From aead94e9a7092cac0c72a195312bc3b3db0b95b6 Mon Sep 17 00:00:00 2001
From: paboyle <paboyle@ph.ed.ac.uk>
Date: Sun, 4 Mar 2018 16:39:29 +0000
Subject: [PATCH] View introduced

---
 benchmarks/Benchmark_ITT.cc              | 17 +++++++----
 benchmarks/Benchmark_dwf.cc              | 26 +++++++++++-----
 benchmarks/Benchmark_dwf_sweep.cc        |  8 +++--
 benchmarks/Benchmark_memory_asynch.cc    |  8 ++---
 benchmarks/Benchmark_memory_bandwidth.cc | 22 ++++++++-----
 benchmarks/Benchmark_wilson.cc           | 39 ++++++++++++++++++------
 6 files changed, 84 insertions(+), 36 deletions(-)

diff --git a/benchmarks/Benchmark_ITT.cc b/benchmarks/Benchmark_ITT.cc
index 266251c5..3e49e2c2 100644
--- a/benchmarks/Benchmark_ITT.cc
+++ b/benchmarks/Benchmark_ITT.cc
@@ -255,8 +255,11 @@ public:
       double start=usecond();
       for(int i=0;i<Nloop;i++){
 	z=a*x-y;
-        x[0]=z[0]; // force serial dependency to prevent optimise away
-        y[4]=z[4];
+	auto x_v = x.View();
+	auto y_v = y.View();
+	auto z_v = z.View();
+        x_v[0]=z_v[0]; // force serial dependency to prevent optimise away
+        y_v[4]=z_v[4];
       }
       double stop=usecond();
       double time = (stop-start)/Nloop*1000;
@@ -532,9 +535,11 @@ public:
     {
       LatticeGaugeField Umu5d(FGrid); 
       std::vector<LatticeColourMatrix> U(4,FGrid);
+      auto Umu_v = Umu.View();
+      auto Umu5d_v = Umu5d.View();
       for(int ss=0;ss<Umu.Grid()->oSites();ss++){
 	for(int s=0;s<Ls;s++){
-	  Umu5d[Ls*ss+s] = Umu[ss];
+	  Umu5d_v[Ls*ss+s] = Umu_v[ss];
 	}
       }
       ref = Zero();
@@ -583,9 +588,9 @@ public:
       }; 
 
       for(int c=0;c<num_cases;c++) {
-
-	 WilsonKernelsStatic::Comms = Cases[c].CommsOverlap;
-	 WilsonKernelsStatic::Opt   = Cases[c].Opt;
+	
+	WilsonKernelsStatic::Comms = Cases[c].CommsOverlap;
+	WilsonKernelsStatic::Opt   = Cases[c].Opt;
 	CartesianCommunicator::SetCommunicatorPolicy(Cases[c].CommsAsynch);
 
 	std::cout<<GridLogMessage << "=================================================================================="<<std::endl;
diff --git a/benchmarks/Benchmark_dwf.cc b/benchmarks/Benchmark_dwf.cc
index 8972a496..6da01b0d 100644
--- a/benchmarks/Benchmark_dwf.cc
+++ b/benchmarks/Benchmark_dwf.cc
@@ -122,9 +122,13 @@ int main (int argc, char ** argv)
   // replicate across fifth dimension
   LatticeGaugeField Umu5d(FGrid); 
   std::vector<LatticeColourMatrix> U(4,FGrid);
-  for(int ss=0;ss<Umu.Grid()->oSites();ss++){
-    for(int s=0;s<Ls;s++){
-      Umu5d[Ls*ss+s] = Umu[ss];
+  {
+    auto Umu5d_v = Umu5d.View();
+    auto Umu_v = Umu.View();
+    for(int ss=0;ss<Umu.Grid()->oSites();ss++){
+      for(int s=0;s<Ls;s++){
+	Umu5d_v[Ls*ss+s] = Umu_v[ss];
+      }
     }
   }
   for(int mu=0;mu<Nd;mu++){
@@ -412,14 +416,22 @@ int main (int argc, char ** argv)
 
       //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu+1,1);
-      for(int i=0;i<ref.size();i++){
-	ref[i]+= tmp[i] + Gamma(Gmu[mu])*tmp[i]; ;
+      {
+	auto ref_v = ref.View();
+	auto tmp_v = tmp.View();
+	for(int i=0;i<ref_v.size();i++){
+	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
+	}
       }
 
