From e6f24de189e4cb9136d6be358fa4d2c6ac4f981f Mon Sep 17 00:00:00 2001
From: Lukas Cejka <lukas.ostatek@gmail.com>
Date: Wed, 13 Mar 2019 19:05:17 +0100
Subject: [PATCH] Fixed implementation of error reporting. Made all formats be
compared with cuSPARSE (format and cuSPARSE compared to CPU, not between each
other, that would require Benchmarks.h to be changed). Added name of mtx
being tested to MetaDataColumns. Code reformatting.
---
src/Benchmarks/SpMV/spmv.h | 406 +++++++++++--------------------------
1 file changed, 115 insertions(+), 291 deletions(-)
diff --git a/src/Benchmarks/SpMV/spmv.h b/src/Benchmarks/SpMV/spmv.h
index aa4995278e..f9648a6452 100644
--- a/src/Benchmarks/SpMV/spmv.h
+++ b/src/Benchmarks/SpMV/spmv.h
@@ -34,13 +34,30 @@ namespace Benchmarks {
template< typename Real, typename Device, typename Index >
using SlicedEllpack = Matrices::SlicedEllpack< Real, Device, Index >;
+std::string getMatrixName( const String& InputFileName )
+{
+ std::string fileName = InputFileName;
+
+ // Remove directory if present.
+ // Do this before extension removal incase directory has a period character.
+ // https://stackoverflow.com/questions/8520560/get-a-file-name-from-a-path
+ // http://www.cplusplus.com/reference/string/string/find_last_of/
+ const size_t last_slash_idx = fileName.find_last_of("/\\");
+ if (std::string::npos != last_slash_idx)
+ {
+ fileName.erase(0, last_slash_idx + 1);
+ }
+
+ return fileName;
+}
+
// Get only the name of the format from getType()
template< typename Matrix >
std::string getMatrixFormat( const Matrix& matrix )
{
std::string mtrxFullType = matrix.getType();
- std::string mtrxType = mtrxFullType.substr(0, mtrxFullType.find("<"));
- std::string format = mtrxType.substr(mtrxType.find(':') + 2);
+ std::string mtrxType = mtrxFullType.substr( 0, mtrxFullType.find( "<" )) ;
+ std::string format = mtrxType.substr( mtrxType.find( ':' ) + 2 );
return format;
}
@@ -62,6 +79,45 @@ bool
benchmarkSpMV( Benchmark & benchmark,
const String & inputFileName )
{
+ // Setup CSR for cuSPARSE
+ typedef Matrices::CSR< Real, Devices::Host, int > CSR_HostMatrix;
+ typedef Matrices::CSR< Real, Devices::Cuda, int > CSR_DeviceMatrix;
+
+ CSR_HostMatrix CSRhostMatrix;
+ CSR_DeviceMatrix CSRdeviceMatrix;
+
+ // Read the matrix for CSR, to setup cuSPARSE
+ try
+ {
+ if( ! MatrixReader< CSR_HostMatrix >::readMtxFile( inputFileName, CSRhostMatrix ) )
+ {
+ benchmark.addErrorMessage( "Failed to read matrix!", 1 );
+ return false;
+ }
+ }
+ catch( std::bad_alloc )
+ {
+ benchmark.addErrorMessage( "Failed to allocate memory for matrix!", 1 );
+ return false;
+ }
+
+ // cuSPARSE handle setup
+ cusparseHandle_t cusparseHandle;
+ cusparseCreate( &cusparseHandle );
+
+#ifdef HAVE_CUDA
+ // FIXME: This doesn't work for ChunkedEllpack, because
+ // its cross-device assignment is not implemented yet
+ CSRdeviceMatrix = CSRhostMatrix;
+
+ // Delete the CSRhostMatrix, so it doesn't take up unnecessary space
+ CSRhostMatrix.reset();
+
+ TNL::CusparseCSR< Real > cusparseCSR;
+ cusparseCSR.init( CSRdeviceMatrix, &cusparseHandle );
+#endif
+
+ // Other formats setup
typedef Matrix< Real, Devices::Host, int > HostMatrix;
typedef Matrix< Real, Devices::Cuda, int > DeviceMatrix;
typedef Containers::Vector< Real, Devices::Host, int > HostVector;
@@ -73,100 +129,19 @@ benchmarkSpMV( Benchmark & benchmark,
CudaVector deviceVector, deviceVector2;
try
- {
- // Start a buffer to capture the output of MatrixReader
- std::stringstream buffer;
- std::streambuf * old = std::cerr.rdbuf( buffer.rdbuf() );
-
+ {
if( ! MatrixReader< HostMatrix >::readMtxFile( inputFileName, hostMatrix ) )
{
- // Capture the original output of MatrixReader, so it isn't printed by console.
