This is an automated email from the ASF dual-hosted git repository. github-merge-queue[bot] pushed a commit to branch gh-readonly-queue/main/pr-11434-989a8fbb8605280eb44b2325e3c8a49ce811d534 in repository https://gitbox.apache.org/repos/asf/arrow-rs.git
commit 551207c86dc7fbb133094cea0d7fb2b02669b67f Author: mightsleep <[email protected]> AuthorDate: Thu Oct 8 02:17:43 2026 +0000 bench(arrow): concat of many small FixedSizeList arrays (#11434) # Rationale for this change No benchmark covers concatenating many small FixedSizeList arrays; "concat fixed size lists" concatenates two large arrays, where the copy dominates. Many one-row arrays is what DataFusion's ScalarValue::iter_to_array does for FixedSizeList scalars (GroupsAccumulatorAdapter::evaluate, first_value/last_value) and what Rerun's chunk batcher does per logged row. Split out so the follow-up fast path can be compared against main. # What changes are included in this PR? - concat 1024 arrays fixed size list f32x3 1 - concat 1024 arrays fixed size list f32x3 nulls 1, built as slices of one array since the generator's fixed seed would give every one-row array the same validity # Are these changes tested? Benchmarks only. # Are there any significant user-facing changes? No. --- arrow/benches/concatenate_kernel.rs | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/arrow/benches/concatenate_kernel.rs b/arrow/benches/concatenate_kernel.rs index 6ae97c90a9..7e4f90f354 100644 --- a/arrow/benches/concatenate_kernel.rs +++ b/arrow/benches/concatenate_kernel.rs @@ -192,6 +192,27 @@ fn add_benchmark(c: &mut Criterion) { b.iter(|| bench_concat(&v1, &v2)) }); + // A column assembled from one-row arrays, as ScalarValue::iter_to_array in DataFusion does + { + let input = (0..1024) + .map(|_| create_primitive_fixed_size_list_array::<Float32Type>(1, 0.0, 0.0, 3)) + .collect::<Vec<_>>(); + let arrays: Vec<_> = input.iter().map(|arr| arr as &dyn Array).collect(); + c.bench_function("concat 1024 arrays fixed size list f32x3 1", |b| { + b.iter(|| bench_concat_arrays(&arrays)) + }); + } + + { + // Slices, as the generator's fixed seed would give every one-row array the same validity + let input = create_primitive_fixed_size_list_array::<Float32Type>(1024, 0.2, 0.0, 3); + let slices = (0..1024).map(|i| input.slice(i, 1)).collect::<Vec<_>>(); + let arrays: Vec<_> = slices.iter().map(|arr| arr as &dyn Array).collect(); + c.bench_function("concat 1024 arrays fixed size list f32x3 nulls 1", |b| { + b.iter(|| bench_concat_arrays(&arrays)) + }); + } + { let batch_size = 1024; let batch_count = 2;
