Седова Ольга. Задача 2. Вариант 16. Ленточная вертикальная схема умножения матрицы на вектор

tasks/mpi/sedova_o_vertical_ribbon_scheme/func_tests/main.cpp

@@ -0,0 +1,192 @@
+#include <gtest/gtest.h>
+
+#include <boost/mpi/communicator.hpp>
+#include <boost/mpi/environment.hpp>
+#include <random>
+#include <vector>
+
+#include "mpi/sedova_o_vertical_ribbon_scheme/include/ops_mpi.hpp"
+
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Test_3) {
+  int rows_ = 5;
+  int cols_ = 3;
+  int count_proc = 5;
+  std::vector<int> proc_(count_proc, 0);
+  std::vector<int> off(count_proc, 0);
+  if (count_proc > rows_) {
+    for (int i = 0; i < rows_; ++i) {
+      off[i] = i * cols_;
+      proc_[i] = cols_;
+    }
+    for (int i = rows_; i < count_proc; ++i) {
+      off[i] = -1;
+      proc_[i] = 0;
+    }
+  } else {
+    int count_proc_ = rows_ / count_proc;
+    int surplus = rows_ % count_proc;
+    int offset = 0;
+    for (int i = 0; i < count_proc; ++i) {
+      if (surplus > 0) {
+        proc_[i] = (count_proc_ + 1) * cols_;
+        --surplus;
+      } else {
+        proc_[i] = count_proc_ * cols_;
+      }
+      off[i] = offset;
+      offset += proc_[i];
+    }
+  }
+  std::vector<int> expected_proc = {3, 3, 3, 3, 3};
+  std::vector<int> expected_off = {0, 3, 6, 9, 12};
+  EXPECT_EQ(proc_, expected_proc);
+  EXPECT_EQ(off, expected_off);
+}
+
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Test_4) {
+  int rows_ = 5;
+  int cols_ = 3;
+  int count_proc = 3;
+  std::vector<int> proc_(count_proc, 0);
+  std::vector<int> off(count_proc, 0);
+  if (count_proc > rows_) {
+    for (int i = 0; i < rows_; ++i) {
+      off[i] = i * cols_;
+      proc_[i] = cols_;
+    }
+    for (int i = rows_; i < count_proc; ++i) {
+      off[i] = -1;
+      proc_[i] = 0;
+    }
+  } else {
+    int count_proc_ = rows_ / count_proc;
+    int surplus = rows_ % count_proc;
+    int offset = 0;
+    for (int i = 0; i < count_proc; ++i) {
+      if (surplus > 0) {
+        proc_[i] = (count_proc_ + 1) * cols_;
+        --surplus;
+      } else {
+        proc_[i] = count_proc_ * cols_;
+      }
+      off[i] = offset;
+      offset += proc_[i];
+    }
+  }
+  std::vector<int> expected_proc = {6, 6, 3};
+  std::vector<int> expected_off = {0, 6, 12};
+  EXPECT_EQ(proc_, expected_proc);
+  EXPECT_EQ(off, expected_off);
+}
+
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Test_5) {
+  int rows_ = 5;
+  int cols_ = 4;
+  int count_proc = 6;
+  std::vector<int> proc_(count_proc, 0);
+  std::vector<int> off(count_proc, 0);
+  if (count_proc > rows_) {
+    for (int i = 0; i < rows_; ++i) {
+      off[i] = i * cols_;
+      proc_[i] = cols_;
+    }
+    for (int i = rows_; i < count_proc; ++i) {
+      off[i] = -1;
+      proc_[i] = 0;
+    }
+  } else {
+    int count_proc_ = rows_ / count_proc;
+    int surplus = rows_ % count_proc;
+    int offset = 0;
+    for (int i = 0; i < count_proc; ++i) {
+      if (surplus > 0) {
+        proc_[i] = (count_proc_ + 1) * cols_;
+        --surplus;
+      } else {
+        proc_[i] = count_proc_ * cols_;
+      }
+      off[i] = offset;
+      offset += proc_[i];
+    }
+  }
+  std::vector<int> expected_proc = {4, 4, 4, 4, 4, 0};
+  std::vector<int> expected_off = {0, 4, 8, 12, 16, -1};
+  EXPECT_EQ(proc_, expected_proc);
+  EXPECT_EQ(off, expected_off);
+}
+
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Test_6) {
+  int rows_ = 10;
+  int cols_ = 4;
+  int count_proc = 8;
