Merge "Refactor codes about motion search" into nextgenv2
diff --git a/vp10/encoder/block.h b/vp10/encoder/block.h
index 88f1236..4c9f8a5 100644
--- a/vp10/encoder/block.h
+++ b/vp10/encoder/block.h
@@ -165,6 +165,9 @@
// Used to store sub partition's choices.
MV pred_mv[MAX_REF_FRAMES];
+ // Store the best motion vector during motion search
+ int_mv best_mv;
+
// Strong color activity detection. Used in RTC coding mode to enhance
// the visual quality at the boundary of moving color objects.
uint8_t color_sensitivity[2];
diff --git a/vp10/encoder/encoder.h b/vp10/encoder/encoder.h
index 519cd03..608cace 100644
--- a/vp10/encoder/encoder.h
+++ b/vp10/encoder/encoder.h
@@ -447,7 +447,7 @@
ActiveMap active_map;
fractional_mv_step_fp *find_fractional_mv_step;
- vp10_full_search_fn_t full_search_sad;
+ vp10_full_search_fn_t full_search_sad; // It is currently unused.
vp10_diamond_search_fn_t diamond_search_sad;
vp10_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
uint64_t time_receive_data;
diff --git a/vp10/encoder/mbgraph.c b/vp10/encoder/mbgraph.c
index 5e66ce5..46cff80 100644
--- a/vp10/encoder/mbgraph.c
+++ b/vp10/encoder/mbgraph.c
@@ -25,7 +25,6 @@
static unsigned int do_16x16_motion_iteration(VP10_COMP *cpi,
const MV *ref_mv,
- MV *dst_mv,
int mb_row,
int mb_col) {
MACROBLOCK *const x = &cpi->td.mb;
@@ -51,8 +50,7 @@
/*cpi->sf.search_method == HEX*/
vp10_hex_search(x, &ref_full, step_param, x->errorperbit, 0,
- cond_cost_list(cpi, cost_list),
- &v_fn_ptr, 0, ref_mv, dst_mv);
+ cond_cost_list(cpi, cost_list), &v_fn_ptr, 0, ref_mv);
// Try sub-pixel MC
// if (bestsme > error_thresh && bestsme < INT_MAX)
@@ -60,7 +58,7 @@
int distortion;
unsigned int sse;
cpi->find_fractional_mv_step(
- x, dst_mv, ref_mv, cpi->common.allow_high_precision_mv, x->errorperbit,
+ x, ref_mv, cpi->common.allow_high_precision_mv, x->errorperbit,
&v_fn_ptr, 0, mv_sf->subpel_iters_per_step,
cond_cost_list(cpi, cost_list),
NULL, NULL,
@@ -74,7 +72,7 @@
#endif // CONFIG_EXT_INTER
xd->mi[0]->mbmi.mode = NEWMV;
- xd->mi[0]->mbmi.mv[0].as_mv = *dst_mv;
+ xd->mi[0]->mbmi.mv[0] = x->best_mv;
#if CONFIG_EXT_INTER
xd->mi[0]->mbmi.ref_frame[1] = NONE;
#endif // CONFIG_EXT_INTER
@@ -92,40 +90,40 @@
}
static int do_16x16_motion_search(VP10_COMP *cpi, const MV *ref_mv,
- int_mv *dst_mv, int mb_row, int mb_col) {
+ int mb_row, int mb_col) {
MACROBLOCK *const x = &cpi->td.mb;
MACROBLOCKD *const xd = &x->e_mbd;
unsigned int err, tmp_err;
- MV tmp_mv;
+ MV best_mv;
// Try zero MV first
// FIXME should really use something like near/nearest MV and/or MV prediction
err = vpx_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride);
- dst_mv->as_int = 0;
+ best_mv.col = best_mv.row = 0;
// Test last reference frame using the previous best mv as the
// starting point (best reference) for the search
- tmp_err = do_16x16_motion_iteration(cpi, ref_mv, &tmp_mv, mb_row, mb_col);
+ tmp_err = do_16x16_motion_iteration(cpi, ref_mv, mb_row, mb_col);
if (tmp_err < err) {
err = tmp_err;
- dst_mv->as_mv = tmp_mv;
+ best_mv = x->best_mv.as_mv;
}
// If the current best reference mv is not centered on 0,0 then do a 0,0
// based search as well.
