X-Git-Url: https://bilbo.iut-bm.univ-fcomte.fr/and/gitweb/GMRES2stage.git/blobdiff_plain/d085467b95f24aa53166c5284f70be0fd24da9fc..91273b11316f5330c60a8ab039e3102501f371ce:/code/ex15.c

diff --git a/code/ex15.c b/code/ex15.c
index 20144dd..7dc050b 100644
--- a/code/ex15.c
+++ b/code/ex15.c
@@ -1,4 +1,8 @@
 
+// /home/couturie/work/petsc-3.5.1/arch-linux2-c-debug/bin/mpirun -np 4    ./ex15 -m 400 -n 400 -ksp_type fgmres
+
+
+
 static char help[] = "Solves a linear system in parallel with KSP.  Also\n\
 illustrates setting a user-defined shell preconditioner and using the\n\
 macro __FUNCT__ to define routine names for use in error handling.\n\
@@ -51,11 +55,11 @@ int KrylovMinimize(Mat A, Vec b, Vec x) {
 
   //Variables
 
-  PetscScalar  gamma, alpha, oldgamma, beta, t2;
+  PetscScalar  gamma, alpha, oldgamma, beta;
   PetscReal norm=20, Eprecision=1e-8, cgprec=1e-40;     
   PetscInt giter=0, ColS=12, col=0, Emaxiter=50000000, iter=0, iterations=15, Iiter=0;
   PetscErrorCode ierr;
-  PetscScalar T1, T2, t1;
+  PetscScalar T1, T2;
   KSP ksp;
   PetscInt total=0;  
   PetscInt size;
@@ -66,7 +70,6 @@ int KrylovMinimize(Mat A, Vec b, Vec x) {
   PetscScalar *array;
   PetscInt *ind_row;
   Vec Alpha, p, ss, vect, r, q, Ax;
-  PetscScalar norm_ksp;
 
 
   PetscInt first=1;
@@ -117,7 +120,7 @@ int KrylovMinimize(Mat A, Vec b, Vec x) {
 
   //Initializations
   //  ierr = KSPGMRESSetRestart(ksp, 16); CHKERRQ(ierr);
-  ierr = KSPSetTolerances(ksp, 1e-10, 1e-10, PETSC_DEFAULT, 16); CHKERRQ(ierr);
+  ierr = KSPSetTolerances(ksp, 1e-13, 1e-13, PETSC_DEFAULT, 16); CHKERRQ(ierr);
   ierr = KSPSetInitialGuessNonzero(ksp, PETSC_TRUE); CHKERRQ(ierr);
 
 
@@ -143,11 +146,13 @@ int KrylovMinimize(Mat A, Vec b, Vec x) {
 
 
 
-      //Error
+      KSPGetResidualNorm(ksp,&norm);
+
+      /*      //Error
       ierr = VecCopy(x, residu); CHKERRQ(ierr);
       ierr = VecAXPY(residu, -1, x_old); CHKERRQ(ierr);
       ierr = VecNorm(residu, NORM_INFINITY, &norm); CHKERRQ(ierr);
-
+       */
 
 
       ierr = PetscPrintf(PETSC_COMM_WORLD, "Norm of error %g, outer iteration %D\n", norm, giter); CHKERRQ(ierr);
@@ -207,7 +212,6 @@ int KrylovMinimize(Mat A, Vec b, Vec x) {
 
   }
   T2 = MPI_Wtime();
-  t1 = T2 - T1;
   ierr = PetscPrintf(PETSC_COMM_WORLD, "\t\t\t -- Execution time : %g (s)\n", T2-T1); CHKERRQ(ierr);
   ierr = PetscPrintf(PETSC_COMM_WORLD, "\t\t\t -- Total number of iterations : %D\n", total); CHKERRQ(ierr);
 
@@ -222,6 +226,245 @@ int KrylovMinimize(Mat A, Vec b, Vec x) {
 
 
 
