summaryrefslogtreecommitdiff
path: root/vp9
diff options
context:
space:
mode:
Diffstat (limited to 'vp9')
-rw-r--r--vp9/common/vp9_postproc.c30
-rw-r--r--vp9/common/vp9_rtcd_defs.pl7
-rw-r--r--vp9/common/x86/vp9_postproc_sse2.asm62
-rw-r--r--vp9/encoder/vp9_encodeframe.c138
-rw-r--r--vp9/encoder/vp9_lookahead.c10
-rw-r--r--vp9/encoder/vp9_lookahead.h8
-rw-r--r--vp9/encoder/vp9_temporal_filter.c8
-rw-r--r--vp9/encoder/x86/vp9_denoiser_sse2.c68
8 files changed, 121 insertions, 210 deletions
diff --git a/vp9/common/vp9_postproc.c b/vp9/common/vp9_postproc.c
index b685d813b..c04cc8f05 100644
--- a/vp9/common/vp9_postproc.c
+++ b/vp9/common/vp9_postproc.c
@@ -12,6 +12,7 @@
#include <stdlib.h>
#include <stdio.h>
+#include "./vpx_dsp_rtcd.h"
#include "./vpx_config.h"
#include "./vpx_scale_rtcd.h"
#include "./vp9_rtcd.h"
@@ -587,32 +588,6 @@ static void fillrd(struct postproc_state *state, int q, int a) {
state->last_noise = a;
}
-void vp9_plane_add_noise_c(uint8_t *start, char *noise,
- char blackclamp[16],
- char whiteclamp[16],
- char bothclamp[16],
- unsigned int width, unsigned int height, int pitch) {
- unsigned int i, j;
-
- // TODO(jbb): why does simd code use both but c doesn't, normalize and
- // fix..
- (void) bothclamp;
- for (i = 0; i < height; i++) {
- uint8_t *pos = start + i * pitch;
- char *ref = (char *)(noise + (rand() & 0xff)); // NOLINT
-
- for (j = 0; j < width; j++) {
- if (pos[j] < blackclamp[0])
- pos[j] = blackclamp[0];
-
- if (pos[j] > 255 + whiteclamp[0])
- pos[j] = 255 + whiteclamp[0];
-
- pos[j] += ref[j];
- }
- }
-}
-
static void swap_mi_and_prev_mi(VP9_COMMON *cm) {
// Current mip will be the prev_mip for the next frame.
MODE_INFO *temp = cm->postproc_state.prev_mip;
@@ -726,8 +701,7 @@ int vp9_post_proc_frame(struct VP9Common *cm,
ppstate->last_noise != noise_level) {
fillrd(ppstate, 63 - q, noise_level);
}
-
- vp9_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
+ vpx_plane_add_noise(ppbuf->y_buffer, ppstate->noise, ppstate->blackclamp,
ppstate->whiteclamp, ppstate->bothclamp,
ppbuf->y_width, ppbuf->y_height, ppbuf->y_stride);
}
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 1cf636c1d..d7f5a2113 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -70,10 +70,6 @@ add_proto qw/void vp9_post_proc_down_and_across/, "const uint8_t *src_ptr, uint8
specialize qw/vp9_post_proc_down_and_across sse2/;
$vp9_post_proc_down_and_across_sse2=vp9_post_proc_down_and_across_xmm;
-add_proto qw/void vp9_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
-specialize qw/vp9_plane_add_noise sse2/;
-$vp9_plane_add_noise_sse2=vp9_plane_add_noise_wmt;
-
add_proto qw/void vp9_filter_by_weight16x16/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp9_filter_by_weight16x16 sse2 msa/;
@@ -169,9 +165,6 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vp9_highbd_post_proc_down_and_across/, "const uint16_t *src_ptr, uint16_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
specialize qw/vp9_highbd_post_proc_down_and_across/;
-
- add_proto qw/void vp9_highbd_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
- specialize qw/vp9_highbd_plane_add_noise/;
}
#
diff --git a/vp9/common/x86/vp9_postproc_sse2.asm b/vp9/common/x86/vp9_postproc_sse2.asm
index ec8bfdb18..430762815 100644
--- a/vp9/common/x86/vp9_postproc_sse2.asm
+++ b/vp9/common/x86/vp9_postproc_sse2.asm
@@ -624,68 +624,6 @@ sym(vp9_mbpost_proc_across_ip_xmm):
%undef flimit4
-;void vp9_plane_add_noise_wmt (unsigned char *start, unsigned char *noise,
-; unsigned char blackclamp[16],
-; unsigned char whiteclamp[16],
