#include "vp9/decoder/vp9_decodemv.h"
#include "vp9/decoder/vp9_dsubexp.h"
#include "vp9/decoder/vp9_dthread.h"
-#include "vp9/decoder/vp9_onyxd_int.h"
+#include "vp9/decoder/vp9_onyxd.h"
#include "vp9/decoder/vp9_read_bit_buffer.h"
#include "vp9/decoder/vp9_reader.h"
#include "vp9/decoder/vp9_thread.h"
static void update_mv_probs(vp9_prob *p, int n, vp9_reader *r) {
int i;
for (i = 0; i < n; ++i)
- if (vp9_read(r, NMV_UPDATE_PROB))
+ if (vp9_read(r, MV_UPDATE_PROB))
p[i] = (vp9_read_literal(r, 7) << 1) | 1;
}
// as they are always compared to values that are in 1/8th pel units
set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
- setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col);
+ vp9_setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col);
}
static void set_ref(VP9_COMMON *const cm, MACROBLOCKD *const xd,
if (!vp9_is_valid_scale(&ref_buffer->sf))
vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
"Invalid scale factors");
- setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col, &ref_buffer->sf);
+ vp9_setup_pre_planes(xd, idx, ref_buffer->buf, mi_row, mi_col,
+ &ref_buffer->sf);
xd->corrupted |= ref_buffer->buf->corrupted;
}
*/
#include "./vpx_config.h"
+
+#include "vpx_mem/vpx_mem.h"
+
#include "vp9/common/vp9_reconinter.h"
+
#include "vp9/decoder/vp9_dthread.h"
-#include "vp9/decoder/vp9_onyxd_int.h"
-#include "vpx_mem/vpx_mem.h"
+#include "vp9/decoder/vp9_onyxd.h"
#if CONFIG_MULTITHREAD
static INLINE void mutex_lock(pthread_mutex_t *const mutex) {
sync_read(lf_sync, r, c);
- setup_dst_planes(xd, frame_buffer, mi_row, mi_col);
+ vp9_setup_dst_planes(xd, frame_buffer, mi_row, mi_col);
vp9_setup_mask(cm, mi_row, mi_col, mi_8x8 + mi_col, cm->mode_info_stride,
&lfm);
mbmi = &xd->mi_8x8[0]->mbmi;
// Set up destination pointers
- setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col);
+ vp9_setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col);
// Set up limit values for MV components
// mv beyond the range do not produce new/different prediction block
const int idx_str = cm->mode_info_stride * mi_row + mi_col;
MODE_INFO **mi_8x8 = cm->mi_grid_visible + idx_str;
MODE_INFO **prev_mi_8x8 = cm->prev_mi_grid_visible + idx_str;
-
cpi->mb.source_variance = UINT_MAX;
if (cpi->sf.partition_search_type == FIXED_PARTITION) {
set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64);
vp9_setup_src_planes(x, cpi->Source, 0, 0);
// TODO(jkoleszar): are these initializations required?
- setup_pre_planes(xd, 0, get_ref_frame_buffer(cpi, LAST_FRAME), 0, 0, NULL);
- setup_dst_planes(xd, get_frame_new_buffer(cm), 0, 0);
+ vp9_setup_pre_planes(xd, 0, get_ref_frame_buffer(cpi, LAST_FRAME), 0, 0,
+ NULL);
+ vp9_setup_dst_planes(xd, get_frame_new_buffer(cm), 0, 0);
vp9_setup_block_planes(&x->e_mbd, cm->subsampling_x, cm->subsampling_y);
static void set_mode_info(MB_MODE_INFO *mbmi, BLOCK_SIZE bsize,
MB_PREDICTION_MODE mode) {
- mbmi->interp_filter = EIGHTTAP;
mbmi->mode = mode;
+ mbmi->uv_mode = mode;
mbmi->mv[0].as_int = 0;
mbmi->mv[1].as_int = 0;
- if (mode < NEARESTMV) {
- mbmi->ref_frame[0] = INTRA_FRAME;
- } else {
- mbmi->ref_frame[0] = LAST_FRAME;
- }
-
- mbmi->ref_frame[1] = INTRA_FRAME;
+ mbmi->ref_frame[0] = INTRA_FRAME;
+ mbmi->ref_frame[1] = NONE;
mbmi->tx_size = max_txsize_lookup[bsize];
- mbmi->uv_mode = mode;
mbmi->skip = 0;
mbmi->sb_type = bsize;
mbmi->segment_id = 0;
vp9_zero(cpi->coef_counts);
vp9_zero(cm->counts.eob_branch);
+ // Set frame level transform size use case
+ select_tx_mode(cpi);
+
cpi->mb.e_mbd.lossless = cm->base_qindex == 0 && cm->y_dc_delta_q == 0
&& cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
switch_lossless_mode(cpi, cpi->mb.e_mbd.lossless);
vp9_initialize_rd_consts(cpi);
vp9_initialize_me_consts(cpi, cm->base_qindex);
- switch_tx_mode(cpi);
if (cpi->oxcf.tuning == VP8_TUNE_SSIM) {
// Initialize encode frame context.
