Merge "Inlining and removing encoder_common_init()."
diff --git a/test/webm_video_source.h b/test/webm_video_source.h
index dc76bb0..11d3d23 100644
--- a/test/webm_video_source.h
+++ b/test/webm_video_source.h
@@ -37,6 +37,7 @@
virtual ~WebMVideoSource() {
if (vpx_ctx_->file != NULL)
fclose(vpx_ctx_->file);
+ webm_free(webm_ctx_);
delete vpx_ctx_;
delete webm_ctx_;
}
diff --git a/vp9/encoder/vp9_aq_cyclicrefresh.c b/vp9/encoder/vp9_aq_cyclicrefresh.c
index 323c103..d1437d3 100644
--- a/vp9/encoder/vp9_aq_cyclicrefresh.c
+++ b/vp9/encoder/vp9_aq_cyclicrefresh.c
@@ -136,7 +136,8 @@
const int xmis = MIN(cm->mi_cols - mi_col, bw);
const int ymis = MIN(cm->mi_rows - mi_row, bh);
const int block_index = mi_row * cm->mi_cols + mi_col;
- const int refresh_this_block = candidate_refresh_aq(cr, mbmi, bsize, use_rd);
+ const int refresh_this_block = cpi->mb.in_static_area ||
+ candidate_refresh_aq(cr, mbmi, bsize, use_rd);
// Default is to not update the refresh map.
int new_map_value = cr->map[block_index];
int x = 0; int y = 0;
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index cad1241..fcf2a04 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -162,6 +162,8 @@
// note that token_costs is the cost when eob node is skipped
vp9_coeff_cost token_costs[TX_SIZES];
+ int in_static_area;
+
int optimize;
// indicate if it is in the rd search loop or encoding process
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index f88ce2d..9956acc 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -1273,6 +1273,7 @@
static int is_background(VP9_COMP *cpi, const TileInfo *const tile,
int mi_row, int mi_col) {
+ MACROBLOCK *x = &cpi->mb;
uint8_t *src, *pre;
int src_stride, pre_stride;
@@ -1304,7 +1305,8 @@
threshold = (row8x8_remaining * col8x8_remaining) << 6;
}
- return (this_sad < 2 * threshold);
+ x->in_static_area = (this_sad < 2 * threshold);
+ return x->in_static_area;
}
static int sb_has_motion(const VP9_COMMON *cm, MODE_INFO **prev_mi_8x8) {
@@ -2433,6 +2435,11 @@
set_offsets(cpi, tile, mi_row, mi_col, bsize);
xd->mi[0]->mbmi.sb_type = bsize;
+ if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && cm->seg.enabled) {
+ if (xd->mi[0]->mbmi.segment_id && x->in_static_area)
+ x->rdmult = vp9_cyclic_refresh_get_rdmult(cpi->cyclic_refresh);
+ }
+
if (!frame_is_intra_only(cm)) {
vp9_pick_inter_mode(cpi, x, tile, mi_row, mi_col,
rate, dist, bsize);
@@ -2856,7 +2863,8 @@
static void encode_nonrd_sb_row(VP9_COMP *cpi, const TileInfo *const tile,
int mi_row, TOKENEXTRA **tp) {
VP9_COMMON *cm = &cpi->common;
- MACROBLOCKD *xd = &cpi->mb.e_mbd;
+ MACROBLOCK *x = &cpi->mb;
+ MACROBLOCKD *xd = &x->e_mbd;
int mi_col;
// Initialize the left context for the new SB row
@@ -2874,8 +2882,9 @@
MODE_INFO **prev_mi_8x8 = cm->prev_mi_grid_visible + idx_str;
BLOCK_SIZE bsize;
- cpi->mb.source_variance = UINT_MAX;
- vp9_zero(cpi->mb.pred_mv);
+ x->in_static_area = 0;
+ x->source_variance = UINT_MAX;
+ vp9_zero(x->pred_mv);
// Set the partition type of the 64X64 block
switch (cpi->sf.partition_search_type) {
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index f9dba96..1dacfaa 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -61,7 +61,7 @@
#define MIN_GF_INTERVAL 4
#endif
-#define DISABLE_RC_LONG_TERM_MEM
+#define LONG_TERM_VBR_CORRECTION
static void swap_yv12(YV12_BUFFER_CONFIG *a, YV12_BUFFER_CONFIG *b) {
YV12_BUFFER_CONFIG temp = *a;
@@ -1033,6 +1033,9 @@
reset_fpf_position(twopass, start_pos);
}
+
+ // Reset the vbr bits off target counter
+ cpi->rc.vbr_bits_off_target = 0;
}
// This function gives an estimate of how badly we believe the prediction
@@ -2192,6 +2195,23 @@
cpi->rc.frames_to_key = INT_MAX;
}
+// For VBR...adjustment to the frame target based on error from previous frames
+void vbr_rate_correction(int * this_frame_target,
+ const int64_t vbr_bits_off_target) {
+ int max_delta = (*this_frame_target * 15) / 100;
+
+ // vbr_bits_off_target > 0 means we have extra bits to spend
+ if (vbr_bits_off_target > 0) {
+ *this_frame_target +=
+ (vbr_bits_off_target > max_delta) ? max_delta
+ : (int)vbr_bits_off_target;
+ } else {
+ *this_frame_target -=
+ (vbr_bits_off_target < -max_delta) ? max_delta
+ : (int)-vbr_bits_off_target;
+ }
+}
+
void vp9_rc_get_second_pass_params(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
RATE_CONTROL *const rc = &cpi->rc;
@@ -2219,8 +2239,15 @@
return;
if (cpi->refresh_alt_ref_frame) {
+ int modified_target = twopass->gf_bits;
+ rc->base_frame_target = twopass->gf_bits;
cm->frame_type = INTER_FRAME;
- vp9_rc_set_frame_target(cpi, twopass->gf_bits);
+#ifdef LONG_TERM_VBR_CORRECTION
+ // Correction to rate target based on prior over or under shoot.
