Skip to content

perf: Optimize floor and ceil scalar performance #19752

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
comphead merged 3 commits into from
Jan 12, 2026
Merged

perf: Optimize floor and ceil scalar performance #19752

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b6090b8
perf: Optimize floor and ceil scalar performance
kumarUjjawal Jan 11, 2026
e7c4f72
fix clippy issue
kumarUjjawal Jan 11, 2026
dcd7444
track if input are scaler
kumarUjjawal Jan 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions datafusion/functions/Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -315,3 +315,8 @@ required-features = ["unicode_expressions"]
harness = false
name = "factorial"
required-features = ["math_expressions"]

[[bench]]
harness = false
name = "floor_ceil"
required-features = ["math_expressions"]
135 changes: 135 additions & 0 deletions datafusion/functions/benches/floor_ceil.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,135 @@
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License. You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied. See the License for the
// specific language governing permissions and limitations
// under the License.

extern crate criterion;

use arrow::datatypes::{DataType, Field, Float64Type};
use arrow::util::bench_util::create_primitive_array;
use criterion::{Criterion, SamplingMode, criterion_group, criterion_main};
use datafusion_common::ScalarValue;
use datafusion_common::config::ConfigOptions;
use datafusion_expr::{ColumnarValue, ScalarFunctionArgs};
use datafusion_functions::math::{ceil, floor};
use std::hint::black_box;
use std::sync::Arc;
use std::time::Duration;

fn criterion_benchmark(c: &mut Criterion) {
let floor_fn = floor();
let ceil_fn = ceil();
let config_options = Arc::new(ConfigOptions::default());

for size in [1024, 4096, 8192] {
let mut group = c.benchmark_group(format!("floor_ceil size={size}"));
group.sampling_mode(SamplingMode::Flat);
group.sample_size(10);
group.measurement_time(Duration::from_secs(10));

// Float64 array benchmark
let f64_array = Arc::new(create_primitive_array::<Float64Type>(size, 0.1));
let batch_len = f64_array.len();
let f64_args = vec![ColumnarValue::Array(f64_array)];

group.bench_function("floor_f64_array", |b| {
b.iter(|| {
let args_cloned = f64_args.clone();
black_box(
floor_fn
.invoke_with_args(ScalarFunctionArgs {
args: args_cloned,
arg_fields: vec![
Field::new("a", DataType::Float64, true).into(),
],
number_rows: batch_len,
return_field: Field::new("f", DataType::Float64, true).into(),
config_options: Arc::clone(&config_options),
})
.unwrap(),
)
})
});

group.bench_function("ceil_f64_array", |b| {
b.iter(|| {
let args_cloned = f64_args.clone();
black_box(
ceil_fn
.invoke_with_args(ScalarFunctionArgs {
args: args_cloned,
arg_fields: vec![
Field::new("a", DataType::Float64, true).into(),
],
number_rows: batch_len,
return_field: Field::new("f", DataType::Float64, true).into(),
config_options: Arc::clone(&config_options),
})
.unwrap(),
)
})
});

// Scalar benchmark (the optimization we added)
let scalar_args = vec![ColumnarValue::Scalar(ScalarValue::Float64(Some(
std::f64::consts::PI,
)))];

group.bench_function("floor_f64_scalar", |b| {
b.iter(|| {
let args_cloned = scalar_args.clone();
black_box(
floor_fn
.invoke_with_args(ScalarFunctionArgs {
args: args_cloned,
arg_fields: vec![
Field::new("a", DataType::Float64, false).into(),
],
number_rows: 1,
return_field: Field::new("f", DataType::Float64, false)
.into(),
config_options: Arc::clone(&config_options),
})
.unwrap(),
)
})
});

group.bench_function("ceil_f64_scalar", |b| {
b.iter(|| {
let args_cloned = scalar_args.clone();
black_box(
ceil_fn
.invoke_with_args(ScalarFunctionArgs {
args: args_cloned,
arg_fields: vec![
Field::new("a", DataType::Float64, false).into(),
],
number_rows: 1,
return_field: Field::new("f", DataType::Float64, false)
.into(),
config_options: Arc::clone(&config_options),
})
.unwrap(),
)
})
});

group.finish();
}
}

criterion_group!(benches, criterion_benchmark);
criterion_main!(benches);
46 changes: 39 additions & 7 deletions datafusion/functions/src/math/ceil.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,8 +95,35 @@ impl ScalarUDFImpl for CeilFunc {
}

fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> {
let args = ColumnarValue::values_to_arrays(&args.args)?;
let value = &args[0];
let arg = &args.args[0];

