use runmat_accelerate_api::GpuTensorHandle;
use runmat_builtins::{CharArray, ComplexTensor, Tensor, Value};
use runmat_macros::runtime_builtin;
use crate::builtins::common::spec::{
BroadcastSemantics, BuiltinFusionSpec, BuiltinGpuSpec, ConstantStrategy, FusionError,
FusionExprContext, FusionKernelTemplate, GpuOpKind, ProviderHook, ReductionNaN,
ResidencyPolicy, ScalarType, ShapeRequirements,
};
use crate::builtins::common::{gpu_helpers, tensor};
use crate::builtins::math::type_resolvers::numeric_unary_type;
use crate::{build_runtime_error, BuiltinResult, RuntimeError};
#[runmat_macros::register_gpu_spec(builtin_path = "crate::builtins::math::rounding::round")]
pub const GPU_SPEC: BuiltinGpuSpec = BuiltinGpuSpec {
name: "round",
op_kind: GpuOpKind::Elementwise,
supported_precisions: &[ScalarType::F32, ScalarType::F64],
broadcast: BroadcastSemantics::Matlab,
provider_hooks: &[ProviderHook::Unary { name: "unary_round" }],
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 execute round directly on the device; digit-aware rounding currently gathers to the host.",
};
#[runmat_macros::register_fusion_spec(builtin_path = "crate::builtins::math::rounding::round")]
pub const FUSION_SPEC: BuiltinFusionSpec = BuiltinFusionSpec {
name: "round",
shape: ShapeRequirements::BroadcastCompatible,
constant_strategy: ConstantStrategy::InlineLiteral,
elementwise: Some(FusionKernelTemplate {
scalar_precisions: &[ScalarType::F32, ScalarType::F64],
wgsl_body: |ctx: &FusionExprContext| {
let input = ctx.inputs.first().ok_or(FusionError::MissingInput(0))?;
Ok(format!("round({input})"))
},
}),
reduction: None,
emits_nan: false,
notes: "Fusion planner emits WGSL `round` calls; providers can substitute custom kernels.",
};
const BUILTIN_NAME: &str = "round";
fn builtin_error(message: impl Into<String>) -> RuntimeError {
build_runtime_error(message)
.with_builtin(BUILTIN_NAME)
.build()
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum RoundStrategy {
Integer,
Decimals(i32),
Significant(i32),
}
impl RoundStrategy {
fn requires_host(&self) -> bool {
!matches!(self, RoundStrategy::Integer)
}
}
#[runtime_builtin(
name = "round",
category = "math/rounding",
summary = "Round values to the nearest integers, decimal places, or significant digits.",
keywords = "round,rounding,significant,decimals,gpu",
accel = "unary",
type_resolver(numeric_unary_type),
builtin_path = "crate::builtins::math::rounding::round"
)]
async fn round_builtin(value: Value, rest: Vec<Value>) -> BuiltinResult<Value> {
let strategy = parse_arguments(&rest)?;
match value {
Value::GpuTensor(handle) => round_gpu(handle, strategy).await,
Value::Complex(re, im) => Ok(Value::Complex(
round_scalar(re, strategy),
round_scalar(im, strategy),
)),
Value::ComplexTensor(ct) => round_complex_tensor(ct, strategy),
Value::CharArray(ca) => round_char_array(ca, strategy),
Value::LogicalArray(logical) => {
let tensor = tensor::logical_to_tensor(&logical).map_err(|err| builtin_error(err))?;
Ok(round_tensor(tensor, strategy).map(tensor::tensor_into_value)?)
}
Value::String(_) | Value::StringArray(_) => {
Err(builtin_error("round: expected numeric or logical input"))
}
other => round_numeric(other, strategy),
}
}
async fn round_gpu(handle: GpuTensorHandle, strategy: RoundStrategy) -> BuiltinResult<Value> {
if !strategy.requires_host() {
if let Some(provider) = runmat_accelerate_api::provider_for_handle(&handle) {
if let Ok(out) = provider.unary_round(&handle).await {
return Ok(Value::GpuTensor(out));
}
}
}
let tensor = gpu_helpers::gather_tensor_async(&handle).await?;
round_tensor(tensor, strategy).map(tensor::tensor_into_value)
}
fn round_numeric(value: Value, strategy: RoundStrategy) -> BuiltinResult<Value> {
match value {
Value::Num(n) => Ok(Value::Num(round_scalar(n, strategy))),
Value::Int(i) => Ok(Value::Num(round_scalar(i.to_f64(), strategy))),
Value::Bool(b) => Ok(Value::Num(round_scalar(
if b { 1.0 } else { 0.0 },
strategy,
))),
Value::Tensor(t) => round_tensor(t, strategy).map(tensor::tensor_into_value),
other => {
let tensor =
tensor::value_into_tensor_for("round", other).map_err(|err| builtin_error(err))?;
Ok(round_tensor(tensor, strategy).map(tensor::tensor_into_value)?)