if (ref_mv->row != 0 || ref_mv->col != 0) {
unsigned int tmp_err;
- MV zero_ref_mv = {0, 0}, tmp_mv;
+ MV zero_ref_mv = {0, 0};
- tmp_err = do_16x16_motion_iteration(cpi, &zero_ref_mv, &tmp_mv,
- mb_row, mb_col);
+ tmp_err = do_16x16_motion_iteration(cpi, &zero_ref_mv, mb_row, mb_col);
if (tmp_err < err) {
- dst_mv->as_mv = tmp_mv;
err = tmp_err;
+ best_mv = x->best_mv.as_mv;
}
}
+ x->best_mv.as_mv = best_mv;
return err;
}
@@ -213,8 +211,8 @@
xd->plane[0].pre[0].stride = golden_ref->y_stride;
g_motion_error = do_16x16_motion_search(cpi,
prev_golden_ref_mv,
- &stats->ref[GOLDEN_FRAME].m.mv,
mb_row, mb_col);
+ stats->ref[GOLDEN_FRAME].m.mv = x->best_mv;
stats->ref[GOLDEN_FRAME].err = g_motion_error;
} else {
stats->ref[GOLDEN_FRAME].err = INT_MAX;
diff --git a/vp10/encoder/mcomp.c b/vp10/encoder/mcomp.c
index e7bd700..afbf3e9 100644
--- a/vp10/encoder/mcomp.c
+++ b/vp10/encoder/mcomp.c
@@ -315,6 +315,7 @@
unsigned int sse; \
unsigned int whichdir; \
int thismse; \
+ MV *bestmv = &x->best_mv.as_mv; \
const unsigned int halfiters = iters_per_step; \
const unsigned int quarteriters = iters_per_step; \
const unsigned int eighthiters = iters_per_step; \
@@ -411,8 +412,8 @@
}
int vp10_find_best_sub_pixel_tree_pruned_evenmore(
- const MACROBLOCK *x,
- MV *bestmv, const MV *ref_mv,
+ MACROBLOCK *x,
+ const MV *ref_mv,
int allow_hp,
int error_per_bit,
const vp10_variance_fn_ptr_t *vfp,
@@ -491,20 +492,20 @@
return besterr;
}
-int vp10_find_best_sub_pixel_tree_pruned_more(const MACROBLOCK *x,
- MV *bestmv, const MV *ref_mv,
- int allow_hp,
- int error_per_bit,
- const vp10_variance_fn_ptr_t *vfp,
- int forced_stop,
- int iters_per_step,
- int *cost_list,
- int *mvjcost, int *mvcost[2],
- int *distortion,
- unsigned int *sse1,
- const uint8_t *second_pred,
- int w, int h,
- int use_upsampled_ref) {
+int vp10_find_best_sub_pixel_tree_pruned_more(MACROBLOCK *x,
+ const MV *ref_mv,
+ int allow_hp,
+ int error_per_bit,
+ const vp10_variance_fn_ptr_t *vfp,
+ int forced_stop,
+ int iters_per_step,
+ int *cost_list,
+ int *mvjcost, int *mvcost[2],
+ int *distortion,
+ unsigned int *sse1,
+ const uint8_t *second_pred,
+ int w, int h,
+ int use_upsampled_ref) {
SETUP_SUBPEL_SEARCH;
(void) use_upsampled_ref;
@@ -568,19 +569,19 @@
return besterr;
}
-int vp10_find_best_sub_pixel_tree_pruned(const MACROBLOCK *x,
- MV *bestmv, const MV *ref_mv,
- int allow_hp,
- int error_per_bit,
- const vp10_variance_fn_ptr_t *vfp,
- int forced_stop,
- int iters_per_step,
- int *cost_list,
- int *mvjcost, int *mvcost[2],
- int *distortion,
- unsigned int *sse1,
- const uint8_t *second_pred,
- int w, int h, int use_upsampled_ref) {
+int vp10_find_best_sub_pixel_tree_pruned(MACROBLOCK *x,
+ const MV *ref_mv,
+ int allow_hp,
+ int error_per_bit,
+ const vp10_variance_fn_ptr_t *vfp,
+ int forced_stop,
+ int iters_per_step,
+ int *cost_list,
+ int *mvjcost, int *mvcost[2],
+ int *distortion,
+ unsigned int *sse1,
+ const uint8_t *second_pred,
+ int w, int h, int use_upsampled_ref) {
SETUP_SUBPEL_SEARCH;
(void) use_upsampled_ref;
@@ -724,19 +725,19 @@
return besterr;
}
-int vp10_find_best_sub_pixel_tree(const MACROBLOCK *x,
- MV *bestmv, const MV *ref_mv,
- int allow_hp,
- int error_per_bit,
- const vp10_variance_fn_ptr_t *vfp,
- int forced_stop,
- int iters_per_step,
- int *cost_list,
- int *mvjcost, int *mvcost[2],
- int *distortion,
- unsigned int *sse1,
- const uint8_t *second_pred,
- int w, int h, int use_upsampled_ref) {
+int vp10_find_best_sub_pixel_tree(MACROBLOCK *x,
+ const MV *ref_mv,
+ int allow_hp,
+ int error_per_bit,
+ const vp10_variance_fn_ptr_t *vfp,
+ int forced_stop,
+ int iters_per_step,
+ int *cost_list,
+ int *mvjcost, int *mvcost[2],
+ int *distortion,
+ unsigned int *sse1,
+ const uint8_t *second_pred,
+ int w, int h, int use_upsampled_ref) {
const uint8_t *const src_address = x->plane[0].src.buf;
const int src_stride = x->plane[0].src.stride;
const MACROBLOCKD *xd = &x->e_mbd;
@@ -744,6 +745,7 @@
unsigned int sse;
unsigned int thismse;
const int y_stride = xd->plane[0].pre[0].stride;
+ MV *bestmv = &x->best_mv.as_mv;
const int offset = bestmv->row * y_stride + bestmv->col;
const uint8_t *const y = xd->plane[0].pre[0].buf;
@@ -1037,19 +1039,18 @@
// candidates as indicated in the num_candidates and candidates arrays
// passed into this function
//
-static int vp10_pattern_search(const MACROBLOCK *x,
- MV *start_mv,
- int search_param,
- int sad_per_bit,
- int do_init_search,
- int *cost_list,
- const vp10_variance_fn_ptr_t *vfp,
- int use_mvcost,
- const MV *center_mv,
