generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

vp9_denoiser_sse2.c: improve code style.

Browse Source 
denoiser_sse2.c: fix typos in comment.

Change-Id: Ic0fb102331b0e533c058da3cab1fbc30de9a0070
This commit is contained in:
JackyChen
2014-10-17 12:16:13 -07:00
parent 0615a196e2
commit 5cba6516aa
2 changed files with 62 additions and 78 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vp8/encoder/x86/denoising_sse2.c
View File

@@ -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 (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH;
if (abs_sum_diff > sum_diff_thresh) { if (abs_sum_diff > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising), // 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 // this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it // 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 // closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be // 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 // The delta is set by the excess of absolute pixel diff over the
// threshold. // 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 (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH_UV;
if (abs_sum_diff > sum_diff_thresh) { if (abs_sum_diff > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising), // 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 // this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it // 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 // closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be // 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 // The delta is set by the excess of absolute pixel diff over the
// threshold. // threshold.

118
vp9/encoder/x86/vp9_denoiser_sse2.c
View File

@@ -23,18 +23,17 @@
// Compute the sum of all pixel differences of this MB. // Compute the sum of all pixel differences of this MB.
static INLINE int sum_diff_16x1(__m128i acc_diff) { static INLINE int sum_diff_16x1(__m128i acc_diff) {
const __m128i k_1 = _mm_set1_epi16(1); const __m128i k_1 = _mm_set1_epi16(1);
const __m128i acc_diff_lo = _mm_srai_epi16( const __m128i acc_diff_lo =
_mm_unpacklo_epi8(acc_diff, acc_diff), 8); _mm_srai_epi16(_mm_unpacklo_epi8(acc_diff, acc_diff), 8);
const __m128i acc_diff_hi = _mm_srai_epi16( const __m128i acc_diff_hi =
_mm_unpackhi_epi8(acc_diff, acc_diff), 8); _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 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 hg_fe_dc_ba = _mm_madd_epi16(acc_diff_16, k_1);
const __m128i hgfe_dcba = _mm_add_epi32(hg_fe_dc_ba, const __m128i hgfe_dcba =
_mm_srli_si128(hg_fe_dc_ba, 8)); _mm_add_epi32(hg_fe_dc_ba, _mm_srli_si128(hg_fe_dc_ba, 8));
const __m128i hgfedcba = _mm_add_epi32(hgfe_dcba, const __m128i hgfedcba =
_mm_srli_si128(hgfe_dcba, 4)); _mm_add_epi32(hgfe_dcba, _mm_srli_si128(hgfe_dcba, 4));
int sum_diff = _mm_cvtsi128_si32(hgfedcba); return _mm_cvtsi128_si32(hgfedcba);
return sum_diff;
} }
// Denoise a 16x1 vector. // Denoise a 16x1 vector.
@@ -51,8 +50,8 @@ static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
__m128i acc_diff) { __m128i acc_diff) {
// Calculate differences // Calculate differences
const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0])); const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0]));
const __m128i v_mc_running_avg_y = _mm_loadu_si128( const __m128i v_mc_running_avg_y =
(const __m128i *)(&mc_running_avg_y[0])); _mm_loadu_si128((const __m128i *)(&mc_running_avg_y[0]));
__m128i v_running_avg_y; __m128i v_running_avg_y;
const __m128i pdiff = _mm_subs_epu8(v_mc_running_avg_y, v_sig); 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); 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); const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, *k_0);
// Clamp absolute difference to 16 to be used to get mask. Doing this // 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. // allows us to use _mm_cmpgt_epi8, which operates on signed byte.
const __m128i clamped_absdiff = _mm_min_epu8( const __m128i clamped_absdiff =
_mm_or_si128(pdiff, ndiff), *k_16); _mm_min_epu8(_mm_or_si128(pdiff, ndiff), *k_16);
// Get masks for l2 l1 and l0 adjustments. // Get masks for l2 l1 and l0 adjustments.
