blob: 0c242190355d4fea1be80303fa10b945b26fe3c3 [file] [log] [blame] [edit]
/*
* Copyright (c) 2021, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 3-Clause Clear License
* and the Alliance for Open Media Patent License 1.0. If the BSD 3-Clause Clear
* License was not distributed with this source code in the LICENSE file, you
* can obtain it at aomedia.org/license/software-license/bsd-3-c-c/. If the
* Alliance for Open Media Patent License 1.0 was not distributed with this
* source code in the PATENTS file, you can obtain it at
* aomedia.org/license/patent-license/.
*/
#include <math.h>
#include "config/aom_config.h"
#include "config/aom_dsp_rtcd.h"
#include "config/av1_rtcd.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
#include "aom_ports/aom_once.h"
#include "aom_ports/mem.h"
#include "aom_ports/system_state.h"
#include "av1/common/av1_common_int.h"
#include "av1/common/cfl.h"
#include "av1/common/reconintra.h"
enum {
NEED_LEFT = 1 << 1,
NEED_ABOVE = 1 << 2,
NEED_ABOVERIGHT = 1 << 3,
NEED_ABOVELEFT = 1 << 4,
NEED_BOTTOMLEFT = 1 << 5,
};
#define INTRA_EDGE_FILT 3
#define INTRA_EDGE_TAPS 5
#define MAX_UPSAMPLE_SZ 16
#if CONFIG_MRLS
#define NUM_INTRA_NEIGHBOUR_PIXELS (MAX_TX_SIZE * 2 + 64)
#else
#define NUM_INTRA_NEIGHBOUR_PIXELS (MAX_TX_SIZE * 2 + 32)
#endif
static const uint8_t extend_modes[INTRA_MODES] = {
NEED_ABOVE | NEED_LEFT
#if CONFIG_ORIP
| NEED_ABOVELEFT
#endif
, // DC
NEED_ABOVE, // V
NEED_LEFT, // H
NEED_ABOVE | NEED_ABOVERIGHT, // D45
NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT, // D135
NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT, // D113
NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT, // D157
NEED_LEFT | NEED_BOTTOMLEFT, // D203
NEED_ABOVE | NEED_ABOVERIGHT, // D67
NEED_LEFT | NEED_ABOVE
#if CONFIG_ORIP
| NEED_ABOVELEFT
#endif
, // SMOOTH
NEED_LEFT | NEED_ABOVE, // SMOOTH_V
NEED_LEFT | NEED_ABOVE, // SMOOTH_H
NEED_LEFT | NEED_ABOVE | NEED_ABOVELEFT, // PAETH
};
// Tables to store if the top-right reference pixels are available. The flags
// are represented with bits, packed into 8-bit integers. E.g., for the 32x32
// blocks in a 128x128 superblock, the index of the "o" block is 10 (in raster
// order), so its flag is stored at the 3rd bit of the 2nd entry in the table,
// i.e. (table[10 / 8] >> (10 % 8)) & 1.
// . . . .
// . . . .
// . . o .
// . . . .
static uint8_t has_tr_4x4[128] = {
255, 255, 255, 255, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
127, 127, 127, 127, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
255, 127, 255, 127, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
127, 127, 127, 127, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
255, 255, 255, 127, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
127, 127, 127, 127, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
255, 127, 255, 127, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
127, 127, 127, 127, 85, 85, 85, 85, 119, 119, 119, 119, 85, 85, 85, 85,
};
static uint8_t has_tr_4x8[64] = {
255, 255, 255, 255, 119, 119, 119, 119, 127, 127, 127, 127, 119,
119, 119, 119, 255, 127, 255, 127, 119, 119, 119, 119, 127, 127,
127, 127, 119, 119, 119, 119, 255, 255, 255, 127, 119, 119, 119,
119, 127, 127, 127, 127, 119, 119, 119, 119, 255, 127, 255, 127,
119, 119, 119, 119, 127, 127, 127, 127, 119, 119, 119, 119,
};
static uint8_t has_tr_8x4[64] = {
255, 255, 0, 0, 85, 85, 0, 0, 119, 119, 0, 0, 85, 85, 0, 0,
127, 127, 0, 0, 85, 85, 0, 0, 119, 119, 0, 0, 85, 85, 0, 0,
255, 127, 0, 0, 85, 85, 0, 0, 119, 119, 0, 0, 85, 85, 0, 0,
127, 127, 0, 0, 85, 85, 0, 0, 119, 119, 0, 0, 85, 85, 0, 0,
};
static uint8_t has_tr_8x8[32] = {
255, 255, 85, 85, 119, 119, 85, 85, 127, 127, 85, 85, 119, 119, 85, 85,
255, 127, 85, 85, 119, 119, 85, 85, 127, 127, 85, 85, 119, 119, 85, 85,
};
static uint8_t has_tr_8x16[16] = {
255, 255, 119, 119, 127, 127, 119, 119,
255, 127, 119, 119, 127, 127, 119, 119,
};
static uint8_t has_tr_16x8[16] = {
255, 0, 85, 0, 119, 0, 85, 0, 127, 0, 85, 0, 119, 0, 85, 0,
};
static uint8_t has_tr_16x16[8] = {
255, 85, 119, 85, 127, 85, 119, 85,
};
static uint8_t has_tr_16x32[4] = { 255, 119, 127, 119 };
static uint8_t has_tr_32x16[4] = { 15, 5, 7, 5 };
static uint8_t has_tr_32x32[2] = { 95, 87 };
static uint8_t has_tr_32x64[1] = { 127 };
static uint8_t has_tr_64x32[1] = { 19 };
static uint8_t has_tr_64x64[1] = { 7 };
static uint8_t has_tr_64x128[1] = { 3 };
static uint8_t has_tr_128x64[1] = { 1 };
static uint8_t has_tr_128x128[1] = { 1 };
static uint8_t has_tr_4x16[32] = {
255, 255, 255, 255, 127, 127, 127, 127, 255, 127, 255,
127, 127, 127, 127, 127, 255, 255, 255, 127, 127, 127,
127, 127, 255, 127, 255, 127, 127, 127, 127, 127,
};
static uint8_t has_tr_16x4[32] = {
255, 0, 0, 0, 85, 0, 0, 0, 119, 0, 0, 0, 85, 0, 0, 0,
127, 0, 0, 0, 85, 0, 0, 0, 119, 0, 0, 0, 85, 0, 0, 0,
};
static uint8_t has_tr_8x32[8] = {
255, 255, 127, 127, 255, 127, 127, 127,
};
static uint8_t has_tr_32x8[8] = {
15, 0, 5, 0, 7, 0, 5, 0,
};
static uint8_t has_tr_16x64[2] = { 255, 127 };
static uint8_t has_tr_64x16[2] = { 3, 1 };
static const uint8_t *const has_tr_tables[BLOCK_SIZES_ALL] = {
// 4X4
has_tr_4x4,
// 4X8, 8X4, 8X8
has_tr_4x8, has_tr_8x4, has_tr_8x8,
// 8X16, 16X8, 16X16
has_tr_8x16, has_tr_16x8, has_tr_16x16,
// 16X32, 32X16, 32X32
has_tr_16x32, has_tr_32x16, has_tr_32x32,
// 32X64, 64X32, 64X64
has_tr_32x64, has_tr_64x32, has_tr_64x64,
// 64x128, 128x64, 128x128
has_tr_64x128, has_tr_128x64, has_tr_128x128,
// 4x16, 16x4, 8x32
has_tr_4x16, has_tr_16x4, has_tr_8x32,
// 32x8, 16x64, 64x16
has_tr_32x8, has_tr_16x64, has_tr_64x16
};
static uint8_t has_tr_vert_8x8[32] = {
255, 255, 0, 0, 119, 119, 0, 0, 127, 127, 0, 0, 119, 119, 0, 0,
255, 127, 0, 0, 119, 119, 0, 0, 127, 127, 0, 0, 119, 119, 0, 0,
};
static uint8_t has_tr_vert_16x16[8] = {
255, 0, 119, 0, 127, 0, 119, 0,
};
static uint8_t has_tr_vert_32x32[2] = { 15, 7 };
static uint8_t has_tr_vert_64x64[1] = { 3 };
// The _vert_* tables are like the ordinary tables above, but describe the
// order we visit square blocks when doing a PARTITION_VERT_A or
// PARTITION_VERT_B. This is the same order as normal except for on the last
// split where we go vertically (TL, BL, TR, BR). We treat the rectangular block
// as a pair of squares, which means that these tables work correctly for both
// mixed vertical partition types.
//
// There are tables for each of the square sizes. Vertical rectangles (like
// BLOCK_16X32) use their respective "non-vert" table
static const uint8_t *const has_tr_vert_tables[BLOCK_SIZES] = {
// 4X4
NULL,
// 4X8, 8X4, 8X8
has_tr_4x8, NULL, has_tr_vert_8x8,
// 8X16, 16X8, 16X16
has_tr_8x16, NULL, has_tr_vert_16x16,
// 16X32, 32X16, 32X32
has_tr_16x32, NULL, has_tr_vert_32x32,
// 32X64, 64X32, 64X64
has_tr_32x64, NULL, has_tr_vert_64x64,
// 64x128, 128x64, 128x128
has_tr_64x128, NULL, has_tr_128x128
};
static const uint8_t *get_has_tr_table(PARTITION_TYPE partition,
BLOCK_SIZE bsize) {
const uint8_t *ret = NULL;
// If this is a mixed vertical partition, look up bsize in orders_vert.
if (partition == PARTITION_VERT_A || partition == PARTITION_VERT_B) {
assert(bsize < BLOCK_SIZES);
ret = has_tr_vert_tables[bsize];
} else {
ret = has_tr_tables[bsize];
}
assert(ret);
return ret;
}
static int has_top_right(const AV1_COMMON *cm, BLOCK_SIZE bsize, int mi_row,
int mi_col, int top_available, int right_available,
PARTITION_TYPE partition, TX_SIZE txsz, int row_off,
int col_off, int ss_x, int ss_y) {
if (!top_available || !right_available) return 0;
const int bw_unit = mi_size_wide[bsize];
const int plane_bw_unit = AOMMAX(bw_unit >> ss_x, 1);
const int top_right_count_unit = tx_size_wide_unit[txsz];
if (row_off > 0) { // Just need to check if enough pixels on the right.
if (block_size_wide[bsize] > block_size_wide[BLOCK_64X64]) {
// Special case: For 128x128 blocks, the transform unit whose
// top-right corner is at the center of the block does in fact have
// pixels available at its top-right corner.
if (row_off == mi_size_high[BLOCK_64X64] >> ss_y &&
col_off + top_right_count_unit == mi_size_wide[BLOCK_64X64] >> ss_x) {
return 1;
}
const int plane_bw_unit_64 = mi_size_wide[BLOCK_64X64] >> ss_x;
const int col_off_64 = col_off % plane_bw_unit_64;
return col_off_64 + top_right_count_unit < plane_bw_unit_64;
}
return col_off + top_right_count_unit < plane_bw_unit;
} else {
// All top-right pixels are in the block above, which is already available.
if (col_off + top_right_count_unit < plane_bw_unit) return 1;
const int bw_in_mi_log2 = mi_size_wide_log2[bsize];
const int bh_in_mi_log2 = mi_size_high_log2[bsize];
const int sb_mi_size = mi_size_high[cm->seq_params.sb_size];
const int blk_row_in_sb = (mi_row & (sb_mi_size - 1)) >> bh_in_mi_log2;
const int blk_col_in_sb = (mi_col & (sb_mi_size - 1)) >> bw_in_mi_log2;
// Top row of superblock: so top-right pixels are in the top and/or
// top-right superblocks, both of which are already available.
if (blk_row_in_sb == 0) return 1;
// Rightmost column of superblock (and not the top row): so top-right pixels
// fall in the right superblock, which is not available yet.
if (((blk_col_in_sb + 1) << bw_in_mi_log2) >= sb_mi_size) {
return 0;
}
// General case (neither top row nor rightmost column): check if the
// top-right block is coded before the current block.
