Google Git
Sign in
cobalt / cobalt / 308ed6b84664d59944cd65f4b4359c28a2443be6 / . / src / third_party / libvpx / vp10 / encoder / encoder.c
blob: e7fff829952be41b5ab7600661c8db23a0d07402 [file] [log] [blame]
/*
* Copyright (c) 2010 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <math.h>
#include <stdio.h>
#include <limits.h>
#include "./vpx_config.h"
#include "vp10/common/alloccommon.h"
#include "vp10/common/filter.h"
#include "vp10/common/idct.h"
#if CONFIG_VP9_POSTPROC
#include "vp10/common/postproc.h"
#endif
#include "vp10/common/reconinter.h"
#include "vp10/common/reconintra.h"
#include "vp10/common/tile_common.h"
#include "vp10/encoder/aq_complexity.h"
#include "vp10/encoder/aq_cyclicrefresh.h"
#include "vp10/encoder/aq_variance.h"
#include "vp10/encoder/bitstream.h"
#include "vp10/encoder/context_tree.h"
#include "vp10/encoder/encodeframe.h"
#include "vp10/encoder/encodemv.h"
#include "vp10/encoder/encoder.h"
#include "vp10/encoder/ethread.h"
#include "vp10/encoder/firstpass.h"
#include "vp10/encoder/mbgraph.h"
#include "vp10/encoder/picklpf.h"
#include "vp10/encoder/ratectrl.h"
#include "vp10/encoder/rd.h"
#include "vp10/encoder/resize.h"
#include "vp10/encoder/segmentation.h"
#include "vp10/encoder/skin_detection.h"
#include "vp10/encoder/speed_features.h"
#include "vp10/encoder/temporal_filter.h"
#include "./vp10_rtcd.h"
#include "./vpx_dsp_rtcd.h"
#include "./vpx_scale_rtcd.h"
#include "vpx/internal/vpx_psnr.h"
#if CONFIG_INTERNAL_STATS
#include "vpx_dsp/ssim.h"
#endif
#include "vpx_dsp/vpx_dsp_common.h"
#include "vpx_dsp/vpx_filter.h"
#include "vpx_ports/mem.h"
#include "vpx_ports/system_state.h"
#include "vpx_ports/vpx_timer.h"
#include "vpx_scale/vpx_scale.h"
#define AM_SEGMENT_ID_INACTIVE 7
#define AM_SEGMENT_ID_ACTIVE 0
#define SHARP_FILTER_QTHRESH 0 /* Q threshold for 8-tap sharp filter */
#define ALTREF_HIGH_PRECISION_MV 1 // Whether to use high precision mv
// for altref computation.
#define HIGH_PRECISION_MV_QTHRESH 200 // Q threshold for high precision
// mv. Choose a very high value for
// now so that HIGH_PRECISION is always
// chosen.
// #define OUTPUT_YUV_REC
#ifdef OUTPUT_YUV_DENOISED
FILE *yuv_denoised_file = NULL;
#endif
#ifdef OUTPUT_YUV_SKINMAP
FILE *yuv_skinmap_file = NULL;
#endif
#ifdef OUTPUT_YUV_REC
FILE *yuv_rec_file;
#endif
#if 0
FILE *framepsnr;
FILE *kf_list;
FILE *keyfile;
#endif
static INLINE void Scale2Ratio(VPX_SCALING mode, int *hr, int *hs) {
switch (mode) {
case NORMAL:
*hr = 1;
*hs = 1;
break;
case FOURFIVE:
*hr = 4;
*hs = 5;
break;
case THREEFIVE:
*hr = 3;
*hs = 5;
break;
case ONETWO:
*hr = 1;
*hs = 2;
break;
default:
*hr = 1;
*hs = 1;
assert(0);
break;
}
}
// Mark all inactive blocks as active. Other segmentation features may be set
// so memset cannot be used, instead only inactive blocks should be reset.
static void suppress_active_map(VP10_COMP *cpi) {
unsigned char *const seg_map = cpi->segmentation_map;
int i;
if (cpi->active_map.enabled || cpi->active_map.update)
for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i)
if (seg_map[i] == AM_SEGMENT_ID_INACTIVE)
seg_map[i] = AM_SEGMENT_ID_ACTIVE;
}
static void apply_active_map(VP10_COMP *cpi) {
struct segmentation *const seg = &cpi->common.seg;
unsigned char *const seg_map = cpi->segmentation_map;
const unsigned char *const active_map = cpi->active_map.map;
int i;
assert(AM_SEGMENT_ID_ACTIVE == CR_SEGMENT_ID_BASE);
if (frame_is_intra_only(&cpi->common)) {
cpi->active_map.enabled = 0;
cpi->active_map.update = 1;
}
if (cpi->active_map.update) {
if (cpi->active_map.enabled) {
for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i)
if (seg_map[i] == AM_SEGMENT_ID_ACTIVE) seg_map[i] = active_map[i];
vp10_enable_segmentation(seg);
vp10_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP);
vp10_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF);
// Setting the data to -MAX_LOOP_FILTER will result in the computed loop
// filter level being zero regardless of the value of seg->abs_delta.
vp10_set_segdata(seg, AM_SEGMENT_ID_INACTIVE,
SEG_LVL_ALT_LF, -MAX_LOOP_FILTER);
} else {
vp10_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP);
vp10_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF);
if (seg->enabled) {
seg->update_data = 1;
seg->update_map = 1;
}
}
cpi->active_map.update = 0;
}
}
int vp10_set_active_map(VP10_COMP* cpi,
unsigned char* new_map_16x16,
int rows,
int cols) {
if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols) {
unsigned char *const active_map_8x8 = cpi->active_map.map;
const int mi_rows = cpi->common.mi_rows;
const int mi_cols = cpi->common.mi_cols;
cpi->active_map.update = 1;
if (new_map_16x16) {
int r, c;
for (r = 0; r < mi_rows; ++r) {
for (c = 0; c < mi_cols; ++c) {
active_map_8x8[r * mi_cols + c] =
new_map_16x16[(r >> 1) * cols + (c >> 1)]
? AM_SEGMENT_ID_ACTIVE
: AM_SEGMENT_ID_INACTIVE;
}
}
cpi->active_map.enabled = 1;
} else {
cpi->active_map.enabled = 0;
}
return 0;
} else {
return -1;
}
}
int vp10_get_active_map(VP10_COMP* cpi,
unsigned char* new_map_16x16,
int rows,
int cols) {
if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols &&
new_map_16x16) {
unsigned char* const seg_map_8x8 = cpi->segmentation_map;
const int mi_rows = cpi->common.mi_rows;
const int mi_cols = cpi->common.mi_cols;
memset(new_map_16x16, !cpi->active_map.enabled, rows * cols);
if (cpi->active_map.enabled) {
int r, c;
for (r = 0; r < mi_rows; ++r) {
for (c = 0; c < mi_cols; ++c) {
// Cyclic refresh segments are considered active despite not having
// AM_SEGMENT_ID_ACTIVE
new_map_16x16[(r >> 1) * cols + (c >> 1)] |=
seg_map_8x8[r * mi_cols + c] != AM_SEGMENT_ID_INACTIVE;
}
}
}
return 0;
} else {
return -1;
}
}
void vp10_set_high_precision_mv(VP10_COMP *cpi, int allow_high_precision_mv) {
MACROBLOCK *const mb = &cpi->td.mb;
cpi->common.allow_high_precision_mv = allow_high_precision_mv;
if (cpi->common.allow_high_precision_mv) {
mb->mvcost = mb->nmvcost_hp;
mb->mvsadcost = mb->nmvsadcost_hp;
} else {
mb->mvcost = mb->nmvcost;
mb->mvsadcost = mb->nmvsadcost;
}
}
static void setup_frame(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
// Set up entropy context depending on frame type. The decoder mandates
// the use of the default context, index 0, for keyframes and inter
// frames where the error_resilient_mode or intra_only flag is set. For
// other inter-frames the encoder currently uses only two contexts;
// context 1 for ALTREF frames and context 0 for the others.
if (frame_is_intra_only(cm) || cm->error_resilient_mode) {
vp10_setup_past_independence(cm);
} else {
cm->frame_context_idx = cpi->refresh_alt_ref_frame;
}
if (cm->frame_type == KEY_FRAME) {
cpi->refresh_golden_frame = 1;
cpi->refresh_alt_ref_frame = 1;
vp10_zero(cpi->interp_filter_selected);
} else {
*cm->fc = cm->frame_contexts[cm->frame_context_idx];
vp10_zero(cpi->interp_filter_selected[0]);
}
}
static void vp10_enc_setup_mi(VP10_COMMON *cm) {
int i;
cm->mi = cm->mip + cm->mi_stride + 1;
memset(cm->mip, 0, cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mip));
cm->prev_mi = cm->prev_mip + cm->mi_stride + 1;
// Clear top border row
memset(cm->prev_mip, 0, sizeof(*cm->prev_mip) * cm->mi_stride);
// Clear left border column
for (i = 1; i < cm->mi_rows + 1; ++i)
memset(&cm->prev_mip[i * cm->mi_stride], 0, sizeof(*cm->prev_mip));
cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mi_stride + 1;
memset(cm->mi_grid_base, 0,
cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mi_grid_base));
}
static int vp10_enc_alloc_mi(VP10_COMMON *cm, int mi_size) {
cm->mip = vpx_calloc(mi_size, sizeof(*cm->mip));
if (!cm->mip)
return 1;
cm->prev_mip = vpx_calloc(mi_size, sizeof(*cm->prev_mip));
if (!cm->prev_mip)
return 1;
cm->mi_alloc_size = mi_size;
cm->mi_grid_base = (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO*));
if (!cm->mi_grid_base)
return 1;
cm->prev_mi_grid_base = (MODE_INFO **)vpx_calloc(mi_size, sizeof(MODE_INFO*));
if (!cm->prev_mi_grid_base)
return 1;
return 0;
}
static void vp10_enc_free_mi(VP10_COMMON *cm) {
vpx_free(cm->mip);
cm->mip = NULL;
vpx_free(cm->prev_mip);
cm->prev_mip = NULL;
vpx_free(cm->mi_grid_base);
cm->mi_grid_base = NULL;
vpx_free(cm->prev_mi_grid_base);
cm->prev_mi_grid_base = NULL;
}
static void vp10_swap_mi_and_prev_mi(VP10_COMMON *cm) {
// Current mip will be the prev_mip for the next frame.
