mbutrovich commented on code in PR #12024: URL: https://github.com/apache/arrow-rs/pull/12024#discussion_r4446729021
########## arrow-buffer/benches/boolean_buffer_builder.rs: ########## @@ -0,0 +1,85 @@ +// 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. + +use arrow_buffer::BooleanBufferBuilder; +use criterion::{BenchmarkId, Criterion, Throughput, criterion_group, criterion_main}; +use rand::rngs::StdRng; +use rand::{RngExt, SeedableRng}; +use std::hint; + +const LEN: usize = 64 * 1024; + +fn criterion_benchmark(c: &mut Criterion) { + let mut rng = StdRng::seed_from_u64(42); + let inputs = [ + ("random", (0..LEN).map(|_| rng.random_bool(0.5)).collect()), + ( + "mostly_true", + (0..LEN).map(|_| rng.random_bool(0.9)).collect(), + ), + ("all_true", vec![true; LEN]), + ]; + + let mut group = c.benchmark_group("BooleanBufferBuilder append_slice"); + group.throughput(Throughput::Elements(LEN as u64)); + + for (name, values) in &inputs { + // An offset of 3 bits makes the slice start in the middle of a byte + for offset in [0, 3] { + let id = BenchmarkId::new(format!("offset_{offset}"), name); + group.bench_with_input(id, values, |b, values: &Vec<bool>| { + b.iter(|| { + let mut builder = BooleanBufferBuilder::new(offset + LEN); + builder.append_n(offset, true); + builder.append_slice(hint::black_box(values)); + hint::black_box(builder.finish()) + }) + }); + } + } + + group.finish(); + + // Many appends of 1 to 7 values each, starting in the middle of a byte + const CALLS: usize = 7 * 1024; + let short_inputs = [ + ("random", (0..8).map(|_| rng.random_bool(0.5)).collect()), + ("all_false", vec![false; 8]), + ]; + + let mut group = c.benchmark_group("BooleanBufferBuilder append_slice short"); + group.throughput(Throughput::Elements(CALLS as u64)); + + for (name, values) in &short_inputs { + group.bench_with_input(*name, values, |b, values: &Vec<bool>| { + b.iter(|| { + let mut builder = BooleanBufferBuilder::new(3 + CALLS * 4); + builder.append_n(3, true); + for i in 0..CALLS { + let len = i % 7 + 1; + builder.append_slice(hint::black_box(&values[..len])); Review Comment: > That would make the random row in the PR description exercise the workload it describes. On an Apple M5 Max this also changes the size of the gain. In a default release build, the current `short/random` case takes 19.1 to 19.3 us on main and 16.9 to 17.0 us on this PR, which matches the description's -12%. With fresh values and the same lengths, main takes 55.3 to 55.9 us and this PR takes 25.9 to 27.2 us. Main's time depends on the build (35.3 us with thin LTO). This PR stays at about 26 us with fresh values against 17 us for the repeated pattern in every build I ran. The repeated pattern also hides most of what the register packing suggestion on line 325 saves: 16.9 to 15.2 us on the repeated pattern, and 26 to 16.4 us with fresh values. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