+ if (cpi->oxcf.end_usage == USAGE_LOCAL_FILE_PLAYBACK)
+ vbr_rate_correction(&modified_target, rc->vbr_bits_off_target);
+#endif
+ vp9_rc_set_frame_target(cpi, modified_target);
return;
}
@@ -2315,6 +2342,13 @@
target = vp9_rc_clamp_iframe_target_size(cpi, rc->this_frame_target);
else
target = vp9_rc_clamp_pframe_target_size(cpi, rc->this_frame_target);
+
+ rc->base_frame_target = target;
+#ifdef LONG_TERM_VBR_CORRECTION
+ // Correction to rate target based on prior over or under shoot.
+ if (cpi->oxcf.end_usage == USAGE_LOCAL_FILE_PLAYBACK)
+ vbr_rate_correction(&target, rc->vbr_bits_off_target);
+#endif
vp9_rc_set_frame_target(cpi, target);
// Update the total stats remaining structure.
@@ -2322,20 +2356,45 @@
}
void vp9_twopass_postencode_update(VP9_COMP *cpi) {
-#ifdef DISABLE_RC_LONG_TERM_MEM
- const uint64_t bits_used = cpi->rc.this_frame_target;
+ RATE_CONTROL *const rc = &cpi->rc;
+#ifdef LONG_TERM_VBR_CORRECTION
+ // In this experimental mode, the VBR correction is done exclusively through
+ // rc->vbr_bits_off_target. Based on the sign of this value, a limited %
+ // adjustment is made to the target rate of subsequent frames, to try and
+ // push it back towards 0. This mode is less likely to suffer from
+ // extreme behaviour at the end of a clip or group of frames.
+ const int bits_used = rc->base_frame_target;
+ rc->vbr_bits_off_target += rc->base_frame_target - rc->projected_frame_size;
#else
- const uint64_t bits_used = cpi->rc.projected_frame_size;
+ // In this mode, VBR correction is acheived by altering bits_left,
+ // kf_group_bits & gf_group_bits to reflect any deviation from the target
+ // rate in this frame. This alters the allocation of bits to the
+ // remaning frames in the group / clip.
+ //
+ // This method can give rise to unstable behaviour near the end of a clip
+ // or kf/gf group of frames where any accumulated error is corrected over an
+ // ever decreasing number of frames. Hence we change the balance of target
+ // vs. actual bitrate gradually as we progress towards the end of the
+ // sequence in order to mitigate this effect.
+ const double progress =
+ (double)(cpi->twopass.stats_in - cpi->twopass.stats_in_start) /
+ (cpi->twopass.stats_in_end - cpi->twopass.stats_in_start);
+ const int bits_used = progress * cpi->rc.this_frame_target +
+ (1.0 - progress) * cpi->rc.projected_frame_size;
#endif
+
cpi->twopass.bits_left -= bits_used;
cpi->twopass.bits_left = MAX(cpi->twopass.bits_left, 0);
- // Update bits left to the kf and gf groups to account for overshoot or
- // undershoot on these frames.
+
+#ifdef LONG_TERM_VBR_CORRECTION
+ if (cpi->common.frame_type != KEY_FRAME) {
+#else
if (cpi->common.frame_type == KEY_FRAME) {
// For key frames kf_group_bits already had the target bits subtracted out.
// So now update to the correct value based on the actual bits used.
cpi->twopass.kf_group_bits += cpi->rc.this_frame_target - bits_used;
} else {
+#endif
cpi->twopass.kf_group_bits -= bits_used;
cpi->twopass.gf_group_bits -= bits_used;
cpi->twopass.gf_group_bits = MAX(cpi->twopass.gf_group_bits, 0);
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index 1bcc955..08a469c 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -131,6 +131,24 @@
cm->fc = cm->frame_contexts[cm->frame_context_idx];
}
+static void setup_frame(VP9_COMP *cpi) {
+ VP9_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 (cm->frame_type == KEY_FRAME) {
+ setup_key_frame(cpi);
+ } else {
+ if (!cm->intra_only && !cm->error_resilient_mode && !cpi->use_svc)
+ cm->frame_context_idx = cpi->refresh_alt_ref_frame;
+ setup_inter_frame(cm);
+ }
+}
+
+
+
void vp9_initialize_enc() {
static int init_done = 0;
@@ -373,122 +391,6 @@
}
}
-static void set_rd_speed_thresholds(VP9_COMP *cpi) {
- int i;
- RD_OPT *const rd = &cpi->rd;
-
- // Set baseline threshold values
- for (i = 0; i < MAX_MODES; ++i)
- rd->thresh_mult[i] = is_best_mode(cpi->oxcf.mode) ? -500 : 0;
-
- rd->thresh_mult[THR_NEARESTMV] = 0;
- rd->thresh_mult[THR_NEARESTG] = 0;
- rd->thresh_mult[THR_NEARESTA] = 0;
-
- rd->thresh_mult[THR_DC] += 1000;
-
- rd->thresh_mult[THR_NEWMV] += 1000;
- rd->thresh_mult[THR_NEWA] += 1000;
- rd->thresh_mult[THR_NEWG] += 1000;
-
- rd->thresh_mult[THR_NEARMV] += 1000;
- rd->thresh_mult[THR_NEARA] += 1000;
- rd->thresh_mult[THR_COMP_NEARESTLA] += 1000;
