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 #if !CONFIG_MISC_FIXES
419 vp10_copy(cc->segment_pred_probs, cm->segp.pred_probs);
422 memcpy(cpi->coding_context.last_frame_seg_map_copy,
423 cm->last_frame_seg_map, (cm->mi_rows * cm->mi_cols));
425 vp10_copy(cc->last_ref_lf_deltas, cm->lf.last_ref_deltas);
426 vp10_copy(cc->last_mode_lf_deltas, cm->lf.last_mode_deltas);
431 static void restore_coding_context(VP10_COMP *cpi) {
432 CODING_CONTEXT *const cc = &cpi->coding_context;
433 VP10_COMMON *cm = &cpi->common;
435 // Restore key state variables to the snapshot state stored in the
436 // previous call to vp10_save_coding_context.
437 vp10_copy(cpi->td.mb.nmvjointcost, cc->nmvjointcost);
439 memcpy(cpi->nmvcosts[0], cc->nmvcosts[0], MV_VALS * sizeof(*cc->nmvcosts[0]));
440 memcpy(cpi->nmvcosts[1], cc->nmvcosts[1], MV_VALS * sizeof(*cc->nmvcosts[1]));
441 memcpy(cpi->nmvcosts_hp[0], cc->nmvcosts_hp[0],
442 MV_VALS * sizeof(*cc->nmvcosts_hp[0]));
443 memcpy(cpi->nmvcosts_hp[1], cc->nmvcosts_hp[1],
444 MV_VALS * sizeof(*cc->nmvcosts_hp[1]));
446 #if !CONFIG_MISC_FIXES
447 vp10_copy(cm->segp.pred_probs, cc->segment_pred_probs);
450 memcpy(cm->last_frame_seg_map,
451 cpi->coding_context.last_frame_seg_map_copy,
452 (cm->mi_rows * cm->mi_cols));
454 vp10_copy(cm->lf.last_ref_deltas, cc->last_ref_lf_deltas);
455 vp10_copy(cm->lf.last_mode_deltas, cc->last_mode_lf_deltas);
460 static void configure_static_seg_features(VP10_COMP *cpi) {
461 VP10_COMMON *const cm = &cpi->common;
462 const RATE_CONTROL *const rc = &cpi->rc;
463 struct segmentation *const seg = &cm->seg;
465 int high_q = (int)(rc->avg_q > 48.0);
468 // Disable and clear down for KF
469 if (cm->frame_type == KEY_FRAME) {
470 // Clear down the global segmentation map
471 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
473 seg->update_data = 0;
474 cpi->static_mb_pct = 0;
476 // Disable segmentation
477 vp10_disable_segmentation(seg);
479 // Clear down the segment features.
480 vp10_clearall_segfeatures(seg);
481 } else if (cpi->refresh_alt_ref_frame) {
482 // If this is an alt ref frame
483 // Clear down the global segmentation map
484 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
486 seg->update_data = 0;
487 cpi->static_mb_pct = 0;
489 // Disable segmentation and individual segment features by default
490 vp10_disable_segmentation(seg);
491 vp10_clearall_segfeatures(seg);
493 // Scan frames from current to arf frame.
494 // This function re-enables segmentation if appropriate.
495 vp10_update_mbgraph_stats(cpi);
497 // If segmentation was enabled set those features needed for the
501 seg->update_data = 1;
503 qi_delta = vp10_compute_qdelta(rc, rc->avg_q, rc->avg_q * 0.875,
505 vp10_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2);
506 vp10_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
508 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
509 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
511 // Where relevant assume segment data is delta data
512 seg->abs_delta = SEGMENT_DELTADATA;
514 } else if (seg->enabled) {
515 // All other frames if segmentation has been enabled
517 // First normal frame in a valid gf or alt ref group
518 if (rc->frames_since_golden == 0) {
519 // Set up segment features for normal frames in an arf group
520 if (rc->source_alt_ref_active) {
522 seg->update_data = 1;
523 seg->abs_delta = SEGMENT_DELTADATA;
525 qi_delta = vp10_compute_qdelta(rc, rc->avg_q, rc->avg_q * 1.125,
527 vp10_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta + 2);
528 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
530 vp10_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
531 vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
533 // Segment coding disabled for compred testing
534 if (high_q || (cpi->static_mb_pct == 100)) {
535 vp10_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
536 vp10_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
537 vp10_enable_segfeature(seg, 1, SEG_LVL_SKIP);
540 // Disable segmentation and clear down features if alt ref
541 // is not active for this group
543 vp10_disable_segmentation(seg);
545 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
548 seg->update_data = 0;
550 vp10_clearall_segfeatures(seg);
552 } else if (rc->is_src_frame_alt_ref) {
553 // Special case where we are coding over the top of a previous
555 // Segment coding disabled for compred testing
557 // Enable ref frame features for segment 0 as well
558 vp10_enable_segfeature(seg, 0, SEG_LVL_REF_FRAME);
559 vp10_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
561 // All mbs should use ALTREF_FRAME
562 vp10_clear_segdata(seg, 0, SEG_LVL_REF_FRAME);
563 vp10_set_segdata(seg, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME);
564 vp10_clear_segdata(seg, 1, SEG_LVL_REF_FRAME);
565 vp10_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
567 // Skip all MBs if high Q (0,0 mv and skip coeffs)
569 vp10_enable_segfeature(seg, 0, SEG_LVL_SKIP);
570 vp10_enable_segfeature(seg, 1, SEG_LVL_SKIP);
572 // Enable data update
573 seg->update_data = 1;
577 // No updates.. leave things as they are.
579 seg->update_data = 0;
584 static void update_reference_segmentation_map(VP10_COMP *cpi) {
585 VP10_COMMON *const cm = &cpi->common;
586 MODE_INFO **mi_8x8_ptr = cm->mi_grid_visible;
587 uint8_t *cache_ptr = cm->last_frame_seg_map;
590 for (row = 0; row < cm->mi_rows; row++) {
591 MODE_INFO **mi_8x8 = mi_8x8_ptr;
592 uint8_t *cache = cache_ptr;
593 for (col = 0; col < cm->mi_cols; col++, mi_8x8++, cache++)
594 cache[0] = mi_8x8[0]->mbmi.segment_id;
595 mi_8x8_ptr += cm->mi_stride;
596 cache_ptr += cm->mi_cols;
600 static void alloc_raw_frame_buffers(VP10_COMP *cpi) {
601 VP10_COMMON *cm = &cpi->common;
602 const VP10EncoderConfig *oxcf = &cpi->oxcf;
605 cpi->lookahead = vp10_lookahead_init(oxcf->width, oxcf->height,
606 cm->subsampling_x, cm->subsampling_y,
607 #if CONFIG_VP9_HIGHBITDEPTH
608 cm->use_highbitdepth,
610 oxcf->lag_in_frames);
612 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
613 "Failed to allocate lag buffers");
615 // TODO(agrange) Check if ARF is enabled and skip allocation if not.
616 if (vpx_realloc_frame_buffer(&cpi->alt_ref_buffer,
617 oxcf->width, oxcf->height,
618 cm->subsampling_x, cm->subsampling_y,
619 #if CONFIG_VP9_HIGHBITDEPTH
620 cm->use_highbitdepth,
622 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
624 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
625 "Failed to allocate altref buffer");
628 static void alloc_util_frame_buffers(VP10_COMP *cpi) {
629 VP10_COMMON *const cm = &cpi->common;
630 if (vpx_realloc_frame_buffer(&cpi->last_frame_uf,
631 cm->width, cm->height,
632 cm->subsampling_x, cm->subsampling_y,
633 #if CONFIG_VP9_HIGHBITDEPTH
634 cm->use_highbitdepth,
636 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
638 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
639 "Failed to allocate last frame buffer");
641 if (vpx_realloc_frame_buffer(&cpi->scaled_source,
642 cm->width, cm->height,
643 cm->subsampling_x, cm->subsampling_y,
644 #if CONFIG_VP9_HIGHBITDEPTH
645 cm->use_highbitdepth,
647 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
649 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
650 "Failed to allocate scaled source buffer");
652 if (vpx_realloc_frame_buffer(&cpi->scaled_last_source,
653 cm->width, cm->height,
654 cm->subsampling_x, cm->subsampling_y,
655 #if CONFIG_VP9_HIGHBITDEPTH
656 cm->use_highbitdepth,
658 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
660 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
661 "Failed to allocate scaled last source buffer");
665 static int alloc_context_buffers_ext(VP10_COMP *cpi) {
666 VP10_COMMON *cm = &cpi->common;
667 int mi_size = cm->mi_cols * cm->mi_rows;
669 cpi->mbmi_ext_base = vpx_calloc(mi_size, sizeof(*cpi->mbmi_ext_base));
670 if (!cpi->mbmi_ext_base)
676 void vp10_alloc_compressor_data(VP10_COMP *cpi) {
677 VP10_COMMON *cm = &cpi->common;
679 vp10_alloc_context_buffers(cm, cm->width, cm->height);
681 alloc_context_buffers_ext(cpi);
683 vpx_free(cpi->tile_tok[0][0]);
686 unsigned int tokens = get_token_alloc(cm->mb_rows, cm->mb_cols);
687 CHECK_MEM_ERROR(cm, cpi->tile_tok[0][0],
688 vpx_calloc(tokens, sizeof(*cpi->tile_tok[0][0])));
691 vp10_setup_pc_tree(&cpi->common, &cpi->td);
694 void vp10_new_framerate(VP10_COMP *cpi, double framerate) {
695 cpi->framerate = framerate < 0.1 ? 30 : framerate;
696 vp10_rc_update_framerate(cpi);
699 static void set_tile_limits(VP10_COMP *cpi) {
700 VP10_COMMON *const cm = &cpi->common;
702 int min_log2_tile_cols, max_log2_tile_cols;
703 vp10_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
705 cm->log2_tile_cols = clamp(cpi->oxcf.tile_columns,
706 min_log2_tile_cols, max_log2_tile_cols);
707 cm->log2_tile_rows = cpi->oxcf.tile_rows;
710 static void update_frame_size(VP10_COMP *cpi) {
711 VP10_COMMON *const cm = &cpi->common;
712 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
714 vp10_set_mb_mi(cm, cm->width, cm->height);
715 vp10_init_context_buffers(cm);
716 vp10_init_macroblockd(cm, xd, NULL);
717 memset(cpi->mbmi_ext_base, 0,
718 cm->mi_rows * cm->mi_cols * sizeof(*cpi->mbmi_ext_base));
720 set_tile_limits(cpi);
723 static void init_buffer_indices(VP10_COMP *cpi) {
729 static void init_config(struct VP10_COMP *cpi, VP10EncoderConfig *oxcf) {
730 VP10_COMMON *const cm = &cpi->common;
733 cpi->framerate = oxcf->init_framerate;
735 cm->profile = oxcf->profile;
736 cm->bit_depth = oxcf->bit_depth;
737 #if CONFIG_VP9_HIGHBITDEPTH
738 cm->use_highbitdepth = oxcf->use_highbitdepth;
740 cm->color_space = oxcf->color_space;
741 cm->color_range = oxcf->color_range;
743 cm->width = oxcf->width;
744 cm->height = oxcf->height;
745 vp10_alloc_compressor_data(cpi);
747 // Single thread case: use counts in common.
748 cpi->td.counts = &cm->counts;
750 // change includes all joint functionality
751 vp10_change_config(cpi, oxcf);
753 cpi->static_mb_pct = 0;
754 cpi->ref_frame_flags = 0;
756 init_buffer_indices(cpi);
759 static void set_rc_buffer_sizes(RATE_CONTROL *rc,
760 const VP10EncoderConfig *oxcf) {
761 const int64_t bandwidth = oxcf->target_bandwidth;
762 const int64_t starting = oxcf->starting_buffer_level_ms;
763 const int64_t optimal = oxcf->optimal_buffer_level_ms;
764 const int64_t maximum = oxcf->maximum_buffer_size_ms;
766 rc->starting_buffer_level = starting * bandwidth / 1000;
767 rc->optimal_buffer_level = (optimal == 0) ? bandwidth / 8
768 : optimal * bandwidth / 1000;
769 rc->maximum_buffer_size = (maximum == 0) ? bandwidth / 8
770 : maximum * bandwidth / 1000;
773 #if CONFIG_VP9_HIGHBITDEPTH
774 #define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF) \
775 cpi->fn_ptr[BT].sdf = SDF; \
776 cpi->fn_ptr[BT].sdaf = SDAF; \
777 cpi->fn_ptr[BT].vf = VF; \
778 cpi->fn_ptr[BT].svf = SVF; \
779 cpi->fn_ptr[BT].svaf = SVAF; \
780 cpi->fn_ptr[BT].sdx3f = SDX3F; \
781 cpi->fn_ptr[BT].sdx8f = SDX8F; \
782 cpi->fn_ptr[BT].sdx4df = SDX4DF;
784 #define MAKE_BFP_SAD_WRAPPER(fnname) \
785 static unsigned int fnname##_bits8(const uint8_t *src_ptr, \
787 const uint8_t *ref_ptr, \
789 return fnname(src_ptr, source_stride, ref_ptr, ref_stride); \
791 static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
793 const uint8_t *ref_ptr, \
795 return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 2; \
797 static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
799 const uint8_t *ref_ptr, \
801 return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 4; \
804 #define MAKE_BFP_SADAVG_WRAPPER(fnname) static unsigned int \
805 fnname##_bits8(const uint8_t *src_ptr, \
807 const uint8_t *ref_ptr, \
809 const uint8_t *second_pred) { \
810 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred); \
812 static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
814 const uint8_t *ref_ptr, \
816 const uint8_t *second_pred) { \
817 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
820 static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
822 const uint8_t *ref_ptr, \
824 const uint8_t *second_pred) { \
825 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
829 #define MAKE_BFP_SAD3_WRAPPER(fnname) \
830 static void fnname##_bits8(const uint8_t *src_ptr, \
832 const uint8_t *ref_ptr, \
834 unsigned int *sad_array) { \
835 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
837 static void fnname##_bits10(const uint8_t *src_ptr, \
839 const uint8_t *ref_ptr, \
841 unsigned int *sad_array) { \
843 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
844 for (i = 0; i < 3; i++) \
845 sad_array[i] >>= 2; \
847 static void fnname##_bits12(const uint8_t *src_ptr, \
849 const uint8_t *ref_ptr, \
851 unsigned int *sad_array) { \
853 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
854 for (i = 0; i < 3; i++) \
855 sad_array[i] >>= 4; \
858 #define MAKE_BFP_SAD8_WRAPPER(fnname) \
859 static void fnname##_bits8(const uint8_t *src_ptr, \
861 const uint8_t *ref_ptr, \
863 unsigned int *sad_array) { \
864 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
866 static void fnname##_bits10(const uint8_t *src_ptr, \
868 const uint8_t *ref_ptr, \
870 unsigned int *sad_array) { \
872 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
873 for (i = 0; i < 8; i++) \
874 sad_array[i] >>= 2; \
876 static void fnname##_bits12(const uint8_t *src_ptr, \
878 const uint8_t *ref_ptr, \
880 unsigned int *sad_array) { \
882 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
883 for (i = 0; i < 8; i++) \
884 sad_array[i] >>= 4; \
886 #define MAKE_BFP_SAD4D_WRAPPER(fnname) \
887 static void fnname##_bits8(const uint8_t *src_ptr, \
889 const uint8_t* const ref_ptr[], \
891 unsigned int *sad_array) { \
892 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
894 static void fnname##_bits10(const uint8_t *src_ptr, \
896 const uint8_t* const ref_ptr[], \
898 unsigned int *sad_array) { \
900 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
901 for (i = 0; i < 4; i++) \
902 sad_array[i] >>= 2; \
904 static void fnname##_bits12(const uint8_t *src_ptr, \
906 const uint8_t* const ref_ptr[], \
908 unsigned int *sad_array) { \
910 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
911 for (i = 0; i < 4; i++) \
912 sad_array[i] >>= 4; \
915 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x16)
916 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x16_avg)
917 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x16x4d)
918 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x32)
919 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x32_avg)
920 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x32x4d)
921 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad64x32)
922 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad64x32_avg)
923 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad64x32x4d)
924 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x64)
925 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x64_avg)
