ref: 58a0f6dbdd69540653d8e13e06948c53a527259a
parent: f0eabfd43226466ebff9ce4a014d0e5bbacd65cd
author: James Zern <jzern@google.com>
date: Fri Oct 25 13:18:04 EDT 2013
vp9: add TileInfo replaces use of cur_tile_mi_(row|col)_(start|end) by VP9_COMMON, making it less stateful and more reusable for parallel tile decoding Change-Id: I1df09382b4567a0e5f4434825d47c79afe2399be
--- a/vp9/common/vp9_findnearmv.c
+++ b/vp9/common/vp9_findnearmv.c
@@ -35,6 +35,7 @@
}
void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
+ const TileInfo *const tile,
int_mv *dst_nearest,
int_mv *dst_near,
int block_idx, int ref_idx,
@@ -46,7 +47,7 @@
assert(ref_idx == 0 || ref_idx == 1);
assert(MAX_MV_REF_CANDIDATES == 2); // makes code here slightly easier
- vp9_find_mv_refs_idx(cm, xd, mi, xd->last_mi,
+ vp9_find_mv_refs_idx(cm, xd, tile, mi, xd->last_mi,
mi->mbmi.ref_frame[ref_idx],
mv_list, block_idx, mi_row, mi_col);
--- a/vp9/common/vp9_findnearmv.h
+++ b/vp9/common/vp9_findnearmv.h
@@ -34,8 +34,8 @@
xd->mb_to_bottom_edge + RIGHT_BOTTOM_MARGIN);
}
-void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm,
- MACROBLOCKD *xd,
+void vp9_append_sub8x8_mvs_for_idx(VP9_COMMON *cm, MACROBLOCKD *xd,
+ const TileInfo *const tile,
int_mv *dst_nearest,
int_mv *dst_near,
int block_idx, int ref_idx,
--- a/vp9/common/vp9_mvref_common.c
+++ b/vp9/common/vp9_mvref_common.c
@@ -170,17 +170,19 @@
// Checks that the given mi_row, mi_col and search point
// are inside the borders of the tile.
-static INLINE int is_inside(const VP9_COMMON *cm, int mi_col, int mi_row,
+static INLINE int is_inside(const TileInfo *const tile,
+ int mi_col, int mi_row, int mi_rows,
const MV *mv) {return !(mi_row + mv->row < 0 ||
- mi_col + mv->col < cm->cur_tile_mi_col_start ||
- mi_row + mv->row >= cm->mi_rows ||
- mi_col + mv->col >= cm->cur_tile_mi_col_end);
+ mi_col + mv->col < tile->mi_col_start ||
+ mi_row + mv->row >= mi_rows ||
+ mi_col + mv->col >= tile->mi_col_end);
}
// This function searches the neighbourhood of a given MB/SB
// to try and find candidate reference vectors.
void vp9_find_mv_refs_idx(const VP9_COMMON *cm, const MACROBLOCKD *xd,
+ const TileInfo *const tile,
MODE_INFO *mi, const MODE_INFO *prev_mi,
MV_REFERENCE_FRAME ref_frame,
int_mv *mv_ref_list,
@@ -201,7 +203,7 @@
// and we also need to keep a mode count.
for (i = 0; i < 2; ++i) {const MV *const mv_ref = &mv_ref_search[i];
- if (is_inside(cm, mi_col, mi_row, mv_ref)) {+ if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {const MODE_INFO *const candidate_mi = xd->mi_8x8[mv_ref->col + mv_ref->row
* xd->mode_info_stride];
const MB_MODE_INFO *const candidate = &candidate_mi->mbmi;
@@ -228,7 +230,7 @@
// mode counts.
for (; i < MVREF_NEIGHBOURS; ++i) {const MV *const mv_ref = &mv_ref_search[i];
- if (is_inside(cm, mi_col, mi_row, mv_ref)) {+ if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {const MB_MODE_INFO *const candidate = &xd->mi_8x8[mv_ref->col +
mv_ref->row
* xd->mode_info_stride]->mbmi;
@@ -258,7 +260,7 @@
if (different_ref_found) { for (i = 0; i < MVREF_NEIGHBOURS; ++i) {const MV *mv_ref = &mv_ref_search[i];
- if (is_inside(cm, mi_col, mi_row, mv_ref)) {+ if (is_inside(tile, mi_col, mi_row, cm->mi_rows, mv_ref)) {const MB_MODE_INFO *const candidate = &xd->mi_8x8[mv_ref->col +
mv_ref->row
* xd->mode_info_stride]->mbmi;
--- a/vp9/common/vp9_mvref_common.h
+++ b/vp9/common/vp9_mvref_common.h
@@ -15,6 +15,7 @@
#define VP9_COMMON_VP9_MVREF_COMMON_H_
void vp9_find_mv_refs_idx(const VP9_COMMON *cm, const MACROBLOCKD *xd,
+ const TileInfo *const tile,
MODE_INFO *mi, const MODE_INFO *prev_mi,
MV_REFERENCE_FRAME ref_frame,
int_mv *mv_ref_list,
@@ -22,11 +23,12 @@
int mi_row, int mi_col);
static INLINE void vp9_find_mv_refs(const VP9_COMMON *cm, const MACROBLOCKD *xd,
+ const TileInfo *const tile,
MODE_INFO *mi, const MODE_INFO *prev_mi,
MV_REFERENCE_FRAME ref_frame,
int_mv *mv_ref_list,
int mi_row, int mi_col) {- vp9_find_mv_refs_idx(cm, xd, mi, prev_mi, ref_frame,
+ vp9_find_mv_refs_idx(cm, xd, tile, mi, prev_mi, ref_frame,
mv_ref_list, -1, mi_row, mi_col);
}
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -19,6 +19,7 @@
#include "vp9/common/vp9_entropy.h"
#include "vp9/common/vp9_entropymode.h"
#include "vp9/common/vp9_quant_common.h"
+#include "vp9/common/vp9_tile_common.h"
#if CONFIG_VP9_POSTPROC
#include "vp9/common/vp9_postproc.h"
@@ -207,8 +208,6 @@
int frame_parallel_decoding_mode;
int log2_tile_cols, log2_tile_rows;
- int cur_tile_mi_col_start, cur_tile_mi_col_end;
- int cur_tile_mi_row_start, cur_tile_mi_row_end;
} VP9_COMMON;
// ref == 0 => LAST_FRAME
@@ -279,17 +278,18 @@
return -1;
}
-static void set_mi_row_col(VP9_COMMON *cm, MACROBLOCKD *xd,
- int mi_row, int bh,
- int mi_col, int bw) {+static void set_mi_row_col(MACROBLOCKD *xd, const TileInfo *const tile,
+ int mi_row, int bh,
+ int mi_col, int bw,
+ int mi_rows, int mi_cols) {xd->mb_to_top_edge = -((mi_row * MI_SIZE) * 8);
- xd->mb_to_bottom_edge = ((cm->mi_rows - bh - mi_row) * MI_SIZE) * 8;
+ xd->mb_to_bottom_edge = ((mi_rows - bh - mi_row) * MI_SIZE) * 8;
xd->mb_to_left_edge = -((mi_col * MI_SIZE) * 8);
- xd->mb_to_right_edge = ((cm->mi_cols - bw - mi_col) * MI_SIZE) * 8;
+ xd->mb_to_right_edge = ((mi_cols - bw - mi_col) * MI_SIZE) * 8;
// Are edges available for intra prediction?
