GetExternalEBField: Use AMReX's CompileTimeOption ParallelFor (#3696)

This greatly improves the performance on AMD and Intel GPUs when
GetExternalEBField is not used.

1 files changed, 12 insertions, 2 deletions

diff --git a/Source/Particles/RigidInjectedParticleContainer.cpp b/Source/Particles/RigidInjectedParticleContainer.cpp
index cc064538b..493b14e55 100644
--- a/Source/Particles/RigidInjectedParticleContainer.cpp
+++ b/Source/Particles/RigidInjectedParticleContainer.cpp
@@ -407,7 +407,14 @@ RigidInjectedParticleContainer::PushP (int lev, Real dt,
             const auto pusher_algo = WarpX::particle_pusher_algo;
             const auto do_crr = do_classical_radiation_reaction;
 
-            amrex::ParallelFor( np, [=] AMREX_GPU_DEVICE (long ip)
+            enum exteb_flags : int { no_exteb, has_exteb };
+
+            int exteb_runtime_flag = getExternalEB.isNoOp() ? no_exteb : has_exteb;
+
+            amrex::ParallelFor(TypeList<CompileTimeOptions<no_exteb,has_exteb>>{},
+                               {exteb_runtime_flag},
+                               np, [=,getExternalEB=getExternalEB]
+                               AMREX_GPU_DEVICE (long ip, auto exteb_control)
             {
                 ux_save[ip] = uxpp[ip];
                 uy_save[ip] = uypp[ip];
@@ -425,7 +432,10 @@ RigidInjectedParticleContainer::PushP (int lev, Real dt,
                                ex_type, ey_type, ez_type, bx_type, by_type, bz_type,
                                dx_arr, xyzmin_arr, lo, n_rz_azimuthal_modes,
                                nox, galerkin_interpolation);
-                getExternalEB(ip, Exp, Eyp, Ezp, Bxp, Byp, Bzp);
+
+                if constexpr (exteb_control == has_exteb) {
+                    getExternalEB(ip, Exp, Eyp, Ezp, Bxp, Byp, Bzp);
+                }
 
                 amrex::ParticleReal qp = q;
                 if (ion_lev) { qp *= ion_lev[ip]; }