Matrix Multiplication Specialization in STAPL

The Standard Template Adaptive Parallel Library (STAPL) is a superset of C++’s Standard Template Library (STL) which allows highproductivity parallel programming in both distributed and shared memory environments. This framework provides parallel equivalents of STL containers and algorithms enabling ease of development for parallel systems. In this paper, we will discuss our methodology for implementing a fast and efficient matrix multiplication algorithm in STAPL. Our implementation employs external linear algebra libraries, specifically the Basic Linear Algebra Subprograms (BLAS) library which includes highly optimized sequential matrix operations. The paper will describe the benefits of creating a parallel matrix multiplication algorithm whose library calls are specialized based on both the matrix storage and traversal. This specialization technique ensures that the most appropriate implementation in terms of data access and structure will be used, resulting in increased efficiency compared to a non-specialized approach.