From bc7f60f78ecc82a4af1d94d4687e5c7a1b8a9ab8 Mon Sep 17 00:00:00 2001
From: Remi Lehe <remi.lehe@normalesup.org>
Date: Mon, 30 Sep 2019 09:11:29 -0700
Subject: More optimization for GPU

---
 Source/Particles/Sorting/Partition.cpp | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

(limited to 'Source/Particles/Sorting/Partition.cpp')

diff --git a/Source/Particles/Sorting/Partition.cpp b/Source/Particles/Sorting/Partition.cpp
index 2b5d661e4..badd35f02 100644
--- a/Source/Particles/Sorting/Partition.cpp
+++ b/Source/Particles/Sorting/Partition.cpp
@@ -43,9 +43,9 @@ PhysicalParticleContainer::PartitionParticlesInBuffers(
     BL_PROFILE("PPC::Evolve::partition");
 
     // Initialize temporary arrays
-    Gpu::ManagedDeviceVector<int> inexflag;
+    Gpu::DeviceVector<int> inexflag;
     inexflag.resize(np);
-    Gpu::ManagedDeviceVector<long> pid;
+    Gpu::DeviceVector<long> pid;
     pid.resize(np);
 
     // First, partition particles into the larger buffer
@@ -138,5 +138,12 @@ PhysicalParticleContainer::PartitionParticlesInBuffers(
         std::swap(uyp, tmp);
         amrex::ParallelFor( np, copyAndReorder<Real>( uzp, tmp, pid ) );
         std::swap(uzp, tmp);
+
+        // Make sure that the temporary arrays are not destroyed before
+        // the GPU kernels finish running
+        Gpu::streamSynchronize();
     }
+    // Make sure that the temporary arrays are not destroyed before
+    // the GPU kernels finish running
+    Gpu::streamSynchronize();
 }
-- 
cgit v1.2.3