- MV *best_mv,
- const int num_candidates[MAX_PATTERN_SCALES],
- const MV candidates[MAX_PATTERN_SCALES]
- [MAX_PATTERN_CANDIDATES]) {
+static int pattern_search(MACROBLOCK *x,
+ MV *start_mv,
+ int search_param,
+ int sad_per_bit,
+ int do_init_search,
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv,
+ const int num_candidates[MAX_PATTERN_SCALES],
+ const MV candidates[MAX_PATTERN_SCALES]
+ [MAX_PATTERN_CANDIDATES]) {
const MACROBLOCKD *const xd = &x->e_mbd;
static const int search_param_to_steps[MAX_MVSEARCH_STEPS] = {
10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
@@ -1294,8 +1295,8 @@
calc_int_cost_list(x, center_mv, sad_per_bit, vfp, &best_mv, cost_list);
}
}
- best_mv->row = br;
- best_mv->col = bc;
+ x->best_mv.as_mv.row = br;
+ x->best_mv.as_mv.col = bc;
return bestsad;
}
@@ -1332,15 +1333,15 @@
x->mvcost, x->errorperbit) : 0);
}
-int vp10_hex_search(const MACROBLOCK *x,
- MV *start_mv,
- int search_param,
- int sad_per_bit,
- int do_init_search,
- int *cost_list,
- const vp10_variance_fn_ptr_t *vfp,
- int use_mvcost,
- const MV *center_mv, MV *best_mv) {
+int vp10_hex_search(MACROBLOCK *x,
+ MV *start_mv,
+ int search_param,
+ int sad_per_bit,
+ int do_init_search,
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv) {
// First scale has 8-closest points, the rest have 6 points in hex shape
// at increasing scales
static const int hex_num_candidates[MAX_PATTERN_SCALES] = {
@@ -1361,22 +1362,20 @@
{{-512, -1024}, {512, -1024}, {1024, 0}, {512, 1024}, { -512, 1024},
{ -1024, 0}},
};
- return vp10_pattern_search(x, start_mv, search_param, sad_per_bit,
- do_init_search, cost_list, vfp, use_mvcost,
- center_mv, best_mv, hex_num_candidates,
- hex_candidates);
+ return pattern_search(x, start_mv, search_param, sad_per_bit,
+ do_init_search, cost_list, vfp, use_mvcost,
+ center_mv, hex_num_candidates, hex_candidates);
}
-int vp10_bigdia_search(const MACROBLOCK *x,
- MV *start_mv,
- int search_param,
- int sad_per_bit,
- int do_init_search,
- int *cost_list,
- const vp10_variance_fn_ptr_t *vfp,
- int use_mvcost,
- const MV *center_mv,
- MV *best_mv) {
+static int bigdia_search(MACROBLOCK *x,
+ MV *start_mv,
+ int search_param,
+ int sad_per_bit,
+ int do_init_search,
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv) {
// First scale has 4-closest points, the rest have 8 points in diamond
// shape at increasing scales
static const int bigdia_num_candidates[MAX_PATTERN_SCALES] = {
@@ -1403,25 +1402,23 @@
{{-512, -512}, {0, -1024}, {512, -512}, {1024, 0}, {512, 512}, {0, 1024},
{-512, 512}, {-1024, 0}},
};
- return vp10_pattern_search(x, start_mv, search_param, sad_per_bit,
- do_init_search, cost_list, vfp, use_mvcost,
- center_mv, best_mv, bigdia_num_candidates,
- bigdia_candidates);
+ return pattern_search(x, start_mv, search_param, sad_per_bit,
+ do_init_search, cost_list, vfp, use_mvcost,
+ center_mv, bigdia_num_candidates, bigdia_candidates);
}
-int vp10_square_search(const MACROBLOCK *x,
- MV *start_mv,
- int search_param,
- int sad_per_bit,
- int do_init_search,
- int *cost_list,
- const vp10_variance_fn_ptr_t *vfp,
- int use_mvcost,
- const MV *center_mv,
- MV *best_mv) {
+static int square_search(MACROBLOCK *x,
+ MV *start_mv,
+ int search_param,
+ int sad_per_bit,
+ int do_init_search,
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv) {
// All scales have 8 closest points in square shape
static const int square_num_candidates[MAX_PATTERN_SCALES] = {
- 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
+ 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
};
// Note that the largest candidate step at each scale is 2^scale
static const MV square_candidates[MAX_PATTERN_SCALES]
@@ -1445,40 +1442,37 @@
{{-1024, -1024}, {0, -1024}, {1024, -1024}, {1024, 0}, {1024, 1024},
{0, 1024}, {-1024, 1024}, {-1024, 0}},
};
- return vp10_pattern_search(x, start_mv, search_param, sad_per_bit,
- do_init_search, cost_list, vfp, use_mvcost,
- center_mv, best_mv, square_num_candidates,
- square_candidates);
+ return pattern_search(x, start_mv, search_param, sad_per_bit,
+ do_init_search, cost_list, vfp, use_mvcost,
+ center_mv, square_num_candidates, square_candidates);
}
-int vp10_fast_hex_search(const MACROBLOCK *x,
- MV *ref_mv,
- int search_param,
- int sad_per_bit,
- int do_init_search, // must be zero for fast_hex
- int *cost_list,
- const vp10_variance_fn_ptr_t *vfp,
- int use_mvcost,
- const MV *center_mv,
- MV *best_mv) {
+static int fast_hex_search(MACROBLOCK *x,
+ MV *ref_mv,
+ int search_param,
+ int sad_per_bit,
