mbutrovich commented on code in PR #12024: URL: https://github.com/apache/arrow-rs/pull/12024#discussion_r4446733181
########## 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])); + } + hint::black_box(builder.finish()) + }) + }); + } + + group.finish(); Review Comment: Could we add 8-value appends at offsets 0 and 3 next to the short group? At offset 3 the five head values and three remainder values all go through the per-bit loops and none is packed. That's the case in https://github.com/apache/arrow-rs/pull/12024#discussion_r4445718870. At offset 0 each call packs one byte and has no head. That's the case where the register packing suggestion on line 325 gains least, because its head ORs 0 into the first byte. With both cases here, the benchmark bot would show whether either one regresses. ```suggestion group.finish(); // Many appends of 8 values each. At offset 3 every value goes to the head // or the remainder, and none is packed. let eight_inputs = [ ( "random", (0..CALLS * 8).map(|_| rng.random_bool(0.5)).collect(), ), ("all_false", vec![false; CALLS * 8]), ]; let mut group = c.benchmark_group("BooleanBufferBuilder append_slice eight"); group.throughput(Throughput::Elements(CALLS as u64)); for (name, values) in &eight_inputs { 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 + values.len()); builder.append_n(offset, true); for chunk in values.as_chunks::<8>().0 { builder.append_slice(hint::black_box(chunk)); } hint::black_box(builder.finish()) }) }); } } group.finish(); ``` -- 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]