926 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x64x4d)
927 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x32)
928 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x32_avg)
929 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad32x32x3)
930 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad32x32x8)
931 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x32x4d)
932 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad64x64)
933 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad64x64_avg)
934 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad64x64x3)
935 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad64x64x8)
936 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad64x64x4d)
937 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x16)
938 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x16_avg)
939 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x16x3)
940 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x16x8)
941 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x16x4d)
942 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x8)
943 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x8_avg)
944 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x8x3)
945 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x8x8)
946 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x8x4d)
947 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x16)
948 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x16_avg)
949 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x16x3)
950 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x16x8)
951 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x16x4d)
952 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x8)
953 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x8_avg)
954 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x8x3)
955 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x8x8)
956 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x8x4d)
957 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x4)
958 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x4_avg)
959 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x4x8)
960 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x4x4d)
961 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x8)
962 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x8_avg)
963 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x8x8)
964 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x8x4d)
965 MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x4)
966 MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x4_avg)
967 MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad4x4x3)
968 MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x4x8)
969 MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x4x4d)
971 static void highbd_set_var_fns(VP10_COMP *const cpi) {
972 VP10_COMMON *const cm = &cpi->common;
973 if (cm->use_highbitdepth) {
974 switch (cm->bit_depth) {
976 HIGHBD_BFP(BLOCK_32X16,
977 vpx_highbd_sad32x16_bits8,
978 vpx_highbd_sad32x16_avg_bits8,
979 vpx_highbd_8_variance32x16,
980 vpx_highbd_8_sub_pixel_variance32x16,
981 vpx_highbd_8_sub_pixel_avg_variance32x16,
984 vpx_highbd_sad32x16x4d_bits8)
986 HIGHBD_BFP(BLOCK_16X32,
987 vpx_highbd_sad16x32_bits8,
988 vpx_highbd_sad16x32_avg_bits8,
989 vpx_highbd_8_variance16x32,
990 vpx_highbd_8_sub_pixel_variance16x32,
991 vpx_highbd_8_sub_pixel_avg_variance16x32,
994 vpx_highbd_sad16x32x4d_bits8)
996 HIGHBD_BFP(BLOCK_64X32,
997 vpx_highbd_sad64x32_bits8,
998 vpx_highbd_sad64x32_avg_bits8,
999 vpx_highbd_8_variance64x32,
1000 vpx_highbd_8_sub_pixel_variance64x32,
1001 vpx_highbd_8_sub_pixel_avg_variance64x32,
1004 vpx_highbd_sad64x32x4d_bits8)
1006 HIGHBD_BFP(BLOCK_32X64,
1007 vpx_highbd_sad32x64_bits8,
1008 vpx_highbd_sad32x64_avg_bits8,
1009 vpx_highbd_8_variance32x64,
1010 vpx_highbd_8_sub_pixel_variance32x64,
1011 vpx_highbd_8_sub_pixel_avg_variance32x64,
1014 vpx_highbd_sad32x64x4d_bits8)
1016 HIGHBD_BFP(BLOCK_32X32,
1017 vpx_highbd_sad32x32_bits8,
1018 vpx_highbd_sad32x32_avg_bits8,
1019 vpx_highbd_8_variance32x32,
1020 vpx_highbd_8_sub_pixel_variance32x32,
1021 vpx_highbd_8_sub_pixel_avg_variance32x32,
1022 vpx_highbd_sad32x32x3_bits8,
1023 vpx_highbd_sad32x32x8_bits8,
1024 vpx_highbd_sad32x32x4d_bits8)
1026 HIGHBD_BFP(BLOCK_64X64,
1027 vpx_highbd_sad64x64_bits8,
1028 vpx_highbd_sad64x64_avg_bits8,
1029 vpx_highbd_8_variance64x64,
1030 vpx_highbd_8_sub_pixel_variance64x64,
1031 vpx_highbd_8_sub_pixel_avg_variance64x64,
1032 vpx_highbd_sad64x64x3_bits8,
1033 vpx_highbd_sad64x64x8_bits8,
1034 vpx_highbd_sad64x64x4d_bits8)
1036 HIGHBD_BFP(BLOCK_16X16,
1037 vpx_highbd_sad16x16_bits8,
1038 vpx_highbd_sad16x16_avg_bits8,
1039 vpx_highbd_8_variance16x16,
1040 vpx_highbd_8_sub_pixel_variance16x16,
1041 vpx_highbd_8_sub_pixel_avg_variance16x16,
1042 vpx_highbd_sad16x16x3_bits8,
1043 vpx_highbd_sad16x16x8_bits8,
1044 vpx_highbd_sad16x16x4d_bits8)
1046 HIGHBD_BFP(BLOCK_16X8,
1047 vpx_highbd_sad16x8_bits8,
1048 vpx_highbd_sad16x8_avg_bits8,
1049 vpx_highbd_8_variance16x8,
1050 vpx_highbd_8_sub_pixel_variance16x8,
1051 vpx_highbd_8_sub_pixel_avg_variance16x8,
1052 vpx_highbd_sad16x8x3_bits8,
1053 vpx_highbd_sad16x8x8_bits8,
1054 vpx_highbd_sad16x8x4d_bits8)
1056 HIGHBD_BFP(BLOCK_8X16,
1057 vpx_highbd_sad8x16_bits8,
1058 vpx_highbd_sad8x16_avg_bits8,
1059 vpx_highbd_8_variance8x16,
1060 vpx_highbd_8_sub_pixel_variance8x16,
1061 vpx_highbd_8_sub_pixel_avg_variance8x16,
1062 vpx_highbd_sad8x16x3_bits8,
1063 vpx_highbd_sad8x16x8_bits8,
1064 vpx_highbd_sad8x16x4d_bits8)
1066 HIGHBD_BFP(BLOCK_8X8,
1067 vpx_highbd_sad8x8_bits8,
1068 vpx_highbd_sad8x8_avg_bits8,
1069 vpx_highbd_8_variance8x8,
1070 vpx_highbd_8_sub_pixel_variance8x8,
1071 vpx_highbd_8_sub_pixel_avg_variance8x8,
1072 vpx_highbd_sad8x8x3_bits8,
1073 vpx_highbd_sad8x8x8_bits8,
1074 vpx_highbd_sad8x8x4d_bits8)
1076 HIGHBD_BFP(BLOCK_8X4,
1077 vpx_highbd_sad8x4_bits8,
1078 vpx_highbd_sad8x4_avg_bits8,
1079 vpx_highbd_8_variance8x4,
1080 vpx_highbd_8_sub_pixel_variance8x4,
1081 vpx_highbd_8_sub_pixel_avg_variance8x4,
1083 vpx_highbd_sad8x4x8_bits8,
1084 vpx_highbd_sad8x4x4d_bits8)
1086 HIGHBD_BFP(BLOCK_4X8,
1087 vpx_highbd_sad4x8_bits8,
1088 vpx_highbd_sad4x8_avg_bits8,
1089 vpx_highbd_8_variance4x8,
1090 vpx_highbd_8_sub_pixel_variance4x8,
1091 vpx_highbd_8_sub_pixel_avg_variance4x8,
1093 vpx_highbd_sad4x8x8_bits8,
1094 vpx_highbd_sad4x8x4d_bits8)
1096 HIGHBD_BFP(BLOCK_4X4,
1097 vpx_highbd_sad4x4_bits8,
1098 vpx_highbd_sad4x4_avg_bits8,
1099 vpx_highbd_8_variance4x4,
1100 vpx_highbd_8_sub_pixel_variance4x4,
1101 vpx_highbd_8_sub_pixel_avg_variance4x4,
1102 vpx_highbd_sad4x4x3_bits8,
1103 vpx_highbd_sad4x4x8_bits8,
1104 vpx_highbd_sad4x4x4d_bits8)
1108 HIGHBD_BFP(BLOCK_32X16,
1109 vpx_highbd_sad32x16_bits10,
1110 vpx_highbd_sad32x16_avg_bits10,
1111 vpx_highbd_10_variance32x16,
1112 vpx_highbd_10_sub_pixel_variance32x16,
1113 vpx_highbd_10_sub_pixel_avg_variance32x16,
1116 vpx_highbd_sad32x16x4d_bits10)
1118 HIGHBD_BFP(BLOCK_16X32,
1119 vpx_highbd_sad16x32_bits10,
1120 vpx_highbd_sad16x32_avg_bits10,
1121 vpx_highbd_10_variance16x32,
1122 vpx_highbd_10_sub_pixel_variance16x32,
1123 vpx_highbd_10_sub_pixel_avg_variance16x32,
1126 vpx_highbd_sad16x32x4d_bits10)
1128 HIGHBD_BFP(BLOCK_64X32,
1129 vpx_highbd_sad64x32_bits10,
1130 vpx_highbd_sad64x32_avg_bits10,
1131 vpx_highbd_10_variance64x32,
1132 vpx_highbd_10_sub_pixel_variance64x32,
1133 vpx_highbd_10_sub_pixel_avg_variance64x32,
1136 vpx_highbd_sad64x32x4d_bits10)
1138 HIGHBD_BFP(BLOCK_32X64,
1139 vpx_highbd_sad32x64_bits10,
1140 vpx_highbd_sad32x64_avg_bits10,
1141 vpx_highbd_10_variance32x64,
1142 vpx_highbd_10_sub_pixel_variance32x64,
1143 vpx_highbd_10_sub_pixel_avg_variance32x64,
1146 vpx_highbd_sad32x64x4d_bits10)
1148 HIGHBD_BFP(BLOCK_32X32,
1149 vpx_highbd_sad32x32_bits10,
1150 vpx_highbd_sad32x32_avg_bits10,
1151 vpx_highbd_10_variance32x32,
1152 vpx_highbd_10_sub_pixel_variance32x32,
1153 vpx_highbd_10_sub_pixel_avg_variance32x32,
1154 vpx_highbd_sad32x32x3_bits10,
1155 vpx_highbd_sad32x32x8_bits10,
1156 vpx_highbd_sad32x32x4d_bits10)
1158 HIGHBD_BFP(BLOCK_64X64,
1159 vpx_highbd_sad64x64_bits10,
1160 vpx_highbd_sad64x64_avg_bits10,
1161 vpx_highbd_10_variance64x64,
1162 vpx_highbd_10_sub_pixel_variance64x64,
1163 vpx_highbd_10_sub_pixel_avg_variance64x64,
1164 vpx_highbd_sad64x64x3_bits10,
1165 vpx_highbd_sad64x64x8_bits10,
1166 vpx_highbd_sad64x64x4d_bits10)
1168 HIGHBD_BFP(BLOCK_16X16,
1169 vpx_highbd_sad16x16_bits10,
1170 vpx_highbd_sad16x16_avg_bits10,
1171 vpx_highbd_10_variance16x16,
1172 vpx_highbd_10_sub_pixel_variance16x16,
1173 vpx_highbd_10_sub_pixel_avg_variance16x16,
1174 vpx_highbd_sad16x16x3_bits10,
1175 vpx_highbd_sad16x16x8_bits10,
1176 vpx_highbd_sad16x16x4d_bits10)
1178 HIGHBD_BFP(BLOCK_16X8,
1179 vpx_highbd_sad16x8_bits10,
1180 vpx_highbd_sad16x8_avg_bits10,
1181 vpx_highbd_10_variance16x8,
1182 vpx_highbd_10_sub_pixel_variance16x8,
1183 vpx_highbd_10_sub_pixel_avg_variance16x8,
1184 vpx_highbd_sad16x8x3_bits10,
1185 vpx_highbd_sad16x8x8_bits10,
1186 vpx_highbd_sad16x8x4d_bits10)
1188 HIGHBD_BFP(BLOCK_8X16,
1189 vpx_highbd_sad8x16_bits10,
1190 vpx_highbd_sad8x16_avg_bits10,
1191 vpx_highbd_10_variance8x16,
1192 vpx_highbd_10_sub_pixel_variance8x16,
1193 vpx_highbd_10_sub_pixel_avg_variance8x16,
1194 vpx_highbd_sad8x16x3_bits10,
1195 vpx_highbd_sad8x16x8_bits10,
1196 vpx_highbd_sad8x16x4d_bits10)
1198 HIGHBD_BFP(BLOCK_8X8,
1199 vpx_highbd_sad8x8_bits10,
1200 vpx_highbd_sad8x8_avg_bits10,
1201 vpx_highbd_10_variance8x8,
1202 vpx_highbd_10_sub_pixel_variance8x8,
1203 vpx_highbd_10_sub_pixel_avg_variance8x8,
1204 vpx_highbd_sad8x8x3_bits10,
1205 vpx_highbd_sad8x8x8_bits10,
1206 vpx_highbd_sad8x8x4d_bits10)
1208 HIGHBD_BFP(BLOCK_8X4,
1209 vpx_highbd_sad8x4_bits10,
1210 vpx_highbd_sad8x4_avg_bits10,
1211 vpx_highbd_10_variance8x4,
1212 vpx_highbd_10_sub_pixel_variance8x4,
1213 vpx_highbd_10_sub_pixel_avg_variance8x4,
1215 vpx_highbd_sad8x4x8_bits10,
1216 vpx_highbd_sad8x4x4d_bits10)
1218 HIGHBD_BFP(BLOCK_4X8,
1219 vpx_highbd_sad4x8_bits10,
1220 vpx_highbd_sad4x8_avg_bits10,
1221 vpx_highbd_10_variance4x8,
1222 vpx_highbd_10_sub_pixel_variance4x8,
1223 vpx_highbd_10_sub_pixel_avg_variance4x8,
1225 vpx_highbd_sad4x8x8_bits10,
1226 vpx_highbd_sad4x8x4d_bits10)
1228 HIGHBD_BFP(BLOCK_4X4,
1229 vpx_highbd_sad4x4_bits10,
1230 vpx_highbd_sad4x4_avg_bits10,
1231 vpx_highbd_10_variance4x4,
1232 vpx_highbd_10_sub_pixel_variance4x4,
1233 vpx_highbd_10_sub_pixel_avg_variance4x4,
1234 vpx_highbd_sad4x4x3_bits10,
1235 vpx_highbd_sad4x4x8_bits10,
1236 vpx_highbd_sad4x4x4d_bits10)
1240 HIGHBD_BFP(BLOCK_32X16,
1241 vpx_highbd_sad32x16_bits12,
1242 vpx_highbd_sad32x16_avg_bits12,
1243 vpx_highbd_12_variance32x16,
1244 vpx_highbd_12_sub_pixel_variance32x16,
1245 vpx_highbd_12_sub_pixel_avg_variance32x16,
1248 vpx_highbd_sad32x16x4d_bits12)
1250 HIGHBD_BFP(BLOCK_16X32,
1251 vpx_highbd_sad16x32_bits12,
1252 vpx_highbd_sad16x32_avg_bits12,
1253 vpx_highbd_12_variance16x32,
1254 vpx_highbd_12_sub_pixel_variance16x32,
1255 vpx_highbd_12_sub_pixel_avg_variance16x32,
1258 vpx_highbd_sad16x32x4d_bits12)
1260 HIGHBD_BFP(BLOCK_64X32,
1261 vpx_highbd_sad64x32_bits12,
1262 vpx_highbd_sad64x32_avg_bits12,
1263 vpx_highbd_12_variance64x32,
1264 vpx_highbd_12_sub_pixel_variance64x32,
1265 vpx_highbd_12_sub_pixel_avg_variance64x32,
1268 vpx_highbd_sad64x32x4d_bits12)
1270 HIGHBD_BFP(BLOCK_32X64,
1271 vpx_highbd_sad32x64_bits12,
1272 vpx_highbd_sad32x64_avg_bits12,
1273 vpx_highbd_12_variance32x64,
1274 vpx_highbd_12_sub_pixel_variance32x64,
1275 vpx_highbd_12_sub_pixel_avg_variance32x64,
1278 vpx_highbd_sad32x64x4d_bits12)
1280 HIGHBD_BFP(BLOCK_32X32,
1281 vpx_highbd_sad32x32_bits12,
1282 vpx_highbd_sad32x32_avg_bits12,
1283 vpx_highbd_12_variance32x32,
1284 vpx_highbd_12_sub_pixel_variance32x32,
1285 vpx_highbd_12_sub_pixel_avg_variance32x32,
1286 vpx_highbd_sad32x32x3_bits12,
1287 vpx_highbd_sad32x32x8_bits12,
1288 vpx_highbd_sad32x32x4d_bits12)
1290 HIGHBD_BFP(BLOCK_64X64,
1291 vpx_highbd_sad64x64_bits12,
1292 vpx_highbd_sad64x64_avg_bits12,
1293 vpx_highbd_12_variance64x64,
1294 vpx_highbd_12_sub_pixel_variance64x64,
1295 vpx_highbd_12_sub_pixel_avg_variance64x64,
1296 vpx_highbd_sad64x64x3_bits12,
1297 vpx_highbd_sad64x64x8_bits12,
1298 vpx_highbd_sad64x64x4d_bits12)
1300 HIGHBD_BFP(BLOCK_16X16,
1301 vpx_highbd_sad16x16_bits12,
1302 vpx_highbd_sad16x16_avg_bits12,
1303 vpx_highbd_12_variance16x16,
1304 vpx_highbd_12_sub_pixel_variance16x16,
1305 vpx_highbd_12_sub_pixel_avg_variance16x16,
1306 vpx_highbd_sad16x16x3_bits12,
1307 vpx_highbd_sad16x16x8_bits12,
1308 vpx_highbd_sad16x16x4d_bits12)
1310 HIGHBD_BFP(BLOCK_16X8,
1311 vpx_highbd_sad16x8_bits12,
1312 vpx_highbd_sad16x8_avg_bits12,
1313 vpx_highbd_12_variance16x8,
1314 vpx_highbd_12_sub_pixel_variance16x8,
1315 vpx_highbd_12_sub_pixel_avg_variance16x8,
1316 vpx_highbd_sad16x8x3_bits12,
1317 vpx_highbd_sad16x8x8_bits12,
1318 vpx_highbd_sad16x8x4d_bits12)
1320 HIGHBD_BFP(BLOCK_8X16,
1321 vpx_highbd_sad8x16_bits12,
1322 vpx_highbd_sad8x16_avg_bits12,
1323 vpx_highbd_12_variance8x16,
1324 vpx_highbd_12_sub_pixel_variance8x16,
1325 vpx_highbd_12_sub_pixel_avg_variance8x16,
1326 vpx_highbd_sad8x16x3_bits12,
1327 vpx_highbd_sad8x16x8_bits12,
1328 vpx_highbd_sad8x16x4d_bits12)
1330 HIGHBD_BFP(BLOCK_8X8,
1331 vpx_highbd_sad8x8_bits12,
1332 vpx_highbd_sad8x8_avg_bits12,
1333 vpx_highbd_12_variance8x8,
1334 vpx_highbd_12_sub_pixel_variance8x8,
1335 vpx_highbd_12_sub_pixel_avg_variance8x8,
1336 vpx_highbd_sad8x8x3_bits12,
1337 vpx_highbd_sad8x8x8_bits12,
1338 vpx_highbd_sad8x8x4d_bits12)
1340 HIGHBD_BFP(BLOCK_8X4,
1341 vpx_highbd_sad8x4_bits12,
1342 vpx_highbd_sad8x4_avg_bits12,
1343 vpx_highbd_12_variance8x4,
1344 vpx_highbd_12_sub_pixel_variance8x4,
1345 vpx_highbd_12_sub_pixel_avg_variance8x4,
1347 vpx_highbd_sad8x4x8_bits12,
1348 vpx_highbd_sad8x4x4d_bits12)
1350 HIGHBD_BFP(BLOCK_4X8,
1351 vpx_highbd_sad4x8_bits12,
1352 vpx_highbd_sad4x8_avg_bits12,
1353 vpx_highbd_12_variance4x8,
1354 vpx_highbd_12_sub_pixel_variance4x8,
1355 vpx_highbd_12_sub_pixel_avg_variance4x8,
1357 vpx_highbd_sad4x8x8_bits12,
1358 vpx_highbd_sad4x8x4d_bits12)
1360 HIGHBD_BFP(BLOCK_4X4,
1361 vpx_highbd_sad4x4_bits12,
1362 vpx_highbd_sad4x4_avg_bits12,
1363 vpx_highbd_12_variance4x4,
1364 vpx_highbd_12_sub_pixel_variance4x4,
1365 vpx_highbd_12_sub_pixel_avg_variance4x4,
1366 vpx_highbd_sad4x4x3_bits12,
1367 vpx_highbd_sad4x4x8_bits12,
1368 vpx_highbd_sad4x4x4d_bits12)
1372 assert(0 && "cm->bit_depth should be VPX_BITS_8, "
1373 "VPX_BITS_10 or VPX_BITS_12");
1377 #endif // CONFIG_VP9_HIGHBITDEPTH
1379 static void realloc_segmentation_maps(VP10_COMP *cpi) {
1380 VP10_COMMON *const cm = &cpi->common;
1382 // Create the encoder segmentation map and set all entries to 0
1383 vpx_free(cpi->segmentation_map);
1384 CHECK_MEM_ERROR(cm, cpi->segmentation_map,
1385 vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
1387 // Create a map used for cyclic background refresh.
