zoomvtools 1.1.0

Video motion vector analysis utilities in pure Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246

#![allow(clippy::undocumented_unsafe_blocks)]

use std::{arch::x86_64::*, num::NonZeroUsize};

use crate::util::Pixel;

#[target_feature(enable = "avx2")]
pub(super) fn reduce_average<T: Pixel>(
    dest: &mut [T],
    src: &[T],
    dest_pitch: NonZeroUsize,
    src_pitch: NonZeroUsize,
    dest_width: NonZeroUsize,
    dest_height: NonZeroUsize,
) {
    // Check the array bounds once at the start of the loop.
    debug_assert!(src.len() >= src_pitch.get() * dest_height.get() * 2);
    debug_assert!(dest.len() >= dest_pitch.get() * dest_height.get());

    match size_of::<T>() {
        1 => unsafe {
            reduce_average_u8(
                dest.as_mut_ptr() as *mut u8,
                src.as_ptr() as *const u8,
                dest_pitch,
                src_pitch,
                dest_width,
                dest_height,
            );
        },
        2 => unsafe {
            reduce_average_u16(
                dest.as_mut_ptr() as *mut u16,
                src.as_ptr() as *const u16,
                dest_pitch,
                src_pitch,
                dest_width,
                dest_height,
            );
        },
        _ => unreachable!(),
    }
}

#[target_feature(enable = "avx2")]
// NOTE: Custom implementation. No C implementation exists.
unsafe fn reduce_average_u8(
    dest: *mut u8,
    src: *const u8,
    dest_pitch: NonZeroUsize,
    src_pitch: NonZeroUsize,
    dest_width: NonZeroUsize,
    dest_height: NonZeroUsize,
) {
    let dest_width = dest_width.get();
    let dest_height = dest_height.get();
    let dest_pitch = dest_pitch.get();
    let src_pitch = src_pitch.get();

    // Process 32 destination pixels at a time (requires 64 source pixels per row).
    const SIMD_WIDTH: usize = 32;
    const FALLBACK_SIMD_WIDTH: usize = 16;
    let ones = _mm256_set1_epi8(1);
    let rounding = _mm256_set1_epi16(2);

    for y in 0..dest_height {
        let dest_row = dest.add(y * dest_pitch);
        let src_row1 = src.add(y * 2 * src_pitch);
        let src_row2 = src.add((y * 2 + 1) * src_pitch);

        let mut x = 0;

        // Process 32-pixel SIMD chunks.
        while x + SIMD_WIDTH <= dest_width {
            let src1_lo = _mm256_loadu_si256(src_row1.add(x * 2) as *const __m256i);
            let src1_hi = _mm256_loadu_si256(src_row1.add(x * 2 + 32) as *const __m256i);
            let src2_lo = _mm256_loadu_si256(src_row2.add(x * 2) as *const __m256i);
            let src2_hi = _mm256_loadu_si256(src_row2.add(x * 2 + 32) as *const __m256i);

            let pairs_lo = _mm256_add_epi16(
                _mm256_maddubs_epi16(src1_lo, ones),
                _mm256_maddubs_epi16(src2_lo, ones),
            );
            let pairs_hi = _mm256_add_epi16(
                _mm256_maddubs_epi16(src1_hi, ones),
                _mm256_maddubs_epi16(src2_hi, ones),
            );

            let result_lo = _mm256_srli_epi16(_mm256_add_epi16(pairs_lo, rounding), 2);
            let result_hi = _mm256_srli_epi16(_mm256_add_epi16(pairs_hi, rounding), 2);
            let packed = _mm256_packus_epi16(result_lo, result_hi);
            let final_result = _mm256_permute4x64_epi64(packed, 0b11_01_10_00);

            _mm256_storeu_si256(dest_row.add(x) as *mut __m256i, final_result);

            x += SIMD_WIDTH;
        }

        if x + FALLBACK_SIMD_WIDTH <= dest_width {
            let src1 = _mm256_loadu_si256(src_row1.add(x * 2) as *const __m256i);
            let src2 = _mm256_loadu_si256(src_row2.add(x * 2) as *const __m256i);

            let pairs = _mm256_add_epi16(
                _mm256_maddubs_epi16(src1, ones),
                _mm256_maddubs_epi16(src2, ones),
            );
            let result = _mm256_srli_epi16(_mm256_add_epi16(pairs, rounding), 2);
            let packed = _mm256_packus_epi16(result, result);
            let final_result = _mm256_permute4x64_epi64(packed, 0b11_01_10_00);

