Merge "Move small fixes and refactoring for obmc pred from AV1" into nextgenv2
diff --git a/aom_scale/generic/aom_scale.c b/aom_scale/generic/aom_scale.c
index 28604ac..9007459 100644
--- a/aom_scale/generic/aom_scale.c
+++ b/aom_scale/generic/aom_scale.c
@@ -68,7 +68,6 @@
unsigned int source_scale, unsigned int source_length,
unsigned char *dest, int dest_step,
unsigned int dest_scale, unsigned int dest_length) {
- const unsigned int source_pitch = source_step;
const unsigned char *const dest_end = dest + dest_length * dest_step;
(void)source_length;
(void)source_scale;
@@ -81,9 +80,9 @@
dest += dest_step;
while (dest < dest_end) {
- const unsigned int a = 3 * source[-source_pitch];
+ const unsigned int a = 3 * source[-source_step];
const unsigned int b = 10 * source[0];
- const unsigned int c = 3 * source[source_pitch];
+ const unsigned int c = 3 * source[source_step];
*dest = (unsigned char)((8 + a + b + c) >> 4);
source += source_step;
dest += dest_step;
diff --git a/aomenc.c b/aomenc.c
index 497c8d5..e32a922 100644
--- a/aomenc.c
+++ b/aomenc.c
@@ -40,12 +40,12 @@
#include "aom/aomdx.h"
#endif
-#include "aom/aom_integer.h"
-#include "aom_ports/mem_ops.h"
-#include "aom_ports/aom_timer.h"
-#include "./rate_hist.h"
#include "./aomstats.h"
+#include "./rate_hist.h"
#include "./warnings.h"
+#include "aom/aom_integer.h"
+#include "aom_ports/aom_timer.h"
+#include "aom_ports/mem_ops.h"
#if CONFIG_WEBM_IO
#include "./webmenc.h"
#endif
@@ -1860,6 +1860,7 @@
uint64_t cx_time = 0;
int stream_cnt = 0;
int res = 0;
+ int profile_updated = 0;
memset(&input, 0, sizeof(input));
exec_name = argv_[0];
@@ -1963,6 +1964,54 @@
{ stream->config.cfg.g_input_bit_depth = input.bit_depth; });
}
+ 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;
+ }
+ if (stream->config.cfg.g_bit_depth > 8) {
+ switch (stream->config.cfg.g_profile) {
+ case 0:
+ stream->config.cfg.g_profile = 2;
+ profile_updated = 1;
+ break;
+ case 1:
+ stream->config.cfg.g_profile = 3;
+ profile_updated = 1;
+ break;
+ default: break;
+ }
+ }
+ 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 "
+ "match input format.\n",
+ stream->config.cfg.g_profile);
+ }
+ });
+
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 63f7618..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);
@@ -953,11 +951,13 @@
// failure condition, encoder setup is done fully in init() currently.
if (res == AOM_CODEC_OK) {
#if CONFIG_EXT_REFS
- data_sz = ctx->cfg.g_w * ctx->cfg.g_h * get_image_bps(img);
+ data_sz = ALIGN_POWER_OF_TWO(ctx->cfg.g_w, 5) *
+ ALIGN_POWER_OF_TWO(ctx->cfg.g_h, 5) * get_image_bps(img);
#else
// There's no codec control for multiple alt-refs so check the encoder
// instance for its status to determine the compressed data size.
- data_sz = ctx->cfg.g_w * ctx->cfg.g_h * get_image_bps(img) / 8 *
+ data_sz = ALIGN_POWER_OF_TWO(ctx->cfg.g_w, 5) *
+ ALIGN_POWER_OF_TWO(ctx->cfg.g_h, 5) * get_image_bps(img) / 8 *
(cpi->multi_arf_allowed ? 8 : 2);
#endif // CONFIG_EXT_REFS
if (data_sz < kMinCompressedSize) data_sz = kMinCompressedSize;
diff --git a/av1/common/blockd.h b/av1/common/blockd.h
index bc1970c..8d7c7f8 100644
--- a/av1/common/blockd.h
+++ b/av1/common/blockd.h
@@ -307,6 +307,8 @@
uint16_t n4_w, n4_h;
// log2 of n4_w, n4_h
uint8_t n4_wl, n4_hl;
+ // block size in pixels
+ uint8_t width, height;
#if CONFIG_AOM_QM
const qm_val_t *seg_iqmatrix[MAX_SEGMENTS][2][TX_SIZES];
diff --git a/av1/common/common_data.h b/av1/common/common_data.h
index 528343b..88bfb0a 100644
--- a/av1/common/common_data.h
+++ b/av1/common/common_data.h
@@ -42,6 +42,12 @@
};
// Width/height lookup tables in units of various block sizes
+static const uint8_t block_size_wide[BLOCK_SIZES] = {
+ 4, 4, 8, 8, 8, 16, 16, 16, 32, 32, 32, 64, 64, IF_EXT_PARTITION(64, 128, 128)
+};
+static const uint8_t block_size_high[BLOCK_SIZES] = {
+ 4, 8, 4, 8, 16, 8, 16, 32, 16, 32, 64, 32, 64, IF_EXT_PARTITION(128, 64, 128)
+};
static const uint8_t num_4x4_blocks_wide_lookup[BLOCK_SIZES] = {
1, 1, 2, 2, 2, 4, 4, 4, 8, 8, 8, 16, 16, IF_EXT_PARTITION(16, 32, 32)
};
diff --git a/av1/common/entropy.c b/av1/common/entropy.c
index 7179e3d..885b782 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) {
@@ -2881,9 +2886,13 @@
}
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) {
update_factor = COEF_MAX_UPDATE_FACTOR_AFTER_KEY_BITS; /* adapt quickly */
@@ -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/enums.h b/av1/common/enums.h
index b02c814..0a1f7a3 100644
--- a/av1/common/enums.h
+++ b/av1/common/enums.h
@@ -58,6 +58,10 @@
#define MAX_TILE_COLS 64
#endif // CONFIG_EXT_TILE
+#if CONFIG_VAR_TX
+#define MAX_VARTX_DEPTH 2
+#endif
+
// Bitstream profiles indicated by 2-3 bits in the uncompressed header.