MODE_INFO **temp_base = cm->prev_mi_grid_base;
MODE_INFO *temp = cm->prev_mip;
cm->prev_mip = cm->mip;
cm->mip = temp;
// Update the upper left visible macroblock ptrs.
cm->mi = cm->mip + cm->mi_stride + 1;
cm->prev_mi = cm->prev_mip + cm->mi_stride + 1;
cm->prev_mi_grid_base = cm->mi_grid_base;
cm->mi_grid_base = temp_base;
cm->mi_grid_visible = cm->mi_grid_base + cm->mi_stride + 1;
cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mi_stride + 1;
}
void vp10_initialize_enc(void) {
static volatile int init_done = 0;
if (!init_done) {
vp10_rtcd();
vpx_dsp_rtcd();
vpx_scale_rtcd();
vp10_init_intra_predictors();
vp10_init_me_luts();
vp10_rc_init_minq_luts();
vp10_entropy_mv_init();
vp10_temporal_filter_init();
vp10_encode_token_init();
init_done = 1;
}
}
static void dealloc_compressor_data(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
vpx_free(cpi->mbmi_ext_base);
cpi->mbmi_ext_base = NULL;
vpx_free(cpi->tile_data);
cpi->tile_data = NULL;
// Delete sementation map
vpx_free(cpi->segmentation_map);
cpi->segmentation_map = NULL;
vpx_free(cpi->coding_context.last_frame_seg_map_copy);
cpi->coding_context.last_frame_seg_map_copy = NULL;
vpx_free(cpi->nmvcosts[0]);
vpx_free(cpi->nmvcosts[1]);
cpi->nmvcosts[0] = NULL;
cpi->nmvcosts[1] = NULL;
vpx_free(cpi->nmvcosts_hp[0]);
vpx_free(cpi->nmvcosts_hp[1]);
cpi->nmvcosts_hp[0] = NULL;
cpi->nmvcosts_hp[1] = NULL;
vpx_free(cpi->nmvsadcosts[0]);
vpx_free(cpi->nmvsadcosts[1]);
cpi->nmvsadcosts[0] = NULL;
cpi->nmvsadcosts[1] = NULL;
vpx_free(cpi->nmvsadcosts_hp[0]);
vpx_free(cpi->nmvsadcosts_hp[1]);
cpi->nmvsadcosts_hp[0] = NULL;
cpi->nmvsadcosts_hp[1] = NULL;
vp10_cyclic_refresh_free(cpi->cyclic_refresh);
cpi->cyclic_refresh = NULL;
vpx_free(cpi->active_map.map);
cpi->active_map.map = NULL;
vp10_free_ref_frame_buffers(cm->buffer_pool);
#if CONFIG_VP9_POSTPROC
vp10_free_postproc_buffers(cm);
#endif
vp10_free_context_buffers(cm);
vpx_free_frame_buffer(&cpi->last_frame_uf);
vpx_free_frame_buffer(&cpi->scaled_source);
vpx_free_frame_buffer(&cpi->scaled_last_source);
vpx_free_frame_buffer(&cpi->alt_ref_buffer);
vp10_lookahead_destroy(cpi->lookahead);
vpx_free(cpi->tile_tok[0][0]);
cpi->tile_tok[0][0] = 0;
vp10_free_pc_tree(&cpi->td);
if (cpi->source_diff_var != NULL) {
vpx_free(cpi->source_diff_var);
cpi->source_diff_var = NULL;
}
}
static void save_coding_context(VP10_COMP *cpi) {
CODING_CONTEXT *const cc = &cpi->coding_context;
VP10_COMMON *cm = &cpi->common;
// Stores a snapshot of key state variables which can subsequently be
// restored with a call to vp10_restore_coding_context. These functions are
// intended for use in a re-code loop in vp10_compress_frame where the
// quantizer value is adjusted between loop iterations.
vp10_copy(cc->nmvjointcost, cpi->td.mb.nmvjointcost);
memcpy(cc->nmvcosts[0], cpi->nmvcosts[0],
MV_VALS * sizeof(*cpi->nmvcosts[0]));
memcpy(cc->nmvcosts[1], cpi->nmvcosts[1],
MV_VALS * sizeof(*cpi->nmvcosts[1]));
memcpy(cc->nmvcosts_hp[0], cpi->nmvcosts_hp[0],
MV_VALS * sizeof(*cpi->nmvcosts_hp[0]));
memcpy(cc->nmvcosts_hp[1], cpi->nmvcosts_hp[1],
MV_VALS * sizeof(*cpi->nmvcosts_hp[1]));
#if !CONFIG_MISC_FIXES
vp10_copy(cc->segment_pred_probs, cm->segp.pred_probs);
#endif
memcpy(cpi->coding_context.last_frame_seg_map_copy,
cm->last_frame_seg_map, (cm->mi_rows * cm->mi_cols));
vp10_copy(cc->last_ref_lf_deltas, cm->lf.last_ref_deltas);
vp10_copy(cc->last_mode_lf_deltas, cm->lf.last_mode_deltas);
cc->fc = *cm->fc;
}
static void restore_coding_context(VP10_COMP *cpi) {
CODING_CONTEXT *const cc = &cpi->coding_context;
VP10_COMMON *cm = &cpi->common;
// Restore key state variables to the snapshot state stored in the
// previous call to vp10_save_coding_context.
vp10_copy(cpi->td.mb.nmvjointcost, cc->nmvjointcost);
memcpy(cpi->nmvcosts[0], cc->nmvcosts[0], MV_VALS * sizeof(*cc->nmvcosts[0]));
memcpy(cpi->nmvcosts[1], cc->nmvcosts[1], MV_VALS * sizeof(*cc->nmvcosts[1]));
memcpy(cpi->nmvcosts_hp[0], cc->nmvcosts_hp[0],
MV_VALS * sizeof(*cc->nmvcosts_hp[0]));
memcpy(cpi->nmvcosts_hp[1], cc->nmvcosts_hp[1],
MV_VALS * sizeof(*cc->nmvcosts_hp[1]));
#if !CONFIG_MISC_FIXES
vp10_copy(cm->segp.pred_probs, cc->segment_pred_probs);
#endif
memcpy(cm->last_frame_seg_map,
cpi->coding_context.last_frame_seg_map_copy,
(cm->mi_rows * cm->mi_cols));
vp10_copy(cm->lf.last_ref_deltas, cc->last_ref_lf_deltas);
vp10_copy(cm->lf.last_mode_deltas, cc->last_mode_lf_deltas);
*cm->fc = cc->fc;
}
static void configure_static_seg_features(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
const RATE_CONTROL *const rc = &cpi->rc;
struct segmentation *const seg = &cm->seg;
int high_q = (int)(rc->avg_q > 48.0);
int qi_delta;
// Disable and clear down for KF
if (cm->frame_type == KEY_FRAME) {
// Clear down the global segmentation map
memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
seg->update_map = 0;
seg->update_data = 0;
cpi->static_mb_pct = 0;
// Disable segmentation
vp10_disable_segmentation(seg);
// Clear down the segment features.
vp10_clearall_segfeatures(seg);
} else if (cpi->refresh_alt_ref_frame) {
// If this is an alt ref frame
// Clear down the global segmentation map
memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
seg->update_map = 0;
seg->update_data = 0;
cpi->static_mb_pct = 0;
// Disable segmentation and individual segment features by default
vp10_disable_segmentation(seg);
vp10_clearall_segfeatures(seg);
// Scan frames from current to arf frame.
// This function re-enables segmentation if appropriate.
vp10_update_mbgraph_stats(cpi);
// If segmentation was enabled set those features needed for the
// arf itself.
if (seg->enabled) {
seg->update_map = 1;
seg->update_data = 1;
qi_delta = vp10_compute_qdelta(rc, rc->avg_q, rc->avg_q * 0.875,
cm->bit_depth);
vp10_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2);
vp10_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
// Where relevant assume segment data is delta data
seg->abs_delta = SEGMENT_DELTADATA;
}
} else if (seg->enabled) {
// All other frames if segmentation has been enabled
// First normal frame in a valid gf or alt ref group
if (rc->frames_since_golden == 0) {
// Set up segment features for normal frames in an arf group
if (rc->source_alt_ref_active) {
seg->update_map = 0;
seg->update_data = 1;
seg->abs_delta = SEGMENT_DELTADATA;
qi_delta = vp10_compute_qdelta(rc, rc->avg_q, rc->avg_q * 1.125,
cm->bit_depth);
vp10_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta + 2);
vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_Q);
vp10_set_segdata(seg, 1, SEG_LVL_ALT_LF, -2);
vp10_enable_segfeature(seg, 1, SEG_LVL_ALT_LF);
// Segment coding disabled for compred testing
if (high_q || (cpi->static_mb_pct == 100)) {
vp10_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
vp10_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
vp10_enable_segfeature(seg, 1, SEG_LVL_SKIP);
}
} else {
// Disable segmentation and clear down features if alt ref
// is not active for this group
vp10_disable_segmentation(seg);
memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols);
seg->update_map = 0;
seg->update_data = 0;
vp10_clearall_segfeatures(seg);
}
} else if (rc->is_src_frame_alt_ref) {
// Special case where we are coding over the top of a previous
// alt ref frame.
// Segment coding disabled for compred testing
// Enable ref frame features for segment 0 as well
vp10_enable_segfeature(seg, 0, SEG_LVL_REF_FRAME);
vp10_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME);
// All mbs should use ALTREF_FRAME
vp10_clear_segdata(seg, 0, SEG_LVL_REF_FRAME);
vp10_set_segdata(seg, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME);
vp10_clear_segdata(seg, 1, SEG_LVL_REF_FRAME);
vp10_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME);
// Skip all MBs if high Q (0,0 mv and skip coeffs)
if (high_q) {
vp10_enable_segfeature(seg, 0, SEG_LVL_SKIP);
vp10_enable_segfeature(seg, 1, SEG_LVL_SKIP);
}
// Enable data update
seg->update_data = 1;
} else {
// All other frames.
// No updates.. leave things as they are.
seg->update_map = 0;
seg->update_data = 0;
}
}
}
static void update_reference_segmentation_map(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
MODE_INFO **mi_8x8_ptr = cm->mi_grid_visible;
uint8_t *cache_ptr = cm->last_frame_seg_map;
int row, col;
for (row = 0; row < cm->mi_rows; row++) {
MODE_INFO **mi_8x8 = mi_8x8_ptr;
uint8_t *cache = cache_ptr;
for (col = 0; col < cm->mi_cols; col++, mi_8x8++, cache++)
cache[0] = mi_8x8[0]->mbmi.segment_id;
mi_8x8_ptr += cm->mi_stride;
cache_ptr += cm->mi_cols;
}
}
static void alloc_raw_frame_buffers(VP10_COMP *cpi) {
VP10_COMMON *cm = &cpi->common;
const VP10EncoderConfig *oxcf = &cpi->oxcf;
if (!cpi->lookahead)
cpi->lookahead = vp10_lookahead_init(oxcf->width, oxcf->height,
cm->subsampling_x, cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif
oxcf->lag_in_frames);
if (!cpi->lookahead)
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate lag buffers");
// TODO(agrange) Check if ARF is enabled and skip allocation if not.
if (vpx_realloc_frame_buffer(&cpi->alt_ref_buffer,
oxcf->width, oxcf->height,
cm->subsampling_x, cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate altref buffer");
}
static void alloc_util_frame_buffers(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
if (vpx_realloc_frame_buffer(&cpi->last_frame_uf,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate last frame buffer");
if (vpx_realloc_frame_buffer(&cpi->scaled_source,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate scaled source buffer");
if (vpx_realloc_frame_buffer(&cpi->scaled_last_source,
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment,
NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate scaled last source buffer");
}
static int alloc_context_buffers_ext(VP10_COMP *cpi) {
VP10_COMMON *cm = &cpi->common;
int mi_size = cm->mi_cols * cm->mi_rows;
cpi->mbmi_ext_base = vpx_calloc(mi_size, sizeof(*cpi->mbmi_ext_base));
if (!cpi->mbmi_ext_base)
return 1;
return 0;
}
void vp10_alloc_compressor_data(VP10_COMP *cpi) {
VP10_COMMON *cm = &cpi->common;
vp10_alloc_context_buffers(cm, cm->width, cm->height);
alloc_context_buffers_ext(cpi);
vpx_free(cpi->tile_tok[0][0]);
{
unsigned int tokens = get_token_alloc(cm->mb_rows, cm->mb_cols);
CHECK_MEM_ERROR(cm, cpi->tile_tok[0][0],
vpx_calloc(tokens, sizeof(*cpi->tile_tok[0][0])));
}
vp10_setup_pc_tree(&cpi->common, &cpi->td);
}
void vp10_new_framerate(VP10_COMP *cpi, double framerate) {
cpi->framerate = framerate < 0.1 ? 30 : framerate;
vp10_rc_update_framerate(cpi);
}
static void set_tile_limits(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
int min_log2_tile_cols, max_log2_tile_cols;
vp10_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
cm->log2_tile_cols = clamp(cpi->oxcf.tile_columns,
min_log2_tile_cols, max_log2_tile_cols);
cm->log2_tile_rows = cpi->oxcf.tile_rows;
}
static void update_frame_size(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
vp10_set_mb_mi(cm, cm->width, cm->height);
vp10_init_context_buffers(cm);
vp10_init_macroblockd(cm, xd, NULL);
memset(cpi->mbmi_ext_base, 0,
cm->mi_rows * cm->mi_cols * sizeof(*cpi->mbmi_ext_base));
set_tile_limits(cpi);
}
static void init_buffer_indices(VP10_COMP *cpi) {
cpi->lst_fb_idx = 0;
cpi->gld_fb_idx = 1;
cpi->alt_fb_idx = 2;
}
static void init_config(struct VP10_COMP *cpi, VP10EncoderConfig *oxcf) {
VP10_COMMON *const cm = &cpi->common;
cpi->oxcf = *oxcf;
cpi->framerate = oxcf->init_framerate;
cm->profile = oxcf->profile;
cm->bit_depth = oxcf->bit_depth;
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth = oxcf->use_highbitdepth;
#endif
cm->color_space = oxcf->color_space;
cm->color_range = oxcf->color_range;
cm->width = oxcf->width;
cm->height = oxcf->height;
vp10_alloc_compressor_data(cpi);
// Single thread case: use counts in common.
