Diagrams.TwoD.Apollonian:descartes from diagrams-contrib-1.3.0.5

Percentage Accurate: 70.2% → 99.2%

Time: 7.5s

Alternatives: 11

Speedup: 0.3×

Specification

Math

\[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (* 2.0 (sqrt (+ (+ (* x y) (* x z)) (* y z)))))

C

double code(double x, double y, double z) {
	return 2.0 * sqrt((((x * y) + (x * z)) + (y * z)));
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    code = 2.0d0 * sqrt((((x * y) + (x * z)) + (y * z)))
end function

Java

public static double code(double x, double y, double z) {
	return 2.0 * Math.sqrt((((x * y) + (x * z)) + (y * z)));
}

Python

def code(x, y, z):
	return 2.0 * math.sqrt((((x * y) + (x * z)) + (y * z)))

Julia

function code(x, y, z)
	return Float64(2.0 * sqrt(Float64(Float64(Float64(x * y) + Float64(x * z)) + Float64(y * z))))
end

MATLAB

function tmp = code(x, y, z)
	tmp = 2.0 * sqrt((((x * y) + (x * z)) + (y * z)));
end

Wolfram

code[x_, y_, z_] := N[(2.0 * N[Sqrt[N[(N[(N[(x * y), $MachinePrecision] + N[(x * z), $MachinePrecision]), $MachinePrecision] + N[(y * z), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]

Initial Program: 70.2% accurate, 1.0× speedup

Math

\[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (* 2.0 (sqrt (+ (+ (* x y) (* x z)) (* y z)))))

C

double code(double x, double y, double z) {
	return 2.0 * sqrt((((x * y) + (x * z)) + (y * z)));
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    code = 2.0d0 * sqrt((((x * y) + (x * z)) + (y * z)))
end function

Java

public static double code(double x, double y, double z) {
	return 2.0 * Math.sqrt((((x * y) + (x * z)) + (y * z)));
}

Python

def code(x, y, z):
	return 2.0 * math.sqrt((((x * y) + (x * z)) + (y * z)))

Julia

function code(x, y, z)
	return Float64(2.0 * sqrt(Float64(Float64(Float64(x * y) + Float64(x * z)) + Float64(y * z))))
end

MATLAB

function tmp = code(x, y, z)
	tmp = 2.0 * sqrt((((x * y) + (x * z)) + (y * z)));
end

Wolfram

code[x_, y_, z_] := N[(2.0 * N[Sqrt[N[(N[(N[(x * y), $MachinePrecision] + N[(x * z), $MachinePrecision]), $MachinePrecision] + N[(y * z), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]