xd->up_available = (mi_row != 0);
- xd->left_available = (mi_col > cm->cur_tile_mi_col_start);
+ xd->left_available = (mi_col > tile->mi_col_start);
}
static void set_prev_mi(VP9_COMMON *cm) {--- a/vp9/common/vp9_tile_common.c
+++ b/vp9/common/vp9_tile_common.c
@@ -10,6 +10,8 @@
#include "vp9/common/vp9_tile_common.h"
+#include "vp9/common/vp9_onyxc_int.h"
+
#define MIN_TILE_WIDTH_B64 4
#define MAX_TILE_WIDTH_B64 64
@@ -17,8 +19,8 @@
return mi_cols_aligned_to_sb(n_mis) >> MI_BLOCK_SIZE_LOG2;
}
-static void vp9_get_tile_offsets(int *min_tile_off, int *max_tile_off,
- int tile_idx, int log2_n_tiles, int n_mis) {+static void get_tile_offsets(int *min_tile_off, int *max_tile_off,
+ int tile_idx, int log2_n_tiles, int n_mis) {const int n_sbs = to_sbs(n_mis);
const int sb_off1 = (tile_idx * n_sbs) >> log2_n_tiles;
const int sb_off2 = ((tile_idx + 1) * n_sbs) >> log2_n_tiles;
@@ -27,16 +29,13 @@
*max_tile_off = MIN(sb_off2 << 3, n_mis);
}
-void vp9_get_tile_col_offsets(VP9_COMMON *cm, int tile_col_idx) {- vp9_get_tile_offsets(&cm->cur_tile_mi_col_start, &cm->cur_tile_mi_col_end,
- tile_col_idx, cm->log2_tile_cols, cm->mi_cols);
+void vp9_tile_init(TileInfo *tile, const VP9_COMMON *cm,
+ int row_idx, int col_idx) {+ get_tile_offsets(&tile->mi_row_start, &tile->mi_row_end,
+ row_idx, cm->log2_tile_rows, cm->mi_rows);
+ get_tile_offsets(&tile->mi_col_start, &tile->mi_col_end,
+ col_idx, cm->log2_tile_cols, cm->mi_cols);
}
-
-void vp9_get_tile_row_offsets(VP9_COMMON *cm, int tile_row_idx) {- vp9_get_tile_offsets(&cm->cur_tile_mi_row_start, &cm->cur_tile_mi_row_end,
- tile_row_idx, cm->log2_tile_rows, cm->mi_rows);
-}
-
void vp9_get_tile_n_bits(int mi_cols,
int *min_log2_tile_cols, int *max_log2_tile_cols) {--- a/vp9/common/vp9_tile_common.h
+++ b/vp9/common/vp9_tile_common.h
@@ -11,11 +11,17 @@
#ifndef VP9_COMMON_VP9_TILE_COMMON_H_
#define VP9_COMMON_VP9_TILE_COMMON_H_
-#include "vp9/common/vp9_onyxc_int.h"
+struct VP9Common;
-void vp9_get_tile_col_offsets(VP9_COMMON *cm, int tile_col_idx);
+typedef struct TileInfo {+ int mi_row_start, mi_row_end;
+ int mi_col_start, mi_col_end;
+} TileInfo;
-void vp9_get_tile_row_offsets(VP9_COMMON *cm, int tile_row_idx);
+// initializes 'tile->mi_(row|col)_(start|end)' for (row_idx, col_idx) based on
+// 'cm->log2_tile_(rows|cols)' & 'cm->mi_(rows|cols)'
+void vp9_tile_init(TileInfo *tile, const struct VP9Common *cm,
+ int row_idx, int col_idx);
void vp9_get_tile_n_bits(int mi_cols,
int *min_log2_tile_cols, int *max_log2_tile_cols);
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -414,6 +414,7 @@
static void read_inter_block_mode_info(VP9_COMMON *const cm,
MACROBLOCKD *const xd,
+ const TileInfo *const tile,
MODE_INFO *const mi,
int mi_row, int mi_col, vp9_reader *r) {MB_MODE_INFO *const mbmi = &mi->mbmi;
@@ -430,7 +431,7 @@
ref0 = mbmi->ref_frame[0];
is_compound = has_second_ref(mbmi);
- vp9_find_mv_refs(cm, xd, mi, xd->last_mi, ref0, mbmi->ref_mvs[ref0],
+ vp9_find_mv_refs(cm, xd, tile, mi, xd->last_mi, ref0, mbmi->ref_mvs[ref0],
mi_row, mi_col);
inter_mode_ctx = mbmi->mode_context[ref0];
@@ -456,7 +457,7 @@
if (is_compound) {const MV_REFERENCE_FRAME ref1 = mbmi->ref_frame[1];
- vp9_find_mv_refs(cm, xd, mi, xd->last_mi,
+ vp9_find_mv_refs(cm, xd, tile, mi, xd->last_mi,
ref1, mbmi->ref_mvs[ref1], mi_row, mi_col);
if (bsize < BLOCK_8X8 || mbmi->mode != ZEROMV) {@@ -482,12 +483,12 @@
b_mode = read_inter_mode(cm, r, inter_mode_ctx);
if (b_mode == NEARESTMV || b_mode == NEARMV) {- vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest[0],
+ vp9_append_sub8x8_mvs_for_idx(cm, xd, tile, &nearest[0],
&nearmv[0], j, 0,
mi_row, mi_col);
if (is_compound)
- vp9_append_sub8x8_mvs_for_idx(cm, xd, &nearest[1],
+ vp9_append_sub8x8_mvs_for_idx(cm, xd, tile, &nearest[1],
&nearmv[1], j, 1,
mi_row, mi_col);
}
@@ -523,6 +524,7 @@
static void read_inter_frame_mode_info(VP9_COMMON *const cm,
MACROBLOCKD *const xd,
+ const TileInfo *const tile,
MODE_INFO *const mi,
int mi_row, int mi_col, vp9_reader *r) {MB_MODE_INFO *const mbmi = &mi->mbmi;
@@ -537,12 +539,13 @@
!mbmi->skip_coeff || !inter_block, r);
if (inter_block)
- read_inter_block_mode_info(cm, xd, mi, mi_row, mi_col, r);
+ read_inter_block_mode_info(cm, xd, tile, mi, mi_row, mi_col, r);
else
read_intra_block_mode_info(cm, mi, r);
}
void vp9_read_mode_info(VP9_COMMON *cm, MACROBLOCKD *xd,
+ const TileInfo *const tile,
int mi_row, int mi_col, vp9_reader *r) {MODE_INFO *const mi = xd->mi_8x8[0];
const BLOCK_SIZE bsize = mi->mbmi.sb_type;
@@ -555,7 +558,7 @@
if (frame_is_intra_only(cm))
read_intra_frame_mode_info(cm, xd, mi, mi_row, mi_col, r);
else
- read_inter_frame_mode_info(cm, xd, mi, mi_row, mi_col, r);
+ read_inter_frame_mode_info(cm, xd, tile, mi, mi_row, mi_col, r);
for (y = 0, z = 0; y < y_mis; y++, z += cm->mode_info_stride) { for (x = !y; x < x_mis; x++) {--- a/vp9/decoder/vp9_decodemv.h
+++ b/vp9/decoder/vp9_decodemv.h
@@ -14,7 +14,10 @@
#include "vp9/decoder/vp9_onyxd_int.h"
#include "vp9/decoder/vp9_dboolhuff.h"
+struct TileInfo;
+
void vp9_read_mode_info(VP9_COMMON *cm, MACROBLOCKD *xd,
+ const struct TileInfo *const tile,
int mi_row, int mi_col, vp9_reader *r);
#endif // VP9_DECODER_VP9_DECODEMV_H_
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -211,9 +211,11 @@
vpx_realloc(pbi->mi_streams, tile_cols *
sizeof(*pbi->mi_streams)));
for (tile_col = 0; tile_col < tile_cols; ++tile_col) {- vp9_get_tile_col_offsets(cm, tile_col);
+ TileInfo tile;
+
+ vp9_tile_init(&tile, cm, 0, tile_col);
pbi->mi_streams[tile_col] =
- &cm->mi[cm->mi_rows * cm->cur_tile_mi_col_start];
+ &cm->mi[cm->mi_rows * tile.mi_col_start];
}
// 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
@@ -327,6 +329,7 @@
}
static void set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+ const TileInfo *const tile,
BLOCK_SIZE bsize, int mi_row, int mi_col) {const int bh = num_8x8_blocks_high_lookup[bsize];
const int bw = num_8x8_blocks_wide_lookup[bsize];
@@ -350,7 +353,7 @@
// Distance of Mb to the various image edges. These are specified to 8th pel
// as they are always compared to values that are in 1/8th pel units
- set_mi_row_col(cm, xd, mi_row, bh, mi_col, bw);
+ set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col);
}
@@ -371,6 +374,7 @@
}
static void decode_modes_b(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+ const TileInfo *const tile,
int mi_row, int mi_col,
vp9_reader *r, BLOCK_SIZE bsize, int index) {const int less8x8 = bsize < BLOCK_8X8;
@@ -381,8 +385,8 @@
if (index > 0)
return;
- set_offsets(cm, xd, bsize, mi_row, mi_col);
