| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * Copyright (c) 2016 Clément Bœsch <u pkh me> | ||
| 3 | * | ||
| 4 | * This file is part of FFmpeg. | ||
| 5 | * | ||
| 6 | * FFmpeg is free software; you can redistribute it and/or | ||
| 7 | * modify it under the terms of the GNU Lesser General Public | ||
| 8 | * License as published by the Free Software Foundation; either | ||
| 9 | * version 2.1 of the License, or (at your option) any later version. | ||
| 10 | * | ||
| 11 | * FFmpeg is distributed in the hope that it will be useful, | ||
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
| 14 | * Lesser General Public License for more details. | ||
| 15 | * | ||
| 16 | * You should have received a copy of the GNU Lesser General Public | ||
| 17 | * License along with FFmpeg; if not, write to the Free Software | ||
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
| 19 | */ | ||
| 20 | |||
| 21 | #ifndef AVFILTER_NLMEANS_INIT_H | ||
| 22 | #define AVFILTER_NLMEANS_INIT_H | ||
| 23 | |||
| 24 | #include <stddef.h> | ||
| 25 | #include <stdint.h> | ||
| 26 | |||
| 27 | #include "config.h" | ||
| 28 | #include "libavutil/avassert.h" | ||
| 29 | #include "libavutil/macros.h" | ||
| 30 | #include "vf_nlmeans.h" | ||
| 31 | |||
| 32 | /** | ||
| 33 | * Compute squared difference of the safe area (the zone where s1 and s2 | ||
| 34 | * overlap). It is likely the largest integral zone, so it is interesting to do | ||
| 35 | * as little checks as possible; contrary to the unsafe version of this | ||
| 36 | * function, we do not need any clipping here. | ||
| 37 | * | ||
| 38 | * The line above dst and the column to its left are always readable. | ||
| 39 | */ | ||
| 40 | 48 | static void compute_safe_ssd_integral_image_c(uint32_t *dst, ptrdiff_t dst_linesize_32, | |
| 41 | const uint8_t *s1, ptrdiff_t linesize1, | ||
| 42 | const uint8_t *s2, ptrdiff_t linesize2, | ||
| 43 | int w, int h) | ||
| 44 | { | ||
| 45 | 48 | const uint32_t *dst_top = dst - dst_linesize_32; | |
| 46 | |||
| 47 | /* SIMD-friendly assumptions allowed here */ | ||
| 48 | av_assert2(!(w & 0xf) && w >= 16 && h >= 1); | ||
| 49 | |||
| 50 |
2/2✓ Branch 0 taken 2100 times.
✓ Branch 1 taken 48 times.
|
2148 | for (int y = 0; y < h; y++) { |
| 51 |
2/2✓ Branch 0 taken 58800 times.
✓ Branch 1 taken 2100 times.
|
60900 | for (int x = 0; x < w; x += 4) { |
| 52 | 58800 | const int d0 = s1[x ] - s2[x ]; | |
| 53 | 58800 | const int d1 = s1[x + 1] - s2[x + 1]; | |
| 54 | 58800 | const int d2 = s1[x + 2] - s2[x + 2]; | |
| 55 | 58800 | const int d3 = s1[x + 3] - s2[x + 3]; | |
| 56 | |||
| 57 | 58800 | dst[x ] = dst_top[x ] - dst_top[x - 1] + d0*d0; | |
| 58 | 58800 | dst[x + 1] = dst_top[x + 1] - dst_top[x ] + d1*d1; | |
| 59 | 58800 | dst[x + 2] = dst_top[x + 2] - dst_top[x + 1] + d2*d2; | |
| 60 | 58800 | dst[x + 3] = dst_top[x + 3] - dst_top[x + 2] + d3*d3; | |
| 61 | |||
| 62 | 58800 | dst[x ] += dst[x - 1]; | |
| 63 | 58800 | dst[x + 1] += dst[x ]; | |
