xref: /petsc/src/mat/tests/ex204.c (revision 609caa7c8c030312b00807b4f015fd827bb80932)

static char help[] = "Test ViennaCL Matrix Conversions";

#include <petscmat.h>

int main(int argc, char **args)
{
  PetscMPIInt rank, size;

  PetscFunctionBeginUser;
  PetscCall(PetscInitialize(&argc, &args, NULL, help));

  PetscCallMPI(MPI_Comm_rank(PETSC_COMM_WORLD, &rank));
  PetscCallMPI(MPI_Comm_size(PETSC_COMM_WORLD, &size));

  /* Construct a sequential ViennaCL matrix and vector */
  if (rank == 0) {
    Mat               A_vcl;
    Vec               v_vcl, r_vcl;
    PetscInt          n = 17, m = 31, nz = 5, i, cnt, j;
    const PetscScalar val = 1.0;

    PetscCall(MatCreateSeqAIJViennaCL(PETSC_COMM_SELF, m, n, nz, NULL, &A_vcl));

    /* Add nz arbitrary entries per row in arbitrary columns */
    for (i = 0; i < m; ++i) {
      for (cnt = 0; cnt < nz; ++cnt) {
        j = (19 * cnt + (7 * i + 3)) % n;
        PetscCall(MatSetValue(A_vcl, i, j, (PetscScalar)(0.3 * i + j), INSERT_VALUES));
      }
    }
    PetscCall(MatAssemblyBegin(A_vcl, MAT_FINAL_ASSEMBLY));
    PetscCall(MatAssemblyEnd(A_vcl, MAT_FINAL_ASSEMBLY));

    PetscCall(VecCreateSeqViennaCL(PETSC_COMM_SELF, n, &v_vcl));
    PetscCall(VecCreateSeqViennaCL(PETSC_COMM_SELF, m, &r_vcl));
    PetscCall(VecSet(v_vcl, val));

    PetscCall(MatMult(A_vcl, v_vcl, r_vcl));

    PetscCall(VecDestroy(&v_vcl));
    PetscCall(VecDestroy(&r_vcl));
    PetscCall(MatDestroy(&A_vcl));
  }

  /* Create a sequential AIJ matrix on rank 0 convert it to a new ViennaCL matrix, and apply it to a ViennaCL vector */
  if (rank == 0) {
    Mat               A, A_vcl;
    Vec               v, r, v_vcl, r_vcl, d_vcl;
    PetscInt          n = 17, m = 31, nz = 5, i, cnt, j;
    const PetscScalar val = 1.0;
    PetscReal         dnorm;
    const PetscReal   tol = 1e-5;

    PetscCall(MatCreateSeqAIJ(PETSC_COMM_SELF, m, n, nz, NULL, &A));

    /* Add nz arbitrary entries per row in arbitrary columns */
    for (i = 0; i < m; ++i) {
      for (cnt = 0; cnt < nz; ++cnt) {
        j = (19 * cnt + (7 * i + 3)) % n;
        PetscCall(MatSetValue(A, i, j, (PetscScalar)(0.3 * i + j), INSERT_VALUES));
      }
    }
    PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
    PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));

    PetscCall(PetscObjectSetName((PetscObject)A, "Sequential CPU Matrix"));

    PetscCall(VecCreateSeq(PETSC_COMM_SELF, n, &v));
    PetscCall(VecCreateSeq(PETSC_COMM_SELF, m, &r));
    PetscCall(PetscObjectSetName((PetscObject)r, "CPU result vector"));
    PetscCall(VecSet(v, val));
    PetscCall(MatMult(A, v, r));

    PetscCall(MatConvert(A, MATSEQAIJVIENNACL, MAT_INITIAL_MATRIX, &A_vcl));
    PetscCall(PetscObjectSetName((PetscObject)A_vcl, "New ViennaCL Matrix"));

    PetscCall(VecCreateSeqViennaCL(PETSC_COMM_SELF, n, &v_vcl));
    PetscCall(VecCreateSeqViennaCL(PETSC_COMM_SELF, m, &r_vcl));
    PetscCall(PetscObjectSetName((PetscObject)r_vcl, "ViennaCL result vector"));
    PetscCall(VecSet(v_vcl, val));
    PetscCall(MatMult(A_vcl, v_vcl, r_vcl));

    PetscCall(VecDuplicate(r_vcl, &d_vcl));
    PetscCall(VecCopy(r_vcl, d_vcl));
    PetscCall(VecAXPY(d_vcl, -1.0, r_vcl));
    PetscCall(VecNorm(d_vcl, NORM_INFINITY, &dnorm));
    PetscCheck(dnorm <= tol, PETSC_COMM_WORLD, PETSC_ERR_ARG_INCOMP, "Sequential CPU and MPI ViennaCL vector results incompatible with inf norm greater than tolerance of %g", tol);

