runmat-runtime 0.4.1

Core runtime for RunMat with builtins, BLAS/LAPACK integration, and execution APIs
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
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
//! MATLAB-compatible `cond` builtin with GPU-aware fallbacks.

use nalgebra::{linalg::SVD, DMatrix};
use num_complex::Complex64;
use runmat_accelerate_api::{GpuTensorHandle, HostTensorView, ProviderCondNorm};
use runmat_builtins::{ComplexTensor, Tensor, Value};
use runmat_macros::runtime_builtin;

use crate::builtins::common::gpu_helpers;
use crate::builtins::common::linalg::matrix_dimensions_for;
use crate::builtins::common::spec::{
    BroadcastSemantics, BuiltinFusionSpec, BuiltinGpuSpec, ConstantStrategy, GpuOpKind,
    ProviderHook, ReductionNaN, ResidencyPolicy, ScalarType, ShapeRequirements,
};
use crate::builtins::common::tensor;
use crate::builtins::math::linalg::type_resolvers::numeric_scalar_type;
use crate::{build_runtime_error, BuiltinResult, RuntimeError};

const NAME: &str = "cond";

#[runmat_macros::register_gpu_spec(builtin_path = "crate::builtins::math::linalg::solve::cond")]
pub const GPU_SPEC: BuiltinGpuSpec = BuiltinGpuSpec {
    name: NAME,
    op_kind: GpuOpKind::Custom("cond"),
    supported_precisions: &[ScalarType::F32, ScalarType::F64],
    broadcast: BroadcastSemantics::None,
    provider_hooks: &[ProviderHook::Custom("cond")],
    constant_strategy: ConstantStrategy::InlineLiteral,
    residency: ResidencyPolicy::NewHandle,
    nan_mode: ReductionNaN::Include,
    two_pass_threshold: None,
    workgroup_size: None,
    accepts_nan_mode: false,
    notes: "Providers may expose a direct condition-number kernel; the reference backends gather to the host, evaluate the shared implementation, and upload the scalar result.",
};

fn builtin_error(message: impl Into<String>) -> RuntimeError {
    build_runtime_error(message).with_builtin(NAME).build()
}

fn map_control_flow(err: RuntimeError) -> RuntimeError {
    if err.message() == "interaction pending..." {
        return build_runtime_error("interaction pending...")
            .with_builtin(NAME)
            .build();
    }
    let mut builder = build_runtime_error(err.message()).with_builtin(NAME);
    if let Some(identifier) = err.identifier() {
        builder = builder.with_identifier(identifier.to_string());
    }
    if let Some(task_id) = err.context.task_id.clone() {
        builder = builder.with_task_id(task_id);
    }
    if !err.context.call_stack.is_empty() {
        builder = builder.with_call_stack(err.context.call_stack.clone());
    }
    if let Some(phase) = err.context.phase.clone() {
        builder = builder.with_phase(phase);
    }
    builder.with_source(err).build()
}

#[runmat_macros::register_fusion_spec(builtin_path = "crate::builtins::math::linalg::solve::cond")]
pub const FUSION_SPEC: BuiltinFusionSpec = BuiltinFusionSpec {
    name: NAME,
    shape: ShapeRequirements::Any,
    constant_strategy: ConstantStrategy::InlineLiteral,
    elementwise: None,
    reduction: None,
    emits_nan: false,
    notes: "Not fusible; cond consumes an entire matrix and returns a scalar diagnostic.",
};

