Apache DataFusion SQL Query Engine
perf: Fix perf regression in `lcm` (#26023)
Neil Conway6e319bdddd9b
diff --git a/datafusion/functions/benches/math_expressions/gcd.rs b/datafusion/functions/benches/math_expressions/gcd.rsindex e702a6a6b0..af60a52a76 100644--- a/datafusion/functions/benches/math_expressions/gcd.rs+++ b/datafusion/functions/benches/math_expressions/gcd.rs@@ -29,8 +29,7 @@ use rand::prelude::*;use std::hint::black_box;use std::sync::Arc;-fn generate_i64_array(n_rows: usize) -> ArrayRef {- let mut rng = StdRng::seed_from_u64(0);+fn generate_i64_array(rng: &mut StdRng, n_rows: usize) -> ArrayRef {let values = (0..n_rows).map(|_| rng.random_range(0..1000)).collect::<Vec<_>>();@@ -39,8 +38,9 @@ fn generate_i64_array(n_rows: usize) -> ArrayRef {fn criterion_benchmark(c: &mut Criterion) {let n_rows = 100000;- let array_a = ColumnarValue::Array(generate_i64_array(n_rows));- let array_b = ColumnarValue::Array(generate_i64_array(n_rows));+ let mut rng = StdRng::seed_from_u64(0);+ let array_a = ColumnarValue::Array(generate_i64_array(&mut rng, n_rows));+ let array_b = ColumnarValue::Array(generate_i64_array(&mut rng, n_rows));let udf = gcd();let config_options = Arc::new(ConfigOptions::default());diff --git a/datafusion/functions/benches/math_expressions/lcm.rs b/datafusion/functions/benches/math_expressions/lcm.rsindex 33f40f16a3..7e40fed78d 100644--- a/datafusion/functions/benches/math_expressions/lcm.rs+++ b/datafusion/functions/benches/math_expressions/lcm.rs@@ -28,8 +28,7 @@ use rand::prelude::*;use std::hint::black_box;use std::sync::Arc;-fn generate_i64_array(n_rows: usize) -> ArrayRef {- let mut rng = StdRng::seed_from_u64(0);+fn generate_i64_array(rng: &mut StdRng, n_rows: usize) -> ArrayRef {let values = (0..n_rows).map(|_| rng.random_range(0..1000)).collect::<Vec<_>>();@@ -38,8 +37,9 @@ fn generate_i64_array(n_rows: usize) -> ArrayRef {fn criterion_benchmark(c: &mut Criterion) {let n_rows = 100000;- let array_a = ColumnarValue::Array(generate_i64_array(n_rows));- let array_b = ColumnarValue::Array(generate_i64_array(n_rows));+ let mut rng = StdRng::seed_from_u64(0);+ let array_a = ColumnarValue::Array(generate_i64_array(&mut rng, n_rows));+ let array_b = ColumnarValue::Array(generate_i64_array(&mut rng, n_rows));let udf = lcm();let config_options = Arc::new(ConfigOptions::default());diff --git a/datafusion/functions/src/math/common.rs b/datafusion/functions/src/math/common.rsindex 9bb6f6fe1e..e9f7c08235 100644--- a/datafusion/functions/src/math/common.rs+++ b/datafusion/functions/src/math/common.rs@@ -24,9 +24,9 @@ use std::ops::RemAssign;/// A gcd helper to compute GCD using Euclidean GCD algorithm/// on non-negative numbers (scalars and decimals)-fn gcd_helper<T>(a: T, b: T) -> Result<T, ArrowError>+fn gcd_helper<T>(a: T, b: T) -> Twhere- T: ArrowNativeTypeOp + RemAssign + CheckedNeg,+ T: ArrowNativeTypeOp + RemAssign,{debug_assert!(a >= T::ZERO);debug_assert!(b >= T::ZERO);@@ -37,7 +37,7 @@ whereb %= a;}- Ok(a)+ a}/// Computes gcd of two unsigned integers using Binary GCD algorithm@@ -84,7 +84,7 @@ where.ok_or_else(|| ArrowError::ComputeError("Signed integer overflow".into()))?};// Call with signed numbers- gcd_helper(a, b)+ Ok(gcd_helper(a, b))}/// Computes gcd of two signed integers@@ -123,7 +123,7 @@ where.ok_or_else(|| ArrowError::ComputeError("Signed integer overflow".into()))?};// Call with signed numbers- let gcd = gcd_helper(a, b)?;+ let gcd = gcd_helper(a, b);// gcd is not zero since both a and b are not zero, so the division is safe.(a / gcd).checked_mul(&b).ok_or_else(|| {ArrowError::ComputeError(format!("Signed integer overflow in LCM({x}, {y})"))@@ -140,7 +140,7 @@ pub(crate) fn lcm_signed_int(x: i64, y: i64) -> Result<i64, ArrowError> {let a = x.unsigned_abs();let b = y.unsigned_abs();- let gcd = gcd_helper::<u64>(a, b)?;+ let gcd = unsigned_gcd(a, b);// gcd is not zero since both a and b are not zero, so the division is safe.(a / gcd).checked_mul(b)