-; unsigned char bothclamp[16],
-; unsigned int width, unsigned int height, int pitch)
-global sym(vp9_plane_add_noise_wmt) PRIVATE
-sym(vp9_plane_add_noise_wmt):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 8
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
-.addnoise_loop:
- call sym(LIBVPX_RAND) WRT_PLT
- mov rcx, arg(1) ;noise
- and rax, 0xff
- add rcx, rax
-
- ; we rely on the fact that the clamping vectors are stored contiguously
- ; in black/white/both order. Note that we have to reload this here because
- ; rdx could be trashed by rand()
- mov rdx, arg(2) ; blackclamp
-
-
- mov rdi, rcx
- movsxd rcx, dword arg(5) ;[Width]
- mov rsi, arg(0) ;Pos
- xor rax,rax
-
-.addnoise_nextset:
- movdqu xmm1,[rsi+rax] ; get the source
-
- psubusb xmm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
- paddusb xmm1, [rdx+32] ;bothclamp
- psubusb xmm1, [rdx+16] ;whiteclamp
-
- movdqu xmm2,[rdi+rax] ; get the noise for this line
- paddb xmm1,xmm2 ; add it in
- movdqu [rsi+rax],xmm1 ; store the result
-
- add rax,16 ; move to the next line
-
- cmp rax, rcx
- jl .addnoise_nextset
-
- movsxd rax, dword arg(7) ; Pitch
- add arg(0), rax ; Start += Pitch
- sub dword arg(6), 1 ; Height -= 1
- jg .addnoise_loop
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
SECTION_RODATA
align 16
rd42:
diff --git a/vp9/encoder/vp9_encodeframe.c b/vp9/encoder/vp9_encodeframe.c
index ae5ca7dd5..e96c96cde 100644
--- a/vp9/encoder/vp9_encodeframe.c
+++ b/vp9/encoder/vp9_encodeframe.c
@@ -662,12 +662,79 @@ static void fill_variance_8x8avg(const uint8_t *s, int sp, const uint8_t *d,
}
}
+#if !CONFIG_VP9_HIGHBITDEPTH
+// Check if most of the superblock is skin content, and if so, force split to
+// 32x32, and set x->sb_is_skin for use in mode selection.
+static int skin_sb_split(VP9_COMP *cpi, MACROBLOCK *x, const int low_res,
+ int mi_row, int mi_col, int *force_split) {
+ VP9_COMMON * const cm = &cpi->common;
+ // Avoid checking superblocks on/near boundary and avoid low resolutions.
+ // Note superblock may still pick 64X64 if y_sad is very small
+ // (i.e., y_sad < cpi->vbp_threshold_sad) below. For now leave this as is.
+ if (!low_res && (mi_col >= 8 && mi_col + 8 < cm->mi_cols && mi_row >= 8 &&
+ mi_row + 8 < cm->mi_rows)) {
+ int num_16x16_skin = 0;
+ int num_16x16_nonskin = 0;
+ uint8_t *ysignal = x->plane[0].src.buf;
+ uint8_t *usignal = x->plane[1].src.buf;
+ uint8_t *vsignal = x->plane[2].src.buf;
+ int sp = x->plane[0].src.stride;
+ int spuv = x->plane[1].src.stride;
+ const int block_index = mi_row * cm->mi_cols + mi_col;
+ const int bw = num_8x8_blocks_wide_lookup[BLOCK_64X64];
+ const int bh = num_8x8_blocks_high_lookup[BLOCK_64X64];
+ const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
+ const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
+ // Loop through the 16x16 sub-blocks.
+ int i, j;
+ for (i = 0; i < ymis; i+=2) {
+ for (j = 0; j < xmis; j+=2) {
+ int bl_index = block_index + i * cm->mi_cols + j;
+ int bl_index1 = bl_index + 1;
+ int bl_index2 = bl_index + cm->mi_cols;
+ int bl_index3 = bl_index2 + 1;
+ int consec_zeromv = VPXMIN(cpi->consec_zero_mv[bl_index],
+ VPXMIN(cpi->consec_zero_mv[bl_index1],
+ VPXMIN(cpi->consec_zero_mv[bl_index2],
+ cpi->consec_zero_mv[bl_index3])));
+ int is_skin = vp9_compute_skin_block(ysignal,
+ usignal,
+ vsignal,
+ sp,
+ spuv,
+ BLOCK_16X16,
+ consec_zeromv,
+ 0);
+ num_16x16_skin += is_skin;
+ num_16x16_nonskin += (1 - is_skin);
+ if (num_16x16_nonskin > 3) {
+ // Exit loop if at least 4 of the 16x16 blocks are not skin.