vp9_tile_init(&tile, cm, tile_row, tile_col);
for (mi_row = tile.mi_row_start;
mi_row < tile.mi_row_end; mi_row += MI_BLOCK_SIZE) {
- if (cpi->sf.use_nonrd_pick_mode)
+ if (cpi->sf.use_nonrd_pick_mode && cm->frame_type != KEY_FRAME)
encode_nonrd_sb_row(cpi, &tile, mi_row, &tp);
else
encode_rd_sb_row(cpi, &tile, mi_row, &tp);
cpi->mb.e_mbd.lossless = cpi->oxcf.lossless;
- /* transform size selection (4x4, 8x8, 16x16 or select-per-mb) */
- select_tx_mode(cpi);
cm->reference_mode = reference_mode;
encode_frame_internal(cpi);
}
}
} else {
+ cpi->mb.e_mbd.lossless = cpi->oxcf.lossless;
+ cm->reference_mode = SINGLE_REFERENCE;
// Force the usage of the BILINEAR interp_filter.
cm->interp_filter = BILINEAR;
encode_frame_internal(cpi);
for (ref = 0; ref < 1 + is_compound; ++ref) {
YV12_BUFFER_CONFIG *cfg = get_ref_frame_buffer(cpi,
mbmi->ref_frame[ref]);
- setup_pre_planes(xd, ref, cfg, mi_row, mi_col, &xd->block_refs[ref]->sf);
+ vp9_setup_pre_planes(xd, ref, cfg, mi_row, mi_col,
+ &xd->block_refs[ref]->sf);
}
vp9_build_inter_predictors_sb(xd, mi_row, mi_col, MAX(bsize, BLOCK_8X8));
#include "vp9/common/vp9_entropymv.h"
#include "vp9/common/vp9_quant_common.h"
- #include "vp9/common/vp9_reconinter.h" // setup_dst_planes()
+ #include "vp9/common/vp9_reconinter.h" // vp9_setup_dst_planes()
#include "vp9/common/vp9_systemdependent.h"
#include "vp9/encoder/vp9_block.h"
// Resets the first pass file to the given position using a relative seek from
// the current position.
static void reset_fpf_position(struct twopass_rc *p,
- FIRSTPASS_STATS *position) {
+ const FIRSTPASS_STATS *position) {
p->stats_in = position;
}
x->sadperbit16, &num00, &v_fn_ptr,
x->nmvjointcost,
x->mvcost, ref_mv);
+ if (tmp_err < INT_MAX)
+ tmp_err = vp9_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
if (tmp_err < INT_MAX - new_mv_mode_penalty)
tmp_err += new_mv_mode_penalty;
&num00, &v_fn_ptr,
x->nmvjointcost,
x->mvcost, ref_mv);
+ if (tmp_err < INT_MAX)
+ tmp_err = vp9_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
if (tmp_err < INT_MAX - new_mv_mode_penalty)
tmp_err += new_mv_mode_penalty;
vp9_clear_system_state();
vp9_setup_src_planes(x, cpi->Source, 0, 0);
- setup_pre_planes(xd, 0, lst_yv12, 0, 0, NULL);
- setup_dst_planes(xd, new_yv12, 0, 0);
+ vp9_setup_pre_planes(xd, 0, lst_yv12, 0, 0, NULL);
+ vp9_setup_dst_planes(xd, new_yv12, 0, 0);
xd->mi_8x8 = cm->mi_grid_visible;
xd->mi_8x8[0] = cm->mi;
mv.as_mv.row *= 8;
mv.as_mv.col *= 8;
this_error = motion_error;
- vp9_set_mbmode_and_mvs(xd, NEWMV, &mv.as_mv);
+ xd->mi_8x8[0]->mbmi.mode = NEWMV;
+ xd->mi_8x8[0]->mbmi.mv[0] = mv;
xd->mi_8x8[0]->mbmi.tx_size = TX_4X4;
xd->mi_8x8[0]->mbmi.ref_frame[0] = LAST_FRAME;
xd->mi_8x8[0]->mbmi.ref_frame[1] = NONE;
void vp9_init_second_pass(VP9_COMP *cpi) {
FIRSTPASS_STATS this_frame;
- FIRSTPASS_STATS *start_pos;
+ const FIRSTPASS_STATS *start_pos;
struct twopass_rc *const twopass = &cpi->twopass;
const VP9_CONFIG *const oxcf = &cpi->oxcf;
loop_decay_rate >= 0.999 &&
last_decay_rate < 0.9) {
int j;
- FIRSTPASS_STATS *position = cpi->twopass.stats_in;
+ const FIRSTPASS_STATS *position = cpi->twopass.stats_in;
FIRSTPASS_STATS tmp_next_frame;
// Look ahead a few frames to see if static condition persists...
// Analyse and define a gf/arf group.
static void define_gf_group(VP9_COMP *cpi, FIRSTPASS_STATS *this_frame) {
FIRSTPASS_STATS next_frame = { 0 };
- FIRSTPASS_STATS *start_pos;
+ const FIRSTPASS_STATS *start_pos;
struct twopass_rc *const twopass = &cpi->twopass;
int i;
double boost_score = 0.0;
((next_frame->intra_error /
DOUBLE_DIVIDE_CHECK(next_frame->coded_error)) > 3.5))))) {
int i;
- FIRSTPASS_STATS *start_pos;
-
- FIRSTPASS_STATS local_next_frame;
-
+ const FIRSTPASS_STATS *start_pos = cpi->twopass.stats_in;
+ FIRSTPASS_STATS local_next_frame = *next_frame;
double boost_score = 0.0;
double old_boost_score = 0.0;
double decay_accumulator = 1.0;
- local_next_frame = *next_frame;
-
- // Note the starting file position so we can reset to it.
