涉及到的优化技术

• 内存布局优化
• 内存访问优化
• 循环展开
• SIMD
• 矩阵分块
• OpenMP
• CUDA
• CUDA+共享内存

运行结果比对

• 测试条件：
  • 两个1024x1024尺寸的float型矩阵相乘
  • 处理器：i5-10210U
  • RAM：8.00GB
  • 显卡：NVIDIA GeForce MX330
• 测试结果：
  • multi1 execution time: 6772ms
  • multi2 execution time: 708ms
  • multi3 execution time: 460ms
  • multi4 execution time: 136ms
  • multi5 execution time: 320ms
  • multi6 execution time: 136ms
  • multi7 execution time: 104ms
  • multi8 execution time: 103ms
  • multi9 execution time: 57ms
  • multi10 execution time: 54ms
  • multi11 execution time: 18ms
  • multi12 execution time: 306ms ---> use openBlas
  • multi13 execution time: 75ms
  • multi14 execution time: 75ms
  • multi15 execution time: 76ms
  • multi16 execution time: 71ms
  • multi17 execution time: 114ms
  • multi18 execution time: 91ms

依赖库

• spdlog
• cuda
• catch2
• openBlas