Alternative 1: 99.2% accurate, 0.2× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_3 := \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ \mathbf{if}\;t\_2 \leq 2.4 \cdot 10^{-308}:\\ \;\;\;\;-2 \cdot \left(t\_0 \cdot \frac{\sqrt{\left(-t\_3\right) - t\_2}}{\sqrt{-t\_0}}\right)\\ \mathbf{elif}\;t\_2 \leq 2.6 \cdot 10^{-24}:\\ \;\;\;\;\left(\sqrt{\left(t\_0 + t\_3\right) + \frac{t\_0 \cdot t\_3}{t\_2}} \cdot \sqrt{t\_2}\right) \cdot 2\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \left(t\_3 \cdot \sqrt{\frac{t\_0 + t\_2}{t\_3}}\right)\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmin (fmax x y) t_1))
        (t_3 (fmax (fmax x y) t_1)))
   (if (<= t_2 2.4e-308)
     (* -2.0 (* t_0 (/ (sqrt (- (- t_3) t_2)) (sqrt (- t_0)))))
     (if (<= t_2 2.6e-24)
       (* (* (sqrt (+ (+ t_0 t_3) (/ (* t_0 t_3) t_2))) (sqrt t_2)) 2.0)
       (* 2.0 (* t_3 (sqrt (/ (+ t_0 t_2) t_3))))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmin(fmax(x, y), t_1);
	double t_3 = fmax(fmax(x, y), t_1);
	double tmp;
	if (t_2 <= 2.4e-308) {
		tmp = -2.0 * (t_0 * (sqrt((-t_3 - t_2)) / sqrt(-t_0)));
	} else if (t_2 <= 2.6e-24) {
		tmp = (sqrt(((t_0 + t_3) + ((t_0 * t_3) / t_2))) * sqrt(t_2)) * 2.0;
	} else {
		tmp = 2.0 * (t_3 * sqrt(((t_0 + t_2) / t_3)));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: t_2
    real(8) :: t_3
    real(8) :: tmp
    t_0 = fmin(fmin(x, y), z)
    t_1 = fmax(fmin(x, y), z)
    t_2 = fmin(fmax(x, y), t_1)
    t_3 = fmax(fmax(x, y), t_1)
    if (t_2 <= 2.4d-308) then
        tmp = (-2.0d0) * (t_0 * (sqrt((-t_3 - t_2)) / sqrt(-t_0)))
    else if (t_2 <= 2.6d-24) then
        tmp = (sqrt(((t_0 + t_3) + ((t_0 * t_3) / t_2))) * sqrt(t_2)) * 2.0d0
    else
        tmp = 2.0d0 * (t_3 * sqrt(((t_0 + t_2) / t_3)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmin(fmax(x, y), t_1);
	double t_3 = fmax(fmax(x, y), t_1);
	double tmp;
	if (t_2 <= 2.4e-308) {
		tmp = -2.0 * (t_0 * (Math.sqrt((-t_3 - t_2)) / Math.sqrt(-t_0)));
	} else if (t_2 <= 2.6e-24) {
		tmp = (Math.sqrt(((t_0 + t_3) + ((t_0 * t_3) / t_2))) * Math.sqrt(t_2)) * 2.0;
	} else {
		tmp = 2.0 * (t_3 * Math.sqrt(((t_0 + t_2) / t_3)));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmin(fmax(x, y), t_1)
	t_3 = fmax(fmax(x, y), t_1)
	tmp = 0
	if t_2 <= 2.4e-308:
		tmp = -2.0 * (t_0 * (math.sqrt((-t_3 - t_2)) / math.sqrt(-t_0)))
	elif t_2 <= 2.6e-24:
		tmp = (math.sqrt(((t_0 + t_3) + ((t_0 * t_3) / t_2))) * math.sqrt(t_2)) * 2.0
	else:
		tmp = 2.0 * (t_3 * math.sqrt(((t_0 + t_2) / t_3)))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmin(fmax(x, y), t_1)
	t_3 = fmax(fmax(x, y), t_1)
	tmp = 0.0
	if (t_2 <= 2.4e-308)
		tmp = Float64(-2.0 * Float64(t_0 * Float64(sqrt(Float64(Float64(-t_3) - t_2)) / sqrt(Float64(-t_0)))));
	elseif (t_2 <= 2.6e-24)
		tmp = Float64(Float64(sqrt(Float64(Float64(t_0 + t_3) + Float64(Float64(t_0 * t_3) / t_2))) * sqrt(t_2)) * 2.0);
	else
		tmp = Float64(2.0 * Float64(t_3 * sqrt(Float64(Float64(t_0 + t_2) / t_3))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = min(min(x, y), z);
	t_1 = max(min(x, y), z);
	t_2 = min(max(x, y), t_1);
	t_3 = max(max(x, y), t_1);
	tmp = 0.0;
	if (t_2 <= 2.4e-308)
		tmp = -2.0 * (t_0 * (sqrt((-t_3 - t_2)) / sqrt(-t_0)));
	elseif (t_2 <= 2.6e-24)
		tmp = (sqrt(((t_0 + t_3) + ((t_0 * t_3) / t_2))) * sqrt(t_2)) * 2.0;
	else
		tmp = 2.0 * (t_3 * sqrt(((t_0 + t_2) / t_3)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$3 = N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, If[LessEqual[t$95$2, 2.4e-308], N[(-2.0 * N[(t$95$0 * N[(N[Sqrt[N[((-t$95$3) - t$95$2), $MachinePrecision]], $MachinePrecision] / N[Sqrt[(-t$95$0)], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$2, 2.6e-24], N[(N[(N[Sqrt[N[(N[(t$95$0 + t$95$3), $MachinePrecision] + N[(N[(t$95$0 * t$95$3), $MachinePrecision] / t$95$2), $MachinePrecision]), $MachinePrecision]], $MachinePrecision] * N[Sqrt[t$95$2], $MachinePrecision]), $MachinePrecision] * 2.0), $MachinePrecision], N[(2.0 * N[(t$95$3 * N[Sqrt[N[(N[(t$95$0 + t$95$2), $MachinePrecision] / t$95$3), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]]]]]

