Skip to content

Fix estimated counts in various kernels #566

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
MrBurmark merged 1 commit into from
Oct 22, 2025
Merged

Fix estimated counts in various kernels #566

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion src/algorithm/HISTOGRAM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,7 +38,7 @@ HISTOGRAM::HISTOGRAM(const RunParams& params)
1*sizeof(Index_type) * getActualProblemSize() );
setBytesWrittenPerRep( 1*sizeof(Data_type) * m_num_bins );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(1 * getActualProblemSize());
setFLOPsPerRep( (std::is_floating_point_v<Data_type> ? 1 : 0) * getActualProblemSize() );

setComplexity(Complexity::N);

Expand Down
4 changes: 2 additions & 2 deletions src/apps/CONVECTION3DPA.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,13 +32,13 @@ CONVECTION3DPA::CONVECTION3DPA(const RunParams& params)

setActualProblemSize( m_NE*CPA_Q1D*CPA_Q1D*CPA_Q1D );

setItsPerRep(getActualProblemSize());
setItsPerRep( m_NE*CPA_Q1D*CPA_Q1D*CPA_Q1D );
setKernelsPerRep(1);

setBytesReadPerRep( 3*sizeof(Real_type) * CPA_Q1D*CPA_D1D + // b, bt, g
2*sizeof(Real_type) * CPA_D1D*CPA_D1D*CPA_D1D*m_NE + // x, y
CPA_VDIM*sizeof(Real_type) * CPA_Q1D*CPA_Q1D*CPA_Q1D*m_NE ); // d
setBytesWrittenPerRep( 1*sizeof(Real_type) + CPA_D1D*CPA_D1D*CPA_D1D*m_NE ); // y
setBytesWrittenPerRep( 1*sizeof(Real_type) * CPA_D1D*CPA_D1D*CPA_D1D*m_NE ); // y
setBytesAtomicModifyWrittenPerRep( 0 );

setFLOPsPerRep(m_NE * (
Expand Down
2 changes: 1 addition & 1 deletion src/apps/DIFFUSION3DPA.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ DIFFUSION3DPA::DIFFUSION3DPA(const RunParams& params)

setActualProblemSize( m_NE*DPA_Q1D*DPA_Q1D*DPA_Q1D );

setItsPerRep(getActualProblemSize());
setItsPerRep( m_NE*DPA_Q1D*DPA_Q1D*DPA_Q1D );
setKernelsPerRep(1);

setBytesReadPerRep( 2*sizeof(Real_type) * DPA_Q1D*DPA_D1D + // b, g
Expand Down
2 changes: 1 addition & 1 deletion src/apps/LTIMES.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ LTIMES::LTIMES(const RunParams& params)

setActualProblemSize( m_psilen );

setItsPerRep( getActualProblemSize() );
setItsPerRep( m_philen );
setKernelsPerRep(1);
// using total data size instead of writes and reads
setBytesReadPerRep( 1*sizeof(Real_type) * m_philen +
Expand Down
2 changes: 1 addition & 1 deletion src/apps/LTIMES_NOVIEW.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ LTIMES_NOVIEW::LTIMES_NOVIEW(const RunParams& params)

setActualProblemSize( m_psilen );

setItsPerRep( getActualProblemSize() );
setItsPerRep( m_philen );
setKernelsPerRep(1);
// using total data size instead of writes and reads
setBytesReadPerRep( 1*sizeof(Real_type) * m_philen +
Expand Down
6 changes: 3 additions & 3 deletions src/apps/MASS3DEA.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,13 +28,13 @@ MASS3DEA::MASS3DEA(const RunParams& params)
setDefaultProblemSize(m_NE_default*MEA_Q1D*MEA_Q1D*MEA_Q1D);
setDefaultReps(1);

const int ea_mat_entries = MEA_D1D*MEA_D1D*MEA_D1D*MEA_D1D*MEA_D1D*MEA_D1D;
const Index_type ea_mat_entries = MEA_D1D*MEA_D1D*MEA_D1D*MEA_D1D*MEA_D1D*MEA_D1D;

m_NE = std::max((getTargetProblemSize() + (ea_mat_entries)/2) / (ea_mat_entries), Index_type(1));

setActualProblemSize( m_NE*ea_mat_entries);
setActualProblemSize( m_NE*ea_mat_entries );

