executorch 0.9.0

Rust bindings for ExecuTorch - On-device AI across mobile, embedded and edge for PyTorch
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
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251

use executorch_sys::ScalarType as CScalarType;

use crate::util::{IntoCpp, IntoRust};

/// Data types (dtypes) that can be used as element types in Tensors.
///
/// The enum contain all the scalar types supported by the Cpp ExecuTorch library.
/// Not all of these types are supported by the Rust library, see [`Scalar`].
#[repr(i8)]
#[derive(Debug, Copy, Clone, Eq, PartialEq)]
pub enum ScalarType {
    /// 8-bit unsigned integer, `u8`
    Byte = CScalarType::ScalarType_Byte as i8,
    /// 8-bit signed, integer, `i8`
    Char = CScalarType::ScalarType_Char as i8,
    /// 16-bit signed integer, `i16`
    Short = CScalarType::ScalarType_Short as i8,
    /// 32-bit signed integer, `i32`
    Int = CScalarType::ScalarType_Int as i8,
    /// 64-bit signed integer, `i64`
    Long = CScalarType::ScalarType_Long as i8,
    /// 16-bit floating point, [`executorch::scalar::f16`](`crate::scalar::f16`).
    Half = CScalarType::ScalarType_Half as i8,
    /// 32-bit floating point, `f32`
    Float = CScalarType::ScalarType_Float as i8,
    /// 64-bit floating point, `f64`
    Double = CScalarType::ScalarType_Double as i8,
    /// 16-bit complex floating point, [`executorch::scalar::Complex<executorch::scalar::f16>`](`crate::scalar::Complex`).
    ComplexHalf = CScalarType::ScalarType_ComplexHalf as i8,
    /// 32-bit complex floating point, [`executorch::scalar::Complex<f32>`](`crate::scalar::Complex`).
    ComplexFloat = CScalarType::ScalarType_ComplexFloat as i8,
    /// 64-bit complex floating point, [`executorch::scalar::Complex<f64>`](`crate::scalar::Complex`).
    ComplexDouble = CScalarType::ScalarType_ComplexDouble as i8,
    /// Boolean, `bool`
    Bool = CScalarType::ScalarType_Bool as i8,
    /// 8-bit quantized integer, [`executorch::scalar::QInt8`](`crate::scalar::QInt8`).
    QInt8 = CScalarType::ScalarType_QInt8 as i8,
    /// 8-bit quantized unsigned integer, [`executorch::scalar::QUInt8`](`crate::scalar::QUInt8`).
    QUInt8 = CScalarType::ScalarType_QUInt8 as i8,
    /// 32-bit quantized integer, [`executorch::scalar::QInt32`](`crate::scalar::QInt32`).
    QInt32 = CScalarType::ScalarType_QInt32 as i8,
    /// 16-bit floating point using the bfloat16 format, [`executorch::scalar::bf16`](`crate::scalar::bf16`).
    BFloat16 = CScalarType::ScalarType_BFloat16 as i8,
    /// Two 4-bit unsigned quantized integers packed into a byte. [`executorch::scalar::QUInt4x2`](`crate::scalar::QUInt4x2`).
    QUInt4x2 = CScalarType::ScalarType_QUInt4x2 as i8,
    /// Four 2-bit unsigned quantized integers packed into a byte. [`executorch::scalar::QUInt2x4`](`crate::scalar::QUInt2x4`).
    QUInt2x4 = CScalarType::ScalarType_QUInt2x4 as i8,
    /// Eight 1-bit values packed into a byte. [`executorch::scalar::Bits1x8`](`crate::scalar::Bits1x8`).
    Bits1x8 = CScalarType::ScalarType_Bits1x8 as i8,
    /// Four 2-bit values packed into a byte. [`executorch::scalar::Bits2x4`](`crate::scalar::Bits2x4`).
    Bits2x4 = CScalarType::ScalarType_Bits2x4 as i8,
    /// Two 4-bit values packed into a byte. [`executorch::scalar::Bits4x2`](`crate::scalar::Bits4x2`).
    Bits4x2 = CScalarType::ScalarType_Bits4x2 as i8,
    /// 8-bit bitfield (1 byte). [`executorch::scalar::Bits8`](`crate::scalar::Bits8`).
    Bits8 = CScalarType::ScalarType_Bits8 as i8,
    /// 16-bit bitfield (2 bytes). [`executorch::scalar::Bits16`](`crate::scalar::Bits16`).
    Bits16 = CScalarType::ScalarType_Bits16 as i8,
    /// 8-bit floating-point with 1 bit for the sign, 5 bits for the exponents, 2 bits for the mantissa.