const int this_blk_index =
((blk_row_in_sb + 0) << (MAX_MIB_SIZE_LOG2 - bw_in_mi_log2)) +
blk_col_in_sb + 0;
const int idx1 = this_blk_index / 8;
const int idx2 = this_blk_index % 8;
const uint8_t *has_tr_table = get_has_tr_table(partition, bsize);
return (has_tr_table[idx1] >> idx2) & 1;
}
}
// Similar to the has_tr_* tables, but store if the bottom-left reference
// pixels are available.
static uint8_t has_bl_4x4[128] = {
84, 85, 85, 85, 16, 17, 17, 17, 84, 85, 85, 85, 0, 1, 1, 1, 84, 85, 85,
85, 16, 17, 17, 17, 84, 85, 85, 85, 0, 0, 1, 0, 84, 85, 85, 85, 16, 17,
17, 17, 84, 85, 85, 85, 0, 1, 1, 1, 84, 85, 85, 85, 16, 17, 17, 17, 84,
85, 85, 85, 0, 0, 0, 0, 84, 85, 85, 85, 16, 17, 17, 17, 84, 85, 85, 85,
0, 1, 1, 1, 84, 85, 85, 85, 16, 17, 17, 17, 84, 85, 85, 85, 0, 0, 1,
0, 84, 85, 85, 85, 16, 17, 17, 17, 84, 85, 85, 85, 0, 1, 1, 1, 84, 85,
85, 85, 16, 17, 17, 17, 84, 85, 85, 85, 0, 0, 0, 0,
};
static uint8_t has_bl_4x8[64] = {
16, 17, 17, 17, 0, 1, 1, 1, 16, 17, 17, 17, 0, 0, 1, 0,
16, 17, 17, 17, 0, 1, 1, 1, 16, 17, 17, 17, 0, 0, 0, 0,
16, 17, 17, 17, 0, 1, 1, 1, 16, 17, 17, 17, 0, 0, 1, 0,
16, 17, 17, 17, 0, 1, 1, 1, 16, 17, 17, 17, 0, 0, 0, 0,
};
static uint8_t has_bl_8x4[64] = {
254, 255, 84, 85, 254, 255, 16, 17, 254, 255, 84, 85, 254, 255, 0, 1,
254, 255, 84, 85, 254, 255, 16, 17, 254, 255, 84, 85, 254, 255, 0, 0,
254, 255, 84, 85, 254, 255, 16, 17, 254, 255, 84, 85, 254, 255, 0, 1,
254, 255, 84, 85, 254, 255, 16, 17, 254, 255, 84, 85, 254, 255, 0, 0,
};
static uint8_t has_bl_8x8[32] = {
84, 85, 16, 17, 84, 85, 0, 1, 84, 85, 16, 17, 84, 85, 0, 0,
84, 85, 16, 17, 84, 85, 0, 1, 84, 85, 16, 17, 84, 85, 0, 0,
};
static uint8_t has_bl_8x16[16] = {
16, 17, 0, 1, 16, 17, 0, 0, 16, 17, 0, 1, 16, 17, 0, 0,
};
static uint8_t has_bl_16x8[16] = {
254, 84, 254, 16, 254, 84, 254, 0, 254, 84, 254, 16, 254, 84, 254, 0,
};
static uint8_t has_bl_16x16[8] = {
84, 16, 84, 0, 84, 16, 84, 0,
};
static uint8_t has_bl_16x32[4] = { 16, 0, 16, 0 };
static uint8_t has_bl_32x16[4] = { 78, 14, 78, 14 };
static uint8_t has_bl_32x32[2] = { 4, 4 };
static uint8_t has_bl_32x64[1] = { 0 };
static uint8_t has_bl_64x32[1] = { 34 };
static uint8_t has_bl_64x64[1] = { 0 };
static uint8_t has_bl_64x128[1] = { 0 };
static uint8_t has_bl_128x64[1] = { 0 };
static uint8_t has_bl_128x128[1] = { 0 };
static uint8_t has_bl_4x16[32] = {
0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0, 0, 0,
0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0, 0, 0,
};
static uint8_t has_bl_16x4[32] = {
254, 254, 254, 84, 254, 254, 254, 16, 254, 254, 254, 84, 254, 254, 254, 0,
254, 254, 254, 84, 254, 254, 254, 16, 254, 254, 254, 84, 254, 254, 254, 0,
};
static uint8_t has_bl_8x32[8] = {
0, 1, 0, 0, 0, 1, 0, 0,
};
static uint8_t has_bl_32x8[8] = {
238, 78, 238, 14, 238, 78, 238, 14,
};
static uint8_t has_bl_16x64[2] = { 0, 0 };
static uint8_t has_bl_64x16[2] = { 42, 42 };
static const uint8_t *const has_bl_tables[BLOCK_SIZES_ALL] = {
// 4X4
has_bl_4x4,
// 4X8, 8X4, 8X8
has_bl_4x8, has_bl_8x4, has_bl_8x8,
// 8X16, 16X8, 16X16
has_bl_8x16, has_bl_16x8, has_bl_16x16,
// 16X32, 32X16, 32X32
has_bl_16x32, has_bl_32x16, has_bl_32x32,
// 32X64, 64X32, 64X64
has_bl_32x64, has_bl_64x32, has_bl_64x64,
// 64x128, 128x64, 128x128
has_bl_64x128, has_bl_128x64, has_bl_128x128,
// 4x16, 16x4, 8x32
has_bl_4x16, has_bl_16x4, has_bl_8x32,
// 32x8, 16x64, 64x16
has_bl_32x8, has_bl_16x64, has_bl_64x16
};
static uint8_t has_bl_vert_8x8[32] = {
254, 255, 16, 17, 254, 255, 0, 1, 254, 255, 16, 17, 254, 255, 0, 0,
254, 255, 16, 17, 254, 255, 0, 1, 254, 255, 16, 17, 254, 255, 0, 0,
};
static uint8_t has_bl_vert_16x16[8] = {
254, 16, 254, 0, 254, 16, 254, 0,
};
static uint8_t has_bl_vert_32x32[2] = { 14, 14 };
static uint8_t has_bl_vert_64x64[1] = { 2 };
// The _vert_* tables are like the ordinary tables above, but describe the
// order we visit square blocks when doing a PARTITION_VERT_A or
// PARTITION_VERT_B. This is the same order as normal except for on the last
// split where we go vertically (TL, BL, TR, BR). We treat the rectangular block
// as a pair of squares, which means that these tables work correctly for both
// mixed vertical partition types.
//
// There are tables for each of the square sizes. Vertical rectangles (like
// BLOCK_16X32) use their respective "non-vert" table
static const uint8_t *const has_bl_vert_tables[BLOCK_SIZES] = {
// 4X4
NULL,
// 4X8, 8X4, 8X8
has_bl_4x8, NULL, has_bl_vert_8x8,
// 8X16, 16X8, 16X16
has_bl_8x16, NULL, has_bl_vert_16x16,
// 16X32, 32X16, 32X32
has_bl_16x32, NULL, has_bl_vert_32x32,
// 32X64, 64X32, 64X64
has_bl_32x64, NULL, has_bl_vert_64x64,
// 64x128, 128x64, 128x128
has_bl_64x128, NULL, has_bl_128x128
};
static const uint8_t *get_has_bl_table(PARTITION_TYPE partition,
BLOCK_SIZE bsize) {
const uint8_t *ret = NULL;
// If this is a mixed vertical partition, look up bsize in orders_vert.
if (partition == PARTITION_VERT_A || partition == PARTITION_VERT_B) {
assert(bsize < BLOCK_SIZES);
ret = has_bl_vert_tables[bsize];
} else {
ret = has_bl_tables[bsize];
}
assert(ret);
return ret;
}
static int has_bottom_left(const AV1_COMMON *cm, BLOCK_SIZE bsize, int mi_row,
int mi_col, int bottom_available, int left_available,
PARTITION_TYPE partition, TX_SIZE txsz, int row_off,
int col_off, int ss_x, int ss_y) {
if (!bottom_available || !left_available) return 0;
// Special case for 128x* blocks, when col_off is half the block width.
// This is needed because 128x* superblocks are divided into 64x* blocks in
// raster order
if (block_size_wide[bsize] > block_size_wide[BLOCK_64X64] && col_off > 0) {
const int plane_bw_unit_64 = mi_size_wide[BLOCK_64X64] >> ss_x;
const int col_off_64 = col_off % plane_bw_unit_64;
if (col_off_64 == 0) {
// We are at the left edge of top-right or bottom-right 64x* block.
const int plane_bh_unit_64 = mi_size_high[BLOCK_64X64] >> ss_y;
const int row_off_64 = row_off % plane_bh_unit_64;
const int plane_bh_unit =
AOMMIN(mi_size_high[bsize] >> ss_y, plane_bh_unit_64);
// Check if all bottom-left pixels are in the left 64x* block (which is
// already coded).
return row_off_64 + tx_size_high_unit[txsz] < plane_bh_unit;
}
}
if (col_off > 0) {
// Bottom-left pixels are in the bottom-left block, which is not available.
return 0;
} else {
const int bh_unit = mi_size_high[bsize];
const int plane_bh_unit = AOMMAX(bh_unit >> ss_y, 1);
const int bottom_left_count_unit = tx_size_high_unit[txsz];
// All bottom-left pixels are in the left block, which is already available.
if (row_off + bottom_left_count_unit < plane_bh_unit) return 1;
const int bw_in_mi_log2 = mi_size_wide_log2[bsize];
const int bh_in_mi_log2 = mi_size_high_log2[bsize];
const int sb_mi_size = mi_size_high[cm->seq_params.sb_size];
const int blk_row_in_sb = (mi_row & (sb_mi_size - 1)) >> bh_in_mi_log2;
const int blk_col_in_sb = (mi_col & (sb_mi_size - 1)) >> bw_in_mi_log2;
// Leftmost column of superblock: so bottom-left pixels maybe in the left
// and/or bottom-left superblocks. But only the left superblock is
// available, so check if all required pixels fall in that superblock.
if (blk_col_in_sb == 0) {
const int blk_start_row_off =
blk_row_in_sb << (bh_in_mi_log2 + MI_SIZE_LOG2 - MI_SIZE_LOG2) >>
ss_y;
const int row_off_in_sb = blk_start_row_off + row_off;
const int sb_height_unit = sb_mi_size >> ss_y;
return row_off_in_sb + bottom_left_count_unit < sb_height_unit;
}
// Bottom row of superblock (and not the leftmost column): so bottom-left
// pixels fall in the bottom superblock, which is not available yet.
if (((blk_row_in_sb + 1) << bh_in_mi_log2) >= sb_mi_size) return 0;
// General case (neither leftmost column nor bottom row): check if the
// bottom-left block is coded before the current block.
