Merge "Fix compile error with --enable-accounting" into nextgenv2
diff --git a/aomenc.c b/aomenc.c
index 742f264..e32a922 100644
--- a/aomenc.c
+++ b/aomenc.c
@@ -1860,9 +1860,7 @@
uint64_t cx_time = 0;
int stream_cnt = 0;
int res = 0;
-#if CONFIG_AOM_HIGHBITDEPTH
int profile_updated = 0;
-#endif
memset(&input, 0, sizeof(input));
exec_name = argv_[0];
@@ -1966,10 +1964,25 @@
{ stream->config.cfg.g_input_bit_depth = input.bit_depth; });
}
-#if CONFIG_AOM_HIGHBITDEPTH
- /* Automatically set the codec bit depth to match the input bit depth.
- * Upgrade the profile if required. */
FOREACH_STREAM({
+ if (input.fmt != AOM_IMG_FMT_I420 && input.fmt != AOM_IMG_FMT_I42016) {
+ /* Automatically upgrade if input is non-4:2:0 but a 4:2:0 profile
+ was selected. */
+ switch (stream->config.cfg.g_profile) {
+ case 0:
+ stream->config.cfg.g_profile = 1;
+ profile_updated = 1;
+ break;
+ case 2:
+ stream->config.cfg.g_profile = 3;
+ profile_updated = 1;
+ break;
+ default: break;
+ }
+ }
+#if CONFIG_AOM_HIGHBITDEPTH
+ /* Automatically set the codec bit depth to match the input bit depth.
+ * Upgrade the profile if required. */
if (stream->config.cfg.g_input_bit_depth >
(unsigned int)stream->config.cfg.g_bit_depth) {
stream->config.cfg.g_bit_depth = stream->config.cfg.g_input_bit_depth;
@@ -1990,6 +2003,7 @@
if (stream->config.cfg.g_profile > 1) {
stream->config.use_16bit_internal = 1;
}
+#endif
if (profile_updated) {
fprintf(stderr,
"Warning: automatically upgrading to profile %d to "
@@ -1997,7 +2011,6 @@
stream->config.cfg.g_profile);
}
});
-#endif
FOREACH_STREAM(set_stream_dimensions(stream, input.width, input.height));
FOREACH_STREAM(validate_stream_config(stream, &global));
diff --git a/av1/av1_cx_iface.c b/av1/av1_cx_iface.c
index bacb23c..a7eb71e 100644
--- a/av1/av1_cx_iface.c
+++ b/av1/av1_cx_iface.c
@@ -425,7 +425,6 @@
oxcf->key_freq = cfg->kf_max_dist;
oxcf->speed = abs(extra_cfg->cpu_used);
- oxcf->encode_breakout = extra_cfg->static_thresh;
oxcf->enable_auto_arf = extra_cfg->enable_auto_alt_ref;
#if CONFIG_EXT_REFS
oxcf->enable_auto_brf = extra_cfg->enable_auto_bwd_ref;
@@ -505,7 +504,6 @@
printf("lag_in_frames: %d\n", oxcf->lag_in_frames);
printf("enable_auto_arf: %d\n", oxcf->enable_auto_arf);
printf("Version: %d\n", oxcf->Version);
- printf("encode_breakout: %d\n", oxcf->encode_breakout);
printf("error resilient: %d\n", oxcf->error_resilient_mode);
printf("frame parallel detokenization: %d\n",
oxcf->frame_parallel_decoding_mode);
diff --git a/av1/common/entropy.c b/av1/common/entropy.c
index 7179e3d..d2de412 100644
--- a/av1/common/entropy.c
+++ b/av1/common/entropy.c
@@ -12,6 +12,7 @@
#include "av1/common/entropy.h"
#include "av1/common/blockd.h"
#include "av1/common/onyxc_int.h"
+#include "av1/common/scan.h"
#include "av1/common/entropymode.h"
#include "aom_mem/aom_mem.h"
#include "aom/aom_integer.h"
@@ -2842,6 +2843,10 @@
#endif // CONFIG_RANS
}
+#if CONFIG_ADAPT_SCAN
+#define ADAPT_SCAN_UPDATE_RATE_16 (1 << 13)
+#endif
+
static void adapt_coef_probs(AV1_COMMON *cm, TX_SIZE tx_size,
unsigned int count_sat,
unsigned int update_factor) {
@@ -2881,11 +2886,15 @@
}
void av1_adapt_coef_probs(AV1_COMMON *cm) {
- TX_SIZE t;
+ TX_SIZE tx_size;
unsigned int count_sat, update_factor;
+#if CONFIG_ADAPT_SCAN
+ TX_TYPE tx_type;
+#endif
+
#if CONFIG_ENTROPY
- if (cm->last_frame_type == KEY_FRAME) {
+ if (!frame_is_intra_only(cm) && cm->last_frame_type == KEY_FRAME) {
update_factor = COEF_MAX_UPDATE_FACTOR_AFTER_KEY_BITS; /* adapt quickly */
count_sat = COEF_COUNT_SAT_AFTER_KEY_BITS;
} else {
@@ -2896,7 +2905,7 @@
update_factor = COEF_MAX_UPDATE_FACTOR_BITS;
}
#else
- if (cm->last_frame_type == KEY_FRAME) {
