From 49d83d653727c19433d45a732ce4ca7d958d06f4 Mon Sep 17 00:00:00 2001
From: Andrea Bocci <andrea.bocci@cern.ch>
Date: Mon, 28 Oct 2019 18:26:42 +0100
Subject: [PATCH] Fix direction of the copies from device to host

---
 HeterogeneousCore/CUDAUtilities/interface/copyAsync.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/HeterogeneousCore/CUDAUtilities/interface/copyAsync.h b/HeterogeneousCore/CUDAUtilities/interface/copyAsync.h
index 5726736166fdc..bfa1bdee9a03d 100644
--- a/HeterogeneousCore/CUDAUtilities/interface/copyAsync.h
+++ b/HeterogeneousCore/CUDAUtilities/interface/copyAsync.h
@@ -26,7 +26,7 @@ namespace cudautils {
                         const cudautils::device::unique_ptr<T>& src,
                         cudaStream_t stream) {
     static_assert(std::is_array<T>::value == false, "For array types, use the other overload with the size parameter");
-    cudaCheck(cudaMemcpyAsync(dst.get(), src.get(), sizeof(T), cudaMemcpyHostToDevice, stream));
+    cudaCheck(cudaMemcpyAsync(dst.get(), src.get(), sizeof(T), cudaMemcpyDeviceToHost, stream));
   }
 
   // Multiple elements
@@ -43,7 +43,7 @@ namespace cudautils {
                         const cudautils::device::unique_ptr<T[]>& src,
                         size_t nelements,
                         cudaStream_t stream) {
-    cudaCheck(cudaMemcpyAsync(dst.get(), src.get(), nelements * sizeof(T), cudaMemcpyHostToDevice, stream));
+    cudaCheck(cudaMemcpyAsync(dst.get(), src.get(), nelements * sizeof(T), cudaMemcpyDeviceToHost, stream));
   }
 }  // namespace cudautils