- start_pos = cpi->twopass.stats_in;
-
// Examine how well the key frame predicts subsequent frames.
for (i = 0; i < 16; ++i) {
double next_iiratio = (IIKFACTOR1 * local_next_frame.intra_error /
FIRSTPASS_STATS last_frame;
FIRSTPASS_STATS first_frame;
FIRSTPASS_STATS next_frame;
- FIRSTPASS_STATS *start_position;
+ const FIRSTPASS_STATS *start_position;
double decay_accumulator = 1.0;
double zero_motion_accumulator = 1.0;
MB_MODE_INFO *mbmi = &xd->mi_8x8[0]->mbmi;
struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0}};
int bestsme = INT_MAX;
- int further_steps, step_param;
+ int step_param;
int sadpb = x->sadperbit16;
MV mvp_full;
int ref = mbmi->ref_frame[0];
- int_mv ref_mv = mbmi->ref_mvs[ref][0];
+ const MV ref_mv = mbmi->ref_mvs[ref][0].as_mv;
int i;
int tmp_col_min = x->mv_col_min;
for (i = 0; i < MAX_MB_PLANE; i++)
backup_yv12[i] = xd->plane[i].pre[0];
- setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
+ vp9_setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
}
- vp9_set_mv_search_range(x, &ref_mv.as_mv);
+ vp9_set_mv_search_range(x, &ref_mv);
// TODO(jingning) exploiting adaptive motion search control in non-RD
// mode decision too.
step_param = 6;
- further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
for (i = LAST_FRAME; i <= LAST_FRAME && cpi->common.show_frame; ++i) {
if ((x->pred_mv_sad[ref] >> 3) > x->pred_mv_sad[i]) {
mvp_full.row >>= 3;
if (cpi->sf.search_method == FAST_HEX) {
- bestsme = vp9_fast_hex_search(x, &mvp_full, step_param, sadpb,
- &cpi->fn_ptr[bsize], 1,
- &ref_mv.as_mv, &tmp_mv->as_mv);
+ // NOTE: this returns SAD
+ vp9_fast_hex_search(x, &mvp_full, step_param, sadpb, 0,
+ &cpi->fn_ptr[bsize], 1,
+ &ref_mv, &tmp_mv->as_mv);
} else if (cpi->sf.search_method == HEX) {
- bestsme = vp9_hex_search(x, &mvp_full, step_param, sadpb, 1,
- &cpi->fn_ptr[bsize], 1,
- &ref_mv.as_mv, &tmp_mv->as_mv);
+ // NOTE: this returns SAD
+ vp9_hex_search(x, &mvp_full, step_param, sadpb, 1,
+ &cpi->fn_ptr[bsize], 1,
+ &ref_mv, &tmp_mv->as_mv);
} else if (cpi->sf.search_method == SQUARE) {
- bestsme = vp9_square_search(x, &mvp_full, step_param, sadpb, 1,
- &cpi->fn_ptr[bsize], 1,
- &ref_mv.as_mv, &tmp_mv->as_mv);
+ // NOTE: this returns SAD
+ vp9_square_search(x, &mvp_full, step_param, sadpb, 1,
+ &cpi->fn_ptr[bsize], 1,
+ &ref_mv, &tmp_mv->as_mv);
} else if (cpi->sf.search_method == BIGDIA) {
- bestsme = vp9_bigdia_search(x, &mvp_full, step_param, sadpb, 1,
- &cpi->fn_ptr[bsize], 1,
- &ref_mv.as_mv, &tmp_mv->as_mv);
+ // NOTE: this returns SAD
+ vp9_bigdia_search(x, &mvp_full, step_param, sadpb, 1,
+ &cpi->fn_ptr[bsize], 1,
+ &ref_mv, &tmp_mv->as_mv);
} else {
- bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
- sadpb, further_steps, 1,
- &cpi->fn_ptr[bsize],
- &ref_mv.as_mv, &tmp_mv->as_mv);
+ int further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
+ // NOTE: this returns variance
+ vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
+ sadpb, further_steps, 1,
+ &cpi->fn_ptr[bsize],
+ &ref_mv, &tmp_mv->as_mv);
}
x->mv_col_min = tmp_col_min;
x->mv_col_max = tmp_col_max;
tmp_mv->as_mv.col = tmp_mv->as_mv.col * 8;
// calculate the bit cost on motion vector
- *rate_mv = vp9_mv_bit_cost(&tmp_mv->as_mv, &ref_mv.as_mv,
+ *rate_mv = vp9_mv_bit_cost(&tmp_mv->as_mv, &ref_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
return bestsme;
}
for (i = 0; i < MAX_MB_PLANE; i++)
backup_yv12[i] = xd->plane[i].pre[0];
- setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
+ vp9_setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
}
tmp_mv->col >>= 3;
}
}
+static void model_rd_for_sb_y(VP9_COMP *cpi, BLOCK_SIZE bsize,
+ MACROBLOCK *x, MACROBLOCKD *xd,
+ int *out_rate_sum, int64_t *out_dist_sum) {
+ // Note our transform coeffs are 8 times an orthogonal transform.
+ // Hence quantizer step is also 8 times. To get effective quantizer
+ // we need to divide by 8 before sending to modeling function.
