New preprint Error analysis of matrix multiplication with narrow range floating-point arithmetic (with Theo Mary).