+ int do_init_search, // must be zero for fast_hex
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv) {
return vp10_hex_search(
x, ref_mv, VPXMAX(MAX_MVSEARCH_STEPS - 2, search_param), sad_per_bit,
- do_init_search, cost_list, vfp, use_mvcost, center_mv, best_mv);
+ do_init_search, cost_list, vfp, use_mvcost, center_mv);
}
-int vp10_fast_dia_search(const MACROBLOCK *x,
- MV *ref_mv,
- int search_param,
- int sad_per_bit,
- int do_init_search,
- int *cost_list,
- const vp10_variance_fn_ptr_t *vfp,
- int use_mvcost,
- const MV *center_mv,
- MV *best_mv) {
- return vp10_bigdia_search(
+static int fast_dia_search(MACROBLOCK *x,
+ MV *ref_mv,
+ int search_param,
+ int sad_per_bit,
+ int do_init_search,
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv) {
+ return bigdia_search(
x, ref_mv, VPXMAX(MAX_MVSEARCH_STEPS - 2, search_param), sad_per_bit,
- do_init_search, cost_list, vfp, use_mvcost, center_mv, best_mv);
+ do_init_search, cost_list, vfp, use_mvcost, center_mv);
}
#undef CHECK_BETTER
@@ -1924,12 +1918,12 @@
/* do_refine: If last step (1-away) of n-step search doesn't pick the center
point as the best match, we will do a final 1-away diamond
refining search */
-int vp10_full_pixel_diamond(const VP10_COMP *cpi, MACROBLOCK *x,
- MV *mvp_full, int step_param,
- int sadpb, int further_steps, int do_refine,
- int *cost_list,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *ref_mv, MV *dst_mv) {
+static int full_pixel_diamond(VP10_COMP *cpi, MACROBLOCK *x,
+ MV *mvp_full, int step_param,
+ int sadpb, int further_steps, int do_refine,
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *fn_ptr,
+ const MV *ref_mv) {
MV temp_mv;
int thissme, n, num00 = 0;
int bestsme = cpi->diamond_search_sad(x, &cpi->ss_cfg, mvp_full, &temp_mv,
@@ -1937,7 +1931,7 @@
fn_ptr, ref_mv);
if (bestsme < INT_MAX)
bestsme = vp10_get_mvpred_var(x, &temp_mv, ref_mv, fn_ptr, 1);
- *dst_mv = temp_mv;
+ x->best_mv.as_mv = temp_mv;
// If there won't be more n-step search, check to see if refining search is
// needed.
@@ -1962,7 +1956,7 @@
if (thissme < bestsme) {
bestsme = thissme;
- *dst_mv = temp_mv;
+ x->best_mv.as_mv = temp_mv;
}
}
}
@@ -1970,20 +1964,20 @@
// final 1-away diamond refining search
if (do_refine) {
const int search_range = 8;
- MV best_mv = *dst_mv;
+ MV best_mv = x->best_mv.as_mv;
thissme = vp10_refining_search_sad(x, &best_mv, sadpb, search_range,
fn_ptr, ref_mv);
if (thissme < INT_MAX)
thissme = vp10_get_mvpred_var(x, &best_mv, ref_mv, fn_ptr, 1);
if (thissme < bestsme) {
bestsme = thissme;
- *dst_mv = best_mv;
+ x->best_mv.as_mv = best_mv;
}
}
// Return cost list.
if (cost_list) {
- calc_int_cost_list(x, ref_mv, sadpb, fn_ptr, dst_mv, cost_list);
+ calc_int_cost_list(x, ref_mv, sadpb, fn_ptr, &x->best_mv.as_mv, cost_list);
}
return bestsme;
}
@@ -1994,7 +1988,8 @@
// Runs an limited range exhaustive mesh search using a pattern set
// according to the encode speed profile.
static int full_pixel_exhaustive(VP10_COMP *cpi, MACROBLOCK *x,
- MV *centre_mv_full, int sadpb, int *cost_list,
+ const MV *centre_mv_full, int sadpb,
+ int *cost_list,
const vp10_variance_fn_ptr_t *fn_ptr,
const MV *ref_mv, MV *dst_mv) {
const SPEED_FEATURES *const sf = &cpi->sf;
@@ -2053,9 +2048,9 @@
}
int vp10_full_search_sad_c(const MACROBLOCK *x, const MV *ref_mv,
- int sad_per_bit, int distance,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, MV *best_mv) {
+ int sad_per_bit, int distance,
+ const vp10_variance_fn_ptr_t *fn_ptr,
+ const MV *center_mv, MV *best_mv) {
int r, c;
const MACROBLOCKD *const xd = &x->e_mbd;
const struct buf_2d *const what = &x->plane[0].src;
@@ -2086,9 +2081,9 @@
}
int vp10_full_search_sadx3(const MACROBLOCK *x, const MV *ref_mv,
- int sad_per_bit, int distance,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, MV *best_mv) {
+ int sad_per_bit, int distance,
+ const vp10_variance_fn_ptr_t *fn_ptr,
+ const MV *center_mv, MV *best_mv) {
int r;
const MACROBLOCKD *const xd = &x->e_mbd;
const struct buf_2d *const what = &x->plane[0].src;
@@ -2151,9 +2146,9 @@
}
int vp10_full_search_sadx8(const MACROBLOCK *x, const MV *ref_mv,
- int sad_per_bit, int distance,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, MV *best_mv) {
+ int sad_per_bit, int distance,
+ const vp10_variance_fn_ptr_t *fn_ptr,
+ const MV *center_mv, MV *best_mv) {
int r;
const MACROBLOCKD *const xd = &x->e_mbd;
const struct buf_2d *const what = &x->plane[0].src;
@@ -2318,29 +2313,30 @@
// This function is called when we do joint motion search in comp_inter_inter
// mode.
