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; 24ffc4695bSBarry Smith ierr = MPI_Comm_rank(PETSC_COMM_WORLD,&rank);CHKERRMPI(ierr); 25ffc4695bSBarry Smith ierr = MPI_Comm_size(PETSC_COMM_WORLD,&size);CHKERRMPI(ierr); 26c4762a1bSJed Brown 27c4762a1bSJed Brown if (size != 2) SETERRQ(PETSC_COMM_WORLD, PETSC_ERR_USER, "Only coded for two MPI processes\n"); 28c4762a1bSJed Brown 29c4762a1bSJed Brown ierr = PetscSFCreate(PETSC_COMM_WORLD,&sf);CHKERRQ(ierr); 30c4762a1bSJed Brown ierr = PetscSFSetFromOptions(sf);CHKERRQ(ierr); 31c4762a1bSJed Brown 32c4762a1bSJed Brown nleaves = 2; 33c4762a1bSJed Brown nroots = 1; 34c4762a1bSJed Brown ierr = PetscMalloc1(nleaves,&ilocal);CHKERRQ(ierr); 35c4762a1bSJed Brown 36c4762a1bSJed Brown for (i = 0; i<nleaves; i++) { 37c4762a1bSJed Brown ilocal[i] = i; 38c4762a1bSJed Brown } 39c4762a1bSJed Brown 40c4762a1bSJed Brown ierr = PetscMalloc1(nleaves,&iremote);CHKERRQ(ierr); 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 } 52c4762a1bSJed Brown ierr = PetscSFSetGraph(sf,nroots,nleaves,ilocal,PETSC_OWN_POINTER,iremote,PETSC_OWN_POINTER);CHKERRQ(ierr); 53c4762a1bSJed Brown ierr = PetscSFSetUp(sf);CHKERRQ(ierr); 54c4762a1bSJed Brown ierr = PetscSFView(sf,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr); 55c4762a1bSJed Brown ierr = VecCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr); 56c4762a1bSJed Brown ierr = VecSetSizes(A,2,PETSC_DETERMINE);CHKERRQ(ierr); 57c4762a1bSJed Brown ierr = VecSetFromOptions(A);CHKERRQ(ierr); 58c4762a1bSJed Brown ierr = VecSetUp(A);CHKERRQ(ierr); 59c4762a1bSJed Brown 60c4762a1bSJed Brown ierr = VecDuplicate(A,&B);CHKERRQ(ierr); 61c4762a1bSJed Brown ierr = VecDuplicate(A,&Aout);CHKERRQ(ierr); 62c4762a1bSJed Brown ierr = VecDuplicate(A,&Bout);CHKERRQ(ierr); 63c4762a1bSJed Brown ierr = VecGetArray(A,&bufA);CHKERRQ(ierr); 64c4762a1bSJed Brown ierr = VecGetArray(B,&bufB);CHKERRQ(ierr); 65c4762a1bSJed Brown for (i=0; i<2; i++) { 66c4762a1bSJed Brown bufA[i] = (PetscScalar)rank; 67c4762a1bSJed Brown bufB[i] = (PetscScalar)(rank) + 10.0; 68c4762a1bSJed Brown } 69c4762a1bSJed Brown ierr = VecRestoreArray(A,&bufA);CHKERRQ(ierr); 70c4762a1bSJed Brown ierr = VecRestoreArray(B,&bufB);CHKERRQ(ierr); 71c4762a1bSJed Brown 72c4762a1bSJed Brown ierr = VecGetArrayRead(A,(const PetscScalar**)&bufA);CHKERRQ(ierr); 73c4762a1bSJed Brown ierr = VecGetArrayRead(B,(const PetscScalar**)&bufB);CHKERRQ(ierr); 74c4762a1bSJed Brown ierr = VecGetArray(Aout,&bufAout);CHKERRQ(ierr); 75c4762a1bSJed Brown ierr = VecGetArray(Bout,&bufBout);CHKERRQ(ierr); 76ad227feaSJunchao Zhang ierr = PetscSFBcastBegin(sf,MPIU_SCALAR,(const void*)bufA,(void *)bufAout,MPI_REPLACE);CHKERRQ(ierr); 77ad227feaSJunchao Zhang ierr = PetscSFBcastBegin(sf,MPIU_SCALAR,(const void*)bufB,(void *)bufBout,MPI_REPLACE);CHKERRQ(ierr); 78ad227feaSJunchao Zhang ierr = PetscSFBcastEnd(sf,MPIU_SCALAR,(const void*)bufA,(void *)bufAout,MPI_REPLACE);CHKERRQ(ierr); 79ad227feaSJunchao Zhang ierr = PetscSFBcastEnd(sf,MPIU_SCALAR,(const void*)bufB,(void *)bufBout,MPI_REPLACE);CHKERRQ(ierr); 80c4762a1bSJed Brown ierr = VecRestoreArrayRead(A,(const PetscScalar**)&bufA);CHKERRQ(ierr); 81c4762a1bSJed Brown ierr = VecRestoreArrayRead(B,(const PetscScalar**)&bufB);CHKERRQ(ierr); 82c4762a1bSJed Brown ierr = VecRestoreArray(Aout,&bufAout);CHKERRQ(ierr); 83c4762a1bSJed Brown ierr = VecRestoreArray(Bout,&bufBout);CHKERRQ(ierr); 84c4762a1bSJed Brown 85c4762a1bSJed Brown ierr = VecView(Aout,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr); 86c4762a1bSJed Brown ierr = VecView(Bout,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr); 87c4762a1bSJed Brown ierr = VecDestroy(&A);CHKERRQ(ierr); 88c4762a1bSJed Brown ierr = VecDestroy(&B);CHKERRQ(ierr); 89c4762a1bSJed Brown ierr = VecDestroy(&Aout);CHKERRQ(ierr); 90c4762a1bSJed Brown ierr = VecDestroy(&Bout);CHKERRQ(ierr); 91c4762a1bSJed Brown ierr = PetscSFDestroy(&sf);CHKERRQ(ierr); 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}} 111*dfd57a17SPierre 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 120*dfd57a17SPierre 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