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.

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

Faster primitive arrays encoding into row format #5858

Merged
merged 3 commits into from
Jun 26, 2024
Merged

Faster primitive arrays encoding into row format #5858

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
519413a
skip iterator removed from primitive encoding
korowa Jun 8, 2024
99e1503
special cases for not-null primitives encoding
korowa Jun 9, 2024
d681c3e
faster iterators for nullable columns
korowa Jun 18, 2024
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
95 changes: 89 additions & 6 deletions arrow-row/src/fixed.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,8 @@ use crate::null_sentinel;
use arrow_array::builder::BufferBuilder;
use arrow_array::{ArrowPrimitiveType, BooleanArray, FixedSizeBinaryArray};
use arrow_buffer::{
bit_util, i256, ArrowNativeType, Buffer, IntervalDayTime, IntervalMonthDayNano, MutableBuffer,
bit_util, i256, ArrowNativeType, BooleanBuffer, Buffer, IntervalDayTime, IntervalMonthDayNano,
MutableBuffer, NullBuffer,
};
use arrow_data::{ArrayData, ArrayDataBuilder};
use arrow_schema::{DataType, SortOptions};
Expand Down Expand Up @@ -216,18 +217,75 @@ where
///
/// - 1 byte `0` if null or `1` if valid
/// - bytes of [`FixedLengthEncoding`]
pub fn encode<T: FixedLengthEncoding, I: IntoIterator<Item = Option<T>>>(
pub fn encode<T: FixedLengthEncoding>(
data: &mut [u8],
offsets: &mut [usize],
i: I,
values: &[T],
nulls: &NullBuffer,
opts: SortOptions,
) {
for (offset, maybe_val) in offsets.iter_mut().skip(1).zip(i) {
for (value_idx, is_valid) in nulls.iter().enumerate() {
let offset = &mut offsets[value_idx + 1];
let end_offset = *offset + T::ENCODED_LEN;
if let Some(val) = maybe_val {
if is_valid {
let to_write = &mut data[*offset..end_offset];
to_write[0] = 1;
let mut encoded = values[value_idx].encode();
if opts.descending {
// Flip bits to reverse order
encoded.as_mut().iter_mut().for_each(|v| *v = !*v)
}
to_write[1..].copy_from_slice(encoded.as_ref())
} else {
data[*offset] = null_sentinel(opts);
}
*offset = end_offset;
}
}

/// Encoding for non-nullable primitive arrays.
/// Iterates directly over the `values`, and skips NULLs-checking.
pub fn encode_not_null<T: FixedLengthEncoding>(
data: &mut [u8],
offsets: &mut [usize],
values: &[T],
opts: SortOptions,
) {
for (value_idx, val) in values.iter().enumerate() {
let offset = &mut offsets[value_idx + 1];
let end_offset = *offset + T::ENCODED_LEN;

let to_write = &mut data[*offset..end_offset];
to_write[0] = 1;
let mut encoded = val.encode();
if opts.descending {
Copy link
Contributor

Choose a reason for hiding this comment

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

Perhaps LLVM is smart enough, but it might be interesting to see if pulling this out of the loop improves performance

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 guess it'll be a follow up (need a bit more time with it) -- as for now after quick check, for non-nullable columns, encoded with descending order

    if opts.descending {
        // Flip bits to reverse order for encoded data skipping validity byte
        for offset_idx in 1..offsets.len() {
            let end_offset = offsets[offset_idx];
            let buf = &mut data[end_offset-T::ENCODED_LEN+1..end_offset];
            buf.iter_mut().for_each(|v| *v = !*v);
        }
    }

after "main" loop shows [9.7890 µs 9.9628 µs 10.196 µs] vs [12.391 µs 12.448 µs 12.545 µs], so there may be an improvement, but it's a bit unclear for nullable columns (don't have any solution with meaningful improvement yet).

// Flip bits to reverse order
encoded.as_mut().iter_mut().for_each(|v| *v = !*v)
}
to_write[1..].copy_from_slice(encoded.as_ref());

*offset = end_offset;
}
}

/// Boolean values are encoded as
///
/// - 1 byte `0` if null or `1` if valid
/// - bytes of [`FixedLengthEncoding`]
pub fn encode_boolean(
data: &mut [u8],
offsets: &mut [usize],
values: &BooleanBuffer,
nulls: &NullBuffer,
opts: SortOptions,
) {
for (idx, is_valid) in nulls.iter().enumerate() {
let offset = &mut offsets[idx + 1];
let end_offset = *offset + bool::ENCODED_LEN;
if is_valid {
let to_write = &mut data[*offset..end_offset];
to_write[0] = 1;
let mut encoded = val.encode();
let mut encoded = values.value(idx).encode();
if opts.descending {
// Flip bits to reverse order
encoded.as_mut().iter_mut().for_each(|v| *v = !*v)
Expand All @@ -240,6 +298,31 @@ pub fn encode<T: FixedLengthEncoding, I: IntoIterator<Item = Option<T>>>(
}
}

/// Encoding for non-nullable boolean arrays.
/// Iterates directly over `values`, and skips NULLs-checking.
pub fn encode_boolean_not_null(
data: &mut [u8],
offsets: &mut [usize],
values: &BooleanBuffer,
opts: SortOptions,
) {
for (value_idx, val) in values.iter().enumerate() {
let offset = &mut offsets[value_idx + 1];
let end_offset = *offset + bool::ENCODED_LEN;

let to_write = &mut data[*offset..end_offset];
to_write[0] = 1;
let mut encoded = val.encode();
if opts.descending {
// Flip bits to reverse order
encoded.as_mut().iter_mut().for_each(|v| *v = !*v)
}
to_write[1..].copy_from_slice(encoded.as_ref());

*offset = end_offset;
}
}

pub fn encode_fixed_size_binary(
data: &mut [u8],
offsets: &mut [usize],
Expand Down
16 changes: 14 additions & 2 deletions arrow-row/src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1146,9 +1146,21 @@ fn encode_column(
match encoder {
Encoder::Stateless => {
downcast_primitive_array! {
column => fixed::encode(data, offsets, column, opts),
column => {
if let Some(nulls) = column.nulls().filter(|n| n.null_count() > 0){
fixed::encode(data, offsets, column.values(), nulls, opts)
} else {
fixed::encode_not_null(data, offsets, column.values(), opts)
}
}
DataType::Null => {}
DataType::Boolean => fixed::encode(data, offsets, column.as_boolean(), opts),
DataType::Boolean => {
if let Some(nulls) = column.nulls().filter(|n| n.null_count() > 0){
fixed::encode_boolean(data, offsets, column.as_boolean().values(), nulls, opts)
} else {
fixed::encode_boolean_not_null(data, offsets, column.as_boolean().values(), opts)
}
}
DataType::Binary => {
variable::encode(data, offsets, as_generic_binary_array::<i32>(column).iter(), opts)
}
Expand Down
Loading