James S. Coan
-
Scalar fused multiply-add instructions produce floating-point matrix arithmetic provably accurate to the penultimate digit
ACM Transactions on Mathematical Software (TOMS)