Commit 51d16ef0 authored by Jakub Klinkovský's avatar Jakub Klinkovský
Browse files

Refactoring benchmarks - common getHardwareMetadata function

parent 76111496

src/Benchmarks/BLAS/tnl-benchmark-blas.h

+2 −37
Original line number Diff line number Diff line
@@ -13,8 +13,7 @@ 
#pragma once



#include <TNL/Devices/Host.h>

#include <TNL/Devices/CudaDeviceInfo.h>

#include <TNL/Devices/SystemInfo.h>

#include <TNL/Devices/Cuda.h>

#include <TNL/Config/ConfigDescription.h>

#include <TNL/Config/ParameterContainer.h>
@@ -26,9 +25,6 @@ using namespace TNL; 
using namespace TNL::Benchmarks;





// TODO: should benchmarks check the result of the computation?





template< typename Real >

void

runBlasBenchmarks( Benchmark & benchmark,
@@ -146,38 +142,7 @@ main( int argc, char* argv[] ) 
   Benchmark benchmark( loops, verbose );



   // prepare global metadata

   const int cpu_id = 0;

   Devices::CacheSizes cacheSizes = Devices::SystemInfo::getCPUCacheSizes( cpu_id );

   String cacheInfo = String( cacheSizes.L1data ) + ", "

                       + String( cacheSizes.L1instruction ) + ", "

                       + String( cacheSizes.L2 ) + ", "

                       + String( cacheSizes.L3 );

#ifdef HAVE_CUDA

   const int activeGPU = Devices::CudaDeviceInfo::getActiveDevice();

   const String deviceArch = String( Devices::CudaDeviceInfo::getArchitectureMajor( activeGPU ) ) + "." +

                             String( Devices::CudaDeviceInfo::getArchitectureMinor( activeGPU ) );

#endif

   Benchmark::MetadataMap metadata {

      { "host name", Devices::SystemInfo::getHostname() },

      { "architecture", Devices::SystemInfo::getArchitecture() },

      { "system", Devices::SystemInfo::getSystemName() },

      { "system release", Devices::SystemInfo::getSystemRelease() },

      { "start time", Devices::SystemInfo::getCurrentTime() },

      { "CPU model name", Devices::SystemInfo::getCPUModelName( cpu_id ) },

      { "CPU cores", Devices::SystemInfo::getNumberOfCores( cpu_id ) },

      { "CPU threads per core", Devices::SystemInfo::getNumberOfThreads( cpu_id ) / Devices::SystemInfo::getNumberOfCores( cpu_id ) },

      { "CPU max frequency (MHz)", Devices::SystemInfo::getCPUMaxFrequency( cpu_id ) / 1e3 },

      { "CPU cache sizes (L1d, L1i, L2, L3) (kiB)", cacheInfo },

#ifdef HAVE_CUDA

      { "GPU name", Devices::CudaDeviceInfo::getDeviceName( activeGPU ) },

      { "GPU architecture", deviceArch },

      { "GPU CUDA cores", Devices::CudaDeviceInfo::getCudaCores( activeGPU ) },

      { "GPU clock rate (MHz)", (double) Devices::CudaDeviceInfo::getClockRate( activeGPU ) / 1e3 },

      { "GPU global memory (GB)", (double) Devices::CudaDeviceInfo::getGlobalMemory( activeGPU ) / 1e9 },

      { "GPU memory clock rate (MHz)", (double) Devices::CudaDeviceInfo::getMemoryClockRate( activeGPU ) / 1e3 },

      { "GPU memory ECC enabled", Devices::CudaDeviceInfo::getECCEnabled( activeGPU ) },

#endif

   };

   Benchmark::MetadataMap metadata = getHardwareMetadata();



   if( precision == "all" || precision == "float" )

      runBlasBenchmarks< float >( benchmark, metadata, minSize, maxSize, sizeStepFactor, loops, elementsPerRow );

src/Benchmarks/Benchmarks.h

+49 −0
Original line number Diff line number Diff line
@@ -21,6 +21,11 @@ 
#include <TNL/String.h>

#include <TNL/Solvers/IterativeSolverMonitor.h>



#include <TNL/Devices/Host.h>

#include <TNL/Devices/SystemInfo.h>

#include <TNL/Devices/CudaDeviceInfo.h>

#include <TNL/Communicators/MpiCommunicator.h>



namespace TNL {

namespace Benchmarks {
@@ -452,5 +457,49 @@ protected: 
   Solvers::IterativeSolverMonitor< double, int > monitor;

};





Benchmark::MetadataMap getHardwareMetadata()

{

   const int cpu_id = 0;

   Devices::CacheSizes cacheSizes = Devices::SystemInfo::getCPUCacheSizes( cpu_id );

   String cacheInfo = String( cacheSizes.L1data ) + ", "

                       + String( cacheSizes.L1instruction ) + ", "

                       + String( cacheSizes.L2 ) + ", "

                       + String( cacheSizes.L3 );

#ifdef HAVE_CUDA

   const int activeGPU = Devices::CudaDeviceInfo::getActiveDevice();

   const String deviceArch = String( Devices::CudaDeviceInfo::getArchitectureMajor( activeGPU ) ) + "." +

                             String( Devices::CudaDeviceInfo::getArchitectureMinor( activeGPU ) );

#endif

   Benchmark::MetadataMap metadata {

       { "host name", Devices::SystemInfo::getHostname() },

       { "architecture", Devices::SystemInfo::getArchitecture() },

       { "system", Devices::SystemInfo::getSystemName() },

       { "system release", Devices::SystemInfo::getSystemRelease() },

       { "start time", Devices::SystemInfo::getCurrentTime() },

#ifdef HAVE_MPI

       { "number of MPI processes", Communicators::MpiCommunicator::GetSize( Communicators::MpiCommunicator::AllGroup ) },