setItsPerRep(getActualProblemSize());
setItsPerRep( m_NE*MEA_Q1D*MEA_Q1D*MEA_Q1D );
setKernelsPerRep(1);

setBytesReadPerRep( 1*sizeof(Real_type) * MEA_Q1D*MEA_D1D + // B
Expand Down
2 changes: 1 addition & 1 deletion src/apps/MASS3DPA.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ MASS3DPA::MASS3DPA(const RunParams& params)

setActualProblemSize( m_NE*MPA_Q1D*MPA_Q1D*MPA_Q1D );

setItsPerRep(getActualProblemSize());
setItsPerRep( m_NE*MPA_Q1D*MPA_Q1D );
setKernelsPerRep(1);

setBytesReadPerRep( 2*sizeof(Real_type) * MPA_Q1D*MPA_D1D + // B, Bt
Expand Down
7 changes: 3 additions & 4 deletions src/apps/NODAL_ACCUMULATION_3D.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,10 +40,9 @@ NODAL_ACCUMULATION_3D::NODAL_ACCUMULATION_3D(const RunParams& params)
setKernelsPerRep(1);
// touched data size, not actual number of stores and loads
setBytesReadPerRep( 1*sizeof(Index_type) * getItsPerRep() +
1*sizeof(Real_type) * getItsPerRep() +
1*sizeof(Real_type) * m_domain->n_real_nodes);
setBytesWrittenPerRep( 1*sizeof(Real_type) * m_domain->n_real_nodes );
setBytesAtomicModifyWrittenPerRep( 0 );
1*sizeof(Real_type) * getItsPerRep() );
setBytesWrittenPerRep( 0 );
setBytesAtomicModifyWrittenPerRep( 1*sizeof(Real_type) * m_domain->n_real_nodes );
setFLOPsPerRep(9 * getItsPerRep());

checksum_scale_factor = 0.001 *
Expand Down
4 changes: 3 additions & 1 deletion src/basic/IF_QUAD.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,9 @@ IF_QUAD::IF_QUAD(const RunParams& params)
setBytesReadPerRep( 3*sizeof(Real_type) * getActualProblemSize() );
setBytesWrittenPerRep( 2*sizeof(Real_type) * getActualProblemSize() );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(11 * getActualProblemSize()); // 1 sqrt
// estimate conditional true half of the time, 1 sqrt
setFLOPsPerRep(4 * getActualProblemSize() +
7 * getActualProblemSize() / 2);

checksum_scale_factor = 0.0001 *
( static_cast<Checksum_type>(getDefaultProblemSize()) /
Expand Down
7 changes: 3 additions & 4 deletions src/basic/MAT_MAT_SHARED.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,19 +25,18 @@ MAT_MAT_SHARED::MAT_MAT_SHARED(const RunParams &params)
setDefaultReps(5);

m_N = std::sqrt(getTargetProblemSize()) + std::sqrt(2)-1;
const Index_type num_tiles = RAJA_DIVIDE_CEILING_INT(m_N, TL_SZ);

setActualProblemSize(m_N * m_N);

setItsPerRep(getActualProblemSize());
setItsPerRep( num_tiles*num_tiles * TL_SZ*TL_SZ );
setKernelsPerRep(1);

setBytesReadPerRep( 2*sizeof(Real_type) * m_N*m_N );
setBytesWrittenPerRep( 1*sizeof(Real_type) * m_N*m_N );
setBytesAtomicModifyWrittenPerRep( 0 );

const Index_type no_tiles = RAJA_DIVIDE_CEILING_INT(m_N, TL_SZ);
const Index_type no_blocks = RAJA_DIVIDE_CEILING_INT(m_N, TL_SZ);
setFLOPsPerRep(2 * TL_SZ * TL_SZ * TL_SZ * no_tiles * no_blocks * no_blocks);
setFLOPsPerRep(2 * TL_SZ * TL_SZ * TL_SZ * num_tiles * num_tiles * num_tiles);

checksum_scale_factor = 1e-6 *
( static_cast<Checksum_type>(getDefaultProblemSize()) /
Expand Down
2 changes: 1 addition & 1 deletion src/basic/REDUCE3_INT.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,7 +37,7 @@ REDUCE3_INT::REDUCE3_INT(const RunParams& params)
1*sizeof(Int_type) * getActualProblemSize() );
setBytesWrittenPerRep( 3*sizeof(Int_type) );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(1 * getActualProblemSize() + 1);
setFLOPsPerRep(0);

setComplexity(Complexity::N);

