1c4762a1bSJed Brown static const char help[] = "Test overlapped communication on a single star forest (PetscSF)\n\n"; 2c4762a1bSJed Brown 3c4762a1bSJed Brown #include <petscvec.h> 4c4762a1bSJed Brown #include <petscsf.h> 5c4762a1bSJed Brown #include <petscviewer.h> 6c4762a1bSJed Brown 7c4762a1bSJed Brown int main(int argc, char **argv) 8c4762a1bSJed Brown { 9c4762a1bSJed Brown PetscInt ierr; 10c4762a1bSJed Brown PetscSF sf; 11c4762a1bSJed Brown Vec A,Aout; 12c4762a1bSJed Brown Vec B,Bout; 13c4762a1bSJed Brown PetscScalar *bufA; 14c4762a1bSJed Brown PetscScalar *bufAout; 15c4762a1bSJed Brown PetscScalar *bufB; 16c4762a1bSJed Brown PetscScalar *bufBout; 17c4762a1bSJed Brown PetscMPIInt rank, size; 18c4762a1bSJed Brown PetscInt nroots, nleaves; 19c4762a1bSJed Brown PetscInt i; 20c4762a1bSJed Brown PetscInt *ilocal; 21c4762a1bSJed Brown PetscSFNode *iremote; 22c4762a1bSJed Brown 23c4762a1bSJed Brown ierr = PetscInitialize(&argc,&argv,NULL,help);if (ierr) return ierr; 24*5f80ce2aSJacob Faibussowitsch CHKERRMPI(MPI_Comm_rank(PETSC_COMM_WORLD,&rank)); 25*5f80ce2aSJacob Faibussowitsch CHKERRMPI(MPI_Comm_size(PETSC_COMM_WORLD,&size)); 26c4762a1bSJed Brown 272c71b3e2SJacob Faibussowitsch PetscCheckFalse(size != 2,PETSC_COMM_WORLD, PETSC_ERR_USER, "Only coded for two MPI processes"); 28c4762a1bSJed Brown 29*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFCreate(PETSC_COMM_WORLD,&sf)); 30*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFSetFromOptions(sf)); 31c4762a1bSJed Brown 32c4762a1bSJed Brown nleaves = 2; 33c4762a1bSJed Brown nroots = 1; 34*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscMalloc1(nleaves,&ilocal)); 35c4762a1bSJed Brown 36c4762a1bSJed Brown for (i = 0; i<nleaves; i++) { 37c4762a1bSJed Brown ilocal[i] = i; 38c4762a1bSJed Brown } 39c4762a1bSJed Brown 40*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscMalloc1(nleaves,&iremote)); 41c4762a1bSJed Brown if (rank == 0) { 42c4762a1bSJed Brown iremote[0].rank = 0; 43c4762a1bSJed Brown iremote[0].index = 0; 44c4762a1bSJed Brown iremote[1].rank = 1; 45c4762a1bSJed Brown iremote[1].index = 0; 46c4762a1bSJed Brown } else { 47c4762a1bSJed Brown iremote[0].rank = 1; 48c4762a1bSJed Brown iremote[0].index = 0; 49c4762a1bSJed Brown iremote[1].rank = 0; 50c4762a1bSJed Brown iremote[1].index = 0; 51c4762a1bSJed Brown } 52*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFSetGraph(sf,nroots,nleaves,ilocal,PETSC_OWN_POINTER,iremote,PETSC_OWN_POINTER)); 53*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFSetUp(sf)); 54*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFView(sf,PETSC_VIEWER_STDOUT_WORLD)); 55*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecCreate(PETSC_COMM_WORLD,&A)); 56*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecSetSizes(A,2,PETSC_DETERMINE)); 57*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecSetFromOptions(A)); 58*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecSetUp(A)); 59c4762a1bSJed Brown 60*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecDuplicate(A,&B)); 61*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecDuplicate(A,&Aout)); 62*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecDuplicate(A,&Bout)); 63*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecGetArray(A,&bufA)); 64*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecGetArray(B,&bufB)); 65c4762a1bSJed Brown for (i=0; i<2; i++) { 66c4762a1bSJed Brown bufA[i] = (PetscScalar)rank; 67c4762a1bSJed Brown bufB[i] = (PetscScalar)(rank) + 10.0; 68c4762a1bSJed Brown } 