Optimized DGEMM: Vectorized Micro-Kernels and Memory-Aware Parallelization

High-performance Double-precision General Matrix Multiplication (DGEMM) in C for N×N doubles, using:

• AVX2 + FMA vectorization
• A custom 6×8 register-blocked micro-kernel
• Thread-level parallelism with static partitioning using POSIX threads
• Block sizes tuned for cache reuse

Build

The program is compiled using gcc compiler on Linux. Make sure to have gcc and make installed and then run:

make

Run

./bin/dgemm

Output example:

Time: 0.850 s

The program initializes A and B with simple deterministic patterns, computes C = A × B, and prints wall-clock time.