- rd->thresh_mult[THR_COMP_NEARESTGA] += 1000;
-
- rd->thresh_mult[THR_TM] += 1000;
-
- rd->thresh_mult[THR_COMP_NEARLA] += 1500;
- rd->thresh_mult[THR_COMP_NEWLA] += 2000;
- rd->thresh_mult[THR_NEARG] += 1000;
- rd->thresh_mult[THR_COMP_NEARGA] += 1500;
- rd->thresh_mult[THR_COMP_NEWGA] += 2000;
-
- rd->thresh_mult[THR_ZEROMV] += 2000;
- rd->thresh_mult[THR_ZEROG] += 2000;
- rd->thresh_mult[THR_ZEROA] += 2000;
- rd->thresh_mult[THR_COMP_ZEROLA] += 2500;
- rd->thresh_mult[THR_COMP_ZEROGA] += 2500;
-
- rd->thresh_mult[THR_H_PRED] += 2000;
- rd->thresh_mult[THR_V_PRED] += 2000;
- rd->thresh_mult[THR_D45_PRED ] += 2500;
- rd->thresh_mult[THR_D135_PRED] += 2500;
- rd->thresh_mult[THR_D117_PRED] += 2500;
- rd->thresh_mult[THR_D153_PRED] += 2500;
- rd->thresh_mult[THR_D207_PRED] += 2500;
- rd->thresh_mult[THR_D63_PRED] += 2500;
-
- /* disable frame modes if flags not set */
- if (!(cpi->ref_frame_flags & VP9_LAST_FLAG)) {
- rd->thresh_mult[THR_NEWMV ] = INT_MAX;
- rd->thresh_mult[THR_NEARESTMV] = INT_MAX;
- rd->thresh_mult[THR_ZEROMV ] = INT_MAX;
- rd->thresh_mult[THR_NEARMV ] = INT_MAX;
- }
- if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG)) {
- rd->thresh_mult[THR_NEARESTG ] = INT_MAX;
- rd->thresh_mult[THR_ZEROG ] = INT_MAX;
- rd->thresh_mult[THR_NEARG ] = INT_MAX;
- rd->thresh_mult[THR_NEWG ] = INT_MAX;
- }
- if (!(cpi->ref_frame_flags & VP9_ALT_FLAG)) {
- rd->thresh_mult[THR_NEARESTA ] = INT_MAX;
- rd->thresh_mult[THR_ZEROA ] = INT_MAX;
- rd->thresh_mult[THR_NEARA ] = INT_MAX;
- rd->thresh_mult[THR_NEWA ] = INT_MAX;
- }
-
- if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
- (VP9_LAST_FLAG | VP9_ALT_FLAG)) {
- rd->thresh_mult[THR_COMP_ZEROLA ] = INT_MAX;
- rd->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
- rd->thresh_mult[THR_COMP_NEARLA ] = INT_MAX;
- rd->thresh_mult[THR_COMP_NEWLA ] = INT_MAX;
- }
- if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
- (VP9_GOLD_FLAG | VP9_ALT_FLAG)) {
- rd->thresh_mult[THR_COMP_ZEROGA ] = INT_MAX;
- rd->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
- rd->thresh_mult[THR_COMP_NEARGA ] = INT_MAX;
- rd->thresh_mult[THR_COMP_NEWGA ] = INT_MAX;
- }
-}
-
-static void set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi) {
- const SPEED_FEATURES *const sf = &cpi->sf;
- RD_OPT *const rd = &cpi->rd;
- int i;
-
- for (i = 0; i < MAX_REFS; ++i)
- rd->thresh_mult_sub8x8[i] = is_best_mode(cpi->oxcf.mode) ? -500 : 0;
-
- rd->thresh_mult_sub8x8[THR_LAST] += 2500;
- rd->thresh_mult_sub8x8[THR_GOLD] += 2500;
- rd->thresh_mult_sub8x8[THR_ALTR] += 2500;
- rd->thresh_mult_sub8x8[THR_INTRA] += 2500;
- rd->thresh_mult_sub8x8[THR_COMP_LA] += 4500;
- rd->thresh_mult_sub8x8[THR_COMP_GA] += 4500;
-
- // Check for masked out split cases.
- for (i = 0; i < MAX_REFS; i++)
- if (sf->disable_split_mask & (1 << i))
- rd->thresh_mult_sub8x8[i] = INT_MAX;
-
- // disable mode test if frame flag is not set
- if (!(cpi->ref_frame_flags & VP9_LAST_FLAG))
- rd->thresh_mult_sub8x8[THR_LAST] = INT_MAX;
- if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG))
- rd->thresh_mult_sub8x8[THR_GOLD] = INT_MAX;
- if (!(cpi->ref_frame_flags & VP9_ALT_FLAG))
- rd->thresh_mult_sub8x8[THR_ALTR] = INT_MAX;
- if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
- (VP9_LAST_FLAG | VP9_ALT_FLAG))
- rd->thresh_mult_sub8x8[THR_COMP_LA] = INT_MAX;
- if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
- (VP9_GOLD_FLAG | VP9_ALT_FLAG))
- rd->thresh_mult_sub8x8[THR_COMP_GA] = INT_MAX;
-}
static void set_speed_features(VP9_COMP *cpi) {
#if CONFIG_INTERNAL_STATS
@@ -500,8 +402,8 @@
vp9_set_speed_features(cpi);
// Set rd thresholds based on mode and speed setting
- set_rd_speed_thresholds(cpi);
- set_rd_speed_thresholds_sub8x8(cpi);
+ vp9_set_rd_speed_thresholds(cpi);
+ vp9_set_rd_speed_thresholds_sub8x8(cpi);
cpi->mb.fwd_txm4x4 = vp9_fdct4x4;
if (cpi->oxcf.lossless || cpi->mb.e_mbd.lossless) {
@@ -567,7 +469,6 @@
vp9_setup_pc_tree(&cpi->common, &cpi->mb);
}
-
static void update_frame_size(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->mb.e_mbd;
@@ -785,8 +686,6 @@
cm->display_width = cpi->oxcf.width;
cm->display_height = cpi->oxcf.height;
- cm->lf.sharpness_level = cpi->oxcf.sharpness;
-
if (cpi->initial_width) {
// Increasing the size of the frame beyond the first seen frame, or some
// otherwise signaled maximum size, is not supported.
