Extend left/above partition context to per mi(8x8)

Update and buffer left/above partition information context per 8x8
block. This allows to further enable recursive partition down to
4x4 block size, and hence deprecating I4X4_PRED and SPLITMV.

This commit also fixes a context buffer swap/restore issue in 32x32
partition type search. This gives 0.1% performance gain for derf/yt.
Will refactor the superblock partition type search into recursion
form.

Change-Id: Ib61975aca5f12b78d8018481d7fa1393d085689b

diff --git a/vp9/common/vp9_alloccommon.c b/vp9/common/vp9_alloccommon.c
index 8179a69..fd39439 100644 (file)
--- a/vp9/common/vp9_alloccommon.c
+++ b/vp9/common/vp9_alloccommon.c
@@ -70,7 +70,7 @@ void vp9_free_frame_buffers(VP9_COMMON *oci) {
 }
 
 int vp9_alloc_frame_buffers(VP9_COMMON *oci, int width, int height) {
-  int i, mb_cols;
+  int i, mi_cols;
 
   // Our internal buffers are always multiples of 16
   const int aligned_width = multiple16(width);
@@ -140,19 +140,19 @@ int vp9_alloc_frame_buffers(VP9_COMMON *oci, int width, int height) {
 
   // FIXME(jkoleszar): allocate subsampled arrays for U/V once subsampling
   // information is exposed at this level
-  mb_cols = mb_cols_aligned_to_sb(oci);
-  oci->above_context[0] = vpx_calloc(sizeof(ENTROPY_CONTEXT) * 12 * mb_cols, 1);
+  mi_cols = mi_cols_aligned_to_sb(oci);
+  oci->above_context[0] = vpx_calloc(sizeof(ENTROPY_CONTEXT) * 6 * mi_cols, 1);
   if (!oci->above_context[0]) {
     vp9_free_frame_buffers(oci);
     return 1;
   }
   oci->above_context[1] =
-    oci->above_context[0] + sizeof(ENTROPY_CONTEXT) * 4 * mb_cols;
+    oci->above_context[0] + sizeof(ENTROPY_CONTEXT) * 2 * mi_cols;
   oci->above_context[2] =
-    oci->above_context[1] + sizeof(ENTROPY_CONTEXT) * 4 * mb_cols;
+    oci->above_context[1] + sizeof(ENTROPY_CONTEXT) * 2 * mi_cols;
 
   oci->above_seg_context =
-    vpx_calloc(sizeof(PARTITION_CONTEXT) * mb_cols_aligned_to_sb(oci), 1);
+    vpx_calloc(sizeof(PARTITION_CONTEXT) * mi_cols, 1);
 
   if (!oci->above_seg_context) {
     vp9_free_frame_buffers(oci);

diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index af8e662..e452684 100644 (file)
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -413,7 +413,7 @@ typedef struct macroblockd {
 static INLINE void update_partition_context(MACROBLOCKD *xd,
                                             BLOCK_SIZE_TYPE sb_type,
                                             BLOCK_SIZE_TYPE sb_size) {
-  int bsl = mi_width_log2(sb_size), bs;
+  int bsl = mi_width_log2(sb_size), bs = 1 << bsl;
   int bwl = mi_width_log2(sb_type);
   int bhl = mi_height_log2(sb_type);
   int boffset = mi_width_log2(BLOCK_SIZE_SB64X64) - bsl;
@@ -422,8 +422,6 @@ static INLINE void update_partition_context(MACROBLOCKD *xd,
   if (bsl == 0)
     return;
 
-  bs = 1 << (bsl - 1);
-
   // update the partition context at the end notes. set partition bits
   // of block sizes larger than the current one to be one, and partition
   // bits of smaller block sizes to be zero.
@@ -454,18 +452,14 @@ static INLINE void update_partition_context(MACROBLOCKD *xd,
 
 static INLINE int partition_plane_context(MACROBLOCKD *xd,
                                           BLOCK_SIZE_TYPE sb_type) {
-  int bsl = mi_width_log2(sb_type), bs;
+  int bsl = mi_width_log2(sb_type), bs = 1 << bsl;
   int above = 0, left = 0, i;
   int boffset = mi_width_log2(BLOCK_SIZE_SB64X64) - bsl;
 
-  bs = 1 << (bsl - 1);
-
   assert(mi_width_log2(sb_type) == mi_height_log2(sb_type));
   assert(bsl >= 0);
   assert(boffset >= 0);
 