#[runtime_builtin(
    name = "cond",
    category = "math/linalg/solve",
    summary = "Compute the matrix condition number with MATLAB-compatible norms.",
    keywords = "cond,condition number,norm,gpu",
    accel = "cond",
    type_resolver(numeric_scalar_type),
    builtin_path = "crate::builtins::math::linalg::solve::cond"
)]
async fn cond_builtin(value: Value, rest: Vec<Value>) -> BuiltinResult<Value> {
    let norm = parse_norm_argument(&rest)?;
    let result = match value {
        Value::GpuTensor(handle) => return cond_gpu(handle, norm).await,
        Value::ComplexTensor(matrix) => cond_complex_tensor_builtin(&matrix, norm)?,
        Value::Complex(re, im) => {
            let tensor = ComplexTensor::new(vec![(re, im)], vec![1, 1]).map_err(builtin_error)?;
            cond_complex_tensor_builtin(&tensor, norm)?
        }
        other => {
            let tensor = tensor::value_into_tensor_for(NAME, other).map_err(builtin_error)?;
            cond_real_tensor_builtin(&tensor, norm)?
        }
    };
    Ok(Value::Num(result))
}

async fn cond_gpu(handle: GpuTensorHandle, norm: CondNorm) -> BuiltinResult<Value> {
    let maybe_provider = runmat_accelerate_api::provider();

    if let Some(provider) = maybe_provider {
        if let Some(value) = cond_gpu_via_provider(provider, &handle, norm).await? {
            return Ok(value);
        }
    }

    let gathered = gpu_helpers::gather_value_async(&Value::GpuTensor(handle.clone()))
        .await
        .map_err(map_control_flow)?;

    let cond_value = match gathered {
        Value::Tensor(tensor) => cond_real_tensor_builtin(&tensor, norm)?,
        Value::ComplexTensor(tensor) => cond_complex_tensor_builtin(&tensor, norm)?,
        Value::Num(n) => {
            if n == 0.0 {
                f64::INFINITY
            } else {
                1.0
            }
        }
        Value::Complex(re, im) => {
            if re == 0.0 && im == 0.0 {
                f64::INFINITY
            } else {
                1.0
            }
        }
        other => {
            let tensor = tensor::value_into_tensor_for(NAME, other).map_err(builtin_error)?;
            cond_real_tensor_builtin(&tensor, norm)?
        }
    };

    if let Some(provider) = maybe_provider {
        match upload_scalar(provider, cond_value) {
            Ok(uploaded) => return Ok(Value::GpuTensor(uploaded)),
            Err(err) => {
                if err.message() == "interaction pending..." {
                    return Err(build_runtime_error("interaction pending...")
                        .with_builtin(NAME)
                        .build());
                }
            }
        }
    }

    Ok(Value::Num(cond_value))
}

async fn cond_gpu_via_provider(
    provider: &'static dyn runmat_accelerate_api::AccelProvider,
    handle: &GpuTensorHandle,
    norm: CondNorm,
) -> BuiltinResult<Option<Value>> {
    let provider_norm = ProviderCondNorm::from(norm);
    match provider.cond(handle, provider_norm).await {
        Ok(result) => Ok(Some(Value::GpuTensor(result))),
        Err(_err) => Ok(None),
    }
}

fn cond_real_tensor_builtin(matrix: &Tensor, norm: CondNorm) -> BuiltinResult<f64> {
    cond_real_tensor(matrix, norm)
}

fn cond_complex_tensor_builtin(matrix: &ComplexTensor, norm: CondNorm) -> BuiltinResult<f64> {
    cond_complex_tensor(matrix, norm)
}

fn cond_real_tensor(matrix: &Tensor, norm: CondNorm) -> BuiltinResult<f64> {
    let (rows, cols) = matrix_dimensions_for(NAME, &matrix.shape).map_err(builtin_error)?;
    if rows == 0 || cols == 0 {
        return Ok(0.0);
    }
    if matrix.data.len() == 1 {
        return Ok(if matrix.data[0] == 0.0 {
            f64::INFINITY
        } else {
            1.0
        });
    }

    match norm {
        CondNorm::Two => cond_two_norm_real(matrix, rows, cols),
        _ => {
            if rows != cols {
                return Err(builtin_error(format!(
                    "{NAME}: matrix must be square for the requested norm."
                )));
            }
            cond_inverse_based_real(matrix, rows, norm)
        }
    }
}

