ref: 3ffa41aae32bdf2372b144cc7a1e65a921501551
parent: ded951793c5f13d62b608f4f964a4c5f1f633150
parent: 7b9ca3caa78bd08fbc4dac8a05396be6c7240143
author: James Zern <[email protected]>
date: Mon Oct 28 08:45:11 EDT 2013
Merge changes If9b16f7d,I75aab21c,I9cbb768c,If5cea3d3,I96940657,I025595d8,Ie0bc3935,I3ebb172d * changes: vp9: remove partition+entropy contexts from common vp9: add above/left_context to MACROBLOCKD vp9: add above/left_seg_context to MACROBLOCKD vp9: add above/left_context to encoder vp9: add above/left_seg_context to encoder vp9: pass entropy context directly to set_skip_context vp9: pass context directly to partition functions vp9/decode: add alloc_tile_storage()
--- a/vp9/common/vp9_alloccommon.c
+++ b/vp9/common/vp9_alloccommon.c
@@ -41,17 +41,12 @@
vpx_free(cm->mip);
vpx_free(cm->prev_mip);
- vpx_free(cm->above_seg_context);
vpx_free(cm->last_frame_seg_map);
vpx_free(cm->mi_grid_base);
vpx_free(cm->prev_mi_grid_base);
- vpx_free(cm->above_context[0]);
- for (i = 0; i < MAX_MB_PLANE; i++)
- cm->above_context[i] = 0;
cm->mip = NULL;
cm->prev_mip = NULL;
- cm->above_seg_context = NULL;
cm->last_frame_seg_map = NULL;
cm->mi_grid_base = NULL;
cm->prev_mi_grid_base = NULL;
@@ -85,7 +80,7 @@
}
int vp9_alloc_frame_buffers(VP9_COMMON *cm, int width, int height) {- int i, mi_cols;
+ int i;
const int aligned_width = ALIGN_POWER_OF_TWO(width, MI_SIZE_LOG2);
const int aligned_height = ALIGN_POWER_OF_TWO(height, MI_SIZE_LOG2);
@@ -140,21 +135,6 @@
setup_mi(cm);
- // FIXME(jkoleszar): allocate subsampled arrays for U/V once subsampling
- // information is exposed at this level
- mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
-
- // 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
- // block where mi unit size is 8x8.
- cm->above_context[0] = vpx_calloc(sizeof(ENTROPY_CONTEXT) * MAX_MB_PLANE *
- (2 * mi_cols), 1);
- if (!cm->above_context[0])
- goto fail;
-
- cm->above_seg_context = vpx_calloc(sizeof(PARTITION_CONTEXT) * mi_cols, 1);
- if (!cm->above_seg_context)
- goto fail;
-
// Create the segmentation map structure and set to 0.
cm->last_frame_seg_map = vpx_calloc(cm->mi_rows * cm->mi_cols, 1);
if (!cm->last_frame_seg_map)
@@ -186,17 +166,11 @@
}
void vp9_update_frame_size(VP9_COMMON *cm) {- int i, mi_cols;
const int aligned_width = ALIGN_POWER_OF_TWO(cm->width, MI_SIZE_LOG2);
const int aligned_height = ALIGN_POWER_OF_TWO(cm->height, MI_SIZE_LOG2);
set_mb_mi(cm, aligned_width, aligned_height);
setup_mi(cm);
-
- mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
- for (i = 1; i < MAX_MB_PLANE; i++)
- cm->above_context[i] =
- cm->above_context[0] + i * sizeof(ENTROPY_CONTEXT) * 2 * mi_cols;
// Initialize the previous frame segment map to 0.