Expand Down
19 changes: 10 additions & 9 deletions src/comm/HALO_EXCHANGE.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,21 +28,22 @@ HALO_EXCHANGE::HALO_EXCHANGE(const RunParams& params)

m_num_vars = params.getHaloNumVars();
m_var_size = m_grid_plus_halo_size ;
const Size_type halo_size = m_var_size - getActualProblemSize();

setItsPerRep( m_num_vars * (m_var_size - getActualProblemSize()) );
setItsPerRep( 2 * m_num_vars * halo_size );
setKernelsPerRep( 2 * s_num_neighbors * m_num_vars );
setBytesReadPerRep( 1*sizeof(Int_type) * getItsPerRep() + // pack
1*sizeof(Real_type) * getItsPerRep() + // pack
setBytesReadPerRep( 1*sizeof(Int_type) * m_num_vars * halo_size + // pack
1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Real_type) * getItsPerRep() + // send
1*sizeof(Real_type) * m_num_vars * halo_size + // send

1*sizeof(Int_type) * getItsPerRep() + // unpack
1*sizeof(Real_type) * getItsPerRep() ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * getItsPerRep() + // pack
1*sizeof(Int_type) * m_num_vars * halo_size + // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Real_type) * getItsPerRep() + // recv
1*sizeof(Real_type) * m_num_vars * halo_size + // recv

1*sizeof(Real_type) * getItsPerRep() ); // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(0);

Expand Down
19 changes: 10 additions & 9 deletions src/comm/HALO_EXCHANGE_FUSED.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,21 +28,22 @@ HALO_EXCHANGE_FUSED::HALO_EXCHANGE_FUSED(const RunParams& params)

m_num_vars = params.getHaloNumVars();
m_var_size = m_grid_plus_halo_size ;
const Size_type halo_size = m_var_size - getActualProblemSize();

setItsPerRep( m_num_vars * (m_var_size - getActualProblemSize()) );
setItsPerRep( 2 * m_num_vars * halo_size );
setKernelsPerRep( 2 );
setBytesReadPerRep( 1*sizeof(Int_type) * getItsPerRep() + // pack
1*sizeof(Real_type) * getItsPerRep() + // pack
setBytesReadPerRep( 1*sizeof(Int_type) * m_num_vars * halo_size + // pack
1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Real_type) * getItsPerRep() + // send
1*sizeof(Real_type) * m_num_vars * halo_size + // send

1*sizeof(Int_type) * getItsPerRep() + // unpack
1*sizeof(Real_type) * getItsPerRep() ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * getItsPerRep() + // pack
1*sizeof(Int_type) * m_num_vars * halo_size + // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Real_type) * getItsPerRep() + // recv
1*sizeof(Real_type) * m_num_vars * halo_size + // recv

1*sizeof(Real_type) * getItsPerRep() ); // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(0);

Expand Down
15 changes: 8 additions & 7 deletions src/comm/HALO_PACKING.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,17 +22,18 @@ HALO_PACKING::HALO_PACKING(const RunParams& params)

m_num_vars = params.getHaloNumVars();
m_var_size = m_grid_plus_halo_size ;
const Size_type halo_size = m_var_size - getActualProblemSize();

setItsPerRep( m_num_vars * (m_var_size - getActualProblemSize()) );
setItsPerRep( 2 * m_num_vars * halo_size );
setKernelsPerRep( 2 * s_num_neighbors * m_num_vars );
setBytesReadPerRep( 1*sizeof(Int_type) * getItsPerRep() + // pack
1*sizeof(Real_type) * getItsPerRep() + // pack
setBytesReadPerRep( 1*sizeof(Int_type) * m_num_vars * halo_size + // pack
1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Int_type) * getItsPerRep() + // unpack
1*sizeof(Real_type) * getItsPerRep() ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * getItsPerRep() + // pack
1*sizeof(Int_type) * m_num_vars * halo_size + // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Real_type) * getItsPerRep() ); // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(0);

