Skip to content
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.

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

Add support StringView / BinaryView in interleave kernel #6779

Open
wants to merge 7 commits into
base: main
Choose a base branch
from
Open
Changes from 6 commits
Commits
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
158 changes: 158 additions & 0 deletions arrow-select/src/interleave.rs
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,9 @@ use arrow_array::types::*;
use arrow_array::*;
use arrow_buffer::{ArrowNativeType, MutableBuffer, NullBuffer, NullBufferBuilder, OffsetBuffer};
use arrow_data::transform::MutableArrayData;
use arrow_data::ByteView;
use arrow_schema::{ArrowError, DataType};
use std::collections::HashMap;
use std::sync::Arc;

macro_rules! primitive_helper {
Expand Down Expand Up @@ -97,6 +99,8 @@ pub fn interleave(
DataType::LargeUtf8 => interleave_bytes::<LargeUtf8Type>(values, indices),
DataType::Binary => interleave_bytes::<BinaryType>(values, indices),
DataType::LargeBinary => interleave_bytes::<LargeBinaryType>(values, indices),
DataType::BinaryView => interleave_views::<BinaryViewType>(values, indices),
DataType::Utf8View => interleave_views::<StringViewType>(values, indices),
DataType::Dictionary(k, _) => downcast_integer! {
k.as_ref() => (dict_helper, values, indices),
_ => unreachable!("illegal dictionary key type {k}")
Expand Down Expand Up @@ -231,6 +235,40 @@ fn interleave_dictionaries<K: ArrowDictionaryKeyType>(
Ok(Arc::new(array))
}

fn interleave_views<T: ByteViewType>(
values: &[&dyn Array],
indices: &[(usize, usize)],
) -> Result<ArrayRef, ArrowError> {
let interleaved = Interleave::<'_, GenericByteViewArray<T>>::new(values, indices);
let mut views_builder = BufferBuilder::new(indices.len());
let mut buffers = Vec::new();

let mut buffer_lookup = HashMap::new();
Copy link
Contributor

@tustvold tustvold Nov 26, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This misunderstands #6780, the issue isn't not skipping buffers that aren't referenced, it is that the arrays being interleaved may contain the same actual buffers, e.g. sourced from the same parquet dictionary page. This needs to deduplicate based on the underlying Buffer pointers.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see, I have filed #6808 to deduplicate the buffers while building the interleaved array in the fallback implementation.

I am not sure if that would completely remove the need for this PR though, it should guarantee that no duplicate buffers should exist on the interleaved array, but the fallback implementation would still have 2 buffers in the test below, because those two buffers are unique. It feels like this PR and #6808 are complementary

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you also please add some comments about what the buffer lookup represents? I think it is

// (input array_index, input buffer_index) --> output array buffer_index

for (array_idx, value_idx) in indices {
let array = interleaved.arrays[*array_idx];
let raw_view = array.views().get(*value_idx).unwrap();
let view = ByteView::from(*raw_view);
onursatici marked this conversation as resolved.
Show resolved Hide resolved

if view.length <= 12 {
views_builder.append(*raw_view);
continue;
}
// value is big enough to be in a variadic buffer
let new_buffer_idx: &mut u32 = buffer_lookup
.entry((*array_idx, view.buffer_index))
.or_insert_with(|| {
buffers.push(array.data_buffers()[view.buffer_index as usize].clone());
(buffers.len() - 1) as u32
});
views_builder.append(view.with_buffer_index(*new_buffer_idx).into());
}

let array = unsafe {
GenericByteViewArray::<T>::new_unchecked(views_builder.into(), buffers, interleaved.nulls)
};
Ok(Arc::new(array))
}

/// Fallback implementation of interleave using [`MutableArrayData`]
fn interleave_fallback(
values: &[&dyn Array],
Expand Down Expand Up @@ -461,4 +499,124 @@ mod tests {
DictionaryArray::<Int32Type>::from_iter(vec![Some("0"), Some("1"), Some("2"), None]);
assert_eq!(array.as_ref(), &expected)
}

#[test]
fn test_interleave_views() {
let values = StringArray::from_iter_values([
"hello",
"world_long_string_not_inlined",
"foo",
"bar",
"baz",
]);
let view_a = StringViewArray::from(&values);

let values = StringArray::from_iter_values([
"test",
"data",
"more_long_string_not_inlined",
"views",
"here",
]);
let view_b = StringViewArray::from(&values);

let indices = &[
(0, 2), // "foo"
(1, 0), // "test"
(0, 4), // "baz"
(1, 3), // "views"
(0, 1), // "world_long_string_not_inlined"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

since the test only picks one long string, only one buffer is going to be copied (the other strings are inlined)

