_sub_pixel_avg_variance4x8, NULL, vpx_sad4x8x8, vpx_sad4x8x4d)
1794 BFP(BLOCK_4X4, vpx_sad4x4, vpx_sad4x4_avg,
1795 vpx_variance4x4, vpx_sub_pixel_variance4x4,
1796 vpx_sub_pixel_avg_variance4x4,
1797 vpx_sad4x4x3, vpx_sad4x4x8, vpx_sad4x4x4d)
1799 #if CONFIG_VP9_HIGHBITDEPTH
1800 highbd_set_var_fns(cpi);
1803 /* vp10_init_quantizer() is first called here. Add check in
1804 * vp10_frame_init_quantizer() so that vp10_init_quantizer is only
1805 * called later when needed. This will avoid unnecessary calls of
1806 * vp10_init_quantizer() for every frame.
1808 vp10_init_quantizer(cpi);
1810 vp10_loop_filter_init(cm);
1812 cm->error.setjmp = 0;
1816 #define SNPRINT(H, T) \
1817 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T))
1819 #define SNPRINT2(H, T, V) \
1820 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V))
1822 void vp10_remove_compressor(VP10_COMP *cpi) {
1823 VP10_COMMON *const cm = &cpi->common;
1830 if (cpi && (cm->current_video_frame > 0)) {
1831 #if CONFIG_INTERNAL_STATS
1832 vpx_clear_system_state();
1834 if (cpi->oxcf.pass != 1) {
1835 char headings[512] = {0};
1836 char results[512] = {0};
1837 FILE *f = fopen("opsnr.stt", "a");
1838 double time_encoded = (cpi->last_end_time_stamp_seen
1839 - cpi->first_time_stamp_ever) / 10000000.000;
1840 double total_encode_time = (cpi->time_receive_data +
1841 cpi->time_compress_data) / 1000.000;
1843 (double)cpi->bytes * (double) 8 / (double)1000 / time_encoded;
1844 const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1);
1846 if (cpi->b_calculate_psnr) {
1847 const double total_psnr =
1848 vpx_sse_to_psnr((double)cpi->total_samples, peak,
1849 (double)cpi->total_sq_error);
1850 const double totalp_psnr =
1851 vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
1852 (double)cpi->totalp_sq_error);
1853 const double total_ssim = 100 * pow(cpi->summed_quality /
1854 cpi->summed_weights, 8.0);
1855 const double totalp_ssim = 100 * pow(cpi->summedp_quality /
1856 cpi->summedp_weights, 8.0);
1858 snprintf(headings, sizeof(headings),
1859 "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t"
1860 "VPXSSIM\tVPSSIMP\tFASTSIM\tPSNRHVS\t"
1861 "WstPsnr\tWstSsim\tWstFast\tWstHVS");
1862 snprintf(results, sizeof(results),
1863 "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1864 "%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1865 "%7.3f\t%7.3f\t%7.3f\t%7.3f",
1866 dr, cpi->psnr.stat[ALL] / cpi->count, total_psnr,
1867 cpi->psnrp.stat[ALL] / cpi->count, totalp_psnr,
1868 total_ssim, totalp_ssim,
1869 cpi->fastssim.stat[ALL] / cpi->count,
1870 cpi->psnrhvs.stat[ALL] / cpi->count,
1871 cpi->psnr.worst, cpi->worst_ssim, cpi->fastssim.worst,
1872 cpi->psnrhvs.worst);
1874 if (cpi->b_calculate_blockiness) {
1875 SNPRINT(headings, "\t Block\tWstBlck");
1876 SNPRINT2(results, "\t%7.3f", cpi->total_blockiness / cpi->count);
1877 SNPRINT2(results, "\t%7.3f", cpi->worst_blockiness);
1880 if (cpi->b_calculate_consistency) {
1881 double consistency =
1882 vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
1883 (double)cpi->total_inconsistency);
1885 SNPRINT(headings, "\tConsist\tWstCons");
1886 SNPRINT2(results, "\t%7.3f", consistency);
1887 SNPRINT2(results, "\t%7.3f", cpi->worst_consistency);
1890 if (cpi->b_calculate_ssimg) {
1891 SNPRINT(headings, "\t SSIMG\tWtSSIMG");
1892 SNPRINT2(results, "\t%7.3f", cpi->ssimg.stat[ALL] / cpi->count);
1893 SNPRINT2(results, "\t%7.3f", cpi->ssimg.worst);
1896 fprintf(f, "%s\t Time\n", headings);
1897 fprintf(f, "%s\t%8.0f\n", results, total_encode_time);
1907 printf("\n_pick_loop_filter_level:%d\n", cpi->time_pick_lpf / 1000);
1908 printf("\n_frames recive_data encod_mb_row compress_frame Total\n");
1909 printf("%6d %10ld %10ld %10ld %10ld\n", cpi->common.current_video_frame,
1910 cpi->time_receive_data / 1000, cpi->time_encode_sb_row / 1000,
1911 cpi->time_compress_data / 1000,
1912 (cpi->time_receive_data + cpi->time_compress_data) / 1000);
1917 #if CONFIG_VP9_TEMPORAL_DENOISING
1918 vp10_denoiser_free(&(cpi->denoiser));
1921 for (t = 0; t < cpi->num_workers; ++t) {
1922 VPxWorker *const worker = &cpi->workers[t];
1923 EncWorkerData *const thread_data = &cpi->tile_thr_data[t];
1925 // Deallocate allocated threads.
1926 vpx_get_worker_interface()->end(worker);
1928 // Deallocate allocated thread data.
1929 if (t < cpi->num_workers - 1) {
1930 vpx_free(thread_data->td->counts);
1931 vp10_free_pc_tree(thread_data->td);
1932 vpx_free(thread_data->td);
1935 vpx_free(cpi->tile_thr_data);
1936 vpx_free(cpi->workers);
1938 if (cpi->num_workers > 1)
1939 vp10_loop_filter_dealloc(&cpi->lf_row_sync);
1941 dealloc_compressor_data(cpi);
1943 for (i = 0; i < sizeof(cpi->mbgraph_stats) /
1944 sizeof(cpi->mbgraph_stats[0]); ++i) {
1945 vpx_free(cpi->mbgraph_stats[i].mb_stats);
1948 #if CONFIG_FP_MB_STATS
1949 if (cpi->use_fp_mb_stats) {
1950 vpx_free(cpi->twopass.frame_mb_stats_buf);
1951 cpi->twopass.frame_mb_stats_buf = NULL;
1955 vp10_remove_common(cm);
1956 vp10_free_ref_frame_buffers(cm->buffer_pool);
1957 #if CONFIG_VP9_POSTPROC
1958 vp10_free_postproc_buffers(cm);
1962 #if CONFIG_VP9_TEMPORAL_DENOISING
1963 #ifdef OUTPUT_YUV_DENOISED
1964 fclose(yuv_denoised_file);
1967 #ifdef OUTPUT_YUV_SKINMAP
1968 fclose(yuv_skinmap_file);
1970 #ifdef OUTPUT_YUV_REC
1971 fclose(yuv_rec_file);
1988 /* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
1989 * and highbd_8_variance(). It should not.
1991 static void encoder_variance(const uint8_t *a, int a_stride,
1992 const uint8_t *b, int b_stride,
1993 int w, int h, unsigned int *sse, int *sum) {
1999 for (i = 0; i < h; i++) {
2000 for (j = 0; j < w; j++) {
2001 const int diff = a[j] - b[j];
2003 *sse += diff * diff;
2011 #if CONFIG_VP9_HIGHBITDEPTH
2012 static void encoder_highbd_variance64(const uint8_t *a8, int a_stride,
2013 const uint8_t *b8, int b_stride,
2014 int w, int h, uint64_t *sse,
2018 uint16_t *a = CONVERT_TO_SHORTPTR(a8);
2019 uint16_t *b = CONVERT_TO_SHORTPTR(b8);
2023 for (i = 0; i < h; i++) {
2024 for (j = 0; j < w; j++) {
2025 const int diff = a[j] - b[j];
2027 *sse += diff * diff;
2034 static void encoder_highbd_8_variance(const uint8_t *a8, int a_stride,
2035 const uint8_t *b8, int b_stride,
2037 unsigned int *sse, int *sum) {
2038 uint64_t sse_long = 0;
2039 uint64_t sum_long = 0;
2040 encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
2041 &sse_long, &sum_long);
2042 *sse = (unsigned int)sse_long;
2043 *sum = (int)sum_long;
2045 #endif // CONFIG_VP9_HIGHBITDEPTH
2047 static int64_t get_sse(const uint8_t *a, int a_stride,
2048 const uint8_t *b, int b_stride,
2049 int width, int height) {
2050 const int dw = width % 16;
2051 const int dh = height % 16;
2052 int64_t total_sse = 0;
2053 unsigned int sse = 0;
2058 encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
2059 dw, height, &sse, &sum);
2064 encoder_variance(&a[(height - dh) * a_stride], a_stride,
2065 &b[(height - dh) * b_stride], b_stride,
2066 width - dw, dh, &sse, &sum);
2070 for (y = 0; y < height / 16; ++y) {
2071 const uint8_t *pa = a;
2072 const uint8_t *pb = b;
2073 for (x = 0; x < width / 16; ++x) {
2074 vpx_mse16x16(pa, a_stride, pb, b_stride, &sse);
2088 #if CONFIG_VP9_HIGHBITDEPTH
2089 static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
2090 const uint8_t *b8, int b_stride,
2091 int width, int height,
2092 unsigned int input_shift) {
2093 const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
2094 const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
2095 int64_t total_sse = 0;
2097 for (y = 0; y < height; ++y) {
2098 for (x = 0; x < width; ++x) {
2100 diff = (a[x] >> input_shift) - (b[x] >> input_shift);
2101 total_sse += diff * diff;
2109 static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
2110 const uint8_t *b, int b_stride,
2111 int width, int height) {
2112 int64_t total_sse = 0;
2114 const int dw = width % 16;
2115 const int dh = height % 16;
2116 unsigned int sse = 0;
2119 encoder_highbd_8_variance(&a[width - dw], a_stride,
2120 &b[width - dw], b_stride,
2121 dw, height, &sse, &sum);
2125 encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
2126 &b[(height - dh) * b_stride], b_stride,
2127 width - dw, dh, &sse, &sum);
2130 for (y = 0; y < height / 16; ++y) {
2131 const uint8_t *pa = a;
2132 const uint8_t *pb = b;
2133 for (x = 0; x < width / 16; ++x) {
2134 vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
2144 #endif // CONFIG_VP9_HIGHBITDEPTH
2147 double psnr[4]; // total/y/u/v
2148 uint64_t sse[4]; // total/y/u/v
2149 uint32_t samples[4]; // total/y/u/v
2152 #if CONFIG_VP9_HIGHBITDEPTH
2153 static void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
2154 const YV12_BUFFER_CONFIG *b,
2156 unsigned int bit_depth,
2157 unsigned int in_bit_depth) {
2158 const int widths[3] =
2159 {a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
2160 const int heights[3] =
2161 {a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
2162 const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer };
2163 const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
2164 const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer };
2165 const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
2167 uint64_t total_sse = 0;
2168 uint32_t total_samples = 0;
2169 const double peak = (double)((1 << in_bit_depth) - 1);
2170 const unsigned int input_shift = bit_depth - in_bit_depth;
2172 for (i = 0; i < 3; ++i) {
2173 const int w = widths[i];
2174 const int h = heights[i];
2175 const uint32_t samples = w * h;
2177 if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
2179 sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
2180 b_planes[i], b_strides[i], w, h,
2183 sse = highbd_get_sse(a_planes[i], a_strides[i],
2184 b_planes[i], b_strides[i], w, h);
2187 sse = get_sse(a_planes[i], a_strides[i],
2188 b_planes[i], b_strides[i],
2191 psnr->sse[1 + i] = sse;
2192 psnr->samples[1 + i] = samples;
2193 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
