footile 0.7.0

A 2D vector graphics library
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

// imgbuf.rs        Functions for blending image buffers.
//
// Copyright (c) 2017-2020  Douglas P Lau
//
use pix::chan::{Ch8, Linear, Premultiplied};
use pix::el::Pixel;
use pix::matte::Matte8;
use std::any::TypeId;
use std::slice::from_raw_parts_mut;

#[cfg(all(target_arch = "x86", feature = "simd"))]
use std::arch::x86::*;
#[cfg(all(target_arch = "x86_64", feature = "simd"))]
use std::arch::x86_64::*;

/// Blend to a Matte8 using a signed area with non-zero fill rule.
/// Source buffer is zeroed upon return.
///
/// * `dst` Destination buffer.
/// * `sgn_area` Signed area.
#[inline]
pub fn matte_src_over_non_zero<P>(dst: &mut [P], sgn_area: &mut [i16])
where
    P: Pixel<Chan = Ch8, Alpha = Premultiplied, Gamma = Linear>,
{
    debug_assert_eq!(TypeId::of::<P>(), TypeId::of::<Matte8>());
    let n_bytes = dst.len() * std::mem::size_of::<Matte8>();
    let ptr = dst.as_mut_ptr() as *mut u8;
    let dst = unsafe { from_raw_parts_mut(ptr, n_bytes) };
    accumulate_non_zero(dst, sgn_area);
}

/// Accumulate signed area with non-zero fill rule.
/// Source buffer is zeroed upon return.
///
/// * `dst` Destination buffer.
/// * `src` Source buffer.
fn accumulate_non_zero(dst: &mut [u8], src: &mut [i16]) {
    assert!(dst.len() <= src.len());
    #[cfg(all(
        any(target_arch = "x86", target_arch = "x86_64"),
        feature = "simd"
    ))]
    {
        if is_x86_feature_detected!("ssse3") {
            unsafe { accumulate_non_zero_x86(dst, src) }
            return;
        }
    }
    accumulate_non_zero_fallback(dst, src)
}

/// Accumulate signed area with non-zero fill rule.
fn accumulate_non_zero_fallback(dst: &mut [u8], src: &mut [i16]) {
    let mut sum = 0;
    for (d, s) in dst.iter_mut().zip(src.iter_mut()) {
        sum += *s;
        *s = 0;
        *d = saturating_cast_i16_u8(sum);
    }
}

/// Cast an i16 to a u8 with saturation
fn saturating_cast_i16_u8(v: i16) -> u8 {
    v.max(0).min(255) as u8
}

/// Accumulate signed area with non-zero fill rule.
#[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), feature = "simd"))]
#[target_feature(enable = "ssse3")]
unsafe fn accumulate_non_zero_x86(dst: &mut [u8], src: &mut [i16]) {
    let zero = _mm_setzero_si128();
    let mut sum = zero;
    let len = dst.len().min(src.len());
    let dst = dst.as_mut_ptr();
    let src = src.as_mut_ptr();
    for i in (0..len).step_by(8) {
        let off = i as isize;
        let d = dst.offset(off) as *mut __m128i;
        let s = src.offset(off) as *mut __m128i;
        // get 8 values from src
        let mut a = _mm_loadu_si128(s);
        // zeroing now is faster than memset later
        _mm_storeu_si128(s, zero);
        // accumulate sum thru 8 pixels
        a = accumulate_i16x8_x86(a);
        // add in previous sum
        a = _mm_add_epi16(a, sum);
        // pack to u8 using saturation
        let b = _mm_packus_epi16(a, a);
        // store result to dest
        _mm_storel_epi64(d, b);
        // shuffle sum into all 16-bit lanes
        sum = _mm_shuffle_epi8(a, _mm_set1_epi16(0x0F_0E));
    }
}

/// Accumulate signed area sum thru 8 pixels.
#[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), feature = "simd"))]
#[target_feature(enable = "ssse3")]
unsafe fn accumulate_i16x8_x86(mut a: __m128i) -> __m128i {
    //   a7 a6 a5 a4 a3 a2 a1 a0
    // + a3 a2 a1 a0 __ __ __ __
    a = _mm_add_epi16(a, _mm_slli_si128(a, 8));
    // + a5 a4 a3 a2 a1 a0 __ __
    // + a1 a0 __ __ __ __ __ __
    a = _mm_add_epi16(a, _mm_slli_si128(a, 4));
    // + a6 a5 a4 a3 a2 a1 a0 __
    // + a2 a1 a0 __ __ __ __ __
    // + a4 a3 a2 a1 a0 __ __ __
    // + a0 __ __ __ __ __ __ __
    _mm_add_epi16(a, _mm_slli_si128(a, 2))
}