const int this_blk_index =
((blk_row_in_sb + 0) << (MAX_MIB_SIZE_LOG2 - bw_in_mi_log2)) +
blk_col_in_sb + 0;
const int idx1 = this_blk_index / 8;
const int idx2 = this_blk_index % 8;
const uint8_t *has_bl_table = get_has_bl_table(partition, bsize);
return (has_bl_table[idx1] >> idx2) & 1;
}
}
typedef void (*intra_pred_fn)(uint8_t *dst, ptrdiff_t stride,
const uint8_t *above, const uint8_t *left);
static intra_pred_fn pred[INTRA_MODES][TX_SIZES_ALL];
static intra_pred_fn dc_pred[2][2][TX_SIZES_ALL];
#if CONFIG_IBP_DC
static intra_pred_fn ibp_dc_pred[2][2][TX_SIZES_ALL];
#endif
typedef void (*intra_high_pred_fn)(uint16_t *dst, ptrdiff_t stride,
const uint16_t *above, const uint16_t *left,
int bd);
static intra_high_pred_fn pred_high[INTRA_MODES][TX_SIZES_ALL];
static intra_high_pred_fn dc_pred_high[2][2][TX_SIZES_ALL];
#if CONFIG_IBP_DC
static intra_high_pred_fn ibp_dc_pred_high[2][2][TX_SIZES_ALL];
#endif
static void init_intra_predictors_internal(void) {
assert(NELEMENTS(mode_to_angle_map) == INTRA_MODES);
#define INIT_RECTANGULAR(p, type) \
p[TX_4X8] = aom_##type##_predictor_4x8; \
p[TX_8X4] = aom_##type##_predictor_8x4; \
p[TX_8X16] = aom_##type##_predictor_8x16; \
p[TX_16X8] = aom_##type##_predictor_16x8; \
p[TX_16X32] = aom_##type##_predictor_16x32; \
p[TX_32X16] = aom_##type##_predictor_32x16; \
p[TX_32X64] = aom_##type##_predictor_32x64; \
p[TX_64X32] = aom_##type##_predictor_64x32; \
p[TX_4X16] = aom_##type##_predictor_4x16; \
p[TX_16X4] = aom_##type##_predictor_16x4; \
p[TX_8X32] = aom_##type##_predictor_8x32; \
p[TX_32X8] = aom_##type##_predictor_32x8; \
p[TX_16X64] = aom_##type##_predictor_16x64; \
p[TX_64X16] = aom_##type##_predictor_64x16;
#define INIT_NO_4X4(p, type) \
p[TX_8X8] = aom_##type##_predictor_8x8; \
p[TX_16X16] = aom_##type##_predictor_16x16; \
p[TX_32X32] = aom_##type##_predictor_32x32; \
p[TX_64X64] = aom_##type##_predictor_64x64; \
INIT_RECTANGULAR(p, type)
#define INIT_ALL_SIZES(p, type) \
p[TX_4X4] = aom_##type##_predictor_4x4; \
INIT_NO_4X4(p, type)
INIT_ALL_SIZES(pred[V_PRED], v);
INIT_ALL_SIZES(pred[H_PRED], h);
INIT_ALL_SIZES(pred[PAETH_PRED], paeth);
INIT_ALL_SIZES(pred[SMOOTH_PRED], smooth);
INIT_ALL_SIZES(pred[SMOOTH_V_PRED], smooth_v);
INIT_ALL_SIZES(pred[SMOOTH_H_PRED], smooth_h);
INIT_ALL_SIZES(dc_pred[0][0], dc_128);
INIT_ALL_SIZES(dc_pred[0][1], dc_top);
INIT_ALL_SIZES(dc_pred[1][0], dc_left);
INIT_ALL_SIZES(dc_pred[1][1], dc);
#if CONFIG_IBP_DC
INIT_ALL_SIZES(ibp_dc_pred[0][0], dc_128);
INIT_ALL_SIZES(ibp_dc_pred[0][1], ibp_dc_top);
INIT_ALL_SIZES(ibp_dc_pred[1][0], ibp_dc_left);
INIT_ALL_SIZES(ibp_dc_pred[1][1], ibp_dc);
#endif
INIT_ALL_SIZES(pred_high[V_PRED], highbd_v);
INIT_ALL_SIZES(pred_high[H_PRED], highbd_h);
INIT_ALL_SIZES(pred_high[PAETH_PRED], highbd_paeth);
INIT_ALL_SIZES(pred_high[SMOOTH_PRED], highbd_smooth);
INIT_ALL_SIZES(pred_high[SMOOTH_V_PRED], highbd_smooth_v);
INIT_ALL_SIZES(pred_high[SMOOTH_H_PRED], highbd_smooth_h);
INIT_ALL_SIZES(dc_pred_high[0][0], highbd_dc_128);
INIT_ALL_SIZES(dc_pred_high[0][1], highbd_dc_top);
INIT_ALL_SIZES(dc_pred_high[1][0], highbd_dc_left);
INIT_ALL_SIZES(dc_pred_high[1][1], highbd_dc);
#if CONFIG_IBP_DC
INIT_ALL_SIZES(ibp_dc_pred_high[0][0], highbd_dc_128);
INIT_ALL_SIZES(ibp_dc_pred_high[0][1], highbd_ibp_dc_top);
INIT_ALL_SIZES(ibp_dc_pred_high[1][0], highbd_ibp_dc_left);
INIT_ALL_SIZES(ibp_dc_pred_high[1][1], highbd_ibp_dc);
#endif
#undef intra_pred_allsizes
}
#if CONFIG_AIMC
// get the context for y_mode_idx
// the context of y_mode_idx depends on the count of directional neighboring
// modes
int get_y_mode_idx_ctx(MACROBLOCKD *const xd) {
const PREDICTION_MODE above_joint_mode =
av1_get_joint_mode(xd->above_right_mbmi);
const PREDICTION_MODE left_joint_mode =
av1_get_joint_mode(xd->bottom_left_mbmi);
const int is_above_angular =
above_joint_mode >= NON_DIRECTIONAL_MODES_COUNT ? 1 : 0;
const int is_left_angular =
left_joint_mode >= NON_DIRECTIONAL_MODES_COUNT ? 1 : 0;
return is_above_angular + is_left_angular;
}
/*! \brief set the luma intra mode and delta angles for a given mode index.
* \param[in] mode_idx mode index in intra mode decision
* process.
* \param[in] mbmi Pointer to structure holding
* the mode info for the current macroblock.
*/
void set_y_mode_and_delta_angle(const int mode_idx, MB_MODE_INFO *const mbmi) {
if (mode_idx < NON_DIRECTIONAL_MODES_COUNT) {
mbmi->mode = mode_idx;
mbmi->angle_delta[PLANE_TYPE_Y] = 0;
} else {
mbmi->mode =
(mode_idx - NON_DIRECTIONAL_MODES_COUNT) / TOTAL_ANGLE_DELTA_COUNT +
NON_DIRECTIONAL_MODES_COUNT;
mbmi->angle_delta[PLANE_TYPE_Y] =
(mode_idx - NON_DIRECTIONAL_MODES_COUNT) % TOTAL_ANGLE_DELTA_COUNT -
MAX_ANGLE_DELTA;
}
mbmi->mode = reordered_y_mode[mbmi->mode];
}
// re-order the intra prediction modes for y component based
// on the neighboring intra prediction modes. The intra prediction
// mode list for 4x4, 4x8, and 8x4 blocks are fixed, and not dependent
// on the intra prediction modes of neighboring blocks
void get_y_intra_mode_set(MB_MODE_INFO *mi, MACROBLOCKD *const xd) {
int neighbor_joint_modes[2];
neighbor_joint_modes[0] = av1_get_joint_mode(xd->bottom_left_mbmi);
neighbor_joint_modes[1] = av1_get_joint_mode(xd->above_right_mbmi);
const int is_left_directional_mode =
neighbor_joint_modes[0] >= NON_DIRECTIONAL_MODES_COUNT ? 1 : 0;
const int is_above_directional_mode =
neighbor_joint_modes[1] >= NON_DIRECTIONAL_MODES_COUNT ? 1 : 0;
// To mark whether each intra prediction mode is added into intra mode list or
// not
int is_mode_selected_list[LUMA_MODE_COUNT];
const int is_small_block = (mi->sb_type[PLANE_TYPE_Y] < BLOCK_8X8);
int i, j;
int mode_idx = 0;
for (i = 0; i < LUMA_MODE_COUNT; i++) {
is_mode_selected_list[i] = -1;
mi->y_intra_mode_list[i] = -1;
}
// always put non-directional modes into the first positions of the mode list
for (i = 0; i < NON_DIRECTIONAL_MODES_COUNT; ++i) {
mi->y_intra_mode_list[mode_idx++] = i;
is_mode_selected_list[i] = 1;
}
if (is_small_block == 0) {
int directional_mode_cnt =
is_above_directional_mode + is_left_directional_mode;
if (directional_mode_cnt == 2 &&
neighbor_joint_modes[0] == neighbor_joint_modes[1])
directional_mode_cnt = 1;
// copy above mode to left mode, if left mode is non-directiona mode and
// above mode is directional mode
if (directional_mode_cnt == 1 && is_left_directional_mode == 0) {
neighbor_joint_modes[0] = neighbor_joint_modes[1];
}
for (i = 0; i < directional_mode_cnt; ++i) {
mi->y_intra_mode_list[mode_idx++] = neighbor_joint_modes[i];
is_mode_selected_list[neighbor_joint_modes[i]] = 1;
}
// Add offsets to derive the neighboring modes
for (i = 0; i < 4; ++i) {
for (j = 0; j < directional_mode_cnt; ++j) {
int left_derived_ode = (neighbor_joint_modes[j] - i +
(56 - NON_DIRECTIONAL_MODES_COUNT - 1)) %
56 +
NON_DIRECTIONAL_MODES_COUNT;
int right_derived_mode =
(neighbor_joint_modes[j] + i - (NON_DIRECTIONAL_MODES_COUNT - 1)) %
56 +
NON_DIRECTIONAL_MODES_COUNT;
if (is_mode_selected_list[left_derived_ode] == -1) {
mi->y_intra_mode_list[mode_idx++] = left_derived_ode;
is_mode_selected_list[left_derived_ode] = 1;
}
if (is_mode_selected_list[right_derived_mode] == -1) {
mi->y_intra_mode_list[mode_idx++] = right_derived_mode;
is_mode_selected_list[right_derived_mode] = 1;
}
}
}
}
// fill the remaining list with default modes
for (i = 0; i < LUMA_MODE_COUNT - NON_DIRECTIONAL_MODES_COUNT &&
mode_idx < LUMA_MODE_COUNT;
++i) {
if (is_mode_selected_list[default_mode_list_y[i] +
NON_DIRECTIONAL_MODES_COUNT] == -1) {
mi->y_intra_mode_list[mode_idx++] =
default_mode_list_y[i] + NON_DIRECTIONAL_MODES_COUNT;
is_mode_selected_list[default_mode_list_y[i] +
NON_DIRECTIONAL_MODES_COUNT] = 1;
}
}
}
// re-order the intra prediction mode of uv component based on the
// intra prediction mode of co-located y block
void get_uv_intra_mode_set(MB_MODE_INFO *mi) {
int is_mode_selected_list[UV_INTRA_MODES];
int i;
int mode_idx = 0;
for (i = 0; i < UV_INTRA_MODES; i++) {
is_mode_selected_list[i] = -1;
mi->uv_intra_mode_list[i] = -1;
}
// check whether co-located y mode is directional mode or not
if (av1_is_directional_mode(mi->mode)) {
mi->uv_intra_mode_list[mode_idx++] = mi->mode;
is_mode_selected_list[mi->mode] = 1;
}
// put non-directional modes into the mode list
mi->uv_intra_mode_list[mode_idx++] = UV_DC_PRED;
is_mode_selected_list[UV_DC_PRED] = 1;
mi->uv_intra_mode_list[mode_idx++] = UV_SMOOTH_PRED;