// 00: Profile 0. 8-bit 4:2:0 only.
// 10: Profile 1. 8-bit 4:4:4, 4:2:2, and 4:4:0.
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/onyxc_int.h b/av1/common/onyxc_int.h
index b6e73cd..be1cbc1 100644
--- a/av1/common/onyxc_int.h
+++ b/av1/common/onyxc_int.h
@@ -531,6 +531,8 @@
xd->plane[i].n4_h = (bh << 1) >> xd->plane[i].subsampling_y;
xd->plane[i].n4_wl = bwl - xd->plane[i].subsampling_x;
xd->plane[i].n4_hl = bhl - xd->plane[i].subsampling_y;
+ xd->plane[i].width = xd->plane[i].n4_w * 4;
+ xd->plane[i].height = xd->plane[i].n4_h * 4;
}
}
diff --git a/av1/common/reconinter.c b/av1/common/reconinter.c
index 8056ead..6c4ae2a 100644
--- a/av1/common/reconinter.c
+++ b/av1/common/reconinter.c
@@ -719,16 +719,16 @@
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
high_inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
subpel_x, subpel_y, sf, x_step, y_step, ref,
- &mi->mbmi.interp_filter, xs, ys, xd->bd);
+ mi->mbmi.interp_filter, xs, ys, xd->bd);
} else {
inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride,
subpel_x, subpel_y, sf, x_step, y_step, ref,
- &mi->mbmi.interp_filter, xs, ys);
+ mi->mbmi.interp_filter, xs, ys);
}
#else
inter_predictor(pre, pre_buf->stride, dst, dst_buf->stride, subpel_x,
subpel_y, sf, x_step, y_step, ref,
- &mi->mbmi.interp_filter, xs, ys);
+ mi->mbmi.interp_filter, xs, ys);
#endif
}
}
@@ -802,7 +802,7 @@
pre_buf->buf0, pre_buf->width, pre_buf->height,
pre_buf->stride, dst, (mi_x >> pd->subsampling_x) + x,
(mi_y >> pd->subsampling_y) + y, w, h, dst_buf->stride,
- pd->subsampling_x, pd->subsampling_y, xs, ys);
+ pd->subsampling_x, pd->subsampling_y, xs, ys, ref);
else
#endif // CONFIG_GLOBAL_MOTION
#endif // CONFIG_EXT_INTER
diff --git a/av1/common/reconintra.c b/av1/common/reconintra.c
index cf4bedf..cfd283f 100644
--- a/av1/common/reconintra.c
+++ b/av1/common/reconintra.c
@@ -283,7 +283,7 @@
const int wl = mi_width_log2_lookup[bsize];
const int hl = mi_height_log2_lookup[bsize];
const int h = 1 << (hl + 1 - ss_y);
- const int step = 1 << txsz;
+ const int step = tx_size_wide_unit[txsz];
const uint8_t *order = orders[bsize];
int my_order, bl_order;
@@ -671,7 +671,7 @@
INTRA_FILTER filter_type) {
const int dx = get_dx(angle);
const int dy = get_dy(angle);
- const int bs = 4 * num_4x4_blocks_wide_txsize_lookup[tx_size];
+ const int bs = tx_size_wide[tx_size];
assert(angle > 0 && angle < 270);
if (angle > 0 && angle < 90) {
@@ -1200,7 +1200,7 @@
DECLARE_ALIGNED(16, uint16_t, above_data[MAX_SB_SIZE + 16]);
uint16_t *above_row = above_data + 16;
const uint16_t *const_above_row = above_row;
- const int bs = 4 * num_4x4_blocks_wide_txsize_lookup[tx_size];
+ const int bs = tx_size_wide[tx_size];
int need_left = extend_modes[mode] & NEED_LEFT;
int need_above = extend_modes[mode] & NEED_ABOVE;
int need_above_left = extend_modes[mode] & NEED_ABOVELEFT;
@@ -1361,7 +1361,7 @@
DECLARE_ALIGNED(16, uint8_t, above_data[MAX_SB_SIZE + 16]);
uint8_t *above_row = above_data + 16;
const uint8_t *const_above_row = above_row;
- const int bs = 4 * num_4x4_blocks_wide_txsize_lookup[tx_size];
+ const int bs = tx_size_wide[tx_size];
int need_left = extend_modes[mode] & NEED_LEFT;
int need_above = extend_modes[mode] & NEED_ABOVE;
int need_above_left = extend_modes[mode] & NEED_ABOVELEFT;
@@ -1507,25 +1507,21 @@
}
}
-void av1_predict_intra_block(const MACROBLOCKD *xd, int bwl_in, int bhl_in,
+void av1_predict_intra_block(const MACROBLOCKD *xd, int wpx, int hpx,
TX_SIZE tx_size, PREDICTION_MODE mode,
const uint8_t *ref, int ref_stride, uint8_t *dst,
int dst_stride, int col_off, int row_off,
int plane) {
const BLOCK_SIZE bsize = xd->mi[0]->mbmi.sb_type;
const struct macroblockd_plane *const pd = &xd->plane[plane];
