1 /*
2 * Copyright (c) 2012 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <arm_neon.h>
12
13 #include "vp8/encoder/denoising.h"
14 #include "vpx_mem/vpx_mem.h"
15 #include "./vp8_rtcd.h"
16
17 /*
18 * The filter function was modified to reduce the computational complexity.
19 *
20 * Step 1:
21 * Instead of applying tap coefficients for each pixel, we calculated the
22 * pixel adjustments vs. pixel diff value ahead of time.
23 * adjustment = filtered_value - current_raw
24 * = (filter_coefficient * diff + 128) >> 8
25 * where
26 * filter_coefficient = (255 << 8) / (256 + ((abs_diff * 330) >> 3));
27 * filter_coefficient += filter_coefficient /
28 * (3 + motion_magnitude_adjustment);
29 * filter_coefficient is clamped to 0 ~ 255.
30 *
31 * Step 2:
32 * The adjustment vs. diff curve becomes flat very quick when diff increases.
33 * This allowed us to use only several levels to approximate the curve without
34 * changing the filtering algorithm too much.
35 * The adjustments were further corrected by checking the motion magnitude.
36 * The levels used are:
37 * diff level adjustment w/o adjustment w/
38 * motion correction motion correction
39 * [-255, -16] 3 -6 -7
40 * [-15, -8] 2 -4 -5
41 * [-7, -4] 1 -3 -4
42 * [-3, 3] 0 diff diff
43 * [4, 7] 1 3 4
44 * [8, 15] 2 4 5
45 * [16, 255] 3 6 7
46 */
47
vp8_denoiser_filter_neon(YV12_BUFFER_CONFIG * mc_running_avg,YV12_BUFFER_CONFIG * running_avg,MACROBLOCK * signal,unsigned int motion_magnitude,int y_offset,int uv_offset)48 int vp8_denoiser_filter_neon(YV12_BUFFER_CONFIG *mc_running_avg,
49 YV12_BUFFER_CONFIG *running_avg,
50 MACROBLOCK *signal, unsigned int motion_magnitude,
51 int y_offset, int uv_offset) {
52 /* If motion_magnitude is small, making the denoiser more aggressive by
53 * increasing the adjustment for each level, level1 adjustment is
54 * increased, the deltas stay the same.
55 */
56 const uint8x16_t v_level1_adjustment = vdupq_n_u8(
57 (motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ? 4 : 3);
58 const uint8x16_t v_delta_level_1_and_2 = vdupq_n_u8(1);
59 const uint8x16_t v_delta_level_2_and_3 = vdupq_n_u8(2);
60 const uint8x16_t v_level1_threshold = vdupq_n_u8(4);
61 const uint8x16_t v_level2_threshold = vdupq_n_u8(8);
62 const uint8x16_t v_level3_threshold = vdupq_n_u8(16);
63
64 /* Local variables for array pointers and strides. */
65 unsigned char *sig = signal->thismb;
66 int sig_stride = 16;
67 unsigned char *mc_running_avg_y = mc_running_avg->y_buffer + y_offset;
68 int mc_running_avg_y_stride = mc_running_avg->y_stride;
69 unsigned char *running_avg_y = running_avg->y_buffer + y_offset;
70 int running_avg_y_stride = running_avg->y_stride;
71
72 /* Go over lines. */
73 int i;
74 int sum_diff = 0;
75 for (i = 0; i < 16; ++i) {
76 int8x16_t v_sum_diff = vdupq_n_s8(0);
77 uint8x16_t v_running_avg_y;
78
79 /* Load inputs. */
80 const uint8x16_t v_sig = vld1q_u8(sig);
81 const uint8x16_t v_mc_running_avg_y = vld1q_u8(mc_running_avg_y);
82
83 /* Calculate absolute difference and sign masks. */
84 const uint8x16_t v_abs_diff = vabdq_u8(v_sig, v_mc_running_avg_y);
85 const uint8x16_t v_diff_pos_mask = vcltq_u8(v_sig, v_mc_running_avg_y);
86 const uint8x16_t v_diff_neg_mask = vcgtq_u8(v_sig, v_mc_running_avg_y);
87
88 /* Figure out which level that put us in. */
89 const uint8x16_t v_level1_mask = vcleq_u8(v_level1_threshold,
90 v_abs_diff);
91 const uint8x16_t v_level2_mask = vcleq_u8(v_level2_threshold,
92 v_abs_diff);
93 const uint8x16_t v_level3_mask = vcleq_u8(v_level3_threshold,
94 v_abs_diff);
95
96 /* Calculate absolute adjustments for level 1, 2 and 3. */
97 const uint8x16_t v_level2_adjustment = vandq_u8(v_level2_mask,
98 v_delta_level_1_and_2);
99 const uint8x16_t v_level3_adjustment = vandq_u8(v_level3_mask,
100 v_delta_level_2_and_3);
101 const uint8x16_t v_level1and2_adjustment = vaddq_u8(v_level1_adjustment,
102 v_level2_adjustment);
103 const uint8x16_t v_level1and2and3_adjustment = vaddq_u8(
104 v_level1and2_adjustment, v_level3_adjustment);
105
106 /* Figure adjustment absolute value by selecting between the absolute
107 * difference if in level0 or the value for level 1, 2 and 3.