is_mode_selected_list[UV_SMOOTH_PRED] = 1;
mi->uv_intra_mode_list[mode_idx++] = UV_SMOOTH_V_PRED;
is_mode_selected_list[UV_SMOOTH_V_PRED] = 1;
mi->uv_intra_mode_list[mode_idx++] = UV_SMOOTH_H_PRED;
is_mode_selected_list[UV_SMOOTH_H_PRED] = 1;
mi->uv_intra_mode_list[mode_idx++] = UV_PAETH_PRED;
is_mode_selected_list[UV_PAETH_PRED] = 1;
// fill the remaining list with default modes
const int directional_mode_count = DIR_MODE_END - DIR_MODE_START;
for (i = 0; i < directional_mode_count; ++i) {
if (is_mode_selected_list[default_mode_list_uv[i]] == -1) {
mi->uv_intra_mode_list[mode_idx++] = default_mode_list_uv[i];
is_mode_selected_list[default_mode_list_uv[i]] = 1;
}
}
// put cfl mode into the mode list
mi->uv_intra_mode_list[mode_idx++] = UV_CFL_PRED;
is_mode_selected_list[UV_CFL_PRED] = 1;
}
#endif // CONFIG_AIMC
// Directional prediction, zone 1: 0 < angle < 90
void av1_dr_prediction_z1_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
const uint8_t *above, const uint8_t *left,
int upsample_above, int dx, int dy
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
int r, c, x, base, shift, val;
(void)left;
(void)dy;
assert(dy == 1);
assert(dx > 0);
#if CONFIG_MRLS
const int max_base_x = ((bw + bh) - 1 + (mrl_index << 1)) << upsample_above;
#else
const int max_base_x = ((bw + bh) - 1) << upsample_above;
#endif
const int frac_bits = 6 - upsample_above;
const int base_inc = 1 << upsample_above;
#if CONFIG_MRLS
x = dx * (1 + mrl_index);
#else
x = dx;
#endif
for (r = 0; r < bh; ++r, dst += stride, x += dx) {
base = x >> frac_bits;
shift = ((x << upsample_above) & 0x3F) >> 1;
if (base >= max_base_x) {
for (int i = r; i < bh; ++i) {
memset(dst, above[max_base_x], bw * sizeof(dst[0]));
dst += stride;
}
return;
}
for (c = 0; c < bw; ++c, base += base_inc) {
if (base < max_base_x) {
val = above[base] * (32 - shift) + above[base + 1] * shift;
dst[c] = ROUND_POWER_OF_TWO(val, 5);
} else {
dst[c] = above[max_base_x];
}
}
}
}
// Directional prediction, zone 2: 90 < angle < 180
void av1_dr_prediction_z2_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
const uint8_t *above, const uint8_t *left,
int upsample_above, int upsample_left, int dx,
int dy
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
assert(dx > 0);
assert(dy > 0);
#if CONFIG_MRLS
const int min_base_x = -((1 + mrl_index) << upsample_above);
const int min_base_y = -((1 + mrl_index) << upsample_left);
#else
const int min_base_x = -(1 << upsample_above);
const int min_base_y = -(1 << upsample_left);
#endif
(void)min_base_y;
const int frac_bits_x = 6 - upsample_above;
const int frac_bits_y = 6 - upsample_left;
for (int r = 0; r < bh; ++r) {
for (int c = 0; c < bw; ++c) {
int val;
int y = r + 1;
#if CONFIG_MRLS
int x = (c << 6) - (y + mrl_index) * dx;
#else
int x = (c << 6) - y * dx;
#endif
const int base_x = x >> frac_bits_x;
if (base_x >= min_base_x) {
const int shift = ((x * (1 << upsample_above)) & 0x3F) >> 1;
val = above[base_x] * (32 - shift) + above[base_x + 1] * shift;
val = ROUND_POWER_OF_TWO(val, 5);
} else {
x = c + 1;
#if CONFIG_MRLS
y = (r << 6) - (x + mrl_index) * dy;
#else
y = (r << 6) - x * dy;
#endif
const int base_y = y >> frac_bits_y;
assert(base_y >= min_base_y);
const int shift = ((y * (1 << upsample_left)) & 0x3F) >> 1;
val = left[base_y] * (32 - shift) + left[base_y + 1] * shift;
val = ROUND_POWER_OF_TWO(val, 5);
}
dst[c] = val;
}
dst += stride;
}
}
// Directional prediction, zone 3: 180 < angle < 270
void av1_dr_prediction_z3_c(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
const uint8_t *above, const uint8_t *left,
int upsample_left, int dx, int dy
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
(void)above;
(void)dx;
assert(dx == 1);
assert(dy > 0);
#if CONFIG_MRLS
const int max_base_y = (bw + bh - 1 + (mrl_index << 1)) << upsample_left;
#else
const int max_base_y = (bw + bh - 1) << upsample_left;
#endif
const int frac_bits = 6 - upsample_left;
const int base_inc = 1 << upsample_left;
#if CONFIG_MRLS
int y = dy * (1 + mrl_index);
#else
int y = dy;
#endif
for (int c = 0; c < bw; ++c, y += dy) {
int base = y >> frac_bits;
const int shift = ((y << upsample_left) & 0x3F) >> 1;
for (int r = 0; r < bh; ++r, base += base_inc) {
if (base < max_base_y) {
const int val = left[base] * (32 - shift) + left[base + 1] * shift;
dst[r * stride + c] = ROUND_POWER_OF_TWO(val, 5);
} else {
for (; r < bh; ++r) dst[r * stride + c] = left[max_base_y];
break;
}
}
}
}
static void dr_predictor(uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size,
const uint8_t *above, const uint8_t *left,
int upsample_above, int upsample_left, int angle
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
const int dx = av1_get_dx(angle);
const int dy = av1_get_dy(angle);
const int bw = tx_size_wide[tx_size];
const int bh = tx_size_high[tx_size];
assert(angle > 0 && angle < 270);
if (angle > 0 && angle < 90) {
av1_dr_prediction_z1(dst, stride, bw, bh, above, left, upsample_above, dx,
dy
#if CONFIG_MRLS
,
mrl_index
#endif
);
} else if (angle > 90 && angle < 180) {
av1_dr_prediction_z2(dst, stride, bw, bh, above, left, upsample_above,
upsample_left, dx, dy
#if CONFIG_MRLS
,
mrl_index
#endif
);
} else if (angle > 180 && angle < 270) {
av1_dr_prediction_z3(dst, stride, bw, bh, above, left, upsample_left, dx, dy
#if CONFIG_MRLS
,
mrl_index
#endif
);
} else if (angle == 90) {
pred[V_PRED][tx_size](dst, stride, above, left);
} else if (angle == 180) {
pred[H_PRED][tx_size](dst, stride, above, left);
}
}
#if CONFIG_IBP_DIR
// Generate the second directional predictor for IBP
static void second_dr_predictor(uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size,
const uint8_t *above, const uint8_t *left,
int upsample_above, int upsample_left,
int angle) {
const int bw = tx_size_wide[tx_size];
const int bh = tx_size_high[tx_size];
if (angle > 0 && angle < 90) {
int dy = second_dr_intra_derivative[angle];
int dx = 1;
#if CONFIG_MRLS
av1_dr_prediction_z3(dst, stride, bw, bh, above, left, upsample_left, dx,
dy, 0);
#else
av1_dr_prediction_z3(dst, stride, bw, bh, above, left, upsample_left, dx,
dy);
#endif
} else if (angle > 180 && angle < 270) {
int dx = second_dr_intra_derivative[270 - angle];
int dy = 1;
#if CONFIG_MRLS
av1_dr_prediction_z1(dst, stride, bw, bh, above, left, upsample_above, dx,
dy, 0);
#else
av1_dr_prediction_z1(dst, stride, bw, bh, above, left, upsample_above, dx,
dy);
#endif
}
}
#endif
// Directional prediction, zone 1: 0 < angle < 90
void av1_highbd_dr_prediction_z1_c(uint16_t *dst, ptrdiff_t stride, int bw,
int bh, const uint16_t *above,
const uint16_t *left, int upsample_above,
int dx, int dy, int bd
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
int r, c, x, base, shift, val;
(void)left;
(void)dy;
(void)bd;
assert(dy == 1);
assert(dx > 0);
#if CONFIG_MRLS
const int max_base_x = ((bw + bh) - 1 + (mrl_index << 1)) << upsample_above;
#else
const int max_base_x = ((bw + bh) - 1) << upsample_above;
#endif
const int frac_bits = 6 - upsample_above;
const int base_inc = 1 << upsample_above;
#if CONFIG_MRLS
x = dx * (1 + mrl_index);
#else
x = dx;
#endif
for (r = 0; r < bh; ++r, dst += stride, x += dx) {
base = x >> frac_bits;
shift = ((x << upsample_above) & 0x3F) >> 1;
if (base >= max_base_x) {
for (int i = r; i < bh; ++i) {
aom_memset16(dst, above[max_base_x], bw);
dst += stride;
}
return;
}
for (c = 0; c < bw; ++c, base += base_inc) {
if (base < max_base_x) {
val = above[base] * (32 - shift) + above[base + 1] * shift;
dst[c] = ROUND_POWER_OF_TWO(val, 5);
} else {
dst[c] = above[max_base_x];
}
}
}
}
// Directional prediction, zone 2: 90 < angle < 180
void av1_highbd_dr_prediction_z2_c(uint16_t *dst, ptrdiff_t stride, int bw,
int bh, const uint16_t *above,
const uint16_t *left, int upsample_above,
int upsample_left, int dx, int dy, int bd
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
(void)bd;
assert(dx > 0);
assert(dy > 0);
#if CONFIG_MRLS
const int min_base_x = -(1 << upsample_above) - mrl_index;
const int min_base_y = -(1 << upsample_left) - mrl_index;
#else
const int min_base_x = -(1 << upsample_above);
const int min_base_y = -(1 << upsample_left);
#endif
(void)min_base_y;
const int frac_bits_x = 6 - upsample_above;
const int frac_bits_y = 6 - upsample_left;
for (int r = 0; r < bh; ++r) {
for (int c = 0; c < bw; ++c) {
int val;
int y = r + 1;
#if CONFIG_MRLS
int x = (c << 6) - (y + mrl_index) * dx;