- const int txw = num_4x4_blocks_wide_txsize_lookup[tx_size];
- const int txh = num_4x4_blocks_high_txsize_lookup[tx_size];
+ const int txw = tx_size_wide_unit[tx_size];
+ const int txh = tx_size_high_unit[tx_size];
const int have_top = row_off || xd->up_available;
const int have_left = col_off || xd->left_available;
const int x = col_off * 4;
const int y = row_off * 4;
- const int bw = pd->subsampling_x ? 1 << bwl_in : AOMMAX(2, 1 << bwl_in);
- const int bh = pd->subsampling_y ? 1 << bhl_in : AOMMAX(2, 1 << bhl_in);
const int mi_row = -xd->mb_to_top_edge >> (3 + MI_SIZE_LOG2);
const int mi_col = -xd->mb_to_left_edge >> (3 + MI_SIZE_LOG2);
- const int wpx = 4 * bw;
- const int hpx = 4 * bh;
const int txwpx = 4 * txw;
const int txhpx = 4 * txh;
// Distance between the right edge of this prediction block to
@@ -1553,8 +1549,8 @@
#if CONFIG_PALETTE
if (xd->mi[0]->mbmi.palette_mode_info.palette_size[plane != 0] > 0) {
- const int bs = 4 * num_4x4_blocks_wide_txsize_lookup[tx_size];
- const int stride = 4 * (1 << bwl_in);
+ const int bs = tx_size_wide[tx_size];
+ const int stride = wpx;
int r, c;
uint8_t *map = NULL;
#if CONFIG_AOM_HIGHBITDEPTH
diff --git a/av1/common/reconintra.h b/av1/common/reconintra.h
index 7778874..23bad1c 100644
--- a/av1/common/reconintra.h
+++ b/av1/common/reconintra.h
@@ -21,7 +21,7 @@
void av1_init_intra_predictors(void);
-void av1_predict_intra_block(const MACROBLOCKD *xd, int bwl_in, int bhl_in,
+void av1_predict_intra_block(const MACROBLOCKD *xd, int bw, int bh,
TX_SIZE tx_size, PREDICTION_MODE mode,
const uint8_t *ref, int ref_stride, uint8_t *dst,
int dst_stride, int aoff, int loff, int plane);
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/common/warped_motion.c b/av1/common/warped_motion.c
index fc632c3..e5ed39d 100644
--- a/av1/common/warped_motion.c
+++ b/av1/common/warped_motion.c
@@ -487,7 +487,7 @@
int p_row, int p_width, int p_height,
int p_stride, int subsampling_x,
int subsampling_y, int x_scale, int y_scale,
- int bd) {
+ int bd, int ref_frm) {
int i, j;
ProjectPointsFunc projectpoints = get_project_points_type(wm->wmtype);
uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
@@ -502,8 +502,15 @@
subsampling_y);
out[0] = ROUND_POWER_OF_TWO_SIGNED(out[0] * x_scale, 4);
out[1] = ROUND_POWER_OF_TWO_SIGNED(out[1] * y_scale, 4);
- pred[(j - p_col) + (i - p_row) * p_stride] = highbd_warp_interpolate(
- ref, out[0], out[1], width, height, stride, bd);
+ if (ref_frm)
+ pred[(j - p_col) + (i - p_row) * p_stride] = ROUND_POWER_OF_TWO(
+ pred[(j - p_col) + (i - p_row) * p_stride] +
+ highbd_warp_interpolate(ref, out[0], out[1], width, height,
+ stride, bd),
+ 1);
+ else
+ pred[(j - p_col) + (i - p_row) * p_stride] = highbd_warp_interpolate(
+ ref, out[0], out[1], width, height, stride, bd);
}
}
}
@@ -542,7 +549,7 @@
int height, int stride, uint8_t *pred, int p_col,
int p_row, int p_width, int p_height, int p_stride,
int subsampling_x, int subsampling_y, int x_scale,
- int y_scale) {
+ int y_scale, int ref_frm) {
int i, j;
ProjectPointsFunc projectpoints = get_project_points_type(wm->wmtype);
if (projectpoints == NULL) return;
@@ -555,8 +562,14 @@
subsampling_y);
out[0] = ROUND_POWER_OF_TWO_SIGNED(out[0] * x_scale, 4);
out[1] = ROUND_POWER_OF_TWO_SIGNED(out[1] * y_scale, 4);
- pred[(j - p_col) + (i - p_row) * p_stride] =
- warp_interpolate(ref, out[0], out[1], width, height, stride);
+ if (ref_frm)
+ pred[(j - p_col) + (i - p_row) * p_stride] = ROUND_POWER_OF_TWO(
+ pred[(j - p_col) + (i - p_row) * p_stride] +
+ warp_interpolate(ref, out[0], out[1], width, height, stride),
+ 1);
+ else
+ pred[(j - p_col) + (i - p_row) * p_stride] =