if (cm->last_frame_seg_map)
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -226,6 +226,13 @@
unsigned char ab_index; // index of 4x4 block inside the 8x8 block
int q_index;
+
+ /* Y,U,V,(A) */
+ ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
+ ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16];
+
+ PARTITION_CONTEXT *above_seg_context;
+ PARTITION_CONTEXT left_seg_context[8];
} MACROBLOCKD;
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -185,14 +185,6 @@
struct loopfilter lf;
struct segmentation seg;
- /* Y,U,V */
- ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
- ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16];
-
- // partition contexts
- PARTITION_CONTEXT *above_seg_context;
- PARTITION_CONTEXT left_seg_context[8];
-
// Context probabilities for reference frame prediction
int allow_comp_inter_inter;
MV_REFERENCE_FRAME comp_fixed_ref;
@@ -254,15 +246,18 @@
return ALIGN_POWER_OF_TWO(n_mis, MI_BLOCK_SIZE_LOG2);
}
-static INLINE void set_skip_context(VP9_COMMON *cm, MACROBLOCKD *xd,
- int mi_row, int mi_col) {+static INLINE void set_skip_context(
+ MACROBLOCKD *xd,
+ ENTROPY_CONTEXT *above_context[MAX_MB_PLANE],
+ ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16],
+ int mi_row, int mi_col) {const int above_idx = mi_col * 2;
const int left_idx = (mi_row * 2) & 15;
int i;
for (i = 0; i < MAX_MB_PLANE; i++) {struct macroblockd_plane *const pd = &xd->plane[i];
- pd->above_context = cm->above_context[i] + (above_idx >> pd->subsampling_x);
- pd->left_context = cm->left_context[i] + (left_idx >> pd->subsampling_y);
+ pd->above_context = above_context[i] + (above_idx >> pd->subsampling_x);
+ pd->left_context = left_context[i] + (left_idx >> pd->subsampling_y);
}
}
@@ -313,12 +308,14 @@
return cm->frame_type == KEY_FRAME || cm->intra_only;
}
-static INLINE void update_partition_context(VP9_COMMON *cm,
- int mi_row, int mi_col,
- BLOCK_SIZE sb_type,
- BLOCK_SIZE sb_size) {- PARTITION_CONTEXT *above_ctx = cm->above_seg_context + mi_col;
- PARTITION_CONTEXT *left_ctx = cm->left_seg_context + (mi_row & MI_MASK);
+static INLINE void update_partition_context(
+ PARTITION_CONTEXT *above_seg_context,
+ PARTITION_CONTEXT left_seg_context[8],
+ int mi_row, int mi_col,
+ BLOCK_SIZE sb_type,
+ BLOCK_SIZE sb_size) {+ PARTITION_CONTEXT *above_ctx = above_seg_context + mi_col;
+ PARTITION_CONTEXT *left_ctx = left_seg_context + (mi_row & MI_MASK);
const int bsl = b_width_log2(sb_size), bs = (1 << bsl) / 2;
const int bwl = b_width_log2(sb_type);
@@ -337,11 +334,13 @@
vpx_memset(left_ctx, pcvalue[bhl == bsl], bs);
}
-static INLINE int partition_plane_context(const VP9_COMMON *cm,
- int mi_row, int mi_col,
- BLOCK_SIZE sb_type) {- const PARTITION_CONTEXT *above_ctx = cm->above_seg_context + mi_col;
- const PARTITION_CONTEXT *left_ctx = cm->left_seg_context + (mi_row & MI_MASK);
+static INLINE int partition_plane_context(
+ const PARTITION_CONTEXT *above_seg_context,
+ const PARTITION_CONTEXT left_seg_context[8],
+ int mi_row, int mi_col,
+ BLOCK_SIZE sb_type) {+ const PARTITION_CONTEXT *above_ctx = above_seg_context + mi_col;
+ const PARTITION_CONTEXT *left_ctx = left_seg_context + (mi_row & MI_MASK);
int bsl = mi_width_log2(sb_type), bs = 1 << bsl;
int above = 0, left = 0, i;
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -199,6 +199,43 @@
xd->plane[i].dequant = cm->uv_dequant[q_index];
}
+// Allocate storage for each tile column.
+// TODO(jzern): when max_threads <= 1 the same storage could be used for each
+// tile.
+static void alloc_tile_storage(VP9D_COMP *pbi, int tile_cols) {+ VP9_COMMON *const cm = &pbi->common;
+ const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
+ int i, tile_col;
+
+ CHECK_MEM_ERROR(cm, pbi->mi_streams,
+ 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);
+ pbi->mi_streams[tile_col] =
+ &cm->mi[cm->mi_rows * cm->cur_tile_mi_col_start];
+ }
+
+ // 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
+ // block where mi unit size is 8x8.