+ unsigned int sse;
+ int rate;
+ int64_t dist;
+
+
+ struct macroblock_plane *const p = &x->plane[0];
+ struct macroblockd_plane *const pd = &xd->plane[0];
+ const BLOCK_SIZE bs = get_plane_block_size(bsize, pd);
+
+ (void) cpi->fn_ptr[bs].vf(p->src.buf, p->src.stride,
+ pd->dst.buf, pd->dst.stride, &sse);
+
+ vp9_model_rd_from_var_lapndz(sse, 1 << num_pels_log2_lookup[bs],
+ pd->dequant[1] >> 3, &rate, &dist);
+
+ *out_rate_sum = rate;
+ *out_dist_sum = dist << 4;
+}
+
// TODO(jingning) placeholder for inter-frame non-RD mode decision.
// this needs various further optimizations. to be continued..
int64_t vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
VP9_ALT_FLAG };
int64_t best_rd = INT64_MAX;
int64_t this_rd = INT64_MAX;
- static const int cost[4]= { 0, 2, 4, 6 };
const int64_t inter_mode_thresh = 300;
const int64_t intra_mode_cost = 50;
+ int rate = INT_MAX;
+ int64_t dist = INT64_MAX;
+
x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
x->skip = 0;
}
for (ref_frame = LAST_FRAME; ref_frame <= LAST_FRAME ; ++ref_frame) {
- int rate_mv = 0;
if (!(cpi->ref_frame_flags & flag_list[ref_frame]))
continue;
mbmi->ref_frame[0] = ref_frame;
for (this_mode = NEARESTMV; this_mode <= NEWMV; ++this_mode) {
- int rate = cost[INTER_OFFSET(this_mode)]
- << (num_pels_log2_lookup[bsize] - 4);
- int64_t dist;
+ int rate_mv = 0;
+
if (cpi->sf.disable_inter_mode_mask[bsize] &
(1 << INTER_OFFSET(this_mode)))
continue;
&frame_mv[NEWMV][ref_frame].as_mv);
}
- if (frame_mv[this_mode][ref_frame].as_int == 0) {
- dist = x->mode_sad[ref_frame][INTER_OFFSET(ZEROMV)];
- } else if (this_mode != NEARESTMV &&
- frame_mv[NEARESTMV][ref_frame].as_int ==
- frame_mv[this_mode][ref_frame].as_int) {
- dist = x->mode_sad[ref_frame][INTER_OFFSET(NEARESTMV)];
- } else {
- mbmi->mode = this_mode;
- mbmi->mv[0].as_int = frame_mv[this_mode][ref_frame].as_int;
- vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
- dist = x->mode_sad[ref_frame][INTER_OFFSET(this_mode)] =
- cpi->fn_ptr[bsize].sdf(p->src.buf, p->src.stride,
- pd->dst.buf, pd->dst.stride, INT_MAX);
- }
+ if (this_mode != NEARESTMV)
+ if (frame_mv[this_mode][ref_frame].as_int ==
+ frame_mv[NEARESTMV][ref_frame].as_int)
+ continue;
- this_rd = rate + dist;
+ mbmi->mode = this_mode;
+ mbmi->mv[0].as_int = frame_mv[this_mode][ref_frame].as_int;
+ vp9_build_inter_predictors_sby(xd, mi_row, mi_col, bsize);
+
+ model_rd_for_sb_y(cpi, bsize, x, xd, &rate, &dist);
+ rate += rate_mv;
+ rate += x->inter_mode_cost[mbmi->mode_context[ref_frame]]
+ [INTER_OFFSET(this_mode)];
+ this_rd = RDCOST(x->rdmult, x->rddiv, rate, dist);
if (this_rd < best_rd) {
best_rd = this_rd;
&p->src.buf[0], p->src.stride,
&pd->dst.buf[0], pd->dst.stride, 0, 0, 0);
- this_rd = cpi->fn_ptr[bsize].sdf(p->src.buf,
- p->src.stride,
- pd->dst.buf,
- pd->dst.stride, INT_MAX);
+ model_rd_for_sb_y(cpi, bsize, x, xd, &rate, &dist);
+ rate += x->mbmode_cost[this_mode];
+ this_rd = RDCOST(x->rdmult, x->rddiv, rate, dist);
if (this_rd + intra_mode_cost < best_rd) {
best_rd = this_rd;
}
}
}
+
return INT64_MAX;
}
#include "vp9/common/vp9_seg_common.h"
#include "vp9/common/vp9_systemdependent.h"
+#include "vp9/encoder/vp9_cost.h"
#include "vp9/encoder/vp9_encodemb.h"
#include "vp9/encoder/vp9_encodemv.h"
#include "vp9/encoder/vp9_mcomp.h"
#include "vp9/encoder/vp9_ratectrl.h"
#include "vp9/encoder/vp9_rdopt.h"
#include "vp9/encoder/vp9_tokenize.h"
-#include "vp9/encoder/vp9_treewriter.h"
#include "vp9/encoder/vp9_variance.h"
/* Factor to weigh the rate for switchable interp filters */
// TODO(rbultje) separate tables for superblock costing?