- std::string errorMsgBuffer = buffer.str();
- // Reset the buffer
- std::cerr.rdbuf( old );
-
- // WHY DID I CAPTURE THE ERROR MESSAGE ONLY TO RUN MatrixReader again? Use the above capture to print into log and console?
-
-
- std::string matrixFormat = getMatrixFormat( hostMatrix );
-
- //https://stackoverflow.com/questions/5419356/redirect-stdout-stderr-to-a-string
- std::stringstream buffer;
- std::streambuf * old = std::cerr.rdbuf( buffer.rdbuf() );
-
- MatrixReader< HostMatrix >::readMtxFile( inputFileName, hostMatrix );
-
- errorMsgBuffer = buffer.str();
-
- // Reset the buffer
- std::cerr.rdbuf( old );
-
- std::string stringErrorMsg = "Benchmark failed: Unable to read the matrix.\n"
- "matrix format: " + matrixFormat +
- "\nFailed to read the matrix file " +
- ( std::string )inputFileName + ".\n" +
- errorMsgBuffer;
-
- //https://stackoverflow.com/questions/1488775/c-remove-new-line-from-multiline-string
- if ( ! stringErrorMsg.empty() && stringErrorMsg[ stringErrorMsg.length() - 1 ] == '\n' )
- stringErrorMsg.erase( stringErrorMsg.length() - 1 );
-
- // https://stackoverflow.com/questions/7352099/stdstring-to-char
- char* errorMsg = &stringErrorMsg[ 0u ];
-
-
- // FIXME: Every other benchmark, the errorMsg doesn't have a "!" as
- // a prefix in the log file.
- // (Try adding more benchmarks in benchmarkSpmvSynthetic(...)
- // and you'll see)
- benchmark.addErrorMessage( errorMsg, 1 );
-
- std::cout << std::endl;
-
+ benchmark.addErrorMessage( "Failed to read matrix!", 1 );
return false;
}
- std::cerr.rdbuf( old );
}
catch( std::bad_alloc )
{
- std::string matrixFormat = getMatrixFormat( hostMatrix );
-
- //https://stackoverflow.com/questions/5419356/redirect-stdout-stderr-to-a-string
- std::stringstream buffer;
- std::streambuf * old = std::cerr.rdbuf(buffer.rdbuf());
-
- MatrixReader< HostMatrix >::readMtxFile( inputFileName, hostMatrix );
-
- std::string errorMsgBuffer = buffer.str();
-
- // Reset the buffer
- std::cerr.rdbuf( old );
-
- std::string stringErrorMsg = "Benchmark failed: Not enough memory.\n"
- "matrix format: " + matrixFormat +
- "\nFailed to allocate memory to read the matrix file " +
- ( std::string )inputFileName + ".\n" +
- errorMsgBuffer;
-
- //https://stackoverflow.com/questions/1488775/c-remove-new-line-from-multiline-string
- if ( ! stringErrorMsg.empty() && stringErrorMsg[ stringErrorMsg.length() - 1 ] == '\n' )
- stringErrorMsg.erase( stringErrorMsg.length() - 1 );
-
- // https://stackoverflow.com/questions/7352099/stdstring-to-char
- char *errorMsg = &stringErrorMsg[ 0u ];
-
- // FIXME: Every other benchmark, the errorMsg doesn't have a "!" as
- // a prefix in the log file.
- // (Try adding more benchmarks in benchmarkSpmvSynthetic(...)