@@ -1862,8 +1761,8 @@
recon_err = vp9_get_y_sse(cpi->Source, get_frame_new_buffer(cm));
if (cpi->twopass.total_left_stats.coded_error != 0.0)
- fprintf(f, "%10u %10d %10d %10d %10d %10d "
- "%10"PRId64" %10"PRId64" %10d "
+ fprintf(f, "%10u %10d %10d %10d %10d"
+ "%10"PRId64" %10"PRId64" %10"PRId64" %10"PRId64" %10d "
"%7.2lf %7.2lf %7.2lf %7.2lf %7.2lf"
"%6d %6d %5d %5d %5d "
"%10"PRId64" %10.3lf"
@@ -1872,6 +1771,7 @@
cpi->rc.projected_frame_size,
cpi->rc.projected_frame_size / cpi->common.MBs,
(cpi->rc.projected_frame_size - cpi->rc.this_frame_target),
+ cpi->rc.vbr_bits_off_target,
cpi->rc.total_target_vs_actual,
(cpi->oxcf.starting_buffer_level - cpi->rc.bits_off_target),
cpi->rc.total_actual_bits, cm->base_qindex,
@@ -1916,20 +1816,7 @@
VP9_COMMON *const cm = &cpi->common;
vp9_clear_system_state();
vp9_set_quantizer(cm, q);
-
- // 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 (cm->frame_type == KEY_FRAME) {
- setup_key_frame(cpi);
- } else {
- if (!cm->intra_only && !cm->error_resilient_mode && !cpi->use_svc)
- cm->frame_context_idx = cpi->refresh_alt_ref_frame;
-
- setup_inter_frame(cm);
- }
+ setup_frame(cpi);
// Variance adaptive and in frame q adjustment experiments are mutually
// exclusive.
if (cpi->oxcf.aq_mode == VARIANCE_AQ) {
@@ -1974,21 +1861,8 @@
vp9_set_quantizer(cm, q);
- if (loop_count == 0) {
- // 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 (cm->frame_type == KEY_FRAME) {
- setup_key_frame(cpi);
- } else {
- if (!cm->intra_only && !cm->error_resilient_mode && !cpi->use_svc)
- cpi->common.frame_context_idx = cpi->refresh_alt_ref_frame;
-
- setup_inter_frame(cm);
- }
- }
+ if (loop_count == 0)
+ setup_frame(cpi);
// Variance adaptive and in frame q adjustment experiments are mutually
// exclusive.
@@ -2279,7 +2153,6 @@
// Set various flags etc to special state if it is a key frame.
if (frame_is_intra_only(cm)) {
- setup_key_frame(cpi);
// Reset the loop filter deltas and segmentation map.
vp9_reset_segment_features(&cm->seg);
diff --git a/vp9/encoder/vp9_ratectrl.c b/vp9/encoder/vp9_ratectrl.c
index b63d0da..b8d0ec4 100644
--- a/vp9/encoder/vp9_ratectrl.c
+++ b/vp9/encoder/vp9_ratectrl.c
@@ -85,7 +85,7 @@
gf_high_motion_minq[i] = get_minq_index(maxq, 0.0000021, -0.00125, 0.50);
afq_low_motion_minq[i] = get_minq_index(maxq, 0.0000015, -0.0009, 0.33);
afq_high_motion_minq[i] = get_minq_index(maxq, 0.0000021, -0.00125, 0.55);
- inter_minq[i] = get_minq_index(maxq, 0.00000271, -0.00113, 0.55);
+ inter_minq[i] = get_minq_index(maxq, 0.00000271, -0.00113, 0.75);
}
}
@@ -1033,6 +1033,7 @@
RATE_CONTROL *const rc = &cpi->rc;
rc->this_frame_target = target;
+
// Target rate per SB64 (including partial SB64s.
rc->sb64_target_rate = ((int64_t)rc->this_frame_target * 64 * 64) /
(cm->width * cm->height);
diff --git a/vp9/encoder/vp9_ratectrl.h b/vp9/encoder/vp9_ratectrl.h
index fbeca82..8203661 100644
--- a/vp9/encoder/vp9_ratectrl.h
+++ b/vp9/encoder/vp9_ratectrl.h
@@ -27,7 +27,9 @@
typedef struct {
// Rate targetting variables
- int this_frame_target;
+ int base_frame_target; // A baseline frame target before adjustment
+ // for previous under or over shoot.
+ int this_frame_target; // Actual frame target after rc adjustment.
int projected_frame_size;
int sb64_target_rate;
int last_q[3]; // Separate values for Intra/Inter/ARF-GF
@@ -67,6 +69,7 @@
int64_t buffer_level;
int64_t bits_off_target;
+ int64_t vbr_bits_off_target;
int decimation_factor;
int decimation_count;
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 63c099d..f7f2ec6 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -243,9 +243,7 @@
cpi->mb.sadperbit4 = sad_per_bit4lut[qindex];
}
-static void set_block_thresholds(VP9_COMP *cpi) {
- const VP9_COMMON *const cm = &cpi->common;
- RD_OPT *const rd = &cpi->rd;
+static void set_block_thresholds(const VP9_COMMON *cm, RD_OPT *rd) {
int i, bsize, segment_id;
for (segment_id = 0; segment_id < MAX_SEGMENTS; ++segment_id) {
@@ -280,20 +278,21 @@
void vp9_initialize_rd_consts(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCK *const x = &cpi->mb;
+ RD_OPT *const rd = &cpi->rd;
int i;
vp9_clear_system_state();
- cpi->rd.RDDIV = RDDIV_BITS; // in bits (to multiply D by 128)
- cpi->rd.RDMULT = vp9_compute_rd_mult(cpi, cm->base_qindex + cm->y_dc_delta_q);
+ rd->RDDIV = RDDIV_BITS; // in bits (to multiply D by 128)
+ rd->RDMULT = vp9_compute_rd_mult(cpi, cm->base_qindex + cm->y_dc_delta_q);
- x->errorperbit = cpi->rd.RDMULT / RD_MULT_EPB_RATIO;
+ x->errorperbit = rd->RDMULT / RD_MULT_EPB_RATIO;
x->errorperbit += (x->errorperbit == 0);
x->select_txfm_size = (cpi->sf.tx_size_search_method == USE_LARGESTALL &&
cm->frame_type != KEY_FRAME) ? 0 : 1;
- set_block_thresholds(cpi);
+ set_block_thresholds(cm, rd);
if (!cpi->sf.use_nonrd_pick_mode || cm->frame_type == KEY_FRAME) {
fill_token_costs(x->token_costs, cm->fc.coef_probs);
@@ -2178,12 +2177,12 @@
x->pred_mv_sad[ref_frame] = best_sad;
}
-static void estimate_ref_frame_costs(VP9_COMP *cpi, int segment_id,
+static void estimate_ref_frame_costs(const VP9_COMMON *cm,
+ const MACROBLOCKD *xd,
+ int segment_id,
unsigned int *ref_costs_single,
unsigned int *ref_costs_comp,
vp9_prob *comp_mode_p) {
- VP9_COMMON *const cm = &cpi->common;
- MACROBLOCKD *const xd = &cpi->mb.e_mbd;
int seg_ref_active = vp9_segfeature_active(&cm->seg, segment_id,
SEG_LVL_REF_FRAME);
if (seg_ref_active) {
@@ -2348,7 +2347,7 @@
int mi_row, int mi_col,
int_mv *tmp_mv, int *rate_mv) {
MACROBLOCKD *xd = &x->e_mbd;
- VP9_COMMON *cm = &cpi->common;
+ const VP9_COMMON *cm = &cpi->common;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
struct buf_2d backup_yv12[MAX_MB_PLANE] = {{0}};
int bestsme = INT_MAX;
@@ -2386,7 +2385,7 @@
// Work out the size of the first step in the mv step search.
