perf: Improve performance of split_part (#19570)

## Which issue does this PR close?

<!--
We generally require a GitHub issue to be filed for all bug fixes and
enhancements and this helps us generate change logs for our releases.
You can link an issue to this PR using the GitHub syntax. For example
`Closes #123` indicates that this PR will close issue #123.
-->

- Closes #.

## Rationale for this change

<!--
Why are you proposing this change? If this is already explained clearly
in the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand
your changes and offer better suggestions for fixes.
-->

I ran microbenchmarks comparing DataFusion with DuckDB for string
functions (see apache/datafusion-benchmarks#26)
and noticed that DF was very slow for `split_part`.

This PR fixes some obvious performance issues. Speedups are:

| Benchmark                         | Before | After | Speedup      |
|-----------------------------------|--------|-------|--------------|
| single_char_delim/pos_first       | 1.27ms | 140µs | 9.1x faster  |
| single_char_delim/pos_middle      | 1.39ms | 396µs | 3.5x faster  |
| single_char_delim/pos_last        | 1.47ms | 738µs | 2.0x faster  |
| single_char_delim/pos_negative    | 1.35ms | 148µs | 9.1x faster  |
| multi_char_delim/pos_first        | 1.22ms | 174µs | 7.0x faster  |
| multi_char_delim/pos_middle       | 1.22ms | 407µs | 3.0x faster  |
| string_view_single_char/pos_first | 1.42ms | 139µs | 10.2x faster |
| many_parts_20/pos_second          | 2.48ms | 201µs | 12.3x faster |
| long_strings_50_parts/pos_first   | 8.18ms | 178µs | 46x faster   |

## What changes are included in this PR?

<!--
There is no need to duplicate the description in the issue here but it
is sometimes worth providing a summary of the individual changes in this
PR.
-->

## Are these changes tested?

<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code

If tests are not included in your PR, please explain why (for example,
are they covered by existing tests)?
-->

## Are there any user-facing changes?

<!--
If there are user-facing changes then we may require documentation to be
updated before approving the PR.
-->

<!--
If there are any breaking changes to public APIs, please add the `api
change` label.
-->

---------

Co-authored-by: Martin Grigorov <martin-g@users.noreply.github.com>
Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>

Author: 3 people

datafusion/functions/Cargo.toml

@@ -296,6 +296,11 @@ harness = false
 name = "levenshtein"
 required-features = ["unicode_expressions"]
 
+[[bench]]
+harness = false
+name = "split_part"
+required-features = ["string_expressions"]
+
 [[bench]]
 harness = false
 name = "left"