#else
int x = (c << 6) - y * dx;
#endif
const int base_x = x >> frac_bits_x;
if (base_x >= min_base_x) {
const int shift = ((x * (1 << upsample_above)) & 0x3F) >> 1;
val = above[base_x] * (32 - shift) + above[base_x + 1] * shift;
val = ROUND_POWER_OF_TWO(val, 5);
} else {
x = c + 1;
#if CONFIG_MRLS
y = (r << 6) - (x + mrl_index) * dy;
#else
y = (r << 6) - x * dy;
#endif
const int base_y = y >> frac_bits_y;
assert(base_y >= min_base_y);
const int shift = ((y * (1 << upsample_left)) & 0x3F) >> 1;
val = left[base_y] * (32 - shift) + left[base_y + 1] * shift;
val = ROUND_POWER_OF_TWO(val, 5);
}
dst[c] = val;
}
dst += stride;
}
}
// Directional prediction, zone 3: 180 < angle < 270
void av1_highbd_dr_prediction_z3_c(uint16_t *dst, ptrdiff_t stride, int bw,
int bh, const uint16_t *above,
const uint16_t *left, int upsample_left,
int dx, int dy, int bd
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
int r, c, y, base, shift, val;
(void)above;
(void)dx;
(void)bd;
assert(dx == 1);
assert(dy > 0);
#if CONFIG_MRLS
const int max_base_y = ((bw + bh - 1) << upsample_left) + (mrl_index << 1);
#else
const int max_base_y = (bw + bh - 1) << upsample_left;
#endif
const int frac_bits = 6 - upsample_left;
const int base_inc = 1 << upsample_left;
#if CONFIG_MRLS
y = dy * (1 + mrl_index);
#else
y = dy;
#endif
for (c = 0; c < bw; ++c, y += dy) {
base = y >> frac_bits;
shift = ((y << upsample_left) & 0x3F) >> 1;
for (r = 0; r < bh; ++r, base += base_inc) {
if (base < max_base_y) {
val = left[base] * (32 - shift) + left[base + 1] * shift;
dst[r * stride + c] = ROUND_POWER_OF_TWO(val, 5);
} else {
for (; r < bh; ++r) dst[r * stride + c] = left[max_base_y];
break;
}
}
}
}
static void highbd_dr_predictor(uint16_t *dst, ptrdiff_t stride,
TX_SIZE tx_size, const uint16_t *above,
const uint16_t *left, int upsample_above,
int upsample_left, int angle, int bd
#if CONFIG_MRLS
,
int mrl_index
#endif
) {
const int dx = av1_get_dx(angle);
const int dy = av1_get_dy(angle);
const int bw = tx_size_wide[tx_size];
const int bh = tx_size_high[tx_size];
assert(angle > 0 && angle < 270);
if (angle > 0 && angle < 90) {
av1_highbd_dr_prediction_z1(dst, stride, bw, bh, above, left,
upsample_above, dx, dy, bd
#if CONFIG_MRLS
,
mrl_index
#endif
);
} else if (angle > 90 && angle < 180) {
av1_highbd_dr_prediction_z2(dst, stride, bw, bh, above, left,
upsample_above, upsample_left, dx, dy, bd
#if CONFIG_MRLS
,
mrl_index
#endif
);
} else if (angle > 180 && angle < 270) {
av1_highbd_dr_prediction_z3(dst, stride, bw, bh, above, left, upsample_left,
dx, dy, bd
#if CONFIG_MRLS
,
mrl_index
#endif
);
} else if (angle == 90) {
pred_high[V_PRED][tx_size](dst, stride, above, left, bd);
} else if (angle == 180) {
pred_high[H_PRED][tx_size](dst, stride, above, left, bd);
}
}
#if CONFIG_IBP_DIR
// Generate the second directional predictor for IBP
static void highbd_second_dr_predictor(uint16_t *dst, ptrdiff_t stride,
TX_SIZE tx_size, const uint16_t *above,
const uint16_t *left, int upsample_above,
int upsample_left, int angle, int bd) {
const int bw = tx_size_wide[tx_size];
const int bh = tx_size_high[tx_size];
if (angle > 0 && angle < 90) {
int dy = second_dr_intra_derivative[angle];
int dx = 1;
#if CONFIG_MRLS
av1_highbd_dr_prediction_z3(dst, stride, bw, bh, above, left, upsample_left,
dx, dy, bd, 0);
#else
av1_highbd_dr_prediction_z3(dst, stride, bw, bh, above, left, upsample_left,
dx, dy, bd);
#endif
} else if (angle > 180 && angle < 270) {
int dx = second_dr_intra_derivative[270 - angle];
int dy = 1;
#if CONFIG_MRLS
av1_highbd_dr_prediction_z1(dst, stride, bw, bh, above, left,
upsample_above, dx, dy, bd, 0);
#else
av1_highbd_dr_prediction_z1(dst, stride, bw, bh, above, left,
upsample_above, dx, dy, bd);
#endif
}
}
#endif
DECLARE_ALIGNED(16, const int8_t,
av1_filter_intra_taps[FILTER_INTRA_MODES][8][8]) = {
{
{ -6, 10, 0, 0, 0, 12, 0, 0 },
{ -5, 2, 10, 0, 0, 9, 0, 0 },
{ -3, 1, 1, 10, 0, 7, 0, 0 },
{ -3, 1, 1, 2, 10, 5, 0, 0 },
{ -4, 6, 0, 0, 0, 2, 12, 0 },
{ -3, 2, 6, 0, 0, 2, 9, 0 },
{ -3, 2, 2, 6, 0, 2, 7, 0 },
{ -3, 1, 2, 2, 6, 3, 5, 0 },
},
{
{ -10, 16, 0, 0, 0, 10, 0, 0 },
{ -6, 0, 16, 0, 0, 6, 0, 0 },
{ -4, 0, 0, 16, 0, 4, 0, 0 },
{ -2, 0, 0, 0, 16, 2, 0, 0 },
{ -10, 16, 0, 0, 0, 0, 10, 0 },
{ -6, 0, 16, 0, 0, 0, 6, 0 },
{ -4, 0, 0, 16, 0, 0, 4, 0 },
{ -2, 0, 0, 0, 16, 0, 2, 0 },
},
{
{ -8, 8, 0, 0, 0, 16, 0, 0 },
{ -8, 0, 8, 0, 0, 16, 0, 0 },
{ -8, 0, 0, 8, 0, 16, 0, 0 },
{ -8, 0, 0, 0, 8, 16, 0, 0 },
{ -4, 4, 0, 0, 0, 0, 16, 0 },
{ -4, 0, 4, 0, 0, 0, 16, 0 },
{ -4, 0, 0, 4, 0, 0, 16, 0 },
{ -4, 0, 0, 0, 4, 0, 16, 0 },
},
{
{ -2, 8, 0, 0, 0, 10, 0, 0 },
{ -1, 3, 8, 0, 0, 6, 0, 0 },
{ -1, 2, 3, 8, 0, 4, 0, 0 },
{ 0, 1, 2, 3, 8, 2, 0, 0 },
{ -1, 4, 0, 0, 0, 3, 10, 0 },
{ -1, 3, 4, 0, 0, 4, 6, 0 },
{ -1, 2, 3, 4, 0, 4, 4, 0 },
{ -1, 2, 2, 3, 4, 3, 3, 0 },
},
{
{ -12, 14, 0, 0, 0, 14, 0, 0 },
{ -10, 0, 14, 0, 0, 12, 0, 0 },
{ -9, 0, 0, 14, 0, 11, 0, 0 },
{ -8, 0, 0, 0, 14, 10, 0, 0 },
{ -10, 12, 0, 0, 0, 0, 14, 0 },
{ -9, 1, 12, 0, 0, 0, 12, 0 },
{ -8, 0, 0, 12, 0, 1, 11, 0 },
{ -7, 0, 0, 1, 12, 1, 9, 0 },
},
};
void av1_filter_intra_predictor_c(uint8_t *dst, ptrdiff_t stride,
TX_SIZE tx_size, const uint8_t *above,
const uint8_t *left, int mode) {
int r, c;
uint8_t buffer[33][33];
const int bw = tx_size_wide[tx_size];
const int bh = tx_size_high[tx_size];
assert(bw <= 32 && bh <= 32);
// The initialization is just for silencing Jenkins static analysis warnings
for (r = 0; r < bh + 1; ++r)
memset(buffer[r], 0, (bw + 1) * sizeof(buffer[0][0]));
for (r = 0; r < bh; ++r) buffer[r + 1][0] = left[r];
memcpy(buffer[0], &above[-1], (bw + 1) * sizeof(uint8_t));
for (r = 1; r < bh + 1; r += 2)
for (c = 1; c < bw + 1; c += 4) {
const uint8_t p0 = buffer[r - 1][c - 1];
const uint8_t p1 = buffer[r - 1][c];
const uint8_t p2 = buffer[r - 1][c + 1];
const uint8_t p3 = buffer[r - 1][c + 2];
const uint8_t p4 = buffer[r - 1][c + 3];
const uint8_t p5 = buffer[r][c - 1];
const uint8_t p6 = buffer[r + 1][c - 1];
for (int k = 0; k < 8; ++k) {
int r_offset = k >> 2;
int c_offset = k & 0x03;
buffer[r + r_offset][c + c_offset] =
clip_pixel(ROUND_POWER_OF_TWO_SIGNED(
av1_filter_intra_taps[mode][k][0] * p0 +
av1_filter_intra_taps[mode][k][1] * p1 +
av1_filter_intra_taps[mode][k][2] * p2 +
av1_filter_intra_taps[mode][k][3] * p3 +
av1_filter_intra_taps[mode][k][4] * p4 +
av1_filter_intra_taps[mode][k][5] * p5 +
av1_filter_intra_taps[mode][k][6] * p6,
FILTER_INTRA_SCALE_BITS));
}
}
for (r = 0; r < bh; ++r) {
memcpy(dst, &buffer[r + 1][1], bw * sizeof(uint8_t));
dst += stride;
}
}
static void highbd_filter_intra_predictor(uint16_t *dst, ptrdiff_t stride,
TX_SIZE tx_size,
const uint16_t *above,
const uint16_t *left, int mode,
int bd) {
int r, c;
uint16_t buffer[33][33];
const int bw = tx_size_wide[tx_size];
const int bh = tx_size_high[tx_size];
assert(bw <= 32 && bh <= 32);
// The initialization is just for silencing Jenkins static analysis warnings
for (r = 0; r < bh + 1; ++r)
memset(buffer[r], 0, (bw + 1) * sizeof(buffer[0][0]));
for (r = 0; r < bh; ++r) buffer[r + 1][0] = left[r];
memcpy(buffer[0], &above[-1], (bw + 1) * sizeof(buffer[0][0]));
for (r = 1; r < bh + 1; r += 2)
for (c = 1; c < bw + 1; c += 4) {
const uint16_t p0 = buffer[r - 1][c - 1];
const uint16_t p1 = buffer[r - 1][c];
const uint16_t p2 = buffer[r - 1][c + 1];
const uint16_t p3 = buffer[r - 1][c + 2];
const uint16_t p4 = buffer[r - 1][c + 3];
const uint16_t p5 = buffer[r][c - 1];
const uint16_t p6 = buffer[r + 1][c - 1];
for (int k = 0; k < 8; ++k) {
int r_offset = k >> 2;
int c_offset = k & 0x03;
buffer[r + r_offset][c + c_offset] =
clip_pixel_highbd(ROUND_POWER_OF_TWO_SIGNED(
av1_filter_intra_taps[mode][k][0] * p0 +
av1_filter_intra_taps[mode][k][1] * p1 +
av1_filter_intra_taps[mode][k][2] * p2 +
av1_filter_intra_taps[mode][k][3] * p3 +
av1_filter_intra_taps[mode][k][4] * p4 +
av1_filter_intra_taps[mode][k][5] * p5 +
av1_filter_intra_taps[mode][k][6] * p6,
FILTER_INTRA_SCALE_BITS),
bd);
}
}
for (r = 0; r < bh; ++r) {
memcpy(dst, &buffer[r + 1][1], bw * sizeof(dst[0]));
dst += stride;
}
}
static int is_smooth(const MB_MODE_INFO *mbmi, int plane) {
if (plane == 0) {
const PREDICTION_MODE mode = mbmi->mode;
return (mode == SMOOTH_PRED || mode == SMOOTH_V_PRED ||
mode == SMOOTH_H_PRED);
} else {
// uv_mode is not set for inter blocks, so need to explicitly
// detect that case.