-int vp10_refining_search_8p_c(const MACROBLOCK *x,
- MV *ref_mv, int error_per_bit,
- int search_range,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv,
- const uint8_t *second_pred) {
+int vp10_refining_search_8p_c(MACROBLOCK *x,
+ int error_per_bit,
+ int search_range,
+ const vp10_variance_fn_ptr_t *fn_ptr,
+ const MV *center_mv,
+ const uint8_t *second_pred) {
const MV neighbors[8] = {{-1, 0}, {0, -1}, {0, 1}, {1, 0},
{-1, -1}, {1, -1}, {-1, 1}, {1, 1}};
const MACROBLOCKD *const xd = &x->e_mbd;
const struct buf_2d *const what = &x->plane[0].src;
const struct buf_2d *const in_what = &xd->plane[0].pre[0];
const MV fcenter_mv = {center_mv->row >> 3, center_mv->col >> 3};
+ MV *best_mv = &x->best_mv.as_mv;
unsigned int best_sad = fn_ptr->sdaf(what->buf, what->stride,
- get_buf_from_mv(in_what, ref_mv), in_what->stride, second_pred) +
- mvsad_err_cost(x, ref_mv, &fcenter_mv, error_per_bit);
+ get_buf_from_mv(in_what, best_mv), in_what->stride, second_pred) +
+ mvsad_err_cost(x, best_mv, &fcenter_mv, error_per_bit);
int i, j;
for (i = 0; i < search_range; ++i) {
int best_site = -1;
for (j = 0; j < 8; ++j) {
- const MV mv = {ref_mv->row + neighbors[j].row,
- ref_mv->col + neighbors[j].col};
+ const MV mv = {best_mv->row + neighbors[j].row,
+ best_mv->col + neighbors[j].col};
if (is_mv_in(x, &mv)) {
unsigned int sad = fn_ptr->sdaf(what->buf, what->stride,
@@ -2358,8 +2354,8 @@
if (best_site == -1) {
break;
} else {
- ref_mv->row += neighbors[best_site].row;
- ref_mv->col += neighbors[best_site].col;
+ best_mv->row += neighbors[best_site].row;
+ best_mv->col += neighbors[best_site].col;
}
}
return best_sad;
@@ -2378,15 +2374,15 @@
}
int vp10_full_pixel_search(VP10_COMP *cpi, MACROBLOCK *x,
- BLOCK_SIZE bsize, MV *mvp_full,
- int step_param, int error_per_bit,
- int *cost_list,
- const MV *ref_mv, MV *tmp_mv,
- int var_max, int rd) {
+ BLOCK_SIZE bsize, MV *mvp_full,
+ int step_param, int error_per_bit,
+ int *cost_list, const MV *ref_mv,
+ int var_max, int rd) {
const SPEED_FEATURES *const sf = &cpi->sf;
const SEARCH_METHODS method = sf->mv.search_method;
vp10_variance_fn_ptr_t *fn_ptr = &cpi->fn_ptr[bsize];
int var = 0;
+
if (cost_list) {
cost_list[0] = INT_MAX;
cost_list[1] = INT_MAX;
@@ -2400,29 +2396,29 @@
switch (method) {
case FAST_DIAMOND:
- var = vp10_fast_dia_search(x, mvp_full, step_param, error_per_bit, 0,
- cost_list, fn_ptr, 1, ref_mv, tmp_mv);
+ var = fast_dia_search(x, mvp_full, step_param, error_per_bit, 0,
+ cost_list, fn_ptr, 1, ref_mv);
break;
case FAST_HEX:
- var = vp10_fast_hex_search(x, mvp_full, step_param, error_per_bit, 0,
- cost_list, fn_ptr, 1, ref_mv, tmp_mv);
+ var = fast_hex_search(x, mvp_full, step_param, error_per_bit, 0,
+ cost_list, fn_ptr, 1, ref_mv);
break;
case HEX:
var = vp10_hex_search(x, mvp_full, step_param, error_per_bit, 1,
- cost_list, fn_ptr, 1, ref_mv, tmp_mv);
+ cost_list, fn_ptr, 1, ref_mv);
break;
case SQUARE:
- var = vp10_square_search(x, mvp_full, step_param, error_per_bit, 1,
- cost_list, fn_ptr, 1, ref_mv, tmp_mv);
+ var = square_search(x, mvp_full, step_param, error_per_bit, 1,
+ cost_list, fn_ptr, 1, ref_mv);
break;
case BIGDIA:
- var = vp10_bigdia_search(x, mvp_full, step_param, error_per_bit, 1,
- cost_list, fn_ptr, 1, ref_mv, tmp_mv);
+ var = bigdia_search(x, mvp_full, step_param, error_per_bit, 1,
+ cost_list, fn_ptr, 1, ref_mv);
break;
case NSTEP:
- var = vp10_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, tmp_mv);
+ 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);
// Should we allow a follow on exhaustive search?
if (is_exhaustive_allowed(cpi, x)) {
@@ -2434,13 +2430,13 @@
if (var > exhuastive_thr) {
int var_ex;
MV tmp_mv_ex;
- var_ex = full_pixel_exhaustive(cpi, x, tmp_mv,
+ var_ex = full_pixel_exhaustive(cpi, x, &x->best_mv.as_mv,
error_per_bit, cost_list, fn_ptr,
ref_mv, &tmp_mv_ex);
if (var_ex < var) {
var = var_ex;
- *tmp_mv = tmp_mv_ex;
+ x->best_mv.as_mv = tmp_mv_ex;
}
}
}
@@ -2452,7 +2448,7 @@
}
if (method != NSTEP && rd && var < var_max)
- var = vp10_get_mvpred_var(x, tmp_mv, ref_mv, fn_ptr, 1);
+ var = vp10_get_mvpred_var(x, &x->best_mv.as_mv, ref_mv, fn_ptr, 1);
return var;
}
diff --git a/vp10/encoder/mcomp.h b/vp10/encoder/mcomp.h
index 704e26c..f97f6c7 100644
--- a/vp10/encoder/mcomp.h
+++ b/vp10/encoder/mcomp.h
@@ -82,31 +82,24 @@
// Perform integral projection based motion estimation.
