Module bgemm_faer 

Batched GEMM backend using the [faer] library. faer-backed batched GEMM kernel on strided views.

Uses faer::linalg::matmul::matmul for SIMD-optimized matrix multiplication. When dimension groups cannot be fused into 2D matrices (non-contiguous strides), copies operands to contiguous column-major buffers before calling faer.

Functions

bgemm_contiguous_into

Batched GEMM on pre-contiguous operands.

bgemm_strided_into

Batched strided GEMM using faer: C = alpha * A * B + beta * C