+ CHECK_MEM_ERROR(cm, pbi->above_context[0],
+ vpx_realloc(pbi->above_context[0],
+ sizeof(*pbi->above_context[0]) * MAX_MB_PLANE *
+ 2 * aligned_mi_cols));
+ for (i = 1; i < MAX_MB_PLANE; ++i) {+ pbi->above_context[i] = pbi->above_context[0] +
+ i * sizeof(*pbi->above_context[0]) *
+ 2 * aligned_mi_cols;
+ }
+
+ // This is sized based on the entire frame. Each tile operates within its
+ // column bounds.
+ CHECK_MEM_ERROR(cm, pbi->above_seg_context,
+ vpx_realloc(pbi->above_seg_context,
+ sizeof(*pbi->above_seg_context) *
+ aligned_mi_cols));
+}
+
static void decode_block(int plane, int block, BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {MACROBLOCKD* const xd = arg;
@@ -311,7 +348,7 @@
// cannot be used.
xd->last_mi = cm->prev_mi ? xd->prev_mi_8x8[0] : NULL;
- set_skip_context(cm, xd, mi_row, mi_col);
+ set_skip_context(xd, xd->above_context, xd->left_context, mi_row, mi_col);
// 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
@@ -387,6 +424,7 @@
static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
vp9_reader* r, BLOCK_SIZE bsize, int index) {VP9_COMMON *const cm = &pbi->common;
+ MACROBLOCKD *const xd = &pbi->mb;
const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
PARTITION_TYPE partition = PARTITION_NONE;
BLOCK_SIZE subsize;
@@ -401,7 +439,8 @@
int pl;
const int idx = check_bsize_coverage(hbs, cm->mi_rows, cm->mi_cols,
mi_row, mi_col);
- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(xd->above_seg_context, xd->left_seg_context,
+ mi_row, mi_col, bsize);
if (idx == 0)
partition = treed_read(r, vp9_partition_tree,
@@ -447,7 +486,8 @@
// update partition context
if (bsize >= BLOCK_8X8 &&
(bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
- update_partition_context(cm, mi_row, mi_col, subsize, bsize);
+ update_partition_context(xd->above_seg_context, xd->left_seg_context,
+ mi_row, mi_col, subsize, bsize);
}
static void setup_token_decoder(const uint8_t *data,
@@ -690,14 +730,24 @@
setup_display_size(cm, rb);
}
-static void decode_tile(VP9D_COMP *pbi, vp9_reader *r, int tile_col) {+static void setup_tile_context(VP9D_COMP *const pbi, MACROBLOCKD *const xd,
+ int tile_col) {+ int i;
+ xd->mi_stream = pbi->mi_streams[tile_col];
+
+ for (i = 0; i < MAX_MB_PLANE; ++i) {+ xd->above_context[i] = pbi->above_context[i];
+ }
+ // see note in alloc_tile_storage().
+ xd->above_seg_context = pbi->above_seg_context;
+}
+
+static void decode_tile(VP9D_COMP *pbi, vp9_reader *r) {const int num_threads = pbi->oxcf.max_threads;
VP9_COMMON *const cm = &pbi->common;
int mi_row, mi_col;
MACROBLOCKD *xd = &pbi->mb;
- xd->mi_stream = pbi->mi_streams[tile_col];
-
if (pbi->do_loopfilter_inline) {LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1;
lf_data->frame_buffer = get_frame_new_buffer(cm);
@@ -711,8 +761,8 @@
for (mi_row = cm->cur_tile_mi_row_start; mi_row < cm->cur_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(cm->left_context);
- vp9_zero(cm->left_seg_context);
+ 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;
mi_col += MI_BLOCK_SIZE)
decode_modes_sb(pbi, mi_row, mi_col, r, BLOCK_64X64, 0);
@@ -768,6 +818,7 @@
vp9_reader residual_bc;
VP9_COMMON *const cm = &pbi->common;
+ MACROBLOCKD *const xd = &pbi->mb;
const uint8_t *const data_end = pbi->source + pbi->source_sz;
const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
@@ -777,11 +828,12 @@
// Note: this memset assumes above_context[0], [1] and [2]
// are allocated as part of the same buffer.
