Merge "Moving get_scan() call out of decode_coeffs() function."
diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c
index 06df850..501bed5 100644
--- a/vp9/common/x86/vp9_idct_intrin_sse2.c
+++ b/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -459,7 +459,9 @@
res1 = _mm_packs_epi32(tmp2, tmp3); \
}
-#define IDCT8_1D \
+#define IDCT8_1D(in0, in1, in2, in3, in4, in5, in6, in7, \
+ out0, out1, out2, out3, out4, out5, out6, out7) \
+ { \
/* Stage1 */ \
{ \
const __m128i lo_17 = _mm_unpacklo_epi16(in1, in7); \
@@ -519,14 +521,15 @@
} \
\
/* Stage4 */ \
- in0 = _mm_adds_epi16(stp1_0, stp2_7); \
- in1 = _mm_adds_epi16(stp1_1, stp1_6); \
- in2 = _mm_adds_epi16(stp1_2, stp1_5); \
- in3 = _mm_adds_epi16(stp1_3, stp2_4); \
- in4 = _mm_subs_epi16(stp1_3, stp2_4); \
- in5 = _mm_subs_epi16(stp1_2, stp1_5); \
- in6 = _mm_subs_epi16(stp1_1, stp1_6); \
- in7 = _mm_subs_epi16(stp1_0, stp2_7);
+ out0 = _mm_adds_epi16(stp1_0, stp2_7); \
+ out1 = _mm_adds_epi16(stp1_1, stp1_6); \
+ out2 = _mm_adds_epi16(stp1_2, stp1_5); \
+ out3 = _mm_adds_epi16(stp1_3, stp2_4); \
+ out4 = _mm_subs_epi16(stp1_3, stp2_4); \
+ out5 = _mm_subs_epi16(stp1_2, stp1_5); \
+ out6 = _mm_subs_epi16(stp1_1, stp1_6); \
+ out7 = _mm_subs_epi16(stp1_0, stp2_7); \
+ }
#define RECON_AND_STORE(dest, in_x) \
{ \
@@ -574,7 +577,8 @@
in0, in1, in2, in3, in4, in5, in6, in7);
// 4-stage 1D idct8x8
- IDCT8_1D
+ IDCT8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
+ in0, in1, in2, in3, in4, in5, in6, in7);
}
// Final rounding and shift
@@ -697,15 +701,8 @@
in0, in1, in2, in3, in4, in5, in6, in7);
// 4-stage 1D idct8x8
- IDCT8_1D
- in[0] = in0;
- in[1] = in1;
- in[2] = in2;
- in[3] = in3;
- in[4] = in4;
- in[5] = in5;
- in[6] = in6;
- in[7] = in7;
+ IDCT8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
+ in[0], in[1], in[2], in[3], in[4], in[5], in[6], in[7]);
}
static void iadst8_1d_sse2(__m128i *in) {
@@ -1112,9 +1109,9 @@
tmp3 = _mm_subs_epi16(stp1_2, stp1_5);
TRANSPOSE_4X8_10(tmp0, tmp1, tmp2, tmp3, in0, in1, in2, in3)
- in4 = in5 = in6 = in7 = zero;
- IDCT8_1D
+ IDCT8_1D(in0, in1, in2, in3, zero, zero, zero, zero,
+ in0, in1, in2, in3, in4, in5, in6, in7);
// Final rounding and shift
in0 = _mm_adds_epi16(in0, final_rounding);
in1 = _mm_adds_epi16(in1, final_rounding);
diff --git a/vp9/encoder/vp9_firstpass.c b/vp9/encoder/vp9_firstpass.c
index 3df9a1f..ba0fea2 100644
--- a/vp9/encoder/vp9_firstpass.c
+++ b/vp9/encoder/vp9_firstpass.c
@@ -271,8 +271,17 @@
const FIRSTPASS_STATS *const stats = &cpi->twopass.total_stats;
const double av_err = stats->ssim_weighted_pred_err / stats->count;
const double this_err = this_frame->ssim_weighted_pred_err;
- return av_err * pow(this_err / DOUBLE_DIVIDE_CHECK(av_err),
- this_err > av_err ? POW1 : POW2);
+ double modified_error;
+
+ modified_error = av_err * pow(this_err / DOUBLE_DIVIDE_CHECK(av_err),
+ this_err > av_err ? POW1 : POW2);
+
+ if (modified_error < cpi->twopass.modified_error_min)
+ modified_error = cpi->twopass.modified_error_min;
+ else if (modified_error > cpi->twopass.modified_error_max)