2196 total_samples += samples;
2199 psnr->sse[0] = total_sse;
2200 psnr->samples[0] = total_samples;
2201 psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
2205 #else // !CONFIG_VP9_HIGHBITDEPTH
2207 static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
2209 static const double peak = 255.0;
2210 const int widths[3] = {
2211 a->y_crop_width, a->uv_crop_width, a->uv_crop_width};
2212 const int heights[3] = {
2213 a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
2214 const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer};
2215 const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
2216 const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer};
2217 const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
2219 uint64_t total_sse = 0;
2220 uint32_t total_samples = 0;
2222 for (i = 0; i < 3; ++i) {
2223 const int w = widths[i];
2224 const int h = heights[i];
2225 const uint32_t samples = w * h;
2226 const uint64_t sse = get_sse(a_planes[i], a_strides[i],
2227 b_planes[i], b_strides[i],
2229 psnr->sse[1 + i] = sse;
2230 psnr->samples[1 + i] = samples;
2231 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
2234 total_samples += samples;
2237 psnr->sse[0] = total_sse;
2238 psnr->samples[0] = total_samples;
2239 psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
2242 #endif // CONFIG_VP9_HIGHBITDEPTH
2244 static void generate_psnr_packet(VP10_COMP *cpi) {
2245 struct vpx_codec_cx_pkt pkt;
2248 #if CONFIG_VP9_HIGHBITDEPTH
2249 calc_highbd_psnr(cpi->Source, cpi->common.frame_to_show, &psnr,
2250 cpi->td.mb.e_mbd.bd, cpi->oxcf.input_bit_depth);
2252 calc_psnr(cpi->Source, cpi->common.frame_to_show, &psnr);
2255 for (i = 0; i < 4; ++i) {
2256 pkt.data.psnr.samples[i] = psnr.samples[i];
2257 pkt.data.psnr.sse[i] = psnr.sse[i];
2258 pkt.data.psnr.psnr[i] = psnr.psnr[i];
2260 pkt.kind = VPX_CODEC_PSNR_PKT;
2261 vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
2264 int vp10_use_as_reference(VP10_COMP *cpi, int ref_frame_flags) {
2265 if (ref_frame_flags > 7)
2268 cpi->ref_frame_flags = ref_frame_flags;
2272 void vp10_update_reference(VP10_COMP *cpi, int ref_frame_flags) {
2273 cpi->ext_refresh_golden_frame = (ref_frame_flags & VP9_GOLD_FLAG) != 0;
2274 cpi->ext_refresh_alt_ref_frame = (ref_frame_flags & VP9_ALT_FLAG) != 0;
2275 cpi->ext_refresh_last_frame = (ref_frame_flags & VP9_LAST_FLAG) != 0;
2276 cpi->ext_refresh_frame_flags_pending = 1;
2279 static YV12_BUFFER_CONFIG *get_vp10_ref_frame_buffer(VP10_COMP *cpi,
2280 VP9_REFFRAME ref_frame_flag) {
2281 MV_REFERENCE_FRAME ref_frame = NONE;
2282 if (ref_frame_flag == VP9_LAST_FLAG)
2283 ref_frame = LAST_FRAME;
2284 else if (ref_frame_flag == VP9_GOLD_FLAG)
2285 ref_frame = GOLDEN_FRAME;
2286 else if (ref_frame_flag == VP9_ALT_FLAG)
2287 ref_frame = ALTREF_FRAME;
2289 return ref_frame == NONE ? NULL : get_ref_frame_buffer(cpi, ref_frame);
2292 int vp10_copy_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
2293 YV12_BUFFER_CONFIG *sd) {
2294 YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
2296 vp8_yv12_copy_frame(cfg, sd);
2303 int vp10_set_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
2304 YV12_BUFFER_CONFIG *sd) {
2305 YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
2307 vp8_yv12_copy_frame(sd, cfg);
2314 int vp10_update_entropy(VP10_COMP * cpi, int update) {
2315 cpi->ext_refresh_frame_context = update;
2316 cpi->ext_refresh_frame_context_pending = 1;
2320 #if defined(OUTPUT_YUV_DENOISED) || defined(OUTPUT_YUV_SKINMAP)
2321 // The denoiser buffer is allocated as a YUV 440 buffer. This function writes it
2322 // as YUV 420. We simply use the top-left pixels of the UV buffers, since we do
2323 // not denoise the UV channels at this time. If ever we implement UV channel
2324 // denoising we will have to modify this.
2325 void vp10_write_yuv_frame_420(YV12_BUFFER_CONFIG *s, FILE *f) {
2326 uint8_t *src = s->y_buffer;
2327 int h = s->y_height;
2330 fwrite(src, s->y_width, 1, f);
2338 fwrite(src, s->uv_width, 1, f);
2339 src += s->uv_stride;
2346 fwrite(src, s->uv_width, 1, f);
2347 src += s->uv_stride;
2352 #ifdef OUTPUT_YUV_REC
2353 void vp10_write_yuv_rec_frame(VP10_COMMON *cm) {
2354 YV12_BUFFER_CONFIG *s = cm->frame_to_show;
2355 uint8_t *src = s->y_buffer;
2358 #if CONFIG_VP9_HIGHBITDEPTH
2359 if (s->flags & YV12_FLAG_HIGHBITDEPTH) {
2360 uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer);
2363 fwrite(src16, s->y_width, 2, yuv_rec_file);
2364 src16 += s->y_stride;
2367 src16 = CONVERT_TO_SHORTPTR(s->u_buffer);
2371 fwrite(src16, s->uv_width, 2, yuv_rec_file);
2372 src16 += s->uv_stride;
2375 src16 = CONVERT_TO_SHORTPTR(s->v_buffer);
2379 fwrite(src16, s->uv_width, 2, yuv_rec_file);
2380 src16 += s->uv_stride;
2383 fflush(yuv_rec_file);
2386 #endif // CONFIG_VP9_HIGHBITDEPTH
2389 fwrite(src, s->y_width, 1, yuv_rec_file);
2397 fwrite(src, s->uv_width, 1, yuv_rec_file);
2398 src += s->uv_stride;
2405 fwrite(src, s->uv_width, 1, yuv_rec_file);
2406 src += s->uv_stride;
2409 fflush(yuv_rec_file);
2413 #if CONFIG_VP9_HIGHBITDEPTH
2414 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
2415 YV12_BUFFER_CONFIG *dst,
2418 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
2419 YV12_BUFFER_CONFIG *dst) {
2420 #endif // CONFIG_VP9_HIGHBITDEPTH
2421 // TODO(dkovalev): replace YV12_BUFFER_CONFIG with vpx_image_t
2423 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
2424 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
2425 const int src_widths[3] = {src->y_crop_width, src->uv_crop_width,
2426 src->uv_crop_width };
2427 const int src_heights[3] = {src->y_crop_height, src->uv_crop_height,
2428 src->uv_crop_height};
2429 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
2430 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
2431 const int dst_widths[3] = {dst->y_crop_width, dst->uv_crop_width,
2432 dst->uv_crop_width};
2433 const int dst_heights[3] = {dst->y_crop_height, dst->uv_crop_height,
2434 dst->uv_crop_height};
2436 for (i = 0; i < MAX_MB_PLANE; ++i) {
2437 #if CONFIG_VP9_HIGHBITDEPTH
2438 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
2439 vp10_highbd_resize_plane(srcs[i], src_heights[i], src_widths[i],
2440 src_strides[i], dsts[i], dst_heights[i],
2441 dst_widths[i], dst_strides[i], bd);
2443 vp10_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
2444 dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
2447 vp10_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
2448 dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
2449 #endif // CONFIG_VP9_HIGHBITDEPTH
2451 vpx_extend_frame_borders(dst);
2454 #if CONFIG_VP9_HIGHBITDEPTH
2455 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
2456 YV12_BUFFER_CONFIG *dst, int bd) {
2458 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
2459 YV12_BUFFER_CONFIG *dst) {
2460 #endif // CONFIG_VP9_HIGHBITDEPTH
2461 const int src_w = src->y_crop_width;
2462 const int src_h = src->y_crop_height;
2463 const int dst_w = dst->y_crop_width;
2464 const int dst_h = dst->y_crop_height;
2465 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
2466 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
2467 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
2468 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
2469 const InterpKernel *const kernel = vp10_filter_kernels[EIGHTTAP];
2472 for (y = 0; y < dst_h; y += 16) {
2473 for (x = 0; x < dst_w; x += 16) {
2474 for (i = 0; i < MAX_MB_PLANE; ++i) {
2475 const int factor = (i == 0 || i == 3 ? 1 : 2);
2476 const int x_q4 = x * (16 / factor) * src_w / dst_w;
2477 const int y_q4 = y * (16 / factor) * src_h / dst_h;
2478 const int src_stride = src_strides[i];
2479 const int dst_stride = dst_strides[i];
2480 const uint8_t *src_ptr = srcs[i] + (y / factor) * src_h / dst_h *
2481 src_stride + (x / factor) * src_w / dst_w;
2482 uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
2484 #if CONFIG_VP9_HIGHBITDEPTH
2485 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
2486 vpx_highbd_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2487 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2488 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2489 16 / factor, 16 / factor, bd);
2491 vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2492 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2493 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2494 16 / factor, 16 / factor);
2497 vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2498 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2499 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2500 16 / factor, 16 / factor);
2501 #endif // CONFIG_VP9_HIGHBITDEPTH
2506 vpx_extend_frame_borders(dst);
2509 static int scale_down(VP10_COMP *cpi, int q) {
2510 RATE_CONTROL *const rc = &cpi->rc;
2511 GF_GROUP *const gf_group = &cpi->twopass.gf_group;
2513 assert(frame_is_kf_gf_arf(cpi));
2515 if (rc->frame_size_selector == UNSCALED &&
2516 q >= rc->rf_level_maxq[gf_group->rf_level[gf_group->index]]) {
2517 const int max_size_thresh = (int)(rate_thresh_mult[SCALE_STEP1]
2518 * VPXMAX(rc->this_frame_target, rc->avg_frame_bandwidth));
2519 scale = rc->projected_frame_size > max_size_thresh ? 1 : 0;
2524 // Function to test for conditions that indicate we should loop
2525 // back and recode a frame.
2526 static int recode_loop_test(VP10_COMP *cpi,
2527 int high_limit, int low_limit,
2528 int q, int maxq, int minq) {
2529 const RATE_CONTROL *const rc = &cpi->rc;
2530 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
2531 const int frame_is_kfgfarf = frame_is_kf_gf_arf(cpi);
2532 int force_recode = 0;
2534 if ((rc->projected_frame_size >= rc->max_frame_bandwidth) ||
2535 (cpi->sf.recode_loop == ALLOW_RECODE) ||
2536 (frame_is_kfgfarf &&
2537 (cpi->sf.recode_loop == ALLOW_RECODE_KFARFGF))) {
2538 if (frame_is_kfgfarf &&
2539 (oxcf->resize_mode == RESIZE_DYNAMIC) &&
2540 scale_down(cpi, q)) {
2541 // Code this group at a lower resolution.
