Actual source code: solve_performance.c

  1: const char help[] = "Profile the performance of MATLMVM MatSolve() in a loop";

  3: #include <petscksp.h>
  4: #include <petscmath.h>

  6: int main(int argc, char **argv)
  7: {
  8:   PetscInt      n        = 1000;
  9:   PetscInt      n_epochs = 10;
 10:   PetscInt      n_iters  = 10;
 11:   Vec           x, g, dx, df, p;
 12:   PetscRandom   rand;
 13:   PetscLogStage matsolve_loop, main_stage;
 14:   Mat           B;

 16:   PetscCall(PetscInitialize(&argc, &argv, NULL, help));
 17:   PetscOptionsBegin(PETSC_COMM_WORLD, NULL, help, "KSP");
 18:   PetscCall(PetscOptionsInt("-n", "Vector size", __FILE__, n, &n, NULL));
 19:   PetscCall(PetscOptionsInt("-epochs", "Number of epochs", __FILE__, n_epochs, &n_epochs, NULL));
 20:   PetscCall(PetscOptionsInt("-iters", "Number of iterations per epoch", __FILE__, n_iters, &n_iters, NULL));
 21:   PetscOptionsEnd();
 22:   PetscCall(VecCreateMPI(PETSC_COMM_WORLD, PETSC_DETERMINE, n, &x));
 23:   PetscCall(VecSetFromOptions(x));
 24:   PetscCall(VecDuplicate(x, &g));
 25:   PetscCall(VecDuplicate(x, &dx));
 26:   PetscCall(VecDuplicate(x, &df));
 27:   PetscCall(VecDuplicate(x, &p));
 28:   PetscCall(MatCreateLMVMBFGS(PETSC_COMM_WORLD, PETSC_DETERMINE, n, &B));
 29:   PetscCall(MatSetFromOptions(B));
 30:   PetscCall(MatLMVMAllocate(B, x, g));
 31:   PetscCall(PetscRandomCreate(PETSC_COMM_WORLD, &rand));
 32:   PetscCall(PetscRandomSetInterval(rand, -1.0, 1.0));
 33:   PetscCall(PetscRandomSetFromOptions(rand));
 34:   PetscCall(PetscLogStageRegister("LMVM MatSolve Loop", &matsolve_loop));
 35:   PetscCall(PetscLogStageGetId("Main Stage", &main_stage));
 36:   PetscCall(PetscLogStageSetVisible(main_stage, PETSC_FALSE));
 37:   for (PetscInt epoch = 0; epoch < n_epochs + 1; epoch++) {
 38:     PetscScalar dot;
 39:     PetscReal   xscale, fscale, absdot;
 40:     PetscInt    history_size;

 42:     PetscCall(VecSetRandom(dx, rand));
 43:     PetscCall(VecSetRandom(df, rand));
 44:     PetscCall(VecDot(dx, df, &dot));
 45:     absdot = PetscAbsScalar(dot);
 46:     PetscCall(VecSetRandom(x, rand));
 47:     PetscCall(VecSetRandom(g, rand));
 48:     xscale = 1.0;
 49:     fscale = absdot / PetscRealPart(dot);
 50:     PetscCall(MatLMVMGetHistorySize(B, &history_size));

 52:     PetscCall(MatLMVMUpdate(B, x, g));
 53:     for (PetscInt iter = 0; iter < history_size; iter++, xscale *= -1.0, fscale *= -1.0) {
 54:       PetscCall(VecAXPY(x, xscale, dx));
 55:       PetscCall(VecAXPY(g, fscale, df));
 56:       PetscCall(MatLMVMUpdate(B, x, g));
 57:       PetscCall(MatSolve(B, g, p));
 58:     }
 59:     if (epoch > 0) PetscCall(PetscLogStagePush(matsolve_loop));
 60:     for (PetscInt iter = 0; iter < n_iters; iter++, xscale *= -1.0, fscale *= -1.0) {
 61:       PetscCall(VecAXPY(x, xscale, dx));
 62:       PetscCall(VecAXPY(g, fscale, df));
 63:       PetscCall(MatLMVMUpdate(B, x, g));
 64:       PetscCall(MatSolve(B, g, p));
 65:     }
 66:     PetscCall(MatLMVMReset(B, PETSC_FALSE));
 67:     if (epoch > 0) PetscCall(PetscLogStagePop());
 68:   }
 69:   PetscCall(MatView(B, PETSC_VIEWER_STDOUT_(PETSC_COMM_WORLD)));
 70:   PetscCall(PetscRandomDestroy(&rand));
 71:   PetscCall(MatDestroy(&B));
 72:   PetscCall(VecDestroy(&p));
 73:   PetscCall(VecDestroy(&df));
 74:   PetscCall(VecDestroy(&dx));
 75:   PetscCall(VecDestroy(&g));
 76:   PetscCall(VecDestroy(&x));
 77:   PetscCall(PetscFinalize());
 78:   return 0;
 79: }

 81: /*TEST

 83:   test:
 84:     suffix: 0
 85:     args: -mat_lmvm_scale_type none

 87: TEST*/