+ warp_interpolate(ref, out[0], out[1], width, height, stride);
}
}
}
@@ -587,17 +600,17 @@
uint8_t *ref, int width, int height, int stride,
uint8_t *pred, int p_col, int p_row, int p_width,
int p_height, int p_stride, int subsampling_x,
- int subsampling_y, int x_scale, int y_scale) {
+ int subsampling_y, int x_scale, int y_scale, int ref_frm) {
#if CONFIG_AOM_HIGHBITDEPTH
if (use_hbd)
highbd_warp_plane(wm, ref, width, height, stride, pred, p_col, p_row,
p_width, p_height, p_stride, subsampling_x, subsampling_y,
- x_scale, y_scale, bd);
+ x_scale, y_scale, bd, ref_frm);
else
#endif // CONFIG_AOM_HIGHBITDEPTH
warp_plane(wm, ref, width, height, stride, pred, p_col, p_row, p_width,
p_height, p_stride, subsampling_x, subsampling_y, x_scale,
- y_scale);
+ y_scale, ref_frm);
}
void av1_integerize_model(const double *model, TransformationType wmtype,
diff --git a/av1/common/warped_motion.h b/av1/common/warped_motion.h
index da92599..e7b9038 100644
--- a/av1/common/warped_motion.h
+++ b/av1/common/warped_motion.h
@@ -64,7 +64,7 @@
uint8_t *ref, int width, int height, int stride,
uint8_t *pred, int p_col, int p_row, int p_width,
int p_height, int p_stride, int subsampling_x,
- int subsampling_y, int x_scale, int y_scale);
+ int subsampling_y, int x_scale, int y_scale, int ref_frm);
// Integerize model into the WarpedMotionParams structure
void av1_integerize_model(const double *model, TransformationType wmtype,
diff --git a/av1/decoder/decodeframe.c b/av1/decoder/decodeframe.c
index 38e2539..e39d08e 100644
--- a/av1/decoder/decodeframe.c
+++ b/av1/decoder/decodeframe.c
@@ -293,7 +293,7 @@
if (mbmi->sb_type < BLOCK_8X8)
if (plane == 0) mode = xd->mi[0]->bmi[(row << 1) + col].as_mode;
- av1_predict_intra_block(xd, pd->n4_wl, pd->n4_hl, tx_size, mode, dst,
+ av1_predict_intra_block(xd, pd->width, pd->height, tx_size, mode, dst,
pd->dst.stride, dst, pd->dst.stride, col, row, plane);
if (!mbmi->skip) {
@@ -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);
@@ -312,7 +315,7 @@
#if CONFIG_VAR_TX
static void decode_reconstruct_tx(AV1_COMMON *cm, MACROBLOCKD *const xd,
aom_reader *r, MB_MODE_INFO *const mbmi,
- int plane, BLOCK_SIZE plane_bsize, int block,
+ int plane, BLOCK_SIZE plane_bsize,
int blk_row, int blk_col, TX_SIZE tx_size,
int *eob_total) {
const struct macroblockd_plane *const pd = &xd->plane[plane];
@@ -334,7 +337,8 @@
if (tx_size == plane_tx_size) {
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
- TX_TYPE tx_type = get_tx_type(plane_type, xd, block, plane_tx_size);
+ int block_idx = (blk_row << 1) + blk_col;
+ TX_TYPE tx_type = get_tx_type(plane_type, xd, block_idx, plane_tx_size);
const SCAN_ORDER *sc = get_scan(cm, plane_tx_size, tx_type, 1);
int16_t max_scan_line = 0;
const int eob =
@@ -355,13 +359,11 @@
for (i = 0; i < 4; ++i) {
const int offsetr = blk_row + ((i >> 1) << bsl);
const int offsetc = blk_col + ((i & 0x01) << bsl);
- int step = num_4x4_blocks_txsize_lookup[tx_size - 1];
if (offsetr >= max_blocks_high || offsetc >= max_blocks_wide) continue;
- decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize,
- block + i * step, offsetr, offsetc, tx_size - 1,
- eob_total);
+ decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, offsetr,
+ offsetc, tx_size - 1, eob_total);
}
}
}
@@ -385,6 +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],
@@ -1286,8 +1291,6 @@
const TX_SIZE max_tx_size = max_txsize_lookup[plane_bsize];
const int bw_var_tx = tx_size_high_unit[max_tx_size];
const int bh_var_tx = tx_size_wide_unit[max_tx_size];
- const int step = num_4x4_blocks_txsize_lookup[max_tx_size];
- int block = 0;
#if CONFIG_EXT_TX && CONFIG_RECT_TX
if (is_rect_tx(mbmi->tx_size)) {