2542 cpi->resize_pending = 1;
2546 // TODO(agrange) high_limit could be greater than the scale-down threshold.
2547 if ((rc->projected_frame_size > high_limit && q < maxq) ||
2548 (rc->projected_frame_size < low_limit && q > minq)) {
2550 } else if (cpi->oxcf.rc_mode == VPX_CQ) {
2551 // Deal with frame undershoot and whether or not we are
2552 // below the automatically set cq level.
2553 if (q > oxcf->cq_level &&
2554 rc->projected_frame_size < ((rc->this_frame_target * 7) >> 3)) {
2559 return force_recode;
2562 void vp10_update_reference_frames(VP10_COMP *cpi) {
2563 VP10_COMMON * const cm = &cpi->common;
2564 BufferPool *const pool = cm->buffer_pool;
2566 // At this point the new frame has been encoded.
2567 // If any buffer copy / swapping is signaled it should be done here.
2568 if (cm->frame_type == KEY_FRAME) {
2569 ref_cnt_fb(pool->frame_bufs,
2570 &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
2571 ref_cnt_fb(pool->frame_bufs,
2572 &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
2573 } else if (vp10_preserve_existing_gf(cpi)) {
2574 // We have decided to preserve the previously existing golden frame as our
2575 // new ARF frame. However, in the short term in function
2576 // vp10_bitstream.c::get_refresh_mask() we left it in the GF slot and, if
2577 // we're updating the GF with the current decoded frame, we save it to the
2578 // ARF slot instead.
2579 // We now have to update the ARF with the current frame and swap gld_fb_idx
2580 // and alt_fb_idx so that, overall, we've stored the old GF in the new ARF
2581 // slot and, if we're updating the GF, the current frame becomes the new GF.
2584 ref_cnt_fb(pool->frame_bufs,
2585 &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
2587 tmp = cpi->alt_fb_idx;
2588 cpi->alt_fb_idx = cpi->gld_fb_idx;
2589 cpi->gld_fb_idx = tmp;
2590 } else { /* For non key/golden frames */
2591 if (cpi->refresh_alt_ref_frame) {
2592 int arf_idx = cpi->alt_fb_idx;
2593 if ((cpi->oxcf.pass == 2) && cpi->multi_arf_allowed) {
2594 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
2595 arf_idx = gf_group->arf_update_idx[gf_group->index];
2598 ref_cnt_fb(pool->frame_bufs,
2599 &cm->ref_frame_map[arf_idx], cm->new_fb_idx);
2600 memcpy(cpi->interp_filter_selected[ALTREF_FRAME],
2601 cpi->interp_filter_selected[0],
2602 sizeof(cpi->interp_filter_selected[0]));
2605 if (cpi->refresh_golden_frame) {
2606 ref_cnt_fb(pool->frame_bufs,
2607 &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
2608 if (!cpi->rc.is_src_frame_alt_ref)
2609 memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
2610 cpi->interp_filter_selected[0],
2611 sizeof(cpi->interp_filter_selected[0]));
2613 memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
2614 cpi->interp_filter_selected[ALTREF_FRAME],
2615 sizeof(cpi->interp_filter_selected[ALTREF_FRAME]));
2619 if (cpi->refresh_last_frame) {
2620 ref_cnt_fb(pool->frame_bufs,
2621 &cm->ref_frame_map[cpi->lst_fb_idx], cm->new_fb_idx);
2622 if (!cpi->rc.is_src_frame_alt_ref)
2623 memcpy(cpi->interp_filter_selected[LAST_FRAME],
2624 cpi->interp_filter_selected[0],
2625 sizeof(cpi->interp_filter_selected[0]));
2627 #if CONFIG_VP9_TEMPORAL_DENOISING
2628 if (cpi->oxcf.noise_sensitivity > 0) {
2629 vp10_denoiser_update_frame_info(&cpi->denoiser,
2631 cpi->common.frame_type,
2632 cpi->refresh_alt_ref_frame,
2633 cpi->refresh_golden_frame,
2634 cpi->refresh_last_frame);
2639 static void loopfilter_frame(VP10_COMP *cpi, VP10_COMMON *cm) {
2640 MACROBLOCKD *xd = &cpi->td.mb.e_mbd;
2641 struct loopfilter *lf = &cm->lf;
2643 lf->filter_level = 0;
2645 struct vpx_usec_timer timer;
2647 vpx_clear_system_state();
2649 vpx_usec_timer_start(&timer);
2651 vp10_pick_filter_level(cpi->Source, cpi, cpi->sf.lpf_pick);
2653 vpx_usec_timer_mark(&timer);
2654 cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
2657 if (lf->filter_level > 0) {
2658 if (cpi->num_workers > 1)
2659 vp10_loop_filter_frame_mt(cm->frame_to_show, cm, xd->plane,
2660 lf->filter_level, 0, 0,
2661 cpi->workers, cpi->num_workers,
2664 vp10_loop_filter_frame(cm->frame_to_show, cm, xd, lf->filter_level, 0, 0);
2667 vpx_extend_frame_inner_borders(cm->frame_to_show);
2670 static INLINE void alloc_frame_mvs(const VP10_COMMON *cm,
2672 RefCntBuffer *const new_fb_ptr = &cm->buffer_pool->frame_bufs[buffer_idx];
2673 if (new_fb_ptr->mvs == NULL ||
2674 new_fb_ptr->mi_rows < cm->mi_rows ||
2675 new_fb_ptr->mi_cols < cm->mi_cols) {
2676 vpx_free(new_fb_ptr->mvs);
2678 (MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols,
2679 sizeof(*new_fb_ptr->mvs));
2680 new_fb_ptr->mi_rows = cm->mi_rows;
2681 new_fb_ptr->mi_cols = cm->mi_cols;
2685 void vp10_scale_references(VP10_COMP *cpi) {
2686 VP10_COMMON *cm = &cpi->common;
2687 MV_REFERENCE_FRAME ref_frame;
2688 const VP9_REFFRAME ref_mask[3] = {VP9_LAST_FLAG, VP9_GOLD_FLAG, VP9_ALT_FLAG};
2690 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
2691 // Need to convert from VP9_REFFRAME to index into ref_mask (subtract 1).
2692 if (cpi->ref_frame_flags & ref_mask[ref_frame - 1]) {
2693 BufferPool *const pool = cm->buffer_pool;
2694 const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi,
2698 cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
2702 #if CONFIG_VP9_HIGHBITDEPTH
2703 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
2704 RefCntBuffer *new_fb_ptr = NULL;
2705 int force_scaling = 0;
2706 int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
2707 if (new_fb == INVALID_IDX) {
2708 new_fb = get_free_fb(cm);
2711 if (new_fb == INVALID_IDX)
2713 new_fb_ptr = &pool->frame_bufs[new_fb];
2714 if (force_scaling ||
2715 new_fb_ptr->buf.y_crop_width != cm->width ||
2716 new_fb_ptr->buf.y_crop_height != cm->height) {
2717 vpx_realloc_frame_buffer(&new_fb_ptr->buf,
2718 cm->width, cm->height,
2719 cm->subsampling_x, cm->subsampling_y,
2720 cm->use_highbitdepth,
2721 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
2723 scale_and_extend_frame(ref, &new_fb_ptr->buf, (int)cm->bit_depth);
2724 cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
2725 alloc_frame_mvs(cm, new_fb);
2728 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
2729 RefCntBuffer *new_fb_ptr = NULL;
2730 int force_scaling = 0;
2731 int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
2732 if (new_fb == INVALID_IDX) {
2733 new_fb = get_free_fb(cm);
2736 if (new_fb == INVALID_IDX)
2738 new_fb_ptr = &pool->frame_bufs[new_fb];
2739 if (force_scaling ||
2740 new_fb_ptr->buf.y_crop_width != cm->width ||
2741 new_fb_ptr->buf.y_crop_height != cm->height) {
2742 vpx_realloc_frame_buffer(&new_fb_ptr->buf,
2743 cm->width, cm->height,
2744 cm->subsampling_x, cm->subsampling_y,
2745 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
2747 scale_and_extend_frame(ref, &new_fb_ptr->buf);
2748 cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
2749 alloc_frame_mvs(cm, new_fb);
2751 #endif // CONFIG_VP9_HIGHBITDEPTH
2753 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
2754 RefCntBuffer *const buf = &pool->frame_bufs[buf_idx];
2755 buf->buf.y_crop_width = ref->y_crop_width;
2756 buf->buf.y_crop_height = ref->y_crop_height;
2757 cpi->scaled_ref_idx[ref_frame - 1] = buf_idx;
2761 if (cpi->oxcf.pass != 0)
2762 cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
2767 static void release_scaled_references(VP10_COMP *cpi) {
2768 VP10_COMMON *cm = &cpi->common;
2770 if (cpi->oxcf.pass == 0) {
2771 // Only release scaled references under certain conditions:
2772 // if reference will be updated, or if scaled reference has same resolution.
2774 refresh[0] = (cpi->refresh_last_frame) ? 1 : 0;
2775 refresh[1] = (cpi->refresh_golden_frame) ? 1 : 0;
2776 refresh[2] = (cpi->refresh_alt_ref_frame) ? 1 : 0;
2777 for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
2778 const int idx = cpi->scaled_ref_idx[i - 1];
2779 RefCntBuffer *const buf = idx != INVALID_IDX ?
2780 &cm->buffer_pool->frame_bufs[idx] : NULL;
2781 const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi, i);
2784 (buf->buf.y_crop_width == ref->y_crop_width &&
2785 buf->buf.y_crop_height == ref->y_crop_height))) {
2787 cpi->scaled_ref_idx[i -1] = INVALID_IDX;
2791 for (i = 0; i < MAX_REF_FRAMES; ++i) {
2792 const int idx = cpi->scaled_ref_idx[i];
