summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--vp8/encoder/x86/denoising_sse2.c8
-rw-r--r--vp9/common/arm/neon/vp9_convolve8_avg_neon.asm16
-rw-r--r--vp9/common/arm/neon/vp9_convolve8_neon.asm16
-rw-r--r--vp9/common/vp9_entropy.c10
-rw-r--r--vp9/common/vp9_entropymode.c8
-rw-r--r--vp9/common/vp9_entropymv.c4
-rw-r--r--vp9/common/vp9_onyxc_int.h6
-rw-r--r--vp9/common/vp9_pred_common.h12
-rw-r--r--vp9/decoder/vp9_decodeframe.c8
-rw-r--r--vp9/decoder/vp9_decodemv.c20
-rw-r--r--vp9/decoder/vp9_decoder.c12
-rw-r--r--vp9/decoder/vp9_detokenize.c2
-rw-r--r--vp9/encoder/vp9_bitstream.c32
-rw-r--r--vp9/encoder/vp9_encodemv.c2
-rw-r--r--vp9/encoder/vp9_encoder.c19
-rw-r--r--vp9/encoder/vp9_pickmode.c2
-rw-r--r--vp9/encoder/vp9_rd.c8
-rw-r--r--vp9/encoder/vp9_rdopt.c2
-rw-r--r--vp9/encoder/vp9_tokenize.c2
-rw-r--r--vp9/encoder/x86/vp9_denoiser_sse2.c132
20 files changed, 164 insertions, 157 deletions
diff --git a/vp8/encoder/x86/denoising_sse2.c b/vp8/encoder/x86/denoising_sse2.c
index 3a4cf7ee7..101d646ef 100644
--- a/vp8/encoder/x86/denoising_sse2.c
+++ b/vp8/encoder/x86/denoising_sse2.c
@@ -121,12 +121,12 @@ int vp8_denoiser_filter_sse2(unsigned char *mc_running_avg_y,
if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH;
if (abs_sum_diff > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising),
- // checK if we can still apply some (weaker) temporal filtering to
+ // check if we can still apply some (weaker) temporal filtering to
// this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it
// closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be
- // within the accceptable range given by sum_diff_thresh.
+ // within the acceptable range given by sum_diff_thresh.
// The delta is set by the excess of absolute pixel diff over the
// threshold.
@@ -302,12 +302,12 @@ int vp8_denoiser_filter_uv_sse2(unsigned char *mc_running_avg,
if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH_UV;
if (abs_sum_diff > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising),
- // checK if we can still apply some (weaker) temporal filtering to
+ // check if we can still apply some (weaker) temporal filtering to
// this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it
// closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be
- // within the accceptable range given by sum_diff_thresh.
+ // within the acceptable range given by sum_diff_thresh.
// The delta is set by the excess of absolute pixel diff over the
// threshold.
diff --git a/vp9/common/arm/neon/vp9_convolve8_avg_neon.asm b/vp9/common/arm/neon/vp9_convolve8_avg_neon.asm
index 6b20cb9bf..4d85846f0 100644
--- a/vp9/common/arm/neon/vp9_convolve8_avg_neon.asm
+++ b/vp9/common/arm/neon/vp9_convolve8_avg_neon.asm
@@ -78,7 +78,7 @@
mov r10, r6 ; w loop counter
-loop_horiz_v
+vp9_convolve8_avg_loop_horiz_v
vld1.8 {d24}, [r0], r1
vld1.8 {d25}, [r0], r1
vld1.8 {d26}, [r0], r1
@@ -101,7 +101,7 @@ loop_horiz_v
add r0, r0, #3
-loop_horiz
+vp9_convolve8_avg_loop_horiz
add r5, r0, #64
vld1.32 {d28[]}, [r0], r1
@@ -170,14 +170,14 @@ loop_horiz
vmov q9, q13
subs r6, r6, #4 ; w -= 4
- bgt loop_horiz
+ bgt vp9_convolve8_avg_loop_horiz
; outer loop
mov r6, r10 ; restore w counter
add r0, r0, r9 ; src += src_stride * 4 - w
add r2, r2, r12 ; dst += dst_stride * 4 - w
subs r7, r7, #4 ; h -= 4
- bgt loop_horiz_v
+ bgt vp9_convolve8_avg_loop_horiz_v
pop {r4-r10, pc}
@@ -203,7 +203,7 @@ loop_horiz
lsl r1, r1, #1
lsl r3, r3, #1
-loop_vert_h
+vp9_convolve8_avg_loop_vert_h
mov r4, r0
add r7, r0, r1, asr #1
mov r5, r2
@@ -223,7 +223,7 @@ loop_vert_h
vmovl.u8 q10, d20
vmovl.u8 q11, d22
-loop_vert
+vp9_convolve8_avg_loop_vert
; always process a 4x4 block at a time
vld1.u32 {d24[0]}, [r7], r1
vld1.u32 {d26[0]}, [r4], r1
@@ -288,13 +288,13 @@ loop_vert
vmov d22, d25
subs r12, r12, #4 ; h -= 4
- bgt loop_vert
+ bgt vp9_convolve8_avg_loop_vert
; outer loop
add r0, r0, #4
add r2, r2, #4
subs r6, r6, #4 ; w -= 4
- bgt loop_vert_h
+ bgt vp9_convolve8_avg_loop_vert_h
pop {r4-r8, pc}
diff --git a/vp9/common/arm/neon/vp9_convolve8_neon.asm b/vp9/common/arm/neon/vp9_convolve8_neon.asm
index 45258454c..184c3ad67 100644
--- a/vp9/common/arm/neon/vp9_convolve8_neon.asm
+++ b/vp9/common/arm/neon/vp9_convolve8_neon.asm
@@ -78,7 +78,7 @@
mov r10, r6 ; w loop counter
-loop_horiz_v
+vp9_convolve8_loop_horiz_v
vld1.8 {d24}, [r0], r1
vld1.8 {d25}, [r0], r1
vld1.8 {d26}, [r0], r1
@@ -101,7 +101,7 @@ loop_horiz_v
add r0, r0, #3
-loop_horiz
+vp9_convolve8_loop_horiz
add r5, r0, #64
vld1.32 {d28[]}, [r0], r1
@@ -159,14 +159,14 @@ loop_horiz
vmov q9, q13
subs r6, r6, #4 ; w -= 4
- bgt loop_horiz
+ bgt vp9_convolve8_loop_horiz
; outer loop
mov r6, r10 ; restore w counter
add r0, r0, r9 ; src += src_stride * 4 - w
add r2, r2, r12 ; dst += dst_stride * 4 - w
subs r7, r7, #4 ; h -= 4
- bgt loop_horiz_v
+ bgt vp9_convolve8_loop_horiz_v
pop {r4-r10, pc}
@@ -192,7 +192,7 @@ loop_horiz
lsl r1, r1, #1
lsl r3, r3, #1
-loop_vert_h
+vp9_convolve8_loop_vert_h
mov r4, r0
add r7, r0, r1, asr #1
mov r5, r2
@@ -212,7 +212,7 @@ loop_vert_h
vmovl.u8 q10, d20
vmovl.u8 q11, d22
-loop_vert
+vp9_convolve8_loop_vert
; always process a 4x4 block at a time
vld1.u32 {d24[0]}, [r7], r1
vld1.u32 {d26[0]}, [r4], r1
@@ -266,13 +266,13 @@ loop_vert
vmov d22, d25
subs r12, r12, #4 ; h -= 4
- bgt loop_vert
+ bgt vp9_convolve8_loop_vert
; outer loop
add r0, r0, #4
