Module gemm 

Non-quantized Backend::gemm comparison against CPU F32. Metal uses F32 buffers (GEMV for m=1, tiled GEMM otherwise); CUDA uses F16 input/output buffers and cuBLAS F32 accumulation. This fixture does not exercise quantized Marlin or the production plan runtime.

Structs

GemmOp

C[m, n] = A[m, k] · B[n, k]^T (row-major, B already transposed to head-major). Matches the Backend::gemm signature used by Linear.