+ modified_error = cpi->twopass.modified_error_max;
+
+ return modified_error;
}
static const double weight_table[256] = {
@@ -1077,13 +1086,6 @@
FIRSTPASS_STATS this_frame;
FIRSTPASS_STATS *start_pos;
- double lower_bounds_min_rate = FRAME_OVERHEAD_BITS * cpi->oxcf.framerate;
- double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth *
- cpi->oxcf.two_pass_vbrmin_section / 100);
-
- if (two_pass_min_rate < lower_bounds_min_rate)
- two_pass_min_rate = lower_bounds_min_rate;
-
zero_stats(&cpi->twopass.total_stats);
zero_stats(&cpi->twopass.total_left_stats);
@@ -1104,8 +1106,6 @@
cpi->output_framerate = cpi->oxcf.framerate;
cpi->twopass.bits_left = (int64_t)(cpi->twopass.total_stats.duration *
cpi->oxcf.target_bandwidth / 10000000.0);
- cpi->twopass.bits_left -= (int64_t)(cpi->twopass.total_stats.duration *
- two_pass_min_rate / 10000000.0);
// Calculate a minimum intra value to be used in determining the IIratio
// scores used in the second pass. We have this minimum to make sure
@@ -1142,9 +1142,16 @@
// Scan the first pass file and calculate a modified total error based upon
// the bias/power function used to allocate bits.
{
+ double av_error = cpi->twopass.total_stats.ssim_weighted_pred_err /
+ DOUBLE_DIVIDE_CHECK(cpi->twopass.total_stats.count);
+
start_pos = cpi->twopass.stats_in; // Note starting "file" position
cpi->twopass.modified_error_total = 0.0;
+ cpi->twopass.modified_error_min =
+ (av_error * cpi->oxcf.two_pass_vbrmin_section) / 100;
+ cpi->twopass.modified_error_max =
+ (av_error * cpi->oxcf.two_pass_vbrmax_section) / 100;
while (input_stats(cpi, &this_frame) != EOF) {
cpi->twopass.modified_error_total +=
@@ -2618,14 +2625,4 @@
#else
cpi->twopass.bits_left -= 8 * bytes_used;
#endif
- if (!cpi->refresh_alt_ref_frame) {
- double lower_bounds_min_rate = FRAME_OVERHEAD_BITS * cpi->oxcf.framerate;
- double two_pass_min_rate = (double)(cpi->oxcf.target_bandwidth *
- cpi->oxcf.two_pass_vbrmin_section
- / 100);
- if (two_pass_min_rate < lower_bounds_min_rate)
- two_pass_min_rate = lower_bounds_min_rate;
- cpi->twopass.bits_left += (int64_t)(two_pass_min_rate /
- cpi->oxcf.framerate);
- }
}
diff --git a/vp9/encoder/vp9_mbgraph.c b/vp9/encoder/vp9_mbgraph.c
index 32ab3fc..a9e0dfd 100644
--- a/vp9/encoder/vp9_mbgraph.c
+++ b/vp9/encoder/vp9_mbgraph.c
@@ -22,7 +22,7 @@
static unsigned int do_16x16_motion_iteration(VP9_COMP *cpi,
- int_mv *ref_mv,
+ const MV *ref_mv,
int_mv *dst_mv,
int mb_row,
int mb_col) {
@@ -42,15 +42,14 @@
(cpi->speed < 8 ? (cpi->speed > 5 ? 1 : 0) : 2);
step_param = MIN(step_param, (cpi->sf.max_step_search_steps - 2));
- vp9_set_mv_search_range(x, &ref_mv->as_mv);
+ vp9_set_mv_search_range(x, ref_mv);
- ref_full.as_mv.col = ref_mv->as_mv.col >> 3;
- ref_full.as_mv.row = ref_mv->as_mv.row >> 3;
+ ref_full.as_mv.col = ref_mv->col >> 3;
+ ref_full.as_mv.row = ref_mv->row >> 3;
/*cpi->sf.search_method == HEX*/
best_err = vp9_hex_search(x, &ref_full.as_mv, step_param, x->errorperbit,
- 0, &v_fn_ptr,
- 0, &ref_mv->as_mv, &dst_mv->as_mv);
+ 0, &v_fn_ptr, 0, ref_mv, &dst_mv->as_mv);