1388 if (cpi->cyclic_refresh)
1389 vp10_cyclic_refresh_free(cpi->cyclic_refresh);
1390 CHECK_MEM_ERROR(cm, cpi->cyclic_refresh,
1391 vp10_cyclic_refresh_alloc(cm->mi_rows, cm->mi_cols));
1393 // Create a map used to mark inactive areas.
1394 vpx_free(cpi->active_map.map);
1395 CHECK_MEM_ERROR(cm, cpi->active_map.map,
1396 vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
1398 // And a place holder structure is the coding context
1399 // for use if we want to save and restore it
1400 vpx_free(cpi->coding_context.last_frame_seg_map_copy);
1401 CHECK_MEM_ERROR(cm, cpi->coding_context.last_frame_seg_map_copy,
1402 vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
1405 void vp10_change_config(struct VP10_COMP *cpi, const VP10EncoderConfig *oxcf) {
1406 VP10_COMMON *const cm = &cpi->common;
1407 RATE_CONTROL *const rc = &cpi->rc;
1409 if (cm->profile != oxcf->profile)
1410 cm->profile = oxcf->profile;
1411 cm->bit_depth = oxcf->bit_depth;
1412 cm->color_space = oxcf->color_space;
1413 cm->color_range = oxcf->color_range;
1415 if (cm->profile <= PROFILE_1)
1416 assert(cm->bit_depth == VPX_BITS_8);
1418 assert(cm->bit_depth > VPX_BITS_8);
1421 #if CONFIG_VP9_HIGHBITDEPTH
1422 cpi->td.mb.e_mbd.bd = (int)cm->bit_depth;
1423 #endif // CONFIG_VP9_HIGHBITDEPTH
1425 rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
1427 cpi->refresh_golden_frame = 0;
1428 cpi->refresh_last_frame = 1;
1429 cm->refresh_frame_context =
1430 oxcf->error_resilient_mode ? REFRESH_FRAME_CONTEXT_OFF :
1431 oxcf->frame_parallel_decoding_mode ? REFRESH_FRAME_CONTEXT_FORWARD
1432 : REFRESH_FRAME_CONTEXT_BACKWARD;
1433 cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
1435 vp10_reset_segment_features(cm);
1436 vp10_set_high_precision_mv(cpi, 0);
1441 for (i = 0; i < MAX_SEGMENTS; i++)
1442 cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout;
1444 cpi->encode_breakout = cpi->oxcf.encode_breakout;
1446 set_rc_buffer_sizes(rc, &cpi->oxcf);
1448 // Under a configuration change, where maximum_buffer_size may change,
1449 // keep buffer level clipped to the maximum allowed buffer size.
1450 rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
1451 rc->buffer_level = VPXMIN(rc->buffer_level, rc->maximum_buffer_size);
1453 // Set up frame rate and related parameters rate control values.
1454 vp10_new_framerate(cpi, cpi->framerate);
1456 // Set absolute upper and lower quality limits
1457 rc->worst_quality = cpi->oxcf.worst_allowed_q;
1458 rc->best_quality = cpi->oxcf.best_allowed_q;
1460 cm->interp_filter = cpi->sf.default_interp_filter;
1462 if (cpi->oxcf.render_width > 0 && cpi->oxcf.render_height > 0) {
1463 cm->render_width = cpi->oxcf.render_width;
1464 cm->render_height = cpi->oxcf.render_height;
1466 cm->render_width = cpi->oxcf.width;
1467 cm->render_height = cpi->oxcf.height;
1469 cm->width = cpi->oxcf.width;
1470 cm->height = cpi->oxcf.height;
1472 if (cpi->initial_width) {
1473 if (cm->width > cpi->initial_width || cm->height > cpi->initial_height) {
1474 vp10_free_context_buffers(cm);
1475 vp10_alloc_compressor_data(cpi);
1476 realloc_segmentation_maps(cpi);
1477 cpi->initial_width = cpi->initial_height = 0;
1480 update_frame_size(cpi);
1482 cpi->alt_ref_source = NULL;
1483 rc->is_src_frame_alt_ref = 0;
1486 // Experimental RD Code
1487 cpi->frame_distortion = 0;
1488 cpi->last_frame_distortion = 0;
1491 set_tile_limits(cpi);
1493 cpi->ext_refresh_frame_flags_pending = 0;
1494 cpi->ext_refresh_frame_context_pending = 0;
1496 #if CONFIG_VP9_HIGHBITDEPTH
1497 highbd_set_var_fns(cpi);
1502 #define M_LOG2_E 0.693147180559945309417
1504 #define log2f(x) (log (x) / (float) M_LOG2_E)
1506 static void cal_nmvjointsadcost(int *mvjointsadcost) {
1507 mvjointsadcost[0] = 600;
1508 mvjointsadcost[1] = 300;
1509 mvjointsadcost[2] = 300;
1510 mvjointsadcost[3] = 300;
1513 static void cal_nmvsadcosts(int *mvsadcost[2]) {
1516 mvsadcost[0][0] = 0;
1517 mvsadcost[1][0] = 0;
1520 double z = 256 * (2 * (log2f(8 * i) + .6));
1521 mvsadcost[0][i] = (int)z;
1522 mvsadcost[1][i] = (int)z;
1523 mvsadcost[0][-i] = (int)z;
1524 mvsadcost[1][-i] = (int)z;
1525 } while (++i <= MV_MAX);
1528 static void cal_nmvsadcosts_hp(int *mvsadcost[2]) {
1531 mvsadcost[0][0] = 0;
1532 mvsadcost[1][0] = 0;
1535 double z = 256 * (2 * (log2f(8 * i) + .6));
1536 mvsadcost[0][i] = (int)z;
1537 mvsadcost[1][i] = (int)z;
1538 mvsadcost[0][-i] = (int)z;
1539 mvsadcost[1][-i] = (int)z;
1540 } while (++i <= MV_MAX);
1544 VP10_COMP *vp10_create_compressor(VP10EncoderConfig *oxcf,
1545 BufferPool *const pool) {
1547 VP10_COMP *volatile const cpi = vpx_memalign(32, sizeof(VP10_COMP));
1548 VP10_COMMON *volatile const cm = cpi != NULL ? &cpi->common : NULL;
1555 if (setjmp(cm->error.jmp)) {
1556 cm->error.setjmp = 0;
1557 vp10_remove_compressor(cpi);
1561 cm->error.setjmp = 1;
1562 cm->alloc_mi = vp10_enc_alloc_mi;
1563 cm->free_mi = vp10_enc_free_mi;
1564 cm->setup_mi = vp10_enc_setup_mi;
1566 CHECK_MEM_ERROR(cm, cm->fc,
1567 (FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
1568 CHECK_MEM_ERROR(cm, cm->frame_contexts,
1569 (FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
1570 sizeof(*cm->frame_contexts)));
1572 cpi->resize_state = 0;
1573 cpi->resize_avg_qp = 0;
1574 cpi->resize_buffer_underflow = 0;
1575 cpi->common.buffer_pool = pool;
1577 init_config(cpi, oxcf);
1578 vp10_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc);
1580 cm->current_video_frame = 0;
1581 cpi->partition_search_skippable_frame = 0;
1582 cpi->tile_data = NULL;
1584 realloc_segmentation_maps(cpi);
1586 CHECK_MEM_ERROR(cm, cpi->nmvcosts[0],
1587 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[0])));
1588 CHECK_MEM_ERROR(cm, cpi->nmvcosts[1],
1589 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[1])));
1590 CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[0],
1591 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[0])));
1592 CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[1],
1593 vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[1])));
1594 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[0],
1595 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[0])));
1596 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[1],
1597 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[1])));
1598 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[0],
1599 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[0])));
1600 CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[1],
1601 vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[1])));
1603 for (i = 0; i < (sizeof(cpi->mbgraph_stats) /
1604 sizeof(cpi->mbgraph_stats[0])); i++) {
1605 CHECK_MEM_ERROR(cm, cpi->mbgraph_stats[i].mb_stats,
1606 vpx_calloc(cm->MBs *
1607 sizeof(*cpi->mbgraph_stats[i].mb_stats), 1));
1610 #if CONFIG_FP_MB_STATS
1611 cpi->use_fp_mb_stats = 0;
1612 if (cpi->use_fp_mb_stats) {
1613 // a place holder used to store the first pass mb stats in the first pass
1614 CHECK_MEM_ERROR(cm, cpi->twopass.frame_mb_stats_buf,
1615 vpx_calloc(cm->MBs * sizeof(uint8_t), 1));
1617 cpi->twopass.frame_mb_stats_buf = NULL;
1621 cpi->refresh_alt_ref_frame = 0;
1622 cpi->multi_arf_last_grp_enabled = 0;
1624 cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
1625 #if CONFIG_INTERNAL_STATS
1626 cpi->b_calculate_ssimg = 0;
1627 cpi->b_calculate_blockiness = 1;
1628 cpi->b_calculate_consistency = 1;
1629 cpi->total_inconsistency = 0;
1630 cpi->psnr.worst = 100.0;
1631 cpi->worst_ssim = 100.0;
1636 if (cpi->b_calculate_psnr) {
1637 cpi->total_sq_error = 0;
1638 cpi->total_samples = 0;
1640 cpi->totalp_sq_error = 0;
1641 cpi->totalp_samples = 0;
1643 cpi->tot_recode_hits = 0;
1644 cpi->summed_quality = 0;
1645 cpi->summed_weights = 0;
1646 cpi->summedp_quality = 0;
1647 cpi->summedp_weights = 0;
1650 if (cpi->b_calculate_ssimg) {
1651 cpi->ssimg.worst= 100.0;
1653 cpi->fastssim.worst = 100.0;
1655 cpi->psnrhvs.worst = 100.0;
1657 if (cpi->b_calculate_blockiness) {
1658 cpi->total_blockiness = 0;
1659 cpi->worst_blockiness = 0.0;
1662 if (cpi->b_calculate_consistency) {
1663 cpi->ssim_vars = vpx_malloc(sizeof(*cpi->ssim_vars) *
1664 4 * cpi->common.mi_rows * cpi->common.mi_cols);
1665 cpi->worst_consistency = 100.0;
1670 cpi->first_time_stamp_ever = INT64_MAX;
1672 cal_nmvjointsadcost(cpi->td.mb.nmvjointsadcost);
1673 cpi->td.mb.nmvcost[0] = &cpi->nmvcosts[0][MV_MAX];
1674 cpi->td.mb.nmvcost[1] = &cpi->nmvcosts[1][MV_MAX];
1675 cpi->td.mb.nmvsadcost[0] = &cpi->nmvsadcosts[0][MV_MAX];
1676 cpi->td.mb.nmvsadcost[1] = &cpi->nmvsadcosts[1][MV_MAX];
1677 cal_nmvsadcosts(cpi->td.mb.nmvsadcost);
1679 cpi->td.mb.nmvcost_hp[0] = &cpi->nmvcosts_hp[0][MV_MAX];
1680 cpi->td.mb.nmvcost_hp[1] = &cpi->nmvcosts_hp[1][MV_MAX];
1681 cpi->td.mb.nmvsadcost_hp[0] = &cpi->nmvsadcosts_hp[0][MV_MAX];
1682 cpi->td.mb.nmvsadcost_hp[1] = &cpi->nmvsadcosts_hp[1][MV_MAX];
1683 cal_nmvsadcosts_hp(cpi->td.mb.nmvsadcost_hp);
1685 #if CONFIG_VP9_TEMPORAL_DENOISING
1686 #ifdef OUTPUT_YUV_DENOISED
1687 yuv_denoised_file = fopen("denoised.yuv", "ab");
1690 #ifdef OUTPUT_YUV_SKINMAP
1691 yuv_skinmap_file = fopen("skinmap.yuv", "ab");
1693 #ifdef OUTPUT_YUV_REC
1694 yuv_rec_file = fopen("rec.yuv", "wb");
1698 framepsnr = fopen("framepsnr.stt", "a");
1699 kf_list = fopen("kf_list.stt", "w");
1702 cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
1704 if (oxcf->pass == 1) {
1705 vp10_init_first_pass(cpi);
1706 } else if (oxcf->pass == 2) {
1707 const size_t packet_sz = sizeof(FIRSTPASS_STATS);
1708 const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz);
1710 #if CONFIG_FP_MB_STATS
1711 if (cpi->use_fp_mb_stats) {
1712 const size_t psz = cpi->common.MBs * sizeof(uint8_t);
1713 const int ps = (int)(oxcf->firstpass_mb_stats_in.sz / psz);
1715 cpi->twopass.firstpass_mb_stats.mb_stats_start =
1716 oxcf->firstpass_mb_stats_in.buf;
1717 cpi->twopass.firstpass_mb_stats.mb_stats_end =
1718 cpi->twopass.firstpass_mb_stats.mb_stats_start +
1719 (ps - 1) * cpi->common.MBs * sizeof(uint8_t);
1723 cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf;
1724 cpi->twopass.stats_in = cpi->twopass.stats_in_start;
1725 cpi->twopass.stats_in_end = &cpi->twopass.stats_in[packets - 1];
1727 vp10_init_second_pass(cpi);
1730 vp10_set_speed_features_framesize_independent(cpi);
1731 vp10_set_speed_features_framesize_dependent(cpi);
1733 // Allocate memory to store variances for a frame.
1734 CHECK_MEM_ERROR(cm, cpi->source_diff_var,
1735 vpx_calloc(cm->MBs, sizeof(diff)));
1736 cpi->source_var_thresh = 0;
1737 cpi->frames_till_next_var_check = 0;
1739 #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF)\
1740 cpi->fn_ptr[BT].sdf = SDF; \
1741 cpi->fn_ptr[BT].sdaf = SDAF; \
1742 cpi->fn_ptr[BT].vf = VF; \
1743 cpi->fn_ptr[BT].svf = SVF; \
1744 cpi->fn_ptr[BT].svaf = SVAF; \
1745 cpi->fn_ptr[BT].sdx3f = SDX3F; \
1746 cpi->fn_ptr[BT].sdx8f = SDX8F; \
1747 cpi->fn_ptr[BT].sdx4df = SDX4DF;
1749 BFP(BLOCK_32X16, vpx_sad32x16, vpx_sad32x16_avg,
1750 vpx_variance32x16, vpx_sub_pixel_variance32x16,
1751 vpx_sub_pixel_avg_variance32x16, NULL, NULL, vpx_sad32x16x4d)
1753 BFP(BLOCK_16X32, vpx_sad16x32, vpx_sad16x32_avg,
1754 vpx_variance16x32, vpx_sub_pixel_variance16x32,
1755 vpx_sub_pixel_avg_variance16x32, NULL, NULL, vpx_sad16x32x4d)
1757 BFP(BLOCK_64X32, vpx_sad64x32, vpx_sad64x32_avg,
1758 vpx_variance64x32, vpx_sub_pixel_variance64x32,
1759 vpx_sub_pixel_avg_variance64x32, NULL, NULL, vpx_sad64x32x4d)
1761 BFP(BLOCK_32X64, vpx_sad32x64, vpx_sad32x64_avg,
1762 vpx_variance32x64, vpx_sub_pixel_variance32x64,
1763 vpx_sub_pixel_avg_variance32x64, NULL, NULL, vpx_sad32x64x4d)
1765 BFP(BLOCK_32X32, vpx_sad32x32, vpx_sad32x32_avg,
1766 vpx_variance32x32, vpx_sub_pixel_variance32x32,
1767 vpx_sub_pixel_avg_variance32x32, vpx_sad32x32x3, vpx_sad32x32x8,
1770 BFP(BLOCK_64X64, vpx_sad64x64, vpx_sad64x64_avg,
1771 vpx_variance64x64, vpx_sub_pixel_variance64x64,
1772 vpx_sub_pixel_avg_variance64x64, vpx_sad64x64x3, vpx_sad64x64x8,
1775 BFP(BLOCK_16X16, vpx_sad16x16, vpx_sad16x16_avg,
1776 vpx_variance16x16, vpx_sub_pixel_variance16x16,
1777 vpx_sub_pixel_avg_variance16x16, vpx_sad16x16x3, vpx_sad16x16x8,
1780 BFP(BLOCK_16X8, vpx_sad16x8, vpx_sad16x8_avg,
1781 vpx_variance16x8, vpx_sub_pixel_variance16x8,
1782 vpx_sub_pixel_avg_variance16x8,
1783 vpx_sad16x8x3, vpx_sad16x8x8, vpx_sad16x8x4d)
1785 BFP(BLOCK_8X16, vpx_sad8x16, vpx_sad8x16_avg,
1786 vpx_variance8x16, vpx_sub_pixel_variance8x16,
1787 vpx_sub_pixel_avg_variance8x16,
1788 vpx_sad8x16x3, vpx_sad8x16x8, vpx_sad8x16x4d)
1790 BFP(BLOCK_8X8, vpx_sad8x8, vpx_sad8x8_avg,
1791 vpx_variance8x8, vpx_sub_pixel_variance8x8,
1792 vpx_sub_pixel_avg_variance8x8,
1793 vpx_sad8x8x3, vpx_sad8x8x8, vpx_sad8x8x4d)
1795 BFP(BLOCK_8X4, vpx_sad8x4, vpx_sad8x4_avg,
1796 vpx_variance8x4, vpx_sub_pixel_variance8x4,
1797 vpx_sub_pixel_avg_variance8x4, NULL, vpx_sad8x4x8, vpx_sad8x4x4d)
1799 BFP(BLOCK_4X8, vpx_sad4x8, vpx_sad4x8_avg,
1800 vpx_variance4x8, vpx_sub_pixel_variance4x8,
1801 vpx_sub_pixel_avg_variance4x8, NULL, vpx_sad4x8x8, vpx_sad4x8x4d)
1803 BFP(BLOCK_4X4, vpx_sad4x4, vpx_sad4x4_avg,
1804 vpx_variance4x4, vpx_sub_pixel_variance4x4,
1805 vpx_sub_pixel_avg_variance4x4,
1806 vpx_sad4x4x3, vpx_sad4x4x8, vpx_sad4x4x4d)
1808 #if CONFIG_VP9_HIGHBITDEPTH
1809 highbd_set_var_fns(cpi);
1812 /* vp10_init_quantizer() is first called here. Add check in
1813 * vp10_frame_init_quantizer() so that vp10_init_quantizer is only
1814 * called later when needed. This will avoid unnecessary calls of
1815 * vp10_init_quantizer() for every frame.
