Optimize PC_TREE to store only winner mode info
PC_TREE is modified to store only the winner reference
mode information. Size of ref_mv_stack, weight,
mode_context and ref_mv_count is reduced in PC_TREE.
Observed memory footprint reduction with similar
encoding time.
Resolution Tile Memory reduction
Single Multi
Thread Thread
640x360 2x1 ~16% ~25% (2 threads)
832x480 2x1 ~13% ~21% (2 threads)
1280x720 2x2 ~10% ~25% (4 threads)
1920x1080 4x2 ~5% ~25% (8 threads)
Memory measuring command:
$ command time -v ./aomenc ...
Change-Id: I5675d289b89b6deabd11274f4210dd81c058b44c
diff --git a/av1/encoder/context_tree.c b/av1/encoder/context_tree.c
index 00f7509..ab245d5 100644
--- a/av1/encoder/context_tree.c
+++ b/av1/encoder/context_tree.c
@@ -223,7 +223,7 @@
void av1_copy_tree_context(PICK_MODE_CONTEXT *dst_ctx,
PICK_MODE_CONTEXT *src_ctx) {
dst_ctx->mic = src_ctx->mic;
- dst_ctx->mbmi_ext = src_ctx->mbmi_ext;
+ dst_ctx->mbmi_ext_best = src_ctx->mbmi_ext_best;
dst_ctx->num_4x4_blk = src_ctx->num_4x4_blk;
dst_ctx->skippable = src_ctx->skippable;
diff --git a/av1/encoder/context_tree.h b/av1/encoder/context_tree.h
index 24f002a..d5b4e83 100644
--- a/av1/encoder/context_tree.h
+++ b/av1/encoder/context_tree.h
@@ -28,7 +28,7 @@
// Structure to hold snapshot of coding context during the mode picking process
typedef struct {
MB_MODE_INFO mic;
- MB_MODE_INFO_EXT mbmi_ext;
+ MB_MODE_INFO_EXT_FRAME mbmi_ext_best;
uint8_t *color_index_map[2];
uint8_t *blk_skip;
diff --git a/av1/encoder/encodeframe.c b/av1/encoder/encodeframe.c
index ee132e6..3e5f966 100644
--- a/av1/encoder/encodeframe.c
+++ b/av1/encoder/encodeframe.c
@@ -507,6 +507,21 @@
x->force_skip = 0;
}
+// This function will copy the best reference mode information from
+// MB_MODE_INFO_EXT_FRAME to MB_MODE_INFO_EXT.
+static INLINE void copy_mbmi_ext_frame_to_mbmi_ext(
+ MB_MODE_INFO_EXT *mbmi_ext,
+ const MB_MODE_INFO_EXT_FRAME *const mbmi_ext_best, uint8_t ref_frame_type) {
+ memcpy(mbmi_ext->ref_mv_stack[ref_frame_type], mbmi_ext_best->ref_mv_stack,
+ sizeof(mbmi_ext->ref_mv_stack[USABLE_REF_MV_STACK_SIZE]));
+ memcpy(mbmi_ext->weight[ref_frame_type], mbmi_ext_best->weight,
+ sizeof(mbmi_ext->weight[USABLE_REF_MV_STACK_SIZE]));
+ mbmi_ext->mode_context[ref_frame_type] = mbmi_ext_best->mode_context;
+ mbmi_ext->ref_mv_count[ref_frame_type] = mbmi_ext_best->ref_mv_count;
+ memcpy(mbmi_ext->global_mvs, mbmi_ext_best->global_mvs,
+ sizeof(mbmi_ext->global_mvs));
+}
+
static AOM_INLINE void update_state(const AV1_COMP *const cpi, ThreadData *td,
const PICK_MODE_CONTEXT *const ctx,
int mi_row, int mi_col, BLOCK_SIZE bsize,
@@ -532,7 +547,8 @@
assert(mi->sb_type == bsize);
*mi_addr = *mi;
- *x->mbmi_ext = ctx->mbmi_ext;
+ copy_mbmi_ext_frame_to_mbmi_ext(x->mbmi_ext, &ctx->mbmi_ext_best,
+ av1_ref_frame_type(ctx->mic.ref_frame));
memcpy(x->blk_skip, ctx->blk_skip, sizeof(x->blk_skip[0]) * ctx->num_4x4_blk);
@@ -687,25 +703,6 @@
return 0 && !frame_is_intra_only(&cpi->common);
}
-// This function will copy the winner reference mode information from block
-// level (x->mbmi_ext) to frame level (cpi->mbmi_ext_frame_base). This frame
-// level buffer (cpi->mbmi_ext_frame_base) will be used during bitstream
-// preparation.