// Try sub-pixel MC
// if (bestsme > error_thresh && bestsme < INT_MAX)
@@ -59,7 +58,7 @@
unsigned int sse;
best_err = cpi->find_fractional_mv_step(
x,
- &dst_mv->as_mv, &ref_mv->as_mv,
+ &dst_mv->as_mv, ref_mv,
cpi->common.allow_high_precision_mv,
x->errorperbit, &v_fn_ptr,
0, cpi->sf.subpel_iters_per_step, NULL, NULL,
@@ -81,8 +80,8 @@
return best_err;
}
-static int do_16x16_motion_search(VP9_COMP *cpi, int_mv *ref_mv, int_mv *dst_mv,
- int mb_row, int mb_col) {
+static int do_16x16_motion_search(VP9_COMP *cpi, const int_mv *ref_mv,
+ int_mv *dst_mv, int mb_row, int mb_col) {
MACROBLOCK *const x = &cpi->mb;
MACROBLOCKD *const xd = &x->e_mbd;
unsigned int err, tmp_err;
@@ -97,7 +96,8 @@
// 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->as_mv, &tmp_mv,
+ mb_row, mb_col);
if (tmp_err < err) {
err = tmp_err;
dst_mv->as_int = tmp_mv.as_int;
@@ -110,7 +110,7 @@
int_mv zero_ref_mv, tmp_mv;
zero_ref_mv.as_int = 0;
- tmp_err = do_16x16_motion_iteration(cpi, &zero_ref_mv, &tmp_mv,
+ tmp_err = do_16x16_motion_iteration(cpi, &zero_ref_mv.as_mv, &tmp_mv,
mb_row, mb_col);
if (tmp_err < err) {
dst_mv->as_int = tmp_mv.as_int;
diff --git a/vp9/encoder/vp9_mcomp.c b/vp9/encoder/vp9_mcomp.c
index 5e74d1a..c199dff 100644
--- a/vp9/encoder/vp9_mcomp.c
+++ b/vp9/encoder/vp9_mcomp.c
@@ -23,7 +23,7 @@
// #define NEW_DIAMOND_SEARCH
-void vp9_set_mv_search_range(MACROBLOCK *x, MV *mv) {
+void vp9_set_mv_search_range(MACROBLOCK *x, const MV *mv) {
const int col_min = (mv->col >> 3) - MAX_FULL_PEL_VAL + (mv->col & 7 ? 1 : 0);
const int row_min = (mv->row >> 3) - MAX_FULL_PEL_VAL + (mv->row & 7 ? 1 : 0);
const int col_max = (mv->col >> 3) + MAX_FULL_PEL_VAL;
@@ -464,20 +464,20 @@
#undef CHECK_BETTER
#undef SP
-static INLINE int check_bounds(MACROBLOCK *x, int row, int col, int range) {
+static INLINE int check_bounds(const MACROBLOCK *x, int row, int col,
+ int range) {
return ((row - range) >= x->mv_row_min) &
((row + range) <= x->mv_row_max) &
((col - range) >= x->mv_col_min) &
((col + range) <= x->mv_col_max);
}
-#define CHECK_POINT \
- {\
- if (this_mv.col < x->mv_col_min) continue;\
- if (this_mv.col > x->mv_col_max) continue;\
- if (this_mv.row < x->mv_row_min) continue;\
- if (this_mv.row > x->mv_row_max) continue;\
- }
+static INLINE int check_point(const MACROBLOCK *x, const MV *mv) {
+ return (mv->col < x->mv_col_min) |
+ (mv->col > x->mv_col_max) |
+ (mv->row < x->mv_row_min) |
+ (mv->row > x->mv_row_max);
+}
#define CHECK_BETTER \
{\
@@ -580,7 +580,8 @@
for (i = 0; i < num_candidates[t]; i++) {
this_mv.row = br + candidates[t][i].row;
this_mv.col = bc + candidates[t][i].col;
- CHECK_POINT
+ if (check_point(x, &this_mv))
+ continue;
this_offset = base_offset + (this_mv.row * in_what_stride) +
this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
@@ -623,7 +624,8 @@
for (i = 0; i < num_candidates[s]; i++) {
this_mv.row = br + candidates[s][i].row;
this_mv.col = bc + candidates[s][i].col;
- CHECK_POINT
+ if (check_point(x, &this_mv))
+ continue;
this_offset = base_offset + (this_mv.row * in_what_stride) +
