1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/x86/denoising_sse2.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,120 @@
1.4 +/*
1.5 + * Copyright (c) 2012 The WebM project authors. All Rights Reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +#include "vp8/encoder/denoising.h"
1.15 +#include "vp8/common/reconinter.h"
1.16 +#include "vpx/vpx_integer.h"
1.17 +#include "vpx_mem/vpx_mem.h"
1.18 +#include "vp8_rtcd.h"
1.19 +
1.20 +#include <emmintrin.h>
1.21 +#include "vpx_ports/emmintrin_compat.h"
1.22 +
1.23 +union sum_union {
1.24 + __m128i v;
1.25 + signed char e[16];
1.26 +};
1.27 +
1.28 +int vp8_denoiser_filter_sse2(YV12_BUFFER_CONFIG *mc_running_avg,
1.29 + YV12_BUFFER_CONFIG *running_avg,
1.30 + MACROBLOCK *signal, unsigned int motion_magnitude,
1.31 + int y_offset, int uv_offset)
1.32 +{
1.33 + unsigned char *sig = signal->thismb;
1.34 + int sig_stride = 16;
1.35 + unsigned char *mc_running_avg_y = mc_running_avg->y_buffer + y_offset;
1.36 + int mc_avg_y_stride = mc_running_avg->y_stride;
1.37 + unsigned char *running_avg_y = running_avg->y_buffer + y_offset;
1.38 + int avg_y_stride = running_avg->y_stride;
1.39 + int r;
1.40 + __m128i acc_diff = _mm_setzero_si128();
1.41 + const __m128i k_0 = _mm_setzero_si128();
1.42 + const __m128i k_4 = _mm_set1_epi8(4);
1.43 + const __m128i k_8 = _mm_set1_epi8(8);
1.44 + const __m128i k_16 = _mm_set1_epi8(16);
1.45 + /* Modify each level's adjustment according to motion_magnitude. */
1.46 + const __m128i l3 = _mm_set1_epi8(
1.47 + (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 7 : 6);
1.48 + /* Difference between level 3 and level 2 is 2. */
1.49 + const __m128i l32 = _mm_set1_epi8(2);
1.50 + /* Difference between level 2 and level 1 is 1. */
1.51 + const __m128i l21 = _mm_set1_epi8(1);
1.52 +
1.53 + for (r = 0; r < 16; ++r)
1.54 + {
1.55 + /* Calculate differences */
1.56 + const __m128i v_sig = _mm_loadu_si128((__m128i *)(&sig[0]));
1.57 + const __m128i v_mc_running_avg_y = _mm_loadu_si128(
1.58 + (__m128i *)(&mc_running_avg_y[0]));
1.59 + __m128i v_running_avg_y;
1.60 + const __m128i pdiff = _mm_subs_epu8(v_mc_running_avg_y, v_sig);
1.61 + const __m128i ndiff = _mm_subs_epu8(v_sig, v_mc_running_avg_y);
1.62 + /* Obtain the sign. FF if diff is negative. */
1.63 + const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, k_0);
1.64 + /* Clamp absolute difference to 16 to be used to get mask. Doing this
1.65 + * allows us to use _mm_cmpgt_epi8, which operates on signed byte. */
1.66 + const __m128i clamped_absdiff = _mm_min_epu8(
1.67 + _mm_or_si128(pdiff, ndiff), k_16);
1.68 + /* Get masks for l2 l1 and l0 adjustments */
1.69 + const __m128i mask2 = _mm_cmpgt_epi8(k_16, clamped_absdiff);
1.70 + const __m128i mask1 = _mm_cmpgt_epi8(k_8, clamped_absdiff);
1.71 + const __m128i mask0 = _mm_cmpgt_epi8(k_4, clamped_absdiff);
1.72 + /* Get adjustments for l2, l1, and l0 */
1.73 + __m128i adj2 = _mm_and_si128(mask2, l32);
1.74 + const __m128i adj1 = _mm_and_si128(mask1, l21);
1.75 + const __m128i adj0 = _mm_and_si128(mask0, clamped_absdiff);
1.76 + __m128i adj, padj, nadj;
1.77 +
1.78 + /* Combine the adjustments and get absolute adjustments. */
1.79 + adj2 = _mm_add_epi8(adj2, adj1);
1.80 + adj = _mm_sub_epi8(l3, adj2);
1.81 + adj = _mm_andnot_si128(mask0, adj);
1.82 + adj = _mm_or_si128(adj, adj0);
1.83 +
1.84 + /* Restore the sign and get positive and negative adjustments. */
1.85 + padj = _mm_andnot_si128(diff_sign, adj);
1.86 + nadj = _mm_and_si128(diff_sign, adj);
1.87 +
1.88 + /* Calculate filtered value. */
1.89 + v_running_avg_y = _mm_adds_epu8(v_sig, padj);
1.90 + v_running_avg_y = _mm_subs_epu8(v_running_avg_y, nadj);
1.91 + _mm_storeu_si128((__m128i *)running_avg_y, v_running_avg_y);
1.92 +
1.93 + /* Adjustments <=7, and each element in acc_diff can fit in signed
1.94 + * char.
1.95 + */
1.96 + acc_diff = _mm_adds_epi8(acc_diff, padj);
1.97 + acc_diff = _mm_subs_epi8(acc_diff, nadj);
1.98 +
1.99 + /* Update pointers for next iteration. */
1.100 + sig += sig_stride;
1.101 + mc_running_avg_y += mc_avg_y_stride;
1.102 + running_avg_y += avg_y_stride;
1.103 + }
1.104 +
1.105 + {
1.106 + /* Compute the sum of all pixel differences of this MB. */
1.107 + union sum_union s;
1.108 + int sum_diff = 0;
1.109 + s.v = acc_diff;
1.110 + sum_diff = s.e[0] + s.e[1] + s.e[2] + s.e[3] + s.e[4] + s.e[5]
1.111 + + s.e[6] + s.e[7] + s.e[8] + s.e[9] + s.e[10] + s.e[11]
1.112 + + s.e[12] + s.e[13] + s.e[14] + s.e[15];
1.113 +
1.114 + if (abs(sum_diff) > SUM_DIFF_THRESHOLD)
1.115 + {
1.116 + return COPY_BLOCK;
1.117 + }
1.118 + }
1.119 +
1.120 + vp8_copy_mem16x16(running_avg->y_buffer + y_offset, avg_y_stride,
1.121 + signal->thismb, sig_stride);
1.122 + return FILTER_BLOCK;
1.123 +}