add r2, r2, #4
subs r6, r6, #4 ; w -= 4
- bgt loop_vert_h
+ bgt vp9_convolve8_loop_vert_h
pop {r4-r8, pc}
diff --git a/vp9/common/vp9_entropy.c b/vp9/common/vp9_entropy.c
index c3fdeb48a..77a8709f0 100644
--- a/vp9/common/vp9_entropy.c
+++ b/vp9/common/vp9_entropy.c
@@ -748,10 +748,10 @@ void vp9_model_to_full_probs(const vp9_prob *model, vp9_prob *full) {
}
void vp9_default_coef_probs(VP9_COMMON *cm) {
- vp9_copy(cm->fc.coef_probs[TX_4X4], default_coef_probs_4x4);
- vp9_copy(cm->fc.coef_probs[TX_8X8], default_coef_probs_8x8);
- vp9_copy(cm->fc.coef_probs[TX_16X16], default_coef_probs_16x16);
- vp9_copy(cm->fc.coef_probs[TX_32X32], default_coef_probs_32x32);
+ vp9_copy(cm->fc->coef_probs[TX_4X4], default_coef_probs_4x4);
+ vp9_copy(cm->fc->coef_probs[TX_8X8], default_coef_probs_8x8);
+ vp9_copy(cm->fc->coef_probs[TX_16X16], default_coef_probs_16x16);
+ vp9_copy(cm->fc->coef_probs[TX_32X32], default_coef_probs_32x32);
}
#define COEF_COUNT_SAT 24
@@ -765,7 +765,7 @@ static void adapt_coef_probs(VP9_COMMON *cm, TX_SIZE tx_size,
unsigned int count_sat,
unsigned int update_factor) {
const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
- vp9_coeff_probs_model *const probs = cm->fc.coef_probs[tx_size];
+ vp9_coeff_probs_model *const probs = cm->fc->coef_probs[tx_size];
const vp9_coeff_probs_model *const pre_probs = pre_fc->coef_probs[tx_size];
vp9_coeff_count_model *counts = cm->counts.coef[tx_size];
unsigned int (*eob_counts)[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] =
diff --git a/vp9/common/vp9_entropymode.c b/vp9/common/vp9_entropymode.c
index 5b00b0082..1a24572ba 100644
--- a/vp9/common/vp9_entropymode.c
+++ b/vp9/common/vp9_entropymode.c
@@ -350,7 +350,7 @@ static void adapt_probs(const vp9_tree_index *tree,
void vp9_adapt_mode_probs(VP9_COMMON *cm) {
int i, j;
- FRAME_CONTEXT *fc = &cm->fc;
+ FRAME_CONTEXT *fc = cm->fc;
const FRAME_CONTEXT *pre_fc = &cm->frame_contexts[cm->frame_context_idx];
const FRAME_COUNTS *counts = &cm->counts;
@@ -451,17 +451,17 @@ void vp9_setup_past_independence(VP9_COMMON *cm) {
lf->last_sharpness_level = -1;
vp9_default_coef_probs(cm);
- vp9_init_mode_probs(&cm->fc);
+ vp9_init_mode_probs(cm->fc);
vp9_init_mv_probs(cm);
if (cm->frame_type == KEY_FRAME ||
cm->error_resilient_mode || cm->reset_frame_context == 3) {
// Reset all frame contexts.
for (i = 0; i < FRAME_CONTEXTS; ++i)
- cm->frame_contexts[i] = cm->fc;
+ cm->frame_contexts[i] = *cm->fc;
} else if (cm->reset_frame_context == 2) {
// Reset only the frame context specified in the frame header.
- cm->frame_contexts[cm->frame_context_idx] = cm->fc;
+ cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
}
if (frame_is_intra_only(cm))
diff --git a/vp9/common/vp9_entropymv.c b/vp9/common/vp9_entropymv.c
index 5bb048202..922c03947 100644
--- a/vp9/common/vp9_entropymv.c
+++ b/vp9/common/vp9_entropymv.c
@@ -196,7 +196,7 @@ static void adapt_probs(const vp9_tree_index *tree, const vp9_prob *pre_probs,
void vp9_adapt_mv_probs(VP9_COMMON *cm, int allow_hp) {
int i, j;
- nmv_context *fc = &cm->fc.nmvc;
+ nmv_context *fc = &cm->fc->nmvc;
const nmv_context *pre_fc = &cm->frame_contexts[cm->frame_context_idx].nmvc;
const nmv_context_counts *counts = &cm->counts.mv;
@@ -229,5 +229,5 @@ void vp9_adapt_mv_probs(VP9_COMMON *cm, int allow_hp) {
}
void vp9_init_mv_probs(VP9_COMMON *cm) {
- cm->fc.nmvc = default_nmv_context;
+ cm->fc->nmvc = default_nmv_context;
}
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index f1eda9117..b818ae818 100644
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -169,8 +169,8 @@ typedef struct VP9Common {
MV_REFERENCE_FRAME comp_var_ref[2];
REFERENCE_MODE reference_mode;
- FRAME_CONTEXT fc; /* this frame entropy */
- FRAME_CONTEXT frame_contexts[FRAME_CONTEXTS];
+ FRAME_CONTEXT *fc; /* this frame entropy */
+ FRAME_CONTEXT *frame_contexts; // FRAME_CONTEXTS
unsigned int frame_context_idx; /* Context to use/update */
FRAME_COUNTS counts;
@@ -261,7 +261,7 @@ static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) {
static INLINE const vp9_prob* get_partition_probs(const VP9_COMMON *cm,
int ctx) {
return frame_is_intra_only(cm) ? vp9_kf_partition_probs[ctx]
- : cm->fc.partition_prob[ctx];
+ : cm->fc->partition_prob[ctx];
}
static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col) {
diff --git a/vp9/common/vp9_pred_common.h b/vp9/common/vp9_pred_common.h
index 39774f142..cf13e4a91 100644
--- a/vp9/common/vp9_pred_common.h
+++ b/vp9/common/vp9_pred_common.h
@@ -54,7 +54,7 @@ static INLINE int vp9_get_skip_context(const MACROBLOCKD *xd) {
static INLINE vp9_prob vp9_get_skip_prob(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
- return cm->fc.skip_probs[vp9_get_skip_context(xd)];
+ return cm->fc->skip_probs[vp9_get_skip_context(xd)];
}
int vp9_get_pred_context_switchable_interp(const MACROBLOCKD *xd);
@@ -63,14 +63,14 @@ int vp9_get_intra_inter_context(const MACROBLOCKD *xd);
static INLINE vp9_prob vp9_get_intra_inter_prob(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
- return cm->fc.intra_inter_prob[vp9_get_intra_inter_context(xd)];
+ return cm->fc->intra_inter_prob[vp9_get_intra_inter_context(xd)];
}
int vp9_get_reference_mode_context(const VP9_COMMON *cm, const MACROBLOCKD *xd);
static INLINE vp9_prob vp9_get_reference_mode_prob(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
