Naive matrix multiply: C = A * B. Each thread computes one element of C: C[row, col] = sum_k A[row, k] * B[k, col] # 2D indexing: derive global row/col from block and thread indices. # blockIdx.y, ...
This is a fork of llama.cpp with a custom ggml backend that offloads matrix multiplication to the AMD XDNA2 NPU found in Ryzen AI MAX processors (e.g. Ryzen AI MAX 385). The NPU backend accelerates ...