OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2012 The WebM project authors. All Rights Reserved. | 2 * Copyright (c) 2012 The WebM project authors. All Rights Reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
(...skipping 94 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
105 /* Compute the sum of all pixel differences of this MB. */ | 105 /* Compute the sum of all pixel differences of this MB. */ |
106 union sum_union s; | 106 union sum_union s; |
107 int sum_diff = 0; | 107 int sum_diff = 0; |
108 s.v = acc_diff; | 108 s.v = acc_diff; |
109 sum_diff = s.e[0] + s.e[1] + s.e[2] + s.e[3] + s.e[4] + s.e[5] | 109 sum_diff = s.e[0] + s.e[1] + s.e[2] + s.e[3] + s.e[4] + s.e[5] |
110 + s.e[6] + s.e[7] + s.e[8] + s.e[9] + s.e[10] + s.e[11] | 110 + s.e[6] + s.e[7] + s.e[8] + s.e[9] + s.e[10] + s.e[11] |
111 + s.e[12] + s.e[13] + s.e[14] + s.e[15]; | 111 + s.e[12] + s.e[13] + s.e[14] + s.e[15]; |
112 | 112 |
113 sum_diff_thresh = SUM_DIFF_THRESHOLD; | 113 sum_diff_thresh = SUM_DIFF_THRESHOLD; |
114 if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH; | 114 if (increase_denoising) sum_diff_thresh = SUM_DIFF_THRESHOLD_HIGH; |
115 if (abs(sum_diff) > sum_diff_thresh) | 115 if (abs(sum_diff) > sum_diff_thresh) { |
116 { | 116 // Before returning to copy the block (i.e., apply no denoising), |
| 117 // checK if we can still apply some (weaker) temporal filtering to |
| 118 // this block, that would otherwise not be denoised at all. Simplest |
| 119 // is to apply an additional adjustment to running_avg_y to bring it |
| 120 // closer to sig. The adjustment is capped by a maximum delta, and |
| 121 // chosen such that in most cases the resulting sum_diff will be |
| 122 // within the accceptable range given by sum_diff_thresh. |
| 123 |
| 124 // The delta is set by the excess of absolute pixel diff over the |
| 125 // threshold. |
| 126 int delta = ((abs(sum_diff) - sum_diff_thresh) >> 8) + 1; |
| 127 // Only apply the adjustment for max delta up to 3. |
| 128 if (delta < 4) { |
| 129 const __m128i k_delta = _mm_set1_epi8(delta); |
| 130 sig -= sig_stride * 16; |
| 131 mc_running_avg_y -= mc_avg_y_stride * 16; |
| 132 running_avg_y -= avg_y_stride * 16; |
| 133 for (r = 0; r < 16; ++r) { |
| 134 __m128i v_running_avg_y = |
| 135 _mm_loadu_si128((__m128i *)(&running_avg_y[0])); |
| 136 // Calculate differences. |
| 137 const __m128i v_sig = _mm_loadu_si128((__m128i *)(&sig[0])); |
| 138 const __m128i v_mc_running_avg_y = |
| 139 _mm_loadu_si128((__m128i *)(&mc_running_avg_y[0])); |
| 140 const __m128i pdiff = _mm_subs_epu8(v_mc_running_avg_y, v_sig); |
| 141 const __m128i ndiff = _mm_subs_epu8(v_sig, v_mc_running_avg_y); |
| 142 // Obtain the sign. FF if diff is negative. |
| 143 const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, k_0); |
| 144 // Clamp absolute difference to delta to get the adjustment. |
| 145 const __m128i adj = |
| 146 _mm_min_epu8(_mm_or_si128(pdiff, ndiff), k_delta); |
| 147 // Restore the sign and get positive and negative adjustments. |
| 148 __m128i padj, nadj; |
| 149 padj = _mm_andnot_si128(diff_sign, adj); |
| 150 nadj = _mm_and_si128(diff_sign, adj); |
| 151 // Calculate filtered value. |
| 152 v_running_avg_y = _mm_subs_epu8(v_running_avg_y, padj); |
| 153 v_running_avg_y = _mm_adds_epu8(v_running_avg_y, nadj); |
| 154 _mm_storeu_si128((__m128i *)running_avg_y, v_running_avg_y); |
| 155 |
| 156 // Accumulate the adjustments. |
| 157 acc_diff = _mm_subs_epi8(acc_diff, padj); |
| 158 acc_diff = _mm_adds_epi8(acc_diff, nadj); |
| 159 |
| 160 // Update pointers for next iteration. |
| 161 sig += sig_stride; |
| 162 mc_running_avg_y += mc_avg_y_stride; |
| 163 running_avg_y += avg_y_stride; |
| 164 } |
| 165 { |
| 166 // Update the sum of all pixel differences of this MB. |
| 167 union sum_union s; |
| 168 s.v = acc_diff; |
| 169 sum_diff = s.e[0] + s.e[1] + s.e[2] + s.e[3] + s.e[4] + s.e[5] |
| 170 + s.e[6] + s.e[7] + s.e[8] + s.e[9] + s.e[10] + s.e[11] |
| 171 + s.e[12] + s.e[13] + s.e[14] + s.e[15]; |
| 172 if (abs(sum_diff) > sum_diff_thresh) { |
| 173 return COPY_BLOCK; |
| 174 } |
| 175 } |
| 176 } else { |
117 return COPY_BLOCK; | 177 return COPY_BLOCK; |
| 178 } |
118 } | 179 } |
119 } | 180 } |
120 | 181 |
121 vp8_copy_mem16x16(running_avg_y_start, avg_y_stride, sig_start, sig_stride); | 182 vp8_copy_mem16x16(running_avg_y_start, avg_y_stride, sig_start, sig_stride); |
122 return FILTER_BLOCK; | 183 return FILTER_BLOCK; |
123 } | 184 } |
OLD | NEW |