Facilitate separate frame buffer offset in diamond search
Frame offset used in Diamond search assumes same stride
for source buffer(lookahead) and scaled source buffer.
This patch will facilitate different frame buffer offset
for lookahead frame and scaled source buffer.
Change-Id: Ib6c2577b114c1cea96868e014bda761d78e97e9a
diff --git a/av1/encoder/encodeframe.c b/av1/encoder/encodeframe.c
index 2dd9027..189c397 100644
--- a/av1/encoder/encodeframe.c
+++ b/av1/encoder/encodeframe.c
@@ -2325,10 +2325,10 @@
// This overwrites the mv_limits so we will need to restore it later.
av1_set_mv_search_range(&x->mv_limits, &ref_mv);
- var = av1_full_pixel_search(cpi, x, bsize, &ref_mv_full, step_param,
- search_methods, do_mesh_search, sadpb,
- cond_cost_list(cpi, cost_list), &ref_mv, INT_MAX,
- 1, mi_col * MI_SIZE, mi_row * MI_SIZE, 0);
+ var = av1_full_pixel_search(
+ cpi, x, bsize, &ref_mv_full, step_param, search_methods, do_mesh_search,
+ sadpb, cond_cost_list(cpi, cost_list), &ref_mv, INT_MAX, 1,
+ mi_col * MI_SIZE, mi_row * MI_SIZE, 0, &cpi->ss_cfg[SS_CFG_SRC]);
// Restore
x->mv_limits = tmp_mv_limits;
diff --git a/av1/encoder/encoder.c b/av1/encoder/encoder.c
index 5e2a3b5..9462ff8 100644
--- a/av1/encoder/encoder.c
+++ b/av1/encoder/encoder.c
@@ -3883,11 +3883,17 @@
static void init_motion_estimation(AV1_COMP *cpi) {
int y_stride = cpi->scaled_source.y_stride;
+ int y_stride_src = (cpi->oxcf.resize_mode || cpi->oxcf.superres_mode)
+ ? y_stride
+ : cpi->lookahead->buf->img.y_stride;
if (cpi->sf.mv.search_method == NSTEP) {
- av1_init3smotion_compensation(&cpi->ss_cfg, y_stride);
+ av1_init3smotion_compensation(&cpi->ss_cfg[SS_CFG_SRC], y_stride);
+ av1_init3smotion_compensation(&cpi->ss_cfg[SS_CFG_LOOKAHEAD], y_stride_src);
} else if (cpi->sf.mv.search_method == DIAMOND) {
- av1_init_dsmotion_compensation(&cpi->ss_cfg, y_stride);
+ av1_init_dsmotion_compensation(&cpi->ss_cfg[SS_CFG_SRC], y_stride);
+ av1_init_dsmotion_compensation(&cpi->ss_cfg[SS_CFG_LOOKAHEAD],
+ y_stride_src);
}
}
diff --git a/av1/encoder/encoder.h b/av1/encoder/encoder.h
index fab81ff..f3b2bf8 100644
--- a/av1/encoder/encoder.h
+++ b/av1/encoder/encoder.h
@@ -148,6 +148,12 @@
kVeryHighSad = 6,
} CONTENT_STATE_SB;
+enum {
+ SS_CFG_SRC = 0,
+ SS_CFG_LOOKAHEAD = 1,
+ SS_CFG_TOTAL = 2
+} UENUM1BYTE(SS_CFG_OFFSET);
+
typedef struct TplDepStats {
int64_t intra_cost;
int64_t inter_cost;
@@ -858,7 +864,9 @@
int frame_flags;
- search_site_config ss_cfg;
+ // ss_cfg[SS_CFG_LOOKAHEAD] : used only in temporal filtering
+ // ss_cfg[SS_CFG_SRC] : used everywhere except temporal filtering
+ search_site_config ss_cfg[SS_CFG_TOTAL];
TileDataEnc *tile_data;
int allocated_tiles; // Keep track of memory allocated for tiles.
diff --git a/av1/encoder/firstpass.c b/av1/encoder/firstpass.c
index 80931aa..25fae11 100644
--- a/av1/encoder/firstpass.c
+++ b/av1/encoder/firstpass.c
@@ -247,8 +247,8 @@
}
// Center the initial step/diamond search on best mv.