/// Blend to a Matte8 using a signed area with even-odd fill rule.
/// Source buffer is zeroed upon return.
///
/// * `dst` Destination buffer.
/// * `sgn_area` Signed area.
#[inline]
pub fn matte_src_over_even_odd<P>(dst: &mut [P], sgn_area: &mut [i16])
where
    P: Pixel<Chan = Ch8, Alpha = Premultiplied, Gamma = Linear>,
{
    debug_assert_eq!(TypeId::of::<P>(), TypeId::of::<Matte8>());
    let n_bytes = dst.len() * std::mem::size_of::<P>();
    let ptr = dst.as_mut_ptr() as *mut u8;
    let dst = unsafe { std::slice::from_raw_parts_mut(ptr, n_bytes) };
    accumulate_even_odd(dst, sgn_area);
}

/// Accumulate signed area with even-odd fill rule.
/// Source buffer is zeroed upon return.
///
/// * `dst` Destination buffer.
/// * `src` Source buffer.
fn accumulate_even_odd(dst: &mut [u8], src: &mut [i16]) {
    assert!(dst.len() <= src.len());
    #[cfg(all(
        any(target_arch = "x86", target_arch = "x86_64"),
        feature = "simd"
    ))]
    {
        if is_x86_feature_detected!("ssse3") {
            unsafe { accumulate_even_odd_x86(dst, src) }
            return;
        }
    }
    accumulate_even_odd_fallback(dst, src)
}

/// Accumulate signed area with even-odd fill rule.
fn accumulate_even_odd_fallback(dst: &mut [u8], src: &mut [i16]) {
    let mut sum = 0;
    for (d, s) in dst.iter_mut().zip(src.iter_mut()) {
        sum += *s;
        *s = 0;
        let v = sum & 0xFF;
        let odd = sum & 0x100;
        let c = (v - odd).abs();
        *d = saturating_cast_i16_u8(c);
    }
}

/// Accumulate signed area with even-odd fill rule.
#[cfg(all(any(target_arch = "x86", target_arch = "x86_64"), feature = "simd"))]
#[target_feature(enable = "ssse3")]
unsafe fn accumulate_even_odd_x86(dst: &mut [u8], src: &mut [i16]) {
    let zero = _mm_setzero_si128();
    let mut sum = zero;
    for (d, s) in dst.chunks_mut(8).zip(src.chunks_mut(8)) {
        let d = d.as_mut_ptr() as *mut __m128i;
        let s = s.as_mut_ptr() as *mut __m128i;
        // get 8 values from src
        let mut a = _mm_loadu_si128(s);
        // zeroing now is faster than memset later
        _mm_storeu_si128(s, zero);
        // accumulate sum thru 8 pixels
        a = accumulate_i16x8_x86(a);
        // add in previous sum
        a = _mm_add_epi16(a, sum);
        let mut val = _mm_and_si128(a, _mm_set1_epi16(0xFF));
        let odd = _mm_and_si128(a, _mm_set1_epi16(0x100));
        val = _mm_sub_epi16(val, odd);
        val = _mm_abs_epi16(val);
        // pack to u8 using saturation
        let b = _mm_packus_epi16(val, val);
        // store result to dest
        _mm_storel_epi64(d, b);
        // shuffle sum into all 16-bit lanes
        sum = _mm_shuffle_epi8(a, _mm_set1_epi16(0x0F_0E));
    }
}

#[cfg(test)]
mod test {
    use super::*;

    #[test]
    fn non_zero() {
        let mut a = [0; 3000];
        let mut b = [0; 3000];
        b[0] = 200;
        accumulate_non_zero(&mut a, &mut b);
        for ai in a.iter() {
            assert_eq!(*ai, 200);
        }
        let mut c = [0; 5000];
        let mut d = [0; 5000];
        d[0] = 300;
        accumulate_non_zero(&mut c, &mut d);
        for ci in c.iter() {
            assert_eq!(*ci, 255);
        }
    }

    #[test]
    fn even_odd() {
        let mut a = [0; 3000];
        let mut b = [0; 3000];
        b[0] = 300;
        accumulate_even_odd(&mut a, &mut b);
        for ai in a.iter() {
            assert_eq!(*ai, 212);
        }
    }
}