refurbished the ReproduceRandomNumbers input parameter

Setting ReproduceRandomNumbers = yes in the input file should now produce really identical random numbers not depending on MPI or scalar. Between scalar and MPI I do see a difference in the 11th digit of deltaH, which is rouding.
Marcogarofalo · Nov 25, 2012 · d9d22ff · d9d22ff
1 parent 35846e5
commit d9d22ff
Show file tree

Hide file tree

Showing 31 changed files with 326 additions and 327 deletions.
diff --git a/P_M_eta.c b/P_M_eta.c
@@ -391,7 +391,7 @@ void Check_Approximation(double const mstar) {
   Sin   =calloc(VOLUMEPLUSRAND, sizeof(spinor));
 #endif
 
-  random_spinor_field(Sin, VOLUME, 1);
+  random_spinor_field_lexic(Sin, 0);
 
   s_ = calloc(4*VOLUMEPLUSRAND+1, sizeof(spinor));
   s  = calloc(4, sizeof(spinor*));

diff --git a/Ptilde_nd.c b/Ptilde_nd.c
@@ -238,7 +238,7 @@ double chebtilde_eval(int M, double *dd, double s){
 void degree_of_Ptilde(int * _degree, double ** coefs,
 		      const double EVMin, const double EVMax,
 		      const int sloppy_degree, const double acc, 
-		      matrix_mult_nd Qsq) {
+		      matrix_mult_nd Qsq, const int repro) {
   int i, j;
   double temp, temp2;
   int degree;
@@ -307,8 +307,8 @@ void degree_of_Ptilde(int * _degree, double ** coefs,
   if(g_debug_level > 0) {
     /* Ptilde P S P  Ptilde X - X */
     /* for random spinor X        */
-    random_spinor_field(ss,VOLUME/2, 1);
-    random_spinor_field(sc,VOLUME/2, 1);
+    random_spinor_field_eo(ss, repro);
+    random_spinor_field_eo(sc, repro);
 
     Ptilde_ndpsi(&auxs[0], &auxc[0], *coefs, degree, &ss[0], &sc[0], Qsq);
     Ptilde_ndpsi(&aux2s[0], &aux2c[0], phmc_dop_cheby_coef, phmc_dop_n_cheby, &auxs[0], &auxc[0], Qsq);

diff --git a/Ptilde_nd.h b/Ptilde_nd.h
@@ -34,6 +34,6 @@ double chebtilde_eval(int M, double *dd, double s);
 void degree_of_Ptilde(int * _degree, double ** coefs, 
 		      const double EVMin, const double EVMax,
 		      const int sloppy_degree, const double acc,
-		      matrix_mult_nd Qsw);
+		      matrix_mult_nd Qsw, const int repro);
 
 #endif
diff --git a/benchmark.c b/benchmark.c
@@ -268,7 +268,7 @@ int main(int argc,char *argv[])
     j_max=2048;
     sdt=0.;
     for (k = 0; k < k_max; k++) {
-      random_spinor_field(g_spinor_field[k], VOLUME/2, 0);
+      random_spinor_field_eo(g_spinor_field[k], reproduce_randomnumber_flag);
     }
 
     while(sdt < 30.) {
@@ -366,7 +366,7 @@ int main(int argc,char *argv[])
     j_max=1;
     sdt=0.;
     for (k=0;k<k_max;k++) {
-      random_spinor_field(g_spinor_field[k], VOLUME, 0);
+      random_spinor_field_lexic(g_spinor_field[k], reproduce_randomnumber_flag);
     }
 
     while(sdt < 3.) {

diff --git a/chebyshev_polynomial.c b/chebyshev_polynomial.c
@@ -241,7 +241,7 @@ double stopeps=5.0e-16;
 int dop_n_cheby=0;
 double * dop_cheby_coef;
 
-void degree_of_polynomial(){
+void degree_of_polynomial(const int repro){
   int i;
   double temp;
   static int ini=0;
@@ -289,11 +289,11 @@ void degree_of_polynomial(){
    aux3c=calloc(VOLUMEPLUSRAND/2, sizeof(spinor));
 #endif
 
