Henrywood and Agarwal, Equation (12)

Percentage Accurate: 35.3% → 69.6%

Time: 11.4s

Alternatives: 11

Speedup: 4.2×

Specification

Math

\[\begin{array}{l} \\ \left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \end{array} \]

FPCore

(FPCore (d h l M D)
 :precision binary64
 (*
  (* (pow (/ d h) (/ 1.0 2.0)) (pow (/ d l) (/ 1.0 2.0)))
  (- 1.0 (* (* (/ 1.0 2.0) (pow (/ (* M D) (* 2.0 d)) 2.0)) (/ h l)))))

C

double code(double d, double h, double l, double M, double D) {
	return (pow((d / h), (1.0 / 2.0)) * pow((d / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * pow(((M * D) / (2.0 * d)), 2.0)) * (h / l)));
}

Fortran

module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d, h, l, m, d_1)
use fmin_fmax_functions
    real(8), intent (in) :: d
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d_1
    code = (((d / h) ** (1.0d0 / 2.0d0)) * ((d / l) ** (1.0d0 / 2.0d0))) * (1.0d0 - (((1.0d0 / 2.0d0) * (((m * d_1) / (2.0d0 * d)) ** 2.0d0)) * (h / l)))
end function

Java

public static double code(double d, double h, double l, double M, double D) {
	return (Math.pow((d / h), (1.0 / 2.0)) * Math.pow((d / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * Math.pow(((M * D) / (2.0 * d)), 2.0)) * (h / l)));
}

Python

def code(d, h, l, M, D):
	return (math.pow((d / h), (1.0 / 2.0)) * math.pow((d / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * math.pow(((M * D) / (2.0 * d)), 2.0)) * (h / l)))

Julia

function code(d, h, l, M, D)
	return Float64(Float64((Float64(d / h) ^ Float64(1.0 / 2.0)) * (Float64(d / l) ^ Float64(1.0 / 2.0))) * Float64(1.0 - Float64(Float64(Float64(1.0 / 2.0) * (Float64(Float64(M * D) / Float64(2.0 * d)) ^ 2.0)) * Float64(h / l))))
end

MATLAB

function tmp = code(d, h, l, M, D)
	tmp = (((d / h) ^ (1.0 / 2.0)) * ((d / l) ^ (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * (((M * D) / (2.0 * d)) ^ 2.0)) * (h / l)));
end

Wolfram

code[d_, h_, l_, M_, D_] := N[(N[(N[Power[N[(d / h), $MachinePrecision], N[(1.0 / 2.0), $MachinePrecision]], $MachinePrecision] * N[Power[N[(d / l), $MachinePrecision], N[(1.0 / 2.0), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[(1.0 - N[(N[(N[(1.0 / 2.0), $MachinePrecision] * N[Power[N[(N[(M * D), $MachinePrecision] / N[(2.0 * d), $MachinePrecision]), $MachinePrecision], 2.0], $MachinePrecision]), $MachinePrecision] * N[(h / l), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Initial Program: 35.3% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \end{array} \]

FPCore

(FPCore (d h l M D)
 :precision binary64
 (*
  (* (pow (/ d h) (/ 1.0 2.0)) (pow (/ d l) (/ 1.0 2.0)))
  (- 1.0 (* (* (/ 1.0 2.0) (pow (/ (* M D) (* 2.0 d)) 2.0)) (/ h l)))))

C

double code(double d, double h, double l, double M, double D) {
	return (pow((d / h), (1.0 / 2.0)) * pow((d / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * pow(((M * D) / (2.0 * d)), 2.0)) * (h / l)));
}

Fortran

module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d, h, l, m, d_1)
use fmin_fmax_functions
    real(8), intent (in) :: d
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d_1
    code = (((d / h) ** (1.0d0 / 2.0d0)) * ((d / l) ** (1.0d0 / 2.0d0))) * (1.0d0 - (((1.0d0 / 2.0d0) * (((m * d_1) / (2.0d0 * d)) ** 2.0d0)) * (h / l)))
end function

Java

public static double code(double d, double h, double l, double M, double D) {
	return (Math.pow((d / h), (1.0 / 2.0)) * Math.pow((d / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * Math.pow(((M * D) / (2.0 * d)), 2.0)) * (h / l)));
}

Python

def code(d, h, l, M, D):
	return (math.pow((d / h), (1.0 / 2.0)) * math.pow((d / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * math.pow(((M * D) / (2.0 * d)), 2.0)) * (h / l)))

Julia

function code(d, h, l, M, D)
	return Float64(Float64((Float64(d / h) ^ Float64(1.0 / 2.0)) * (Float64(d / l) ^ Float64(1.0 / 2.0))) * Float64(1.0 - Float64(Float64(Float64(1.0 / 2.0) * (Float64(Float64(M * D) / Float64(2.0 * d)) ^ 2.0)) * Float64(h / l))))
end

MATLAB

function tmp = code(d, h, l, M, D)
	tmp = (((d / h) ^ (1.0 / 2.0)) * ((d / l) ^ (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * (((M * D) / (2.0 * d)) ^ 2.0)) * (h / l)));
end