- // and you'll see)
- benchmark.addErrorMessage( errorMsg, 1 );
-
- std::cout << std::endl;
-
+ benchmark.addErrorMessage( "Failed to allocate memory for matrix!", 1 );
return false;
}
- // printMatrixInfo is redundant, because all the information is in the Benchmark's MetadataColumns
-// printMatrixInfo( hostMatrix, std::cout );
+
#ifdef HAVE_CUDA
// FIXME: This doesn't work for ChunkedEllpack, because
// its cross-device assignment is not implemented yet
@@ -175,6 +150,7 @@ benchmarkSpMV( Benchmark & benchmark,
benchmark.setMetadataColumns( Benchmark::MetadataColumns({
{ "matrix format", convertToString( getMatrixFormat( hostMatrix ) ) },
+ { "matrix name", convertToString( getMatrixName( inputFileName ) ) },
{ "non-zeros", convertToString( hostMatrix.getNumberOfNonzeroMatrixElements() ) },
{ "rows", convertToString( hostMatrix.getRows() ) },
{ "columns", convertToString( hostMatrix.getColumns() ) }
@@ -209,6 +185,9 @@ benchmarkSpMV( Benchmark & benchmark,
auto spmvCuda = [&]() {
deviceMatrix.vectorProduct( deviceVector, deviceVector2 );
};
+ auto spmvCusparse = [&]() {
+ cusparseCSR.vectorProduct( deviceVector, deviceVector2 );
+ };
benchmark.setOperation( datasetSize );
benchmark.time< Devices::Host >( reset, "CPU", spmvHost );
@@ -217,234 +196,83 @@ benchmarkSpMV( Benchmark & benchmark,
HostVector resultHostVector2;
resultHostVector2.setSize( hostVector2.getSize() );
resultHostVector2.setValue( 0.0 );
+
// Copy the values
- for( int i = 0; i < hostVector2.getSize(); i++ )
- resultHostVector2.setElement( i, hostVector2.getElement( i ) );
+ resultHostVector2 = hostVector2;
#ifdef HAVE_CUDA
benchmark.time< Devices::Cuda >( reset, "GPU", spmvCuda );
- #endif
// Setup the device vector to be compared
HostVector resultDeviceVector2;
- resultDeviceVector2.setSize( hostVector2.getSize() );
+ resultDeviceVector2.setSize( deviceVector2.getSize() );
resultDeviceVector2.setValue( 0.0 );
-// resultDeviceVector2 += deviceVector2; // Throws a segfault.
-
- // Copy the values
- for( int i = 0; i < deviceVector2.getSize(); i++ )
- resultDeviceVector2.setElement( i, deviceVector2.getElement( i ) );
-
- Real differenceAbsMax = resultHostVector2.differenceAbsMax( resultDeviceVector2 );
- Real differenceLpNorm = resultHostVector2.differenceLpNorm( resultDeviceVector2, 1 );
-
- std::string resultDifferenceAbsMax = "differenceAbsMax = " + std::to_string( differenceAbsMax );
- std::string resultDifferenceLpNorm = "differenceLpNorm = " + std::to_string( differenceLpNorm );
-
- char *absMax = &resultDifferenceAbsMax[ 0u ];
- char *lpNorm = &resultDifferenceLpNorm[ 0u ];
-
- // FIXME: THIS ISN'T AN ELEGANT SOLUTION, IT MAKES THE LOG FILE VERY LONG
-// benchmark.addErrorMessage( absMax, 1 );
-// benchmark.addErrorMessage( lpNorm, 1 );
-
- std::cout << std::endl;
- return true;
-}
-
-// Compares only CSR on GPU and Cusparse on GPU.
-template< typename Real,
- template< typename, typename, typename > class Vector = Containers::Vector >
-bool
-benchmarkCusparseSpMV( Benchmark & benchmark,
- const String & inputFileName )
-{
- typedef Matrices::CSR< Real, Devices::Host, int > CSR_HostMatrix;
- typedef Matrices::CSR< Real, Devices::Cuda, int > CSR_DeviceMatrix;
- typedef Containers::Vector< Real, Devices::Host, int > HostVector;
- typedef Containers::Vector< Real, Devices::Cuda, int > CudaVector;
-
- CSR_HostMatrix CSRhostMatrix;
- CSR_DeviceMatrix CSRdeviceMatrix;
- CudaVector deviceVector, deviceVector2;
-
- try
- {
- // Start a buffer to capture the output of MatrixReader
- std::stringstream buffer;
- std::streambuf * old = std::cerr.rdbuf( buffer.rdbuf() );
-
- if( ! MatrixReader< CSR_HostMatrix >::readMtxFile( inputFileName, CSRhostMatrix ) )
- {
- // Capture the original output of MatrixReader, so it isn't printed by console.
