Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(114)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: source/libvpx/vp9/encoder/x86/vp9_denoiser_sse2.c

Issue 756673003: libvpx: Pull from upstream (Closed) Base URL: svn://svn.chromium.org/chrome/trunk/deps/third_party/libvpx/
Patch Set: Created 6 years, 1 month ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « source/libvpx/vp9/encoder/x86/vp9_dct_ssse3.c ('k') | source/libvpx/vp9/encoder/x86/vp9_highbd_block_error_intrin_sse2.c » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: source/libvpx/vp9/encoder/x86/vp9_denoiser_sse2.c
===================================================================
--- source/libvpx/vp9/encoder/x86/vp9_denoiser_sse2.c (revision 293081)
+++ source/libvpx/vp9/encoder/x86/vp9_denoiser_sse2.c (working copy)
@@ -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 @@
__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 @@
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);
@@ -95,17 +94,15 @@
}
// 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.
@@ -112,7 +109,7 @@
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);
@@ -128,19 +125,16 @@
return acc_diff;
}
-static int vp9_denoiser_4xM_sse2(const uint8_t *sig, int sig_stride,
- const uint8_t *mc_running_avg_y,
- int mc_avg_y_stride,
- uint8_t *running_avg_y, int avg_y_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];
+// Denoiser for 4xM and 8xM blocks.
+static int vp9_denoiser_NxM_sse2_small(
+ const uint8_t *sig, int sig_stride, const uint8_t *mc_running_avg_y,
+ int mc_avg_y_stride, uint8_t *running_avg_y, int avg_y_stride,
+ int increase_denoising, BLOCK_SIZE bs, int motion_magnitude, int width) {
+ 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);
@@ -148,48 +142,51 @@
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;
+ const uint8_t shift = (width == 4) ? 2 : 1;
- for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> 2); ++r) {
- vpx_memcpy(sig_buffer[r], sig, 4);
- vpx_memcpy(sig_buffer[r] + 4, sig + sig_stride, 4);
- vpx_memcpy(sig_buffer[r] + 8, sig + sig_stride * 2, 4);
- vpx_memcpy(sig_buffer[r] + 12, sig + sig_stride * 3, 4);
- vpx_memcpy(mc_running_buffer[r], mc_running_avg_y, 4);
- vpx_memcpy(mc_running_buffer[r] + 4, mc_running_avg_y +
- mc_avg_y_stride, 4);
- vpx_memcpy(mc_running_buffer[r] + 8, mc_running_avg_y +
- mc_avg_y_stride * 2, 4);
- vpx_memcpy(mc_running_buffer[r] + 12, mc_running_avg_y +
- mc_avg_y_stride * 3, 4);
- vpx_memcpy(running_buffer[r], running_avg_y, 4);
- vpx_memcpy(running_buffer[r] + 4, running_avg_y +
- avg_y_stride, 4);
- vpx_memcpy(running_buffer[r] + 8, running_avg_y +
- avg_y_stride * 2, 4);
- vpx_memcpy(running_buffer[r] + 12, running_avg_y +
- avg_y_stride * 3, 4);
+ for (r = 0; r < ((4 << b_height_log2_lookup[bs]) >> shift); ++r) {
+ vpx_memcpy(sig_buffer[r], sig, width);
+ vpx_memcpy(sig_buffer[r] + width, sig + sig_stride, width);
+ vpx_memcpy(mc_running_buffer[r], mc_running_avg_y, width);
+ vpx_memcpy(mc_running_buffer[r] + width,
+ mc_running_avg_y + mc_avg_y_stride, width);
+ vpx_memcpy(running_buffer[r], running_avg_y, width);
+ vpx_memcpy(running_buffer[r] + width,
+ running_avg_y + avg_y_stride, width);
+ if (width == 4) {
+ vpx_memcpy(sig_buffer[r] + width * 2, sig + sig_stride * 2, width);
+ vpx_memcpy(sig_buffer[r] + width * 3, sig + sig_stride * 3, width);