Wolfram

code[d_, h_, l_, M_, D_] := N[(N[(N[Power[N[(d / h), $MachinePrecision], N[(1.0 / 2.0), $MachinePrecision]], $MachinePrecision] * N[Power[N[(d / l), $MachinePrecision], N[(1.0 / 2.0), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[(1.0 - N[(N[(N[(1.0 / 2.0), $MachinePrecision] * N[Power[N[(N[(M * D), $MachinePrecision] / N[(2.0 * d), $MachinePrecision]), $MachinePrecision], 2.0], $MachinePrecision]), $MachinePrecision] * N[(h / l), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Alternative 1: 69.6% accurate, 1.5× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} t_0 := \frac{M \cdot D}{d\_m + d\_m}\\ \mathbf{if}\;d\_m \leq 3.1 \cdot 10^{-287}:\\ \;\;\;\;-0.125 \cdot \frac{{D}^{2} \cdot \left({M}^{2} \cdot \left(h \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)\right)}{d\_m \cdot \ell}\\ \mathbf{else}:\\ \;\;\;\;\left(d\_m \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(t\_0 \cdot \left(t\_0 \cdot 0.5\right)\right) \cdot \frac{h}{\ell}\right)\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (let* ((t_0 (/ (* M D) (+ d_m d_m))))
   (if (<= d_m 3.1e-287)
     (*
      -0.125
      (/
       (* (pow D 2.0) (* (pow M 2.0) (* h (sqrt (/ 1.0 (* h l))))))
       (* d_m l)))
     (*
      (* d_m (sqrt (/ (/ 1.0 l) h)))
      (- 1.0 (* (* t_0 (* t_0 0.5)) (/ h l)))))))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (M * D) / (d_m + d_m);
	double tmp;
	if (d_m <= 3.1e-287) {
		tmp = -0.125 * ((pow(D, 2.0) * (pow(M, 2.0) * (h * sqrt((1.0 / (h * l)))))) / (d_m * l));
	} else {
		tmp = (d_m * sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: t_0
    real(8) :: tmp
    t_0 = (m * d) / (d_m + d_m)
    if (d_m <= 3.1d-287) then
        tmp = (-0.125d0) * (((d ** 2.0d0) * ((m ** 2.0d0) * (h * sqrt((1.0d0 / (h * l)))))) / (d_m * l))
    else
        tmp = (d_m * sqrt(((1.0d0 / l) / h))) * (1.0d0 - ((t_0 * (t_0 * 0.5d0)) * (h / l)))
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (M * D) / (d_m + d_m);
	double tmp;
	if (d_m <= 3.1e-287) {
		tmp = -0.125 * ((Math.pow(D, 2.0) * (Math.pow(M, 2.0) * (h * Math.sqrt((1.0 / (h * l)))))) / (d_m * l));
	} else {
		tmp = (d_m * Math.sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	t_0 = (M * D) / (d_m + d_m)
	tmp = 0
	if d_m <= 3.1e-287:
		tmp = -0.125 * ((math.pow(D, 2.0) * (math.pow(M, 2.0) * (h * math.sqrt((1.0 / (h * l)))))) / (d_m * l))
	else:
		tmp = (d_m * math.sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)))
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	t_0 = Float64(Float64(M * D) / Float64(d_m + d_m))
	tmp = 0.0
	if (d_m <= 3.1e-287)
		tmp = Float64(-0.125 * Float64(Float64((D ^ 2.0) * Float64((M ^ 2.0) * Float64(h * sqrt(Float64(1.0 / Float64(h * l)))))) / Float64(d_m * l)));
	else
		tmp = Float64(Float64(d_m * sqrt(Float64(Float64(1.0 / l) / h))) * Float64(1.0 - Float64(Float64(t_0 * Float64(t_0 * 0.5)) * Float64(h / l))));
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	t_0 = (M * D) / (d_m + d_m);
	tmp = 0.0;
	if (d_m <= 3.1e-287)
		tmp = -0.125 * (((D ^ 2.0) * ((M ^ 2.0) * (h * sqrt((1.0 / (h * l)))))) / (d_m * l));
	else
		tmp = (d_m * sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := Block[{t$95$0 = N[(N[(M * D), $MachinePrecision] / N[(d$95$m + d$95$m), $MachinePrecision]), $MachinePrecision]}, If[LessEqual[d$95$m, 3.1e-287], N[(-0.125 * N[(N[(N[Power[D, 2.0], $MachinePrecision] * N[(N[Power[M, 2.0], $MachinePrecision] * N[(h * N[Sqrt[N[(1.0 / N[(h * l), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] / N[(d$95$m * l), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], N[(N[(d$95$m * N[Sqrt[N[(N[(1.0 / l), $MachinePrecision] / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[(1.0 - N[(N[(t$95$0 * N[(t$95$0 * 0.5), $MachinePrecision]), $MachinePrecision] * N[(h / l), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]

Derivation

1. Split input into 2 regimes

2. if d < 3.1000000000000001e-287

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\frac{-1}{8} \cdot \frac{{D}^{2} \cdot \left({M}^{2} \cdot \left(h \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)\right)}{d \cdot \ell}} \]

   6. Applied rewrites 32.5%

      \[\leadsto \color{blue}{-0.125 \cdot \frac{{D}^{2} \cdot \left({M}^{2} \cdot \left(h \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)\right)}{d \cdot \ell}} \]

   if 3.1000000000000001e-287 < d

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   8. Applied rewrites 69.6%

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   10. Applied rewrites 69.6%

       \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \color{blue}{\left(\frac{M \cdot D}{d + d} \cdot \left(\frac{M \cdot D}{d + d} \cdot 0.5\right)\right)} \cdot \frac{h}{\ell}\right) \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 2: 69.6% accurate, 1.8× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} t_0 := \frac{M \cdot D}{d\_m + d\_m}\\ \mathbf{if}\;h \leq 5.6 \cdot 10^{+222}:\\ \;\;\;\;\left(d\_m \cdot \sqrt{\frac{1}{h \cdot \ell}}\right) \cdot \left(1 - \left(t\_0 \cdot \left(t\_0 \cdot 0.5\right)\right) \cdot \frac{h}{\ell}\right)\\ \mathbf{else}:\\ \;\;\;\;\frac{\left(-1 \cdot \left(d\_m \cdot \sqrt{\frac{h}{d\_m}}\right)\right) \cdot \sqrt{\frac{d\_m}{\ell}}}{h}\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (let* ((t_0 (/ (* M D) (+ d_m d_m))))
   (if (<= h 5.6e+222)
     (* (* d_m (sqrt (/ 1.0 (* h l)))) (- 1.0 (* (* t_0 (* t_0 0.5)) (/ h l))))
     (/ (* (* -1.0 (* d_m (sqrt (/ h d_m)))) (sqrt (/ d_m l))) h))))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (M * D) / (d_m + d_m);
	double tmp;
	if (h <= 5.6e+222) {
		tmp = (d_m * sqrt((1.0 / (h * l)))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
	} else {
		tmp = ((-1.0 * (d_m * sqrt((h / d_m)))) * sqrt((d_m / l))) / h;
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: t_0
    real(8) :: tmp
    t_0 = (m * d) / (d_m + d_m)
    if (h <= 5.6d+222) then
        tmp = (d_m * sqrt((1.0d0 / (h * l)))) * (1.0d0 - ((t_0 * (t_0 * 0.5d0)) * (h / l)))
    else
        tmp = (((-1.0d0) * (d_m * sqrt((h / d_m)))) * sqrt((d_m / l))) / h
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (M * D) / (d_m + d_m);
	double tmp;
	if (h <= 5.6e+222) {
		tmp = (d_m * Math.sqrt((1.0 / (h * l)))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
	} else {
		tmp = ((-1.0 * (d_m * Math.sqrt((h / d_m)))) * Math.sqrt((d_m / l))) / h;
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	t_0 = (M * D) / (d_m + d_m)
	tmp = 0
	if h <= 5.6e+222:
		tmp = (d_m * math.sqrt((1.0 / (h * l)))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)))
	else:
		tmp = ((-1.0 * (d_m * math.sqrt((h / d_m)))) * math.sqrt((d_m / l))) / h
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	t_0 = Float64(Float64(M * D) / Float64(d_m + d_m))
	tmp = 0.0
	if (h <= 5.6e+222)
		tmp = Float64(Float64(d_m * sqrt(Float64(1.0 / Float64(h * l)))) * Float64(1.0 - Float64(Float64(t_0 * Float64(t_0 * 0.5)) * Float64(h / l))));
	else
		tmp = Float64(Float64(Float64(-1.0 * Float64(d_m * sqrt(Float64(h / d_m)))) * sqrt(Float64(d_m / l))) / h);
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	t_0 = (M * D) / (d_m + d_m);
	tmp = 0.0;
	if (h <= 5.6e+222)
		tmp = (d_m * sqrt((1.0 / (h * l)))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
	else
		tmp = ((-1.0 * (d_m * sqrt((h / d_m)))) * sqrt((d_m / l))) / h;
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := Block[{t$95$0 = N[(N[(M * D), $MachinePrecision] / N[(d$95$m + d$95$m), $MachinePrecision]), $MachinePrecision]}, If[LessEqual[h, 5.6e+222], N[(N[(d$95$m * N[Sqrt[N[(1.0 / N[(h * l), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[(1.0 - N[(N[(t$95$0 * N[(t$95$0 * 0.5), $MachinePrecision]), $MachinePrecision] * N[(h / l), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], N[(N[(N[(-1.0 * N[(d$95$m * N[Sqrt[N[(h / d$95$m), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * N[Sqrt[N[(d$95$m / l), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] / h), $MachinePrecision]]]