            _mm_storeu_si128(
                dest_row.add(x) as *mut __m128i,
                _mm256_castsi256_si128(final_result),
            );

            x += FALLBACK_SIMD_WIDTH;
        }

        // Handle remaining pixels with scalar code
        while x < dest_width {
            let src_x = x * 2;
            let a = *src_row1.add(src_x) as u16;
            let b = *src_row1.add(src_x + 1) as u16;
            let c = *src_row2.add(src_x) as u16;
            let d = *src_row2.add(src_x + 1) as u16;

            let avg = ((a + b + c + d + 2) / 4) as u8;
            *dest_row.add(x) = avg;
            x += 1;
        }
    }
}

#[target_feature(enable = "avx2")]
// NOTE: Custom implementation. No C implementation exists.
unsafe fn reduce_average_u16(
    dest: *mut u16,
    src: *const u16,
    dest_pitch: NonZeroUsize,
    src_pitch: NonZeroUsize,
    dest_width: NonZeroUsize,
    dest_height: NonZeroUsize,
) {
    let dest_width = dest_width.get();
    let dest_height = dest_height.get();
    let dest_pitch = dest_pitch.get();
    let src_pitch = src_pitch.get();

    // Process 16 destination pixels at a time (requires 32 source pixels per row).
    const SIMD_WIDTH: usize = 16;
    const FALLBACK_SIMD_WIDTH: usize = 8;
    let signed_bias = _mm256_set1_epi16(i16::MIN);
    let ones = _mm256_set1_epi16(1);
    let bias_correction = _mm256_set1_epi32(131_074);

    for y in 0..dest_height {
        let dest_row = dest.add(y * dest_pitch);
        let src_row1 = src.add(y * 2 * src_pitch);
        let src_row2 = src.add((y * 2 + 1) * src_pitch);

        let mut x = 0;

        // Process 16-pixel SIMD chunks.
        while x + SIMD_WIDTH <= dest_width {
            let src1_lo = _mm256_xor_si256(
                _mm256_loadu_si256(src_row1.add(x * 2) as *const __m256i),
                signed_bias,
            );
            let src1_hi = _mm256_xor_si256(
                _mm256_loadu_si256(src_row1.add(x * 2 + 16) as *const __m256i),
                signed_bias,
            );
            let src2_lo = _mm256_xor_si256(
                _mm256_loadu_si256(src_row2.add(x * 2) as *const __m256i),
                signed_bias,
            );
            let src2_hi = _mm256_xor_si256(
                _mm256_loadu_si256(src_row2.add(x * 2 + 16) as *const __m256i),
                signed_bias,
            );

            let sum_lo = _mm256_add_epi32(
                _mm256_add_epi32(
                    _mm256_madd_epi16(src1_lo, ones),
                    _mm256_madd_epi16(src2_lo, ones),
                ),
                bias_correction,
            );
            let sum_hi = _mm256_add_epi32(
                _mm256_add_epi32(
                    _mm256_madd_epi16(src1_hi, ones),
                    _mm256_madd_epi16(src2_hi, ones),
                ),
                bias_correction,
            );

            let result_lo = _mm256_srli_epi32(sum_lo, 2);
            let result_hi = _mm256_srli_epi32(sum_hi, 2);
            let packed = _mm256_packus_epi32(result_lo, result_hi);
            let final_result = _mm256_permute4x64_epi64(packed, 0b11_01_10_00);

            _mm256_storeu_si256(dest_row.add(x) as *mut __m256i, final_result);

            x += SIMD_WIDTH;
        }

        if x + FALLBACK_SIMD_WIDTH <= dest_width {
            let src1 = _mm256_xor_si256(
                _mm256_loadu_si256(src_row1.add(x * 2) as *const __m256i),
                signed_bias,
            );
            let src2 = _mm256_xor_si256(
                _mm256_loadu_si256(src_row2.add(x * 2) as *const __m256i),
                signed_bias,
            );

            let sum = _mm256_add_epi32(
                _mm256_add_epi32(_mm256_madd_epi16(src1, ones), _mm256_madd_epi16(src2, ones)),
                bias_correction,
            );
            let result = _mm256_srli_epi32(sum, 2);
            let packed = _mm256_packus_epi32(result, result);
            let final_result = _mm256_permute4x64_epi64(packed, 0b11_01_10_00);

            _mm_storeu_si128(
                dest_row.add(x) as *mut __m128i,
                _mm256_castsi256_si128(final_result),
            );

            x += FALLBACK_SIMD_WIDTH;
        }

        // Handle remaining pixels with scalar code
        while x < dest_width {
            let src_x = x * 2;
            let a = *src_row1.add(src_x) as u32;
            let b = *src_row1.add(src_x + 1) as u32;
            let c = *src_row2.add(src_x) as u32;
            let d = *src_row2.add(src_x + 1) as u32;

            let avg = ((a + b + c + d + 2) / 4) as u16;
            *dest_row.add(x) = avg;
            x += 1;
        }
    }
}