Commit 8bb94c12 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Used lambda functions to slightly generalize benchmarks

parent 2771e103

tests/benchmarks/tnl-cuda-benchmarks.h

+92 −84
Original line number Diff line number Diff line
@@ -37,6 +37,12 @@ using SlicedEllpackMatrix = tnlSlicedEllpackMatrix< Real, Device, Index >; 
const double oneGB = 1024.0 * 1024.0 * 1024.0;





// TODO:

// check operations with the timer:

//   - reset() clears the timer and starts it again

//   - getTime() stops the timer and starts it again !!!





template< typename Matrix >

int setHostTestMatrix( Matrix& matrix,

                       const int elementsPerRow )
@@ -101,6 +107,21 @@ void setCudaTestMatrix( Matrix& matrix, 
   tnlCuda::freeFromDevice( kernel_matrix );

}



template<typename Function, typename... Args>

double time_void_function(int loops, Function & f, Args & ...args)

{

    tnlTimerRT timer;

    timer.reset();



    for(int i = 0; i < loops; ++i) {

        timer.start();

        f(args...);

        timer.stop();

    }



    return timer.getTime();

}



template< typename Real,

          template< typename, typename, typename > class Matrix,

          template< typename, typename, typename > class Vector = tnlVector >
@@ -148,7 +169,6 @@ benchmarkSpMV( const int & loops, 
      return false;

   }



   tnlTimerRT timer;

   double bandwidth( 0.0 ), datasetSize( 0.0 ), timeHost( 0.0 ), timeDevice( 0.0 );



   tnlList< tnlString > parsedType;
@@ -161,31 +181,26 @@ benchmarkSpMV( const int & loops, 
   hostVector.setValue( 1.0 );

   deviceVector.setValue( 1.0 );



   // TODO: check operations with the timer:

   //   -   reset() clears the timer and starts it again

   //   -   getTime() stops the timer and starts it again !!!

   timer.reset();

   for( int i = 0; i < loops; i++ )

      hostMatrix.vectorProduct( hostVector, hostVector2 );

   timer.stop();

   timeHost = timer.getTime();

   bandwidth = datasetSize / timer.getTime();

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

   

   timer.reset();

   for( int i = 0; i < loops; i++ )

      deviceMatrix.vectorProduct( deviceVector, deviceVector2 );

   timer.stop();

   timeDevice = timer.getTime();

   bandwidth = datasetSize / timer.getTime();

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

   auto spmvHost = []( const HostMatrix & m, const HostVector & x, HostVector & y ) {

      m.vectorProduct( x, y );

   };

   timeHost = time_void_function( loops, spmvHost, hostMatrix, hostVector, hostVector2 );

   bandwidth = datasetSize / timeHost;

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeHost << " sec." << endl;

   

   auto spmvCuda = []( const DeviceMatrix & m, const CudaVector & x, CudaVector & y ) {

      m.vectorProduct( x, y );

   };

   timeDevice = time_void_function( loops, spmvCuda, deviceMatrix, deviceVector, deviceVector2 );

   bandwidth = datasetSize / timeDevice;

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeDevice << " sec." << endl;

   cout << "  CPU/GPU speedup: " << timeHost / timeDevice << endl;



   //cout << hostVector2 << endl << deviceVector2 << endl;

      

   if( hostVector2 != deviceVector2 )

   {      

      cerr << "Error in SliceEllpack Spmv kernel at positions" << endl;

      cerr << "Error in Spmv kernel" << endl;

      //for( int i = 0; i < size; i++ )

      //   if( hostVector2.getElement( i ) != deviceVector2.getElement( i ) )

      //      cerr << " " << i;
@@ -236,40 +251,35 @@ int main( int argc, char* argv[] ) 
   hostVector2.setValue( 1.0 );

   deviceVector2.setValue( 1.0 );



   tnlTimerRT timer;

   double bandwidth( 0.0 );

   Real resultHost, resultDevice, timeHost, timeDevice;





   cout << "Benchmarking CPU-GPU memory bandwidth: ";

   timer.reset();

   timer.start();

   for( int i = 0; i < loops; i++ )

     deviceVector = hostVector;

   timer.stop();    

   bandwidth = datasetSize / timer.getTime();

   auto copyAssign = []( CudaVector & v1, const HostVector & v2 ) {

      v1 = v2;

   };

   timeHost = time_void_function( loops, copyAssign, deviceVector, hostVector );

   bandwidth = datasetSize / timeHost;

   cout << bandwidth << " GB/sec." << endl;

    



   cout << "Benchmarking vector addition:" << endl;

   timer.reset();

   timer.start();

   for( int i = 0; i < loops; i++ )

     hostVector.addVector( hostVector2 );

   timer.stop();

   timeHost = timer.getTime();

   bandwidth = 3 * datasetSize / timer.getTime();

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

    

   timer.reset();

   timer.start();

   for( int i = 0; i < loops; i++ )

     deviceVector.addVector( deviceVector2 );

   auto addVectorHost = []( HostVector & v1, const HostVector & v2 ) {

      v1.addVector( v2 );