1817 vp10_init_quantizer(cpi);
1819 vp10_loop_filter_init(cm);
1821 cm->error.setjmp = 0;
1825 #define SNPRINT(H, T) \
1826 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T))
1828 #define SNPRINT2(H, T, V) \
1829 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V))
1831 void vp10_remove_compressor(VP10_COMP *cpi) {
1840 if (cm->current_video_frame > 0) {
1841 #if CONFIG_INTERNAL_STATS
1842 vpx_clear_system_state();
1844 if (cpi->oxcf.pass != 1) {
1845 char headings[512] = {0};
1846 char results[512] = {0};
1847 FILE *f = fopen("opsnr.stt", "a");
1848 double time_encoded = (cpi->last_end_time_stamp_seen
1849 - cpi->first_time_stamp_ever) / 10000000.000;
1850 double total_encode_time = (cpi->time_receive_data +
1851 cpi->time_compress_data) / 1000.000;
1853 (double)cpi->bytes * (double) 8 / (double)1000 / time_encoded;
1854 const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1);
1856 if (cpi->b_calculate_psnr) {
1857 const double total_psnr =
1858 vpx_sse_to_psnr((double)cpi->total_samples, peak,
1859 (double)cpi->total_sq_error);
1860 const double totalp_psnr =
1861 vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
1862 (double)cpi->totalp_sq_error);
1863 const double total_ssim = 100 * pow(cpi->summed_quality /
1864 cpi->summed_weights, 8.0);
1865 const double totalp_ssim = 100 * pow(cpi->summedp_quality /
1866 cpi->summedp_weights, 8.0);
1868 snprintf(headings, sizeof(headings),
1869 "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t"
1870 "VPXSSIM\tVPSSIMP\tFASTSIM\tPSNRHVS\t"
1871 "WstPsnr\tWstSsim\tWstFast\tWstHVS");
1872 snprintf(results, sizeof(results),
1873 "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1874 "%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
1875 "%7.3f\t%7.3f\t%7.3f\t%7.3f",
1876 dr, cpi->psnr.stat[ALL] / cpi->count, total_psnr,
1877 cpi->psnrp.stat[ALL] / cpi->count, totalp_psnr,
1878 total_ssim, totalp_ssim,
1879 cpi->fastssim.stat[ALL] / cpi->count,
1880 cpi->psnrhvs.stat[ALL] / cpi->count,
1881 cpi->psnr.worst, cpi->worst_ssim, cpi->fastssim.worst,
1882 cpi->psnrhvs.worst);
1884 if (cpi->b_calculate_blockiness) {
1885 SNPRINT(headings, "\t Block\tWstBlck");
1886 SNPRINT2(results, "\t%7.3f", cpi->total_blockiness / cpi->count);
1887 SNPRINT2(results, "\t%7.3f", cpi->worst_blockiness);
1890 if (cpi->b_calculate_consistency) {
1891 double consistency =
1892 vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
1893 (double)cpi->total_inconsistency);
1895 SNPRINT(headings, "\tConsist\tWstCons");
1896 SNPRINT2(results, "\t%7.3f", consistency);
1897 SNPRINT2(results, "\t%7.3f", cpi->worst_consistency);
1900 if (cpi->b_calculate_ssimg) {
1901 SNPRINT(headings, "\t SSIMG\tWtSSIMG");
1902 SNPRINT2(results, "\t%7.3f", cpi->ssimg.stat[ALL] / cpi->count);
1903 SNPRINT2(results, "\t%7.3f", cpi->ssimg.worst);
1906 fprintf(f, "%s\t Time\n", headings);
1907 fprintf(f, "%s\t%8.0f\n", results, total_encode_time);
1917 printf("\n_pick_loop_filter_level:%d\n", cpi->time_pick_lpf / 1000);
1918 printf("\n_frames recive_data encod_mb_row compress_frame Total\n");
1919 printf("%6d %10ld %10ld %10ld %10ld\n", cpi->common.current_video_frame,
1920 cpi->time_receive_data / 1000, cpi->time_encode_sb_row / 1000,
1921 cpi->time_compress_data / 1000,
1922 (cpi->time_receive_data + cpi->time_compress_data) / 1000);
1927 #if CONFIG_VP9_TEMPORAL_DENOISING
1928 vp10_denoiser_free(&(cpi->denoiser));
1931 for (t = 0; t < cpi->num_workers; ++t) {
1932 VPxWorker *const worker = &cpi->workers[t];
1933 EncWorkerData *const thread_data = &cpi->tile_thr_data[t];
1935 // Deallocate allocated threads.
1936 vpx_get_worker_interface()->end(worker);
1938 // Deallocate allocated thread data.
1939 if (t < cpi->num_workers - 1) {
1940 vpx_free(thread_data->td->counts);
1941 vp10_free_pc_tree(thread_data->td);
1942 vpx_free(thread_data->td);
1945 vpx_free(cpi->tile_thr_data);
1946 vpx_free(cpi->workers);
1948 if (cpi->num_workers > 1)
1949 vp10_loop_filter_dealloc(&cpi->lf_row_sync);
1951 dealloc_compressor_data(cpi);
1953 for (i = 0; i < sizeof(cpi->mbgraph_stats) /
1954 sizeof(cpi->mbgraph_stats[0]); ++i) {
1955 vpx_free(cpi->mbgraph_stats[i].mb_stats);
1958 #if CONFIG_FP_MB_STATS
1959 if (cpi->use_fp_mb_stats) {
1960 vpx_free(cpi->twopass.frame_mb_stats_buf);
1961 cpi->twopass.frame_mb_stats_buf = NULL;
1965 vp10_remove_common(cm);
1966 vp10_free_ref_frame_buffers(cm->buffer_pool);
1967 #if CONFIG_VP9_POSTPROC
1968 vp10_free_postproc_buffers(cm);
1972 #if CONFIG_VP9_TEMPORAL_DENOISING
1973 #ifdef OUTPUT_YUV_DENOISED
1974 fclose(yuv_denoised_file);
1977 #ifdef OUTPUT_YUV_SKINMAP
1978 fclose(yuv_skinmap_file);
1980 #ifdef OUTPUT_YUV_REC
1981 fclose(yuv_rec_file);
1998 /* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
1999 * and highbd_8_variance(). It should not.
2001 static void encoder_variance(const uint8_t *a, int a_stride,
2002 const uint8_t *b, int b_stride,
2003 int w, int h, unsigned int *sse, int *sum) {
2009 for (i = 0; i < h; i++) {
2010 for (j = 0; j < w; j++) {
2011 const int diff = a[j] - b[j];
2013 *sse += diff * diff;
2021 #if CONFIG_VP9_HIGHBITDEPTH
2022 static void encoder_highbd_variance64(const uint8_t *a8, int a_stride,
2023 const uint8_t *b8, int b_stride,
2024 int w, int h, uint64_t *sse,
2028 uint16_t *a = CONVERT_TO_SHORTPTR(a8);
2029 uint16_t *b = CONVERT_TO_SHORTPTR(b8);
2033 for (i = 0; i < h; i++) {
2034 for (j = 0; j < w; j++) {
2035 const int diff = a[j] - b[j];
2037 *sse += diff * diff;
2044 static void encoder_highbd_8_variance(const uint8_t *a8, int a_stride,
2045 const uint8_t *b8, int b_stride,
2047 unsigned int *sse, int *sum) {
2048 uint64_t sse_long = 0;
2049 uint64_t sum_long = 0;
2050 encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
2051 &sse_long, &sum_long);
2052 *sse = (unsigned int)sse_long;
2053 *sum = (int)sum_long;
2055 #endif // CONFIG_VP9_HIGHBITDEPTH
2057 static int64_t get_sse(const uint8_t *a, int a_stride,
2058 const uint8_t *b, int b_stride,
2059 int width, int height) {
2060 const int dw = width % 16;
2061 const int dh = height % 16;
2062 int64_t total_sse = 0;
2063 unsigned int sse = 0;
2068 encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
2069 dw, height, &sse, &sum);
2074 encoder_variance(&a[(height - dh) * a_stride], a_stride,
2075 &b[(height - dh) * b_stride], b_stride,
2076 width - dw, dh, &sse, &sum);
2080 for (y = 0; y < height / 16; ++y) {
2081 const uint8_t *pa = a;
2082 const uint8_t *pb = b;
2083 for (x = 0; x < width / 16; ++x) {
2084 vpx_mse16x16(pa, a_stride, pb, b_stride, &sse);
2098 #if CONFIG_VP9_HIGHBITDEPTH
2099 static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
2100 const uint8_t *b8, int b_stride,
2101 int width, int height,
2102 unsigned int input_shift) {
2103 const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
2104 const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
2105 int64_t total_sse = 0;
2107 for (y = 0; y < height; ++y) {
2108 for (x = 0; x < width; ++x) {
2110 diff = (a[x] >> input_shift) - (b[x] >> input_shift);
2111 total_sse += diff * diff;
2119 static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
2120 const uint8_t *b, int b_stride,
2121 int width, int height) {
2122 int64_t total_sse = 0;
2124 const int dw = width % 16;
2125 const int dh = height % 16;
2126 unsigned int sse = 0;
2129 encoder_highbd_8_variance(&a[width - dw], a_stride,
2130 &b[width - dw], b_stride,
2131 dw, height, &sse, &sum);
2135 encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
2136 &b[(height - dh) * b_stride], b_stride,
2137 width - dw, dh, &sse, &sum);
2140 for (y = 0; y < height / 16; ++y) {
2141 const uint8_t *pa = a;
2142 const uint8_t *pb = b;
2143 for (x = 0; x < width / 16; ++x) {
2144 vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
2154 #endif // CONFIG_VP9_HIGHBITDEPTH
2157 double psnr[4]; // total/y/u/v
2158 uint64_t sse[4]; // total/y/u/v
2159 uint32_t samples[4]; // total/y/u/v
2162 #if CONFIG_VP9_HIGHBITDEPTH
2163 static void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
2164 const YV12_BUFFER_CONFIG *b,
2166 unsigned int bit_depth,
2167 unsigned int in_bit_depth) {
2168 const int widths[3] =
2169 {a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
2170 const int heights[3] =
2171 {a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
2172 const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer };
2173 const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
2174 const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer };
2175 const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
2177 uint64_t total_sse = 0;
2178 uint32_t total_samples = 0;
2179 const double peak = (double)((1 << in_bit_depth) - 1);
2180 const unsigned int input_shift = bit_depth - in_bit_depth;
2182 for (i = 0; i < 3; ++i) {
2183 const int w = widths[i];
2184 const int h = heights[i];
2185 const uint32_t samples = w * h;
2187 if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
2189 sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
2190 b_planes[i], b_strides[i], w, h,
2193 sse = highbd_get_sse(a_planes[i], a_strides[i],
2194 b_planes[i], b_strides[i], w, h);
2197 sse = get_sse(a_planes[i], a_strides[i],
2198 b_planes[i], b_strides[i],
2201 psnr->sse[1 + i] = sse;
2202 psnr->samples[1 + i] = samples;
2203 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
2206 total_samples += samples;
2209 psnr->sse[0] = total_sse;
2210 psnr->samples[0] = total_samples;
2211 psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
2215 #else // !CONFIG_VP9_HIGHBITDEPTH
2217 static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
2219 static const double peak = 255.0;
2220 const int widths[3] = {
2221 a->y_crop_width, a->uv_crop_width, a->uv_crop_width};
2222 const int heights[3] = {
2223 a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
2224 const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer};
2225 const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
2226 const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer};
2227 const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
2229 uint64_t total_sse = 0;
2230 uint32_t total_samples = 0;
2232 for (i = 0; i < 3; ++i) {
2233 const int w = widths[i];
2234 const int h = heights[i];
2235 const uint32_t samples = w * h;
2236 const uint64_t sse = get_sse(a_planes[i], a_strides[i],
2237 b_planes[i], b_strides[i],
2239 psnr->sse[1 + i] = sse;
2240 psnr->samples[1 + i] = samples;
2241 psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
2244 total_samples += samples;
2247 psnr->sse[0] = total_sse;
2248 psnr->samples[0] = total_samples;
2249 psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
2252 #endif // CONFIG_VP9_HIGHBITDEPTH
2254 static void generate_psnr_packet(VP10_COMP *cpi) {
2255 struct vpx_codec_cx_pkt pkt;
2258 #if CONFIG_VP9_HIGHBITDEPTH
2259 calc_highbd_psnr(cpi->Source, cpi->common.frame_to_show, &psnr,
2260 cpi->td.mb.e_mbd.bd, cpi->oxcf.input_bit_depth);
2262 calc_psnr(cpi->Source, cpi->common.frame_to_show, &psnr);
2265 for (i = 0; i < 4; ++i) {
2266 pkt.data.psnr.samples[i] = psnr.samples[i];
2267 pkt.data.psnr.sse[i] = psnr.sse[i];
2268 pkt.data.psnr.psnr[i] = psnr.psnr[i];
2270 pkt.kind = VPX_CODEC_PSNR_PKT;
2271 vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
2274 int vp10_use_as_reference(VP10_COMP *cpi, int ref_frame_flags) {
2275 if (ref_frame_flags > 7)
2278 cpi->ref_frame_flags = ref_frame_flags;
2282 void vp10_update_reference(VP10_COMP *cpi, int ref_frame_flags) {
2283 cpi->ext_refresh_golden_frame = (ref_frame_flags & VP9_GOLD_FLAG) != 0;
2284 cpi->ext_refresh_alt_ref_frame = (ref_frame_flags & VP9_ALT_FLAG) != 0;
2285 cpi->ext_refresh_last_frame = (ref_frame_flags & VP9_LAST_FLAG) != 0;
2286 cpi->ext_refresh_frame_flags_pending = 1;
2289 static YV12_BUFFER_CONFIG *get_vp10_ref_frame_buffer(VP10_COMP *cpi,
2290 VP9_REFFRAME ref_frame_flag) {
2291 MV_REFERENCE_FRAME ref_frame = NONE;
2292 if (ref_frame_flag == VP9_LAST_FLAG)
2293 ref_frame = LAST_FRAME;
2294 else if (ref_frame_flag == VP9_GOLD_FLAG)
2295 ref_frame = GOLDEN_FRAME;
2296 else if (ref_frame_flag == VP9_ALT_FLAG)
2297 ref_frame = ALTREF_FRAME;
2299 return ref_frame == NONE ? NULL : get_ref_frame_buffer(cpi, ref_frame);
2302 int vp10_copy_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
2303 YV12_BUFFER_CONFIG *sd) {
2304 YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
2306 vp8_yv12_copy_frame(cfg, sd);
2313 int vp10_set_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
2314 YV12_BUFFER_CONFIG *sd) {
2315 YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
2317 vp8_yv12_copy_frame(sd, cfg);
2324 int vp10_update_entropy(VP10_COMP * cpi, int update) {
2325 cpi->ext_refresh_frame_context = update;
2326 cpi->ext_refresh_frame_context_pending = 1;
2330 #if defined(OUTPUT_YUV_DENOISED) || defined(OUTPUT_YUV_SKINMAP)
2331 // The denoiser buffer is allocated as a YUV 440 buffer. This function writes it
2332 // as YUV 420. We simply use the top-left pixels of the UV buffers, since we do
2333 // not denoise the UV channels at this time. If ever we implement UV channel
2334 // denoising we will have to modify this.