| 64 | 58800 | dst[x + 2] += dst[x + 1]; | |
| 65 | 58800 | dst[x + 3] += dst[x + 2]; | |
| 66 | } | ||
| 67 | 2100 | s1 += linesize1; | |
| 68 | 2100 | s2 += linesize2; | |
| 69 | 2100 | dst += dst_linesize_32; | |
| 70 | 2100 | dst_top += dst_linesize_32; | |
| 71 | } | ||
| 72 | 48 | } | |
| 73 | |||
| 74 | ✗ | static void compute_weights_line_c(const uint32_t *const iia, | |
| 75 | const uint32_t *const iib, | ||
| 76 | const uint32_t *const iid, | ||
| 77 | const uint32_t *const iie, | ||
| 78 | const uint8_t *const src, | ||
| 79 | float *total_weight, | ||
| 80 | float *sum, | ||
| 81 | const float *const weight_lut, | ||
| 82 | int max_meaningful_diff, | ||
| 83 | int startx, int endx) | ||
| 84 | { | ||
| 85 | ✗ | for (int x = startx; x < endx; x++) { | |
| 86 | /* | ||
| 87 | * M is a discrete map where every entry contains the sum of all the entries | ||
| 88 | * in the rectangle from the top-left origin of M to its coordinate. In the | ||
| 89 | * following schema, "i" contains the sum of the whole map: | ||
| 90 | * | ||
| 91 | * M = +----------+-----------------+----+ | ||
| 92 | * | | | | | ||
| 93 | * | | | | | ||
| 94 | * | a| b| c| | ||
| 95 | * +----------+-----------------+----+ | ||
| 96 | * | | | | | ||
| 97 | * | | | | | ||
| 98 | * | | X | | | ||
| 99 | * | | | | | ||
| 100 | * | d| e| f| | ||
| 101 | * +----------+-----------------+----+ | ||
| 102 | * | | | | | ||
| 103 | * | g| h| i| | ||
| 104 | * +----------+-----------------+----+ | ||
| 105 | * | ||
| 106 | * The sum of the X box can be calculated with: | ||
| 107 | * X = e-d-b+a | ||
| 108 | * | ||
| 109 | * See https://en.wikipedia.org/wiki/Summed_area_table | ||
| 110 | * | ||
| 111 | * The compute*_ssd functions compute the integral image M where every entry | ||
| 112 | * contains the sum of the squared difference of every corresponding pixels of | ||
| 113 | * two input planes of the same size as M. | ||
| 114 | */ | ||
| 115 | ✗ | const uint32_t a = iia[x]; | |
| 116 | ✗ | const uint32_t b = iib[x]; | |
| 117 | ✗ | const uint32_t d = iid[x]; | |
| 118 | ✗ | const uint32_t e = iie[x]; | |
| 119 | ✗ | const uint32_t patch_diff_sq = FFMIN(e - d - b + a, max_meaningful_diff); | |
| 120 | ✗ | const float weight = weight_lut[patch_diff_sq]; // exp(-patch_diff_sq * s->pdiff_scale) | |
| 121 | |||
| 122 | ✗ | total_weight[x] += weight; | |
| 123 | ✗ | sum[x] += weight * src[x]; | |
| 124 | } | ||
| 125 | ✗ | } | |
| 126 | |||
| 127 | 13 | static av_unused void ff_nlmeans_init(NLMeansDSPContext *dsp) | |
| 128 | { | ||
| 129 | 13 | dsp->compute_safe_ssd_integral_image = compute_safe_ssd_integral_image_c; | |
| 130 | 13 | dsp->compute_weights_line = compute_weights_line_c; | |
| 131 | |||
| 132 | #if ARCH_AARCH64 | ||
| 133 | ff_nlmeans_init_aarch64(dsp); | ||
| 134 | #elif ARCH_X86 | ||
| 135 | 13 | ff_nlmeans_init_x86(dsp); | |
| 136 | #endif | ||
| 137 | 13 | } | |
| 138 | |||
| 139 | #endif /* AVFILTER_NLMEANS_INIT_H */ | ||
| 140 |