2793 RefCntBuffer *const buf = idx != INVALID_IDX ?
2794 &cm->buffer_pool->frame_bufs[idx] : NULL;
2797 cpi->scaled_ref_idx[i] = INVALID_IDX;
2803 static void full_to_model_count(unsigned int *model_count,
2804 unsigned int *full_count) {
2806 model_count[ZERO_TOKEN] = full_count[ZERO_TOKEN];
2807 model_count[ONE_TOKEN] = full_count[ONE_TOKEN];
2808 model_count[TWO_TOKEN] = full_count[TWO_TOKEN];
2809 for (n = THREE_TOKEN; n < EOB_TOKEN; ++n)
2810 model_count[TWO_TOKEN] += full_count[n];
2811 model_count[EOB_MODEL_TOKEN] = full_count[EOB_TOKEN];
2814 static void full_to_model_counts(vp10_coeff_count_model *model_count,
2815 vp10_coeff_count *full_count) {
2818 for (i = 0; i < PLANE_TYPES; ++i)
2819 for (j = 0; j < REF_TYPES; ++j)
2820 for (k = 0; k < COEF_BANDS; ++k)
2821 for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
2822 full_to_model_count(model_count[i][j][k][l], full_count[i][j][k][l]);
2825 #if 0 && CONFIG_INTERNAL_STATS
2826 static void output_frame_level_debug_stats(VP10_COMP *cpi) {
2827 VP10_COMMON *const cm = &cpi->common;
2828 FILE *const f = fopen("tmp.stt", cm->current_video_frame ? "a" : "w");
2831 vpx_clear_system_state();
2833 recon_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
2835 if (cpi->twopass.total_left_stats.coded_error != 0.0)
2836 fprintf(f, "%10u %dx%d %d %d %10d %10d %10d %10d"
2837 "%10"PRId64" %10"PRId64" %5d %5d %10"PRId64" "
2838 "%10"PRId64" %10"PRId64" %10d "
2839 "%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
2840 "%6d %6d %5d %5d %5d "
2841 "%10"PRId64" %10.3lf"
2842 "%10lf %8u %10"PRId64" %10d %10d %10d\n",
2843 cpi->common.current_video_frame,
2844 cm->width, cm->height,
2845 cpi->rc.source_alt_ref_pending,
2846 cpi->rc.source_alt_ref_active,
2847 cpi->rc.this_frame_target,
2848 cpi->rc.projected_frame_size,
2849 cpi->rc.projected_frame_size / cpi->common.MBs,
2850 (cpi->rc.projected_frame_size - cpi->rc.this_frame_target),
2851 cpi->rc.vbr_bits_off_target,
2852 cpi->rc.vbr_bits_off_target_fast,
2853 cpi->twopass.extend_minq,
2854 cpi->twopass.extend_minq_fast,
2855 cpi->rc.total_target_vs_actual,
2856 (cpi->rc.starting_buffer_level - cpi->rc.bits_off_target),
2857 cpi->rc.total_actual_bits, cm->base_qindex,
2858 vp10_convert_qindex_to_q(cm->base_qindex, cm->bit_depth),
2859 (double)vp10_dc_quant(cm->base_qindex, 0, cm->bit_depth) / 4.0,
2860 vp10_convert_qindex_to_q(cpi->twopass.active_worst_quality,
2863 vp10_convert_qindex_to_q(cpi->oxcf.cq_level, cm->bit_depth),
2864 cpi->refresh_last_frame, cpi->refresh_golden_frame,
2865 cpi->refresh_alt_ref_frame, cm->frame_type, cpi->rc.gfu_boost,
2866 cpi->twopass.bits_left,
2867 cpi->twopass.total_left_stats.coded_error,
2868 cpi->twopass.bits_left /
2869 (1 + cpi->twopass.total_left_stats.coded_error),
2870 cpi->tot_recode_hits, recon_err, cpi->rc.kf_boost,
2871 cpi->twopass.kf_zeromotion_pct,
2872 cpi->twopass.fr_content_type);
2877 FILE *const fmodes = fopen("Modes.stt", "a");
2880 fprintf(fmodes, "%6d:%1d:%1d:%1d ", cpi->common.current_video_frame,
2881 cm->frame_type, cpi->refresh_golden_frame,
2882 cpi->refresh_alt_ref_frame);
2884 for (i = 0; i < MAX_MODES; ++i)
2885 fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
2887 fprintf(fmodes, "\n");
2894 static void set_mv_search_params(VP10_COMP *cpi) {
2895 const VP10_COMMON *const cm = &cpi->common;
2896 const unsigned int max_mv_def = VPXMIN(cm->width, cm->height);
2898 // Default based on max resolution.
2899 cpi->mv_step_param = vp10_init_search_range(max_mv_def);
2901 if (cpi->sf.mv.auto_mv_step_size) {
2902 if (frame_is_intra_only(cm)) {
2903 // Initialize max_mv_magnitude for use in the first INTER frame
2904 // after a key/intra-only frame.
2905 cpi->max_mv_magnitude = max_mv_def;
2907 if (cm->show_frame) {
2908 // Allow mv_steps to correspond to twice the max mv magnitude found
2909 // in the previous frame, capped by the default max_mv_magnitude based
2911 cpi->mv_step_param = vp10_init_search_range(
2912 VPXMIN(max_mv_def, 2 * cpi->max_mv_magnitude));
2914 cpi->max_mv_magnitude = 0;
2919 static void set_size_independent_vars(VP10_COMP *cpi) {
2920 vp10_set_speed_features_framesize_independent(cpi);
2921 vp10_set_rd_speed_thresholds(cpi);
2922 vp10_set_rd_speed_thresholds_sub8x8(cpi);
2923 cpi->common.interp_filter = cpi->sf.default_interp_filter;
2926 static void set_size_dependent_vars(VP10_COMP *cpi, int *q,
2927 int *bottom_index, int *top_index) {
2928 VP10_COMMON *const cm = &cpi->common;
2929 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
2931 // Setup variables that depend on the dimensions of the frame.
2932 vp10_set_speed_features_framesize_dependent(cpi);
2934 // Decide q and q bounds.
2935 *q = vp10_rc_pick_q_and_bounds(cpi, bottom_index, top_index);
2937 if (!frame_is_intra_only(cm)) {
2938 vp10_set_high_precision_mv(cpi, (*q) < HIGH_PRECISION_MV_QTHRESH);
2941 // Configure experimental use of segmentation for enhanced coding of
2942 // static regions if indicated.
2943 // Only allowed in the second pass of a two pass encode, as it requires
2944 // lagged coding, and if the relevant speed feature flag is set.
2945 if (oxcf->pass == 2 && cpi->sf.static_segmentation)
2946 configure_static_seg_features(cpi);
2948 #if CONFIG_VP9_POSTPROC
2949 if (oxcf->noise_sensitivity > 0) {
2951 switch (oxcf->noise_sensitivity) {
2969 vp10_denoise(cpi->Source, cpi->Source, l);
2971 #endif // CONFIG_VP9_POSTPROC
2974 static void init_motion_estimation(VP10_COMP *cpi) {
2975 int y_stride = cpi->scaled_source.y_stride;
2977 if (cpi->sf.mv.search_method == NSTEP) {
2978 vp10_init3smotion_compensation(&cpi->ss_cfg, y_stride);
2979 } else if (cpi->sf.mv.search_method == DIAMOND) {
2980 vp10_init_dsmotion_compensation(&cpi->ss_cfg, y_stride);
2984 static void set_frame_size(VP10_COMP *cpi) {
2986 VP10_COMMON *const cm = &cpi->common;
2987 VP10EncoderConfig *const oxcf = &cpi->oxcf;
2988 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
2990 if (oxcf->pass == 2 &&
2991 oxcf->rc_mode == VPX_VBR &&
2992 ((oxcf->resize_mode == RESIZE_FIXED && cm->current_video_frame == 0) ||
2993 (oxcf->resize_mode == RESIZE_DYNAMIC && cpi->resize_pending))) {
2994 vp10_calculate_coded_size(
2995 cpi, &oxcf->scaled_frame_width, &oxcf->scaled_frame_height);
2997 // There has been a change in frame size.
2998 vp10_set_size_literal(cpi, oxcf->scaled_frame_width,
2999 oxcf->scaled_frame_height);
3002 if (oxcf->pass == 0 &&
3003 oxcf->rc_mode == VPX_CBR &&
3004 oxcf->resize_mode == RESIZE_DYNAMIC) {
3005 if (cpi->resize_pending == 1) {
3006 oxcf->scaled_frame_width =
3007 (cm->width * cpi->resize_scale_num) / cpi->resize_scale_den;
3008 oxcf->scaled_frame_height =
3009 (cm->height * cpi->resize_scale_num) /cpi->resize_scale_den;
3010 } else if (cpi->resize_pending == -1) {
3011 // Go back up to original size.
3012 oxcf->scaled_frame_width = oxcf->width;
3013 oxcf->scaled_frame_height = oxcf->height;
3015 if (cpi->resize_pending != 0) {
3016 // There has been a change in frame size.
3017 vp10_set_size_literal(cpi,
3018 oxcf->scaled_frame_width,
3019 oxcf->scaled_frame_height);
3021 // TODO(agrange) Scale cpi->max_mv_magnitude if frame-size has changed.
3022 set_mv_search_params(cpi);
3026 if (oxcf->pass == 2) {
3027 vp10_set_target_rate(cpi);
3030 alloc_frame_mvs(cm, cm->new_fb_idx);
3032 // Reset the frame pointers to the current frame size.
3033 vpx_realloc_frame_buffer(get_frame_new_buffer(cm),
3034 cm->width, cm->height,
3035 cm->subsampling_x, cm->subsampling_y,
3036 #if CONFIG_VP9_HIGHBITDEPTH
3037 cm->use_highbitdepth,
3039 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
3042 alloc_util_frame_buffers(cpi);
3043 init_motion_estimation(cpi);
3045 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
3046 RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - 1];
3047 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
3049 ref_buf->idx = buf_idx;
3051 if (buf_idx != INVALID_IDX) {
3052 YV12_BUFFER_CONFIG *const buf = &cm->buffer_pool->frame_bufs[buf_idx].buf;
3054 #if CONFIG_VP9_HIGHBITDEPTH
3055 vp10_setup_scale_factors_for_frame(&ref_buf->sf,
3056 buf->y_crop_width, buf->y_crop_height,
3057 cm->width, cm->height,
3058 (buf->flags & YV12_FLAG_HIGHBITDEPTH) ?
3061 vp10_setup_scale_factors_for_frame(&ref_buf->sf,
3062 buf->y_crop_width, buf->y_crop_height,
3063 cm->width, cm->height);
3064 #endif // CONFIG_VP9_HIGHBITDEPTH
3065 if (vp10_is_scaled(&ref_buf->sf))
3066 vpx_extend_frame_borders(buf);
3068 ref_buf->buf = NULL;
3072 set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME);
3075 static void encode_without_recode_loop(VP10_COMP *cpi) {
3076 VP10_COMMON *const cm = &cpi->common;
3077 int q = 0, bottom_index = 0, top_index = 0; // Dummy variables.
3079 vpx_clear_system_state();
3081 set_frame_size(cpi);
3083 // For 1 pass CBR under dynamic resize mode: use faster scaling for source.
3084 // Only for 2x2 scaling for now.
3085 if (cpi->oxcf.pass == 0 &&
3086 cpi->oxcf.rc_mode == VPX_CBR &&
3087 cpi->oxcf.resize_mode == RESIZE_DYNAMIC &&
3088 cpi->un_scaled_source->y_width == (cm->width << 1) &&
3089 cpi->un_scaled_source->y_height == (cm->height << 1)) {
3090 cpi->Source = vp10_scale_if_required_fast(cm,
3091 cpi->un_scaled_source,
3092 &cpi->scaled_source);
3093 if (cpi->unscaled_last_source != NULL)
3094 cpi->Last_Source = vp10_scale_if_required_fast(cm,
3095 cpi->unscaled_last_source,
3096 &cpi->scaled_last_source);
3098 cpi->Source = vp10_scale_if_required(cm, cpi->un_scaled_source,
3099 &cpi->scaled_source);
3100 if (cpi->unscaled_last_source != NULL)
3101 cpi->Last_Source = vp10_scale_if_required(cm, cpi->unscaled_last_source,
3102 &cpi->scaled_last_source);
3105 if (frame_is_intra_only(cm) == 0) {
3106 vp10_scale_references(cpi);
3109 set_size_independent_vars(cpi);
3110 set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
3112 vp10_set_quantizer(cm, q);
3113 vp10_set_variance_partition_thresholds(cpi, q);
3117 suppress_active_map(cpi);
3118 // Variance adaptive and in frame q adjustment experiments are mutually
3120 if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
3121 vp10_vaq_frame_setup(cpi);
3122 } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
3123 vp10_setup_in_frame_q_adj(cpi);
3124 } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
3125 vp10_cyclic_refresh_setup(cpi);
3127 apply_active_map(cpi);
3129 // transform / motion compensation build reconstruction frame
3130 vp10_encode_frame(cpi);
3132 // Update some stats from cyclic refresh, and check if we should not update
3133 // golden reference, for 1 pass CBR.
