(arrow-rs) branch main updated: bench(arrow): sweep byte-view scalar comparison across a size range (#10728)

[alamb]

This is an automated email from the ASF dual-hosted git repository.

alamb pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/arrow-rs.git


The following commit(s) were added to refs/heads/main by this push:
     new 7c306e401c bench(arrow): sweep byte-view scalar comparison across a 
size range (#10728)
7c306e401c is described below

commit 7c306e401c0f22a0c39f34989148c018af95ace8
Author: Giladi <dkgs1...@gmail.com>
AuthorDate: Tue Aug 18 20:02:10 2026 +0300

    bench(arrow): sweep byte-view scalar comparison across a size range (#10728)
    
    # Which issue does this PR close?
    
    - Closes #10727.
    
    # Rationale for this change
    
    The byte-view scalar comparison benchmarks in
    arrow/benches/comparison_kernels.rs all run at one size: 1024 * 1024 * 8
    = 8,388,608 rows, or 134 MB of views. That is far past cache, so they
    measure memory bandwidth more than the per-row comparison, and only at a
    single point on the size curve. Split out of #10689 at @alamb's request
    so it can go through the automated benchmarking scripts first.
    
    # What changes are included in this PR?
    
    A new stringview_scalar_eq benchmark group running eq(StringViewArray,
    StringViewArray::new_scalar("xxxx")) at 65,536 / 1,048,576 / 8,388,608
    rows, with Throughput::Elements so criterion reports elements/sec across
    the sizes. Existing benchmarks are untouched.
    
    # Are these changes tested?
    
    yes
    
    # Are there any user-facing changes?
    
    no
---
 arrow/benches/comparison_kernels.rs | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/arrow/benches/comparison_kernels.rs 
b/arrow/benches/comparison_kernels.rs
index ce890b7f49..69523026e4 100644
--- a/arrow/benches/comparison_kernels.rs
+++ b/arrow/benches/comparison_kernels.rs
@@ -25,7 +25,7 @@ use arrow::{array::*, datatypes::Float32Type, 
datatypes::Int32Type};
 use arrow_buffer::IntervalMonthDayNano;
 use arrow_string::like::*;
 use arrow_string::regexp::regexp_is_match_scalar;
-use criterion::Criterion;
+use criterion::{Criterion, Throughput};
 use rand::RngExt;
 use rand::rngs::StdRng;
 use std::hint;
@@ -530,6 +530,23 @@ fn add_benchmark(c: &mut Criterion) {
         b.iter(|| eq(&dict_arr_a, &dict_arr_b).unwrap())
     });
 
+    // eq scalar benchmarks across sizes: 16 bytes of view a row, so the 
largest is bandwidth-bound
+
+    let mut group = c.benchmark_group("stringview_scalar_eq");
+
+    for rows in [65_536usize, 1024 * 1024, 1024 * 1024 * 8] {
+        let mut rng = seedable_rng();
+        let values = StringViewArray::from_iter(make_string_array(rows, &mut 
rng));
+        let scalar = StringViewArray::new_scalar("xxxx");
+
+        group.throughput(Throughput::Elements(rows as u64));
+        group.bench_function(format!("eq_scalar(rows={rows})"), |b| {
+            b.iter(|| eq(&values, &scalar).unwrap())
+        });
+    }
+
+    group.finish();
+
     // RunEndEncoded benchmarks
 
     let mut group = c.benchmark_group("ree_comparison");