Merge "Inlining set_partition_seg_context function."
diff --git a/vp9/encoder/vp9_block.h b/vp9/encoder/vp9_block.h
index fe76804..e40a609 100644
--- a/vp9/encoder/vp9_block.h
+++ b/vp9/encoder/vp9_block.h
@@ -26,7 +26,8 @@
// Structure to hold snapshot of coding context during the mode picking process
typedef struct {
MODE_INFO mic;
- uint8_t zcoeff_blk[256];
+ uint8_t *zcoeff_blk;
+ int num_4x4_blk;
int skip;
int_mv best_ref_mv;
int_mv second_best_ref_mv;
@@ -177,6 +178,45 @@
int y_blocks);
};
+// TODO(jingning): the variables used here are little complicated. need further
+// refactoring on organizing the temporary buffers, when recursive
+// partition down to 4x4 block size is enabled.
+static PICK_MODE_CONTEXT *get_block_context(MACROBLOCK *x, BLOCK_SIZE bsize) {
+ MACROBLOCKD *const xd = &x->e_mbd;
+
+ switch (bsize) {
+ case BLOCK_64X64:
+ return &x->sb64_context;
+ case BLOCK_64X32:
+ return &x->sb64x32_context[xd->sb_index];
+ case BLOCK_32X64:
+ return &x->sb32x64_context[xd->sb_index];
+ case BLOCK_32X32:
+ return &x->sb32_context[xd->sb_index];
+ case BLOCK_32X16:
+ return &x->sb32x16_context[xd->sb_index][xd->mb_index];
+ case BLOCK_16X32:
+ return &x->sb16x32_context[xd->sb_index][xd->mb_index];
+ case BLOCK_16X16:
+ return &x->mb_context[xd->sb_index][xd->mb_index];
+ case BLOCK_16X8:
+ return &x->sb16x8_context[xd->sb_index][xd->mb_index][xd->b_index];
+ case BLOCK_8X16:
+ return &x->sb8x16_context[xd->sb_index][xd->mb_index][xd->b_index];
+ case BLOCK_8X8:
+ return &x->sb8x8_context[xd->sb_index][xd->mb_index][xd->b_index];
+ case BLOCK_8X4:
+ return &x->sb8x4_context[xd->sb_index][xd->mb_index][xd->b_index];
+ case BLOCK_4X8:
+ return &x->sb4x8_context[xd->sb_index][xd->mb_index][xd->b_index];
+ case BLOCK_4X4:
+ return &x->ab4x4_context[xd->sb_index][xd->mb_index][xd->b_index];
+ default:
+ assert(0);
+ return NULL;
+ }
+}
+
struct rdcost_block_args {
MACROBLOCK *x;
ENTROPY_CONTEXT t_above[16];
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index f48aa5a..98284a6 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -410,7 +410,7 @@
x->skip = ctx->skip;
vpx_memcpy(x->zcoeff_blk[mbmi->tx_size], ctx->zcoeff_blk,
- sizeof(ctx->zcoeff_blk));
+ sizeof(uint8_t) * ctx->num_4x4_blk);
if (!output_enabled)
return;
@@ -689,45 +689,6 @@
}
}
-// TODO(jingning): the variables used here are little complicated. need further
-// refactoring on organizing the temporary buffers, when recursive
-// partition down to 4x4 block size is enabled.
-static PICK_MODE_CONTEXT *get_block_context(MACROBLOCK *x, BLOCK_SIZE bsize) {
- MACROBLOCKD *const xd = &x->e_mbd;
-
- switch (bsize) {
- case BLOCK_64X64:
- return &x->sb64_context;
- case BLOCK_64X32:
- return &x->sb64x32_context[xd->sb_index];
- case BLOCK_32X64:
- return &x->sb32x64_context[xd->sb_index];
- case BLOCK_32X32:
- return &x->sb32_context[xd->sb_index];
- case BLOCK_32X16:
- return &x->sb32x16_context[xd->sb_index][xd->mb_index];
- case BLOCK_16X32:
- return &x->sb16x32_context[xd->sb_index][xd->mb_index];
- case BLOCK_16X16:
- return &x->mb_context[xd->sb_index][xd->mb_index];
- case BLOCK_16X8:
- return &x->sb16x8_context[xd->sb_index][xd->mb_index][xd->b_index];
- case BLOCK_8X16:
- return &x->sb8x16_context[xd->sb_index][xd->mb_index][xd->b_index];
- case BLOCK_8X8:
- return &x->sb8x8_context[xd->sb_index][xd->mb_index][xd->b_index];
- case BLOCK_8X4:
- return &x->sb8x4_context[xd->sb_index][xd->mb_index][xd->b_index];
- case BLOCK_4X8:
- return &x->sb4x8_context[xd->sb_index][xd->mb_index][xd->b_index];
- case BLOCK_4X4:
- return &x->ab4x4_context[xd->sb_index][xd->mb_index][xd->b_index];
- default:
- assert(0);
- return NULL;
- }
-}
-
static BLOCK_SIZE *get_sb_partitioning(MACROBLOCK *x, BLOCK_SIZE bsize) {
MACROBLOCKD *const xd = &x->e_mbd;
switch (bsize) {
diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c
index db74df7..f6b2a28 100644
--- a/vp9/encoder/vp9_onyx_if.c
+++ b/vp9/encoder/vp9_onyx_if.c
@@ -1416,6 +1416,94 @@
} while (++i <= MV_MAX);
}
