onnx-ir 0.19.0

ONNX-IR is a pure Rust library for parsing ONNX models into an intermediate representation that can be used to generate code for various ML/DL frameworks
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
252
253
254
255

use crate::{
    ir::{ArgType, ElementType, Node, TensorType},
    util::{compute_broadcast_rank, compute_broadcast_static_shape},
};

/// Get element type from ArgType, handling Shape types specially
fn get_elem_type(arg_type: &ArgType) -> ElementType {
    match arg_type {
        ArgType::Scalar(elem_type) => elem_type.clone(),
        ArgType::Tensor(tensor) => tensor.elem_type.clone(),
        ArgType::Shape(_) => ElementType::Int64, // Shape types are always i64
    }
}

/// Check if output should be a Shape type
fn should_output_shape(
    x: &ArgType,
    y: &ArgType,
    output_rank: usize,
    elem_type: &ElementType,
) -> bool {
    // Output Shape if both inputs are Shape and output would be 1D int64
    matches!(x, ArgType::Shape(_))
        && matches!(y, ArgType::Shape(_))
        && output_rank == 1
        && *elem_type == ElementType::Int64
}

/// Get size of Shape type, or 1 for other types
fn get_shape_size(arg_type: &ArgType) -> usize {
    match arg_type {
        ArgType::Shape(size) => *size,
        _ => 1,
    }
}

/// Update output type for Where operation.
///
/// The Where operation selects elements from x or y based on condition.
/// Output shape is the broadcasted shape of all three inputs.
/// Output element type is taken from x and y (which must match).
pub fn where_update_outputs(node: &mut Node) {
    log::debug!("Where rank inference for node {}", node.name);

    let condition = &node.inputs[0].ty;
    let x = &node.inputs[1].ty;
    let y = &node.inputs[2].ty;

    // Get element types, handling Shape types specially
    let x_elem_type = get_elem_type(x);
    let y_elem_type = get_elem_type(y);
    let condition_elem_type = get_elem_type(condition);

    if !matches!(condition, ArgType::Shape(_)) {
        assert_eq!(
            condition_elem_type,
            ElementType::Bool,
            "Where condition must be boolean!"
        );
    }

    let elem_type = if x_elem_type == y_elem_type {
        x_elem_type
    } else if matches!(x, ArgType::Shape(_)) {
        y_elem_type
    } else if matches!(y, ArgType::Shape(_)) {
        x_elem_type
    } else {
        panic!(
            "Where x and y have different element types! ({:?} vs {:?})",
            x_elem_type, y_elem_type
        );
    };

    log::debug!(
        "Where input ranks for {}: condition={}, x={}, y={}",
        node.name,
        condition.rank(),
        x.rank(),
        y.rank()
    );

    let output_rank = compute_broadcast_rank(&node.inputs);
    log::debug!("Where output rank for {}: {}", node.name, output_rank);

    // Determine output type
    if output_rank == 0 {
        node.outputs[0].ty = ArgType::Scalar(elem_type);
        log::debug!("Where result for {} is scalar", node.name);
    } else if should_output_shape(x, y, output_rank, &elem_type) {
        // If both inputs are Shape types and output is 1D int64, preserve Shape type
        let shape_size = get_shape_size(x).max(get_shape_size(y));
        node.outputs[0].ty = ArgType::Shape(shape_size);
        log::debug!(
            "Where result for {} is Shape({}) type",
            node.name,
            shape_size
        );
    } else {
        // Try to propagate static shape using the shared broadcast helper
        let static_shape = compute_broadcast_static_shape(&node.inputs);

        node.outputs[0].ty = ArgType::Tensor(TensorType {
            elem_type,
            rank: output_rank,
            static_shape,
        });
        log::debug!(
            "Where result for {} is tensor with rank {}, static_shape: {:?}",
            node.name,
            output_rank,
            node.outputs[0].ty.static_shape()
        );
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::ir::NodeType;
    use crate::node::test_utils::NodeBuilder;

    fn create_test_node(condition_rank: usize, x_rank: usize, y_rank: usize) -> Node {
        NodeBuilder::new(NodeType::Where, "test_where")
            .input_tensor_bool("condition", condition_rank, None)
            .input_tensor_f32("X", x_rank, None)
            .input_tensor_f32("Y", y_rank, None)
            .output_tensor_f32("output", 0, None) // Rank will be updated
            .build()
    }