3134 if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
3135 cm->frame_type != KEY_FRAME &&
3136 (cpi->oxcf.pass == 0 && cpi->oxcf.rc_mode == VPX_CBR))
3137 vp10_cyclic_refresh_check_golden_update(cpi);
3139 // Update the skip mb flag probabilities based on the distribution
3140 // seen in the last encoder iteration.
3141 // update_base_skip_probs(cpi);
3142 vpx_clear_system_state();
3145 static void encode_with_recode_loop(VP10_COMP *cpi,
3148 VP10_COMMON *const cm = &cpi->common;
3149 RATE_CONTROL *const rc = &cpi->rc;
3150 int bottom_index, top_index;
3152 int loop_at_this_size = 0;
3154 int overshoot_seen = 0;
3155 int undershoot_seen = 0;
3156 int frame_over_shoot_limit;
3157 int frame_under_shoot_limit;
3158 int q = 0, q_low = 0, q_high = 0;
3160 set_size_independent_vars(cpi);
3163 vpx_clear_system_state();
3165 set_frame_size(cpi);
3167 if (loop_count == 0 || cpi->resize_pending != 0) {
3168 set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
3170 // TODO(agrange) Scale cpi->max_mv_magnitude if frame-size has changed.
3171 set_mv_search_params(cpi);
3173 // Reset the loop state for new frame size.
3175 undershoot_seen = 0;
3177 // Reconfiguration for change in frame size has concluded.
3178 cpi->resize_pending = 0;
3180 q_low = bottom_index;
3183 loop_at_this_size = 0;
3186 // Decide frame size bounds first time through.
3187 if (loop_count == 0) {
3188 vp10_rc_compute_frame_size_bounds(cpi, rc->this_frame_target,
3189 &frame_under_shoot_limit,
3190 &frame_over_shoot_limit);
3193 cpi->Source = vp10_scale_if_required(cm, cpi->un_scaled_source,
3194 &cpi->scaled_source);
3196 if (cpi->unscaled_last_source != NULL)
3197 cpi->Last_Source = vp10_scale_if_required(cm, cpi->unscaled_last_source,
3198 &cpi->scaled_last_source);
3200 if (frame_is_intra_only(cm) == 0) {
3201 if (loop_count > 0) {
3202 release_scaled_references(cpi);
3204 vp10_scale_references(cpi);
3207 vp10_set_quantizer(cm, q);
3209 if (loop_count == 0)
3212 // Variance adaptive and in frame q adjustment experiments are mutually
3214 if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
3215 vp10_vaq_frame_setup(cpi);
3216 } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
3217 vp10_setup_in_frame_q_adj(cpi);
3220 // transform / motion compensation build reconstruction frame
3221 vp10_encode_frame(cpi);
3223 // Update the skip mb flag probabilities based on the distribution
3224 // seen in the last encoder iteration.
3225 // update_base_skip_probs(cpi);
3227 vpx_clear_system_state();
3229 // Dummy pack of the bitstream using up to date stats to get an
3230 // accurate estimate of output frame size to determine if we need
3232 if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) {
3233 save_coding_context(cpi);
3234 vp10_pack_bitstream(cpi, dest, size);
3236 rc->projected_frame_size = (int)(*size) << 3;
3237 restore_coding_context(cpi);
3239 if (frame_over_shoot_limit == 0)
3240 frame_over_shoot_limit = 1;
3243 if (cpi->oxcf.rc_mode == VPX_Q) {
3246 if ((cm->frame_type == KEY_FRAME) &&
3247 rc->this_key_frame_forced &&
3248 (rc->projected_frame_size < rc->max_frame_bandwidth)) {
3252 int64_t high_err_target = cpi->ambient_err;
3253 int64_t low_err_target = cpi->ambient_err >> 1;
3255 #if CONFIG_VP9_HIGHBITDEPTH
3256 if (cm->use_highbitdepth) {
3257 kf_err = vp10_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3259 kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3262 kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3263 #endif // CONFIG_VP9_HIGHBITDEPTH
3265 // Prevent possible divide by zero error below for perfect KF
3268 // The key frame is not good enough or we can afford
3269 // to make it better without undue risk of popping.
3270 if ((kf_err > high_err_target &&
3271 rc->projected_frame_size <= frame_over_shoot_limit) ||
3272 (kf_err > low_err_target &&
3273 rc->projected_frame_size <= frame_under_shoot_limit)) {
3275 q_high = q > q_low ? q - 1 : q_low;
3278 q = (int)((q * high_err_target) / kf_err);
3279 q = VPXMIN(q, (q_high + q_low) >> 1);
3280 } else if (kf_err < low_err_target &&
3281 rc->projected_frame_size >= frame_under_shoot_limit) {
3282 // The key frame is much better than the previous frame
3284 q_low = q < q_high ? q + 1 : q_high;
3287 q = (int)((q * low_err_target) / kf_err);
3288 q = VPXMIN(q, (q_high + q_low + 1) >> 1);
3291 // Clamp Q to upper and lower limits:
3292 q = clamp(q, q_low, q_high);
3295 } else if (recode_loop_test(
3296 cpi, frame_over_shoot_limit, frame_under_shoot_limit,
3297 q, VPXMAX(q_high, top_index), bottom_index)) {
3298 // Is the projected frame size out of range and are we allowed
3299 // to attempt to recode.
3303 if (cpi->resize_pending == 1) {
3304 // Change in frame size so go back around the recode loop.
3305 cpi->rc.frame_size_selector =
3306 SCALE_STEP1 - cpi->rc.frame_size_selector;
3307 cpi->rc.next_frame_size_selector = cpi->rc.frame_size_selector;
3309 #if CONFIG_INTERNAL_STATS
3310 ++cpi->tot_recode_hits;
3317 // Frame size out of permitted range:
3318 // Update correction factor & compute new Q to try...
3320 // Frame is too large
3321 if (rc->projected_frame_size > rc->this_frame_target) {
3322 // Special case if the projected size is > the max allowed.
3323 if (rc->projected_frame_size >= rc->max_frame_bandwidth)
3324 q_high = rc->worst_quality;
3326 // Raise Qlow as to at least the current value
3327 q_low = q < q_high ? q + 1 : q_high;
3329 if (undershoot_seen || loop_at_this_size > 1) {
3330 // Update rate_correction_factor unless
3331 vp10_rc_update_rate_correction_factors(cpi);
3333 q = (q_high + q_low + 1) / 2;
3335 // Update rate_correction_factor unless
3336 vp10_rc_update_rate_correction_factors(cpi);
3338 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3339 bottom_index, VPXMAX(q_high, top_index));
3341 while (q < q_low && retries < 10) {
3342 vp10_rc_update_rate_correction_factors(cpi);
3343 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3344 bottom_index, VPXMAX(q_high, top_index));
3351 // Frame is too small
3352 q_high = q > q_low ? q - 1 : q_low;
3354 if (overshoot_seen || loop_at_this_size > 1) {
3355 vp10_rc_update_rate_correction_factors(cpi);
3356 q = (q_high + q_low) / 2;
3358 vp10_rc_update_rate_correction_factors(cpi);
3359 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3360 bottom_index, top_index);
3361 // Special case reset for qlow for constrained quality.
3362 // This should only trigger where there is very substantial
3363 // undershoot on a frame and the auto cq level is above
3364 // the user passsed in value.
3365 if (cpi->oxcf.rc_mode == VPX_CQ &&
3370 while (q > q_high && retries < 10) {
3371 vp10_rc_update_rate_correction_factors(cpi);
3372 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3373 bottom_index, top_index);
3378 undershoot_seen = 1;
3381 // Clamp Q to upper and lower limits:
3382 q = clamp(q, q_low, q_high);
3384 loop = (q != last_q);
3390 // Special case for overlay frame.
3391 if (rc->is_src_frame_alt_ref &&
3392 rc->projected_frame_size < rc->max_frame_bandwidth)
3397 ++loop_at_this_size;
3399 #if CONFIG_INTERNAL_STATS
3400 ++cpi->tot_recode_hits;
3406 static int get_ref_frame_flags(const VP10_COMP *cpi) {
3407 const int *const map = cpi->common.ref_frame_map;
3408 const int gold_is_last = map[cpi->gld_fb_idx] == map[cpi->lst_fb_idx];
3409 const int alt_is_last = map[cpi->alt_fb_idx] == map[cpi->lst_fb_idx];
3410 const int gold_is_alt = map[cpi->gld_fb_idx] == map[cpi->alt_fb_idx];
3411 int flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
3414 flags &= ~VP9_GOLD_FLAG;
3416 if (cpi->rc.frames_till_gf_update_due == INT_MAX)
3417 flags &= ~VP9_GOLD_FLAG;
3420 flags &= ~VP9_ALT_FLAG;
3423 flags &= ~VP9_ALT_FLAG;
3428 static void set_ext_overrides(VP10_COMP *cpi) {
3429 // Overrides the defaults with the externally supplied values with
3430 // vp10_update_reference() and vp10_update_entropy() calls
3431 // Note: The overrides are valid only for the next frame passed
3432 // to encode_frame_to_data_rate() function
3433 if (cpi->ext_refresh_frame_context_pending) {
3434 cpi->common.refresh_frame_context = cpi->ext_refresh_frame_context;
3435 cpi->ext_refresh_frame_context_pending = 0;
3437 if (cpi->ext_refresh_frame_flags_pending) {
3438 cpi->refresh_last_frame = cpi->ext_refresh_last_frame;
3439 cpi->refresh_golden_frame = cpi->ext_refresh_golden_frame;
3440 cpi->refresh_alt_ref_frame = cpi->ext_refresh_alt_ref_frame;
3441 cpi->ext_refresh_frame_flags_pending = 0;
3445 YV12_BUFFER_CONFIG *vp10_scale_if_required_fast(VP10_COMMON *cm,
3446 YV12_BUFFER_CONFIG *unscaled,
3447 YV12_BUFFER_CONFIG *scaled) {
3448 if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
3449 cm->mi_rows * MI_SIZE != unscaled->y_height) {
3450 // For 2x2 scaling down.
