The GAMS taxonomy for classifying mathematical software already exists. GAMS is a single-faceted classification scheme which differentiates by problem area only. Additional facets, such as algorithm/method and programming model/machine architecture, may be necessary for classifying high performance math software.
Linear algebra:
Gallivan, Heath, Ng, Ortega, Peyton, Plemmons, Romine, Sameh, and Voigt, Parallel Algorithms for Matrix Computations, SIAM, 1990.
Demmel, Heath, and Van der Vorst, Parallel Numerical Linear Algebra, Acta Numerica 1993 (also available at http://www.netlib.org/lapack/lawns/lawn60.ps)
Optimization:
Optimization Software Guide by Jorge J. More' and Stephen J. Wright
Jorge J. More' and Brett M. Averick, Evaluation of Large-Scale Optimization Problems on Vector and Parallel Architectures, SIAM Journal on Optimization 4:4 (nov 1994), pp. 708-721.
PDEs:
Johnson, Numerical solution of partial differential equations by the finite element method, Cambridge University Press, 1987
Proc. 7th SIAM Conf. on Parallel Processing for Scientific Computing, Chapter 6: Partial Differential Equations
FFTs:
Swarztrauber, Multiprocessor FFTs, Parallel Computing 5 (1987), pp. 197-210
Briggs, Hart, Sweet, and O'Gallagher, Multiprocessor FFT Methods, SIAM J. Sci. Statist. Comput., 8, pp. 27-42.
Math libraries typically allow for small-scale horizontal reuse, in that single routines may be incorporated into a user's program as needed.