To provoke the issue described in the ticket, I think you need to interleave that same long string from the two different buffers. Something like

        let indices = &[
            (0, 1), // "world_long_string_not_inlined" (view_a)
            (1, 2), // "world_long_string_not_inlined" (view_b)
            (0, 1), // "world_long_string_not_inlined" (view_a)
            (1, 2), // "world_long_string_not_inlined" (view_b)
            (0, 1), // "world_long_string_not_inlined" (view_a)
            (1, 2), // "world_long_string_not_inlined" (view_b)
            (0, 1), // "world_long_string_not_inlined" (view_a)
            (1, 2), // "world_long_string_not_inlined" (view_b)
            (0, 1), // "world_long_string_not_inlined" (view_a)
            (1, 2), // "world_long_string_not_inlined" (view_b)

And make sure the output view only has 2 buffers (the one from view_a and the one from view_b)

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the issue with the fallback implementation was that it did copy all buffers from all inputs, even if no views on the output referred them. I believe this test shows that issue well, when we run the fallback implementation, it does end up having two buffers although only one is referenced in the output array. When run with the new implementation specific to views, it does correctly prune the unused buffer from the second input array.

];

// Test specialized implementation
let values = interleave(&[&view_a, &view_b], indices).unwrap();
let result = values.as_string_view();
assert_eq!(result.data_buffers().len(), 1);

// Test fallback implementation
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Since interleave_fallback isn't pub I don't think we should be testing it directly. Intead you should arrange the paramters to call it directly

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I removed the comment, it was a bit misleading. I was not trying to test the fallback implementation here, but was trying to use it to show that it does produce the same results with the new implementation, but with more buffers.

With the new implementation I don't think it is straightforward to call the fallback implementation using the public methods now

let fallback = interleave_fallback(&[&view_a, &view_b], indices).unwrap();
let fallback_result = fallback.as_string_view();
// note that fallback_result has 2 buffers, but only one long enough string to warrant a buffer
assert_eq!(fallback_result.data_buffers().len(), 2);

// Convert to strings for easier assertion
let collected: Vec<_> = result.iter().map(|x| x.map(|s| s.to_string())).collect();

let fallback_collected: Vec<_> = fallback_result
.iter()
.map(|x| x.map(|s| s.to_string()))
.collect();

assert_eq!(&collected, &fallback_collected);

assert_eq!(
&collected,
&[
Some("foo".to_string()),
Some("test".to_string()),
Some("baz".to_string()),
Some("views".to_string()),
Some("world_long_string_not_inlined".to_string()),
]
);
}

#[test]
onursatici marked this conversation as resolved.
Show resolved Hide resolved
fn test_interleave_views_with_nulls() {
let values = StringArray::from_iter([
Some("hello"),
None,
Some("foo_long_string_not_inlined"),
Some("bar"),
None,
]);
let view_a = StringViewArray::from(&values);

let values = StringArray::from_iter([
Some("test"),
Some("data_long_string_not_inlined"),
None,
None,
Some("here"),
]);
let view_b = StringViewArray::from(&values);

let indices = &[
(0, 1), // null
(1, 2), // null
(0, 2), // "foo_long_string_not_inlined"
(1, 3), // null
(0, 4), // null
];

// Test specialized implementation
let values = interleave(&[&view_a, &view_b], indices).unwrap();
let result = values.as_string_view();
assert_eq!(result.data_buffers().len(), 1);

// Test fallback implementation
let fallback = interleave_fallback(&[&view_a, &view_b], indices).unwrap();
let fallback_result = fallback.as_string_view();

// Convert to strings for easier assertion
let collected: Vec<_> = result.iter().map(|x| x.map(|s| s.to_string())).collect();

let fallback_collected: Vec<_> = fallback_result
.iter()
.map(|x| x.map(|s| s.to_string()))
.collect();

assert_eq!(&collected, &fallback_collected);

assert_eq!(
&collected,
&[
None,
None,
Some("foo_long_string_not_inlined".to_string()),
None,
None,
]
);
}
}
Loading