+ vpx_memcpy(mc_running_buffer[r] + width * 2,
+ mc_running_avg_y + mc_avg_y_stride * 2, width);
+ vpx_memcpy(mc_running_buffer[r] + width * 3,
+ mc_running_avg_y + mc_avg_y_stride * 3, width);
+ vpx_memcpy(running_buffer[r] + width * 2,
+ running_avg_y + avg_y_stride * 2, width);
+ vpx_memcpy(running_buffer[r] + width * 3,
+ running_avg_y + avg_y_stride * 3, width);
+ }
acc_diff = vp9_denoiser_16x1_sse2(sig_buffer[r],
mc_running_buffer[r],
running_buffer[r],
&k_0, &k_4, &k_8, &k_16,
&l3, &l32, &l21, 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,
- running_buffer[r] + 8, 4);
- vpx_memcpy(running_avg_y + avg_y_stride * 3,
- running_buffer[r] + 12, 4);
+ vpx_memcpy(running_avg_y, running_buffer[r], width);
+ vpx_memcpy(running_avg_y + avg_y_stride, running_buffer[r] + width, width);
+ if (width == 4) {
+ vpx_memcpy(running_avg_y + avg_y_stride * 2,
+ running_buffer[r] + width * 2, width);
+ vpx_memcpy(running_avg_y + avg_y_stride * 3,
+ running_buffer[r] + width * 3, width);
+ }
// Update pointers for next iteration.
- sig += (sig_stride << 2);
- mc_running_avg_y += (mc_avg_y_stride << 2);
- running_avg_y += (avg_y_stride << 2);
+ sig += (sig_stride << shift);
+ mc_running_avg_y += (mc_avg_y_stride << shift);
+ running_avg_y += (avg_y_stride << shift);
}
{
@@ -197,35 +194,36 @@
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]);
- 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]) >> shift); ++r) {
acc_diff = vp9_denoiser_adj_16x1_sse2(
- 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,
- running_buffer[r] + 8, 4);
- vpx_memcpy(running_avg_y + avg_y_stride * 3,
- running_buffer[r] + 12, 4);
+ sig_buffer[r], mc_running_buffer[r], running_buffer[r],
+ k_0, k_delta, acc_diff);
+ vpx_memcpy(running_avg_y, running_buffer[r], width);
+ vpx_memcpy(running_avg_y + avg_y_stride,
+ running_buffer[r] + width, width);
+ if (width == 4) {
+ vpx_memcpy(running_avg_y + avg_y_stride * 2,
+ running_buffer[r] + width * 2, width);
+ vpx_memcpy(running_avg_y + avg_y_stride * 3,
+ running_buffer[r] + width * 3, width);
+ }
// Update pointers for next iteration.
- running_avg_y += (avg_y_stride << 2);
+ running_avg_y += (avg_y_stride << shift);
}
sum_diff = sum_diff_16x1(acc_diff);
if (abs(sum_diff) > sum_diff_thresh) {
@@ -239,109 +237,18 @@
return FILTER_BLOCK;
}
-static int vp9_denoiser_8xM_sse2(const uint8_t *sig, int sig_stride,
- const uint8_t *mc_running_avg_y,
- int mc_avg_y_stride,
- uint8_t *running_avg_y, int avg_y_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];
- __m128i acc_diff = _mm_setzero_si128();
- const __m128i k_0 = _mm_setzero_si128();
- const __m128i k_4 = _mm_set1_epi8(4 + shift_inc);
- const __m128i k_8 = _mm_set1_epi8(8);
- 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);
- // 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);
- vpx_memcpy(sig_buffer[r] + 8, sig + sig_stride, 8);
- vpx_memcpy(mc_running_buffer[r], mc_running_avg_y, 8);
- vpx_memcpy(mc_running_buffer[r] + 8, mc_running_avg_y +
- mc_avg_y_stride, 8);
- vpx_memcpy(running_buffer[r], running_avg_y, 8);
- vpx_memcpy(running_buffer[r] + 8, running_avg_y +
- avg_y_stride, 8);
- acc_diff = vp9_denoiser_16x1_sse2(sig_buffer[r],
- mc_running_buffer[r],
- running_buffer[r],
- &k_0, &k_4, &k_8, &k_16,
- &l3, &l32, &l21, 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.