const __m128i mask2 = _mm_cmpgt_epi8(*k_16, clamped_absdiff); const __m128i mask2 = _mm_cmpgt_epi8(*k_16, clamped_absdiff);
const __m128i mask1 = _mm_cmpgt_epi8(*k_8, clamped_absdiff); const __m128i mask1 = _mm_cmpgt_epi8(*k_8, clamped_absdiff);
@@ -95,12 +94,10 @@ static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
} }
// Denoise a 16x1 vector with a weaker filter. // Denoise a 16x1 vector with a weaker filter.
static INLINE __m128i vp9_denoiser_adj_16x1_sse2(const uint8_t *sig, static INLINE __m128i vp9_denoiser_adj_16x1_sse2(
const uint8_t *mc_running_avg_y, const uint8_t *sig, const uint8_t *mc_running_avg_y,
uint8_t *running_avg_y, uint8_t *running_avg_y, const __m128i k_0,
const __m128i k_0, const __m128i k_delta, __m128i acc_diff) {
const __m128i k_delta,
__m128i acc_diff) {
__m128i v_running_avg_y = _mm_loadu_si128((__m128i *)(&running_avg_y[0])); __m128i v_running_avg_y = _mm_loadu_si128((__m128i *)(&running_avg_y[0]));
// Calculate differences. // Calculate differences.
const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0])); const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0]));
@@ -135,12 +132,11 @@ static int vp9_denoiser_4xM_sse2(const uint8_t *sig, int sig_stride,
int increase_denoising, int increase_denoising,
BLOCK_SIZE bs, BLOCK_SIZE bs,
int motion_magnitude) { int motion_magnitude) {
int sum_diff_thresh; int sum_diff_thresh, r, sum_diff = 0;
int r; const int shift_inc = (increase_denoising &&
int shift_inc = (increase_denoising && motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 1 : 0; 1 : 0;
unsigned char sig_buffer[2][16], mc_running_buffer[2][16], uint8_t sig_buffer[2][16], mc_running_buffer[2][16], running_buffer[2][16];
running_buffer[2][16];
__m128i acc_diff = _mm_setzero_si128(); __m128i acc_diff = _mm_setzero_si128();
const __m128i k_0 = _mm_setzero_si128(); const __m128i k_0 = _mm_setzero_si128();
const __m128i k_4 = _mm_set1_epi8(4 + shift_inc); 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); const __m128i k_16 = _mm_set1_epi8(16);
// Modify each level's adjustment according to motion_magnitude. // Modify each level's adjustment according to motion_magnitude.
const __m128i l3 = _mm_set1_epi8( const __m128i l3 = _mm_set1_epi8(
(motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 + shift_inc : 6);
7 + shift_inc : 6);
// Difference between level 3 and level 2 is 2. // Difference between level 3 and level 2 is 2.
const __m128i l32 = _mm_set1_epi8(2); const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1. // Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1); const __m128i l21 = _mm_set1_epi8(1);
int sum_diff = 0;
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 2); ++r) { for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 2); ++r) {
vpx_memcpy(sig_buffer[r], sig, 4); 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); sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising);
if (abs(sum_diff) > sum_diff_thresh) { if (abs(sum_diff) > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising), // 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 // this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it // 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 // closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be // 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 // The delta is set by the excess of absolute pixel diff over the
// threshold. // threshold.
int delta = ((abs(sum_diff) - sum_diff_thresh) const int delta = ((abs(sum_diff) - sum_diff_thresh) >>
>> num_pels_log2_lookup[bs]) + 1; num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3. // Only apply the adjustment for max delta up to 3.