+  std::vector<int> proc_(count_proc, 0);
+  std::vector<int> off(count_proc, 0);
+  if (count_proc > rows_) {
+    for (int i = 0; i < rows_; ++i) {
+      off[i] = i * cols_;
+      proc_[i] = cols_;
+    }
+    for (int i = rows_; i < count_proc; ++i) {
+      off[i] = -1;
+      proc_[i] = 0;
+    }
+  } else {
+    int count_proc_ = rows_ / count_proc;
+    int surplus = rows_ % count_proc;
+    int offset = 0;
+    for (int i = 0; i < count_proc; ++i) {
+      if (surplus > 0) {
+        proc_[i] = (count_proc_ + 1) * cols_;
+        --surplus;
+      } else {
+        proc_[i] = count_proc_ * cols_;
+      }
+      off[i] = offset;
+      offset += proc_[i];
+    }
+  }
+  std::vector<int> expected_proc = {8, 8, 4, 4, 4, 4, 4, 4};
+  std::vector<int> expected_off = {0, 8, 16, 20, 24, 28, 32, 36};
+  EXPECT_EQ(proc_, expected_proc);
+  EXPECT_EQ(off, expected_off);
+}
+
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Test_7) {
+  std::vector<int> matrix = {1, 2, 3};
+  std::vector<int> vector = {7, 8};
+  std::vector<int> result(3, 0);
+
+  std::shared_ptr<ppc::core::TaskData> taskDataSeq = std::make_shared<ppc::core::TaskData>();
+  taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(matrix.data()));
+  taskDataSeq->inputs_count.emplace_back(matrix.size());
+  taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(vector.data()));
+  taskDataSeq->inputs_count.emplace_back(vector.size());
+  taskDataSeq->outputs.emplace_back(reinterpret_cast<uint8_t*>(result.data()));
+  taskDataSeq->outputs_count.emplace_back(result.size());
+
+  sedova_o_vertical_ribbon_scheme_mpi::SequentialMPI TestSequential(taskDataSeq);
+  ASSERT_EQ(TestSequential.validation(), false);
+}
+
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Test_8) {
+  std::vector<int> matrix = {1, 2, 3, 4, 5, 6};
+  std::vector<int> vector = {7, 8};
+  std::vector<int> result(3, 0);
+
+  std::shared_ptr<ppc::core::TaskData> taskDataSeq = std::make_shared<ppc::core::TaskData>();
+  taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(matrix.data()));
+  taskDataSeq->inputs_count.emplace_back(matrix.size());
+  taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(vector.data()));
+  taskDataSeq->inputs_count.emplace_back(vector.size());
+  taskDataSeq->outputs.emplace_back(reinterpret_cast<uint8_t*>(result.data()));
+  taskDataSeq->outputs_count.emplace_back(result.size());
+
+  sedova_o_vertical_ribbon_scheme_mpi::SequentialMPI TestSequential(taskDataSeq);
+  ASSERT_EQ(TestSequential.validation(), true);
+  TestSequential.pre_processing();
+  TestSequential.run();
+  TestSequential.post_processing();
+
+  std::vector<int> expected_result = {39, 54, 69};
+  ASSERT_EQ(result, expected_result);
+}

tasks/mpi/sedova_o_vertical_ribbon_scheme/include/ops_mpi.hpp

@@ -0,0 +1,53 @@
+// Copyright 2024 Sedova Olga
+#pragma once
+
+#include <gtest/gtest.h>
+
+#include <boost/mpi/collectives.hpp>
+#include <boost/mpi/communicator.hpp>
+#include <memory>
+#include <numeric>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "core/task/include/task.hpp"
+
+namespace sedova_o_vertical_ribbon_scheme_mpi {
+
+class ParallelMPI : public ppc::core::Task {
+ public:
+  explicit ParallelMPI(std::shared_ptr<ppc::core::TaskData> taskData_) : Task(std::move(taskData_)) {}
+  bool pre_processing() override;
+  bool validation() override;
+  bool run() override;
+  bool post_processing() override;
+
+ private:
+  int rows_{};
+  int cols_{};