+ if (!frame_is_intra_only(cm) && cm->last_frame_type == KEY_FRAME) {
update_factor = COEF_MAX_UPDATE_FACTOR_AFTER_KEY; /* adapt quickly */
count_sat = COEF_COUNT_SAT_AFTER_KEY;
} else {
@@ -2904,11 +2913,19 @@
count_sat = COEF_COUNT_SAT;
}
#endif // CONFIG_ENTROPY
- for (t = TX_4X4; t <= TX_32X32; t++)
- adapt_coef_probs(cm, t, count_sat, update_factor);
+ for (tx_size = TX_4X4; tx_size <= TX_32X32; tx_size++)
+ adapt_coef_probs(cm, tx_size, count_sat, update_factor);
#if CONFIG_RANS
av1_coef_pareto_cdfs(cm->fc);
#endif // CONFIG_RANS
+
+#if CONFIG_ADAPT_SCAN
+ for (tx_size = TX_4X4; tx_size < TX_SIZES; ++tx_size)
+ for (tx_type = TX_4X4; tx_type < TX_TYPES; ++tx_type) {
+ av1_update_scan_prob(cm, tx_size, tx_type, ADAPT_SCAN_UPDATE_RATE_16);
+ av1_update_scan_order_facade(cm, tx_size, tx_type);
+ }
+#endif
}
#if CONFIG_ENTROPY
diff --git a/av1/common/entropy.h b/av1/common/entropy.h
index 15b50db..28556fc 100644
--- a/av1/common/entropy.h
+++ b/av1/common/entropy.h
@@ -288,6 +288,10 @@
#endif // CONFIG_ENTROPY
+#if CONFIG_ADAPT_SCAN
+#define ADAPT_SCAN_UPDATE_RATE_16 (1 << 13)
+#endif
+
static INLINE aom_prob av1_merge_probs(aom_prob pre_prob,
const unsigned int ct[2],
unsigned int count_sat,
diff --git a/av1/common/entropymode.c b/av1/common/entropymode.c
index e812f15..e25dcf8 100644
--- a/av1/common/entropymode.c
+++ b/av1/common/entropymode.c
@@ -12,6 +12,7 @@
#include "aom_mem/aom_mem.h"
#include "av1/common/reconinter.h"
+#include "av1/common/scan.h"
#include "av1/common/onyxc_int.h"
#include "av1/common/seg_common.h"
@@ -1755,6 +1756,9 @@
av1_default_coef_probs(cm);
init_mode_probs(cm->fc);
av1_init_mv_probs(cm);
+#if CONFIG_ADAPT_SCAN
+ av1_init_scan_order(cm);
+#endif
cm->fc->initialized = 1;
if (cm->frame_type == KEY_FRAME || cm->error_resilient_mode ||
diff --git a/av1/common/filter.h b/av1/common/filter.h
index 0ac52a9..15fc806 100644
--- a/av1/common/filter.h
+++ b/av1/common/filter.h
@@ -33,8 +33,12 @@
#define SUPPORT_NONINTERPOLATING_FILTERS 0 /* turn on for experimentation */
#define SWITCHABLE_FILTERS 5 /* Number of switchable filters */
+#define LOG_SWITCHABLE_FILTERS \
+ 3 /* (1 << LOG_SWITCHABLE_FILTERS) > SWITCHABLE_FILTERS */
#else
#define SWITCHABLE_FILTERS 3 /* Number of switchable filters */
+#define LOG_SWITCHABLE_FILTERS \
+ 2 /* (1 << LOG_SWITCHABLE_FILTERS) > SWITCHABLE_FILTERS */
#endif // CONFIG_EXT_INTERP
#define USE_TEMPORALFILTER_12TAP 1
diff --git a/av1/common/idct.c b/av1/common/idct.c
index eedbc79..4f33f9b 100644
--- a/av1/common/idct.c
+++ b/av1/common/idct.c
@@ -837,8 +837,10 @@
if (eob == 1)
// DC only DCT coefficient
aom_idct8x8_1_add(input, dest, stride);
+#if !CONFIG_ADAPT_SCAN
else if (eob <= 12)
aom_idct8x8_12_add(input, dest, stride);
+#endif
else
aom_idct8x8_64_add(input, dest, stride);
}
@@ -849,19 +851,22 @@
* coefficients. Use eobs to separate different cases. */
if (eob == 1) /* DC only DCT coefficient. */
aom_idct16x16_1_add(input, dest, stride);
+#if !CONFIG_ADAPT_SCAN
else if (eob <= 10)
aom_idct16x16_10_add(input, dest, stride);
+#endif
else
aom_idct16x16_256_add(input, dest, stride);
}
void av1_idct32x32_add(const tran_low_t *input, uint8_t *dest, int stride,
int eob) {
- if (eob == 1)
- aom_idct32x32_1_add(input, dest, stride);
+ if (eob == 1) aom_idct32x32_1_add(input, dest, stride);
+#if !CONFIG_ADAPT_SCAN
else if (eob <= 34)
// non-zero coeff only in upper-left 8x8
aom_idct32x32_34_add(input, dest, stride);
+#endif
else
aom_idct32x32_1024_add(input, dest, stride);
}
@@ -1659,13 +1664,13 @@
// TODO(yunqingwang): "eobs = 1" case is also handled in av1_short_idct8x8_c.
// Combine that with code here.
