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 cpi->td.mb.mbmi_ext_base = cpi->mbmi_ext_base;
714 memset(cpi->mbmi_ext_base, 0,
715 cm->mi_rows * cm->mi_cols * sizeof(*cpi->mbmi_ext_base));
717 set_tile_limits(cpi);
720 static void init_buffer_indices(VP10_COMP *cpi) {
726 static void init_config(struct VP10_COMP *cpi, VP10EncoderConfig *oxcf) {
727 VP10_COMMON *const cm = &cpi->common;
730 cpi->framerate = oxcf->init_framerate;
732 cm->profile = oxcf->profile;
733 cm->bit_depth = oxcf->bit_depth;
734 #if CONFIG_VP9_HIGHBITDEPTH
735 cm->use_highbitdepth = oxcf->use_highbitdepth;
737 cm->color_space = oxcf->color_space;
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;
1410 if (cm->profile <= PROFILE_1)
1411 assert(cm->bit_depth == VPX_BITS_8);
1413 assert(cm->bit_depth > VPX_BITS_8);
1416 #if CONFIG_VP9_HIGHBITDEPTH
1417 cpi->td.mb.e_mbd.bd = (int)cm->bit_depth;
1418 #endif // CONFIG_VP9_HIGHBITDEPTH
1420 rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
1422 cpi->refresh_golden_frame = 0;
1423 cpi->refresh_last_frame = 1;
1424 cm->refresh_frame_context = 1;
1425 cm->reset_frame_context = 0;
1427 vp10_reset_segment_features(&cm->seg);
1428 vp10_set_high_precision_mv(cpi, 0);
1433 for (i = 0; i < MAX_SEGMENTS; i++)
1434 cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout;
1436 cpi->encode_breakout = cpi->oxcf.encode_breakout;
1438 set_rc_buffer_sizes(rc, &cpi->oxcf);
1440 // Under a configuration change, where maximum_buffer_size may change,
1441 // keep buffer level clipped to the maximum allowed buffer size.
1442 rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
1443 rc->buffer_level = VPXMIN(rc->buffer_level, rc->maximum_buffer_size);
1445 // Set up frame rate and related parameters rate control values.
1446 vp10_new_framerate(cpi, cpi->framerate);
1448 // Set absolute upper and lower quality limits
1449 rc->worst_quality = cpi->oxcf.worst_allowed_q;
1450 rc->best_quality = cpi->oxcf.best_allowed_q;
1452 cm->interp_filter = cpi->sf.default_interp_filter;
1454 cm->display_width = cpi->oxcf.width;
1455 cm->display_height = cpi->oxcf.height;
1456 cm->width = cpi->oxcf.width;
1457 cm->height = cpi->oxcf.height;
1459 if (cpi->initial_width) {
1460 if (cm->width > cpi->initial_width || cm->height > cpi->initial_height) {
1461 vp10_free_context_buffers(cm);
1462 vp10_alloc_compressor_data(cpi);
1463 realloc_segmentation_maps(cpi);
1464 cpi->initial_width = cpi->initial_height = 0;
1467 update_frame_size(cpi);
1469 cpi->alt_ref_source = NULL;
1470 rc->is_src_frame_alt_ref = 0;
1473 // Experimental RD Code
1474 cpi->frame_distortion = 0;
1475 cpi->last_frame_distortion = 0;
1478 set_tile_limits(cpi);
1480 cpi->ext_refresh_frame_flags_pending = 0;
1481 cpi->ext_refresh_frame_context_pending = 0;
1483 #if CONFIG_VP9_HIGHBITDEPTH
1484 highbd_set_var_fns(cpi);
1489 #define M_LOG2_E 0.693147180559945309417
1491 #define log2f(x) (log (x) / (float) M_LOG2_E)
1493 static void cal_nmvjointsadcost(int *mvjointsadcost) {
1494 mvjointsadcost[0] = 600;
1495 mvjointsadcost[1] = 300;
1496 mvjointsadcost[2] = 300;
1497 mvjointsadcost[3] = 300;
1500 static void cal_nmvsadcosts(int *mvsadcost[2]) {
1503 mvsadcost[0][0] = 0;
1504 mvsadcost[1][0] = 0;
1507 double z = 256 * (2 * (log2f(8 * i) + .6));
1508 mvsadcost[0][i] = (int)z;
1509 mvsadcost[1][i] = (int)z;
1510 mvsadcost[0][-i] = (int)z;
1511 mvsadcost[1][-i] = (int)z;
1512 } while (++i <= MV_MAX);
1515 static void cal_nmvsadcosts_hp(int *mvsadcost[2]) {
1518 mvsadcost[0][0] = 0;
1519 mvsadcost[1][0] = 0;
1522 double z = 256 * (2 * (log2f(8 * i) + .6));
1523 mvsadcost[0][i] = (int)z;
1524 mvsadcost[1][i] = (int)z;
1525 mvsadcost[0][-i] = (int)z;
1526 mvsadcost[1][-i] = (int)z;
1527 } while (++i <= MV_MAX);
1531 VP10_COMP *vp10_create_compressor(VP10EncoderConfig *oxcf,
1532 BufferPool *const pool) {
1534 VP10_COMP *volatile const cpi = vpx_memalign(32, sizeof(VP10_COMP));
1535 VP10_COMMON *volatile const cm = cpi != NULL ? &cpi->common : NULL;
1542 if (setjmp(cm->error.jmp)) {
1543 cm->error.setjmp = 0;
1544 vp10_remove_compressor(cpi);
1548 cm->error.setjmp = 1;
1549 cm->alloc_mi = vp10_enc_alloc_mi;
1550 cm->free_mi = vp10_enc_free_mi;
1551 cm->setup_mi = vp10_enc_setup_mi;
1553 CHECK_MEM_ERROR(cm, cm->fc,
1554 (FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
1555 CHECK_MEM_ERROR(cm, cm->frame_contexts,
1556 (FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
1557 sizeof(*cm->frame_contexts)));
1559 cpi->resize_state = 0;
1560 cpi->resize_avg_qp = 0;
1561 cpi->resize_buffer_underflow = 0;
1562 cpi->common.buffer_pool = pool;
1564 init_config(cpi, oxcf);
1565 vp10_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc);
1567 cm->current_video_frame = 0;
1568 cpi->partition_search_skippable_frame = 0;
1569 cpi->tile_data = NULL;
1571 realloc_segmentation_maps(cpi);
1573 CHECK_MEM_ERROR(cm, cpi->nmvcosts[0],
1574 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[0])));
1575 CHECK_MEM_ERROR(cm, cpi->nmvcosts[1],
1576 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[1])));
1577 CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[0],
1578 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[0])));
1579 CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[1],
1580 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[1])));
1581 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[0],
1582 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[0])));
1583 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[1],
1584 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[1])));
1585 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[0],
1586 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[0])));
1587 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[1],
1588 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[1])));
1590 for (i = 0; i < (sizeof(cpi->mbgraph_stats) /
1591 sizeof(cpi->mbgraph_stats[0])); i++) {
1592 CHECK_MEM_ERROR(cm, cpi->mbgraph_stats[i].mb_stats,
1593 vpx_calloc(cm->MBs *
1594 sizeof(*cpi->mbgraph_stats[i].mb_stats), 1));
1597 #if CONFIG_FP_MB_STATS
1598 cpi->use_fp_mb_stats = 0;
1599 if (cpi->use_fp_mb_stats) {
1600 // a place holder used to store the first pass mb stats in the first pass
1601 CHECK_MEM_ERROR(cm, cpi->twopass.frame_mb_stats_buf,
1602 vpx_calloc(cm->MBs * sizeof(uint8_t), 1));
1604 cpi->twopass.frame_mb_stats_buf = NULL;
1608 cpi->refresh_alt_ref_frame = 0;
1609 cpi->multi_arf_last_grp_enabled = 0;
1611 cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
1612 #if CONFIG_INTERNAL_STATS
1613 cpi->b_calculate_ssimg = 0;
1614 cpi->b_calculate_blockiness = 1;
1615 cpi->b_calculate_consistency = 1;
1616 cpi->total_inconsistency = 0;
1617 cpi->psnr.worst = 100.0;
1618 cpi->worst_ssim = 100.0;
1623 if (cpi->b_calculate_psnr) {
1624 cpi->total_sq_error = 0;
1625 cpi->total_samples = 0;
1627 cpi->totalp_sq_error = 0;
1628 cpi->totalp_samples = 0;
1630 cpi->tot_recode_hits = 0;
1631 cpi->summed_quality = 0;
1632 cpi->summed_weights = 0;
1633 cpi->summedp_quality = 0;
1634 cpi->summedp_weights = 0;
1637 if (cpi->b_calculate_ssimg) {
1638 cpi->ssimg.worst= 100.0;
1640 cpi->fastssim.worst = 100.0;
1642 cpi->psnrhvs.worst = 100.0;
1644 if (cpi->b_calculate_blockiness) {
1645 cpi->total_blockiness = 0;
1646 cpi->worst_blockiness = 0.0;
1649 if (cpi->b_calculate_consistency) {
1650 cpi->ssim_vars = vpx_malloc(sizeof(*cpi->ssim_vars) *
1651 4 * cpi->common.mi_rows * cpi->common.mi_cols);
1652 cpi->worst_consistency = 100.0;
1657 cpi->first_time_stamp_ever = INT64_MAX;
1659 cal_nmvjointsadcost(cpi->td.mb.nmvjointsadcost);
1660 cpi->td.mb.nmvcost[0] = &cpi->nmvcosts[0][MV_MAX];
1661 cpi->td.mb.nmvcost[1] = &cpi->nmvcosts[1][MV_MAX];
1662 cpi->td.mb.nmvsadcost[0] = &cpi->nmvsadcosts[0][MV_MAX];
1663 cpi->td.mb.nmvsadcost[1] = &cpi->nmvsadcosts[1][MV_MAX];
1664 cal_nmvsadcosts(cpi->td.mb.nmvsadcost);
1666 cpi->td.mb.nmvcost_hp[0] = &cpi->nmvcosts_hp[0][MV_MAX];
1667 cpi->td.mb.nmvcost_hp[1] = &cpi->nmvcosts_hp[1][MV_MAX];
1668 cpi->td.mb.nmvsadcost_hp[0] = &cpi->nmvsadcosts_hp[0][MV_MAX];
1669 cpi->td.mb.nmvsadcost_hp[1] = &cpi->nmvsadcosts_hp[1][MV_MAX];
1670 cal_nmvsadcosts_hp(cpi->td.mb.nmvsadcost_hp);
1672 #if CONFIG_VP9_TEMPORAL_DENOISING
1673 #ifdef OUTPUT_YUV_DENOISED
1674 yuv_denoised_file = fopen("denoised.yuv", "ab");
1677 #ifdef OUTPUT_YUV_SKINMAP
1678 yuv_skinmap_file = fopen("skinmap.yuv", "ab");
1680 #ifdef OUTPUT_YUV_REC
1681 yuv_rec_file = fopen("rec.yuv", "wb");
1685 framepsnr = fopen("framepsnr.stt", "a");
1686 kf_list = fopen("kf_list.stt", "w");
1689 cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
1691 if (oxcf->pass == 1) {
1692 vp10_init_first_pass(cpi);
1693 } else if (oxcf->pass == 2) {
1694 const size_t packet_sz = sizeof(FIRSTPASS_STATS);
1695 const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz);
1697 #if CONFIG_FP_MB_STATS
1698 if (cpi->use_fp_mb_stats) {
1699 const size_t psz = cpi->common.MBs * sizeof(uint8_t);
1700 const int ps = (int)(oxcf->firstpass_mb_stats_in.sz / psz);
1702 cpi->twopass.firstpass_mb_stats.mb_stats_start =
1703 oxcf->firstpass_mb_stats_in.buf;
1704 cpi->twopass.firstpass_mb_stats.mb_stats_end =
1705 cpi->twopass.firstpass_mb_stats.mb_stats_start +
1706 (ps - 1) * cpi->common.MBs * sizeof(uint8_t);
1710 cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf;
1711 cpi->twopass.stats_in = cpi->twopass.stats_in_start;
1712 cpi->twopass.stats_in_end = &cpi->twopass.stats_in[packets - 1];
1714 vp10_init_second_pass(cpi);
1717 vp10_set_speed_features_framesize_independent(cpi);
1718 vp10_set_speed_features_framesize_dependent(cpi);
1720 // Allocate memory to store variances for a frame.