2335 void vp10_write_yuv_frame_420(YV12_BUFFER_CONFIG *s, FILE *f) {
2336 uint8_t *src = s->y_buffer;
2337 int h = s->y_height;
2340 fwrite(src, s->y_width, 1, f);
2348 fwrite(src, s->uv_width, 1, f);
2349 src += s->uv_stride;
2356 fwrite(src, s->uv_width, 1, f);
2357 src += s->uv_stride;
2362 #ifdef OUTPUT_YUV_REC
2363 void vp10_write_yuv_rec_frame(VP10_COMMON *cm) {
2364 YV12_BUFFER_CONFIG *s = cm->frame_to_show;
2365 uint8_t *src = s->y_buffer;
2368 #if CONFIG_VP9_HIGHBITDEPTH
2369 if (s->flags & YV12_FLAG_HIGHBITDEPTH) {
2370 uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer);
2373 fwrite(src16, s->y_width, 2, yuv_rec_file);
2374 src16 += s->y_stride;
2377 src16 = CONVERT_TO_SHORTPTR(s->u_buffer);
2381 fwrite(src16, s->uv_width, 2, yuv_rec_file);
2382 src16 += s->uv_stride;
2385 src16 = CONVERT_TO_SHORTPTR(s->v_buffer);
2389 fwrite(src16, s->uv_width, 2, yuv_rec_file);
2390 src16 += s->uv_stride;
2393 fflush(yuv_rec_file);
2396 #endif // CONFIG_VP9_HIGHBITDEPTH
2399 fwrite(src, s->y_width, 1, yuv_rec_file);
2407 fwrite(src, s->uv_width, 1, yuv_rec_file);
2408 src += s->uv_stride;
2415 fwrite(src, s->uv_width, 1, yuv_rec_file);
2416 src += s->uv_stride;
2419 fflush(yuv_rec_file);
2423 #if CONFIG_VP9_HIGHBITDEPTH
2424 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
2425 YV12_BUFFER_CONFIG *dst,
2428 static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
2429 YV12_BUFFER_CONFIG *dst) {
2430 #endif // CONFIG_VP9_HIGHBITDEPTH
2431 // TODO(dkovalev): replace YV12_BUFFER_CONFIG with vpx_image_t
2433 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
2434 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
2435 const int src_widths[3] = {src->y_crop_width, src->uv_crop_width,
2436 src->uv_crop_width };
2437 const int src_heights[3] = {src->y_crop_height, src->uv_crop_height,
2438 src->uv_crop_height};
2439 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
2440 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
2441 const int dst_widths[3] = {dst->y_crop_width, dst->uv_crop_width,
2442 dst->uv_crop_width};
2443 const int dst_heights[3] = {dst->y_crop_height, dst->uv_crop_height,
2444 dst->uv_crop_height};
2446 for (i = 0; i < MAX_MB_PLANE; ++i) {
2447 #if CONFIG_VP9_HIGHBITDEPTH
2448 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
2449 vp10_highbd_resize_plane(srcs[i], src_heights[i], src_widths[i],
2450 src_strides[i], dsts[i], dst_heights[i],
2451 dst_widths[i], dst_strides[i], bd);
2453 vp10_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
2454 dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
2457 vp10_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
2458 dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
2459 #endif // CONFIG_VP9_HIGHBITDEPTH
2461 vpx_extend_frame_borders(dst);
2464 #if CONFIG_VP9_HIGHBITDEPTH
2465 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
2466 YV12_BUFFER_CONFIG *dst, int bd) {
2468 static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
2469 YV12_BUFFER_CONFIG *dst) {
2470 #endif // CONFIG_VP9_HIGHBITDEPTH
2471 const int src_w = src->y_crop_width;
2472 const int src_h = src->y_crop_height;
2473 const int dst_w = dst->y_crop_width;
2474 const int dst_h = dst->y_crop_height;
2475 const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
2476 const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
2477 uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
2478 const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
2479 const InterpKernel *const kernel = vp10_filter_kernels[EIGHTTAP];
2482 for (y = 0; y < dst_h; y += 16) {
2483 for (x = 0; x < dst_w; x += 16) {
2484 for (i = 0; i < MAX_MB_PLANE; ++i) {
2485 const int factor = (i == 0 || i == 3 ? 1 : 2);
2486 const int x_q4 = x * (16 / factor) * src_w / dst_w;
2487 const int y_q4 = y * (16 / factor) * src_h / dst_h;
2488 const int src_stride = src_strides[i];
2489 const int dst_stride = dst_strides[i];
2490 const uint8_t *src_ptr = srcs[i] + (y / factor) * src_h / dst_h *
2491 src_stride + (x / factor) * src_w / dst_w;
2492 uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
2494 #if CONFIG_VP9_HIGHBITDEPTH
2495 if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
2496 vpx_highbd_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2497 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2498 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2499 16 / factor, 16 / factor, bd);
2501 vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2502 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2503 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2504 16 / factor, 16 / factor);
2507 vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
2508 kernel[x_q4 & 0xf], 16 * src_w / dst_w,
2509 kernel[y_q4 & 0xf], 16 * src_h / dst_h,
2510 16 / factor, 16 / factor);
2511 #endif // CONFIG_VP9_HIGHBITDEPTH
2516 vpx_extend_frame_borders(dst);
2519 static int scale_down(VP10_COMP *cpi, int q) {
2520 RATE_CONTROL *const rc = &cpi->rc;
2521 GF_GROUP *const gf_group = &cpi->twopass.gf_group;
2523 assert(frame_is_kf_gf_arf(cpi));
2525 if (rc->frame_size_selector == UNSCALED &&
2526 q >= rc->rf_level_maxq[gf_group->rf_level[gf_group->index]]) {
2527 const int max_size_thresh = (int)(rate_thresh_mult[SCALE_STEP1]
2528 * VPXMAX(rc->this_frame_target, rc->avg_frame_bandwidth));
2529 scale = rc->projected_frame_size > max_size_thresh ? 1 : 0;
2534 // Function to test for conditions that indicate we should loop
2535 // back and recode a frame.
2536 static int recode_loop_test(VP10_COMP *cpi,
2537 int high_limit, int low_limit,
2538 int q, int maxq, int minq) {
2539 const RATE_CONTROL *const rc = &cpi->rc;
2540 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
2541 const int frame_is_kfgfarf = frame_is_kf_gf_arf(cpi);
2542 int force_recode = 0;
2544 if ((rc->projected_frame_size >= rc->max_frame_bandwidth) ||
2545 (cpi->sf.recode_loop == ALLOW_RECODE) ||
2546 (frame_is_kfgfarf &&
2547 (cpi->sf.recode_loop == ALLOW_RECODE_KFARFGF))) {
2548 if (frame_is_kfgfarf &&
2549 (oxcf->resize_mode == RESIZE_DYNAMIC) &&
2550 scale_down(cpi, q)) {
2551 // Code this group at a lower resolution.
2552 cpi->resize_pending = 1;
2556 // TODO(agrange) high_limit could be greater than the scale-down threshold.
2557 if ((rc->projected_frame_size > high_limit && q < maxq) ||
2558 (rc->projected_frame_size < low_limit && q > minq)) {
2560 } else if (cpi->oxcf.rc_mode == VPX_CQ) {
2561 // Deal with frame undershoot and whether or not we are
2562 // below the automatically set cq level.
2563 if (q > oxcf->cq_level &&
2564 rc->projected_frame_size < ((rc->this_frame_target * 7) >> 3)) {
2569 return force_recode;
2572 void vp10_update_reference_frames(VP10_COMP *cpi) {
2573 VP10_COMMON * const cm = &cpi->common;
2574 BufferPool *const pool = cm->buffer_pool;
2576 // At this point the new frame has been encoded.
2577 // If any buffer copy / swapping is signaled it should be done here.
2578 if (cm->frame_type == KEY_FRAME) {
2579 ref_cnt_fb(pool->frame_bufs,
2580 &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
2581 ref_cnt_fb(pool->frame_bufs,
2582 &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
2583 } else if (vp10_preserve_existing_gf(cpi)) {
2584 // We have decided to preserve the previously existing golden frame as our
2585 // new ARF frame. However, in the short term in function
2586 // vp10_bitstream.c::get_refresh_mask() we left it in the GF slot and, if
2587 // we're updating the GF with the current decoded frame, we save it to the
2588 // ARF slot instead.
2589 // We now have to update the ARF with the current frame and swap gld_fb_idx
2590 // and alt_fb_idx so that, overall, we've stored the old GF in the new ARF
2591 // slot and, if we're updating the GF, the current frame becomes the new GF.
2594 ref_cnt_fb(pool->frame_bufs,
2595 &cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
2597 tmp = cpi->alt_fb_idx;
2598 cpi->alt_fb_idx = cpi->gld_fb_idx;
2599 cpi->gld_fb_idx = tmp;
2600 } else { /* For non key/golden frames */
2601 if (cpi->refresh_alt_ref_frame) {
2602 int arf_idx = cpi->alt_fb_idx;
2603 if ((cpi->oxcf.pass == 2) && cpi->multi_arf_allowed) {
2604 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
2605 arf_idx = gf_group->arf_update_idx[gf_group->index];
2608 ref_cnt_fb(pool->frame_bufs,
2609 &cm->ref_frame_map[arf_idx], cm->new_fb_idx);
2610 memcpy(cpi->interp_filter_selected[ALTREF_FRAME],
2611 cpi->interp_filter_selected[0],
2612 sizeof(cpi->interp_filter_selected[0]));
2615 if (cpi->refresh_golden_frame) {
2616 ref_cnt_fb(pool->frame_bufs,
2617 &cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
2618 if (!cpi->rc.is_src_frame_alt_ref)
2619 memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
2620 cpi->interp_filter_selected[0],
2621 sizeof(cpi->interp_filter_selected[0]));
2623 memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
2624 cpi->interp_filter_selected[ALTREF_FRAME],
2625 sizeof(cpi->interp_filter_selected[ALTREF_FRAME]));
2629 if (cpi->refresh_last_frame) {
2630 ref_cnt_fb(pool->frame_bufs,
2631 &cm->ref_frame_map[cpi->lst_fb_idx], cm->new_fb_idx);
2632 if (!cpi->rc.is_src_frame_alt_ref)
2633 memcpy(cpi->interp_filter_selected[LAST_FRAME],
2634 cpi->interp_filter_selected[0],
2635 sizeof(cpi->interp_filter_selected[0]));
2637 #if CONFIG_VP9_TEMPORAL_DENOISING
2638 if (cpi->oxcf.noise_sensitivity > 0) {
2639 vp10_denoiser_update_frame_info(&cpi->denoiser,
2641 cpi->common.frame_type,
2642 cpi->refresh_alt_ref_frame,
2643 cpi->refresh_golden_frame,
2644 cpi->refresh_last_frame);
2649 static void loopfilter_frame(VP10_COMP *cpi, VP10_COMMON *cm) {
2650 MACROBLOCKD *xd = &cpi->td.mb.e_mbd;
2651 struct loopfilter *lf = &cm->lf;
2652 if (is_lossless_requested(&cpi->oxcf)) {
2653 lf->filter_level = 0;
2655 struct vpx_usec_timer timer;
2657 vpx_clear_system_state();
2659 vpx_usec_timer_start(&timer);
2661 vp10_pick_filter_level(cpi->Source, cpi, cpi->sf.lpf_pick);
2663 vpx_usec_timer_mark(&timer);
2664 cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
2667 if (lf->filter_level > 0) {
2668 if (cpi->num_workers > 1)
2669 vp10_loop_filter_frame_mt(cm->frame_to_show, cm, xd->plane,
2670 lf->filter_level, 0, 0,
2671 cpi->workers, cpi->num_workers,
2674 vp10_loop_filter_frame(cm->frame_to_show, cm, xd, lf->filter_level, 0, 0);
2677 vpx_extend_frame_inner_borders(cm->frame_to_show);
2680 static INLINE void alloc_frame_mvs(const VP10_COMMON *cm,
2682 RefCntBuffer *const new_fb_ptr = &cm->buffer_pool->frame_bufs[buffer_idx];
2683 if (new_fb_ptr->mvs == NULL ||
2684 new_fb_ptr->mi_rows < cm->mi_rows ||
2685 new_fb_ptr->mi_cols < cm->mi_cols) {
2686 vpx_free(new_fb_ptr->mvs);
2688 (MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols,
2689 sizeof(*new_fb_ptr->mvs));
2690 new_fb_ptr->mi_rows = cm->mi_rows;
2691 new_fb_ptr->mi_cols = cm->mi_cols;
2695 void vp10_scale_references(VP10_COMP *cpi) {
2696 VP10_COMMON *cm = &cpi->common;
2697 MV_REFERENCE_FRAME ref_frame;
2698 const VP9_REFFRAME ref_mask[3] = {VP9_LAST_FLAG, VP9_GOLD_FLAG, VP9_ALT_FLAG};
2700 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
2701 // Need to convert from VP9_REFFRAME to index into ref_mask (subtract 1).
2702 if (cpi->ref_frame_flags & ref_mask[ref_frame - 1]) {
2703 BufferPool *const pool = cm->buffer_pool;
2704 const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi,
2708 cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
2712 #if CONFIG_VP9_HIGHBITDEPTH
2713 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
2714 RefCntBuffer *new_fb_ptr = NULL;
2715 int force_scaling = 0;
2716 int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
2717 if (new_fb == INVALID_IDX) {
2718 new_fb = get_free_fb(cm);
2721 if (new_fb == INVALID_IDX)
2723 new_fb_ptr = &pool->frame_bufs[new_fb];
2724 if (force_scaling ||
2725 new_fb_ptr->buf.y_crop_width != cm->width ||
2726 new_fb_ptr->buf.y_crop_height != cm->height) {
2727 vpx_realloc_frame_buffer(&new_fb_ptr->buf,
2728 cm->width, cm->height,
2729 cm->subsampling_x, cm->subsampling_y,
2730 cm->use_highbitdepth,
2731 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
2733 scale_and_extend_frame(ref, &new_fb_ptr->buf, (int)cm->bit_depth);
2734 cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
2735 alloc_frame_mvs(cm, new_fb);
2738 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
2739 RefCntBuffer *new_fb_ptr = NULL;
2740 int force_scaling = 0;
2741 int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
2742 if (new_fb == INVALID_IDX) {
2743 new_fb = get_free_fb(cm);
2746 if (new_fb == INVALID_IDX)
2748 new_fb_ptr = &pool->frame_bufs[new_fb];
2749 if (force_scaling ||
2750 new_fb_ptr->buf.y_crop_width != cm->width ||
2751 new_fb_ptr->buf.y_crop_height != cm->height) {
2752 vpx_realloc_frame_buffer(&new_fb_ptr->buf,
2753 cm->width, cm->height,
2754 cm->subsampling_x, cm->subsampling_y,
2755 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
2757 scale_and_extend_frame(ref, &new_fb_ptr->buf);
2758 cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
2759 alloc_frame_mvs(cm, new_fb);
2761 #endif // CONFIG_VP9_HIGHBITDEPTH
2763 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
2764 RefCntBuffer *const buf = &pool->frame_bufs[buf_idx];
2765 buf->buf.y_crop_width = ref->y_crop_width;
2766 buf->buf.y_crop_height = ref->y_crop_height;
2767 cpi->scaled_ref_idx[ref_frame - 1] = buf_idx;
2771 if (cpi->oxcf.pass != 0)
2772 cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
2777 static void release_scaled_references(VP10_COMP *cpi) {
2778 VP10_COMMON *cm = &cpi->common;
2780 if (cpi->oxcf.pass == 0) {
2781 // Only release scaled references under certain conditions:
2782 // if reference will be updated, or if scaled reference has same resolution.
2784 refresh[0] = (cpi->refresh_last_frame) ? 1 : 0;
2785 refresh[1] = (cpi->refresh_golden_frame) ? 1 : 0;
2786 refresh[2] = (cpi->refresh_alt_ref_frame) ? 1 : 0;
2787 for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) {
2788 const int idx = cpi->scaled_ref_idx[i - 1];
2789 RefCntBuffer *const buf = idx != INVALID_IDX ?
2790 &cm->buffer_pool->frame_bufs[idx] : NULL;
2791 const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi, i);
2794 (buf->buf.y_crop_width == ref->y_crop_width &&
2795 buf->buf.y_crop_height == ref->y_crop_height))) {
2797 cpi->scaled_ref_idx[i -1] = INVALID_IDX;
2801 for (i = 0; i < MAX_REF_FRAMES; ++i) {
2802 const int idx = cpi->scaled_ref_idx[i];
2803 RefCntBuffer *const buf = idx != INVALID_IDX ?
2804 &cm->buffer_pool->frame_bufs[idx] : NULL;
2807 cpi->scaled_ref_idx[i] = INVALID_IDX;
2813 static void full_to_model_count(unsigned int *model_count,
2814 unsigned int *full_count) {
2816 model_count[ZERO_TOKEN] = full_count[ZERO_TOKEN];
2817 model_count[ONE_TOKEN] = full_count[ONE_TOKEN];
2818 model_count[TWO_TOKEN] = full_count[TWO_TOKEN];
2819 for (n = THREE_TOKEN; n < EOB_TOKEN; ++n)
2820 model_count[TWO_TOKEN] += full_count[n];
2821 model_count[EOB_MODEL_TOKEN] = full_count[EOB_TOKEN];
2824 static void full_to_model_counts(vp10_coeff_count_model *model_count,
2825 vp10_coeff_count *full_count) {
2828 for (i = 0; i < PLANE_TYPES; ++i)
2829 for (j = 0; j < REF_TYPES; ++j)
2830 for (k = 0; k < COEF_BANDS; ++k)
2831 for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l)
2832 full_to_model_count(model_count[i][j][k][l], full_count[i][j][k][l]);
2835 #if 0 && CONFIG_INTERNAL_STATS
2836 static void output_frame_level_debug_stats(VP10_COMP *cpi) {
2837 VP10_COMMON *const cm = &cpi->common;
2838 FILE *const f = fopen("tmp.stt", cm->current_video_frame ? "a" : "w");
2841 vpx_clear_system_state();
2843 recon_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
2845 if (cpi->twopass.total_left_stats.coded_error != 0.0)
2846 fprintf(f, "%10u %dx%d %d %d %10d %10d %10d %10d"
2847 "%10"PRId64" %10"PRId64" %5d %5d %10"PRId64" "
2848 "%10"PRId64" %10"PRId64" %10d "
2849 "%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
2850 "%6d %6d %5d %5d %5d "
2851 "%10"PRId64" %10.3lf"
2852 "%10lf %8u %10"PRId64" %10d %10d %10d\n",
2853 cpi->common.current_video_frame,
2854 cm->width, cm->height,
2855 cpi->rc.source_alt_ref_pending,
2856 cpi->rc.source_alt_ref_active,
2857 cpi->rc.this_frame_target,
2858 cpi->rc.projected_frame_size,
2859 cpi->rc.projected_frame_size / cpi->common.MBs,
2860 (cpi->rc.projected_frame_size - cpi->rc.this_frame_target),
2861 cpi->rc.vbr_bits_off_target,
2862 cpi->rc.vbr_bits_off_target_fast,
2863 cpi->twopass.extend_minq,
2864 cpi->twopass.extend_minq_fast,
2865 cpi->rc.total_target_vs_actual,
2866 (cpi->rc.starting_buffer_level - cpi->rc.bits_off_target),
2867 cpi->rc.total_actual_bits, cm->base_qindex,
2868 vp10_convert_qindex_to_q(cm->base_qindex, cm->bit_depth),
2869 (double)vp10_dc_quant(cm->base_qindex, 0, cm->bit_depth) / 4.0,
2870 vp10_convert_qindex_to_q(cpi->twopass.active_worst_quality,
2873 vp10_convert_qindex_to_q(cpi->oxcf.cq_level, cm->bit_depth),
2874 cpi->refresh_last_frame, cpi->refresh_golden_frame,
2875 cpi->refresh_alt_ref_frame, cm->frame_type, cpi->rc.gfu_boost,
2876 cpi->twopass.bits_left,
2877 cpi->twopass.total_left_stats.coded_error,
2878 cpi->twopass.bits_left /
2879 (1 + cpi->twopass.total_left_stats.coded_error),
2880 cpi->tot_recode_hits, recon_err, cpi->rc.kf_boost,
2881 cpi->twopass.kf_zeromotion_pct,
2882 cpi->twopass.fr_content_type);
2887 FILE *const fmodes = fopen("Modes.stt", "a");
2890 fprintf(fmodes, "%6d:%1d:%1d:%1d ", cpi->common.current_video_frame,
2891 cm->frame_type, cpi->refresh_golden_frame,
2892 cpi->refresh_alt_ref_frame);
2894 for (i = 0; i < MAX_MODES; ++i)
2895 fprintf(fmodes, "%5d ", cpi->mode_chosen_counts[i]);
2897 fprintf(fmodes, "\n");
2904 static void set_mv_search_params(VP10_COMP *cpi) {
2905 const VP10_COMMON *const cm = &cpi->common;
2906 const unsigned int max_mv_def = VPXMIN(cm->width, cm->height);
2908 // Default based on max resolution.
