fix CUDA build?

Signed-off-by: Steven Hahn <hahnse@ornl.gov>
project-asgard · Jul 3, 2024 · a7f44ff · a7f44ff
1 parent d52242f
commit a7f44ff
Show file tree

Hide file tree

Showing 3 changed files with 9 additions and 21 deletions.
diff --git a/src/solver.cpp b/src/solver.cpp
@@ -805,12 +805,6 @@ bicgstab_euler(const double dt, imex_flag imex,
 
 #ifdef ASGARD_USE_CUDA
 template gmres_info<double>
-simple_gmres_euler(const double dt, imex_flag imex,
-                   kron_operators<double> const &ops,
-                   fk::vector<double, mem_type::owner, resource::device> &x,
-                   fk::vector<double, mem_type::owner, resource::device> const &b,
-                   int const restart, int const max_iter, double const tolerance);
-template gmres_info<double>
 bicgstab_euler(const double dt, imex_flag imex,
                kron_operators<double> const &ops,
                fk::vector<double, mem_type::owner, resource::device> &x,
@@ -819,17 +813,11 @@ bicgstab_euler(const double dt, imex_flag imex,
 
 template gmres_info<double>
 simple_gmres_euler(adapt::distributed_grid<double> const &adaptive_grid, int const elem_size,
-                   const double dt, kron_operators<double> const &ops,
-                   fk::vector<double> &x, fk::vector<double> const &b,
+                   const double dt, imex_flag imex, kron_operators<double> const &ops,
+                   fk::vector<double, mem_type::owner, resource::device> &x,
+                   fk::vector<double, mem_type::owner, resource::device> const &b,
                    int const restart, int const max_iter,
                    double const tolerance);
-
-template gmres_info<double> simple_gmres_euler(
-    adapt::distributed_grid<double> const &adaptive_grid, int const elem_size,
-    double const dt, kron_operators<double> const &ops,
-    fk::vector<double, mem_type::owner, resource::device> &x,
-    fk::vector<double, mem_type::owner, resource::device> const &b,
-    int const restart, int const max_iter, double const tolerance);
 #endif
 
 template int default_gmres_restarts<double>(int num_cols);
@@ -874,7 +862,7 @@ bicgstab_euler(const float dt, imex_flag imex,
 
 #ifdef ASGARD_USE_CUDA
 template gmres_info<float>
-simple_gmres_euler(adapt::distributed_grid<double> const &adaptive_grid, int const elem_size,
+simple_gmres_euler(adapt::distributed_grid<float> const &adaptive_grid, int const elem_size,
                    const float dt, imex_flag imex,
                    kron_operators<float> const &ops,
                    fk::vector<float, mem_type::owner, resource::device> &x,

diff --git a/src/solver_tests.cpp b/src/solver_tests.cpp
@@ -160,7 +160,7 @@ void test_kronmult(parser const &parse, P const tol_factor)
     int const restart  = parser::DEFAULT_GMRES_INNER_ITERATIONS;
     int const max_iter = parser::DEFAULT_GMRES_OUTER_ITERATIONS;
     P const tolerance  = std::is_same_v<float, P> ? 1e-6 : 1e-12;
-    solver::simple_gmres_euler(adaptive_grid, dt, imex_flag::unspecified, operator_matrices,
+    solver::simple_gmres_euler(adaptive_grid, elem_size, dt, imex_flag::unspecified, operator_matrices,
                                x_d, b_d, restart, max_iter, tolerance);
     return x_d.clone_onto_host();
   }();

diff --git a/src/time_advance.cpp b/src/time_advance.cpp
@@ -528,22 +528,21 @@ imex_advance(PDE<P> &pde, kron_operators<P> &operator_matrices,
   P const max          = pde.get_dimensions()[0].domain_max;
   int const N_elements = fm::two_raised_to(level);
 
-  auto nodes          = gen_realspace_nodes(degree, level, min, max);
-  int const elem_size = static_cast<int>(std::pow(degree, pde.num_dims()));
+  auto nodes = gen_realspace_nodes(degree, level, min, max);
 #ifdef ASGARD_USE_CUDA
   fk::vector<P, mem_type::owner, imex_resrc> f = f_0.clone_onto_device();
   fk::vector<P, mem_type::owner, imex_resrc> f_orig_dev =
       f_0.clone_onto_device();
 #else
   fk::vector<P, mem_type::owner, imex_resrc> f          = f_0;
   fk::vector<P, mem_type::owner, imex_resrc> f_orig_dev = f_0;
-
+#endif
   auto const &plan       = adaptive_grid.get_distrib_plan();
   auto const &grid       = adaptive_grid.get_subgrid(get_rank());
+  int const elem_size    = static_cast<int>(std::pow(degree, pde.num_dims()));
   int const A_local_rows = elem_size * grid.nrows();
 
   fk::vector<P, mem_type::owner, imex_resrc> reduced_fx(A_local_rows);
-#endif
 
   // Create moment matrices that take DG function in (x,v) and transfer to DG
   // function in x
@@ -818,6 +817,7 @@ imex_advance(PDE<P> &pde, kron_operators<P> &operator_matrices,
   exchange_results(reduced_fx, fx, elem_size, plan, get_rank());
   fm::axpy(fx, f, dt); // f here is f_1s
 #else
+  ignore(plan);
   fm::axpy(fx, f, dt);   // f here is f_1s
 #endif