number of flops changed to 1608 from 1320

GPU/mixed_solve.cu

@@ -2775,13 +2775,13 @@ void benchmark(spinor * const Q,MixedsolveParameter<RealT>& mixedsolveParameter)
     assert((stop = clock())!=-1);
     timeelapsed = (double) (stop-start)/CLOCKS_PER_SEC;
     // x2 because 2x Hopping per iteration
-    double benchres = 1320.0*2*(VOLUME/2)* 1000 / timeelapsed / 1.0e9;
+    double benchres = 1608.0*2*(VOLUME/2)* 1000 / timeelapsed / 1.0e9;
     printf("Benchmark: %f Gflops\n", benchres); 
   #else
     stop = MPI_Wtime();
     timeelapsed = (double) (stop-start);
     // x2 because 2x Hopping per iteration
-    double benchres = 1320.0*2*(g_nproc*VOLUME/2)* 1000 / timeelapsed / 1.0e9;
+    double benchres = 1608.0*2*(g_nproc*VOLUME/2)* 1000 / timeelapsed / 1.0e9;
     if (g_proc_id == 0) {
       printf("Benchmark: %f Gflops\n", benchres); 
     }
@@ -2894,13 +2894,13 @@ void benchmark2(spinor * const Q,MixedsolveParameter<RealT>& mixedsolveParameter
     assert((stop = clock())!=-1);
     timeelapsed = (double) (stop-start)/CLOCKS_PER_SEC;
     // x8 because 8x Hopping per iteration
-    double benchres = 1320.0*8*(VOLUME/2)* 100 / timeelapsed / 1.0e9;
+    double benchres = 1608.0*8*(VOLUME/2)* 100 / timeelapsed / 1.0e9;
     printf("Benchmark: %f Gflops\n", benchres); 
   #else
     stop = MPI_Wtime();
     timeelapsed = (double) (stop-start);
     // 8 because 8x Hopping per iteration
-    double benchres = 1320.0*8*(g_nproc*VOLUME/2)* 100 / timeelapsed / 1.0e9;
+    double benchres = 1608.0*8*(g_nproc*VOLUME/2)* 100 / timeelapsed / 1.0e9;
     if (g_proc_id == 0) {
       printf("Benchmark: %f Gflops\n", benchres); 
     }

GPU/mixed_solve_eo_nd.cuh

@@ -3486,7 +3486,7 @@ extern "C" int mixedsolve_eo_nd (spinor * P_up, spinor * P_dn,
   #ifdef ALGORITHM_BENCHMARK
     double effectiveflops;			// will used to count the "effective" flop's (from the algorithmic perspective)
     // double hoppingflops = 1488.0;
-    double hoppingflops = 1320.0;
+    double hoppingflops = 1608.0;
     double matrixflops  = 2  *  (  2 * ( (2*hoppingflops+12+3) + (2*hoppingflops+3) + (12+2) + 12 )  );
     #ifdef MPI
       double allflops;				// flops added for all processes

