use runmat_builtins::{Tensor, Value};
use runmat_macros::runtime_builtin;
use crate::build_runtime_error;
use crate::builtins::common::random;
use crate::builtins::common::random_args::extract_dims;
use crate::builtins::common::tensor;
use runmat_builtins::ResolveContext;
use runmat_builtins::Type;
fn builtin_error(message: impl Into<String>) -> crate::RuntimeError {
build_runtime_error(message).with_builtin("normrnd").build()
}
fn normrnd_type(args: &[Type], _ctx: &ResolveContext) -> Type {
if args.len() <= 2 {
Type::Num
} else {
Type::Unknown
}
}
#[runtime_builtin(
name = "normrnd",
category = "stats/random",
summary = "Normally-distributed random numbers with mean mu and standard deviation sigma.",
keywords = "normrnd,normal,gaussian,random,distribution,statistics",
type_resolver(normrnd_type),
builtin_path = "crate::builtins::stats::random::normrnd"
)]
async fn normrnd_builtin(args: Vec<Value>) -> crate::BuiltinResult<Value> {
let (mu, sigma, shape) = parse_args(args).await?;
if sigma < 0.0 {
return Err(builtin_error("normrnd: sigma must be non-negative"));
}
if let Some(value) = try_gpu_normrnd(mu, sigma, &shape)? {
return Ok(value);
}
let len = tensor::element_count(&shape);
let data = random::generate_normal_scaled(mu, sigma, len, "normrnd")?;
let t = Tensor::new(data, shape).map_err(|e| builtin_error(format!("normrnd: {e}")))?;
Ok(tensor::tensor_into_value(t))
}
async fn parse_args(args: Vec<Value>) -> crate::BuiltinResult<(f64, f64, Vec<usize>)> {
if args.len() < 2 {
return Err(builtin_error(
"normrnd: requires at least two arguments (mu, sigma)",
));
}
let mu = scalar_f64(&args[0])?;
let sigma = scalar_f64(&args[1])?;
let shape = parse_shape_args(&args[2..]).await?;
Ok((mu, sigma, shape))
}
fn scalar_f64(value: &Value) -> crate::BuiltinResult<f64> {
match value {
Value::Num(v) => Ok(*v),
Value::Int(i) => Ok(i.to_f64()),
Value::Bool(b) => Ok(if *b { 1.0 } else { 0.0 }),
other => Err(builtin_error(format!(
"normrnd: expected scalar parameter, got {other:?}"
))),
}
}
async fn parse_shape_args(rest: &[Value]) -> crate::BuiltinResult<Vec<usize>> {
if rest.is_empty() {
return Ok(vec![1, 1]);
}
let mut dims: Vec<usize> = Vec::new();
for arg in rest {
match extract_dims(arg, "normrnd").await? {
Some(d) => dims.extend(d),
None => {
return Err(builtin_error(format!(
"normrnd: invalid size argument: {arg:?}"
)))
}
}
}
Ok(normalize_dims(dims))
}
fn normalize_dims(dims: Vec<usize>) -> Vec<usize> {
if dims.is_empty() {
vec![0, 0]
} else if dims.len() == 1 {
vec![dims[0], dims[0]]
} else {
dims
}
}
fn try_gpu_normrnd(mu: f64, sigma: f64, shape: &[usize]) -> crate::BuiltinResult<Option<Value>> {
let Some(provider) = runmat_accelerate_api::provider() else {
return Ok(None);
};
if provider.precision() != runmat_accelerate_api::ProviderPrecision::F64 {
return Ok(None);
}
match provider.random_normrnd(mu, sigma, shape) {
Ok(handle) => {
let len = tensor::element_count(shape);
let uniform_count = len.saturating_add(1) / 2 * 2;
random::skip_uniform(uniform_count, "normrnd")?;
Ok(Some(Value::GpuTensor(handle)))
}
Err(_) => Ok(None),
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::builtins::common::random;
use futures::executor::block_on;
fn reset() {
runmat_accelerate_api::clear_provider();
random::reset_rng();
}
#[test]
fn normrnd_scalar_deterministic() {
let _guard = random::test_lock().lock().unwrap();
reset();
let result =
block_on(normrnd_builtin(vec![Value::Num(0.0), Value::Num(1.0)])).expect("normrnd");
let expected = random::expected_normal_scaled_sequence(0.0, 1.0, 1)[0];
match result {
Value::Num(v) => assert!((v - expected).abs() < 1e-12),
other => panic!("expected scalar, got {other:?}"),
}
}
#[test]
fn normrnd_matrix_dims() {
let _guard = random::test_lock().lock().unwrap();
reset();
let args = vec![
Value::Num(5.0),
Value::Num(2.0),
Value::Num(3.0),
Value::Num(4.0),
];
let result = block_on(normrnd_builtin(args)).expect("normrnd");
match result {
Value::Tensor(t) => assert_eq!(t.shape, vec![3, 4]),
other => panic!("expected tensor, got {other:?}"),
}
}
#[test]
fn normrnd_size_vec() {
let _guard = random::test_lock().lock().unwrap();
reset();
let size = Tensor::new(vec![3.0, 4.0], vec![1, 2]).unwrap();
let args = vec![Value::Num(0.0), Value::Num(1.0), Value::Tensor(size)];
let result = block_on(normrnd_builtin(args)).expect("normrnd");
match result {
Value::Tensor(t) => assert_eq!(t.shape, vec![3, 4]),
other => panic!("expected tensor, got {other:?}"),
}
}
#[test]
fn normrnd_rejects_negative_sigma() {
let args = vec![Value::Num(0.0), Value::Num(-1.0)];
assert!(block_on(normrnd_builtin(args)).is_err());
}
#[test]
fn normrnd_distribution_mean_and_std() {
let _guard = random::test_lock().lock().unwrap();
reset();
let mu = 5.0_f64;
let sigma = 2.0_f64;
let n = 50_000_usize;
let args = vec![
Value::Num(mu),
Value::Num(sigma),
Value::Num(n as f64),
Value::Num(1.0),
];
let result = block_on(normrnd_builtin(args)).expect("normrnd");
let data = match result {
Value::Tensor(t) => t.data,
other => panic!("expected tensor, got {other:?}"),
};
let mean = data.iter().sum::<f64>() / data.len() as f64;
let variance = data.iter().map(|&x| (x - mean).powi(2)).sum::<f64>() / data.len() as f64;
let std_dev = variance.sqrt();
assert!(
(mean - mu).abs() / sigma.max(1.0) < 0.05,
"sample mean {mean:.4} not within 5% tolerance of mu={mu}"
);
assert!(
(std_dev - sigma).abs() / sigma < 0.05,
"sample std {std_dev:.4} not within 5% of sigma={sigma}"
);
}
}