    /// [`executorch::scalar::Float8_e5m2`](`crate::scalar::Float8_e5m2`).
    #[allow(non_camel_case_types)]
    Float8_e5m2 = CScalarType::ScalarType_Float8_e5m2 as i8,
    /// 8-bit floating-point with 1 bit for the sign, 4 bits for the exponents, 3 bits for the mantissa,
    /// only nan values and no infinite values (FN).
    /// [`executorch::scalar::Float8_e4m3fn`](`crate::scalar::Float8_e4m3fn`).
    #[allow(non_camel_case_types)]
    Float8_e4m3fn = CScalarType::ScalarType_Float8_e4m3fn as i8,
    /// 8-bit floating-point with 1 bit for the sign, 5 bits for the exponents, 2 bits for the mantissa,
    /// only nan values and no infinite values (FN), no negative zero (UZ).
    /// [`executorch::scalar::Float8_e5m2fnuz`](`crate::scalar::Float8_e5m2fnuz`).
    #[allow(non_camel_case_types)]
    Float8_e5m2fnuz = CScalarType::ScalarType_Float8_e5m2fnuz as i8,
    /// 8-bit floating-point with 1 bit for the sign, 4 bits for the exponents, 3 bits for the mantissa,
    /// only nan values and no infinite values (FN), no negative zero (UZ).
    /// [`executorch::scalar::Float8_e4m3fnuz`](`crate::scalar::Float8_e4m3fnuz`).
    #[allow(non_camel_case_types)]
    Float8_e4m3fnuz = CScalarType::ScalarType_Float8_e4m3fnuz as i8,
    /// 16-bit unsigned integer, `u16`
    UInt16 = CScalarType::ScalarType_UInt16 as i8,
    /// 32-bit unsigned integer, `u32`
    UInt32 = CScalarType::ScalarType_UInt32 as i8,
    /// 64-bit unsigned integer, `u64`
    UInt64 = CScalarType::ScalarType_UInt64 as i8,
}
impl IntoRust for CScalarType {
    type RsType = ScalarType;
    fn rs(self) -> Self::RsType {
        match self {
            CScalarType::ScalarType_Byte => ScalarType::Byte,
            CScalarType::ScalarType_Char => ScalarType::Char,
            CScalarType::ScalarType_Short => ScalarType::Short,
            CScalarType::ScalarType_Int => ScalarType::Int,
            CScalarType::ScalarType_Long => ScalarType::Long,
            CScalarType::ScalarType_Half => ScalarType::Half,
            CScalarType::ScalarType_Float => ScalarType::Float,
            CScalarType::ScalarType_Double => ScalarType::Double,
            CScalarType::ScalarType_ComplexHalf => ScalarType::ComplexHalf,
            CScalarType::ScalarType_ComplexFloat => ScalarType::ComplexFloat,
            CScalarType::ScalarType_ComplexDouble => ScalarType::ComplexDouble,
            CScalarType::ScalarType_Bool => ScalarType::Bool,
            CScalarType::ScalarType_QInt8 => ScalarType::QInt8,
            CScalarType::ScalarType_QUInt8 => ScalarType::QUInt8,
            CScalarType::ScalarType_QInt32 => ScalarType::QInt32,
            CScalarType::ScalarType_BFloat16 => ScalarType::BFloat16,
            CScalarType::ScalarType_QUInt4x2 => ScalarType::QUInt4x2,
            CScalarType::ScalarType_QUInt2x4 => ScalarType::QUInt2x4,
            CScalarType::ScalarType_Bits1x8 => ScalarType::Bits1x8,
            CScalarType::ScalarType_Bits2x4 => ScalarType::Bits2x4,
            CScalarType::ScalarType_Bits4x2 => ScalarType::Bits4x2,
            CScalarType::ScalarType_Bits8 => ScalarType::Bits8,
            CScalarType::ScalarType_Bits16 => ScalarType::Bits16,
            CScalarType::ScalarType_Float8_e5m2 => ScalarType::Float8_e5m2,
            CScalarType::ScalarType_Float8_e4m3fn => ScalarType::Float8_e4m3fn,
            CScalarType::ScalarType_Float8_e5m2fnuz => ScalarType::Float8_e5m2fnuz,
            CScalarType::ScalarType_Float8_e4m3fnuz => ScalarType::Float8_e4m3fnuz,
            CScalarType::ScalarType_UInt16 => ScalarType::UInt16,
            CScalarType::ScalarType_UInt32 => ScalarType::UInt32,
            CScalarType::ScalarType_UInt64 => ScalarType::UInt64,
        }
    }
}
impl IntoCpp for ScalarType {
    type CppType = CScalarType;