Derivation

1. Split input into 2 regimes

2. if h < 5.6000000000000003e222

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   8. Applied rewrites 69.1%

      \[\leadsto \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right) \cdot \left(1 - \color{blue}{\left(\frac{M \cdot D}{d + d} \cdot \left(\frac{M \cdot D}{d + d} \cdot 0.5\right)\right)} \cdot \frac{h}{\ell}\right) \]

   if 5.6000000000000003e222 < h

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   2. Taylor expanded in h around 0

      \[\leadsto \color{blue}{\frac{\sqrt{d \cdot h} \cdot \sqrt{\frac{d}{\ell}}}{h}} \]

   4. Applied rewrites 17.9%

      \[\leadsto \color{blue}{\frac{\sqrt{d \cdot h} \cdot \sqrt{\frac{d}{\ell}}}{h}} \]

   5. Taylor expanded in d around -inf

      \[\leadsto \frac{\left(-1 \cdot \left(d \cdot \sqrt{\frac{h}{d}}\right)\right) \cdot \sqrt{\frac{d}{\ell}}}{h} \]

   7. Applied rewrites 5.9%

      \[\leadsto \frac{\left(-1 \cdot \left(d \cdot \sqrt{\frac{h}{d}}\right)\right) \cdot \sqrt{\frac{d}{\ell}}}{h} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 3: 68.5% accurate, 1.9× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} t_0 := \frac{M \cdot D}{d\_m + d\_m}\\ \left(d\_m \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(t\_0 \cdot \left(t\_0 \cdot 0.5\right)\right) \cdot \frac{h}{\ell}\right) \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (let* ((t_0 (/ (* M D) (+ d_m d_m))))
   (* (* d_m (sqrt (/ (/ 1.0 l) h))) (- 1.0 (* (* t_0 (* t_0 0.5)) (/ h l))))))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (M * D) / (d_m + d_m);
	return (d_m * sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: t_0
    t_0 = (m * d) / (d_m + d_m)
    code = (d_m * sqrt(((1.0d0 / l) / h))) * (1.0d0 - ((t_0 * (t_0 * 0.5d0)) * (h / l)))
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (M * D) / (d_m + d_m);
	return (d_m * Math.sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	t_0 = (M * D) / (d_m + d_m)
	return (d_m * math.sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)))

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	t_0 = Float64(Float64(M * D) / Float64(d_m + d_m))
	return Float64(Float64(d_m * sqrt(Float64(Float64(1.0 / l) / h))) * Float64(1.0 - Float64(Float64(t_0 * Float64(t_0 * 0.5)) * Float64(h / l))))
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp = code(d_m, h, l, M, D)
	t_0 = (M * D) / (d_m + d_m);
	tmp = (d_m * sqrt(((1.0 / l) / h))) * (1.0 - ((t_0 * (t_0 * 0.5)) * (h / l)));
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := Block[{t$95$0 = N[(N[(M * D), $MachinePrecision] / N[(d$95$m + d$95$m), $MachinePrecision]), $MachinePrecision]}, N[(N[(d$95$m * N[Sqrt[N[(N[(1.0 / l), $MachinePrecision] / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[(1.0 - N[(N[(t$95$0 * N[(t$95$0 * 0.5), $MachinePrecision]), $MachinePrecision] * N[(h / l), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]

Derivation

1. Initial program 35.3%

   \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

3. Applied rewrites 42.6%

   \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

4. Taylor expanded in d around 0

   \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

6. Applied rewrites 69.1%

   \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

8. Applied rewrites 69.6%

   \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

10. Applied rewrites 69.6%

    \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \color{blue}{\left(\frac{M \cdot D}{d + d} \cdot \left(\frac{M \cdot D}{d + d} \cdot 0.5\right)\right)} \cdot \frac{h}{\ell}\right) \]
11. Add Preprocessing

