nevi-me commented on a change in pull request #8116:
URL: https://github.com/apache/arrow/pull/8116#discussion_r483954060
##########
File path: rust/datafusion/src/physical_plan/math_expressions.rs
##########
@@ -19,23 +19,25 @@
use crate::error::{ExecutionError, Result};
-use arrow::array::{Array, ArrayRef, Float32Array, Float64Array,
Float64Builder};
+use arrow::array::{Array, ArrayRef, Float32Array, Float64Array};
use arrow::datatypes::DataType;
use std::sync::Arc;
macro_rules! compute_op {
($ARRAY:expr, $FUNC:ident, $TYPE:ident) => {{
- let mut builder = Float64Builder::new($ARRAY.len());
- for i in 0..$ARRAY.len() {
- if $ARRAY.is_null(i) {
- builder.append_null()?;
- } else {
- builder.append_value($ARRAY.value(i).$FUNC() as f64)?;
- }
- }
- Ok(Arc::new(builder.finish()))
+ let result: Float64Array = (0..$ARRAY.len())
+ .map(|i| {
Review comment:
You could likely improve performance further by computing without
checking nullity of array, and then building the array using the input bitmask.
I'm away from home, so I don't have a PC to check with; but look at the
pattern that most of the compute functions follow.
Removing the if/else allows the compiler to vectorise the loop.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