- vpx_memset(cm->above_context[0], 0,
- sizeof(ENTROPY_CONTEXT) * MAX_MB_PLANE * (2 * aligned_mi_cols));
+ vpx_memset(pbi->above_context[0], 0,
+ sizeof(*pbi->above_context[0]) * MAX_MB_PLANE *
+ 2 * aligned_mi_cols);
- vpx_memset(cm->above_seg_context, 0,
- sizeof(PARTITION_CONTEXT) * aligned_mi_cols);
+ vpx_memset(pbi->above_seg_context, 0,
+ sizeof(*pbi->above_seg_context) * aligned_mi_cols);
if (pbi->oxcf.inv_tile_order) {const uint8_t *data_ptr2[4][1 << 6];
@@ -811,7 +863,8 @@
setup_token_decoder(data_ptr2[tile_row][tile_col], data_end,
data_end - data_ptr2[tile_row][tile_col],
&cm->error, &residual_bc);
- decode_tile(pbi, &residual_bc, tile_col);
+ setup_tile_context(pbi, xd, tile_col);
+ decode_tile(pbi, &residual_bc);
if (tile_row == tile_rows - 1 && tile_col == tile_cols - 1)
bc_bak = residual_bc;
}
@@ -840,7 +893,8 @@
}
setup_token_decoder(data, data_end, size, &cm->error, &residual_bc);
- decode_tile(pbi, &residual_bc, tile_col);
+ setup_tile_context(pbi, xd, tile_col);
+ decode_tile(pbi, &residual_bc);
data += size;
}
}
@@ -1092,7 +1146,6 @@
const int keyframe = cm->frame_type == KEY_FRAME;
YV12_BUFFER_CONFIG *new_fb = get_frame_new_buffer(cm);
const int tile_cols = 1 << cm->log2_tile_cols;
- int tile_col;
if (!first_partition_size) { if (!keyframe) {@@ -1132,14 +1185,7 @@
xd->mi_8x8 = cm->mi_grid_visible;
xd->mode_info_stride = cm->mode_info_stride;
- CHECK_MEM_ERROR(cm, pbi->mi_streams,
- 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);
- pbi->mi_streams[tile_col] =
- &cm->mi[cm->mi_rows * cm->cur_tile_mi_col_start];
- }
+ alloc_tile_storage(pbi, tile_cols);
cm->fc = cm->frame_contexts[cm->frame_context_idx];
--- a/vp9/decoder/vp9_onyxd_if.c
+++ b/vp9/decoder/vp9_onyxd_if.c
@@ -156,6 +156,8 @@
vp9_worker_end(&pbi->lf_worker);
vpx_free(pbi->lf_worker.data1);
vpx_free(pbi->mi_streams);
+ vpx_free(pbi->above_context[0]);
+ vpx_free(pbi->above_seg_context);
vpx_free(pbi);
}
--- a/vp9/decoder/vp9_onyxd_int.h
+++ b/vp9/decoder/vp9_onyxd_int.h
@@ -43,6 +43,9 @@
/* Each tile column has its own MODE_INFO stream. This array indexes them by
tile column index. */
MODE_INFO **mi_streams;
+
+ ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
+ PARTITION_CONTEXT *above_seg_context;
} VP9D_COMP;
#endif // VP9_DECODER_VP9_ONYXD_INT_H_
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -618,7 +618,8 @@
int pl;
const int idx = check_bsize_coverage(bs, cm->mi_rows, cm->mi_cols,
mi_row, mi_col);
- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context, cpi->left_seg_context,
+ mi_row, mi_col, bsize);
// encode the partition information
if (idx == 0)
write_token(bc, vp9_partition_tree,
@@ -661,7 +662,8 @@
// update partition context
if (bsize >= BLOCK_8X8 &&
(bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
- update_partition_context(cm, mi_row, mi_col, subsize, bsize);
+ update_partition_context(cpi->above_seg_context, cpi->left_seg_context,
+ mi_row, mi_col, subsize, bsize);
}
static void write_modes(VP9_COMP *cpi, vp9_writer* const bc,
@@ -677,7 +679,7 @@
for (mi_row = cm->cur_tile_mi_row_start; mi_row < cm->cur_tile_mi_row_end;
mi_row += 8, mi_8x8 += 8 * mis) {m_8x8 = mi_8x8;
- vp9_zero(cm->left_seg_context);
+ vp9_zero(cpi->left_seg_context);
for (mi_col = cm->cur_tile_mi_col_start; mi_col < cm->cur_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,
@@ -1201,7 +1203,7 @@
const int tile_cols = 1 << cm->log2_tile_cols;
const int tile_rows = 1 << cm->log2_tile_rows;
- vpx_memset(cm->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