const TX_SIZE tx_size =
@@ -1309,13 +1312,10 @@
plane, row, col, tx_size);
} else {
#endif
- for (row = 0; row < num_4x4_h; row += bh_var_tx) {
- for (col = 0; col < num_4x4_w; col += bw_var_tx) {
- decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, block,
- row, col, max_tx_size, &eobtotal);
- block += step;
- }
- }
+ for (row = 0; row < num_4x4_h; row += bh_var_tx)
+ for (col = 0; col < num_4x4_w; col += bw_var_tx)
+ decode_reconstruct_tx(cm, xd, r, mbmi, plane, plane_bsize, row,
+ col, max_tx_size, &eobtotal);
#if CONFIG_EXT_TX && CONFIG_RECT_TX
}
#endif
diff --git a/av1/decoder/decodemv.c b/av1/decoder/decodemv.c
index 0ad53fa..3993e72 100644
--- a/av1/decoder/decodemv.c
+++ b/av1/decoder/decodemv.c
@@ -294,7 +294,7 @@
if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
- if (depth == 2) {
+ if (depth == MAX_VARTX_DEPTH) {
int idx, idy;
inter_tx_size[0][0] = tx_size;
for (idy = 0; idy < num_4x4_blocks_high_txsize_lookup[tx_size] / 2; ++idy)
diff --git a/av1/encoder/bitstream.c b/av1/encoder/bitstream.c
index d8e12d5..df27fbc 100644
--- a/av1/encoder/bitstream.c
+++ b/av1/encoder/bitstream.c
@@ -70,37 +70,9 @@
};
#endif // CONFIG_EXT_INTER
#if CONFIG_PALETTE
-static const struct av1_token palette_size_encodings[] = {
- { 0, 1 }, { 2, 2 }, { 6, 3 }, { 14, 4 }, { 30, 5 }, { 62, 6 }, { 63, 6 },
-};
-static const struct av1_token
- palette_color_encodings[PALETTE_MAX_SIZE - 1][PALETTE_MAX_SIZE] = {
- { { 0, 1 }, { 1, 1 } }, // 2 colors
- { { 0, 1 }, { 2, 2 }, { 3, 2 } }, // 3 colors
- { { 0, 1 }, { 2, 2 }, { 6, 3 }, { 7, 3 } }, // 4 colors
- { { 0, 1 }, { 2, 2 }, { 6, 3 }, { 14, 4 }, { 15, 4 } }, // 5 colors
- { { 0, 1 },
- { 2, 2 },
- { 6, 3 },
- { 14, 4 },
- { 30, 5 },
- { 31, 5 } }, // 6 colors
- { { 0, 1 },
- { 2, 2 },
- { 6, 3 },
- { 14, 4 },
- { 30, 5 },
- { 62, 6 },
- { 63, 6 } }, // 7 colors
- { { 0, 1 },
- { 2, 2 },
- { 6, 3 },
- { 14, 4 },
- { 30, 5 },
- { 62, 6 },
- { 126, 7 },
- { 127, 7 } }, // 8 colors
- };
+static struct av1_token palette_size_encodings[PALETTE_MAX_SIZE - 1];
+static struct av1_token palette_color_encodings[PALETTE_MAX_SIZE - 1]
+ [PALETTE_MAX_SIZE];
#endif // CONFIG_PALETTE
static const struct av1_token tx_size_encodings[TX_SIZES - 1][TX_SIZES] = {
{ { 0, 1 }, { 1, 1 } }, // Max tx_size is 8X8
@@ -145,8 +117,10 @@
#endif // CONFIG_LOOP_RESTORATION
void av1_encode_token_init(void) {
-#if CONFIG_EXT_TX
+#if CONFIG_EXT_TX || CONFIG_PALETTE
int s;
+#endif // CONFIG_EXT_TX || CONFIG_PALETTE
+#if CONFIG_EXT_TX
for (s = 1; s < EXT_TX_SETS_INTER; ++s) {
av1_tokens_from_tree(ext_tx_inter_encodings[s], av1_ext_tx_inter_tree[s]);
}
@@ -163,6 +137,13 @@
av1_tokens_from_tree(inter_mode_encodings, av1_inter_mode_tree);
#endif
+#if CONFIG_PALETTE
+ av1_tokens_from_tree(palette_size_encodings, av1_palette_size_tree);
+ for (s = 0; s < PALETTE_MAX_SIZE - 1; ++s) {
+ av1_tokens_from_tree(palette_color_encodings[s], av1_palette_color_tree[s]);
+ }
+#endif // CONFIG_PALETTE
+
#if CONFIG_EXT_INTRA
av1_tokens_from_tree(intra_filter_encodings, av1_intra_filter_tree);
#endif // CONFIG_EXT_INTRA
@@ -379,7 +360,7 @@
if (blk_row >= max_blocks_high || blk_col >= max_blocks_wide) return;
- if (depth == 2) {
+ if (depth == MAX_VARTX_DEPTH) {
txfm_partition_update(xd->above_txfm_context + tx_col,
xd->left_txfm_context + tx_row, tx_size);
return;
diff --git a/av1/encoder/block.h b/av1/encoder/block.h
index 0f9ef46..5c8a2f6 100644
--- a/av1/encoder/block.h
+++ b/av1/encoder/block.h
@@ -80,8 +80,7 @@
MACROBLOCKD e_mbd;
MB_MODE_INFO_EXT *mbmi_ext;
int skip_block;
- int select_tx_size;
- int q_index;
+ int qindex;
// The equivalent error at the current rdmult of one whole bit (not one
// bitcost unit).