- sig += (sig_stride << 1);
- mc_running_avg_y += (mc_avg_y_stride << 1);
- running_avg_y += (avg_y_stride << 1);
- }
-
- {
- sum_diff = sum_diff_16x1(acc_diff);
- 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
- // 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.
-
- // 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;
- // 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);
- 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.
- running_avg_y += (avg_y_stride << 1);
- }
- sum_diff = sum_diff_16x1(acc_diff);
- if (abs(sum_diff) > sum_diff_thresh) {
- return COPY_BLOCK;
- }
- } else {
- return COPY_BLOCK;
- }
- }
- }
- return FILTER_BLOCK;
-}
-
-static int vp9_denoiser_64_32_16xM_sse2(const uint8_t *sig, int sig_stride,
- const uint8_t *mc_running_avg_y,
- int mc_avg_y_stride,
- uint8_t *running_avg_y,
- 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;
+// Denoiser for 16xM, 32xM and 64xM blocks
+static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
+ const uint8_t *mc_running_avg_y,
+ int mc_avg_y_stride,
+ uint8_t *running_avg_y,
+ int avg_y_stride,
+ int increase_denoising, BLOCK_SIZE bs,
+ int motion_magnitude) {
+ 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 +256,11 @@
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 +268,11 @@
}
}
- 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 +288,8 @@
// 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 +298,8 @@
{
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 +311,8 @@
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;
@@ -449,25 +351,25 @@
BLOCK_SIZE bs,
int motion_magnitude) {
if (bs == BLOCK_4X4 || bs == BLOCK_4X8) {
- return vp9_denoiser_4xM_sse2(sig, sig_stride,
- mc_avg, mc_avg_stride,
- avg, avg_stride,
- increase_denoising,
- bs, motion_magnitude);
+ return vp9_denoiser_NxM_sse2_small(sig, sig_stride,
+ mc_avg, mc_avg_stride,
+ avg, avg_stride,
+ increase_denoising,
+ bs, motion_magnitude, 4);
} else if (bs == BLOCK_8X4 || bs == BLOCK_8X8 || bs == BLOCK_8X16) {
- return vp9_denoiser_8xM_sse2(sig, sig_stride,
- mc_avg, mc_avg_stride,
- avg, avg_stride,
- increase_denoising,
- bs, motion_magnitude);
+ return vp9_denoiser_NxM_sse2_small(sig, sig_stride,
+ mc_avg, mc_avg_stride,
+ avg, avg_stride,
+ increase_denoising,
+ bs, motion_magnitude, 8);
} else if (bs == BLOCK_16X8 || bs == BLOCK_16X16 || bs == BLOCK_16X32 ||
bs == BLOCK_32X16|| bs == BLOCK_32X32 || bs == BLOCK_32X64 ||
bs == BLOCK_64X32 || bs == BLOCK_64X64) {
- return vp9_denoiser_64_32_16xM_sse2(sig, sig_stride,
- mc_avg, mc_avg_stride,
- avg, avg_stride,
- increase_denoising,
- bs, motion_magnitude);
+ return vp9_denoiser_NxM_sse2_big(sig, sig_stride,
+ mc_avg, mc_avg_stride,
+ avg, avg_stride,
+ increase_denoising,
+ bs, motion_magnitude);
} else {
return COPY_BLOCK;
}
« no previous file with comments | « source/libvpx/vp9/encoder/x86/vp9_dct_ssse3.c ('k') | source/libvpx/vp9/encoder/x86/vp9_highbd_block_error_intrin_sse2.c » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698