1721 CHECK_MEM_ERROR(cm, cpi->source_diff_var,
1722 vpx_calloc(cm->MBs, sizeof(diff)));
1723 cpi->source_var_thresh = 0;
1724 cpi->frames_till_next_var_check = 0;
1726 #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF)\
1727 cpi->fn_ptr[BT].sdf = SDF; \
1728 cpi->fn_ptr[BT].sdaf = SDAF; \
1729 cpi->fn_ptr[BT].vf = VF; \
1730 cpi->fn_ptr[BT].svf = SVF; \
1731 cpi->fn_ptr[BT].svaf = SVAF; \
1732 cpi->fn_ptr[BT].sdx3f = SDX3F; \
1733 cpi->fn_ptr[BT].sdx8f = SDX8F; \
1734 cpi->fn_ptr[BT].sdx4df = SDX4DF;
1736 BFP(BLOCK_32X16, vpx_sad32x16, vpx_sad32x16_avg,
1737 vpx_variance32x16, vpx_sub_pixel_variance32x16,
1738 vpx_sub_pixel_avg_variance32x16, NULL, NULL, vpx_sad32x16x4d)
1740 BFP(BLOCK_16X32, vpx_sad16x32, vpx_sad16x32_avg,
1741 vpx_variance16x32, vpx_sub_pixel_variance16x32,
1742 vpx_sub_pixel_avg_variance16x32, NULL, NULL, vpx_sad16x32x4d)
1744 BFP(BLOCK_64X32, vpx_sad64x32, vpx_sad64x32_avg,
1745 vpx_variance64x32, vpx_sub_pixel_variance64x32,
1746 vpx_sub_pixel_avg_variance64x32, NULL, NULL, vpx_sad64x32x4d)
1748 BFP(BLOCK_32X64, vpx_sad32x64, vpx_sad32x64_avg,
1749 vpx_variance32x64, vpx_sub_pixel_variance32x64,
1750 vpx_sub_pixel_avg_variance32x64, NULL, NULL, vpx_sad32x64x4d)
1752 BFP(BLOCK_32X32, vpx_sad32x32, vpx_sad32x32_avg,
1753 vpx_variance32x32, vpx_sub_pixel_variance32x32,
1754 vpx_sub_pixel_avg_variance32x32, vpx_sad32x32x3, vpx_sad32x32x8,
1757 BFP(BLOCK_64X64, vpx_sad64x64, vpx_sad64x64_avg,
1758 vpx_variance64x64, vpx_sub_pixel_variance64x64,
1759 vpx_sub_pixel_avg_variance64x64, vpx_sad64x64x3, vpx_sad64x64x8,
1762 BFP(BLOCK_16X16, vpx_sad16x16, vpx_sad16x16_avg,
1763 vpx_variance16x16, vpx_sub_pixel_variance16x16,
1764 vpx_sub_pixel_avg_variance16x16, vpx_sad16x16x3, vpx_sad16x16x8,
1767 BFP(BLOCK_16X8, vpx_sad16x8, vpx_sad16x8_avg,
1768 vpx_variance16x8, vpx_sub_pixel_variance16x8,
1769 vpx_sub_pixel_avg_variance16x8,
1770 vpx_sad16x8x3, vpx_sad16x8x8, vpx_sad16x8x4d)
1772 BFP(BLOCK_8X16, vpx_sad8x16, vpx_sad8x16_avg,
1773 vpx_variance8x16, vpx_sub_pixel_variance8x16,
1774 vpx_sub_pixel_avg_variance8x16,
1775 vpx_sad8x16x3, vpx_sad8x16x8, vpx_sad8x16x4d)
1777 BFP(BLOCK_8X8, vpx_sad8x8, vpx_sad8x8_avg,
1778 vpx_variance8x8, vpx_sub_pixel_variance8x8,
1779 vpx_sub_pixel_avg_variance8x8,
1780 vpx_sad8x8x3, vpx_sad8x8x8, vpx_sad8x8x4d)
1782 BFP(BLOCK_8X4, vpx_sad8x4, vpx_sad8x4_avg,
1783 vpx_variance8x4, vpx_sub_pixel_variance8x4,
1784 vpx_sub_pixel_avg_variance8x4, NULL, vpx_sad8x4x8, vpx_sad8x4x4d)
1786 BFP(BLOCK_4X8, vpx_sad4x8, vpx_sad4x8_avg,
1787 vpx_variance4x8, vpx_sub_pixel_variance4x8,
1788 vpx_sub_pixel_avg_variance4x8, NULL, vpx_sad4x8x8, vpx_sad4x8x4d)
1790 BFP(BLOCK_4X4, vpx_sad4x4, vpx_sad4x4_avg,
1791 vpx_variance4x4, vpx_sub_pixel_variance4x4,
1792 vpx_sub_pixel_avg_variance4x4,
1793 vpx_sad4x4x3, vpx_sad4x4x8, vpx_sad4x4x4d)
1795 #if CONFIG_VP9_HIGHBITDEPTH
1796 highbd_set_var_fns(cpi);
1799 /* vp10_init_quantizer() is first called here. Add check in
1800 * vp10_frame_init_quantizer() so that vp10_init_quantizer is only
1801 * called later when needed. This will avoid unnecessary calls of
1802 * vp10_init_quantizer() for every frame.
1804 vp10_init_quantizer(cpi);
1806 vp10_loop_filter_init(cm);
1808 cm->error.setjmp = 0;
1812 #define SNPRINT(H, T) \
1813 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T))
1815 #define SNPRINT2(H, T, V) \
1816 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V))
1818 void vp10_remove_compressor(VP10_COMP *cpi) {
1819 VP10_COMMON *const cm = &cpi->common;
1826 if (cpi && (cm->current_video_frame > 0)) {
1827 #if CONFIG_INTERNAL_STATS
1828 vpx_clear_system_state();
1830 if (cpi->oxcf.pass != 1) {
1831 char headings[512] = {0};
1832 char results[512] = {0};
1833 FILE *f = fopen("opsnr.stt", "a");
1834 double time_encoded = (cpi->last_end_time_stamp_seen
1835 - cpi->first_time_stamp_ever) / 10000000.000;
1836 double total_encode_time = (cpi->time_receive_data +
1837 cpi->time_compress_data) / 1000.000;
1839 (double)cpi->bytes * (double) 8 / (double)1000 / time_encoded;
1840 const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1);
1842 if (cpi->b_calculate_psnr) {
1843 const double total_psnr =
1844 vpx_sse_to_psnr((double)cpi->total_samples, peak,
1845 (double)cpi->total_sq_error);
1846 const double totalp_psnr =
1847 vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
1848 (double)cpi->totalp_sq_error);
1849 const double total_ssim = 100 * pow(cpi->summed_quality /
1850 cpi->summed_weights, 8.0);
1851 const double totalp_ssim = 100 * pow(cpi->summedp_quality /
1852 cpi->summedp_weights, 8.0);
1854 snprintf(headings, sizeof(headings),
1855 "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t"
1856 "VPXSSIM\tVPSSIMP\tFASTSIM\tPSNRHVS\t"
1857 "WstPsnr\tWstSsim\tWstFast\tWstHVS");
1858 snprintf(results, sizeof(results),
1859 "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1860 "%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1861 "%7.3f\t%7.3f\t%7.3f\t%7.3f",
1862 dr, cpi->psnr.stat[ALL] / cpi->count, total_psnr,
1863 cpi->psnrp.stat[ALL] / cpi->count, totalp_psnr,
1864 total_ssim, totalp_ssim,
1865 cpi->fastssim.stat[ALL] / cpi->count,
1866 cpi->psnrhvs.stat[ALL] / cpi->count,
1867 cpi->psnr.worst, cpi->worst_ssim, cpi->fastssim.worst,
1868 cpi->psnrhvs.worst);
1870 if (cpi->b_calculate_blockiness) {
1871 SNPRINT(headings, "\t Block\tWstBlck");
1872 SNPRINT2(results, "\t%7.3f", cpi->total_blockiness / cpi->count);
1873 SNPRINT2(results, "\t%7.3f", cpi->worst_blockiness);
1876 if (cpi->b_calculate_consistency) {
1877 double consistency =
1878 vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
1879 (double)cpi->total_inconsistency);
1881 SNPRINT(headings, "\tConsist\tWstCons");
1882 SNPRINT2(results, "\t%7.3f", consistency);
1883 SNPRINT2(results, "\t%7.3f", cpi->worst_consistency);
1886 if (cpi->b_calculate_ssimg) {
1887 SNPRINT(headings, "\t SSIMG\tWtSSIMG");
1888 SNPRINT2(results, "\t%7.3f", cpi->ssimg.stat[ALL] / cpi->count);
1889 SNPRINT2(results, "\t%7.3f", cpi->ssimg.worst);
1892 fprintf(f, "%s\t Time\n", headings);
1893 fprintf(f, "%s\t%8.0f\n", results, total_encode_time);
1903 printf("\n_pick_loop_filter_level:%d\n", cpi->time_pick_lpf / 1000);
1904 printf("\n_frames recive_data encod_mb_row compress_frame Total\n");
1905 printf("%6d %10ld %10ld %10ld %10ld\n", cpi->common.current_video_frame,
1906 cpi->time_receive_data / 1000, cpi->time_encode_sb_row / 1000,
1907 cpi->time_compress_data / 1000,
1908 (cpi->time_receive_data + cpi->time_compress_data) / 1000);
1913 #if CONFIG_VP9_TEMPORAL_DENOISING
1914 vp10_denoiser_free(&(cpi->denoiser));
1917 for (t = 0; t < cpi->num_workers; ++t) {
1918 VPxWorker *const worker = &cpi->workers[t];
1919 EncWorkerData *const thread_data = &cpi->tile_thr_data[t];
1921 // Deallocate allocated threads.
1922 vpx_get_worker_interface()->end(worker);
1924 // Deallocate allocated thread data.
1925 if (t < cpi->num_workers - 1) {
1926 vpx_free(thread_data->td->counts);
1927 vp10_free_pc_tree(thread_data->td);
1928 vpx_free(thread_data->td);
1931 vpx_free(cpi->tile_thr_data);
1932 vpx_free(cpi->workers);
1934 if (cpi->num_workers > 1)
1935 vp10_loop_filter_dealloc(&cpi->lf_row_sync);
1937 dealloc_compressor_data(cpi);
1939 for (i = 0; i < sizeof(cpi->mbgraph_stats) /
1940 sizeof(cpi->mbgraph_stats[0]); ++i) {
1941 vpx_free(cpi->mbgraph_stats[i].mb_stats);
1944 #if CONFIG_FP_MB_STATS
1945 if (cpi->use_fp_mb_stats) {
1946 vpx_free(cpi->twopass.frame_mb_stats_buf);
1947 cpi->twopass.frame_mb_stats_buf = NULL;
1951 vp10_remove_common(cm);
1952 vp10_free_ref_frame_buffers(cm->buffer_pool);
1953 #if CONFIG_VP9_POSTPROC
1954 vp10_free_postproc_buffers(cm);
1958 #if CONFIG_VP9_TEMPORAL_DENOISING
1959 #ifdef OUTPUT_YUV_DENOISED
1960 fclose(yuv_denoised_file);
1963 #ifdef OUTPUT_YUV_SKINMAP
1964 fclose(yuv_skinmap_file);
1966 #ifdef OUTPUT_YUV_REC
1967 fclose(yuv_rec_file);
1984 /* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
1985 * and highbd_8_variance(). It should not.