// DC only DCT coefficient
- if (eob == 1) {
- aom_highbd_idct8x8_1_add(input, dest, stride, bd);
- } else if (eob <= 10) {
+ if (eob == 1) aom_highbd_idct8x8_1_add(input, dest, stride, bd);
+#if !CONFIG_ADAPT_SCAN
+ else if (eob <= 10)
aom_highbd_idct8x8_10_add(input, dest, stride, bd);
- } else {
+#endif
+ else
aom_highbd_idct8x8_64_add(input, dest, stride, bd);
- }
}
void av1_highbd_idct16x16_add(const tran_low_t *input, uint8_t *dest,
@@ -1673,25 +1678,25 @@
// The calculation can be simplified if there are not many non-zero dct
// coefficients. Use eobs to separate different cases.
// DC only DCT coefficient.
- if (eob == 1) {
- aom_highbd_idct16x16_1_add(input, dest, stride, bd);
- } else if (eob <= 10) {
+ if (eob == 1) aom_highbd_idct16x16_1_add(input, dest, stride, bd);
+#if !CONFIG_ADAPT_SCAN
+ else if (eob <= 10)
aom_highbd_idct16x16_10_add(input, dest, stride, bd);
- } else {
+#endif
+ else
aom_highbd_idct16x16_256_add(input, dest, stride, bd);
- }
}
void av1_highbd_idct32x32_add(const tran_low_t *input, uint8_t *dest,
int stride, int eob, int bd) {
// Non-zero coeff only in upper-left 8x8
- if (eob == 1) {
- aom_highbd_idct32x32_1_add(input, dest, stride, bd);
- } else if (eob <= 34) {
+ if (eob == 1) aom_highbd_idct32x32_1_add(input, dest, stride, bd);
+#if !CONFIG_ADAPT_SCAN
+ else if (eob <= 34)
aom_highbd_idct32x32_34_add(input, dest, stride, bd);
- } else {
+#endif
+ else
aom_highbd_idct32x32_1024_add(input, dest, stride, bd);
- }
}
void av1_highbd_inv_txfm_add_4x4(const tran_low_t *input, uint8_t *dest,
diff --git a/av1/common/reconinter.c b/av1/common/reconinter.c
index ad9b462..6c4ae2a 100644
--- a/av1/common/reconinter.c
+++ b/av1/common/reconinter.c
@@ -546,6 +546,7 @@
struct macroblockd_plane *const pd = &xd->plane[plane];
#if CONFIG_MOTION_VAR
const MODE_INFO *mi = xd->mi[mi_col_offset + xd->mi_stride * mi_row_offset];
+ const int build_for_obmc = !(mi_col_offset == 0 && mi_row_offset == 0);
#else
const MODE_INFO *mi = xd->mi[0];
#endif // CONFIG_MOTION_VAR
@@ -567,7 +568,11 @@
// TODO(sarahparker) enable the use of DUAL_FILTER in warped motion functions
// in order to allow GLOBAL_MOTION and DUAL_FILTER to work together
#if CONFIG_DUAL_FILTER
+#if CONFIG_MOTION_VAR
+ if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0 && !build_for_obmc) {
+#else
if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0) {
+#endif // CONFIG_MOTION_VAR
// block size in log2
const int b4_wl = b_width_log2_lookup[mi->mbmi.sb_type];
const int b4_hl = b_height_log2_lookup[mi->mbmi.sb_type];
@@ -648,7 +653,11 @@
#endif
#if CONFIG_SUB8X8_MC
+#if CONFIG_MOTION_VAR
+ if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0 && !build_for_obmc) {
+#else
if (mi->mbmi.sb_type < BLOCK_8X8 && plane > 0) {
+#endif // CONFIG_MOTION_VAR
// block size in log2
const int b4_wl = b_width_log2_lookup[mi->mbmi.sb_type];
const int b4_hl = b_height_log2_lookup[mi->mbmi.sb_type];
@@ -734,7 +743,12 @@
struct buf_2d *const dst_buf = &pd->dst;
uint8_t *const dst = dst_buf->buf + dst_buf->stride * y + x;
const MV mv = mi->mbmi.sb_type < BLOCK_8X8
+#if CONFIG_MOTION_VAR
+ ? (build_for_obmc ? mi->bmi[block].as_mv[ref].as_mv
+ : average_split_mvs(pd, mi, ref, block))
+#else
? average_split_mvs(pd, mi, ref, block)
+#endif // CONFIG_MOTION_VAR
: mi->mbmi.mv[ref].as_mv;
// TODO(jkoleszar): This clamping is done in the incorrect place for the
@@ -1371,9 +1385,11 @@
const TileInfo *const tile = &xd->tile;
BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
int i, j, mi_step, ref;
+ int mb_to_right_edge_base = xd->mb_to_right_edge;
if (mi_row <= tile->mi_row_start) return;
+ xd->mb_to_bottom_edge += xd->n8_h * 32;
for (i = 0; i < AOMMIN(xd->n8_w, cm->mi_cols - mi_col); i += mi_step) {
int mi_row_offset = -1;
int mi_col_offset = i;
@@ -1412,6 +1428,8 @@
}
xd->mb_to_left_edge = -(((mi_col + i) * MI_SIZE) * 8);
+ xd->mb_to_right_edge =
+ mb_to_right_edge_base + (xd->n8_w - i - mi_step) * 64;
mi_x = (mi_col + i) << MI_SIZE_LOG2;
mi_y = mi_row << MI_SIZE_LOG2;
@@ -1425,19 +1443,19 @@
const PARTITION_TYPE bp = BLOCK_8X8 - above_mbmi->sb_type;
const int have_vsplit = bp != PARTITION_HORZ;
const int have_hsplit = bp != PARTITION_VERT;