#if CONFIG_SDP
if (is_inter_block(mbmi, SHARED_PART)) return 0;
#else
if (is_inter_block(mbmi)) return 0;
#endif
const UV_PREDICTION_MODE uv_mode = mbmi->uv_mode;
return (uv_mode == UV_SMOOTH_PRED || uv_mode == UV_SMOOTH_V_PRED ||
uv_mode == UV_SMOOTH_H_PRED);
}
}
static int get_filt_type(const MACROBLOCKD *xd, int plane) {
int ab_sm, le_sm;
if (plane == 0) {
const MB_MODE_INFO *ab = xd->above_mbmi;
const MB_MODE_INFO *le = xd->left_mbmi;
ab_sm = ab ? is_smooth(ab, plane) : 0;
le_sm = le ? is_smooth(le, plane) : 0;
} else {
const MB_MODE_INFO *ab = xd->chroma_above_mbmi;
const MB_MODE_INFO *le = xd->chroma_left_mbmi;
ab_sm = ab ? is_smooth(ab, plane) : 0;
le_sm = le ? is_smooth(le, plane) : 0;
}
return (ab_sm || le_sm) ? 1 : 0;
}
static int intra_edge_filter_strength(int bs0, int bs1, int delta, int type) {
const int d = abs(delta);
int strength = 0;
const int blk_wh = bs0 + bs1;
if (type == 0) {
if (blk_wh <= 8) {
if (d >= 56) strength = 1;
} else if (blk_wh <= 12) {
if (d >= 40) strength = 1;
} else if (blk_wh <= 16) {
if (d >= 40) strength = 1;
} else if (blk_wh <= 24) {
if (d >= 8) strength = 1;
if (d >= 16) strength = 2;
if (d >= 32) strength = 3;
} else if (blk_wh <= 32) {
if (d >= 1) strength = 1;
if (d >= 4) strength = 2;
if (d >= 32) strength = 3;
} else {
if (d >= 1) strength = 3;
}
} else {
if (blk_wh <= 8) {
if (d >= 40) strength = 1;
if (d >= 64) strength = 2;
} else if (blk_wh <= 16) {
if (d >= 20) strength = 1;
if (d >= 48) strength = 2;
} else if (blk_wh <= 24) {
if (d >= 4) strength = 3;
} else {
if (d >= 1) strength = 3;
}
}
return strength;
}
void av1_filter_intra_edge_c(uint8_t *p, int sz, int strength) {
if (!strength) return;
const int kernel[INTRA_EDGE_FILT][INTRA_EDGE_TAPS] = { { 0, 4, 8, 4, 0 },
{ 0, 5, 6, 5, 0 },
{ 2, 4, 4, 4, 2 } };
const int filt = strength - 1;
uint8_t edge[129];
memcpy(edge, p, sz * sizeof(*p));
for (int i = 1; i < sz; i++) {
int s = 0;
for (int j = 0; j < INTRA_EDGE_TAPS; j++) {
int k = i - 2 + j;
k = (k < 0) ? 0 : k;
k = (k > sz - 1) ? sz - 1 : k;
s += edge[k] * kernel[filt][j];
}
s = (s + 8) >> 4;
p[i] = s;
}
}
static void filter_intra_edge_corner(uint8_t *p_above, uint8_t *p_left) {
const int kernel[3] = { 5, 6, 5 };
int s = (p_left[0] * kernel[0]) + (p_above[-1] * kernel[1]) +
(p_above[0] * kernel[2]);
s = (s + 8) >> 4;
p_above[-1] = s;
p_left[-1] = s;
}
void av1_filter_intra_edge_high_c(uint16_t *p, int sz, int strength) {
if (!strength) return;
const int kernel[INTRA_EDGE_FILT][INTRA_EDGE_TAPS] = { { 0, 4, 8, 4, 0 },
{ 0, 5, 6, 5, 0 },
{ 2, 4, 4, 4, 2 } };
const int filt = strength - 1;
uint16_t edge[129];
memcpy(edge, p, sz * sizeof(*p));
for (int i = 1; i < sz; i++) {
int s = 0;
for (int j = 0; j < INTRA_EDGE_TAPS; j++) {
int k = i - 2 + j;
k = (k < 0) ? 0 : k;
k = (k > sz - 1) ? sz - 1 : k;
s += edge[k] * kernel[filt][j];
}
s = (s + 8) >> 4;
p[i] = s;
}
}
static void filter_intra_edge_corner_high(uint16_t *p_above, uint16_t *p_left) {
const int kernel[3] = { 5, 6, 5 };
int s = (p_left[0] * kernel[0]) + (p_above[-1] * kernel[1]) +
(p_above[0] * kernel[2]);
s = (s + 8) >> 4;
p_above[-1] = s;
p_left[-1] = s;
}
void av1_upsample_intra_edge_c(uint8_t *p, int sz) {
// interpolate half-sample positions
assert(sz <= MAX_UPSAMPLE_SZ);
uint8_t in[MAX_UPSAMPLE_SZ + 3];
// copy p[-1..(sz-1)] and extend first and last samples
in[0] = p[-1];
in[1] = p[-1];
for (int i = 0; i < sz; i++) {
in[i + 2] = p[i];
}
in[sz + 2] = p[sz - 1];
// interpolate half-sample edge positions
p[-2] = in[0];
for (int i = 0; i < sz; i++) {
int s = -in[i] + (9 * in[i + 1]) + (9 * in[i + 2]) - in[i + 3];
s = clip_pixel((s + 8) >> 4);
p[2 * i - 1] = s;
p[2 * i] = in[i + 2];
}
}
void av1_upsample_intra_edge_high_c(uint16_t *p, int sz, int bd) {
// interpolate half-sample positions
assert(sz <= MAX_UPSAMPLE_SZ);
uint16_t in[MAX_UPSAMPLE_SZ + 3];
// copy p[-1..(sz-1)] and extend first and last samples
in[0] = p[-1];
in[1] = p[-1];
for (int i = 0; i < sz; i++) {
in[i + 2] = p[i];
}
in[sz + 2] = p[sz - 1];
// interpolate half-sample edge positions
p[-2] = in[0];
for (int i = 0; i < sz; i++) {
int s = -in[i] + (9 * in[i + 1]) + (9 * in[i + 2]) - in[i + 3];
s = (s + 8) >> 4;
s = clip_pixel_highbd(s, bd);
p[2 * i - 1] = s;
p[2 * i] = in[i + 2];
}
}
#if CONFIG_IBP_DIR
void av1_highbd_ibp_dr_prediction_z1_c(uint8_t *weights, uint16_t *dst,
ptrdiff_t stride, uint16_t *second_pred,
ptrdiff_t second_stride, int bw,
int bh) {
int r, c;
for (r = 0; r < bh; ++r) {
for (c = 0; c < bw; ++c) {
dst[c] = ROUND_POWER_OF_TWO(
dst[c] * weights[c] + second_pred[c] * (IBP_WEIGHT_MAX - weights[c]),
IBP_WEIGHT_SHIFT);
}
weights += bw;
dst += stride;
second_pred += second_stride;
}
}
void av1_highbd_ibp_dr_prediction_z3_c(uint8_t *weights, uint16_t *dst,
ptrdiff_t stride, uint16_t *second_pred,
ptrdiff_t second_stride, int bw,
int bh) {
int r, c;
for (c = 0; c < bw; ++c) {
uint16_t *tmp_dst = dst + c;
uint16_t *tmp_second = second_pred + c;
for (r = 0; r < bh; ++r) {
tmp_dst[0] =
ROUND_POWER_OF_TWO(tmp_dst[0] * weights[r] +
tmp_second[0] * (IBP_WEIGHT_MAX - weights[r]),
IBP_WEIGHT_SHIFT);
tmp_dst += stride;
tmp_second += second_stride;
}
weights += bh;
}
}
void av1_ibp_dr_prediction_z1_c(uint8_t *weights, uint8_t *dst,
ptrdiff_t stride, uint8_t *second_pred,
ptrdiff_t second_stride, int bw, int bh) {
int r, c;
for (r = 0; r < bh; ++r) {
for (c = 0; c < bw; ++c) {
dst[c] = ROUND_POWER_OF_TWO(
dst[c] * weights[c] + second_pred[c] * (IBP_WEIGHT_MAX - weights[c]),
IBP_WEIGHT_SHIFT);
}
weights += bw;
dst += stride;
second_pred += second_stride;
}
}
void av1_ibp_dr_prediction_z3_c(uint8_t *weights, uint8_t *dst,
ptrdiff_t stride, uint8_t *second_pred,
ptrdiff_t second_stride, int bw, int bh) {
int r, c;
for (c = 0; c < bw; ++c) {
uint8_t *tmp_dst = dst + c;
uint8_t *tmp_second = second_pred + c;
for (r = 0; r < bh; ++r) {
tmp_dst[0] =
ROUND_POWER_OF_TWO(tmp_dst[0] * weights[r] +
tmp_second[0] * (IBP_WEIGHT_MAX - weights[r]),
IBP_WEIGHT_SHIFT);
tmp_dst += stride;
tmp_second += second_stride;
}
weights += bh;
}
}
#endif
static void build_intra_predictors_high(
const MACROBLOCKD *xd, const uint8_t *ref8, int ref_stride, uint8_t *dst8,
int dst_stride, PREDICTION_MODE mode, int angle_delta,
FILTER_INTRA_MODE filter_intra_mode, TX_SIZE tx_size,
int disable_edge_filter, int n_top_px, int n_topright_px, int n_left_px,
int n_bottomleft_px, int plane
#if CONFIG_MRLS
,
int is_sb_boundary
#endif
#if CONFIG_ORIP
,
const int seq_intra_pred_filter_flag
#endif
#if CONFIG_IBP_DIR || CONFIG_IBP_DC
,
const int seq_ibp_flag
#endif
#if CONFIG_IBP_DIR
,
uint8_t *const ibp_weights[TX_SIZES_ALL][DIR_MODES_0_90]
#endif
) {
int i;
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
DECLARE_ALIGNED(16, uint16_t, left_data[NUM_INTRA_NEIGHBOUR_PIXELS]);
DECLARE_ALIGNED(16, uint16_t, above_data[NUM_INTRA_NEIGHBOUR_PIXELS]);
#if CONFIG_IBP_DIR
DECLARE_ALIGNED(16, uint16_t, second_pred_data[MAX_TX_SQUARE + 32]);
#endif
#if CONFIG_MRLS
uint16_t *const above_row = above_data + 32;
uint16_t *const left_col = left_data + 32;
#else
uint16_t *const above_row = above_data + 16;
uint16_t *const left_col = left_data + 16;
#endif
#if CONFIG_IBP_DIR
uint16_t *const second_pred = second_pred_data + 16;
#endif
const int txwpx = tx_size_wide[tx_size];
const int txhpx = tx_size_high[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;
#if CONFIG_MRLS
const uint8_t mrl_index =
(plane == PLANE_TYPE_Y && is_inter_block(xd->mi[0]
#if CONFIG_SDP
,
xd->tree_type
#endif
) == 0)
? xd->mi[0]->mrl_index
: 0;
const int above_mrl_idx = is_sb_boundary ? 0 : mrl_index;
const uint16_t *above_ref = ref - ref_stride * (above_mrl_idx + 1);
const uint16_t *left_ref = ref - 1 - mrl_index;
#else
const uint16_t *above_ref = ref - ref_stride;
const uint16_t *left_ref = ref - 1;
#endif
int p_angle = 0;
const int is_dr_mode = av1_is_directional_mode(mode);
const int use_filter_intra = filter_intra_mode != FILTER_INTRA_MODES;
int base = 128 << (xd->bd - 8);
// The left_data, above_data buffers must be zeroed to fix some intermittent
// valgrind errors. Uninitialized reads in intra pred modules (e.g. width = 4
// path in av1_highbd_dr_prediction_z2_avx2()) from left_data, above_data are
// seen to be the potential reason for this issue.