this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
@@ -659,7 +661,8 @@
for (i = 0; i < PATTERN_CANDIDATES_REF; i++) {
this_mv.row = br + candidates[s][next_chkpts_indices[i]].row;
this_mv.col = bc + candidates[s][next_chkpts_indices[i]].col;
- CHECK_POINT
+ if (check_point(x, &this_mv))
+ continue;
this_offset = base_offset + (this_mv.row * (in_what_stride)) +
this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
@@ -699,7 +702,8 @@
for (i = 0; i < 4; i++) {
this_mv.row = br + neighbors[i].row;
this_mv.col = bc + neighbors[i].col;
- CHECK_POINT
+ if (check_point(x, &this_mv))
+ continue;
this_offset = base_offset + (this_mv.row * (in_what_stride)) +
this_mv.col;
thissad = vfp->sdf(what, what_stride, this_offset, in_what_stride,
@@ -852,7 +856,6 @@
square_num_candidates, square_candidates);
};
-#undef CHECK_POINT
#undef CHECK_BETTER
int vp9_full_range_search_c(MACROBLOCK *x, MV *ref_mv, MV *best_mv,
@@ -1303,16 +1306,16 @@
refining search */
int vp9_full_pixel_diamond(VP9_COMP *cpi, MACROBLOCK *x,
- int_mv *mvp_full, int step_param,
+ MV *mvp_full, int step_param,
int sadpb, int further_steps,
int do_refine, vp9_variance_fn_ptr_t *fn_ptr,
- int_mv *ref_mv, int_mv *dst_mv) {
+ const MV *ref_mv, int_mv *dst_mv) {
int_mv temp_mv;
int thissme, n, num00;
- int bestsme = cpi->diamond_search_sad(x, &mvp_full->as_mv, &temp_mv.as_mv,
+ int bestsme = cpi->diamond_search_sad(x, mvp_full, &temp_mv.as_mv,
step_param, sadpb, &num00,
fn_ptr, x->nmvjointcost,
- x->mvcost, &ref_mv->as_mv);
+ x->mvcost, ref_mv);
dst_mv->as_int = temp_mv.as_int;
n = num00;
@@ -1329,10 +1332,10 @@
if (num00) {
num00--;
} else {
- thissme = cpi->diamond_search_sad(x, &mvp_full->as_mv, &temp_mv.as_mv,
+ thissme = cpi->diamond_search_sad(x, mvp_full, &temp_mv.as_mv,
step_param + n, sadpb, &num00,
fn_ptr, x->nmvjointcost, x->mvcost,
- &ref_mv->as_mv);
+ ref_mv);
/* check to see if refining search is needed. */
if (num00 > (further_steps - n))
@@ -1352,7 +1355,7 @@
best_mv.as_int = dst_mv->as_int;
thissme = cpi->refining_search_sad(x, &best_mv.as_mv, sadpb, search_range,
fn_ptr, x->nmvjointcost, x->mvcost,
- &ref_mv->as_mv);
+ ref_mv);
if (thissme < bestsme) {
bestsme = thissme;
diff --git a/vp9/encoder/vp9_mcomp.h b/vp9/encoder/vp9_mcomp.h
index 1cb2977..b3d8975 100644
--- a/vp9/encoder/vp9_mcomp.h
+++ b/vp9/encoder/vp9_mcomp.h
@@ -28,7 +28,7 @@
#define BORDER_MV_PIXELS_B16 (16 + VP9_INTERP_EXTEND)
-void vp9_set_mv_search_range(MACROBLOCK *x, MV *mv);
+void vp9_set_mv_search_range(MACROBLOCK *x, const MV *mv);
int vp9_mv_bit_cost(const MV *mv, const MV *ref,
const int *mvjcost, int *mvcost[2], int weight);
void vp9_init_dsmotion_compensation(MACROBLOCK *x, int stride);
@@ -39,10 +39,10 @@
// Runs sequence of diamond searches in smaller steps for RD
int vp9_full_pixel_diamond(struct VP9_COMP *cpi, MACROBLOCK *x,
- int_mv *mvp_full, int step_param,
+ MV *mvp_full, int step_param,
int sadpb, int further_steps, int do_refine,
vp9_variance_fn_ptr_t *fn_ptr,
- int_mv *ref_mv, int_mv *dst_mv);
+ const MV *ref_mv, int_mv *dst_mv);
