Skip to main content

runmat_runtime/builtins/strings/core/
str2double.rs

1//! MATLAB-compatible `str2double` builtin with GPU-aware semantics for RunMat.
2
3use std::borrow::Cow;
4
5use runmat_builtins::{CellArray, CharArray, StringArray, Tensor, Value};
6use runmat_macros::runtime_builtin;
7
8use crate::builtins::common::map_control_flow_with_builtin;
9use crate::builtins::common::spec::{
10    BroadcastSemantics, BuiltinFusionSpec, BuiltinGpuSpec, ConstantStrategy, GpuOpKind,
11    ReductionNaN, ResidencyPolicy, ShapeRequirements,
12};
13use crate::builtins::common::tensor;
14use crate::builtins::strings::type_resolvers::numeric_text_scalar_or_tensor_type;
15use crate::{build_runtime_error, gather_if_needed_async, BuiltinResult, RuntimeError};
16
17#[runmat_macros::register_gpu_spec(builtin_path = "crate::builtins::strings::core::str2double")]
18pub const GPU_SPEC: BuiltinGpuSpec = BuiltinGpuSpec {
19    name: "str2double",
20    op_kind: GpuOpKind::Custom("conversion"),
21    supported_precisions: &[],
22    broadcast: BroadcastSemantics::None,
23    provider_hooks: &[],
24    constant_strategy: ConstantStrategy::InlineLiteral,
25    residency: ResidencyPolicy::GatherImmediately,
26    nan_mode: ReductionNaN::Include,
27    two_pass_threshold: None,
28    workgroup_size: None,
29    accepts_nan_mode: false,
30    notes: "Parses text on the CPU; GPU-resident inputs are gathered before conversion.",
31};
32
33#[runmat_macros::register_fusion_spec(builtin_path = "crate::builtins::strings::core::str2double")]
34pub const FUSION_SPEC: BuiltinFusionSpec = BuiltinFusionSpec {
35    name: "str2double",
36    shape: ShapeRequirements::Any,
37    constant_strategy: ConstantStrategy::InlineLiteral,
38    elementwise: None,
39    reduction: None,
40    emits_nan: true,
41    notes: "Conversion builtin; not eligible for fusion and materialises host-side doubles.",
42};
43
44const ARG_TYPE_ERROR: &str =
45    "str2double: input must be a string array, character array, or cell array of character vectors";
46const CELL_ELEMENT_ERROR: &str =
47    "str2double: cell array elements must be character vectors or string scalars";
48
49fn str2double_flow(message: impl Into<String>) -> RuntimeError {
50    build_runtime_error(message)
51        .with_builtin("str2double")
52        .build()
53}
54
55fn remap_str2double_flow(err: RuntimeError) -> RuntimeError {
56    map_control_flow_with_builtin(err, "str2double")
57}
58
59#[runtime_builtin(
60    name = "str2double",
61    category = "strings/core",
62    summary = "Convert strings, character arrays, or cell arrays of text into doubles.",
63    keywords = "str2double,string to double,text conversion,gpu",
64    accel = "sink",
65    type_resolver(numeric_text_scalar_or_tensor_type),
66    builtin_path = "crate::builtins::strings::core::str2double"
67)]
68async fn str2double_builtin(value: Value) -> crate::BuiltinResult<Value> {
69    let gathered = gather_if_needed_async(&value)
70        .await
71        .map_err(remap_str2double_flow)?;
72    match gathered {
73        Value::String(text) => Ok(Value::Num(parse_numeric_scalar(&text))),
74        Value::StringArray(array) => str2double_string_array(array),
75        Value::CharArray(array) => str2double_char_array(array),
76        Value::Cell(cell) => str2double_cell_array(cell),
77        _ => Err(str2double_flow(ARG_TYPE_ERROR)),
78    }
79}
80
81fn str2double_string_array(array: StringArray) -> BuiltinResult<Value> {
82    let StringArray { data, shape, .. } = array;
83    let mut values = Vec::with_capacity(data.len());
84    for text in &data {
85        values.push(parse_numeric_scalar(text));
86    }
87    let tensor =
88        Tensor::new(values, shape).map_err(|e| str2double_flow(format!("str2double: {e}")))?;
89    Ok(tensor::tensor_into_value(tensor))
90}
91
92fn str2double_char_array(array: CharArray) -> BuiltinResult<Value> {
93    let rows = array.rows;
94    let cols = array.cols;
95    let mut values = Vec::with_capacity(rows);
96    for row in 0..rows {
97        let start = row * cols;
98        let end = start + cols;
99        let row_text: String = array.data[start..end].iter().collect();
100        values.push(parse_numeric_scalar(&row_text));
101    }
102    let tensor = Tensor::new(values, vec![rows, 1])
103        .map_err(|e| str2double_flow(format!("str2double: {e}")))?;
104    Ok(tensor::tensor_into_value(tensor))
105}
106
107fn str2double_cell_array(cell: CellArray) -> BuiltinResult<Value> {
108    let CellArray {
109        data, rows, cols, ..