aom_memset16(left_data, base + 1, NUM_INTRA_NEIGHBOUR_PIXELS);
aom_memset16(above_data, base - 1, NUM_INTRA_NEIGHBOUR_PIXELS);
// The default values if ref pixels are not available:
// base base-1 base-1 .. base-1 base-1 base-1 base-1 base-1 base-1
// base+1 A B .. Y Z
// base+1 C D .. W X
// base+1 E F .. U V
// base+1 G H .. S T T T T T
#if CONFIG_ORIP
#if CONFIG_MRLS
int apply_sub_block_based_refinement_filter =
seq_intra_pred_filter_flag && (mrl_index == 0);
#else
int apply_sub_block_based_refinement_filter = seq_intra_pred_filter_flag;
#endif
#endif
if (is_dr_mode) {
p_angle = mode_to_angle_map[mode] + angle_delta;
if (p_angle <= 90)
need_above = 1, need_left = 0, need_above_left = 1;
else if (p_angle < 180)
need_above = 1, need_left = 1, need_above_left = 1;
else
need_above = 0, need_left = 1, need_above_left = 1;
#if CONFIG_IBP_DIR
if (seq_ibp_flag) {
need_above = 1, need_left = 1, need_above_left = 1;
}
#endif
#if CONFIG_ORIP && !CONFIG_ORIP_NONDC_DISABLED
if (apply_sub_block_based_refinement_filter &&
(p_angle == 90 || p_angle == 180)) {
need_above = 1;
need_left = 1;
need_above_left = 1;
}
#endif
}
if (use_filter_intra) need_left = need_above = need_above_left = 1;
assert(n_top_px >= 0);
assert(n_topright_px >= 0);
assert(n_left_px >= 0);
assert(n_bottomleft_px >= 0);
if ((!need_above && n_left_px == 0) || (!need_left && n_top_px == 0)) {
int val;
if (need_left) {
val = (n_top_px > 0) ? above_ref[0] : base + 1;
} else {
val = (n_left_px > 0) ? left_ref[0] : base - 1;
}
for (i = 0; i < txhpx; ++i) {
aom_memset16(dst, val, txwpx);
dst += dst_stride;
}
return;
}
// NEED_LEFT
if (need_left) {
int need_bottom = extend_modes[mode] & NEED_BOTTOMLEFT;
if (use_filter_intra) need_bottom = 0;
#if CONFIG_IBP_DIR
if (is_dr_mode)
need_bottom =
seq_ibp_flag ? (p_angle < 90) || (p_angle > 180) : p_angle > 180;
#else
if (is_dr_mode) need_bottom = p_angle > 180;
#endif
#if CONFIG_MRLS
const int num_left_pixels_needed =
txhpx + (need_bottom ? txwpx : 3) + (mrl_index << 1);
#else
const int num_left_pixels_needed = txhpx + (need_bottom ? txwpx : 0);
#endif
i = 0;
if (n_left_px > 0) {
for (; i < n_left_px; i++) left_col[i] = left_ref[i * ref_stride];
if (need_bottom && n_bottomleft_px > 0) {
assert(i == txhpx);
for (; i < txhpx + n_bottomleft_px; i++)
left_col[i] = left_ref[i * ref_stride];
}
if (i < num_left_pixels_needed)
aom_memset16(&left_col[i], left_col[i - 1], num_left_pixels_needed - i);
} else if (n_top_px > 0) {
aom_memset16(left_col, above_ref[0], num_left_pixels_needed);
}
}
// NEED_ABOVE
if (need_above) {
int need_right = extend_modes[mode] & NEED_ABOVERIGHT;
if (use_filter_intra) need_right = 0;
#if CONFIG_IBP_DIR
if (is_dr_mode)
need_right =
seq_ibp_flag ? (p_angle < 90) || (p_angle > 180) : p_angle < 90;
#else
if (is_dr_mode) need_right = p_angle < 90;
#endif
#if CONFIG_MRLS
const int num_top_pixels_needed =
txwpx + (need_right ? txhpx : 0) + (mrl_index << 1);
#else
const int num_top_pixels_needed = txwpx + (need_right ? txhpx : 0);
#endif
if (n_top_px > 0) {
memcpy(above_row, above_ref, n_top_px * sizeof(above_ref[0]));
i = n_top_px;
if (need_right && n_topright_px > 0) {
assert(n_top_px == txwpx);
memcpy(above_row + txwpx, above_ref + txwpx,
n_topright_px * sizeof(above_ref[0]));
i += n_topright_px;
}
if (i < num_top_pixels_needed)
aom_memset16(&above_row[i], above_row[i - 1],
num_top_pixels_needed - i);
} else if (n_left_px > 0) {
aom_memset16(above_row, left_ref[0], num_top_pixels_needed);
}
}
if (need_above_left) {
#if CONFIG_MRLS
for (i = 1; i <= mrl_index + 1; i++) {
if (n_top_px > 0 && n_left_px > 0) {
above_row[-i] = above_ref[-i];
if (is_sb_boundary)
left_col[-i] = left_ref[-ref_stride];
else
left_col[-i] = left_ref[-i * ref_stride];
} else if (n_top_px > 0) {
above_row[-i] = left_col[-i] = above_ref[0];
} else if (n_left_px > 0) {
above_row[-i] = left_col[-i] = left_ref[0];
} else {
above_row[-i] = left_col[-i] = base;
}
}
#else
if (n_top_px > 0 && n_left_px > 0) {
above_row[-1] = above_ref[-1];
} else if (n_top_px > 0) {
above_row[-1] = above_ref[0];
} else if (n_left_px > 0) {
above_row[-1] = left_ref[0];
} else {
above_row[-1] = base;
}
left_col[-1] = above_row[-1];
#endif
}
if (use_filter_intra) {
highbd_filter_intra_predictor(dst, dst_stride, tx_size, above_row, left_col,
filter_intra_mode, xd->bd);
return;
}
if (is_dr_mode) {
int upsample_above = 0;
int upsample_left = 0;
#if CONFIG_MRLS
if (!disable_edge_filter && mrl_index == 0) {
#else
if (!disable_edge_filter) {
#endif
#if CONFIG_IBP_DIR
int need_right = p_angle < 90;
int need_bottom = p_angle > 180;
int filt_type_above = get_filt_type(xd, plane);
int filt_type_left = filt_type_above;
int angle_above = p_angle - 90;
int angle_left = p_angle - 180;
if (seq_ibp_flag) {
need_right |= p_angle > 180;
need_bottom |= p_angle < 90;
const MB_MODE_INFO *ab =
(plane == 0) ? xd->above_mbmi : xd->chroma_above_mbmi;
const MB_MODE_INFO *le =
(plane == 0) ? xd->left_mbmi : xd->chroma_left_mbmi;
filt_type_above = ab ? is_smooth(ab, plane) : 0;
filt_type_left = le ? is_smooth(le, plane) : 0;
angle_above = p_angle > 180 ? (p_angle - 180 - 90) : angle_above;
angle_left = p_angle < 90 ? p_angle : angle_left;
}
#else
const int need_right = p_angle < 90;
const int need_bottom = p_angle > 180;
const int filt_type = get_filt_type(xd, plane);
#endif
if (p_angle != 90 && p_angle != 180) {
const int ab_le = need_above_left ? 1 : 0;
if (need_above && need_left && (txwpx + txhpx >= 24)) {
filter_intra_edge_corner_high(above_row, left_col);
}
if (need_above && n_top_px > 0) {
#if CONFIG_IBP_DIR
const int strength = intra_edge_filter_strength(
txwpx, txhpx, angle_above, filt_type_above);
#else
const int strength =
intra_edge_filter_strength(txwpx, txhpx, p_angle - 90, filt_type);
#endif
const int n_px = n_top_px + ab_le + (need_right ? txhpx : 0);
av1_filter_intra_edge_high(above_row - ab_le, n_px, strength);
}
if (need_left && n_left_px > 0) {
#if CONFIG_IBP_DIR
const int strength = intra_edge_filter_strength(
txhpx, txwpx, angle_left, filt_type_left);
#else
const int strength = intra_edge_filter_strength(
txhpx, txwpx, p_angle - 180, filt_type);
#endif
const int n_px = n_left_px + ab_le + (need_bottom ? txwpx : 0);
av1_filter_intra_edge_high(left_col - ab_le, n_px, strength);
}
}
#if CONFIG_IBP_DIR
upsample_above = av1_use_intra_edge_upsample(txwpx, txhpx, angle_above,
filt_type_above);
#else
upsample_above =
av1_use_intra_edge_upsample(txwpx, txhpx, p_angle - 90, filt_type);
#endif
if (need_above && upsample_above) {
const int n_px = txwpx + (need_right ? txhpx : 0);
av1_upsample_intra_edge_high(above_row, n_px, xd->bd);
}
#if CONFIG_IBP_DIR
upsample_left =
av1_use_intra_edge_upsample(txhpx, txwpx, angle_left, filt_type_left);
#else
upsample_left =
av1_use_intra_edge_upsample(txhpx, txwpx, p_angle - 180, filt_type);
#endif
if (need_left && upsample_left) {
const int n_px = txhpx + (need_bottom ? txwpx : 0);
av1_upsample_intra_edge_high(left_col, n_px, xd->bd);
}
}
highbd_dr_predictor(dst, dst_stride, tx_size, above_row, left_col,
upsample_above, upsample_left, p_angle, xd->bd
#if CONFIG_MRLS
,
mrl_index
#endif
);
#if CONFIG_IBP_DIR
if (seq_ibp_flag) {
#if CONFIG_MRLS
if (mrl_index == 0) {
#endif
if (p_angle > 0 && p_angle < 90) {
int mode_index = angle_to_mode_index[p_angle];
uint8_t *weights = ibp_weights[tx_size][mode_index];
highbd_second_dr_predictor(second_pred, txwpx, tx_size, above_row,
left_col, upsample_above, upsample_left,
p_angle, xd->bd);
av1_highbd_ibp_dr_prediction_z1_c(weights, dst, dst_stride,
second_pred, txwpx, txwpx, txhpx);
}
if (p_angle > 180 && p_angle < 270) {
int mode_index = angle_to_mode_index[270 - p_angle];
int transpose_tsize = transpose_tx_size[tx_size];
uint8_t *weights = ibp_weights[transpose_tsize][mode_index];
highbd_second_dr_predictor(second_pred, txwpx, tx_size, above_row,
left_col, upsample_above, upsample_left,
p_angle, xd->bd);
av1_highbd_ibp_dr_prediction_z3_c(weights, dst, dst_stride,
second_pred, txwpx, txwpx, txhpx);
}
#if CONFIG_MRLS
}
#endif
}
#endif
#if CONFIG_ORIP
#if !CONFIG_ORIP_NONDC_DISABLED
// Apply sub-block based filter for horizontal/vertical intra mode
if (apply_sub_block_based_refinement_filter &&
av1_allow_orip_dir(p_angle)) {
av1_apply_orip_4x4subblock_hbd(dst, dst_stride, tx_size, above_row,
left_col, mode, xd->bd);
}
#endif
#endif
return;
}
// predict
if (mode == DC_PRED) {
dc_pred_high[n_left_px > 0][n_top_px > 0][tx_size](
dst, dst_stride, above_row, left_col, xd->bd);
#if CONFIG_IBP_DC
if (seq_ibp_flag && ((plane == 0) || (xd->mi[0]->uv_mode != UV_CFL_PRED)) &&
((n_left_px > 0) || (n_top_px > 0))) {
ibp_dc_pred_high[n_left_px > 0][n_top_px > 0][tx_size](
dst, dst_stride, above_row, left_col, xd->bd);
}
#endif
} else {
pred_high[mode][tx_size](dst, dst_stride, above_row, left_col, xd->bd);
}
#if CONFIG_ORIP
// Apply sub-block based filter for DC/smooth intra mode
apply_sub_block_based_refinement_filter &=
av1_allow_orip_smooth_dc(mode, plane);
if (apply_sub_block_based_refinement_filter) {
av1_apply_orip_4x4subblock_hbd(dst, dst_stride, tx_size, above_row,
left_col, mode, xd->bd);
}
#endif
}
static void build_intra_predictors(
const MACROBLOCKD *xd, const uint8_t *ref, int ref_stride, uint8_t *dst,
int dst_stride, PREDICTION_MODE mode, int angle_delta,
FILTER_INTRA_MODE filter_intra_mode, TX_SIZE tx_size,
int disable_edge_filter, int n_top_px, int n_topright_px, int n_left_px,
int n_bottomleft_px, int plane
#if CONFIG_MRLS
,
int is_sb_boundary
#endif
#if CONFIG_ORIP
,
const int seq_intra_pred_filter_flag
#endif
#if CONFIG_IBP_DIR || CONFIG_IBP_DC
,
const int seq_ibp_flag
#endif
#if CONFIG_IBP_DIR
,
uint8_t *const ibp_weights[TX_SIZES_ALL][DIR_MODES_0_90]
#endif
) {
int i;
#if CONFIG_MRLS
const uint8_t mrl_index =
(plane == PLANE_TYPE_Y && is_inter_block(xd->mi[0]
#if CONFIG_SDP
,
xd->tree_type
#endif
) == 0)
? xd->mi[0]->mrl_index
: 0;
const int above_mrl_idx = is_sb_boundary ? 0 : mrl_index;
const uint8_t *above_ref = ref - ref_stride * (above_mrl_idx + 1);
const uint8_t *left_ref = ref - 1 - mrl_index;
#else
const uint8_t *above_ref = ref - ref_stride;
const uint8_t *left_ref = ref - 1;
#endif
DECLARE_ALIGNED(16, uint8_t, left_data[NUM_INTRA_NEIGHBOUR_PIXELS]);
DECLARE_ALIGNED(16, uint8_t, above_data[NUM_INTRA_NEIGHBOUR_PIXELS]);
#if CONFIG_IBP_DIR
DECLARE_ALIGNED(16, uint8_t, second_pred_data[MAX_TX_SQUARE + 32]);
#endif
#if CONFIG_MRLS
uint8_t *const above_row = above_data + 32;
uint8_t *const left_col = left_data + 32;
#else
uint8_t *const above_row = above_data + 16;
uint8_t *const left_col = left_data + 16;
#endif
#if CONFIG_IBP_DIR
uint8_t *const second_pred = second_pred_data + 16;
#endif
const int txwpx = tx_size_wide[tx_size];
const int txhpx = tx_size_high[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;
int p_angle = 0;
const int is_dr_mode = av1_is_directional_mode(mode);
const int use_filter_intra = filter_intra_mode != FILTER_INTRA_MODES;
// The left_data, above_data buffers must be zeroed to fix some intermittent
// valgrind errors. Uninitialized reads in intra pred modules (e.g. width = 4
// path in av1_dr_prediction_z1_avx2()) from left_data, above_data are seen to
// be the potential reason for this issue.
memset(left_data, 129, NUM_INTRA_NEIGHBOUR_PIXELS);
memset(above_data, 127, NUM_INTRA_NEIGHBOUR_PIXELS);
// The default values if ref pixels are not available:
// 128 127 127 .. 127 127 127 127 127 127
// 129 A B .. Y Z
// 129 C D .. W X
// 129 E F .. U V
// 129 G H .. S T T T T T
// ..
