From 8f44c799a69d3041f00af7b7785a268a914ed6c5 Mon Sep 17 00:00:00 2001
From: Peter Boyle <paboyle@ph.ed.ac.uk>
Date: Thu, 26 Apr 2018 14:48:03 +0100
Subject: [PATCH] Saving the benchmarking tests for Cshift

---
 benchmarks/Benchmark_su3.cc | 64 +++++++++++++++++++++++++++++++------
 1 file changed, 55 insertions(+), 9 deletions(-)

diff --git a/benchmarks/Benchmark_su3.cc b/benchmarks/Benchmark_su3.cc
index b31af942..656f816a 100644
--- a/benchmarks/Benchmark_su3.cc
+++ b/benchmarks/Benchmark_su3.cc
@@ -35,24 +35,25 @@ using namespace Grid::QCD;
 int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
+#define LMIN (16)
 #define LMAX (40)
-#define LINC (4)
+#define LINC (8)
 
-  int64_t Nloop=20;
+  int64_t Nloop=200;
 
   std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
   std::vector<int> mpi_layout  = GridDefaultMpi();
 
   int64_t threads = GridThread::GetThreads();
   std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
-
+#if 0
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  x= x*y"<<std::endl;
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=4;lat<=LMAX;lat+=LINC){
+  for(int lat=LMIN;lat<=LMAX;lat+=LINC){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@@ -84,7 +85,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=LINC){
+  for(int lat=LMIN;lat<=LMAX;lat+=LINC){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@@ -115,7 +116,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=LINC){
+  for(int lat=LMIN;lat<=LMAX;lat+=LINC){
 
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@@ -146,7 +147,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=LINC){
+  for(int lat=LMIN;lat<=LMAX;lat+=LINC){
     
     std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
     int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
@@ -170,7 +171,6 @@ int main (int argc, char ** argv)
     std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
     
   }
-  
 
   std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
   std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  CovShiftForward(z,x,y)"<<std::endl;
@@ -178,7 +178,7 @@ int main (int argc, char ** argv)
   std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
   std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
-  for(int lat=2;lat<=LMAX;lat+=LINC){
+  for(int lat=LMIN;lat<=LMAX;lat+=LINC){
       std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
       int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
 
@@ -203,6 +203,52 @@ int main (int argc, char ** argv)
 	std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
       }
     }
+#endif  
+
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  z= x * Cshift(y)"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
+
+  for(int lat=LMIN;lat<=LMAX;lat+=LINC){
+      std::vector<int> latt_size  ({lat*mpi_layout[0],lat*mpi_layout[1],lat*mpi_layout[2],lat*mpi_layout[3]});
+      int64_t vol = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3];
+
+      GridCartesian     Grid(latt_size,simd_layout,mpi_layout);
+      GridParallelRNG          pRNG(&Grid);      pRNG.SeedFixedIntegers(std::vector<int>({45,12,81,9}));
+
+      LatticeColourMatrix z(&Grid); random(pRNG,z);
+      LatticeColourMatrix x(&Grid); random(pRNG,x);
+      LatticeColourMatrix y(&Grid); random(pRNG,y);
+      LatticeColourMatrix tmp(&Grid);
+
+      for(int mu=0;mu<4;mu++){
+	double tshift=0;
+	double tmult =0;
+
+	double start=usecond();
+	for(int64_t i=0;i<Nloop;i++){
+	  tshift-=usecond();
+	  tmp = Cshift(y,mu,-1);
+	  tshift+=usecond();
+	  tmult-=usecond();
+	  z   = x*tmp;
+	  tmult+=usecond();
+	}
+	double stop=usecond();
+	double time = (stop-start)/Nloop;
+	tshift = tshift/Nloop;
+	tmult  = tmult /Nloop;
+	
+	double bytes=3*vol*Nc*Nc*sizeof(Complex);
+	double flops=Nc*Nc*(6+8+8)*vol;
+	std::cout<<GridLogMessage<<std::setprecision(3) << "total us "<<time<<" shift "<<tshift <<" mult "<<tmult<<std::endl;
+	time = time * 1000; // convert to NS for GB/s
+	std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
+      }
+    }
+
 
   Grid_finalize();
 }