datafusion/functions/benches/split_part.rs

@@ -0,0 +1,382 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements.  See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership.  The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License.  You may obtain a copy of the License at
+//
+//   http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.  See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+extern crate criterion;
+
+use arrow::array::{ArrayRef, Int64Array, StringArray, StringViewArray};
+use arrow::datatypes::{DataType, Field};
+use criterion::{BenchmarkId, Criterion, criterion_group, criterion_main};
+use datafusion_common::config::ConfigOptions;
+use datafusion_expr::{ColumnarValue, ScalarFunctionArgs};
+use datafusion_functions::string::split_part;
+use rand::distr::Alphanumeric;
+use rand::prelude::StdRng;
+use rand::{Rng, SeedableRng};
+use std::hint::black_box;
+use std::sync::Arc;
+
+const N_ROWS: usize = 8192;
+
+/// Generate test data for split_part benchmarks
+/// Creates strings with multiple parts separated by the delimiter
+fn gen_split_part_data(
+    n_rows: usize,
+    num_parts: usize, // number of parts in each string (separated by delimiter)
+    part_len: usize,  // length of each part
+    delimiter: &str,  // the delimiter to use
+    use_string_view: bool, // false -> StringArray, true -> StringViewArray
+) -> (ColumnarValue, ColumnarValue) {
+    let mut rng = StdRng::seed_from_u64(42);
+
+    let mut strings: Vec<String> = Vec::with_capacity(n_rows);
+    for _ in 0..n_rows {
+        let mut parts: Vec<String> = Vec::with_capacity(num_parts);
+        for _ in 0..num_parts {
+            let part: String = (&mut rng)
+                .sample_iter(&Alphanumeric)
+                .take(part_len)
+                .map(char::from)
+                .collect();
+            parts.push(part);
+        }
+        strings.push(parts.join(delimiter));
+    }
+
+    let delimiters: Vec<String> = vec![delimiter.to_string(); n_rows];
+
+    if use_string_view {
+        let string_array: StringViewArray = strings.into_iter().map(Some).collect();
+        let delimiter_array: StringViewArray = delimiters.into_iter().map(Some).collect();
+        (
+            ColumnarValue::Array(Arc::new(string_array) as ArrayRef),
+            ColumnarValue::Array(Arc::new(delimiter_array) as ArrayRef),
+        )
+    } else {
+        let string_array: StringArray = strings.into_iter().map(Some).collect();
+        let delimiter_array: StringArray = delimiters.into_iter().map(Some).collect();
+        (
+            ColumnarValue::Array(Arc::new(string_array) as ArrayRef),
+            ColumnarValue::Array(Arc::new(delimiter_array) as ArrayRef),
+        )
+    }
+}
+
+fn gen_positions(n_rows: usize, position: i64) -> ColumnarValue {
+    let positions: Vec<i64> = vec![position; n_rows];
+    ColumnarValue::Array(Arc::new(Int64Array::from(positions)) as ArrayRef)
+}
+
+fn criterion_benchmark(c: &mut Criterion) {
+    let split_part_func = split_part();
+    let config_options = Arc::new(ConfigOptions::default());
+
+    let mut group = c.benchmark_group("split_part");
+
+    // Test different scenarios
+    // Scenario 1: Single-char delimiter, first position (should be fastest with optimization)
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 10, 8, ".", false);
+        let positions = gen_positions(N_ROWS, 1);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(BenchmarkId::new("single_char_delim", "pos_first"), |b| {
+            b.iter(|| {
+                black_box(
+                    split_part_func
+                        .invoke_with_args(ScalarFunctionArgs {
+                            args: args.clone(),
+                            arg_fields: arg_fields.clone(),
+                            number_rows: N_ROWS,
+                            return_field: Arc::clone(&return_field),
+                            config_options: Arc::clone(&config_options),
+                        })
+                        .expect("split_part should work"),
+                )
+            })
+        });
+    }
+
+    // Scenario 2: Single-char delimiter, middle position
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 10, 8, ".", false);
+        let positions = gen_positions(N_ROWS, 5);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(BenchmarkId::new("single_char_delim", "pos_middle"), |b| {
+            b.iter(|| {
+                black_box(
+                    split_part_func
+                        .invoke_with_args(ScalarFunctionArgs {
+                            args: args.clone(),
+                            arg_fields: arg_fields.clone(),
+                            number_rows: N_ROWS,
+                            return_field: Arc::clone(&return_field),
+                            config_options: Arc::clone(&config_options),
+                        })
+                        .expect("split_part should work"),
+                )
+            })
+        });
+    }
+
+    // Scenario 3: Single-char delimiter, last position
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 10, 8, ".", false);
+        let positions = gen_positions(N_ROWS, 10);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(BenchmarkId::new("single_char_delim", "pos_last"), |b| {
+            b.iter(|| {
+                black_box(
+                    split_part_func
+                        .invoke_with_args(ScalarFunctionArgs {
+                            args: args.clone(),
+                            arg_fields: arg_fields.clone(),
+                            number_rows: N_ROWS,
+                            return_field: Arc::clone(&return_field),
+                            config_options: Arc::clone(&config_options),
+                        })
+                        .expect("split_part should work"),
+                )
+            })
+        });
+    }
+
+    // Scenario 4: Single-char delimiter, negative position (last element)
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 10, 8, ".", false);