vp9_cost_tokens(x->mbmode_cost, fc->y_mode_prob[1], vp9_intra_mode_tree);
- vp9_cost_tokens(x->intra_uv_mode_cost[1],
- fc->uv_mode_prob[INTRA_MODES - 1], vp9_intra_mode_tree);
- vp9_cost_tokens(x->intra_uv_mode_cost[0],
- vp9_kf_uv_mode_prob[INTRA_MODES - 1], vp9_intra_mode_tree);
+ vp9_cost_tokens(x->intra_uv_mode_cost[KEY_FRAME],
+ vp9_kf_uv_mode_prob[TM_PRED], vp9_intra_mode_tree);
+ vp9_cost_tokens(x->intra_uv_mode_cost[INTER_FRAME],
+ fc->uv_mode_prob[TM_PRED], vp9_intra_mode_tree);
for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i)
vp9_cost_tokens((int *)x->switchable_interp_costs[i],
x->errorperbit = cpi->RDMULT / RD_MULT_EPB_RATIO;
x->errorperbit += (x->errorperbit == 0);
- vp9_set_speed_features(cpi);
-
x->select_txfm_size = (cpi->sf.tx_size_search_method == USE_LARGESTALL &&
cm->frame_type != KEY_FRAME) ? 0 : 1;
set_block_thresholds(cpi);
- if (!cpi->sf.use_nonrd_pick_mode) {
+ if (!cpi->sf.use_nonrd_pick_mode || cm->frame_type == KEY_FRAME) {
fill_token_costs(x->token_costs, cm->fc.coef_probs);
for (i = 0; i < PARTITION_CONTEXTS; i++)
vp9_partition_tree);
}
- if (!cpi->sf.use_nonrd_pick_mode || (cm->current_video_frame & 0x07) == 1) {
+ if (!cpi->sf.use_nonrd_pick_mode || (cm->current_video_frame & 0x07) == 1 ||
+ cm->frame_type == KEY_FRAME) {
fill_mode_costs(cpi);
if (!frame_is_intra_only(cm)) {
*d_q10 = (dist_tab_q10[xq] * b_q10 + dist_tab_q10[xq + 1] * a_q10) >> 10;
}
-static void model_rd_from_var_lapndz(unsigned int var, unsigned int n,
- unsigned int qstep, int *rate,
- int64_t *dist) {
+void vp9_model_rd_from_var_lapndz(unsigned int var, unsigned int n,
+ unsigned int qstep, int *rate,
+ int64_t *dist) {
// This function models the rate and distortion for a Laplacian
// source with given variance when quantized with a uniform quantizer
// with given stepsize. The closed form expressions are in:
} else {
int rate;
int64_t dist;
- model_rd_from_var_lapndz(sse, 1 << num_pels_log2_lookup[bs],
- pd->dequant[1] >> 3, &rate, &dist);
+ vp9_model_rd_from_var_lapndz(sse, 1 << num_pels_log2_lookup[bs],
+ pd->dequant[1] >> 3, &rate, &dist);
rate_sum += rate;
dist_sum += dist;
}
&pd->dst.buf[j * pd->dst.stride + k], pd->dst.stride,
&sse);
// sse works better than var, since there is no dc prediction used
- model_rd_from_var_lapndz(sse, t * t, pd->dequant[1] >> 3, &rate, &dist);
+ vp9_model_rd_from_var_lapndz(sse, t * t, pd->dequant[1] >> 3,
+ &rate, &dist);
rate_sum += rate;
dist_sum += dist;
*out_skip &= (rate < 1024);
static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib,
MB_PREDICTION_MODE *best_mode,
- int *bmode_costs,
+ const int *bmode_costs,
ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,
int *bestrate, int *bestratey,
int64_t *bestdistortion,
return best_rd;
}
-static int64_t rd_pick_intra_sub_8x8_y_mode(VP9_COMP * const cpi,
- MACROBLOCK * const mb,
- int * const rate,
- int * const rate_y,
- int64_t * const distortion,
+static int64_t rd_pick_intra_sub_8x8_y_mode(VP9_COMP *cpi, MACROBLOCK *mb,
+ int *rate, int *rate_y,
+ int64_t *distortion,
int64_t best_rd) {
int i, j;
- MACROBLOCKD *const xd = &mb->e_mbd;
+ const MACROBLOCKD *const xd = &mb->e_mbd;
MODE_INFO *const mic = xd->mi_8x8[0];
const MODE_INFO *above_mi = xd->mi_8x8[-xd->mode_info_stride];
const MODE_INFO *left_mi = xd->left_available ? xd->mi_8x8[-1] : NULL;
int tot_rate_y = 0;
int64_t total_rd = 0;
ENTROPY_CONTEXT t_above[4], t_left[4];
- int *bmode_costs;
+ const int *bmode_costs = mb->mbmode_cost;
vpx_memcpy(t_above, xd->plane[0].above_context, sizeof(t_above));
vpx_memcpy(t_left, xd->plane[0].left_context, sizeof(t_left));
- bmode_costs = mb->mbmode_cost;
-
// Pick modes for each sub-block (of size 4x4, 4x8, or 8x4) in an 8x8 block.