1987 static void encoder_variance(const uint8_t *a, int a_stride,
1988 const uint8_t *b, int b_stride,
1989 int w, int h, unsigned int *sse, int *sum) {
1995 for (i = 0; i < h; i++) {
1996 for (j = 0; j < w; j++) {
1997 const int diff = a[j] - b[j];
1999 *sse += diff * diff;
2007 #if CONFIG_VP9_HIGHBITDEPTH
2008 static void encoder_highbd_variance64(const uint8_t *a8, int a_stride,
2009 const uint8_t *b8, int b_stride,
2010 int w, int h, uint64_t *sse,
2014 uint16_t *a = CONVERT_TO_SHORTPTR(a8);
2015 uint16_t *b = CONVERT_TO_SHORTPTR(b8);
2019 for (i = 0; i < h; i++) {
2020 for (j = 0; j < w; j++) {
2021 const int diff = a[j] - b[j];
2023 *sse += diff * diff;
2030 static void encoder_highbd_8_variance(const uint8_t *a8, int a_stride,
2031 const uint8_t *b8, int b_stride,
2033 unsigned int *sse, int *sum) {
2034 uint64_t sse_long = 0;
2035 uint64_t sum_long = 0;
2036 encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
2037 &sse_long, &sum_long);
2038 *sse = (unsigned int)sse_long;
2039 *sum = (int)sum_long;
2041 #endif // CONFIG_VP9_HIGHBITDEPTH
2043 static int64_t get_sse(const uint8_t *a, int a_stride,
2044 const uint8_t *b, int b_stride,
2045 int width, int height) {
2046 const int dw = width % 16;
2047 const int dh = height % 16;
2048 int64_t total_sse = 0;
2049 unsigned int sse = 0;
2054 encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
2055 dw, height, &sse, &sum);
2060 encoder_variance(&a[(height - dh) * a_stride], a_stride,
2061 &b[(height - dh) * b_stride], b_stride,
2062 width - dw, dh, &sse, &sum);
2066 for (y = 0; y < height / 16; ++y) {
2067 const uint8_t *pa = a;
2068 const uint8_t *pb = b;
2069 for (x = 0; x < width / 16; ++x) {
2070 vpx_mse16x16(pa, a_stride, pb, b_stride, &sse);
2084 #if CONFIG_VP9_HIGHBITDEPTH
2085 static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
2086 const uint8_t *b8, int b_stride,
2087 int width, int height,
2088 unsigned int input_shift) {
2089 const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
2090 const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
2091 int64_t total_sse = 0;
2093 for (y = 0; y < height; ++y) {
2094 for (x = 0; x < width; ++x) {
2096 diff = (a[x] >> input_shift) - (b[x] >> input_shift);
2097 total_sse += diff * diff;
2105 static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
2106 const uint8_t *b, int b_stride,
2107 int width, int height) {
2108 int64_t total_sse = 0;
2110 const int dw = width % 16;
2111 const int dh = height % 16;
2112 unsigned int sse = 0;
2115 encoder_highbd_8_variance(&a[width - dw], a_stride,
2116 &b[width - dw], b_stride,
2117 dw, height, &sse, &sum);
2121 encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
2122 &b[(height - dh) * b_stride], b_stride,
2123 width - dw, dh, &sse, &sum);
2126 for (y = 0; y < height / 16; ++y) {
2127 const uint8_t *pa = a;
2128 const uint8_t *pb = b;
2129 for (x = 0; x < width / 16; ++x) {
2130 vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
2140 #endif // CONFIG_VP9_HIGHBITDEPTH
2143 double psnr[4]; // total/y/u/v
2144 uint64_t sse[4]; // total/y/u/v
2145 uint32_t samples[4]; // total/y/u/v
2148 #if CONFIG_VP9_HIGHBITDEPTH
2149 static void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
2150 const YV12_BUFFER_CONFIG *b,
2152 unsigned int bit_depth,
2153 unsigned int in_bit_depth) {
2154 const int widths[3] =
2155 {a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
2156 const int heights[3] =
2157 {a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
2158 const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer };
2159 const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
2160 const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer };
2161 const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
2163 uint64_t total_sse = 0;
2164 uint32_t total_samples = 0;
2165 const double peak = (double)((1 << in_bit_depth) - 1);
2166 const unsigned int input_shift = bit_depth - in_bit_depth;
2168 for (i = 0; i < 3; ++i) {
2169 const int w = widths[i];
2170 const int h = heights[i];
2171 const uint32_t samples = w * h;
2173 if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
2175 sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
2176 b_planes[i], b_strides[i], w, h,
2179 sse = highbd_get_sse(a_planes[i], a_strides[i],
2180 b_planes[i], b_strides[i], w, h);
2183 sse = get_sse(a_planes[i], a_strides[i],
2184 b_planes[i], b_strides[i],
2187 psnr->sse[1 + i] = sse;
2188 psnr->samples[1 + i] = samples;
2189 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
2192 total_samples += samples;
2195 psnr->sse[0] = total_sse;
2196 psnr->samples[0] = total_samples;
2197 psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
2201 #else // !CONFIG_VP9_HIGHBITDEPTH
2203 static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
2205 static const double peak = 255.0;
2206 const int widths[3] = {
2207 a->y_crop_width, a->uv_crop_width, a->uv_crop_width};
2208 const int heights[3] = {
2209 a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
2210 const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer};
2211 const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
2212 const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer};
2213 const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
2215 uint64_t total_sse = 0;
2216 uint32_t total_samples = 0;
2218 for (i = 0; i < 3; ++i) {
2219 const int w = widths[i];
2220 const int h = heights[i];
2221 const uint32_t samples = w * h;
2222 const uint64_t sse = get_sse(a_planes[i], a_strides[i],
2223 b_planes[i], b_strides[i],
2225 psnr->sse[1 + i] = sse;
2226 psnr->samples[1 + i] = samples;
2227 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
2230 total_samples += samples;
2233 psnr->sse[0] = total_sse;
2234 psnr->samples[0] = total_samples;
2235 psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
2238 #endif // CONFIG_VP9_HIGHBITDEPTH
2240 static void generate_psnr_packet(VP10_COMP *cpi) {
2241 struct vpx_codec_cx_pkt pkt;
2244 #if CONFIG_VP9_HIGHBITDEPTH
2245 calc_highbd_psnr(cpi->Source, cpi->common.frame_to_show, &psnr,
2246 cpi->td.mb.e_mbd.bd, cpi->oxcf.input_bit_depth);
2248 calc_psnr(cpi->Source, cpi->common.frame_to_show, &psnr);
2251 for (i = 0; i < 4; ++i) {
2252 pkt.data.psnr.samples[i] = psnr.samples[i];
2253 pkt.data.psnr.sse[i] = psnr.sse[i];
2254 pkt.data.psnr.psnr[i] = psnr.psnr[i];
2256 pkt.kind = VPX_CODEC_PSNR_PKT;
2257 vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
2260 int vp10_use_as_reference(VP10_COMP *cpi, int ref_frame_flags) {
2261 if (ref_frame_flags > 7)
2264 cpi->ref_frame_flags = ref_frame_flags;
2268 void vp10_update_reference(VP10_COMP *cpi, int ref_frame_flags) {
2269 cpi->ext_refresh_golden_frame = (ref_frame_flags & VP9_GOLD_FLAG) != 0;
2270 cpi->ext_refresh_alt_ref_frame = (ref_frame_flags & VP9_ALT_FLAG) != 0;
2271 cpi->ext_refresh_last_frame = (ref_frame_flags & VP9_LAST_FLAG) != 0;
2272 cpi->ext_refresh_frame_flags_pending = 1;
2275 static YV12_BUFFER_CONFIG *get_vp10_ref_frame_buffer(VP10_COMP *cpi,
2276 VP9_REFFRAME ref_frame_flag) {
2277 MV_REFERENCE_FRAME ref_frame = NONE;
2278 if (ref_frame_flag == VP9_LAST_FLAG)
2279 ref_frame = LAST_FRAME;
2280 else if (ref_frame_flag == VP9_GOLD_FLAG)
2281 ref_frame = GOLDEN_FRAME;
2282 else if (ref_frame_flag == VP9_ALT_FLAG)
2283 ref_frame = ALTREF_FRAME;
2285 return ref_frame == NONE ? NULL : get_ref_frame_buffer(cpi, ref_frame);
2288 int vp10_copy_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
2289 YV12_BUFFER_CONFIG *sd) {
2290 YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
2292 vp8_yv12_copy_frame(cfg, sd);
2299 int vp10_set_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
2300 YV12_BUFFER_CONFIG *sd) {
2301 YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
2303 vp8_yv12_copy_frame(sd, cfg);
2310 int vp10_update_entropy(VP10_COMP * cpi, int update) {
2311 cpi->ext_refresh_frame_context = update;
2312 cpi->ext_refresh_frame_context_pending = 1;
2316 #if defined(OUTPUT_YUV_DENOISED) || defined(OUTPUT_YUV_SKINMAP)
2317 // The denoiser buffer is allocated as a YUV 440 buffer. This function writes it
2318 // as YUV 420. We simply use the top-left pixels of the UV buffers, since we do
2319 // not denoise the UV channels at this time. If ever we implement UV channel
2320 // denoising we will have to modify this.
2321 void vp10_write_yuv_frame_420(YV12_BUFFER_CONFIG *s, FILE *f) {
2322 uint8_t *src = s->y_buffer;
2323 int h = s->y_height;
2326 fwrite(src, s->y_width, 1, f);
2334 fwrite(src, s->uv_width, 1, f);
2335 src += s->uv_stride;
2342 fwrite(src, s->uv_width, 1, f);
2343 src += s->uv_stride;
2348 #ifdef OUTPUT_YUV_REC
2349 void vp10_write_yuv_rec_frame(VP10_COMMON *cm) {
2350 YV12_BUFFER_CONFIG *s = cm->frame_to_show;
2351 uint8_t *src = s->y_buffer;
2354 #if CONFIG_VP9_HIGHBITDEPTH
2355 if (s->flags & YV12_FLAG_HIGHBITDEPTH) {
2356 uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer);
2359 fwrite(src16, s->y_width, 2, yuv_rec_file);
2360 src16 += s->y_stride;
2363 src16 = CONVERT_TO_SHORTPTR(s->u_buffer);
2367 fwrite(src16, s->uv_width, 2, yuv_rec_file);
2368 src16 += s->uv_stride;
2371 src16 = CONVERT_TO_SHORTPTR(s->v_buffer);
2375 fwrite(src16, s->uv_width, 2, yuv_rec_file);
2376 src16 += s->uv_stride;
2379 fflush(yuv_rec_file);
2382 #endif // CONFIG_VP9_HIGHBITDEPTH
2385 fwrite(src, s->y_width, 1, yuv_rec_file);
2393 fwrite(src, s->uv_width, 1, yuv_rec_file);
2394 src += s->uv_stride;
2401 fwrite(src, s->uv_width, 1, yuv_rec_file);
2402 src += s->uv_stride;
2405 fflush(yuv_rec_file);
2409 #if CONFIG_VP9_HIGHBITDEPTH
2410 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
2411 YV12_BUFFER_CONFIG *dst,
2414 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
2415 YV12_BUFFER_CONFIG *dst) {
2416 #endif // CONFIG_VP9_HIGHBITDEPTH
2417 // TODO(dkovalev): replace YV12_BUFFER_CONFIG with vpx_image_t
2419 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
2420 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
2421 const int src_widths[3] = {src->y_crop_width, src->uv_crop_width,
2422 src->uv_crop_width };
2423 const int src_heights[3] = {src->y_crop_height, src->uv_crop_height,
2424 src->uv_crop_height};
2425 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
2426 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
2427 const int dst_widths[3] = {dst->y_crop_width, dst->uv_crop_width,
2428 dst->uv_crop_width};
2429 const int dst_heights[3] = {dst->y_crop_height, dst->uv_crop_height,
2430 dst->uv_crop_height};
2432 for (i = 0; i < MAX_MB_PLANE; ++i) {
2433 #if CONFIG_VP9_HIGHBITDEPTH
2434 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
2435 vp10_highbd_resize_plane(srcs[i], src_heights[i], src_widths[i],
2436 src_strides[i], dsts[i], dst_heights[i],
2437 dst_widths[i], dst_strides[i], bd);
2439 vp10_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
2440 dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
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]);
2445 #endif // CONFIG_VP9_HIGHBITDEPTH
2447 vpx_extend_frame_borders(dst);
2450 #if CONFIG_VP9_HIGHBITDEPTH
2451 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
2452 YV12_BUFFER_CONFIG *dst, int bd) {
2454 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
2455 YV12_BUFFER_CONFIG *dst) {
2456 #endif // CONFIG_VP9_HIGHBITDEPTH
2457 const int src_w = src->y_crop_width;
2458 const int src_h = src->y_crop_height;
2459 const int dst_w = dst->y_crop_width;
2460 const int dst_h = dst->y_crop_height;
2461 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
2462 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
2463 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
2464 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
2465 const InterpKernel *const kernel = vp10_filter_kernels[EIGHTTAP];
2468 for (y = 0; y < dst_h; y += 16) {
2469 for (x = 0; x < dst_w; x += 16) {
2470 for (i = 0; i < MAX_MB_PLANE; ++i) {
2471 const int factor = (i == 0 || i == 3 ? 1 : 2);
2472 const int x_q4 = x * (16 / factor) * src_w / dst_w;
2473 const int y_q4 = y * (16 / factor) * src_h / dst_h;
2474 const int src_stride = src_strides[i];
2475 const int dst_stride = dst_strides[i];
2476 const uint8_t *src_ptr = srcs[i] + (y / factor) * src_h / dst_h *
2477 src_stride + (x / factor) * src_w / dst_w;
2478 uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
2480 #if CONFIG_VP9_HIGHBITDEPTH
2481 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
2482 vpx_highbd_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2483 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2484 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2485 16 / factor, 16 / factor, bd);
2487 vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2488 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2489 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2490 16 / factor, 16 / factor);
2493 vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2494 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2495 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2496 16 / factor, 16 / factor);
2497 #endif // CONFIG_VP9_HIGHBITDEPTH
2502 vpx_extend_frame_borders(dst);
2505 static int scale_down(VP10_COMP *cpi, int q) {
2506 RATE_CONTROL *const rc = &cpi->rc;
2507 GF_GROUP *const gf_group = &cpi->twopass.gf_group;
2509 assert(frame_is_kf_gf_arf(cpi));
2511 if (rc->frame_size_selector == UNSCALED &&
2512 q >= rc->rf_level_maxq[gf_group->rf_level[gf_group->index]]) {
2513 const int max_size_thresh = (int)(rate_thresh_mult[SCALE_STEP1]
2514 * VPXMAX(rc->this_frame_target, rc->avg_frame_bandwidth));
2515 scale = rc->projected_frame_size > max_size_thresh ? 1 : 0;
2520 // Function to test for conditions that indicate we should loop
2521 // back and recode a frame.
2522 static int recode_loop_test(VP10_COMP *cpi,
2523 int high_limit, int low_limit,
2524 int q, int maxq, int minq) {
2525 const RATE_CONTROL *const rc = &cpi->rc;
2526 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
2527 const int frame_is_kfgfarf = frame_is_kf_gf_arf(cpi);
2528 int force_recode = 0;
2530 if ((rc->projected_frame_size >= rc->max_frame_bandwidth) ||
2531 (cpi->sf.recode_loop == ALLOW_RECODE) ||
2532 (frame_is_kfgfarf &&
2533 (cpi->sf.recode_loop == ALLOW_RECODE_KFARFGF))) {
2534 if (frame_is_kfgfarf &&
2535 (oxcf->resize_mode == RESIZE_DYNAMIC) &&
2536 scale_down(cpi, q)) {
2537 // Code this group at a lower resolution.
2538 cpi->resize_pending = 1;
2542 // TODO(agrange) high_limit could be greater than the scale-down threshold.
2543 if ((rc->projected_frame_size > high_limit && q < maxq) ||
2544 (rc->projected_frame_size < low_limit && q > minq)) {
2546 } else if (cpi->oxcf.rc_mode == VPX_CQ) {
2547 // Deal with frame undershoot and whether or not we are
2548 // below the automatically set cq level.
2549 if (q > oxcf->cq_level &&
2550 rc->projected_frame_size < ((rc->this_frame_target * 7) >> 3)) {
2555 return force_recode;
2558 void vp10_update_reference_frames(VP10_COMP *cpi) {
2559 VP10_COMMON * const cm = &cpi->common;
2560 BufferPool *const pool = cm->buffer_pool;
2562 // At this point the new frame has been encoded.
