tools: Add codec benchmark

This code can be used to benchmark native vs .js implementations of vertex codec (and, in the future, vertex filters). Note: right now the SIMD variant of the JS benchmark doesn't run because compiler generates some instructions that Chrome doesn't support...
zeux · Feb 12, 2020 · e452b03 · e452b03
1 parent c4314d1
commit e452b03
Show file tree

Hide file tree

Showing 2 changed files with 146 additions and 0 deletions.
diff --git a/Makefile b/Makefile
@@ -99,6 +99,15 @@ $(EXECUTABLE): $(DEMO_OBJECTS) $(LIBRARY)
 vcachetuner: tools/vcachetuner.cpp $(BUILD)/tools/meshloader.cpp.o $(BUILD)/demo/miniz.cpp.o $(LIBRARY)
 	$(CXX) $^ -fopenmp $(CXXFLAGS) -std=c++11 $(LDFLAGS) -o $@
 
+codecbench: tools/codecbench.cpp $(LIBRARY)
+	$(CXX) $^ $(CXXFLAGS) $(LDFLAGS) -o $@
+
+codecbench.js: tools/codecbench.cpp ${LIBRARY_SOURCES}
+	emcc $^ -O3 -DNDEBUG -s ALLOW_MEMORY_GROWTH=1 -o $@
+
+codecbench-simd.js: tools/codecbench.cpp ${LIBRARY_SOURCES}
+	emcc $^ -O3 -DNDEBUG -s ALLOW_MEMORY_GROWTH=1 -munimplemented-simd128 -o $@
+
 $(LIBRARY): $(LIBRARY_OBJECTS)
 	ar rcs $@ $^
 

diff --git a/tools/codecbench.cpp b/tools/codecbench.cpp
@@ -0,0 +1,137 @@
+#include "../src/meshoptimizer.h"
+
+#include <vector>
+
+#include <time.h>
+#include <stdint.h>
+#include <stdio.h>
+
+#ifdef __EMSCRIPTEN__
+double timestamp()
+{
+	return double(clock()) / double(CLOCKS_PER_SEC);
+}
+#else
+double timestamp()
+{
+	timespec ts;
+	clock_gettime(CLOCK_MONOTONIC, &ts);
+	return double(ts.tv_sec) + 1e-9 * double(ts.tv_nsec);
+}
+#endif
+
+struct Vertex
+{
+	uint16_t data[16];
+};
+
+uint32_t murmur3(uint32_t h)
+{
+	h ^= h >> 16;
+	h *= 0x85ebca6bu;
+	h ^= h >> 13;
+	h *= 0xc2b2ae35u;
+	h ^= h >> 16;
+
+	return h;
+}
+
+void bench(const std::vector<Vertex>& vertices, const std::vector<unsigned int>& indices)
+{
+	std::vector<Vertex> vb(vertices.size());
+	std::vector<unsigned int> ib(indices.size());
+
+	std::vector<unsigned char> vc(meshopt_encodeVertexBufferBound(vertices.size(), sizeof(Vertex)));
+	std::vector<unsigned char> ic(meshopt_encodeIndexBufferBound(indices.size(), vertices.size()));
+
+	printf("source: vertex data %d bytes, index data %d bytes\n", int(vertices.size() * sizeof(Vertex)), int(indices.size() * 4));
+
+	for (int pass = 0; pass < 2; ++pass)
+	{
+		if (pass == 1)
+			meshopt_optimizeVertexCacheStrip(&ib[0], &indices[0], indices.size(), vertices.size());
+		else
+			meshopt_optimizeVertexCache(&ib[0], &indices[0], indices.size(), vertices.size());
+
+		meshopt_optimizeVertexFetch(&vb[0], &ib[0], indices.size(), &vertices[0], vertices.size(), sizeof(Vertex));
+
+		vc.resize(vc.capacity());
+		vc.resize(meshopt_encodeVertexBuffer(&vc[0], vc.size(), &vb[0], vertices.size(), sizeof(Vertex)));
+
+		ic.resize(ic.capacity());
+		ic.resize(meshopt_encodeIndexBuffer(&ic[0], ic.size(), &ib[0], indices.size()));
+
+		printf("pass %d: vertex data %d bytes, index data %d bytes\n", pass, int(vc.size()), int(ic.size()));
+
+		for (int attempt = 0; attempt < 10; ++attempt)
+		{
+			double t0 = timestamp();
+
+			int rv = meshopt_decodeVertexBuffer(&vb[0], vertices.size(), sizeof(Vertex), &vc[0], vc.size());
+			assert(rv == 0);
+			(void)rv;
+
+			double t1 = timestamp();
+
+			int ri = meshopt_decodeIndexBuffer(&ib[0], indices.size(), 4, &ic[0], ic.size());
+			assert(ri == 0);
+			(void)ri;
+
+			double t2 = timestamp();
+
+			double GB = 1024 * 1024 * 1024;
+
+			printf("decode: vertex %.2f ms (%.2f GB/sec), index %.2f ms (%.2f GB/sec)\n",
+				(t1 - t0) * 1000, double(vertices.size() * sizeof(Vertex)) / GB / (t1 - t0),
+				(t2 - t1) * 1000, double(indices.size() * 4) / GB / (t2 - t1));
+		}
+	}
+}
+
+int main()
+{
+	meshopt_encodeIndexVersion(1);
+
+	const int N = 2000;
+
+	std::vector<Vertex> vertices;
+	vertices.reserve((N + 1) * (N + 1));
+
+	for (int x = 0; x <= N; ++x)
+	{
+		for (int y = 0; y <= N; ++y)
+		{
+			Vertex v;
+
+			for (int k = 0; k < 16; ++k)
+			{
+				uint32_t h = murmur3((x * (N + 1) + y) * 16 + k);
+
+				// use random k-bit sequence for each word to test all encoding types
+				// note: this doesn't stress the sentinel logic too much but it's all branchless so it's probably fine?
+				v.data[k] = h & ((1 << k) - 1);
+			}
+
+			vertices.push_back(v);
+		}
+	}
+
+	std::vector<unsigned int> indices;
+	indices.reserve(N * N * 6);
+
+	for (int x = 0; x < N; ++x)
+	{
+		for (int y = 0; y < N; ++y)
+		{
+			indices.push_back((x + 0) * N + (y + 0));
+			indices.push_back((x + 1) * N + (y + 0));
+			indices.push_back((x + 0) * N + (y + 1));
+
+			indices.push_back((x + 0) * N + (y + 1));
+			indices.push_back((x + 1) * N + (y + 0));
+			indices.push_back((x + 1) * N + (y + 1));
+		}
+	}
+
+	bench(vertices, indices);
+}