- return cm->fc.comp_inter_prob[vp9_get_reference_mode_context(cm, xd)];
+ return cm->fc->comp_inter_prob[vp9_get_reference_mode_context(cm, xd)];
}
int vp9_get_pred_context_comp_ref_p(const VP9_COMMON *cm,
@@ -79,21 +79,21 @@ int vp9_get_pred_context_comp_ref_p(const VP9_COMMON *cm,
static INLINE vp9_prob vp9_get_pred_prob_comp_ref_p(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
const int pred_context = vp9_get_pred_context_comp_ref_p(cm, xd);
- return cm->fc.comp_ref_prob[pred_context];
+ return cm->fc->comp_ref_prob[pred_context];
}
int vp9_get_pred_context_single_ref_p1(const MACROBLOCKD *xd);
static INLINE vp9_prob vp9_get_pred_prob_single_ref_p1(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
- return cm->fc.single_ref_prob[vp9_get_pred_context_single_ref_p1(xd)][0];
+ return cm->fc->single_ref_prob[vp9_get_pred_context_single_ref_p1(xd)][0];
}
int vp9_get_pred_context_single_ref_p2(const MACROBLOCKD *xd);
static INLINE vp9_prob vp9_get_pred_prob_single_ref_p2(const VP9_COMMON *cm,
const MACROBLOCKD *xd) {
- return cm->fc.single_ref_prob[vp9_get_pred_context_single_ref_p2(xd)][1];
+ return cm->fc->single_ref_prob[vp9_get_pred_context_single_ref_p2(xd)][1];
}
int vp9_get_tx_size_context(const MACROBLOCKD *xd);
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index dc712f045..fb041b02d 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -127,7 +127,7 @@ static REFERENCE_MODE read_frame_reference_mode(const VP9_COMMON *cm,
}
static void read_frame_reference_mode_probs(VP9_COMMON *cm, vp9_reader *r) {
- FRAME_CONTEXT *const fc = &cm->fc;
+ FRAME_CONTEXT *const fc = cm->fc;
int i;
if (cm->reference_mode == REFERENCE_MODE_SELECT)
@@ -1386,7 +1386,7 @@ static int read_compressed_header(VP9Decoder *pbi, const uint8_t *data,
size_t partition_size) {
VP9_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
- FRAME_CONTEXT *const fc = &cm->fc;
+ FRAME_CONTEXT *const fc = cm->fc;
vp9_reader r;
int k;
@@ -1540,7 +1540,7 @@ void vp9_decode_frame(VP9Decoder *pbi,
setup_plane_dequants(cm, xd, cm->base_qindex);
vp9_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y);
- cm->fc = cm->frame_contexts[cm->frame_context_idx];
+ *cm->fc = cm->frame_contexts[cm->frame_context_idx];
vp9_zero(cm->counts);
vp9_zero(xd->dqcoeff);
@@ -1580,5 +1580,5 @@ void vp9_decode_frame(VP9Decoder *pbi,
}
if (cm->refresh_frame_context)
- cm->frame_contexts[cm->frame_context_idx] = cm->fc;
+ cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
}
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index a01fe842e..d0e0b76da 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -30,7 +30,7 @@ static PREDICTION_MODE read_intra_mode(vp9_reader *r, const vp9_prob *p) {
static PREDICTION_MODE read_intra_mode_y(VP9_COMMON *cm, vp9_reader *r,
int size_group) {
const PREDICTION_MODE y_mode =
- read_intra_mode(r, cm->fc.y_mode_prob[size_group]);
+ read_intra_mode(r, cm->fc->y_mode_prob[size_group]);
if (!cm->frame_parallel_decoding_mode)
++cm->counts.y_mode[size_group][y_mode];
return y_mode;
@@ -39,7 +39,7 @@ static PREDICTION_MODE read_intra_mode_y(VP9_COMMON *cm, vp9_reader *r,
static PREDICTION_MODE read_intra_mode_uv(VP9_COMMON *cm, vp9_reader *r,
PREDICTION_MODE y_mode) {
const PREDICTION_MODE uv_mode = read_intra_mode(r,
- cm->fc.uv_mode_prob[y_mode]);
+ cm->fc->uv_mode_prob[y_mode]);
if (!cm->frame_parallel_decoding_mode)
++cm->counts.uv_mode[y_mode][uv_mode];
return uv_mode;
@@ -47,7 +47,7 @@ static PREDICTION_MODE read_intra_mode_uv(VP9_COMMON *cm, vp9_reader *r,
static PREDICTION_MODE read_inter_mode(VP9_COMMON *cm, vp9_reader *r, int ctx) {
const int mode = vp9_read_tree(r, vp9_inter_mode_tree,
- cm->fc.inter_mode_probs[ctx]);
+ cm->fc->inter_mode_probs[ctx]);
if (!cm->frame_parallel_decoding_mode)
++cm->counts.inter_mode[ctx][mode];
@@ -61,7 +61,7 @@ static int read_segment_id(vp9_reader *r, const struct segmentation *seg) {
static TX_SIZE read_selected_tx_size(VP9_COMMON *cm, MACROBLOCKD *xd,
TX_SIZE max_tx_size, vp9_reader *r) {
const int ctx = vp9_get_tx_size_context(xd);
- const vp9_prob *tx_probs = get_tx_probs(max_tx_size, ctx, &cm->fc.tx_probs);
+ const vp9_prob *tx_probs = get_tx_probs(max_tx_size, ctx, &cm->fc->tx_probs);
int tx_size = vp9_read(r, tx_probs[0]);
if (tx_size != TX_4X4 && max_tx_size >= TX_16X16) {
tx_size += vp9_read(r, tx_probs[1]);
@@ -150,7 +150,7 @@ static int read_skip(VP9_COMMON *cm, const MACROBLOCKD *xd,
return 1;
} else {
const int ctx = vp9_get_skip_context(xd);
- const int skip = vp9_read(r, cm->fc.skip_probs[ctx]);
+ const int skip = vp9_read(r, cm->fc->skip_probs[ctx]);
if (!cm->frame_parallel_decoding_mode)
++cm->counts.skip[ctx][skip];
return skip;
@@ -258,7 +258,7 @@ static REFERENCE_MODE read_block_reference_mode(VP9_COMMON *cm,
if (cm->reference_mode == REFERENCE_MODE_SELECT) {
const int ctx = vp9_get_reference_mode_context(cm, xd);
const REFERENCE_MODE mode =
- (REFERENCE_MODE)vp9_read(r, cm->fc.comp_inter_prob[ctx]);
+ (REFERENCE_MODE)vp9_read(r, cm->fc->comp_inter_prob[ctx]);
if (!cm->frame_parallel_decoding_mode)
++cm->counts.comp_inter[ctx][mode];
return mode; // SINGLE_REFERENCE or COMPOUND_REFERENCE
@@ -271,7 +271,7 @@ static REFERENCE_MODE read_block_reference_mode(VP9_COMMON *cm,
static void read_ref_frames(VP9_COMMON *const cm, MACROBLOCKD *const xd,
vp9_reader *r,
int segment_id, MV_REFERENCE_FRAME ref_frame[2]) {
- FRAME_CONTEXT *const fc = &cm->fc;