+  std::vector<int> input_matrix_1;
+  std::vector<int> input_vector_1;
+  std::vector<int> result_vector_;
+  std::vector<int> proc;
+  std::vector<int> off;
+  boost::mpi::communicator world;
+};
+
+class SequentialMPI : public ppc::core::Task {
+ public:
+  explicit SequentialMPI(std::shared_ptr<ppc::core::TaskData> taskData_) : Task(std::move(taskData_)) {}
+  bool pre_processing() override;
+  bool validation() override;
+  bool run() override;
+  bool post_processing() override;
+
+ private:
+  int* matrix_;
+  int* vector_;
+  std::vector<int> result_vector_;
+  int count;
+  int rows_;
+  int cols_;
+};
+}  // namespace sedova_o_vertical_ribbon_scheme_mpi

tasks/mpi/sedova_o_vertical_ribbon_scheme/perf_tests/main.cpp

@@ -0,0 +1,134 @@
+#include <gtest/gtest.h>
+
+#include <boost/mpi/timer.hpp>
+#include <random>
+
+#include "core/perf/include/perf.hpp"
+#include "mpi/sedova_o_vertical_ribbon_scheme/include/ops_mpi.hpp"
+
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Performance_Pipeline_Run) {
+  boost::mpi::environment env;
+  boost::mpi::communicator world;
+  std::vector<int> global_matrix;
+  std::vector<int> global_vector;
+  std::vector<int> global_result;
+  std::shared_ptr<ppc::core::TaskData> taskDataPar = std::make_shared<ppc::core::TaskData>();
+  int rows_;
+  int cols_;
+  if (world.rank() == 0) {
+    rows_ = 2024;
+    cols_ = 2024;
+    global_vector.resize(cols_);
+    global_matrix.resize(rows_ * cols_);
+    for (int j = 0; j < rows_; ++j) {
+      for (int i = 0; i < cols_; ++i) {
+        global_matrix[j * cols_ + i] = (rand() % 101) - 50;
+      }
+    }
+    for (int i = 0; i < rows_; ++i) {
+      global_vector[i] = (rand() % 100) - 50;
+    }
+    global_result.resize(cols_, 0);
+    taskDataPar->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_matrix.data()));
+    taskDataPar->inputs_count.emplace_back(global_matrix.size());
+    taskDataPar->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_vector.data()));
+    taskDataPar->inputs_count.emplace_back(global_vector.size());
+    taskDataPar->outputs.emplace_back(reinterpret_cast<uint8_t*>(global_result.data()));
+    taskDataPar->outputs_count.emplace_back(global_result.size());
+  }
+  auto taskParallel = std::make_shared<sedova_o_vertical_ribbon_scheme_mpi::ParallelMPI>(taskDataPar);
+  ASSERT_TRUE(taskParallel->validation());
+  taskParallel->pre_processing();
+  taskParallel->run();
+  taskParallel->post_processing();
+  auto perfAttr = std::make_shared<ppc::core::PerfAttr>();
+  perfAttr->num_running = 10;
+  const boost::mpi::timer current_timer;
+  perfAttr->current_timer = [&] { return current_timer.elapsed(); };
+  auto perfResults = std::make_shared<ppc::core::PerfResults>();
+  auto perfAnalyzer = std::make_shared<ppc::core::Perf>(taskParallel);
+  perfAnalyzer->pipeline_run(perfAttr, perfResults);
+  if (world.rank() == 0) {
+    ppc::core::Perf::print_perf_statistic(perfResults);
+    std::vector<int> seq_result(global_result.size(), 0);
+    auto taskDataSeq = std::make_shared<ppc::core::TaskData>();
+    taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_matrix.data()));
+    taskDataSeq->inputs_count.emplace_back(global_matrix.size());
+    taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_vector.data()));
+    taskDataSeq->inputs_count.emplace_back(global_vector.size());
+    taskDataSeq->outputs.emplace_back(reinterpret_cast<uint8_t*>(seq_result.data()));
+    taskDataSeq->outputs_count.emplace_back(seq_result.size());