    #[test]
    fn test_where_basic() {
        let mut node = create_test_node(2, 3, 2);
        where_update_outputs(&mut node);

        match &node.outputs[0].ty {
            ArgType::Tensor(tensor) => {
                assert_eq!(tensor.elem_type, ElementType::Float32);
                assert_eq!(tensor.rank, 3); // max(2, max(3, 2)) = 3
            }
            _ => panic!("Expected tensor output"),
        }
    }

    #[test]
    fn test_where_scalar_result() {
        let mut node = create_test_node(0, 0, 0);
        where_update_outputs(&mut node);

        match &node.outputs[0].ty {
            ArgType::Scalar(elem_type) => {
                assert_eq!(*elem_type, ElementType::Float32);
            }
            _ => panic!("Expected scalar output"),
        }
    }

    #[test]
    #[should_panic(expected = "Where condition must be boolean!")]
    fn test_where_invalid_condition() {
        let mut node = create_test_node(2, 2, 2);

        // Replace condition with non-boolean tensor
        let non_bool_input = NodeBuilder::new(NodeType::Identity, "temp")
            .input_tensor_f32("x", 2, None)
            .build()
            .inputs
            .pop()
            .unwrap();

        node.inputs[0] = non_bool_input;
        where_update_outputs(&mut node);
    }

    #[test]
    #[should_panic(expected = "Where x and y have different element types!")]
    fn test_where_mismatched_types() {
        let mut node = create_test_node(2, 2, 2);

        // Replace Y with int64 tensor (different from X's float32)
        let int64_input = NodeBuilder::new(NodeType::Identity, "temp")
            .input_tensor_i64("y", 2, None)
            .build()
            .inputs
            .pop()
            .unwrap();

        node.inputs[2] = int64_input;
        where_update_outputs(&mut node);
    }

    #[test]
    fn test_where_with_shape_inputs() {
        let mut node = create_test_node(1, 0, 0);

        // Replace X and Y with Shape types
        node.inputs[1].ty = ArgType::Shape(3);
        node.inputs[2].ty = ArgType::Shape(3);

        where_update_outputs(&mut node);

        match &node.outputs[0].ty {
            ArgType::Shape(size) => {
                assert_eq!(*size, 3); // Should preserve Shape type
            }
            _ => panic!("Expected Shape output"),
        }
    }

    #[test]
    fn test_where_static_shape_propagation() {
        // Test that static shapes propagate correctly through Where
        let mut node = NodeBuilder::new(NodeType::Where, "test_where")
            .input_tensor_bool("condition", 2, Some(vec![2, 2]))
            .input_tensor_f32("X", 2, Some(vec![2, 2]))
            .input_tensor_f32("Y", 2, Some(vec![2, 2]))
            .output_tensor_f32("output", 0, None)
            .build();

        where_update_outputs(&mut node);

        match &node.outputs[0].ty {
            ArgType::Tensor(tensor) => {
                assert_eq!(tensor.elem_type, ElementType::Float32);
                assert_eq!(tensor.rank, 2);
                assert_eq!(tensor.static_shape, Some(vec![2, 2]));
            }
            _ => panic!("Expected tensor output"),
        }
    }

    #[test]
    fn test_where_static_shape_propagation_partial() {
        // Test that static shape propagates even when only some inputs have it
        let mut node = NodeBuilder::new(NodeType::Where, "test_where")
            .input_tensor_bool("condition", 2, None) // No static shape
            .input_tensor_f32("X", 2, Some(vec![3, 4])) // Has static shape
            .input_tensor_f32("Y", 2, Some(vec![3, 4])) // Has static shape
            .output_tensor_f32("output", 0, None)
            .build();

        where_update_outputs(&mut node);

        match &node.outputs[0].ty {
            ArgType::Tensor(tensor) => {
                assert_eq!(tensor.elem_type, ElementType::Float32);
                assert_eq!(tensor.rank, 2);
                // Since X and Y have the same static shape, it should propagate
                assert_eq!(tensor.static_shape, Some(vec![3, 4]));
            }
            _ => panic!("Expected tensor output"),
        }
    }
}