// 0 here is maximum length first step. 1 is MAX >> 1 etc.
- if (cpi->sf.auto_mv_step_size && cpi->common.show_frame) {
+ if (cpi->sf.auto_mv_step_size && cm->show_frame) {
// Take wtd average of the step_params based on the last frame's
// max mv magnitude and that based on the best ref mvs of the current
// block for the given reference.
@@ -2397,7 +2396,7 @@
}
if (cpi->sf.adaptive_motion_search && bsize < BLOCK_64X64 &&
- cpi->common.show_frame) {
+ cm->show_frame) {
int boffset = 2 * (b_width_log2(BLOCK_64X64) - MIN(b_height_log2(bsize),
b_width_log2(bsize)));
step_param = MAX(step_param, boffset);
@@ -2412,7 +2411,7 @@
if (tlevel < 5)
step_param += 2;
- for (i = LAST_FRAME; i <= ALTREF_FRAME && cpi->common.show_frame; ++i) {
+ for (i = LAST_FRAME; i <= ALTREF_FRAME && cm->show_frame; ++i) {
if ((x->pred_mv_sad[ref] >> 3) > x->pred_mv_sad[i]) {
x->pred_mv[ref].as_int = 0;
tmp_mv->as_int = INVALID_MV;
@@ -2496,7 +2495,7 @@
*rate_mv = vp9_mv_bit_cost(&tmp_mv->as_mv, &ref_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
- if (cpi->sf.adaptive_motion_search && cpi->common.show_frame)
+ if (cpi->sf.adaptive_motion_search && cm->show_frame)
x->pred_mv[ref].as_int = tmp_mv->as_int;
if (scaled_ref_frame) {
@@ -3104,13 +3103,13 @@
// combination that wins out.
static void update_rd_thresh_fact(VP9_COMP *cpi, int bsize,
int best_mode_index) {
- if (cpi->sf.adaptive_rd_thresh) {
+ if (cpi->sf.adaptive_rd_thresh > 0) {
const int top_mode = bsize < BLOCK_8X8 ? MAX_REFS : MAX_MODES;
- int mode_index;
- for (mode_index = 0; mode_index < top_mode; ++mode_index) {
- int *const fact = &cpi->rd.thresh_freq_fact[bsize][mode_index];
+ int mode;
+ for (mode = 0; mode < top_mode; ++mode) {
+ int *const fact = &cpi->rd.thresh_freq_fact[bsize][mode];
- if (mode_index == best_mode_index) {
+ if (mode == best_mode_index) {
*fact -= (*fact >> 3);
} else {
*fact = MIN(*fact + RD_THRESH_INC,
@@ -3178,7 +3177,7 @@
x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
- estimate_ref_frame_costs(cpi, segment_id, ref_costs_single, ref_costs_comp,
+ estimate_ref_frame_costs(cm, xd, segment_id, ref_costs_single, ref_costs_comp,
&comp_mode_p);
for (i = 0; i < REFERENCE_MODES; ++i)
@@ -3792,7 +3791,7 @@
seg_mvs[i][j].as_int = INVALID_MV;
}
- estimate_ref_frame_costs(cpi, segment_id, ref_costs_single, ref_costs_comp,
+ estimate_ref_frame_costs(cm, xd, segment_id, ref_costs_single, ref_costs_comp,
&comp_mode_p);
for (i = 0; i < REFERENCE_MODES; ++i)
@@ -4374,3 +4373,120 @@
return best_rd;
}
+
+void vp9_set_rd_speed_thresholds(VP9_COMP *cpi) {
+ int i;
+ RD_OPT *const rd = &cpi->rd;
+
+ // Set baseline threshold values
+ for (i = 0; i < MAX_MODES; ++i)
+ rd->thresh_mult[i] = is_best_mode(cpi->oxcf.mode) ? -500 : 0;
+
+ rd->thresh_mult[THR_NEARESTMV] = 0;
+ rd->thresh_mult[THR_NEARESTG] = 0;
+ rd->thresh_mult[THR_NEARESTA] = 0;
+
+ rd->thresh_mult[THR_DC] += 1000;
+
+ rd->thresh_mult[THR_NEWMV] += 1000;
+ rd->thresh_mult[THR_NEWA] += 1000;
+ rd->thresh_mult[THR_NEWG] += 1000;
+
+ rd->thresh_mult[THR_NEARMV] += 1000;
+ rd->thresh_mult[THR_NEARA] += 1000;
+ rd->thresh_mult[THR_COMP_NEARESTLA] += 1000;
+ rd->thresh_mult[THR_COMP_NEARESTGA] += 1000;
+
+ rd->thresh_mult[THR_TM] += 1000;
+
+ rd->thresh_mult[THR_COMP_NEARLA] += 1500;
+ rd->thresh_mult[THR_COMP_NEWLA] += 2000;
+ rd->thresh_mult[THR_NEARG] += 1000;
+ rd->thresh_mult[THR_COMP_NEARGA] += 1500;
+ rd->thresh_mult[THR_COMP_NEWGA] += 2000;
+
+ rd->thresh_mult[THR_ZEROMV] += 2000;
+ rd->thresh_mult[THR_ZEROG] += 2000;
+ rd->thresh_mult[THR_ZEROA] += 2000;
+ rd->thresh_mult[THR_COMP_ZEROLA] += 2500;
+ rd->thresh_mult[THR_COMP_ZEROGA] += 2500;
+
+ rd->thresh_mult[THR_H_PRED] += 2000;
+ rd->thresh_mult[THR_V_PRED] += 2000;
+ rd->thresh_mult[THR_D45_PRED ] += 2500;
+ rd->thresh_mult[THR_D135_PRED] += 2500;
+ rd->thresh_mult[THR_D117_PRED] += 2500;
+ rd->thresh_mult[THR_D153_PRED] += 2500;
+ rd->thresh_mult[THR_D207_PRED] += 2500;
+ rd->thresh_mult[THR_D63_PRED] += 2500;
