Add a general sparse iterative solver: CGNR
This adds a new LinearOperator which implements symmetric
products of a matrix, and a new CGNR solver to leverage
CG to directly solve the normal equations. This also
includes a block diagonal preconditioner. In experiments
on problem-16, the non-preconditioned version is about
1/5 the speed of SPARSE_SCHUR, and the preconditioned
version using block cholesky is about 20% slower than
SPARSE_SCHUR.
diff --git a/internal/ceres/linear_solver.cc b/internal/ceres/linear_solver.cc
index e3912eb..fd40252 100644
--- a/internal/ceres/linear_solver.cc
+++ b/internal/ceres/linear_solver.cc
@@ -31,7 +31,7 @@
#include "ceres/linear_solver.h"
#include <glog/logging.h>
-#include "ceres/conjugate_gradients_solver.h"
+#include "ceres/cgnr_solver.h"
#include "ceres/dense_qr_solver.h"
#include "ceres/iterative_schur_complement_solver.h"
#include "ceres/schur_complement_solver.h"
@@ -47,7 +47,7 @@
LinearSolver* LinearSolver::Create(const LinearSolver::Options& options) {
switch (options.type) {
case CONJUGATE_GRADIENTS:
- return new ConjugateGradientsSolver(options);
+ return new CgnrSolver(options);
case SPARSE_NORMAL_CHOLESKY:
#ifndef CERES_NO_SUITESPARSE