if (delta < 4) { if (delta < 4) {
const __m128i k_delta = _mm_set1_epi8(delta); 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; sum_diff = 0;
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 2); ++r) { for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 2); ++r) {
acc_diff = vp9_denoiser_adj_16x1_sse2( acc_diff = vp9_denoiser_adj_16x1_sse2(
sig_buffer[r], mc_running_buffer[r], sig_buffer[r], mc_running_buffer[r], running_buffer[r],
running_buffer[r], k_0, k_delta, k_0, k_delta, acc_diff);
acc_diff);
vpx_memcpy(running_avg_y, running_buffer[r], 4); 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, running_buffer[r] + 4, 4);
vpx_memcpy(running_avg_y + avg_y_stride * 2, 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, int increase_denoising,
BLOCK_SIZE bs, BLOCK_SIZE bs,
int motion_magnitude) { int motion_magnitude) {
int sum_diff_thresh; int sum_diff_thresh, r, sum_diff = 0;
int r; const int shift_inc = (increase_denoising &&
int shift_inc = (increase_denoising && motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 1 : 0; 1 : 0;
unsigned char sig_buffer[8][16], mc_running_buffer[8][16], uint8_t sig_buffer[8][16], mc_running_buffer[8][16], running_buffer[8][16];
running_buffer[8][16];
__m128i acc_diff = _mm_setzero_si128(); __m128i acc_diff = _mm_setzero_si128();
const __m128i k_0 = _mm_setzero_si128(); const __m128i k_0 = _mm_setzero_si128();
const __m128i k_4 = _mm_set1_epi8(4 + shift_inc); 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); const __m128i k_16 = _mm_set1_epi8(16);
// Modify each level's adjustment according to motion_magnitude. // Modify each level's adjustment according to motion_magnitude.
const __m128i l3 = _mm_set1_epi8( const __m128i l3 = _mm_set1_epi8(
(motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 + shift_inc : 6);
7 + shift_inc : 6);
// Difference between level 3 and level 2 is 2. // Difference between level 3 and level 2 is 2.
const __m128i l32 = _mm_set1_epi8(2); const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1. // Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1); const __m128i l21 = _mm_set1_epi8(1);
int sum_diff = 0;
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 1); ++r) { for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 1); ++r) {
vpx_memcpy(sig_buffer[r], sig, 8); vpx_memcpy(sig_buffer[r], sig, 8);
@@ -294,16 +284,16 @@ static int vp9_denoiser_8xM_sse2(const uint8_t *sig, int sig_stride,
sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising); sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising);
if (abs(sum_diff) > sum_diff_thresh) { if (abs(sum_diff) > sum_diff_thresh) {
// Before returning to copy the block (i.e., apply no denoising), // 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 // this block, that would otherwise not be denoised at all. Simplest
// is to apply an additional adjustment to running_avg_y to bring it // 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 // closer to sig. The adjustment is capped by a maximum delta, and
// chosen such that in most cases the resulting sum_diff will be // 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 // The delta is set by the excess of absolute pixel diff over the
// threshold. // threshold.
int delta = ((abs(sum_diff) - sum_diff_thresh) const int delta = ((abs(sum_diff) - sum_diff_thresh)
>> num_pels_log2_lookup[bs]) + 1; >> num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3. // Only apply the adjustment for max delta up to 3.
if (delta < 4) { if (delta < 4) {
@@ -311,9 +301,8 @@ static int vp9_denoiser_8xM_sse2(const uint8_t *sig, int sig_stride,
running_avg_y -= avg_y_stride * (4 << b_height_log2_lookup[bs]); running_avg_y -= avg_y_stride * (4 << b_height_log2_lookup[bs]);
for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 1); ++r) { for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 1); ++r) {
acc_diff = vp9_denoiser_adj_16x1_sse2( acc_diff = vp9_denoiser_adj_16x1_sse2(
sig_buffer[r], mc_running_buffer[r], sig_buffer[r], mc_running_buffer[r], running_buffer[r],
running_buffer[r], k_0, k_delta, k_0, k_delta, acc_diff);
acc_diff);
vpx_memcpy(running_avg_y, running_buffer[r], 8); vpx_memcpy(running_avg_y, running_buffer[r], 8);
vpx_memcpy(running_avg_y + avg_y_stride, running_buffer[r] + 8, 8); vpx_memcpy(running_avg_y + avg_y_stride, running_buffer[r] + 8, 8);
// Update pointers for next iteration. // Update pointers for next iteration.