Expand Down
15 changes: 8 additions & 7 deletions src/comm/HALO_PACKING_FUSED.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,17 +22,18 @@ HALO_PACKING_FUSED::HALO_PACKING_FUSED(const RunParams& params)

m_num_vars = params.getHaloNumVars();
m_var_size = m_grid_plus_halo_size ;
const Size_type halo_size = m_var_size - getActualProblemSize();

setItsPerRep( m_num_vars * (m_var_size - getActualProblemSize()) );
setItsPerRep( 2 * m_num_vars * halo_size );
setKernelsPerRep( 2 );
setBytesReadPerRep( 1*sizeof(Int_type) * getItsPerRep() + // pack
1*sizeof(Real_type) * getItsPerRep() + // pack
setBytesReadPerRep( 1*sizeof(Int_type) * m_num_vars * halo_size + // pack
1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Int_type) * getItsPerRep() + // unpack
1*sizeof(Real_type) * getItsPerRep() ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * getItsPerRep() + // pack
1*sizeof(Int_type) * m_num_vars * halo_size + // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesWrittenPerRep( 1*sizeof(Real_type) * m_num_vars * halo_size + // pack

1*sizeof(Real_type) * getItsPerRep() ); // unpack
1*sizeof(Real_type) * m_num_vars * halo_size ); // unpack
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(0);

Expand Down
7 changes: 4 additions & 3 deletions src/comm/HALO_SENDRECV.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,11 +28,12 @@ HALO_SENDRECV::HALO_SENDRECV(const RunParams& params)

m_num_vars = params.getHaloNumVars();
m_var_size = m_grid_plus_halo_size ;
const Size_type halo_size = m_var_size - getActualProblemSize();

setItsPerRep( m_num_vars * (m_var_size - getActualProblemSize()) );
setItsPerRep( 0 );
setKernelsPerRep( 0 );
setBytesReadPerRep( 1*sizeof(Real_type) * getItsPerRep() ); // send
setBytesWrittenPerRep( 1*sizeof(Real_type) * getItsPerRep() ); // recv
setBytesReadPerRep( 1*sizeof(Real_type) * m_num_vars * halo_size ); // send
setBytesWrittenPerRep( 1*sizeof(Real_type) * m_num_vars * halo_size ); // recv
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(0);

Expand Down
2 changes: 1 addition & 1 deletion src/lcals/EOS.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ EOS::EOS(const RunParams& params)

setActualProblemSize( getTargetProblemSize() );

m_array_length = getActualProblemSize() + 7;
m_array_length = getActualProblemSize() + 6;

setItsPerRep( getActualProblemSize() );
setItsPerRep( getActualProblemSize() );
Expand Down
10 changes: 5 additions & 5 deletions src/lcals/FIRST_SUM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,16 +24,16 @@ FIRST_SUM::FIRST_SUM(const RunParams& params)
setDefaultProblemSize(1000000);
setDefaultReps(2000);

setActualProblemSize( getTargetProblemSize() );
setActualProblemSize( std::max(getTargetProblemSize(), Index_type(2)) );

m_N = getActualProblemSize();

setItsPerRep( getActualProblemSize() );
setItsPerRep( m_N-1 );
setKernelsPerRep(1);
setBytesReadPerRep( 1*sizeof(Real_type ) * (m_N-1) );
setBytesWrittenPerRep( 1*sizeof(Real_type ) * m_N );
setBytesReadPerRep( 1*sizeof(Real_type ) * m_N );
setBytesWrittenPerRep( 1*sizeof(Real_type ) * (m_N-1) );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(1 * (getActualProblemSize()-1));
setFLOPsPerRep(1 * (m_N-1));

setComplexity(Complexity::N);

Expand Down
4 changes: 2 additions & 2 deletions src/lcals/GEN_LIN_RECUR.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,15 +28,15 @@ GEN_LIN_RECUR::GEN_LIN_RECUR(const RunParams& params)

m_N = getActualProblemSize();

setItsPerRep( getActualProblemSize() );
setItsPerRep( 2 * m_N );
setKernelsPerRep(2);
setBytesReadPerRep( 3*sizeof(Real_type ) * m_N +
3*sizeof(Real_type ) * m_N );
setBytesWrittenPerRep( 2*sizeof(Real_type ) * m_N +
2*sizeof(Real_type ) * m_N );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep((3 +
3 ) * getActualProblemSize());
3 ) * m_N);

checksum_scale_factor = 0.01 *
( static_cast<Checksum_type>(getDefaultProblemSize()) /
Expand Down
14 changes: 10 additions & 4 deletions src/lcals/HYDRO_2D.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,15 +36,21 @@ HYDRO_2D::HYDRO_2D(const RunParams& params)
m_jn = m_kn = std::sqrt(getTargetProblemSize()) + std::sqrt(2)-1;
m_array_length = m_kn * m_jn;

