Add an InstanceVertices partition perf test

Summary:
* Given a trivially simple condition, like a bounding box, over 1
  million items can be partitioned in under 3ms.
* Parallel algorithms reduce that a little but are only effective with
  volumes in excess ~200k, this might be better with a more complex
  condition/predicate.

1 files changed, 43 insertions, 0 deletions

diff --git a/test/perf-instancing.cpp b/test/perf-instancing.cpp
new file mode 100644
index 0000000..829ea50
--- /dev/null
+++ b/test/perf-instancing.cpp
@@ -0,0 +1,43 @@
+#include "gfx/gl/instanceVertices.h"
+#include "testMainWindow.h"
+#include <benchmark/benchmark.h>
+#include <random>
+
+struct Instance {
+	GlobalPosition3D pos;
+	glm::mat3 rot;
+};
+
+struct data {
+	explicit data(size_t n)
+	{
+		std::mt19937 gen(std::random_device {}());
+		std::uniform_int_distribution<GlobalDistance> xy(0, 1000000);
+		std::uniform_int_distribution<GlobalDistance> z(0, 10000);
+		while (n--) {
+			proxies.emplace_back(instances.acquire(GlobalPosition3D {xy(gen), xy(gen), z(gen)}, glm::mat3 {}));
+		}
+	}
+
+	InstanceVertices<Instance> instances;
+	std::vector<InstanceVertices<Instance>::InstanceProxy> proxies;
+};
+
+static void
+partition(benchmark::State & state)
+{
+	TestMainWindow window;
+	data d(static_cast<size_t>(state.range()));
+	GlobalPosition2D pos {};
+	for (auto _ : state) {
+		d.instances.partition([&pos](const auto & i) {
+			return std::abs(i.pos.x - pos.x) < 5 && std::abs(i.pos.y - pos.y) < 5;
+		});
+		pos += GlobalPosition2D {33, 17};
+		pos %= 1000000;
+	}
+}
+
+BENCHMARK(partition)->Range(0, 1 << 20);
+
+BENCHMARK_MAIN();