cpi->td.counts = &cm->counts;
// change includes all joint functionality
vp10_change_config(cpi, oxcf);
cpi->static_mb_pct = 0;
cpi->ref_frame_flags = 0;
init_buffer_indices(cpi);
}
static void set_rc_buffer_sizes(RATE_CONTROL *rc,
const VP10EncoderConfig *oxcf) {
const int64_t bandwidth = oxcf->target_bandwidth;
const int64_t starting = oxcf->starting_buffer_level_ms;
const int64_t optimal = oxcf->optimal_buffer_level_ms;
const int64_t maximum = oxcf->maximum_buffer_size_ms;
rc->starting_buffer_level = starting * bandwidth / 1000;
rc->optimal_buffer_level = (optimal == 0) ? bandwidth / 8
: optimal * bandwidth / 1000;
rc->maximum_buffer_size = (maximum == 0) ? bandwidth / 8
: maximum * bandwidth / 1000;
}
#if CONFIG_VP9_HIGHBITDEPTH
#define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF) \
cpi->fn_ptr[BT].sdf = SDF; \
cpi->fn_ptr[BT].sdaf = SDAF; \
cpi->fn_ptr[BT].vf = VF; \
cpi->fn_ptr[BT].svf = SVF; \
cpi->fn_ptr[BT].svaf = SVAF; \
cpi->fn_ptr[BT].sdx3f = SDX3F; \
cpi->fn_ptr[BT].sdx8f = SDX8F; \
cpi->fn_ptr[BT].sdx4df = SDX4DF;
#define MAKE_BFP_SAD_WRAPPER(fnname) \
static unsigned int fnname##_bits8(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride) { \
return fnname(src_ptr, source_stride, ref_ptr, ref_stride); \
} \
static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride) { \
return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 2; \
} \
static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride) { \
return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 4; \
}
#define MAKE_BFP_SADAVG_WRAPPER(fnname) static unsigned int \
fnname##_bits8(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
const uint8_t *second_pred) { \
return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred); \
} \
static unsigned int fnname##_bits10(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
const uint8_t *second_pred) { \
return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
second_pred) >> 2; \
} \
static unsigned int fnname##_bits12(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
const uint8_t *second_pred) { \
return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \
second_pred) >> 4; \
}
#define MAKE_BFP_SAD3_WRAPPER(fnname) \
static void fnname##_bits8(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
unsigned int *sad_array) { \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
} \
static void fnname##_bits10(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
unsigned int *sad_array) { \
int i; \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
for (i = 0; i < 3; i++) \
sad_array[i] >>= 2; \
} \
static void fnname##_bits12(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
unsigned int *sad_array) { \
int i; \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
for (i = 0; i < 3; i++) \
sad_array[i] >>= 4; \
}
#define MAKE_BFP_SAD8_WRAPPER(fnname) \
static void fnname##_bits8(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
unsigned int *sad_array) { \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
} \
static void fnname##_bits10(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
unsigned int *sad_array) { \
int i; \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
for (i = 0; i < 8; i++) \
sad_array[i] >>= 2; \
} \
static void fnname##_bits12(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t *ref_ptr, \
int ref_stride, \
unsigned int *sad_array) { \
int i; \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
for (i = 0; i < 8; i++) \
sad_array[i] >>= 4; \
}
#define MAKE_BFP_SAD4D_WRAPPER(fnname) \
static void fnname##_bits8(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t* const ref_ptr[], \
int ref_stride, \
unsigned int *sad_array) { \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
} \
static void fnname##_bits10(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t* const ref_ptr[], \
int ref_stride, \
unsigned int *sad_array) { \
int i; \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
for (i = 0; i < 4; i++) \
sad_array[i] >>= 2; \
} \
static void fnname##_bits12(const uint8_t *src_ptr, \
int source_stride, \
const uint8_t* const ref_ptr[], \
int ref_stride, \
unsigned int *sad_array) { \
int i; \
fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \
for (i = 0; i < 4; i++) \
sad_array[i] >>= 4; \
}
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x16)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x16_avg)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x16x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x32)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x32_avg)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x32x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad64x32)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad64x32_avg)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad64x32x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x64)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x64_avg)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x64x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad32x32)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad32x32_avg)
MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad32x32x3)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad32x32x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad32x32x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad64x64)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad64x64_avg)
MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad64x64x3)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad64x64x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad64x64x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x16)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x16_avg)
MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x16x3)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x16x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x16x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad16x8)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad16x8_avg)
MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad16x8x3)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad16x8x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad16x8x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x16)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x16_avg)
MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x16x3)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x16x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x16x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x8)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x8_avg)
MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad8x8x3)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x8x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x8x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad8x4)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad8x4_avg)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad8x4x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad8x4x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x8)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x8_avg)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x8x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x8x4d)
MAKE_BFP_SAD_WRAPPER(vpx_highbd_sad4x4)
MAKE_BFP_SADAVG_WRAPPER(vpx_highbd_sad4x4_avg)
MAKE_BFP_SAD3_WRAPPER(vpx_highbd_sad4x4x3)
MAKE_BFP_SAD8_WRAPPER(vpx_highbd_sad4x4x8)
MAKE_BFP_SAD4D_WRAPPER(vpx_highbd_sad4x4x4d)
static void highbd_set_var_fns(VP10_COMP *const cpi) {
VP10_COMMON *const cm = &cpi->common;
if (cm->use_highbitdepth) {
switch (cm->bit_depth) {
case VPX_BITS_8:
HIGHBD_BFP(BLOCK_32X16,
vpx_highbd_sad32x16_bits8,
vpx_highbd_sad32x16_avg_bits8,
vpx_highbd_8_variance32x16,
vpx_highbd_8_sub_pixel_variance32x16,
vpx_highbd_8_sub_pixel_avg_variance32x16,
NULL,
NULL,
vpx_highbd_sad32x16x4d_bits8)
HIGHBD_BFP(BLOCK_16X32,
vpx_highbd_sad16x32_bits8,
vpx_highbd_sad16x32_avg_bits8,
vpx_highbd_8_variance16x32,
vpx_highbd_8_sub_pixel_variance16x32,
vpx_highbd_8_sub_pixel_avg_variance16x32,
NULL,
NULL,
vpx_highbd_sad16x32x4d_bits8)
HIGHBD_BFP(BLOCK_64X32,
vpx_highbd_sad64x32_bits8,
vpx_highbd_sad64x32_avg_bits8,
vpx_highbd_8_variance64x32,
vpx_highbd_8_sub_pixel_variance64x32,
vpx_highbd_8_sub_pixel_avg_variance64x32,
NULL,
NULL,
vpx_highbd_sad64x32x4d_bits8)
HIGHBD_BFP(BLOCK_32X64,
vpx_highbd_sad32x64_bits8,
vpx_highbd_sad32x64_avg_bits8,
vpx_highbd_8_variance32x64,
vpx_highbd_8_sub_pixel_variance32x64,
vpx_highbd_8_sub_pixel_avg_variance32x64,
NULL,
NULL,
vpx_highbd_sad32x64x4d_bits8)
HIGHBD_BFP(BLOCK_32X32,
vpx_highbd_sad32x32_bits8,
vpx_highbd_sad32x32_avg_bits8,
vpx_highbd_8_variance32x32,
vpx_highbd_8_sub_pixel_variance32x32,
vpx_highbd_8_sub_pixel_avg_variance32x32,
vpx_highbd_sad32x32x3_bits8,
vpx_highbd_sad32x32x8_bits8,
vpx_highbd_sad32x32x4d_bits8)
HIGHBD_BFP(BLOCK_64X64,
vpx_highbd_sad64x64_bits8,
vpx_highbd_sad64x64_avg_bits8,
vpx_highbd_8_variance64x64,
vpx_highbd_8_sub_pixel_variance64x64,
vpx_highbd_8_sub_pixel_avg_variance64x64,
vpx_highbd_sad64x64x3_bits8,
vpx_highbd_sad64x64x8_bits8,
vpx_highbd_sad64x64x4d_bits8)
HIGHBD_BFP(BLOCK_16X16,
vpx_highbd_sad16x16_bits8,
vpx_highbd_sad16x16_avg_bits8,