for (idy = 0; idy < 2; idy += num_4x4_blocks_high) {
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
}
}
-void vp9_set_mbmode_and_mvs(MACROBLOCKD *xd, MB_PREDICTION_MODE mode,
- const MV *mv) {
- xd->mi_8x8[0]->mbmi.mode = mode;
- xd->mi_8x8[0]->mbmi.mv[0].as_mv = *mv;
-}
-
static void joint_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
BLOCK_SIZE bsize,
int_mv *frame_mv,
int_mv single_newmv[MAX_REF_FRAMES],
int *rate_mv);
-static int labels2mode(MACROBLOCK *x, int i,
+static int labels2mode(VP9_COMP *cpi, MACROBLOCKD *xd, int i,
MB_PREDICTION_MODE mode,
- int_mv *this_mv, int_mv *this_second_mv,
+ int_mv this_mv[2],
int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES],
int_mv seg_mvs[MAX_REF_FRAMES],
- int_mv *best_ref_mv,
- int_mv *second_best_ref_mv,
- int *mvjcost, int *mvcost[2], VP9_COMP *cpi) {
- MACROBLOCKD *const xd = &x->e_mbd;
+ int_mv *best_ref_mv[2],
+ const int *mvjcost, int *mvcost[2]) {
MODE_INFO *const mic = xd->mi_8x8[0];
- MB_MODE_INFO *mbmi = &mic->mbmi;
+ const MB_MODE_INFO *const mbmi = &mic->mbmi;
int thismvcost = 0;
int idx, idy;
const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[mbmi->sb_type];
const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[mbmi->sb_type];
- const int has_second_rf = has_second_ref(mbmi);
+ const int is_compound = has_second_ref(mbmi);
// the only time we should do costing for new motion vector or mode
// is when we are on a new label (jbb May 08, 2007)
switch (mode) {
case NEWMV:
- this_mv->as_int = seg_mvs[mbmi->ref_frame[0]].as_int;
- thismvcost += vp9_mv_bit_cost(&this_mv->as_mv, &best_ref_mv->as_mv,
+ this_mv[0].as_int = seg_mvs[mbmi->ref_frame[0]].as_int;
+ thismvcost += vp9_mv_bit_cost(&this_mv[0].as_mv, &best_ref_mv[0]->as_mv,
mvjcost, mvcost, MV_COST_WEIGHT_SUB);
- if (has_second_rf) {
- this_second_mv->as_int = seg_mvs[mbmi->ref_frame[1]].as_int;
- thismvcost += vp9_mv_bit_cost(&this_second_mv->as_mv,
- &second_best_ref_mv->as_mv,
+ if (is_compound) {
+ this_mv[1].as_int = seg_mvs[mbmi->ref_frame[1]].as_int;
+ thismvcost += vp9_mv_bit_cost(&this_mv[1].as_mv, &best_ref_mv[1]->as_mv,
mvjcost, mvcost, MV_COST_WEIGHT_SUB);
}
break;
case NEARESTMV:
- this_mv->as_int = frame_mv[NEARESTMV][mbmi->ref_frame[0]].as_int;
- if (has_second_rf)
- this_second_mv->as_int = frame_mv[NEARESTMV][mbmi->ref_frame[1]].as_int;
+ this_mv[0].as_int = frame_mv[NEARESTMV][mbmi->ref_frame[0]].as_int;
+ if (is_compound)
+ this_mv[1].as_int = frame_mv[NEARESTMV][mbmi->ref_frame[1]].as_int;
break;
case NEARMV:
- this_mv->as_int = frame_mv[NEARMV][mbmi->ref_frame[0]].as_int;
- if (has_second_rf)
- this_second_mv->as_int = frame_mv[NEARMV][mbmi->ref_frame[1]].as_int;
+ this_mv[0].as_int = frame_mv[NEARMV][mbmi->ref_frame[0]].as_int;
+ if (is_compound)
+ this_mv[1].as_int = frame_mv[NEARMV][mbmi->ref_frame[1]].as_int;
break;
case ZEROMV:
- this_mv->as_int = 0;
- if (has_second_rf)
- this_second_mv->as_int = 0;
+ this_mv[0].as_int = 0;
+ if (is_compound)
+ this_mv[1].as_int = 0;
break;
default:
break;
}
- mic->bmi[i].as_mv[0].as_int = this_mv->as_int;
- if (has_second_rf)
- mic->bmi[i].as_mv[1].as_int = this_second_mv->as_int;
+ mic->bmi[i].as_mv[0].as_int = this_mv[0].as_int;
+ if (is_compound)
+ mic->bmi[i].as_mv[1].as_int = this_mv[1].as_int;
mic->bmi[i].as_mode = mode;
int mode_idx;
int subpelmv = 1, have_ref = 0;
const int has_second_rf = has_second_ref(mbmi);
+ const int disable_inter_mode_mask = cpi->sf.disable_inter_mode_mask[bsize];
vpx_memcpy(t_above, pd->above_context, sizeof(t_above));
vpx_memcpy(t_left, pd->left_context, sizeof(t_left));
for (idx = 0; idx < 2; idx += num_4x4_blocks_wide) {
// TODO(jingning,rbultje): rewrite the rate-distortion optimization
// loop for 4x4/4x8/8x4 block coding. to be replaced with new rd loop
- int_mv mode_mv[MB_MODE_COUNT], second_mode_mv[MB_MODE_COUNT];
+ int_mv mode_mv[MB_MODE_COUNT][2];
int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES];
MB_PREDICTION_MODE mode_selected = ZEROMV;
int64_t best_rd = INT64_MAX;
mode_idx = INTER_OFFSET(this_mode);
bsi->rdstat[i][mode_idx].brdcost = INT64_MAX;