+
+int KrylovMinimizeLSQR(Mat A, Vec b, Vec x) {
+  
+
+  //Variables
+
+  PetscScalar  alpha, beta;
+  PetscReal norm=20, Eprecision=1e-8, tol=1e-40;     
+  PetscInt giter=0, ColS=12, col=0, Emaxiter=50000000, iter=0, iterations=20, Iiter=0;
+  PetscErrorCode ierr;
+  PetscScalar T1, T2;
+  KSP ksp;
+  PetscInt total=0;  
+  PetscInt size;
+  PetscInt Istart,Iend;
+  PetscInt i,its;
+  Vec       x_old, residu;
+  Mat S, AS;
+  PetscScalar *array;
+  PetscInt *ind_row;
+  Vec  Ax;
+  PetscScalar norm_ksp;
+  Vec u,v,d,uu,vt,zero_long,zero_short,x_lsqr;
+
+  PetscInt first=1;
+
+  ierr = KSPCreate(PETSC_COMM_WORLD,&ksp);CHKERRQ(ierr);
+  ierr = KSPSetOperators(ksp,A,A);CHKERRQ(ierr);
+  ierr = KSPSetFromOptions(ksp);CHKERRQ(ierr);
+
+
+
+
+  VecGetSize(b,&size);
+
+  ierr = PetscPrintf(PETSC_COMM_WORLD, "Size of vector %D\n", size); CHKERRQ(ierr);  
+
+  PetscInt aa,bb;
+  MatGetOwnershipRange(A,&aa,&bb);
+
+  //  ierr = PetscSynchronizedPrintf(PETSC_COMM_WORLD, "%D %D\n", aa,bb); CHKERRQ(ierr);  
+  //PetscSynchronizedFlush(PETSC_COMM_WORLD);
+
+
+  ierr = MatCreate(PETSC_COMM_WORLD, &S);CHKERRQ(ierr);
+  ierr = MatSetSizes(S, bb-aa,  PETSC_DECIDE, size, ColS); CHKERRQ(ierr);
+  ierr = MatSetType(S, MATMPIDENSE); CHKERRQ(ierr);
+  ierr = MatSetUp(S); CHKERRQ(ierr);
+
+  ierr = MatGetOwnershipRange(S, &Istart, &Iend); CHKERRQ(ierr);
+
+
+
+  ierr = VecDuplicate(b, &Ax); CHKERRQ(ierr);
+
+  ierr = VecDuplicate(b,&x_old);CHKERRQ(ierr);
+  ierr = VecDuplicate(b,&residu);CHKERRQ(ierr);
+
+
+  //long vector
+  ierr = VecDuplicate(b,&u);CHKERRQ(ierr);
+
+
+  ierr = VecDuplicate(b,&uu);CHKERRQ(ierr);
+  ierr = VecDuplicate(b,&zero_long);CHKERRQ(ierr);
+  ierr = VecSet(zero_long,0);CHKERRQ(ierr);
+
+  //small vector
+  ierr = VecCreate(PETSC_COMM_WORLD, &v); CHKERRQ(ierr);
+  ierr = VecSetSizes(v, PETSC_DECIDE, ColS); CHKERRQ(ierr);
+  ierr = VecSetFromOptions(v); CHKERRQ(ierr);
+  ierr = VecDuplicate(v,&zero_short);CHKERRQ(ierr);
+  ierr = VecSet(zero_short,0);CHKERRQ(ierr);
+  ierr = VecDuplicate(v,&d);CHKERRQ(ierr);
+  ierr = VecDuplicate(v,&vt);CHKERRQ(ierr);
+  ierr = VecDuplicate(v,&x_lsqr);CHKERRQ(ierr);
+
+
+  //indexes of row (these indexes are global)
+  ind_row = (PetscInt*)malloc(sizeof(PetscInt)*(Iend-Istart));
+  for(i=0; i<Iend-Istart; i++) ind_row[i] = i + Istart;
+
+  //Initializations
+  //  ierr = KSPGMRESSetRestart(ksp, 16); CHKERRQ(ierr);
+  ierr = KSPSetTolerances(ksp, 1e-10, 1e-10, PETSC_DEFAULT, 16); CHKERRQ(ierr);
+  ierr = KSPSetInitialGuessNonzero(ksp, PETSC_TRUE); CHKERRQ(ierr);
+
+
+
+
+  //GMRES WITH MINIMIZATION
+  T1 = MPI_Wtime();
+  while(giter<Emaxiter && norm>Eprecision ){
+    for(col=0; col<ColS  &&  norm>Eprecision; col++){
+
+
+      //Solve
+      ierr = KSPSolve(ksp, b, x); CHKERRQ(ierr);
+
+      ierr = KSPGetIterationNumber(ksp, &its); CHKERRQ(ierr);
+      total += its; Iiter ++;
+
+
+
+      //Build S'
+      ierr = VecGetArray(x, &array);
+      ierr = MatSetValues(S, Iend-Istart, ind_row, 1, &col, array, INSERT_VALUES);
+      VecRestoreArray(x, &array);
+
+
+
+      KSPGetResidualNorm(ksp,&norm);
+
+      /*
+      //Error
+      ierr = VecCopy(x, residu); CHKERRQ(ierr);
+      ierr = VecAXPY(residu, -1, x_old); CHKERRQ(ierr);
+      ierr = VecNorm(residu, NORM_INFINITY, &norm); CHKERRQ(ierr);
+       */
+
+
+      ierr = PetscPrintf(PETSC_COMM_WORLD, "Norm of error %g, outer iteration %D\n", norm, giter); CHKERRQ(ierr);
+      ierr = VecCopy(x, x_old); CHKERRQ(ierr);
+
+
+    }
+
+
+    //minimization step
+    if( norm>Eprecision) {
+
+      MatAssemblyBegin(S, MAT_FINAL_ASSEMBLY);
+      MatAssemblyEnd(S, MAT_FINAL_ASSEMBLY);
+
+
+
+
+      //Build AS
+      if(first) {
+        MatMatMult(A,S, MAT_INITIAL_MATRIX,PETSC_DEFAULT,&AS);
+        
+        first=0;
+      }
+      else