@@ -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 410498b..4041b15 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
@@ -852,7 +847,6 @@
if (cyclic_refresh_segment_id_boosted(segment_id)) {
int q = av1_get_qindex(&cm->seg, segment_id, cm->base_qindex);
- assert(q == xd->qindex[segment_id]);
set_vbp_thresholds(cpi, thresholds, q);
}
}
@@ -1606,7 +1600,6 @@
av1_init_plane_quantizers(cpi, x, segment_id);
aom_clear_system_state();
segment_qindex = av1_get_qindex(&cm->seg, segment_id, cm->base_qindex);
- assert(segment_qindex == x->e_mbd.qindex[segment_id]);
return av1_compute_rd_mult(cpi, segment_qindex + cm->y_dc_delta_q);
}
@@ -1690,7 +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) {
@@ -4675,7 +4667,6 @@
: cm->base_qindex;
xd->lossless[i] = qindex == 0 && cm->y_dc_delta_q == 0 &&
cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
- xd->qindex[i] = qindex;
}
if (!cm->seg.enabled && xd->lossless[0]) x->optimize = 0;
diff --git a/av1/encoder/encodemb.c b/av1/encoder/encodemb.c
index 8914ba5..6b7e72c 100644
--- a/av1/encoder/encodemb.c
+++ b/av1/encoder/encodemb.c
@@ -95,8 +95,7 @@
#endif
const int shift = get_tx_scale(xd, tx_type, tx_size);
#if CONFIG_NEW_QUANT
- int dq = get_dq_profile_from_ctx(xd->qindex[xd->mi[0]->mbmi.segment_id], ctx,
- ref, plane_type);
+ int dq = get_dq_profile_from_ctx(mb->qindex, ctx, ref, plane_type);
const dequant_val_type_nuq *dequant_val = pd->dequant_val_nuq[dq];
#else
const int dq_step[2] = { dequant_ptr[0] >> shift, dequant_ptr[1] >> shift };
@@ -123,8 +122,7 @@
int shortcut = 0;
int next_shortcut = 0;
- assert((xd->qindex[xd->mi[0]->mbmi.segment_id] == 0) ^
- (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
+ assert((mb->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
token_costs += band;
@@ -518,8 +516,7 @@
tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
- int dq = get_dq_profile_from_ctx(xd->qindex[xd->mi[0]->mbmi.segment_id], ctx,
- is_inter, plane_type);
+ int dq = get_dq_profile_from_ctx(x->qindex, ctx, is_inter, plane_type);
uint16_t *const eob = &p->eobs[block];
const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
const int16_t *src_diff;
@@ -527,8 +524,7 @@
FWD_TXFM_PARAM fwd_txfm_param;
- assert((xd->qindex[xd->mi[0]->mbmi.segment_id] == 0) ^
- (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
+ assert((x->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
fwd_txfm_param.tx_type = tx_type;
fwd_txfm_param.tx_size = tx_size;
@@ -588,8 +584,7 @@
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
const SCAN_ORDER *const scan_order = get_scan(cm, tx_size, tx_type, is_inter);
- int dq = get_dq_profile_from_ctx(xd->qindex[xd->mi[0]->mbmi.segment_id], ctx,
- is_inter, plane_type);
+ int dq = get_dq_profile_from_ctx(x->qindex, ctx, is_inter, plane_type);
tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);
tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);
tran_low_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
@@ -600,8 +595,7 @@
FWD_TXFM_PARAM fwd_txfm_param;
- assert((xd->qindex[xd->mi[0]->mbmi.segment_id] == 0) ^
- (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
+ assert((x->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
fwd_txfm_param.tx_type = tx_type;
fwd_txfm_param.tx_size = tx_size;
@@ -665,13 +659,11 @@
const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
const int16_t *src_diff;
const int is_inter = is_inter_block(&xd->mi[0]->mbmi);
- int dq = get_dq_profile_from_ctx(xd->qindex[xd->mi[0]->mbmi.segment_id], ctx,
- is_inter, plane_type);
+ int dq = get_dq_profile_from_ctx(x->qindex, ctx, is_inter, plane_type);
FWD_TXFM_PARAM fwd_txfm_param;
- assert((xd->qindex[xd->mi[0]->mbmi.segment_id] == 0) ^
