Change GPU sync to GPU stream sync (Exawind#1507)

erf-model · Feb 27, 2025 · d012d58 · d012d58
1 parent 23b314b
commit d012d58
Show file tree

Hide file tree

Showing 78 changed files with 149 additions and 149 deletions.
diff --git a/amr-wind/core/Field.cpp b/amr-wind/core/Field.cpp
@@ -414,7 +414,7 @@ void Field::to_uniform_space() noexcept
                     detJ[nbx](i, j, k) / fac[nbx](i, j, k, n);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
     m_mesh_mapped = true;
 }
 
@@ -445,7 +445,7 @@ void Field::to_stretched_space() noexcept
                     fac[nbx](i, j, k, n) / detJ[nbx](i, j, k);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
     m_mesh_mapped = false;
 }
 

diff --git a/amr-wind/core/field_ops.H b/amr-wind/core/field_ops.H
@@ -390,7 +390,7 @@ lower_bound(FType& field, const amrex::Real min_value, const int icomp = 0)
                     amrex::max(min_value, farrs[nbx](i, j, k, icomp));
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 /** Computes the global maximum of a field from all levels
@@ -465,7 +465,7 @@ inline void normalize(FType& field)
                 }
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 } // namespace amr_wind::field_ops

diff --git a/amr-wind/diffusion/incflo_diffusion.cpp b/amr-wind/diffusion/incflo_diffusion.cpp
@@ -259,7 +259,7 @@ void viscosity_to_uniform_space(
                                         std::pow(fac_arrs[nbx](i, j, k, 2), 2);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 } // namespace diffusion
diff --git a/amr-wind/equation_systems/CompRHSOps.H b/amr-wind/equation_systems/CompRHSOps.H
@@ -135,7 +135,7 @@ struct ComputeRHSOp
                     });
             }
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 
     /** Compute right-hand side for corrector steps
@@ -259,7 +259,7 @@ struct ComputeRHSOp
                     });
             }
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 
     // data members

diff --git a/amr-wind/equation_systems/DiffusionOps.cpp b/amr-wind/equation_systems/DiffusionOps.cpp
@@ -135,7 +135,7 @@ void DiffSolverIface<LinOp>::linsys_solve_impl()
                     rho_arrs[nbx](i, j, k) * fld_arrs[nbx](i, j, k, n);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 
     amrex::MLMG mlmg(*this->m_solver);
     this->setup_solver(mlmg);

diff --git a/amr-wind/equation_systems/PDEOps.H b/amr-wind/equation_systems/PDEOps.H
@@ -107,7 +107,7 @@ struct SrcTermOpBase
                     vf_arrs[nbx](i, j, k, n) *= rho_arrs[nbx](i, j, k);
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 
     //! Update source terms during time-integration procedure

diff --git a/amr-wind/equation_systems/density/density_ops.H b/amr-wind/equation_systems/density/density_ops.H
@@ -41,7 +41,7 @@ struct ComputeRHSOp<Density, Scheme>
                     // k));
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 
     void corrector_rhs(
@@ -73,7 +73,7 @@ struct ComputeRHSOp<Density, Scheme>
                     //     0.5 * (rho(i, j, k) + rho_o(i, j, k));
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 
     // data members

diff --git a/amr-wind/equation_systems/icns/icns_advection.cpp b/amr-wind/equation_systems/icns/icns_advection.cpp
@@ -454,7 +454,7 @@ void MacProjOp::mac_proj_to_uniform_space(
                     rho_arrs[nbx](i, j, k);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 } // namespace amr_wind::pde
diff --git a/amr-wind/equation_systems/icns/icns_diffusion.H b/amr-wind/equation_systems/icns/icns_diffusion.H
@@ -178,7 +178,7 @@ public:
                         divtau_arrs[nbx](i, j, k, 2) *= rhoinv;
                     });
             }
-            amrex::Gpu::synchronize();
+            amrex::Gpu::streamSynchronize();
         }
     }
 
@@ -244,7 +244,7 @@ public:
                         rho_arrs[nbx](i, j, k) * fld_arrs[nbx](i, j, k, n);
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
 
         amrex::MLMG mlmg(*m_solver_scalar);
         m_options(mlmg);
@@ -401,7 +401,7 @@ public:
                         divtau_arrs[nbx](i, j, k, 2) *= rhoinv;
                     });
             }
-            amrex::Gpu::synchronize();
+            amrex::Gpu::streamSynchronize();
         }
     }
 