Alternative 4: 45.1% accurate, 2.0× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} \mathbf{if}\;M \leq 2.85 \cdot 10^{-75}:\\ \;\;\;\;\left(d\_m \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot 1\\ \mathbf{else}:\\ \;\;\;\;\frac{\left(\ell - \frac{\left(\frac{\left(M \cdot M\right) \cdot \left(D \cdot D\right)}{d\_m} \cdot 0.125\right) \cdot h}{d\_m}\right) \cdot \sqrt{\frac{d\_m \cdot d\_m}{\ell \cdot h}}}{\ell}\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (if (<= M 2.85e-75)
   (* (* d_m (sqrt (/ (/ 1.0 l) h))) 1.0)
   (/
    (*
     (- l (/ (* (* (/ (* (* M M) (* D D)) d_m) 0.125) h) d_m))
     (sqrt (/ (* d_m d_m) (* l h))))
    l)))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if (M <= 2.85e-75) {
		tmp = (d_m * sqrt(((1.0 / l) / h))) * 1.0;
	} else {
		tmp = ((l - ((((((M * M) * (D * D)) / d_m) * 0.125) * h) / d_m)) * sqrt(((d_m * d_m) / (l * h)))) / l;
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: tmp
    if (m <= 2.85d-75) then
        tmp = (d_m * sqrt(((1.0d0 / l) / h))) * 1.0d0
    else
        tmp = ((l - ((((((m * m) * (d * d)) / d_m) * 0.125d0) * h) / d_m)) * sqrt(((d_m * d_m) / (l * h)))) / l
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if (M <= 2.85e-75) {
		tmp = (d_m * Math.sqrt(((1.0 / l) / h))) * 1.0;
	} else {
		tmp = ((l - ((((((M * M) * (D * D)) / d_m) * 0.125) * h) / d_m)) * Math.sqrt(((d_m * d_m) / (l * h)))) / l;
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	tmp = 0
	if M <= 2.85e-75:
		tmp = (d_m * math.sqrt(((1.0 / l) / h))) * 1.0
	else:
		tmp = ((l - ((((((M * M) * (D * D)) / d_m) * 0.125) * h) / d_m)) * math.sqrt(((d_m * d_m) / (l * h)))) / l
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	tmp = 0.0
	if (M <= 2.85e-75)
		tmp = Float64(Float64(d_m * sqrt(Float64(Float64(1.0 / l) / h))) * 1.0);
	else
		tmp = Float64(Float64(Float64(l - Float64(Float64(Float64(Float64(Float64(Float64(M * M) * Float64(D * D)) / d_m) * 0.125) * h) / d_m)) * sqrt(Float64(Float64(d_m * d_m) / Float64(l * h)))) / l);
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	tmp = 0.0;
	if (M <= 2.85e-75)
		tmp = (d_m * sqrt(((1.0 / l) / h))) * 1.0;
	else
		tmp = ((l - ((((((M * M) * (D * D)) / d_m) * 0.125) * h) / d_m)) * sqrt(((d_m * d_m) / (l * h)))) / l;
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := If[LessEqual[M, 2.85e-75], N[(N[(d$95$m * N[Sqrt[N[(N[(1.0 / l), $MachinePrecision] / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * 1.0), $MachinePrecision], N[(N[(N[(l - N[(N[(N[(N[(N[(N[(M * M), $MachinePrecision] * N[(D * D), $MachinePrecision]), $MachinePrecision] / d$95$m), $MachinePrecision] * 0.125), $MachinePrecision] * h), $MachinePrecision] / d$95$m), $MachinePrecision]), $MachinePrecision] * N[Sqrt[N[(N[(d$95$m * d$95$m), $MachinePrecision] / N[(l * h), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] / l), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if M < 2.84999999999999983e-75

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   8. Applied rewrites 69.6%

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   9. Taylor expanded in d around inf

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   11. Applied rewrites 42.5%

       \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   if 2.84999999999999983e-75 < M

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 31.1%

      \[\leadsto \color{blue}{\frac{\left(\ell - \frac{\left(\left(M \cdot M\right) \cdot \left(D \cdot D\right)\right) \cdot 0.25}{\left(d + d\right) \cdot d} \cdot h\right) \cdot \sqrt{\frac{d \cdot d}{\ell \cdot h}}}{\ell}} \]

   5. Applied rewrites 37.4%

      \[\leadsto \frac{\left(\ell - \color{blue}{\frac{\left(\frac{\left(M \cdot M\right) \cdot \left(D \cdot D\right)}{d} \cdot 0.125\right) \cdot h}{d}}\right) \cdot \sqrt{\frac{d \cdot d}{\ell \cdot h}}}{\ell} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 5: 44.7% accurate, 2.0× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} \mathbf{if}\;M \leq 2.9 \cdot 10^{-75}:\\ \;\;\;\;\left(d\_m \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot 1\\ \mathbf{else}:\\ \;\;\;\;\frac{\left(\ell - \left(\frac{\left(M \cdot M\right) \cdot \left(D \cdot D\right)}{d\_m} \cdot 0.125\right) \cdot \frac{h}{d\_m}\right) \cdot \sqrt{\frac{d\_m \cdot d\_m}{\ell \cdot h}}}{\ell}\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (if (<= M 2.9e-75)
   (* (* d_m (sqrt (/ (/ 1.0 l) h))) 1.0)
   (/
    (*
     (- l (* (* (/ (* (* M M) (* D D)) d_m) 0.125) (/ h d_m)))
     (sqrt (/ (* d_m d_m) (* l h))))
    l)))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if (M <= 2.9e-75) {
		tmp = (d_m * sqrt(((1.0 / l) / h))) * 1.0;
	} else {
		tmp = ((l - (((((M * M) * (D * D)) / d_m) * 0.125) * (h / d_m))) * sqrt(((d_m * d_m) / (l * h)))) / l;
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: tmp
    if (m <= 2.9d-75) then
        tmp = (d_m * sqrt(((1.0d0 / l) / h))) * 1.0d0
    else
        tmp = ((l - (((((m * m) * (d * d)) / d_m) * 0.125d0) * (h / d_m))) * sqrt(((d_m * d_m) / (l * h)))) / l
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if (M <= 2.9e-75) {
		tmp = (d_m * Math.sqrt(((1.0 / l) / h))) * 1.0;
	} else {
		tmp = ((l - (((((M * M) * (D * D)) / d_m) * 0.125) * (h / d_m))) * Math.sqrt(((d_m * d_m) / (l * h)))) / l;
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	tmp = 0
	if M <= 2.9e-75:
		tmp = (d_m * math.sqrt(((1.0 / l) / h))) * 1.0
	else:
		tmp = ((l - (((((M * M) * (D * D)) / d_m) * 0.125) * (h / d_m))) * math.sqrt(((d_m * d_m) / (l * h)))) / l
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	tmp = 0.0
	if (M <= 2.9e-75)
		tmp = Float64(Float64(d_m * sqrt(Float64(Float64(1.0 / l) / h))) * 1.0);
	else
		tmp = Float64(Float64(Float64(l - Float64(Float64(Float64(Float64(Float64(M * M) * Float64(D * D)) / d_m) * 0.125) * Float64(h / d_m))) * sqrt(Float64(Float64(d_m * d_m) / Float64(l * h)))) / l);
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	tmp = 0.0;
	if (M <= 2.9e-75)
		tmp = (d_m * sqrt(((1.0 / l) / h))) * 1.0;
	else
		tmp = ((l - (((((M * M) * (D * D)) / d_m) * 0.125) * (h / d_m))) * sqrt(((d_m * d_m) / (l * h)))) / l;
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := If[LessEqual[M, 2.9e-75], N[(N[(d$95$m * N[Sqrt[N[(N[(1.0 / l), $MachinePrecision] / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * 1.0), $MachinePrecision], N[(N[(N[(l - N[(N[(N[(N[(N[(M * M), $MachinePrecision] * N[(D * D), $MachinePrecision]), $MachinePrecision] / d$95$m), $MachinePrecision] * 0.125), $MachinePrecision] * N[(h / d$95$m), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * N[Sqrt[N[(N[(d$95$m * d$95$m), $MachinePrecision] / N[(l * h), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] / l), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if M < 2.9000000000000002e-75