vpx_highbd_8_variance16x16,
vpx_highbd_8_sub_pixel_variance16x16,
vpx_highbd_8_sub_pixel_avg_variance16x16,
vpx_highbd_sad16x16x3_bits8,
vpx_highbd_sad16x16x8_bits8,
vpx_highbd_sad16x16x4d_bits8)
HIGHBD_BFP(BLOCK_16X8,
vpx_highbd_sad16x8_bits8,
vpx_highbd_sad16x8_avg_bits8,
vpx_highbd_8_variance16x8,
vpx_highbd_8_sub_pixel_variance16x8,
vpx_highbd_8_sub_pixel_avg_variance16x8,
vpx_highbd_sad16x8x3_bits8,
vpx_highbd_sad16x8x8_bits8,
vpx_highbd_sad16x8x4d_bits8)
HIGHBD_BFP(BLOCK_8X16,
vpx_highbd_sad8x16_bits8,
vpx_highbd_sad8x16_avg_bits8,
vpx_highbd_8_variance8x16,
vpx_highbd_8_sub_pixel_variance8x16,
vpx_highbd_8_sub_pixel_avg_variance8x16,
vpx_highbd_sad8x16x3_bits8,
vpx_highbd_sad8x16x8_bits8,
vpx_highbd_sad8x16x4d_bits8)
HIGHBD_BFP(BLOCK_8X8,
vpx_highbd_sad8x8_bits8,
vpx_highbd_sad8x8_avg_bits8,
vpx_highbd_8_variance8x8,
vpx_highbd_8_sub_pixel_variance8x8,
vpx_highbd_8_sub_pixel_avg_variance8x8,
vpx_highbd_sad8x8x3_bits8,
vpx_highbd_sad8x8x8_bits8,
vpx_highbd_sad8x8x4d_bits8)
HIGHBD_BFP(BLOCK_8X4,
vpx_highbd_sad8x4_bits8,
vpx_highbd_sad8x4_avg_bits8,
vpx_highbd_8_variance8x4,
vpx_highbd_8_sub_pixel_variance8x4,
vpx_highbd_8_sub_pixel_avg_variance8x4,
NULL,
vpx_highbd_sad8x4x8_bits8,
vpx_highbd_sad8x4x4d_bits8)
HIGHBD_BFP(BLOCK_4X8,
vpx_highbd_sad4x8_bits8,
vpx_highbd_sad4x8_avg_bits8,
vpx_highbd_8_variance4x8,
vpx_highbd_8_sub_pixel_variance4x8,
vpx_highbd_8_sub_pixel_avg_variance4x8,
NULL,
vpx_highbd_sad4x8x8_bits8,
vpx_highbd_sad4x8x4d_bits8)
HIGHBD_BFP(BLOCK_4X4,
vpx_highbd_sad4x4_bits8,
vpx_highbd_sad4x4_avg_bits8,
vpx_highbd_8_variance4x4,
vpx_highbd_8_sub_pixel_variance4x4,
vpx_highbd_8_sub_pixel_avg_variance4x4,
vpx_highbd_sad4x4x3_bits8,
vpx_highbd_sad4x4x8_bits8,
vpx_highbd_sad4x4x4d_bits8)
break;
case VPX_BITS_10:
HIGHBD_BFP(BLOCK_32X16,
vpx_highbd_sad32x16_bits10,
vpx_highbd_sad32x16_avg_bits10,
vpx_highbd_10_variance32x16,
vpx_highbd_10_sub_pixel_variance32x16,
vpx_highbd_10_sub_pixel_avg_variance32x16,
NULL,
NULL,
vpx_highbd_sad32x16x4d_bits10)
HIGHBD_BFP(BLOCK_16X32,
vpx_highbd_sad16x32_bits10,
vpx_highbd_sad16x32_avg_bits10,
vpx_highbd_10_variance16x32,
vpx_highbd_10_sub_pixel_variance16x32,
vpx_highbd_10_sub_pixel_avg_variance16x32,
NULL,
NULL,
vpx_highbd_sad16x32x4d_bits10)
HIGHBD_BFP(BLOCK_64X32,
vpx_highbd_sad64x32_bits10,
vpx_highbd_sad64x32_avg_bits10,
vpx_highbd_10_variance64x32,
vpx_highbd_10_sub_pixel_variance64x32,
vpx_highbd_10_sub_pixel_avg_variance64x32,
NULL,
NULL,
vpx_highbd_sad64x32x4d_bits10)
HIGHBD_BFP(BLOCK_32X64,
vpx_highbd_sad32x64_bits10,
vpx_highbd_sad32x64_avg_bits10,
vpx_highbd_10_variance32x64,
vpx_highbd_10_sub_pixel_variance32x64,
vpx_highbd_10_sub_pixel_avg_variance32x64,
NULL,
NULL,
vpx_highbd_sad32x64x4d_bits10)
HIGHBD_BFP(BLOCK_32X32,
vpx_highbd_sad32x32_bits10,
vpx_highbd_sad32x32_avg_bits10,
vpx_highbd_10_variance32x32,
vpx_highbd_10_sub_pixel_variance32x32,
vpx_highbd_10_sub_pixel_avg_variance32x32,
vpx_highbd_sad32x32x3_bits10,
vpx_highbd_sad32x32x8_bits10,
vpx_highbd_sad32x32x4d_bits10)
HIGHBD_BFP(BLOCK_64X64,
vpx_highbd_sad64x64_bits10,
vpx_highbd_sad64x64_avg_bits10,
vpx_highbd_10_variance64x64,
vpx_highbd_10_sub_pixel_variance64x64,
vpx_highbd_10_sub_pixel_avg_variance64x64,
vpx_highbd_sad64x64x3_bits10,
vpx_highbd_sad64x64x8_bits10,
vpx_highbd_sad64x64x4d_bits10)
HIGHBD_BFP(BLOCK_16X16,
vpx_highbd_sad16x16_bits10,
vpx_highbd_sad16x16_avg_bits10,
vpx_highbd_10_variance16x16,
vpx_highbd_10_sub_pixel_variance16x16,
vpx_highbd_10_sub_pixel_avg_variance16x16,
vpx_highbd_sad16x16x3_bits10,
vpx_highbd_sad16x16x8_bits10,
vpx_highbd_sad16x16x4d_bits10)
HIGHBD_BFP(BLOCK_16X8,
vpx_highbd_sad16x8_bits10,
vpx_highbd_sad16x8_avg_bits10,
vpx_highbd_10_variance16x8,
vpx_highbd_10_sub_pixel_variance16x8,
vpx_highbd_10_sub_pixel_avg_variance16x8,
vpx_highbd_sad16x8x3_bits10,
vpx_highbd_sad16x8x8_bits10,
vpx_highbd_sad16x8x4d_bits10)
HIGHBD_BFP(BLOCK_8X16,
vpx_highbd_sad8x16_bits10,
vpx_highbd_sad8x16_avg_bits10,
vpx_highbd_10_variance8x16,
vpx_highbd_10_sub_pixel_variance8x16,
vpx_highbd_10_sub_pixel_avg_variance8x16,
vpx_highbd_sad8x16x3_bits10,
vpx_highbd_sad8x16x8_bits10,
vpx_highbd_sad8x16x4d_bits10)
HIGHBD_BFP(BLOCK_8X8,
vpx_highbd_sad8x8_bits10,
vpx_highbd_sad8x8_avg_bits10,
vpx_highbd_10_variance8x8,
vpx_highbd_10_sub_pixel_variance8x8,
vpx_highbd_10_sub_pixel_avg_variance8x8,
vpx_highbd_sad8x8x3_bits10,
vpx_highbd_sad8x8x8_bits10,
vpx_highbd_sad8x8x4d_bits10)
HIGHBD_BFP(BLOCK_8X4,
vpx_highbd_sad8x4_bits10,
vpx_highbd_sad8x4_avg_bits10,
vpx_highbd_10_variance8x4,
vpx_highbd_10_sub_pixel_variance8x4,
vpx_highbd_10_sub_pixel_avg_variance8x4,
NULL,
vpx_highbd_sad8x4x8_bits10,
vpx_highbd_sad8x4x4d_bits10)
HIGHBD_BFP(BLOCK_4X8,
vpx_highbd_sad4x8_bits10,
vpx_highbd_sad4x8_avg_bits10,
vpx_highbd_10_variance4x8,
vpx_highbd_10_sub_pixel_variance4x8,
vpx_highbd_10_sub_pixel_avg_variance4x8,
NULL,
vpx_highbd_sad4x8x8_bits10,
vpx_highbd_sad4x8x4d_bits10)
HIGHBD_BFP(BLOCK_4X4,
vpx_highbd_sad4x4_bits10,
vpx_highbd_sad4x4_avg_bits10,
vpx_highbd_10_variance4x4,
vpx_highbd_10_sub_pixel_variance4x4,
vpx_highbd_10_sub_pixel_avg_variance4x4,
vpx_highbd_sad4x4x3_bits10,
vpx_highbd_sad4x4x8_bits10,
vpx_highbd_sad4x4x4d_bits10)
break;
case VPX_BITS_12:
HIGHBD_BFP(BLOCK_32X16,
vpx_highbd_sad32x16_bits12,
vpx_highbd_sad32x16_avg_bits12,
vpx_highbd_12_variance32x16,
vpx_highbd_12_sub_pixel_variance32x16,
vpx_highbd_12_sub_pixel_avg_variance32x16,
NULL,
NULL,
vpx_highbd_sad32x16x4d_bits12)
HIGHBD_BFP(BLOCK_16X32,
vpx_highbd_sad16x32_bits12,
vpx_highbd_sad16x32_avg_bits12,
vpx_highbd_12_variance16x32,
vpx_highbd_12_sub_pixel_variance16x32,
vpx_highbd_12_sub_pixel_avg_variance16x32,
NULL,
NULL,
vpx_highbd_sad16x32x4d_bits12)
HIGHBD_BFP(BLOCK_64X32,
vpx_highbd_sad64x32_bits12,
vpx_highbd_sad64x32_avg_bits12,
vpx_highbd_12_variance64x32,
vpx_highbd_12_sub_pixel_variance64x32,
vpx_highbd_12_sub_pixel_avg_variance64x32,
NULL,
NULL,
vpx_highbd_sad64x32x4d_bits12)
HIGHBD_BFP(BLOCK_32X64,
vpx_highbd_sad32x64_bits12,
vpx_highbd_sad32x64_avg_bits12,
vpx_highbd_12_variance32x64,
vpx_highbd_12_sub_pixel_variance32x64,
vpx_highbd_12_sub_pixel_avg_variance32x64,
NULL,
NULL,
vpx_highbd_sad32x64x4d_bits12)
HIGHBD_BFP(BLOCK_32X32,
vpx_highbd_sad32x32_bits12,
vpx_highbd_sad32x32_avg_bits12,
vpx_highbd_12_variance32x32,
vpx_highbd_12_sub_pixel_variance32x32,
vpx_highbd_12_sub_pixel_avg_variance32x32,
vpx_highbd_sad32x32x3_bits12,
vpx_highbd_sad32x32x8_bits12,
vpx_highbd_sad32x32x4d_bits12)
HIGHBD_BFP(BLOCK_64X64,
vpx_highbd_sad64x64_bits12,
vpx_highbd_sad64x64_avg_bits12,
vpx_highbd_12_variance64x64,
vpx_highbd_12_sub_pixel_variance64x64,
vpx_highbd_12_sub_pixel_avg_variance64x64,
vpx_highbd_sad64x64x3_bits12,
vpx_highbd_sad64x64x8_bits12,
vpx_highbd_sad64x64x4d_bits12)
HIGHBD_BFP(BLOCK_16X16,
vpx_highbd_sad16x16_bits12,
vpx_highbd_sad16x16_avg_bits12,
vpx_highbd_12_variance16x16,
vpx_highbd_12_sub_pixel_variance16x16,
vpx_highbd_12_sub_pixel_avg_variance16x16,
vpx_highbd_sad16x16x3_bits12,
vpx_highbd_sad16x16x8_bits12,
vpx_highbd_sad16x16x4d_bits12)
HIGHBD_BFP(BLOCK_16X8,
vpx_highbd_sad16x8_bits12,
vpx_highbd_sad16x8_avg_bits12,
vpx_highbd_12_variance16x8,
vpx_highbd_12_sub_pixel_variance16x8,
vpx_highbd_12_sub_pixel_avg_variance16x8,
vpx_highbd_sad16x8x3_bits12,
vpx_highbd_sad16x8x8_bits12,
vpx_highbd_sad16x8x4d_bits12)
HIGHBD_BFP(BLOCK_8X16,
vpx_highbd_sad8x16_bits12,
vpx_highbd_sad8x16_avg_bits12,
vpx_highbd_12_variance8x16,
vpx_highbd_12_sub_pixel_variance8x16,
vpx_highbd_12_sub_pixel_avg_variance8x16,
vpx_highbd_sad8x16x3_bits12,
vpx_highbd_sad8x16x8_bits12,
vpx_highbd_sad8x16x4d_bits12)
HIGHBD_BFP(BLOCK_8X8,
vpx_highbd_sad8x8_bits12,
vpx_highbd_sad8x8_avg_bits12,
vpx_highbd_12_variance8x8,
vpx_highbd_12_sub_pixel_variance8x8,
vpx_highbd_12_sub_pixel_avg_variance8x8,
vpx_highbd_sad8x8x3_bits12,
vpx_highbd_sad8x8x8_bits12,
vpx_highbd_sad8x8x4d_bits12)
HIGHBD_BFP(BLOCK_8X4,
vpx_highbd_sad8x4_bits12,
vpx_highbd_sad8x4_avg_bits12,
vpx_highbd_12_variance8x4,
vpx_highbd_12_sub_pixel_variance8x4,
vpx_highbd_12_sub_pixel_avg_variance8x4,
NULL,
vpx_highbd_sad8x4x8_bits12,
vpx_highbd_sad8x4x4d_bits12)
HIGHBD_BFP(BLOCK_4X8,
vpx_highbd_sad4x8_bits12,
vpx_highbd_sad4x8_avg_bits12,
vpx_highbd_12_variance4x8,
vpx_highbd_12_sub_pixel_variance4x8,
vpx_highbd_12_sub_pixel_avg_variance4x8,
NULL,
vpx_highbd_sad4x8x8_bits12,
vpx_highbd_sad4x8x4d_bits12)
HIGHBD_BFP(BLOCK_4X4,
vpx_highbd_sad4x4_bits12,
vpx_highbd_sad4x4_avg_bits12,
vpx_highbd_12_variance4x4,
vpx_highbd_12_sub_pixel_variance4x4,
vpx_highbd_12_sub_pixel_avg_variance4x4,
vpx_highbd_sad4x4x3_bits12,
vpx_highbd_sad4x4x8_bits12,
vpx_highbd_sad4x4x4d_bits12)
break;
default:
assert(0 && "cm->bit_depth should be VPX_BITS_8, "
"VPX_BITS_10 or VPX_BITS_12");
}
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
static void realloc_segmentation_maps(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
// Create the encoder segmentation map and set all entries to 0
vpx_free(cpi->segmentation_map);
CHECK_MEM_ERROR(cm, cpi->segmentation_map,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
// Create a map used for cyclic background refresh.
if (cpi->cyclic_refresh)
vp10_cyclic_refresh_free(cpi->cyclic_refresh);
CHECK_MEM_ERROR(cm, cpi->cyclic_refresh,
vp10_cyclic_refresh_alloc(cm->mi_rows, cm->mi_cols));
// Create a map used to mark inactive areas.