int vp9_hex_search(MACROBLOCK *x,
MV *ref_mv,
diff --git a/vp9/encoder/vp9_onyx_int.h b/vp9/encoder/vp9_onyx_int.h
index f5b6562..d32711c 100644
--- a/vp9/encoder/vp9_onyx_int.h
+++ b/vp9/encoder/vp9_onyx_int.h
@@ -525,6 +525,8 @@
int64_t bits_left;
int64_t clip_bits_total;
double avg_iiratio;
+ double modified_error_min;
+ double modified_error_max;
double modified_error_total;
double modified_error_left;
double kf_intra_err_min;
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index 22e480b..46fe163 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1742,7 +1742,7 @@
int further_steps;
int thissme, bestsme = INT_MAX;
int sadpb = x->sadperbit4;
- int_mv mvp_full;
+ MV mvp_full;
int max_mv;
/* Is the best so far sufficiently good that we cant justify doing
@@ -1773,12 +1773,12 @@
step_param = cpi->mv_step_param;
}
- mvp_full.as_mv.row = bsi->mvp.as_mv.row >> 3;
- mvp_full.as_mv.col = bsi->mvp.as_mv.col >> 3;
+ mvp_full.row = bsi->mvp.as_mv.row >> 3;
+ mvp_full.col = bsi->mvp.as_mv.col >> 3;
if (cpi->sf.adaptive_motion_search && cpi->common.show_frame) {
- mvp_full.as_mv.row = x->pred_mv[mbmi->ref_frame[0]].as_mv.row >> 3;
- mvp_full.as_mv.col = x->pred_mv[mbmi->ref_frame[0]].as_mv.col >> 3;
+ mvp_full.row = x->pred_mv[mbmi->ref_frame[0]].as_mv.row >> 3;
+ mvp_full.col = x->pred_mv[mbmi->ref_frame[0]].as_mv.col >> 3;
step_param = MAX(step_param, 8);
}
@@ -1789,19 +1789,19 @@
vp9_set_mv_search_range(x, &bsi->ref_mv->as_mv);
if (cpi->sf.search_method == HEX) {
- bestsme = vp9_hex_search(x, &mvp_full.as_mv,
+ bestsme = vp9_hex_search(x, &mvp_full,
step_param,
sadpb, 1, v_fn_ptr, 1,
&bsi->ref_mv->as_mv,
&mode_mv[NEWMV].as_mv);
} else if (cpi->sf.search_method == SQUARE) {
- bestsme = vp9_square_search(x, &mvp_full.as_mv,
+ bestsme = vp9_square_search(x, &mvp_full,
step_param,
sadpb, 1, v_fn_ptr, 1,
&bsi->ref_mv->as_mv,
&mode_mv[NEWMV].as_mv);
} else if (cpi->sf.search_method == BIGDIA) {
- bestsme = vp9_bigdia_search(x, &mvp_full.as_mv,
+ bestsme = vp9_bigdia_search(x, &mvp_full,
step_param,
sadpb, 1, v_fn_ptr, 1,
&bsi->ref_mv->as_mv,
@@ -1809,16 +1809,17 @@
} else {
bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
sadpb, further_steps, 0, v_fn_ptr,
- bsi->ref_mv, &mode_mv[NEWMV]);
+ &bsi->ref_mv->as_mv,
+ &mode_mv[NEWMV]);
}
// Should we do a full search (best quality only)
if (cpi->compressor_speed == 0) {
/* Check if mvp_full is within the range. */
- clamp_mv(&mvp_full.as_mv, x->mv_col_min, x->mv_col_max,
+ clamp_mv(&mvp_full, x->mv_col_min, x->mv_col_max,
x->mv_row_min, x->mv_row_max);
- thissme = cpi->full_search_sad(x, &mvp_full.as_mv,
+ thissme = cpi->full_search_sad(x, &mvp_full,
sadpb, 16, v_fn_ptr,
x->nmvjointcost, x->mvcost,
&bsi->ref_mv->as_mv, i);
@@ -2439,10 +2440,10 @@
&cpi->fn_ptr[bsize], 1,
&ref_mv.as_mv, &tmp_mv->as_mv);
} else {
- bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full, step_param,
+ bestsme = vp9_full_pixel_diamond(cpi, x, &mvp_full.as_mv, step_param,
sadpb, further_steps, 1,
&cpi->fn_ptr[bsize],
- &ref_mv, tmp_mv);
+ &ref_mv.as_mv, tmp_mv);
}
x->mv_col_min = tmp_col_min;