- if (cpi->sf.disable_inter_mode_mask[bsize] & (1 << mode_idx))
+ if (disable_inter_mode_mask & (1 << mode_idx))
continue;
// if we're near/nearest and mv == 0,0, compare to zeromv
- if ((this_mode == NEARMV || this_mode == NEARESTMV ||
+ if (!(disable_inter_mode_mask & (1 << INTER_OFFSET(ZEROMV))) &&
+ (this_mode == NEARMV || this_mode == NEARESTMV ||
this_mode == ZEROMV) &&
frame_mv[this_mode][mbmi->ref_frame[0]].as_int == 0 &&
(!has_second_rf ||
// motion search for newmv (single predictor case only)
if (!has_second_rf && this_mode == NEWMV &&
seg_mvs[i][mbmi->ref_frame[0]].as_int == INVALID_MV) {
- int_mv *const new_mv = &mode_mv[NEWMV];
+ int_mv *const new_mv = &mode_mv[NEWMV][0];
int step_param = 0;
int further_steps;
int thissme, bestsme = INT_MAX;
sadpb, 1, v_fn_ptr, 1,
&bsi->ref_mv[0]->as_mv,
&new_mv->as_mv);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_var(x, &new_mv->as_mv,
+ &bsi->ref_mv[0]->as_mv,
+ v_fn_ptr, 1);
} else if (cpi->sf.search_method == SQUARE) {
bestsme = vp9_square_search(x, &mvp_full,
step_param,
sadpb, 1, v_fn_ptr, 1,
&bsi->ref_mv[0]->as_mv,
&new_mv->as_mv);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_var(x, &new_mv->as_mv,
+ &bsi->ref_mv[0]->as_mv,
+ v_fn_ptr, 1);
} else if (cpi->sf.search_method == BIGDIA) {
bestsme = vp9_bigdia_search(x, &mvp_full,
step_param,
sadpb, 1, v_fn_ptr, 1,
&bsi->ref_mv[0]->as_mv,
&new_mv->as_mv);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_var(x, &new_mv->as_mv,
+ &bsi->ref_mv[0]->as_mv,
+ v_fn_ptr, 1);
} else {
bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
sadpb, further_steps, 0, v_fn_ptr,
}
bsi->rdstat[i][mode_idx].brate =
- labels2mode(x, i, this_mode, &mode_mv[this_mode],
- &second_mode_mv[this_mode], frame_mv, seg_mvs[i],
- bsi->ref_mv[0], bsi->ref_mv[1], x->nmvjointcost,
- x->mvcost, cpi);
-
-
- bsi->rdstat[i][mode_idx].mvs[0].as_int = mode_mv[this_mode].as_int;
- if (num_4x4_blocks_wide > 1)
- bsi->rdstat[i + 1][mode_idx].mvs[0].as_int =
- mode_mv[this_mode].as_int;
- if (num_4x4_blocks_high > 1)
- bsi->rdstat[i + 2][mode_idx].mvs[0].as_int =
- mode_mv[this_mode].as_int;
- if (has_second_rf) {
- bsi->rdstat[i][mode_idx].mvs[1].as_int =
- second_mode_mv[this_mode].as_int;
+ labels2mode(cpi, xd, i, this_mode, mode_mv[this_mode], frame_mv,
+ seg_mvs[i], bsi->ref_mv, x->nmvjointcost, x->mvcost);
+
+ for (ref = 0; ref < 1 + has_second_rf; ++ref) {
+ bsi->rdstat[i][mode_idx].mvs[ref].as_int =
+ mode_mv[this_mode][ref].as_int;
if (num_4x4_blocks_wide > 1)
- bsi->rdstat[i + 1][mode_idx].mvs[1].as_int =
- second_mode_mv[this_mode].as_int;
+ bsi->rdstat[i + 1][mode_idx].mvs[ref].as_int =
+ mode_mv[this_mode][ref].as_int;
if (num_4x4_blocks_high > 1)
- bsi->rdstat[i + 2][mode_idx].mvs[1].as_int =
- second_mode_mv[this_mode].as_int;
+ bsi->rdstat[i + 2][mode_idx].mvs[ref].as_int =
+ mode_mv[this_mode][ref].as_int;
}
// Trap vectors that reach beyond the UMV borders
- if (mv_check_bounds(x, &mode_mv[this_mode].as_mv) ||
+ if (mv_check_bounds(x, &mode_mv[this_mode][0].as_mv) ||
(has_second_rf &&
- mv_check_bounds(x, &second_mode_mv[this_mode].as_mv)))
+ mv_check_bounds(x, &mode_mv[this_mode][1].as_mv)))
continue;
if (filter_idx > 0) {
BEST_SEG_INFO *ref_bsi = bsi_buf;
- subpelmv = mv_has_subpel(&mode_mv[this_mode].as_mv);
- have_ref = mode_mv[this_mode].as_int ==
- ref_bsi->rdstat[i][mode_idx].mvs[0].as_int;
- if (has_second_rf) {
- subpelmv |= mv_has_subpel(&second_mode_mv[this_mode].as_mv);
- have_ref &= second_mode_mv[this_mode].as_int ==
- ref_bsi->rdstat[i][mode_idx].mvs[1].as_int;
+ subpelmv = 0;
+ have_ref = 1;
+
+ for (ref = 0; ref < 1 + has_second_rf; ++ref) {
+ subpelmv |= mv_has_subpel(&mode_mv[this_mode][ref].as_mv);
+ have_ref &= mode_mv[this_mode][ref].as_int ==
+ ref_bsi->rdstat[i][mode_idx].mvs[ref].as_int;
}
if (filter_idx > 1 && !subpelmv && !have_ref) {
ref_bsi = bsi_buf + 1;
- have_ref = mode_mv[this_mode].as_int ==
- ref_bsi->rdstat[i][mode_idx].mvs[0].as_int;
- if (has_second_rf) {
- have_ref &= second_mode_mv[this_mode].as_int ==
- ref_bsi->rdstat[i][mode_idx].mvs[1].as_int;
- }
+ have_ref = 1;