108 */
109 const uint8x16_t v_abs_adjustment = vbslq_u8(v_level1_mask,
110 v_level1and2and3_adjustment, v_abs_diff);
111
112 /* Calculate positive and negative adjustments. Apply them to the signal
113 * and accumulate them. Adjustments are less than eight and the maximum
114 * sum of them (7 * 16) can fit in a signed char.
115 */
116 const uint8x16_t v_pos_adjustment = vandq_u8(v_diff_pos_mask,
117 v_abs_adjustment);
118 const uint8x16_t v_neg_adjustment = vandq_u8(v_diff_neg_mask,
119 v_abs_adjustment);
120 v_running_avg_y = vqaddq_u8(v_sig, v_pos_adjustment);
121 v_running_avg_y = vqsubq_u8(v_running_avg_y, v_neg_adjustment);
122 v_sum_diff = vqaddq_s8(v_sum_diff,
123 vreinterpretq_s8_u8(v_pos_adjustment));
124 v_sum_diff = vqsubq_s8(v_sum_diff,
125 vreinterpretq_s8_u8(v_neg_adjustment));
126
127 /* Store results. */
128 vst1q_u8(running_avg_y, v_running_avg_y);
129
130 /* Sum all the accumulators to have the sum of all pixel differences
131 * for this macroblock.
132 */
133 {
134 int s0 = vgetq_lane_s8(v_sum_diff, 0) +
135 vgetq_lane_s8(v_sum_diff, 1) +
136 vgetq_lane_s8(v_sum_diff, 2) +
137 vgetq_lane_s8(v_sum_diff, 3);
138 int s1 = vgetq_lane_s8(v_sum_diff, 4) +
139 vgetq_lane_s8(v_sum_diff, 5) +
140 vgetq_lane_s8(v_sum_diff, 6) +
141 vgetq_lane_s8(v_sum_diff, 7);
142 int s2 = vgetq_lane_s8(v_sum_diff, 8) +
143 vgetq_lane_s8(v_sum_diff, 9) +
144 vgetq_lane_s8(v_sum_diff, 10) +
145 vgetq_lane_s8(v_sum_diff, 11);
146 int s3 = vgetq_lane_s8(v_sum_diff, 12) +
147 vgetq_lane_s8(v_sum_diff, 13) +
148 vgetq_lane_s8(v_sum_diff, 14) +
149 vgetq_lane_s8(v_sum_diff, 15);
150 sum_diff += s0 + s1+ s2 + s3;
151 }
152
153 /* Update pointers for next iteration. */
154 sig += sig_stride;
155 mc_running_avg_y += mc_running_avg_y_stride;
156 running_avg_y += running_avg_y_stride;
157 }
158
159 /* Too much adjustments => copy block. */
160 if (abs(sum_diff) > SUM_DIFF_THRESHOLD)
161 return COPY_BLOCK;
162
163 /* Tell above level that block was filtered. */
164 vp8_copy_mem16x16(running_avg->y_buffer + y_offset, running_avg_y_stride,
165 signal->thismb, sig_stride);
166 return FILTER_BLOCK;
167 }
168