@@ -478,7 +478,7 @@ public:
                         rho_arrs[nbx](i, j, k) * fld_arrs[nbx](i, j, k, n);
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
 
         for (int i = 0; i < AMREX_SPACEDIM; ++i) {
             auto vel_comp = m_pdefields.field.subview(i);

diff --git a/amr-wind/equation_systems/levelset/levelset_ops.H b/amr-wind/equation_systems/levelset/levelset_ops.H
@@ -80,7 +80,7 @@ struct ComputeRHSOp<Levelset, Scheme>
                     // k));
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 
     void corrector_rhs(
@@ -112,7 +112,7 @@ struct ComputeRHSOp<Levelset, Scheme>
                     //     0.5 * (phi(i, j, k) + phi_o(i, j, k));
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 
     // data members

diff --git a/amr-wind/equation_systems/tke/tke_ops.H b/amr-wind/equation_systems/tke/tke_ops.H
@@ -51,7 +51,7 @@ struct PostSolveOp<TKE>
                                                    : field_arrs[nbx](i, j, k);
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
         m_fields.field.fillpatch(time);
     }
 

diff --git a/amr-wind/equation_systems/vof/vof_hybridsolver_ops.H b/amr-wind/equation_systems/vof/vof_hybridsolver_ops.H
@@ -28,7 +28,7 @@ static void replace_masked_vof(
                         : vfmasked_arrs[nbx](i, j, k);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 } // namespace amr_wind::multiphase

diff --git a/amr-wind/immersed_boundary/bluff_body/bluff_body_ops.cpp b/amr-wind/immersed_boundary/bluff_body/bluff_body_ops.cpp
@@ -68,7 +68,7 @@ void apply_mms_vel(CFDSim& sim)
                 }
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 void apply_dirichlet_vel(CFDSim& sim, const amrex::Vector<amrex::Real>& vel_bc)
@@ -129,7 +129,7 @@ void apply_dirichlet_vel(CFDSim& sim, const amrex::Vector<amrex::Real>& vel_bc)
                 }
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 void prepare_netcdf_file(

diff --git a/amr-wind/mesh_mapping_models/ChannelFlowMap.cpp b/amr-wind/mesh_mapping_models/ChannelFlowMap.cpp
@@ -238,7 +238,7 @@ void ChannelFlowMap::create_face_map(int lev, const amrex::Geometry& geom)
                 scale_fac_zf_arrs[nbx](i, j, k, 1) *
                 scale_fac_zf_arrs[nbx](i, j, k, 2);
         });
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
     // TODO: Call fill patch operators ?
 }
 

diff --git a/amr-wind/mesh_mapping_models/ConstantMap.cpp b/amr-wind/mesh_mapping_models/ConstantMap.cpp
@@ -114,7 +114,7 @@ void ConstantMap::create_face_map(int lev)
                 scale_fac_zf_arrs[nbx](i, j, k, 2);
         });
 
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 /** Construct the non-uniform mesh field

diff --git a/amr-wind/ocean_waves/relaxation_zones/linear_waves_ops.H b/amr-wind/ocean_waves/relaxation_zones/linear_waves_ops.H
@@ -132,7 +132,7 @@ struct InitDataOp<LinearWaves>
                     phi_arrs[nbx](i, j, k) = phi;
                 }
             });
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 };
 
@@ -214,7 +214,7 @@ struct UpdateTargetFieldsOp<LinearWaves>
                     }
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 };
 

diff --git a/amr-wind/ocean_waves/relaxation_zones/relaxation_zones_ops.cpp b/amr-wind/ocean_waves/relaxation_zones/relaxation_zones_ops.cpp
@@ -67,7 +67,7 @@ void update_target_vof(CFDSim& sim)
                     multiphase::levelset_to_vof(i, j, k, eps, target_phi[nbx]);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 void apply_relaxation_zones(CFDSim& sim, const RelaxZonesBaseData& wdata)
@@ -265,7 +265,7 @@ void apply_relaxation_zones(CFDSim& sim, const RelaxZonesBaseData& wdata)
                     rho1 * volfrac(i, j, k) + rho2 * (1. - volfrac(i, j, k));
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 
     vof.fillpatch(time);
     velocity.fillpatch(time);

diff --git a/amr-wind/ocean_waves/relaxation_zones/stokes_waves_ops.H b/amr-wind/ocean_waves/relaxation_zones/stokes_waves_ops.H
@@ -162,7 +162,7 @@ struct InitDataOp<StokesWaves>
                     phi_arrs[nbx](i, j, k) = phi;
                 }
             });
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 };
 