+ FRAME_CONTEXT *const fc = cm->fc;
FRAME_COUNTS *const counts = &cm->counts;
if (vp9_segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
@@ -317,7 +317,7 @@ static INLINE INTERP_FILTER read_switchable_interp_filter(
const int ctx = vp9_get_pred_context_switchable_interp(xd);
const INTERP_FILTER type =
(INTERP_FILTER)vp9_read_tree(r, vp9_switchable_interp_tree,
- cm->fc.switchable_interp_prob[ctx]);
+ cm->fc->switchable_interp_prob[ctx]);
if (!cm->frame_parallel_decoding_mode)
++cm->counts.switchable_interp[ctx][type];
return type;
@@ -372,7 +372,7 @@ static INLINE int assign_mv(VP9_COMMON *cm, PREDICTION_MODE mode,
nmv_context_counts *const mv_counts = cm->frame_parallel_decoding_mode ?
NULL : &cm->counts.mv;
for (i = 0; i < 1 + is_compound; ++i) {
- read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc.nmvc, mv_counts,
+ read_mv(r, &mv[i].as_mv, &ref_mv[i].as_mv, &cm->fc->nmvc, mv_counts,
allow_hp);
ret = ret && is_mv_valid(&mv[i].as_mv);
}
@@ -410,7 +410,7 @@ static int read_is_inter_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
INTRA_FRAME;
} else {
const int ctx = vp9_get_intra_inter_context(xd);
- const int is_inter = vp9_read(r, cm->fc.intra_inter_prob[ctx]);
+ const int is_inter = vp9_read(r, cm->fc->intra_inter_prob[ctx]);
if (!cm->frame_parallel_decoding_mode)
++cm->counts.intra_inter[ctx][is_inter];
return is_inter;
diff --git a/vp9/decoder/vp9_decoder.c b/vp9/decoder/vp9_decoder.c
index baf6ab7ef..3c41f44f5 100644
--- a/vp9/decoder/vp9_decoder.c
+++ b/vp9/decoder/vp9_decoder.c
@@ -59,6 +59,13 @@ VP9Decoder *vp9_decoder_create() {
}
cm->error.setjmp = 1;
+
+ CHECK_MEM_ERROR(cm, cm->fc,
+ (FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
+ CHECK_MEM_ERROR(cm, cm->frame_contexts,
+ (FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
+ sizeof(*cm->frame_contexts)));
+
pbi->need_resync = 1;
initialize_dec();
@@ -88,6 +95,11 @@ void vp9_decoder_remove(VP9Decoder *pbi) {
VP9_COMMON *const cm = &pbi->common;
int i;
+ vpx_free(cm->fc);
+ cm->fc = NULL;
+ vpx_free(cm->frame_contexts);
+ cm->frame_contexts = NULL;
+
vp9_get_worker_interface()->end(&pbi->lf_worker);
vpx_free(pbi->lf_worker.data1);
vpx_free(pbi->tile_data);
diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c
index 421229a28..8704fddac 100644
--- a/vp9/decoder/vp9_detokenize.c
+++ b/vp9/decoder/vp9_detokenize.c
@@ -58,7 +58,7 @@ static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd, PLANE_TYPE type,
int ctx, const int16_t *scan, const int16_t *nb,
vp9_reader *r) {
const int max_eob = 16 << (tx_size << 1);
- const FRAME_CONTEXT *const fc = &cm->fc;
+ const FRAME_CONTEXT *const fc = cm->fc;
FRAME_COUNTS *const counts = &cm->counts;
const int ref = is_inter_block(&xd->mi[0].src_mi->mbmi);
int band, c = 0;
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index 421e04969..4166d3106 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -84,7 +84,7 @@ static void write_selected_tx_size(const VP9_COMMON *cm,
vp9_writer *w) {
const TX_SIZE max_tx_size = max_txsize_lookup[bsize];
const vp9_prob *const tx_probs = get_tx_probs2(max_tx_size, xd,
- &cm->fc.tx_probs);
+ &cm->fc->tx_probs);
vp9_write(w, tx_size != TX_4X4, tx_probs[0]);
if (tx_size != TX_4X4 && max_tx_size >= TX_16X16) {
vp9_write(w, tx_size != TX_8X8, tx_probs[1]);
@@ -108,14 +108,14 @@ static void update_skip_probs(VP9_COMMON *cm, vp9_writer *w) {
int k;
for (k = 0; k < SKIP_CONTEXTS; ++k)
- vp9_cond_prob_diff_update(w, &cm->fc.skip_probs[k], cm->counts.skip[k]);
+ vp9_cond_prob_diff_update(w, &cm->fc->skip_probs[k], cm->counts.skip[k]);
}
static void update_switchable_interp_probs(VP9_COMMON *cm, vp9_writer *w) {
int j;
for (j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j)
prob_diff_update(vp9_switchable_interp_tree,
- cm->fc.switchable_interp_prob[j],
+ cm->fc->switchable_interp_prob[j],
cm->counts.switchable_interp[j], SWITCHABLE_FILTERS, w);
}
@@ -237,7 +237,7 @@ static void write_ref_frames(const VP9_COMMON *cm, const MACROBLOCKD *xd,
static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
vp9_writer *w) {
VP9_COMMON *const cm = &cpi->common;
- const nmv_context *nmvc = &cm->fc.nmvc;
+ const nmv_context *nmvc = &cm->fc->nmvc;
const MACROBLOCK *const x = &cpi->mb;
const MACROBLOCKD *const xd = &x->e_mbd;
const struct segmentation *const seg = &cm->seg;
@@ -275,7 +275,7 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
if (!is_inter) {
if (bsize >= BLOCK_8X8) {
- write_intra_mode(w, mode, cm->fc.y_mode_prob[size_group_lookup[bsize]]);
+ write_intra_mode(w, mode, cm->fc->y_mode_prob[size_group_lookup[bsize]]);
} else {
int idx, idy;
const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize];
@@ -283,14 +283,14 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
const PREDICTION_MODE b_mode = mi->bmi[idy * 2 + idx].as_mode;
- write_intra_mode(w, b_mode, cm->fc.y_mode_prob[0]);
+ write_intra_mode(w, b_mode, cm->fc->y_mode_prob[0]);
}
}
}
- write_intra_mode(w, mbmi->uv_mode, cm->fc.uv_mode_prob[mode]);
+ write_intra_mode(w, mbmi->uv_mode, cm->fc->uv_mode_prob[mode]);
} else {
const int mode_ctx = mbmi->mode_context[mbmi->ref_frame[0]];
- const vp9_prob *const inter_probs = cm->fc.inter_mode_probs[mode_ctx];
+ const vp9_prob *const inter_probs = cm->fc->inter_mode_probs[mode_ctx];
write_ref_frames(cm, xd, w);
// If segment skip is not enabled code the mode.
