60% more MFlop/s

Tom


git-svn-id: http://svn.cactuscode.org/arrangements/CactusWave/WaveToyCXX/trunk@21 50555cc7-fb31-491a-85db-9a2874240742

1 files changed, 35 insertions, 21 deletions

diff --git a/src/WaveToy.cc b/src/WaveToy.cc
index 9762425..23843fc 100644
--- a/src/WaveToy.cc
+++ b/src/WaveToy.cc
@@ -41,39 +41,52 @@ extern "C" void  WaveToyCXX_Evolution(CCTK_ARGUMENTS)
 
   // Set up shorthands 
 
-  CCTK_REAL dx = CCTK_DELTA_SPACE(0),
-            dy = CCTK_DELTA_SPACE(1),
-            dz = CCTK_DELTA_SPACE(2),
-            dt = CCTK_DELTA_TIME,
+  CCTK_REAL dx = CCTK_DELTA_SPACE(0);
+  CCTK_REAL dy = CCTK_DELTA_SPACE(1);
+  CCTK_REAL dz = CCTK_DELTA_SPACE(2);
+  CCTK_REAL dt = CCTK_DELTA_TIME;
       
-            dx2=dx*dx,
-	    dy2=dy*dy,
-	    dz2=dz*dz,
-	    dt2=dt*dt,
+  CCTK_REAL dx2 = dx*dx;
+  CCTK_REAL dy2 = dy*dy;
+  CCTK_REAL dz2 = dz*dz;
+  CCTK_REAL dt2 = dt*dt;
 
-            factor = 2*(1 - (dt2)*(1/dx2 + 1/dy2 + 1/dz2));
+  CCTK_REAL dx2i = 1.0/dx2;
+  CCTK_REAL dy2i = 1.0/dy2;
+  CCTK_REAL dz2i = 1.0/dz2;
 
-  int   iend = cctk_lsh[0]-1,
-        jend = cctk_lsh[1]-1,
-        kend = cctk_lsh[2]-1; 
+  CCTK_REAL factor = 2*(1 - (dt2)*(dx2i + dy2i + dz2i));
+
+  int istart = 1;
+  int jstart = 1;
+  int kstart = 1;
+
+  int iend = cctk_lsh[0]-1;
+  int jend = cctk_lsh[1]-1;
+  int kend = cctk_lsh[2]-1; 
 
   //
   // Do the evolution 
   // 
 
   for (int k=1; k<kend; k++)
+  {
     for (int j=1; j<jend; j++)
+    {
       for (int i=1; i<iend; i++)
       {
-        val(phi_new, i,j,k) =
-		     factor* val( phi,i,j,k ) - val( phi_old,i,j,k) 
+        int index = CCTK_GFINDEX3D(cctkGH,i,j,k);
+
+        phi_new[index] =
+		     factor*phi[index] - phi_old[index] 
               + dt2 * 
-                   ( ( val( phi, i+1,j  ,k  ) + val( phi, i-1,j  ,k)   )/dx2
-                    +( val( phi, i  ,j+1,k  ) + val( phi, i  ,j-1,k)   )/dy2
-                    +( val( phi, i  ,j  ,k+1) + val( phi, i  ,j,  k-1) )/dz2
+                   ( ( val( phi, i+1,j  ,k  ) + val( phi, i-1,j  ,k)   )*dx2i
+                    +( val( phi, i  ,j+1,k  ) + val( phi, i  ,j-1,k)   )*dy2i
+                    +( val( phi, i  ,j  ,k+1) + val( phi, i  ,j,  k-1) )*dz2i
 		   );
       } 
-
+    }
+  }
 
   // 
   // Synchronize before applying boundary conditions 
@@ -93,9 +106,10 @@ extern "C" void  WaveToyCXX_Evolution(CCTK_ARGUMENTS)
       for (int j=0; j<cctk_lsh[1]; j++)
 	for (int i=0; i<cctk_lsh[0]; i++)
 	{
-          int     index = CCTK_GFINDEX3D(cctkGH,i,j,k); 
-                  phi_old[index]  = phi[index]; 
-		  phi[index]      = phi_new[index];
+          int index = CCTK_GFINDEX3D(cctkGH,i,j,k); 
+
+          phi_old[index] = phi[index]; 
+          phi[index]     = phi_new[index];
 	}
    }