vpx_free(cpi->active_map.map);
CHECK_MEM_ERROR(cm, cpi->active_map.map,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
// And a place holder structure is the coding context
// for use if we want to save and restore it
vpx_free(cpi->coding_context.last_frame_seg_map_copy);
CHECK_MEM_ERROR(cm, cpi->coding_context.last_frame_seg_map_copy,
vpx_calloc(cm->mi_rows * cm->mi_cols, 1));
}
void vp10_change_config(struct VP10_COMP *cpi, const VP10EncoderConfig *oxcf) {
VP10_COMMON *const cm = &cpi->common;
RATE_CONTROL *const rc = &cpi->rc;
if (cm->profile != oxcf->profile)
cm->profile = oxcf->profile;
cm->bit_depth = oxcf->bit_depth;
cm->color_space = oxcf->color_space;
cm->color_range = oxcf->color_range;
if (cm->profile <= PROFILE_1)
assert(cm->bit_depth == VPX_BITS_8);
else
assert(cm->bit_depth > VPX_BITS_8);
cpi->oxcf = *oxcf;
#if CONFIG_VP9_HIGHBITDEPTH
cpi->td.mb.e_mbd.bd = (int)cm->bit_depth;
#endif // CONFIG_VP9_HIGHBITDEPTH
if ((oxcf->pass == 0) && (oxcf->rc_mode == VPX_Q)) {
rc->baseline_gf_interval = FIXED_GF_INTERVAL;
} else {
rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2;
}
cpi->refresh_golden_frame = 0;
cpi->refresh_last_frame = 1;
cm->refresh_frame_context =
oxcf->error_resilient_mode ? REFRESH_FRAME_CONTEXT_OFF :
oxcf->frame_parallel_decoding_mode ? REFRESH_FRAME_CONTEXT_FORWARD
: REFRESH_FRAME_CONTEXT_BACKWARD;
cm->reset_frame_context = RESET_FRAME_CONTEXT_NONE;
vp10_reset_segment_features(cm);
vp10_set_high_precision_mv(cpi, 0);
{
int i;
for (i = 0; i < MAX_SEGMENTS; i++)
cpi->segment_encode_breakout[i] = cpi->oxcf.encode_breakout;
}
cpi->encode_breakout = cpi->oxcf.encode_breakout;
set_rc_buffer_sizes(rc, &cpi->oxcf);
// Under a configuration change, where maximum_buffer_size may change,
// keep buffer level clipped to the maximum allowed buffer size.
rc->bits_off_target = VPXMIN(rc->bits_off_target, rc->maximum_buffer_size);
rc->buffer_level = VPXMIN(rc->buffer_level, rc->maximum_buffer_size);
// Set up frame rate and related parameters rate control values.
vp10_new_framerate(cpi, cpi->framerate);
// Set absolute upper and lower quality limits
rc->worst_quality = cpi->oxcf.worst_allowed_q;
rc->best_quality = cpi->oxcf.best_allowed_q;
cm->interp_filter = cpi->sf.default_interp_filter;
if (cpi->oxcf.render_width > 0 && cpi->oxcf.render_height > 0) {
cm->render_width = cpi->oxcf.render_width;
cm->render_height = cpi->oxcf.render_height;
} else {
cm->render_width = cpi->oxcf.width;
cm->render_height = cpi->oxcf.height;
}
cm->width = cpi->oxcf.width;
cm->height = cpi->oxcf.height;
if (cpi->initial_width) {
if (cm->width > cpi->initial_width || cm->height > cpi->initial_height) {
vp10_free_context_buffers(cm);
vp10_alloc_compressor_data(cpi);
realloc_segmentation_maps(cpi);
cpi->initial_width = cpi->initial_height = 0;
}
}
update_frame_size(cpi);
cpi->alt_ref_source = NULL;
rc->is_src_frame_alt_ref = 0;
#if 0
// Experimental RD Code
cpi->frame_distortion = 0;
cpi->last_frame_distortion = 0;
#endif
set_tile_limits(cpi);
cpi->ext_refresh_frame_flags_pending = 0;
cpi->ext_refresh_frame_context_pending = 0;
#if CONFIG_VP9_HIGHBITDEPTH
highbd_set_var_fns(cpi);
#endif
}
#ifndef M_LOG2_E
#define M_LOG2_E 0.693147180559945309417
#endif
#define log2f(x) (log (x) / (float) M_LOG2_E)
static void cal_nmvjointsadcost(int *mvjointsadcost) {
mvjointsadcost[0] = 600;
mvjointsadcost[1] = 300;
mvjointsadcost[2] = 300;
mvjointsadcost[3] = 300;
}
static void cal_nmvsadcosts(int *mvsadcost[2]) {
int i = 1;
mvsadcost[0][0] = 0;
mvsadcost[1][0] = 0;
do {
double z = 256 * (2 * (log2f(8 * i) + .6));
mvsadcost[0][i] = (int)z;
mvsadcost[1][i] = (int)z;
mvsadcost[0][-i] = (int)z;
mvsadcost[1][-i] = (int)z;
} while (++i <= MV_MAX);
}
static void cal_nmvsadcosts_hp(int *mvsadcost[2]) {
int i = 1;
mvsadcost[0][0] = 0;
mvsadcost[1][0] = 0;
do {
double z = 256 * (2 * (log2f(8 * i) + .6));
mvsadcost[0][i] = (int)z;
mvsadcost[1][i] = (int)z;
mvsadcost[0][-i] = (int)z;
mvsadcost[1][-i] = (int)z;
} while (++i <= MV_MAX);
}
VP10_COMP *vp10_create_compressor(VP10EncoderConfig *oxcf,
BufferPool *const pool) {
unsigned int i;
VP10_COMP *volatile const cpi = vpx_memalign(32, sizeof(VP10_COMP));
VP10_COMMON *volatile const cm = cpi != NULL ? &cpi->common : NULL;
if (!cm)
return NULL;
vp10_zero(*cpi);
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp10_remove_compressor(cpi);
return 0;
}
cm->error.setjmp = 1;
cm->alloc_mi = vp10_enc_alloc_mi;
cm->free_mi = vp10_enc_free_mi;
cm->setup_mi = vp10_enc_setup_mi;
CHECK_MEM_ERROR(cm, cm->fc,
(FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
CHECK_MEM_ERROR(cm, cm->frame_contexts,
(FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
sizeof(*cm->frame_contexts)));
cpi->resize_state = 0;
cpi->resize_avg_qp = 0;
cpi->resize_buffer_underflow = 0;
cpi->common.buffer_pool = pool;
init_config(cpi, oxcf);
vp10_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc);
cm->current_video_frame = 0;
cpi->partition_search_skippable_frame = 0;
cpi->tile_data = NULL;
realloc_segmentation_maps(cpi);
CHECK_MEM_ERROR(cm, cpi->nmvcosts[0],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[0])));
CHECK_MEM_ERROR(cm, cpi->nmvcosts[1],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts[1])));
CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[0],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[0])));
CHECK_MEM_ERROR(cm, cpi->nmvcosts_hp[1],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvcosts_hp[1])));
CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[0],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[0])));
CHECK_MEM_ERROR(cm, cpi->nmvsadcosts[1],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts[1])));
CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[0],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[0])));
CHECK_MEM_ERROR(cm, cpi->nmvsadcosts_hp[1],
vpx_calloc(MV_VALS, sizeof(*cpi->nmvsadcosts_hp[1])));
for (i = 0; i < (sizeof(cpi->mbgraph_stats) /
sizeof(cpi->mbgraph_stats[0])); i++) {
CHECK_MEM_ERROR(cm, cpi->mbgraph_stats[i].mb_stats,
vpx_calloc(cm->MBs *
sizeof(*cpi->mbgraph_stats[i].mb_stats), 1));
}
#if CONFIG_FP_MB_STATS
cpi->use_fp_mb_stats = 0;
if (cpi->use_fp_mb_stats) {
// a place holder used to store the first pass mb stats in the first pass
CHECK_MEM_ERROR(cm, cpi->twopass.frame_mb_stats_buf,
vpx_calloc(cm->MBs * sizeof(uint8_t), 1));
} else {
cpi->twopass.frame_mb_stats_buf = NULL;
}
#endif
cpi->refresh_alt_ref_frame = 0;
cpi->multi_arf_last_grp_enabled = 0;
cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS;
#if CONFIG_INTERNAL_STATS
cpi->b_calculate_ssimg = 0;
cpi->b_calculate_blockiness = 1;
cpi->b_calculate_consistency = 1;
cpi->total_inconsistency = 0;
cpi->psnr.worst = 100.0;
cpi->worst_ssim = 100.0;
cpi->count = 0;
cpi->bytes = 0;
if (cpi->b_calculate_psnr) {
cpi->total_sq_error = 0;
cpi->total_samples = 0;
cpi->totalp_sq_error = 0;
cpi->totalp_samples = 0;
cpi->tot_recode_hits = 0;
cpi->summed_quality = 0;
cpi->summed_weights = 0;
cpi->summedp_quality = 0;
cpi->summedp_weights = 0;
}
if (cpi->b_calculate_ssimg) {
cpi->ssimg.worst= 100.0;
}
cpi->fastssim.worst = 100.0;
cpi->psnrhvs.worst = 100.0;
if (cpi->b_calculate_blockiness) {
cpi->total_blockiness = 0;
cpi->worst_blockiness = 0.0;
}
if (cpi->b_calculate_consistency) {
CHECK_MEM_ERROR(cm, cpi->ssim_vars,
vpx_malloc(sizeof(*cpi->ssim_vars) * 4 *
cpi->common.mi_rows * cpi->common.mi_cols));
cpi->worst_consistency = 100.0;
}
#endif
cpi->first_time_stamp_ever = INT64_MAX;
cal_nmvjointsadcost(cpi->td.mb.nmvjointsadcost);
cpi->td.mb.nmvcost[0] = &cpi->nmvcosts[0][MV_MAX];
cpi->td.mb.nmvcost[1] = &cpi->nmvcosts[1][MV_MAX];
cpi->td.mb.nmvsadcost[0] = &cpi->nmvsadcosts[0][MV_MAX];
cpi->td.mb.nmvsadcost[1] = &cpi->nmvsadcosts[1][MV_MAX];
cal_nmvsadcosts(cpi->td.mb.nmvsadcost);
cpi->td.mb.nmvcost_hp[0] = &cpi->nmvcosts_hp[0][MV_MAX];
cpi->td.mb.nmvcost_hp[1] = &cpi->nmvcosts_hp[1][MV_MAX];
cpi->td.mb.nmvsadcost_hp[0] = &cpi->nmvsadcosts_hp[0][MV_MAX];
cpi->td.mb.nmvsadcost_hp[1] = &cpi->nmvsadcosts_hp[1][MV_MAX];
cal_nmvsadcosts_hp(cpi->td.mb.nmvsadcost_hp);
#if CONFIG_VP9_TEMPORAL_DENOISING
#ifdef OUTPUT_YUV_DENOISED
yuv_denoised_file = fopen("denoised.yuv", "ab");
#endif
#endif
#ifdef OUTPUT_YUV_SKINMAP
yuv_skinmap_file = fopen("skinmap.yuv", "ab");
#endif
#ifdef OUTPUT_YUV_REC
yuv_rec_file = fopen("rec.yuv", "wb");
#endif
#if 0
framepsnr = fopen("framepsnr.stt", "a");
kf_list = fopen("kf_list.stt", "w");
#endif
cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
if (oxcf->pass == 1) {
vp10_init_first_pass(cpi);
} else if (oxcf->pass == 2) {
const size_t packet_sz = sizeof(FIRSTPASS_STATS);
const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz);
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
const size_t psz = cpi->common.MBs * sizeof(uint8_t);
const int ps = (int)(oxcf->firstpass_mb_stats_in.sz / psz);
cpi->twopass.firstpass_mb_stats.mb_stats_start =
oxcf->firstpass_mb_stats_in.buf;
cpi->twopass.firstpass_mb_stats.mb_stats_end =
cpi->twopass.firstpass_mb_stats.mb_stats_start +
(ps - 1) * cpi->common.MBs * sizeof(uint8_t);
}
#endif
cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf;
cpi->twopass.stats_in = cpi->twopass.stats_in_start;
cpi->twopass.stats_in_end = &cpi->twopass.stats_in[packets - 1];
vp10_init_second_pass(cpi);
}
vp10_set_speed_features_framesize_independent(cpi);
vp10_set_speed_features_framesize_dependent(cpi);
// Allocate memory to store variances for a frame.