+        MatMatMult(A,S, MAT_REUSE_MATRIX,PETSC_DEFAULT,&AS);
+
+
+
+
+
+      //LSQR
+      //LSQR
+      //LSQR
+
+
+
+      PetscScalar n2b,tolb,normr,c,s,phibar,normar,norma,thet,rhot,rho,phi;
+      PetscInt stag;
+      tolb = tol * n2b;
+      VecNorm(b, NORM_2, &n2b); //n2b = norm(b);
+      ierr = VecCopy(b, u); CHKERRQ(ierr);   //u=b
+      VecNorm(u, NORM_2, &beta); // beta=norm(u)
+      normr=beta;
+      if (beta != 0) {
+        VecAYPX(u,1/beta,zero_long); //  u = u / beta;
+      }
+      c=1;
+      s=0;
+      phibar=beta;
+      MatMultTranspose(AS, u, v); //v=A'*u
+      ierr = VecSet(x_lsqr,0);CHKERRQ(ierr);
+      VecNorm(v, NORM_2, &alpha); // alpha=norm(v)
+      if (alpha != 0) {
+        VecAYPX(v,1/alpha,zero_short); //  v = v / alpha;
+      }
+      ierr = VecSet(d,0);CHKERRQ(ierr);
+      normar = alpha * beta;
+      norma=0;
+      //stag=0;
+      for(i=0;i<iterations;i++) {
+        MatMult(AS, v, uu); //uu=A*v
+        VecAYPX(u, -alpha, uu); //u = uu-alpha*u;
+        VecNorm(u, NORM_2, &beta); // beta=norm(u)
+        VecAYPX(u,1/beta,zero_long); //  u = u / beta;
+        norma=sqrt(norma*norma+alpha*alpha+beta*beta); // norma = norm([norma alpha beta]);
+        thet = - s * alpha;
+        rhot = c * alpha;
+        rho = sqrt(rhot*rhot + beta*beta);
+        c = rhot / rho;
+        s = - beta / rho;
+        phi = c * phibar;
+        if (phi == 0) {             // stagnation of the method
+          stag = 1;
+        }
+        phibar = s * phibar;
+        VecAYPX(d,-thet,v);       //d = (v - thet * d);
+        VecAYPX(d,1/rho,zero_short);     //d=d/ rho;
+
+        /*
+        if (normar/(norma*normr) <= tol) { // check for convergence in min{|b-A*x|}
+          break;
+        }
+        if (normr <= tolb) {           // check for convergence in A*x=b
+          break;
+        }
+         */
+
+        VecAXPY(x_lsqr,phi,d);     // x_lsqr=x_lsqr+phi*d
+        normr = abs(s) * normr;
+        MatMultTranspose(AS, u, vt); //vt=A'*u;
+        VecAYPX(v,-beta,vt);  // v = vt - beta * v;
+        VecNorm(v, NORM_2, &alpha); // alpha=norm(v)
+        VecAYPX(v,1/alpha,zero_short); //  v = v / alpha;
+        normar = alpha * abs( s * phi);
+      }	
+      
+
+
+      iter = 0; giter ++;
+      //Minimizer
+      MatMult(S, x_lsqr, x); //x = S*x_small;
+    }
+
+  }
+  T2 = MPI_Wtime();
+  ierr = PetscPrintf(PETSC_COMM_WORLD, "\t\t\t -- Execution time LSQR : %g (s)\n", T2-T1); CHKERRQ(ierr);
+  ierr = PetscPrintf(PETSC_COMM_WORLD, "\t\t\t -- Total number of iterations LSQR : %D\n", total); CHKERRQ(ierr);
+
+  return 0;
+
+}
+
+
+
+
+
+
+
+
 int main(int argc,char **args)
 {
   Vec            x,b,u;   /* approx solution, RHS, exact solution */
@@ -229,11 +472,9 @@ int main(int argc,char **args)
   KSP            ksp;      /* linear solver context */
   PC             pc;        /* preconditioner context */
   PetscReal      norm;      /* norm of solution error */
-  SampleShellPC  *shell;    /* user-defined preconditioner context */
-  PetscScalar    v,one = 1.0,none = -1.0;
+  PetscScalar    v,one = 1.0;
   PetscInt       i,j,Ii,J,Istart,Iend,m = 8,n = 7,its;
   PetscErrorCode ierr;
-  PetscBool      user_defined_pc = PETSC_FALSE;
 
   PetscInitialize(&argc,&args,(char*)0,help);
   ierr = PetscOptionsGetInt(NULL,"-m",&m,NULL);CHKERRQ(ierr);
@@ -387,8 +628,25 @@ int main(int argc,char **args)
   }
 
 
+ {
 
+    Vec x2;
+    Vec sol;
+    VecDuplicate(b,&x2);
+    VecDuplicate(b,&sol);
+    
+    KrylovMinimizeLSQR(A, b, x2);
+
+
+
+    MatMult(A,x2,sol);
+    VecAXPY(sol,-1,b);
+    VecNorm(sol, NORM_2, &norm);
+    ierr = PetscPrintf(PETSC_COMM_WORLD, "\t\t\t -- Error Krylov Minimization LSQR %g\n",norm);
 
+  }
+
+  
 
   /*
      Free work space.  All PETSc objects should be destroyed when they