}
}
}
fn round_tensor(mut tensor: Tensor, strategy: RoundStrategy) -> BuiltinResult<Tensor> {
for value in &mut tensor.data {
*value = round_scalar(*value, strategy);
}
Ok(tensor)
}
fn round_complex_tensor(ct: ComplexTensor, strategy: RoundStrategy) -> BuiltinResult<Value> {
let data = ct
.data
.iter()
.map(|&(re, im)| (round_scalar(re, strategy), round_scalar(im, strategy)))
.collect::<Vec<_>>();
let tensor = ComplexTensor::new(data, ct.shape.clone())
.map_err(|e| builtin_error(format!("round: {e}")))?;
Ok(Value::ComplexTensor(tensor))
}
fn round_char_array(ca: CharArray, strategy: RoundStrategy) -> BuiltinResult<Value> {
let mut data = Vec::with_capacity(ca.data.len());
for ch in ca.data {
data.push(round_scalar(ch as u32 as f64, strategy));
}
let tensor = Tensor::new(data, vec![ca.rows, ca.cols])
.map_err(|e| builtin_error(format!("round: {e}")))?;
Ok(Value::Tensor(tensor))
}
fn round_scalar(value: f64, strategy: RoundStrategy) -> f64 {
if !value.is_finite() {
return value;
}
match strategy {
RoundStrategy::Integer => value.round(),
RoundStrategy::Decimals(n) => round_with_decimals(value, n),
RoundStrategy::Significant(n) => round_with_significant(value, n),
}
}
fn round_with_decimals(value: f64, digits: i32) -> f64 {
if digits == 0 {
return value.round();
}
let factor = 10f64.powi(digits);
if !factor.is_finite() || factor == 0.0 {
return value;
}
(value * factor).round() / factor
}
fn round_with_significant(value: f64, digits: i32) -> f64 {
if value == 0.0 {
return 0.0;
}
let abs_val = value.abs();
let order = abs_val.log10().floor();
let scale_power = digits - 1 - order as i32;
let scale = 10f64.powi(scale_power);
if !scale.is_finite() || scale == 0.0 {
return value;
}
(value * scale).round() / scale
}
fn parse_arguments(args: &[Value]) -> BuiltinResult<RoundStrategy> {
match args.len() {
0 => Ok(RoundStrategy::Integer),
1 => {
let digits = parse_digits(&args[0])?;
Ok(RoundStrategy::Decimals(digits))
}
2 => {
let digits = parse_digits(&args[0])?;
let mode = parse_mode(&args[1])?;
match mode {
RoundMode::Decimals => Ok(RoundStrategy::Decimals(digits)),
RoundMode::Significant => {
if digits <= 0 {
return Err(builtin_error(
"round: N must be a positive integer for 'significant' rounding",
));
}
Ok(RoundStrategy::Significant(digits))
}
}
}
_ => Err(builtin_error("round: too many input arguments")),
}
}
fn parse_digits(value: &Value) -> BuiltinResult<i32> {
let err = || builtin_error("round: N must be an integer scalar");
let raw = match value {
Value::Int(i) => i.to_i64(),
Value::Num(n) => {
if !n.is_finite() {
return Err(err());
}
let rounded = n.round();
if (rounded - n).abs() > f64::EPSILON {
return Err(err());
}
rounded as i64
}
Value::Bool(b) => {
if *b {
1
} else {
0
}
}
other => {
return Err(builtin_error(format!(
"round: N must be numeric, got {:?}",
other
)))
}
};
if raw > i32::MAX as i64 || raw < i32::MIN as i64 {
return Err(builtin_error("round: integer overflow in N"));
}
Ok(raw as i32)
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum RoundMode {
Decimals,
Significant,
}
fn parse_mode(value: &Value) -> BuiltinResult<RoundMode> {
let Some(text) = tensor::value_to_string(value) else {
return Err(builtin_error(
"round: mode must be a character vector or string scalar",
));
};
let lowered = text.trim().to_ascii_lowercase();
match lowered.as_str() {
"significant" => Ok(RoundMode::Significant),
"decimal" | "decimals" => Ok(RoundMode::Decimals),
other => Err(builtin_error(format!(
"round: unknown rounding mode '{other}'"
))),
}
}
#[cfg(test)]
pub(crate) mod tests {
use super::*;
use crate::builtins::common::test_support;
use futures::executor::block_on;
use runmat_builtins::{IntValue, ResolveContext, Tensor, Type};
