Skip to content

Commit

Permalink
perf: improve binview filter (pola-rs#13878)
Browse files Browse the repository at this point in the history
  • Loading branch information
ritchie46 authored and r-brink committed Jan 24, 2024
1 parent 1b70f68 commit c6c2ae6
Show file tree
Hide file tree
Showing 2 changed files with 28 additions and 6 deletions.
22 changes: 22 additions & 0 deletions crates/polars-arrow/src/array/growable/binview.rs
Original file line number Diff line number Diff line change
Expand Up @@ -114,6 +114,28 @@ impl<'a, T: ViewType + ?Sized> GrowableBinaryViewArray<'a, T> {
}
}));
}

#[inline]
pub(crate) unsafe fn extend_unchecked_no_buffers(
&mut self,
index: usize,
start: usize,
len: usize,
) {
let array = *self.arrays.get_unchecked(index);

extend_validity(&mut self.validity, array, start, len);

let range = start..start + len;

self.views
.extend(array.views().get_unchecked(range).iter().map(|view| {
let len = (*view as u32) as usize;
self.total_bytes_len += len;

*view
}))
}
}

impl<'a, T: ViewType + ?Sized> Growable<'a> for GrowableBinaryViewArray<'a, T> {
Expand Down
12 changes: 6 additions & 6 deletions crates/polars-arrow/src/compute/filter.rs
Original file line number Diff line number Diff line change
Expand Up @@ -274,14 +274,14 @@ pub fn filter(array: &dyn Array, filter: &BooleanArray) -> PolarsResult<Box<dyn
}),
BinaryView => {
let iter = SlicesIterator::new(filter.values());
let mut mutable = growable::GrowableBinaryViewArray::new(
vec![array.as_any().downcast_ref::<BinaryViewArray>().unwrap()],
false,
iter.slots(),
);
let array = array.as_any().downcast_ref::<BinaryViewArray>().unwrap();
let mut mutable =
growable::GrowableBinaryViewArray::new(vec![array], false, iter.slots());
unsafe {
iter.for_each(|(start, len)| mutable.extend_unchecked(0, start, len));
// We don't have to correct buffers as there is only one array.
iter.for_each(|(start, len)| mutable.extend_unchecked_no_buffers(0, start, len));
}

Ok(mutable.as_box())
},
// Should go via BinaryView
Expand Down

0 comments on commit c6c2ae6

Please sign in to comment.