+ vpx_memset(cpi->above_seg_context, 0, sizeof(*cpi->above_seg_context) *
mi_cols_aligned_to_sb(cm->mi_cols));
tok[0][0] = cpi->tok;
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -499,7 +499,7 @@
const int idx_map = mb_row * cm->mb_cols + mb_col;
const struct segmentation *const seg = &cm->seg;
- set_skip_context(cm, xd, mi_row, mi_col);
+ set_skip_context(xd, cpi->above_context, cpi->left_context, mi_row, mi_col);
// Activity map pointer
x->mb_activity_ptr = &cpi->mb_activity_map[idx_map];
@@ -711,7 +711,6 @@
ENTROPY_CONTEXT l[16 * MAX_MB_PLANE],
PARTITION_CONTEXT sa[8], PARTITION_CONTEXT sl[8],
BLOCK_SIZE bsize) {- VP9_COMMON *const cm = &cpi->common;
MACROBLOCK *const x = &cpi->mb;
MACROBLOCKD *const xd = &x->e_mbd;
int p;
@@ -721,21 +720,21 @@
int mi_height = num_8x8_blocks_high_lookup[bsize];
for (p = 0; p < MAX_MB_PLANE; p++) {vpx_memcpy(
- cm->above_context[p] + ((mi_col * 2) >> xd->plane[p].subsampling_x),
+ cpi->above_context[p] + ((mi_col * 2) >> xd->plane[p].subsampling_x),
a + num_4x4_blocks_wide * p,
(sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_wide) >>
xd->plane[p].subsampling_x);
vpx_memcpy(
- cm->left_context[p]
+ cpi->left_context[p]
+ ((mi_row & MI_MASK) * 2 >> xd->plane[p].subsampling_y),
l + num_4x4_blocks_high * p,
(sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_high) >>
xd->plane[p].subsampling_y);
}
- vpx_memcpy(cm->above_seg_context + mi_col, sa,
- sizeof(PARTITION_CONTEXT) * mi_width);
- vpx_memcpy(cm->left_seg_context + (mi_row & MI_MASK), sl,
- sizeof(PARTITION_CONTEXT) * mi_height);
+ vpx_memcpy(cpi->above_seg_context + mi_col, sa,
+ sizeof(*cpi->above_seg_context) * mi_width);
+ vpx_memcpy(cpi->left_seg_context + (mi_row & MI_MASK), sl,
+ sizeof(cpi->left_seg_context[0]) * mi_height);
}
static void save_context(VP9_COMP *cpi, int mi_row, int mi_col,
ENTROPY_CONTEXT a[16 * MAX_MB_PLANE],
@@ -742,7 +741,6 @@
ENTROPY_CONTEXT l[16 * MAX_MB_PLANE],
PARTITION_CONTEXT sa[8], PARTITION_CONTEXT sl[8],
BLOCK_SIZE bsize) {- const VP9_COMMON *const cm = &cpi->common;
const MACROBLOCK *const x = &cpi->mb;
const MACROBLOCKD *const xd = &x->e_mbd;
int p;
@@ -755,20 +753,20 @@
for (p = 0; p < MAX_MB_PLANE; ++p) {vpx_memcpy(
a + num_4x4_blocks_wide * p,
- cm->above_context[p] + (mi_col * 2 >> xd->plane[p].subsampling_x),
+ cpi->above_context[p] + (mi_col * 2 >> xd->plane[p].subsampling_x),
(sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_wide) >>
xd->plane[p].subsampling_x);
vpx_memcpy(
l + num_4x4_blocks_high * p,
- cm->left_context[p]
+ cpi->left_context[p]
+ ((mi_row & MI_MASK) * 2 >> xd->plane[p].subsampling_y),
(sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_high) >>
xd->plane[p].subsampling_y);
}
- vpx_memcpy(sa, cm->above_seg_context + mi_col,
- sizeof(PARTITION_CONTEXT) * mi_width);
- vpx_memcpy(sl, cm->left_seg_context + (mi_row & MI_MASK),
- sizeof(PARTITION_CONTEXT) * mi_height);
+ vpx_memcpy(sa, cpi->above_seg_context + mi_col,
+ sizeof(*cpi->above_seg_context) * mi_width);
+ vpx_memcpy(sl, cpi->left_seg_context + (mi_row & MI_MASK),
+ sizeof(cpi->left_seg_context[0]) * mi_height);
}
static void encode_b(VP9_COMP *cpi, TOKENEXTRA **tp, int mi_row, int mi_col,
@@ -818,7 +816,8 @@
c1 = BLOCK_4X4;
if (bsize >= BLOCK_8X8) {- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context, cpi->left_seg_context,
+ mi_row, mi_col, bsize);
c1 = *(get_sb_partitioning(x, bsize));
}
partition = partition_lookup[bsl][c1];
@@ -861,7 +860,8 @@
}
if (partition != PARTITION_SPLIT || bsize == BLOCK_8X8)
