From 0a83ab329a14d6d36ca3d8b6ea25d4c528c44730 Mon Sep 17 00:00:00 2001
From: Dmitri Naumov <dmitri.naumov@ufz.de>
Date: Thu, 8 Aug 2024 17:44:41 +0200
Subject: [PATCH] [PL] Storing indices locally is measurably faster
A little bit :) 0.5% maybe. Measured on TRM/Mockup ctest.
---
.../Assembly/ParallelVectorMatrixAssembler.cpp | 16 +++++++---------
1 file changed, 7 insertions(+), 9 deletions(-)
diff --git a/ProcessLib/Assembly/ParallelVectorMatrixAssembler.cpp b/ProcessLib/Assembly/ParallelVectorMatrixAssembler.cpp
index 3944d02f627..e46377bb59d 100644
--- a/ProcessLib/Assembly/ParallelVectorMatrixAssembler.cpp
+++ b/ProcessLib/Assembly/ParallelVectorMatrixAssembler.cpp
@@ -33,11 +33,11 @@ void assembleWithJacobianOneElement(
const NumLib::LocalToGlobalIndexMap& dof_table, const double t,
const double dt, const GlobalVector& x, const GlobalVector& x_prev,
std::vector<double>& local_b_data, std::vector<double>& local_Jac_data,
- std::vector<GlobalIndexType>& indices,
ProcessLib::AbstractJacobianAssembler& jacobian_assembler,
ProcessLib::Assembly::MultiMatrixElementCache& cache)
{
- indices = NumLib::getIndices(mesh_item_id, dof_table);
+ std::vector<GlobalIndexType> const& indices =
+ NumLib::getIndices(mesh_item_id, dof_table);
local_b_data.clear();
local_Jac_data.clear();
@@ -95,7 +95,6 @@ void runAssemblyForEachLocalAssembler(
NumLib::LocalToGlobalIndexMap const& dof_table, double const t,
double const dt, GlobalVector const& x, GlobalVector const& x_prev,
std::vector<double>& local_b_data, std::vector<double>& local_Jac_data,
- std::vector<GlobalIndexType>& indices,
ProcessLib::AbstractJacobianAssembler& jac_asm, ThreadException& exception,
ProcessLib::Assembly::MultiMatrixElementCache& cache,
auto local_matrix_output)
@@ -116,9 +115,9 @@ void runAssemblyForEachLocalAssembler(
try
{
- assembleWithJacobianOneElement(
- element_id, loc_asm, dof_table, t, dt, x, x_prev, local_b_data,
- local_Jac_data, indices, jac_asm, cache);
+ assembleWithJacobianOneElement(element_id, loc_asm, dof_table, t,
+ dt, x, x_prev, local_b_data,
+ local_Jac_data, jac_asm, cache);
}
catch (...)
{
@@ -225,7 +224,6 @@ void ParallelVectorMatrixAssembler::assembleWithJacobian(
// reallocations.
std::vector<double> local_b_data;
std::vector<double> local_Jac_data;
- std::vector<GlobalIndexType> indices;
// copy to avoid concurrent access
auto const jac_asm = jacobian_assembler_.copy();
@@ -243,8 +241,8 @@ void ParallelVectorMatrixAssembler::assembleWithJacobian(
// TODO corner case: what if all elements on a submesh are deactivated?
runAssemblyForEachLocalAssembler(
collectActiveLocalAssemblers(local_assemblers, active_elements),
- dof_table, t, dt, x, x_prev, local_b_data, local_Jac_data, indices,
- *jac_asm, exception, cache, local_matrix_output);
+ dof_table, t, dt, x, x_prev, local_b_data, local_Jac_data, *jac_asm,
+ exception, cache, local_matrix_output);
}
stats->print();
--
GitLab