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

use half::f16;
use ndarray::ArrayViewD;

use re_types::tensor_data::TensorDataType;

/// Stats about a tensor or image.
#[derive(Clone, Copy, Debug)]
pub struct TensorStats {
    /// The range of values, ignoring `NaN`s.
    ///
    /// `None` for empty tensors.
    pub range: Option<(f64, f64)>,

    /// Like `range`, but ignoring all `NaN`/inf values.
    ///
    /// If no finite values are present, this takes the maximum finite range
    /// of the underlying data type.
    pub finite_range: (f64, f64),
}

impl TensorStats {
    pub fn from_tensor(tensor: &re_types::datatypes::TensorData) -> Self {
        re_tracing::profile_function!();

        macro_rules! declare_tensor_range_int {
            ($name:ident, $typ:ty) => {
                fn $name(tensor: ndarray::ArrayViewD<'_, $typ>) -> (f64, f64) {
                    re_tracing::profile_function!();
                    let (min, max) = tensor
                        .fold((<$typ>::MAX, <$typ>::MIN), |(min, max), &value| {
                            (min.min(value), max.max(value))
                        });
                    (min as f64, max as f64)
                }
            };
        }

        macro_rules! declare_tensor_range_float {
            ($name:ident, $typ:ty) => {
                fn $name(tensor: ndarray::ArrayViewD<'_, $typ>) -> (f64, f64) {
                    re_tracing::profile_function!();
                    let (min, max) = tensor.fold(
                        (<$typ>::INFINITY, <$typ>::NEG_INFINITY),
                        |(min, max), &value| (min.min(value), max.max(value)),
                    );
                    #[allow(trivial_numeric_casts)]
                    (min as f64, max as f64)
                }
            };
        }

        declare_tensor_range_int!(tensor_range_u8, u8);
        declare_tensor_range_int!(tensor_range_u16, u16);
        declare_tensor_range_int!(tensor_range_u32, u32);
        declare_tensor_range_int!(tensor_range_u64, u64);

        declare_tensor_range_int!(tensor_range_i8, i8);
        declare_tensor_range_int!(tensor_range_i16, i16);
        declare_tensor_range_int!(tensor_range_i32, i32);
        declare_tensor_range_int!(tensor_range_i64, i64);

        // declare_tensor_range_float!(tensor_range_f16, half::f16);
        declare_tensor_range_float!(tensor_range_f32, f32);
        declare_tensor_range_float!(tensor_range_f64, f64);

        #[allow(clippy::needless_pass_by_value)]
        fn tensor_range_f16(tensor: ndarray::ArrayViewD<'_, f16>) -> (f64, f64) {
            re_tracing::profile_function!();
            let (min, max) = tensor
                .fold((f16::INFINITY, f16::NEG_INFINITY), |(min, max), &value| {
                    (min.min(value), max.max(value))
                });
            (min.to_f64(), max.to_f64())
        }

        macro_rules! declare_tensor_finite_range_float {
            ($name:ident, $typ:ty) => {
                fn $name(tensor: ndarray::ArrayViewD<'_, $typ>) -> (f64, f64) {
                    re_tracing::profile_function!();
                    let (min, max) = tensor.fold(
                        (<$typ>::INFINITY, <$typ>::NEG_INFINITY),
                        |(min, max), &value| {
                            if value.is_finite() {
                                (min.min(value), max.max(value))
                            } else {
                                (min, max)
                            }
                        },
                    );
                    #[allow(trivial_numeric_casts)]
                    (min as f64, max as f64)
                }
            };
        }

        // declare_tensor_range_float!(tensor_range_f16, half::f16);
        declare_tensor_finite_range_float!(tensor_finite_range_f32, f32);
        declare_tensor_finite_range_float!(tensor_finite_range_f64, f64);

        #[allow(clippy::needless_pass_by_value)]
        fn tensor_finite_range_f16(tensor: ndarray::ArrayViewD<'_, f16>) -> (f64, f64) {
            re_tracing::profile_function!();
            let (min, max) =
                tensor.fold((f16::INFINITY, f16::NEG_INFINITY), |(min, max), &value| {
                    if value.is_finite() {
                        (min.min(value), max.max(value))
                    } else {
                        (min, max)
                    }
                });
            (min.to_f64(), max.to_f64())
        }

        let range = match tensor.dtype() {
            TensorDataType::U8 => ArrayViewD::<u8>::try_from(tensor).map(tensor_range_u8),
            TensorDataType::U16 => ArrayViewD::<u16>::try_from(tensor).map(tensor_range_u16),
            TensorDataType::U32 => ArrayViewD::<u32>::try_from(tensor).map(tensor_range_u32),
            TensorDataType::U64 => ArrayViewD::<u64>::try_from(tensor).map(tensor_range_u64),

            TensorDataType::I8 => ArrayViewD::<i8>::try_from(tensor).map(tensor_range_i8),
            TensorDataType::I16 => ArrayViewD::<i16>::try_from(tensor).map(tensor_range_i16),
            TensorDataType::I32 => ArrayViewD::<i32>::try_from(tensor).map(tensor_range_i32),
            TensorDataType::I64 => ArrayViewD::<i64>::try_from(tensor).map(tensor_range_i64),

            TensorDataType::F16 => ArrayViewD::<f16>::try_from(tensor).map(tensor_range_f16),
            TensorDataType::F32 => ArrayViewD::<f32>::try_from(tensor).map(tensor_range_f32),
            TensorDataType::F64 => ArrayViewD::<f64>::try_from(tensor).map(tensor_range_f64),
        }
        .ok();

        if let Some((min, max)) = range {
            if max < min {
                // Empty tensor
                return Self {
                    range: None,
                    finite_range: (tensor.dtype().min_value(), tensor.dtype().max_value()),
                };
            }
        }

        let finite_range = if range
            .as_ref()
            .map_or(true, |r| r.0.is_finite() && r.1.is_finite())
        {
            range
        } else {
            let finite_range = match tensor.dtype() {
                TensorDataType::U8
                | TensorDataType::U16
                | TensorDataType::U32
                | TensorDataType::U64
                | TensorDataType::I8
                | TensorDataType::I16
                | TensorDataType::I32
                | TensorDataType::I64 => range,

                TensorDataType::F16 => ArrayViewD::<f16>::try_from(tensor)
                    .ok()
                    .map(tensor_finite_range_f16),
                TensorDataType::F32 => ArrayViewD::<f32>::try_from(tensor)
                    .ok()
                    .map(tensor_finite_range_f32),
                TensorDataType::F64 => ArrayViewD::<f64>::try_from(tensor)
                    .ok()
                    .map(tensor_finite_range_f64),
            };

            // If we didn't find a finite range, set it to None.
            finite_range.and_then(|r| {
                if r.0.is_finite() && r.1.is_finite() {
                    Some(r)
                } else {
                    None
                }
            })
        }
        .unwrap_or_else(|| (tensor.dtype().min_value(), tensor.dtype().max_value()));

        Self {
            range,
            finite_range,
        }
    }
}