tbb-matrix-multiplication

Parallel matrix multiplication using Intel TBB library. This approached achieved the fastest running time in the Parallel programming class in 2020.

Solution

• Intel C++ compiler v18
• Second matrix was transposed
  • Enables vectorization
  • Better cache efficiency
• AVX 256 instructions
  • Doubles the speed of vectorized functions compared to the VS compiler
  • Uses 256bit registers
• std::inner_product
  • Standard library function that is already optimized and vectorizes pretty well
• Tbb tasks achieved the best time
  • Tree like hierarchy was created with tasks

Hardware Specification

• Windows 10
• Intel i7 9750h processor (6 cores, 12 logical) @2.56GHz (4.5GHz boost)
• Nvidia GTX1660TI
• 16GB ddr4 ram (dual channel)
• 512 GB M2-SSD

Results

Ilustration 1 - Achieved results.

Ilustration 2 - Execution time.

Ilustration 3 - Speedup compared to serial baseline.