110    } = cell;
111    let mut values = Vec::with_capacity(rows * cols);
112    for col in 0..cols {
113        for row in 0..rows {
114            let idx = row * cols + col;
115            let element: &Value = &data[idx];
116            let numeric = match element {
117                Value::String(text) => parse_numeric_scalar(text),
118                Value::StringArray(sa) if sa.data.len() == 1 => parse_numeric_scalar(&sa.data[0]),
119                Value::CharArray(char_vec) if char_vec.rows == 1 => {
120                    let row_text: String = char_vec.data.iter().collect();
121                    parse_numeric_scalar(&row_text)
122                }
123                Value::CharArray(_) => return Err(str2double_flow(CELL_ELEMENT_ERROR)),
124                _ => return Err(str2double_flow(CELL_ELEMENT_ERROR)),
125            };
126            values.push(numeric);
127        }
128    }
129    let tensor = Tensor::new(values, vec![rows, cols])
130        .map_err(|e| str2double_flow(format!("str2double: {e}")))?;
131    Ok(tensor::tensor_into_value(tensor))
132}
133
134fn parse_numeric_scalar(text: &str) -> f64 {
135    let trimmed = text.trim();
136    if trimmed.is_empty() {
137        return f64::NAN;
138    }
139
140    let lowered = trimmed.to_ascii_lowercase();
141    match lowered.as_str() {
142        "nan" => return f64::NAN,
143        "inf" | "+inf" | "infinity" | "+infinity" => return f64::INFINITY,
144        "-inf" | "-infinity" => return f64::NEG_INFINITY,
145        _ => {}
146    }
147
148    let normalized: Cow<'_, str> = if trimmed.chars().any(|c| c == 'd' || c == 'D') {
149        Cow::Owned(
150            trimmed
151                .chars()
152                .map(|c| if c == 'd' || c == 'D' { 'e' } else { c })
153                .collect(),
154        )
155    } else {
156        Cow::Borrowed(trimmed)
157    };
158
159    normalized.parse::<f64>().unwrap_or(f64::NAN)
160}
161
162#[cfg(test)]
163pub(crate) mod tests {
164    use super::*;
165    use runmat_builtins::{ResolveContext, Type};
166
167    fn str2double_builtin(value: Value) -> BuiltinResult<Value> {
168        futures::executor::block_on(super::str2double_builtin(value))
169    }
170
171    fn error_message(err: crate::RuntimeError) -> String {
172        err.message().to_string()
173    }
174
175    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
176    #[test]
177    fn str2double_string_scalar() {
178        let result = str2double_builtin(Value::String("42.5".into())).expect("str2double");
179        assert_eq!(result, Value::Num(42.5));
180    }
181
182    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
183    #[test]
184    fn str2double_string_scalar_invalid_returns_nan() {
185        let result = str2double_builtin(Value::String("abc".into())).expect("str2double");
186        match result {
187            Value::Num(v) => assert!(v.is_nan()),
188            other => panic!("expected scalar result, got {other:?}"),
189        }
190    }
191
192    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
193    #[test]
194    fn str2double_string_array_preserves_shape() {
195        let array =
196            StringArray::new(vec!["1".into(), " 2.5 ".into(), "foo".into()], vec![3, 1]).unwrap();
197        let result = str2double_builtin(Value::StringArray(array)).expect("str2double");
198        match result {
199            Value::Tensor(tensor) => {
200                assert_eq!(tensor.shape, vec![3, 1]);
201                assert_eq!(tensor.data[0], 1.0);
202                assert_eq!(tensor.data[1], 2.5);
203                assert!(tensor.data[2].is_nan());
204            }
205            Value::Num(_) => panic!("expected tensor"),
206            other => panic!("unexpected result {other:?}"),
207        }
208    }
209
210    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
211    #[test]
212    fn str2double_char_array_multiple_rows() {
213        let data: Vec<char> = vec!['4', '2', ' ', ' ', '1', '0', '0', ' '];
