Efficiently, same size square two matrix multiplication using Assembly and SSE2 Instructions.
NOTE: This function is able to multiply 4 and multiples of square matrix size like 4x4-8x8-12x12-16x16-20x20-...
Matrix is an array for numbers, characters, or other types of data, arranged in rows and columns. A matrix can be modified by operations with scalars, vectors, or matrices. Each matrix has row and column size. A matrix with m rows and n columns is called an m x n matrix. If matrix has same row and column size, that matrix is called square matrix. Single-instruction, multiple data (SIMD) instructions are used for computing vector operations.
In this work, I had built a function that multiplies two square matrices whose has integer numbers efficiently. I used SSE2 (Streaming SIMD Extensions 2) instructions for parallel operations as it is intended it to be faster than C compiled only SISD architecture-based code. Besides that, I used the SISD instructions (pure assembly) to flow control like loops in function. At the final, the function I built, and C compiled code was compared on CPU cycles.
10000 iterations with random integer numbers were tested on each size of square matrix. Average results of C compiled code and SIMD and Pure Assembly code had compared each other. The minimum size of matrix was 4x4 and maximum was 224x224 on the testing. Function had problems after more than 224x244 size in executing. 100% stacked results chart are given on below.
As can be seen, it has been observed that it is at least 40 percent more efficient in tests. Maximum efficiency is around 60 percent.
This project is licensed under the GPL License - see the LICENSE.md file for details