- (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
+ assert((x->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
fwd_txfm_param.tx_type = tx_type;
fwd_txfm_param.tx_size = tx_size;
@@ -730,13 +722,11 @@
const int diff_stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
const int16_t *src_diff;
const int is_inter = is_inter_block(&xd->mi[0]->mbmi);
- int dq = get_dq_profile_from_ctx(xd->qindex[xd->mi[0]->mbmi.segment_id], ctx,
- is_inter, plane_type);
+ int dq = get_dq_profile_from_ctx(x->qindex, ctx, is_inter, plane_type);
FWD_TXFM_PARAM fwd_txfm_param;
- assert((xd->qindex[xd->mi[0]->mbmi.segment_id] == 0) ^
- (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
+ assert((x->qindex == 0) ^ (xd->lossless[xd->mi[0]->mbmi.segment_id] == 0));
fwd_txfm_param.tx_type = tx_type;
fwd_txfm_param.tx_size = tx_size;
@@ -1077,16 +1067,14 @@
PLANE_TYPE plane_type = (plane == 0) ? PLANE_TYPE_Y : PLANE_TYPE_UV;
const TX_TYPE tx_type = get_tx_type(plane_type, xd, block, tx_size);
PREDICTION_MODE mode;
- const int bwl = b_width_log2_lookup[plane_bsize];
- const int bhl = b_height_log2_lookup[plane_bsize];
- const int diff_stride = 4 * (1 << bwl);
+ const int diff_stride = block_size_wide[plane_bsize];
uint8_t *src, *dst;
int16_t *src_diff;
uint16_t *eob = &p->eobs[block];
const int src_stride = p->src.stride;
const int dst_stride = pd->dst.stride;
- const int tx1d_width = num_4x4_blocks_wide_txsize_lookup[tx_size] << 2;
- const int tx1d_height = num_4x4_blocks_high_txsize_lookup[tx_size] << 2;
+ const int tx1d_width = tx_size_wide[tx_size];
+ const int tx1d_height = tx_size_high[tx_size];
ENTROPY_CONTEXT *a = NULL, *l = NULL;
int ctx;
@@ -1097,10 +1085,9 @@
dst = &pd->dst.buf[4 * (blk_row * dst_stride + blk_col)];
src = &p->src.buf[4 * (blk_row * src_stride + blk_col)];
src_diff = &p->src_diff[4 * (blk_row * diff_stride + blk_col)];
-
mode = plane == 0 ? get_y_mode(xd->mi[0], block) : mbmi->uv_mode;
- av1_predict_intra_block(xd, bwl, bhl, tx_size, mode, dst, dst_stride, dst,
- dst_stride, blk_col, blk_row, plane);
+ av1_predict_intra_block(xd, pd->width, pd->height, tx_size, mode, dst,
+ dst_stride, dst, dst_stride, blk_col, blk_row, plane);
#if CONFIG_AOM_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
aom_highbd_subtract_block(tx1d_height, tx1d_width, src_diff, diff_stride,
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 0c66905..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;
@@ -751,12 +730,10 @@
#endif // CONFIG_EXT_REFS
static INLINE unsigned int get_token_alloc(int mb_rows, int mb_cols) {
- // TODO(JBB): double check we can't exceed this token count if we have a
- // 32x32 transform crossing a boundary at a multiple of 16.
- // mb_rows, cols are in units of 16 pixels. We assume 3 planes all at full
- // resolution. We assume up to 1 token per pixel, and then allow
- // a head room of 1 EOSB token per 8x8 block per plane.
- return mb_rows * mb_cols * (16 * 16 + 4) * 3;
+ // We assume 3 planes all at full resolution. We assume up to 1 token per
+ // pixel, and then allow a head room of 1 EOSB token per 4x4 block per plane,
+ // plus EOSB_TOKEN per plane.
+ return mb_rows * mb_cols * (16 * 16 + 17) * 3;
}
// Get the allocated token size for a tile. It does the same calculation as in
diff --git a/av1/encoder/firstpass.c b/av1/encoder/firstpass.c
index 466cb9c..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.
@@ -579,13 +582,15 @@
set_mi_row_col(xd, &tile, mb_row << 1, num_8x8_blocks_high_lookup[bsize],
mb_col << 1, num_8x8_blocks_wide_lookup[bsize],
cm->mi_rows, cm->mi_cols);
+ set_plane_n4(xd, num_8x8_blocks_wide_lookup[bsize],
+ num_8x8_blocks_high_lookup[bsize],
+ mi_width_log2_lookup[bsize], mi_height_log2_lookup[bsize]);
// Do intra 16x16 prediction.