@@ -304,7 +304,7 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
if (cm->interp_filter == SWITCHABLE) {
const int ctx = vp9_get_pred_context_switchable_interp(xd);
vp9_write_token(w, vp9_switchable_interp_tree,
- cm->fc.switchable_interp_prob[ctx],
+ cm->fc->switchable_interp_prob[ctx],
&switchable_interp_encodings[mbmi->interp_filter]);
++cpi->interp_filter_selected[0][mbmi->interp_filter];
} else {
@@ -528,7 +528,7 @@ static void update_coef_probs_common(vp9_writer* const bc, VP9_COMP *cpi,
TX_SIZE tx_size,
vp9_coeff_stats *frame_branch_ct,
vp9_coeff_probs_model *new_coef_probs) {
- vp9_coeff_probs_model *old_coef_probs = cpi->common.fc.coef_probs[tx_size];
+ vp9_coeff_probs_model *old_coef_probs = cpi->common.fc->coef_probs[tx_size];
const vp9_prob upd = DIFF_UPDATE_PROB;
const int entropy_nodes_update = UNCONSTRAINED_NODES;
int i, j, k, l, t;
@@ -830,20 +830,20 @@ static void encode_txfm_probs(VP9_COMMON *cm, vp9_writer *w) {
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
tx_counts_to_branch_counts_8x8(cm->counts.tx.p8x8[i], ct_8x8p);
for (j = 0; j < TX_SIZES - 3; j++)
- vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p8x8[i][j], ct_8x8p[j]);
+ vp9_cond_prob_diff_update(w, &cm->fc->tx_probs.p8x8[i][j], ct_8x8p[j]);
}
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
tx_counts_to_branch_counts_16x16(cm->counts.tx.p16x16[i], ct_16x16p);
for (j = 0; j < TX_SIZES - 2; j++)
- vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p16x16[i][j],
+ vp9_cond_prob_diff_update(w, &cm->fc->tx_probs.p16x16[i][j],
ct_16x16p[j]);
}
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
tx_counts_to_branch_counts_32x32(cm->counts.tx.p32x32[i], ct_32x32p);
for (j = 0; j < TX_SIZES - 1; j++)
- vp9_cond_prob_diff_update(w, &cm->fc.tx_probs.p32x32[i][j],
+ vp9_cond_prob_diff_update(w, &cm->fc->tx_probs.p32x32[i][j],
ct_32x32p[j]);
}
}
@@ -1161,7 +1161,7 @@ static void write_uncompressed_header(VP9_COMP *cpi,
static size_t write_compressed_header(VP9_COMP *cpi, uint8_t *data) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->mb.e_mbd;
- FRAME_CONTEXT *const fc = &cm->fc;
+ FRAME_CONTEXT *const fc = cm->fc;
vp9_writer header_bc;
vp9_start_encode(&header_bc, data);
@@ -1178,7 +1178,7 @@ static size_t write_compressed_header(VP9_COMP *cpi, uint8_t *data) {
int i;
for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
- prob_diff_update(vp9_inter_mode_tree, cm->fc.inter_mode_probs[i],
+ prob_diff_update(vp9_inter_mode_tree, cm->fc->inter_mode_probs[i],
cm->counts.inter_mode[i], INTER_MODES, &header_bc);
vp9_zero(cm->counts.inter_mode);
@@ -1219,7 +1219,7 @@ static size_t write_compressed_header(VP9_COMP *cpi, uint8_t *data) {
cm->counts.comp_ref[i]);
for (i = 0; i < BLOCK_SIZE_GROUPS; ++i)
- prob_diff_update(vp9_intra_mode_tree, cm->fc.y_mode_prob[i],
+ prob_diff_update(vp9_intra_mode_tree, cm->fc->y_mode_prob[i],
cm->counts.y_mode[i], INTRA_MODES, &header_bc);
for (i = 0; i < PARTITION_CONTEXTS; ++i)
diff --git a/vp9/encoder/vp9_encodemv.c b/vp9/encoder/vp9_encodemv.c
index 089839567..f36d76e3d 100644
--- a/vp9/encoder/vp9_encodemv.c
+++ b/vp9/encoder/vp9_encodemv.c
@@ -163,7 +163,7 @@ static void write_mv_update(const vp9_tree_index *tree,
void vp9_write_nmv_probs(VP9_COMMON *cm, int usehp, vp9_writer *w) {
int i, j;
- nmv_context *const mvc = &cm->fc.nmvc;
+ nmv_context *const mvc = &cm->fc->nmvc;
nmv_context_counts *const counts = &cm->counts.mv;
write_mv_update(vp9_mv_joint_tree, mvc->joints, counts->joints, MV_JOINTS, w);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index c5e872607..9b6e9a290 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -134,7 +134,7 @@ static void setup_frame(VP9_COMP *cpi) {
cpi->refresh_alt_ref_frame = 1;
vp9_zero(cpi->interp_filter_selected);
} else {
- cm->fc = cm->frame_contexts[cm->frame_context_idx];
+ *cm->fc = cm->frame_contexts[cm->frame_context_idx];
vp9_zero(cpi->interp_filter_selected[0]);
}
}
@@ -160,6 +160,11 @@ static void dealloc_compressor_data(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
int i;
+ vpx_free(cm->fc);
+ cm->fc = NULL;
+ vpx_free(cm->frame_contexts);
+ cm->frame_contexts = NULL;
+
// Delete sementation map
vpx_free(cpi->segmentation_map);
cpi->segmentation_map = NULL;
@@ -257,7 +262,7 @@ static void save_coding_context(VP9_COMP *cpi) {
vp9_copy(cc->last_ref_lf_deltas, cm->lf.last_ref_deltas);
vp9_copy(cc->last_mode_lf_deltas, cm->lf.last_mode_deltas);
- cc->fc = cm->fc;
+ cc->fc = *cm->fc;
}
static void restore_coding_context(VP9_COMP *cpi) {
@@ -286,7 +291,7 @@ static void restore_coding_context(VP9_COMP *cpi) {
vp9_copy(cm->lf.last_ref_deltas, cc->last_ref_lf_deltas);
vp9_copy(cm->lf.last_mode_deltas, cc->last_mode_lf_deltas);
- cm->fc = cc->fc;
+ *cm->fc = cc->fc;
}
static void configure_static_seg_features(VP9_COMP *cpi) {
@@ -1374,6 +1379,12 @@ VP9_COMP *vp9_create_compressor(VP9EncoderConfig *oxcf) {
cm->error.setjmp = 1;
+ CHECK_MEM_ERROR(cm, cm->fc,
+ (FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
+ CHECK_MEM_ERROR(cm, cm->frame_contexts,
+ (FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
+ sizeof(*cm->frame_contexts)));
+
cpi->use_svc = 0;
init_config(cpi, oxcf);
@@ -3638,7 +3649,7 @@ int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
}
if (cm->refresh_frame_context)
- cm->frame_contexts[cm->frame_context_idx] = cm->fc;
+ cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
// Frame was dropped, release scaled references.