214        let array = CharArray::new(data, 2, 4).unwrap();
215        let result = str2double_builtin(Value::CharArray(array)).expect("str2double");
216        match result {
217            Value::Tensor(tensor) => {
218                assert_eq!(tensor.shape, vec![2, 1]);
219                assert_eq!(tensor.data[0], 42.0);
220                assert_eq!(tensor.data[1], 100.0);
221            }
222            other => panic!("expected tensor result, got {other:?}"),
223        }
224    }
225
226    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
227    #[test]
228    fn str2double_char_array_empty_rows() {
229        let array = CharArray::new(Vec::new(), 0, 0).unwrap();
230        let result = str2double_builtin(Value::CharArray(array)).expect("str2double");
231        match result {
232            Value::Tensor(tensor) => {
233                assert_eq!(tensor.shape, vec![0, 1]);
234                assert_eq!(tensor.data.len(), 0);
235            }
236            other => panic!("expected empty tensor, got {other:?}"),
237        }
238    }
239
240    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
241    #[test]
242    #[allow(
243        clippy::approx_constant,
244        reason = "Test ensures literal 3.14 text stays 3.14, not π"
245    )]
246    fn str2double_cell_array_of_text() {
247        let cell = CellArray::new(
248            vec![
249                Value::String("3.14".into()),
250                Value::CharArray(CharArray::new_row("NaN")),
251                Value::String("-Inf".into()),
252            ],
253            1,
254            3,
255        )
256        .unwrap();
257        let result = str2double_builtin(Value::Cell(cell)).expect("str2double");
258        match result {
259            Value::Tensor(tensor) => {
260                assert_eq!(tensor.shape, vec![1, 3]);
261                assert_eq!(tensor.data[0], 3.14);
262                assert!(tensor.data[1].is_nan());
263                assert_eq!(tensor.data[2], f64::NEG_INFINITY);
264            }
265            other => panic!("expected tensor result, got {other:?}"),
266        }
267    }
268
269    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
270    #[test]
271    fn str2double_cell_array_invalid_element_errors() {
272        let cell = CellArray::new(vec![Value::Num(5.0)], 1, 1).unwrap();
273        let err = error_message(str2double_builtin(Value::Cell(cell)).unwrap_err());
274        assert!(
275            err.contains("str2double"),
276            "unexpected error message: {err}"
277        );
278    }
279
280    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
281    #[test]
282    fn str2double_supports_d_exponent() {
283        let result = str2double_builtin(Value::String("1.5D3".into())).expect("str2double");
284        match result {
285            Value::Num(v) => assert_eq!(v, 1500.0),
286            other => panic!("expected scalar result, got {other:?}"),
287        }
288    }
289
290    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
291    #[test]
292    fn str2double_recognises_infinity_forms() {
293        let array = StringArray::new(
294            vec!["Inf".into(), "-Infinity".into(), "+inf".into()],
295            vec![3, 1],
296        )
297        .unwrap();
298        let result = str2double_builtin(Value::StringArray(array)).expect("str2double");
299        match result {
300            Value::Tensor(tensor) => {
301                assert_eq!(tensor.data[0], f64::INFINITY);
302                assert_eq!(tensor.data[1], f64::NEG_INFINITY);
303                assert_eq!(tensor.data[2], f64::INFINITY);
304            }
305            other => panic!("expected tensor result, got {other:?}"),
306        }
307    }
308
309    #[test]
310    fn str2double_type_is_numeric_text_scalar_or_tensor() {
311        assert_eq!(
312            numeric_text_scalar_or_tensor_type(&[Type::String], &ResolveContext::new(Vec::new())),
313            Type::Num
314        );
315    }
316}