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   8. Applied rewrites 69.6%

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   9. Taylor expanded in d around inf

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   11. Applied rewrites 42.5%

       \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   if 2.9000000000000002e-75 < M

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 31.1%

      \[\leadsto \color{blue}{\frac{\left(\ell - \frac{\left(\left(M \cdot M\right) \cdot \left(D \cdot D\right)\right) \cdot 0.25}{\left(d + d\right) \cdot d} \cdot h\right) \cdot \sqrt{\frac{d \cdot d}{\ell \cdot h}}}{\ell}} \]

   5. Applied rewrites 34.6%

      \[\leadsto \frac{\left(\ell - \color{blue}{\left(\frac{\left(M \cdot M\right) \cdot \left(D \cdot D\right)}{d} \cdot 0.125\right) \cdot \frac{h}{d}}\right) \cdot \sqrt{\frac{d \cdot d}{\ell \cdot h}}}{\ell} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 6: 43.9% accurate, 3.3× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} \mathbf{if}\;M \cdot D \leq 2.15 \cdot 10^{+173}:\\ \;\;\;\;\left(d\_m \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot 1\\ \mathbf{else}:\\ \;\;\;\;\frac{-1 \cdot \left(d\_m \cdot \left(\ell \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)\right)}{\ell}\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (if (<= (* M D) 2.15e+173)
   (* (* d_m (sqrt (/ (/ 1.0 l) h))) 1.0)
   (/ (* -1.0 (* d_m (* l (sqrt (/ 1.0 (* h l)))))) l)))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = (d_m * sqrt(((1.0 / l) / h))) * 1.0;
	} else {
		tmp = (-1.0 * (d_m * (l * sqrt((1.0 / (h * l)))))) / l;
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: tmp
    if ((m * d) <= 2.15d+173) then
        tmp = (d_m * sqrt(((1.0d0 / l) / h))) * 1.0d0
    else
        tmp = ((-1.0d0) * (d_m * (l * sqrt((1.0d0 / (h * l)))))) / l
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = (d_m * Math.sqrt(((1.0 / l) / h))) * 1.0;
	} else {
		tmp = (-1.0 * (d_m * (l * Math.sqrt((1.0 / (h * l)))))) / l;
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	tmp = 0
	if (M * D) <= 2.15e+173:
		tmp = (d_m * math.sqrt(((1.0 / l) / h))) * 1.0
	else:
		tmp = (-1.0 * (d_m * (l * math.sqrt((1.0 / (h * l)))))) / l
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	tmp = 0.0
	if (Float64(M * D) <= 2.15e+173)
		tmp = Float64(Float64(d_m * sqrt(Float64(Float64(1.0 / l) / h))) * 1.0);
	else
		tmp = Float64(Float64(-1.0 * Float64(d_m * Float64(l * sqrt(Float64(1.0 / Float64(h * l)))))) / l);
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	tmp = 0.0;
	if ((M * D) <= 2.15e+173)
		tmp = (d_m * sqrt(((1.0 / l) / h))) * 1.0;
	else
		tmp = (-1.0 * (d_m * (l * sqrt((1.0 / (h * l)))))) / l;
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := If[LessEqual[N[(M * D), $MachinePrecision], 2.15e+173], N[(N[(d$95$m * N[Sqrt[N[(N[(1.0 / l), $MachinePrecision] / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * 1.0), $MachinePrecision], N[(N[(-1.0 * N[(d$95$m * N[(l * N[Sqrt[N[(1.0 / N[(h * l), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] / l), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if (*.f64 M D) < 2.15000000000000013e173

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   8. Applied rewrites 69.6%

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   9. Taylor expanded in d around inf

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   11. Applied rewrites 42.5%

       \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   if 2.15000000000000013e173 < (*.f64 M D)

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 31.1%

      \[\leadsto \color{blue}{\frac{\left(\ell - \frac{\left(\left(M \cdot M\right) \cdot \left(D \cdot D\right)\right) \cdot 0.25}{\left(d + d\right) \cdot d} \cdot h\right) \cdot \sqrt{\frac{d \cdot d}{\ell \cdot h}}}{\ell}} \]

   4. Taylor expanded in d around -inf

      \[\leadsto \frac{\color{blue}{-1 \cdot \left(d \cdot \left(\ell \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)\right)}}{\ell} \]

   6. Applied rewrites 10.6%

      \[\leadsto \frac{\color{blue}{-1 \cdot \left(d \cdot \left(\ell \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)\right)}}{\ell} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 7: 43.7% accurate, 4.2× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} t_0 := d\_m \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\\ \mathbf{if}\;M \cdot D \leq 2.15 \cdot 10^{+173}:\\ \;\;\;\;t\_0 \cdot 1\\ \mathbf{else}:\\ \;\;\;\;-1 \cdot t\_0\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (let* ((t_0 (* d_m (sqrt (/ (/ 1.0 l) h)))))
   (if (<= (* M D) 2.15e+173) (* t_0 1.0) (* -1.0 t_0))))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double t_0 = d_m * sqrt(((1.0 / l) / h));
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = t_0 * 1.0;
	} else {
		tmp = -1.0 * t_0;
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: t_0
    real(8) :: tmp
    t_0 = d_m * sqrt(((1.0d0 / l) / h))
    if ((m * d) <= 2.15d+173) then
        tmp = t_0 * 1.0d0
    else
        tmp = (-1.0d0) * t_0
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double t_0 = d_m * Math.sqrt(((1.0 / l) / h));
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = t_0 * 1.0;
	} else {
		tmp = -1.0 * t_0;
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	t_0 = d_m * math.sqrt(((1.0 / l) / h))
	tmp = 0
	if (M * D) <= 2.15e+173:
		tmp = t_0 * 1.0
	else:
		tmp = -1.0 * t_0
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	t_0 = Float64(d_m * sqrt(Float64(Float64(1.0 / l) / h)))
	tmp = 0.0
	if (Float64(M * D) <= 2.15e+173)
		tmp = Float64(t_0 * 1.0);
	else
		tmp = Float64(-1.0 * t_0);
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	t_0 = d_m * sqrt(((1.0 / l) / h));
	tmp = 0.0;
	if ((M * D) <= 2.15e+173)
		tmp = t_0 * 1.0;
	else
		tmp = -1.0 * t_0;
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := Block[{t$95$0 = N[(d$95$m * N[Sqrt[N[(N[(1.0 / l), $MachinePrecision] / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]}, If[LessEqual[N[(M * D), $MachinePrecision], 2.15e+173], N[(t$95$0 * 1.0), $MachinePrecision], N[(-1.0 * t$95$0), $MachinePrecision]]]