- vp9_read_mode_info(cm, xd, mi_row, mi_col, r);
+ set_offsets(cm, xd, tile, bsize, mi_row, mi_col);
+ vp9_read_mode_info(cm, xd, tile, mi_row, mi_col, r);
if (less8x8)
bsize = BLOCK_8X8;
@@ -419,6 +423,7 @@
}
static void decode_modes_sb(VP9_COMMON *const cm, MACROBLOCKD *const xd,
+ const TileInfo *const tile,
int mi_row, int mi_col,
vp9_reader* r, BLOCK_SIZE bsize, int index) {const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
@@ -455,23 +460,23 @@
switch (partition) {case PARTITION_NONE:
- decode_modes_b(cm, xd, mi_row, mi_col, r, subsize, 0);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, 0);
break;
case PARTITION_HORZ:
- decode_modes_b(cm, xd, mi_row, mi_col, r, subsize, 0);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, 0);
if (mi_row + hbs < cm->mi_rows)
- decode_modes_b(cm, xd, mi_row + hbs, mi_col, r, subsize, 1);
+ decode_modes_b(cm, xd, tile, mi_row + hbs, mi_col, r, subsize, 1);
break;
case PARTITION_VERT:
- decode_modes_b(cm, xd, mi_row, mi_col, r, subsize, 0);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col, r, subsize, 0);
if (mi_col + hbs < cm->mi_cols)
- decode_modes_b(cm, xd, mi_row, mi_col + hbs, r, subsize, 1);
+ decode_modes_b(cm, xd, tile, mi_row, mi_col + hbs, r, subsize, 1);
break;
case PARTITION_SPLIT: {int n;
for (n = 0; n < 4; n++) {const int j = n >> 1, i = n & 1;
- decode_modes_sb(cm, xd, mi_row + j * hbs, mi_col + i * hbs,
+ decode_modes_sb(cm, xd, tile, mi_row + j * hbs, mi_col + i * hbs,
r, subsize, n);
}
} break;
@@ -737,7 +742,8 @@
xd->above_seg_context = pbi->above_seg_context;
}
-static void decode_tile(VP9D_COMP *pbi, vp9_reader *r) {+static void decode_tile(VP9D_COMP *pbi, const TileInfo *const tile,
+ vp9_reader *r) {const int num_threads = pbi->oxcf.max_threads;
VP9_COMMON *const cm = &pbi->common;
int mi_row, mi_col;
@@ -753,14 +759,14 @@
vp9_loop_filter_frame_init(cm, cm->lf.filter_level);
}
- for (mi_row = cm->cur_tile_mi_row_start; mi_row < cm->cur_tile_mi_row_end;
+ for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
mi_row += MI_BLOCK_SIZE) {// For a SB there are 2 left contexts, each pertaining to a MB row within
vp9_zero(xd->left_context);
vp9_zero(xd->left_seg_context);
- for (mi_col = cm->cur_tile_mi_col_start; mi_col < cm->cur_tile_mi_col_end;
+ for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE)
- decode_modes_sb(cm, xd, mi_row, mi_col, r, BLOCK_64X64, 0);
+ decode_modes_sb(cm, xd, tile, mi_row, mi_col, r, BLOCK_64X64, 0);
if (pbi->do_loopfilter_inline) {const int lf_start = mi_row - MI_BLOCK_SIZE;
@@ -770,7 +776,7 @@
if (lf_start < 0) continue;
// decoding has completed: finish up the loop filter in this thread.
- if (mi_row + MI_BLOCK_SIZE >= cm->cur_tile_mi_row_end) continue;
+ if (mi_row + MI_BLOCK_SIZE >= tile->mi_row_end) continue;
vp9_worker_sync(&pbi->lf_worker);
lf_data->start = lf_start;
@@ -852,14 +858,15 @@
}
for (tile_row = 0; tile_row < tile_rows; tile_row++) {- vp9_get_tile_row_offsets(cm, tile_row);
for (tile_col = tile_cols - 1; tile_col >= 0; tile_col--) {- vp9_get_tile_col_offsets(cm, tile_col);
+ TileInfo tile;
+
+ vp9_tile_init(&tile, cm, tile_row, tile_col);
setup_token_decoder(data_ptr2[tile_row][tile_col], data_end,
data_end - data_ptr2[tile_row][tile_col],
&cm->error, &residual_bc);
setup_tile_context(pbi, xd, tile_col);
- decode_tile(pbi, &residual_bc);
+ decode_tile(pbi, &tile, &residual_bc);
if (tile_row == tile_rows - 1 && tile_col == tile_cols - 1)
bc_bak = residual_bc;
}
@@ -869,11 +876,11 @@
int has_more;
for (tile_row = 0; tile_row < tile_rows; tile_row++) {- vp9_get_tile_row_offsets(cm, tile_row);
for (tile_col = 0; tile_col < tile_cols; tile_col++) {+ TileInfo tile;
size_t size;
- vp9_get_tile_col_offsets(cm, tile_col);
+ vp9_tile_init(&tile, cm, tile_row, tile_col);
has_more = tile_col < tile_cols - 1 || tile_row < tile_rows - 1;
if (has_more) {@@ -889,7 +896,7 @@
setup_token_decoder(data, data_end, size, &cm->error, &residual_bc);
setup_tile_context(pbi, xd, tile_col);
- decode_tile(pbi, &residual_bc);
+ decode_tile(pbi, &tile, &residual_bc);
data += size;
}
}
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -561,7 +561,8 @@
write_intra_mode(bc, m->mbmi.uv_mode, vp9_kf_uv_mode_prob[ym]);
}
-static void write_modes_b(VP9_COMP *cpi, MODE_INFO **mi_8x8, vp9_writer *bc,
+static void write_modes_b(VP9_COMP *cpi, const TileInfo *const tile,
+ MODE_INFO **mi_8x8, vp9_writer *bc,
TOKENEXTRA **tok, TOKENEXTRA *tok_end,
int mi_row, int mi_col, int index) {VP9_COMMON *const cm = &cpi->common;
@@ -574,9 +575,10 @@
xd->mi_8x8 = mi_8x8;
- set_mi_row_col(&cpi->common, xd,
+ set_mi_row_col(xd, tile,
mi_row, num_8x8_blocks_high_lookup[m->mbmi.sb_type],
- mi_col, num_8x8_blocks_wide_lookup[m->mbmi.sb_type]);
+ mi_col, num_8x8_blocks_wide_lookup[m->mbmi.sb_type],
+ cm->mi_rows, cm->mi_cols);
if (frame_is_intra_only(cm)) {write_mb_modes_kf(cpi, mi_8x8, bc);
#ifdef ENTROPY_STATS
@@ -593,7 +595,8 @@
pack_mb_tokens(bc, tok, tok_end);
}
-static void write_modes_sb(VP9_COMP *cpi, MODE_INFO **mi_8x8, vp9_writer *bc,
+static void write_modes_sb(VP9_COMP *cpi, const TileInfo *const tile,
+ MODE_INFO **mi_8x8, vp9_writer *bc,
TOKENEXTRA **tok, TOKENEXTRA *tok_end,
int mi_row, int mi_col, BLOCK_SIZE bsize,
int index) {@@ -634,24 +637,25 @@
switch (partition) {case PARTITION_NONE:
- write_modes_b(cpi, mi_8x8, bc, tok, tok_end, mi_row, mi_col, 0);
+ write_modes_b(cpi, tile, mi_8x8, bc, tok, tok_end, mi_row, mi_col, 0);
break;
case PARTITION_HORZ:
- write_modes_b(cpi, mi_8x8, bc, tok, tok_end, mi_row, mi_col, 0);
+ write_modes_b(cpi, tile, mi_8x8, bc, tok, tok_end, mi_row, mi_col, 0);
if ((mi_row + bs) < cm->mi_rows)
- write_modes_b(cpi, mi_8x8 + bs * mis, bc, tok, tok_end, mi_row + bs,
- mi_col, 1);
+ write_modes_b(cpi, tile, mi_8x8 + bs * mis, bc, tok, tok_end,
+ mi_row + bs, mi_col, 1);
break;
case PARTITION_VERT:
- write_modes_b(cpi, mi_8x8, bc, tok, tok_end, mi_row, mi_col, 0);
+ write_modes_b(cpi, tile, mi_8x8, bc, tok, tok_end, mi_row, mi_col, 0);
if ((mi_col + bs) < cm->mi_cols)
- write_modes_b(cpi, mi_8x8 + bs, bc, tok, tok_end, mi_row, mi_col + bs,
- 1);
+ write_modes_b(cpi, tile, mi_8x8 + bs, bc, tok, tok_end,
+ mi_row, mi_col + bs, 1);
break;
case PARTITION_SPLIT:
for (n = 0; n < 4; n++) {const int j = n >> 1, i = n & 1;
- write_modes_sb(cpi, mi_8x8 + j * bs * mis + i * bs, bc, tok, tok_end,
+ write_modes_sb(cpi, tile, mi_8x8 + j * bs * mis + i * bs, bc,
+ tok, tok_end,
mi_row + j * bs, mi_col + i * bs, subsize, n);
}
break;
@@ -666,7 +670,8 @@
mi_row, mi_col, subsize, bsize);
}
-static void write_modes(VP9_COMP *cpi, vp9_writer* const bc,