2563 // If any buffer copy / swapping is signaled it should be done here.
2564 if (cm->frame_type == KEY_FRAME) {
2565 ref_cnt_fb(pool->frame_bufs,
2566 &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
2567 ref_cnt_fb(pool->frame_bufs,
2568 &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
2569 } else if (vp10_preserve_existing_gf(cpi)) {
2570 // We have decided to preserve the previously existing golden frame as our
2571 // new ARF frame. However, in the short term in function
2572 // vp10_bitstream.c::get_refresh_mask() we left it in the GF slot and, if
2573 // we're updating the GF with the current decoded frame, we save it to the
2574 // ARF slot instead.
2575 // We now have to update the ARF with the current frame and swap gld_fb_idx
2576 // and alt_fb_idx so that, overall, we've stored the old GF in the new ARF
2577 // slot and, if we're updating the GF, the current frame becomes the new GF.
2580 ref_cnt_fb(pool->frame_bufs,
2581 &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
2583 tmp = cpi->alt_fb_idx;
2584 cpi->alt_fb_idx = cpi->gld_fb_idx;
2585 cpi->gld_fb_idx = tmp;
2586 } else { /* For non key/golden frames */
2587 if (cpi->refresh_alt_ref_frame) {
2588 int arf_idx = cpi->alt_fb_idx;
2589 if ((cpi->oxcf.pass == 2) && cpi->multi_arf_allowed) {
2590 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
2591 arf_idx = gf_group->arf_update_idx[gf_group->index];
2594 ref_cnt_fb(pool->frame_bufs,
2595 &cm->ref_frame_map[arf_idx], cm->new_fb_idx);
2596 memcpy(cpi->interp_filter_selected[ALTREF_FRAME],
2597 cpi->interp_filter_selected[0],
2598 sizeof(cpi->interp_filter_selected[0]));
2601 if (cpi->refresh_golden_frame) {
2602 ref_cnt_fb(pool->frame_bufs,
2603 &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
2604 if (!cpi->rc.is_src_frame_alt_ref)
2605 memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
2606 cpi->interp_filter_selected[0],
2607 sizeof(cpi->interp_filter_selected[0]));
2609 memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
2610 cpi->interp_filter_selected[ALTREF_FRAME],
2611 sizeof(cpi->interp_filter_selected[ALTREF_FRAME]));
2615 if (cpi->refresh_last_frame) {
2616 ref_cnt_fb(pool->frame_bufs,
2617 &cm->ref_frame_map[cpi->lst_fb_idx], cm->new_fb_idx);
2618 if (!cpi->rc.is_src_frame_alt_ref)
2619 memcpy(cpi->interp_filter_selected[LAST_FRAME],
2620 cpi->interp_filter_selected[0],
2621 sizeof(cpi->interp_filter_selected[0]));
2623 #if CONFIG_VP9_TEMPORAL_DENOISING
2624 if (cpi->oxcf.noise_sensitivity > 0) {
2625 vp10_denoiser_update_frame_info(&cpi->denoiser,
2627 cpi->common.frame_type,
2628 cpi->refresh_alt_ref_frame,
2629 cpi->refresh_golden_frame,
2630 cpi->refresh_last_frame);
2635 static void loopfilter_frame(VP10_COMP *cpi, VP10_COMMON *cm) {
2636 MACROBLOCKD *xd = &cpi->td.mb.e_mbd;
2637 struct loopfilter *lf = &cm->lf;
2639 lf->filter_level = 0;
2641 struct vpx_usec_timer timer;
2643 vpx_clear_system_state();
2645 vpx_usec_timer_start(&timer);
2647 vp10_pick_filter_level(cpi->Source, cpi, cpi->sf.lpf_pick);
2649 vpx_usec_timer_mark(&timer);
2650 cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
2653 if (lf->filter_level > 0) {
2654 if (cpi->num_workers > 1)
2655 vp10_loop_filter_frame_mt(cm->frame_to_show, cm, xd->plane,
2656 lf->filter_level, 0, 0,
2657 cpi->workers, cpi->num_workers,
2660 vp10_loop_filter_frame(cm->frame_to_show, cm, xd, lf->filter_level, 0, 0);
2663 vpx_extend_frame_inner_borders(cm->frame_to_show);
2666 static INLINE void alloc_frame_mvs(const VP10_COMMON *cm,
2668 RefCntBuffer *const new_fb_ptr = &cm->buffer_pool->frame_bufs[buffer_idx];
2669 if (new_fb_ptr->mvs == NULL ||
2670 new_fb_ptr->mi_rows < cm->mi_rows ||
2671 new_fb_ptr->mi_cols < cm->mi_cols) {
2672 vpx_free(new_fb_ptr->mvs);
2674 (MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols,
2675 sizeof(*new_fb_ptr->mvs));
2676 new_fb_ptr->mi_rows = cm->mi_rows;
2677 new_fb_ptr->mi_cols = cm->mi_cols;
2681 void vp10_scale_references(VP10_COMP *cpi) {
2682 VP10_COMMON *cm = &cpi->common;
2683 MV_REFERENCE_FRAME ref_frame;
2684 const VP9_REFFRAME ref_mask[3] = {VP9_LAST_FLAG, VP9_GOLD_FLAG, VP9_ALT_FLAG};
2686 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
2687 // Need to convert from VP9_REFFRAME to index into ref_mask (subtract 1).
2688 if (cpi->ref_frame_flags & ref_mask[ref_frame - 1]) {
2689 BufferPool *const pool = cm->buffer_pool;
2690 const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi,
2694 cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
2698 #if CONFIG_VP9_HIGHBITDEPTH
2699 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
2700 RefCntBuffer *new_fb_ptr = NULL;
2701 int force_scaling = 0;
2702 int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
2703 if (new_fb == INVALID_IDX) {
2704 new_fb = get_free_fb(cm);
2707 if (new_fb == INVALID_IDX)
2709 new_fb_ptr = &pool->frame_bufs[new_fb];
2710 if (force_scaling ||
2711 new_fb_ptr->buf.y_crop_width != cm->width ||
2712 new_fb_ptr->buf.y_crop_height != cm->height) {
2713 vpx_realloc_frame_buffer(&new_fb_ptr->buf,
2714 cm->width, cm->height,
2715 cm->subsampling_x, cm->subsampling_y,
2716 cm->use_highbitdepth,
2717 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
2719 scale_and_extend_frame(ref, &new_fb_ptr->buf, (int)cm->bit_depth);
2720 cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
2721 alloc_frame_mvs(cm, new_fb);
2724 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
2725 RefCntBuffer *new_fb_ptr = NULL;
2726 int force_scaling = 0;
2727 int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
2728 if (new_fb == INVALID_IDX) {
2729 new_fb = get_free_fb(cm);
2732 if (new_fb == INVALID_IDX)
2734 new_fb_ptr = &pool->frame_bufs[new_fb];
2735 if (force_scaling ||
2736 new_fb_ptr->buf.y_crop_width != cm->width ||
2737 new_fb_ptr->buf.y_crop_height != cm->height) {
2738 vpx_realloc_frame_buffer(&new_fb_ptr->buf,
2739 cm->width, cm->height,
2740 cm->subsampling_x, cm->subsampling_y,
2741 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
2743 scale_and_extend_frame(ref, &new_fb_ptr->buf);
2744 cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
2745 alloc_frame_mvs(cm, new_fb);
2747 #endif // CONFIG_VP9_HIGHBITDEPTH
2749 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
2750 RefCntBuffer *const buf = &pool->frame_bufs[buf_idx];
2751 buf->buf.y_crop_width = ref->y_crop_width;
2752 buf->buf.y_crop_height = ref->y_crop_height;
2753 cpi->scaled_ref_idx[ref_frame - 1] = buf_idx;
2757 if (cpi->oxcf.pass != 0)
2758 cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
2763 static void release_scaled_references(VP10_COMP *cpi) {
2764 VP10_COMMON *cm = &cpi->common;
2766 if (cpi->oxcf.pass == 0) {
2767 // Only release scaled references under certain conditions:
2768 // if reference will be updated, or if scaled reference has same resolution.
2770 refresh[0] = (cpi->refresh_last_frame) ? 1 : 0;
2771 refresh[1] = (cpi->refresh_golden_frame) ? 1 : 0;
2772 refresh[2] = (cpi->refresh_alt_ref_frame) ? 1 : 0;
2773 for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
2774 const int idx = cpi->scaled_ref_idx[i - 1];
2775 RefCntBuffer *const buf = idx != INVALID_IDX ?
2776 &cm->buffer_pool->frame_bufs[idx] : NULL;
2777 const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi, i);
2780 (buf->buf.y_crop_width == ref->y_crop_width &&
2781 buf->buf.y_crop_height == ref->y_crop_height))) {
2783 cpi->scaled_ref_idx[i -1] = INVALID_IDX;
2787 for (i = 0; i < MAX_REF_FRAMES; ++i) {
2788 const int idx = cpi->scaled_ref_idx[i];
2789 RefCntBuffer *const buf = idx != INVALID_IDX ?
2790 &cm->buffer_pool->frame_bufs[idx] : NULL;
2793 cpi->scaled_ref_idx[i] = INVALID_IDX;
2799 static void full_to_model_count(unsigned int *model_count,
2800 unsigned int *full_count) {
2802 model_count[ZERO_TOKEN] = full_count[ZERO_TOKEN];
2803 model_count[ONE_TOKEN] = full_count[ONE_TOKEN];
2804 model_count[TWO_TOKEN] = full_count[TWO_TOKEN];
2805 for (n = THREE_TOKEN; n < EOB_TOKEN; ++n)
2806 model_count[TWO_TOKEN] += full_count[n];
2807 model_count[EOB_MODEL_TOKEN] = full_count[EOB_TOKEN];
2810 static void full_to_model_counts(vp10_coeff_count_model *model_count,
2811 vp10_coeff_count *full_count) {
2814 for (i = 0; i < PLANE_TYPES; ++i)
2815 for (j = 0; j < REF_TYPES; ++j)
2816 for (k = 0; k < COEF_BANDS; ++k)
2817 for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
2818 full_to_model_count(model_count[i][j][k][l], full_count[i][j][k][l]);
2821 #if 0 && CONFIG_INTERNAL_STATS
2822 static void output_frame_level_debug_stats(VP10_COMP *cpi) {
2823 VP10_COMMON *const cm = &cpi->common;
2824 FILE *const f = fopen("tmp.stt", cm->current_video_frame ? "a" : "w");
2827 vpx_clear_system_state();
2829 recon_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
2831 if (cpi->twopass.total_left_stats.coded_error != 0.0)
2832 fprintf(f, "%10u %dx%d %d %d %10d %10d %10d %10d"
2833 "%10"PRId64" %10"PRId64" %5d %5d %10"PRId64" "
2834 "%10"PRId64" %10"PRId64" %10d "
2835 "%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
2836 "%6d %6d %5d %5d %5d "
2837 "%10"PRId64" %10.3lf"
2838 "%10lf %8u %10"PRId64" %10d %10d %10d\n",
2839 cpi->common.current_video_frame,
2840 cm->width, cm->height,
2841 cpi->rc.source_alt_ref_pending,
2842 cpi->rc.source_alt_ref_active,
2843 cpi->rc.this_frame_target,
2844 cpi->rc.projected_frame_size,
2845 cpi->rc.projected_frame_size / cpi->common.MBs,
2846 (cpi->rc.projected_frame_size - cpi->rc.this_frame_target),
2847 cpi->rc.vbr_bits_off_target,
2848 cpi->rc.vbr_bits_off_target_fast,
2849 cpi->twopass.extend_minq,
2850 cpi->twopass.extend_minq_fast,
2851 cpi->rc.total_target_vs_actual,
2852 (cpi->rc.starting_buffer_level - cpi->rc.bits_off_target),
2853 cpi->rc.total_actual_bits, cm->base_qindex,
2854 vp10_convert_qindex_to_q(cm->base_qindex, cm->bit_depth),
2855 (double)vp10_dc_quant(cm->base_qindex, 0, cm->bit_depth) / 4.0,
2856 vp10_convert_qindex_to_q(cpi->twopass.active_worst_quality,
2859 vp10_convert_qindex_to_q(cpi->oxcf.cq_level, cm->bit_depth),
2860 cpi->refresh_last_frame, cpi->refresh_golden_frame,
2861 cpi->refresh_alt_ref_frame, cm->frame_type, cpi->rc.gfu_boost,
2862 cpi->twopass.bits_left,
2863 cpi->twopass.total_left_stats.coded_error,
2864 cpi->twopass.bits_left /
2865 (1 + cpi->twopass.total_left_stats.coded_error),
2866 cpi->tot_recode_hits, recon_err, cpi->rc.kf_boost,
2867 cpi->twopass.kf_zeromotion_pct,
2868 cpi->twopass.fr_content_type);
2873 FILE *const fmodes = fopen("Modes.stt", "a");
2876 fprintf(fmodes, "%6d:%1d:%1d:%1d ", cpi->common.current_video_frame,
2877 cm->frame_type, cpi->refresh_golden_frame,
2878 cpi->refresh_alt_ref_frame);
2880 for (i = 0; i < MAX_MODES; ++i)
2881 fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
2883 fprintf(fmodes, "\n");
2890 static void set_mv_search_params(VP10_COMP *cpi) {
2891 const VP10_COMMON *const cm = &cpi->common;
2892 const unsigned int max_mv_def = VPXMIN(cm->width, cm->height);
2894 // Default based on max resolution.