Derivation

1. Split input into 3 regimes

2. if y < 2.40000000000000008e-308

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around -inf

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]

      2. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \color{blue}{\sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}}\right) \]

      3. lower-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      4. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      5. lower-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      7. lower-*.f64 30.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

   4. Applied rewrites 30.0%

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]
   5. Step-by-step derivation

      1. lift-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      2. lift-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      3. mul-1-neg N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{\mathsf{neg}\left(\frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}\right)}\right) \]

      4. lift-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{\mathsf{neg}\left(\frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}\right)}\right) \]

      5. distribute-neg-frac2 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{\frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{\mathsf{neg}\left(x\right)}}\right) \]

      6. sqrt-div N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}}{\color{blue}{\sqrt{\mathsf{neg}\left(x\right)}}}\right) \]

      7. lower-unsound-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}}{\color{blue}{\sqrt{\mathsf{neg}\left(x\right)}}}\right) \]

      8. lower-unsound-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}}{\sqrt{\color{blue}{\mathsf{neg}\left(x\right)}}}\right) \]

      9. lift-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot y + -1 \cdot z}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      10. +-commutative N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z + -1 \cdot y}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      11. mul-1-neg N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z + \left(\mathsf{neg}\left(y\right)\right)}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      12. sub-flip-reverse N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z - y}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      13. lower--.f64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z - y}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      14. lift-*.f64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      15. mul-1-neg N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(\mathsf{neg}\left(z\right)\right) - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      16. lower-neg.f64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      17. lower-unsound-sqrt.64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      18. lower-neg.f64 33.0%

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\sqrt{-x}}\right) \]

   6. Applied rewrites 33.0%

      \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\color{blue}{\sqrt{-x}}}\right) \]

   if 2.40000000000000008e-308 < y < 2.6e-24

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]
   2. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto \color{blue}{2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z}} \]

      2. *-commutative N/A

         \[\leadsto \color{blue}{\sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \cdot 2} \]

      3. lower-*.f64 70.2%

         \[\leadsto \color{blue}{\sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \cdot 2} \]

   3. Applied rewrites 70.3%

      \[\leadsto \color{blue}{\sqrt{\mathsf{fma}\left(z + x, y, z \cdot x\right)} \cdot 2} \]

   5. Applied rewrites 39.3%

      \[\leadsto \color{blue}{\left(\sqrt{\left(x + z\right) + \frac{x \cdot z}{y}} \cdot \sqrt{y}\right)} \cdot 2 \]

   if 2.6e-24 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in z around inf

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \color{blue}{\sqrt{\frac{x + y}{z}}}\right) \]

      2. lower-sqrt.64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      3. lower-/.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      4. lower-+.f64 29.6%

