From ab894186589224d570e0ecef8eea06443194a8ab Mon Sep 17 00:00:00 2001
From: paboyle <paboyle@ph.ed.ac.uk>
Date: Wed, 11 May 2016 15:18:47 +0100
Subject: [PATCH] Precision change going in; useful for mixed precision
 algorithms for example.

---
 lib/lattice/Lattice_transfer.h | 49 +++++++++++++++++++++++++++++-----
 1 file changed, 43 insertions(+), 6 deletions(-)

diff --git a/lib/lattice/Lattice_transfer.h b/lib/lattice/Lattice_transfer.h
index f361f7d4..178b5937 100644
--- a/lib/lattice/Lattice_transfer.h
+++ b/lib/lattice/Lattice_transfer.h
@@ -325,6 +325,41 @@ inline void blockPromote(const Lattice<iVector<CComplex,nbasis > > &coarseData,
   
 }
 
+// Useful for precision conversion, or indeed anything where an operator= does a conversion on scalars.
+// Simd layouts need not match since we use peek/poke Local
+template<class vobj,class vvobj>
+void localConvert(const Lattice<vobj> &in,Lattice<vvobj> &out)
+{
+  typedef typename vobj::scalar_object sobj;
+  typedef typename vvobj::scalar_object ssobj;
+
+  sobj s;
+  ssobj ss;
+
+  GridBase *ig = in._grid;
+  GridBase *og = out._grid;
+
+  int ni = ig->_ndimension;
+  int no = og->_ndimension;
+
+  assert(ni == no);
+
+  for(int d=0;d<no;d++){
+    assert(ig->_processors[d]  == og->_processors[d]);
+    assert(ig->_ldimensions[d] == og->_ldimensions[d]);
+  }
+
+PARALLEL_FOR_LOOP
+  for(int idx=0;idx<ig->lSites();idx++){
+    std::vector<int> lcoor(ni);
+    ig->LocalIndexToLocalCoor(idx,lcoor);
+    peekLocalSite(s,in,lcoor);
+    ss=s;
+    pokeLocalSite(ss,out,lcoor);
+  }
+}
+
+
 template<class vobj>
 void InsertSlice(const Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice, int orthog)
 {
@@ -351,12 +386,13 @@ void InsertSlice(const Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice
   }
 
   // the above should guarantee that the operations are local
-  std::vector<int> lcoor(nl);
-  std::vector<int> hcoor(nh);
-  hcoor[orthog] = slice;
+PARALLEL_FOR_LOOP
   for(int idx=0;idx<lg->lSites();idx++){
+    std::vector<int> lcoor(nl);
+    std::vector<int> hcoor(nh);
     lg->LocalIndexToLocalCoor(idx,lcoor);
     dl=0;
+    hcoor[orthog] = slice;
     for(int d=0;d<nh;d++){
       if ( d!=orthog ) { 
 	hcoor[d]=lcoor[dl++];
@@ -392,12 +428,13 @@ void ExtractSlice(Lattice<vobj> &lowDim,const Lattice<vobj> & higherDim,int slic
     }
   }
   // the above should guarantee that the operations are local
-  std::vector<int> lcoor(nl);
-  std::vector<int> hcoor(nh);
-  hcoor[orthog] = slice;
+PARALLEL_FOR_LOOP
   for(int idx=0;idx<lg->lSites();idx++){
+    std::vector<int> lcoor(nl);
+    std::vector<int> hcoor(nh);
     lg->LocalIndexToLocalCoor(idx,lcoor);
     dl=0;
+    hcoor[orthog] = slice;
     for(int d=0;d<nh;d++){
       if ( d!=orthog ) { 
 	hcoor[d]=lcoor[dl++];