@@ -227,7 +227,7 @@ struct UpdateTargetFieldsOp<StokesWaves>
                     }
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
     }
 };
 

diff --git a/amr-wind/ocean_waves/relaxation_zones/waves2amr_ops.H b/amr-wind/ocean_waves/relaxation_zones/waves2amr_ops.H
@@ -96,7 +96,7 @@ void postprocess_velocity_mfab_liquid(
                 vel[nbx](i, j, k, 2) = 0.0;
             }
         });
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 void postprocess_velocity_field_liquid(
@@ -509,7 +509,7 @@ struct InitDataOp<W2AWaves>
                 w2a_vel[nbx](i, j, k, 1) = quiescent[0];
                 w2a_vel[nbx](i, j, k, 2) = quiescent[0];
             });
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
 #else
         amrex::ignore_unused(data, level, geom, multiphase_mode);
 #endif
@@ -718,7 +718,7 @@ struct UpdateTargetFieldsOp<W2AWaves>
                         (time - t_last) / (W2A_t - t_last + 1e-16);
                 });
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
 #else
         amrex::ignore_unused(data, time);
 #endif

diff --git a/amr-wind/overset/OversetOps.cpp b/amr-wind/overset/OversetOps.cpp
@@ -144,7 +144,7 @@ void OversetOps::update_gradp()
                 gp_lev_arrs[nbx](i, j, k, n) = gp_proj_arrs[nbx](i, j, k, n);
             });
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 
     // Averaging down here would be unnecessary; it is built into calcGradPhi
 }
@@ -249,7 +249,7 @@ void OversetOps::sharpen_nalu_data()
                 iblank_cell(lev), dx, max_pvscale);
         pvscale = std::min(pvscale, pvscale_lev);
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
     amrex::ParallelDescriptor::ReduceRealMin(pvscale);
 
     // Convert levelset to vof to get target_vof
@@ -260,14 +260,14 @@ void OversetOps::sharpen_nalu_data()
         // A tolerance of 0 should do nothing
         overset_ops::process_vof((*target_vof)(lev), m_target_cutoff);
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 
     // Replace vof with original values in amr domain
     for (int lev = 0; lev < nlevels; ++lev) {
         overset_ops::harmonize_vof(
             (*target_vof)(lev), vof(lev), iblank_cell(lev));
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 
     // Put fluxes in vector for averaging down during iterations
     amrex::Vector<amrex::Array<amrex::MultiFab*, AMREX_SPACEDIM>> fluxes(
@@ -319,7 +319,7 @@ void OversetOps::sharpen_nalu_data()
                 err = amrex::max(err, err_lev);
             }
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
 
         // Average down fluxes across levels for consistency
         for (int lev = nlevels - 1; lev > 0; --lev) {
@@ -338,7 +338,7 @@ void OversetOps::sharpen_nalu_data()
                 m_convg_tol);
             ptfac = amrex::min(ptfac, ptfac_lev);
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
         amrex::ParallelDescriptor::ReduceRealMin(ptfac);
 
         // Conform pseudo dt (dtau) to pseudo CFL
@@ -357,7 +357,7 @@ void OversetOps::sharpen_nalu_data()
             velocity(lev).FillBoundary(geom[lev].periodicity());
             gp(lev).FillBoundary(geom[lev].periodicity());
         }
-        amrex::Gpu::synchronize();
+        amrex::Gpu::streamSynchronize();
 
         // Update density (fillpatch built in)
         m_mphase->set_density_via_vof();
@@ -382,7 +382,7 @@ void OversetOps::sharpen_nalu_data()
     for (int lev = 0; lev < nlevels; ++lev) {
         overset_ops::equate_field(levelset(lev), velocity(lev));
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 void OversetOps::form_perturb_pressure()
@@ -421,7 +421,7 @@ void OversetOps::replace_masked_gradp()
         // Reapply pressure gradient term
         overset_ops::apply_pressure_gradient(vel(lev), rho(lev), gp(lev), dt);
     }
-    amrex::Gpu::synchronize();
+    amrex::Gpu::streamSynchronize();
 }
 
 } // namespace amr_wind