3451 vpx_scale_frame(unscaled, scaled, unscaled->y_buffer, 9, 2, 1,
3453 vpx_extend_frame_borders(scaled);
3460 YV12_BUFFER_CONFIG *vp10_scale_if_required(VP10_COMMON *cm,
3461 YV12_BUFFER_CONFIG *unscaled,
3462 YV12_BUFFER_CONFIG *scaled) {
3463 if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
3464 cm->mi_rows * MI_SIZE != unscaled->y_height) {
3465 #if CONFIG_VP9_HIGHBITDEPTH
3466 scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
3468 scale_and_extend_frame_nonnormative(unscaled, scaled);
3469 #endif // CONFIG_VP9_HIGHBITDEPTH
3476 static void set_arf_sign_bias(VP10_COMP *cpi) {
3477 VP10_COMMON *const cm = &cpi->common;
3480 if ((cpi->oxcf.pass == 2) && cpi->multi_arf_allowed) {
3481 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3482 arf_sign_bias = cpi->rc.source_alt_ref_active &&
3483 (!cpi->refresh_alt_ref_frame ||
3484 (gf_group->rf_level[gf_group->index] == GF_ARF_LOW));
3487 (cpi->rc.source_alt_ref_active && !cpi->refresh_alt_ref_frame);
3489 cm->ref_frame_sign_bias[ALTREF_FRAME] = arf_sign_bias;
3492 static int setup_interp_filter_search_mask(VP10_COMP *cpi) {
3493 INTERP_FILTER ifilter;
3494 int ref_total[MAX_REF_FRAMES] = {0};
3495 MV_REFERENCE_FRAME ref;
3497 if (cpi->common.last_frame_type == KEY_FRAME ||
3498 cpi->refresh_alt_ref_frame)
3500 for (ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref)
3501 for (ifilter = EIGHTTAP; ifilter <= EIGHTTAP_SHARP; ++ifilter)
3502 ref_total[ref] += cpi->interp_filter_selected[ref][ifilter];
3504 for (ifilter = EIGHTTAP; ifilter <= EIGHTTAP_SHARP; ++ifilter) {
3505 if ((ref_total[LAST_FRAME] &&
3506 cpi->interp_filter_selected[LAST_FRAME][ifilter] == 0) &&
3507 (ref_total[GOLDEN_FRAME] == 0 ||
3508 cpi->interp_filter_selected[GOLDEN_FRAME][ifilter] * 50
3509 < ref_total[GOLDEN_FRAME]) &&
3510 (ref_total[ALTREF_FRAME] == 0 ||
3511 cpi->interp_filter_selected[ALTREF_FRAME][ifilter] * 50
3512 < ref_total[ALTREF_FRAME]))
3513 mask |= 1 << ifilter;
3518 static void encode_frame_to_data_rate(VP10_COMP *cpi,
3521 unsigned int *frame_flags) {
3522 VP10_COMMON *const cm = &cpi->common;
3523 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
3524 struct segmentation *const seg = &cm->seg;
3527 set_ext_overrides(cpi);
3528 vpx_clear_system_state();
3530 // Set the arf sign bias for this frame.
3531 set_arf_sign_bias(cpi);
3533 // Set default state for segment based loop filter update flags.
3534 cm->lf.mode_ref_delta_update = 0;
3536 if (cpi->oxcf.pass == 2 &&
3537 cpi->sf.adaptive_interp_filter_search)
3538 cpi->sf.interp_filter_search_mask =
3539 setup_interp_filter_search_mask(cpi);
3541 // Set various flags etc to special state if it is a key frame.
3542 if (frame_is_intra_only(cm)) {
3543 // Reset the loop filter deltas and segmentation map.
3544 vp10_reset_segment_features(&cm->seg);
3546 // If segmentation is enabled force a map update for key frames.
3548 seg->update_map = 1;
3549 seg->update_data = 1;
3552 // The alternate reference frame cannot be active for a key frame.
3553 cpi->rc.source_alt_ref_active = 0;
3555 cm->error_resilient_mode = oxcf->error_resilient_mode;
3557 // By default, encoder assumes decoder can use prev_mi.
3558 if (cm->error_resilient_mode) {
3559 cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
3560 cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_OFF;
3561 } else if (cm->intra_only) {
3562 // Only reset the current context.
3563 cm->reset_frame_context = RESET_FRAME_CONTEXT_CURRENT;
3567 // For 1 pass CBR, check if we are dropping this frame.
3568 // Never drop on key frame.
3569 if (oxcf->pass == 0 &&
3570 oxcf->rc_mode == VPX_CBR &&
3571 cm->frame_type != KEY_FRAME) {
3572 if (vp10_rc_drop_frame(cpi)) {
3573 vp10_rc_postencode_update_drop_frame(cpi);
3574 ++cm->current_video_frame;
3579 vpx_clear_system_state();
3581 #if CONFIG_INTERNAL_STATS
3582 memset(cpi->mode_chosen_counts, 0,
3583 MAX_MODES * sizeof(*cpi->mode_chosen_counts));
3586 if (cpi->sf.recode_loop == DISALLOW_RECODE) {
3587 encode_without_recode_loop(cpi);
3589 encode_with_recode_loop(cpi, size, dest);
3592 #if CONFIG_VP9_TEMPORAL_DENOISING
3593 #ifdef OUTPUT_YUV_DENOISED
3594 if (oxcf->noise_sensitivity > 0) {
3595 vp10_write_yuv_frame_420(&cpi->denoiser.running_avg_y[INTRA_FRAME],
3600 #ifdef OUTPUT_YUV_SKINMAP
3601 if (cpi->common.current_video_frame > 1) {
3602 vp10_compute_skin_map(cpi, yuv_skinmap_file);
3606 // Special case code to reduce pulsing when key frames are forced at a
3607 // fixed interval. Note the reconstruction error if it is the frame before
3608 // the force key frame
3609 if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
3610 #if CONFIG_VP9_HIGHBITDEPTH
3611 if (cm->use_highbitdepth) {
3612 cpi->ambient_err = vp10_highbd_get_y_sse(cpi->Source,
3613 get_frame_new_buffer(cm));
3615 cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3618 cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3619 #endif // CONFIG_VP9_HIGHBITDEPTH
3622 // If the encoder forced a KEY_FRAME decision
3623 if (cm->frame_type == KEY_FRAME)
3624 cpi->refresh_last_frame = 1;
3626 cm->frame_to_show = get_frame_new_buffer(cm);
3627 cm->frame_to_show->color_space = cm->color_space;
3628 cm->frame_to_show->color_range = cm->color_range;
3630 // Pick the loop filter level for the frame.
3631 loopfilter_frame(cpi, cm);
3633 // build the bitstream
3634 vp10_pack_bitstream(cpi, dest, size);
3636 if (cm->seg.update_map)
3637 update_reference_segmentation_map(cpi);
3639 if (frame_is_intra_only(cm) == 0) {
3640 release_scaled_references(cpi);
3642 vp10_update_reference_frames(cpi);
3644 for (t = TX_4X4; t <= TX_32X32; t++)
3645 full_to_model_counts(cpi->td.counts->coef[t],
3646 cpi->td.rd_counts.coef_counts[t]);
3648 if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD)
3649 vp10_adapt_coef_probs(cm);
3651 if (!frame_is_intra_only(cm)) {
3652 if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
3653 vp10_adapt_mode_probs(cm);
3654 vp10_adapt_mv_probs(cm, cm->allow_high_precision_mv);
3658 if (cpi->refresh_golden_frame == 1)
3659 cpi->frame_flags |= FRAMEFLAGS_GOLDEN;
3661 cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN;
3663 if (cpi->refresh_alt_ref_frame == 1)
3664 cpi->frame_flags |= FRAMEFLAGS_ALTREF;
3666 cpi->frame_flags &= ~FRAMEFLAGS_ALTREF;
3668 cpi->ref_frame_flags = get_ref_frame_flags(cpi);
3670 cm->last_frame_type = cm->frame_type;
3672 vp10_rc_postencode_update(cpi, *size);
3675 output_frame_level_debug_stats(cpi);
3678 if (cm->frame_type == KEY_FRAME) {
3679 // Tell the caller that the frame was coded as a key frame
3680 *frame_flags = cpi->frame_flags | FRAMEFLAGS_KEY;
3682 *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
3685 // Clear the one shot update flags for segmentation map and mode/ref loop
3687 cm->seg.update_map = 0;
3688 cm->seg.update_data = 0;
3689 cm->lf.mode_ref_delta_update = 0;
3691 // keep track of the last coded dimensions
3692 cm->last_width = cm->width;
3693 cm->last_height = cm->height;
3695 // reset to normal state now that we are done.
3696 if (!cm->show_existing_frame)
3697 cm->last_show_frame = cm->show_frame;
3699 if (cm->show_frame) {
3700 vp10_swap_mi_and_prev_mi(cm);
3701 // Don't increment frame counters if this was an altref buffer
3702 // update not a real frame
3703 ++cm->current_video_frame;
3705 cm->prev_frame = cm->cur_frame;
3708 static void Pass0Encode(VP10_COMP *cpi, size_t *size, uint8_t *dest,
3709 unsigned int *frame_flags) {
3710 if (cpi->oxcf.rc_mode == VPX_CBR) {
3711 vp10_rc_get_one_pass_cbr_params(cpi);
3713 vp10_rc_get_one_pass_vbr_params(cpi);
3715 encode_frame_to_data_rate(cpi, size, dest, frame_flags);
3718 static void Pass2Encode(VP10_COMP *cpi, size_t *size,
3719 uint8_t *dest, unsigned int *frame_flags) {
3720 cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
3721 encode_frame_to_data_rate(cpi, size, dest, frame_flags);
3723 vp10_twopass_postencode_update(cpi);
3726 static void init_ref_frame_bufs(VP10_COMMON *cm) {
3728 BufferPool *const pool = cm->buffer_pool;
3729 cm->new_fb_idx = INVALID_IDX;
3730 for (i = 0; i < REF_FRAMES; ++i) {
3731 cm->ref_frame_map[i] = INVALID_IDX;
3732 pool->frame_bufs[i].ref_count = 0;
3736 static void check_initial_width(VP10_COMP *cpi,
3737 #if CONFIG_VP9_HIGHBITDEPTH
3738 int use_highbitdepth,
3740 int subsampling_x, int subsampling_y) {
3741 VP10_COMMON *const cm = &cpi->common;
3743 if (!cpi->initial_width ||
3744 #if CONFIG_VP9_HIGHBITDEPTH
3745 cm->use_highbitdepth != use_highbitdepth ||
3747 cm->subsampling_x != subsampling_x ||
3748 cm->subsampling_y != subsampling_y) {
3749 cm->subsampling_x = subsampling_x;
3750 cm->subsampling_y = subsampling_y;
3751 #if CONFIG_VP9_HIGHBITDEPTH
3752 cm->use_highbitdepth = use_highbitdepth;
3755 alloc_raw_frame_buffers(cpi);
3756 init_ref_frame_bufs(cm);
3757 alloc_util_frame_buffers(cpi);
3759 init_motion_estimation(cpi); // TODO(agrange) This can be removed.
