Add coroutine net2 benchmarks

Benchmark Time CPU Time Old Time New CPU Old CPU New --------------------------------------------------------------------------------------------------------------------------------------------- [bench_delay vs. coroutine_delay]<DELAY>/0 +0.0418 +0.0418 102 106 102 106 [bench_delay vs. coroutine_delay]<DELAY>/1 +0.0424 +0.0424 102 107 102 107 [bench_delay vs. coroutine_delay]<DELAY>/8 +0.0393 +0.0393 102 106 102 106 [bench_delay vs. coroutine_delay]<DELAY>/64 +0.0638 +0.0638 115 122 115 122 [bench_delay vs. coroutine_delay]<DELAY>/512 +0.0429 +0.0429 127 132 127 132 [bench_delay vs. coroutine_delay]<DELAY>/4096 +0.0369 +0.0369 137 142 137 142 [bench_delay vs. coroutine_delay]<DELAY>/32768 +0.0498 +0.0498 149 157 149 157 [bench_delay vs. coroutine_delay]<DELAY>/65536 +0.0343 +0.0343 174 180 174 180 [bench_delay vs. coroutine_delay]<YIELD>/0 +0.0976 +0.0976 46 50 46 50 [bench_delay vs. coroutine_delay]<YIELD>/1 +0.0331 +0.0331 42 43 42 43 [bench_delay vs. coroutine_delay]<YIELD>/8 +0.0293 +0.0293 42 43 42 43 [bench_delay vs. coroutine_delay]<YIELD>/64 +0.0300 +0.0300 42 43 42 43 [bench_delay vs. coroutine_delay]<YIELD>/512 +0.0294 +0.0294 42 43 42 43 [bench_delay vs. coroutine_delay]<YIELD>/4096 +0.0271 +0.0271 42 43 42 43 [bench_delay vs. coroutine_delay]<YIELD>/32768 +0.0320 +0.0320 43 45 43 45 [bench_delay vs. coroutine_delay]<YIELD>/65536 +0.0264 +0.0264 46 47 46 47 OVERALL_GEOMEAN +0.0409 +0.0409 0 0 0 0 Comparing bench_net2 to coroutine_net2 (from ./cbuild_output/bin/flowbench) Benchmark Time CPU Time Old Time New CPU Old CPU New ------------------------------------------------------------------------------------------------------------------------------------ [bench_net2 vs. coroutine_net2]/1 +0.0317 +0.0317 1660 1712 1660 1712 [bench_net2 vs. coroutine_net2]/8 +0.0627 +0.0627 2900 3082 2900 3082 [bench_net2 vs. coroutine_net2]/64 +0.1096 +0.1096 13499 14979 13498 14978 [bench_net2 vs. coroutine_net2]/512 +0.1434 +0.1434 119422 136551 119420 136549 [bench_net2 vs. coroutine_net2]/4096 +0.1966 +0.1966 1269104 1518596 1269076 1518563 [bench_net2 vs. coroutine_net2]/32768 +0.1929 +0.1929 13250215 15805923 13249663 15805633 [bench_net2 vs. coroutine_net2]/65536 +0.3265 +0.3265 30584645 40571416 30584104 40569898 OVERALL_GEOMEAN +0.1484 +0.1484 0 0 0 0
apple · Aug 26, 2024 · d939f97 · d939f97
1 parent b07498b
commit d939f97
Showing 1 changed file with 90 additions and 0 deletions.
diff --git a/flowbench/BenchNet2.cpp b/flowbench/BenchNet2.cpp
@@ -0,0 +1,90 @@
+/*
+ * BenchNet2.cpp
+ *
+ * This source file is part of the FoundationDB open source project
+ *
+ * Copyright 2013-2024 Apple Inc. and the FoundationDB project authors
+ *
+ * Licensed 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 "benchmark/benchmark.h"
+
+#include "flow/IRandom.h"
+#include "flow/flow.h"
+#include "flow/DeterministicRandom.h"
+#include "flow/network.h"
+#include "flow/ThreadHelper.actor.h"
+
+#include "flow/actorcompiler.h" // This must be the last #include.
+
+static Future<Void> increment(TaskPriority priority, uint32_t* sum) {
+	co_await delay(0, priority);
+	++(*sum);
+}
+
+static inline TaskPriority getRandomTaskPriority(DeterministicRandom& rand) {
+	return static_cast<TaskPriority>(rand.randomInt(0, 100));
+}
+
+static Future<Void> benchNet2Actor(benchmark::State* benchState) {
+	size_t actorCount = benchState->range(0);
+	uint32_t sum;
+	int seed = platform::getRandomSeed();
+	while (benchState->KeepRunning()) {
+		sum = 0;
+		std::vector<Future<Void>> futures;
+		futures.reserve(actorCount);
+		DeterministicRandom rand(seed);
+		for (int i = 0; i < actorCount; ++i) {
+			futures.push_back(increment(getRandomTaskPriority(rand), &sum));
+		}
+		co_await waitForAll(futures);
+		benchmark::DoNotOptimize(sum);
+	}
+	benchState->SetItemsProcessed(actorCount * static_cast<long>(benchState->iterations()));
+}
+
+static void coroutine_net2(benchmark::State& benchState) {
+	onMainThread([&benchState] { return benchNet2Actor(&benchState); }).blockUntilReady();
+}
+
+BENCHMARK(coroutine_net2)->Range(1, 1 << 16)->ReportAggregatesOnly(true);
+
+static constexpr bool DELAY = false;
+static constexpr bool YIELD = true;
+
+template <bool useYield>
+static Future<Void> benchDelay(benchmark::State* benchState) {
+	// Number of random delays to start to just to populate the run loop
+	// priority queue
+	int64_t timerCount = benchState->range(0);
+	std::vector<Future<Void>> futures;
+	DeterministicRandom rand(platform::getRandomSeed());
+	while (--timerCount > 0) {
+		futures.push_back(delay(1.0 + rand.random01(), getRandomTaskPriority(rand)));
+	}
+
+	while (benchState->KeepRunning()) {
+		co_await (useYield ? yield() : delay(0));
+	}
+	benchState->SetItemsProcessed(static_cast<long>(benchState->iterations()));
+}
+
+template <bool useYield>
+static void coroutine_delay(benchmark::State& benchState) {
+	onMainThread([&benchState] { return benchDelay<useYield>(&benchState); }).blockUntilReady();
+}
+
+BENCHMARK_TEMPLATE(coroutine_delay, DELAY)->Range(0, 1 << 16)->ReportAggregatesOnly(true);
+BENCHMARK_TEMPLATE(coroutine_delay, YIELD)->Range(0, 1 << 16)->ReportAggregatesOnly(true);