fn cond_complex_tensor(matrix: &ComplexTensor, norm: CondNorm) -> BuiltinResult<f64> {
    let (rows, cols) = matrix_dimensions_for(NAME, &matrix.shape).map_err(builtin_error)?;
    if rows == 0 || cols == 0 {
        return Ok(0.0);
    }
    if matrix.data.len() == 1 {
        let (re, im) = matrix.data[0];
        let magnitude = re.hypot(im);
        return Ok(if magnitude == 0.0 { f64::INFINITY } else { 1.0 });
    }

    match norm {
        CondNorm::Two => cond_two_norm_complex(matrix, rows, cols),
        _ => {
            if rows != cols {
                return Err(builtin_error(format!(
                    "{NAME}: matrix must be square for the requested norm."
                )));
            }
            cond_inverse_based_complex(matrix, rows, norm)
        }
    }
}

fn cond_two_norm_real(matrix: &Tensor, rows: usize, cols: usize) -> BuiltinResult<f64> {
    let a = DMatrix::from_column_slice(rows, cols, &matrix.data);
    let svd = SVD::new(a, false, false);
    Ok(singular_value_cond(svd.singular_values.as_slice()))
}

fn cond_two_norm_complex(matrix: &ComplexTensor, rows: usize, cols: usize) -> BuiltinResult<f64> {
    let data: Vec<Complex64> = matrix
        .data
        .iter()
        .map(|&(re, im)| Complex64::new(re, im))
        .collect();
    let a = DMatrix::from_column_slice(rows, cols, &data);
    let svd = SVD::new(a, false, false);
    Ok(singular_value_cond(svd.singular_values.as_slice()))
}

fn cond_inverse_based_real(matrix: &Tensor, order: usize, norm: CondNorm) -> BuiltinResult<f64> {
    let dm = DMatrix::from_column_slice(order, order, &matrix.data);
    if let Some(inv) = dm.try_inverse() {
        let norm_a = matrix_norm_real(matrix.data.as_slice(), order, order, norm);
        let norm_inv = matrix_norm_real(inv.as_slice(), order, order, norm);
        let cond = norm_a * norm_inv;
        if cond.is_finite() {
            Ok(cond)
        } else {
            Ok(f64::INFINITY)
        }
    } else {
        Ok(f64::INFINITY)
    }
}

fn cond_inverse_based_complex(
    matrix: &ComplexTensor,
    order: usize,
    norm: CondNorm,
) -> BuiltinResult<f64> {
    let data: Vec<Complex64> = matrix
        .data
        .iter()
        .map(|&(re, im)| Complex64::new(re, im))
        .collect();
    let dm = DMatrix::from_column_slice(order, order, &data);
    if let Some(inv) = dm.try_inverse() {
        let norm_a = matrix_norm_complex(&data, order, order, norm);
        let norm_inv = matrix_norm_complex(inv.as_slice(), order, order, norm);
        let cond = norm_a * norm_inv;
        if cond.is_finite() {
            Ok(cond)
        } else {
            Ok(f64::INFINITY)
        }
    } else {
        Ok(f64::INFINITY)
    }
}

fn matrix_norm_real(data: &[f64], rows: usize, cols: usize, norm: CondNorm) -> f64 {
    match norm {
        CondNorm::One => {
            let mut max_sum: f64 = 0.0;
            for c in 0..cols {
                let mut sum = 0.0;
                for r in 0..rows {
                    sum += data[r + c * rows].abs();
                }
                max_sum = max_sum.max(sum);
            }
            max_sum
        }
        CondNorm::Inf => {
            let mut max_sum: f64 = 0.0;
            for r in 0..rows {
                let mut sum = 0.0;
                for c in 0..cols {
                    sum += data[r + c * rows].abs();
                }
                max_sum = max_sum.max(sum);
            }
            max_sum
        }
        CondNorm::Fro => {
            let sum_sq: f64 = data.iter().map(|v| v * v).sum();
            sum_sq.sqrt()
        }
        CondNorm::Two => unreachable!("matrix_norm_real not used for 2-norm"),
    }
}