+static void write_modes(VP9_COMP *cpi, const TileInfo *const tile,
+ vp9_writer* const bc,
TOKENEXTRA **tok, TOKENEXTRA *tok_end) {VP9_COMMON *const cm = &cpi->common;
const int mis = cm->mode_info_stride;
@@ -674,15 +679,15 @@
MODE_INFO **mi_8x8 = cm->mi_grid_visible;
MODE_INFO **m_8x8;
- mi_8x8 += cm->cur_tile_mi_col_start + cm->cur_tile_mi_row_start * mis;
+ mi_8x8 += tile->mi_col_start + tile->mi_row_start * mis;
- for (mi_row = cm->cur_tile_mi_row_start; mi_row < cm->cur_tile_mi_row_end;
+ for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
mi_row += 8, mi_8x8 += 8 * mis) {m_8x8 = mi_8x8;
vp9_zero(cpi->left_seg_context);
- for (mi_col = cm->cur_tile_mi_col_start; mi_col < cm->cur_tile_mi_col_end;
+ for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE, m_8x8 += MI_BLOCK_SIZE) {- write_modes_sb(cpi, m_8x8, bc, tok, tok_end, mi_row, mi_col,
+ write_modes_sb(cpi, tile, m_8x8, bc, tok, tok_end, mi_row, mi_col,
BLOCK_64X64, 0);
}
}
@@ -1218,9 +1223,10 @@
}
for (tile_row = 0; tile_row < tile_rows; tile_row++) {- vp9_get_tile_row_offsets(cm, tile_row);
for (tile_col = 0; tile_col < tile_cols; tile_col++) {- vp9_get_tile_col_offsets(cm, tile_col);
+ TileInfo tile;
+
+ vp9_tile_init(&tile, cm, 0, tile_col);
tok_end = tok[tile_row][tile_col] + cpi->tok_count[tile_row][tile_col];
if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1)
@@ -1228,7 +1234,7 @@
else
vp9_start_encode(&residual_bc, data_ptr + total_size);
- write_modes(cpi, &residual_bc, &tok[tile_row][tile_col], tok_end);
+ write_modes(cpi, &tile, &residual_bc, &tok[tile_row][tile_col], tok_end);
assert(tok[tile_row][tile_col] == tok_end);
vp9_stop_encode(&residual_bc);
if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1) {--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -484,8 +484,8 @@
x->e_mbd.plane[i].subsampling_y);
}
-static void set_offsets(VP9_COMP *cpi, int mi_row, int mi_col,
- BLOCK_SIZE bsize) {+static void set_offsets(VP9_COMP *cpi, const TileInfo *const tile,
+ int mi_row, int mi_col, BLOCK_SIZE bsize) {MACROBLOCK *const x = &cpi->mb;
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &x->e_mbd;
@@ -528,7 +528,8 @@
// Set up distance of MB to edge of frame in 1/8th pel units
assert(!(mi_col & (mi_width - 1)) && !(mi_row & (mi_height - 1)));
- set_mi_row_col(cm, xd, mi_row, mi_height, mi_col, mi_width);
+ set_mi_row_col(xd, tile, mi_row, mi_height, mi_col, mi_width,
+ cm->mi_rows, cm->mi_cols);
/* set up source buffers */
vp9_setup_src_planes(x, cpi->Source, mi_row, mi_col);
@@ -555,9 +556,8 @@
const int x = mb_col & ~3;
const int p16 = ((mb_row & 1) << 1) + (mb_col & 1);
const int p32 = ((mb_row & 2) << 2) + ((mb_col & 2) << 1);
- const int tile_progress = cm->cur_tile_mi_col_start * cm->mb_rows >> 1;
- const int mb_cols = (cm->cur_tile_mi_col_end - cm->cur_tile_mi_col_start)
- >> 1;
+ const int tile_progress = tile->mi_col_start * cm->mb_rows >> 1;
+ const int mb_cols = (tile->mi_col_end - tile->mi_col_start) >> 1;
cpi->seg0_progress = ((y * mb_cols + x * 4 + p32 + p16 + tile_progress)
<< 16) / cm->MBs;
@@ -570,7 +570,8 @@
}
}
-static void pick_sb_modes(VP9_COMP *cpi, int mi_row, int mi_col,
+static void pick_sb_modes(VP9_COMP *cpi, const TileInfo *const tile,
+ int mi_row, int mi_col,
int *totalrate, int64_t *totaldist,
BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx,
int64_t best_rd) {@@ -596,7 +597,7 @@
}
}
- set_offsets(cpi, mi_row, mi_col, bsize);
+ set_offsets(cpi, tile, mi_row, mi_col, bsize);
xd->mi_8x8[0]->mbmi.sb_type = bsize;
// Set to zero to make sure we do not use the previous encoded frame stats
@@ -632,10 +633,10 @@
best_rd);
} else {if (bsize >= BLOCK_8X8)
- vp9_rd_pick_inter_mode_sb(cpi, x, mi_row, mi_col, totalrate, totaldist,
- bsize, ctx, best_rd);
+ vp9_rd_pick_inter_mode_sb(cpi, x, tile, mi_row, mi_col,
+ totalrate, totaldist, bsize, ctx, best_rd);
else
- vp9_rd_pick_inter_mode_sub8x8(cpi, x, mi_row, mi_col, totalrate,
+ vp9_rd_pick_inter_mode_sub8x8(cpi, x, tile, mi_row, mi_col, totalrate,
totaldist, bsize, ctx, best_rd);
}
@@ -769,7 +770,8 @@
sizeof(cpi->left_seg_context[0]) * mi_height);
}
-static void encode_b(VP9_COMP *cpi, TOKENEXTRA **tp, int mi_row, int mi_col,
+static void encode_b(VP9_COMP *cpi, const TileInfo *const tile,
+ TOKENEXTRA **tp, int mi_row, int mi_col,
int output_enabled, BLOCK_SIZE bsize, int sub_index) {VP9_COMMON * const cm = &cpi->common;
MACROBLOCK * const x = &cpi->mb;
@@ -787,7 +789,7 @@
if (xd->ab_index > 0)
return;
}
- set_offsets(cpi, mi_row, mi_col, bsize);
+ set_offsets(cpi, tile, mi_row, mi_col, bsize);
update_state(cpi, get_block_context(x, bsize), bsize, output_enabled);
encode_superblock(cpi, tp, output_enabled, mi_row, mi_col, bsize);
@@ -799,7 +801,8 @@
}
}
-static void encode_sb(VP9_COMP *cpi, TOKENEXTRA **tp, int mi_row, int mi_col,
+static void encode_sb(VP9_COMP *cpi, const TileInfo *const tile,
+ TOKENEXTRA **tp, int mi_row, int mi_col,
int output_enabled, BLOCK_SIZE bsize) {VP9_COMMON * const cm = &cpi->common;
MACROBLOCK * const x = &cpi->mb;
@@ -826,19 +829,19 @@
case PARTITION_NONE:
if (output_enabled && bsize >= BLOCK_8X8)
cpi->partition_count[pl][PARTITION_NONE]++;
- encode_b(cpi, tp, mi_row, mi_col, output_enabled, c1, -1);
+ encode_b(cpi, tile, tp, mi_row, mi_col, output_enabled, c1, -1);
break;
case PARTITION_VERT:
if (output_enabled)
cpi->partition_count[pl][PARTITION_VERT]++;
- encode_b(cpi, tp, mi_row, mi_col, output_enabled, c1, 0);
- encode_b(cpi, tp, mi_row, mi_col + bs, output_enabled, c1, 1);
+ encode_b(cpi, tile, tp, mi_row, mi_col, output_enabled, c1, 0);
+ encode_b(cpi, tile, tp, mi_row, mi_col + bs, output_enabled, c1, 1);
break;
case PARTITION_HORZ:
if (output_enabled)
cpi->partition_count[pl][PARTITION_HORZ]++;
- encode_b(cpi, tp, mi_row, mi_col, output_enabled, c1, 0);
- encode_b(cpi, tp, mi_row + bs, mi_col, output_enabled, c1, 1);
+ encode_b(cpi, tile, tp, mi_row, mi_col, output_enabled, c1, 0);
+ encode_b(cpi, tile, tp, mi_row + bs, mi_col, output_enabled, c1, 1);
break;
case PARTITION_SPLIT:
subsize = get_subsize(bsize, PARTITION_SPLIT);
@@ -850,7 +853,7 @@
const int x_idx = i & 1, y_idx = i >> 1;
*get_sb_index(xd, subsize) = i;
- encode_sb(cpi, tp, mi_row + y_idx * bs, mi_col + x_idx * bs,
+ encode_sb(cpi, tile, tp, mi_row + y_idx * bs, mi_col + x_idx * bs,
output_enabled, subsize);
}
break;
@@ -889,13 +892,13 @@
// However, at the bottom and right borders of the image the requested size
// may not be allowed in which case this code attempts to choose the largest
// allowable partition.