69*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecRestoreArray(A,&bufA)); 70*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecRestoreArray(B,&bufB)); 71c4762a1bSJed Brown 72*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecGetArrayRead(A,(const PetscScalar**)&bufA)); 73*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecGetArrayRead(B,(const PetscScalar**)&bufB)); 74*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecGetArray(Aout,&bufAout)); 75*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecGetArray(Bout,&bufBout)); 76*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFBcastBegin(sf,MPIU_SCALAR,(const void*)bufA,(void *)bufAout,MPI_REPLACE)); 77*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFBcastBegin(sf,MPIU_SCALAR,(const void*)bufB,(void *)bufBout,MPI_REPLACE)); 78*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFBcastEnd(sf,MPIU_SCALAR,(const void*)bufA,(void *)bufAout,MPI_REPLACE)); 79*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFBcastEnd(sf,MPIU_SCALAR,(const void*)bufB,(void *)bufBout,MPI_REPLACE)); 80*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecRestoreArrayRead(A,(const PetscScalar**)&bufA)); 81*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecRestoreArrayRead(B,(const PetscScalar**)&bufB)); 82*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecRestoreArray(Aout,&bufAout)); 83*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecRestoreArray(Bout,&bufBout)); 84c4762a1bSJed Brown 85*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecView(Aout,PETSC_VIEWER_STDOUT_WORLD)); 86*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecView(Bout,PETSC_VIEWER_STDOUT_WORLD)); 87*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecDestroy(&A)); 88*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecDestroy(&B)); 89*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecDestroy(&Aout)); 90*5f80ce2aSJacob Faibussowitsch CHKERRQ(VecDestroy(&Bout)); 91*5f80ce2aSJacob Faibussowitsch CHKERRQ(PetscSFDestroy(&sf)); 92c4762a1bSJed Brown 93c4762a1bSJed Brown ierr = PetscFinalize(); 94c4762a1bSJed Brown return ierr; 95c4762a1bSJed Brown } 96c4762a1bSJed Brown 97c4762a1bSJed Brown /*TEST 98c4762a1bSJed Brown 99c4762a1bSJed Brown test: 100c4762a1bSJed Brown suffix: basic 101c4762a1bSJed Brown nsize: 2 102c4762a1bSJed Brown filter: grep -v "type" | grep -v "sort" 103c4762a1bSJed Brown args: -sf_type basic 104c4762a1bSJed Brown 105c4762a1bSJed Brown test: 106c4762a1bSJed Brown suffix: window 107c4762a1bSJed Brown nsize: 2 108c4762a1bSJed Brown filter: grep -v "type" | grep -v "sort" 109c4762a1bSJed Brown output_file: output/ex2_basic.out 110c4762a1bSJed Brown args: -sf_type window -sf_window_sync {{fence active lock}} -sf_window_flavor {{create dynamic allocate}} 111dfd57a17SPierre Jolivet requires: defined(PETSC_HAVE_MPI_ONE_SIDED) defined(PETSC_HAVE_MPI_FEATURE_DYNAMIC_WINDOW) 112c4762a1bSJed Brown 113c4762a1bSJed Brown # The nightly test suite with MPICH uses ch3:sock, which is broken when winsize == 0 in some of the processes 114c4762a1bSJed Brown test: 115c4762a1bSJed Brown suffix: window_shared 116c4762a1bSJed Brown nsize: 2 117c4762a1bSJed Brown filter: grep -v "type" | grep -v "sort" 118c4762a1bSJed Brown output_file: output/ex2_basic.out 119c4762a1bSJed Brown args: -sf_type window -sf_window_sync {{fence active lock}} -sf_window_flavor shared 120dfd57a17SPierre Jolivet requires: defined(PETSC_HAVE_MPI_PROCESS_SHARED_MEMORY) !defined(PETSC_HAVE_MPICH_NUMVERSION) defined(PETSC_HAVE_MPI_ONE_SIDED) 121c4762a1bSJed Brown 122c4762a1bSJed Brown TEST*/ 123