From 19f9378b9886f2d0ef2028185ab37311080987c2 Mon Sep 17 00:00:00 2001
From: Peter Boyle <paboyle@ed.ac.uk>
Date: Tue, 11 Mar 2025 13:49:53 +0000
Subject: [PATCH] Should work on Aurora nowb

---
 Grid/allocator/AlignedAllocator.h      | 2 +-
 Grid/allocator/MemoryManagerCache.cc   | 3 +++
 Grid/communicator/Communicator_mpi3.cc | 3 ---
 Grid/cshift/Cshift_mpi.h               | 8 ++++----
 4 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/Grid/allocator/AlignedAllocator.h b/Grid/allocator/AlignedAllocator.h
index a32d69db..316f201c 100644
--- a/Grid/allocator/AlignedAllocator.h
+++ b/Grid/allocator/AlignedAllocator.h
@@ -69,7 +69,7 @@ public:
   }
 
   // FIXME: hack for the copy constructor: it must be avoided to avoid single thread loop
-  void construct(pointer __p, const _Tp& __val) { assert(0);};
+  void construct(pointer __p, const _Tp& __val) { };
   void construct(pointer __p) { };
   void destroy(pointer __p) { };
 };
diff --git a/Grid/allocator/MemoryManagerCache.cc b/Grid/allocator/MemoryManagerCache.cc
index eb8c6d38..09afbcf7 100644
--- a/Grid/allocator/MemoryManagerCache.cc
+++ b/Grid/allocator/MemoryManagerCache.cc
@@ -234,6 +234,9 @@ void *MemoryManager::ViewOpen(void* _CpuPtr,size_t bytes,ViewMode mode,ViewAdvis
 }
 void  MemoryManager::EvictVictims(uint64_t bytes)
 {
+  if(bytes>=DeviceMaxBytes) {
+    printf("EvictVictims bytes %ld DeviceMaxBytes %ld\n",bytes,DeviceMaxBytes);
+  }
   assert(bytes<DeviceMaxBytes);
   while(bytes+DeviceLRUBytes > DeviceMaxBytes){
     if ( DeviceLRUBytes > 0){
diff --git a/Grid/communicator/Communicator_mpi3.cc b/Grid/communicator/Communicator_mpi3.cc
index 38b9f9c6..8de29669 100644
--- a/Grid/communicator/Communicator_mpi3.cc
+++ b/Grid/communicator/Communicator_mpi3.cc
@@ -759,9 +759,6 @@ void CartesianCommunicator::StencilSendToRecvFromComplete(std::vector<CommsReque
 
   int nreq=MpiRequests.size();
 
-  std::cout << GridLogMessage << " StencilSendToRevFromComplete "<<nreq<<" Mpi Requests"<<std::endl;
-
-
   if (nreq>0) {
     status.resize(MpiRequests.size());
     int ierr = MPI_Waitall(MpiRequests.size(),&MpiRequests[0],&status[0]); // Sends are guaranteed in order. No harm in not completing.
diff --git a/Grid/cshift/Cshift_mpi.h b/Grid/cshift/Cshift_mpi.h
index 710792ee..4b000035 100644
--- a/Grid/cshift/Cshift_mpi.h
+++ b/Grid/cshift/Cshift_mpi.h
@@ -126,8 +126,8 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
   static deviceVector<vobj> send_buf; send_buf.resize(buffer_size);
   static deviceVector<vobj> recv_buf; recv_buf.resize(buffer_size);
 #ifndef ACCELERATOR_AWARE_MPI
-  static hostVector<vobj> hsend_buf;  hsend_buf.resize(buffer_size);
-  static hostVector<vobj> hrecv_buf;  hrecv_buf.resize(buffer_size);
+  static std::vector<vobj> hsend_buf;  hsend_buf.resize(buffer_size);
+  static std::vector<vobj> hrecv_buf;  hrecv_buf.resize(buffer_size);
 #endif
   
   int cb= (cbmask==0x2)? Odd : Even;
@@ -250,8 +250,8 @@ template<class vobj> void  Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
     recv_buf_extract[s].resize(buffer_size);
   }
 #ifndef ACCELERATOR_AWARE_MPI
-  hostVector<scalar_object> hsend_buf; hsend_buf.resize(buffer_size);
-  hostVector<scalar_object> hrecv_buf; hrecv_buf.resize(buffer_size);
+  std::vector<scalar_object> hsend_buf; hsend_buf.resize(buffer_size);
+  std::vector<scalar_object> hrecv_buf; hrecv_buf.resize(buffer_size);
 #endif
   
   int bytes = buffer_size*sizeof(scalar_object);