tenflowers-core 0.2.0

Core tensor operations and execution engine for TenfloweRS
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

//! Auto-generated module
//!
//! 🤖 Generated with [SplitRS](https://github.com/cool-japan/splitrs)

use crate::graph::{Graph, NodeId};
use crate::{Result, TensorError};

use super::pass_support::{get_node_inputs, OptimizationPass};

/// Constant folding optimization pass
/// Evaluates constant expressions at compile time
pub struct ConstantFoldingPass;
impl ConstantFoldingPass {
    pub fn new() -> Self {
        Self
    }
    /// True only when `node_id` is itself a literal `Constant` node (not
    /// merely "part of a constant subtree" — see the comment in `apply`).
    pub(super) fn is_literal_constant(&self, graph: &Graph, node_id: NodeId) -> bool {
        graph
            .get_node(node_id)
            .is_some_and(|node| matches!(node.op_type, crate::graph::NodeType::Constant))
    }
    pub(super) fn can_fold_binary_op(&self, graph: &Graph, node_id: NodeId) -> bool {
        let inputs = get_node_inputs(graph, node_id);
        inputs.len() == 2 && inputs.iter().all(|&id| self.is_literal_constant(graph, id))
    }
    pub(super) fn can_fold_matmul(&self, graph: &Graph, node_id: NodeId) -> bool {
        let inputs = get_node_inputs(graph, node_id);
        inputs.len() == 2 && inputs.iter().all(|&id| self.is_literal_constant(graph, id))
    }
    pub(super) fn evaluate_constant_operation(
        &self,
        graph: &Graph,
        node_id: NodeId,
    ) -> Result<Option<crate::tensor::Tensor<f32>>> {
        use crate::ops::{binary, matmul};
        let node = graph.get_node(node_id).ok_or_else(|| {
            TensorError::invalid_argument(format!("Node {node_id} does not exist"))
        })?;
        if let crate::graph::NodeType::Operation(op_name) = &node.op_type {
            let input_node_ids = get_node_inputs(graph, node_id);
            let input_tensors: std::result::Result<Vec<_>, crate::error::TensorError> =
                input_node_ids
                    .iter()
                    .map(|&input_id| self.get_constant_tensor(graph, input_id))
                    .collect();
            let inputs = input_tensors?;
            match op_name.as_str() {
                "Add" if inputs.len() == 2 => {
                    let result = binary::add(&inputs[0], &inputs[1])?;
                    Ok(Some(result))
                }
                "Sub" if inputs.len() == 2 => {
                    let result = binary::sub(&inputs[0], &inputs[1])?;
                    Ok(Some(result))
                }
                "Mul" if inputs.len() == 2 => {
                    let result = binary::mul(&inputs[0], &inputs[1])?;
                    Ok(Some(result))
                }
                "Div" if inputs.len() == 2 => {
                    let result = binary::div(&inputs[0], &inputs[1])?;
                    Ok(Some(result))
                }
                "MatMul" if inputs.len() == 2 => {
                    let result = matmul::matmul(&inputs[0], &inputs[1])?;
                    Ok(Some(result))
                }
                _ => Ok(None),
            }
        } else {
            Ok(None)
        }
    }
    pub(super) fn get_constant_tensor(
        &self,
        graph: &Graph,
        node_id: NodeId,
    ) -> std::result::Result<crate::tensor::Tensor<f32>, crate::error::TensorError> {
        let node = graph.get_node(node_id).ok_or_else(|| {
            TensorError::invalid_argument(format!("Node {node_id} does not exist"))
        })?;
        if let crate::graph::NodeType::Constant = &node.op_type {
            if let Some(crate::graph::AttributeValue::Tensor(tensor)) = node.attributes.get("value")
            {
                Ok(tensor.clone())
            } else {
                Err(TensorError::invalid_argument(
                    "Constant node missing tensor value".to_string(),
                ))
            }
        } else {
            Err(TensorError::invalid_argument(format!(
                "Node {node_id} is not a constant"
            )))
        }
    }
}
impl OptimizationPass for ConstantFoldingPass {
    fn apply(&self, graph: &mut Graph) -> Result<bool> {
        let mut changed = false;
        let mut nodes_to_fold = Vec::new();
        for node in graph.nodes() {
            if let crate::graph::NodeType::Operation(op_name) = &node.op_type {
                let foldable = match op_name.as_str() {
                    "Add" | "Mul" | "Sub" | "Div" => self.can_fold_binary_op(graph, node.id),
                    "MatMul" => self.can_fold_matmul(graph, node.id),
                    _ => false,
                };
                if foldable {
                    nodes_to_fold.push(node.id);
                }
            }
        }
        for node_id in nodes_to_fold {
            if let Some(result_tensor) = self.evaluate_constant_operation(graph, node_id)? {
                graph.replace_with_constant(node_id, result_tensor)?;
                changed = true;
            }
        }
        Ok(changed)
    }
    fn name(&self) -> &str {
        "ConstantFolding"
    }
    fn is_applicable(&self, graph: &Graph) -> bool {
        graph.node_count() > 0
    }
    fn priority(&self) -> u32 {
        200
    }
}
impl Default for ConstantFoldingPass {
    fn default() -> Self {
        Self::new()
    }
}