CHECK_MEM_ERROR(cm, cpi->source_diff_var,
vpx_calloc(cm->MBs, sizeof(diff)));
cpi->source_var_thresh = 0;
cpi->frames_till_next_var_check = 0;
#define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX3F, SDX8F, SDX4DF)\
cpi->fn_ptr[BT].sdf = SDF; \
cpi->fn_ptr[BT].sdaf = SDAF; \
cpi->fn_ptr[BT].vf = VF; \
cpi->fn_ptr[BT].svf = SVF; \
cpi->fn_ptr[BT].svaf = SVAF; \
cpi->fn_ptr[BT].sdx3f = SDX3F; \
cpi->fn_ptr[BT].sdx8f = SDX8F; \
cpi->fn_ptr[BT].sdx4df = SDX4DF;
BFP(BLOCK_32X16, vpx_sad32x16, vpx_sad32x16_avg,
vpx_variance32x16, vpx_sub_pixel_variance32x16,
vpx_sub_pixel_avg_variance32x16, NULL, NULL, vpx_sad32x16x4d)
BFP(BLOCK_16X32, vpx_sad16x32, vpx_sad16x32_avg,
vpx_variance16x32, vpx_sub_pixel_variance16x32,
vpx_sub_pixel_avg_variance16x32, NULL, NULL, vpx_sad16x32x4d)
BFP(BLOCK_64X32, vpx_sad64x32, vpx_sad64x32_avg,
vpx_variance64x32, vpx_sub_pixel_variance64x32,
vpx_sub_pixel_avg_variance64x32, NULL, NULL, vpx_sad64x32x4d)
BFP(BLOCK_32X64, vpx_sad32x64, vpx_sad32x64_avg,
vpx_variance32x64, vpx_sub_pixel_variance32x64,
vpx_sub_pixel_avg_variance32x64, NULL, NULL, vpx_sad32x64x4d)
BFP(BLOCK_32X32, vpx_sad32x32, vpx_sad32x32_avg,
vpx_variance32x32, vpx_sub_pixel_variance32x32,
vpx_sub_pixel_avg_variance32x32, vpx_sad32x32x3, vpx_sad32x32x8,
vpx_sad32x32x4d)
BFP(BLOCK_64X64, vpx_sad64x64, vpx_sad64x64_avg,
vpx_variance64x64, vpx_sub_pixel_variance64x64,
vpx_sub_pixel_avg_variance64x64, vpx_sad64x64x3, vpx_sad64x64x8,
vpx_sad64x64x4d)
BFP(BLOCK_16X16, vpx_sad16x16, vpx_sad16x16_avg,
vpx_variance16x16, vpx_sub_pixel_variance16x16,
vpx_sub_pixel_avg_variance16x16, vpx_sad16x16x3, vpx_sad16x16x8,
vpx_sad16x16x4d)
BFP(BLOCK_16X8, vpx_sad16x8, vpx_sad16x8_avg,
vpx_variance16x8, vpx_sub_pixel_variance16x8,
vpx_sub_pixel_avg_variance16x8,
vpx_sad16x8x3, vpx_sad16x8x8, vpx_sad16x8x4d)
BFP(BLOCK_8X16, vpx_sad8x16, vpx_sad8x16_avg,
vpx_variance8x16, vpx_sub_pixel_variance8x16,
vpx_sub_pixel_avg_variance8x16,
vpx_sad8x16x3, vpx_sad8x16x8, vpx_sad8x16x4d)
BFP(BLOCK_8X8, vpx_sad8x8, vpx_sad8x8_avg,
vpx_variance8x8, vpx_sub_pixel_variance8x8,
vpx_sub_pixel_avg_variance8x8,
vpx_sad8x8x3, vpx_sad8x8x8, vpx_sad8x8x4d)
BFP(BLOCK_8X4, vpx_sad8x4, vpx_sad8x4_avg,
vpx_variance8x4, vpx_sub_pixel_variance8x4,
vpx_sub_pixel_avg_variance8x4, NULL, vpx_sad8x4x8, vpx_sad8x4x4d)
BFP(BLOCK_4X8, vpx_sad4x8, vpx_sad4x8_avg,
vpx_variance4x8, vpx_sub_pixel_variance4x8,
vpx_sub_pixel_avg_variance4x8, NULL, vpx_sad4x8x8, vpx_sad4x8x4d)
BFP(BLOCK_4X4, vpx_sad4x4, vpx_sad4x4_avg,
vpx_variance4x4, vpx_sub_pixel_variance4x4,
vpx_sub_pixel_avg_variance4x4,
vpx_sad4x4x3, vpx_sad4x4x8, vpx_sad4x4x4d)
#if CONFIG_VP9_HIGHBITDEPTH
highbd_set_var_fns(cpi);
#endif
/* vp10_init_quantizer() is first called here. Add check in
* vp10_frame_init_quantizer() so that vp10_init_quantizer is only
* called later when needed. This will avoid unnecessary calls of
* vp10_init_quantizer() for every frame.
*/
vp10_init_quantizer(cpi);
vp10_loop_filter_init(cm);
cm->error.setjmp = 0;
return cpi;
}
#define SNPRINT(H, T) \
snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T))
#define SNPRINT2(H, T, V) \
snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V))
void vp10_remove_compressor(VP10_COMP *cpi) {
VP10_COMMON *cm;
unsigned int i;
int t;
if (!cpi)
return;
cm = &cpi->common;
if (cm->current_video_frame > 0) {
#if CONFIG_INTERNAL_STATS
vpx_clear_system_state();
if (cpi->oxcf.pass != 1) {
char headings[512] = {0};
char results[512] = {0};
FILE *f = fopen("opsnr.stt", "a");
double time_encoded = (cpi->last_end_time_stamp_seen
- cpi->first_time_stamp_ever) / 10000000.000;
double total_encode_time = (cpi->time_receive_data +
cpi->time_compress_data) / 1000.000;
const double dr =
(double)cpi->bytes * (double) 8 / (double)1000 / time_encoded;
const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1);
const double target_rate = (double)cpi->oxcf.target_bandwidth / 1000;
const double rate_err = ((100.0 * (dr - target_rate)) / target_rate);
if (cpi->b_calculate_psnr) {
const double total_psnr =
vpx_sse_to_psnr((double)cpi->total_samples, peak,
(double)cpi->total_sq_error);
const double totalp_psnr =
vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
(double)cpi->totalp_sq_error);
const double total_ssim = 100 * pow(cpi->summed_quality /
cpi->summed_weights, 8.0);
const double totalp_ssim = 100 * pow(cpi->summedp_quality /
cpi->summedp_weights, 8.0);
snprintf(headings, sizeof(headings),
"Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t"
"VPXSSIM\tVPSSIMP\tFASTSIM\tPSNRHVS\t"
"WstPsnr\tWstSsim\tWstFast\tWstHVS");
snprintf(results, sizeof(results),
"%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
"%7.3f\t%7.3f\t%7.3f\t%7.3f\t"
"%7.3f\t%7.3f\t%7.3f\t%7.3f",
dr, cpi->psnr.stat[ALL] / cpi->count, total_psnr,
cpi->psnrp.stat[ALL] / cpi->count, totalp_psnr,
total_ssim, totalp_ssim,
cpi->fastssim.stat[ALL] / cpi->count,
cpi->psnrhvs.stat[ALL] / cpi->count,
cpi->psnr.worst, cpi->worst_ssim, cpi->fastssim.worst,
cpi->psnrhvs.worst);
if (cpi->b_calculate_blockiness) {
SNPRINT(headings, "\t Block\tWstBlck");
SNPRINT2(results, "\t%7.3f", cpi->total_blockiness / cpi->count);
SNPRINT2(results, "\t%7.3f", cpi->worst_blockiness);
}
if (cpi->b_calculate_consistency) {
double consistency =
vpx_sse_to_psnr((double)cpi->totalp_samples, peak,
(double)cpi->total_inconsistency);
SNPRINT(headings, "\tConsist\tWstCons");
SNPRINT2(results, "\t%7.3f", consistency);
SNPRINT2(results, "\t%7.3f", cpi->worst_consistency);
}
if (cpi->b_calculate_ssimg) {
SNPRINT(headings, "\t SSIMG\tWtSSIMG");
SNPRINT2(results, "\t%7.3f", cpi->ssimg.stat[ALL] / cpi->count);
SNPRINT2(results, "\t%7.3f", cpi->ssimg.worst);
}
fprintf(f, "%s\t Time Rc-Err Abs Err\n", headings);
fprintf(f, "%s\t%8.0f %7.2f %7.2f\n", results,
total_encode_time, rate_err, fabs(rate_err));
}
fclose(f);
}
#endif
#if 0
{
printf("\n_pick_loop_filter_level:%d\n", cpi->time_pick_lpf / 1000);
printf("\n_frames recive_data encod_mb_row compress_frame Total\n");
printf("%6d %10ld %10ld %10ld %10ld\n", cpi->common.current_video_frame,
cpi->time_receive_data / 1000, cpi->time_encode_sb_row / 1000,
cpi->time_compress_data / 1000,
(cpi->time_receive_data + cpi->time_compress_data) / 1000);
}
#endif
}
#if CONFIG_VP9_TEMPORAL_DENOISING
vp10_denoiser_free(&(cpi->denoiser));
#endif
for (t = 0; t < cpi->num_workers; ++t) {
VPxWorker *const worker = &cpi->workers[t];
EncWorkerData *const thread_data = &cpi->tile_thr_data[t];
// Deallocate allocated threads.
vpx_get_worker_interface()->end(worker);
// Deallocate allocated thread data.
if (t < cpi->num_workers - 1) {
vpx_free(thread_data->td->counts);
vp10_free_pc_tree(thread_data->td);
vpx_free(thread_data->td);
}
}
vpx_free(cpi->tile_thr_data);
vpx_free(cpi->workers);
if (cpi->num_workers > 1)
vp10_loop_filter_dealloc(&cpi->lf_row_sync);
dealloc_compressor_data(cpi);
for (i = 0; i < sizeof(cpi->mbgraph_stats) /
sizeof(cpi->mbgraph_stats[0]); ++i) {
vpx_free(cpi->mbgraph_stats[i].mb_stats);
}
#if CONFIG_FP_MB_STATS
if (cpi->use_fp_mb_stats) {
vpx_free(cpi->twopass.frame_mb_stats_buf);
cpi->twopass.frame_mb_stats_buf = NULL;
}
#endif
vp10_remove_common(cm);
vp10_free_ref_frame_buffers(cm->buffer_pool);
#if CONFIG_VP9_POSTPROC
vp10_free_postproc_buffers(cm);
#endif
vpx_free(cpi);
#if CONFIG_VP9_TEMPORAL_DENOISING
#ifdef OUTPUT_YUV_DENOISED
fclose(yuv_denoised_file);
#endif
#endif
#ifdef OUTPUT_YUV_SKINMAP
fclose(yuv_skinmap_file);
#endif
#ifdef OUTPUT_YUV_REC
fclose(yuv_rec_file);
#endif
#if 0
if (keyfile)
fclose(keyfile);
if (framepsnr)
fclose(framepsnr);
if (kf_list)
fclose(kf_list);
#endif
}
/* TODO(yaowu): The block_variance calls the unoptimized versions of variance()
* and highbd_8_variance(). It should not.