#endif

       { "OpenMP enabled", Devices::Host::isOMPEnabled() },

       { "OpenMP threads", Devices::Host::getMaxThreadsCount() },

       { "CPU model name", Devices::SystemInfo::getCPUModelName( cpu_id ) },

       { "CPU cores", Devices::SystemInfo::getNumberOfCores( cpu_id ) },

       { "CPU threads per core", Devices::SystemInfo::getNumberOfThreads( cpu_id ) / Devices::SystemInfo::getNumberOfCores( cpu_id ) },

       { "CPU max frequency (MHz)", Devices::SystemInfo::getCPUMaxFrequency( cpu_id ) / 1e3 },

       { "CPU cache sizes (L1d, L1i, L2, L3) (kiB)", cacheInfo },

#ifdef HAVE_CUDA

       { "GPU name", Devices::CudaDeviceInfo::getDeviceName( activeGPU ) },

       { "GPU architecture", deviceArch },

       { "GPU CUDA cores", Devices::CudaDeviceInfo::getCudaCores( activeGPU ) },

       { "GPU clock rate (MHz)", (double) Devices::CudaDeviceInfo::getClockRate( activeGPU ) / 1e3 },

       { "GPU global memory (GB)", (double) Devices::CudaDeviceInfo::getGlobalMemory( activeGPU ) / 1e9 },

       { "GPU memory clock rate (MHz)", (double) Devices::CudaDeviceInfo::getMemoryClockRate( activeGPU ) / 1e3 },

       { "GPU memory ECC enabled", Devices::CudaDeviceInfo::getECCEnabled( activeGPU ) },

#endif

   };



   return metadata;

}



} // namespace Benchmarks

} // namespace TNL

src/Benchmarks/DistSpMV/tnl-benchmark-distributed-spmv.h

+3 −41
Original line number Diff line number Diff line
@@ -16,12 +16,10 @@ 
#include <TNL/Debugging/FPE.h>

#endif



#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

#include <TNL/Devices/SystemInfo.h>

#include <TNL/Devices/CudaDeviceInfo.h>

#include <TNL/Config/ConfigDescription.h>

#include <TNL/Config/ParameterContainer.h>

#include <TNL/Devices/Host.h>

#include <TNL/Devices/Cuda.h>

#include <TNL/Communicators/MpiCommunicator.h>

#include <TNL/Communicators/NoDistrCommunicator.h>

#include <TNL/Communicators/ScopedInitializer.h>
@@ -340,43 +338,7 @@ main( int argc, char* argv[] ) 
   Benchmark benchmark( loops, verbose );



   // prepare global metadata

   const int cpu_id = 0;

   Devices::CacheSizes cacheSizes = Devices::SystemInfo::getCPUCacheSizes( cpu_id );

   String cacheInfo = String( cacheSizes.L1data ) + ", "

                       + String( cacheSizes.L1instruction ) + ", "

                       + String( cacheSizes.L2 ) + ", "

                       + String( cacheSizes.L3 );

#ifdef HAVE_CUDA

   const int activeGPU = Devices::CudaDeviceInfo::getActiveDevice();

   const String deviceArch = String( Devices::CudaDeviceInfo::getArchitectureMajor( activeGPU ) ) + "." +

                             String( Devices::CudaDeviceInfo::getArchitectureMinor( activeGPU ) );

#endif

   Benchmark::MetadataMap metadata {

       { "host name", Devices::SystemInfo::getHostname() },

       { "architecture", Devices::SystemInfo::getArchitecture() },

       { "system", Devices::SystemInfo::getSystemName() },

       { "system release", Devices::SystemInfo::getSystemRelease() },

       { "start time", Devices::SystemInfo::getCurrentTime() },

#ifdef HAVE_MPI

       { "number of MPI processes", CommunicatorType::GetSize( CommunicatorType::AllGroup ) },

#endif

       { "OpenMP enabled", Devices::Host::isOMPEnabled() },

       { "OpenMP threads", Devices::Host::getMaxThreadsCount() },

       { "CPU model name", Devices::SystemInfo::getCPUModelName( cpu_id ) },

       { "CPU cores", Devices::SystemInfo::getNumberOfCores( cpu_id ) },

       { "CPU threads per core", Devices::SystemInfo::getNumberOfThreads( cpu_id ) / Devices::SystemInfo::getNumberOfCores( cpu_id ) },

       { "CPU max frequency (MHz)", Devices::SystemInfo::getCPUMaxFrequency( cpu_id ) / 1e3 },

       { "CPU cache sizes (L1d, L1i, L2, L3) (kiB)", cacheInfo },

#ifdef HAVE_CUDA

       { "GPU name", Devices::CudaDeviceInfo::getDeviceName( activeGPU ) },

       { "GPU architecture", deviceArch },

       { "GPU CUDA cores", Devices::CudaDeviceInfo::getCudaCores( activeGPU ) },

       { "GPU clock rate (MHz)", (double) Devices::CudaDeviceInfo::getClockRate( activeGPU ) / 1e3 },

       { "GPU global memory (GB)", (double) Devices::CudaDeviceInfo::getGlobalMemory( activeGPU ) / 1e9 },

       { "GPU memory clock rate (MHz)", (double) Devices::CudaDeviceInfo::getMemoryClockRate( activeGPU ) / 1e3 },

       { "GPU memory ECC enabled", Devices::CudaDeviceInfo::getECCEnabled( activeGPU ) },

#endif

   };

   Benchmark::MetadataMap metadata = getHardwareMetadata();



   // TODO: implement resolveMatrixType

//   return ! Matrices::resolveMatrixType< MainConfig,