Added more benchmarks for vector operations (07d193cf) · Commits · TNL / tnl-dev

tests/benchmarks/vector-operations.h

+134 −28

Original line number	Diff line number	Diff line
		@@ -64,12 +64,48 @@ benchmarkVectorOperations( const int & loops,

		reset12();

		cout << "Benchmarking CPU-CPU memory transfer:" << endl;
		auto copyAssignHostHost = [&]() {
		hostVector = hostVector2;
		};
		cout << " ";
		benchmarkSingle( loops, datasetSize, copyAssignHostHost, trueFunc, reset1 );

		cout << "Benchmarking CPU-GPU memory transfer:" << endl;
		auto copyAssign = [&]() {
		auto copyAssignHostCuda = [&]() {
		deviceVector = hostVector;
		};
		cout << " ";
		benchmarkSingle( loops, datasetSize, copyAssignHostCuda, compare1, reset1 );

		cout << "Benchmarking GPU-GPU memory transfer:" << endl;
		auto copyAssignCudaCuda = [&]() {
		deviceVector = hostVector;
		};
		cout << " ";
		benchmarkSingle( loops, datasetSize, copyAssign, compare1, reset1 );
		benchmarkSingle( loops, datasetSize, copyAssignCudaCuda, trueFunc, reset1 );

		cout << endl;


		cout << "Benchmarking tnlVector.operator==" << endl;
		auto compareHost = [&]() {
		resultHost = (int) hostVector == hostVector2;
		};
		auto compareCuda = [&]() {
		resultDevice = (int) deviceVector == deviceVector2;
		};
		benchmarkCuda( loops, 2 * datasetSize, compareHost, compareCuda, compareScalars, voidFunc );


		cout << "Benchmarking scalar multiplication:" << endl;
		auto multiplyHost = [&]() {
		hostVector *= 0.5;
		};
		auto multiplyCuda = [&]() {
		deviceVector *= 0.5;
		};
		benchmarkCuda( loops, 2 * datasetSize, multiplyHost, multiplyCuda, compare1, reset1 );


		cout << "Benchmarking vector addition:" << endl;
		@@ -82,6 +118,86 @@ benchmarkVectorOperations( const int & loops,
		benchmarkCuda( loops, 3 * datasetSize, addVectorHost, addVectorCuda, compare1, reset1 );


		cout << "Benchmarking max:" << endl;
		auto maxHost = [&]() {
		resultHost = hostVector.max();
		};
		auto maxCuda = [&]() {
		resultDevice = deviceVector.max();
		};
		benchmarkCuda( loops, datasetSize, maxHost, maxCuda, compareScalars, voidFunc );


		cout << "Benchmarking min:" << endl;
		auto minHost = [&]() {
		resultHost = hostVector.min();
		};
		auto minCuda = [&]() {
		resultDevice = deviceVector.min();
		};
		benchmarkCuda( loops, datasetSize, minHost, minCuda, compareScalars, voidFunc );


		cout << "Benchmarking absMax:" << endl;
		auto absMaxHost = [&]() {
		resultHost = hostVector.absMax();
		};
		auto absMaxCuda = [&]() {
		resultDevice = deviceVector.absMax();
		};
		benchmarkCuda( loops, datasetSize, absMaxHost, absMaxCuda, compareScalars, voidFunc );


		cout << "Benchmarking absMin:" << endl;
		auto absMinHost = [&]() {
		resultHost = hostVector.absMin();
		};
		auto absMinCuda = [&]() {
		resultDevice = deviceVector.absMin();
		};
		benchmarkCuda( loops, datasetSize, absMinHost, absMinCuda, compareScalars, voidFunc );


		cout << "Benchmarking sum:" << endl;
		auto sumHost = [&]() {
		resultHost = hostVector.sum();
		};
		auto sumCuda = [&]() {
		resultDevice = deviceVector.sum();
		};
		benchmarkCuda( loops, datasetSize, sumHost, sumCuda, compareScalars, voidFunc );


		cout << "Benchmarking l1 norm: " << endl;
		auto l1normHost = [&]() {
		resultHost = hostVector.lpNorm( 1.0 );
		};
		auto l1normCuda = [&]() {
		resultDevice = deviceVector.lpNorm( 1.0 );
		};
		benchmarkCuda( loops, datasetSize, l1normHost, l1normCuda, compareScalars, voidFunc );


		cout << "Benchmarking l2 norm: " << endl;
		auto l2normHost = [&]() {
		resultHost = hostVector.lpNorm( 2.0 );
		};
		auto l2normCuda = [&]() {
		resultDevice = deviceVector.lpNorm( 2.0 );
		};
		benchmarkCuda( loops, datasetSize, l2normHost, l2normCuda, compareScalars, voidFunc );


		cout << "Benchmarking l3 norm: " << endl;
		auto l3normHost = [&]() {
		resultHost = hostVector.lpNorm( 3.0 );
		};
		auto l3normCuda = [&]() {
		resultDevice = deviceVector.lpNorm( 3.0 );
		};
		benchmarkCuda( loops, datasetSize, l3normHost, l3normCuda, compareScalars, voidFunc );


		cout << "Benchmarking scalar product:" << endl;
		auto scalarProductHost = [&]() {
		resultHost = hostVector.scalarProduct( hostVector2 );
		@@ -111,16 +227,6 @@ benchmarkVectorOperations( const int & loops,
		#endif
		*/

		cout << "Benchmarking L2 norm: " << endl;
		auto l2normHost = [&]() {
		resultHost = hostVector.lpNorm( 2.0 );
		};
		auto l2normCuda = [&]() {
		resultDevice = deviceVector.lpNorm( 2.0 );
		};
		benchmarkCuda( loops, datasetSize, l2normHost, l2normCuda, compareScalars, voidFunc );


		/*
		cout << "Benchmarking prefix-sum:" << endl;
		timer.reset();