-chebyshev_polynomial(cheb_evmin, cheb_evmax, dop_cheby_coef, N_CHEBYMAX, 0.25);
+   chebyshev_polynomial(cheb_evmin, cheb_evmax, dop_cheby_coef, N_CHEBYMAX, 0.25);
 
    temp=1.0;
-   random_spinor_field(ss,VOLUME/2);
-   random_spinor_field(sc,VOLUME/2);
+   random_spinor_field_eo(ss, repro);
+   random_spinor_field_eo(sc, repro);
 /*   assign(&sc[0], &ss[0],VOLUME/2);
 
   Qtm_pm_psi(&auxs[0], &ss[0]);

diff --git a/chebyshev_polynomial.h b/chebyshev_polynomial.h
@@ -31,6 +31,6 @@ void chebyshev_polynomial(double a, double b, double c[], int n, double exponent
 
 void QdaggerQ_power(spinor *R_s, spinor *R_c, double *c, int n, spinor *S_s, spinor *S_c);
 
-void degree_of_polynomial();
+void degree_of_polynomial(const int repro);
 
 #endif
diff --git a/chebyshev_polynomial_nd.c b/chebyshev_polynomial_nd.c
@@ -104,7 +104,7 @@ double cheb_eval(int M, double *c, double s){
 
 void degree_of_polynomial_nd(int * _degree_of_p, double ** coefs,
 			     const double EVMin, const double EVMax,
-			     matrix_mult_nd Qsq) { 
+			     matrix_mult_nd Qsq, const int repro) { 
   double temp, temp2;
   int degree_of_p = *_degree_of_p + 1;
 
@@ -130,8 +130,8 @@ void degree_of_polynomial_nd(int * _degree_of_p, double ** coefs,
 
   chebyshev_coefs(EVMin, EVMax, *coefs, degree_of_p, -0.5);
 
-  random_spinor_field(ss,VOLUME/2, 1);
-  random_spinor_field(sc,VOLUME/2, 1);
+  random_spinor_field_eo(ss, repro);
+  random_spinor_field_eo(sc, repro);
 
   if((g_proc_id == g_stdio_proc) && (g_debug_level > 0)){
     printf("# NDPOLY MD Polynomial: EVmin = %e  EVmax = %e  \n", EVMin, EVMax);

diff --git a/chebyshev_polynomial_nd.h b/chebyshev_polynomial_nd.h
@@ -29,6 +29,6 @@ double cheb_eval(int M, double *c, double s);
 
 void degree_of_polynomial_nd(int * _degree_of_p, double ** coefs,
 			     const double EVMin, const double EVMax,
-			     matrix_mult_nd Qsq);
+			     matrix_mult_nd Qsq, const int repro);
 
 #endif
diff --git a/hopping_test.c b/hopping_test.c
@@ -270,7 +270,7 @@ int main(int argc,char *argv[])
     /*initialize the pseudo-fermion fields*/
     j_max=1;
     for (k = 0; k < k_max; k++) {
-      random_spinor_field(g_spinor_field[k], VOLUME/2, 0);
+      random_spinor_field_eo(g_spinor_field[k], reproduce_randomnumber_flag);
     }
 
     if (read_source_flag == 2) { /* save */

diff --git a/hybrid_update.c b/hybrid_update.c
@@ -105,120 +105,88 @@ double moment_energy(su3adj ** const momenta) {
 double init_momenta(const int repro, su3adj ** const momenta) {
 
   su3adj *xm;
-  int i, mu;
+  int i, mu, t0, x, y, z, X, Y, Z, t, id = 0;
+  int coords[4];
 #ifdef MPI
   int k;
   int rlxd_state[105];
 #endif
-  static double y[8];
-  static double tt,tr,ts,kc,ks,sum;
+  double ALIGN yy[8];
+  double ALIGN tt, tr, ts, kc = 0., ks = 0., sum;
 