- std::string errorMsgBuffer = buffer.str();
- // Reset the buffer
- std::cerr.rdbuf( old );
-
-
- std::string matrixFormat = getMatrixFormat( CSRhostMatrix );
-
- //https://stackoverflow.com/questions/5419356/redirect-stdout-stderr-to-a-string
- std::stringstream buffer;
- std::streambuf * old = std::cerr.rdbuf( buffer.rdbuf() );
-
- MatrixReader< CSR_HostMatrix >::readMtxFile( inputFileName, CSRhostMatrix );
-
- errorMsgBuffer = buffer.str();
-
- // Reset the buffer
- std::cerr.rdbuf( old );
-
- std::string stringErrorMsg = "Benchmark failed: Unable to read the matrix.\n"
- "matrix format: " + matrixFormat +
- "\nFailed to read the matrix file " +
- ( std::string )inputFileName + ".\n" +
- errorMsgBuffer;
-
- //https://stackoverflow.com/questions/1488775/c-remove-new-line-from-multiline-string
- if ( ! stringErrorMsg.empty() && stringErrorMsg[ stringErrorMsg.length() - 1 ] == '\n' )
- stringErrorMsg.erase( stringErrorMsg.length() - 1 );
-
- // https://stackoverflow.com/questions/7352099/stdstring-to-char
- char* errorMsg = &stringErrorMsg[ 0u ];
-
-
- // FIXME: Every other benchmark, the errorMsg doesn't have a "!" as
- // a prefix in the log file.
- // (Try adding more benchmarks in benchmarkSpmvSynthetic(...)
- // and you'll see)
- benchmark.addErrorMessage( errorMsg, 1 );
-
- std::cout << std::endl;
-
- return false;
- }
- std::cerr.rdbuf( old );
- }
- catch( std::bad_alloc )
- {
- std::string matrixFormat = getMatrixFormat( CSRhostMatrix );
-
- //https://stackoverflow.com/questions/5419356/redirect-stdout-stderr-to-a-string
- std::stringstream buffer;
- std::streambuf * old = std::cerr.rdbuf(buffer.rdbuf());
-
- MatrixReader< CSR_HostMatrix >::readMtxFile( inputFileName, CSRhostMatrix );
-
- std::string errorMsgBuffer = buffer.str();
-
- // Reset the buffer
- std::cerr.rdbuf( old );
-
- std::string stringErrorMsg = "Benchmark failed: Not enough memory.\n"
- "matrix format: " + matrixFormat +
- "\nFailed to allocate memory to read the matrix file " +
- ( std::string )inputFileName + ".\n" +
- errorMsgBuffer;
-
- //https://stackoverflow.com/questions/1488775/c-remove-new-line-from-multiline-string
- if ( ! stringErrorMsg.empty() && stringErrorMsg[ stringErrorMsg.length() - 1 ] == '\n' )
- stringErrorMsg.erase( stringErrorMsg.length() - 1 );
-
- // https://stackoverflow.com/questions/7352099/stdstring-to-char
- char *errorMsg = &stringErrorMsg[ 0u ];
-
- // FIXME: Every other benchmark, the errorMsg doesn't have a "!" as
- // a prefix in the log file.
- // (Try adding more benchmarks in benchmarkSpmvSynthetic(...)
- // and you'll see)
- benchmark.addErrorMessage( errorMsg, 1 );
-
- std::cout << std::endl;
-
- return false;
- }
+ resultDeviceVector2 = deviceVector2;
+#endif
+ // FIXME: How to include benchmark with different name under the same header as the current format being benchmarked???
+ // FIXME: Does it matter that speedup show difference only between current test and first test?