unsigned int vp10_int_pro_motion_estimation(const struct VP10_COMP *cpi,
- MACROBLOCK *x,
- BLOCK_SIZE bsize,
- int mi_row, int mi_col);
+ MACROBLOCK *x,
+ BLOCK_SIZE bsize,
+ int mi_row, int mi_col);
-typedef int (integer_mv_pattern_search_fn) (
- const MACROBLOCK *x,
- MV *ref_mv,
- int search_param,
- int error_per_bit,
- int do_init_search,
- int *cost_list,
- const vp10_variance_fn_ptr_t *vf,
- int use_mvcost,
- const MV *center_mv,
- MV *best_mv);
-integer_mv_pattern_search_fn vp10_hex_search;
-integer_mv_pattern_search_fn vp10_bigdia_search;
-integer_mv_pattern_search_fn vp10_square_search;
-integer_mv_pattern_search_fn vp10_fast_hex_search;
-integer_mv_pattern_search_fn vp10_fast_dia_search;
+int vp10_hex_search(MACROBLOCK *x,
+ MV *start_mv,
+ int search_param,
+ int sad_per_bit,
+ int do_init_search,
+ int *cost_list,
+ const vp10_variance_fn_ptr_t *vfp,
+ int use_mvcost,
+ const MV *center_mv);
typedef int (fractional_mv_step_fp) (
- const MACROBLOCK *x,
- MV *bestmv, const MV *ref_mv,
+ MACROBLOCK *x,
+ const MV *ref_mv,
int allow_hp,
int error_per_bit,
const vp10_variance_fn_ptr_t *vfp,
@@ -124,39 +117,32 @@
extern fractional_mv_step_fp vp10_find_best_sub_pixel_tree_pruned_evenmore;
typedef int (*vp10_full_search_fn_t)(const MACROBLOCK *x,
- const MV *ref_mv, int sad_per_bit,
- int distance,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, MV *best_mv);
+ const MV *ref_mv, int sad_per_bit,
+ int distance,
+ const vp10_variance_fn_ptr_t *fn_ptr,
+ const MV *center_mv, MV *best_mv);
-typedef int (*vp10_refining_search_fn_t)(const MACROBLOCK *x,
- MV *ref_mv, int sad_per_bit,
- int distance,
+typedef int (*vp10_diamond_search_fn_t)(const MACROBLOCK *x,
+ const search_site_config *cfg,
+ MV *ref_mv, MV *best_mv,
+ int search_param, int sad_per_bit,
+ int *num00,
const vp10_variance_fn_ptr_t *fn_ptr,
const MV *center_mv);
-typedef int (*vp10_diamond_search_fn_t)(const MACROBLOCK *x,
- const search_site_config *cfg,
- MV *ref_mv, MV *best_mv,
- int search_param, int sad_per_bit,
- int *num00,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv);
-
-int vp10_refining_search_8p_c(const MACROBLOCK *x,
- MV *ref_mv, int error_per_bit,
- int search_range,
- const vp10_variance_fn_ptr_t *fn_ptr,
- const MV *center_mv, const uint8_t *second_pred);
+int vp10_refining_search_8p_c(MACROBLOCK *x,
+ int error_per_bit,
+ int search_range,
+ const vp10_variance_fn_ptr_t *fn_ptr,
+ const MV *center_mv, const uint8_t *second_pred);
struct VP10_COMP;
int vp10_full_pixel_search(struct VP10_COMP *cpi, MACROBLOCK *x,
- BLOCK_SIZE bsize, MV *mvp_full,
- int step_param, int error_per_bit,
- int *cost_list,
- const MV *ref_mv, MV *tmp_mv,
- int var_max, int rd);
+ BLOCK_SIZE bsize, MV *mvp_full,
+ int step_param, int error_per_bit,
+ int *cost_list, const MV *ref_mv,
+ int var_max, int rd);
#if CONFIG_EXT_INTER
int vp10_find_best_masked_sub_pixel_tree(const MACROBLOCK *x,
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index 0ebdfd7..45c10cd 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -4754,7 +4754,7 @@
struct buf_2d ref_yv12[2];
int bestsme = INT_MAX;
int sadpb = x->sadperbit16;
- MV tmp_mv;
+ MV *const best_mv = &x->best_mv.as_mv;
int search_range = 3;
int tmp_col_min = x->mv_col_min;
@@ -4815,23 +4815,22 @@
vp10_set_mv_search_range(x, &ref_mv[id].as_mv);
// Use the mv result from the single mode as mv predictor.
- tmp_mv = frame_mv[refs[id]].as_mv;
+ *best_mv = frame_mv[refs[id]].as_mv;
- tmp_mv.col >>= 3;
- tmp_mv.row >>= 3;
+ best_mv->col >>= 3;
+ best_mv->row >>= 3;
#if CONFIG_REF_MV
vp10_set_mvcost(x, refs[id]);
#endif
// Small-range full-pixel motion search.