+static void init_pick_mode_context(VP9_COMP *cpi) {
+ int i;
+ MACROBLOCK *x = &cpi->mb;
+ MACROBLOCKD *xd = &x->e_mbd;
+ VP9_COMMON *cm = &cpi->common;
+
+ for (i = 0; i < BLOCK_SIZES; ++i) {
+ const int num_4x4_w = num_4x4_blocks_wide_lookup[i];
+ const int num_4x4_h = num_4x4_blocks_high_lookup[i];
+ const int num_4x4_blk = MAX(4, num_4x4_w * num_4x4_h);
+ if (i < BLOCK_16X16) {
+ for (xd->sb_index = 0; xd->sb_index < 4; ++xd->sb_index) {
+ for (xd->mb_index = 0; xd->mb_index < 4; ++xd->mb_index) {
+ for (xd->b_index = 0; xd->b_index < 16 / num_4x4_blk; ++xd->b_index) {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ ctx->num_4x4_blk = num_4x4_blk;
+ CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
+ vpx_calloc(num_4x4_blk, sizeof(uint8_t)));
+ }
+ }
+ }
+ } else if (i < BLOCK_32X32) {
+ for (xd->sb_index = 0; xd->sb_index < 4; ++xd->sb_index) {
+ for (xd->mb_index = 0; xd->mb_index < 64 / num_4x4_blk;
+ ++xd->mb_index) {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ ctx->num_4x4_blk = num_4x4_blk;
+ CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
+ vpx_calloc(num_4x4_blk, sizeof(uint8_t)));
+ }
+ }
+ } else if (i < BLOCK_64X64) {
+ for (xd->sb_index = 0; xd->sb_index < 256 / num_4x4_blk; ++xd->sb_index) {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ ctx->num_4x4_blk = num_4x4_blk;
+ CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
+ vpx_calloc(num_4x4_blk, sizeof(uint8_t)));
+ }
+ } else {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ ctx->num_4x4_blk = num_4x4_blk;
+ CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
+ vpx_calloc(num_4x4_blk, sizeof(uint8_t)));
+ }
+ }
+}
+
+static void free_pick_mode_context(MACROBLOCK *x) {
+ int i;
+ MACROBLOCKD *xd = &x->e_mbd;
+
+ for (i = 0; i < BLOCK_SIZES; ++i) {
+ const int num_4x4_w = num_4x4_blocks_wide_lookup[i];
+ const int num_4x4_h = num_4x4_blocks_high_lookup[i];
+ const int num_4x4_blk = MAX(4, num_4x4_w * num_4x4_h);
+ if (i < BLOCK_16X16) {
+ for (xd->sb_index = 0; xd->sb_index < 4; ++xd->sb_index) {
+ for (xd->mb_index = 0; xd->mb_index < 4; ++xd->mb_index) {
+ for (xd->b_index = 0; xd->b_index < 16 / num_4x4_blk; ++xd->b_index) {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ vpx_free(ctx->zcoeff_blk);
+ ctx->zcoeff_blk = 0;
+ }
+ }
+ }
+ } else if (i < BLOCK_32X32) {
+ for (xd->sb_index = 0; xd->sb_index < 4; ++xd->sb_index) {
+ for (xd->mb_index = 0; xd->mb_index < 64 / num_4x4_blk;
+ ++xd->mb_index) {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ vpx_free(ctx->zcoeff_blk);
+ ctx->zcoeff_blk = 0;
+ }
+ }
+ } else if (i < BLOCK_64X64) {
+ for (xd->sb_index = 0; xd->sb_index < 256 / num_4x4_blk; ++xd->sb_index) {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ vpx_free(ctx->zcoeff_blk);
+ ctx->zcoeff_blk = 0;
+ }
+ } else {
+ PICK_MODE_CONTEXT *ctx = get_block_context(x, i);
+ vpx_free(ctx->zcoeff_blk);
+ ctx->zcoeff_blk = 0;
+ }
+ }
+}
+
VP9_PTR vp9_create_compressor(VP9_CONFIG *oxcf) {
int i, j;
volatile union {
@@ -1452,6 +1540,8 @@
init_config((VP9_PTR)cpi, oxcf);
+ init_pick_mode_context(cpi);
+
cm->current_video_frame = 0;
cpi->kf_overspend_bits = 0;
cpi->kf_bitrate_adjustment = 0;
@@ -1915,6 +2005,7 @@
#endif
}
+ free_pick_mode_context(&cpi->mb);
dealloc_compressor_data(cpi);
vpx_free(cpi->mb.ss);
vpx_free(cpi->tok);
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 0fc7152..9ef5076 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -248,23 +248,22 @@
set_block_thresholds(cpi);
- fill_token_costs(cpi->mb.token_costs, cpi->common.fc.coef_probs);
+ fill_token_costs(cpi->mb.token_costs, cm->fc.coef_probs);
for (i = 0; i < NUM_PARTITION_CONTEXTS; i++)
vp9_cost_tokens(cpi->mb.partition_cost[i],
- cpi->common.fc.partition_prob[cpi->common.frame_type][i],
+ cm->fc.partition_prob[cm->frame_type][i],
vp9_partition_tree);
/*rough estimate for costing*/
vp9_init_mode_costs(cpi);
- if (!frame_is_intra_only(&cpi->common)) {
+ if (!frame_is_intra_only(cm)) {
vp9_build_nmv_cost_table(
cpi->mb.nmvjointcost,
- cpi->common.allow_high_precision_mv ?