Derivation

1. Split input into 2 regimes

2. if (*.f64 M D) < 2.15000000000000013e173

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   8. Applied rewrites 69.6%

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   9. Taylor expanded in d around inf

      \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   11. Applied rewrites 42.5%

       \[\leadsto \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \cdot \color{blue}{1} \]

   if 2.15000000000000013e173 < (*.f64 M D)

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around -inf

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   6. Applied rewrites 10.4%

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   8. Applied rewrites 10.4%

      \[\leadsto -1 \cdot \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 8: 43.4% accurate, 4.2× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} \mathbf{if}\;M \cdot D \leq 2.15 \cdot 10^{+173}:\\ \;\;\;\;\left(d\_m \cdot \sqrt{\frac{1}{h \cdot \ell}}\right) \cdot 1\\ \mathbf{else}:\\ \;\;\;\;-1 \cdot \left(d\_m \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right)\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (if (<= (* M D) 2.15e+173)
   (* (* d_m (sqrt (/ 1.0 (* h l)))) 1.0)
   (* -1.0 (* d_m (sqrt (/ (/ 1.0 l) h))))))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = (d_m * sqrt((1.0 / (h * l)))) * 1.0;
	} else {
		tmp = -1.0 * (d_m * sqrt(((1.0 / l) / h)));
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: tmp
    if ((m * d) <= 2.15d+173) then
        tmp = (d_m * sqrt((1.0d0 / (h * l)))) * 1.0d0
    else
        tmp = (-1.0d0) * (d_m * sqrt(((1.0d0 / l) / h)))
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = (d_m * Math.sqrt((1.0 / (h * l)))) * 1.0;
	} else {
		tmp = -1.0 * (d_m * Math.sqrt(((1.0 / l) / h)));
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	tmp = 0
	if (M * D) <= 2.15e+173:
		tmp = (d_m * math.sqrt((1.0 / (h * l)))) * 1.0
	else:
		tmp = -1.0 * (d_m * math.sqrt(((1.0 / l) / h)))
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	tmp = 0.0
	if (Float64(M * D) <= 2.15e+173)
		tmp = Float64(Float64(d_m * sqrt(Float64(1.0 / Float64(h * l)))) * 1.0);
	else
		tmp = Float64(-1.0 * Float64(d_m * sqrt(Float64(Float64(1.0 / l) / h))));
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	tmp = 0.0;
	if ((M * D) <= 2.15e+173)
		tmp = (d_m * sqrt((1.0 / (h * l)))) * 1.0;
	else
		tmp = -1.0 * (d_m * sqrt(((1.0 / l) / h)));
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := If[LessEqual[N[(M * D), $MachinePrecision], 2.15e+173], N[(N[(d$95$m * N[Sqrt[N[(1.0 / N[(h * l), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * 1.0), $MachinePrecision], N[(-1.0 * N[(d$95$m * N[Sqrt[N[(N[(1.0 / l), $MachinePrecision] / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]

Derivation

1. Split input into 2 regimes

2. if (*.f64 M D) < 2.15000000000000013e173

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   7. Taylor expanded in d around inf

      \[\leadsto \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right) \cdot \color{blue}{1} \]

   9. Applied rewrites 42.2%

      \[\leadsto \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right) \cdot \color{blue}{1} \]

   if 2.15000000000000013e173 < (*.f64 M D)

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around -inf

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   6. Applied rewrites 10.4%

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   8. Applied rewrites 10.4%

      \[\leadsto -1 \cdot \left(d \cdot \sqrt{\frac{\frac{1}{\ell}}{h}}\right) \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 9: 43.4% accurate, 4.2× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} t_0 := \sqrt{\frac{1}{h \cdot \ell}}\\ \mathbf{if}\;M \cdot D \leq 2.15 \cdot 10^{+173}:\\ \;\;\;\;\left(d\_m \cdot t\_0\right) \cdot 1\\ \mathbf{else}:\\ \;\;\;\;t\_0 \cdot \left(-d\_m\right)\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (let* ((t_0 (sqrt (/ 1.0 (* h l)))))
   (if (<= (* M D) 2.15e+173) (* (* d_m t_0) 1.0) (* t_0 (- d_m)))))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double t_0 = sqrt((1.0 / (h * l)));
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = (d_m * t_0) * 1.0;
	} else {
		tmp = t_0 * -d_m;
	}
	return tmp;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    real(8) :: t_0
    real(8) :: tmp
    t_0 = sqrt((1.0d0 / (h * l)))
    if ((m * d) <= 2.15d+173) then
        tmp = (d_m * t_0) * 1.0d0
    else
        tmp = t_0 * -d_m
    end if
    code = tmp
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double t_0 = Math.sqrt((1.0 / (h * l)));
	double tmp;
	if ((M * D) <= 2.15e+173) {
		tmp = (d_m * t_0) * 1.0;
	} else {
		tmp = t_0 * -d_m;
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	t_0 = math.sqrt((1.0 / (h * l)))
	tmp = 0
	if (M * D) <= 2.15e+173:
		tmp = (d_m * t_0) * 1.0
	else:
		tmp = t_0 * -d_m
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	t_0 = sqrt(Float64(1.0 / Float64(h * l)))
	tmp = 0.0
	if (Float64(M * D) <= 2.15e+173)
		tmp = Float64(Float64(d_m * t_0) * 1.0);
	else
		tmp = Float64(t_0 * Float64(-d_m));
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	t_0 = sqrt((1.0 / (h * l)));
	tmp = 0.0;
	if ((M * D) <= 2.15e+173)
		tmp = (d_m * t_0) * 1.0;
	else
		tmp = t_0 * -d_m;
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := Block[{t$95$0 = N[Sqrt[N[(1.0 / N[(h * l), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]}, If[LessEqual[N[(M * D), $MachinePrecision], 2.15e+173], N[(N[(d$95$m * t$95$0), $MachinePrecision] * 1.0), $MachinePrecision], N[(t$95$0 * (-d$95$m)), $MachinePrecision]]]

Derivation

1. Split input into 2 regimes

2. if (*.f64 M D) < 2.15000000000000013e173

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around 0

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   6. Applied rewrites 69.1%

      \[\leadsto \color{blue}{\left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   7. Taylor expanded in d around inf

      \[\leadsto \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right) \cdot \color{blue}{1} \]

   9. Applied rewrites 42.2%

      \[\leadsto \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right) \cdot \color{blue}{1} \]

   if 2.15000000000000013e173 < (*.f64 M D)