- bestsme = vp10_refining_search_8p_c(x, &tmp_mv, sadpb,
- search_range,
- &cpi->fn_ptr[bsize],
- &ref_mv[id].as_mv, second_pred);
+ bestsme = vp10_refining_search_8p_c(x, sadpb, search_range,
+ &cpi->fn_ptr[bsize],
+ &ref_mv[id].as_mv, second_pred);
if (bestsme < INT_MAX)
- bestsme = vp10_get_mvpred_av_var(x, &tmp_mv, &ref_mv[id].as_mv,
- second_pred, &cpi->fn_ptr[bsize], 1);
+ bestsme = vp10_get_mvpred_av_var(x, best_mv, &ref_mv[id].as_mv,
+ second_pred, &cpi->fn_ptr[bsize], 1);
x->mv_col_min = tmp_col_min;
x->mv_col_max = tmp_col_max;
@@ -4860,8 +4859,7 @@
pd->pre[0].stride)) << 3];
bestsme = cpi->find_fractional_mv_step(
- x, &tmp_mv,
- &ref_mv[id].as_mv,
+ x, &ref_mv[id].as_mv,
cpi->common.allow_high_precision_mv,
x->errorperbit,
&cpi->fn_ptr[bsize],
@@ -4876,8 +4874,7 @@
} else {
(void) block;
bestsme = cpi->find_fractional_mv_step(
- x, &tmp_mv,
- &ref_mv[id].as_mv,
+ x, &ref_mv[id].as_mv,
cpi->common.allow_high_precision_mv,
x->errorperbit,
&cpi->fn_ptr[bsize],
@@ -4894,7 +4891,7 @@
xd->plane[0].pre[0] = ref_yv12[0];
if (bestsme < last_besterr[id]) {
- frame_mv[refs[id]].as_mv = tmp_mv;
+ frame_mv[refs[id]].as_mv = *best_mv;
last_besterr[id] = bestsme;
} else {
break;
@@ -5197,11 +5194,6 @@
run_mv_search)
#endif // CONFIG_EXT_INTER
) {
-#if CONFIG_EXT_INTER
- MV *const new_mv = &mode_mv[this_mode][0].as_mv;
-#else
- MV *const new_mv = &mode_mv[NEWMV][0].as_mv;
-#endif // CONFIG_EXT_INTER
int step_param = 0;
int bestsme = INT_MAX;
int sadpb = x->sadperbit4;
@@ -5269,8 +5261,7 @@
bestsme = vp10_full_pixel_search(
cpi, x, bsize, &mvp_full, step_param, sadpb,
cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? cost_list : NULL,
- &bsi->ref_mv[0]->as_mv, new_mv,
- INT_MAX, 1);
+ &bsi->ref_mv[0]->as_mv, INT_MAX, 1);
if (bestsme < INT_MAX) {
int distortion;
@@ -5295,9 +5286,7 @@
pd->pre[0].stride)) << 3];
cpi->find_fractional_mv_step(
- x,
- new_mv,
- &bsi->ref_mv[0]->as_mv,
+ x, &bsi->ref_mv[0]->as_mv,
cm->allow_high_precision_mv,
x->errorperbit, &cpi->fn_ptr[bsize],
cpi->sf.mv.subpel_force_stop,
@@ -5312,9 +5301,7 @@
pd->pre[0] = backup_pred;
} else {
cpi->find_fractional_mv_step(
- x,
- new_mv,
- &bsi->ref_mv[0]->as_mv,
+ x, &bsi->ref_mv[0]->as_mv,
cm->allow_high_precision_mv,
x->errorperbit, &cpi->fn_ptr[bsize],
cpi->sf.mv.subpel_force_stop,
@@ -5328,14 +5315,20 @@
// save motion search result for use in compound prediction
#if CONFIG_EXT_INTER
- seg_mvs[i][mv_idx][mbmi->ref_frame[0]].as_mv = *new_mv;
+ seg_mvs[i][mv_idx][mbmi->ref_frame[0]].as_mv = x->best_mv.as_mv;
#else
- seg_mvs[i][mbmi->ref_frame[0]].as_mv = *new_mv;
+ seg_mvs[i][mbmi->ref_frame[0]].as_mv = x->best_mv.as_mv;
#endif // CONFIG_EXT_INTER
}
if (cpi->sf.adaptive_motion_search)
- x->pred_mv[mbmi->ref_frame[0]] = *new_mv;
+ x->pred_mv[mbmi->ref_frame[0]] = x->best_mv.as_mv;
+
+#if CONFIG_EXT_INTER
+ mode_mv[this_mode][0] = x->best_mv;
+#else
+ mode_mv[NEWMV][0] = x->best_mv;
+#endif // CONFIG_EXT_INTER
// restore src pointers
mi_buf_restore(x, orig_src, orig_pre);
@@ -5904,7 +5897,7 @@
int ref_idx,
int mv_idx,
#endif // CONFIG_EXT_INTER
- int_mv *tmp_mv, int *rate_mv) {
+ int *rate_mv) {
MACROBLOCKD *xd = &x->e_mbd;
const VP10_COMMON *cm = &cpi->common;
MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
@@ -5986,7 +5979,7 @@
if ((x->pred_mv_sad[ref] >> 3) > x->pred_mv_sad[i]) {
x->pred_mv[ref].row = 0;
x->pred_mv[ref].col = 0;
- tmp_mv->as_int = INVALID_MV;
+ x->best_mv.as_int = INVALID_MV;
if (scaled_ref_frame) {
int i;
@@ -6006,7 +5999,7 @@
bestsme = vp10_full_pixel_search(cpi, x, bsize, &mvp_full, step_param, sadpb,
cond_cost_list(cpi, cost_list),
- &ref_mv, &tmp_mv->as_mv, INT_MAX, 1);
+ &ref_mv, INT_MAX, 1);
x->mv_col_min = tmp_col_min;
x->mv_col_max = tmp_col_max;
@@ -6028,7 +6021,7 @@
upsampled_ref->y_stride, (mi_row << 3), (mi_col << 3),
NULL, pd->subsampling_x, pd->subsampling_y);
- bestsme = cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
+ bestsme = cpi->find_fractional_mv_step(x, &ref_mv,
cm->allow_high_precision_mv,
x->errorperbit,
&cpi->fn_ptr[bsize],
@@ -6042,7 +6035,7 @@
// Restore the reference frames.