+ // Speedup shows difference between CPU and GPU-cuSPARSE, because in Benchmarks.h:
+ // * If there is no baseTime, the resulting test time is set to baseTime.
+ // * However, if there is a baseTime (from the CPU compared to GPU test),
+ // baseTime isn't changed. If we change it in Benchmarks.h to compare
+ // the speedup from the last test, it will mess up BLAS benchmarks etc.
benchmark.setMetadataColumns( Benchmark::MetadataColumns({
- { "matrix format", convertToString( getMatrixFormat( CSRhostMatrix ) ) },
- { "non-zeros", convertToString( CSRhostMatrix.getNumberOfNonzeroMatrixElements() ) },
- { "rows", convertToString( CSRhostMatrix.getRows() ) },
- { "columns", convertToString( CSRhostMatrix.getColumns() ) }
+ { "matrix format", convertToString( "CSR-cuSPARSE" ) },
+ { "matrix name", convertToString( getMatrixName( inputFileName ) ) },
+ { "non-zeros", convertToString( hostMatrix.getNumberOfNonzeroMatrixElements() ) },
+ { "rows", convertToString( hostMatrix.getRows() ) },
+ { "columns", convertToString( hostMatrix.getColumns() ) }
} ));
- cusparseHandle_t cusparseHandle;
- cusparseCreate( &cusparseHandle );
-
#ifdef HAVE_CUDA
- // FIXME: This doesn't work for ChunkedEllpack, because
- // its cross-device assignment is not implemented yet
- CSRdeviceMatrix = CSRhostMatrix;
+ benchmark.time< Devices::Cuda >( reset, "GPU-Cusparse", spmvCusparse );
- TNL::CusparseCSR< Real > cusparseCSR;
- cusparseCSR.init( CSRdeviceMatrix, &cusparseHandle );
-#endif
-
-#ifdef HAVE_CUDA
- deviceVector.setSize( CSRhostMatrix.getColumns() );
- deviceVector2.setSize( CSRhostMatrix.getRows() );
-#endif
-
- // reset function
- auto reset = [&]() {
- #ifdef HAVE_CUDA
- deviceVector.setValue( 1.0 );
- deviceVector2.setValue( 0.0 );
+ HostVector resultcuSPARSEDeviceVector2;
+ resultcuSPARSEDeviceVector2.setSize( deviceVector2.getSize() );
+ resultcuSPARSEDeviceVector2.setValue( 0.0 );
+
+ resultcuSPARSEDeviceVector2 = deviceVector2;
#endif
- };
-
- const int elements = CSRhostMatrix.getNumberOfNonzeroMatrixElements();
-
- const double datasetSize = (double) elements * ( 2 * sizeof( Real ) + sizeof( int ) ) / oneGB;
-
- // compute functions
- auto spmvCuda = [&]() {
- CSRdeviceMatrix.vectorProduct( deviceVector, deviceVector2 );
- };
- auto spmvCusparse = [&]() {
- cusparseCSR.vectorProduct( deviceVector, deviceVector2 );
- };
-
- benchmark.setOperation( datasetSize );
- #ifdef HAVE_CUDA
- benchmark.time< Devices::Cuda >( reset, "GPU", spmvCuda );
+#ifdef RESULTS
+ // Difference between GPU (curent format) and GPU-cuSPARSE results
+ Real cuSPARSEdifferenceAbsMax = resultDeviceVector2.differenceAbsMax( resultcuSPARSEDeviceVector2 );
+ Real cuSPARSEdifferenceLpNorm = resultDeviceVector2.differenceLpNorm( resultcuSPARSEDeviceVector2, 1 );
- // Initialize the cuda vector to be compared. (The values in hostVector2 will be reset when spmvCuda starts)
- HostVector resultCusparseVector2;
- resultCusparseVector2.setSize( deviceVector2.getSize() );
- resultCusparseVector2.setValue( 0.0 );
- // Copy the values
- for( int i = 0; i < deviceVector2.getSize(); i++ )
- resultCusparseVector2.setElement( i, deviceVector2.getElement( i ) );
+ std::string GPUxGPUcuSPARSE_resultDifferenceAbsMax = "GPUxGPUcuSPARSE differenceAbsMax = " + std::to_string( cuSPARSEdifferenceAbsMax );