*/
static void encoder_variance(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
int w, int h, unsigned int *sse, int *sum) {
int i, j;
*sum = 0;
*sse = 0;
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
const int diff = a[j] - b[j];
*sum += diff;
*sse += diff * diff;
}
a += a_stride;
b += b_stride;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
static void encoder_highbd_variance64(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h, uint64_t *sse,
uint64_t *sum) {
int i, j;
uint16_t *a = CONVERT_TO_SHORTPTR(a8);
uint16_t *b = CONVERT_TO_SHORTPTR(b8);
*sum = 0;
*sse = 0;
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
const int diff = a[j] - b[j];
*sum += diff;
*sse += diff * diff;
}
a += a_stride;
b += b_stride;
}
}
static void encoder_highbd_8_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h,
unsigned int *sse, int *sum) {
uint64_t sse_long = 0;
uint64_t sum_long = 0;
encoder_highbd_variance64(a8, a_stride, b8, b_stride, w, h,
&sse_long, &sum_long);
*sse = (unsigned int)sse_long;
*sum = (int)sum_long;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
static int64_t get_sse(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
int width, int height) {
const int dw = width % 16;
const int dh = height % 16;
int64_t total_sse = 0;
unsigned int sse = 0;
int sum = 0;
int x, y;
if (dw > 0) {
encoder_variance(&a[width - dw], a_stride, &b[width - dw], b_stride,
dw, height, &sse, &sum);
total_sse += sse;
}
if (dh > 0) {
encoder_variance(&a[(height - dh) * a_stride], a_stride,
&b[(height - dh) * b_stride], b_stride,
width - dw, dh, &sse, &sum);
total_sse += sse;
}
for (y = 0; y < height / 16; ++y) {
const uint8_t *pa = a;
const uint8_t *pb = b;
for (x = 0; x < width / 16; ++x) {
vpx_mse16x16(pa, a_stride, pb, b_stride, &sse);
total_sse += sse;
pa += 16;
pb += 16;
}
a += 16 * a_stride;
b += 16 * b_stride;
}
return total_sse;
}
#if CONFIG_VP9_HIGHBITDEPTH
static int64_t highbd_get_sse_shift(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int width, int height,
unsigned int input_shift) {
const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
int64_t total_sse = 0;
int x, y;
for (y = 0; y < height; ++y) {
for (x = 0; x < width; ++x) {
int64_t diff;
diff = (a[x] >> input_shift) - (b[x] >> input_shift);
total_sse += diff * diff;
}
a += a_stride;
b += b_stride;
}
return total_sse;
}
static int64_t highbd_get_sse(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
int width, int height) {
int64_t total_sse = 0;
int x, y;
const int dw = width % 16;
const int dh = height % 16;
unsigned int sse = 0;
int sum = 0;
if (dw > 0) {
encoder_highbd_8_variance(&a[width - dw], a_stride,
&b[width - dw], b_stride,
dw, height, &sse, &sum);
total_sse += sse;
}
if (dh > 0) {
encoder_highbd_8_variance(&a[(height - dh) * a_stride], a_stride,
&b[(height - dh) * b_stride], b_stride,
width - dw, dh, &sse, &sum);
total_sse += sse;
}
for (y = 0; y < height / 16; ++y) {
const uint8_t *pa = a;
const uint8_t *pb = b;
for (x = 0; x < width / 16; ++x) {
vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
total_sse += sse;
pa += 16;
pb += 16;
}
a += 16 * a_stride;
b += 16 * b_stride;
}
return total_sse;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
typedef struct {
double psnr[4]; // total/y/u/v
uint64_t sse[4]; // total/y/u/v
uint32_t samples[4]; // total/y/u/v
} PSNR_STATS;
#if CONFIG_VP9_HIGHBITDEPTH
static void calc_highbd_psnr(const YV12_BUFFER_CONFIG *a,
const YV12_BUFFER_CONFIG *b,
PSNR_STATS *psnr,
unsigned int bit_depth,
unsigned int in_bit_depth) {
const int widths[3] =
{a->y_crop_width, a->uv_crop_width, a->uv_crop_width };
const int heights[3] =
{a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer };
const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer };
const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
int i;
uint64_t total_sse = 0;
uint32_t total_samples = 0;
const double peak = (double)((1 << in_bit_depth) - 1);
const unsigned int input_shift = bit_depth - in_bit_depth;
for (i = 0; i < 3; ++i) {
const int w = widths[i];
const int h = heights[i];
const uint32_t samples = w * h;
uint64_t sse;
if (a->flags & YV12_FLAG_HIGHBITDEPTH) {
if (input_shift) {
sse = highbd_get_sse_shift(a_planes[i], a_strides[i],
b_planes[i], b_strides[i], w, h,
input_shift);
} else {
sse = highbd_get_sse(a_planes[i], a_strides[i],
b_planes[i], b_strides[i], w, h);
}
} else {
sse = get_sse(a_planes[i], a_strides[i],
b_planes[i], b_strides[i],
w, h);
}
psnr->sse[1 + i] = sse;
psnr->samples[1 + i] = samples;
psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
total_sse += sse;
total_samples += samples;
}
psnr->sse[0] = total_sse;
psnr->samples[0] = total_samples;
psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
(double)total_sse);
}
#else // !CONFIG_VP9_HIGHBITDEPTH
static void calc_psnr(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b,
PSNR_STATS *psnr) {
static const double peak = 255.0;
const int widths[3] = {
a->y_crop_width, a->uv_crop_width, a->uv_crop_width};
const int heights[3] = {
a->y_crop_height, a->uv_crop_height, a->uv_crop_height};
const uint8_t *a_planes[3] = {a->y_buffer, a->u_buffer, a->v_buffer};
const int a_strides[3] = {a->y_stride, a->uv_stride, a->uv_stride};
const uint8_t *b_planes[3] = {b->y_buffer, b->u_buffer, b->v_buffer};
const int b_strides[3] = {b->y_stride, b->uv_stride, b->uv_stride};
int i;
uint64_t total_sse = 0;
uint32_t total_samples = 0;
for (i = 0; i < 3; ++i) {
const int w = widths[i];
const int h = heights[i];
const uint32_t samples = w * h;
const uint64_t sse = get_sse(a_planes[i], a_strides[i],
b_planes[i], b_strides[i],
w, h);
psnr->sse[1 + i] = sse;
psnr->samples[1 + i] = samples;
psnr->psnr[1 + i] = vpx_sse_to_psnr(samples, peak, (double)sse);
total_sse += sse;
total_samples += samples;
}
psnr->sse[0] = total_sse;
psnr->samples[0] = total_samples;
psnr->psnr[0] = vpx_sse_to_psnr((double)total_samples, peak,
(double)total_sse);
}
#endif // CONFIG_VP9_HIGHBITDEPTH
static void generate_psnr_packet(VP10_COMP *cpi) {
struct vpx_codec_cx_pkt pkt;
int i;
PSNR_STATS psnr;
#if CONFIG_VP9_HIGHBITDEPTH
calc_highbd_psnr(cpi->Source, cpi->common.frame_to_show, &psnr,
cpi->td.mb.e_mbd.bd, cpi->oxcf.input_bit_depth);
#else
calc_psnr(cpi->Source, cpi->common.frame_to_show, &psnr);
#endif
for (i = 0; i < 4; ++i) {
pkt.data.psnr.samples[i] = psnr.samples[i];
pkt.data.psnr.sse[i] = psnr.sse[i];
pkt.data.psnr.psnr[i] = psnr.psnr[i];
}
pkt.kind = VPX_CODEC_PSNR_PKT;
vpx_codec_pkt_list_add(cpi->output_pkt_list, &pkt);
}
int vp10_use_as_reference(VP10_COMP *cpi, int ref_frame_flags) {
if (ref_frame_flags > 7)
return -1;
cpi->ref_frame_flags = ref_frame_flags;
return 0;
}
void vp10_update_reference(VP10_COMP *cpi, int ref_frame_flags) {
cpi->ext_refresh_golden_frame = (ref_frame_flags & VP9_GOLD_FLAG) != 0;
cpi->ext_refresh_alt_ref_frame = (ref_frame_flags & VP9_ALT_FLAG) != 0;
cpi->ext_refresh_last_frame = (ref_frame_flags & VP9_LAST_FLAG) != 0;
cpi->ext_refresh_frame_flags_pending = 1;
}
static YV12_BUFFER_CONFIG *get_vp10_ref_frame_buffer(VP10_COMP *cpi,
VP9_REFFRAME ref_frame_flag) {
MV_REFERENCE_FRAME ref_frame = NONE;
if (ref_frame_flag == VP9_LAST_FLAG)
ref_frame = LAST_FRAME;
else if (ref_frame_flag == VP9_GOLD_FLAG)
ref_frame = GOLDEN_FRAME;
else if (ref_frame_flag == VP9_ALT_FLAG)
ref_frame = ALTREF_FRAME;
return ref_frame == NONE ? NULL : get_ref_frame_buffer(cpi, ref_frame);
}
int vp10_copy_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
if (cfg) {
vp8_yv12_copy_frame(cfg, sd);
return 0;
} else {
return -1;
}
}
int vp10_set_reference_enc(VP10_COMP *cpi, VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
YV12_BUFFER_CONFIG *cfg = get_vp10_ref_frame_buffer(cpi, ref_frame_flag);
if (cfg) {
vp8_yv12_copy_frame(sd, cfg);
return 0;
} else {
return -1;
}
}
int vp10_update_entropy(VP10_COMP * cpi, int update) {
cpi->ext_refresh_frame_context = update;
cpi->ext_refresh_frame_context_pending = 1;
return 0;
}
#if defined(OUTPUT_YUV_DENOISED) || defined(OUTPUT_YUV_SKINMAP)
// The denoiser buffer is allocated as a YUV 440 buffer. This function writes it
// as YUV 420. We simply use the top-left pixels of the UV buffers, since we do
// not denoise the UV channels at this time. If ever we implement UV channel
// denoising we will have to modify this.
void vp10_write_yuv_frame_420(YV12_BUFFER_CONFIG *s, FILE *f) {
uint8_t *src = s->y_buffer;
int h = s->y_height;
do {
fwrite(src, s->y_width, 1, f);
src += s->y_stride;
} while (--h);
src = s->u_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, f);
src += s->uv_stride;
} while (--h);
src = s->v_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, f);
src += s->uv_stride;
} while (--h);
}
#endif
#ifdef OUTPUT_YUV_REC
void vp10_write_yuv_rec_frame(VP10_COMMON *cm) {
YV12_BUFFER_CONFIG *s = cm->frame_to_show;
uint8_t *src = s->y_buffer;
int h = cm->height;
#if CONFIG_VP9_HIGHBITDEPTH
if (s->flags & YV12_FLAG_HIGHBITDEPTH) {
uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer);
do {
fwrite(src16, s->y_width, 2, yuv_rec_file);
src16 += s->y_stride;
} while (--h);
src16 = CONVERT_TO_SHORTPTR(s->u_buffer);
h = s->uv_height;
do {
fwrite(src16, s->uv_width, 2, yuv_rec_file);
src16 += s->uv_stride;
} while (--h);
src16 = CONVERT_TO_SHORTPTR(s->v_buffer);
h = s->uv_height;
do {
fwrite(src16, s->uv_width, 2, yuv_rec_file);
src16 += s->uv_stride;
} while (--h);
fflush(yuv_rec_file);
return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
do {
fwrite(src, s->y_width, 1, yuv_rec_file);
src += s->y_stride;
} while (--h);
src = s->u_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, yuv_rec_file);
src += s->uv_stride;
} while (--h);
src = s->v_buffer;
h = s->uv_height;
do {
fwrite(src, s->uv_width, 1, yuv_rec_file);
src += s->uv_stride;
} while (--h);
fflush(yuv_rec_file);
}
#endif
#if CONFIG_VP9_HIGHBITDEPTH
static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst,
int bd) {
#else
static void scale_and_extend_frame_nonnormative(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst) {
#endif // CONFIG_VP9_HIGHBITDEPTH
// TODO(dkovalev): replace YV12_BUFFER_CONFIG with vpx_image_t
int i;
const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
const int src_widths[3] = {src->y_crop_width, src->uv_crop_width,
src->uv_crop_width };
const int src_heights[3] = {src->y_crop_height, src->uv_crop_height,
src->uv_crop_height};
uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
const int dst_widths[3] = {dst->y_crop_width, dst->uv_crop_width,
dst->uv_crop_width};
const int dst_heights[3] = {dst->y_crop_height, dst->uv_crop_height,
dst->uv_crop_height};
for (i = 0; i < MAX_MB_PLANE; ++i) {
#if CONFIG_VP9_HIGHBITDEPTH
if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
vp10_highbd_resize_plane(srcs[i], src_heights[i], src_widths[i],
src_strides[i], dsts[i], dst_heights[i],
dst_widths[i], dst_strides[i], bd);
} else {
vp10_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
}
#else
vp10_resize_plane(srcs[i], src_heights[i], src_widths[i], src_strides[i],
dsts[i], dst_heights[i], dst_widths[i], dst_strides[i]);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
vpx_extend_frame_borders(dst);
}
#if CONFIG_VP9_HIGHBITDEPTH
static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst, int bd) {
#else
static void scale_and_extend_frame(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst) {
#endif // CONFIG_VP9_HIGHBITDEPTH
const int src_w = src->y_crop_width;
const int src_h = src->y_crop_height;
const int dst_w = dst->y_crop_width;
const int dst_h = dst->y_crop_height;
const uint8_t *const srcs[3] = {src->y_buffer, src->u_buffer, src->v_buffer};
const int src_strides[3] = {src->y_stride, src->uv_stride, src->uv_stride};
uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
const InterpKernel *const kernel = vp10_filter_kernels[EIGHTTAP];
int x, y, i;
for (y = 0; y < dst_h; y += 16) {
for (x = 0; x < dst_w; x += 16) {
for (i = 0; i < MAX_MB_PLANE; ++i) {
const int factor = (i == 0 || i == 3 ? 1 : 2);
const int x_q4 = x * (16 / factor) * src_w / dst_w;
const int y_q4 = y * (16 / factor) * src_h / dst_h;
const int src_stride = src_strides[i];
const int dst_stride = dst_strides[i];
const uint8_t *src_ptr = srcs[i] + (y / factor) * src_h / dst_h *
src_stride + (x / factor) * src_w / dst_w;
uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
#if CONFIG_VP9_HIGHBITDEPTH
if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
vpx_highbd_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h,
16 / factor, 16 / factor, bd);
} else {
vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h,
16 / factor, 16 / factor);
}
#else
vpx_convolve8(src_ptr, src_stride, dst_ptr, dst_stride,
kernel[x_q4 & 0xf], 16 * src_w / dst_w,
kernel[y_q4 & 0xf], 16 * src_h / dst_h,
16 / factor, 16 / factor);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}
}
vpx_extend_frame_borders(dst);
}
static int scale_down(VP10_COMP *cpi, int q) {
RATE_CONTROL *const rc = &cpi->rc;
GF_GROUP *const gf_group = &cpi->twopass.gf_group;
int scale = 0;
assert(frame_is_kf_gf_arf(cpi));
if (rc->frame_size_selector == UNSCALED &&
q >= rc->rf_level_maxq[gf_group->rf_level[gf_group->index]]) {
const int max_size_thresh = (int)(rate_thresh_mult[SCALE_STEP1]
* VPXMAX(rc->this_frame_target, rc->avg_frame_bandwidth));
scale = rc->projected_frame_size > max_size_thresh ? 1 : 0;
}
return scale;
}
// Function to test for conditions that indicate we should loop
// back and recode a frame.