3761 cpi->initial_width = cm->width;
3762 cpi->initial_height = cm->height;
3763 cpi->initial_mbs = cm->MBs;
3767 #if CONFIG_VP9_TEMPORAL_DENOISING
3768 static void setup_denoiser_buffer(VP10_COMP *cpi) {
3769 VP10_COMMON *const cm = &cpi->common;
3770 if (cpi->oxcf.noise_sensitivity > 0 &&
3771 !cpi->denoiser.frame_buffer_initialized) {
3772 vp10_denoiser_alloc(&(cpi->denoiser), cm->width, cm->height,
3773 cm->subsampling_x, cm->subsampling_y,
3774 #if CONFIG_VP9_HIGHBITDEPTH
3775 cm->use_highbitdepth,
3777 VP9_ENC_BORDER_IN_PIXELS);
3782 int vp10_receive_raw_frame(VP10_COMP *cpi, unsigned int frame_flags,
3783 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
3785 VP10_COMMON *cm = &cpi->common;
3786 struct vpx_usec_timer timer;
3788 const int subsampling_x = sd->subsampling_x;
3789 const int subsampling_y = sd->subsampling_y;
3790 #if CONFIG_VP9_HIGHBITDEPTH
3791 const int use_highbitdepth = sd->flags & YV12_FLAG_HIGHBITDEPTH;
3792 check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y);
3794 check_initial_width(cpi, subsampling_x, subsampling_y);
3795 #endif // CONFIG_VP9_HIGHBITDEPTH
3797 #if CONFIG_VP9_TEMPORAL_DENOISING
3798 setup_denoiser_buffer(cpi);
3800 vpx_usec_timer_start(&timer);
3802 if (vp10_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
3803 #if CONFIG_VP9_HIGHBITDEPTH
3805 #endif // CONFIG_VP9_HIGHBITDEPTH
3808 vpx_usec_timer_mark(&timer);
3809 cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
3811 if ((cm->profile == PROFILE_0 || cm->profile == PROFILE_2) &&
3812 (subsampling_x != 1 || subsampling_y != 1)) {
3813 vpx_internal_error(&cm->error, VPX_CODEC_INVALID_PARAM,
3814 "Non-4:2:0 color format requires profile 1 or 3");
3817 if ((cm->profile == PROFILE_1 || cm->profile == PROFILE_3) &&
3818 (subsampling_x == 1 && subsampling_y == 1)) {
3819 vpx_internal_error(&cm->error, VPX_CODEC_INVALID_PARAM,
3820 "4:2:0 color format requires profile 0 or 2");
3828 static int frame_is_reference(const VP10_COMP *cpi) {
3829 const VP10_COMMON *cm = &cpi->common;
3831 return cm->frame_type == KEY_FRAME ||
3832 cpi->refresh_last_frame ||
3833 cpi->refresh_golden_frame ||
3834 cpi->refresh_alt_ref_frame ||
3835 cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_OFF ||
3836 cm->lf.mode_ref_delta_update ||
3837 cm->seg.update_map ||
3838 cm->seg.update_data;
3841 static void adjust_frame_rate(VP10_COMP *cpi,
3842 const struct lookahead_entry *source) {
3843 int64_t this_duration;
3846 if (source->ts_start == cpi->first_time_stamp_ever) {
3847 this_duration = source->ts_end - source->ts_start;
3850 int64_t last_duration = cpi->last_end_time_stamp_seen
3851 - cpi->last_time_stamp_seen;
3853 this_duration = source->ts_end - cpi->last_end_time_stamp_seen;
3855 // do a step update if the duration changes by 10%
3857 step = (int)((this_duration - last_duration) * 10 / last_duration);
3860 if (this_duration) {
3862 vp10_new_framerate(cpi, 10000000.0 / this_duration);
3864 // Average this frame's rate into the last second's average
3865 // frame rate. If we haven't seen 1 second yet, then average
3866 // over the whole interval seen.
3867 const double interval = VPXMIN(
3868 (double)(source->ts_end - cpi->first_time_stamp_ever), 10000000.0);
3869 double avg_duration = 10000000.0 / cpi->framerate;
3870 avg_duration *= (interval - avg_duration + this_duration);
3871 avg_duration /= interval;
3873 vp10_new_framerate(cpi, 10000000.0 / avg_duration);
3876 cpi->last_time_stamp_seen = source->ts_start;
3877 cpi->last_end_time_stamp_seen = source->ts_end;
3880 // Returns 0 if this is not an alt ref else the offset of the source frame
3881 // used as the arf midpoint.
3882 static int get_arf_src_index(VP10_COMP *cpi) {
3883 RATE_CONTROL *const rc = &cpi->rc;
3884 int arf_src_index = 0;
3885 if (is_altref_enabled(cpi)) {
3886 if (cpi->oxcf.pass == 2) {
3887 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3888 if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
3889 arf_src_index = gf_group->arf_src_offset[gf_group->index];
3891 } else if (rc->source_alt_ref_pending) {
3892 arf_src_index = rc->frames_till_gf_update_due;
3895 return arf_src_index;
3898 static void check_src_altref(VP10_COMP *cpi,
3899 const struct lookahead_entry *source) {
3900 RATE_CONTROL *const rc = &cpi->rc;
3902 if (cpi->oxcf.pass == 2) {
3903 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3904 rc->is_src_frame_alt_ref =
3905 (gf_group->update_type[gf_group->index] == OVERLAY_UPDATE);
3907 rc->is_src_frame_alt_ref = cpi->alt_ref_source &&
3908 (source == cpi->alt_ref_source);
3911 if (rc->is_src_frame_alt_ref) {
3912 // Current frame is an ARF overlay frame.
3913 cpi->alt_ref_source = NULL;
3915 // Don't refresh the last buffer for an ARF overlay frame. It will
3916 // become the GF so preserve last as an alternative prediction option.
3917 cpi->refresh_last_frame = 0;
3921 #if CONFIG_INTERNAL_STATS
3922 extern double vp10_get_blockiness(const unsigned char *img1, int img1_pitch,
3923 const unsigned char *img2, int img2_pitch,
3924 int width, int height);
3926 static void adjust_image_stat(double y, double u, double v, double all,
3931 s->stat[ALL] += all;
3932 s->worst = VPXMIN(s->worst, all);
3934 #endif // CONFIG_INTERNAL_STATS
3936 int vp10_get_compressed_data(VP10_COMP *cpi, unsigned int *frame_flags,
3937 size_t *size, uint8_t *dest,
3938 int64_t *time_stamp, int64_t *time_end, int flush) {
3939 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
3940 VP10_COMMON *const cm = &cpi->common;
3941 BufferPool *const pool = cm->buffer_pool;
3942 RATE_CONTROL *const rc = &cpi->rc;
3943 struct vpx_usec_timer cmptimer;
3944 YV12_BUFFER_CONFIG *force_src_buffer = NULL;
3945 struct lookahead_entry *last_source = NULL;
3946 struct lookahead_entry *source = NULL;
3950 vpx_usec_timer_start(&cmptimer);
3952 vp10_set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV);
3954 // Is multi-arf enabled.
3955 // Note that at the moment multi_arf is only configured for 2 pass VBR
3956 if ((oxcf->pass == 2) && (cpi->oxcf.enable_auto_arf > 1))
3957 cpi->multi_arf_allowed = 1;
3959 cpi->multi_arf_allowed = 0;
3962 cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
3963 cm->refresh_frame_context =
3964 oxcf->error_resilient_mode ? REFRESH_FRAME_CONTEXT_OFF :
3965 oxcf->frame_parallel_decoding_mode ? REFRESH_FRAME_CONTEXT_FORWARD
3966 : REFRESH_FRAME_CONTEXT_BACKWARD;
3968 cpi->refresh_last_frame = 1;
3969 cpi->refresh_golden_frame = 0;
3970 cpi->refresh_alt_ref_frame = 0;
3972 // Should we encode an arf frame.
3973 arf_src_index = get_arf_src_index(cpi);
3975 if (arf_src_index) {
3976 assert(arf_src_index <= rc->frames_to_key);
3978 if ((source = vp10_lookahead_peek(cpi->lookahead, arf_src_index)) != NULL) {
3979 cpi->alt_ref_source = source;
3981 if (oxcf->arnr_max_frames > 0) {
3982 // Produce the filtered ARF frame.
3983 vp10_temporal_filter(cpi, arf_src_index);
3984 vpx_extend_frame_borders(&cpi->alt_ref_buffer);
3985 force_src_buffer = &cpi->alt_ref_buffer;
3990 cpi->refresh_alt_ref_frame = 1;
3991 cpi->refresh_golden_frame = 0;
3992 cpi->refresh_last_frame = 0;
3993 rc->is_src_frame_alt_ref = 0;
3994 rc->source_alt_ref_pending = 0;
3996 rc->source_alt_ref_pending = 0;
4001 // Get last frame source.
4002 if (cm->current_video_frame > 0) {
4003 if ((last_source = vp10_lookahead_peek(cpi->lookahead, -1)) == NULL)
4007 // Read in the source frame.
4008 source = vp10_lookahead_pop(cpi->lookahead, flush);
4010 if (source != NULL) {
4014 // Check to see if the frame should be encoded as an arf overlay.
4015 check_src_altref(cpi, source);
4020 cpi->un_scaled_source = cpi->Source = force_src_buffer ? force_src_buffer
4023 cpi->unscaled_last_source = last_source != NULL ? &last_source->img : NULL;
4025 *time_stamp = source->ts_start;
4026 *time_end = source->ts_end;
4027 *frame_flags = (source->flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
4031 if (flush && oxcf->pass == 1 && !cpi->twopass.first_pass_done) {
4032 vp10_end_first_pass(cpi); /* get last stats packet */
4033 cpi->twopass.first_pass_done = 1;
4038 if (source->ts_start < cpi->first_time_stamp_ever) {
4039 cpi->first_time_stamp_ever = source->ts_start;
4040 cpi->last_end_time_stamp_seen = source->ts_start;
4043 // Clear down mmx registers
4044 vpx_clear_system_state();
4046 // adjust frame rates based on timestamps given
4047 if (cm->show_frame) {
4048 adjust_frame_rate(cpi, source);
4051 // Find a free buffer for the new frame, releasing the reference previously
4053 if (cm->new_fb_idx != INVALID_IDX) {
4054 --pool->frame_bufs[cm->new_fb_idx].ref_count;
4056 cm->new_fb_idx = get_free_fb(cm);
4058 if (cm->new_fb_idx == INVALID_IDX)
4061 cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx];
4063 if (cpi->multi_arf_allowed) {
4064 if (cm->frame_type == KEY_FRAME) {
4065 init_buffer_indices(cpi);
4066 } else if (oxcf->pass == 2) {
4067 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
4068 cpi->alt_fb_idx = gf_group->arf_ref_idx[gf_group->index];
4072 // Start with a 0 size frame.
4075 cpi->frame_flags = *frame_flags;
4077 if (oxcf->pass == 2) {
4078 vp10_rc_get_second_pass_params(cpi);
4079 } else if (oxcf->pass == 1) {
4080 set_frame_size(cpi);
4083 if (cpi->oxcf.pass != 0 || frame_is_intra_only(cm) == 1) {
4084 for (i = 0; i < MAX_REF_FRAMES; ++i)
4085 cpi->scaled_ref_idx[i] = INVALID_IDX;
4088 if (oxcf->pass == 1) {
4089 const int lossless = is_lossless_requested(oxcf);
4090 #if CONFIG_VP9_HIGHBITDEPTH
4091 if (cpi->oxcf.use_highbitdepth)
4092 cpi->td.mb.fwd_txm4x4 = lossless ?
4093 vp10_highbd_fwht4x4 : vpx_highbd_fdct4x4;
4095 cpi->td.mb.fwd_txm4x4 = lossless ? vp10_fwht4x4 : vpx_fdct4x4;
4096 cpi->td.mb.highbd_itxm_add = lossless ? vp10_highbd_iwht4x4_add :
4097 vp10_highbd_idct4x4_add;
4099 cpi->td.mb.fwd_txm4x4 = lossless ? vp10_fwht4x4 : vpx_fdct4x4;
4100 #endif // CONFIG_VP9_HIGHBITDEPTH
4101 cpi->td.mb.itxm_add = lossless ? vp10_iwht4x4_add : vp10_idct4x4_add;
4102 vp10_first_pass(cpi, source);
4103 } else if (oxcf->pass == 2) {
4104 Pass2Encode(cpi, size, dest, frame_flags);
4107 Pass0Encode(cpi, size, dest, frame_flags);
4110 if (cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_OFF)
4111 cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
4113 // No frame encoded, or frame was dropped, release scaled references.