2909 cpi->mv_step_param = vp10_init_search_range(max_mv_def);
2911 if (cpi->sf.mv.auto_mv_step_size) {
2912 if (frame_is_intra_only(cm)) {
2913 // Initialize max_mv_magnitude for use in the first INTER frame
2914 // after a key/intra-only frame.
2915 cpi->max_mv_magnitude = max_mv_def;
2917 if (cm->show_frame) {
2918 // Allow mv_steps to correspond to twice the max mv magnitude found
2919 // in the previous frame, capped by the default max_mv_magnitude based
2921 cpi->mv_step_param = vp10_init_search_range(
2922 VPXMIN(max_mv_def, 2 * cpi->max_mv_magnitude));
2924 cpi->max_mv_magnitude = 0;
2929 static void set_size_independent_vars(VP10_COMP *cpi) {
2930 vp10_set_speed_features_framesize_independent(cpi);
2931 vp10_set_rd_speed_thresholds(cpi);
2932 vp10_set_rd_speed_thresholds_sub8x8(cpi);
2933 cpi->common.interp_filter = cpi->sf.default_interp_filter;
2936 static void set_size_dependent_vars(VP10_COMP *cpi, int *q,
2937 int *bottom_index, int *top_index) {
2938 VP10_COMMON *const cm = &cpi->common;
2939 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
2941 // Setup variables that depend on the dimensions of the frame.
2942 vp10_set_speed_features_framesize_dependent(cpi);
2944 // Decide q and q bounds.
2945 *q = vp10_rc_pick_q_and_bounds(cpi, bottom_index, top_index);
2947 if (!frame_is_intra_only(cm)) {
2948 vp10_set_high_precision_mv(cpi, (*q) < HIGH_PRECISION_MV_QTHRESH);
2951 // Configure experimental use of segmentation for enhanced coding of
2952 // static regions if indicated.
2953 // Only allowed in the second pass of a two pass encode, as it requires
2954 // lagged coding, and if the relevant speed feature flag is set.
2955 if (oxcf->pass == 2 && cpi->sf.static_segmentation)
2956 configure_static_seg_features(cpi);
2958 #if CONFIG_VP9_POSTPROC
2959 if (oxcf->noise_sensitivity > 0) {
2961 switch (oxcf->noise_sensitivity) {
2979 vp10_denoise(cpi->Source, cpi->Source, l);
2981 #endif // CONFIG_VP9_POSTPROC
2984 static void init_motion_estimation(VP10_COMP *cpi) {
2985 int y_stride = cpi->scaled_source.y_stride;
2987 if (cpi->sf.mv.search_method == NSTEP) {
2988 vp10_init3smotion_compensation(&cpi->ss_cfg, y_stride);
2989 } else if (cpi->sf.mv.search_method == DIAMOND) {
2990 vp10_init_dsmotion_compensation(&cpi->ss_cfg, y_stride);
2994 static void set_frame_size(VP10_COMP *cpi) {
2996 VP10_COMMON *const cm = &cpi->common;
2997 VP10EncoderConfig *const oxcf = &cpi->oxcf;
2998 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
3000 if (oxcf->pass == 2 &&
3001 oxcf->rc_mode == VPX_VBR &&
3002 ((oxcf->resize_mode == RESIZE_FIXED && cm->current_video_frame == 0) ||
3003 (oxcf->resize_mode == RESIZE_DYNAMIC && cpi->resize_pending))) {
3004 vp10_calculate_coded_size(
3005 cpi, &oxcf->scaled_frame_width, &oxcf->scaled_frame_height);
3007 // There has been a change in frame size.
3008 vp10_set_size_literal(cpi, oxcf->scaled_frame_width,
3009 oxcf->scaled_frame_height);
3012 if (oxcf->pass == 0 &&
3013 oxcf->rc_mode == VPX_CBR &&
3014 oxcf->resize_mode == RESIZE_DYNAMIC) {
3015 if (cpi->resize_pending == 1) {
3016 oxcf->scaled_frame_width =
3017 (cm->width * cpi->resize_scale_num) / cpi->resize_scale_den;
3018 oxcf->scaled_frame_height =
3019 (cm->height * cpi->resize_scale_num) /cpi->resize_scale_den;
3020 } else if (cpi->resize_pending == -1) {
3021 // Go back up to original size.
3022 oxcf->scaled_frame_width = oxcf->width;
3023 oxcf->scaled_frame_height = oxcf->height;
3025 if (cpi->resize_pending != 0) {
3026 // There has been a change in frame size.
3027 vp10_set_size_literal(cpi,
3028 oxcf->scaled_frame_width,
3029 oxcf->scaled_frame_height);
3031 // TODO(agrange) Scale cpi->max_mv_magnitude if frame-size has changed.
3032 set_mv_search_params(cpi);
3036 if (oxcf->pass == 2) {
3037 vp10_set_target_rate(cpi);
3040 alloc_frame_mvs(cm, cm->new_fb_idx);
3042 // Reset the frame pointers to the current frame size.
3043 vpx_realloc_frame_buffer(get_frame_new_buffer(cm),
3044 cm->width, cm->height,
3045 cm->subsampling_x, cm->subsampling_y,
3046 #if CONFIG_VP9_HIGHBITDEPTH
3047 cm->use_highbitdepth,
3049 VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
3052 alloc_util_frame_buffers(cpi);
3053 init_motion_estimation(cpi);
3055 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
3056 RefBuffer *const ref_buf = &cm->frame_refs[ref_frame - 1];
3057 const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
3059 ref_buf->idx = buf_idx;
3061 if (buf_idx != INVALID_IDX) {
3062 YV12_BUFFER_CONFIG *const buf = &cm->buffer_pool->frame_bufs[buf_idx].buf;
3064 #if CONFIG_VP9_HIGHBITDEPTH
3065 vp10_setup_scale_factors_for_frame(&ref_buf->sf,
3066 buf->y_crop_width, buf->y_crop_height,
3067 cm->width, cm->height,
3068 (buf->flags & YV12_FLAG_HIGHBITDEPTH) ?
3071 vp10_setup_scale_factors_for_frame(&ref_buf->sf,
3072 buf->y_crop_width, buf->y_crop_height,
3073 cm->width, cm->height);
3074 #endif // CONFIG_VP9_HIGHBITDEPTH
3075 if (vp10_is_scaled(&ref_buf->sf))
3076 vpx_extend_frame_borders(buf);
3078 ref_buf->buf = NULL;
3082 set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME);
3085 static void encode_without_recode_loop(VP10_COMP *cpi) {
3086 VP10_COMMON *const cm = &cpi->common;
3087 int q = 0, bottom_index = 0, top_index = 0; // Dummy variables.
3089 vpx_clear_system_state();
3091 set_frame_size(cpi);
3093 // For 1 pass CBR under dynamic resize mode: use faster scaling for source.
3094 // Only for 2x2 scaling for now.
3095 if (cpi->oxcf.pass == 0 &&
3096 cpi->oxcf.rc_mode == VPX_CBR &&
3097 cpi->oxcf.resize_mode == RESIZE_DYNAMIC &&
3098 cpi->un_scaled_source->y_width == (cm->width << 1) &&
3099 cpi->un_scaled_source->y_height == (cm->height << 1)) {
3100 cpi->Source = vp10_scale_if_required_fast(cm,
3101 cpi->un_scaled_source,
3102 &cpi->scaled_source);
3103 if (cpi->unscaled_last_source != NULL)
3104 cpi->Last_Source = vp10_scale_if_required_fast(cm,
3105 cpi->unscaled_last_source,
3106 &cpi->scaled_last_source);
3108 cpi->Source = vp10_scale_if_required(cm, cpi->un_scaled_source,
3109 &cpi->scaled_source);
3110 if (cpi->unscaled_last_source != NULL)
3111 cpi->Last_Source = vp10_scale_if_required(cm, cpi->unscaled_last_source,
3112 &cpi->scaled_last_source);
3115 if (frame_is_intra_only(cm) == 0) {
3116 vp10_scale_references(cpi);
3119 set_size_independent_vars(cpi);
3120 set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
3122 vp10_set_quantizer(cm, q);
3123 vp10_set_variance_partition_thresholds(cpi, q);
3127 suppress_active_map(cpi);
3128 // Variance adaptive and in frame q adjustment experiments are mutually
3130 if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
3131 vp10_vaq_frame_setup(cpi);
3132 } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
3133 vp10_setup_in_frame_q_adj(cpi);
3134 } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
3135 vp10_cyclic_refresh_setup(cpi);
3137 apply_active_map(cpi);
3139 // transform / motion compensation build reconstruction frame
3140 vp10_encode_frame(cpi);
3142 // Update some stats from cyclic refresh, and check if we should not update
3143 // golden reference, for 1 pass CBR.
3144 if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
3145 cm->frame_type != KEY_FRAME &&
3146 (cpi->oxcf.pass == 0 && cpi->oxcf.rc_mode == VPX_CBR))
3147 vp10_cyclic_refresh_check_golden_update(cpi);
3149 // Update the skip mb flag probabilities based on the distribution
3150 // seen in the last encoder iteration.
3151 // update_base_skip_probs(cpi);
3152 vpx_clear_system_state();
3155 static void encode_with_recode_loop(VP10_COMP *cpi,
3158 VP10_COMMON *const cm = &cpi->common;
3159 RATE_CONTROL *const rc = &cpi->rc;
3160 int bottom_index, top_index;
3162 int loop_at_this_size = 0;
3164 int overshoot_seen = 0;
3165 int undershoot_seen = 0;
3166 int frame_over_shoot_limit;
3167 int frame_under_shoot_limit;
3168 int q = 0, q_low = 0, q_high = 0;
3170 set_size_independent_vars(cpi);
3173 vpx_clear_system_state();
3175 set_frame_size(cpi);
3177 if (loop_count == 0 || cpi->resize_pending != 0) {
3178 set_size_dependent_vars(cpi, &q, &bottom_index, &top_index);
3180 // TODO(agrange) Scale cpi->max_mv_magnitude if frame-size has changed.
3181 set_mv_search_params(cpi);
3183 // Reset the loop state for new frame size.
3185 undershoot_seen = 0;
3187 // Reconfiguration for change in frame size has concluded.
3188 cpi->resize_pending = 0;
3190 q_low = bottom_index;
3193 loop_at_this_size = 0;
3196 // Decide frame size bounds first time through.
3197 if (loop_count == 0) {
3198 vp10_rc_compute_frame_size_bounds(cpi, rc->this_frame_target,
3199 &frame_under_shoot_limit,
3200 &frame_over_shoot_limit);
3203 cpi->Source = vp10_scale_if_required(cm, cpi->un_scaled_source,
3204 &cpi->scaled_source);
3206 if (cpi->unscaled_last_source != NULL)
3207 cpi->Last_Source = vp10_scale_if_required(cm, cpi->unscaled_last_source,
3208 &cpi->scaled_last_source);
3210 if (frame_is_intra_only(cm) == 0) {
3211 if (loop_count > 0) {
3212 release_scaled_references(cpi);
3214 vp10_scale_references(cpi);
3217 vp10_set_quantizer(cm, q);
3219 if (loop_count == 0)
3222 // Variance adaptive and in frame q adjustment experiments are mutually
3224 if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
3225 vp10_vaq_frame_setup(cpi);
3226 } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
3227 vp10_setup_in_frame_q_adj(cpi);
3230 // transform / motion compensation build reconstruction frame
3231 vp10_encode_frame(cpi);
3233 // Update the skip mb flag probabilities based on the distribution
3234 // seen in the last encoder iteration.
3235 // update_base_skip_probs(cpi);
3237 vpx_clear_system_state();
3239 // Dummy pack of the bitstream using up to date stats to get an
3240 // accurate estimate of output frame size to determine if we need
3242 if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) {
3243 save_coding_context(cpi);
3244 vp10_pack_bitstream(cpi, dest, size);
3246 rc->projected_frame_size = (int)(*size) << 3;
3247 restore_coding_context(cpi);
3249 if (frame_over_shoot_limit == 0)
3250 frame_over_shoot_limit = 1;
3253 if (cpi->oxcf.rc_mode == VPX_Q) {
3256 if ((cm->frame_type == KEY_FRAME) &&
3257 rc->this_key_frame_forced &&
3258 (rc->projected_frame_size < rc->max_frame_bandwidth)) {
3262 int64_t high_err_target = cpi->ambient_err;
3263 int64_t low_err_target = cpi->ambient_err >> 1;
3265 #if CONFIG_VP9_HIGHBITDEPTH
3266 if (cm->use_highbitdepth) {
3267 kf_err = vp10_highbd_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3269 kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3272 kf_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3273 #endif // CONFIG_VP9_HIGHBITDEPTH
3275 // Prevent possible divide by zero error below for perfect KF
3278 // The key frame is not good enough or we can afford
3279 // to make it better without undue risk of popping.
3280 if ((kf_err > high_err_target &&
3281 rc->projected_frame_size <= frame_over_shoot_limit) ||
3282 (kf_err > low_err_target &&
3283 rc->projected_frame_size <= frame_under_shoot_limit)) {
3285 q_high = q > q_low ? q - 1 : q_low;
3288 q = (int)((q * high_err_target) / kf_err);
3289 q = VPXMIN(q, (q_high + q_low) >> 1);
3290 } else if (kf_err < low_err_target &&
3291 rc->projected_frame_size >= frame_under_shoot_limit) {
3292 // The key frame is much better than the previous frame
3294 q_low = q < q_high ? q + 1 : q_high;
3297 q = (int)((q * low_err_target) / kf_err);
3298 q = VPXMIN(q, (q_high + q_low + 1) >> 1);
3301 // Clamp Q to upper and lower limits:
3302 q = clamp(q, q_low, q_high);
3305 } else if (recode_loop_test(
3306 cpi, frame_over_shoot_limit, frame_under_shoot_limit,
3307 q, VPXMAX(q_high, top_index), bottom_index)) {
3308 // Is the projected frame size out of range and are we allowed
3309 // to attempt to recode.
3313 if (cpi->resize_pending == 1) {
3314 // Change in frame size so go back around the recode loop.
3315 cpi->rc.frame_size_selector =
3316 SCALE_STEP1 - cpi->rc.frame_size_selector;
3317 cpi->rc.next_frame_size_selector = cpi->rc.frame_size_selector;
3319 #if CONFIG_INTERNAL_STATS
3320 ++cpi->tot_recode_hits;
3327 // Frame size out of permitted range:
3328 // Update correction factor & compute new Q to try...
3330 // Frame is too large
3331 if (rc->projected_frame_size > rc->this_frame_target) {
3332 // Special case if the projected size is > the max allowed.
3333 if (rc->projected_frame_size >= rc->max_frame_bandwidth)
3334 q_high = rc->worst_quality;
3336 // Raise Qlow as to at least the current value
3337 q_low = q < q_high ? q + 1 : q_high;
3339 if (undershoot_seen || loop_at_this_size > 1) {
3340 // Update rate_correction_factor unless
3341 vp10_rc_update_rate_correction_factors(cpi);
3343 q = (q_high + q_low + 1) / 2;
3345 // Update rate_correction_factor unless
3346 vp10_rc_update_rate_correction_factors(cpi);
3348 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3349 bottom_index, VPXMAX(q_high, top_index));
3351 while (q < q_low && retries < 10) {
3352 vp10_rc_update_rate_correction_factors(cpi);
3353 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3354 bottom_index, VPXMAX(q_high, top_index));
3361 // Frame is too small
3362 q_high = q > q_low ? q - 1 : q_low;
3364 if (overshoot_seen || loop_at_this_size > 1) {
3365 vp10_rc_update_rate_correction_factors(cpi);
3366 q = (q_high + q_low) / 2;
3368 vp10_rc_update_rate_correction_factors(cpi);
3369 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3370 bottom_index, top_index);
3371 // Special case reset for qlow for constrained quality.
3372 // This should only trigger where there is very substantial
3373 // undershoot on a frame and the auto cq level is above
3374 // the user passsed in value.
3375 if (cpi->oxcf.rc_mode == VPX_CQ &&
3380 while (q > q_high && retries < 10) {
3381 vp10_rc_update_rate_correction_factors(cpi);
3382 q = vp10_rc_regulate_q(cpi, rc->this_frame_target,
3383 bottom_index, top_index);
3388 undershoot_seen = 1;
3391 // Clamp Q to upper and lower limits:
3392 q = clamp(q, q_low, q_high);
3394 loop = (q != last_q);
3400 // Special case for overlay frame.
3401 if (rc->is_src_frame_alt_ref &&
3402 rc->projected_frame_size < rc->max_frame_bandwidth)
3407 ++loop_at_this_size;
3409 #if CONFIG_INTERNAL_STATS
3410 ++cpi->tot_recode_hits;
3416 static int get_ref_frame_flags(const VP10_COMP *cpi) {
3417 const int *const map = cpi->common.ref_frame_map;
3418 const int gold_is_last = map[cpi->gld_fb_idx] == map[cpi->lst_fb_idx];
3419 const int alt_is_last = map[cpi->alt_fb_idx] == map[cpi->lst_fb_idx];
3420 const int gold_is_alt = map[cpi->gld_fb_idx] == map[cpi->alt_fb_idx];
3421 int flags = VP9_ALT_FLAG | VP9_GOLD_FLAG | VP9_LAST_FLAG;
3424 flags &= ~VP9_GOLD_FLAG;
3426 if (cpi->rc.frames_till_gf_update_due == INT_MAX)
3427 flags &= ~VP9_GOLD_FLAG;
3430 flags &= ~VP9_ALT_FLAG;
3433 flags &= ~VP9_ALT_FLAG;
3438 static void set_ext_overrides(VP10_COMP *cpi) {
3439 // Overrides the defaults with the externally supplied values with
3440 // vp10_update_reference() and vp10_update_entropy() calls
3441 // Note: The overrides are valid only for the next frame passed
3442 // to encode_frame_to_data_rate() function
3443 if (cpi->ext_refresh_frame_context_pending) {
3444 cpi->common.refresh_frame_context = cpi->ext_refresh_frame_context;
3445 cpi->ext_refresh_frame_context_pending = 0;
3447 if (cpi->ext_refresh_frame_flags_pending) {
3448 cpi->refresh_last_frame = cpi->ext_refresh_last_frame;
3449 cpi->refresh_golden_frame = cpi->ext_refresh_golden_frame;
3450 cpi->refresh_alt_ref_frame = cpi->ext_refresh_alt_ref_frame;
3451 cpi->ext_refresh_frame_flags_pending = 0;
3455 YV12_BUFFER_CONFIG *vp10_scale_if_required_fast(VP10_COMMON *cm,
3456 YV12_BUFFER_CONFIG *unscaled,
3457 YV12_BUFFER_CONFIG *scaled) {
3458 if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
3459 cm->mi_rows * MI_SIZE != unscaled->y_height) {
3460 // For 2x2 scaling down.