2895 cpi->mv_step_param = vp10_init_search_range(max_mv_def);
2897 if (cpi->sf.mv.auto_mv_step_size) {
2898 if (frame_is_intra_only(cm)) {
2899 // Initialize max_mv_magnitude for use in the first INTER frame
2900 // after a key/intra-only frame.
2901 cpi->max_mv_magnitude = max_mv_def;
2903 if (cm->show_frame) {
2904 // Allow mv_steps to correspond to twice the max mv magnitude found
2905 // in the previous frame, capped by the default max_mv_magnitude based
2907 cpi->mv_step_param = vp10_init_search_range(
2908 VPXMIN(max_mv_def, 2 * cpi->max_mv_magnitude));
2910 cpi->max_mv_magnitude = 0;
2915 static void set_size_independent_vars(VP10_COMP *cpi) {
2916 vp10_set_speed_features_framesize_independent(cpi);
2917 vp10_set_rd_speed_thresholds(cpi);
2918 vp10_set_rd_speed_thresholds_sub8x8(cpi);
2919 cpi->common.interp_filter = cpi->sf.default_interp_filter;
2922 static void set_size_dependent_vars(VP10_COMP *cpi, int *q,
2923 int *bottom_index, int *top_index) {
2924 VP10_COMMON *const cm = &cpi->common;
2925 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
2927 // Setup variables that depend on the dimensions of the frame.
2928 vp10_set_speed_features_framesize_dependent(cpi);
2930 // Decide q and q bounds.
2931 *q = vp10_rc_pick_q_and_bounds(cpi, bottom_index, top_index);
2933 if (!frame_is_intra_only(cm)) {
2934 vp10_set_high_precision_mv(cpi, (*q) < HIGH_PRECISION_MV_QTHRESH);
2937 // Configure experimental use of segmentation for enhanced coding of
2938 // static regions if indicated.
2939 // Only allowed in the second pass of a two pass encode, as it requires
2940 // lagged coding, and if the relevant speed feature flag is set.
2941 if (oxcf->pass == 2 && cpi->sf.static_segmentation)
2942 configure_static_seg_features(cpi);
2944 #if CONFIG_VP9_POSTPROC
2945 if (oxcf->noise_sensitivity > 0) {
2947 switch (oxcf->noise_sensitivity) {
2965 vp10_denoise(cpi->Source, cpi->Source, l);
2967 #endif // CONFIG_VP9_POSTPROC
2970 static void init_motion_estimation(VP10_COMP *cpi) {
2971 int y_stride = cpi->scaled_source.y_stride;
2973 if (cpi->sf.mv.search_method == NSTEP) {
2974 vp10_init3smotion_compensation(&cpi->ss_cfg, y_stride);
2975 } else if (cpi->sf.mv.search_method == DIAMOND) {
2976 vp10_init_dsmotion_compensation(&cpi->ss_cfg, y_stride);
2980 static void set_frame_size(VP10_COMP *cpi) {
2982 VP10_COMMON *const cm = &cpi->common;
2983 VP10EncoderConfig *const oxcf = &cpi->oxcf;
2984 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
2986 if (oxcf->pass == 2 &&
2987 oxcf->rc_mode == VPX_VBR &&
2988 ((oxcf->resize_mode == RESIZE_FIXED && cm->current_video_frame == 0) ||
2989 (oxcf->resize_mode == RESIZE_DYNAMIC && cpi->resize_pending))) {
2990 vp10_calculate_coded_size(
2991 cpi, &oxcf->scaled_frame_width, &oxcf->scaled_frame_height);
2993 // There has been a change in frame size.
2994 vp10_set_size_literal(cpi, oxcf->scaled_frame_width,
2995 oxcf->scaled_frame_height);
2998 if (oxcf->pass == 0 &&
2999 oxcf->rc_mode == VPX_CBR &&
3000 oxcf->resize_mode == RESIZE_DYNAMIC) {
3001 if (cpi->resize_pending == 1) {
3002 oxcf->scaled_frame_width =
3003 (cm->width * cpi->resize_scale_num) / cpi->resize_scale_den;
3004 oxcf->scaled_frame_height =
3005 (cm->height * cpi->resize_scale_num) /cpi->resize_scale_den;
3006 } else if (cpi->resize_pending == -1) {
3007 // Go back up to original size.
3008 oxcf->scaled_frame_width = oxcf->width;
3009 oxcf->scaled_frame_height = oxcf->height;
3011 if (cpi->resize_pending != 0) {
3012 // There has been a change in frame size.
3013 vp10_set_size_literal(cpi,
3014 oxcf->scaled_frame_width,
3015 oxcf->scaled_frame_height);
3017 // TODO(agrange) Scale cpi->max_mv_magnitude if frame-size has changed.
3018 set_mv_search_params(cpi);
3022 if (oxcf->pass == 2) {
3023 vp10_set_target_rate(cpi);
3026 alloc_frame_mvs(cm, cm->new_fb_idx);
3028 // Reset the frame pointers to the current frame size.
3029 vpx_realloc_frame_buffer(get_frame_new_buffer(cm),
3030 cm->width, cm->height,
3031 cm->subsampling_x, cm->subsampling_y,
3032 #if CONFIG_VP9_HIGHBITDEPTH
3033 cm->use_highbitdepth,
3035 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
3038 alloc_util_frame_buffers(cpi);
3039 init_motion_estimation(cpi);
3041 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
3042 RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - 1];
3043 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
3045 ref_buf->idx = buf_idx;
3047 if (buf_idx != INVALID_IDX) {
3048 YV12_BUFFER_CONFIG *const buf = &cm->buffer_pool->frame_bufs[buf_idx].buf;
3050 #if CONFIG_VP9_HIGHBITDEPTH
3051 vp10_setup_scale_factors_for_frame(&ref_buf->sf,
3052 buf->y_crop_width, buf->y_crop_height,
3053 cm->width, cm->height,
3054 (buf->flags & YV12_FLAG_HIGHBITDEPTH) ?
3057 vp10_setup_scale_factors_for_frame(&ref_buf->sf,
3058 buf->y_crop_width, buf->y_crop_height,
3059 cm->width, cm->height);
3060 #endif // CONFIG_VP9_HIGHBITDEPTH
3061 if (vp10_is_scaled(&ref_buf->sf))
3062 vpx_extend_frame_borders(buf);
3064 ref_buf->buf = NULL;
3068 set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME);
3071 static void encode_without_recode_loop(VP10_COMP *cpi) {
3072 VP10_COMMON *const cm = &cpi->common;
3073 int q = 0, bottom_index = 0, top_index = 0; // Dummy variables.
3075 vpx_clear_system_state();
3077 set_frame_size(cpi);
3079 // For 1 pass CBR under dynamic resize mode: use faster scaling for source.
3080 // Only for 2x2 scaling for now.
3081 if (cpi->oxcf.pass == 0 &&
3082 cpi->oxcf.rc_mode == VPX_CBR &&
3083 cpi->oxcf.resize_mode == RESIZE_DYNAMIC &&
3084 cpi->un_scaled_source->y_width == (cm->width << 1) &&
3085 cpi->un_scaled_source->y_height == (cm->height << 1)) {
3086 cpi->Source = vp10_scale_if_required_fast(cm,
3087 cpi->un_scaled_source,
3088 &cpi->scaled_source);
3089 if (cpi->unscaled_last_source != NULL)
3090 cpi->Last_Source = vp10_scale_if_required_fast(cm,
3091 cpi->unscaled_last_source,
3092 &cpi->scaled_last_source);
3094 cpi->Source = vp10_scale_if_required(cm, cpi->un_scaled_source,
3095 &cpi->scaled_source);
3096 if (cpi->unscaled_last_source != NULL)
3097 cpi->Last_Source = vp10_scale_if_required(cm, cpi->unscaled_last_source,
3098 &cpi->scaled_last_source);
3101 if (frame_is_intra_only(cm) == 0) {
3102 vp10_scale_references(cpi);
3105 set_size_independent_vars(cpi);
3106 set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
3108 vp10_set_quantizer(cm, q);
3109 vp10_set_variance_partition_thresholds(cpi, q);
3113 suppress_active_map(cpi);
3114 // Variance adaptive and in frame q adjustment experiments are mutually
3116 if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
3117 vp10_vaq_frame_setup(cpi);
3118 } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
3119 vp10_setup_in_frame_q_adj(cpi);
3120 } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
3121 vp10_cyclic_refresh_setup(cpi);
3123 apply_active_map(cpi);
3125 // transform / motion compensation build reconstruction frame
3126 vp10_encode_frame(cpi);
3128 // Update some stats from cyclic refresh, and check if we should not update
3129 // golden reference, for 1 pass CBR.
3130 if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
3131 cm->frame_type != KEY_FRAME &&
3132 (cpi->oxcf.pass == 0 && cpi->oxcf.rc_mode == VPX_CBR))
3133 vp10_cyclic_refresh_check_golden_update(cpi);
3135 // Update the skip mb flag probabilities based on the distribution
3136 // seen in the last encoder iteration.
3137 // update_base_skip_probs(cpi);
3138 vpx_clear_system_state();
3141 static void encode_with_recode_loop(VP10_COMP *cpi,
3144 VP10_COMMON *const cm = &cpi->common;
3145 RATE_CONTROL *const rc = &cpi->rc;
3146 int bottom_index, top_index;
3148 int loop_at_this_size = 0;
3150 int overshoot_seen = 0;
3151 int undershoot_seen = 0;
3152 int frame_over_shoot_limit;
3153 int frame_under_shoot_limit;
3154 int q = 0, q_low = 0, q_high = 0;
3156 set_size_independent_vars(cpi);
3159 vpx_clear_system_state();
3161 set_frame_size(cpi);
3163 if (loop_count == 0 || cpi->resize_pending != 0) {
3164 set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
3166 // TODO(agrange) Scale cpi->max_mv_magnitude if frame-size has changed.
3167 set_mv_search_params(cpi);
3169 // Reset the loop state for new frame size.
3171 undershoot_seen = 0;
3173 // Reconfiguration for change in frame size has concluded.
3174 cpi->resize_pending = 0;
3176 q_low = bottom_index;
3179 loop_at_this_size = 0;
3182 // Decide frame size bounds first time through.
3183 if (loop_count == 0) {
3184 vp10_rc_compute_frame_size_bounds(cpi, rc->this_frame_target,
3185 &frame_under_shoot_limit,
3186 &frame_over_shoot_limit);
3189 cpi->Source = vp10_scale_if_required(cm, cpi->un_scaled_source,
3190 &cpi->scaled_source);
3192 if (cpi->unscaled_last_source != NULL)
3193 cpi->Last_Source = vp10_scale_if_required(cm, cpi->unscaled_last_source,
3194 &cpi->scaled_last_source);
3196 if (frame_is_intra_only(cm) == 0) {
3197 if (loop_count > 0) {
3198 release_scaled_references(cpi);
3200 vp10_scale_references(cpi);
3203 vp10_set_quantizer(cm, q);
3205 if (loop_count == 0)
3208 // Variance adaptive and in frame q adjustment experiments are mutually
3210 if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
3211 vp10_vaq_frame_setup(cpi);
3212 } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
3213 vp10_setup_in_frame_q_adj(cpi);
3216 // transform / motion compensation build reconstruction frame
3217 vp10_encode_frame(cpi);
3219 // Update the skip mb flag probabilities based on the distribution
3220 // seen in the last encoder iteration.
3221 // update_base_skip_probs(cpi);
3223 vpx_clear_system_state();
3225 // Dummy pack of the bitstream using up to date stats to get an
3226 // accurate estimate of output frame size to determine if we need
3228 if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) {
3229 save_coding_context(cpi);
3230 vp10_pack_bitstream(cpi, dest, size);
3232 rc->projected_frame_size = (int)(*size) << 3;
3233 restore_coding_context(cpi);
3235 if (frame_over_shoot_limit == 0)
3236 frame_over_shoot_limit = 1;
3239 if (cpi->oxcf.rc_mode == VPX_Q) {
3242 if ((cm->frame_type == KEY_FRAME) &&
3243 rc->this_key_frame_forced &&
3244 (rc->projected_frame_size < rc->max_frame_bandwidth)) {
3248 int64_t high_err_target = cpi->ambient_err;
3249 int64_t low_err_target = cpi->ambient_err >> 1;
3251 #if CONFIG_VP9_HIGHBITDEPTH
3252 if (cm->use_highbitdepth) {
3253 kf_err = vp10_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3255 kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3258 kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3259 #endif // CONFIG_VP9_HIGHBITDEPTH
3261 // Prevent possible divide by zero error below for perfect KF
3264 // The key frame is not good enough or we can afford
3265 // to make it better without undue risk of popping.
3266 if ((kf_err > high_err_target &&
3267 rc->projected_frame_size <= frame_over_shoot_limit) ||
3268 (kf_err > low_err_target &&
3269 rc->projected_frame_size <= frame_under_shoot_limit)) {
3271 q_high = q > q_low ? q - 1 : q_low;
3274 q = (int)((q * high_err_target) / kf_err);
3275 q = VPXMIN(q, (q_high + q_low) >> 1);
3276 } else if (kf_err < low_err_target &&
3277 rc->projected_frame_size >= frame_under_shoot_limit) {
3278 // The key frame is much better than the previous frame
3280 q_low = q < q_high ? q + 1 : q_high;
3283 q = (int)((q * low_err_target) / kf_err);
3284 q = VPXMIN(q, (q_high + q_low + 1) >> 1);
3287 // Clamp Q to upper and lower limits:
3288 q = clamp(q, q_low, q_high);
3291 } else if (recode_loop_test(
3292 cpi, frame_over_shoot_limit, frame_under_shoot_limit,
3293 q, VPXMAX(q_high, top_index), bottom_index)) {
3294 // Is the projected frame size out of range and are we allowed
3295 // to attempt to recode.