+ for (ref = 0; ref < 1 + has_second_rf; ++ref)
+ have_ref &= mode_mv[this_mode][ref].as_int ==
+ ref_bsi->rdstat[i][mode_idx].mvs[ref].as_int;
}
if (!subpelmv && have_ref &&
vpx_memcpy(t_above, bsi->rdstat[i][mode_idx].ta, sizeof(t_above));
vpx_memcpy(t_left, bsi->rdstat[i][mode_idx].tl, sizeof(t_left));
- labels2mode(x, i, mode_selected, &mode_mv[mode_selected],
- &second_mode_mv[mode_selected], frame_mv, seg_mvs[i],
- bsi->ref_mv[0], bsi->ref_mv[1], x->nmvjointcost,
- x->mvcost, cpi);
+ labels2mode(cpi, xd, i, mode_selected, mode_mv[mode_selected],
+ frame_mv, seg_mvs[i], bsi->ref_mv, x->nmvjointcost,
+ x->mvcost);
br += bsi->rdstat[i][mode_idx].brate;
bd += bsi->rdstat[i][mode_idx].bdist;
for (i = 0; i < MAX_MB_PLANE; i++)
backup_yv12[i] = xd->plane[i].pre[0];
- setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
+ vp9_setup_pre_planes(xd, 0, scaled_ref_frame, mi_row, mi_col, NULL);
}
vp9_set_mv_search_range(x, &ref_mv);
further_steps = (cpi->sf.max_step_search_steps - 1) - step_param;
if (cpi->sf.search_method == FAST_HEX) {
- bestsme = vp9_fast_hex_search(x, &mvp_full, step_param, sadpb,
+ bestsme = vp9_fast_hex_search(x, &mvp_full, step_param, sadpb, 0,
&cpi->fn_ptr[bsize], 1,
&ref_mv, &tmp_mv->as_mv);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_var(x, &tmp_mv->as_mv, &ref_mv,
+ &cpi->fn_ptr[bsize], 1);
} else if (cpi->sf.search_method == HEX) {
bestsme = vp9_hex_search(x, &mvp_full, step_param, sadpb, 1,
&cpi->fn_ptr[bsize], 1,
&ref_mv, &tmp_mv->as_mv);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_var(x, &tmp_mv->as_mv, &ref_mv,
+ &cpi->fn_ptr[bsize], 1);
} else if (cpi->sf.search_method == SQUARE) {
bestsme = vp9_square_search(x, &mvp_full, step_param, sadpb, 1,
&cpi->fn_ptr[bsize], 1,
&ref_mv, &tmp_mv->as_mv);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_var(x, &tmp_mv->as_mv, &ref_mv,
+ &cpi->fn_ptr[bsize], 1);
} else if (cpi->sf.search_method == BIGDIA) {
bestsme = vp9_bigdia_search(x, &mvp_full, step_param, sadpb, 1,
&cpi->fn_ptr[bsize], 1,
&ref_mv, &tmp_mv->as_mv);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_var(x, &tmp_mv->as_mv, &ref_mv,
+ &cpi->fn_ptr[bsize], 1);
} else {
bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
sadpb, further_steps, 1,
// motion search code to be used without additional modifications.
for (i = 0; i < MAX_MB_PLANE; i++)
backup_yv12[ref][i] = xd->plane[i].pre[ref];
- setup_pre_planes(xd, ref, scaled_ref_frame[ref], mi_row, mi_col, NULL);
+ vp9_setup_pre_planes(xd, ref, scaled_ref_frame[ref], mi_row, mi_col,
+ NULL);
}
frame_mv[refs[ref]].as_int = single_newmv[refs[ref]].as_int;
x->nmvjointcost, x->mvcost,
&ref_mv[id].as_mv, second_pred,
pw, ph);
+ if (bestsme < INT_MAX)
+ bestsme = vp9_get_mvpred_av_var(x, &tmp_mv.as_mv, &ref_mv[id].as_mv,
+ second_pred, &cpi->fn_ptr[bsize], 1);
x->mv_col_min = tmp_col_min;
x->mv_col_max = tmp_col_max;
const int mode_search_skip_flags = cpi->sf.mode_search_skip_flags;
const int intra_y_mode_mask =
cpi->sf.intra_y_mode_mask[max_txsize_lookup[bsize]];
+ const int disable_inter_mode_mask = cpi->sf.disable_inter_mode_mask[bsize];
x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
mode_skip_mask |= new_modes_mask;
}
+ if (bsize > cpi->sf.max_intra_bsize) {
+ mode_skip_mask |= 0xFF30808;
+ }
+
for (mode_index = 0; mode_index < MAX_MODES; ++mode_index) {
int mode_excluded = 0;
int64_t this_rd = INT64_MAX;
this_mode = vp9_mode_order[mode_index].mode;
ref_frame = vp9_mode_order[mode_index].ref_frame[0];
if (ref_frame != INTRA_FRAME &&
- cpi->sf.disable_inter_mode_mask[bsize] & (1 << INTER_OFFSET(this_mode)))
+ disable_inter_mode_mask & (1 << INTER_OFFSET(this_mode)))
continue;
second_ref_frame = vp9_mode_order[mode_index].ref_frame[1];
}
} else {
// if we're near/nearest and mv == 0,0, compare to zeromv
- if ((this_mode == NEARMV || this_mode == NEARESTMV ||
+ if (!(disable_inter_mode_mask & (1 << INTER_OFFSET(ZEROMV))) &&
+ (this_mode == NEARMV || this_mode == NEARESTMV ||
this_mode == ZEROMV) &&
frame_mv[this_mode][ref_frame].as_int == 0 &&
!vp9_segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP) &&