-static void set_partitioning(VP9_COMP *cpi, MODE_INFO **mi_8x8,
- int mi_row, int mi_col) {+static void set_partitioning(VP9_COMP *cpi, const TileInfo *const tile,
+ MODE_INFO **mi_8x8, int mi_row, int mi_col) {VP9_COMMON *const cm = &cpi->common;
BLOCK_SIZE bsize = cpi->sf.always_this_block_size;
const int mis = cm->mode_info_stride;
- int row8x8_remaining = cm->cur_tile_mi_row_end - mi_row;
- int col8x8_remaining = cm->cur_tile_mi_col_end - mi_col;
+ int row8x8_remaining = tile->mi_row_end - mi_row;
+ int col8x8_remaining = tile->mi_col_end - mi_col;
int block_row, block_col;
MODE_INFO * mi_upper_left = cm->mi + mi_row * mis + mi_col;
int bh = num_8x8_blocks_high_lookup[bsize];
@@ -970,7 +973,9 @@
return 0;
}
-static void rd_use_partition(VP9_COMP *cpi, MODE_INFO **mi_8x8,
+static void rd_use_partition(VP9_COMP *cpi,
+ const TileInfo *const tile,
+ MODE_INFO **mi_8x8,
TOKENEXTRA **tp, int mi_row, int mi_col,
BLOCK_SIZE bsize, int *rate, int64_t *dist,
int do_recon) {@@ -1022,7 +1027,7 @@
save_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
if (bsize == BLOCK_16X16) {- set_offsets(cpi, mi_row, mi_col, bsize);
+ set_offsets(cpi, tile, mi_row, mi_col, bsize);
x->mb_energy = vp9_block_energy(cpi, x, bsize);
}
@@ -1049,7 +1054,7 @@
mi_row + (ms >> 1) < cm->mi_rows &&
mi_col + (ms >> 1) < cm->mi_cols) {*(get_sb_partitioning(x, bsize)) = bsize;
- pick_sb_modes(cpi, mi_row, mi_col, &none_rate, &none_dist, bsize,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &none_rate, &none_dist, bsize,
get_block_context(x, bsize), INT64_MAX);
pl = partition_plane_context(cpi->above_seg_context,
@@ -1065,12 +1070,12 @@
switch (partition) {case PARTITION_NONE:
- pick_sb_modes(cpi, mi_row, mi_col, &last_part_rate, &last_part_dist,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &last_part_rate, &last_part_dist,
bsize, get_block_context(x, bsize), INT64_MAX);
break;
case PARTITION_HORZ:
*get_sb_index(xd, subsize) = 0;
- pick_sb_modes(cpi, mi_row, mi_col, &last_part_rate, &last_part_dist,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &last_part_rate, &last_part_dist,
subsize, get_block_context(x, subsize), INT64_MAX);
if (last_part_rate != INT_MAX &&
bsize >= BLOCK_8X8 && mi_row + (mh >> 1) < cm->mi_rows) {@@ -1079,7 +1084,7 @@
update_state(cpi, get_block_context(x, subsize), subsize, 0);
encode_superblock(cpi, tp, 0, mi_row, mi_col, subsize);
*get_sb_index(xd, subsize) = 1;
- pick_sb_modes(cpi, mi_row + (ms >> 1), mi_col, &rt, &dt, subsize,
+ pick_sb_modes(cpi, tile, mi_row + (ms >> 1), mi_col, &rt, &dt, subsize,
get_block_context(x, subsize), INT64_MAX);
if (rt == INT_MAX || dt == INT_MAX) {last_part_rate = INT_MAX;
@@ -1093,7 +1098,7 @@
break;
case PARTITION_VERT:
*get_sb_index(xd, subsize) = 0;
- pick_sb_modes(cpi, mi_row, mi_col, &last_part_rate, &last_part_dist,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &last_part_rate, &last_part_dist,
subsize, get_block_context(x, subsize), INT64_MAX);
if (last_part_rate != INT_MAX &&
bsize >= BLOCK_8X8 && mi_col + (ms >> 1) < cm->mi_cols) {@@ -1102,7 +1107,7 @@
update_state(cpi, get_block_context(x, subsize), subsize, 0);
encode_superblock(cpi, tp, 0, mi_row, mi_col, subsize);
*get_sb_index(xd, subsize) = 1;
- pick_sb_modes(cpi, mi_row, mi_col + (ms >> 1), &rt, &dt, subsize,
+ pick_sb_modes(cpi, tile, mi_row, mi_col + (ms >> 1), &rt, &dt, subsize,
get_block_context(x, subsize), INT64_MAX);
if (rt == INT_MAX || dt == INT_MAX) {last_part_rate = INT_MAX;
@@ -1129,7 +1134,7 @@
*get_sb_index(xd, subsize) = i;
- rd_use_partition(cpi, mi_8x8 + jj * bss * mis + ii * bss, tp,
+ rd_use_partition(cpi, tile, mi_8x8 + jj * bss * mis + ii * bss, tp,
mi_row + y_idx, mi_col + x_idx, subsize, &rt, &dt,
i != 3);
if (rt == INT_MAX || dt == INT_MAX) {@@ -1178,7 +1183,7 @@
save_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
- pick_sb_modes(cpi, mi_row + y_idx, mi_col + x_idx, &rt, &dt,
+ pick_sb_modes(cpi, tile, mi_row + y_idx, mi_col + x_idx, &rt, &dt,
split_subsize, get_block_context(x, split_subsize),
INT64_MAX);
@@ -1191,7 +1196,7 @@
}
if (i != 3)
- encode_sb(cpi, tp, mi_row + y_idx, mi_col + x_idx, 0,
+ encode_sb(cpi, tile, tp, mi_row + y_idx, mi_col + x_idx, 0,
split_subsize);
split_rate += rt;
@@ -1237,7 +1242,7 @@
assert(chosen_rate < INT_MAX && chosen_dist < INT_MAX);
if (do_recon)
- encode_sb(cpi, tp, mi_row, mi_col, bsize == BLOCK_64X64, bsize);
+ encode_sb(cpi, tile, tp, mi_row, mi_col, bsize == BLOCK_64X64, bsize);
*rate = chosen_rate;
*dist = chosen_dist;
@@ -1285,7 +1290,8 @@
// Look at neighboring blocks and set a min and max partition size based on
// what they chose.
-static void rd_auto_partition_range(VP9_COMP *cpi, int row, int col,
+static void rd_auto_partition_range(VP9_COMP *cpi, const TileInfo *const tile,
+ int row, int col,
BLOCK_SIZE *min_block_size,
BLOCK_SIZE *max_block_size) {VP9_COMMON * const cm = &cpi->common;
@@ -1299,8 +1305,8 @@
MODE_INFO ** above_sb64_mi_8x8;
MODE_INFO ** left_sb64_mi_8x8;
- int row8x8_remaining = cm->cur_tile_mi_row_end - row;
- int col8x8_remaining = cm->cur_tile_mi_col_end - col;
+ int row8x8_remaining = tile->mi_row_end - row;
+ int col8x8_remaining = tile->mi_col_end - col;
int bh, bw;
// Trap case where we do not have a prediction.
