burn-cubecl 0.21.0-pre.3

Generic backend that can be compiled just-in-time to any shader language target
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

use cubecl::{calculate_cube_count_elemwise, prelude::*};
use cubecl::{
    num_traits::Zero,
    std::{
        FastDivmod,
        tensor::layout::{linear::LinearLayout, *},
    },
};

use crate::{
    CubeRuntime,
    kernel::utils::{address_type, linear_layout, shape_divmod},
    ops::max_vector_size,
    tensor::CubeTensor,
};

#[cube(launch, address_type = "dynamic")]
fn interpolate_lanczos3_kernel<F: Float, N: Size>(
    input: &Tensor<Vector<F, N>>,
    output: &mut Tensor<Vector<F, N>>,
    shape_out: Sequence<FastDivmod<usize>>,
    out_layout: LinearLayout,
    #[comptime] align_corners: bool,
    #[define(F)] _dtype: StorageType,
) {
    if ABSOLUTE_POS >= output.len() {
        terminate!();
    }

    let vector_size = input.vector_size();
    let out_idx = out_layout.to_source_pos(ABSOLUTE_POS);

    let (rem, c) = shape_out[3].div_mod(ABSOLUTE_POS * vector_size);
    let (rem, x) = shape_out[2].div_mod(rem);
    let (b, y) = shape_out[1].div_mod(rem);

    let input_height = input.shape(1) - 1;
    let input_height_f = input_height as f32;

    let y_frac = if align_corners {
        let output_height = clamp_min(output.shape(1) - 1, 1) as f32;
        (y * input_height) as f32 / output_height
    } else {
        let in_size = (input_height + 1) as f32;
        let out_size = output.shape(1) as f32;
        (y as f32 + 0.5) * (in_size / out_size) - 0.5
    };
    let y0 = f32::floor(y_frac);

    let input_width = input.shape(2) - 1;
    let input_width_f = input_width as f32;

    let x_frac = if align_corners {
        let output_width = clamp_min(output.shape(2) - 1, 1) as f32;
        (x * input_width) as f32 / output_width
    } else {
        let in_size = (input_width + 1) as f32;
        let out_size = output.shape(2) as f32;
        (x as f32 + 0.5) * (in_size / out_size) - 0.5
    };
    let x0 = f32::floor(x_frac);

    let index_base = b * input.stride(0) + c * input.stride(3);
    let in_stride_y = input.stride(1);
    let in_stride_x = input.stride(2);

    let mut result = Vector::zero();
    let mut weight_sum = 0.0f32;

    // 6-tap separable Lanczos3 filter: ky in -2..=3, kx in -2..=3
    // Skip out-of-bounds positions instead of clamping (matches TF/JAX/PIL)
    #[unroll]
    for ky in -2..4i32 {
        let y_pos = y0 + ky as f32;
        if y_pos >= 0.0 && y_pos <= input_height_f {
            let y_idx = y_pos as usize;
            let wy = lanczos3_weight(y_frac - y_pos);

            #[unroll]
            for kx in -2..4i32 {
                let x_pos = x0 + kx as f32;
                if x_pos >= 0.0 && x_pos <= input_width_f {
                    let x_idx = x_pos as usize;
                    let wx = lanczos3_weight(x_frac - x_pos);

                    let wt = wy * wx;
                    let idx = index_base + y_idx * in_stride_y + x_idx * in_stride_x;
                    let pixel = input[idx / vector_size];
                    let w = Vector::new(F::cast_from(wt));
                    result += pixel * w;
                    weight_sum += wt;
                }
            }
        }
    }

    if weight_sum != 0.0 {
        let inv_w = Vector::new(F::cast_from(1.0 / weight_sum));
        result *= inv_w;
    }

    output[out_idx] = result;
}

#[cube]
fn lanczos3_weight(x: f32) -> f32 {
    let abs_x = f32::abs(x);
    let mut result = 0.0f32;
    if abs_x < 1e-7 {
        result = 1.0;
    } else if abs_x < 3.0 {
        let pi = core::f32::consts::PI;
        let pi_x = pi * x;
        let pi_x_over_3 = pi_x / 3.0;
        result = (f32::sin(pi_x) * f32::sin(pi_x_over_3)) / (pi_x * pi_x_over_3);
    }
    result
}

pub(crate) fn interpolate_lanczos3_launch<R: CubeRuntime>(
    input: CubeTensor<R>,
    output: CubeTensor<R>,
    align_corners: bool,
) -> CubeTensor<R> {
    let vector_size = max_vector_size(&input);
    let out_shape = shape_divmod(&output);
    let out_layout = linear_layout(&output, vector_size);

    let working_units = output.meta.num_elements() / vector_size as usize;
    let cube_dim = CubeDim::new(&input.client, working_units);
    let cube_count = calculate_cube_count_elemwise(&input.client, working_units, cube_dim);

    interpolate_lanczos3_kernel::launch(
        &output.client,
        cube_count,
        cube_dim,
        address_type!(input, output),
        vector_size,
        input.into_tensor_arg(),
        output.clone().into_tensor_arg(),
        out_shape,
        out_layout,
        align_corners,
        output.dtype.into(),
    );

    output
}