fn matrix_norm_complex(data: &[Complex64], rows: usize, cols: usize, norm: CondNorm) -> f64 {
    match norm {
        CondNorm::One => {
            let mut max_sum: f64 = 0.0;
            for c in 0..cols {
                let mut sum = 0.0;
                for r in 0..rows {
                    sum += data[r + c * rows].norm();
                }
                max_sum = max_sum.max(sum);
            }
            max_sum
        }
        CondNorm::Inf => {
            let mut max_sum: f64 = 0.0;
            for r in 0..rows {
                let mut sum = 0.0;
                for c in 0..cols {
                    sum += data[r + c * rows].norm();
                }
                max_sum = max_sum.max(sum);
            }
            max_sum
        }
        CondNorm::Fro => {
            let sum_sq: f64 = data.iter().map(|v| v.norm_sqr()).sum();
            sum_sq.sqrt()
        }
        CondNorm::Two => unreachable!("matrix_norm_complex not used for 2-norm"),
    }
}

fn singular_value_cond(singular_values: &[f64]) -> f64 {
    if singular_values.is_empty() {
        return 0.0;
    }
    let mut min_sv = f64::INFINITY;
    let mut max_sv = 0.0_f64;
    for &sv in singular_values {
        let abs = sv.abs();
        if !abs.is_finite() {
            return f64::INFINITY;
        }
        min_sv = min_sv.min(abs);
        max_sv = max_sv.max(abs);
    }
    if min_sv == 0.0 {
        f64::INFINITY
    } else {
        max_sv / min_sv
    }
}

fn parse_norm_argument(args: &[Value]) -> BuiltinResult<CondNorm> {
    match args.len() {
        0 => Ok(CondNorm::Two),
        1 => parse_norm_value(&args[0]),
        _ => Err(builtin_error(format!("{NAME}: too many input arguments"))),
    }
}

fn parse_norm_value(value: &Value) -> BuiltinResult<CondNorm> {
    if let Some(text) = tensor::value_to_string(value) {
        return parse_norm_string(&text);
    }
    match value {
        Value::Num(n) => parse_norm_numeric(*n),
        Value::Int(i) => parse_norm_numeric(i.to_f64()),
        Value::Tensor(t) if tensor::is_scalar_tensor(t) => parse_norm_numeric(t.data[0]),
        Value::Bool(b) => {
            if *b {
                Ok(CondNorm::One)
            } else {
                Err(builtin_error(format!(
                    "{NAME}: norm must be 1, 2, Inf, or 'fro'"
                )))
            }
        }
        Value::LogicalArray(logical) if logical.len() == 1 => {
            if logical.data[0] != 0 {
                Ok(CondNorm::One)
            } else {
                Err(builtin_error(format!(
                    "{NAME}: norm must be 1, 2, Inf, or 'fro'"
                )))
            }
        }
        _ => Err(builtin_error(format!(
            "{NAME}: norm must be 1, 2, Inf, or 'fro'"
        ))),
    }
}

fn parse_norm_numeric(raw: f64) -> BuiltinResult<CondNorm> {
    if raw == 1.0 {
        Ok(CondNorm::One)
    } else if raw == 2.0 {
        Ok(CondNorm::Two)
    } else if raw.is_infinite() && raw.is_sign_positive() {
        Ok(CondNorm::Inf)
    } else {
        Err(builtin_error(format!(
            "{NAME}: norm must be 1, 2, Inf, or 'fro'"
        )))
    }
}

fn parse_norm_string(text: &str) -> BuiltinResult<CondNorm> {
    let lowered = text.trim().to_ascii_lowercase();
    match lowered.as_str() {
        "2" | "two" => Ok(CondNorm::Two),
        "1" | "one" => Ok(CondNorm::One),
        "inf" | "infinity" => Ok(CondNorm::Inf),
        "fro" | "frobenius" => Ok(CondNorm::Fro),
        _ => Err(builtin_error(format!("{NAME}: unrecognised norm '{text}'"))),
    }
}