+
+ /* disable frame modes if flags not set */
+ if (!(cpi->ref_frame_flags & VP9_LAST_FLAG)) {
+ rd->thresh_mult[THR_NEWMV ] = INT_MAX;
+ rd->thresh_mult[THR_NEARESTMV] = INT_MAX;
+ rd->thresh_mult[THR_ZEROMV ] = INT_MAX;
+ rd->thresh_mult[THR_NEARMV ] = INT_MAX;
+ }
+ if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG)) {
+ rd->thresh_mult[THR_NEARESTG ] = INT_MAX;
+ rd->thresh_mult[THR_ZEROG ] = INT_MAX;
+ rd->thresh_mult[THR_NEARG ] = INT_MAX;
+ rd->thresh_mult[THR_NEWG ] = INT_MAX;
+ }
+ if (!(cpi->ref_frame_flags & VP9_ALT_FLAG)) {
+ rd->thresh_mult[THR_NEARESTA ] = INT_MAX;
+ rd->thresh_mult[THR_ZEROA ] = INT_MAX;
+ rd->thresh_mult[THR_NEARA ] = INT_MAX;
+ rd->thresh_mult[THR_NEWA ] = INT_MAX;
+ }
+
+ if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
+ (VP9_LAST_FLAG | VP9_ALT_FLAG)) {
+ rd->thresh_mult[THR_COMP_ZEROLA ] = INT_MAX;
+ rd->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
+ rd->thresh_mult[THR_COMP_NEARLA ] = INT_MAX;
+ rd->thresh_mult[THR_COMP_NEWLA ] = INT_MAX;
+ }
+ if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
+ (VP9_GOLD_FLAG | VP9_ALT_FLAG)) {
+ rd->thresh_mult[THR_COMP_ZEROGA ] = INT_MAX;
+ rd->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
+ rd->thresh_mult[THR_COMP_NEARGA ] = INT_MAX;
+ rd->thresh_mult[THR_COMP_NEWGA ] = INT_MAX;
+ }
+}
+
+void vp9_set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi) {
+ const SPEED_FEATURES *const sf = &cpi->sf;
+ RD_OPT *const rd = &cpi->rd;
+ int i;
+
+ for (i = 0; i < MAX_REFS; ++i)
+ rd->thresh_mult_sub8x8[i] = is_best_mode(cpi->oxcf.mode) ? -500 : 0;
+
+ rd->thresh_mult_sub8x8[THR_LAST] += 2500;
+ rd->thresh_mult_sub8x8[THR_GOLD] += 2500;
+ rd->thresh_mult_sub8x8[THR_ALTR] += 2500;
+ rd->thresh_mult_sub8x8[THR_INTRA] += 2500;
+ rd->thresh_mult_sub8x8[THR_COMP_LA] += 4500;
+ rd->thresh_mult_sub8x8[THR_COMP_GA] += 4500;
+
+ // Check for masked out split cases.
+ for (i = 0; i < MAX_REFS; i++)
+ if (sf->disable_split_mask & (1 << i))
+ rd->thresh_mult_sub8x8[i] = INT_MAX;
+
+ // disable mode test if frame flag is not set
+ if (!(cpi->ref_frame_flags & VP9_LAST_FLAG))
+ rd->thresh_mult_sub8x8[THR_LAST] = INT_MAX;
+ if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG))
+ rd->thresh_mult_sub8x8[THR_GOLD] = INT_MAX;
+ if (!(cpi->ref_frame_flags & VP9_ALT_FLAG))
+ rd->thresh_mult_sub8x8[THR_ALTR] = INT_MAX;
+ if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
+ (VP9_LAST_FLAG | VP9_ALT_FLAG))
+ rd->thresh_mult_sub8x8[THR_COMP_LA] = INT_MAX;
+ if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
+ (VP9_GOLD_FLAG | VP9_ALT_FLAG))
+ rd->thresh_mult_sub8x8[THR_COMP_GA] = INT_MAX;
+}
diff --git a/vp9/encoder/vp9_rdopt.h b/vp9/encoder/vp9_rdopt.h
index a01dbd4..e485664 100644
--- a/vp9/encoder/vp9_rdopt.h
+++ b/vp9/encoder/vp9_rdopt.h
@@ -83,6 +83,10 @@
ENTROPY_CONTEXT t_above[16],
ENTROPY_CONTEXT t_left[16]);
+void vp9_set_rd_speed_thresholds(VP9_COMP *cpi);
+
+void vp9_set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi);
+
#ifdef __cplusplus
} // extern "C"
#endif
diff --git a/vp9/encoder/vp9_variance.c b/vp9/encoder/vp9_variance.c
index 71867a9..a31a476 100644
--- a/vp9/encoder/vp9_variance.c
+++ b/vp9/encoder/vp9_variance.c
@@ -18,63 +18,34 @@
#include "vp9/encoder/vp9_variance.h"
-void variance(const uint8_t *src_ptr,
- int source_stride,
- const uint8_t *ref_ptr,
- int recon_stride,
- int w,
- int h,
- unsigned int *sse,
- int *sum) {
+void 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;
- int diff;
*sum = 0;
*sse = 0;
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
- diff = src_ptr[j] - ref_ptr[j];
+ const int diff = a[j] - b[j];
*sum += diff;
*sse += diff * diff;
}
- src_ptr += source_stride;
- ref_ptr += recon_stride;
+ a += a_stride;
+ b += b_stride;
}
}
-/****************************************************************************
- *
- * ROUTINE : filter_block2d_bil_first_pass
- *
- * INPUTS : uint8_t *src_ptr : Pointer to source block.