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

   4. Applied rewrites 29.6%

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 2: 96.9% accurate, 0.2× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_3 := t\_0 + t\_2\\ t_4 := \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ \mathbf{if}\;t\_2 \leq -2.95 \cdot 10^{-248}:\\ \;\;\;\;-2 \cdot \left(t\_0 \cdot \frac{\sqrt{\left(-t\_4\right) - t\_2}}{\sqrt{-t\_0}}\right)\\ \mathbf{elif}\;t\_2 \leq 4.8 \cdot 10^{+23}:\\ \;\;\;\;2 \cdot \sqrt{t\_4 \cdot t\_3}\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \left(t\_4 \cdot \sqrt{\frac{t\_3}{t\_4}}\right)\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmin (fmax x y) t_1))
        (t_3 (+ t_0 t_2))
        (t_4 (fmax (fmax x y) t_1)))
   (if (<= t_2 -2.95e-248)
     (* -2.0 (* t_0 (/ (sqrt (- (- t_4) t_2)) (sqrt (- t_0)))))
     (if (<= t_2 4.8e+23)
       (* 2.0 (sqrt (* t_4 t_3)))
       (* 2.0 (* t_4 (sqrt (/ t_3 t_4))))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmin(fmax(x, y), t_1);
	double t_3 = t_0 + t_2;
	double t_4 = fmax(fmax(x, y), t_1);
	double tmp;
	if (t_2 <= -2.95e-248) {
		tmp = -2.0 * (t_0 * (sqrt((-t_4 - t_2)) / sqrt(-t_0)));
	} else if (t_2 <= 4.8e+23) {
		tmp = 2.0 * sqrt((t_4 * t_3));
	} else {
		tmp = 2.0 * (t_4 * sqrt((t_3 / t_4)));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: t_2
    real(8) :: t_3
    real(8) :: t_4
    real(8) :: tmp
    t_0 = fmin(fmin(x, y), z)
    t_1 = fmax(fmin(x, y), z)
    t_2 = fmin(fmax(x, y), t_1)
    t_3 = t_0 + t_2
    t_4 = fmax(fmax(x, y), t_1)
    if (t_2 <= (-2.95d-248)) then
        tmp = (-2.0d0) * (t_0 * (sqrt((-t_4 - t_2)) / sqrt(-t_0)))
    else if (t_2 <= 4.8d+23) then
        tmp = 2.0d0 * sqrt((t_4 * t_3))
    else
        tmp = 2.0d0 * (t_4 * sqrt((t_3 / t_4)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmin(fmax(x, y), t_1);
	double t_3 = t_0 + t_2;
	double t_4 = fmax(fmax(x, y), t_1);
	double tmp;
	if (t_2 <= -2.95e-248) {
		tmp = -2.0 * (t_0 * (Math.sqrt((-t_4 - t_2)) / Math.sqrt(-t_0)));
	} else if (t_2 <= 4.8e+23) {
		tmp = 2.0 * Math.sqrt((t_4 * t_3));
	} else {
		tmp = 2.0 * (t_4 * Math.sqrt((t_3 / t_4)));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmin(fmax(x, y), t_1)
	t_3 = t_0 + t_2
	t_4 = fmax(fmax(x, y), t_1)
	tmp = 0
	if t_2 <= -2.95e-248:
		tmp = -2.0 * (t_0 * (math.sqrt((-t_4 - t_2)) / math.sqrt(-t_0)))
	elif t_2 <= 4.8e+23:
		tmp = 2.0 * math.sqrt((t_4 * t_3))
	else:
		tmp = 2.0 * (t_4 * math.sqrt((t_3 / t_4)))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmin(fmax(x, y), t_1)
	t_3 = Float64(t_0 + t_2)
	t_4 = fmax(fmax(x, y), t_1)
	tmp = 0.0
	if (t_2 <= -2.95e-248)
		tmp = Float64(-2.0 * Float64(t_0 * Float64(sqrt(Float64(Float64(-t_4) - t_2)) / sqrt(Float64(-t_0)))));
	elseif (t_2 <= 4.8e+23)
		tmp = Float64(2.0 * sqrt(Float64(t_4 * t_3)));
	else
		tmp = Float64(2.0 * Float64(t_4 * sqrt(Float64(t_3 / t_4))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = min(min(x, y), z);
	t_1 = max(min(x, y), z);
	t_2 = min(max(x, y), t_1);
	t_3 = t_0 + t_2;
	t_4 = max(max(x, y), t_1);
	tmp = 0.0;
	if (t_2 <= -2.95e-248)
		tmp = -2.0 * (t_0 * (sqrt((-t_4 - t_2)) / sqrt(-t_0)));
	elseif (t_2 <= 4.8e+23)
		tmp = 2.0 * sqrt((t_4 * t_3));
	else
		tmp = 2.0 * (t_4 * sqrt((t_3 / t_4)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$3 = N[(t$95$0 + t$95$2), $MachinePrecision]}, Block[{t$95$4 = N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, If[LessEqual[t$95$2, -2.95e-248], N[(-2.0 * N[(t$95$0 * N[(N[Sqrt[N[((-t$95$4) - t$95$2), $MachinePrecision]], $MachinePrecision] / N[Sqrt[(-t$95$0)], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$2, 4.8e+23], N[(2.0 * N[Sqrt[N[(t$95$4 * t$95$3), $MachinePrecision]], $MachinePrecision]), $MachinePrecision], N[(2.0 * N[(t$95$4 * N[Sqrt[N[(t$95$3 / t$95$4), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]]]]]]

Derivation

1. Split input into 3 regimes

2. if y < -2.94999999999999993e-248

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around -inf

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]

      2. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \color{blue}{\sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}}\right) \]

      3. lower-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      4. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      5. lower-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      7. lower-*.f64 30.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

   4. Applied rewrites 30.0%

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]
   5. Step-by-step derivation

      1. lift-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      2. lift-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      3. mul-1-neg N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{\mathsf{neg}\left(\frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}\right)}\right) \]