fn upload_scalar(
    provider: &'static dyn runmat_accelerate_api::AccelProvider,
    value: f64,
) -> BuiltinResult<GpuTensorHandle> {
    let data = [value];
    let shape = [1usize, 1usize];
    let view = HostTensorView {
        data: &data,
        shape: &shape,
    };
    provider
        .upload(&view)
        .map_err(|e| builtin_error(format!("{NAME}: {e}")))
}

/// Helper for provider backends that reuse the host implementation.
pub fn cond_host_real_for_provider(matrix: &Tensor, norm: ProviderCondNorm) -> BuiltinResult<f64> {
    cond_real_tensor(matrix, CondNorm::from(norm))
}

#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum CondNorm {
    Two,
    One,
    Inf,
    Fro,
}

impl From<CondNorm> for ProviderCondNorm {
    fn from(value: CondNorm) -> Self {
        match value {
            CondNorm::Two => ProviderCondNorm::Two,
            CondNorm::One => ProviderCondNorm::One,
            CondNorm::Inf => ProviderCondNorm::Inf,
            CondNorm::Fro => ProviderCondNorm::Fro,
        }
    }
}

impl From<ProviderCondNorm> for CondNorm {
    fn from(value: ProviderCondNorm) -> Self {
        match value {
            ProviderCondNorm::Two => CondNorm::Two,
            ProviderCondNorm::One => CondNorm::One,
            ProviderCondNorm::Inf => CondNorm::Inf,
            ProviderCondNorm::Fro => CondNorm::Fro,
        }
    }
}