+ std::string GPUxGPUcuSPARSE_resultDifferenceLpNorm = "GPUxGPUcuSPARSE differenceLpNorm = " + std::to_string( cuSPARSEdifferenceLpNorm );
- benchmark.time< Devices::Cuda >( reset, "GPU-Cusparse", spmvCusparse );
- #endif
-
- // Setup the device vector to be compared
- HostVector resultDeviceVector2;
- resultDeviceVector2.setSize( resultCusparseVector2.getSize() );
- resultDeviceVector2.setValue( 0.0 );
+ char *GPUcuSPARSE_absMax = &GPUxGPUcuSPARSE_resultDifferenceAbsMax[ 0u ];
+ char *GPUcuSPARSE_lpNorm = &GPUxGPUcuSPARSE_resultDifferenceLpNorm[ 0u ];
- // Copy the values
- for( int i = 0; i < deviceVector2.getSize(); i++ )
- resultDeviceVector2.setElement( i, deviceVector2.getElement( i ) );
- Real differenceAbsMax = resultCusparseVector2.differenceAbsMax( resultDeviceVector2 );
- Real differenceLpNorm = resultCusparseVector2.differenceLpNorm( resultDeviceVector2, 1 );
+ // Difference between CPU and GPU results for the current format
+ Real differenceAbsMax = resultHostVector2.differenceAbsMax( resultDeviceVector2 );
+ Real differenceLpNorm = resultHostVector2.differenceLpNorm( resultDeviceVector2, 1 );
- std::string resultDifferenceAbsMax = "differenceAbsMax = " + std::to_string( differenceAbsMax );
- std::string resultDifferenceLpNorm = "differenceLpNorm = " + std::to_string( differenceLpNorm );
+ std::string CPUxGPU_resultDifferenceAbsMax = "CPUxGPU differenceAbsMax = " + std::to_string( differenceAbsMax );
+ std::string CPUxGPU_resultDifferenceLpNorm = "CPUxGPU differenceLpNorm = " + std::to_string( differenceLpNorm );
- char *absMax = &resultDifferenceAbsMax[ 0u ];
- char *lpNorm = &resultDifferenceLpNorm[ 0u ];
+ char *CPUxGPU_absMax = &CPUxGPU_resultDifferenceAbsMax[ 0u ];
+ char *CPUxGPU_lpNorm = &CPUxGPU_resultDifferenceLpNorm[ 0u ];
+
+ // Print result differences of CPU and GPU of current format
+ std::cout << CPUxGPU_absMax << std::endl;
+ std::cout << CPUxGPU_lpNorm << std::endl;
+
+ // Print result differences of GPU of current format and GPU with cuSPARSE.
+ std::cout << GPUcuSPARSE_absMax << std::endl;
+ std::cout << GPUcuSPARSE_lpNorm << std::endl;
// FIXME: THIS ISN'T AN ELEGANT SOLUTION, IT MAKES THE LOG FILE VERY LONG
// benchmark.addErrorMessage( absMax, 1 );
// benchmark.addErrorMessage( lpNorm, 1 );
+#endif
+
std::cout << std::endl;
- cusparseDestroy( cusparseHandle );
return true;
}
@@ -456,11 +284,7 @@ benchmarkSpmvSynthetic( Benchmark & benchmark,
{
bool result = true;
// TODO: benchmark all formats from tnl-benchmark-spmv (different parameters of the base formats)
- result |= benchmarkSpMV< Real, Matrices::CSR >( benchmark, inputFileName );
-
- // This doesn't have a titles (matrix format, rows, cols, etc.) in the output, because the header is the same as before (CSR).
- result |= benchmarkCusparseSpMV< Real, Matrices::CSR >( benchmark, inputFileName );
-
+ result |= benchmarkSpMV< Real, Matrices::CSR >( benchmark, inputFileName );
result |= benchmarkSpMV< Real, Matrices::Ellpack >( benchmark, inputFileName );
result |= benchmarkSpMV< Real, SlicedEllpack >( benchmark, inputFileName );
// result |= benchmarkSpMV< Real, Matrices::ChunkedEllpack >( benchmark, inputFileName );
--
GitLab