4114 if ((*size == 0) && (frame_is_intra_only(cm) == 0)) {
4115 release_scaled_references(cpi);
4119 cpi->droppable = !frame_is_reference(cpi);
4122 vpx_usec_timer_mark(&cmptimer);
4123 cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
4125 if (cpi->b_calculate_psnr && oxcf->pass != 1 && cm->show_frame)
4126 generate_psnr_packet(cpi);
4128 #if CONFIG_INTERNAL_STATS
4130 if (oxcf->pass != 1) {
4131 double samples = 0.0;
4132 cpi->bytes += (int)(*size);
4134 if (cm->show_frame) {
4137 if (cpi->b_calculate_psnr) {
4138 YV12_BUFFER_CONFIG *orig = cpi->Source;
4139 YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
4140 YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer;
4142 #if CONFIG_VP9_HIGHBITDEPTH
4143 calc_highbd_psnr(orig, recon, &psnr, cpi->td.mb.e_mbd.bd,
4144 cpi->oxcf.input_bit_depth);
4146 calc_psnr(orig, recon, &psnr);
4147 #endif // CONFIG_VP9_HIGHBITDEPTH
4149 adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3],
4150 psnr.psnr[0], &cpi->psnr);
4151 cpi->total_sq_error += psnr.sse[0];
4152 cpi->total_samples += psnr.samples[0];
4153 samples = psnr.samples[0];
4157 double frame_ssim2 = 0, weight = 0;
4158 #if CONFIG_VP9_POSTPROC
4159 if (vpx_alloc_frame_buffer(&cm->post_proc_buffer,
4160 recon->y_crop_width, recon->y_crop_height,
4161 cm->subsampling_x, cm->subsampling_y,
4162 #if CONFIG_VP9_HIGHBITDEPTH
4163 cm->use_highbitdepth,
4165 VP9_ENC_BORDER_IN_PIXELS,
4166 cm->byte_alignment) < 0) {
4167 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
4168 "Failed to allocate post processing buffer");
4171 vp10_deblock(cm->frame_to_show, &cm->post_proc_buffer,
4172 cm->lf.filter_level * 10 / 6);
4174 vpx_clear_system_state();
4176 #if CONFIG_VP9_HIGHBITDEPTH
4177 calc_highbd_psnr(orig, pp, &psnr2, cpi->td.mb.e_mbd.bd,
4178 cpi->oxcf.input_bit_depth);
4180 calc_psnr(orig, pp, &psnr2);
4181 #endif // CONFIG_VP9_HIGHBITDEPTH
4183 cpi->totalp_sq_error += psnr2.sse[0];
4184 cpi->totalp_samples += psnr2.samples[0];
4185 adjust_image_stat(psnr2.psnr[1], psnr2.psnr[2], psnr2.psnr[3],
4186 psnr2.psnr[0], &cpi->psnrp);
4188 #if CONFIG_VP9_HIGHBITDEPTH
4189 if (cm->use_highbitdepth) {
4190 frame_ssim2 = vpx_highbd_calc_ssim(orig, recon, &weight,
4191 (int)cm->bit_depth);
4193 frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
4196 frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
4197 #endif // CONFIG_VP9_HIGHBITDEPTH
4199 cpi->worst_ssim= VPXMIN(cpi->worst_ssim, frame_ssim2);
4200 cpi->summed_quality += frame_ssim2 * weight;
4201 cpi->summed_weights += weight;
4203 #if CONFIG_VP9_HIGHBITDEPTH
4204 if (cm->use_highbitdepth) {
4205 frame_ssim2 = vpx_highbd_calc_ssim(
4206 orig, &cm->post_proc_buffer, &weight, (int)cm->bit_depth);
4208 frame_ssim2 = vpx_calc_ssim(orig, &cm->post_proc_buffer, &weight);
4211 frame_ssim2 = vpx_calc_ssim(orig, &cm->post_proc_buffer, &weight);
4212 #endif // CONFIG_VP9_HIGHBITDEPTH
4214 cpi->summedp_quality += frame_ssim2 * weight;
4215 cpi->summedp_weights += weight;
4218 FILE *f = fopen("q_used.stt", "a");
4219 fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
4220 cpi->common.current_video_frame, y2, u2, v2,
4221 frame_psnr2, frame_ssim2);
4227 if (cpi->b_calculate_blockiness) {
4228 #if CONFIG_VP9_HIGHBITDEPTH
4229 if (!cm->use_highbitdepth)
4232 double frame_blockiness = vp10_get_blockiness(
4233 cpi->Source->y_buffer, cpi->Source->y_stride,
4234 cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
4235 cpi->Source->y_width, cpi->Source->y_height);
4236 cpi->worst_blockiness =
4237 VPXMAX(cpi->worst_blockiness, frame_blockiness);
4238 cpi->total_blockiness += frame_blockiness;
4242 if (cpi->b_calculate_consistency) {
4243 #if CONFIG_VP9_HIGHBITDEPTH
4244 if (!cm->use_highbitdepth)
4247 double this_inconsistency = vpx_get_ssim_metrics(
4248 cpi->Source->y_buffer, cpi->Source->y_stride,
4249 cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
4250 cpi->Source->y_width, cpi->Source->y_height, cpi->ssim_vars,
4253 const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1);
4254 double consistency = vpx_sse_to_psnr(samples, peak,
4255 (double)cpi->total_inconsistency);
4256 if (consistency > 0.0)
4257 cpi->worst_consistency =
4258 VPXMIN(cpi->worst_consistency, consistency);
4259 cpi->total_inconsistency += this_inconsistency;
4263 if (cpi->b_calculate_ssimg) {
4264 double y, u, v, frame_all;
4265 #if CONFIG_VP9_HIGHBITDEPTH
4266 if (cm->use_highbitdepth) {
4267 frame_all = vpx_highbd_calc_ssimg(cpi->Source, cm->frame_to_show, &y,
4268 &u, &v, (int)cm->bit_depth);
4270 frame_all = vpx_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u,
4274 frame_all = vpx_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u, &v);
4275 #endif // CONFIG_VP9_HIGHBITDEPTH
4276 adjust_image_stat(y, u, v, frame_all, &cpi->ssimg);
4278 #if CONFIG_VP9_HIGHBITDEPTH
4279 if (!cm->use_highbitdepth)
4282 double y, u, v, frame_all;
4283 frame_all = vpx_calc_fastssim(cpi->Source, cm->frame_to_show, &y, &u,
4285 adjust_image_stat(y, u, v, frame_all, &cpi->fastssim);
4286 /* TODO(JBB): add 10/12 bit support */
4288 #if CONFIG_VP9_HIGHBITDEPTH
4289 if (!cm->use_highbitdepth)
4292 double y, u, v, frame_all;
4293 frame_all = vpx_psnrhvs(cpi->Source, cm->frame_to_show, &y, &u, &v);
4294 adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs);
4300 vpx_clear_system_state();
4304 int vp10_get_preview_raw_frame(VP10_COMP *cpi, YV12_BUFFER_CONFIG *dest,
4305 vp10_ppflags_t *flags) {
4306 VP10_COMMON *cm = &cpi->common;
4307 #if !CONFIG_VP9_POSTPROC
4311 if (!cm->show_frame) {
4315 #if CONFIG_VP9_POSTPROC
4316 ret = vp10_post_proc_frame(cm, dest, flags);
4318 if (cm->frame_to_show) {
4319 *dest = *cm->frame_to_show;
4320 dest->y_width = cm->width;
4321 dest->y_height = cm->height;
4322 dest->uv_width = cm->width >> cm->subsampling_x;
4323 dest->uv_height = cm->height >> cm->subsampling_y;
4328 #endif // !CONFIG_VP9_POSTPROC
4329 vpx_clear_system_state();
4334 int vp10_set_internal_size(VP10_COMP *cpi,
4335 VPX_SCALING horiz_mode, VPX_SCALING vert_mode) {
4336 VP10_COMMON *cm = &cpi->common;
4337 int hr = 0, hs = 0, vr = 0, vs = 0;
4339 if (horiz_mode > ONETWO || vert_mode > ONETWO)
4342 Scale2Ratio(horiz_mode, &hr, &hs);
4343 Scale2Ratio(vert_mode, &vr, &vs);
4345 // always go to the next whole number
4346 cm->width = (hs - 1 + cpi->oxcf.width * hr) / hs;
4347 cm->height = (vs - 1 + cpi->oxcf.height * vr) / vs;
4348 assert(cm->width <= cpi->initial_width);
4349 assert(cm->height <= cpi->initial_height);
4351 update_frame_size(cpi);
4356 int vp10_set_size_literal(VP10_COMP *cpi, unsigned int width,
4357 unsigned int height) {
4358 VP10_COMMON *cm = &cpi->common;
4359 #if CONFIG_VP9_HIGHBITDEPTH
4360 check_initial_width(cpi, cm->use_highbitdepth, 1, 1);
4362 check_initial_width(cpi, 1, 1);
4363 #endif // CONFIG_VP9_HIGHBITDEPTH
4365 #if CONFIG_VP9_TEMPORAL_DENOISING
4366 setup_denoiser_buffer(cpi);
4371 if (cm->width > cpi->initial_width) {
4372 cm->width = cpi->initial_width;
4373 printf("Warning: Desired width too large, changed to %d\n", cm->width);
4378 cm->height = height;
4379 if (cm->height > cpi->initial_height) {
4380 cm->height = cpi->initial_height;
4381 printf("Warning: Desired height too large, changed to %d\n", cm->height);
4384 assert(cm->width <= cpi->initial_width);
4385 assert(cm->height <= cpi->initial_height);
4387 update_frame_size(cpi);
4392 int64_t vp10_get_y_sse(const YV12_BUFFER_CONFIG *a,
4393 const YV12_BUFFER_CONFIG *b) {
4394 assert(a->y_crop_width == b->y_crop_width);
4395 assert(a->y_crop_height == b->y_crop_height);
4397 return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
4398 a->y_crop_width, a->y_crop_height);
4401 #if CONFIG_VP9_HIGHBITDEPTH
4402 int64_t vp10_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
4403 const YV12_BUFFER_CONFIG *b) {
4404 assert(a->y_crop_width == b->y_crop_width);
4405 assert(a->y_crop_height == b->y_crop_height);
4406 assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
4407 assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
4409 return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
4410 a->y_crop_width, a->y_crop_height);
4412 #endif // CONFIG_VP9_HIGHBITDEPTH
4414 int vp10_get_quantizer(VP10_COMP *cpi) {
4415 return cpi->common.base_qindex;
4418 void vp10_apply_encoding_flags(VP10_COMP *cpi, vpx_enc_frame_flags_t flags) {
4419 if (flags & (VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF |
4420 VP8_EFLAG_NO_REF_ARF)) {
4423 if (flags & VP8_EFLAG_NO_REF_LAST)
4424 ref ^= VP9_LAST_FLAG;
4426 if (flags & VP8_EFLAG_NO_REF_GF)
4427 ref ^= VP9_GOLD_FLAG;
4429 if (flags & VP8_EFLAG_NO_REF_ARF)
4430 ref ^= VP9_ALT_FLAG;
4432 vp10_use_as_reference(cpi, ref);
4435 if (flags & (VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
4436 VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_FORCE_GF |
4437 VP8_EFLAG_FORCE_ARF)) {
4440 if (flags & VP8_EFLAG_NO_UPD_LAST)
4441 upd ^= VP9_LAST_FLAG;
4443 if (flags & VP8_EFLAG_NO_UPD_GF)
4444 upd ^= VP9_GOLD_FLAG;
4446 if (flags & VP8_EFLAG_NO_UPD_ARF)
4447 upd ^= VP9_ALT_FLAG;
4449 vp10_update_reference(cpi, upd);
4452 if (flags & VP8_EFLAG_NO_UPD_ENTROPY) {
4453 vp10_update_entropy(cpi, 0);