-  if(repro == 1) {
-    if(g_proc_id==0){
-      kc=0.; 
-      ks=0.;
-      for(i=0;i<VOLUME;i++){ 
-	for(mu=0;mu<4;mu++){
-	  sum=0.;
-	  xm=&momenta[i][mu];
-	  gauss_vector(y,8);
-	  /* from the previous line we get exp(-y^2) distribution */
-	  /* this means that <y^2> = sigma^2 = 1/2 */
-	  /* in order to get <y^2> = 1 distribution ==> *sqrt(2) */
-	  (*xm).d1=1.4142135623731*y[0];
-	  (*xm).d2=1.4142135623731*y[1];
-	  sum+=(*xm).d1*(*xm).d1+(*xm).d2*(*xm).d2;
-	  (*xm).d3=1.4142135623731*y[2];
-	  (*xm).d4=1.4142135623731*y[3];
-	  sum+=(*xm).d3*(*xm).d3+(*xm).d4*(*xm).d4;
-	  (*xm).d5=1.4142135623731*y[4];
-	  (*xm).d6=1.4142135623731*y[5];
-	  sum+=(*xm).d5*(*xm).d5+(*xm).d6*(*xm).d6;
-	  (*xm).d7=1.4142135623731*y[6];
-	  (*xm).d8=1.4142135623731*y[7];
-	  sum+=(*xm).d7*(*xm).d7+(*xm).d8*(*xm).d8;
-	  tr=sum+kc;
-	  ts=tr+ks;
-	  tt=ts-ks;
-	  ks=ts;
-	  kc=tr-tt;
-	}
-      }
+  if(repro) {
 #ifdef MPI
-      /* send the state for the random-number generator to 1 */
+    if(g_proc_id == 0) {
       rlxd_get(rlxd_state);
-      MPI_Send(&rlxd_state[0], 105, MPI_INT, 1, 101, MPI_COMM_WORLD);
-#endif
     }
-
+    MPI_Bcast(rlxd_state, 105, MPI_INT, 0, MPI_COMM_WORLD);
+    rlxd_reset(rlxd_state);
+#endif
+    for(t0 = 0; t0 < g_nproc_t*T; t0++) {
+      t = t0 - T*g_proc_coords[0];
+      coords[0] = t0 / T;
+      for(x = 0; x < g_nproc_x*LX; x++) {
+	X = x - g_proc_coords[1]*LX;
+	coords[1] = x / LX;
+	for(y = 0; y < g_nproc_y*LY; y++) {
+	  Y = y - g_proc_coords[2]*LY;
+	  coords[2] = y / LY;
+	  for(z = 0; z < g_nproc_z*LZ; z++) {
+	    Z = z - g_proc_coords[3]*LZ;
+	    coords[3] = z / LZ;
 #ifdef MPI
-    if(g_proc_id != 0){
-      MPI_Recv(&rlxd_state[0], 105, MPI_INT, g_proc_id-1, 101, MPI_COMM_WORLD, &status);
-      rlxd_reset(rlxd_state);
-      kc=0.; ks=0.;
-      for(i=0;i<VOLUME;i++){ 
-	for(mu=0;mu<4;mu++){
-	  sum=0.;
-	  xm=&momenta[i][mu];
-	  gauss_vector(y,8);
-	  (*xm).d1=1.4142135623731*y[0];
-	  (*xm).d2=1.4142135623731*y[1];
-	  sum+=(*xm).d1*(*xm).d1+(*xm).d2*(*xm).d2;
-	  (*xm).d3=1.4142135623731*y[2];
-	  (*xm).d4=1.4142135623731*y[3];
-	  sum+=(*xm).d3*(*xm).d3+(*xm).d4*(*xm).d4;
-	  (*xm).d5=1.4142135623731*y[4];
-	  (*xm).d6=1.4142135623731*y[5];
-	  sum+=(*xm).d5*(*xm).d5+(*xm).d6*(*xm).d6;
-	  (*xm).d7=1.4142135623731*y[6];
-	  (*xm).d8=1.4142135623731*y[7];
-	  sum+=(*xm).d7*(*xm).d7+(*xm).d8*(*xm).d8;
-	  tr=sum+kc;
-	  ts=tr+ks;
-	  tt=ts-ks;
-	  ks=ts;
-	  kc=tr-tt;