- update_partition_context(cm, mi_row, mi_col, c1, bsize);
+ update_partition_context(cpi->above_seg_context, cpi->left_seg_context,
+ mi_row, mi_col, c1, bsize);
}
// Check to see if the given partition size is allowed for a specified number
@@ -1052,7 +1052,9 @@
pick_sb_modes(cpi, mi_row, mi_col, &none_rate, &none_dist, bsize,
get_block_context(x, bsize), INT64_MAX);
- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context,
+ cpi->left_seg_context,
+ mi_row, mi_col, bsize);
none_rate += x->partition_cost[pl][PARTITION_NONE];
restore_context(cpi, mi_row, mi_col, a, l, sa, sl, bsize);
@@ -1143,7 +1145,8 @@
assert(0);
}
- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context, cpi->left_seg_context,
+ mi_row, mi_col, bsize);
if (last_part_rate < INT_MAX)
last_part_rate += x->partition_cost[pl][partition];
@@ -1193,10 +1196,13 @@
split_rate += rt;
split_dist += dt;
- pl = partition_plane_context(cm, mi_row + y_idx, mi_col + x_idx, bsize);
+ pl = partition_plane_context(cpi->above_seg_context,
+ cpi->left_seg_context,
+ mi_row + y_idx, mi_col + x_idx, bsize);
split_rate += x->partition_cost[pl][PARTITION_NONE];
}
- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context, cpi->left_seg_context,
+ mi_row, mi_col, bsize);
if (split_rate < INT_MAX) {split_rate += x->partition_cost[pl][PARTITION_SPLIT];
@@ -1525,7 +1531,9 @@
get_block_context(x, bsize), best_rd);
if (this_rate != INT_MAX) { if (bsize >= BLOCK_8X8) {- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context,
+ cpi->left_seg_context,
+ mi_row, mi_col, bsize);
this_rate += x->partition_cost[pl][PARTITION_NONE];
}
sum_rd = RDCOST(x->rdmult, x->rddiv, this_rate, this_dist);
@@ -1585,7 +1593,9 @@
}
}
if (sum_rd < best_rd && i == 4) {- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context,
+ cpi->left_seg_context,
+ mi_row, mi_col, bsize);
sum_rate += x->partition_cost[pl][PARTITION_SPLIT];
sum_rd = RDCOST(x->rdmult, x->rddiv, sum_rate, sum_dist);
if (sum_rd < best_rd) {@@ -1641,7 +1651,9 @@
}
}
if (sum_rd < best_rd) {- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context,
+ cpi->left_seg_context,
+ mi_row, mi_col, bsize);
sum_rate += x->partition_cost[pl][PARTITION_HORZ];
sum_rd = RDCOST(x->rdmult, x->rddiv, sum_rate, sum_dist);
if (sum_rd < best_rd) {@@ -1683,7 +1695,9 @@
}
}
if (sum_rd < best_rd) {- pl = partition_plane_context(cm, mi_row, mi_col, bsize);
+ pl = partition_plane_context(cpi->above_seg_context,
+ cpi->left_seg_context,
+ mi_row, mi_col, bsize);
sum_rate += x->partition_cost[pl][PARTITION_VERT];
sum_rd = RDCOST(x->rdmult, x->rddiv, sum_rate, sum_dist);
if (sum_rd < best_rd) {@@ -1734,7 +1748,8 @@
cpi->set_ref_frame_mask = 1;
pick_sb_modes(cpi, mi_row, mi_col, &r, &d, BLOCK_64X64,
get_block_context(x, BLOCK_64X64), INT64_MAX);
- pl = partition_plane_context(cm, mi_row, mi_col, BLOCK_64X64);
+ pl = partition_plane_context(cpi->above_seg_context, cpi->left_seg_context,
+ mi_row, mi_col, BLOCK_64X64);
r += x->partition_cost[pl][PARTITION_NONE];
*(get_sb_partitioning(x, BLOCK_64X64)) = BLOCK_64X64;
@@ -1750,8 +1765,8 @@
int mi_col;
// Initialize the left context for the new SB row
- vpx_memset(&cm->left_context, 0, sizeof(cm->left_context));
- vpx_memset(cm->left_seg_context, 0, sizeof(cm->left_seg_context));
+ vpx_memset(&cpi->left_context, 0, sizeof(cpi->left_context));
+ 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;
@@ -1856,10 +1871,11 @@
// Note: this memset assumes above_context[0], [1] and [2]
// are allocated as part of the same buffer.