      4. lift-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{\mathsf{neg}\left(\frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}\right)}\right) \]

      5. distribute-neg-frac2 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{\frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{\mathsf{neg}\left(x\right)}}\right) \]

      6. sqrt-div N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}}{\color{blue}{\sqrt{\mathsf{neg}\left(x\right)}}}\right) \]

      7. lower-unsound-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}}{\color{blue}{\sqrt{\mathsf{neg}\left(x\right)}}}\right) \]

      8. lower-unsound-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}}{\sqrt{\color{blue}{\mathsf{neg}\left(x\right)}}}\right) \]

      9. lift-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot y + -1 \cdot z}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      10. +-commutative N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z + -1 \cdot y}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      11. mul-1-neg N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z + \left(\mathsf{neg}\left(y\right)\right)}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      12. sub-flip-reverse N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z - y}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      13. lower--.f64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z - y}}{\sqrt{\mathsf{neg}\left(\color{blue}{x}\right)}}\right) \]

      14. lift-*.f64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{-1 \cdot z - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      15. mul-1-neg N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(\mathsf{neg}\left(z\right)\right) - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      16. lower-neg.f64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      17. lower-unsound-sqrt.64 N/A

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\sqrt{\mathsf{neg}\left(x\right)}}\right) \]

      18. lower-neg.f64 33.0%

          \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\sqrt{-x}}\right) \]

   6. Applied rewrites 33.0%

      \[\leadsto -2 \cdot \left(x \cdot \frac{\sqrt{\left(-z\right) - y}}{\color{blue}{\sqrt{-x}}}\right) \]

   if -2.94999999999999993e-248 < y < 4.8e23

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in z around inf

      \[\leadsto 2 \cdot \sqrt{\color{blue}{z \cdot \left(x + y\right)}} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \sqrt{z \cdot \color{blue}{\left(x + y\right)}} \]

      2. lower-+.f64 47.2%

         \[\leadsto 2 \cdot \sqrt{z \cdot \left(x + \color{blue}{y}\right)} \]

   4. Applied rewrites 47.2%

      \[\leadsto 2 \cdot \sqrt{\color{blue}{z \cdot \left(x + y\right)}} \]

   if 4.8e23 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in z around inf

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \color{blue}{\sqrt{\frac{x + y}{z}}}\right) \]

      2. lower-sqrt.64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      3. lower-/.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      4. lower-+.f64 29.6%

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

   4. Applied rewrites 29.6%

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 3: 96.8% accurate, 0.2× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_3 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ \mathbf{if}\;t\_3 \leq -220000:\\ \;\;\;\;\left(\sqrt{\frac{t\_3 + t\_2}{t\_0}} \cdot t\_0\right) \cdot -2\\ \mathbf{elif}\;t\_3 \leq 1.25 \cdot 10^{+16}:\\ \;\;\;\;\sqrt{\mathsf{fma}\left(t\_2 + t\_0, t\_3, t\_2 \cdot t\_0\right)} \cdot 2\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \left(t\_2 \cdot \sqrt{\frac{t\_0 + t\_3}{t\_2}}\right)\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmax (fmax x y) t_1))
        (t_3 (fmin (fmax x y) t_1)))
   (if (<= t_3 -220000.0)
     (* (* (sqrt (/ (+ t_3 t_2) t_0)) t_0) -2.0)
     (if (<= t_3 1.25e+16)
       (* (sqrt (fma (+ t_2 t_0) t_3 (* t_2 t_0))) 2.0)
       (* 2.0 (* t_2 (sqrt (/ (+ t_0 t_3) t_2))))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_3 <= -220000.0) {
		tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	} else if (t_3 <= 1.25e+16) {
		tmp = sqrt(fma((t_2 + t_0), t_3, (t_2 * t_0))) * 2.0;
	} else {
		tmp = 2.0 * (t_2 * sqrt(((t_0 + t_3) / t_2)));
	}
	return tmp;
}

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	tmp = 0.0
	if (t_3 <= -220000.0)
		tmp = Float64(Float64(sqrt(Float64(Float64(t_3 + t_2) / t_0)) * t_0) * -2.0);
	elseif (t_3 <= 1.25e+16)
		tmp = Float64(sqrt(fma(Float64(t_2 + t_0), t_3, Float64(t_2 * t_0))) * 2.0);
	else
		tmp = Float64(2.0 * Float64(t_2 * sqrt(Float64(Float64(t_0 + t_3) / t_2))));
	end
	return tmp
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$3 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, If[LessEqual[t$95$3, -220000.0], N[(N[(N[Sqrt[N[(N[(t$95$3 + t$95$2), $MachinePrecision] / t$95$0), $MachinePrecision]], $MachinePrecision] * t$95$0), $MachinePrecision] * -2.0), $MachinePrecision], If[LessEqual[t$95$3, 1.25e+16], N[(N[Sqrt[N[(N[(t$95$2 + t$95$0), $MachinePrecision] * t$95$3 + N[(t$95$2 * t$95$0), $MachinePrecision]), $MachinePrecision]], $MachinePrecision] * 2.0), $MachinePrecision], N[(2.0 * N[(t$95$2 * N[Sqrt[N[(N[(t$95$0 + t$95$3), $MachinePrecision] / t$95$2), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]]]]]

Derivation

1. Split input into 3 regimes

2. if y < -2.2e5

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around -inf

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]

      2. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \color{blue}{\sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}}\right) \]

      3. lower-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      4. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      5. lower-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      7. lower-*.f64 30.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

   4. Applied rewrites 30.0%

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]
   5. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]

      2. *-commutative N/A

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

      3. lower-*.f64 30.0%

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

   6. Applied rewrites 30.0%

      \[\leadsto \left(\sqrt{\frac{y + z}{x}} \cdot x\right) \cdot \color{blue}{-2} \]

   if -2.2e5 < y < 1.25e16

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]
   2. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto \color{blue}{2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z}} \]

      2. *-commutative N/A

         \[\leadsto \color{blue}{\sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \cdot 2} \]

      3. lower-*.f64 70.2%

         \[\leadsto \color{blue}{\sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \cdot 2} \]

   3. Applied rewrites 70.3%

      \[\leadsto \color{blue}{\sqrt{\mathsf{fma}\left(z + x, y, z \cdot x\right)} \cdot 2} \]

   if 1.25e16 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in z around inf

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \color{blue}{\sqrt{\frac{x + y}{z}}}\right) \]

      2. lower-sqrt.64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      3. lower-/.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      4. lower-+.f64 29.6%

