zoomvtools 2.0.0

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

#![allow(clippy::undocumented_unsafe_blocks)]
#![allow(unsafe_op_in_unsafe_fn)]

#[cfg(any(test, feature = "experimental"))]
use std::{arch::x86_64::*, mem::size_of, num::NonZeroUsize};

#[cfg(any(test, feature = "experimental"))]
use cpudetect::target_family;
#[cfg(any(test, feature = "experimental"))]
use semisafe::result::unwrap as semisafe_res_unwrap;

#[target_family("x86_64_v3")]
#[cfg(any(test, feature = "experimental"))]
// NOTE: Ported from C implementation
pub(super) unsafe fn limit_changes_u8(
    dest: *mut u8,
    dest_stride_bytes: NonZeroUsize,
    src: *const u8,
    src_stride_bytes: NonZeroUsize,
    width: NonZeroUsize,
    height: NonZeroUsize,
    limit: u16,
) {
    let width = width.get();
    let height = height.get();
    let dest_stride = dest_stride_bytes.get();
    let src_stride = src_stride_bytes.get();
    let limit = semisafe_res_unwrap(u8::try_from(limit));

    let limit256 = _mm256_set1_epi8(i8::from_ne_bytes([limit]));
    let limit128 = _mm_set1_epi8(i8::from_ne_bytes([limit]));

    for y in 0..height {
        let src_row = src.add(y * src_stride);
        let dest_row = dest.add(y * dest_stride);
        let mut x = 0;

        while x + 32 <= width {
            let src256 = _mm256_loadu_si256(src_row.add(x).cast::<__m256i>());
            let dest256 = _mm256_loadu_si256(dest_row.add(x).cast::<__m256i>());

            let lower = _mm256_subs_epu8(src256, limit256);
            let upper = _mm256_adds_epu8(src256, limit256);
            let clamped = _mm256_min_epu8(_mm256_max_epu8(dest256, lower), upper);

            _mm256_storeu_si256(dest_row.add(x).cast::<__m256i>(), clamped);
            x += 32;
        }

        if x + 16 <= width {
            let src128 = _mm_loadu_si128(src_row.add(x).cast::<__m128i>());
            let dest128 = _mm_loadu_si128(dest_row.add(x).cast::<__m128i>());

            let lower = _mm_subs_epu8(src128, limit128);
            let upper = _mm_adds_epu8(src128, limit128);
            let clamped = _mm_min_epu8(_mm_max_epu8(dest128, lower), upper);

            _mm_storeu_si128(dest_row.add(x).cast::<__m128i>(), clamped);
            x += 16;
        }

        while x < width {
            let src = *src_row.add(x);
            let dest = *dest_row.add(x);
            let lower = src.saturating_sub(limit);
            let upper = src.saturating_add(limit);
            *dest_row.add(x) = lower.max(dest).min(upper);
            x += 1;
        }
    }
}

#[target_family("x86_64_v3")]
#[cfg(any(test, feature = "experimental"))]
// NOTE: Custom implementation. No C implementation exists.
pub(super) unsafe fn limit_changes_u16(
    dest: *mut u8,
    dest_stride_bytes: NonZeroUsize,
    src: *const u8,
    src_stride_bytes: NonZeroUsize,
    width: NonZeroUsize,
    height: NonZeroUsize,
    limit: u16,
) {
    let width = width.get();
    let height = height.get();
    let dest_stride = dest_stride_bytes.get() / size_of::<u16>();
    let src_stride = src_stride_bytes.get() / size_of::<u16>();
    let src = src.cast::<u16>();
    let dest = dest.cast::<u16>();

    let limit256 = _mm256_set1_epi16(i16::from_ne_bytes(limit.to_ne_bytes()));
    let limit128 = _mm_set1_epi16(i16::from_ne_bytes(limit.to_ne_bytes()));

    for y in 0..height {
        let src_row = src.add(y * src_stride);
        let dest_row = dest.add(y * dest_stride);
        let mut x = 0;

        while x + 16 <= width {
            let src256 = _mm256_loadu_si256(src_row.add(x).cast::<__m256i>());
            let dest256 = _mm256_loadu_si256(dest_row.add(x).cast::<__m256i>());

            let lower = _mm256_subs_epu16(src256, limit256);
            let upper = _mm256_adds_epu16(src256, limit256);
            let clamped = _mm256_min_epu16(_mm256_max_epu16(dest256, lower), upper);

            _mm256_storeu_si256(dest_row.add(x).cast::<__m256i>(), clamped);
            x += 16;
        }

        if x + 8 <= width {
            let src128 = _mm_loadu_si128(src_row.add(x).cast::<__m128i>());
            let dest128 = _mm_loadu_si128(dest_row.add(x).cast::<__m128i>());

            let lower = _mm_subs_epu16(src128, limit128);
            let upper = _mm_adds_epu16(src128, limit128);
            let clamped = _mm_min_epu16(_mm_max_epu16(dest128, lower), upper);

            _mm_storeu_si128(dest_row.add(x).cast::<__m128i>(), clamped);
            x += 8;
        }

        while x < width {
            let src = *src_row.add(x);
            let dest = *dest_row.add(x);
            let lower = src.saturating_sub(limit);
            let upper = src.saturating_add(limit);
            *dest_row.add(x) = lower.max(dest).min(upper);
            x += 1;
        }
    }
}