3461 vpx_scale_frame(unscaled, scaled, unscaled->y_buffer, 9, 2, 1,
3463 vpx_extend_frame_borders(scaled);
3470 YV12_BUFFER_CONFIG *vp10_scale_if_required(VP10_COMMON *cm,
3471 YV12_BUFFER_CONFIG *unscaled,
3472 YV12_BUFFER_CONFIG *scaled) {
3473 if (cm->mi_cols * MI_SIZE != unscaled->y_width ||
3474 cm->mi_rows * MI_SIZE != unscaled->y_height) {
3475 #if CONFIG_VP9_HIGHBITDEPTH
3476 scale_and_extend_frame_nonnormative(unscaled, scaled, (int)cm->bit_depth);
3478 scale_and_extend_frame_nonnormative(unscaled, scaled);
3479 #endif // CONFIG_VP9_HIGHBITDEPTH
3486 static void set_arf_sign_bias(VP10_COMP *cpi) {
3487 VP10_COMMON *const cm = &cpi->common;
3490 if ((cpi->oxcf.pass == 2) && cpi->multi_arf_allowed) {
3491 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3492 arf_sign_bias = cpi->rc.source_alt_ref_active &&
3493 (!cpi->refresh_alt_ref_frame ||
3494 (gf_group->rf_level[gf_group->index] == GF_ARF_LOW));
3497 (cpi->rc.source_alt_ref_active && !cpi->refresh_alt_ref_frame);
3499 cm->ref_frame_sign_bias[ALTREF_FRAME] = arf_sign_bias;
3502 static int setup_interp_filter_search_mask(VP10_COMP *cpi) {
3503 INTERP_FILTER ifilter;
3504 int ref_total[MAX_REF_FRAMES] = {0};
3505 MV_REFERENCE_FRAME ref;
3507 if (cpi->common.last_frame_type == KEY_FRAME ||
3508 cpi->refresh_alt_ref_frame)
3510 for (ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref)
3511 for (ifilter = EIGHTTAP; ifilter <= EIGHTTAP_SHARP; ++ifilter)
3512 ref_total[ref] += cpi->interp_filter_selected[ref][ifilter];
3514 for (ifilter = EIGHTTAP; ifilter <= EIGHTTAP_SHARP; ++ifilter) {
3515 if ((ref_total[LAST_FRAME] &&
3516 cpi->interp_filter_selected[LAST_FRAME][ifilter] == 0) &&
3517 (ref_total[GOLDEN_FRAME] == 0 ||
3518 cpi->interp_filter_selected[GOLDEN_FRAME][ifilter] * 50
3519 < ref_total[GOLDEN_FRAME]) &&
3520 (ref_total[ALTREF_FRAME] == 0 ||
3521 cpi->interp_filter_selected[ALTREF_FRAME][ifilter] * 50
3522 < ref_total[ALTREF_FRAME]))
3523 mask |= 1 << ifilter;
3528 static void encode_frame_to_data_rate(VP10_COMP *cpi,
3531 unsigned int *frame_flags) {
3532 VP10_COMMON *const cm = &cpi->common;
3533 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
3534 struct segmentation *const seg = &cm->seg;
3537 set_ext_overrides(cpi);
3538 vpx_clear_system_state();
3540 // Set the arf sign bias for this frame.
3541 set_arf_sign_bias(cpi);
3543 // Set default state for segment based loop filter update flags.
3544 cm->lf.mode_ref_delta_update = 0;
3546 if (cpi->oxcf.pass == 2 &&
3547 cpi->sf.adaptive_interp_filter_search)
3548 cpi->sf.interp_filter_search_mask =
3549 setup_interp_filter_search_mask(cpi);
3551 // Set various flags etc to special state if it is a key frame.
3552 if (frame_is_intra_only(cm)) {
3553 // Reset the loop filter deltas and segmentation map.
3554 vp10_reset_segment_features(cm);
3556 // If segmentation is enabled force a map update for key frames.
3558 seg->update_map = 1;
3559 seg->update_data = 1;
3562 // The alternate reference frame cannot be active for a key frame.
3563 cpi->rc.source_alt_ref_active = 0;
3565 cm->error_resilient_mode = oxcf->error_resilient_mode;
3567 // By default, encoder assumes decoder can use prev_mi.
3568 if (cm->error_resilient_mode) {
3569 cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
3570 cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_OFF;
3571 } else if (cm->intra_only) {
3572 // Only reset the current context.
3573 cm->reset_frame_context = RESET_FRAME_CONTEXT_CURRENT;
3577 // For 1 pass CBR, check if we are dropping this frame.
3578 // Never drop on key frame.
3579 if (oxcf->pass == 0 &&
3580 oxcf->rc_mode == VPX_CBR &&
3581 cm->frame_type != KEY_FRAME) {
3582 if (vp10_rc_drop_frame(cpi)) {
3583 vp10_rc_postencode_update_drop_frame(cpi);
3584 ++cm->current_video_frame;
3589 vpx_clear_system_state();
3591 #if CONFIG_INTERNAL_STATS
3592 memset(cpi->mode_chosen_counts, 0,
3593 MAX_MODES * sizeof(*cpi->mode_chosen_counts));
3596 if (cpi->sf.recode_loop == DISALLOW_RECODE) {
3597 encode_without_recode_loop(cpi);
3599 encode_with_recode_loop(cpi, size, dest);
3602 #if CONFIG_VP9_TEMPORAL_DENOISING
3603 #ifdef OUTPUT_YUV_DENOISED
3604 if (oxcf->noise_sensitivity > 0) {
3605 vp10_write_yuv_frame_420(&cpi->denoiser.running_avg_y[INTRA_FRAME],
3610 #ifdef OUTPUT_YUV_SKINMAP
3611 if (cpi->common.current_video_frame > 1) {
3612 vp10_compute_skin_map(cpi, yuv_skinmap_file);
3616 // Special case code to reduce pulsing when key frames are forced at a
3617 // fixed interval. Note the reconstruction error if it is the frame before
3618 // the force key frame
3619 if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) {
3620 #if CONFIG_VP9_HIGHBITDEPTH
3621 if (cm->use_highbitdepth) {
3622 cpi->ambient_err = vp10_highbd_get_y_sse(cpi->Source,
3623 get_frame_new_buffer(cm));
3625 cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3628 cpi->ambient_err = vp10_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
3629 #endif // CONFIG_VP9_HIGHBITDEPTH
3632 // If the encoder forced a KEY_FRAME decision
3633 if (cm->frame_type == KEY_FRAME)
3634 cpi->refresh_last_frame = 1;
3636 cm->frame_to_show = get_frame_new_buffer(cm);
3637 cm->frame_to_show->color_space = cm->color_space;
3638 cm->frame_to_show->color_range = cm->color_range;
3639 cm->frame_to_show->render_width = cm->render_width;
3640 cm->frame_to_show->render_height = cm->render_height;
3642 // Pick the loop filter level for the frame.
3643 loopfilter_frame(cpi, cm);
3645 // build the bitstream
3646 vp10_pack_bitstream(cpi, dest, size);
3648 if (cm->seg.update_map)
3649 update_reference_segmentation_map(cpi);
3651 if (frame_is_intra_only(cm) == 0) {
3652 release_scaled_references(cpi);
3654 vp10_update_reference_frames(cpi);
3656 for (t = TX_4X4; t <= TX_32X32; t++)
3657 full_to_model_counts(cpi->td.counts->coef[t],
3658 cpi->td.rd_counts.coef_counts[t]);
3660 if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
3661 vp10_adapt_coef_probs(cm);
3662 #if CONFIG_MISC_FIXES
3663 vp10_adapt_intra_frame_probs(cm);
3665 if (!frame_is_intra_only(cm))
3666 vp10_adapt_intra_frame_probs(cm);
3670 if (!frame_is_intra_only(cm)) {
3671 if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) {
3672 vp10_adapt_inter_frame_probs(cm);
3673 vp10_adapt_mv_probs(cm, cm->allow_high_precision_mv);
3677 if (cpi->refresh_golden_frame == 1)
3678 cpi->frame_flags |= FRAMEFLAGS_GOLDEN;
3680 cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN;
3682 if (cpi->refresh_alt_ref_frame == 1)
3683 cpi->frame_flags |= FRAMEFLAGS_ALTREF;
3685 cpi->frame_flags &= ~FRAMEFLAGS_ALTREF;
3687 cpi->ref_frame_flags = get_ref_frame_flags(cpi);
3689 cm->last_frame_type = cm->frame_type;
3691 vp10_rc_postencode_update(cpi, *size);
3694 output_frame_level_debug_stats(cpi);
3697 if (cm->frame_type == KEY_FRAME) {
3698 // Tell the caller that the frame was coded as a key frame
3699 *frame_flags = cpi->frame_flags | FRAMEFLAGS_KEY;
3701 *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY;
3704 // Clear the one shot update flags for segmentation map and mode/ref loop
3706 cm->seg.update_map = 0;
3707 cm->seg.update_data = 0;
3708 cm->lf.mode_ref_delta_update = 0;
3710 // keep track of the last coded dimensions
3711 cm->last_width = cm->width;
3712 cm->last_height = cm->height;
3714 // reset to normal state now that we are done.
3715 if (!cm->show_existing_frame)
3716 cm->last_show_frame = cm->show_frame;
3718 if (cm->show_frame) {
3719 vp10_swap_mi_and_prev_mi(cm);
3720 // Don't increment frame counters if this was an altref buffer
3721 // update not a real frame
3722 ++cm->current_video_frame;
3724 cm->prev_frame = cm->cur_frame;
3727 static void Pass0Encode(VP10_COMP *cpi, size_t *size, uint8_t *dest,
3728 unsigned int *frame_flags) {
3729 if (cpi->oxcf.rc_mode == VPX_CBR) {
3730 vp10_rc_get_one_pass_cbr_params(cpi);
3732 vp10_rc_get_one_pass_vbr_params(cpi);
3734 encode_frame_to_data_rate(cpi, size, dest, frame_flags);
3737 static void Pass2Encode(VP10_COMP *cpi, size_t *size,
3738 uint8_t *dest, unsigned int *frame_flags) {
3739 cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
3740 encode_frame_to_data_rate(cpi, size, dest, frame_flags);
3742 vp10_twopass_postencode_update(cpi);
3745 static void init_ref_frame_bufs(VP10_COMMON *cm) {
3747 BufferPool *const pool = cm->buffer_pool;
3748 cm->new_fb_idx = INVALID_IDX;
3749 for (i = 0; i < REF_FRAMES; ++i) {
3750 cm->ref_frame_map[i] = INVALID_IDX;
3751 pool->frame_bufs[i].ref_count = 0;
3755 static void check_initial_width(VP10_COMP *cpi,
3756 #if CONFIG_VP9_HIGHBITDEPTH
3757 int use_highbitdepth,
3759 int subsampling_x, int subsampling_y) {
3760 VP10_COMMON *const cm = &cpi->common;
3762 if (!cpi->initial_width ||
3763 #if CONFIG_VP9_HIGHBITDEPTH
3764 cm->use_highbitdepth != use_highbitdepth ||
3766 cm->subsampling_x != subsampling_x ||
3767 cm->subsampling_y != subsampling_y) {
3768 cm->subsampling_x = subsampling_x;
3769 cm->subsampling_y = subsampling_y;
3770 #if CONFIG_VP9_HIGHBITDEPTH
3771 cm->use_highbitdepth = use_highbitdepth;
3774 alloc_raw_frame_buffers(cpi);
3775 init_ref_frame_bufs(cm);
3776 alloc_util_frame_buffers(cpi);
3778 init_motion_estimation(cpi); // TODO(agrange) This can be removed.
3780 cpi->initial_width = cm->width;
3781 cpi->initial_height = cm->height;
3782 cpi->initial_mbs = cm->MBs;
3786 #if CONFIG_VP9_TEMPORAL_DENOISING
3787 static void setup_denoiser_buffer(VP10_COMP *cpi) {
3788 VP10_COMMON *const cm = &cpi->common;
3789 if (cpi->oxcf.noise_sensitivity > 0 &&
3790 !cpi->denoiser.frame_buffer_initialized) {
3791 vp10_denoiser_alloc(&(cpi->denoiser), cm->width, cm->height,
3792 cm->subsampling_x, cm->subsampling_y,
3793 #if CONFIG_VP9_HIGHBITDEPTH
3794 cm->use_highbitdepth,
3796 VP9_ENC_BORDER_IN_PIXELS);
3801 int vp10_receive_raw_frame(VP10_COMP *cpi, unsigned int frame_flags,
3802 YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
3804 VP10_COMMON *cm = &cpi->common;
3805 struct vpx_usec_timer timer;
3807 const int subsampling_x = sd->subsampling_x;
3808 const int subsampling_y = sd->subsampling_y;
3809 #if CONFIG_VP9_HIGHBITDEPTH
3810 const int use_highbitdepth = sd->flags & YV12_FLAG_HIGHBITDEPTH;
3811 check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y);
3813 check_initial_width(cpi, subsampling_x, subsampling_y);
3814 #endif // CONFIG_VP9_HIGHBITDEPTH
3816 #if CONFIG_VP9_TEMPORAL_DENOISING
3817 setup_denoiser_buffer(cpi);
3819 vpx_usec_timer_start(&timer);
3821 if (vp10_lookahead_push(cpi->lookahead, sd, time_stamp, end_time,
3822 #if CONFIG_VP9_HIGHBITDEPTH
3824 #endif // CONFIG_VP9_HIGHBITDEPTH
3827 vpx_usec_timer_mark(&timer);
3828 cpi->time_receive_data += vpx_usec_timer_elapsed(&timer);
3830 if ((cm->profile == PROFILE_0 || cm->profile == PROFILE_2) &&
3831 (subsampling_x != 1 || subsampling_y != 1)) {
3832 vpx_internal_error(&cm->error, VPX_CODEC_INVALID_PARAM,
3833 "Non-4:2:0 color format requires profile 1 or 3");
3836 if ((cm->profile == PROFILE_1 || cm->profile == PROFILE_3) &&
3837 (subsampling_x == 1 && subsampling_y == 1)) {
3838 vpx_internal_error(&cm->error, VPX_CODEC_INVALID_PARAM,
3839 "4:2:0 color format requires profile 0 or 2");
3847 static int frame_is_reference(const VP10_COMP *cpi) {
3848 const VP10_COMMON *cm = &cpi->common;
3850 return cm->frame_type == KEY_FRAME ||
3851 cpi->refresh_last_frame ||
3852 cpi->refresh_golden_frame ||
3853 cpi->refresh_alt_ref_frame ||
3854 cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_OFF ||
3855 cm->lf.mode_ref_delta_update ||
3856 cm->seg.update_map ||
3857 cm->seg.update_data;
3860 static void adjust_frame_rate(VP10_COMP *cpi,
3861 const struct lookahead_entry *source) {
3862 int64_t this_duration;
3865 if (source->ts_start == cpi->first_time_stamp_ever) {
3866 this_duration = source->ts_end - source->ts_start;
3869 int64_t last_duration = cpi->last_end_time_stamp_seen
3870 - cpi->last_time_stamp_seen;
3872 this_duration = source->ts_end - cpi->last_end_time_stamp_seen;
3874 // do a step update if the duration changes by 10%
3876 step = (int)((this_duration - last_duration) * 10 / last_duration);
3879 if (this_duration) {
3881 vp10_new_framerate(cpi, 10000000.0 / this_duration);
3883 // Average this frame's rate into the last second's average
3884 // frame rate. If we haven't seen 1 second yet, then average
3885 // over the whole interval seen.
3886 const double interval = VPXMIN(
3887 (double)(source->ts_end - cpi->first_time_stamp_ever), 10000000.0);
3888 double avg_duration = 10000000.0 / cpi->framerate;
3889 avg_duration *= (interval - avg_duration + this_duration);
3890 avg_duration /= interval;
3892 vp10_new_framerate(cpi, 10000000.0 / avg_duration);
3895 cpi->last_time_stamp_seen = source->ts_start;
3896 cpi->last_end_time_stamp_seen = source->ts_end;
3899 // Returns 0 if this is not an alt ref else the offset of the source frame
3900 // used as the arf midpoint.
3901 static int get_arf_src_index(VP10_COMP *cpi) {
3902 RATE_CONTROL *const rc = &cpi->rc;
3903 int arf_src_index = 0;
3904 if (is_altref_enabled(cpi)) {
3905 if (cpi->oxcf.pass == 2) {
3906 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3907 if (gf_group->update_type[gf_group->index] == ARF_UPDATE) {
3908 arf_src_index = gf_group->arf_src_offset[gf_group->index];
3910 } else if (rc->source_alt_ref_pending) {
3911 arf_src_index = rc->frames_till_gf_update_due;
3914 return arf_src_index;
3917 static void check_src_altref(VP10_COMP *cpi,
3918 const struct lookahead_entry *source) {
3919 RATE_CONTROL *const rc = &cpi->rc;
3921 if (cpi->oxcf.pass == 2) {
3922 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
3923 rc->is_src_frame_alt_ref =
3924 (gf_group->update_type[gf_group->index] == OVERLAY_UPDATE);
3926 rc->is_src_frame_alt_ref = cpi->alt_ref_source &&
3927 (source == cpi->alt_ref_source);
3930 if (rc->is_src_frame_alt_ref) {
3931 // Current frame is an ARF overlay frame.
3932 cpi->alt_ref_source = NULL;
3934 // Don't refresh the last buffer for an ARF overlay frame. It will
3935 // become the GF so preserve last as an alternative prediction option.