setActualProblemSize( getTargetProblemSize() );
setActualProblemSize( m_array_length );

setItsPerRep( 3 * getActualProblemSize() );
setItsPerRep( 3 * (m_kn-2) * (m_jn-2) );
setKernelsPerRep(3);
setBytesReadPerRep( 4*sizeof(Real_type ) * m_array_length +
4*sizeof(Real_type ) * m_array_length +
setBytesReadPerRep( 4*sizeof(Real_type ) * ((m_kn-1) * (m_jn-1) - 1) +

2*sizeof(Real_type ) * (m_kn-2) * (m_jn-2) +
2*sizeof(Real_type ) * (m_kn-2) * (m_jn-1) +
2*sizeof(Real_type ) * ((m_kn) * (m_jn) - 4) +

4*sizeof(Real_type ) * (m_kn-2) * (m_jn-2) );
setBytesWrittenPerRep( 2*sizeof(Real_type ) * (m_kn-2) * (m_jn-2) +

2*sizeof(Real_type ) * (m_kn-2) * (m_jn-2) +

2*sizeof(Real_type ) * (m_kn-2) * (m_jn-2) );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep((14 +
Expand Down
8 changes: 4 additions & 4 deletions src/lcals/TRIDIAG_ELIM.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,16 +24,16 @@ TRIDIAG_ELIM::TRIDIAG_ELIM(const RunParams& params)
setDefaultProblemSize(1000000);
setDefaultReps(1000);

setActualProblemSize( getTargetProblemSize() );
setActualProblemSize( std::max(getTargetProblemSize(), Index_type(2)) );

m_N = getActualProblemSize() + 1;
m_N = getActualProblemSize();

setItsPerRep( getActualProblemSize() );
setItsPerRep( m_N-1 );
setKernelsPerRep(1);
setBytesReadPerRep( 3*sizeof(Real_type ) * (m_N-1) );
setBytesWrittenPerRep( 1*sizeof(Real_type ) * (m_N-1) );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep(2 * (getActualProblemSize()-1));
setFLOPsPerRep(2 * (m_N-1));

setComplexity(Complexity::N);

Expand Down
20 changes: 13 additions & 7 deletions src/polybench/POLYBENCH_ADI.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,19 +28,25 @@ POLYBENCH_ADI::POLYBENCH_ADI(const RunParams& params)

m_n = std::sqrt( getTargetProblemSize() ) + 2 + std::sqrt(2)-1;

setItsPerRep( (m_n-2) + (m_n-2) );
setItsPerRep( 2 * (m_n-2) + (m_n-2) );


setActualProblemSize( (m_n-2) * (m_n-2) );

setKernelsPerRep( 2 );
setBytesReadPerRep( 3*sizeof(Real_type ) * m_n * (m_n-2) +
3*sizeof(Real_type ) * m_n * (m_n-2) );
setBytesWrittenPerRep( 3*sizeof(Real_type ) * m_n * (m_n-2) +
3*sizeof(Real_type ) * m_n * (m_n-2) );
setBytesReadPerRep( 1*sizeof(Real_type ) * (m_n-2) * (m_n ) +
2*sizeof(Real_type ) * (m_n-2) * (m_n-2) +

1*sizeof(Real_type ) * (m_n-2) * (m_n ) +
2*sizeof(Real_type ) * (m_n-2) * (m_n-2) );
setBytesWrittenPerRep( 2*sizeof(Real_type ) * (m_n-2) * (m_n-1) +
1*sizeof(Real_type ) * (m_n-2) * (m_n ) +

2*sizeof(Real_type ) * (m_n-2) * (m_n-1) +
1*sizeof(Real_type ) * (m_n-2) * (m_n ) );
setBytesAtomicModifyWrittenPerRep( 0 );
setFLOPsPerRep( (15 + 2) * (m_n-2)*(m_n-2) +
(15 + 2) * (m_n-2)*(m_n-2) );
setFLOPsPerRep( (13 + 2) * (m_n-2)*(m_n-2) +
(13 + 2) * (m_n-2)*(m_n-2) );

checksum_scale_factor = 0.0000001 *
( static_cast<Checksum_type>(getDefaultProblemSize()) /
Expand Down
Loading