-  bs = 1 << (bsl - 1);
-
   for (i = 0; i < bs; i++)
     above |= (xd->above_seg_context[i] & (1 << boffset));
   for (i = 0; i < bs; i++)

diff --git a/vp9/common/vp9_enums.h b/vp9/common/vp9_enums.h
index b00d892..1663195 100644 (file)
--- a/vp9/common/vp9_enums.h
+++ b/vp9/common/vp9_enums.h
@@ -18,6 +18,8 @@
 #define MI_SIZE (1 << LOG2_MI_SIZE)
 #define MI_UV_SIZE (1 << (LOG2_MI_SIZE - 1))
 
+#define MI_MASK ((64 >> LOG2_MI_SIZE) - 1)
+
 typedef enum BLOCK_SIZE_TYPE {
   BLOCK_SIZE_AB4X4,
 #if CONFIG_AB4X4

diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index d9d2989..de2cace 100644 (file)
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -217,7 +217,7 @@ typedef struct VP9Common {
 
   // partition contexts
   PARTITION_CONTEXT *above_seg_context;
-  PARTITION_CONTEXT left_seg_context[4];
+  PARTITION_CONTEXT left_seg_context[8];
 
   /* keyframe block modes are predicted by their above, left neighbors */
 
@@ -297,8 +297,8 @@ static void ref_cnt_fb(int *buf, int *idx, int new_idx) {
   buf[new_idx]++;
 }
 
-static int mb_cols_aligned_to_sb(VP9_COMMON *cm) {
-  return (cm->mb_cols + 3) & ~3;
+static int mi_cols_aligned_to_sb(VP9_COMMON *cm) {
+  return 2 * ((cm->mb_cols + 3) & ~3);
 }
 
 static void set_mi_row_col(VP9_COMMON *cm, MACROBLOCKD *xd,

diff --git a/vp9/decoder/vp9_decodframe.c b/vp9/decoder/vp9_decodframe.c
index 3007863..f9c2a51 100644 (file)
--- a/vp9/decoder/vp9_decodframe.c
+++ b/vp9/decoder/vp9_decodframe.c
@@ -372,8 +372,8 @@ static void set_offsets(VP9D_COMP *pbi, BLOCK_SIZE_TYPE bsize,
     xd->plane[i].left_context = cm->left_context[i] +
         (((mi_row * 2) & 15) >> xd->plane[i].subsampling_y);
   }
-  xd->above_seg_context = cm->above_seg_context + (mi_col >> 1);
-  xd->left_seg_context  = cm->left_seg_context + ((mi_row >> 1) & 3);
+  xd->above_seg_context = cm->above_seg_context + mi_col;
+  xd->left_seg_context  = cm->left_seg_context + (mi_row & MI_MASK);
 
   // 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
@@ -443,9 +443,8 @@ static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
   if (bsize > BLOCK_SIZE_SB8X8) {
     int pl;
     // read the partition information
-    xd->left_seg_context =
-        pc->left_seg_context + ((mi_row >> 1) & 3);
-    xd->above_seg_context = pc->above_seg_context + (mi_col >> 1);
+    xd->left_seg_context = pc->left_seg_context + (mi_row & MI_MASK);
+    xd->above_seg_context = pc->above_seg_context + mi_col;
     pl = partition_plane_context(xd, bsize);
     partition = treed_read(r, vp9_partition_tree,
                            pc->fc.partition_prob[pl]);
@@ -486,8 +485,8 @@ static void decode_modes_sb(VP9D_COMP *pbi, int mi_row, int mi_col,
   if ((partition == PARTITION_SPLIT) && (bsize > BLOCK_SIZE_MB16X16))
     return;
 
-  xd->left_seg_context = pc->left_seg_context + ((mi_row >> 1) & 3);
-  xd->above_seg_context = pc->above_seg_context + (mi_col >> 1);
+  xd->left_seg_context = pc->left_seg_context + (mi_row & MI_MASK);
+  xd->above_seg_context = pc->above_seg_context + mi_col;
   update_partition_context(xd, subsize, bsize);
 }
 