+ i = ymis;
+ break;
+ }
+ ysignal += 16;
+ usignal += 8;
+ vsignal += 8;
+ }
+ ysignal += (sp << 4) - 64;
+ usignal += (spuv << 3) - 32;
+ vsignal += (spuv << 3) - 32;
+ }
+ if (num_16x16_skin > 12) {
+ *force_split = 1;
+ return 1;
+ }
+ }
+ return 0;
+}
+#endif
+
// This function chooses partitioning based on the variance between source and
// reconstructed last, where variance is computed for down-sampled inputs.
static int choose_partitioning(VP9_COMP *cpi,
- const TileInfo *const tile,
- MACROBLOCK *x,
- int mi_row, int mi_col) {
+ const TileInfo *const tile,
+ MACROBLOCK *x,
+ int mi_row, int mi_col) {
VP9_COMMON * const cm = &cpi->common;
MACROBLOCKD *xd = &x->e_mbd;
int i, j, k, m;
@@ -771,70 +838,13 @@ static int choose_partitioning(VP9_COMP *cpi,
set_ref_ptrs(cm, xd, mi->ref_frame[0], mi->ref_frame[1]);
vp9_build_inter_predictors_sb(xd, mi_row, mi_col, BLOCK_64X64);
- // Check if most of the superblock is skin content, and if so, force split
- // to 32x32, and set x->sb_is_skin for use in mode selection.
- // Avoid checking superblocks on/near boundary and avoid low resolutions.
- // Note superblock may still pick 64X64 if y_sad is very small
- // (i.e., y_sad < cpi->vbp_threshold_sad) below. For now leave this as is.
x->sb_is_skin = 0;
#if !CONFIG_VP9_HIGHBITDEPTH
- if (cpi->use_skin_detection && !low_res && (mi_col >= 8 &&
- mi_col + 8 < cm->mi_cols && mi_row >= 8 && mi_row + 8 < cm->mi_rows)) {
- int bl_index1, bl_index2, bl_index3;
- int num_16x16_skin = 0;
- int num_16x16_nonskin = 0;
- int is_skin = 0;
- int consec_zeromv = 0;
- uint8_t *ysignal = x->plane[0].src.buf;
- uint8_t *usignal = x->plane[1].src.buf;
- uint8_t *vsignal = x->plane[2].src.buf;
- int spuv = x->plane[1].src.stride;
- const int block_index = mi_row * cm->mi_cols + mi_col;
- const int bw = num_8x8_blocks_wide_lookup[BLOCK_64X64];
- const int bh = num_8x8_blocks_high_lookup[BLOCK_64X64];
- const int xmis = VPXMIN(cm->mi_cols - mi_col, bw);
- const int ymis = VPXMIN(cm->mi_rows - mi_row, bh);
- // Loop through the 16x16 sub-blocks.
- int j, i;
- for (i = 0; i < ymis; i+=2) {
- for (j = 0; j < xmis; j+=2) {
- int bl_index = block_index + i * cm->mi_cols + j;
- bl_index1 = bl_index + 1;
- bl_index2 = bl_index + cm->mi_cols;
- bl_index3 = bl_index2 + 1;
- consec_zeromv = VPXMIN(cpi->consec_zero_mv[bl_index],
- VPXMIN(cpi->consec_zero_mv[bl_index1],
- VPXMIN(cpi->consec_zero_mv[bl_index2],
- cpi->consec_zero_mv[bl_index3])));
- is_skin = vp9_compute_skin_block(ysignal,
- usignal,
- vsignal,
- sp,
- spuv,
- BLOCK_16X16,
- consec_zeromv,
- 0);
- num_16x16_skin += is_skin;
- num_16x16_nonskin += (1 - is_skin);
- if (num_16x16_nonskin > 3) {
- // Exit loop if at least 4 of the 16x16 blocks are not skin.