- cpi->mb.nmvcost_hp : cpi->mb.nmvcost,
- &cpi->common.fc.nmvc,
- cpi->common.allow_high_precision_mv, 1, 1);
+ cm->allow_high_precision_mv ? cpi->mb.nmvcost_hp : cpi->mb.nmvcost,
+ &cm->fc.nmvc,
+ cm->allow_high_precision_mv, 1, 1);
for (i = 0; i < INTER_MODE_CONTEXTS; i++) {
MB_PREDICTION_MODE m;
@@ -272,8 +271,8 @@
for (m = NEARESTMV; m < MB_MODE_COUNT; m++)
cpi->mb.inter_mode_cost[i][inter_mode_offset(m)] =
cost_token(vp9_inter_mode_tree,
- cpi->common.fc.inter_mode_probs[i],
- vp9_inter_mode_encodings + inter_mode_offset(m));
+ cm->fc.inter_mode_probs[i],
+ &vp9_inter_mode_encodings[inter_mode_offset(m)]);
}
}
}
@@ -3139,8 +3138,7 @@
unsigned int mode_mask = 0;
int64_t mode_distortions[MB_MODE_COUNT] = {-1};
int64_t frame_distortions[MAX_REF_FRAMES] = {-1};
- int intra_cost_penalty = 20 * vp9_dc_quant(cpi->common.base_qindex,
- cpi->common.y_dc_delta_q);
+ int intra_cost_penalty = 20 * vp9_dc_quant(cm->base_qindex, cm->y_dc_delta_q);
const int bws = num_8x8_blocks_wide_lookup[bsize] / 2;
const int bhs = num_8x8_blocks_high_lookup[bsize] / 2;
int best_skip2 = 0;
@@ -3321,7 +3319,7 @@
// Evaluate all sub-pel filters irrespective of whether we can use
// them for this frame.
mbmi->interp_filter = cm->mcomp_filter_type;
- vp9_setup_interp_filters(xd, mbmi->interp_filter, &cpi->common);
+ vp9_setup_interp_filters(xd, mbmi->interp_filter, cm);
if (comp_pred) {
if (!(cpi->ref_frame_flags & flag_list[second_ref_frame]))
@@ -3458,7 +3456,7 @@
continue;
}
- if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
+ if (cm->comp_pred_mode == HYBRID_PREDICTION) {
rate2 += compmode_cost;
}
@@ -3577,7 +3575,7 @@
best_mbmode = *mbmi;
best_skip2 = this_skip2;
vpx_memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
- sizeof(ctx->zcoeff_blk));
+ sizeof(uint8_t) * ctx->num_4x4_blk);
// TODO(debargha): enhance this test with a better distortion prediction
// based on qp, activity mask and history
@@ -3602,7 +3600,7 @@
if (!disable_skip && ref_frame != INTRA_FRAME) {
int single_rd, hybrid_rd, single_rate, hybrid_rate;
- if (cpi->common.comp_pred_mode == HYBRID_PREDICTION) {
+ if (cm->comp_pred_mode == HYBRID_PREDICTION) {
single_rate = rate2 - compmode_cost;
hybrid_rate = rate2;
} else {
@@ -4317,7 +4315,7 @@
best_mbmode = *mbmi;
best_skip2 = this_skip2;
vpx_memcpy(ctx->zcoeff_blk, x->zcoeff_blk[mbmi->tx_size],
- sizeof(ctx->zcoeff_blk));
+ sizeof(uint8_t) * ctx->num_4x4_blk);
for (i = 0; i < 4; i++)
best_bmodes[i] = xd->mi_8x8[0]->bmi[i];
