Common LAPACK/Blas utility functions

General Usage
1

LAPACK and Blas have all these handy little routines like

dlascl - scale an array between two normalized quantities handling overflow/underflow/etc.

# _dlascl(from, to, x)
# This will compute x .*= x (to/from) and handle special cases 
julia> _dlascl(1e-300, 1e300, [1e-300, 1e-301, 1e-299]) 
3-element Vector{Float64}:
 1.0000000000000002e300
 1.0000000000000002e299
 1.0000000000000002e301

dlapy2 - computer sqrt(x^2+y^2) handling underflow/overflow

julia> _dlapy2(1e300,1e200)
1.0e300

julia> sqrt(1e300^2+1e200^2)
Inf

Am I missing some package/scenario in Julia where these exist? (I realize I can call BLAS/LAPACK directly, but I want to know if there’s another place I should be looking for these little micro-utilities.)

Each routine isn’t very complicated, but they are often somewhat subtle, e.g.

@inline function _dlapy2(x::T, y::T) where T 
  !isnan(x) || return x
  !isnan(y) || return y
  w = max(abs(x),abs(y))
  z = min(abs(x),abs(y))
  if z == 0
    return w
  else
    return w*sqrt(1+(z/w)^2)
  end
end

Apologies if I haven’t found just the right google search or my list of what I would search for wasn’t quite expansive enough.

PS - Edited for code blocks

2

This is the hypot function.

dlascl would be easy enough to implement, but there doesn’t seem to have been any demand for it.

PS. Technically, these are LAPACK auxiliary routines. The generally useful ones (e.g. dlamch, dsecnd, dlisnan, dlapy2, dlapy3, dlarnv, etc.) seem to mostly already have Julia equivalents.

3 Likes