xd->mi[0]->mbmi.segment_id = 0;
#if CONFIG_SUPERTX
xd->mi[0]->mbmi.segment_id_supertx = 0;
#endif // CONFIG_SUPERTX
- xd->qindex[xd->mi[0]->mbmi.segment_id] = qindex;
xd->lossless[xd->mi[0]->mbmi.segment_id] = (qindex == 0);
xd->mi[0]->mbmi.mode = DC_PRED;
xd->mi[0]->mbmi.tx_size =
diff --git a/av1/encoder/mbgraph.c b/av1/encoder/mbgraph.c
index 9bbed2b..1fd1682 100644
--- a/av1/encoder/mbgraph.c
+++ b/av1/encoder/mbgraph.c
@@ -149,7 +149,7 @@
unsigned int err;
xd->mi[0]->mbmi.mode = mode;
- av1_predict_intra_block(xd, 2, 2, TX_16X16, mode, x->plane[0].src.buf,
+ av1_predict_intra_block(xd, 16, 16, TX_16X16, mode, x->plane[0].src.buf,
x->plane[0].src.stride, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, 0, 0, 0);
err = aom_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
diff --git a/av1/encoder/quantize.c b/av1/encoder/quantize.c
index db2fdb8..771f94b 100644
--- a/av1/encoder/quantize.c
+++ b/av1/encoder/quantize.c
@@ -1293,11 +1293,11 @@
}
x->skip_block = segfeature_active(&cm->seg, segment_id, SEG_LVL_SKIP);
- x->q_index = qindex;
+ x->qindex = qindex;
set_error_per_bit(x, rdmult);
- av1_initialize_me_consts(cpi, x, x->q_index);
+ av1_initialize_me_consts(cpi, x, qindex);
}
void av1_frame_init_quantizer(AV1_COMP *cpi) {
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index faf330b..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,
@@ -1944,8 +1960,9 @@
int16_t *const src_diff =
av1_raster_block_offset_int16(BLOCK_8X8, block, p->src_diff);
xd->mi[0]->bmi[block].as_mode = mode;
- av1_predict_intra_block(xd, 1, 1, TX_4X4, mode, dst, dst_stride, dst,
- dst_stride, col + idx, row + idy, 0);
+ av1_predict_intra_block(xd, pd->width, pd->height, TX_4X4, mode, dst,
+ dst_stride, dst, dst_stride, col + idx,
+ row + idy, 0);
aom_highbd_subtract_block(4, 4, src_diff, 8, src, src_stride, dst,
dst_stride, xd->bd);
if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
@@ -2064,8 +2081,9 @@
int16_t *const src_diff =
av1_raster_block_offset_int16(BLOCK_8X8, block, p->src_diff);
xd->mi[0]->bmi[block].as_mode = mode;
- av1_predict_intra_block(xd, 1, 1, TX_4X4, mode, dst, dst_stride, dst,
- dst_stride, col + idx, row + idy, 0);
+ av1_predict_intra_block(xd, pd->width, pd->height, TX_4X4, mode, dst,
+ dst_stride, dst, dst_stride, col + idx,
+ row + idy, 0);
aom_subtract_block(4, 4, src_diff, 8, src, src_stride, dst, dst_stride);
if (xd->lossless[xd->mi[0]->mbmi.segment_id]) {
@@ -3116,13 +3134,13 @@
*skip = 0;
}
- if (tx_size > TX_4X4 && depth < 2)
+ if (tx_size > TX_4X4 && depth < MAX_VARTX_DEPTH)
*rate += av1_cost_bit(cpi->common.fc->txfm_partition_prob[ctx], 0);
this_rd = RDCOST(x->rdmult, x->rddiv, *rate, *dist);
tmp_eob = p->eobs[block];
}
- if (tx_size > TX_4X4 && depth < 2) {
+ if (tx_size > TX_4X4 && depth < MAX_VARTX_DEPTH) {
BLOCK_SIZE bsize = txsize_to_bsize[tx_size];
int bsl = b_height_log2_lookup[bsize];
int sub_step = num_4x4_blocks_txsize_lookup[tx_size - 1];
@@ -9373,7 +9391,7 @@
int best_rate_nocoef;
#endif
int64_t distortion2 = 0, distortion_y = 0, dummy_rd = best_rd, this_rd;
- int skippable = 0, rate_overhead = 0;
+ int skippable = 0, rate_overhead_palette = 0;
TX_SIZE best_tx_size, uv_tx;
TX_TYPE best_tx_type;
PALETTE_MODE_INFO palette_mode_info;
@@ -9381,13 +9399,12 @@
x->palette_buffer->best_palette_color_map;
uint8_t *const color_map = xd->plane[0].color_index_map;
- rate_overhead = 0;
mbmi->mode = DC_PRED;
mbmi->uv_mode = DC_PRED;
mbmi->ref_frame[0] = INTRA_FRAME;
mbmi->ref_frame[1] = NONE;
palette_mode_info.palette_size[0] = 0;
- rate_overhead = rd_pick_palette_intra_sby(
+ rate_overhead_palette = rd_pick_palette_intra_sby(
cpi, x, bsize, palette_ctx, intra_mode_cost[DC_PRED],
&palette_mode_info, best_palette_color_map, &best_tx_size,
&best_tx_type, &mode_selected, &dummy_rd);
@@ -9436,7 +9453,7 @@
#endif // CONFIG_FILTER_INTRA
skippable = skippable && skip_uvs[uv_tx];
distortion2 = distortion_y + dist_uvs[uv_tx];
- rate2 = rate_y + rate_overhead + rate_uv_intra[uv_tx];
+ rate2 = rate_y + rate_overhead_palette + rate_uv_intra[uv_tx];
rate2 += ref_costs_single[INTRA_FRAME];
if (skippable) {
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
