| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "eremin_v_hypercube/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <cmath> | ||
| 6 | #include <tuple> | ||
| 7 | |||
| 8 | #include "eremin_v_hypercube/common/include/common.hpp" | ||
| 9 | |||
| 10 | namespace eremin_v_hypercube { | ||
| 11 | |||
| 12 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | EreminVHypercubeMPI::EreminVHypercubeMPI(const InType &in) { |
| 13 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 14 | GetInput() = in; | ||
| 15 | 12 | GetOutput() = 0; | |
| 16 | 12 | } | |
| 17 | |||
| 18 | 12 | bool EreminVHypercubeMPI::ValidationImpl() { | |
| 19 | auto &input = GetInput(); | ||
| 20 |
3/6✗ Branch 0 not taken.
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 12 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 12 times.
|
12 | return (std::get<0>(input) < std::get<1>(input)) && (std::get<2>(input) > 0) && (std::get<2>(input) <= 100000000) && |
| 21 |
3/6✗ Branch 0 not taken.
✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 12 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 12 times.
|
12 | (std::get<0>(input) >= -1e9) && (std::get<0>(input) <= 1e9) && (std::get<1>(input) >= -1e9) && |
| 22 |
2/4✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 12 times.
|
24 | (std::get<1>(input) <= 1e9) && (GetOutput() == 0); |
| 23 | } | ||
| 24 | |||
| 25 | 12 | bool EreminVHypercubeMPI::PreProcessingImpl() { | |
| 26 | 12 | return true; | |
| 27 | } | ||
| 28 | |||
| 29 | 12 | bool EreminVHypercubeMPI::RunImpl() { | |
| 30 | 12 | int world_rank = 0; | |
| 31 | 12 | int world_size = 0; | |
| 32 | 12 | MPI_Comm_rank(MPI_COMM_WORLD, &world_rank); | |
| 33 | 12 | MPI_Comm_size(MPI_COMM_WORLD, &world_size); | |
| 34 | |||
| 35 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | int ndims = static_cast<int>(std::floor(std::log2(world_size))); |
| 36 | 12 | int hypercube_size = (1 << ndims); | |
| 37 | |||
| 38 | 12 | double lower_bound = 0.0; | |
| 39 | 12 | double upper_bound = 0.0; | |
| 40 | 12 | double final_result = 0.0; | |
| 41 | 12 | int steps = 0; | |
| 42 | |||
| 43 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | if (world_rank == 0) { |
| 44 | auto &input = GetInput(); | ||
| 45 | 6 | lower_bound = std::get<0>(input); | |
| 46 | 6 | upper_bound = std::get<1>(input); | |
| 47 | 6 | steps = std::get<2>(input); | |
| 48 | } | ||
| 49 | |||
| 50 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | if (world_rank < hypercube_size) { |
| 51 | int cube_rank = world_rank; | ||
| 52 | |||
| 53 | 12 | BroadcastBoundsOverHypercube(cube_rank, ndims, lower_bound, upper_bound, steps); | |
| 54 | |||
| 55 | 12 | const auto in_function = std::get<3>(GetInput()); | |
| 56 | 12 | double step_size = (upper_bound - lower_bound) / static_cast<double>(steps); | |
| 57 | double local_result = 0.0; | ||
| 58 | |||
| 59 |
2/2✓ Branch 0 taken 311000 times.
✓ Branch 1 taken 12 times.
|
311012 | for (int i = cube_rank; i < steps; i += hypercube_size) { |
| 60 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 311000 times.
|
622000 | local_result += in_function(lower_bound + ((static_cast<double>(i) + 0.5) * step_size)); |
| 61 | } | ||
| 62 | 12 | local_result *= step_size; | |
| 63 | |||
| 64 | 12 | double current_sum = local_result; | |
| 65 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | for (int dim = 0; dim < ndims; ++dim) { |
| 66 | 12 | int neighbor = cube_rank ^ (1 << dim); | |
| 67 | 12 | double received_sum = 0.0; | |
| 68 | MPI_Status status; | ||
| 69 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Sendrecv(¤t_sum, 1, MPI_DOUBLE, neighbor, 10, &received_sum, 1, MPI_DOUBLE, neighbor, 10, |
| 70 | MPI_COMM_WORLD, &status); | ||
| 71 | 12 | current_sum += received_sum; | |
| 72 | } | ||
| 73 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | final_result = current_sum; |
| 74 | } | ||
| 75 | |||
| 76 | 12 | MPI_Bcast(&final_result, 1, MPI_DOUBLE, 0, MPI_COMM_WORLD); | |
| 77 | |||
| 78 | 12 | GetOutput() = final_result; | |
| 79 | 12 | return true; | |
| 80 | } | ||
| 81 | |||
| 82 | 12 | bool EreminVHypercubeMPI::PostProcessingImpl() { | |
| 83 | 12 | return true; | |
| 84 | } | ||
| 85 | |||
| 86 | 12 | void EreminVHypercubeMPI::BroadcastBoundsOverHypercube(int cube_rank, int ndims, double &lower_bound, | |
| 87 | double &upper_bound, int &steps) { | ||
| 88 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | for (int dim = ndims - 1; dim >= 0; --dim) { |
| 89 | 12 | int neighbor = cube_rank ^ (1 << dim); | |
| 90 | 12 | const bool is_active = (cube_rank & ((1 << dim) - 1)) == 0; | |
| 91 | 12 | const bool is_sender = (cube_rank & (1 << dim)) == 0; | |
| 92 | |||
| 93 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 12 times.
|
12 | if (!is_active) { |
| 94 | ✗ | continue; | |
| 95 | } | ||
| 96 | |||
| 97 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | if (is_sender) { |
| 98 | 6 | MPI_Send(&lower_bound, 1, MPI_DOUBLE, neighbor, 0, MPI_COMM_WORLD); | |
| 99 | 6 | MPI_Send(&upper_bound, 1, MPI_DOUBLE, neighbor, 1, MPI_COMM_WORLD); | |
| 100 | 6 | MPI_Send(&steps, 1, MPI_INT, neighbor, 2, MPI_COMM_WORLD); | |
| 101 | } else { | ||
| 102 | 6 | MPI_Recv(&lower_bound, 1, MPI_DOUBLE, neighbor, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | |
| 103 | 6 | MPI_Recv(&upper_bound, 1, MPI_DOUBLE, neighbor, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | |
| 104 | 6 | MPI_Recv(&steps, 1, MPI_INT, neighbor, 2, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | |
| 105 | } | ||
| 106 | } | ||
| 107 | 12 | } | |
| 108 | |||
| 109 | } // namespace eremin_v_hypercube | ||
| 110 |