- tmp_err = cpi->diamond_search_sad(x, &cpi->ss_cfg, &ref_mv_full, &tmp_mv,
- step_param, x->sadperbit16, &num00,
+ tmp_err = cpi->diamond_search_sad(x, &cpi->ss_cfg[SS_CFG_SRC], &ref_mv_full,
+ &tmp_mv, step_param, x->sadperbit16, &num00,
&v_fn_ptr, ref_mv);
if (tmp_err < INT_MAX)
tmp_err = av1_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
@@ -269,9 +269,9 @@
if (num00) {
--num00;
} else {
- tmp_err = cpi->diamond_search_sad(x, &cpi->ss_cfg, &ref_mv_full, &tmp_mv,
- step_param + n, x->sadperbit16, &num00,
- &v_fn_ptr, ref_mv);
+ tmp_err = cpi->diamond_search_sad(
+ x, &cpi->ss_cfg[SS_CFG_SRC], &ref_mv_full, &tmp_mv, step_param + n,
+ x->sadperbit16, &num00, &v_fn_ptr, ref_mv);
if (tmp_err < INT_MAX)
tmp_err = av1_get_mvpred_var(x, &tmp_mv, ref_mv, &v_fn_ptr, 1);
if (tmp_err < INT_MAX - new_mv_mode_penalty)
diff --git a/av1/encoder/mcomp.c b/av1/encoder/mcomp.c
index 1bab45c..3317a93 100644
--- a/av1/encoder/mcomp.c
+++ b/av1/encoder/mcomp.c
@@ -1798,11 +1798,11 @@
MV *mvp_full, int step_param, int sadpb,
int further_steps, int do_refine, int *cost_list,
const aom_variance_fn_ptr_t *fn_ptr,
- const MV *ref_mv) {
+ const MV *ref_mv, const search_site_config *cfg) {
MV temp_mv;
int thissme, n, num00 = 0;
- int bestsme = cpi->diamond_search_sad(x, &cpi->ss_cfg, mvp_full, &temp_mv,
- step_param, sadpb, &n, fn_ptr, ref_mv);
+ int bestsme = cpi->diamond_search_sad(x, cfg, mvp_full, &temp_mv, step_param,
+ sadpb, &n, fn_ptr, ref_mv);
if (bestsme < INT_MAX)
bestsme = av1_get_mvpred_var(x, &temp_mv, ref_mv, fn_ptr, 1);
x->best_mv.as_mv = temp_mv;
@@ -1817,9 +1817,9 @@
if (num00) {
num00--;
} else {
- thissme = cpi->diamond_search_sad(x, &cpi->ss_cfg, mvp_full, &temp_mv,
- step_param + n, sadpb, &num00, fn_ptr,
- ref_mv);
+ thissme =
+ cpi->diamond_search_sad(x, cfg, mvp_full, &temp_mv, step_param + n,
+ sadpb, &num00, fn_ptr, ref_mv);
if (thissme < INT_MAX)
thissme = av1_get_mvpred_var(x, &temp_mv, ref_mv, fn_ptr, 1);
@@ -2099,7 +2099,8 @@
MV *mvp_full, int step_param, int method,
int run_mesh_search, int error_per_bit,
int *cost_list, const MV *ref_mv, int var_max, int rd,
- int x_pos, int y_pos, int intra) {
+ int x_pos, int y_pos, int intra,
+ const search_site_config *cfg) {
const SPEED_FEATURES *const sf = &cpi->sf;
const aom_variance_fn_ptr_t *fn_ptr = &cpi->fn_ptr[bsize];
int var = 0;
@@ -2139,7 +2140,7 @@
case NSTEP:
var = full_pixel_diamond(cpi, x, mvp_full, step_param, error_per_bit,
MAX_MVSEARCH_STEPS - 1 - step_param, 1,
- cost_list, fn_ptr, ref_mv);
+ cost_list, fn_ptr, ref_mv, cfg);
// Should we allow a follow on exhaustive search?