- const int num_4x4_w = 2 >> ((!have_vsplit) | pd->subsampling_x);
- const int num_4x4_h = 2 >> ((!have_hsplit) | pd->subsampling_y);
- const int pw = 8 >> (have_vsplit | pd->subsampling_x);
+ const int num_4x4_w = 2 >> !have_vsplit;
+ const int num_4x4_h = 2 >> !have_hsplit;
+ const int pw = 8 >> (have_vsplit + pd->subsampling_x);
int x, y;
for (y = 0; y < num_4x4_h; ++y)
for (x = 0; x < num_4x4_w; ++x) {
- if ((bp == PARTITION_HORZ || bp == PARTITION_SPLIT) && y == 0 &&
- !pd->subsampling_y)
+ if ((bp == PARTITION_HORZ || bp == PARTITION_SPLIT) && y == 0)
continue;
build_inter_predictors(xd, j, mi_col_offset, mi_row_offset,
- y * 2 + x, bw, bh, 4 * x, 0, pw, bh,
+ y * 2 + x, bw, bh,
+ (4 * x) >> pd->subsampling_x, 0, pw, bh,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
0, 0,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
@@ -1457,6 +1475,8 @@
#endif // CONFIG_EXT_INTER
}
xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8);
+ xd->mb_to_right_edge = mb_to_right_edge_base;
+ xd->mb_to_bottom_edge -= xd->n8_h * 32;
}
void av1_build_prediction_by_left_preds(const AV1_COMMON *cm, MACROBLOCKD *xd,
@@ -1468,9 +1488,11 @@
const TileInfo *const tile = &xd->tile;
BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
int i, j, mi_step, ref;
+ int mb_to_bottom_edge_base = xd->mb_to_bottom_edge;
if (mi_col == 0 || (mi_col - 1 < tile->mi_col_start)) return;
+ xd->mb_to_right_edge += xd->n8_w * 32;
for (i = 0; i < AOMMIN(xd->n8_h, cm->mi_rows - mi_row); i += mi_step) {
int mi_row_offset = i;
int mi_col_offset = -1;
@@ -1509,6 +1531,8 @@
}
xd->mb_to_top_edge = -(((mi_row + i) * MI_SIZE) * 8);
+ xd->mb_to_bottom_edge =
+ mb_to_bottom_edge_base + (xd->n8_h - i - mi_step) * 64;
mi_x = mi_col << MI_SIZE_LOG2;
mi_y = (mi_row + i) << MI_SIZE_LOG2;
@@ -1522,19 +1546,19 @@
const PARTITION_TYPE bp = BLOCK_8X8 - left_mbmi->sb_type;
const int have_vsplit = bp != PARTITION_HORZ;
const int have_hsplit = bp != PARTITION_VERT;
- const int num_4x4_w = 2 >> ((!have_vsplit) | pd->subsampling_x);
- const int num_4x4_h = 2 >> ((!have_hsplit) | pd->subsampling_y);
- const int ph = 8 >> (have_hsplit | pd->subsampling_y);
+ const int num_4x4_w = 2 >> !have_vsplit;
+ const int num_4x4_h = 2 >> !have_hsplit;
+ const int ph = 8 >> (have_hsplit + pd->subsampling_y);
int x, y;
for (y = 0; y < num_4x4_h; ++y)
for (x = 0; x < num_4x4_w; ++x) {
- if ((bp == PARTITION_VERT || bp == PARTITION_SPLIT) && x == 0 &&
- !pd->subsampling_x)
+ if ((bp == PARTITION_VERT || bp == PARTITION_SPLIT) && x == 0)
continue;
build_inter_predictors(xd, j, mi_col_offset, mi_row_offset,
- y * 2 + x, bw, bh, 0, 4 * y, bw, ph,
+ y * 2 + x, bw, bh, 0,
+ (4 * y) >> pd->subsampling_y, bw, ph,
#if CONFIG_SUPERTX && CONFIG_EXT_INTER
0, 0,
#endif // CONFIG_SUPERTX && CONFIG_EXT_INTER
@@ -1554,6 +1578,54 @@
#endif // CONFIG_EXT_INTER
}
xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8);
+ xd->mb_to_bottom_edge = mb_to_bottom_edge_base;
+ xd->mb_to_right_edge -= xd->n8_w * 32;
+}
+
+void av1_build_obmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
+ int mi_row, int mi_col) {
+#if CONFIG_AOM_HIGHBITDEPTH
+ DECLARE_ALIGNED(16, uint8_t, tmp_buf1[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
+ DECLARE_ALIGNED(16, uint8_t, tmp_buf2[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
+#else
+ DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_SB_SQUARE]);
+ DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_SB_SQUARE]);
+#endif // CONFIG_AOM_HIGHBITDEPTH
+ uint8_t *dst_buf1[MAX_MB_PLANE], *dst_buf2[MAX_MB_PLANE];
+ int dst_stride1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+ int dst_stride2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+ int dst_width1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+ int dst_width2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+ int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+ int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
+
+#if CONFIG_AOM_HIGHBITDEPTH
+ if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
+ int len = sizeof(uint16_t);
+ dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
+ dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * len);
+ dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * 2 * len);
+ dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
+ dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * len);
+ dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * 2 * len);
+ } else {
+#endif // CONFIG_AOM_HIGHBITDEPTH
+ dst_buf1[0] = tmp_buf1;
+ dst_buf1[1] = tmp_buf1 + MAX_SB_SQUARE;
+ dst_buf1[2] = tmp_buf1 + MAX_SB_SQUARE * 2;
+ dst_buf2[0] = tmp_buf2;
+ dst_buf2[1] = tmp_buf2 + MAX_SB_SQUARE;
+ dst_buf2[2] = tmp_buf2 + MAX_SB_SQUARE * 2;
+#if CONFIG_AOM_HIGHBITDEPTH
+ }
+#endif // CONFIG_AOM_HIGHBITDEPTH
+ av1_build_prediction_by_above_preds(cm, xd, mi_row, mi_col, dst_buf1,
+ dst_width1, dst_height1, dst_stride1);
+ av1_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
+ dst_width2, dst_height2, dst_stride2);
+ av1_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
+ av1_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1, dst_stride1,
+ dst_buf2, dst_stride2);
}
#endif // CONFIG_MOTION_VAR
diff --git a/av1/common/reconinter.h b/av1/common/reconinter.h
index 5f62f0a..3eec384 100644
--- a/av1/common/reconinter.h
+++ b/av1/common/reconinter.h
@@ -535,6 +535,8 @@
int tmp_width[MAX_MB_PLANE],
int tmp_height[MAX_MB_PLANE],
int tmp_stride[MAX_MB_PLANE]);
+void av1_build_obmc_inter_predictors_sb(const AV1_COMMON *cm, MACROBLOCKD *xd,
+ int mi_row, int mi_col);
#endif // CONFIG_MOTION_VAR
#if CONFIG_EXT_INTER
diff --git a/av1/common/scan.h b/av1/common/scan.h
index 407c9ec..af39993 100644
--- a/av1/common/scan.h
+++ b/av1/common/scan.h
@@ -82,6 +82,10 @@
static INLINE const SCAN_ORDER *get_scan(const AV1_COMMON *cm, TX_SIZE tx_size,
TX_TYPE tx_type, int is_inter) {
+#if CONFIG_ADAPT_SCAN
+ (void)is_inter;
+ return &cm->fc->sc[tx_size][tx_type];
+#else // CONFIG_ADAPT_SCAN
(void)cm;
#if CONFIG_EXT_TX
return is_inter ? &av1_inter_scan_orders[tx_size][tx_type]
@@ -90,6 +94,7 @@
(void)is_inter;
return &av1_intra_scan_orders[tx_size][tx_type];
#endif // CONFIG_EXT_TX
+#endif // CONFIG_ADAPT_SCAN
}
#ifdef __cplusplus
diff --git a/av1/decoder/decodeframe.c b/av1/decoder/decodeframe.c
index dc9f5e4..efb6465 100644
--- a/av1/decoder/decodeframe.c
+++ b/av1/decoder/decodeframe.c
@@ -303,6 +303,9 @@
const int eob =
av1_decode_block_tokens(xd, plane, scan_order, col, row, tx_size,
tx_type, &max_scan_line, r, mbmi->segment_id);
+#if CONFIG_ADAPT_SCAN
+ av1_update_scan_count_facade(cm, tx_size, tx_type, pd->dqcoeff, eob);
+#endif
if (eob)
inverse_transform_block(xd, plane, tx_type, tx_size, dst, pd->dst.stride,
max_scan_line, eob);
@@ -384,6 +387,9 @@
const int eob =
av1_decode_block_tokens(xd, plane, scan_order, col, row, tx_size, tx_type,
&max_scan_line, r, segment_id);
+#if CONFIG_ADAPT_SCAN
+ av1_update_scan_count_facade(cm, tx_size, tx_type, pd->dqcoeff, eob);
+#endif
if (eob)
inverse_transform_block(xd, plane, tx_type, tx_size,
&pd->dst.buf[4 * row * pd->dst.stride + 4 * col],
@@ -1264,49 +1270,7 @@
av1_build_inter_predictors_sb(xd, mi_row, mi_col, AOMMAX(bsize, BLOCK_8X8));
#if CONFIG_MOTION_VAR
if (mbmi->motion_mode == OBMC_CAUSAL) {
-#if CONFIG_AOM_HIGHBITDEPTH
- DECLARE_ALIGNED(16, uint8_t, tmp_buf1[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
- DECLARE_ALIGNED(16, uint8_t, tmp_buf2[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
-#else
- DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_SB_SQUARE]);
- DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_SB_SQUARE]);
-#endif // CONFIG_AOM_HIGHBITDEPTH
- uint8_t *dst_buf1[MAX_MB_PLANE], *dst_buf2[MAX_MB_PLANE];
- int dst_width1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_width2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_stride1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_stride2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
-
- assert(mbmi->sb_type >= BLOCK_8X8);
-#if CONFIG_AOM_HIGHBITDEPTH
- if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
- int len = sizeof(uint16_t);
- dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
- dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * len);
- dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * 2 * len);
- dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
- dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * len);
- dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * 2 * len);
- } else {
-#endif // CONFIG_AOM_HIGHBITDEPTH
- dst_buf1[0] = tmp_buf1;
- dst_buf1[1] = tmp_buf1 + MAX_SB_SQUARE;
- dst_buf1[2] = tmp_buf1 + MAX_SB_SQUARE * 2;
- dst_buf2[0] = tmp_buf2;
- dst_buf2[1] = tmp_buf2 + MAX_SB_SQUARE;
- dst_buf2[2] = tmp_buf2 + MAX_SB_SQUARE * 2;
-#if CONFIG_AOM_HIGHBITDEPTH
- }
-#endif // CONFIG_AOM_HIGHBITDEPTH
- av1_build_prediction_by_above_preds(cm, xd, mi_row, mi_col, dst_buf1,
- dst_width1, dst_height1, dst_stride1);
- av1_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
- dst_width2, dst_height2, dst_stride2);
- av1_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
- av1_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1,
- dst_stride1, dst_buf2, dst_stride2);
+ av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
}
#endif // CONFIG_MOTION_VAR
@@ -2275,7 +2239,7 @@
static InterpFilter read_interp_filter(struct aom_read_bit_buffer *rb) {
return aom_rb_read_bit(rb) ? SWITCHABLE
- : aom_rb_read_literal(rb, 2 + CONFIG_EXT_INTERP);
+ : aom_rb_read_literal(rb, LOG_SWITCHABLE_FILTERS);
}
static void setup_render_size(AV1_COMMON *cm, struct aom_read_bit_buffer *rb) {
diff --git a/av1/encoder/bitstream.c b/av1/encoder/bitstream.c
index df27fbc..5660e06 100644
--- a/av1/encoder/bitstream.c
+++ b/av1/encoder/bitstream.c
@@ -2915,7 +2915,7 @@
struct aom_write_bit_buffer *wb) {
aom_wb_write_bit(wb, filter == SWITCHABLE);
if (filter != SWITCHABLE)
- aom_wb_write_literal(wb, filter, 2 + CONFIG_EXT_INTERP);
+ aom_wb_write_literal(wb, filter, LOG_SWITCHABLE_FILTERS);
}
static void fix_interp_filter(AV1_COMMON *cm, FRAME_COUNTS *counts) {
diff --git a/av1/encoder/block.h b/av1/encoder/block.h
index dcdf97e..5c8a2f6 100644
--- a/av1/encoder/block.h
+++ b/av1/encoder/block.h
@@ -80,7 +80,6 @@
MACROBLOCKD e_mbd;
MB_MODE_INFO_EXT *mbmi_ext;
int skip_block;
- int select_tx_size;
int qindex;
// The equivalent error at the current rdmult of one whole bit (not one
@@ -157,8 +156,6 @@
int skip;
- int encode_breakout;
-
// note that token_costs is the cost when eob node is skipped
av1_coeff_cost token_costs[TX_SIZES];
diff --git a/av1/encoder/encodeframe.c b/av1/encoder/encodeframe.c
index acca6f1..ffb5d46 100644
--- a/av1/encoder/encodeframe.c
+++ b/av1/encoder/encodeframe.c
@@ -324,11 +324,8 @@
mbmi->segment_id = get_segment_id(cm, map, bsize, mi_row, mi_col);
}
av1_init_plane_quantizers(cpi, x, mbmi->segment_id);
-
- x->encode_breakout = cpi->segment_encode_breakout[mbmi->segment_id];
} else {
mbmi->segment_id = 0;
- x->encode_breakout = cpi->encode_breakout;
}
#if CONFIG_SUPERTX
@@ -404,7 +401,6 @@
if (!seg->enabled) {
seg_id_supertx = 0;
- x->encode_breakout = cpi->encode_breakout;
} else {
// Find the minimum segment_id
for (r = 0; r < mih; r++)
@@ -415,7 +411,6 @@
// Initialize plane quantisers
av1_init_plane_quantizers(cpi, x, seg_id_supertx);
- x->encode_breakout = cpi->segment_encode_breakout[seg_id_supertx];
}
// Assign the the segment_id back to segment_id_supertx
@@ -1332,8 +1327,8 @@
#if CONFIG_DUAL_FILTER
update_filter_type_count(td->counts, xd, mbmi);
#else
- const int ctx = av1_get_pred_context_switchable_interp(xd);
- ++td->counts->switchable_interp[ctx][mbmi->interp_filter];
+ const int pred_ctx = av1_get_pred_context_switchable_interp(xd);
+ ++td->counts->switchable_interp[pred_ctx][mbmi->interp_filter];
#endif
}
@@ -1688,7 +1683,6 @@
mbmi->segment_id = av1_vaq_segment_id(energy);
// Re-initialise quantiser
av1_init_plane_quantizers(cpi, x, mbmi->segment_id);
- x->encode_breakout = cpi->segment_encode_breakout[mbmi->segment_id];
}
x->rdmult = set_segment_rdmult(cpi, x, mbmi->segment_id);
} else if (aq_mode == COMPLEXITY_AQ) {
@@ -5183,50 +5177,7 @@
#if CONFIG_MOTION_VAR
if (mbmi->motion_mode == OBMC_CAUSAL) {
-#if CONFIG_AOM_HIGHBITDEPTH
- DECLARE_ALIGNED(16, uint8_t, tmp_buf1[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
- DECLARE_ALIGNED(16, uint8_t, tmp_buf2[2 * MAX_MB_PLANE * MAX_SB_SQUARE]);
-#else
- DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * MAX_SB_SQUARE]);
- DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * MAX_SB_SQUARE]);
-#endif // CONFIG_AOM_HIGHBITDEPTH