+    auto taskSequential = std::make_shared<sedova_o_vertical_ribbon_scheme_mpi::SequentialMPI>(taskDataSeq);
+    ASSERT_TRUE(taskSequential->validation());
+    taskSequential->pre_processing();
+    taskSequential->run();
+    taskSequential->post_processing();
+    ASSERT_EQ(global_result.size(), seq_result.size());
+    for (size_t i = 0; i < global_result.size(); ++i) {
+      ASSERT_EQ(global_result[i], seq_result[i]);
+    }
+  }
+}
+TEST(sedova_o_vertical_ribbon_scheme_mpi, Performance_Task_Run) {
+  boost::mpi::environment env;
+  boost::mpi::communicator world;
+  std::vector<int> global_matrix;
+  std::vector<int> global_vector;
+  std::vector<int> global_result;
+  std::shared_ptr<ppc::core::TaskData> taskDataPar = std::make_shared<ppc::core::TaskData>();
+  int rows_;
+  int cols_;
+  if (world.rank() == 0) {
+    rows_ = 2000;
+    cols_ = 2000;
+    global_matrix.resize(rows_ * cols_);
+    global_vector.resize(cols_);
+    for (int j = 0; j < rows_; ++j) {
+      for (int i = 0; i < cols_; ++i) {
+        global_matrix[j * cols_ + i] = (rand() % 101) - 50;
+      }
+    }
+    for (int i = 0; i < rows_; ++i) {
+      global_vector[i] = (rand() % 100) - 50;
+    }
+    global_result.resize(cols_, 0);
+    taskDataPar->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_matrix.data()));
+    taskDataPar->inputs_count.emplace_back(global_matrix.size());
+    taskDataPar->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_vector.data()));
+    taskDataPar->inputs_count.emplace_back(global_vector.size());
+    taskDataPar->outputs.emplace_back(reinterpret_cast<uint8_t*>(global_result.data()));
+    taskDataPar->outputs_count.emplace_back(global_result.size());
+  }
+  auto taskParallel = std::make_shared<sedova_o_vertical_ribbon_scheme_mpi::ParallelMPI>(taskDataPar);
+  ASSERT_TRUE(taskParallel->validation());
+  taskParallel->pre_processing();
+  taskParallel->run();
+  taskParallel->post_processing();
+  auto perfAttr = std::make_shared<ppc::core::PerfAttr>();
+  perfAttr->num_running = 10;
+  const boost::mpi::timer current_timer;
+  perfAttr->current_timer = [&] { return current_timer.elapsed(); };
+  auto perfResults = std::make_shared<ppc::core::PerfResults>();
+  auto perfAnalyzer = std::make_shared<ppc::core::Perf>(taskParallel);
+  perfAnalyzer->task_run(perfAttr, perfResults);
+  if (world.rank() == 0) {
+    ppc::core::Perf::print_perf_statistic(perfResults);
+    std::vector<int> seq_result(global_result.size(), 0);
+    auto taskDataSeq = std::make_shared<ppc::core::TaskData>();
+    taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_matrix.data()));
+    taskDataSeq->inputs_count.emplace_back(global_matrix.size());
+    taskDataSeq->inputs.emplace_back(reinterpret_cast<uint8_t*>(global_vector.data()));
+    taskDataSeq->inputs_count.emplace_back(global_vector.size());
+    taskDataSeq->outputs.emplace_back(reinterpret_cast<uint8_t*>(seq_result.data()));
+    taskDataSeq->outputs_count.emplace_back(seq_result.size());
+    auto taskSequential = std::make_shared<sedova_o_vertical_ribbon_scheme_mpi::SequentialMPI>(taskDataSeq);
+    ASSERT_TRUE(taskSequential->validation());
+    taskSequential->pre_processing();
+    taskSequential->run();
+    taskSequential->post_processing();
+    ASSERT_EQ(global_result.size(), seq_result.size());
+    for (size_t i = 0; i < global_result.size(); ++i) {
+      ASSERT_EQ(global_result[i], seq_result[i]);
+    }
+  }
+}