use amrex removeInvalidParticles for the beam (#1067)

Hi-PACE · Feb 22, 2024 · 4a257f4 · 4a257f4
1 parent c2f50ff
commit 4a257f4
Showing 1 changed file with 17 additions and 76 deletions.
diff --git a/src/particles/sorting/SliceSort.cpp b/src/particles/sorting/SliceSort.cpp
@@ -19,39 +19,21 @@ shiftSlippedParticles (BeamParticleContainer& beam, const int slice, amrex::Geom
         return;
     }
 
-    const auto ptd = beam.getBeamSlice(WhichBeamSlice::This).getParticleTileData();
+    // remove all invalid particles from WhichBeamSlice::This (including slipped)
+    amrex::removeInvalidParticles(beam.getBeamSlice(WhichBeamSlice::This));
 
     // min_z is the lower end of WhichBeamSlice::This
     const amrex::Real min_z = geom.ProbLo(2) + (slice-geom.Domain().smallEnd(2))*geom.CellSize(2);
 
-    amrex::ReduceOps<amrex::ReduceOpSum, amrex::ReduceOpSum> reduce_op;
-    amrex::ReduceData<int, int> reduce_data(reduce_op);
-    using ReduceTuple = typename decltype(reduce_data)::Type;
-
-    const int num_particles = beam.getNumParticlesIncludingSlipped(WhichBeamSlice::This);
-
-    // count the number of invalid and slipped particles
-    reduce_op.eval(
-        num_particles, reduce_data,
-        [=] AMREX_GPU_DEVICE (const int ip) -> ReduceTuple
-        {
-            if (ptd.id(ip) < 0) {
-                return {1, 0};
-            } else if (ptd.pos(2, ip) < min_z) {
-                return {0, 1};
-            } else {
-                return {0, 0};
-            }
+    // put non slipped particles at the start of the slice
+    const int num_stay = amrex::partitionParticles(beam.getBeamSlice(WhichBeamSlice::This),
+        [=] AMREX_GPU_DEVICE (auto& ptd, int i) {
+            return ptd.pos(2, i) >= min_z;
         });
 
-    ReduceTuple t = reduce_data.value();
-
-    const int num_invalid = amrex::get<0>(t);
-    const int num_slipped = amrex::get<1>(t);
-    const int num_stay = beam.getNumParticlesIncludingSlipped(WhichBeamSlice::This)
-                         - num_invalid - num_slipped;
+    const int num_slipped = beam.getBeamSlice(WhichBeamSlice::This).size() - num_stay;
 
-    if (num_invalid == 0 && num_slipped == 0) {
+    if (num_slipped == 0) {
         // nothing to do
         beam.resize(WhichBeamSlice::This, num_stay, 0);
         return;
@@ -64,60 +46,19 @@ shiftSlippedParticles (BeamParticleContainer& beam, const int slice, amrex::Geom
 
     beam.resize(WhichBeamSlice::Next, next_size, num_slipped);
 
-    BeamTile tmp{};
-    tmp.resize(num_stay);
-
-    const auto ptd_tmp = tmp.getParticleTileData();
-
-    // copy valid non slipped particles to the tmp tile
-    const int num_stay2 = amrex::Scan::PrefixSum<int> (num_particles,
-        [=] AMREX_GPU_DEVICE (const int ip) -> int
-        {
-            return ptd.id(ip) >= 0 && ptd.pos(2, ip) >= min_z;
-        },
-        [=] AMREX_GPU_DEVICE (const int ip, const int s)
-        {
-            if (ptd.id(ip) >= 0 && ptd.pos(2, ip) >= min_z) {
-                ptd_tmp.idcpu(s) = ptd.idcpu(ip);
-                for (int j=0; j<ptd_tmp.NAR; ++j) {
-                    ptd_tmp.rdata(j)[s] = ptd.rdata(j)[ip];
-                }
-                for (int j=0; j<ptd_tmp.NAI; ++j) {
-                    ptd_tmp.idata(j)[s] = ptd.idata(j)[ip];
-                }
-            }
-        },
-        amrex::Scan::Type::exclusive);
-
-    AMREX_ALWAYS_ASSERT(num_stay == num_stay2);
-
+    const auto ptd_this = beam.getBeamSlice(WhichBeamSlice::This).getParticleTileData();
     const auto ptd_next = beam.getBeamSlice(WhichBeamSlice::Next).getParticleTileData();
 
-    // copy valid slipped particles to WhichBeamSlice::Next
-    const int num_slipped2 = amrex::Scan::PrefixSum<int> (num_particles,
-        [=] AMREX_GPU_DEVICE (const int ip) -> int
+    amrex::ParallelFor(num_slipped,
+        [=] AMREX_GPU_DEVICE (int i)
         {
-            return ptd.id(ip) >= 0 && ptd.pos(2, ip) < min_z;
-        },
-        [=] AMREX_GPU_DEVICE (const int ip, const int s)
-        {
-            if (ptd.id(ip) >= 0 && ptd.pos(2, ip) < min_z) {
-                ptd_next.idcpu(s+next_size) = ptd.idcpu(ip);
-                for (int j=0; j<ptd_next.NAR; ++j) {
-                    ptd_next.rdata(j)[s+next_size] = ptd.rdata(j)[ip];
-                }
-                for (int j=0; j<ptd_next.NAI; ++j) {
-                    ptd_next.idata(j)[s+next_size] = ptd.idata(j)[ip];
-                }
-            }
-        },
-        amrex::Scan::Type::exclusive);
-
-    AMREX_ALWAYS_ASSERT(num_slipped == num_slipped2);
+            // copy particles from WhichBeamSlice::This to WhichBeamSlice::Next
+            amrex::copyParticle(ptd_next, ptd_this, num_stay + i, next_size + i);
+        });
 
-    beam.getBeamSlice(WhichBeamSlice::This).swap(tmp);
-    beam.resize(WhichBeamSlice::This, num_stay, 0);
 
-    // stream sync before tmp is deallocated
+    // stream sync before WhichBeamSlice::This is resized
     amrex::Gpu::streamSynchronize();
+
+    beam.resize(WhichBeamSlice::This, num_stay, 0);
 }