if (*size == 0) {
diff --git a/vp9/encoder/vp9_pickmode.c b/vp9/encoder/vp9_pickmode.c
index 56240ca95..1826885c7 100644
--- a/vp9/encoder/vp9_pickmode.c
+++ b/vp9/encoder/vp9_pickmode.c
@@ -634,7 +634,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
// motion vector is at sub-pixel accuracy level for luma component, i.e.,
// the last three bits are all zeros.
if (cpi->sf.reuse_inter_pred_sby) {
- if (this_mode == NEARESTMV) {
+ if (!this_mode_pred) {
this_mode_pred = &tmp[3];
} else {
this_mode_pred = &tmp[get_pred_buffer(tmp, 3)];
diff --git a/vp9/encoder/vp9_rd.c b/vp9/encoder/vp9_rd.c
index 7f526fc42..d758430c3 100644
--- a/vp9/encoder/vp9_rd.c
+++ b/vp9/encoder/vp9_rd.c
@@ -65,7 +65,7 @@ static const uint8_t rd_thresh_block_size_factor[BLOCK_SIZES] = {
};
static void fill_mode_costs(VP9_COMP *cpi) {
- const FRAME_CONTEXT *const fc = &cpi->common.fc;
+ const FRAME_CONTEXT *const fc = cpi->common.fc;
int i, j;
for (i = 0; i < INTRA_MODES; ++i)
@@ -280,7 +280,7 @@ void vp9_initialize_rd_consts(VP9_COMP *cpi) {
set_block_thresholds(cm, rd);
if (!cpi->sf.use_nonrd_pick_mode || cm->frame_type == KEY_FRAME) {
- fill_token_costs(x->token_costs, cm->fc.coef_probs);
+ fill_token_costs(x->token_costs, cm->fc->coef_probs);
for (i = 0; i < PARTITION_CONTEXTS; ++i)
vp9_cost_tokens(cpi->partition_cost[i], get_partition_probs(cm, i),
@@ -295,11 +295,11 @@ void vp9_initialize_rd_consts(VP9_COMP *cpi) {
vp9_build_nmv_cost_table(x->nmvjointcost,
cm->allow_high_precision_mv ? x->nmvcost_hp
: x->nmvcost,
- &cm->fc.nmvc, cm->allow_high_precision_mv);
+ &cm->fc->nmvc, cm->allow_high_precision_mv);
for (i = 0; i < INTER_MODE_CONTEXTS; ++i)
vp9_cost_tokens((int *)cpi->inter_mode_cost[i],
- cm->fc.inter_mode_probs[i], vp9_inter_mode_tree);
+ cm->fc->inter_mode_probs[i], vp9_inter_mode_tree);
}
}
}
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index eca8e5880..97d2c25e1 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -639,7 +639,7 @@ static void choose_tx_size_from_rd(VP9_COMP *cpi, MACROBLOCK *x,
int64_t best_rd = INT64_MAX;
TX_SIZE best_tx = max_tx_size;
- const vp9_prob *tx_probs = get_tx_probs2(max_tx_size, xd, &cm->fc.tx_probs);
+ const vp9_prob *tx_probs = get_tx_probs2(max_tx_size, xd, &cm->fc->tx_probs);
assert(skip_prob > 0);
s0 = vp9_cost_bit(skip_prob, 0);
s1 = vp9_cost_bit(skip_prob, 1);
diff --git a/vp9/encoder/vp9_tokenize.c b/vp9/encoder/vp9_tokenize.c
index adf01bf35..0166a50a0 100644
--- a/vp9/encoder/vp9_tokenize.c
+++ b/vp9/encoder/vp9_tokenize.c
@@ -313,7 +313,7 @@ static void tokenize_b(int plane, int block, BLOCK_SIZE plane_bsize,
unsigned int (*const counts)[COEFF_CONTEXTS][ENTROPY_TOKENS] =
cpi->coef_counts[tx_size][type][ref];
vp9_prob (*const coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] =
- cpi->common.fc.coef_probs[tx_size][type][ref];
+ cpi->common.fc->coef_probs[tx_size][type][ref];
unsigned int (*const eob_branch)[COEFF_CONTEXTS] =
cpi->common.counts.eob_branch[tx_size][type][ref];
const uint8_t *const band = get_band_translate(tx_size);
diff --git a/vp9/encoder/x86/vp9_denoiser_sse2.c b/vp9/encoder/x86/vp9_denoiser_sse2.c
index bf5fa889f..09969b0d9 100644
--- a/vp9/encoder/x86/vp9_denoiser_sse2.c
+++ b/vp9/encoder/x86/vp9_denoiser_sse2.c
@@ -23,18 +23,17 @@
// Compute the sum of all pixel differences of this MB.
static INLINE int sum_diff_16x1(__m128i acc_diff) {
const __m128i k_1 = _mm_set1_epi16(1);
- const __m128i acc_diff_lo = _mm_srai_epi16(
- _mm_unpacklo_epi8(acc_diff, acc_diff), 8);
- const __m128i acc_diff_hi = _mm_srai_epi16(
- _mm_unpackhi_epi8(acc_diff, acc_diff), 8);
+ const __m128i acc_diff_lo =
+ _mm_srai_epi16(_mm_unpacklo_epi8(acc_diff, acc_diff), 8);
+ const __m128i acc_diff_hi =
+ _mm_srai_epi16(_mm_unpackhi_epi8(acc_diff, acc_diff), 8);
const __m128i acc_diff_16 = _mm_add_epi16(acc_diff_lo, acc_diff_hi);
const __m128i hg_fe_dc_ba = _mm_madd_epi16(acc_diff_16, k_1);
- const __m128i hgfe_dcba = _mm_add_epi32(hg_fe_dc_ba,
- _mm_srli_si128(hg_fe_dc_ba, 8));
- const __m128i hgfedcba = _mm_add_epi32(hgfe_dcba,
- _mm_srli_si128(hgfe_dcba, 4));
- int sum_diff = _mm_cvtsi128_si32(hgfedcba);
- return sum_diff;
+ const __m128i hgfe_dcba =
+ _mm_add_epi32(hg_fe_dc_ba, _mm_srli_si128(hg_fe_dc_ba, 8));
+ const __m128i hgfedcba =
+ _mm_add_epi32(hgfe_dcba, _mm_srli_si128(hgfe_dcba, 4));
+ return _mm_cvtsi128_si32(hgfedcba);
}
// Denoise a 16x1 vector.