+        let positions = gen_positions(N_ROWS, -1);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(
+            BenchmarkId::new("single_char_delim", "pos_negative"),
+            |b| {
+                b.iter(|| {
+                    black_box(
+                        split_part_func
+                            .invoke_with_args(ScalarFunctionArgs {
+                                args: args.clone(),
+                                arg_fields: arg_fields.clone(),
+                                number_rows: N_ROWS,
+                                return_field: Arc::clone(&return_field),
+                                config_options: Arc::clone(&config_options),
+                            })
+                            .expect("split_part should work"),
+                    )
+                })
+            },
+        );
+    }
+
+    // Scenario 5: Multi-char delimiter, first position
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 10, 8, "~@~", false);
+        let positions = gen_positions(N_ROWS, 1);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(BenchmarkId::new("multi_char_delim", "pos_first"), |b| {
+            b.iter(|| {
+                black_box(
+                    split_part_func
+                        .invoke_with_args(ScalarFunctionArgs {
+                            args: args.clone(),
+                            arg_fields: arg_fields.clone(),
+                            number_rows: N_ROWS,
+                            return_field: Arc::clone(&return_field),
+                            config_options: Arc::clone(&config_options),
+                        })
+                        .expect("split_part should work"),
+                )
+            })
+        });
+    }
+
+    // Scenario 6: Multi-char delimiter, middle position
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 10, 8, "~@~", false);
+        let positions = gen_positions(N_ROWS, 5);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(BenchmarkId::new("multi_char_delim", "pos_middle"), |b| {
+            b.iter(|| {
+                black_box(
+                    split_part_func
+                        .invoke_with_args(ScalarFunctionArgs {
+                            args: args.clone(),
+                            arg_fields: arg_fields.clone(),
+                            number_rows: N_ROWS,
+                            return_field: Arc::clone(&return_field),
+                            config_options: Arc::clone(&config_options),
+                        })
+                        .expect("split_part should work"),
+                )
+            })
+        });
+    }
+
+    // Scenario 7: StringViewArray, single-char delimiter, first position
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 10, 8, ".", true);
+        let positions = gen_positions(N_ROWS, 1);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(
+            BenchmarkId::new("string_view_single_char", "pos_first"),
+            |b| {
+                b.iter(|| {
+                    black_box(
+                        split_part_func
+                            .invoke_with_args(ScalarFunctionArgs {
+                                args: args.clone(),
+                                arg_fields: arg_fields.clone(),
+                                number_rows: N_ROWS,
+                                return_field: Arc::clone(&return_field),
+                                config_options: Arc::clone(&config_options),
+                            })
+                            .expect("split_part should work"),
+                    )
+                })
+            },
+        );
+    }
+
+    // Scenario 8: Many parts (20), position near end - shows benefit of early termination
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 20, 8, ".", false);
+        let positions = gen_positions(N_ROWS, 2);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(BenchmarkId::new("many_parts_20", "pos_second"), |b| {
+            b.iter(|| {
+                black_box(
+                    split_part_func
+                        .invoke_with_args(ScalarFunctionArgs {
+                            args: args.clone(),
+                            arg_fields: arg_fields.clone(),
+                            number_rows: N_ROWS,
+                            return_field: Arc::clone(&return_field),
+                            config_options: Arc::clone(&config_options),
+                        })
+                        .expect("split_part should work"),
+                )
+            })
+        });
+    }
+
+    // Scenario 9: Long strings with many parts - worst case for old implementation
+    {
+        let (strings, delimiters) = gen_split_part_data(N_ROWS, 50, 16, "/", false);
+        let positions = gen_positions(N_ROWS, 1);
+        let args = vec![strings, delimiters, positions];
+        let arg_fields: Vec<_> = args
+            .iter()
+            .enumerate()
+            .map(|(idx, arg)| {
+                Field::new(format!("arg_{idx}"), arg.data_type(), true).into()
+            })
+            .collect();
+        let return_field = Field::new("f", DataType::Utf8, true).into();
+
+        group.bench_function(
+            BenchmarkId::new("long_strings_50_parts", "pos_first"),
+            |b| {
+                b.iter(|| {
+                    black_box(
+                        split_part_func
+                            .invoke_with_args(ScalarFunctionArgs {
+                                args: args.clone(),
+                                arg_fields: arg_fields.clone(),
+                                number_rows: N_ROWS,
+                                return_field: Arc::clone(&return_field),
+                                config_options: Arc::clone(&config_options),
+                            })
+                            .expect("split_part should work"),
+                    )
+                })
+            },
+        );
+    }
+
+    group.finish();
+}
+
+criterion_group!(benches, criterion_benchmark);
+criterion_main!(benches);