@@ -849,11 +848,11 @@ static void decode_tiles(VP9D_COMP *pbi,
 
   // Note: this memset assumes above_context[0], [1] and [2]
   // are allocated as part of the same buffer.
-  vpx_memset(pc->above_context[0], 0, sizeof(ENTROPY_CONTEXT) * 4 *
-                                      MAX_MB_PLANE * mb_cols_aligned_to_sb(pc));
+  vpx_memset(pc->above_context[0], 0, sizeof(ENTROPY_CONTEXT) * 2 *
+                                      MAX_MB_PLANE * mi_cols_aligned_to_sb(pc));
 
   vpx_memset(pc->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
-                                       mb_cols_aligned_to_sb(pc));
+                                       mi_cols_aligned_to_sb(pc));
 
   if (pbi->oxcf.inv_tile_order) {
     const int n_cols = pc->tile_columns;

diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index a7c26a4..050821d 100644 (file)
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -878,9 +878,8 @@ static void write_modes_sb(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
 
   if (bsize > BLOCK_SIZE_SB8X8) {
     int pl;
-    xd->left_seg_context =
-        cm->left_seg_context + ((mi_row >> 1) & 3);
-    xd->above_seg_context = cm->above_seg_context + (mi_col >> 1);
+    xd->left_seg_context = cm->left_seg_context + (mi_row & MI_MASK);
+    xd->above_seg_context = cm->above_seg_context + mi_col;
     pl = partition_plane_context(xd, bsize);
     // encode the partition information
     write_token(bc, vp9_partition_tree, cm->fc.partition_prob[pl],
@@ -918,8 +917,8 @@ static void write_modes_sb(VP9_COMP *cpi, MODE_INFO *m, vp9_writer *bc,
   if ((partition == PARTITION_SPLIT) && (bsize > BLOCK_SIZE_MB16X16))
     return;
 
-  xd->left_seg_context = cm->left_seg_context + ((mi_row >> 1) & 3);
-  xd->above_seg_context = cm->above_seg_context + (mi_col >> 1);
+  xd->left_seg_context = cm->left_seg_context + (mi_row & MI_MASK);
+  xd->above_seg_context = cm->above_seg_context + mi_col;
   update_partition_context(xd, subsize, bsize);
 }
 
@@ -932,7 +931,7 @@ static void write_modes(VP9_COMP *cpi, vp9_writer* const bc,
 
   m_ptr += c->cur_tile_mi_col_start + c->cur_tile_mi_row_start * mis;
   vpx_memset(c->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
-             mb_cols_aligned_to_sb(c));
+             mi_cols_aligned_to_sb(c));
 
   for (mi_row = c->cur_tile_mi_row_start;
        mi_row < c->cur_tile_mi_row_end;

diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index 8ddad26..49e8cce 100644 (file)
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -544,8 +544,8 @@ static INLINE void set_partition_seg_context(VP9_COMP *cpi,
   VP9_COMMON *const cm = &cpi->common;
   MACROBLOCKD *const xd = &cpi->mb.e_mbd;
 
-  xd->above_seg_context = cm->above_seg_context + (mi_col >> 1);
-  xd->left_seg_context  = cm->left_seg_context + ((mi_row >> 1) & 3);
+  xd->above_seg_context = cm->above_seg_context + mi_col;
+  xd->left_seg_context  = cm->left_seg_context + (mi_row & MI_MASK);
 }
 
 static void set_offsets(VP9_COMP *cpi,
@@ -878,7 +878,7 @@ static void encode_sb_row(VP9_COMP *cpi,
     int sb64_rate = 0, sb64_dist = 0;
     int sb64_skip = 0;
     ENTROPY_CONTEXT l[16 * MAX_MB_PLANE], a[16 * MAX_MB_PLANE];
-    PARTITION_CONTEXT seg_l[4], seg_a[4];
+    PARTITION_CONTEXT seg_l[64 / MI_SIZE], seg_a[64 / MI_SIZE];
     TOKENEXTRA *tp_orig = *tp;
 
     for (p = 0; p < MAX_MB_PLANE; p++) {
@@ -888,9 +888,8 @@ static void encode_sb_row(VP9_COMP *cpi,
       memcpy(l + 16 * p, cm->left_context[p],
              sizeof(ENTROPY_CONTEXT) * 16 >> xd->plane[p].subsampling_y);
     }
-    memcpy(&seg_a, cm->above_seg_context + (mi_col >> 1),
-           sizeof(seg_a));
-    memcpy(&seg_l, cm->left_seg_context, sizeof(seg_l));
+    vpx_memcpy(&seg_a, cm->above_seg_context + mi_col, sizeof(seg_a));
+    vpx_memcpy(&seg_l, cm->left_seg_context, sizeof(seg_l));
 