pd->pre[ref_idx] = backup_pred;
} else {
- cpi->find_fractional_mv_step(x, &tmp_mv->as_mv, &ref_mv,
+ cpi->find_fractional_mv_step(x, &ref_mv,
cm->allow_high_precision_mv,
x->errorperbit,
&cpi->fn_ptr[bsize],
@@ -6053,11 +6046,11 @@
&dis, &x->pred_sse[ref], NULL, 0, 0, 0);
}
}
- *rate_mv = vp10_mv_bit_cost(&tmp_mv->as_mv, &ref_mv,
+ *rate_mv = vp10_mv_bit_cost(&x->best_mv.as_mv, &ref_mv,
x->nmvjointcost, x->mvcost, MV_COST_WEIGHT);
if (cpi->sf.adaptive_motion_search)
- x->pred_mv[ref] = tmp_mv->as_mv;
+ x->pred_mv[ref] = x->best_mv.as_mv;
if (scaled_ref_frame) {
int i;
@@ -6991,34 +6984,32 @@
}
#endif // CONFIG_EXT_INTER
} else {
- int_mv tmp_mv;
-
#if CONFIG_EXT_INTER
if (is_comp_interintra_pred) {
- tmp_mv = single_newmvs[mv_idx][refs[0]];
+ x->best_mv = single_newmvs[mv_idx][refs[0]];
rate_mv = single_newmvs_rate[mv_idx][refs[0]];
} else {
single_motion_search(cpi, x, bsize, mi_row, mi_col,
- 0, mv_idx, &tmp_mv, &rate_mv);
- single_newmvs[mv_idx][refs[0]] = tmp_mv;
+ 0, mv_idx, &rate_mv);
+ single_newmvs[mv_idx][refs[0]] = x->best_mv;
single_newmvs_rate[mv_idx][refs[0]] = rate_mv;
}
#else
- single_motion_search(cpi, x, bsize, mi_row, mi_col, &tmp_mv, &rate_mv);
- single_newmv[refs[0]] = tmp_mv;
+ single_motion_search(cpi, x, bsize, mi_row, mi_col, &rate_mv);
+ single_newmv[refs[0]] = x->best_mv;
#endif // CONFIG_EXT_INTER
- if (tmp_mv.as_int == INVALID_MV)
+ if (x->best_mv.as_int == INVALID_MV)
return INT64_MAX;
- frame_mv[refs[0]] = tmp_mv;
- xd->mi[0]->bmi[0].as_mv[0] = tmp_mv;
+ frame_mv[refs[0]] = x->best_mv;
+ xd->mi[0]->bmi[0].as_mv[0] = x->best_mv;
// Estimate the rate implications of a new mv but discount this
// under certain circumstances where we want to help initiate a weak
// motion field, where the distortion gain for a single block may not
// be enough to overcome the cost of a new mv.
- if (discount_newmv_test(cpi, this_mode, tmp_mv, mode_mv, refs[0])) {
+ if (discount_newmv_test(cpi, this_mode, x->best_mv, mode_mv, refs[0])) {
rate_mv = VPXMAX((rate_mv / NEW_MV_DISCOUNT_FACTOR), 1);
}
}
diff --git a/vp10/encoder/temporal_filter.c b/vp10/encoder/temporal_filter.c
index 39be057..d125dae 100644
--- a/vp10/encoder/temporal_filter.c
+++ b/vp10/encoder/temporal_filter.c
@@ -288,7 +288,6 @@
MV best_ref_mv1 = {0, 0};
MV best_ref_mv1_full; /* full-pixel value of best_ref_mv1 */
- MV *ref_mv = &x->e_mbd.mi[0]->bmi[0].as_mv[0].as_mv;
// Save input state
struct buf_2d src = x->plane[0].src;
@@ -315,12 +314,11 @@
// Ignore mv costing by sending NULL pointer instead of cost arrays
vp10_hex_search(x, &best_ref_mv1_full, step_param, sadpb, 1,
- cond_cost_list(cpi, cost_list),
- &cpi->fn_ptr[BLOCK_16X16], 0, &best_ref_mv1, ref_mv);
+ cond_cost_list(cpi, cost_list),
+ &cpi->fn_ptr[BLOCK_16X16], 0, &best_ref_mv1);
// Ignore mv costing by sending NULL pointer instead of cost array
- bestsme = cpi->find_fractional_mv_step(x, ref_mv,
- &best_ref_mv1,
+ bestsme = cpi->find_fractional_mv_step(x, &best_ref_mv1,
cpi->common.allow_high_precision_mv,
x->errorperbit,
&cpi->fn_ptr[BLOCK_16X16],
@@ -329,6 +327,8 @@
NULL, NULL,
&distortion, &sse, NULL, 0, 0, 0);
+ x->e_mbd.mi[0]->bmi[0].as_mv[0] = x->best_mv;
+
// Restore input state
x->plane[0].src = src;
xd->plane[0].pre[0] = pre;