-static INLINE void copy_winner_ref_mode_from_mbmi_ext(MACROBLOCK *const x) {
- MACROBLOCKD *const xd = &x->e_mbd;
- MB_MODE_INFO *mbmi = xd->mi[0];
- uint8_t ref_frame_type = av1_ref_frame_type(mbmi->ref_frame);
- memcpy(x->mbmi_ext_frame->ref_mv_stack,
- x->mbmi_ext->ref_mv_stack[ref_frame_type],
- sizeof(x->mbmi_ext->ref_mv_stack[USABLE_REF_MV_STACK_SIZE]));
- memcpy(x->mbmi_ext_frame->weight, x->mbmi_ext->weight[ref_frame_type],
- sizeof(x->mbmi_ext->weight[USABLE_REF_MV_STACK_SIZE]));
- x->mbmi_ext_frame->mode_context = x->mbmi_ext->mode_context[ref_frame_type];
- x->mbmi_ext_frame->ref_mv_count = x->mbmi_ext->ref_mv_count[ref_frame_type];
- memcpy(x->mbmi_ext_frame->global_mvs, x->mbmi_ext->global_mvs,
- sizeof(x->mbmi_ext->global_mvs));
-}
-
static void hybrid_intra_mode_search(AV1_COMP *cpi, MACROBLOCK *const x,
RD_STATS *rd_cost, BLOCK_SIZE bsize,
PICK_MODE_CONTEXT *ctx) {
@@ -1677,7 +1674,12 @@
}
}
// TODO(Ravi/Remya): Move this copy function to a better logical place
- copy_winner_ref_mode_from_mbmi_ext(x);
+ // This function will copy the winner reference mode information from block
+ // level (x->mbmi_ext) to frame level (cpi->mbmi_ext_frame_base). This frame
+ // level buffer (cpi->mbmi_ext_frame_base) will be used during bitstream
+ // preparation.
+ av1_copy_mbmi_ext_to_mbmi_ext_frame(x->mbmi_ext_frame, x->mbmi_ext,
+ av1_ref_frame_type(xd->mi[0]->ref_frame));
x->rdmult = origin_mult;
}
diff --git a/av1/encoder/nonrd_pickmode.c b/av1/encoder/nonrd_pickmode.c
index 4e81806..902420d 100644
--- a/av1/encoder/nonrd_pickmode.c
+++ b/av1/encoder/nonrd_pickmode.c
@@ -940,7 +940,8 @@
#endif // CONFIG_INTERNAL_STATS
ctx->mic = *xd->mi[0];
ctx->skippable = x->force_skip;
- ctx->mbmi_ext = *x->mbmi_ext;
+ av1_copy_mbmi_ext_to_mbmi_ext_frame(&ctx->mbmi_ext_best, x->mbmi_ext,
+ av1_ref_frame_type(xd->mi[0]->ref_frame));
ctx->comp_pred_diff = 0;
ctx->hybrid_pred_diff = 0;
ctx->single_pred_diff = 0;
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index b381d95..8022137 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -905,7 +905,8 @@
ctx->best_mode_index = mode_index;
#endif // CONFIG_INTERNAL_STATS
ctx->mic = *xd->mi[0];
- ctx->mbmi_ext = *x->mbmi_ext;
+ av1_copy_mbmi_ext_to_mbmi_ext_frame(&ctx->mbmi_ext_best, x->mbmi_ext,
+ av1_ref_frame_type(xd->mi[0]->ref_frame));
ctx->single_pred_diff = (int)comp_pred_diff[SINGLE_REFERENCE];
ctx->comp_pred_diff = (int)comp_pred_diff[COMPOUND_REFERENCE];
ctx->hybrid_pred_diff = (int)comp_pred_diff[REFERENCE_MODE_SELECT];
@@ -2880,7 +2881,8 @@
if (rd_cost->rate == INT_MAX) return;
ctx->mic = *xd->mi[0];
- ctx->mbmi_ext = *x->mbmi_ext;
+ av1_copy_mbmi_ext_to_mbmi_ext_frame(&ctx->mbmi_ext_best, x->mbmi_ext,
+ av1_ref_frame_type(xd->mi[0]->ref_frame));
av1_copy_array(ctx->tx_type_map, xd->tx_type_map, ctx->num_4x4_blk);
}
diff --git a/av1/encoder/rdopt.h b/av1/encoder/rdopt.h
index 821c485..c7c99ac 100644
--- a/av1/encoder/rdopt.h
+++ b/av1/encoder/rdopt.h
@@ -221,6 +221,22 @@
return 0;
}
+
+// This function will copy the best reference mode information from
+// MB_MODE_INFO_EXT to MB_MODE_INFO_EXT_FRAME.
+static INLINE void av1_copy_mbmi_ext_to_mbmi_ext_frame(
+ MB_MODE_INFO_EXT_FRAME *mbmi_ext_best,
+ const MB_MODE_INFO_EXT *const mbmi_ext, uint8_t ref_frame_type) {
+ memcpy(mbmi_ext_best->ref_mv_stack, mbmi_ext->ref_mv_stack[ref_frame_type],
+ sizeof(mbmi_ext->ref_mv_stack[USABLE_REF_MV_STACK_SIZE]));
+ memcpy(mbmi_ext_best->weight, mbmi_ext->weight[ref_frame_type],
+ sizeof(mbmi_ext->weight[USABLE_REF_MV_STACK_SIZE]));
+ mbmi_ext_best->mode_context = mbmi_ext->mode_context[ref_frame_type];
+ mbmi_ext_best->ref_mv_count = mbmi_ext->ref_mv_count[ref_frame_type];
+ memcpy(mbmi_ext_best->global_mvs, mbmi_ext->global_mvs,
+ sizeof(mbmi_ext->global_mvs));
+}
+
#ifdef __cplusplus
} // extern "C"
#endif