     // FIXME(rbultje): this function should probably be rewritten to be
     // recursive at some point in the future.
@@ -902,6 +901,7 @@ static void encode_sb_row(VP9_COMP *cpi,
       int sb32_skip = 0;
       int j;
       ENTROPY_CONTEXT l2[8 * MAX_MB_PLANE], a2[8 * MAX_MB_PLANE];
+      PARTITION_CONTEXT sl32[32 / MI_SIZE], sa32[32 / MI_SIZE];
 
       sb_partitioning[i] = BLOCK_SIZE_MB16X16;
       if (mi_row + y_idx >= cm->mi_rows || mi_col + x_idx >= cm->mi_cols)
@@ -920,6 +920,8 @@ static void encode_sb_row(VP9_COMP *cpi,
                        ((mi_col + x_idx) * 2 >> xd->plane[p].subsampling_x),
                    sizeof(ENTROPY_CONTEXT) * 8 >> xd->plane[p].subsampling_x);
       }
+      vpx_memcpy(&sa32, cm->above_seg_context + mi_col + x_idx, sizeof(sa32));
+      vpx_memcpy(&sl32, cm->left_seg_context + y_idx, sizeof(sl32));
 
       /* Encode MBs in raster order within the SB */
       for (j = 0; j < 4; j++) {
@@ -928,6 +930,7 @@ static void encode_sb_row(VP9_COMP *cpi,
         int r, d;
         int r2, d2, mb16_rate = 0, mb16_dist = 0, k;
         ENTROPY_CONTEXT l3[4 * MAX_MB_PLANE], a3[4 * MAX_MB_PLANE];
+        PARTITION_CONTEXT sl16[16 / MI_SIZE], sa16[16 / MI_SIZE];
 
         mb_partitioning[i][j] = BLOCK_SIZE_SB8X8;
 
@@ -950,6 +953,9 @@ static void encode_sb_row(VP9_COMP *cpi,
                          ((mi_col + x_idx_m) * 2 >> xd->plane[p].subsampling_x),
                      sizeof(ENTROPY_CONTEXT) * 4 >> xd->plane[p].subsampling_x);
         }
+        vpx_memcpy(&sa16, cm->above_seg_context + mi_col + x_idx_m,
+                   sizeof(sa16));
+        vpx_memcpy(&sl16, cm->left_seg_context + y_idx_m, sizeof(sl16));
 
         for (k = 0; k < 4; k++) {
           xd->b_index = k;
@@ -983,6 +989,9 @@ static void encode_sb_row(VP9_COMP *cpi,
                      a3 + 4 * p,
                      sizeof(ENTROPY_CONTEXT) * 4 >> xd->plane[p].subsampling_x);
         }
+        vpx_memcpy(cm->above_seg_context + mi_col + x_idx_m,
+                   sa16, sizeof(sa16));
+        vpx_memcpy(cm->left_seg_context + y_idx_m, sl16, sizeof(sl16));
 
         // try 8x16 coding
         r2 = 0;
@@ -1102,6 +1111,9 @@ static void encode_sb_row(VP9_COMP *cpi,
                    a2 + 8 * p,
                    sizeof(ENTROPY_CONTEXT) * 8 >> xd->plane[p].subsampling_x);
       }
+      // restore partition information context
+      vpx_memcpy(cm->above_seg_context + mi_col + x_idx, sa32, sizeof(sa32));
+      vpx_memcpy(cm->left_seg_context + y_idx, sl32, sizeof(sl32));
 
       set_partition_seg_context(cpi, mi_row + y_idx, mi_col + x_idx);
       pl = partition_plane_context(xd, BLOCK_SIZE_SB32X32);
@@ -1258,8 +1270,7 @@ static void encode_sb_row(VP9_COMP *cpi,
       memcpy(cm->left_context[p], l + 16 * p,
              sizeof(ENTROPY_CONTEXT) * 16 >> xd->plane[p].subsampling_y);
     }
-    memcpy(cm->above_seg_context + (mi_col >> 1), &seg_a,
-           sizeof(seg_a));
+    memcpy(cm->above_seg_context + mi_col, &seg_a, sizeof(seg_a));
     memcpy(cm->left_seg_context, &seg_l, sizeof(seg_l));
 
     set_partition_seg_context(cpi, mi_row, mi_col);
@@ -1428,10 +1439,10 @@ static void init_encode_frame_mb_context(VP9_COMP *cpi) {
 
   // 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) * 4 *
-                                      MAX_MB_PLANE * mb_cols_aligned_to_sb(cm));
+  vpx_memset(cm->above_context[0], 0, sizeof(ENTROPY_CONTEXT) * 2 *
+                                      MAX_MB_PLANE * mi_cols_aligned_to_sb(cm));
   vpx_memset(cm->above_seg_context, 0, sizeof(PARTITION_CONTEXT) *
-                                       mb_cols_aligned_to_sb(cm));
+                                       mi_cols_aligned_to_sb(cm));
 }
 
 static void switch_lossless_mode(VP9_COMP *cpi, int lossless) {