// Scalar fast path for float types - avoid array conversion overhead entirely
if let ColumnarValue::Scalar(scalar) = arg {
match scalar {
ScalarValue::Float64(v) => {
return Ok(ColumnarValue::Scalar(ScalarValue::Float64(
v.map(f64::ceil),
)));
}
ScalarValue::Float32(v) => {
return Ok(ColumnarValue::Scalar(ScalarValue::Float32(
v.map(f32::ceil),
)));
}
ScalarValue::Null => {
return Ok(ColumnarValue::Scalar(ScalarValue::Float64(None)));
}
// For decimals: convert to array of size 1, process, then extract scalar
// This ensures we don't expand the array while reusing overflow validation
_ => {}
}
}

// Track if input was a scalar to convert back at the end
let is_scalar = matches!(arg, ColumnarValue::Scalar(_));

// Array path (also handles decimal scalars converted to size-1 arrays)
let value = arg.to_array(args.number_rows)?;

let result: ArrayRef = match value.data_type() {
DataType::Float64 => Arc::new(
Expand All @@ -114,7 +141,7 @@ impl ScalarUDFImpl for CeilFunc {
}
DataType::Decimal32(precision, scale) => {
apply_decimal_op::<Decimal32Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -123,7 +150,7 @@ impl ScalarUDFImpl for CeilFunc {
}
DataType::Decimal64(precision, scale) => {
apply_decimal_op::<Decimal64Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -132,7 +159,7 @@ impl ScalarUDFImpl for CeilFunc {
}
DataType::Decimal128(precision, scale) => {
apply_decimal_op::<Decimal128Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -141,7 +168,7 @@ impl ScalarUDFImpl for CeilFunc {
}
DataType::Decimal256(precision, scale) => {
apply_decimal_op::<Decimal256Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -156,7 +183,12 @@ impl ScalarUDFImpl for CeilFunc {
}
};

Ok(ColumnarValue::Array(result))
// If input was a scalar, convert result back to scalar
if is_scalar {
ScalarValue::try_from_array(&result, 0).map(ColumnarValue::Scalar)
} else {
Ok(ColumnarValue::Array(result))
}
}

fn output_ordering(&self, input: &[ExprProperties]) -> Result<SortProperties> {
Expand Down
46 changes: 39 additions & 7 deletions datafusion/functions/src/math/floor.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,8 +95,35 @@ impl ScalarUDFImpl for FloorFunc {
}

fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> {
let args = ColumnarValue::values_to_arrays(&args.args)?;
let value = &args[0];
let arg = &args.args[0];

// Scalar fast path for float types - avoid array conversion overhead entirely
if let ColumnarValue::Scalar(scalar) = arg {
match scalar {
ScalarValue::Float64(v) => {
return Ok(ColumnarValue::Scalar(ScalarValue::Float64(
v.map(f64::floor),
)));
}
ScalarValue::Float32(v) => {
return Ok(ColumnarValue::Scalar(ScalarValue::Float32(
v.map(f32::floor),
)));
}
ScalarValue::Null => {
return Ok(ColumnarValue::Scalar(ScalarValue::Float64(None)));
}
// For decimals: convert to array of size 1, process, then extract scalar
// This ensures we don't expand the array while reusing overflow validation
_ => {}
}
}

// Track if input was a scalar to convert back at the end
let is_scalar = matches!(arg, ColumnarValue::Scalar(_));

// Array path (also handles decimal scalars converted to size-1 arrays)
let value = arg.to_array(args.number_rows)?;

let result: ArrayRef = match value.data_type() {
DataType::Float64 => Arc::new(
Expand All @@ -114,7 +141,7 @@ impl ScalarUDFImpl for FloorFunc {
}
DataType::Decimal32(precision, scale) => {
apply_decimal_op::<Decimal32Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -123,7 +150,7 @@ impl ScalarUDFImpl for FloorFunc {
}
DataType::Decimal64(precision, scale) => {
apply_decimal_op::<Decimal64Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -132,7 +159,7 @@ impl ScalarUDFImpl for FloorFunc {
}
DataType::Decimal128(precision, scale) => {
apply_decimal_op::<Decimal128Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -141,7 +168,7 @@ impl ScalarUDFImpl for FloorFunc {
}
DataType::Decimal256(precision, scale) => {
apply_decimal_op::<Decimal256Type, _>(
value,
&value,
*precision,
*scale,
self.name(),
Expand All @@ -156,7 +183,12 @@ impl ScalarUDFImpl for FloorFunc {
}
};

Ok(ColumnarValue::Array(result))
// If input was a scalar, convert result back to scalar
if is_scalar {
ScalarValue::try_from_array(&result, 0).map(ColumnarValue::Scalar)
} else {
Ok(ColumnarValue::Array(result))
}
}

fn output_ordering(&self, input: &[ExprProperties]) -> Result<SortProperties> {
Expand Down