#!/usr/bin/env python3 """ MPI-OpenMP矩阵乘法实验数据分析脚本 用于读取实验数据并生成性能分析图表 """ import pandas as pd import matplotlib.pyplot as plt import numpy as np import seaborn as sns # 设置中文字体和样式 plt.rcParams['font.sans-serif'] = ['SimHei', 'DejaVu Sans'] plt.rcParams['axes.unicode_minus'] = False sns.set_style("whitegrid") def load_data(filename='experiment_results.csv'): """加载实验数据""" df = pd.read_csv(filename) return df def load_serial_data(filename='serial_results.csv'): """加载串行基准数据""" df = pd.read_csv(filename) return df def plot_experiment1(df): """绘制实验一：MPI进程数扩展性""" exp1 = df[df['Experiment'] == 'Exp1'].copy() if exp1.empty: print("警告：没有找到实验一的数据") return fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 6)) # 绘制加速比 for size in exp1['M'].unique(): data = exp1[exp1['M'] == size].sort_values('MPI_Processes') ax1.plot(data['MPI_Processes'], data['Speedup'], marker='o', label=f'{size}×{size}', linewidth=2) ax1.set_xlabel('MPI进程数', fontsize=12) ax1.set_ylabel('加速比', fontsize=12) ax1.set_title('实验一：MPI进程数扩展性（OpenMP=1）', fontsize=14) ax1.legend(fontsize=10) ax1.grid(True, alpha=0.3) ax1.plot([1, 16], [1, 16], 'k--', alpha=0.3, label='理想线性加速比') # 绘制效率 for size in exp1['M'].unique(): data = exp1[exp1['M'] == size].sort_values('MPI_Processes') ax2.plot(data['MPI_Processes'], data['Efficiency'] * 100, marker='s', label=f'{size}×{size}', linewidth=2) ax2.set_xlabel('MPI进程数', fontsize=12) ax2.set_ylabel('效率 (%)', fontsize=12) ax2.set_title('实验一：并行效率', fontsize=14) ax2.legend(fontsize=10) ax2.grid(True, alpha=0.3) ax2.axhline(y=100, color='k', linestyle='--', alpha=0.3, label='理想效率100%') plt.tight_layout() plt.savefig('exp1_mpi_scaling.png', dpi=300, bbox_inches='tight') print("已保存: exp1_mpi_scaling.png") plt.close() def plot_experiment2(df): """绘制实验二：MPI-OpenMP混合并行扩展性""" exp2 = df[df['Experiment'] == 'Exp2'].copy() if exp2.empty: print("警告：没有找到实验二的数据") return exp2['Total_Processors'] = exp2['MPI_Processes'] * exp2['OpenMP_Threads'] fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 6)) # 绘制加速比（按OpenMP线程数分组） for nthreads in exp2['OpenMP_Threads'].unique(): data = exp2[exp2['OpenMP_Threads'] == nthreads].copy() # 对相同总处理器数的数据取平均 avg_data = data.groupby('Total_Processors').agg({ 'Speedup': 'mean', 'Efficiency': 'mean' }).reset_index() ax1.plot(avg_data['Total_Processors'], avg_data['Speedup'], marker='o', label=f'OpenMP={nthreads}', linewidth=2) ax1.set_xlabel('总处理器数', fontsize=12) ax1.set_ylabel('加速比', fontsize=12) ax1.set_title('实验二：混合并行扩展性', fontsize=14) ax1.legend(fontsize=10) ax1.grid(True, alpha=0.3) # 绘制效率 for nthreads in exp2['OpenMP_Threads'].unique(): data = exp2[exp2['OpenMP_Threads'] == nthreads].copy() avg_data = data.groupby('Total_Processors').agg({ 'Speedup': 'mean', 'Efficiency': 'mean' }).reset_index() ax2.plot(avg_data['Total_Processors'], avg_data['Efficiency'] * 100, marker='s', label=f'OpenMP={nthreads}', linewidth=2) ax2.set_xlabel('总处理器数', fontsize=12) ax2.set_ylabel('效率 (%)', fontsize=12) ax2.set_title('实验二：并行效率', fontsize=14) ax2.legend(fontsize=10) ax2.grid(True, alpha=0.3) ax2.axhline(y=100, color='k', linestyle='--', alpha=0.3) plt.tight_layout() plt.savefig('exp2_hybrid_scaling.png', dpi=300, bbox_inches='tight') print("已保存: exp2_hybrid_scaling.png") plt.close() def plot_experiment3(df): """绘制实验三：MPI/OpenMP组合优化""" exp3 = df[df['Experiment'] == 'Exp3'].copy() if exp3.empty: print("警告：没有找到实验三的数据") return exp3['Total_Processors'] = exp3['MPI_Processes'] * exp3['OpenMP_Threads'] fig, ax = plt.subplots(figsize=(12, 6)) # 绘制效率热图 for size in exp3['M'].unique(): data = exp3[exp3['M'] == size] ax.plot(data['MPI_Processes'], data['Efficiency'] * 100, marker='o', label=f'{size}×{size}', linewidth=2, markersize=8) ax.set_xlabel('MPI进程数', fontsize=12) ax.set_ylabel('效率 (%)', fontsize=12) ax.set_title('实验三：不同MPI/OpenMP组合的效率（总处理器数=16）', fontsize=14) ax.legend(fontsize=10) ax.grid(True, alpha=0.3) ax.axhline(y=100, color='k', linestyle='--', alpha=0.3) # 添加x轴标签显示OpenMP线程数 ax2 = ax.twiny() ax2.set_xlim(ax.get_xlim()) ax2.set_xlabel('OpenMP线程数', fontsize=12) ax2.set_xticks([1, 2, 4, 8, 16]) ax2.set_xticklabels([16, 8, 4, 2, 1]) plt.tight_layout() plt.savefig('exp3_mpi_openmp_combo.png', dpi=300, bbox_inches='tight') print("已保存: exp3_mpi_openmp_combo.png") plt.close() def plot_efficiency_heatmap(df): """绘制效率热图""" exp2 = df[df['Experiment'] == 'Exp2'].copy() if exp2.empty: print("警告：没有找到实验二的数据") return # 选择一个中等规模的矩阵尺寸 sizes = sorted(exp2['M'].unique()) if len(sizes) > 2: target_size = sizes[len(sizes)//2] else: target_size = sizes[0] if sizes else 1024 data = exp2[exp2['M'] == target_size].copy() if data.empty: print("警告：没有足够的数据绘制热图") return # 创建数据透视表 pivot_data = data.pivot_table( values='Efficiency', index='MPI_Processes', columns='OpenMP_Threads', aggfunc='mean' ) * 100 fig, ax = plt.subplots(figsize=(10, 8)) sns.heatmap(pivot_data, annot=True, fmt='.1f', cmap='YlOrRd', cbar_kws={'label': '效率 (%)'}, ax=ax) ax.set_title(f'并行效率热图（矩阵尺寸: {target_size}×{target_size}）', fontsize=14) ax.set_xlabel('OpenMP线程数', fontsize=12) ax.set_ylabel('MPI进程数', fontsize=12) plt.tight_layout() plt.savefig('efficiency_heatmap.png', dpi=300, bbox_inches='tight') print("已保存: efficiency_heatmap.png") plt.close() def print_summary(df): """打印实验结果摘要""" print("\n" + "="*80) print("实验结果摘要") print("="*80) # 实验一摘要 exp1 = df[df['Experiment'] == 'Exp1'] if not exp1.empty: print("\n实验一：MPI进程数扩展性（OpenMP=1）") print("-" * 80) for size in sorted(exp1['M'].unique()): data = exp1[exp1['M'] == size] max_speedup = data['Speedup'].max() max_eff = data['Efficiency'].max() best_np = data.loc[data['Speedup'].idxmax(), 'MPI_Processes'] print(f"矩阵 {size}×{size}: 最大加速比={max_speedup:.2f} (NP={best_np}), " f"最高效率={max_eff*100:.1f}%") # 实验二摘要 exp2 = df[df['Experiment'] == 'Exp2'] if not exp2.empty: exp2['Total_Processors'] = exp2['MPI_Processes'] * exp2['OpenMP_Threads'] print("\n实验二：混合并行扩展性") print("-" * 80) for nthreads in sorted(exp2['OpenMP_Threads'].unique()): data = exp2[exp2['OpenMP_Threads'] == nthreads] max_speedup = data['Speedup'].max() max_eff = data['Efficiency'].max() best_total = data.loc[data['Speedup'].idxmax(), 'Total_Processors'] print(f"OpenMP={nthreads}: 最大加速比={max_speedup:.2f} " f"(总处理器={best_total}), 最高效率={max_eff*100:.1f}%") # 实验三摘要 exp3 = df[df['Experiment'] == 'Exp3'] if not exp3.empty: print("\n实验三：MPI/OpenMP组合优化（总处理器=16）") print("-" * 80) for size in sorted(exp3['M'].unique()): data = exp3[exp3['M'] == size] max_eff = data['Efficiency'].max() best_config = data.loc[data['Efficiency'].idxmax()] print(f"矩阵 {size}×{size}: 最高效率={max_eff*100:.1f}% " f"(MPI={best_config['MPI_Processes']}, " f"OpenMP={best_config['OpenMP_Threads']})") print("\n" + "="*80) def main(): """主函数""" import sys filename = sys.argv[1] if len(sys.argv) > 1 else 'experiment_results.csv' print(f"加载数据文件: {filename}") try: df = load_data(filename) print(f"数据加载成功，共 {len(df)} 条记录") except FileNotFoundError: print(f"错误：找不到文件 {filename}") print("请先运行 ./run_experiments.sh 生成实验数据") return # 打印摘要 print_summary(df) # 生成图表 print("\n生成性能分析图表...") plot_experiment1(df) plot_experiment2(df) plot_experiment3(df) plot_efficiency_heatmap(df) print("\n所有图表已生成完成！") print("\n建议：") print("1. 查看 exp1_mpi_scaling.png 了解MPI扩展性") print("2. 查看 exp2_hybrid_scaling.png 了解混合并行性能") print("3. 查看 exp3_mpi_openmp_combo.png 了解MPI/OpenMP组合优化") print("4. 查看 efficiency_heatmap.png 了解不同配置的效率分布") if __name__ == '__main__': main()