@@ -1450,7 +1456,8 @@
// TODO(jingning,jimbankoski,rbultje): properly skip partition types that are
// unlikely to be selected depending on previous rate-distortion optimization
// results, for encoding speed-up.
-static void rd_pick_partition(VP9_COMP *cpi, TOKENEXTRA **tp, int mi_row,
+static void rd_pick_partition(VP9_COMP *cpi, const TileInfo *const tile,
+ TOKENEXTRA **tp, int mi_row,
int mi_col, BLOCK_SIZE bsize, int *rate,
int64_t *dist, int do_recon, int64_t best_rd) {VP9_COMMON * const cm = &cpi->common;
@@ -1490,7 +1497,7 @@
assert(mi_height_log2(bsize) == mi_width_log2(bsize));
if (bsize == BLOCK_16X16) {- set_offsets(cpi, mi_row, mi_col, bsize);
+ set_offsets(cpi, tile, mi_row, mi_col, bsize);
x->mb_energy = vp9_block_energy(cpi, x, bsize);
}
@@ -1527,7 +1534,7 @@
// PARTITION_NONE
if (partition_none_allowed) {- pick_sb_modes(cpi, mi_row, mi_col, &this_rate, &this_dist, bsize,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &this_rate, &this_dist, bsize,
get_block_context(x, bsize), best_rd);
if (this_rate != INT_MAX) { if (bsize >= BLOCK_8X8) {@@ -1581,7 +1588,7 @@
*get_sb_index(xd, subsize) = i;
if (cpi->sf.adaptive_motion_search)
load_pred_mv(x, get_block_context(x, bsize));
- rd_pick_partition(cpi, tp, mi_row + y_idx, mi_col + x_idx, subsize,
+ rd_pick_partition(cpi, tile, tp, mi_row + y_idx, mi_col + x_idx, subsize,
&this_rate, &this_dist, i != 3, best_rd - sum_rd);
if (this_rate == INT_MAX) {@@ -1628,7 +1635,7 @@
*get_sb_index(xd, subsize) = 0;
if (cpi->sf.adaptive_motion_search)
load_pred_mv(x, get_block_context(x, bsize));
- pick_sb_modes(cpi, mi_row, mi_col, &sum_rate, &sum_dist, subsize,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &sum_rate, &sum_dist, subsize,
get_block_context(x, subsize), best_rd);
sum_rd = RDCOST(x->rdmult, x->rddiv, sum_rate, sum_dist);
@@ -1639,7 +1646,7 @@
*get_sb_index(xd, subsize) = 1;
if (cpi->sf.adaptive_motion_search)
load_pred_mv(x, get_block_context(x, bsize));
- pick_sb_modes(cpi, mi_row + ms, mi_col, &this_rate,
+ pick_sb_modes(cpi, tile, mi_row + ms, mi_col, &this_rate,
&this_dist, subsize, get_block_context(x, subsize),
best_rd - sum_rd);
if (this_rate == INT_MAX) {@@ -1673,7 +1680,7 @@
*get_sb_index(xd, subsize) = 0;
if (cpi->sf.adaptive_motion_search)
load_pred_mv(x, get_block_context(x, bsize));
- pick_sb_modes(cpi, mi_row, mi_col, &sum_rate, &sum_dist, subsize,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &sum_rate, &sum_dist, subsize,
get_block_context(x, subsize), best_rd);
sum_rd = RDCOST(x->rdmult, x->rddiv, sum_rate, sum_dist);
if (sum_rd < best_rd && mi_col + ms < cm->mi_cols) {@@ -1683,7 +1690,7 @@
*get_sb_index(xd, subsize) = 1;
if (cpi->sf.adaptive_motion_search)
load_pred_mv(x, get_block_context(x, bsize));
- pick_sb_modes(cpi, mi_row, mi_col + ms, &this_rate,
+ pick_sb_modes(cpi, tile, mi_row, mi_col + ms, &this_rate,
&this_dist, subsize, get_block_context(x, subsize),
best_rd - sum_rd);
if (this_rate == INT_MAX) {@@ -1715,7 +1722,7 @@
*dist = best_dist;
if (best_rate < INT_MAX && best_dist < INT64_MAX && do_recon)
- encode_sb(cpi, tp, mi_row, mi_col, bsize == BLOCK_64X64, bsize);
+ encode_sb(cpi, tile, tp, mi_row, mi_col, bsize == BLOCK_64X64, bsize);
if (bsize == BLOCK_64X64) {assert(tp_orig < *tp);
assert(best_rate < INT_MAX);
@@ -1726,7 +1733,8 @@
}
// Examines 64x64 block and chooses a best reference frame
-static void rd_pick_reference_frame(VP9_COMP *cpi, int mi_row, int mi_col) {+static void rd_pick_reference_frame(VP9_COMP *cpi, const TileInfo *const tile,
+ int mi_row, int mi_col) {VP9_COMMON * const cm = &cpi->common;
MACROBLOCK * const x = &cpi->mb;
int bsl = b_width_log2(BLOCK_64X64), bs = 1 << bsl;
@@ -1746,7 +1754,7 @@
if ((mi_row + (ms >> 1) < cm->mi_rows) &&
(mi_col + (ms >> 1) < cm->mi_cols)) {cpi->set_ref_frame_mask = 1;
- pick_sb_modes(cpi, mi_row, mi_col, &r, &d, BLOCK_64X64,
+ pick_sb_modes(cpi, tile, mi_row, mi_col, &r, &d, BLOCK_64X64,
get_block_context(x, BLOCK_64X64), INT64_MAX);
pl = partition_plane_context(cpi->above_seg_context, cpi->left_seg_context,
mi_row, mi_col, BLOCK_64X64);
@@ -1759,8 +1767,8 @@
restore_context(cpi, mi_row, mi_col, a, l, sa, sl, BLOCK_64X64);
}
-static void encode_sb_row(VP9_COMP *cpi, int mi_row, TOKENEXTRA **tp,
- int *totalrate) {+static void encode_sb_row(VP9_COMP *cpi, const TileInfo *const tile,
+ int mi_row, TOKENEXTRA **tp, int *totalrate) {VP9_COMMON * const cm = &cpi->common;
int mi_col;
@@ -1769,7 +1777,7 @@
vpx_memset(cpi->left_seg_context, 0, sizeof(cpi->left_seg_context));
// Code each SB in the row
- for (mi_col = cm->cur_tile_mi_col_start; mi_col < cm->cur_tile_mi_col_end;
+ for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE) {int dummy_rate;
int64_t dummy_dist;
@@ -1777,7 +1785,7 @@
vp9_zero(cpi->mb.pred_mv);
if (cpi->sf.reference_masking)
- rd_pick_reference_frame(cpi, mi_row, mi_col);
+ rd_pick_reference_frame(cpi, tile, mi_row, mi_col);
if (cpi->sf.use_lastframe_partitioning ||
cpi->sf.use_one_partition_size_always ) {@@ -1787,9 +1795,9 @@
cpi->mb.source_variance = UINT_MAX;
if (cpi->sf.use_one_partition_size_always) {- set_offsets(cpi, mi_row, mi_col, BLOCK_64X64);
- set_partitioning(cpi, mi_8x8, mi_row, mi_col);
- rd_use_partition(cpi, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64,
+ set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64);
+ set_partitioning(cpi, tile, mi_8x8, mi_row, mi_col);
+ rd_use_partition(cpi, tile, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64,
&dummy_rate, &dummy_dist, 1);
} else {if ((cpi->common.current_video_frame
@@ -1803,16 +1811,16 @@
sb_has_motion(cpi, prev_mi_8x8))) {// If required set upper and lower partition size limits
if (cpi->sf.auto_min_max_partition_size) {- set_offsets(cpi, mi_row, mi_col, BLOCK_64X64);
- rd_auto_partition_range(cpi, mi_row, mi_col,
+ set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64);
+ rd_auto_partition_range(cpi, tile, mi_row, mi_col,
&cpi->sf.min_partition_size,
&cpi->sf.max_partition_size);
}
- rd_pick_partition(cpi, tp, mi_row, mi_col, BLOCK_64X64,
+ rd_pick_partition(cpi, tile, tp, mi_row, mi_col, BLOCK_64X64,