3299 if (cpi->resize_pending == 1) {
3300 // Change in frame size so go back around the recode loop.
3301 cpi->rc.frame_size_selector =
3302 SCALE_STEP1 - cpi->rc.frame_size_selector;
3303 cpi->rc.next_frame_size_selector = cpi->rc.frame_size_selector;
3305 #if CONFIG_INTERNAL_STATS
3306 ++cpi->tot_recode_hits;
3313 // Frame size out of permitted range:
3314 // Update correction factor & compute new Q to try...
3316 // Frame is too large
3317 if (rc->projected_frame_size > rc->this_frame_target) {
3318 // Special case if the projected size is > the max allowed.
3319 if (rc->projected_frame_size >= rc->max_frame_bandwidth)
3320 q_high = rc->worst_quality;
3322 // Raise Qlow as to at least the current value
3323 q_low = q < q_high ? q + 1 : q_high;
3325 if (undershoot_seen || loop_at_this_size > 1) {
3326 // Update rate_correction_factor unless
3327 vp10_rc_update_rate_correction_factors(cpi);
3329 q = (q_high + q_low + 1) / 2;
3331 // Update rate_correction_factor unless
3332 vp10_rc_update_rate_correction_factors(cpi);
3334 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3335 bottom_index, VPXMAX(q_high, top_index));
3337 while (q < q_low && retries < 10) {
3338 vp10_rc_update_rate_correction_factors(cpi);
3339 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3340 bottom_index, VPXMAX(q_high, top_index));
3347 // Frame is too small
3348 q_high = q > q_low ? q - 1 : q_low;
3350 if (overshoot_seen || loop_at_this_size > 1) {
3351 vp10_rc_update_rate_correction_factors(cpi);
3352 q = (q_high + q_low) / 2;
3354 vp10_rc_update_rate_correction_factors(cpi);
3355 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3356 bottom_index, top_index);
3357 // Special case reset for qlow for constrained quality.
3358 // This should only trigger where there is very substantial
3359 // undershoot on a frame and the auto cq level is above
3360 // the user passsed in value.
3361 if (cpi->oxcf.rc_mode == VPX_CQ &&
3366 while (q > q_high && retries < 10) {
3367 vp10_rc_update_rate_correction_factors(cpi);
3368 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3369 bottom_index, top_index);
3374 undershoot_seen = 1;
3377 // Clamp Q to upper and lower limits:
3378 q = clamp(q, q_low, q_high);
3380 loop = (q != last_q);
3386 // Special case for overlay frame.
3387 if (rc->is_src_frame_alt_ref &&
3388 rc->projected_frame_size < rc->max_frame_bandwidth)
3393 ++loop_at_this_size;
3395 #if CONFIG_INTERNAL_STATS
3396 ++cpi->tot_recode_hits;
3402 static int get_ref_frame_flags(const VP10_COMP *cpi) {
3403 const int *const map = cpi->common.ref_frame_map;
3404 const int gold_is_last = map[cpi->gld_fb_idx] == map[cpi->lst_fb_idx];
3405 const int alt_is_last = map[cpi->alt_fb_idx] == map[cpi->lst_fb_idx];
3406 const int gold_is_alt = map[cpi->gld_fb_idx] == map[cpi->alt_fb_idx];
3407 int flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
3410 flags &= ~VP9_GOLD_FLAG;
3412 if (cpi->rc.frames_till_gf_update_due == INT_MAX)
3413 flags &= ~VP9_GOLD_FLAG;
3416 flags &= ~VP9_ALT_FLAG;
3419 flags &= ~VP9_ALT_FLAG;
3424 static void set_ext_overrides(VP10_COMP *cpi) {
3425 // Overrides the defaults with the externally supplied values with
3426 // vp10_update_reference() and vp10_update_entropy() calls
3427 // Note: The overrides are valid only for the next frame passed
3428 // to encode_frame_to_data_rate() function
3429 if (cpi->ext_refresh_frame_context_pending) {
3430 cpi->common.refresh_frame_context = cpi->ext_refresh_frame_context;
3431 cpi->ext_refresh_frame_context_pending = 0;
3433 if (cpi->ext_refresh_frame_flags_pending) {
3434 cpi->refresh_last_frame = cpi->ext_refresh_last_frame;
3435 cpi->refresh_golden_frame = cpi->ext_refresh_golden_frame;
3436 cpi->refresh_alt_ref_frame = cpi->ext_refresh_alt_ref_frame;
3437 cpi->ext_refresh_frame_flags_pending = 0;
3441 YV12_BUFFER_CONFIG *vp10_scale_if_required_fast(VP10_COMMON *cm,
3442 YV12_BUFFER_CONFIG *unscaled,
3443 YV12_BUFFER_CONFIG *scaled) {
3444 if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
3445 cm->mi_rows * MI_SIZE != unscaled->y_height) {
3446 // For 2x2 scaling down.
3447 vpx_scale_frame(unscaled, scaled, unscaled->y_buffer, 9, 2, 1,
3449 vpx_extend_frame_borders(scaled);
3456 YV12_BUFFER_CONFIG *vp10_scale_if_required(VP10_COMMON *cm,
3457 YV12_BUFFER_CONFIG *unscaled,
3458 YV12_BUFFER_CONFIG *scaled) {
3459 if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
3460 cm->mi_rows * MI_SIZE != unscaled->y_height) {
3461 #if CONFIG_VP9_HIGHBITDEPTH
3462 scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
3464 scale_and_extend_frame_nonnormative(unscaled, scaled);
3465 #endif // CONFIG_VP9_HIGHBITDEPTH
3472 static void set_arf_sign_bias(VP10_COMP *cpi) {
3473 VP10_COMMON *const cm = &cpi->common;
3476 if ((cpi->oxcf.pass == 2) && cpi->multi_arf_allowed) {
3477 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3478 arf_sign_bias = cpi->rc.source_alt_ref_active &&
3479 (!cpi->refresh_alt_ref_frame ||
3480 (gf_group->rf_level[gf_group->index] == GF_ARF_LOW));
3483 (cpi->rc.source_alt_ref_active && !cpi->refresh_alt_ref_frame);
3485 cm->ref_frame_sign_bias[ALTREF_FRAME] = arf_sign_bias;
3488 static int setup_interp_filter_search_mask(VP10_COMP *cpi) {
3489 INTERP_FILTER ifilter;
3490 int ref_total[MAX_REF_FRAMES] = {0};
3491 MV_REFERENCE_FRAME ref;
3493 if (cpi->common.last_frame_type == KEY_FRAME ||
3494 cpi->refresh_alt_ref_frame)
3496 for (ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref)
3497 for (ifilter = EIGHTTAP; ifilter <= EIGHTTAP_SHARP; ++ifilter)
3498 ref_total[ref] += cpi->interp_filter_selected[ref][ifilter];
3500 for (ifilter = EIGHTTAP; ifilter <= EIGHTTAP_SHARP; ++ifilter) {
3501 if ((ref_total[LAST_FRAME] &&
3502 cpi->interp_filter_selected[LAST_FRAME][ifilter] == 0) &&
3503 (ref_total[GOLDEN_FRAME] == 0 ||
3504 cpi->interp_filter_selected[GOLDEN_FRAME][ifilter] * 50
3505 < ref_total[GOLDEN_FRAME]) &&
3506 (ref_total[ALTREF_FRAME] == 0 ||
3507 cpi->interp_filter_selected[ALTREF_FRAME][ifilter] * 50
3508 < ref_total[ALTREF_FRAME]))
3509 mask |= 1 << ifilter;
3514 static void encode_frame_to_data_rate(VP10_COMP *cpi,
3517 unsigned int *frame_flags) {
3518 VP10_COMMON *const cm = &cpi->common;
3519 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
3520 struct segmentation *const seg = &cm->seg;
3523 set_ext_overrides(cpi);
3524 vpx_clear_system_state();
3526 // Set the arf sign bias for this frame.
3527 set_arf_sign_bias(cpi);
3529 // Set default state for segment based loop filter update flags.
3530 cm->lf.mode_ref_delta_update = 0;
3532 if (cpi->oxcf.pass == 2 &&
3533 cpi->sf.adaptive_interp_filter_search)
3534 cpi->sf.interp_filter_search_mask =
3535 setup_interp_filter_search_mask(cpi);
3537 // Set various flags etc to special state if it is a key frame.
3538 if (frame_is_intra_only(cm)) {
3539 // Reset the loop filter deltas and segmentation map.
3540 vp10_reset_segment_features(&cm->seg);
3542 // If segmentation is enabled force a map update for key frames.
3544 seg->update_map = 1;
3545 seg->update_data = 1;
3548 // The alternate reference frame cannot be active for a key frame.
3549 cpi->rc.source_alt_ref_active = 0;
3551 cm->error_resilient_mode = oxcf->error_resilient_mode;
3552 cm->frame_parallel_decoding_mode = oxcf->frame_parallel_decoding_mode;
3554 // By default, encoder assumes decoder can use prev_mi.
3555 if (cm->error_resilient_mode) {
3556 cm->frame_parallel_decoding_mode = 1;
3557 cm->reset_frame_context = 0;
3558 cm->refresh_frame_context = 0;
3559 } else if (cm->intra_only) {
3560 // Only reset the current context.
3561 cm->reset_frame_context = 2;
3565 // For 1 pass CBR, check if we are dropping this frame.
3566 // Never drop on key frame.
3567 if (oxcf->pass == 0 &&
3568 oxcf->rc_mode == VPX_CBR &&
3569 cm->frame_type != KEY_FRAME) {
3570 if (vp10_rc_drop_frame(cpi)) {
3571 vp10_rc_postencode_update_drop_frame(cpi);
3572 ++cm->current_video_frame;
3577 vpx_clear_system_state();
3579 #if CONFIG_INTERNAL_STATS
3580 memset(cpi->mode_chosen_counts, 0,
3581 MAX_MODES * sizeof(*cpi->mode_chosen_counts));
3584 if (cpi->sf.recode_loop == DISALLOW_RECODE) {
3585 encode_without_recode_loop(cpi);
3587 encode_with_recode_loop(cpi, size, dest);
3590 #if CONFIG_VP9_TEMPORAL_DENOISING
3591 #ifdef OUTPUT_YUV_DENOISED
3592 if (oxcf->noise_sensitivity > 0) {
3593 vp10_write_yuv_frame_420(&cpi->denoiser.running_avg_y[INTRA_FRAME],
3598 #ifdef OUTPUT_YUV_SKINMAP
3599 if (cpi->common.current_video_frame > 1) {
3600 vp10_compute_skin_map(cpi, yuv_skinmap_file);
3604 // Special case code to reduce pulsing when key frames are forced at a
3605 // fixed interval. Note the reconstruction error if it is the frame before
3606 // the force key frame
3607 if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
3608 #if CONFIG_VP9_HIGHBITDEPTH
3609 if (cm->use_highbitdepth) {
3610 cpi->ambient_err = vp10_highbd_get_y_sse(cpi->Source,
3611 get_frame_new_buffer(cm));
3613 cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3616 cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3617 #endif // CONFIG_VP9_HIGHBITDEPTH
3620 // If the encoder forced a KEY_FRAME decision
3621 if (cm->frame_type == KEY_FRAME)
3622 cpi->refresh_last_frame = 1;
3624 cm->frame_to_show = get_frame_new_buffer(cm);
3626 // Pick the loop filter level for the frame.
3627 loopfilter_frame(cpi, cm);
3629 // build the bitstream
3630 vp10_pack_bitstream(cpi, dest, size);
3632 if (cm->seg.update_map)
3633 update_reference_segmentation_map(cpi);
3635 if (frame_is_intra_only(cm) == 0) {
3636 release_scaled_references(cpi);
3638 vp10_update_reference_frames(cpi);
3640 for (t = TX_4X4; t <= TX_32X32; t++)
3641 full_to_model_counts(cpi->td.counts->coef[t],
3642 cpi->td.rd_counts.coef_counts[t]);
3644 if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode)
3645 vp10_adapt_coef_probs(cm);
3647 if (!frame_is_intra_only(cm)) {
3648 if (!cm->error_resilient_mode && !cm->frame_parallel_decoding_mode) {
3649 vp10_adapt_mode_probs(cm);
3650 vp10_adapt_mv_probs(cm, cm->allow_high_precision_mv);
3654 if (cpi->refresh_golden_frame == 1)
3655 cpi->frame_flags |= FRAMEFLAGS_GOLDEN;
3657 cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN;
3659 if (cpi->refresh_alt_ref_frame == 1)
3660 cpi->frame_flags |= FRAMEFLAGS_ALTREF;
3662 cpi->frame_flags &= ~FRAMEFLAGS_ALTREF;
3664 cpi->ref_frame_flags = get_ref_frame_flags(cpi);
3666 cm->last_frame_type = cm->frame_type;
3668 vp10_rc_postencode_update(cpi, *size);
3671 output_frame_level_debug_stats(cpi);
3674 if (cm->frame_type == KEY_FRAME) {
3675 // Tell the caller that the frame was coded as a key frame
3676 *frame_flags = cpi->frame_flags | FRAMEFLAGS_KEY;
3678 *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
3681 // Clear the one shot update flags for segmentation map and mode/ref loop
3683 cm->seg.update_map = 0;
3684 cm->seg.update_data = 0;
3685 cm->lf.mode_ref_delta_update = 0;
3687 // keep track of the last coded dimensions
3688 cm->last_width = cm->width;
3689 cm->last_height = cm->height;
3691 // reset to normal state now that we are done.