+	    MPI_Cart_rank(g_cart_grid, coords, &id);
+#endif
+	    if(g_cart_id == id) i = g_ipt[t][X][Y][Z];
+	    for(mu = 0; mu < 4; mu++) {
+	      gauss_vector(yy,8);
+	      if(g_cart_id == id) {
+		sum = 0.;
+		xm = &momenta[i][mu];
+		(*xm).d1 = 1.4142135623731*yy[0];
+		(*xm).d2 = 1.4142135623731*yy[1];
+		sum += (*xm).d1*(*xm).d1+(*xm).d2*(*xm).d2;
+		(*xm).d3 = 1.4142135623731*yy[2];
+		(*xm).d4 = 1.4142135623731*yy[3];
+		sum += (*xm).d3*(*xm).d3+(*xm).d4*(*xm).d4;
+		(*xm).d5 = 1.4142135623731*yy[4];
+		(*xm).d6 = 1.4142135623731*yy[5];
+		sum += (*xm).d5*(*xm).d5+(*xm).d6*(*xm).d6;
+		(*xm).d7 = 1.4142135623731*yy[6];
+		(*xm).d8 = 1.4142135623731*yy[7];
+		sum += (*xm).d7*(*xm).d7+(*xm).d8*(*xm).d8;
+		tr = sum+kc;
+		ts = tr+ks;
+		tt = ts-ks;
+		ks = ts;
+		kc = tr-tt;
+	      }
+	    }
+	  }
 	}
       }
-      /* send the state fo the random-number 
-	 generator to next processor */
-
-      k=g_proc_id+1; 
-      if(k==g_nproc){ 
-	k=0;
-      }
-      rlxd_get(rlxd_state);
-      MPI_Send(&rlxd_state[0], 105, MPI_INT, k, 101, MPI_COMM_WORLD);
     }
-#endif
     kc=0.5*(ks+kc);
-
-#ifdef MPI
-    if(g_proc_id == 0){
-      MPI_Recv(&rlxd_state[0], 105, MPI_INT, g_nproc-1, 101, MPI_COMM_WORLD, &status);
-      rlxd_reset(rlxd_state);
-    }
-#endif
   }
   else {
     kc=0.; 
     ks=0.;
-    for(i=0;i<VOLUME;i++){ 
-      for(mu=0;mu<4;mu++){
+    for(i = 0; i < VOLUME; i++) { 
+      for(mu = 0; mu < 4; mu++) {
 	sum=0.;
 	xm=&momenta[i][mu];
-	gauss_vector(y,8);
-	(*xm).d1=1.4142135623731*y[0];
-	(*xm).d2=1.4142135623731*y[1];
+	gauss_vector(yy,8);
+	(*xm).d1=1.4142135623731*yy[0];
+	(*xm).d2=1.4142135623731*yy[1];
 	sum+=(*xm).d1*(*xm).d1+(*xm).d2*(*xm).d2;
-	(*xm).d3=1.4142135623731*y[2];
-	(*xm).d4=1.4142135623731*y[3];
+	(*xm).d3=1.4142135623731*yy[2];
+	(*xm).d4=1.4142135623731*yy[3];
 	sum+=(*xm).d3*(*xm).d3+(*xm).d4*(*xm).d4;
-	(*xm).d5=1.4142135623731*y[4];
-	(*xm).d6=1.4142135623731*y[5];
+	(*xm).d5=1.4142135623731*yy[4];
+	(*xm).d6=1.4142135623731*yy[5];
 	sum+=(*xm).d5*(*xm).d5+(*xm).d6*(*xm).d6;
-	(*xm).d7=1.4142135623731*y[6];
-	(*xm).d8=1.4142135623731*y[7];
+	(*xm).d7=1.4142135623731*yy[6];
+	(*xm).d8=1.4142135623731*yy[7];
 	sum+=(*xm).d7*(*xm).d7+(*xm).d8*(*xm).d8;
 	tr=sum+kc;
 	ts=tr+ks;
@@ -228,8 +196,6 @@ double init_momenta(const int repro, su3adj ** const momenta) {
       }
     }
     kc=0.5*(ks+kc);
-
-
   }
 #ifdef MPI
   MPI_Allreduce(&kc, &ks, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);