- vpx_memset(cm->above_context[0], 0,
- sizeof(ENTROPY_CONTEXT) * 2 * MAX_MB_PLANE * aligned_mi_cols);
- vpx_memset(cm->above_seg_context, 0,
- sizeof(PARTITION_CONTEXT) * aligned_mi_cols);
+ vpx_memset(cpi->above_context[0], 0,
+ sizeof(*cpi->above_context[0]) *
+ 2 * aligned_mi_cols * MAX_MB_PLANE);
+ vpx_memset(cpi->above_seg_context, 0,
+ sizeof(*cpi->above_seg_context) * aligned_mi_cols);
}
static void switch_lossless_mode(VP9_COMP *cpi, int lossless) {--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -312,6 +312,12 @@
cpi->mb_activity_map = 0;
vpx_free(cpi->mb_norm_activity_map);
cpi->mb_norm_activity_map = 0;
+
+ vpx_free(cpi->above_context[0]);
+ cpi->above_context[0] = NULL;
+
+ vpx_free(cpi->above_seg_context);
+ cpi->above_seg_context = NULL;
}
// Computes a q delta (in "q index" terms) to get from a starting q value
@@ -1040,6 +1046,19 @@
CHECK_MEM_ERROR(cm, cpi->mb_norm_activity_map,
vpx_calloc(sizeof(unsigned int),
cm->mb_rows * cm->mb_cols));
+
+ // 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
+ // block where mi unit size is 8x8.
+ vpx_free(cpi->above_context[0]);
+ CHECK_MEM_ERROR(cm, cpi->above_context[0],
+ vpx_calloc(2 * mi_cols_aligned_to_sb(cm->mi_cols) *
+ MAX_MB_PLANE,
+ sizeof(*cpi->above_context[0])));
+
+ vpx_free(cpi->above_seg_context);
+ CHECK_MEM_ERROR(cm, cpi->above_seg_context,
+ vpx_calloc(mi_cols_aligned_to_sb(cm->mi_cols),
+ sizeof(*cpi->above_seg_context)));
}
@@ -1070,6 +1089,15 @@
vp9_init3smotion_compensation(&cpi->mb, y_stride);
} else if (cpi->sf.search_method == DIAMOND) {vp9_init_dsmotion_compensation(&cpi->mb, y_stride);
+ }
+ }
+
+ {+ int i;
+ for (i = 1; i < MAX_MB_PLANE; ++i) {+ cpi->above_context[i] = cpi->above_context[0] +
+ i * sizeof(*cpi->above_context[0]) * 2 *
+ mi_cols_aligned_to_sb(cm->mi_cols);
}
}
}
--- a/vp9/encoder/vp9_onyx_int.h
+++ b/vp9/encoder/vp9_onyx_int.h
@@ -675,6 +675,13 @@
// Debug / test stats
int64_t mode_test_hits[BLOCK_SIZES];
#endif
+
+ /* Y,U,V,(A) */
+ ENTROPY_CONTEXT *above_context[MAX_MB_PLANE];
+ ENTROPY_CONTEXT left_context[MAX_MB_PLANE][16];
+
+ PARTITION_CONTEXT *above_seg_context;
+ PARTITION_CONTEXT left_seg_context[8];
} VP9_COMP;
static int get_ref_frame_idx(VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {