small refinements serial/ omp/

jacobi/openmp/not_opt/Makefile

@@ -51,7 +51,7 @@ valgrind_callgrind: $(PROG_CALLGRIND)
 
 valgrind_cachegrind: $(PROG_CACHEGRIND)
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'
-	valgrind --tool=cachegrind --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128 2
+	valgrind --tool=cachegrind --cache-sim=yes --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128 2
 	@echo '$$ cg_annotate --auto=yes cachegrind.out.<pid> | less'
 	@echo '(kcachegrind is required in order to visualize the output using the GUI)'
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'

jacobi/openmp/not_opt/src/jacobi_2D_omp_not_opt.c

@@ -234,9 +234,6 @@ void JacobiAlgorithm(MyData      **const restrict Phi,
     #pragma omp barrier
 #endif /* DEBUG */
 
-
-    /***** MISSING LOOP(s) PARALLELIZATION *****/
-    
     for (int j=jbeg ; j<=jend ; j++)
       {
 	for (int i=ibeg ; i<=iend ; i++)
@@ -244,10 +241,7 @@ void JacobiAlgorithm(MyData      **const restrict Phi,
 	    Phi[j][i] = 0.25 * (Phi0[j][i-1] + Phi0[j][i+1] +
 				Phi0[j-1][i] + Phi0[j+1][i]);
 
-	    #pragma omp critical
-	    {
 	    *error += delta[X] * delta[Y] * fabs(Phi[j][i] - Phi0[j][i]);
-	    }
 	  } /* loop over columns */
       } /* loop over rows */
   } /* omp parallel */

jacobi/serial/not_opt/Makefile

@@ -48,7 +48,7 @@ valgrind_callgrind: $(PROG_CALLGRIND)
 
 valgrind_cachegrind: $(PROG_CACHEGRIND)
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'
-	valgrind --tool=cachegrind --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128
+	valgrind --tool=cachegrind --cache-sim=yes --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128
 	@echo '$$ cg_annotate --auto=yes cachegrind.out.<pid> | less'
 	@echo '(kcachegrind is required in order to visualize the output using the GUI)'
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'

jacobi_solutions/serial/not_opt/Makefile

@@ -48,7 +48,7 @@ valgrind_callgrind: $(PROG_CALLGRIND)
 
 valgrind_cachegrind: $(PROG_CACHEGRIND)
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'
-	valgrind --tool=cachegrind --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128
+	valgrind --tool=cachegrind --cache-sim=yes --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128
 	@echo '$$ cg_annotate --auto=yes cachegrind.out.<pid> | less'
 	@echo '(kcachegrind is required in order to visualize the output using the GUI)'
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'

jacobi_solutions/serial/not_opt/src/jacobi_2D_serial_not_opt.c

@@ -214,10 +214,13 @@ void JacobiAlgorithm(MyData      **const restrict Phi,
     {
       for (int i=ibeg ; i<=iend ; i++)
 	{
-	  Phi[j][i] = 0.25 * (Phi0[j][i-1] + Phi0[j][i+1] +
-			      Phi0[j-1][i] + Phi0[j+1][i]);
+	  Phi[j][i] = 0.25 * (Phi0[j][i-1] +
+			      Phi0[j][i+1] +
+			      Phi0[j-1][i] +
+			      Phi0[j+1][i]);
                 
-	  *error += delta[X] * delta[Y] * fabs(Phi[j][i] - Phi0[j][i]);
+	  *error += (delta[X] * delta[Y]) *
+	            fabs(Phi[j][i] - Phi0[j][i]);
 	} /* loop over columns */
     } /* loop over rows */
   

jacobi_solutions/serial/not_opt/src/tools.c

@@ -51,9 +51,8 @@ void Show_2DdblArray(const MyData **const A,
 
 double seconds()
 {
-  struct timeval tmp;
-  gettimeofday(&tmp, (struct timezone *)0);
-  double sec = tmp.tv_sec + ((double)tmp.tv_usec)/1000000.0;
-
-  return sec;
+  struct timespec ts;
+  return (clock_gettime( CLOCK_PROCESS_CPUTIME_ID, &ts ),
+	   (double)ts.tv_sec +
+	   (double)ts.tv_nsec * 1e-9);
 }

jacobi_solutions/serial/opt/Makefile

@@ -48,7 +48,7 @@ valgrind_callgrind: $(PROG_CALLGRIND)
 
 valgrind_cachegrind: $(PROG_CACHEGRIND)
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'
-	valgrind --tool=cachegrind --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128
+	valgrind --tool=cachegrind --cache-sim=yes --log-file=valgrind_cachegrind_log_.%p.txt ./$< 128 128
 	@echo '$$ cg_annotate --auto=yes cachegrind.out.<pid> | less'
 	@echo '(kcachegrind is required in order to visualize the output using the GUI)'
 	@echo 'oooOOO... valgrind_cachegrind ...OOOooo'

jacobi_solutions/serial/opt/src/jacobi_2D_serial_opt.c

@@ -212,13 +212,19 @@ void JacobiAlgorithm(MyData      **const restrict Phi,
     {
       for (int i=ibeg ; i<=iend ; i++)
 	{
-	  Phi[j][i] = 0.25 * (Phi0[j][i-1] + Phi0[j][i+1] +
-			      Phi0[j-1][i] + Phi0[j+1][i]);
-                
-	  *error += delta[X] * delta[Y] * fabs(Phi[j][i] - Phi0[j][i]);
+	  Phi[j][i] = 0.25 * (Phi0[j  ][i-1] +
+			      Phi0[j  ][i+1] +
+			      Phi0[j-1][i  ] +
+			      Phi0[j+1][i  ]);
+
+	  /* avoid fabs from math library */
+	  const MyData diff = (Phi[j][i] - Phi0[j][i]);
+	  *error += ((diff > 0) ? diff : -diff);
 	} /* loop over columns */
     } /* loop over rows */
   
+  *error *= (delta[X] * delta[Y]);
+  
   return;
 }
 

jacobi_solutions/serial/opt/src/tools.c

@@ -51,9 +51,8 @@ void Show_2DdblArray(const MyData **const A,
 
 double seconds()
 {
-  struct timeval tmp;
-  gettimeofday(&tmp, (struct timezone *)0);
-  double sec = tmp.tv_sec + ((double)tmp.tv_usec)/1000000.0;
-
-  return sec;
+  struct timespec ts;
+  return (clock_gettime( CLOCK_PROCESS_CPUTIME_ID, &ts ),
+	   (double)ts.tv_sec +
+	   (double)ts.tv_nsec * 1e-9);
 }