runmat_runtime/builtins/math/interpolation/
ppval.rs

1//! Evaluate piecewise-polynomial structs produced by `spline` and `pchip`.
2
3use runmat_builtins::{ResolveContext, Type, Value};
4use runmat_macros::runtime_builtin;
5
6use crate::builtins::common::spec::{
7    BroadcastSemantics, BuiltinFusionSpec, BuiltinGpuSpec, ConstantStrategy, GpuOpKind,
8    ReductionNaN, ResidencyPolicy, ScalarType, ShapeRequirements,
9};
10
11use super::pp::{evaluate_pp, pp_from_value, query_points, Extrapolation};
12
13const NAME: &str = "ppval";
14
15#[runmat_macros::register_gpu_spec(builtin_path = "crate::builtins::math::interpolation::ppval")]
16pub const GPU_SPEC: BuiltinGpuSpec = BuiltinGpuSpec {
17    name: NAME,
18    op_kind: GpuOpKind::Custom("piecewise-polynomial-eval"),
19    supported_precisions: &[ScalarType::F32, ScalarType::F64],
20    broadcast: BroadcastSemantics::Matlab,
21    provider_hooks: &[],
22    constant_strategy: ConstantStrategy::UniformBuffer,
23    residency: ResidencyPolicy::GatherImmediately,
24    nan_mode: ReductionNaN::Include,
25    two_pass_threshold: None,
26    workgroup_size: None,
27    accepts_nan_mode: false,
28    notes:
29        "Initial implementation evaluates pp structs on the CPU after gathering GPU query points.",
30};
31
32#[runmat_macros::register_fusion_spec(builtin_path = "crate::builtins::math::interpolation::ppval")]
33pub const FUSION_SPEC: BuiltinFusionSpec = BuiltinFusionSpec {
34    name: NAME,
35    shape: ShapeRequirements::Any,
36    constant_strategy: ConstantStrategy::UniformBuffer,
37    elementwise: None,
38    reduction: None,
39    emits_nan: true,
40    notes: "ppval is currently a runtime sink.",
41};
42
43fn ppval_type(args: &[Type], _ctx: &ResolveContext) -> Type {
44    match args.get(1) {
45        Some(Type::Num | Type::Int | Type::Bool) => Type::Num,
46        Some(Type::Tensor { shape }) | Some(Type::Logical { shape }) => Type::Tensor {
47            shape: shape.clone(),
48        },
49        _ => Type::tensor(),
50    }
51}
52
53#[runtime_builtin(
54    name = "ppval",
55    category = "math/interpolation",
56    summary = "Evaluate a piecewise-polynomial structure at query points.",
57    keywords = "ppval,piecewise polynomial,spline,pchip",
58    accel = "sink",
59    sink = true,
60    type_resolver(ppval_type),
61    builtin_path = "crate::builtins::math::interpolation::ppval"
62)]
63async fn ppval_builtin(pp: Value, xq: Value) -> crate::BuiltinResult<Value> {
64    let parsed = pp_from_value(pp, NAME).await?;
65    let query = query_points(xq, NAME).await?;
66    evaluate_pp(&parsed, &query, &Extrapolation::Extrapolate, NAME)
67}
68
69#[cfg(test)]
70mod tests {
71    use super::*;
72    use crate::builtins::math::interpolation::pp::{build_spline_pp, pp_to_value, NumericSeries};
73    use futures::executor::block_on;
74    use runmat_builtins::Tensor;
75
76    #[test]
77    fn ppval_evaluates_spline_struct() {
78        let series = NumericSeries {
79            x: vec![1.0, 2.0, 3.0],
80            y: vec![1.0, 4.0, 9.0],
81            series: 1,
82            trailing_shape: Vec::new(),
83        };
84        let pp = pp_to_value(
85            build_spline_pp(&series, "spline").expect("spline"),
86            "spline",
87        )
88        .expect("pp");
89        let query = Value::Tensor(Tensor::new(vec![1.5, 2.5], vec![1, 2]).expect("tensor"));
90        let value = block_on(ppval_builtin(pp, query)).expect("ppval");
91        let Value::Tensor(tensor) = value else {
92            panic!("expected tensor");
93        };
94        assert!((tensor.data[0] - 2.25).abs() < 1e-10);
95        assert!((tensor.data[1] - 6.25).abs() < 1e-10);
96    }
97}
runmat_runtime/builtins/math/interpolation/ppval.rs

runmat_runtime/builtins/math/interpolation/
ppval.rs