- * uint32_t src_pixels_per_line : Stride of input block.
- * uint32_t pixel_step : Offset between filter input
- * samples (see notes).
- * uint32_t output_height : Input block height.
- * uint32_t output_width : Input block width.
- * int32_t *vp9_filter : Array of 2 bi-linear filter
- * taps.
- *
- * OUTPUTS : int32_t *output_ptr : Pointer to filtered block.
- *
- * RETURNS : void
- *
- * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
- * either horizontal or vertical direction to produce the
- * filtered output block. Used to implement first-pass
- * of 2-D separable filter.
- *
- * SPECIAL NOTES : Produces int32_t output to retain precision for next pass.
- * Two filter taps should sum to VP9_FILTER_WEIGHT.
- * pixel_step defines whether the filter is applied
- * horizontally (pixel_step=1) or vertically (pixel_step=
- * stride).
- * It defines the offset required to move from one input
- * to the next.
- *
- ****************************************************************************/
+// Applies a 1-D 2-tap bi-linear filter to the source block in either horizontal
+// or vertical direction to produce the filtered output block. Used to implement
+// first-pass of 2-D separable filter.
+//
+// Produces int32_t output to retain precision for next pass. Two filter taps
+// should sum to VP9_FILTER_WEIGHT. pixel_step defines whether the filter is
+// applied horizontally (pixel_step=1) or vertically (pixel_step=stride). It
+// defines the offset required to move from one input to the next.
static void var_filter_block2d_bil_first_pass(const uint8_t *src_ptr,
uint16_t *output_ptr,
unsigned int src_pixels_per_line,
@@ -99,38 +70,14 @@
}
}
-/****************************************************************************
- *
- * ROUTINE : filter_block2d_bil_second_pass
- *
- * INPUTS : int32_t *src_ptr : Pointer to source block.
- * uint32_t src_pixels_per_line : Stride of input block.
- * uint32_t pixel_step : Offset between filter input
- * samples (see notes).
- * uint32_t output_height : Input block height.
- * uint32_t output_width : Input block width.
- * int32_t *vp9_filter : Array of 2 bi-linear filter
- * taps.
- *
- * OUTPUTS : uint16_t *output_ptr : Pointer to filtered block.
- *
- * RETURNS : void
- *
- * FUNCTION : Applies a 1-D 2-tap bi-linear filter to the source block in
- * either horizontal or vertical direction to produce the
- * filtered output block. Used to implement second-pass
- * of 2-D separable filter.
- *
- * SPECIAL NOTES : Requires 32-bit input as produced by
- * filter_block2d_bil_first_pass.
- * Two filter taps should sum to VP9_FILTER_WEIGHT.
- * pixel_step defines whether the filter is applied
- * horizontally (pixel_step=1) or vertically (pixel_step=
- * stride).
- * It defines the offset required to move from one input
- * to the next.
- *
- ****************************************************************************/
+// Applies a 1-D 2-tap bi-linear filter to the source block in either horizontal
+// or vertical direction to produce the filtered output block. Used to implement
+// second-pass of 2-D separable filter.
+//
+// Requires 32-bit input as produced by filter_block2d_bil_first_pass. Two
+// filter taps should sum to VP9_FILTER_WEIGHT. pixel_step defines whether the
+// filter is applied horizontally (pixel_step=1) or vertically (pixel_step=
+// stride). It defines the offset required to move from one input to the next.
static void var_filter_block2d_bil_second_pass(const uint16_t *src_ptr,
uint8_t *output_ptr,
unsigned int src_pixels_per_line,
@@ -156,9 +103,8 @@
unsigned int vp9_get_mb_ss_c(const int16_t *src_ptr) {
unsigned int i, sum = 0;
- for (i = 0; i < 256; i++) {
- sum += (src_ptr[i] * src_ptr[i]);
- }
+ for (i = 0; i < 256; i++)
+ sum += src_ptr[i] * src_ptr[i];
return sum;
}
@@ -183,12 +129,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
+ uint16_t fdata3[65 * 64];
uint8_t temp2[68 * 64];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 33, 64, hfilter);
@@ -205,13 +149,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
+ uint16_t fdata3[65 * 64];
uint8_t temp2[68 * 64];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 33, 64, hfilter);
@@ -240,12 +182,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
+ uint16_t fdata3[65 * 64];
uint8_t temp2[68 * 64];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 65, 32, hfilter);
@@ -262,13 +202,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
+ uint16_t fdata3[65 * 64];
uint8_t temp2[68 * 64];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 64); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 64);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 65, 32, hfilter);
@@ -297,12 +235,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
+ uint16_t fdata3[33 * 32];
uint8_t temp2[36 * 32];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 17, 32, hfilter);
@@ -319,13 +255,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
+ uint16_t fdata3[33 * 32];
uint8_t temp2[36 * 32];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 16); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 16);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 17, 32, hfilter);
@@ -354,12 +288,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
+ uint16_t fdata3[33 * 32];
uint8_t temp2[36 * 32];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 33, 16, hfilter);
@@ -376,13 +308,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
+ uint16_t fdata3[33 * 32];
uint8_t temp2[36 * 32];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 32); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 32);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 33, 16, hfilter);
@@ -582,17 +512,12 @@
int dst_pixels_per_line,
unsigned int *sse) {
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
- uint16_t fdata3[5 * 4]; // Temp data buffer used in filtering
+ uint16_t fdata3[5 * 4];
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
-
- // First filter 1d Horizontal
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 5, 4, hfilter);
-
- // Now filter Verticaly
var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, vfilter);
return vp9_variance4x4(temp2, 4, dst_ptr, dst_pixels_per_line, sse);
@@ -607,18 +532,13 @@
unsigned int *sse,
const uint8_t *second_pred) {
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 4); // compound pred buffer
- uint16_t fdata3[5 * 4]; // Temp data buffer used in filtering
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 4);