- i = ymis;
- j = xmis;
- }
- ysignal += 16;
- usignal += 8;
- vsignal += 8;
- }
- ysignal += (sp << 4) - 64;
- usignal += (spuv << 3) - 32;
- vsignal += (spuv << 3) - 32;
- }
- if (num_16x16_skin > 12) {
- x->sb_is_skin = 1;
- force_split[0] = 1;
- }
- }
+ if (cpi->use_skin_detection)
+ x->sb_is_skin = skin_sb_split(cpi, x, low_res, mi_row, mi_col,
+ &force_split[0]);
#endif
+
for (i = 1; i <= 2; ++i) {
struct macroblock_plane *p = &x->plane[i];
struct macroblockd_plane *pd = &xd->plane[i];
diff --git a/vp9/encoder/vp9_lookahead.c b/vp9/encoder/vp9_lookahead.c
index def9b8c1d..441280cb1 100644
--- a/vp9/encoder/vp9_lookahead.c
+++ b/vp9/encoder/vp9_lookahead.c
@@ -20,8 +20,8 @@
/* Return the buffer at the given absolute index and increment the index */
static struct lookahead_entry *pop(struct lookahead_ctx *ctx,
- unsigned int *idx) {
- unsigned int index = *idx;
+ int *idx) {
+ int index = *idx;
struct lookahead_entry *buf = ctx->buf + index;
assert(index < ctx->max_sz);
@@ -35,7 +35,7 @@ static struct lookahead_entry *pop(struct lookahead_ctx *ctx,
void vp9_lookahead_destroy(struct lookahead_ctx *ctx) {
if (ctx) {
if (ctx->buf) {
- unsigned int i;
+ int i;
for (i = 0; i < ctx->max_sz; i++)
vpx_free_frame_buffer(&ctx->buf[i].img);
@@ -221,9 +221,9 @@ struct lookahead_entry *vp9_lookahead_peek(struct lookahead_ctx *ctx,
if (index >= 0) {
// Forward peek
- if (index < (int)ctx->sz) {
+ if (index < ctx->sz) {
index += ctx->read_idx;
- if (index >= (int)ctx->max_sz)
+ if (index >= ctx->max_sz)
index -= ctx->max_sz;
buf = ctx->buf + index;
}
diff --git a/vp9/encoder/vp9_lookahead.h b/vp9/encoder/vp9_lookahead.h
index 13820380f..db0fd1cde 100644
--- a/vp9/encoder/vp9_lookahead.h
+++ b/vp9/encoder/vp9_lookahead.h
@@ -36,10 +36,10 @@ struct lookahead_entry {
#define MAX_PRE_FRAMES 1
struct lookahead_ctx {
- unsigned int max_sz; /* Absolute size of the queue */
- unsigned int sz; /* Number of buffers currently in the queue */
- unsigned int read_idx; /* Read index */
- unsigned int write_idx; /* Write index */
+ int max_sz; /* Absolute size of the queue */
+ int sz; /* Number of buffers currently in the queue */
+ int read_idx; /* Read index */
+ int write_idx; /* Write index */
struct lookahead_entry *buf; /* Buffer list */
};
diff --git a/vp9/encoder/vp9_temporal_filter.c b/vp9/encoder/vp9_temporal_filter.c
index ebe28b8bf..2ba2750e4 100644
--- a/vp9/encoder/vp9_temporal_filter.c
+++ b/vp9/encoder/vp9_temporal_filter.c
@@ -143,8 +143,8 @@ void vp9_temporal_filter_apply_c(uint8_t *frame1,
for (idy = -1; idy <= 1; ++idy) {
for (idx = -1; idx <= 1; ++idx) {
- int row = i + idy;
- int col = j + idx;
+ int row = (int)i + idy;
+ int col = (int)j + idx;
if (row >= 0 && row < (int)block_height &&
col >= 0 && col < (int)block_width) {
@@ -211,8 +211,8 @@ void vp9_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
for (idy = -1; idy <= 1; ++idy) {
for (idx = -1; idx <= 1; ++idx) {
- int row = i + idy;
- int col = j + idx;
+ int row = (int)i + idy;
+ int col = (int)j + idx;
if (row >= 0 && row < (int)block_height &&
col >= 0 && col < (int)block_width) {
diff --git a/vp9/encoder/x86/vp9_denoiser_sse2.c b/vp9/encoder/x86/vp9_denoiser_sse2.c
index f4a149d9e..883507af3 100644
--- a/vp9/encoder/x86/vp9_denoiser_sse2.c
+++ b/vp9/encoder/x86/vp9_denoiser_sse2.c
@@ -147,8 +147,9 @@ static int vp9_denoiser_NxM_sse2_small(
const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1);
+ const int b_height = (4 << b_height_log2_lookup[bs]) >> 1;