if (is_exhaustive_allowed(cpi, x)) {
@@ -2676,14 +2677,15 @@
MV *mvp_full, int step_param, int sadpb,
int further_steps, int do_refine,
const aom_variance_fn_ptr_t *fn_ptr,
- const MV *ref_mv, MV *dst_mv,
- int is_second) {
+ const MV *ref_mv, MV *dst_mv, int is_second,
+ const search_site_config *cfg) {
+ (void)cpi; // to silence compiler warning
const int32_t *wsrc = x->wsrc_buf;
const int32_t *mask = x->mask_buf;
MV temp_mv;
int thissme, n, num00 = 0;
int bestsme =
- obmc_diamond_search_sad(x, &cpi->ss_cfg, wsrc, mask, mvp_full, &temp_mv,
+ obmc_diamond_search_sad(x, cfg, wsrc, mask, mvp_full, &temp_mv,
step_param, sadpb, &n, fn_ptr, ref_mv, is_second);
if (bestsme < INT_MAX)
bestsme = get_obmc_mvpred_var(x, wsrc, mask, &temp_mv, ref_mv, fn_ptr, 1,
@@ -2700,9 +2702,9 @@
if (num00) {
num00--;
} else {
- thissme = obmc_diamond_search_sad(x, &cpi->ss_cfg, wsrc, mask, mvp_full,
- &temp_mv, step_param + n, sadpb, &num00,
- fn_ptr, ref_mv, is_second);
+ thissme = obmc_diamond_search_sad(x, cfg, wsrc, mask, mvp_full, &temp_mv,
+ step_param + n, sadpb, &num00, fn_ptr,
+ ref_mv, is_second);
if (thissme < INT_MAX)
thissme = get_obmc_mvpred_var(x, wsrc, mask, &temp_mv, ref_mv, fn_ptr,
1, is_second);
@@ -2738,11 +2740,12 @@
int step_param, int sadpb, int further_steps,
int do_refine,
const aom_variance_fn_ptr_t *fn_ptr,
- const MV *ref_mv, MV *dst_mv, int is_second) {
+ const MV *ref_mv, MV *dst_mv, int is_second,
+ const search_site_config *cfg) {
if (cpi->sf.obmc_full_pixel_search_level == 0) {
return obmc_full_pixel_diamond(cpi, x, mvp_full, step_param, sadpb,
further_steps, do_refine, fn_ptr, ref_mv,
- dst_mv, is_second);
+ dst_mv, is_second, cfg);
} else {
const int32_t *wsrc = x->wsrc_buf;
const int32_t *mask = x->mask_buf;
diff --git a/av1/encoder/mcomp.h b/av1/encoder/mcomp.h
index 3f8b3b1..1284fa4 100644
--- a/av1/encoder/mcomp.h
+++ b/av1/encoder/mcomp.h
@@ -132,13 +132,15 @@
BLOCK_SIZE bsize, MV *mvp_full, int step_param,
int method, int run_mesh_search, int error_per_bit,
int *cost_list, const MV *ref_mv, int var_max, int rd,
- int x_pos, int y_pos, int intra);
+ int x_pos, int y_pos, int intra,
+ const search_site_config *cfg);
int av1_obmc_full_pixel_search(const struct AV1_COMP *cpi, MACROBLOCK *x,
MV *mvp_full, int step_param, int sadpb,
int further_steps, int do_refine,
const aom_variance_fn_ptr_t *fn_ptr,
- const MV *ref_mv, MV *dst_mv, int is_second);
+ const MV *ref_mv, MV *dst_mv, int is_second,
+ const search_site_config *cfg);
int av1_find_best_obmc_sub_pixel_tree_up(
MACROBLOCK *x, const AV1_COMMON *const cm, int mi_row, int mi_col,
MV *bestmv, const MV *ref_mv, int allow_hp, int error_per_bit,
diff --git a/av1/encoder/rdopt.c b/av1/encoder/rdopt.c
index 10c75b1..eb51457 100644
--- a/av1/encoder/rdopt.c