- uint8_t *dst_buf1[MAX_MB_PLANE], *dst_buf2[MAX_MB_PLANE];
- int dst_stride1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_stride2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_width1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_width2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_height1[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
- int dst_height2[MAX_MB_PLANE] = { MAX_SB_SIZE, MAX_SB_SIZE, MAX_SB_SIZE };
-
- assert(mbmi->sb_type >= BLOCK_8X8);
-
-#if CONFIG_AOM_HIGHBITDEPTH
- if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
- int len = sizeof(uint16_t);
- dst_buf1[0] = CONVERT_TO_BYTEPTR(tmp_buf1);
- dst_buf1[1] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * len);
- dst_buf1[2] = CONVERT_TO_BYTEPTR(tmp_buf1 + MAX_SB_SQUARE * 2 * len);
- dst_buf2[0] = CONVERT_TO_BYTEPTR(tmp_buf2);
- dst_buf2[1] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * len);
- dst_buf2[2] = CONVERT_TO_BYTEPTR(tmp_buf2 + MAX_SB_SQUARE * 2 * len);
- } else {
-#endif // CONFIG_AOM_HIGHBITDEPTH
- dst_buf1[0] = tmp_buf1;
- dst_buf1[1] = tmp_buf1 + MAX_SB_SQUARE;
- dst_buf1[2] = tmp_buf1 + MAX_SB_SQUARE * 2;
- dst_buf2[0] = tmp_buf2;
- dst_buf2[1] = tmp_buf2 + MAX_SB_SQUARE;
- dst_buf2[2] = tmp_buf2 + MAX_SB_SQUARE * 2;
-#if CONFIG_AOM_HIGHBITDEPTH
- }
-#endif // CONFIG_AOM_HIGHBITDEPTH
- av1_build_prediction_by_above_preds(cm, xd, mi_row, mi_col, dst_buf1,
- dst_width1, dst_height1, dst_stride1);
- av1_build_prediction_by_left_preds(cm, xd, mi_row, mi_col, dst_buf2,
- dst_width2, dst_height2, dst_stride2);
- av1_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
- av1_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, dst_buf1,
- dst_stride1, dst_buf2, dst_stride2);
+ av1_build_obmc_inter_predictors_sb(cm, xd, mi_row, mi_col);
}
#endif // CONFIG_MOTION_VAR
diff --git a/av1/encoder/encoder.c b/av1/encoder/encoder.c
index 33c536d..629eb46 100644
--- a/av1/encoder/encoder.c
+++ b/av1/encoder/encoder.c
@@ -1897,14 +1897,6 @@
av1_reset_segment_features(cm);
av1_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,
@@ -2167,8 +2159,6 @@
kf_list = fopen("kf_list.stt", "w");
#endif
- cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
-
if (oxcf->pass == 1) {
av1_init_first_pass(cpi);
} else if (oxcf->pass == 2) {
@@ -4760,8 +4750,6 @@
static void Pass2Encode(AV1_COMP *cpi, size_t *size, uint8_t *dest,
unsigned int *frame_flags) {
- cpi->allow_encode_breakout = ENCODE_BREAKOUT_ENABLED;
-
encode_frame_to_data_rate(cpi, size, dest, frame_flags);
#if CONFIG_EXT_REFS
diff --git a/av1/encoder/encoder.h b/av1/encoder/encoder.h
index 1bf7ff4..8738609 100644
--- a/av1/encoder/encoder.h
+++ b/av1/encoder/encoder.h
@@ -86,15 +86,6 @@
} FRAME_CONTEXT_INDEX;
typedef enum {
- // encode_breakout is disabled.
- ENCODE_BREAKOUT_DISABLED = 0,
- // encode_breakout is enabled.
- ENCODE_BREAKOUT_ENABLED = 1,
- // encode_breakout is enabled with small max_thresh limit.
- ENCODE_BREAKOUT_LIMITED = 2
-} ENCODE_BREAKOUT_TYPE;
-
-typedef enum {
NORMAL = 0,
FOURFIVE = 1,
THREEFIVE = 2,
@@ -223,8 +214,6 @@
int enable_auto_brf; // (b)ackward (r)ef (f)rame
#endif // CONFIG_EXT_REFS
- int encode_breakout; // early breakout : for video conf recommend 800
-
/* Bitfield defining the error resiliency features to enable.
* Can provide decodable frames after losses in previous
* frames and decodable partitions after losses in the same frame.
@@ -449,18 +438,8 @@
int allow_comp_inter_inter;
- // Default value is 1. From first pass stats, encode_breakout may be disabled.
- ENCODE_BREAKOUT_TYPE allow_encode_breakout;
-
- // Get threshold from external input. A suggested threshold is 800 for HD
- // clips, and 300 for < HD clips.
- int encode_breakout;
-
uint8_t *segmentation_map;
- // segment threashold for encode breakout
- int segment_encode_breakout[MAX_SEGMENTS];
-
CYCLIC_REFRESH *cyclic_refresh;
ActiveMap active_map;
diff --git a/av1/encoder/firstpass.c b/av1/encoder/firstpass.c
index dc97ddf..0f7fcca 100644
--- a/av1/encoder/firstpass.c
+++ b/av1/encoder/firstpass.c
@@ -535,6 +535,9 @@
}
av1_init_mv_probs(cm);
+#if CONFIG_ADAPT_SCAN
+ av1_init_scan_order(cm);
+#endif
av1_initialize_rd_consts(cpi);
// Tiling is ignored in the first pass.