- for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 1); ++r) {
+ for (r = 0; r < b_height; ++r) {
memcpy(sig_buffer[r], sig, width);
memcpy(sig_buffer[r] + width, sig + sig_stride, width);
memcpy(mc_running_buffer[r], mc_running_avg_y, width);
@@ -188,8 +189,8 @@ static int vp9_denoiser_NxM_sse2_small(
// 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) {
+ running_avg_y -= avg_y_stride * (b_height << 1);
+ for (r = 0; r < b_height; ++r) {
acc_diff = vp9_denoiser_adj_16x1_sse2(
sig_buffer[r], mc_running_buffer[r], running_buffer[r],
k_0, k_delta, acc_diff);
@@ -235,38 +236,37 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1);
+ const int b_width = (4 << b_width_log2_lookup[bs]);
+ const int b_height = (4 << b_height_log2_lookup[bs]);
+ const int b_width_shift4 = b_width >> 4;
- for (c = 0; c < 4; ++c) {
- for (r = 0; r < 4; ++r) {
+ for (r = 0; r < 4; ++r) {
+ for (c = 0; c < b_width_shift4; ++c) {
acc_diff[c][r] = _mm_setzero_si128();
}
}
- 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(
+ for (r = 0; r < b_height; ++r) {
+ for (c = 0; c < b_width_shift4; ++c) {
+ acc_diff[c][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]);
+ &k_8, &k_16, &l3, &l32, &l21, acc_diff[c][r>>4]);
// Update pointers for next iteration.
sig += 16;
mc_running_avg_y += 16;
running_avg_y += 16;
}
- if ((r + 1) % 16 == 0 || (bs == BLOCK_16X8 && r == 7)) {
- for (c = 0; c < (4 << b_width_log2_lookup[bs]); c += 16) {
- sum_diff += sum_diff_16x1(acc_diff[c>>4][r>>4]);
+ if ((r & 0xf) == 0xf || (bs == BLOCK_16X8 && r == 7)) {
+ for (c = 0; c < b_width_shift4; ++c) {
+ sum_diff += sum_diff_16x1(acc_diff[c][r>>4]);
}
}
// 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;
- running_avg_y = running_avg_y -
- 16 * ((4 << b_width_log2_lookup[bs]) >> 4) +
- avg_y_stride;
+ sig = sig - b_width + sig_stride;
+ mc_running_avg_y = mc_running_avg_y - b_width + mc_avg_y_stride;
+ running_avg_y = running_avg_y - b_width + avg_y_stride;
}
{
@@ -278,33 +278,29 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
const __m128i k_delta = _mm_set1_epi8(delta);
- sig -= sig_stride * (4 << b_height_log2_lookup[bs]);
- mc_running_avg_y -= mc_avg_y_stride * (4 << b_height_log2_lookup[bs]);
- running_avg_y -= avg_y_stride * (4 << b_height_log2_lookup[bs]);
+ sig -= sig_stride * b_height;
+ mc_running_avg_y -= mc_avg_y_stride * b_height;
+ running_avg_y -= avg_y_stride * b_height;
sum_diff = 0;
- 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(
+ for (r = 0; r < b_height; ++r) {
+ for (c = 0; c < b_width_shift4; ++c) {
+ acc_diff[c][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]);
+ k_delta, acc_diff[c][r>>4]);
// Update pointers for next iteration.
sig += 16;
mc_running_avg_y += 16;
running_avg_y += 16;
}
- if ((r + 1) % 16 == 0 || (bs == BLOCK_16X8 && r == 7)) {
- for (c = 0; c < (4 << b_width_log2_lookup[bs]); c += 16) {
- sum_diff += sum_diff_16x1(acc_diff[c>>4][r>>4]);
+ if ((r & 0xf) == 0xf || (bs == BLOCK_16X8 && r == 7)) {
+ for (c = 0; c < b_width_shift4; ++c) {
+ sum_diff += sum_diff_16x1(acc_diff[c][r>>4]);
}
}
- 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;
- running_avg_y = running_avg_y -
- 16 * ((4 << b_width_log2_lookup[bs]) >> 4) +
- avg_y_stride;
+ sig = sig - b_width + sig_stride;
+ mc_running_avg_y = mc_running_avg_y - b_width + mc_avg_y_stride;
+ running_avg_y = running_avg_y - b_width + avg_y_stride;
}
if (abs(sum_diff) > sum_diff_thresh) {
return COPY_BLOCK;
generated by cgit v1.2.3 (git 2.25.1) at 2024-10-16 08:08:40 +0000