3936 cpi->refresh_last_frame = 0;
3940 #if CONFIG_INTERNAL_STATS
3941 extern double vp10_get_blockiness(const unsigned char *img1, int img1_pitch,
3942 const unsigned char *img2, int img2_pitch,
3943 int width, int height);
3945 static void adjust_image_stat(double y, double u, double v, double all,
3950 s->stat[ALL] += all;
3951 s->worst = VPXMIN(s->worst, all);
3953 #endif // CONFIG_INTERNAL_STATS
3955 int vp10_get_compressed_data(VP10_COMP *cpi, unsigned int *frame_flags,
3956 size_t *size, uint8_t *dest,
3957 int64_t *time_stamp, int64_t *time_end, int flush) {
3958 const VP10EncoderConfig *const oxcf = &cpi->oxcf;
3959 VP10_COMMON *const cm = &cpi->common;
3960 BufferPool *const pool = cm->buffer_pool;
3961 RATE_CONTROL *const rc = &cpi->rc;
3962 struct vpx_usec_timer cmptimer;
3963 YV12_BUFFER_CONFIG *force_src_buffer = NULL;
3964 struct lookahead_entry *last_source = NULL;
3965 struct lookahead_entry *source = NULL;
3969 vpx_usec_timer_start(&cmptimer);
3971 vp10_set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV);
3973 // Is multi-arf enabled.
3974 // Note that at the moment multi_arf is only configured for 2 pass VBR
3975 if ((oxcf->pass == 2) && (cpi->oxcf.enable_auto_arf > 1))
3976 cpi->multi_arf_allowed = 1;
3978 cpi->multi_arf_allowed = 0;
3981 cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
3982 cm->refresh_frame_context =
3983 oxcf->error_resilient_mode ? REFRESH_FRAME_CONTEXT_OFF :
3984 oxcf->frame_parallel_decoding_mode ? REFRESH_FRAME_CONTEXT_FORWARD
3985 : REFRESH_FRAME_CONTEXT_BACKWARD;
3987 cpi->refresh_last_frame = 1;
3988 cpi->refresh_golden_frame = 0;
3989 cpi->refresh_alt_ref_frame = 0;
3991 // Should we encode an arf frame.
3992 arf_src_index = get_arf_src_index(cpi);
3994 if (arf_src_index) {
3995 assert(arf_src_index <= rc->frames_to_key);
3997 if ((source = vp10_lookahead_peek(cpi->lookahead, arf_src_index)) != NULL) {
3998 cpi->alt_ref_source = source;
4000 if (oxcf->arnr_max_frames > 0) {
4001 // Produce the filtered ARF frame.
4002 vp10_temporal_filter(cpi, arf_src_index);
4003 vpx_extend_frame_borders(&cpi->alt_ref_buffer);
4004 force_src_buffer = &cpi->alt_ref_buffer;
4009 cpi->refresh_alt_ref_frame = 1;
4010 cpi->refresh_golden_frame = 0;
4011 cpi->refresh_last_frame = 0;
4012 rc->is_src_frame_alt_ref = 0;
4013 rc->source_alt_ref_pending = 0;
4015 rc->source_alt_ref_pending = 0;
4020 // Get last frame source.
4021 if (cm->current_video_frame > 0) {
4022 if ((last_source = vp10_lookahead_peek(cpi->lookahead, -1)) == NULL)
4026 // Read in the source frame.
4027 source = vp10_lookahead_pop(cpi->lookahead, flush);
4029 if (source != NULL) {
4033 // Check to see if the frame should be encoded as an arf overlay.
4034 check_src_altref(cpi, source);
4039 cpi->un_scaled_source = cpi->Source = force_src_buffer ? force_src_buffer
4042 cpi->unscaled_last_source = last_source != NULL ? &last_source->img : NULL;
4044 *time_stamp = source->ts_start;
4045 *time_end = source->ts_end;
4046 *frame_flags = (source->flags & VPX_EFLAG_FORCE_KF) ? FRAMEFLAGS_KEY : 0;
4050 if (flush && oxcf->pass == 1 && !cpi->twopass.first_pass_done) {
4051 vp10_end_first_pass(cpi); /* get last stats packet */
4052 cpi->twopass.first_pass_done = 1;
4057 if (source->ts_start < cpi->first_time_stamp_ever) {
4058 cpi->first_time_stamp_ever = source->ts_start;
4059 cpi->last_end_time_stamp_seen = source->ts_start;
4062 // Clear down mmx registers
4063 vpx_clear_system_state();
4065 // adjust frame rates based on timestamps given
4066 if (cm->show_frame) {
4067 adjust_frame_rate(cpi, source);
4070 // Find a free buffer for the new frame, releasing the reference previously
4072 if (cm->new_fb_idx != INVALID_IDX) {
4073 --pool->frame_bufs[cm->new_fb_idx].ref_count;
4075 cm->new_fb_idx = get_free_fb(cm);
4077 if (cm->new_fb_idx == INVALID_IDX)
4080 cm->cur_frame = &pool->frame_bufs[cm->new_fb_idx];
4082 if (cpi->multi_arf_allowed) {
4083 if (cm->frame_type == KEY_FRAME) {
4084 init_buffer_indices(cpi);
4085 } else if (oxcf->pass == 2) {
4086 const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
4087 cpi->alt_fb_idx = gf_group->arf_ref_idx[gf_group->index];
4091 // Start with a 0 size frame.
4094 cpi->frame_flags = *frame_flags;
4096 if (oxcf->pass == 2) {
4097 vp10_rc_get_second_pass_params(cpi);
4098 } else if (oxcf->pass == 1) {
4099 set_frame_size(cpi);
4102 if (cpi->oxcf.pass != 0 || frame_is_intra_only(cm) == 1) {
4103 for (i = 0; i < MAX_REF_FRAMES; ++i)
4104 cpi->scaled_ref_idx[i] = INVALID_IDX;
4107 if (oxcf->pass == 1) {
4108 cpi->td.mb.e_mbd.lossless[0] = is_lossless_requested(oxcf);
4109 vp10_first_pass(cpi, source);
4110 } else if (oxcf->pass == 2) {
4111 Pass2Encode(cpi, size, dest, frame_flags);
4114 Pass0Encode(cpi, size, dest, frame_flags);
4117 if (cm->refresh_frame_context != REFRESH_FRAME_CONTEXT_OFF)
4118 cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
4120 // No frame encoded, or frame was dropped, release scaled references.
4121 if ((*size == 0) && (frame_is_intra_only(cm) == 0)) {
4122 release_scaled_references(cpi);
4126 cpi->droppable = !frame_is_reference(cpi);
4129 vpx_usec_timer_mark(&cmptimer);
4130 cpi->time_compress_data += vpx_usec_timer_elapsed(&cmptimer);
4132 if (cpi->b_calculate_psnr && oxcf->pass != 1 && cm->show_frame)
4133 generate_psnr_packet(cpi);
4135 #if CONFIG_INTERNAL_STATS
4137 if (oxcf->pass != 1) {
4138 double samples = 0.0;
4139 cpi->bytes += (int)(*size);
4141 if (cm->show_frame) {
4144 if (cpi->b_calculate_psnr) {
4145 YV12_BUFFER_CONFIG *orig = cpi->Source;
4146 YV12_BUFFER_CONFIG *recon = cpi->common.frame_to_show;
4147 YV12_BUFFER_CONFIG *pp = &cm->post_proc_buffer;
4149 #if CONFIG_VP9_HIGHBITDEPTH
4150 calc_highbd_psnr(orig, recon, &psnr, cpi->td.mb.e_mbd.bd,
4151 cpi->oxcf.input_bit_depth);
4153 calc_psnr(orig, recon, &psnr);
4154 #endif // CONFIG_VP9_HIGHBITDEPTH
4156 adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3],
4157 psnr.psnr[0], &cpi->psnr);
4158 cpi->total_sq_error += psnr.sse[0];
4159 cpi->total_samples += psnr.samples[0];
4160 samples = psnr.samples[0];
4164 double frame_ssim2 = 0, weight = 0;
4165 #if CONFIG_VP9_POSTPROC
4166 if (vpx_alloc_frame_buffer(&cm->post_proc_buffer,
4167 recon->y_crop_width, recon->y_crop_height,
4168 cm->subsampling_x, cm->subsampling_y,
4169 #if CONFIG_VP9_HIGHBITDEPTH
4170 cm->use_highbitdepth,
4172 VP9_ENC_BORDER_IN_PIXELS,
4173 cm->byte_alignment) < 0) {
4174 vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
4175 "Failed to allocate post processing buffer");
4178 vp10_deblock(cm->frame_to_show, &cm->post_proc_buffer,
4179 cm->lf.filter_level * 10 / 6);
4181 vpx_clear_system_state();
4183 #if CONFIG_VP9_HIGHBITDEPTH
4184 calc_highbd_psnr(orig, pp, &psnr2, cpi->td.mb.e_mbd.bd,
4185 cpi->oxcf.input_bit_depth);
4187 calc_psnr(orig, pp, &psnr2);
4188 #endif // CONFIG_VP9_HIGHBITDEPTH
4190 cpi->totalp_sq_error += psnr2.sse[0];
4191 cpi->totalp_samples += psnr2.samples[0];
4192 adjust_image_stat(psnr2.psnr[1], psnr2.psnr[2], psnr2.psnr[3],
4193 psnr2.psnr[0], &cpi->psnrp);
4195 #if CONFIG_VP9_HIGHBITDEPTH
4196 if (cm->use_highbitdepth) {
4197 frame_ssim2 = vpx_highbd_calc_ssim(orig, recon, &weight,
4198 (int)cm->bit_depth);
4200 frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
4203 frame_ssim2 = vpx_calc_ssim(orig, recon, &weight);
4204 #endif // CONFIG_VP9_HIGHBITDEPTH
4206 cpi->worst_ssim= VPXMIN(cpi->worst_ssim, frame_ssim2);
4207 cpi->summed_quality += frame_ssim2 * weight;
4208 cpi->summed_weights += weight;
4210 #if CONFIG_VP9_HIGHBITDEPTH
4211 if (cm->use_highbitdepth) {
4212 frame_ssim2 = vpx_highbd_calc_ssim(
4213 orig, &cm->post_proc_buffer, &weight, (int)cm->bit_depth);
4215 frame_ssim2 = vpx_calc_ssim(orig, &cm->post_proc_buffer, &weight);
4218 frame_ssim2 = vpx_calc_ssim(orig, &cm->post_proc_buffer, &weight);
4219 #endif // CONFIG_VP9_HIGHBITDEPTH
4221 cpi->summedp_quality += frame_ssim2 * weight;
4222 cpi->summedp_weights += weight;
4225 FILE *f = fopen("q_used.stt", "a");
4226 fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n",
4227 cpi->common.current_video_frame, y2, u2, v2,
4228 frame_psnr2, frame_ssim2);
4234 if (cpi->b_calculate_blockiness) {
4235 #if CONFIG_VP9_HIGHBITDEPTH
4236 if (!cm->use_highbitdepth)
4239 double frame_blockiness = vp10_get_blockiness(
4240 cpi->Source->y_buffer, cpi->Source->y_stride,
4241 cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
4242 cpi->Source->y_width, cpi->Source->y_height);
4243 cpi->worst_blockiness =
4244 VPXMAX(cpi->worst_blockiness, frame_blockiness);
4245 cpi->total_blockiness += frame_blockiness;
4249 if (cpi->b_calculate_consistency) {
4250 #if CONFIG_VP9_HIGHBITDEPTH
4251 if (!cm->use_highbitdepth)
4254 double this_inconsistency = vpx_get_ssim_metrics(
4255 cpi->Source->y_buffer, cpi->Source->y_stride,
4256 cm->frame_to_show->y_buffer, cm->frame_to_show->y_stride,
4257 cpi->Source->y_width, cpi->Source->y_height, cpi->ssim_vars,
4260 const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1);
4261 double consistency = vpx_sse_to_psnr(samples, peak,
4262 (double)cpi->total_inconsistency);
4263 if (consistency > 0.0)
4264 cpi->worst_consistency =
4265 VPXMIN(cpi->worst_consistency, consistency);
4266 cpi->total_inconsistency += this_inconsistency;
4270 if (cpi->b_calculate_ssimg) {
4271 double y, u, v, frame_all;
4272 #if CONFIG_VP9_HIGHBITDEPTH
4273 if (cm->use_highbitdepth) {
4274 frame_all = vpx_highbd_calc_ssimg(cpi->Source, cm->frame_to_show, &y,
4275 &u, &v, (int)cm->bit_depth);
4277 frame_all = vpx_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u,
4281 frame_all = vpx_calc_ssimg(cpi->Source, cm->frame_to_show, &y, &u, &v);
4282 #endif // CONFIG_VP9_HIGHBITDEPTH
4283 adjust_image_stat(y, u, v, frame_all, &cpi->ssimg);
4285 #if CONFIG_VP9_HIGHBITDEPTH
4286 if (!cm->use_highbitdepth)
4289 double y, u, v, frame_all;
4290 frame_all = vpx_calc_fastssim(cpi->Source, cm->frame_to_show, &y, &u,
4292 adjust_image_stat(y, u, v, frame_all, &cpi->fastssim);
4293 /* TODO(JBB): add 10/12 bit support */
4295 #if CONFIG_VP9_HIGHBITDEPTH
4296 if (!cm->use_highbitdepth)
4299 double y, u, v, frame_all;
4300 frame_all = vpx_psnrhvs(cpi->Source, cm->frame_to_show, &y, &u, &v);
4301 adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs);
4307 vpx_clear_system_state();
4311 int vp10_get_preview_raw_frame(VP10_COMP *cpi, YV12_BUFFER_CONFIG *dest,
4312 vp10_ppflags_t *flags) {
4313 VP10_COMMON *cm = &cpi->common;
4314 #if !CONFIG_VP9_POSTPROC
4318 if (!cm->show_frame) {
4322 #if CONFIG_VP9_POSTPROC
4323 ret = vp10_post_proc_frame(cm, dest, flags);
4325 if (cm->frame_to_show) {
4326 *dest = *cm->frame_to_show;
4327 dest->y_width = cm->width;
4328 dest->y_height = cm->height;
4329 dest->uv_width = cm->width >> cm->subsampling_x;
4330 dest->uv_height = cm->height >> cm->subsampling_y;
4335 #endif // !CONFIG_VP9_POSTPROC
4336 vpx_clear_system_state();
4341 int vp10_set_internal_size(VP10_COMP *cpi,
4342 VPX_SCALING horiz_mode, VPX_SCALING vert_mode) {
4343 VP10_COMMON *cm = &cpi->common;
4344 int hr = 0, hs = 0, vr = 0, vs = 0;
4346 if (horiz_mode > ONETWO || vert_mode > ONETWO)
4349 Scale2Ratio(horiz_mode, &hr, &hs);
4350 Scale2Ratio(vert_mode, &vr, &vs);
4352 // always go to the next whole number
4353 cm->width = (hs - 1 + cpi->oxcf.width * hr) / hs;
4354 cm->height = (vs - 1 + cpi->oxcf.height * vr) / vs;
4355 assert(cm->width <= cpi->initial_width);
4356 assert(cm->height <= cpi->initial_height);
4358 update_frame_size(cpi);
4363 int vp10_set_size_literal(VP10_COMP *cpi, unsigned int width,
4364 unsigned int height) {
4365 VP10_COMMON *cm = &cpi->common;
4366 #if CONFIG_VP9_HIGHBITDEPTH
4367 check_initial_width(cpi, cm->use_highbitdepth, 1, 1);
4369 check_initial_width(cpi, 1, 1);
4370 #endif // CONFIG_VP9_HIGHBITDEPTH
4372 #if CONFIG_VP9_TEMPORAL_DENOISING
4373 setup_denoiser_buffer(cpi);
4378 if (cm->width > cpi->initial_width) {
4379 cm->width = cpi->initial_width;
4380 printf("Warning: Desired width too large, changed to %d\n", cm->width);
4385 cm->height = height;
4386 if (cm->height > cpi->initial_height) {
4387 cm->height = cpi->initial_height;
4388 printf("Warning: Desired height too large, changed to %d\n", cm->height);
4391 assert(cm->width <= cpi->initial_width);
4392 assert(cm->height <= cpi->initial_height);
4394 update_frame_size(cpi);
4399 int64_t vp10_get_y_sse(const YV12_BUFFER_CONFIG *a,
4400 const YV12_BUFFER_CONFIG *b) {
4401 assert(a->y_crop_width == b->y_crop_width);
4402 assert(a->y_crop_height == b->y_crop_height);
4404 return get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
4405 a->y_crop_width, a->y_crop_height);
4408 #if CONFIG_VP9_HIGHBITDEPTH
4409 int64_t vp10_highbd_get_y_sse(const YV12_BUFFER_CONFIG *a,
4410 const YV12_BUFFER_CONFIG *b) {
4411 assert(a->y_crop_width == b->y_crop_width);
4412 assert(a->y_crop_height == b->y_crop_height);
4413 assert((a->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
4414 assert((b->flags & YV12_FLAG_HIGHBITDEPTH) != 0);
4416 return highbd_get_sse(a->y_buffer, a->y_stride, b->y_buffer, b->y_stride,
4417 a->y_crop_width, a->y_crop_height);
4419 #endif // CONFIG_VP9_HIGHBITDEPTH
4421 int vp10_get_quantizer(VP10_COMP *cpi) {
4422 return cpi->common.base_qindex;
4425 void vp10_apply_encoding_flags(VP10_COMP *cpi, vpx_enc_frame_flags_t flags) {
4426 if (flags & (VP8_EFLAG_NO_REF_LAST | VP8_EFLAG_NO_REF_GF |
4427 VP8_EFLAG_NO_REF_ARF)) {
4430 if (flags & VP8_EFLAG_NO_REF_LAST)
4431 ref ^= VP9_LAST_FLAG;
4433 if (flags & VP8_EFLAG_NO_REF_GF)
4434 ref ^= VP9_GOLD_FLAG;
4436 if (flags & VP8_EFLAG_NO_REF_ARF)
4437 ref ^= VP9_ALT_FLAG;
4439 vp10_use_as_reference(cpi, ref);
4442 if (flags & (VP8_EFLAG_NO_UPD_LAST | VP8_EFLAG_NO_UPD_GF |
4443 VP8_EFLAG_NO_UPD_ARF | VP8_EFLAG_FORCE_GF |
4444 VP8_EFLAG_FORCE_ARF)) {
4447 if (flags & VP8_EFLAG_NO_UPD_LAST)
4448 upd ^= VP9_LAST_FLAG;
4450 if (flags & VP8_EFLAG_NO_UPD_GF)
4451 upd ^= VP9_GOLD_FLAG;
4453 if (flags & VP8_EFLAG_NO_UPD_ARF)
4454 upd ^= VP9_ALT_FLAG;
4456 vp10_update_reference(cpi, upd);
4459 if (flags & VP8_EFLAG_NO_UPD_ENTROPY) {
4460 vp10_update_entropy(cpi, 0);