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index a1bac85..fdf9b51 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -1528,6 +1528,10 @@
#if CONFIG_EXT_TX
int ext_tx_set;
#endif // CONFIG_EXT_TX
+ *distortion = INT64_MAX;
+ *rate = INT_MAX;
+ *skip = 0;
+ *sse = INT64_MAX;
mbmi->tx_size = tx_size_from_tx_mode(bs, cm->tx_mode, is_inter);
@@ -1591,10 +1595,17 @@
if (this_rd < best_rd) {
best_rd = this_rd;
best_tx_type = mbmi->tx_type;
+ *distortion = d;
+ *rate = r;
+ *skip = s;
+ *sse = psse;
}
}
+ } else {
+ mbmi->tx_type = DCT_DCT;
+ txfm_rd_in_plane(x, cpi, rate, distortion, skip, sse, ref_best_rd, 0, bs,
+ mbmi->tx_size, cpi->sf.use_fast_coef_costing);
}
-
#else // CONFIG_EXT_TX
if (mbmi->tx_size < TX_32X32 && !xd->lossless[mbmi->segment_id]) {
for (tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
@@ -1628,14 +1639,19 @@
if (this_rd < best_rd) {
best_rd = this_rd;
best_tx_type = mbmi->tx_type;
+ *distortion = d;
+ *rate = r;
+ *skip = s;
+ *sse = psse;
}
}
+ } else {
+ mbmi->tx_type = DCT_DCT;
+ txfm_rd_in_plane(x, cpi, rate, distortion, skip, sse, ref_best_rd, 0, bs,
+ mbmi->tx_size, cpi->sf.use_fast_coef_costing);
}
#endif // CONFIG_EXT_TX
mbmi->tx_type = best_tx_type;
-
- txfm_rd_in_plane(x, cpi, rate, distortion, skip, sse, ref_best_rd, 0, bs,
- mbmi->tx_size, cpi->sf.use_fast_coef_costing);
}
static void choose_smallest_tx_size(const AV1_COMP *const cpi, MACROBLOCK *x,
diff --git a/av1/encoder/speed_features.c b/av1/encoder/speed_features.c
index d32ff7d..acdc13b 100644
--- a/av1/encoder/speed_features.c
+++ b/av1/encoder/speed_features.c
@@ -270,11 +270,6 @@
sf->partition_search_breakout_dist_thr = (1 << 23);
}
}
-
- if (speed >= 7) {
- sf->encode_breakout_thresh =
- (AOMMIN(cm->width, cm->height) >= 720) ? 800 : 300;
- }
}
static void set_rt_speed_feature(AV1_COMP *cpi, SPEED_FEATURES *sf, int speed,
@@ -454,11 +449,6 @@
sf->adaptive_pred_interp_filter = 0;
}
- if (cpi->encode_breakout && oxcf->mode == REALTIME &&
- sf->encode_breakout_thresh > cpi->encode_breakout) {
- cpi->encode_breakout = sf->encode_breakout_thresh;
- }
-
// Check for masked out split cases.
for (i = 0; i < MAX_REFS; ++i) {
if (sf->disable_split_mask & (1 << i)) {
@@ -542,7 +532,6 @@
// to FIXED_PARTITION.
sf->always_this_block_size = BLOCK_16X16;
sf->search_type_check_frequency = 50;
- sf->encode_breakout_thresh = 0;
// Recode loop tolerance %.
sf->recode_tolerance = 25;
sf->default_interp_filter = SWITCHABLE;
diff --git a/av1/encoder/speed_features.h b/av1/encoder/speed_features.h
index 5c34808..c6821bf 100644
--- a/av1/encoder/speed_features.h
+++ b/av1/encoder/speed_features.h
@@ -443,10 +443,6 @@
// time mode speed 6.
int reuse_inter_pred_sby;
- // This variable sets the encode_breakout threshold. Currently, it is only
- // enabled in real time mode.
- int encode_breakout_thresh;
-
// default interp filter choice
InterpFilter default_interp_filter;
diff --git a/av1/encoder/tokenize.c b/av1/encoder/tokenize.c
index 7707e8f..821adba 100644
--- a/av1/encoder/tokenize.c
+++ b/av1/encoder/tokenize.c
@@ -534,6 +534,14 @@
*tp = t;
+#if CONFIG_ADAPT_SCAN
+ // Since dqcoeff is not available here, we pass qcoeff into
+ // av1_update_scan_count_facade(). The update behavior should be the same
+ // because av1_update_scan_count_facade() only cares if coefficients are zero
+ // or not.
+ av1_update_scan_count_facade((AV1_COMMON *)cm, tx_size, tx_type, qcoeff, c);
+#endif
+
av1_set_contexts(xd, pd, tx_size, c > 0, blk_col, blk_row);
}
diff --git a/test/av1_inv_txfm_test.cc b/test/av1_inv_txfm_test.cc
index ff358b6..84e2402 100644
--- a/test/av1_inv_txfm_test.cc
+++ b/test/av1_inv_txfm_test.cc
@@ -137,6 +137,7 @@
InvTxfmFunc partial_itxfm_;
};
+#if !CONFIG_ADAPT_SCAN
TEST_P(AV1PartialIDctTest, RunQuantCheck) {
int size;
switch (tx_size_) {
@@ -256,6 +257,7 @@
EXPECT_EQ(0, max_error)
<< "Error: partial inverse transform produces different results";
}
+#endif
using std::tr1::make_tuple;
INSTANTIATE_TEST_CASE_P(
diff --git a/test/test.mk b/test/test.mk
index 61ea1d0..2d18f69 100644
--- a/test/test.mk
+++ b/test/test.mk
@@ -113,6 +113,7 @@
#LIBAOM_TEST_SRCS-yes += encoder_parms_get_to_decoder.cc
endif
+LIBAOM_TEST_SRCS-$(CONFIG_ADAPT_SCAN) += scan_test.cc
#LIBAOM_TEST_SRCS-yes += convolve_test.cc
LIBAOM_TEST_SRCS-yes += lpf_8_test.cc
LIBAOM_TEST_SRCS-$(CONFIG_CLPF) += clpf_test.cc
