Improve performance of CPR preconditioner.

This restores the performance to approximately the level it had before
the change to support non-diagonal well jacobians, for SPE9. All changes
are to the eliminateVariable() method.
 - Explicitly compute and apply the inverse.
 - Change loop ordering to apply inverse only num_eq - 1 times, instead
   of (numeq - 1)^2.
 - Use UmfPackLU instead of SparseLU.
This commit is contained in:
Atgeirr Flø Rasmussen 2014-11-13 16:56:11 +01:00
parent f9b479a354
commit 933cfaf666

View File

@ -44,7 +44,7 @@
#include <opm/core/utility/platform_dependent/reenable_warnings.h>
#include <Eigen/SparseLU>
#include <Eigen/UmfPackSupport>
namespace Opm
{
@ -249,7 +249,10 @@ namespace Opm
const std::vector<M>& Jn = eqs[n].derivative();
// Use sparse LU to solve the block submatrices i.e compute inv(D)
const Eigen::SparseLU< M > solver(Jn[n]);
const Eigen::UmfPackLU< M > solver(Jn[n]);
M id(Jn[n].rows(), Jn[n].cols());
id.setIdentity();
const M Di = solver.solve(id);
// compute inv(D)*bn for the update of the right hand side
const Eigen::VectorXd& Dibn = solver.solve(eqs[n].value().matrix());
@ -257,29 +260,32 @@ namespace Opm
std::vector<V> vals(num_eq); // Number n will remain empty.
std::vector<std::vector<M>> jacs(num_eq); // Number n will remain empty.
for (int eq = 0; eq < num_eq; ++eq) {
if (eq == n) {
continue;
}
jacs[eq].reserve(num_eq - 1);
const std::vector<M>& Je = eqs[eq].derivative();
const M& B = Je[n];
jacs[eq].reserve(num_eq - 1);
for (int var = 0; var < num_eq; ++var) {
if (var == n) {
// Update right hand side.
vals[eq] = eqs[eq].value().matrix() - B * Dibn;
}
for (int var = 0; var < num_eq; ++var) {
if (var == n) {
continue;
}
// solve Du = C
const M u = Di * Jn[var]; // solver.solve(Jn[var]);
for (int eq = 0; eq < num_eq; ++eq) {
if (eq == n) {
continue;
}
const std::vector<M>& Je = eqs[eq].derivative();
const M& B = Je[n];
// Create new jacobians.
// Add A
jacs[eq].push_back(Je[var]);
M& J = jacs[eq].back();
// solve Du = C
const M& u = solver.solve(Jn[var]);
// Subtract Bu (B*inv(D))
// Subtract Bu (B*inv(D)*C)
J -= B * u;
}
// Update right hand side.
vals[eq] = eqs[eq].value().matrix() - B * Dibn;
}
// Create return value.
@ -317,7 +323,7 @@ namespace Opm
const M& C = eq_coll.derivative()[0];
// Use sparse LU to solve the block submatrices
const Eigen::SparseLU< M > solver(D);
const Eigen::UmfPackLU< M > solver(D);
// Compute value of eliminated variable.
const Eigen::VectorXd b = (equation.value().matrix() - C * partial_solution.matrix());