static int recode_loop_test(VP10_COMP *cpi,
int high_limit, int low_limit,
int q, int maxq, int minq) {
const RATE_CONTROL *const rc = &cpi->rc;
const VP10EncoderConfig *const oxcf = &cpi->oxcf;
const int frame_is_kfgfarf = frame_is_kf_gf_arf(cpi);
int force_recode = 0;
if ((rc->projected_frame_size >= rc->max_frame_bandwidth) ||
(cpi->sf.recode_loop == ALLOW_RECODE) ||
(frame_is_kfgfarf &&
(cpi->sf.recode_loop == ALLOW_RECODE_KFARFGF))) {
if (frame_is_kfgfarf &&
(oxcf->resize_mode == RESIZE_DYNAMIC) &&
scale_down(cpi, q)) {
// Code this group at a lower resolution.
cpi->resize_pending = 1;
return 1;
}
// TODO(agrange) high_limit could be greater than the scale-down threshold.
if ((rc->projected_frame_size > high_limit && q < maxq) ||
(rc->projected_frame_size < low_limit && q > minq)) {
force_recode = 1;
} else if (cpi->oxcf.rc_mode == VPX_CQ) {
// Deal with frame undershoot and whether or not we are
// below the automatically set cq level.
if (q > oxcf->cq_level &&
rc->projected_frame_size < ((rc->this_frame_target * 7) >> 3)) {
force_recode = 1;
}
}
}
return force_recode;
}
void vp10_update_reference_frames(VP10_COMP *cpi) {
VP10_COMMON * const cm = &cpi->common;
BufferPool *const pool = cm->buffer_pool;
// At this point the new frame has been encoded.
// If any buffer copy / swapping is signaled it should be done here.
if (cm->frame_type == KEY_FRAME) {
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
} else if (vp10_preserve_existing_gf(cpi)) {
// We have decided to preserve the previously existing golden frame as our
// new ARF frame. However, in the short term in function
// vp10_bitstream.c::get_refresh_mask() we left it in the GF slot and, if
// we're updating the GF with the current decoded frame, we save it to the
// ARF slot instead.
// We now have to update the ARF with the current frame and swap gld_fb_idx
// and alt_fb_idx so that, overall, we've stored the old GF in the new ARF
// slot and, if we're updating the GF, the current frame becomes the new GF.
int tmp;
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->alt_fb_idx], cm->new_fb_idx);
tmp = cpi->alt_fb_idx;
cpi->alt_fb_idx = cpi->gld_fb_idx;
cpi->gld_fb_idx = tmp;
} else { /* For non key/golden frames */
if (cpi->refresh_alt_ref_frame) {
int arf_idx = cpi->alt_fb_idx;
if ((cpi->oxcf.pass == 2) && cpi->multi_arf_allowed) {
const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
arf_idx = gf_group->arf_update_idx[gf_group->index];
}
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[arf_idx], cm->new_fb_idx);
memcpy(cpi->interp_filter_selected[ALTREF_FRAME],
cpi->interp_filter_selected[0],
sizeof(cpi->interp_filter_selected[0]));
}
if (cpi->refresh_golden_frame) {
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->gld_fb_idx], cm->new_fb_idx);
if (!cpi->rc.is_src_frame_alt_ref)
memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
cpi->interp_filter_selected[0],
sizeof(cpi->interp_filter_selected[0]));
else
memcpy(cpi->interp_filter_selected[GOLDEN_FRAME],
cpi->interp_filter_selected[ALTREF_FRAME],
sizeof(cpi->interp_filter_selected[ALTREF_FRAME]));
}
}
if (cpi->refresh_last_frame) {
ref_cnt_fb(pool->frame_bufs,
&cm->ref_frame_map[cpi->lst_fb_idx], cm->new_fb_idx);
if (!cpi->rc.is_src_frame_alt_ref)
memcpy(cpi->interp_filter_selected[LAST_FRAME],
cpi->interp_filter_selected[0],
sizeof(cpi->interp_filter_selected[0]));
}
#if CONFIG_VP9_TEMPORAL_DENOISING
if (cpi->oxcf.noise_sensitivity > 0) {
vp10_denoiser_update_frame_info(&cpi->denoiser,
*cpi->Source,
cpi->common.frame_type,
cpi->refresh_alt_ref_frame,
cpi->refresh_golden_frame,
cpi->refresh_last_frame);
}
#endif
}
static void loopfilter_frame(VP10_COMP *cpi, VP10_COMMON *cm) {
MACROBLOCKD *xd = &cpi->td.mb.e_mbd;
struct loopfilter *lf = &cm->lf;
if (is_lossless_requested(&cpi->oxcf)) {
lf->filter_level = 0;
} else {
struct vpx_usec_timer timer;
vpx_clear_system_state();
vpx_usec_timer_start(&timer);
vp10_pick_filter_level(cpi->Source, cpi, cpi->sf.lpf_pick);
vpx_usec_timer_mark(&timer);
cpi->time_pick_lpf += vpx_usec_timer_elapsed(&timer);
}
if (lf->filter_level > 0) {
if (cpi->num_workers > 1)
vp10_loop_filter_frame_mt(cm->frame_to_show, cm, xd->plane,
lf->filter_level, 0, 0,
cpi->workers, cpi->num_workers,
&cpi->lf_row_sync);
else
vp10_loop_filter_frame(cm->frame_to_show, cm, xd, lf->filter_level, 0, 0);
}
vpx_extend_frame_inner_borders(cm->frame_to_show);
}
static INLINE void alloc_frame_mvs(VP10_COMMON *const cm,
int buffer_idx) {
RefCntBuffer *const new_fb_ptr = &cm->buffer_pool->frame_bufs[buffer_idx];
if (new_fb_ptr->mvs == NULL ||
new_fb_ptr->mi_rows < cm->mi_rows ||
new_fb_ptr->mi_cols < cm->mi_cols) {
vpx_free(new_fb_ptr->mvs);
CHECK_MEM_ERROR(cm, new_fb_ptr->mvs,
(MV_REF *)vpx_calloc(cm->mi_rows * cm->mi_cols,
sizeof(*new_fb_ptr->mvs)));
new_fb_ptr->mi_rows = cm->mi_rows;
new_fb_ptr->mi_cols = cm->mi_cols;
}
}
void vp10_scale_references(VP10_COMP *cpi) {
VP10_COMMON *cm = &cpi->common;
MV_REFERENCE_FRAME ref_frame;
const VP9_REFFRAME ref_mask[3] = {VP9_LAST_FLAG, VP9_GOLD_FLAG, VP9_ALT_FLAG};
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
// Need to convert from VP9_REFFRAME to index into ref_mask (subtract 1).
if (cpi->ref_frame_flags & ref_mask[ref_frame - 1]) {
BufferPool *const pool = cm->buffer_pool;
const YV12_BUFFER_CONFIG *const ref = get_ref_frame_buffer(cpi,
ref_frame);
if (ref == NULL) {
cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
continue;
}
#if CONFIG_VP9_HIGHBITDEPTH
if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
RefCntBuffer *new_fb_ptr = NULL;
int force_scaling = 0;
int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
if (new_fb == INVALID_IDX) {
new_fb = get_free_fb(cm);
force_scaling = 1;
}
if (new_fb == INVALID_IDX)
return;
new_fb_ptr = &pool->frame_bufs[new_fb];
if (force_scaling ||
new_fb_ptr->buf.y_crop_width != cm->width ||
new_fb_ptr->buf.y_crop_height != cm->height) {
if (vpx_realloc_frame_buffer(&new_fb_ptr->buf, cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
cm->use_highbitdepth,
VP9_ENC_BORDER_IN_PIXELS,
cm->byte_alignment, NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffer");
scale_and_extend_frame(ref, &new_fb_ptr->buf, (int)cm->bit_depth);
cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
alloc_frame_mvs(cm, new_fb);
}
#else
if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) {
RefCntBuffer *new_fb_ptr = NULL;
int force_scaling = 0;
int new_fb = cpi->scaled_ref_idx[ref_frame - 1];
if (new_fb == INVALID_IDX) {
new_fb = get_free_fb(cm);
force_scaling = 1;
}
if (new_fb == INVALID_IDX)
return;
new_fb_ptr = &pool->frame_bufs[new_fb];
if (force_scaling ||
new_fb_ptr->buf.y_crop_width != cm->width ||
new_fb_ptr->buf.y_crop_height != cm->height) {
if (vpx_realloc_frame_buffer(&new_fb_ptr->buf, cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS,
cm->byte_alignment, NULL, NULL, NULL))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffer");
scale_and_extend_frame(ref, &new_fb_ptr->buf);
cpi->scaled_ref_idx[ref_frame - 1] = new_fb;
alloc_frame_mvs(cm, new_fb);
}
#endif // CONFIG_VP9_HIGHBITDEPTH
} else {
const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
RefCntBuffer *const buf = &pool->frame_bufs[buf_idx];
buf->buf.y_crop_width = ref->y_crop_width;
buf->buf.y_crop_height = ref->y_crop_height;
cpi->scaled_ref_idx[ref_frame - 1] = buf_idx;
++buf->ref_count;
}
} else {
if (cpi->oxcf.pass != 0)
cpi->scaled_ref_idx[ref_frame - 1] = INVALID_IDX;
}
}
}
static void release_scaled_references(VP10_COMP *cpi) {
VP10_COMMON *cm = &cpi->common;
int i;
if (cpi->oxcf.pass == 0) {
// Only release scaled references under certain conditions:
// if reference will be updated, or if scaled reference has same resolution.
int refresh[3];
refresh[0] = (cpi->refresh_last_frame) ? 1 : 0;
refresh[1] =<