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu+1,-1);
-      for(int i=0;i<ref.size();i++){
-	ref[i]+= tmp[i] - Gamma(Gmu[mu])*tmp[i]; ;
+      {
+	auto ref_v = ref.View();
+	auto tmp_v = tmp.View();
+	for(int i=0;i<ref_v.size();i++){
+	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
+	}
       }
     }
     ref = -0.5*ref;
diff --git a/benchmarks/Benchmark_dwf_sweep.cc b/benchmarks/Benchmark_dwf_sweep.cc
index 679e9244..2166d657 100644
--- a/benchmarks/Benchmark_dwf_sweep.cc
+++ b/benchmarks/Benchmark_dwf_sweep.cc
@@ -132,9 +132,11 @@ void benchDw(std::vector<int> & latt4, int Ls, int threads,int report )
   LatticeGaugeField Umu5d(FGrid); 
 
   // replicate across fifth dimension
+  auto Umu5d_v = Umu5d.View();
+  auto Umu_v   = Umu.View();
   for(int ss=0;ss<Umu.Grid()->oSites();ss++){
     for(int s=0;s<Ls;s++){
-      Umu5d[Ls*ss+s] = Umu[ss];
+      Umu5d_v[Ls*ss+s] = Umu_v[ss];
     }
   }
 
@@ -272,9 +274,11 @@ void benchsDw(std::vector<int> & latt4, int Ls, int threads, int report )
   LatticeGaugeField Umu5d(FGrid); 
 
   // replicate across fifth dimension
+  auto Umu5d_v = Umu5d.View();
+  auto Umu_v = Umu.View();
   for(int ss=0;ss<Umu.Grid()->oSites();ss++){
     for(int s=0;s<Ls;s++){
-      Umu5d[Ls*ss+s] = Umu[ss];
+      Umu5d_v[Ls*ss+s] = Umu_v[ss];
     }
   }
 
diff --git a/benchmarks/Benchmark_memory_asynch.cc b/benchmarks/Benchmark_memory_asynch.cc
index c13eec16..5d14ef51 100644
--- a/benchmarks/Benchmark_memory_asynch.cc
+++ b/benchmarks/Benchmark_memory_asynch.cc
@@ -79,11 +79,11 @@ int main (int argc, char ** argv)
 
     double start=usecond();
     thread_loop( (int t=0;t<threads;t++),{
-
-      sum[t] = x[t][0];
+      auto x_t = x[t].View();
+      sum[t] = x_t[0];
       for(int i=0;i<Nloop;i++){
-	for(auto ss=x[t].begin();ss<x[t].end();ss++){
-	  sum[t] = sum[t]+x[t][ss];
+	for(auto ss=x_t.begin();ss<x_t.end();ss++){
+	  sum[t] = sum[t]+x_t[ss];
 	}
       }
       stop[t]=usecond();
diff --git a/benchmarks/Benchmark_memory_bandwidth.cc b/benchmarks/Benchmark_memory_bandwidth.cc
index 7312ed67..bc8ff76d 100644
--- a/benchmarks/Benchmark_memory_bandwidth.cc
+++ b/benchmarks/Benchmark_memory_bandwidth.cc
@@ -75,9 +75,12 @@ int main (int argc, char ** argv)
 
       double start=usecond();
       for(int i=0;i<Nloop;i++){
+	auto x_v = x.View();
+	auto y_v = y.View();
+	auto z_v = z.View();
 	axpy(z,a,x,y);
-        x[0]=z[0]; // serial loop dependence to prevent optimise
-        y[4]=z[4];
+        x_v[0]=z_v[0]; // serial loop dependence to prevent optimise
+        y_v[4]=z_v[4];
       }
       double stop=usecond();
       double time = (stop-start)/Nloop*1000;
@@ -111,9 +114,12 @@ int main (int argc, char ** argv)
 
       double start=usecond();
       for(int i=0;i<Nloop;i++){
+	auto x_v = x.View();
+	auto y_v = y.View();
+	auto z_v = z.View();
 	z=a*x-y;
-        x[0]=z[0]; // force serial dependency to prevent optimise away
-        y[4]=z[4];
+        x_v[0]=z_v[0]; // force serial dependency to prevent optimise away
+        y_v[4]=z_v[4];
       }
       double stop=usecond();
       double time = (stop-start)/Nloop*1000;
@@ -145,11 +151,12 @@ int main (int argc, char ** argv)
       LatticeVec y(&Grid);// random(pRNG,y);
       RealD a=2.0;
 