   };

   timeHost = time_void_function( loops, addVectorHost, hostVector, hostVector2 );

   bandwidth = 3 * datasetSize / timeHost;

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeHost << " sec." << endl;

    

   auto addVectorCuda = []( CudaVector & v1, const CudaVector & v2 ) {

      v1.addVector( v2 );

      // TODO: synchronization should be part of addVector

      cudaThreadSynchronize();

   timer.stop();

   timeDevice = timer.getTime();

   bandwidth = 3 * datasetSize / timer.getTime();

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

   };

   timeDevice = time_void_function( loops, addVectorCuda, deviceVector, deviceVector2 );

   bandwidth = 3 * datasetSize / timeDevice;

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeDevice << " sec." << endl;

   cout << "  CPU/GPU speedup: " << timeHost / timeDevice << endl;
@@ -280,30 +290,31 @@ int main( int argc, char* argv[] ) 




   cout << "Benchmarking scalar product:" << endl;

   timer.reset();

   timer.start();

   for( int i = 0; i < loops; i++ )

     resultHost = hostVector.scalarProduct( hostVector2 );

   timer.stop();

   timeHost = timer.getTime();

   bandwidth = 2 * datasetSize / timer.getTime();

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

    

   timer.reset();

   timer.start();

   for( int i = 0; i < loops; i++ )

      resultDevice = deviceVector.scalarProduct( deviceVector2 );

   timer.stop();

   timeDevice = timer.getTime();

   bandwidth = 2 * datasetSize / timer.getTime();

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

   // FIXME: scalarProduct is not const method

//   auto scalarProductHost = []( const HostVector & v1, const HostVector & v2 ) {

   auto scalarProductHost = []( HostVector & v1, const HostVector & v2 ) {

      return v1.scalarProduct( v2 );

   };

   timeHost = time_void_function( loops, scalarProductHost, hostVector, hostVector2 );

   bandwidth = 2 * datasetSize / timeHost;

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeHost << " sec." << endl;

    

   // FIXME: scalarProduct is not const method

//   auto scalarProductCuda = []( const CudaVector & v1, const CudaVector & v2 ) {

   auto scalarProductCuda = []( CudaVector & v1, const CudaVector & v2 ) {

      return v1.scalarProduct( v2 );

   };

   timeDevice = time_void_function( loops, scalarProductCuda, deviceVector, deviceVector2 );

   bandwidth = 2 * datasetSize / timeDevice;

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeDevice << " sec." << endl;

   cout << "  CPU/GPU speedup: " << timeHost / timeDevice << endl;



   if( resultHost != resultDevice )

   {

      cerr << "Error. " << resultHost << " != " << resultDevice << endl;

   // TODO: devise a way to check the result of the timed function

//   if( resultHost != resultDevice )

//   {

//      cerr << "Error. " << resultHost << " != " << resultDevice << endl;

      //return EXIT_FAILURE;

   }

//   }



#ifdef HAVE_CUBLAS

   cout << "Benchmarking scalar product on GPU with Cublas: " << endl;
@@ -323,31 +334,28 @@ int main( int argc, char* argv[] ) 
   cout << "bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

#endif    



   cout << "Benchmarking L2 norm: ";

   timer.reset();

   timer.start();

   for( int i = 0; i < loops; i++ )

     resultHost = hostVector.lpNorm( 2.0 );

   timer.stop();

   timeHost = timer.getTime();

   bandwidth = datasetSize / timer.getTime();

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

    

   timer.reset();

   timer.start();

   for( int i = 0; i < loops; i++ )

      resultDevice = deviceVector.lpNorm( 2.0 );

   timer.stop();

   timeDevice = timer.getTime();

   bandwidth = datasetSize / timer.getTime();

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timer.getTime() << " sec." << endl;

   cout << "Benchmarking L2 norm: " << endl;

   auto l2normHost = []( const HostVector & v ) {

      return v.lpNorm( 2.0 );

   };

   timeHost = time_void_function( loops, l2normHost, hostVector );

   bandwidth = datasetSize / timeHost;

   cout << "  CPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeHost << " sec." << endl;

    

   auto l2normCuda = []( const CudaVector & v ) {

      return v.lpNorm( 2.0 );

   };

   timeDevice = time_void_function( loops, l2normCuda, deviceVector );

   bandwidth = datasetSize / timeDevice;

   cout << "  GPU: bandwidth: " << bandwidth << " GB/sec, time: " << timeDevice << " sec." << endl;

   cout << "  CPU/GPU speedup: " << timeHost / timeDevice << endl;



   if( resultHost != resultDevice )

   {

      cerr << "Error. " << resultHost << " != " << resultDevice << endl;

   // TODO: devise a way to check the result of the timed function

//   if( resultHost != resultDevice )

//   {

//      cerr << "Error. " << resultHost << " != " << resultDevice << endl;

      //return EXIT_FAILURE;

   }

//   }





   /*