Something I find myself wanting a lot is a few core factorization utilities, particularly the QR, LU, Cholesky/LDLt, and EVD factorizations. These factorizations are extremely useful for computing properties about matrices or avoiding the computation of inverses while maintaining numerical stability. Fast, size-specific, implementations of these core factorization routines would be extremely useful for a number of tasks.
Something I find myself wanting a lot is a few core factorization utilities, particularly the QR, LU, Cholesky/LDLt, and EVD factorizations. These factorizations are extremely useful for computing properties about matrices or avoiding the computation of inverses while maintaining numerical stability. Fast, size-specific, implementations of these core factorization routines would be extremely useful for a number of tasks.