General matrix multiplication

Introducation

This repository contains a simple implementation of matrix multiplication using OpenMP and the NEON instruction set. The goal is to demonstrate the use of parallel processing and optimized instructions for matrix operations.

Requirements

1. A compatible ARM-based processor with NEON support
2. OpenMP installed and configured on your system
3. A C compiler (e.g., GCC/CLANG)

Compiliation

To compile the code, use the following command:

 gcc -o gemm gemm.c -O3 -ffast-math -fopenmp -march=native

Matrix Multiplication Benchmark on M2 Pro Processor

This benchmark compares the performance of four different matrix multiplication implementations on an M2 Pro processor. Implementations:

1. Optimized Neon Parallel BLOCKED
2. Standard Neon Parallel BLOCKED
3. Normal Parallel NEON
4. Normal Parallel matmul

Results:

1. N = 1024, BLOCK_SIZE = 16

• Optimized: 87.17 GFLOP/S
• Standard: 69.49 GFLOP/S
• Normal NEON: 76.44 GFLOP/S
• Normal matmul: 4.85 GFLOP/S s

2. N = 8192, BLOCK_SIZE = 16

• Optimized: 122.27 GFLOP/S ms
• Standard: 72.04 GFLOP/S ms
• Normal NEON: 60.23 GFLOP/S ms
• Normal matmul: Not applicable

Note

Our old best performance was ~76 GFLOP/S, and we have now achieved a significant improvement of 122.27 GFLOP/S with our optimized implementation.