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
use png::ColorType;
use mask::Mask;
use pixel;
#[cfg(target_arch = "x86")]
use std::arch::x86::*;
#[cfg(target_arch = "x86_64")]
use std::arch::x86_64::*;
const X86: bool = cfg!(any(target_arch="x86", target_arch="x86_64"));
#[derive(Clone,Copy,Debug,Default)]
#[repr(C)]
pub struct Rgba8 {
red: u8,
green: u8,
blue: u8,
alpha: u8,
}
impl From<Rgba8> for i32 {
fn from(c: Rgba8) -> i32 {
let red = (c.red() as i32) << 0;
let green = (c.green() as i32) << 8;
let blue = (c.blue() as i32) << 16;
let alpha = (c.alpha() as i32) << 24;
red | green | blue | alpha
}
}
impl Rgba8 {
pub fn new(red: u8, green: u8, blue: u8, alpha: u8) -> Self {
Rgba8 { red, green, blue, alpha }
}
pub fn rgb(red: u8, green: u8, blue: u8) -> Self {
Rgba8::new(red, green, blue, 0xFF)
}
fn divide_alpha(self) -> Self {
let alpha = self.alpha();
let red = unscale_u8(self.red(), alpha);
let green = unscale_u8(self.green(), alpha);
let blue = unscale_u8(self.blue(), alpha);
Rgba8::new(red, green, blue, alpha)
}
pub fn red(self) -> u8 {
self.red
}
pub fn green(self) -> u8 {
self.green
}
pub fn blue(self) -> u8 {
self.blue
}
pub fn alpha(self) -> u8 {
self.alpha
}
fn over_alpha(self, bot: Rgba8, alpha: u8) -> Self {
let r = self.red() as i32 - bot.red() as i32;
let g = self.green() as i32 - bot.green() as i32;
let b = self.blue() as i32 - bot.blue() as i32;
let a = self.alpha() as i32 - bot.alpha() as i32;
let red = (bot.red() as i32 + scale_i32(r, alpha)) as u8;
let green = (bot.green() as i32 + scale_i32(g, alpha)) as u8;
let blue = (bot.blue() as i32 + scale_i32(b, alpha)) as u8;
let alpha = (bot.alpha() as i32 + scale_i32(a, alpha)) as u8;
Rgba8::new(red, green, blue, alpha)
}
}
fn scale_i32(a: i32, b: u8) -> i32 {
let c = a * b as i32;
(((c + 1) + (c >> 8)) >> 8) as i32
}
fn unscale_u8(a: u8, b: u8) -> u8 {
if b > 0 {
let aa = (a as u32) << 8;
let bb = b as u32;
(aa / bb).min(255) as u8
} else {
0
}
}
impl pixel::PixFmt for Rgba8 {
fn color_type() -> ColorType {
ColorType::RGBA
}
fn over(pix: &mut [Self], mask: &Mask, clr: Self) {
if X86 && is_x86_feature_detected!("ssse3") {
unsafe { over_x86(pix, mask, clr) }
} else {
over_fallback(pix, mask, clr);
}
}
fn divide_alpha(pix: &mut [Self]) {
for p in pix.iter_mut() {
*p = p.divide_alpha();
}
}
}
#[cfg(any(target_arch="x86", target_arch="x86_64"))]
unsafe fn over_x86(pix: &mut [Rgba8], mask: &Mask, clr: Rgba8) {
let clr = _mm_set1_epi32(clr.into());
let src = mask.pixels();
let dst = pix;
let len = src.len().min(dst.len());
let dst = dst.as_mut_ptr();
let src = src.as_ptr();
for i in (0..len).step_by(4) {
let off = i as isize;
let dst = dst.offset(off) as *mut __m128i;
let src = src.offset(off) as *const i32;
let alpha = swizzle_mask_x86(_mm_set1_epi32(*src));
let bot = _mm_loadu_si128(dst);
let out = over_alpha_u8x16_x86(clr, bot, alpha);
_mm_storeu_si128(dst, out);
}
}
#[cfg(any(target_arch="x86", target_arch="x86_64"))]
unsafe fn swizzle_mask_x86(v: __m128i) -> __m128i {
_mm_shuffle_epi8(v, _mm_set_epi8(3, 3, 3, 3,
2, 2, 2, 2,
1, 1, 1, 1,
0, 0, 0, 0))
}
#[cfg(any(target_arch="x86", target_arch="x86_64"))]
unsafe fn over_alpha_u8x16_x86(t: __m128i, b: __m128i, a: __m128i) -> __m128i {
let t_even = _mm_unpacklo_epi8(t, _mm_setzero_si128());
let b_even = _mm_unpacklo_epi8(b, _mm_setzero_si128());
let a_even = _mm_unpacklo_epi8(a, _mm_setzero_si128());
let a_even = _mm_srli_epi16(a_even, 1);
let even = _mm_mullo_epi16(a_even, _mm_sub_epi16(t_even, b_even));
let even = scale_i16_to_u8_x86(even);
let even = _mm_add_epi16(b_even, even);
let t_odd = _mm_unpackhi_epi8(t, _mm_setzero_si128());
let b_odd = _mm_unpackhi_epi8(b, _mm_setzero_si128());
let a_odd = _mm_unpackhi_epi8(a, _mm_setzero_si128());
let a_odd = _mm_srli_epi16(a_odd, 1);
let odd = _mm_mullo_epi16(a_odd, _mm_sub_epi16(t_odd, b_odd));
let odd = scale_i16_to_u8_x86(odd);
let odd = _mm_add_epi16(b_odd, odd);
_mm_packus_epi16(even, odd)
}
#[cfg(any(target_arch="x86", target_arch="x86_64"))]
unsafe fn scale_i16_to_u8_x86(v: __m128i) -> __m128i {
_mm_srai_epi16(_mm_add_epi16(_mm_add_epi16(v,
_mm_set1_epi16(1)),
_mm_srai_epi16(v, 8)),
7)
}
fn over_fallback(pix: &mut [Rgba8], mask: &Mask, clr: Rgba8) {
for (bot, m) in pix.iter_mut().zip(mask.pixels()) {
let mut out = clr.over_alpha(*bot, *m);
*bot = out;
}
}