Module kernels 

SIMD microkernels for the inner loop of matrix multiplication.

These kernels compute small tiles of C += A × B using AVX2 or AVX-512 intrinsics. They’re called by the blocked GEMM implementations after packing the input matrices into cache-friendly layouts.

Available kernels:

• kernel_4x4: 4×4 tile, AVX2 (4 registers)
• kernel_12x4: 12×4 tile, AVX2 (12 registers, better throughput)
• kernel_8x8: 8×8 tile, AVX-512 (8 registers, 64 outputs per iteration)

Modules

kernel_4x4

4×4 AVX2 microkernel for matrix multiplication.

kernel_8x8

8×8 AVX-512 microkernel for matrix multiplication.

kernel_12x4

12×4 AVX2 microkernel for matrix multiplication.