3692 if (!cm->show_existing_frame)
3693 cm->last_show_frame = cm->show_frame;
3695 if (cm->show_frame) {
3696 vp10_swap_mi_and_prev_mi(cm);
3697 // Don't increment frame counters if this was an altref buffer
3698 // update not a real frame
3699 ++cm->current_video_frame;
3701 cm->prev_frame = cm->cur_frame;
3704 static void Pass0Encode(VP10_COMP *cpi, size_t *size, uint8_t *dest,
3705 unsigned int *frame_flags) {
3706 if (cpi->oxcf.rc_mode == VPX_CBR) {
3707 vp10_rc_get_one_pass_cbr_params(cpi);
3709 vp10_rc_get_one_pass_vbr_params(cpi);
3711 encode_frame_to_data_rate(cpi, size, dest, frame_flags);
3714 static void Pass2Encode(VP10_COMP *cpi, size_t *size,
3715 uint8_t *dest, unsigned int *frame_flags) {
3716 cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
3717 encode_frame_to_data_rate(cpi, size, dest, frame_flags);
3719 vp10_twopass_postencode_update(cpi);
3722 static void init_ref_frame_bufs(VP10_COMMON *cm) {
3724 BufferPool *const pool = cm->buffer_pool;
3725 cm->new_fb_idx = INVALID_IDX;
3726 for (i = 0; i < REF_FRAMES; ++i) {
3727 cm->ref_frame_map[i] = INVALID_IDX;
3728 pool->frame_bufs[i].ref_count = 0;
3732 static void check_initial_width(VP10_COMP *cpi,
3733 #if CONFIG_VP9_HIGHBITDEPTH
3734 int use_highbitdepth,
3736 int subsampling_x, int subsampling_y) {
3737 VP10_COMMON *const cm = &cpi->common;
3739 if (!cpi->initial_width ||
3740 #if CONFIG_VP9_HIGHBITDEPTH
3741 cm->use_highbitdepth != use_highbitdepth ||
3743 cm->subsampling_x != subsampling_x ||
3744 cm->subsampling_y != subsampling_y) {
3745 cm->subsampling_x = subsampling_x;
3746 cm->subsampling_y = subsampling_y;
3747 #if CONFIG_VP9_HIGHBITDEPTH
3748 cm->use_highbitdepth = use_highbitdepth;
3751 alloc_raw_frame_buffers(cpi);
3752 init_ref_frame_bufs(cm);
3753 alloc_util_frame_buffers(cpi);
3755 init_motion_estimation(cpi); // TODO(agrange) This can be removed.
3757 cpi->initial_width = cm->width;
3758 cpi->initial_height = cm->height;
3759 cpi->initial_mbs = cm->MBs;
3763 #if CONFIG_VP9_TEMPORAL_DENOISING
3764 static void setup_denoiser_buffer(VP10_COMP *cpi) {
3765 VP10_COMMON *const cm = &cpi->common;
3766 if (cpi->oxcf.noise_sensitivity > 0 &&
3767 !cpi->denoiser.frame_buffer_initialized) {
3768 vp10_denoiser_alloc(&(cpi->denoiser), cm->width, cm->height,
3769 cm->subsampling_x, cm->subsampling_y,
3770 #if CONFIG_VP9_HIGHBITDEPTH
3771 cm->use_highbitdepth,
3773 VP9_ENC_BORDER_IN_PIXELS);
3778 int vp10_receive_raw_frame(VP10_COMP *cpi, unsigned int frame_flags,
3779 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
3781 VP10_COMMON *cm = &cpi->common;
3782 struct vpx_usec_timer timer;
3784 const int subsampling_x = sd->subsampling_x;
3785 const int subsampling_y = sd->subsampling_y;
3786 #if CONFIG_VP9_HIGHBITDEPTH
3787 const int use_highbitdepth = sd->flags & YV12_FLAG_HIGHBITDEPTH;
3788 check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y);
3790 check_initial_width(cpi, subsampling_x, subsampling_y);
3791 #endif // CONFIG_VP9_HIGHBITDEPTH
3793 #if CONFIG_VP9_TEMPORAL_DENOISING
3794 setup_denoiser_buffer(cpi);
3796 vpx_usec_timer_start(&timer);
3798 if (vp10_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
3799 #if CONFIG_VP9_HIGHBITDEPTH
3801 #endif // CONFIG_VP9_HIGHBITDEPTH
3804 vpx_usec_timer_mark(&timer);
3805 cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
3807 if ((cm->profile == PROFILE_0 || cm->profile == PROFILE_2) &&
3808 (subsampling_x != 1 || subsampling_y != 1)) {
3809 vpx_internal_error(&cm->error, VPX_CODEC_INVALID_PARAM,
3810 "Non-4:2:0 color format requires profile 1 or 3");
3813 if ((cm->profile == PROFILE_1 || cm->profile == PROFILE_3) &&
3814 (subsampling_x == 1 && subsampling_y == 1)) {
3815 vpx_internal_error(&cm->error, VPX_CODEC_INVALID_PARAM,
3816 "4:2:0 color format requires profile 0 or 2");
3824 static int frame_is_reference(const VP10_COMP *cpi) {
3825 const VP10_COMMON *cm = &cpi->common;
3827 return cm->frame_type == KEY_FRAME ||
3828 cpi->refresh_last_frame ||
3829 cpi->refresh_golden_frame ||
3830 cpi->refresh_alt_ref_frame ||
3831 cm->refresh_frame_context ||
3832 cm->lf.mode_ref_delta_update ||
3833 cm->seg.update_map ||
3834 cm->seg.update_data;
3837 static void adjust_frame_rate(VP10_COMP *cpi,
3838 const struct lookahead_entry *source) {
3839 int64_t this_duration;
3842 if (source->ts_start == cpi->first_time_stamp_ever) {
3843 this_duration = source->ts_end - source->ts_start;
3846 int64_t last_duration = cpi->last_end_time_stamp_seen
3847 - cpi->last_time_stamp_seen;
3849 this_duration = source->ts_end - cpi->last_end_time_stamp_seen;
3851 // do a step update if the duration changes by 10%
3853 step = (int)((this_duration - last_duration) * 10 / last_duration);
3856 if (this_duration) {
3858 vp10_new_framerate(cpi, 10000000.0 / this_duration);
3860 // Average this frame's rate into the last second's average
3861 // frame rate. If we haven't seen 1 second yet, then average
3862 // over the whole interval seen.
3863 const double interval = VPXMIN(
3864 (double)(source->ts_end - cpi->first_time_stamp_ever), 10000000.0);
3865 double avg_duration = 10000000.0 / cpi->framerate;
3866 avg_duration *= (interval - avg_duration + this_duration);
3867 avg_duration /= interval;
3869 vp10_new_framerate(cpi, 10000000.0 / avg_duration);
3872 cpi->last_time_stamp_seen = source->ts_start;
3873 cpi->last_end_time_stamp_seen = source->ts_end;
3876 // Returns 0 if this is not an alt ref else the offset of the source frame
3877 // used as the arf midpoint.
3878 static int get_arf_src_index(VP10_COMP *cpi) {
3879 RATE_CONTROL *const rc = &cpi->rc;
3880 int arf_src_index = 0;
3881 if (is_altref_enabled(cpi)) {
3882 if (cpi->oxcf.pass == 2) {
3883 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3884 if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
3885 arf_src_index = gf_group->arf_src_offset[gf_group->index];
3887 } else if (rc->source_alt_ref_pending) {
3888 arf_src_index = rc->frames_till_gf_update_due;
3891 return arf_src_index;
3894 static void check_src_altref(VP10_COMP *cpi,
3895 const struct lookahead_entry *source) {
3896 RATE_CONTROL *const rc = &cpi->rc;
3898 if (cpi->oxcf.pass == 2) {
3899 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3900 rc->is_src_frame_alt_ref =
3901 (gf_group->update_type[gf_group->index] == OVERLAY_UPDATE);
3903 rc->is_src_frame_alt_ref = cpi->alt_ref_source &&
3904 (source == cpi->alt_ref_source);
3907 if (rc->is_src_frame_alt_ref) {
3908 // Current frame is an ARF overlay frame.
3909 cpi->alt_ref_source = NULL;
3911 // Don't refresh the last buffer for an ARF overlay frame. It will
3912 // become the GF so preserve last as an alternative prediction option.
3913 cpi->refresh_last_frame = 0;
3917 #if CONFIG_INTERNAL_STATS
3918 extern double vp10_get_blockiness(const unsigned char *img1, int img1_pitch,
3919 const unsigned char *img2, int img2_pitch,
3920 int width, int height);
3922 static void adjust_image_stat(double y, double u, double v, double all,
3927 s->stat[ALL] += all;
3928 s->worst = VPXMIN(s->worst, all);
3930 #endif // CONFIG_INTERNAL_STATS
3932 int vp10_get_compressed_data(VP10_COMP *cpi, unsigned int *frame_flags,
3933 size_t *size, uint8_t *dest,
3934 int64_t *time_stamp, int64_t *time_end, int flush) {
3935 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
3936 VP10_COMMON *const cm = &cpi->common;
3937 BufferPool *const pool = cm->buffer_pool;
3938 RATE_CONTROL *const rc = &cpi->rc;
3939 struct vpx_usec_timer cmptimer;
3940 YV12_BUFFER_CONFIG *force_src_buffer = NULL;
3941 struct lookahead_entry *last_source = NULL;
3942 struct lookahead_entry *source = NULL;
3946 vpx_usec_timer_start(&cmptimer);
3948 vp10_set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV);
3950 // Is multi-arf enabled.
3951 // Note that at the moment multi_arf is only configured for 2 pass VBR
3952 if ((oxcf->pass == 2) && (cpi->oxcf.enable_auto_arf > 1))
3953 cpi->multi_arf_allowed = 1;
3955 cpi->multi_arf_allowed = 0;
3958 cm->reset_frame_context = 0;
3959 cm->refresh_frame_context = 1;
3961 cpi->refresh_last_frame = 1;
3962 cpi->refresh_golden_frame = 0;
3963 cpi->refresh_alt_ref_frame = 0;
3965 // Should we encode an arf frame.
3966 arf_src_index = get_arf_src_index(cpi);
3968 if (arf_src_index) {
3969 assert(arf_src_index <= rc->frames_to_key);
3971 if ((source = vp10_lookahead_peek(cpi->lookahead, arf_src_index)) != NULL) {
3972 cpi->alt_ref_source = source;
3974 if (oxcf->arnr_max_frames > 0) {
3975 // Produce the filtered ARF frame.
3976 vp10_temporal_filter(cpi, arf_src_index);
3977 vpx_extend_frame_borders(&cpi->alt_ref_buffer);
3978 force_src_buffer = &cpi->alt_ref_buffer;
3983 cpi->refresh_alt_ref_frame = 1;
3984 cpi->refresh_golden_frame = 0;
3985 cpi->refresh_last_frame = 0;
3986 rc->is_src_frame_alt_ref = 0;
3987 rc->source_alt_ref_pending = 0;
3989 rc->source_alt_ref_pending = 0;
3994 // Get last frame source.
3995 if (cm->current_video_frame > 0) {
3996 if ((last_source = vp10_lookahead_peek(cpi->lookahead, -1)) == NULL)
4000 // Read in the source frame.
4001 source = vp10_lookahead_pop(cpi->lookahead, flush);
4003 if (source != NULL) {
4007 // Check to see if the frame should be encoded as an arf overlay.
4008 check_src_altref(cpi, source);
4013 cpi->un_scaled_source = cpi->Source = force_src_buffer ? force_src_buffer
4016 cpi->unscaled_last_source = last_source != NULL ? &last_source->img : NULL;
4018 *time_stamp = source->ts_start;
4019 *time_end = source->ts_end;
4020 *frame_flags = (source->flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
4024 if (flush && oxcf->pass == 1 && !cpi->twopass.first_pass_done) {
4025 vp10_end_first_pass(cpi); /* get last stats packet */
4026 cpi->twopass.first_pass_done = 1;
4031 if (source->ts_start < cpi->first_time_stamp_ever) {
4032 cpi->first_time_stamp_ever = source->ts_start;
4033 cpi->last_end_time_stamp_seen = source->ts_start;
4036 // Clear down mmx registers
4037 vpx_clear_system_state();
4039 // adjust frame rates based on timestamps given
4040 if (cm->show_frame) {
4041 adjust_frame_rate(cpi, source);
4044 // Find a free buffer for the new frame, releasing the reference previously
4046 if (cm->new_fb_idx != INVALID_IDX) {
4047 --pool->frame_bufs[cm->new_fb_idx].ref_count;
4049 cm->new_fb_idx = get_free_fb(cm);
4051 if (cm->new_fb_idx == INVALID_IDX)
4054 cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx];
4056 if (cpi->multi_arf_allowed) {
4057 if (cm->frame_type == KEY_FRAME) {
4058 init_buffer_indices(cpi);
4059 } else if (oxcf->pass == 2) {
4060 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
4061 cpi->alt_fb_idx = gf_group->arf_ref_idx[gf_group->index];
4065 // Start with a 0 size frame.