+ uint16_t fdata3[5 * 4];
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
-
- // First filter 1d Horizontal
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 5, 4, hfilter);
-
- // Now filter Verticaly
var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, vfilter);
vp9_comp_avg_pred(temp3, second_pred, 4, 4, temp2, 4);
return vp9_variance4x4(temp3, 4, dst_ptr, dst_pixels_per_line, sse);
@@ -631,17 +551,14 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[9 * 8]; // Temp data buffer used in filtering
+ uint16_t fdata3[9 * 8];
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 9, 8, hfilter);
var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 8, 8, vfilter);
-
return vp9_variance8x8(temp2, 8, dst_ptr, dst_pixels_per_line, sse);
}
@@ -653,18 +570,17 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[9 * 8]; // Temp data buffer used in filtering
+ uint16_t fdata3[9 * 8];
uint8_t temp2[20 * 16];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 8); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 8);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 9, 8, hfilter);
var_filter_block2d_bil_second_pass(fdata3, temp2, 8, 8, 8, 8, vfilter);
vp9_comp_avg_pred(temp3, second_pred, 8, 8, temp2, 8);
+
return vp9_variance8x8(temp3, 8, dst_ptr, dst_pixels_per_line, sse);
}
@@ -675,12 +591,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[17 * 16]; // Temp data buffer used in filtering
+ uint16_t fdata3[17 * 16];
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 17, 16, hfilter);
@@ -699,11 +613,9 @@
const uint8_t *second_pred) {
uint16_t fdata3[17 * 16];
uint8_t temp2[20 * 16];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 16); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 16);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 17, 16, hfilter);
@@ -722,10 +634,8 @@
unsigned int *sse) {
uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
uint8_t temp2[68 * 64];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 65, 64, hfilter);
@@ -742,13 +652,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[65 * 64]; // Temp data buffer used in filtering
+ uint16_t fdata3[65 * 64];
uint8_t temp2[68 * 64];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 64 * 64);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 65, 64, hfilter);
@@ -764,12 +672,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
+ uint16_t fdata3[33 * 32];
uint8_t temp2[36 * 32];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 33, 32, hfilter);
@@ -786,13 +692,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[33 * 32]; // Temp data buffer used in filtering
+ uint16_t fdata3[33 * 32];
uint8_t temp2[36 * 32];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 32); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 32 * 32);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 33, 32, hfilter);
@@ -928,12 +832,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[16 * 9]; // Temp data buffer used in filtering
+ uint16_t fdata3[16 * 9];
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 9, 16, hfilter);
@@ -950,13 +852,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[16 * 9]; // Temp data buffer used in filtering
+ uint16_t fdata3[16 * 9];
uint8_t temp2[20 * 16];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 8); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 16 * 8);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 9, 16, hfilter);
@@ -974,10 +874,8 @@
unsigned int *sse) {
uint16_t fdata3[9 * 16]; // Temp data buffer used in filtering
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 17, 8, hfilter);
@@ -994,13 +892,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[9 * 16]; // Temp data buffer used in filtering
+ uint16_t fdata3[9 * 16];
uint8_t temp2[20 * 16];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 16); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 16);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 17, 8, hfilter);
@@ -1016,12 +912,10 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[8 * 5]; // Temp data buffer used in filtering
+ uint16_t fdata3[8 * 5];
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 5, 8, hfilter);
@@ -1038,13 +932,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[8 * 5]; // Temp data buffer used in filtering
+ uint16_t fdata3[8 * 5];
uint8_t temp2[20 * 16];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 4); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 8 * 4);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 5, 8, hfilter);
@@ -1060,14 +952,12 @@
const uint8_t *dst_ptr,
int dst_pixels_per_line,
unsigned int *sse) {
- uint16_t fdata3[5 * 8]; // Temp data buffer used in filtering
+ uint16_t fdata3[5 * 8];
// FIXME(jingning,rbultje): this temp2 buffer probably doesn't need to be
// of this big? same issue appears in all other block size settings.
uint8_t temp2[20 * 16];
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 9, 4, hfilter);
@@ -1084,13 +974,11 @@
int dst_pixels_per_line,
unsigned int *sse,
const uint8_t *second_pred) {
- uint16_t fdata3[5 * 8]; // Temp data buffer used in filtering
+ uint16_t fdata3[5 * 8];
uint8_t temp2[20 * 16];
- DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 8); // compound pred buffer
- const int16_t *hfilter, *vfilter;
-
- hfilter = BILINEAR_FILTERS_2TAP(xoffset);
- vfilter = BILINEAR_FILTERS_2TAP(yoffset);
+ DECLARE_ALIGNED_ARRAY(16, uint8_t, temp3, 4 * 8);
+ const int16_t *const hfilter = BILINEAR_FILTERS_2TAP(xoffset);
+ const int16_t *const vfilter = BILINEAR_FILTERS_2TAP(yoffset);
var_filter_block2d_bil_first_pass(src_ptr, fdata3, src_pixels_per_line,
1, 9, 4, hfilter);
@@ -1106,9 +994,8 @@
for (i = 0; i < height; i++) {
for (j = 0; j < width; j++) {
- int tmp;
- tmp = pred[j] + ref[j];
- comp_pred[j] = (tmp + 1) >> 1;
+ const int tmp = pred[j] + ref[j];
+ comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
}
comp_pred += width;
pred += width;
diff --git a/vp9/encoder/vp9_variance.h b/vp9/encoder/vp9_variance.h
index c9e39a1..4c8be71 100644
--- a/vp9/encoder/vp9_variance.h
+++ b/vp9/encoder/vp9_variance.h
@@ -17,14 +17,10 @@
extern "C" {
#endif
-void variance(const uint8_t *src_ptr,
- int source_stride,
- const uint8_t *ref_ptr,
- int recon_stride,
- int w,
- int h,
- unsigned int *sse,
- int *sum);
+void variance(const uint8_t *a, int a_stride,
+ const uint8_t *b, int b_stride,
+ int w, int h,
+ unsigned int *sse, int *sum);
typedef unsigned int(*vp9_sad_fn_t)(const uint8_t *src_ptr,
int source_stride,