@@ -51,8 +50,8 @@ static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
__m128i acc_diff) {
// Calculate differences
const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0]));
- const __m128i v_mc_running_avg_y = _mm_loadu_si128(
- (const __m128i *)(&mc_running_avg_y[0]));
+ const __m128i v_mc_running_avg_y =
+ _mm_loadu_si128((const __m128i *)(&mc_running_avg_y[0]));
__m128i v_running_avg_y;
const __m128i pdiff = _mm_subs_epu8(v_mc_running_avg_y, v_sig);
const __m128i ndiff = _mm_subs_epu8(v_sig, v_mc_running_avg_y);
@@ -60,8 +59,8 @@ static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, *k_0);
// Clamp absolute difference to 16 to be used to get mask. Doing this
// allows us to use _mm_cmpgt_epi8, which operates on signed byte.
- const __m128i clamped_absdiff = _mm_min_epu8(
- _mm_or_si128(pdiff, ndiff), *k_16);
+ const __m128i clamped_absdiff =
+ _mm_min_epu8(_mm_or_si128(pdiff, ndiff), *k_16);
// Get masks for l2 l1 and l0 adjustments.
const __m128i mask2 = _mm_cmpgt_epi8(*k_16, clamped_absdiff);
const __m128i mask1 = _mm_cmpgt_epi8(*k_8, clamped_absdiff);
@@ -95,24 +94,22 @@ static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
}
// Denoise a 16x1 vector with a weaker filter.
-static INLINE __m128i vp9_denoiser_adj_16x1_sse2(const uint8_t *sig,
- const uint8_t *mc_running_avg_y,
- uint8_t *running_avg_y,
- const __m128i k_0,
- const __m128i k_delta,
- __m128i acc_diff) {
+static INLINE __m128i vp9_denoiser_adj_16x1_sse2(
+ const uint8_t *sig, const uint8_t *mc_running_avg_y,
+ uint8_t *running_avg_y, const __m128i k_0,
+ const __m128i k_delta, __m128i acc_diff) {
__m128i v_running_avg_y = _mm_loadu_si128((__m128i *)(&running_avg_y[0]));
// Calculate differences.
const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0]));
const __m128i v_mc_running_avg_y =
- _mm_loadu_si128((const __m128i *)(&mc_running_avg_y[0]));
+ _mm_loadu_si128((const __m128i *)(&mc_running_avg_y[0]));
const __m128i pdiff = _mm_subs_epu8(v_mc_running_avg_y, v_sig);
const __m128i ndiff = _mm_subs_epu8(v_sig, v_mc_running_avg_y);
// Obtain the sign. FF if diff is negative.
const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, k_0);
// Clamp absolute difference to delta to get the adjustment.
const __m128i adj =
- _mm_min_epu8(_mm_or_si128(pdiff, ndiff), k_delta);
+ _mm_min_epu8(_mm_or_si128(pdiff, ndiff), k_delta);
// Restore the sign and get positive and negative adjustments.
__m128i padj, nadj;
padj = _mm_andnot_si128(diff_sign, adj);
@@ -135,12 +132,11 @@ static int vp9_denoiser_4xM_sse2(const uint8_t *sig, int sig_stride,
int increase_denoising,
BLOCK_SIZE bs,
int motion_magnitude) {
- int sum_diff_thresh;
- int r;
- int shift_inc = (increase_denoising &&
- motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 1 : 0;
- unsigned char sig_buffer[2][16], mc_running_buffer[2][16],
- running_buffer[2][16];
+ int sum_diff_thresh, r, sum_diff = 0;
+ const int shift_inc = (increase_denoising &&
+ motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
+ 1 : 0;
+ uint8_t sig_buffer[2][16], mc_running_buffer[2][16], running_buffer[2][16];
__m128i acc_diff = _mm_setzero_si128();
const __m128i k_0 = _mm_setzero_si128();
const __m128i k_4 = _mm_set1_epi8(4 + shift_inc);
@@ -148,13 +144,11 @@ static int vp9_denoiser_4xM_sse2(const uint8_t *sig, int sig_stride,
const __m128i k_16 = _mm_set1_epi8(16);
// Modify each level's adjustment according to motion_magnitude.
const __m128i l3 = _mm_set1_epi8(
- (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
- 7 + shift_inc : 6);
+ (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 + shift_inc : 6);
// Difference between level 3 and level 2 is 2.
const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1);
- int sum_diff = 0;
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 2); ++r) {
vpx_memcpy(sig_buffer[r], sig, 4);
@@ -197,17 +191,17 @@ static int vp9_denoiser_4xM_sse2(const uint8_t *sig, int sig_stride,
sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising);
if (abs(sum_diff) > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising),
- // checK if we can still apply some (weaker) temporal filtering to
+ // check if we can still apply some (weaker) temporal filtering to
// this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it
// closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be
- // within the accceptable range given by sum_diff_thresh.
+ // within the acceptable range given by sum_diff_thresh.
// The delta is set by the excess of absolute pixel diff over the
// threshold.