diff --git a/invert.c b/invert.c
@@ -450,11 +450,11 @@ int main(int argc, char *argv[])
       /* Compute little Dirac operators */
       /*       alt_block_compute_little_D(); */
       if (g_debug_level > 0) {
-        check_projectors();
-        check_local_D();
+        check_projectors(reproduce_randomnumber_flag);
+        check_local_D(reproduce_randomnumber_flag);
       }
       if (g_debug_level > 1) {
-        check_little_D_inversion();
+        check_little_D_inversion(reproduce_randomnumber_flag);
       }
 
     }

diff --git a/monomial/clover_trlog_monomial.c b/monomial/clover_trlog_monomial.c
@@ -67,7 +67,7 @@ double clover_trlog_acc(const int id, hamiltonian_field_t * const hf) {
   /*compute the contribution from the clover trlog term */
   mnl->energy1 = -sw_trace(EO, mnl->mu);
   if(g_proc_id == 0 && g_debug_level > 3) {
-    printf("called clover_trlog_acc for id %d dH = %1.4e\n", 
+    printf("called clover_trlog_acc for id %d dH = %1.10e\n", 
 	   id, mnl->energy1 - mnl->energy0);
   }
   return(mnl->energy1 - mnl->energy0);

diff --git a/monomial/cloverdet_monomial.c b/monomial/cloverdet_monomial.c
@@ -149,7 +149,7 @@ void cloverdet_heatbath(const int id, hamiltonian_field_t * const hf) {
   sw_term( (const su3**) hf->gaugefield, mnl->kappa, mnl->c_sw); 
   sw_invert(EE, mnl->mu);
 
-  random_spinor_field(mnl->w_fields[0], VOLUME/2, mnl->rngrepro);
+  random_spinor_field_eo(mnl->w_fields[0], mnl->rngrepro);
   mnl->energy0 = square_norm(mnl->w_fields[0], VOLUME/2, 1);
 
   mnl->Qp(mnl->pf, mnl->w_fields[0]);
@@ -193,7 +193,7 @@ double cloverdet_acc(const int id, hamiltonian_field_t * const hf) {
   g_mu3 = 0.;
   boundary(g_kappa);
   if(g_proc_id == 0 && g_debug_level > 3) {
-    printf("called cloverdet_acc for id %d %d dH = %1.4e\n", 
+    printf("called cloverdet_acc for id %d %d dH = %1.10e\n", 
 	   id, mnl->even_odd_flag, mnl->energy1 - mnl->energy0);
   }
   return(mnl->energy1 - mnl->energy0);

diff --git a/monomial/cloverdetratio_monomial.c b/monomial/cloverdetratio_monomial.c
@@ -240,7 +240,7 @@ void cloverdetratio_heatbath(const int id, hamiltonian_field_t * const hf) {
   sw_term( (const su3**) hf->gaugefield, mnl->kappa, mnl->c_sw); 
   sw_invert(EE, mnl->mu);
 
-  random_spinor_field(mnl->w_fields[0], VOLUME/2, mnl->rngrepro);
+  random_spinor_field_eo(mnl->w_fields[0], mnl->rngrepro);
   mnl->energy0  = square_norm(mnl->w_fields[0], VOLUME/2, 1);
 
   g_mu3 = mnl->rho;
@@ -291,7 +291,7 @@ double cloverdetratio_acc(const int id, hamiltonian_field_t * const hf) {
   g_mu3 = 0.;
   boundary(g_kappa);
   if(g_proc_id == 0 && g_debug_level > 3) {
-    printf("called cloverdetratio_acc for id %d dH = %1.4e\n", 
+    printf("called cloverdetratio_acc for id %d dH = %1.10e\n", 
 	   id, mnl->energy1 - mnl->energy0);
   }
   return(mnl->energy1 - mnl->energy0);

diff --git a/monomial/clovernd_trlog_monomial.c b/monomial/clovernd_trlog_monomial.c
@@ -67,7 +67,7 @@ double clovernd_trlog_acc(const int id, hamiltonian_field_t * const hf) {
   /*compute the contribution from the clover trlog term */
   mnl->energy1 = -sw_trace_nd(EE, mnl->mubar, mnl->epsbar);
   if(g_proc_id == 0 && g_debug_level > 3) {
-    printf("called clovernd_trlog_acc for id %d dH = %1.4e\n", 
+    printf("called clovernd_trlog_acc for id %d dH = %1.10e\n", 
 	   id, mnl->energy1 - mnl->energy0);
   }
   return(mnl->energy1 - mnl->energy0);