    fn cpp(self) -> Self::CppType {
        match self {
            ScalarType::Byte => CScalarType::ScalarType_Byte,
            ScalarType::Char => CScalarType::ScalarType_Char,
            ScalarType::Short => CScalarType::ScalarType_Short,
            ScalarType::Int => CScalarType::ScalarType_Int,
            ScalarType::Long => CScalarType::ScalarType_Long,
            ScalarType::Half => CScalarType::ScalarType_Half,
            ScalarType::Float => CScalarType::ScalarType_Float,
            ScalarType::Double => CScalarType::ScalarType_Double,
            ScalarType::ComplexHalf => CScalarType::ScalarType_ComplexHalf,
            ScalarType::ComplexFloat => CScalarType::ScalarType_ComplexFloat,
            ScalarType::ComplexDouble => CScalarType::ScalarType_ComplexDouble,
            ScalarType::Bool => CScalarType::ScalarType_Bool,
            ScalarType::QInt8 => CScalarType::ScalarType_QInt8,
            ScalarType::QUInt8 => CScalarType::ScalarType_QUInt8,
            ScalarType::QInt32 => CScalarType::ScalarType_QInt32,
            ScalarType::BFloat16 => CScalarType::ScalarType_BFloat16,
            ScalarType::QUInt4x2 => CScalarType::ScalarType_QUInt4x2,
            ScalarType::QUInt2x4 => CScalarType::ScalarType_QUInt2x4,
            ScalarType::Bits1x8 => CScalarType::ScalarType_Bits1x8,
            ScalarType::Bits2x4 => CScalarType::ScalarType_Bits2x4,
            ScalarType::Bits4x2 => CScalarType::ScalarType_Bits4x2,
            ScalarType::Bits8 => CScalarType::ScalarType_Bits8,
            ScalarType::Bits16 => CScalarType::ScalarType_Bits16,
            ScalarType::Float8_e5m2 => CScalarType::ScalarType_Float8_e5m2,
            ScalarType::Float8_e4m3fn => CScalarType::ScalarType_Float8_e4m3fn,
            ScalarType::Float8_e5m2fnuz => CScalarType::ScalarType_Float8_e5m2fnuz,
            ScalarType::Float8_e4m3fnuz => CScalarType::ScalarType_Float8_e4m3fnuz,
            ScalarType::UInt16 => CScalarType::ScalarType_UInt16,
            ScalarType::UInt32 => CScalarType::ScalarType_UInt32,
            ScalarType::UInt64 => CScalarType::ScalarType_UInt64,
        }
    }
}