-
       double start=usecond();
       for(int i=0;i<Nloop;i++){
+	auto x_v = x.View();
+	auto z_v = z.View();
 	z=a*x;
-        x[0]=z[0]*2.0;
+        x_v[0]=z_v[0]*2.0;
       }
       double stop=usecond();
       double time = (stop-start)/Nloop*1000;
@@ -181,8 +188,9 @@ int main (int argc, char ** argv)
       Real nn;      
       double start=usecond();
       for(int i=0;i<Nloop;i++){
+	auto x_v = x.View();
 	nn=norm2(x);
-	vsplat(x[0]._internal[0],nn);
+	vsplat(x_v[0]._internal[0],nn);
       }
       double stop=usecond();
       double time = (stop-start)/Nloop*1000;
diff --git a/benchmarks/Benchmark_wilson.cc b/benchmarks/Benchmark_wilson.cc
index ad550bb8..6e30bbb9 100644
--- a/benchmarks/Benchmark_wilson.cc
+++ b/benchmarks/Benchmark_wilson.cc
@@ -108,14 +108,22 @@ int main (int argc, char ** argv)
     for(int mu=0;mu<Nd;mu++){
       //    ref =  src + Gamma(Gamma::Algebra::GammaX)* src ; // 1-gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
-      for(int i=0;i<ref.size();i++){
-	ref[i]+= tmp[i] - Gamma(Gmu[mu])*tmp[i]; ;
+      {
+	auto ref_v = ref.View();
+	auto tmp_v = tmp.View();
+	for(int i=0;i<ref_v.size();i++){
+	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
+	}
       }
 
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu,-1);
-      for(int i=0;i<ref.size();i++){
-	ref[i]+= tmp[i] + Gamma(Gmu[mu])*tmp[i]; ;
+      {
+	auto ref_v = ref.View();
+	auto tmp_v = tmp.View();
+	for(int i=0;i<ref_v.size();i++){
+	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
+	}
       }
     }
   }
@@ -148,8 +156,10 @@ int main (int argc, char ** argv)
   for(int ss=0;ss<0;ss++ ){
     for(int i=0;i<Ns;i++){
       for(int j=0;j<Nc;j++){
-	ComplexF * ref_p = (ComplexF *)&ref[ss]()(i)(j);
-	ComplexF * res_p = (ComplexF *)&result[ss]()(i)(j);
+	auto ref_v = ref.View();
+	auto result_v = result.View();
+	ComplexF * ref_p = (ComplexF *)&ref_v[ss]()(i)(j);
+	ComplexF * res_p = (ComplexF *)&result_v[ss]()(i)(j);
 	std::cout<<GridLogMessage << ss<< " "<<i<<" "<<j<<" "<< (*ref_p)<<" " <<(*res_p)<<std::endl;
       }
     }
@@ -159,16 +169,25 @@ int main (int argc, char ** argv)
     ref = Zero();
     for(int mu=0;mu<Nd;mu++){
 
+
       //    ref =  src - Gamma(Gamma::Algebra::GammaX)* src ; // 1+gamma_x
       tmp = U[mu]*Cshift(src,mu,1);
-      for(int i=0;i<ref.size();i++){
-	ref[i]+= tmp[i] + Gamma(Gmu[mu])*tmp[i]; ;
+      {
+	auto ref_v = ref.View();
+	auto tmp_v = tmp.View();
+	for(int i=0;i<ref_v.size();i++){
+	  ref_v[i]+= tmp_v[i] + Gamma(Gmu[mu])*tmp_v[i]; ;
+	}
       }
 
       tmp =adj(U[mu])*src;
       tmp =Cshift(tmp,mu,-1);
-      for(int i=0;i<ref.size();i++){
-	ref[i]+= tmp[i] - Gamma(Gmu[mu])*tmp[i]; ;
+      {
+	auto ref_v = ref.View();
+	auto tmp_v = tmp.View();
+	for(int i=0;i<ref_v.size();i++){
+	  ref_v[i]+= tmp_v[i] - Gamma(Gmu[mu])*tmp_v[i]; ;
+	}
       }
     }
   }