#if CONFIG_ORIP
#if CONFIG_MRLS
int apply_sub_block_based_refinement_filter =
seq_intra_pred_filter_flag && (mrl_index == 0);
#else
int apply_sub_block_based_refinement_filter = seq_intra_pred_filter_flag;
#endif
#endif
if (is_dr_mode) {
p_angle = mode_to_angle_map[mode] + angle_delta;
if (p_angle <= 90)
need_above = 1, need_left = 0, need_above_left = 1;
else if (p_angle < 180)
need_above = 1, need_left = 1, need_above_left = 1;
else
need_above = 0, need_left = 1, need_above_left = 1;
#if CONFIG_IBP_DIR
if (seq_ibp_flag) {
need_above = 1, need_left = 1, need_above_left = 1;
}
#endif
#if CONFIG_ORIP && !CONFIG_ORIP_NONDC_DISABLED
if (apply_sub_block_based_refinement_filter &&
(p_angle == 90 || p_angle == 180)) {
need_above = 1;
need_left = 1;
need_above_left = 1;
}
#endif
}
if (use_filter_intra) need_left = need_above = need_above_left = 1;
assert(n_top_px >= 0);
assert(n_topright_px >= 0);
assert(n_left_px >= 0);
assert(n_bottomleft_px >= 0);
if ((!need_above && n_left_px == 0) || (!need_left && n_top_px == 0)) {
int val;
if (need_left) {
val = (n_top_px > 0) ? above_ref[0] : 129;
} else {
val = (n_left_px > 0) ? left_ref[0] : 127;
}
for (i = 0; i < txhpx; ++i) {
memset(dst, val, txwpx);
dst += dst_stride;
}
return;
}
// NEED_LEFT
if (need_left) {
int need_bottom = extend_modes[mode] & NEED_BOTTOMLEFT;
if (use_filter_intra) need_bottom = 0;
#if CONFIG_IBP_DIR
if (is_dr_mode)
need_bottom =
seq_ibp_flag ? (p_angle < 90) || (p_angle > 180) : p_angle > 180;
#else
if (is_dr_mode) need_bottom = p_angle > 180;
#endif
#if CONFIG_MRLS
const int num_left_pixels_needed =
txhpx + (need_bottom ? txwpx : 3) + (mrl_index << 1);
#else
const int num_left_pixels_needed = txhpx + (need_bottom ? txwpx : 0);
#endif
i = 0;
if (n_left_px > 0) {
for (; i < n_left_px; i++) left_col[i] = left_ref[i * ref_stride];
if (need_bottom && n_bottomleft_px > 0) {
assert(i == txhpx);
for (; i < txhpx + n_bottomleft_px; i++)
left_col[i] = left_ref[i * ref_stride];
}
if (i < num_left_pixels_needed)
memset(&left_col[i], left_col[i - 1], num_left_pixels_needed - i);
} else if (n_top_px > 0) {
memset(left_col, above_ref[0], num_left_pixels_needed);
}
}
// NEED_ABOVE
if (need_above) {
int need_right = extend_modes[mode] & NEED_ABOVERIGHT;
if (use_filter_intra) need_right = 0;
#if CONFIG_IBP_DIR
if (is_dr_mode)
need_right =
seq_ibp_flag ? (p_angle < 90) || (p_angle > 180) : p_angle < 90;
#else
if (is_dr_mode) need_right = p_angle < 90;
#endif
#if CONFIG_MRLS
const int num_top_pixels_needed =
txwpx + (need_right ? txhpx : 0) + (mrl_index << 1);
#else
const int num_top_pixels_needed = txwpx + (need_right ? txhpx : 0);
#endif
if (n_top_px > 0) {
memcpy(above_row, above_ref, n_top_px);
i = n_top_px;
if (need_right && n_topright_px > 0) {
assert(n_top_px == txwpx);
memcpy(above_row + txwpx, above_ref + txwpx, n_topright_px);
i += n_topright_px;
}
if (i < num_top_pixels_needed)
memset(&above_row[i], above_row[i - 1], num_top_pixels_needed - i);
} else if (n_left_px > 0) {
memset(above_row, left_ref[0], num_top_pixels_needed);
}
}
if (need_above_left) {
#if CONFIG_MRLS
for (i = 1; i <= mrl_index + 1; i++) {
if (n_top_px > 0 && n_left_px > 0) {
above_row[-i] = above_ref[-i];
if (is_sb_boundary)
left_col[-i] = left_ref[-ref_stride];
else
left_col[-i] = left_ref[-i * ref_stride];
} else if (n_top_px > 0) {
above_row[-i] = left_col[-i] = above_ref[0];
} else if (n_left_px > 0) {
above_row[-i] = left_col[-i] = left_ref[0];
} else {
above_row[-i] = left_col[-i] = 128;
}
}
#else
if (n_top_px > 0 && n_left_px > 0) {
above_row[-1] = above_ref[-1];
} else if (n_top_px > 0) {
above_row[-1] = above_ref[0];
} else if (n_left_px > 0) {
above_row[-1] = left_ref[0];
} else {
above_row[-1] = 128;
}
left_col[-1] = above_row[-1];
#endif
}
if (use_filter_intra) {
av1_filter_intra_predictor(dst, dst_stride, tx_size, above_row, left_col,
filter_intra_mode);
return;
}
if (is_dr_mode) {
int upsample_above = 0;
int upsample_left = 0;
#if CONFIG_MRLS
if (!disable_edge_filter && mrl_index == 0) {
#else
if (!disable_edge_filter) {
#endif
#if CONFIG_IBP_DIR
int need_right = p_angle < 90;
int need_bottom = p_angle > 180;
int filt_type_above = get_filt_type(xd, plane);
int filt_type_left = filt_type_above;
int angle_above = p_angle - 90;
int angle_left = p_angle - 180;
if (seq_ibp_flag) {
need_right |= p_angle > 180;
need_bottom |= p_angle < 90;
const MB_MODE_INFO *ab =
(plane == 0) ? xd->above_mbmi : xd->chroma_above_mbmi;
const MB_MODE_INFO *le =
(plane == 0) ? xd->left_mbmi : xd->chroma_left_mbmi;
filt_type_above = ab ? is_smooth(ab, plane) : 0;
filt_type_left = le ? is_smooth(le, plane) : 0;
angle_above = p_angle > 180 ? (p_angle - 180 - 90) : angle_above;
angle_left = p_angle < 90 ? p_angle : angle_left;
}
#else
const int need_right = p_angle < 90;
const int need_bottom = p_angle > 180;
const int filt_type = get_filt_type(xd, plane);
#endif
if (p_angle != 90 && p_angle != 180) {
const int ab_le = need_above_left ? 1 : 0;
if (need_above && need_left && (txwpx + txhpx >= 24)) {
filter_intra_edge_corner(above_row, left_col);
}
if (need_above && n_top_px > 0) {
#if CONFIG_IBP_DIR
const int strength = intra_edge_filter_strength(
txwpx, txhpx, angle_above, filt_type_above);
#else
const int strength =
intra_edge_filter_strength(txwpx, txhpx, p_angle - 90, filt_type);
#endif
const int n_px = n_top_px + ab_le + (need_right ? txhpx : 0);
av1_filter_intra_edge(above_row - ab_le, n_px, strength);
}
if (need_left && n_left_px > 0) {
#if CONFIG_IBP_DIR
const int strength = intra_edge_filter_strength(
txhpx, txwpx, angle_left, filt_type_left);
#else
const int strength = intra_edge_filter_strength(
txhpx, txwpx, p_angle - 180, filt_type);
#endif
const int n_px = n_left_px + ab_le + (need_bottom ? txwpx : 0);
av1_filter_intra_edge(left_col - ab_le, n_px, strength);
}
}
#if CONFIG_IBP_DIR
upsample_above = av1_use_intra_edge_upsample(txwpx, txhpx, angle_above,
filt_type_above);
#else
upsample_above =
av1_use_intra_edge_upsample(txwpx, txhpx, p_angle - 90, filt_type);
#endif
if (need_above && upsample_above) {
const int n_px = txwpx + (need_right ? txhpx : 0);
av1_upsample_intra_edge(above_row, n_px);
}
#if CONFIG_IBP_DIR
upsample_left =
av1_use_intra_edge_upsample(txhpx, txwpx, angle_left, filt_type_left);
#else
upsample_left =
av1_use_intra_edge_upsample(txhpx, txwpx, p_angle - 180, filt_type);
#endif
if (need_left && upsample_left) {
const int n_px = txhpx + (need_bottom ? txwpx : 0);
av1_upsample_intra_edge(left_col, n_px);
}
}
dr_predictor(dst, dst_stride, tx_size, above_row, left_col, upsample_above,
upsample_left, p_angle
#if CONFIG_MRLS
,
mrl_index
#endif
);
#if CONFIG_IBP_DIR
if (seq_ibp_flag) {
#if CONFIG_MRLS
if (mrl_index == 0) {
#endif
if (p_angle > 0 && p_angle < 90) {
int mode_index = angle_to_mode_index[p_angle];
uint8_t *weights = ibp_weights[tx_size][mode_index];
second_dr_predictor(second_pred, txwpx, tx_size, above_row, left_col,
upsample_above, upsample_left, p_angle);
av1_ibp_dr_prediction_z1_c(weights, dst, dst_stride, second_pred,
txwpx, txwpx, txhpx);
}
if (p_angle > 180 && p_angle < 270) {
int mode_index = angle_to_mode_index[270 - p_angle];
int transpose_tsize = transpose_tx_size[tx_size];
uint8_t *weights = ibp_weights[transpose_tsize][mode_index];
second_dr_predictor(second_pred, txwpx, tx_size, above_row, left_col,
upsample_above, upsample_left, p_angle);
av1_ibp_dr_prediction_z3_c(weights, dst, dst_stride, second_pred,
txwpx, txwpx, txhpx);
}
#if CONFIG_MRLS
}
#endif
}
#endif
#if CONFIG_ORIP
#if !CONFIG_ORIP_NONDC_DISABLED
// Apply sub-block based filter for horizontal/vertical intra mode
if (apply_sub_block_based_refinement_filter &&
av1_allow_orip_dir(p_angle)) {
av1_apply_orip_4x4subblock(dst, dst_stride, tx_size, above_row, left_col,
mode);
}
#endif
#endif
return;
}
// predict
if (mode == DC_PRED) {
dc_pred[n_left_px > 0][n_top_px > 0][tx_size](dst, dst_stride, above_row,
left_col);
#if CONFIG_IBP_DC
if (seq_ibp_flag && ((plane == 0) || (xd->mi[0]->uv_mode != UV_CFL_PRED)) &&
((n_left_px > 0) || (n_top_px > 0))) {
ibp_dc_pred[n_left_px > 0][n_top_px > 0][tx_size](dst, dst_stride,
above_row, left_col);
}
#endif
} else {
pred[mode][tx_size](dst, dst_stride, above_row, left_col);
}
#if CONFIG_ORIP
apply_sub_block_based_refinement_filter &=
av1_allow_orip_smooth_dc(mode, plane);
if (apply_sub_block_based_refinement_filter) {
av1_apply_orip_4x4subblock(dst, dst_stride, tx_size, above_row, left_col,
mode);
}
#endif
}
static INLINE BLOCK_SIZE scale_chroma_bsize(BLOCK_SIZE bsize, int subsampling_x,
int subsampling_y) {
assert(subsampling_x >= 0 && subsampling_x < 2);
assert(subsampling_y >= 0 && subsampling_y < 2);