+++ b/av1/encoder/rdopt.c
@@ -7131,13 +7131,13 @@
bestsme = av1_full_pixel_search(
cpi, x, bsize, &mvp_full, step_param, cpi->sf.mv.search_method, 0,
sadpb, cond_cost_list(cpi, cost_list), &ref_mv, INT_MAX, 1,
- (MI_SIZE * mi_col), (MI_SIZE * mi_row), 0);
+ (MI_SIZE * mi_col), (MI_SIZE * mi_row), 0, &cpi->ss_cfg[SS_CFG_SRC]);
break;
case OBMC_CAUSAL:
- bestsme = av1_obmc_full_pixel_search(cpi, x, &mvp_full, step_param, sadpb,
- MAX_MVSEARCH_STEPS - 1 - step_param,
- 1, &cpi->fn_ptr[bsize], &ref_mv,
- &(x->best_mv.as_mv), 0);
+ bestsme = av1_obmc_full_pixel_search(
+ cpi, x, &mvp_full, step_param, sadpb,
+ MAX_MVSEARCH_STEPS - 1 - step_param, 1, &cpi->fn_ptr[bsize], &ref_mv,
+ &(x->best_mv.as_mv), 0, &cpi->ss_cfg[SS_CFG_SRC]);
break;
default: assert(0 && "Invalid motion mode!\n");
}
@@ -10540,7 +10540,7 @@
const int bestsme = av1_full_pixel_search(
cpi, x, bsize, &mvp_full, step_param, cpi->sf.mv.search_method, 0,
sadpb, cond_cost_list(cpi, cost_list), &dv_ref.as_mv, INT_MAX, 1,
- (MI_SIZE * mi_col), (MI_SIZE * mi_row), 1);
+ (MI_SIZE * mi_col), (MI_SIZE * mi_row), 1, &cpi->ss_cfg[SS_CFG_SRC]);
x->mv_limits = tmp_mv_limits;
if (bestsme == INT_MAX) continue;
diff --git a/av1/encoder/temporal_filter.c b/av1/encoder/temporal_filter.c
index d151762..3339faf 100644
--- a/av1/encoder/temporal_filter.c
+++ b/av1/encoder/temporal_filter.c
@@ -661,7 +661,7 @@
// x->best_mv.
av1_full_pixel_search(cpi, x, TF_BLOCK, &best_ref_mv1_full, step_param, NSTEP,
1, sadpb, cond_cost_list(cpi, cost_list), &best_ref_mv1,
- 0, 0, x_pos, y_pos, 0);
+ 0, 0, x_pos, y_pos, 0, &cpi->ss_cfg[SS_CFG_LOOKAHEAD]);
x->mv_limits = tmp_mv_limits;
// Ignore mv costing by sending NULL pointer instead of cost array
@@ -715,9 +715,10 @@
xd->plane[0].pre[0].stride = stride;
av1_set_mv_search_range(&x->mv_limits, &best_ref_mv1);
- av1_full_pixel_search(
- cpi, x, TF_SUB_BLOCK, &best_ref_mv1_full, step_param, NSTEP, 1, sadpb,
- cond_cost_list(cpi, cost_list), &best_ref_mv1, 0, 0, x_pos, y_pos, 0);
+ av1_full_pixel_search(cpi, x, TF_SUB_BLOCK, &best_ref_mv1_full,
+ step_param, NSTEP, 1, sadpb,
+ cond_cost_list(cpi, cost_list), &best_ref_mv1, 0, 0,
+ x_pos, y_pos, 0, &cpi->ss_cfg[SS_CFG_LOOKAHEAD]);
x->mv_limits = tmp_mv_limits;
blk_bestsme[k] = cpi->find_fractional_mv_step(
diff --git a/av1/encoder/tpl_model.c b/av1/encoder/tpl_model.c
index 3399116..36d342a 100644
--- a/av1/encoder/tpl_model.c
+++ b/av1/encoder/tpl_model.c
@@ -97,7 +97,7 @@
av1_full_pixel_search(cpi, x, bsize, &best_ref_mv1_full, step_param,
search_method, 0, sadpb, cond_cost_list(cpi, cost_list),
&best_ref_mv1, INT_MAX, 0, (MI_SIZE * mi_col),
- (MI_SIZE * mi_row), 0);
+ (MI_SIZE * mi_row), 0, &cpi->ss_cfg[SS_CFG_SRC]);
/* restore UMV window */
x->mv_limits = tmp_mv_limits;