4068 cpi->frame_flags = *frame_flags;
4070 if (oxcf->pass == 2) {
4071 vp10_rc_get_second_pass_params(cpi);
4072 } else if (oxcf->pass == 1) {
4073 set_frame_size(cpi);
4076 if (cpi->oxcf.pass != 0 || frame_is_intra_only(cm) == 1) {
4077 for (i = 0; i < MAX_REF_FRAMES; ++i)
4078 cpi->scaled_ref_idx[i] = INVALID_IDX;
4081 if (oxcf->pass == 1) {
4082 const int lossless = is_lossless_requested(oxcf);
4083 #if CONFIG_VP9_HIGHBITDEPTH
4084 if (cpi->oxcf.use_highbitdepth)
4085 cpi->td.mb.fwd_txm4x4 = lossless ?
4086 vp10_highbd_fwht4x4 : vpx_highbd_fdct4x4;
4088 cpi->td.mb.fwd_txm4x4 = lossless ? vp10_fwht4x4 : vpx_fdct4x4;
4089 cpi->td.mb.highbd_itxm_add = lossless ? vp10_highbd_iwht4x4_add :
4090 vp10_highbd_idct4x4_add;
4092 cpi->td.mb.fwd_txm4x4 = lossless ? vp10_fwht4x4 : vpx_fdct4x4;
4093 #endif // CONFIG_VP9_HIGHBITDEPTH
4094 cpi->td.mb.itxm_add = lossless ? vp10_iwht4x4_add : vp10_idct4x4_add;
4095 vp10_first_pass(cpi, source);
4096 } else if (oxcf->pass == 2) {
4097 Pass2Encode(cpi, size, dest, frame_flags);
4100 Pass0Encode(cpi, size, dest, frame_flags);
4103 if (cm->refresh_frame_context)
4104 cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
4106 // No frame encoded, or frame was dropped, release scaled references.
4107 if ((*size == 0) && (frame_is_intra_only(cm) == 0)) {
4108 release_scaled_references(cpi);
4112 cpi->droppable = !frame_is_reference(cpi);
4115 vpx_usec_timer_mark(&cmptimer);
4116 cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
4118 if (cpi->b_calculate_psnr && oxcf->pass != 1 && cm->show_frame)
4119 generate_psnr_packet(cpi);
4121 #if CONFIG_INTERNAL_STATS
4123 if (oxcf->pass != 1) {
4124 double samples = 0.0;
4125 cpi->bytes += (int)(*size);
4127 if (cm->show_frame) {
4130 if (cpi->b_calculate_psnr) {
4131 YV12_BUFFER_CONFIG *orig = cpi->Source;
4132 YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
4133 YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer;
4135 #if CONFIG_VP9_HIGHBITDEPTH
4136 calc_highbd_psnr(orig, recon, &psnr, cpi->td.mb.e_mbd.bd,
4137 cpi->oxcf.input_bit_depth);
4139 calc_psnr(orig, recon, &psnr);
4140 #endif // CONFIG_VP9_HIGHBITDEPTH
4142 adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3],
4143 psnr.psnr[0], &cpi->psnr);
4144 cpi->total_sq_error += psnr.sse[0];
4145 cpi->total_samples += psnr.samples[0];
4146 samples = psnr.samples[0];
4150 double frame_ssim2 = 0, weight = 0;
4151 #if CONFIG_VP9_POSTPROC
4152 if (vpx_alloc_frame_buffer(&cm->post_proc_buffer,
4153 recon->y_crop_width, recon->y_crop_height,
4154 cm->subsampling_x, cm->subsampling_y,
4155 #if CONFIG_VP9_HIGHBITDEPTH
4156 cm->use_highbitdepth,
4158 VP9_ENC_BORDER_IN_PIXELS,
4159 cm->byte_alignment) < 0) {
4160 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
4161 "Failed to allocate post processing buffer");
4164 vp10_deblock(cm->frame_to_show, &cm->post_proc_buffer,
4165 cm->lf.filter_level * 10 / 6);
4167 vpx_clear_system_state();
4169 #if CONFIG_VP9_HIGHBITDEPTH
4170 calc_highbd_psnr(orig, pp, &psnr2, cpi->td.mb.e_mbd.bd,
4171 cpi->oxcf.input_bit_depth);
4173 calc_psnr(orig, pp, &psnr2);
4174 #endif // CONFIG_VP9_HIGHBITDEPTH
4176 cpi->totalp_sq_error += psnr2.sse[0];
4177 cpi->totalp_samples += psnr2.samples[0];
4178 adjust_image_stat(psnr2.psnr[1], psnr2.psnr[2], psnr2.psnr[3],
4179 psnr2.psnr[0], &cpi->psnrp);
4181 #if CONFIG_VP9_HIGHBITDEPTH
4182 if (cm->use_highbitdepth) {
4183 frame_ssim2 = vpx_highbd_calc_ssim(orig, recon, &weight,
4184 (int)cm->bit_depth);
4186 frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
4189 frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
4190 #endif // CONFIG_VP9_HIGHBITDEPTH
4192 cpi->worst_ssim= VPXMIN(cpi->worst_ssim, frame_ssim2);
4193 cpi->summed_quality += frame_ssim2 * weight;
4194 cpi->summed_weights += weight;
4196 #if CONFIG_VP9_HIGHBITDEPTH
4197 if (cm->use_highbitdepth) {
4198 frame_ssim2 = vpx_highbd_calc_ssim(
4199 orig, &cm->post_proc_buffer, &weight, (int)cm->bit_depth);
4201 frame_ssim2 = vpx_calc_ssim(orig, &cm->post_proc_buffer, &weight);
4204 frame_ssim2 = vpx_calc_ssim(orig, &cm->post_proc_buffer, &weight);
4205 #endif // CONFIG_VP9_HIGHBITDEPTH
4207 cpi->summedp_quality += frame_ssim2 * weight;
4208 cpi->summedp_weights += weight;
4211 FILE *f = fopen("q_used.stt", "a");
4212 fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
4213 cpi->common.current_video_frame, y2, u2, v2,
4214 frame_psnr2, frame_ssim2);
4220 if (cpi->b_calculate_blockiness) {
4221 #if CONFIG_VP9_HIGHBITDEPTH
4222 if (!cm->use_highbitdepth)
4225 double frame_blockiness = vp10_get_blockiness(
4226 cpi->Source->y_buffer, cpi->Source->y_stride,
4227 cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
4228 cpi->Source->y_width, cpi->Source->y_height);
4229 cpi->worst_blockiness =
4230 VPXMAX(cpi->worst_blockiness, frame_blockiness);
4231 cpi->total_blockiness += frame_blockiness;
4235 if (cpi->b_calculate_consistency) {
4236 #if CONFIG_VP9_HIGHBITDEPTH
4237 if (!cm->use_highbitdepth)
4240 double this_inconsistency = vpx_get_ssim_metrics(
4241 cpi->Source->y_buffer, cpi->Source->y_stride,
4242 cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
4243 cpi->Source->y_width, cpi->Source->y_height, cpi->ssim_vars,
4246 const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1);
4247 double consistency = vpx_sse_to_psnr(samples, peak,
4248 (double)cpi->total_inconsistency);
4249 if (consistency > 0.0)
4250 cpi->worst_consistency =
4251 VPXMIN(cpi->worst_consistency, consistency);
4252 cpi->total_inconsistency += this_inconsistency;
4256 if (cpi->b_calculate_ssimg) {
4257 double y, u, v, frame_all;
4258 #if CONFIG_VP9_HIGHBITDEPTH
4259 if (cm->use_highbitdepth) {
4260 frame_all = vpx_highbd_calc_ssimg(cpi->Source, cm->frame_to_show, &y,
4261 &u, &v, (int)cm->bit_depth);
4263 frame_all = vpx_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u,
4267 frame_all = vpx_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u, &v);
4268 #endif // CONFIG_VP9_HIGHBITDEPTH
4269 adjust_image_stat(y, u, v, frame_all, &cpi->ssimg);
4271 #if CONFIG_VP9_HIGHBITDEPTH
4272 if (!cm->use_highbitdepth)
4275 double y, u, v, frame_all;
4276 frame_all = vpx_calc_fastssim(cpi->Source, cm->frame_to_show, &y, &u,
4278 adjust_image_stat(y, u, v, frame_all, &cpi->fastssim);
4279 /* TODO(JBB): add 10/12 bit support */
4281 #if CONFIG_VP9_HIGHBITDEPTH
4282 if (!cm->use_highbitdepth)
4285 double y, u, v, frame_all;
4286 frame_all = vpx_psnrhvs(cpi->Source, cm->frame_to_show, &y, &u, &v);
4287 adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs);
4293 vpx_clear_system_state();
4297 int vp10_get_preview_raw_frame(VP10_COMP *cpi, YV12_BUFFER_CONFIG *dest,
4298 vp10_ppflags_t *flags) {
4299 VP10_COMMON *cm = &cpi->common;
4300 #if !CONFIG_VP9_POSTPROC
4304 if (!cm->show_frame) {
4308 #if CONFIG_VP9_POSTPROC
4309 ret = vp10_post_proc_frame(cm, dest, flags);
4311 if (cm->frame_to_show) {
4312 *dest = *cm->frame_to_show;
4313 dest->y_width = cm->width;
4314 dest->y_height = cm->height;
4315 dest->uv_width = cm->width >> cm->subsampling_x;
4316 dest->uv_height = cm->height >> cm->subsampling_y;
4321 #endif // !CONFIG_VP9_POSTPROC
4322 vpx_clear_system_state();
4327 int vp10_set_internal_size(VP10_COMP *cpi,
4328 VPX_SCALING horiz_mode, VPX_SCALING vert_mode) {
4329 VP10_COMMON *cm = &cpi->common;
4330 int hr = 0, hs = 0, vr = 0, vs = 0;
4332 if (horiz_mode > ONETWO || vert_mode > ONETWO)
4335 Scale2Ratio(horiz_mode, &hr, &hs);
4336 Scale2Ratio(vert_mode, &vr, &vs);
4338 // always go to the next whole number
4339 cm->width = (hs - 1 + cpi->oxcf.width * hr) / hs;
4340 cm->height = (vs - 1 + cpi->oxcf.height * vr) / vs;
4341 assert(cm->width <= cpi->initial_width);
4342 assert(cm->height <= cpi->initial_height);
4344 update_frame_size(cpi);
4349 int vp10_set_size_literal(VP10_COMP *cpi, unsigned int width,
4350 unsigned int height) {
4351 VP10_COMMON *cm = &cpi->common;
4352 #if CONFIG_VP9_HIGHBITDEPTH
4353 check_initial_width(cpi, cm->use_highbitdepth, 1, 1);
4355 check_initial_width(cpi, 1, 1);
4356 #endif // CONFIG_VP9_HIGHBITDEPTH
4358 #if CONFIG_VP9_TEMPORAL_DENOISING
4359 setup_denoiser_buffer(cpi);
4364 if (cm->width > cpi->initial_width) {
4365 cm->width = cpi->initial_width;
4366 printf("Warning: Desired width too large, changed to %d\n", cm->width);
4371 cm->height = height;
4372 if (cm->height > cpi->initial_height) {
4373 cm->height = cpi->initial_height;
4374 printf("Warning: Desired height too large, changed to %d\n", cm->height);
4377 assert(cm->width <= cpi->initial_width);
4378 assert(cm->height <= cpi->initial_height);
4380 update_frame_size(cpi);
4385 int64_t vp10_get_y_sse(const YV12_BUFFER_CONFIG *a,
4386 const YV12_BUFFER_CONFIG *b) {
4387 assert(a->y_crop_width == b->y_crop_width);
4388 assert(a->y_crop_height == b->y_crop_height);
4390 return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
4391 a->y_crop_width, a->y_crop_height);
4394 #if CONFIG_VP9_HIGHBITDEPTH
4395 int64_t vp10_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
4396 const YV12_BUFFER_CONFIG *b) {
4397 assert(a->y_crop_width == b->y_crop_width);
4398 assert(a->y_crop_height == b->y_crop_height);
4399 assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
4400 assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
4402 return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
4403 a->y_crop_width, a->y_crop_height);
4405 #endif // CONFIG_VP9_HIGHBITDEPTH
4407 int vp10_get_quantizer(VP10_COMP *cpi) {
4408 return cpi->common.base_qindex;
4411 void vp10_apply_encoding_flags(VP10_COMP *cpi, vpx_enc_frame_flags_t flags) {
4412 if (flags & (VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF |
4413 VP8_EFLAG_NO_REF_ARF)) {
4416 if (flags & VP8_EFLAG_NO_REF_LAST)
4417 ref ^= VP9_LAST_FLAG;
4419 if (flags & VP8_EFLAG_NO_REF_GF)
4420 ref ^= VP9_GOLD_FLAG;
4422 if (flags & VP8_EFLAG_NO_REF_ARF)
4423 ref ^= VP9_ALT_FLAG;
4425 vp10_use_as_reference(cpi, ref);
4428 if (flags & (VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
4429 VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_FORCE_GF |
4430 VP8_EFLAG_FORCE_ARF)) {
4433 if (flags & VP8_EFLAG_NO_UPD_LAST)
4434 upd ^= VP9_LAST_FLAG;
4436 if (flags & VP8_EFLAG_NO_UPD_GF)
4437 upd ^= VP9_GOLD_FLAG;
4439 if (flags & VP8_EFLAG_NO_UPD_ARF)
4440 upd ^= VP9_ALT_FLAG;
4442 vp10_update_reference(cpi, upd);
4445 if (flags & VP8_EFLAG_NO_UPD_ENTROPY) {
4446 vp10_update_entropy(cpi, 0);