jxl 0.1.0

High performance Rust implementation of a JPEG XL decoder
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

// Copyright (c) the JPEG XL Project Authors. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

use std::{
    arch::x86_64::{
        __m256, _mm256_add_ps, _mm256_andnot_si256, _mm256_castps_si256, _mm256_castsi256_ps,
        _mm256_div_ps, _mm256_fmadd_ps, _mm256_loadu_ps, _mm256_max_ps, _mm256_mul_ps,
        _mm256_set1_epi32, _mm256_set1_ps, _mm256_storeu_ps, _mm256_sub_ps,
    },
    ops::{Add, AddAssign, Div, DivAssign, Mul, MulAssign, Sub, SubAssign},
};

use super::super::{F32SimdVec, SimdDescriptor};

// Safety invariant: this type is only ever constructed if avx2 and fma are available.
#[derive(Clone, Copy, Debug)]
pub struct AvxDescriptor;

impl SimdDescriptor for AvxDescriptor {
    type F32Vec = F32VecAvx;
    fn new() -> Option<Self> {
        if is_x86_feature_detected!("avx2") && is_x86_feature_detected!("fma") {
            // SAFETY: we just checked avx2 and fma.
            Some(Self)
        } else {
            None
        }
    }
}

// TODO(veluca): retire this macro once we have #[unsafe(target_feature)].
macro_rules! fn_avx {
    (
        $this:ident: $self_ty:ty,
        fn $name:ident($($arg:ident: $ty:ty),* $(,)?) $(-> $ret:ty )? $body: block) => {
        #[inline(always)]
        fn $name(self: $self_ty, $($arg: $ty),*) $(-> $ret)? {
            #[target_feature(enable = "fma,avx2")]
            #[inline]
            fn inner($this: $self_ty, $($arg: $ty),*) $(-> $ret)? {
                $body
            }
            // SAFETY: `self.1` is constructed iff avx2 and fma are available.
            unsafe { inner(self, $($arg),*) }
        }
    };
}

#[derive(Clone, Copy, Debug)]
#[repr(transparent)]
pub struct F32VecAvx(__m256, AvxDescriptor);

impl F32SimdVec for F32VecAvx {
    type Descriptor = AvxDescriptor;

    const LEN: usize = 8;

    #[inline(always)]
    fn load(d: Self::Descriptor, mem: &[f32]) -> Self {
        assert!(mem.len() >= Self::LEN);
        // SAFETY: we just checked that `mem` has enough space. Moreover, we know avx is available
        // from the safety invariant on `d`.
        Self(unsafe { _mm256_loadu_ps(mem.as_ptr()) }, d)
    }

    #[inline(always)]
    fn store(&self, mem: &mut [f32]) {
        assert!(mem.len() >= Self::LEN);
        // SAFETY: we just checked that `mem` has enough space. Moreover, we know avx is available
        // from the safety invariant on `self.1`.
        unsafe { _mm256_storeu_ps(mem.as_mut_ptr(), self.0) }
    }

    fn_avx!(this: F32VecAvx, fn mul_add(mul: F32VecAvx, add: F32VecAvx) -> F32VecAvx {
        F32VecAvx(_mm256_fmadd_ps(this.0, mul.0, add.0), this.1)
    });

    fn splat(d: Self::Descriptor, v: f32) -> Self {
        // SAFETY: We know avx is available from the safety invariant on `d`.
        unsafe { Self(_mm256_set1_ps(v), d) }
    }

    fn_avx!(this: F32VecAvx, fn abs() -> F32VecAvx {
        F32VecAvx(
            _mm256_castsi256_ps(_mm256_andnot_si256(
                _mm256_set1_epi32(0b10000000000000000000000000000000u32 as i32),
                _mm256_castps_si256(this.0),
            )),
            this.1)
    });

    fn_avx!(this: F32VecAvx, fn max(other: F32VecAvx) -> F32VecAvx {
        F32VecAvx(_mm256_max_ps(this.0, other.0), this.1)
    });
}

impl Add<F32VecAvx> for F32VecAvx {
    type Output = F32VecAvx;
    fn_avx!(this: F32VecAvx, fn add(rhs: F32VecAvx) -> F32VecAvx {
        F32VecAvx(_mm256_add_ps(this.0, rhs.0), this.1)
    });
}

impl Sub<F32VecAvx> for F32VecAvx {
    type Output = F32VecAvx;
    fn_avx!(this: F32VecAvx, fn sub(rhs: F32VecAvx) -> F32VecAvx {
        F32VecAvx(_mm256_sub_ps(this.0, rhs.0), this.1)
    });
}

impl Mul<F32VecAvx> for F32VecAvx {
    type Output = F32VecAvx;
    fn_avx!(this: F32VecAvx, fn mul(rhs: F32VecAvx) -> F32VecAvx {
        F32VecAvx(_mm256_mul_ps(this.0, rhs.0), this.1)
    });
}

impl Div<F32VecAvx> for F32VecAvx {
    type Output = F32VecAvx;
    fn_avx!(this: F32VecAvx, fn div(rhs: F32VecAvx) -> F32VecAvx {
        F32VecAvx(_mm256_div_ps(this.0, rhs.0), this.1)
    });
}

impl AddAssign<F32VecAvx> for F32VecAvx {
    fn_avx!(this: &mut F32VecAvx, fn add_assign(rhs: F32VecAvx) {
        this.0 = _mm256_add_ps(this.0, rhs.0)
    });
}

impl SubAssign<F32VecAvx> for F32VecAvx {
    fn_avx!(this: &mut F32VecAvx, fn sub_assign(rhs: F32VecAvx) {
        this.0 = _mm256_sub_ps(this.0, rhs.0)
    });
}

impl MulAssign<F32VecAvx> for F32VecAvx {
    fn_avx!(this: &mut F32VecAvx, fn mul_assign(rhs: F32VecAvx) {
        this.0 = _mm256_mul_ps(this.0, rhs.0)
    });
}

impl DivAssign<F32VecAvx> for F32VecAvx {
    fn_avx!(this: &mut F32VecAvx, fn div_assign(rhs: F32VecAvx) {
        this.0 = _mm256_div_ps(this.0, rhs.0)
    });
}