#[cfg(test)]
pub(crate) mod tests {
    use super::*;
    use crate::builtins::common::test_support;
    use futures::executor::block_on;
    use runmat_builtins::{IntValue, ResolveContext, Type};
    fn unwrap_error(err: crate::RuntimeError) -> crate::RuntimeError {
        err
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_identity_is_one() {
        let tensor = Tensor::new(vec![1.0, 0.0, 0.0, 1.0], vec![2, 2]).unwrap();
        let result = cond_builtin(Value::Tensor(tensor), Vec::new()).expect("cond");
        match result {
            Value::Num(value) => assert!((value - 1.0).abs() < 1e-12),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[test]
    fn cond_type_returns_scalar() {
        let out = numeric_scalar_type(
            &[Type::Tensor {
                shape: Some(vec![Some(2), Some(2)]),
            }],
            &ResolveContext::new(Vec::new()),
        );
        assert_eq!(out, Type::Num);
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_zero_is_infinite() {
        let tensor = Tensor::new(vec![0.0], vec![1, 1]).unwrap();
        let result = cond_builtin(Value::Tensor(tensor), Vec::new()).expect("cond");
        match result {
            Value::Num(value) => assert!(value.is_infinite()),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_rectangular_two_norm() {
        let tensor = Tensor::new(vec![1.0, 0.0, 0.0, 1.0, 1.0, 1.0], vec![3, 2]).unwrap();
        let result = cond_builtin(Value::Tensor(tensor), Vec::new()).expect("cond");
        match result {
            Value::Num(value) => assert!((value - 2.414213562).abs() < 1e-9),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_one_norm_matches_manual() {
        let tensor = Tensor::new(vec![4.0, 2.0, -1.0, 3.0], vec![2, 2]).unwrap();
        let result =
            cond_builtin(Value::Tensor(tensor), vec![Value::Int(IntValue::I32(1))]).expect("cond");
        match result {
            Value::Num(value) => assert!((value - 2.142_857_142_857_143).abs() < 1e-9),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_infinity_norm() {
        let tensor = Tensor::new(vec![4.0, 2.0, -1.0, 3.0], vec![2, 2]).unwrap();
        let result = cond_builtin(Value::Tensor(tensor), vec![Value::from("inf")]).expect("cond");
        match result {
            Value::Num(value) => assert!((value - 2.142_857_142_857_143).abs() < 1e-12),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_frobenius_norm() {
        let tensor = Tensor::new(vec![5.0, 0.0, 0.0, 2.0], vec![2, 2]).unwrap();
        let result = cond_builtin(Value::Tensor(tensor), vec![Value::from("fro")]).expect("cond");
        match result {
            Value::Num(value) => assert!((value - 2.9).abs() < 1e-12),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_complex_matrix_supported() {
        let data = vec![(1.0, 2.0), (0.0, 0.0), (0.0, 3.0), (2.0, -1.0)];
        let tensor = ComplexTensor::new(data, vec![2, 2]).unwrap();
        let result = cond_builtin(Value::ComplexTensor(tensor), Vec::new()).expect("cond");
        match result {
            Value::Num(value) => assert!(value.is_finite() && value >= 1.0),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_rejects_non_square_for_other_norms() {
        let tensor = Tensor::new(vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0], vec![2, 3]).unwrap();
        let err = unwrap_error(
            cond_builtin(Value::Tensor(tensor), vec![Value::from("inf")]).unwrap_err(),
        );
        assert_eq!(
            err.message(),
            "cond: matrix must be square for the requested norm."
        );
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_empty_returns_zero() {
        let tensor = Tensor::new(vec![], vec![0, 0]).unwrap();
        let result = cond_builtin(Value::Tensor(tensor), Vec::new()).expect("cond");
        match result {
            Value::Num(value) => assert_eq!(value, 0.0),
            other => panic!("expected scalar result, got {other:?}"),
        }
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    fn cond_gpu_round_trip_matches_cpu() {
        test_support::with_test_provider(|provider| {
            let tensor = Tensor::new(vec![4.0, 1.0, 2.0, 3.0], vec![2, 2]).unwrap();
            let view = HostTensorView {
                data: &tensor.data,
                shape: &tensor.shape,
            };
            let handle = provider.upload(&view).expect("upload");
            let gpu_value = cond_builtin(Value::GpuTensor(handle), Vec::new()).expect("cond");
            let gathered = test_support::gather(gpu_value).expect("gather");
            assert_eq!(gathered.shape, vec![1, 1]);
            let expected = cond_builtin(Value::Tensor(tensor.clone()), Vec::new())
                .map(|v| match v {
                    Value::Num(n) => n,
                    _ => unreachable!(),
                })
                .expect("cpu cond");
            assert!((gathered.data[0] - expected).abs() < 1e-12);
        });
    }

    #[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
    #[test]
    #[cfg(feature = "wgpu")]
    fn cond_wgpu_matches_cpu() {
        let _ = runmat_accelerate::backend::wgpu::provider::register_wgpu_provider(
            runmat_accelerate::backend::wgpu::provider::WgpuProviderOptions::default(),
        );
        let tensor = Tensor::new(vec![2.0, 0.0, 0.0, 0.2], vec![2, 2]).unwrap();
        let cpu = cond_builtin(Value::Tensor(tensor.clone()), Vec::new()).expect("cpu");
        let cpu_value = match cpu {
            Value::Num(n) => n,
            _ => unreachable!(),
        };
        let provider = runmat_accelerate_api::provider().expect("wgpu provider");
        let view = HostTensorView {
            data: &tensor.data,
            shape: &tensor.shape,
        };
        let handle = provider.upload(&view).expect("upload");
        let gpu = cond_builtin(Value::GpuTensor(handle), Vec::new()).expect("cond");
        let gathered = test_support::gather(gpu).expect("gather");
        assert!((gathered.data[0] - cpu_value).abs() < 1e-9);
    }

    fn cond_builtin(value: Value, rest: Vec<Value>) -> BuiltinResult<Value> {
        block_on(super::cond_builtin(value, rest))
    }
}