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around -inf

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   6. Applied rewrites 10.4%

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   8. Applied rewrites 10.4%

      \[\leadsto \sqrt{\frac{1}{h \cdot \ell}} \cdot \color{blue}{\left(-d\right)} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 10: 26.1% accurate, 0.5× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \begin{array}{l} t_0 := \left({\left(\frac{d\_m}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d\_m}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d\_m}\right)}^{2}\right) \cdot \frac{h}{\ell}\right)\\ t_1 := \sqrt{\frac{1}{h \cdot \ell}} \cdot \left(-d\_m\right)\\ \mathbf{if}\;t\_0 \leq -2 \cdot 10^{-117}:\\ \;\;\;\;t\_1\\ \mathbf{elif}\;t\_0 \leq \infty:\\ \;\;\;\;\frac{d\_m \cdot \sqrt{\frac{\ell}{h}}}{\ell}\\ \mathbf{else}:\\ \;\;\;\;t\_1\\ \end{array} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D)
 :precision binary64
 (let* ((t_0
         (*
          (* (pow (/ d_m h) (/ 1.0 2.0)) (pow (/ d_m l) (/ 1.0 2.0)))
          (-
           1.0
           (* (* (/ 1.0 2.0) (pow (/ (* M D) (* 2.0 d_m)) 2.0)) (/ h l)))))
        (t_1 (* (sqrt (/ 1.0 (* h l))) (- d_m))))
   (if (<= t_0 -2e-117)
     t_1
     (if (<= t_0 INFINITY) (/ (* d_m (sqrt (/ l h))) l) t_1))))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (pow((d_m / h), (1.0 / 2.0)) * pow((d_m / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * pow(((M * D) / (2.0 * d_m)), 2.0)) * (h / l)));
	double t_1 = sqrt((1.0 / (h * l))) * -d_m;
	double tmp;
	if (t_0 <= -2e-117) {
		tmp = t_1;
	} else if (t_0 <= ((double) INFINITY)) {
		tmp = (d_m * sqrt((l / h))) / l;
	} else {
		tmp = t_1;
	}
	return tmp;
}

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	double t_0 = (Math.pow((d_m / h), (1.0 / 2.0)) * Math.pow((d_m / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * Math.pow(((M * D) / (2.0 * d_m)), 2.0)) * (h / l)));
	double t_1 = Math.sqrt((1.0 / (h * l))) * -d_m;
	double tmp;
	if (t_0 <= -2e-117) {
		tmp = t_1;
	} else if (t_0 <= Double.POSITIVE_INFINITY) {
		tmp = (d_m * Math.sqrt((l / h))) / l;
	} else {
		tmp = t_1;
	}
	return tmp;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	t_0 = (math.pow((d_m / h), (1.0 / 2.0)) * math.pow((d_m / l), (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * math.pow(((M * D) / (2.0 * d_m)), 2.0)) * (h / l)))
	t_1 = math.sqrt((1.0 / (h * l))) * -d_m
	tmp = 0
	if t_0 <= -2e-117:
		tmp = t_1
	elif t_0 <= math.inf:
		tmp = (d_m * math.sqrt((l / h))) / l
	else:
		tmp = t_1
	return tmp

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	t_0 = Float64(Float64((Float64(d_m / h) ^ Float64(1.0 / 2.0)) * (Float64(d_m / l) ^ Float64(1.0 / 2.0))) * Float64(1.0 - Float64(Float64(Float64(1.0 / 2.0) * (Float64(Float64(M * D) / Float64(2.0 * d_m)) ^ 2.0)) * Float64(h / l))))
	t_1 = Float64(sqrt(Float64(1.0 / Float64(h * l))) * Float64(-d_m))
	tmp = 0.0
	if (t_0 <= -2e-117)
		tmp = t_1;
	elseif (t_0 <= Inf)
		tmp = Float64(Float64(d_m * sqrt(Float64(l / h))) / l);
	else
		tmp = t_1;
	end
	return tmp
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp_2 = code(d_m, h, l, M, D)
	t_0 = (((d_m / h) ^ (1.0 / 2.0)) * ((d_m / l) ^ (1.0 / 2.0))) * (1.0 - (((1.0 / 2.0) * (((M * D) / (2.0 * d_m)) ^ 2.0)) * (h / l)));
	t_1 = sqrt((1.0 / (h * l))) * -d_m;
	tmp = 0.0;
	if (t_0 <= -2e-117)
		tmp = t_1;
	elseif (t_0 <= Inf)
		tmp = (d_m * sqrt((l / h))) / l;
	else
		tmp = t_1;
	end
	tmp_2 = tmp;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := Block[{t$95$0 = N[(N[(N[Power[N[(d$95$m / h), $MachinePrecision], N[(1.0 / 2.0), $MachinePrecision]], $MachinePrecision] * N[Power[N[(d$95$m / l), $MachinePrecision], N[(1.0 / 2.0), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[(1.0 - N[(N[(N[(1.0 / 2.0), $MachinePrecision] * N[Power[N[(N[(M * D), $MachinePrecision] / N[(2.0 * d$95$m), $MachinePrecision]), $MachinePrecision], 2.0], $MachinePrecision]), $MachinePrecision] * N[(h / l), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]}, Block[{t$95$1 = N[(N[Sqrt[N[(1.0 / N[(h * l), $MachinePrecision]), $MachinePrecision]], $MachinePrecision] * (-d$95$m)), $MachinePrecision]}, If[LessEqual[t$95$0, -2e-117], t$95$1, If[LessEqual[t$95$0, Infinity], N[(N[(d$95$m * N[Sqrt[N[(l / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] / l), $MachinePrecision], t$95$1]]]]

Derivation

1. Split input into 2 regimes

2. if (*.f64 (*.f64 (pow.f64 (/.f64 d h) (/.f64 #s(literal 1 binary64) #s(literal 2 binary64))) (pow.f64 (/.f64 d l) (/.f64 #s(literal 1 binary64) #s(literal 2 binary64)))) (-.f64 #s(literal 1 binary64) (*.f64 (*.f64 (/.f64 #s(literal 1 binary64) #s(literal 2 binary64)) (pow.f64 (/.f64 (*.f64 M D) (*.f64 #s(literal 2 binary64) d)) #s(literal 2 binary64))) (/.f64 h l)))) < -2.00000000000000006e-117 or +inf.0 < (*.f64 (*.f64 (pow.f64 (/.f64 d h) (/.f64 #s(literal 1 binary64) #s(literal 2 binary64))) (pow.f64 (/.f64 d l) (/.f64 #s(literal 1 binary64) #s(literal 2 binary64)))) (-.f64 #s(literal 1 binary64) (*.f64 (*.f64 (/.f64 #s(literal 1 binary64) #s(literal 2 binary64)) (pow.f64 (/.f64 (*.f64 M D) (*.f64 #s(literal 2 binary64) d)) #s(literal 2 binary64))) (/.f64 h l))))