- int delta = ((abs(sum_diff) - sum_diff_thresh)
- >> num_pels_log2_lookup[bs]) + 1;
+ const int delta = ((abs(sum_diff) - sum_diff_thresh) >>
+ num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
const __m128i k_delta = _mm_set1_epi8(delta);
@@ -215,9 +209,8 @@ static int vp9_denoiser_4xM_sse2(const uint8_t *sig, int sig_stride,
sum_diff = 0;
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 2); ++r) {
acc_diff = vp9_denoiser_adj_16x1_sse2(
- sig_buffer[r], mc_running_buffer[r],
- running_buffer[r], k_0, k_delta,
- acc_diff);
+ sig_buffer[r], mc_running_buffer[r], running_buffer[r],
+ k_0, k_delta, acc_diff);
vpx_memcpy(running_avg_y, running_buffer[r], 4);
vpx_memcpy(running_avg_y + avg_y_stride, running_buffer[r] + 4, 4);
vpx_memcpy(running_avg_y + avg_y_stride * 2,
@@ -246,12 +239,11 @@ static int vp9_denoiser_8xM_sse2(const uint8_t *sig, int sig_stride,
int increase_denoising,
BLOCK_SIZE bs,
int motion_magnitude) {
- int sum_diff_thresh;
- int r;
- int shift_inc = (increase_denoising &&
- motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 1 : 0;
- unsigned char sig_buffer[8][16], mc_running_buffer[8][16],
- running_buffer[8][16];
+ int sum_diff_thresh, r, sum_diff = 0;
+ const int shift_inc = (increase_denoising &&
+ motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
+ 1 : 0;
+ uint8_t sig_buffer[8][16], mc_running_buffer[8][16], running_buffer[8][16];
__m128i acc_diff = _mm_setzero_si128();
const __m128i k_0 = _mm_setzero_si128();
const __m128i k_4 = _mm_set1_epi8(4 + shift_inc);
@@ -259,13 +251,11 @@ static int vp9_denoiser_8xM_sse2(const uint8_t *sig, int sig_stride,
const __m128i k_16 = _mm_set1_epi8(16);
// Modify each level's adjustment according to motion_magnitude.
const __m128i l3 = _mm_set1_epi8(
- (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
- 7 + shift_inc : 6);
+ (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 + shift_inc : 6);
// Difference between level 3 and level 2 is 2.
const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1);
- int sum_diff = 0;
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 1); ++r) {
vpx_memcpy(sig_buffer[r], sig, 8);
@@ -294,26 +284,25 @@ static int vp9_denoiser_8xM_sse2(const uint8_t *sig, int sig_stride,
sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising);
if (abs(sum_diff) > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising),
- // checK if we can still apply some (weaker) temporal filtering to
+ // check if we can still apply some (weaker) temporal filtering to
// this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it
// closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be
- // within the accceptable range given by sum_diff_thresh.
+ // within the acceptable range given by sum_diff_thresh.
// The delta is set by the excess of absolute pixel diff over the
// threshold.
- int delta = ((abs(sum_diff) - sum_diff_thresh)
- >> num_pels_log2_lookup[bs]) + 1;
+ const int delta = ((abs(sum_diff) - sum_diff_thresh)
+ >> num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
const __m128i k_delta = _mm_set1_epi8(delta);
running_avg_y -= avg_y_stride * (4 << b_height_log2_lookup[bs]);
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 1); ++r) {
acc_diff = vp9_denoiser_adj_16x1_sse2(
- sig_buffer[r], mc_running_buffer[r],
- running_buffer[r], k_0, k_delta,
- acc_diff);
+ sig_buffer[r], mc_running_buffer[r], running_buffer[r],
+ k_0, k_delta, acc_diff);
vpx_memcpy(running_avg_y, running_buffer[r], 8);
vpx_memcpy(running_avg_y + avg_y_stride, running_buffer[r] + 8, 8);
// Update pointers for next iteration.
@@ -324,7 +313,7 @@ static int vp9_denoiser_8xM_sse2(const uint8_t *sig, int sig_stride,
return COPY_BLOCK;
}
} else {
- return COPY_BLOCK;
+ return COPY_BLOCK;
}
}
}
@@ -338,10 +327,10 @@ static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
int avg_y_stride,
int increase_denoising, BLOCK_SIZE bs,
int motion_magnitude) {
- int sum_diff_thresh;
- int r, c;
- int shift_inc = (increase_denoising &&
- motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 1 : 0;
+ int sum_diff_thresh, r, c, sum_diff = 0;
+ const int shift_inc = (increase_denoising &&
+ motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
+ 1 : 0;
__m128i acc_diff[4][4];
const __m128i k_0 = _mm_setzero_si128();
const __m128i k_4 = _mm_set1_epi8(4 + shift_inc);
@@ -349,13 +338,11 @@ static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
const __m128i k_16 = _mm_set1_epi8(16);
// Modify each level's adjustment according to motion_magnitude.
const __m128i l3 = _mm_set1_epi8(
- (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
- 7 + shift_inc : 6);
+ (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 + shift_inc : 6);
// Difference between level 3 and level 2 is 2.
const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1);
- int sum_diff = 0;
for (c = 0; c < 4; ++c) {
for (r = 0; r < 4; ++r) {
@@ -363,13 +350,11 @@ static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
}
}
- for (r = 0; r < (4 << b_height_log2_lookup[bs]); r++) {
+ for (r = 0; r < (4 << b_height_log2_lookup[bs]); ++r) {
for (c = 0; c < (4 << b_width_log2_lookup[bs]); c += 16) {
acc_diff[c>>4][r>>4] = vp9_denoiser_16x1_sse2(
- sig, mc_running_avg_y,
- running_avg_y,
- &k_0, &k_4, &k_8, &k_16,
- &l3, &l32, &l21, acc_diff[c>>4][r>>4]);
+ sig, mc_running_avg_y, running_avg_y, &k_0, &k_4,
+ &k_8, &k_16, &l3, &l32, &l21, acc_diff[c>>4][r>>4]);
// Update pointers for next iteration.
sig += 16;
mc_running_avg_y += 16;
@@ -385,8 +370,8 @@ static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
// Update pointers for next iteration.
sig = sig - 16 * ((4 << b_width_log2_lookup[bs]) >> 4) + sig_stride;
mc_running_avg_y = mc_running_avg_y -
- 16 * ((4 << b_width_log2_lookup[bs]) >> 4) +
- mc_avg_y_stride;
+ 16 * ((4 << b_width_log2_lookup[bs]) >> 4) +
+ mc_avg_y_stride;
running_avg_y = running_avg_y -
16 * ((4 << b_width_log2_lookup[bs]) >> 4) +
avg_y_stride;
@@ -395,8 +380,8 @@ static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
{
sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising);
if (abs(sum_diff) > sum_diff_thresh) {
- int delta = ((abs(sum_diff) - sum_diff_thresh)
- >> num_pels_log2_lookup[bs]) + 1;
+ const int delta = ((abs(sum_diff) - sum_diff_thresh) >>
+ num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
@@ -408,9 +393,8 @@ static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
for (r = 0; r < (4 << b_height_log2_lookup[bs]); ++r) {
for (c = 0; c < (4 << b_width_log2_lookup[bs]); c += 16) {
acc_diff[c>>4][r>>4] = vp9_denoiser_adj_16x1_sse2(
- sig, mc_running_avg_y,
- running_avg_y, k_0,
- k_delta, acc_diff[c>>4][r>>4]);
+ sig, mc_running_avg_y, running_avg_y, k_0,
+ k_delta, acc_diff[c>>4][r>>4]);
// Update pointers for next iteration.
sig += 16;
mc_running_avg_y += 16;
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-11 01:28:45 +0000