benchmark.c

@@ -315,9 +315,9 @@ int main(int argc,char *argv[])
     if(g_proc_id==0) {
       printf("# The following result is just to make sure that the calculation is not optimized away: %e\n", antioptaway);
       printf("# Total compute time %e sec, variance of the time %e sec. (%d iterations).\n", sdt, sqdt, j_max);
-      printf("# Communication switched on:\n# (%d Mflops [%d bit arithmetic])\n", (int)(1320.0f/sdt),(int)sizeof(spinor)/3);
+      printf("# Communication switched on:\n# (%d Mflops [%d bit arithmetic])\n", (int)(1608.0f/sdt),(int)sizeof(spinor)/3);
 #ifdef OMP
-      printf("# Mflops per OpenMP thread ~ %d\n",(int)(1320.0f/(omp_num_threads*sdt)));
+      printf("# Mflops per OpenMP thread ~ %d\n",(int)(1608.0f/(omp_num_threads*sdt)));
 #endif
       printf("\n");
       fflush(stdout);
@@ -345,9 +345,9 @@ int main(int argc,char *argv[])
     dt=1.0e6f*dt/((double)(k_max*j_max*(VOLUME)));
     if(g_proc_id==0) {
       printf("# The following result is printed just to make sure that the calculation is not optimized away: %e\n",antioptaway);
-      printf("# Communication switched off: \n# (%d Mflops [%d bit arithmetic])\n", (int)(1320.0f/dt),(int)sizeof(spinor)/3);
+      printf("# Communication switched off: \n# (%d Mflops [%d bit arithmetic])\n", (int)(1608.0f/dt),(int)sizeof(spinor)/3);
 #ifdef OMP
-      printf("# Mflops per OpenMP thread ~ %d\n",(int)(1320.0f/(omp_num_threads*dt)));
+      printf("# Mflops per OpenMP thread ~ %d\n",(int)(1608.0f/(omp_num_threads*dt)));
 #endif
       printf("\n"); 
       fflush(stdout);
@@ -411,9 +411,9 @@ int main(int argc,char *argv[])
     if(g_proc_id==0) {
       printf("# The following result is just to make sure that the calculation is not optimized away: %e\n", antioptaway);
       printf("# Total compute time %e sec, variance of the time %e sec. (%d iterations).\n", sdt, sqdt, j_max);
-      printf("\n# (%d Mflops [%d bit arithmetic])\n", (int)(1392.0f/sdt),(int)sizeof(spinor)/3);
+      printf("\n# (%d Mflops [%d bit arithmetic])\n", (int)(1680.0f/sdt),(int)sizeof(spinor)/3);
 #ifdef OMP
-      printf("# Mflops per OpenMP thread ~ %d\n",(int)(1320.0f/(omp_num_threads*sdt)));
+      printf("# Mflops per OpenMP thread ~ %d\n",(int)(1680.0f/(omp_num_threads*sdt)));
 #endif
       printf("\n"); 
       fflush(stdout);

linsolve.c

@@ -146,8 +146,8 @@ int solve_cg(spinor * const k, spinor * const l, double eps_sq, const int rel_pr
   }
   etime = gettime();
   /* 2 A + 2 Nc Ns + N_Count ( 2 A + 10 Nc Ns ) */
-  /* 2*1320.0 because the linalg is over VOLUME/2 */
-  flops = (2*(2*1320.0+2*3*4) + 2*3*4 + iteration*(2.*(2*1320.0+2*3*4) + 10*3*4))*VOLUME/2/1.0e6f;
+  /* 2*1608.0 because the linalg is over VOLUME/2 */
+  flops = (2*(2*1608.0+2*3*4) + 2*3*4 + iteration*(2.*(2*1608.0+2*3*4) + 10*3*4))*VOLUME/2/1.0e6f;
   if(g_proc_id==0 && g_debug_level > 0) {
     printf("CG: iter: %d eps_sq: %1.4e t/s: %1.4e\n", iteration, eps_sq, etime-atime); 
     printf("CG: flopcount: t/s: %1.4e mflops_local: %.1f mflops: %.1f\n", 

sample-input/benchmark.input

@@ -0,0 +1,7 @@
+T=48
+L=24
+NrXProcs = 1
+NrYProcs = 1
+NrZProcs = 1
+
+OMPNumThreads = 1

solver/cg_her.c

@@ -127,11 +127,11 @@ int cg_her(spinor * const P, spinor * const Q, const int max_iter,
   etime = gettime();
   g_sloppy_precision = save_sloppy;
   /* 2 A + 2 Nc Ns + N_Count ( 2 A + 10 Nc Ns ) */
-  /* 2*1320.0 because the linalg is over VOLUME/2 */
-  flops = (2*(2*1320.0+2*3*4) + 2*3*4 + iteration*(2.*(2*1320.0+2*3*4) + 10*3*4))*N/1.0e6f;
+  /* 2*1608.0 because the linalg is over VOLUME/2 */
+  flops = (2*(2*1608.0+2*3*4) + 2*3*4 + iteration*(2.*(2*1608.0+2*3*4) + 10*3*4))*N/1.0e6f;
   if(g_debug_level > 0 && g_proc_id == 0 && N != VOLUME) {
     printf("# CG: iter: %d eps_sq: %1.4e t/s: %1.4e\n", iteration, eps_sq, etime-atime); 
-    printf("# CG: flopcount (for tmWilson only): t/s: %1.4e mflops_local: %.1f mflops: %.1f\n", 
+    printf("# CG: flopcount (for e/o tmWilson only): t/s: %1.4e mflops_local: %.1f mflops: %.1f\n", 
            etime-atime, flops/(etime-atime), g_nproc*flops/(etime-atime));
   }
   finalize_solver(solver_field, nr_sf);