    PetscCall(VecDestroy(&v));
    PetscCall(VecDestroy(&r));
    PetscCall(VecDestroy(&v_vcl));
    PetscCall(VecDestroy(&r_vcl));
    PetscCall(VecDestroy(&d_vcl));
    PetscCall(MatDestroy(&A));
    PetscCall(MatDestroy(&A_vcl));
  }

  /* Create a sequential AIJ matrix on rank 0 convert it inplace to a new ViennaCL matrix, and apply it to a ViennaCL vector */
  if (rank == 0) {
    Mat               A;
    Vec               v, r, v_vcl, r_vcl, d_vcl;
    PetscInt          n = 17, m = 31, nz = 5, i, cnt, j;
    const PetscScalar val = 1.0;
    PetscReal         dnorm;
    const PetscReal   tol = 1e-5;

    PetscCall(MatCreateSeqAIJ(PETSC_COMM_SELF, m, n, nz, NULL, &A));

    /* Add nz arbitrary entries per row in arbitrary columns */
    for (i = 0; i < m; ++i) {
      for (cnt = 0; cnt < nz; ++cnt) {
        j = (19 * cnt + (7 * i + 3)) % n;
        PetscCall(MatSetValue(A, i, j, (PetscScalar)(0.3 * i + j), INSERT_VALUES));
      }
    }
    PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
    PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));

    PetscCall(PetscObjectSetName((PetscObject)A, "Sequential CPU Matrix"));

    PetscCall(VecCreateSeq(PETSC_COMM_SELF, n, &v));
    PetscCall(VecCreateSeq(PETSC_COMM_SELF, m, &r));
    PetscCall(PetscObjectSetName((PetscObject)r, "CPU result vector"));
    PetscCall(VecSet(v, val));
    PetscCall(MatMult(A, v, r));

    PetscCall(MatConvert(A, MATSEQAIJVIENNACL, MAT_INPLACE_MATRIX, &A));
    PetscCall(PetscObjectSetName((PetscObject)A, "Converted ViennaCL Matrix"));

    PetscCall(VecCreateSeqViennaCL(PETSC_COMM_SELF, n, &v_vcl));
    PetscCall(VecCreateSeqViennaCL(PETSC_COMM_SELF, m, &r_vcl));
    PetscCall(PetscObjectSetName((PetscObject)r_vcl, "ViennaCL result vector"));
    PetscCall(VecSet(v_vcl, val));
    PetscCall(MatMult(A, v_vcl, r_vcl));

    PetscCall(VecDuplicate(r_vcl, &d_vcl));
    PetscCall(VecCopy(r_vcl, d_vcl));
    PetscCall(VecAXPY(d_vcl, -1.0, r_vcl));
    PetscCall(VecNorm(d_vcl, NORM_INFINITY, &dnorm));
    PetscCheck(dnorm <= tol, PETSC_COMM_WORLD, PETSC_ERR_ARG_INCOMP, "MPI CPU and MPI ViennaCL Vector results incompatible with inf norm greater than tolerance of %g", tol);

    PetscCall(VecDestroy(&v));
    PetscCall(VecDestroy(&r));
    PetscCall(VecDestroy(&v_vcl));
    PetscCall(VecDestroy(&r_vcl));
    PetscCall(VecDestroy(&d_vcl));
    PetscCall(MatDestroy(&A));
  }

  /* Create a parallel AIJ matrix, convert it to a new ViennaCL matrix, and apply it to a ViennaCL vector */
  if (size > 1) {
    Mat               A, A_vcl;
    Vec               v, r, v_vcl, r_vcl, d_vcl;
    PetscInt          N = 17, M = 31, nz = 5, i, cnt, j, rlow, rhigh;
    const PetscScalar val = 1.0;
    PetscReal         dnorm;
    const PetscReal   tol = 1e-5;

    PetscCall(MatCreateAIJ(PETSC_COMM_WORLD, PETSC_DETERMINE, PETSC_DETERMINE, M, N, nz, NULL, nz, NULL, &A));

    /* Add nz arbitrary entries per row in arbitrary columns */
    PetscCall(MatGetOwnershipRange(A, &rlow, &rhigh));
    for (i = rlow; i < rhigh; ++i) {
      for (cnt = 0; cnt < nz; ++cnt) {
        j = (19 * cnt + (7 * i + 3)) % N;
        PetscCall(MatSetValue(A, i, j, (PetscScalar)(0.3 * i + j), INSERT_VALUES));
      }
    }
    PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
    PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));

    PetscCall(PetscObjectSetName((PetscObject)A, "MPI CPU Matrix"));

    PetscCall(MatCreateVecs(A, &v, &r));
    PetscCall(PetscObjectSetName((PetscObject)r, "MPI CPU result vector"));
    PetscCall(VecSet(v, val));
    PetscCall(MatMult(A, v, r));

