Floating-Point Multiply-Add with Approximate Normalization for Low-Cost Matrix Engines | IEEE Conference Publication | IEEE Xplore