&dummy_rate, &dummy_dist, 1, INT64_MAX);
} else {copy_partitioning(cpi, mi_8x8, prev_mi_8x8);
- rd_use_partition(cpi, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64,
+ rd_use_partition(cpi, tile, mi_8x8, tp, mi_row, mi_col, BLOCK_64X64,
&dummy_rate, &dummy_dist, 1);
}
}
@@ -1819,12 +1827,12 @@
} else {// If required set upper and lower partition size limits
if (cpi->sf.auto_min_max_partition_size) {- set_offsets(cpi, mi_row, mi_col, BLOCK_64X64);
- rd_auto_partition_range(cpi, mi_row, mi_col,
+ set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64);
+ rd_auto_partition_range(cpi, tile, mi_row, mi_col,
&cpi->sf.min_partition_size,
&cpi->sf.max_partition_size);
}
- rd_pick_partition(cpi, tp, mi_row, mi_col, BLOCK_64X64,
+ rd_pick_partition(cpi, tile, tp, mi_row, mi_col, BLOCK_64X64,
&dummy_rate, &dummy_dist, 1, INT64_MAX);
}
}
@@ -1979,16 +1987,15 @@
const int tile_rows = 1 << cm->log2_tile_rows;
for (tile_row = 0; tile_row < tile_rows; tile_row++) {- vp9_get_tile_row_offsets(cm, tile_row);
-
for (tile_col = 0; tile_col < tile_cols; tile_col++) {+ TileInfo tile;
TOKENEXTRA *tp_old = tp;
// For each row of SBs in the frame
- vp9_get_tile_col_offsets(cm, tile_col);
- for (mi_row = cm->cur_tile_mi_row_start;
- mi_row < cm->cur_tile_mi_row_end; mi_row += 8)
- encode_sb_row(cpi, mi_row, &tp, &totalrate);
+ vp9_tile_init(&tile, cm, tile_row, tile_col);
+ for (mi_row = tile.mi_row_start;
+ mi_row < tile.mi_row_end; mi_row += 8)
+ encode_sb_row(cpi, &tile, mi_row, &tp, &totalrate);
cpi->tok_count[tile_row][tile_col] = (unsigned int)(tp - tp_old);
assert(tp - cpi->tok <= get_token_alloc(cm->mb_rows, cm->mb_cols));
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -481,6 +481,7 @@
MACROBLOCK *const x = &cpi->mb;
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &x->e_mbd;
+ TileInfo tile;
int recon_yoffset, recon_uvoffset;
const int lst_yv12_idx = cm->ref_frame_map[cpi->lst_fb_idx];
@@ -532,6 +533,9 @@
vp9_initialize_rd_consts(cpi);
}
+ // tiling is ignored in the first pass
+ vp9_tile_init(&tile, cm, 0, 0);
+
// for each macroblock row in image
for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {int_mv best_ref_mv;
@@ -578,11 +582,12 @@
}
}
xd->mi_8x8[0]->mbmi.ref_frame[0] = INTRA_FRAME;
- set_mi_row_col(cm, xd,
+ set_mi_row_col(xd, &tile,
mb_row << 1,
1 << mi_height_log2(xd->mi_8x8[0]->mbmi.sb_type),
mb_col << 1,
- 1 << mi_width_log2(xd->mi_8x8[0]->mbmi.sb_type));
+ 1 << mi_width_log2(xd->mi_8x8[0]->mbmi.sb_type),
+ cm->mi_rows, cm->mi_cols);
if (cpi->sf.variance_adaptive_quantization) {int energy = vp9_block_energy(cpi, x, xd->mi_8x8[0]->mbmi.sb_type);
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1432,10 +1432,6 @@
int mi_row, int mi_col,
int_mv single_newmv[MAX_REF_FRAMES],
int *rate_mv);
-static void single_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
- BLOCK_SIZE bsize,
- int mi_row, int mi_col,
- int_mv *tmp_mv, int *rate_mv);
static int labels2mode(MACROBLOCK *x, int i,
MB_PREDICTION_MODE this_mode,
@@ -1646,6 +1642,7 @@
}
static void rd_check_segment_txsize(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
BEST_SEG_INFO *bsi_buf, int filter_idx,
int_mv seg_mvs[4][MAX_REF_FRAMES],
int mi_row, int mi_col) {@@ -1691,13 +1688,13 @@
i = idy * 2 + idx;
frame_mv[ZEROMV][mbmi->ref_frame[0]].as_int = 0;
- vp9_append_sub8x8_mvs_for_idx(&cpi->common, &x->e_mbd,
+ vp9_append_sub8x8_mvs_for_idx(&cpi->common, &x->e_mbd, tile,
&frame_mv[NEARESTMV][mbmi->ref_frame[0]],
&frame_mv[NEARMV][mbmi->ref_frame[0]],
i, 0, mi_row, mi_col);
if (has_second_rf) {frame_mv[ZEROMV][mbmi->ref_frame[1]].as_int = 0;
- vp9_append_sub8x8_mvs_for_idx(&cpi->common, &x->e_mbd,
+ vp9_append_sub8x8_mvs_for_idx(&cpi->common, &x->e_mbd, tile,
&frame_mv[NEARESTMV][mbmi->ref_frame[1]],
&frame_mv[NEARMV][mbmi->ref_frame[1]],
i, 1, mi_row, mi_col);
@@ -2027,6 +2024,7 @@
}
static int64_t rd_pick_best_mbsegmentation(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
int_mv *best_ref_mv,
int_mv *second_best_ref_mv,
int64_t best_rd,
@@ -2057,7 +2055,8 @@
for (i = 0; i < 4; i++)
bsi->modes[i] = ZEROMV;
- rd_check_segment_txsize(cpi, x, bsi_buf, filter_idx, seg_mvs, mi_row, mi_col);
+ rd_check_segment_txsize(cpi, x, tile, bsi_buf, filter_idx, seg_mvs,
+ mi_row, mi_col);
if (bsi->segment_rd > best_rd)
return INT64_MAX;
@@ -2253,6 +2252,7 @@
}
static void setup_buffer_inter(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
int idx, MV_REFERENCE_FRAME frame_type,
BLOCK_SIZE block_size,
int mi_row, int mi_col,
@@ -2281,7 +2281,7 @@
&scale[frame_type], &scale[frame_type]);
// Gets an initial list of candidate vectors from neighbours and orders them
- vp9_find_mv_refs(&cpi->common, xd, xd->mi_8x8[0],
+ vp9_find_mv_refs(cm, xd, tile, xd->mi_8x8[0],
xd->last_mi,
frame_type,
mbmi->ref_mvs[frame_type], mi_row, mi_col);
@@ -2318,6 +2318,7 @@
}
static void single_motion_search(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
BLOCK_SIZE bsize,
int mi_row, int mi_col,
int_mv *tmp_mv, int *rate_mv) {@@ -2614,6 +2615,7 @@
}
static int64_t handle_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
BLOCK_SIZE bsize,
int64_t txfm_cache[],
int *rate2, int64_t *distortion,
@@ -2672,7 +2674,8 @@
*rate2 += rate_mv;
} else {int_mv tmp_mv;
- single_motion_search(cpi, x, bsize, mi_row, mi_col, &tmp_mv, &rate_mv);
+ single_motion_search(cpi, x, tile, bsize, mi_row, mi_col,
+ &tmp_mv, &rate_mv);
*rate2 += rate_mv;
frame_mv[refs[0]].as_int =
xd->mi_8x8[0]->bmi[0].as_mv[0].as_int = tmp_mv.as_int;
@@ -3083,6 +3086,7 @@
}
int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
int mi_row, int mi_col,
int *returnrate,
int64_t *returndistortion,
@@ -3193,8 +3197,9 @@
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) { if (cpi->ref_frame_flags & flag_list[ref_frame]) {- setup_buffer_inter(cpi, x, idx_list[ref_frame], ref_frame, block_size,
- mi_row, mi_col, frame_mv[NEARESTMV], frame_mv[NEARMV],
+ setup_buffer_inter(cpi, x, tile, idx_list[ref_frame], ref_frame,
+ block_size, mi_row, mi_col,
+ frame_mv[NEARESTMV], frame_mv[NEARMV],
yv12_mb, scale_factor);