fn round_builtin(value: Value, rest: Vec<Value>) -> BuiltinResult<Value> {
block_on(super::round_builtin(value, rest))
}
fn assert_error_contains(err: crate::RuntimeError, needle: &str) {
assert!(
err.message().contains(needle),
"unexpected error: {}",
err.message()
);
}
#[test]
fn round_type_preserves_tensor_shape() {
let out = numeric_unary_type(
&[Type::Tensor {
shape: Some(vec![Some(2), Some(3)]),
}],
&ResolveContext::new(Vec::new()),
);
assert_eq!(
out,
Type::Tensor {
shape: Some(vec![Some(2), Some(3)])
}
);
}
#[test]
fn round_type_scalar_tensor_returns_num() {
let out = numeric_unary_type(
&[Type::Tensor {
shape: Some(vec![Some(1), Some(1)]),
}],
&ResolveContext::new(Vec::new()),
);
assert_eq!(out, Type::Num);
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_scalar_defaults() {
let result = round_builtin(Value::Num(1.7), Vec::new()).expect("round");
match result {
Value::Num(v) => assert_eq!(v, 2.0),
other => panic!("expected scalar result, got {other:?}"),
}
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_scalar_negative_half() {
let result = round_builtin(Value::Num(-2.5), Vec::new()).expect("round");
match result {
Value::Num(v) => assert_eq!(v, -3.0),
other => panic!("expected scalar result, got {other:?}"),
}
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_tensor_decimals() {
let tensor = Tensor::new(vec![1.2345, 2.499, 3.5001], vec![3, 1]).unwrap();
let result = round_builtin(Value::Tensor(tensor), vec![Value::Int(IntValue::I32(2))])
.expect("round");
match result {
Value::Tensor(t) => {
assert_eq!(t.shape, vec![3, 1]);
let expected = [1.23, 2.5, 3.5];
for (a, b) in t.data.iter().zip(expected.iter()) {
assert!((a - b).abs() < 1e-12, "expected {b}, got {a}");
}
}
other => panic!("expected tensor result, got {other:?}"),
}
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_tensor_negative_decimals() {
let tensor = Tensor::new(vec![123.0, 149.9, 150.0], vec![3, 1]).unwrap();
let result = round_builtin(Value::Tensor(tensor), vec![Value::Int(IntValue::I32(-2))])
.expect("round");
match result {
Value::Tensor(t) => {
assert_eq!(t.data, vec![100.0, 100.0, 200.0]);
}
other => panic!("expected tensor result, got {other:?}"),
}
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_scalar_significant() {
let result = round_builtin(
Value::Num(0.0012345),
vec![Value::Int(IntValue::I32(3)), Value::from("significant")],
)
.expect("round");
match result {
Value::Num(v) => assert!((v - 0.00123).abs() < 1e-12),
other => panic!("expected scalar result, got {other:?}"),
}
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_complex_value() {
let result = round_builtin(Value::Complex(1.2, -3.6), Vec::new()).expect("round");
match result {
Value::Complex(re, im) => {
assert_eq!(re, 1.0);
assert_eq!(im, -4.0);
}
other => panic!("expected complex result, got {other:?}"),
}
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_invalid_mode_errors() {
let err = round_builtin(
Value::Num(1.0),
vec![Value::Int(IntValue::I32(2)), Value::from("approx")],
)
.unwrap_err();
assert_error_contains(err, "unknown rounding mode");
}
#[cfg_attr(target_arch = "wasm32", wasm_bindgen_test::wasm_bindgen_test)]
#[test]
fn round_gpu_provider_roundtrip() {
test_support::with_test_provider(|provider| {
let tensor = Tensor::new(vec![-2.5, -0.2, 0.5, 1.8], vec![4, 1]).unwrap();
let view = runmat_accelerate_api::HostTensorView {
data: &tensor.data,
shape: &tensor.shape,
};
let handle = provider.upload(&view).expect("upload");
let result = round_builtin(Value::GpuTensor(handle), Vec::new()).expect("round");
let gathered = test_support::gather(result).expect("gather");
assert_eq!(gathered.shape, vec![4, 1]);
assert_eq!(gathered.data, vec![-3.0, 0.0, 1.0, 2.0]);
});
}
}