@@ -338,10 +327,10 @@ static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
int avg_y_stride, int avg_y_stride,
int increase_denoising, BLOCK_SIZE bs, int increase_denoising, BLOCK_SIZE bs,
int motion_magnitude) { int motion_magnitude) {
int sum_diff_thresh; int sum_diff_thresh, r, c, sum_diff = 0;
int r, c; const int shift_inc = (increase_denoising &&
int shift_inc = (increase_denoising && motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 1 : 0; 1 : 0;
__m128i acc_diff[4][4]; __m128i acc_diff[4][4];
const __m128i k_0 = _mm_setzero_si128(); const __m128i k_0 = _mm_setzero_si128();
const __m128i k_4 = _mm_set1_epi8(4 + shift_inc); 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); const __m128i k_16 = _mm_set1_epi8(16);
// Modify each level's adjustment according to motion_magnitude. // Modify each level's adjustment according to motion_magnitude.
const __m128i l3 = _mm_set1_epi8( const __m128i l3 = _mm_set1_epi8(
(motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 + shift_inc : 6);
7 + shift_inc : 6);
// Difference between level 3 and level 2 is 2. // Difference between level 3 and level 2 is 2.
const __m128i l32 = _mm_set1_epi8(2); const __m128i l32 = _mm_set1_epi8(2);
// Difference between level 2 and level 1 is 1. // Difference between level 2 and level 1 is 1.
const __m128i l21 = _mm_set1_epi8(1); const __m128i l21 = _mm_set1_epi8(1);
int sum_diff = 0;
for (c = 0; c < 4; ++c) { for (c = 0; c < 4; ++c) {
for (r = 0; r < 4; ++r) { 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) { for (c = 0; c < (4 << b_width_log2_lookup[bs]); c += 16) {
acc_diff[c>>4][r>>4] = vp9_denoiser_16x1_sse2( acc_diff[c>>4][r>>4] = vp9_denoiser_16x1_sse2(
sig, mc_running_avg_y, sig, mc_running_avg_y, running_avg_y, &k_0, &k_4,
running_avg_y, &k_8, &k_16, &l3, &l32, &l21, acc_diff[c>>4][r>>4]);
&k_0, &k_4, &k_8, &k_16,
&l3, &l32, &l21, acc_diff[c>>4][r>>4]);
// Update pointers for next iteration. // Update pointers for next iteration.
sig += 16; sig += 16;
mc_running_avg_y += 16; mc_running_avg_y += 16;
@@ -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); sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising);
if (abs(sum_diff) > sum_diff_thresh) { if (abs(sum_diff) > sum_diff_thresh) {
int delta = ((abs(sum_diff) - sum_diff_thresh) const int delta = ((abs(sum_diff) - sum_diff_thresh) >>
>> num_pels_log2_lookup[bs]) + 1; num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3. // Only apply the adjustment for max delta up to 3.
if (delta < 4) { if (delta < 4) {
@@ -408,8 +393,7 @@ 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) { for (c = 0; c < (4 << b_width_log2_lookup[bs]); c += 16) {
acc_diff[c>>4][r>>4] = vp9_denoiser_adj_16x1_sse2( acc_diff[c>>4][r>>4] = vp9_denoiser_adj_16x1_sse2(
sig, mc_running_avg_y, sig, mc_running_avg_y, running_avg_y, k_0,
running_avg_y, k_0,
k_delta, acc_diff[c>>4][r>>4]); k_delta, acc_diff[c>>4][r>>4]);
// Update pointers for next iteration. // Update pointers for next iteration.
sig += 16; sig += 16;