    PetscCall(MatConvert(A, MATMPIAIJVIENNACL, MAT_INITIAL_MATRIX, &A_vcl));
    PetscCall(PetscObjectSetName((PetscObject)A_vcl, "New MPI ViennaCL Matrix"));

    PetscCall(MatCreateVecs(A_vcl, &v_vcl, &r_vcl));
    PetscCall(PetscObjectSetName((PetscObject)r_vcl, "ViennaCL result vector"));
    PetscCall(VecSet(v_vcl, val));
    PetscCall(MatMult(A_vcl, v_vcl, r_vcl));

    PetscCall(VecDuplicate(r_vcl, &d_vcl));
    PetscCall(VecCopy(r_vcl, d_vcl));
    PetscCall(VecAXPY(d_vcl, -1.0, r_vcl));
    PetscCall(VecNorm(d_vcl, NORM_INFINITY, &dnorm));
    PetscCheck(dnorm <= tol, PETSC_COMM_WORLD, PETSC_ERR_ARG_INCOMP, "MPI CPU and MPI ViennaCL Vector results incompatible with inf norm greater than tolerance of %g", tol);

    PetscCall(VecDestroy(&v));
    PetscCall(VecDestroy(&r));
    PetscCall(VecDestroy(&v_vcl));
    PetscCall(VecDestroy(&r_vcl));
    PetscCall(VecDestroy(&d_vcl));
    PetscCall(MatDestroy(&A));
    PetscCall(MatDestroy(&A_vcl));
  }

  /* Create a parallel AIJ matrix, convert it in-place to a ViennaCL matrix, and apply it to a ViennaCL vector */
  if (size > 1) {
    Mat               A;
    Vec               v, r, v_vcl, r_vcl, d_vcl;
    PetscInt          N = 17, M = 31, nz = 5, i, cnt, j, rlow, rhigh;
    const PetscScalar val = 1.0;
    PetscReal         dnorm;
    const PetscReal   tol = 1e-5;

    PetscCall(MatCreateAIJ(PETSC_COMM_WORLD, PETSC_DETERMINE, PETSC_DETERMINE, M, N, nz, NULL, nz, NULL, &A));

    /* Add nz arbitrary entries per row in arbitrary columns */
    PetscCall(MatGetOwnershipRange(A, &rlow, &rhigh));
    for (i = rlow; i < rhigh; ++i) {
      for (cnt = 0; cnt < nz; ++cnt) {
        j = (19 * cnt + (7 * i + 3)) % N;
        PetscCall(MatSetValue(A, i, j, (PetscScalar)(0.3 * i + j), INSERT_VALUES));
      }
    }
    PetscCall(MatAssemblyBegin(A, MAT_FINAL_ASSEMBLY));
    PetscCall(MatAssemblyEnd(A, MAT_FINAL_ASSEMBLY));

    PetscCall(PetscObjectSetName((PetscObject)A, "MPI CPU Matrix"));

    PetscCall(MatCreateVecs(A, &v, &r));
    PetscCall(PetscObjectSetName((PetscObject)r, "MPI CPU result vector"));
    PetscCall(VecSet(v, val));
    PetscCall(MatMult(A, v, r));

    PetscCall(MatConvert(A, MATMPIAIJVIENNACL, MAT_INPLACE_MATRIX, &A));
    PetscCall(PetscObjectSetName((PetscObject)A, "Converted MPI ViennaCL Matrix"));

    PetscCall(MatCreateVecs(A, &v_vcl, &r_vcl));
    PetscCall(PetscObjectSetName((PetscObject)r_vcl, "ViennaCL result vector"));
    PetscCall(VecSet(v_vcl, val));
    PetscCall(MatMult(A, v_vcl, r_vcl));

    PetscCall(VecDuplicate(r_vcl, &d_vcl));
    PetscCall(VecCopy(r_vcl, d_vcl));
    PetscCall(VecAXPY(d_vcl, -1.0, r_vcl));
    PetscCall(VecNorm(d_vcl, NORM_INFINITY, &dnorm));
    PetscCheck(dnorm <= tol, PETSC_COMM_WORLD, PETSC_ERR_ARG_INCOMP, "MPI CPU and MPI ViennaCL Vector results incompatible with inf norm greater than tolerance of %g", tol);

    PetscCall(VecDestroy(&v));
    PetscCall(VecDestroy(&r));
    PetscCall(VecDestroy(&v_vcl));
    PetscCall(VecDestroy(&r_vcl));
    PetscCall(VecDestroy(&d_vcl));
    PetscCall(MatDestroy(&A));
  }

  PetscCall(PetscFinalize());
  return 0;
}

/*TEST

   build:
      requires: viennacl

   test:
      output_file: output/empty.out

   test:
      suffix: 2
      nsize: 2
      output_file: output/empty.out

TEST*/