fsaintjacques commented on a change in pull request #6997: URL: https://github.com/apache/arrow/pull/6997#discussion_r412951352
########## File path: cpp/src/arrow/memory_pool_benchmark.cc ########## @@ -0,0 +1,127 @@ +// 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 "arrow/memory_pool.h" +#include "arrow/result.h" +#include "arrow/util/logging.h" + +#include "benchmark/benchmark.h" + +namespace arrow { + +struct SystemAlloc { + static Result<MemoryPool*> GetAllocator() { return system_memory_pool(); } +}; + +#ifdef ARROW_JEMALLOC +struct Jemalloc { + static Result<MemoryPool*> GetAllocator() { + MemoryPool* pool; + RETURN_NOT_OK(jemalloc_memory_pool(&pool)); + return pool; + } +}; +#endif + +#ifdef ARROW_MIMALLOC +struct Mimalloc { + static Result<MemoryPool*> GetAllocator() { + MemoryPool* pool; + RETURN_NOT_OK(mimalloc_memory_pool(&pool)); + return pool; + } +}; +#endif + +static void TouchCacheLines(uint8_t* data, int64_t nbytes) { + uint8_t total = 0; + while (nbytes > 0) { + total += *data; + data += 64; + nbytes -= 64; + } + benchmark::DoNotOptimize(total); +} + +// Benchmark the raw cost of allocating memory. +// Note this is a best case situation: we always allocate and deallocate exactly +// the same size, without any other allocator traffic. However, it can be +// representative of workloads where we routinely create and destroy +// temporary buffers for intermediate computation results. +template <typename Alloc> +static void AllocateDeallocate(benchmark::State& state) { // NOLINT non-const reference + const int64_t nbytes = state.range(0); + MemoryPool* pool = *Alloc::GetAllocator(); + + for (auto _ : state) { + uint8_t* data; + ARROW_CHECK_OK(pool->Allocate(nbytes, &data)); + pool->Free(data, nbytes); + } +} + +// Benchmark the cost of allocating memory plus accessing it. +template <typename Alloc> +static void AllocateTouchDeallocate( + benchmark::State& state) { // NOLINT non-const reference + const int64_t nbytes = state.range(0); + MemoryPool* pool = *Alloc::GetAllocator(); + + for (auto _ : state) { + uint8_t* data; + ARROW_CHECK_OK(pool->Allocate(nbytes, &data)); + TouchCacheLines(data, nbytes); + pool->Free(data, nbytes); + } +} + +// Benchmark the cost of accessing always the same memory area. +static void TouchArea(benchmark::State& state) { // NOLINT non-const reference Review comment: I didn't understand at first, is this because you want to measure the difference between `AllocateDeallocate` and `AllocateTouchDeallocate` and this test gives you a rough number to validate said gap? If so, add a comment on top of it and move this as the first test to help reading. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
