pitrou commented on code in PR #36702: URL: https://github.com/apache/arrow/pull/36702#discussion_r1264681973
########## cpp/src/parquet/page_index_benchmark.cc: ########## @@ -0,0 +1,153 @@ +// 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. + +#include <cstdint> +#include <vector> + +#include "benchmark/benchmark.h" + +#include "parquet/metadata.h" +#include "parquet/page_index.h" +#include "parquet/schema.h" +#include "parquet/test_util.h" +#include "parquet/thrift_internal.h" + +namespace parquet { + +void BM_ReadOffsetIndex(::benchmark::State& state) { + auto builder = OffsetIndexBuilder::Make(); + const size_t num_pages = state.range(0); + constexpr size_t page_size = 1024; + constexpr size_t first_row_index = 10000; + for (size_t i = 0; i < num_pages; ++i) { + builder->AddPage(page_size * i, page_size, first_row_index * i); + } + const int64_t final_position = 4096; + builder->Finish(final_position); + auto sink = CreateOutputStream(); + builder->WriteTo(sink.get()); + auto buffer = sink->Finish().ValueOrDie(); + ReaderProperties properties; + for (auto _ : state) { + auto offset_index = OffsetIndex::Make(buffer->data() + 0, buffer->size(), properties); + ::benchmark::DoNotOptimize(offset_index); + } + state.counters["num_pages"] = num_pages; +} + +BENCHMARK(BM_ReadOffsetIndex)->Range(8, 1024); + +// The sample string length for FLBA and ByteArray benchmarks +constexpr static uint32_t kDataStringLength = 8; + +void GenerateRandomString(uint32_t length, uint32_t seed, std::vector<uint8_t>* heap) { + // Character set used to generate random string + const std::string charset = + "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"; + + std::default_random_engine gen(seed); + std::uniform_int_distribution<uint32_t> dist(0, static_cast<int>(charset.size() - 1)); + + for (uint32_t i = 0; i < length; i++) { + heap->push_back(charset[dist(gen)]); + } +} + +template <typename T> +void GenerateBenchmarkData(uint32_t size, uint32_t seed, T* data, + [[maybe_unused]] std::vector<uint8_t>* heap = nullptr) { + if constexpr (std::is_integral_v<T>) { + std::default_random_engine gen(seed); + std::uniform_int_distribution<T> d(std::numeric_limits<T>::min(), + std::numeric_limits<T>::max()); + for (uint32_t i = 0; i < size; ++i) { + data[i] = d(gen); + } + } else if constexpr (std::is_floating_point_v<T>) { + std::default_random_engine gen(seed); + std::uniform_real_distribution<T> d(std::numeric_limits<T>::lowest(), + std::numeric_limits<T>::max()); + for (uint32_t i = 0; i < size; ++i) { + data[i] = d(gen); + } + } else if constexpr (std::is_same_v<FLBA, T>) { + GenerateRandomString(kDataStringLength * size, seed, heap); + for (uint32_t i = 0; i < size; ++i) { + data[i].ptr = heap->data() + i * kDataStringLength; + } + } else if constexpr (std::is_same_v<ByteArray, T>) { + GenerateRandomString(kDataStringLength * size, seed, heap); + for (uint32_t i = 0; i < size; ++i) { + data[i].ptr = heap->data() + i * kDataStringLength; + data[i].len = kDataStringLength; + } + } else if constexpr (std::is_same_v<Int96, T>) { + std::default_random_engine gen(seed); + std::uniform_int_distribution<int> d(std::numeric_limits<int>::min(), + std::numeric_limits<int>::max()); + for (uint32_t i = 0; i < size; ++i) { + data[i].value[0] = d(gen); + data[i].value[1] = d(gen); + data[i].value[2] = d(gen); + } + } +} + +template <typename DType> +void BM_ReadColumnIndex(::benchmark::State& state) { + schema::NodePtr type = ::parquet::schema::PrimitiveNode::Make( + "b", Repetition::OPTIONAL, DType::type_num, ConvertedType::NONE, 8); + auto descr_ptr = + std::make_unique<ColumnDescriptor>(type, /*def_level=*/1, /*rep_level=*/0); + auto descr = descr_ptr.get(); + + const size_t num_pages = state.range(0); + + auto builder = ColumnIndexBuilder::Make(descr); + + const size_t values_per_page = 100; + for (int i = 0; i < num_pages; ++i) { + auto stats = MakeStatistics<DType>(descr); + std::vector<uint8_t> heap; + std::vector<typename DType::c_type> values; + values.resize(values_per_page); + GenerateBenchmarkData(values_per_page, /*seed=*/0, values.data(), &heap); + stats->Update(values.data(), values_per_page, 0); + builder->AddPage(stats->Encode()); + } + + builder->Finish(); + auto sink = CreateOutputStream(); + builder->WriteTo(sink.get()); + auto buffer = sink->Finish().ValueOrDie(); + ReaderProperties properties; + for (auto _ : state) { + auto column_index = + ColumnIndex::Make(*descr, buffer->data() + 0, buffer->size(), properties); + ::benchmark::DoNotOptimize(column_index); + } + state.counters["num_pages"] = num_pages; +} + +BENCHMARK_TEMPLATE(BM_ReadColumnIndex, Int32Type)->Range(8, 1024); +BENCHMARK_TEMPLATE(BM_ReadColumnIndex, Int64Type)->Range(8, 1024); +BENCHMARK_TEMPLATE(BM_ReadColumnIndex, FloatType)->Range(8, 1024); +BENCHMARK_TEMPLATE(BM_ReadColumnIndex, DoubleType)->Range(8, 1024); +BENCHMARK_TEMPLATE(BM_ReadColumnIndex, FLBAType)->Range(8, 1024); +BENCHMARK_TEMPLATE(BM_ReadColumnIndex, ByteArrayType)->Range(8, 1024); Review Comment: Hmm, why so many types? We don't expect significant differences between Int32 and Int64, or Float and Double, I think. Therefore we can probably reduce it to: Int64, Double, FLBA, ByteArray. -- 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]