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 42.6%

      \[\leadsto \color{blue}{\sqrt{\frac{d \cdot d}{\ell \cdot h}}} \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   4. Taylor expanded in d around -inf

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   6. Applied rewrites 10.4%

      \[\leadsto \color{blue}{-1 \cdot \left(d \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)} \]

   8. Applied rewrites 10.4%

      \[\leadsto \sqrt{\frac{1}{h \cdot \ell}} \cdot \color{blue}{\left(-d\right)} \]

   if -2.00000000000000006e-117 < (*.f64 (*.f64 (pow.f64 (/.f64 d h) (/.f64 #s(literal 1 binary64) #s(literal 2 binary64))) (pow.f64 (/.f64 d l) (/.f64 #s(literal 1 binary64) #s(literal 2 binary64)))) (-.f64 #s(literal 1 binary64) (*.f64 (*.f64 (/.f64 #s(literal 1 binary64) #s(literal 2 binary64)) (pow.f64 (/.f64 (*.f64 M D) (*.f64 #s(literal 2 binary64) d)) #s(literal 2 binary64))) (/.f64 h l)))) < +inf.0

   1. Initial program 35.3%

      \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

   3. Applied rewrites 31.1%

      \[\leadsto \color{blue}{\frac{\left(\ell - \frac{\left(\left(M \cdot M\right) \cdot \left(D \cdot D\right)\right) \cdot 0.25}{\left(d + d\right) \cdot d} \cdot h\right) \cdot \sqrt{\frac{d \cdot d}{\ell \cdot h}}}{\ell}} \]

   4. Taylor expanded in d around inf

      \[\leadsto \frac{\color{blue}{d \cdot \left(\ell \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)}}{\ell} \]

   6. Applied rewrites 36.9%

      \[\leadsto \frac{\color{blue}{d \cdot \left(\ell \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)}}{\ell} \]

   7. Taylor expanded in l around 0

      \[\leadsto \frac{d \cdot \sqrt{\frac{\ell}{h}}}{\ell} \]

   9. Applied rewrites 21.2%

      \[\leadsto \frac{d \cdot \sqrt{\frac{\ell}{h}}}{\ell} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 11: 21.2% accurate, 7.4× speedup

Math

\[\begin{array}{l} d_m = \left|d\right| \\ [d_m, h, l, M, D] = \mathsf{sort}([d_m, h, l, M, D])\\ \\ \frac{d\_m \cdot \sqrt{\frac{\ell}{h}}}{\ell} \end{array} \]

FPCore

d_m = (fabs.f64 d)
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
(FPCore (d_m h l M D) :precision binary64 (/ (* d_m (sqrt (/ l h))) l))

C

d_m = fabs(d);
assert(d_m < h && h < l && l < M && M < D);
double code(double d_m, double h, double l, double M, double D) {
	return (d_m * sqrt((l / h))) / l;
}

Fortran

d_m =     private
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(d_m, h, l, m, d)
use fmin_fmax_functions
    real(8), intent (in) :: d_m
    real(8), intent (in) :: h
    real(8), intent (in) :: l
    real(8), intent (in) :: m
    real(8), intent (in) :: d
    code = (d_m * sqrt((l / h))) / l
end function

Java

d_m = Math.abs(d);
assert d_m < h && h < l && l < M && M < D;
public static double code(double d_m, double h, double l, double M, double D) {
	return (d_m * Math.sqrt((l / h))) / l;
}

Python

d_m = math.fabs(d)
[d_m, h, l, M, D] = sort([d_m, h, l, M, D])
def code(d_m, h, l, M, D):
	return (d_m * math.sqrt((l / h))) / l

Julia

d_m = abs(d)
d_m, h, l, M, D = sort([d_m, h, l, M, D])
function code(d_m, h, l, M, D)
	return Float64(Float64(d_m * sqrt(Float64(l / h))) / l)
end

MATLAB

d_m = abs(d);
d_m, h, l, M, D = num2cell(sort([d_m, h, l, M, D])){:}
function tmp = code(d_m, h, l, M, D)
	tmp = (d_m * sqrt((l / h))) / l;
end

Wolfram

d_m = N[Abs[d], $MachinePrecision]
NOTE: d_m, h, l, M, and D should be sorted in increasing order before calling this function.
code[d$95$m_, h_, l_, M_, D_] := N[(N[(d$95$m * N[Sqrt[N[(l / h), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] / l), $MachinePrecision]

Derivation

1. Initial program 35.3%

   \[\left({\left(\frac{d}{h}\right)}^{\left(\frac{1}{2}\right)} \cdot {\left(\frac{d}{\ell}\right)}^{\left(\frac{1}{2}\right)}\right) \cdot \left(1 - \left(\frac{1}{2} \cdot {\left(\frac{M \cdot D}{2 \cdot d}\right)}^{2}\right) \cdot \frac{h}{\ell}\right) \]

3. Applied rewrites 31.1%

   \[\leadsto \color{blue}{\frac{\left(\ell - \frac{\left(\left(M \cdot M\right) \cdot \left(D \cdot D\right)\right) \cdot 0.25}{\left(d + d\right) \cdot d} \cdot h\right) \cdot \sqrt{\frac{d \cdot d}{\ell \cdot h}}}{\ell}} \]

4. Taylor expanded in d around inf

   \[\leadsto \frac{\color{blue}{d \cdot \left(\ell \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)}}{\ell} \]

6. Applied rewrites 36.9%

   \[\leadsto \frac{\color{blue}{d \cdot \left(\ell \cdot \sqrt{\frac{1}{h \cdot \ell}}\right)}}{\ell} \]

7. Taylor expanded in l around 0

   \[\leadsto \frac{d \cdot \sqrt{\frac{\ell}{h}}}{\ell} \]

9. Applied rewrites 21.2%

   \[\leadsto \frac{d \cdot \sqrt{\frac{\ell}{h}}}{\ell} \]
10. Add Preprocessing

Reproduce

herbie shell --seed 2025161 
(FPCore (d h l M D)
  :name "Henrywood and Agarwal, Equation (12)"
  :precision binary64
  (* (* (pow (/ d h) (/ 1.0 2.0)) (pow (/ d l) (/ 1.0 2.0))) (- 1.0 (* (* (/ 1.0 2.0) (pow (/ (* M D) (* 2.0 d)) 2.0)) (/ h l)))))