/// A trait for types that can be used as scalar types in Tensors.
pub trait Scalar: 'static {
    /// The [`ScalarType`] enum variant of the implementing type.
    const TYPE: ScalarType;
    private_decl! {}
}
macro_rules! impl_scalar {
    ($rust_type:path, $scalar_type_variant:ident) => {
        impl Scalar for $rust_type {
            const TYPE: ScalarType = ScalarType::$scalar_type_variant;
            private_impl! {}
        }
    };
}

impl_scalar!(u8, Byte);
impl_scalar!(i8, Char);
impl_scalar!(i16, Short);
impl_scalar!(i32, Int);
impl_scalar!(i64, Long);
impl_scalar!(crate::scalar::f16, Half);
impl_scalar!(f32, Float);
impl_scalar!(f64, Double);
impl_scalar!(crate::scalar::Complex<crate::scalar::f16>, ComplexHalf);
impl_scalar!(crate::scalar::Complex<f32>, ComplexFloat);
impl_scalar!(crate::scalar::Complex<f64>, ComplexDouble);
impl_scalar!(bool, Bool);
impl_scalar!(crate::scalar::QInt8, QInt8);
impl_scalar!(crate::scalar::QUInt8, QUInt8);
impl_scalar!(crate::scalar::QInt32, QInt32);
impl_scalar!(crate::scalar::bf16, BFloat16);
impl_scalar!(crate::scalar::QUInt4x2, QUInt4x2);
impl_scalar!(crate::scalar::QUInt2x4, QUInt2x4);
impl_scalar!(crate::scalar::Bits1x8, Bits1x8);
impl_scalar!(crate::scalar::Bits2x4, Bits2x4);
impl_scalar!(crate::scalar::Bits4x2, Bits4x2);
impl_scalar!(crate::scalar::Bits8, Bits8);
impl_scalar!(crate::scalar::Bits16, Bits16);
impl_scalar!(crate::scalar::Float8_e5m2, Float8_e5m2);
impl_scalar!(crate::scalar::Float8_e4m3fn, Float8_e4m3fn);
impl_scalar!(crate::scalar::Float8_e5m2fnuz, Float8_e5m2fnuz);
impl_scalar!(crate::scalar::Float8_e4m3fnuz, Float8_e4m3fnuz);
impl_scalar!(u16, UInt16);
impl_scalar!(u32, UInt32);
impl_scalar!(u64, UInt64);

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

    #[test]
    fn rust_cpp_conversions() {
        type CType = CScalarType;
        type RType = ScalarType;
        let scalars = [
            (CType::ScalarType_Byte, RType::Byte),
            (CType::ScalarType_Char, RType::Char),
            (CType::ScalarType_Short, RType::Short),
            (CType::ScalarType_Int, RType::Int),
            (CType::ScalarType_Long, RType::Long),
            (CType::ScalarType_Half, RType::Half),
            (CType::ScalarType_Float, RType::Float),
            (CType::ScalarType_Double, RType::Double),
            (CType::ScalarType_ComplexHalf, RType::ComplexHalf),
            (CType::ScalarType_ComplexFloat, RType::ComplexFloat),
            (CType::ScalarType_ComplexDouble, RType::ComplexDouble),
            (CType::ScalarType_Bool, RType::Bool),
            (CType::ScalarType_QInt8, RType::QInt8),
            (CType::ScalarType_QUInt8, RType::QUInt8),
            (CType::ScalarType_QInt32, RType::QInt32),
            (CType::ScalarType_BFloat16, RType::BFloat16),
            (CType::ScalarType_QUInt4x2, RType::QUInt4x2),
            (CType::ScalarType_QUInt2x4, RType::QUInt2x4),
            (CType::ScalarType_Bits1x8, RType::Bits1x8),
            (CType::ScalarType_Bits2x4, RType::Bits2x4),
            (CType::ScalarType_Bits4x2, RType::Bits4x2),
            (CType::ScalarType_Bits8, RType::Bits8),
            (CType::ScalarType_Bits16, RType::Bits16),
            (CType::ScalarType_Float8_e5m2, RType::Float8_e5m2),
            (CType::ScalarType_Float8_e4m3fn, RType::Float8_e4m3fn),
            (CType::ScalarType_Float8_e5m2fnuz, RType::Float8_e5m2fnuz),
            (CType::ScalarType_Float8_e4m3fnuz, RType::Float8_e4m3fnuz),
            (CType::ScalarType_UInt16, RType::UInt16),
            (CType::ScalarType_UInt32, RType::UInt32),
            (CType::ScalarType_UInt64, RType::UInt64),
        ];
        for (cpp, rust) in scalars {
            assert_eq!(cpp.rs(), rust);
            assert_eq!(rust.cpp(), cpp);
        }
    }
}