}
frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
@@ -3438,7 +3443,7 @@
} else {mbmi->mode = this_mode;
compmode_cost = vp9_cost_bit(comp_mode_p, second_ref_frame > INTRA_FRAME);
- this_rd = handle_inter_mode(cpi, x, bsize,
+ this_rd = handle_inter_mode(cpi, x, tile, bsize,
tx_cache,
&rate2, &distortion2, &skippable,
&rate_y, &distortion_y,
@@ -3780,6 +3785,7 @@
int64_t vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x,
+ const TileInfo *const tile,
int mi_row, int mi_col,
int *returnrate,
int64_t *returndistortion,
@@ -3864,8 +3870,9 @@
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) { if (cpi->ref_frame_flags & flag_list[ref_frame]) {- setup_buffer_inter(cpi, x, idx_list[ref_frame], ref_frame, block_size,
- mi_row, mi_col, frame_mv[NEARESTMV], frame_mv[NEARMV],
+ setup_buffer_inter(cpi, x, tile, idx_list[ref_frame], ref_frame,
+ block_size, mi_row, mi_col,
+ frame_mv[NEARESTMV], frame_mv[NEARMV],
yv12_mb, scale_factor);
}
frame_mv[NEWMV][ref_frame].as_int = INVALID_MV;
@@ -4095,7 +4102,7 @@
mbmi->interp_filter = switchable_filter_index;
vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
- tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
+ tmp_rd = rd_pick_best_mbsegmentation(cpi, x, tile,
&mbmi->ref_mvs[ref_frame][0],
second_ref,
best_yrd,
@@ -4159,7 +4166,7 @@
if (!pred_exists) {// Handles the special case when a filter that is not in the
// switchable list (bilinear, 6-tap) is indicated at the frame level
- tmp_rd = rd_pick_best_mbsegmentation(cpi, x,
+ tmp_rd = rd_pick_best_mbsegmentation(cpi, x, tile,
&mbmi->ref_mvs[ref_frame][0],
second_ref,
best_yrd,
--- a/vp9/encoder/vp9_rdopt.h
+++ b/vp9/encoder/vp9_rdopt.h
@@ -18,6 +18,8 @@
(((128 + ((int64_t)R) * (RM)) >> 8) + (D << DM))
#define QIDX_SKIP_THRESH 115
+struct TileInfo;
+
int vp9_compute_rd_mult(VP9_COMP *cpi, int qindex);
void vp9_initialize_rd_consts(VP9_COMP *cpi);
@@ -29,14 +31,22 @@
PICK_MODE_CONTEXT *ctx, int64_t best_rd);
int64_t vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, MACROBLOCK *x,
+ const struct TileInfo *const tile,
int mi_row, int mi_col,
- int *r, int64_t *d, BLOCK_SIZE bsize,
- PICK_MODE_CONTEXT *ctx, int64_t best_rd);
+ int *returnrate,
+ int64_t *returndistortion,
+ BLOCK_SIZE bsize,
+ PICK_MODE_CONTEXT *ctx,
+ int64_t best_rd_so_far);
int64_t vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x,
+ const struct TileInfo *const tile,
int mi_row, int mi_col,
- int *r, int64_t *d, BLOCK_SIZE bsize,
- PICK_MODE_CONTEXT *ctx, int64_t best_rd);
+ int *returnrate,
+ int64_t *returndistortion,
+ BLOCK_SIZE bsize,
+ PICK_MODE_CONTEXT *ctx,
+ int64_t best_rd_so_far);
void vp9_init_me_luts();
--- a/vp9/encoder/vp9_segmentation.c
+++ b/vp9/encoder/vp9_segmentation.c
@@ -117,7 +117,8 @@
return cost;
}
-static void count_segs(VP9_COMP *cpi, MODE_INFO **mi_8x8,
+static void count_segs(VP9_COMP *cpi, const TileInfo *const tile,
+ MODE_INFO **mi_8x8,
int *no_pred_segcounts,
int (*temporal_predictor_count)[2],
int *t_unpred_seg_counts,
@@ -132,7 +133,7 @@
xd->mi_8x8 = mi_8x8;
segment_id = xd->mi_8x8[0]->mbmi.segment_id;
- set_mi_row_col(cm, xd, mi_row, bh, mi_col, bw);
+ set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols);
// Count the number of hits on each segment with no prediction
no_pred_segcounts[segment_id]++;
@@ -157,7 +158,8 @@
}
}
-static void count_segs_sb(VP9_COMP *cpi, MODE_INFO **mi_8x8,
+static void count_segs_sb(VP9_COMP *cpi, const TileInfo *const tile,
+ MODE_INFO **mi_8x8,
int *no_pred_segcounts,
int (*temporal_predictor_count)[2],
int *t_unpred_seg_counts,
@@ -175,19 +177,20 @@
bh = num_8x8_blocks_high_lookup[mi_8x8[0]->mbmi.sb_type];
if (bw == bs && bh == bs) {- count_segs(cpi, mi_8x8, no_pred_segcounts, temporal_predictor_count,
+ count_segs(cpi, tile, mi_8x8, no_pred_segcounts, temporal_predictor_count,
t_unpred_seg_counts, bs, bs, mi_row, mi_col);
} else if (bw == bs && bh < bs) {- count_segs(cpi, mi_8x8, no_pred_segcounts, temporal_predictor_count,
+ count_segs(cpi, tile, mi_8x8, no_pred_segcounts, temporal_predictor_count,
t_unpred_seg_counts, bs, hbs, mi_row, mi_col);
- count_segs(cpi, mi_8x8 + hbs * mis, no_pred_segcounts,
+ count_segs(cpi, tile, mi_8x8 + hbs * mis, no_pred_segcounts,
temporal_predictor_count, t_unpred_seg_counts, bs, hbs,
mi_row + hbs, mi_col);
} else if (bw < bs && bh == bs) {- count_segs(cpi, mi_8x8, no_pred_segcounts, temporal_predictor_count,
+ count_segs(cpi, tile, mi_8x8, no_pred_segcounts, temporal_predictor_count,
t_unpred_seg_counts, hbs, bs, mi_row, mi_col);
- count_segs(cpi, mi_8x8 + hbs, no_pred_segcounts, temporal_predictor_count,
- t_unpred_seg_counts, hbs, bs, mi_row, mi_col + hbs);
+ count_segs(cpi, tile, mi_8x8 + hbs,
+ no_pred_segcounts, temporal_predictor_count, t_unpred_seg_counts,
+ hbs, bs, mi_row, mi_col + hbs);
} else {const BLOCK_SIZE subsize = subsize_lookup[PARTITION_SPLIT][bsize];
int n;
@@ -198,7 +201,7 @@
const int mi_dc = hbs * (n & 1);
const int mi_dr = hbs * (n >> 1);
- count_segs_sb(cpi, &mi_8x8[mi_dr * mis + mi_dc],
+ count_segs_sb(cpi, tile, &mi_8x8[mi_dr * mis + mi_dc],
no_pred_segcounts, temporal_predictor_count,
t_unpred_seg_counts,
mi_row + mi_dr, mi_col + mi_dc, subsize);
@@ -234,15 +237,18 @@
// First of all generate stats regarding how well the last segment map
// predicts this one
for (tile_col = 0; tile_col < 1 << cm->log2_tile_cols; tile_col++) {- vp9_get_tile_col_offsets(cm, tile_col);
- mi_ptr = cm->mi_grid_visible + cm->cur_tile_mi_col_start;
+ TileInfo tile;
+
+ vp9_tile_init(&tile, cm, 0, tile_col);
+ mi_ptr = cm->mi_grid_visible + tile.mi_col_start;
for (mi_row = 0; mi_row < cm->mi_rows;
mi_row += 8, mi_ptr += 8 * mis) {mi = mi_ptr;
- for (mi_col = cm->cur_tile_mi_col_start; mi_col < cm->cur_tile_mi_col_end;
+ for (mi_col = tile.mi_col_start; mi_col < tile.mi_col_end;
mi_col += 8, mi += 8)
- count_segs_sb(cpi, mi, no_pred_segcounts, temporal_predictor_count,
- t_unpred_seg_counts, mi_row, mi_col, BLOCK_64X64);
+ count_segs_sb(cpi, &tile, mi, no_pred_segcounts,
+ temporal_predictor_count, t_unpred_seg_counts,
+ mi_row, mi_col, BLOCK_64X64);
}
}