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

   4. Applied rewrites 29.6%

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 4: 95.1% accurate, 0.2× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_3 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ \mathbf{if}\;t\_3 \leq -2.5 \cdot 10^{-24}:\\ \;\;\;\;\left(\sqrt{\frac{t\_3 + t\_2}{t\_0}} \cdot t\_0\right) \cdot -2\\ \mathbf{elif}\;t\_3 \leq 6.8 \cdot 10^{+19}:\\ \;\;\;\;2 \cdot \sqrt{t\_3 \cdot \left(t\_0 + t\_2\right)}\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \left(t\_2 \cdot \sqrt{\frac{t\_0 + t\_3}{t\_2}}\right)\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmax (fmax x y) t_1))
        (t_3 (fmin (fmax x y) t_1)))
   (if (<= t_3 -2.5e-24)
     (* (* (sqrt (/ (+ t_3 t_2) t_0)) t_0) -2.0)
     (if (<= t_3 6.8e+19)
       (* 2.0 (sqrt (* t_3 (+ t_0 t_2))))
       (* 2.0 (* t_2 (sqrt (/ (+ t_0 t_3) t_2))))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_3 <= -2.5e-24) {
		tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	} else if (t_3 <= 6.8e+19) {
		tmp = 2.0 * sqrt((t_3 * (t_0 + t_2)));
	} else {
		tmp = 2.0 * (t_2 * sqrt(((t_0 + t_3) / t_2)));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: t_2
    real(8) :: t_3
    real(8) :: tmp
    t_0 = fmin(fmin(x, y), z)
    t_1 = fmax(fmin(x, y), z)
    t_2 = fmax(fmax(x, y), t_1)
    t_3 = fmin(fmax(x, y), t_1)
    if (t_3 <= (-2.5d-24)) then
        tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * (-2.0d0)
    else if (t_3 <= 6.8d+19) then
        tmp = 2.0d0 * sqrt((t_3 * (t_0 + t_2)))
    else
        tmp = 2.0d0 * (t_2 * sqrt(((t_0 + t_3) / t_2)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_3 <= -2.5e-24) {
		tmp = (Math.sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	} else if (t_3 <= 6.8e+19) {
		tmp = 2.0 * Math.sqrt((t_3 * (t_0 + t_2)));
	} else {
		tmp = 2.0 * (t_2 * Math.sqrt(((t_0 + t_3) / t_2)));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	tmp = 0
	if t_3 <= -2.5e-24:
		tmp = (math.sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0
	elif t_3 <= 6.8e+19:
		tmp = 2.0 * math.sqrt((t_3 * (t_0 + t_2)))
	else:
		tmp = 2.0 * (t_2 * math.sqrt(((t_0 + t_3) / t_2)))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	tmp = 0.0
	if (t_3 <= -2.5e-24)
		tmp = Float64(Float64(sqrt(Float64(Float64(t_3 + t_2) / t_0)) * t_0) * -2.0);
	elseif (t_3 <= 6.8e+19)
		tmp = Float64(2.0 * sqrt(Float64(t_3 * Float64(t_0 + t_2))));
	else
		tmp = Float64(2.0 * Float64(t_2 * sqrt(Float64(Float64(t_0 + t_3) / t_2))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = min(min(x, y), z);
	t_1 = max(min(x, y), z);
	t_2 = max(max(x, y), t_1);
	t_3 = min(max(x, y), t_1);
	tmp = 0.0;
	if (t_3 <= -2.5e-24)
		tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	elseif (t_3 <= 6.8e+19)
		tmp = 2.0 * sqrt((t_3 * (t_0 + t_2)));
	else
		tmp = 2.0 * (t_2 * sqrt(((t_0 + t_3) / t_2)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$3 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, If[LessEqual[t$95$3, -2.5e-24], N[(N[(N[Sqrt[N[(N[(t$95$3 + t$95$2), $MachinePrecision] / t$95$0), $MachinePrecision]], $MachinePrecision] * t$95$0), $MachinePrecision] * -2.0), $MachinePrecision], If[LessEqual[t$95$3, 6.8e+19], N[(2.0 * N[Sqrt[N[(t$95$3 * N[(t$95$0 + t$95$2), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision], N[(2.0 * N[(t$95$2 * N[Sqrt[N[(N[(t$95$0 + t$95$3), $MachinePrecision] / t$95$2), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]]]]]

Derivation

1. Split input into 3 regimes

2. if y < -2.4999999999999999e-24

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around -inf

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]

      2. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \color{blue}{\sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}}\right) \]

      3. lower-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      4. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      5. lower-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      7. lower-*.f64 30.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

   4. Applied rewrites 30.0%

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]
   5. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]

      2. *-commutative N/A

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

      3. lower-*.f64 30.0%

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

   6. Applied rewrites 30.0%

      \[\leadsto \left(\sqrt{\frac{y + z}{x}} \cdot x\right) \cdot \color{blue}{-2} \]

   if -2.4999999999999999e-24 < y < 6.8e19

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in y around inf

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \sqrt{y \cdot \color{blue}{\left(x + z\right)}} \]

      2. lower-+.f64 47.8%

         \[\leadsto 2 \cdot \sqrt{y \cdot \left(x + \color{blue}{z}\right)} \]

   4. Applied rewrites 47.8%

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]

   if 6.8e19 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in z around inf

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \color{blue}{\sqrt{\frac{x + y}{z}}}\right) \]

      2. lower-sqrt.64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      3. lower-/.f64 N/A

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

      4. lower-+.f64 29.6%

         \[\leadsto 2 \cdot \left(z \cdot \sqrt{\frac{x + y}{z}}\right) \]

   4. Applied rewrites 29.6%

      \[\leadsto 2 \cdot \color{blue}{\left(z \cdot \sqrt{\frac{x + y}{z}}\right)} \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 5: 94.7% accurate, 0.2× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_3 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_4 := t\_0 + t\_2\\ \mathbf{if}\;t\_3 \leq -2.5 \cdot 10^{-24}:\\ \;\;\;\;\left(\sqrt{\frac{t\_3 + t\_2}{t\_0}} \cdot t\_0\right) \cdot -2\\ \mathbf{elif}\;t\_3 \leq 2 \cdot 10^{+33}:\\ \;\;\;\;2 \cdot \sqrt{t\_3 \cdot t\_4}\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \left(t\_3 \cdot \sqrt{\frac{t\_4}{t\_3}}\right)\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmax (fmax x y) t_1))
        (t_3 (fmin (fmax x y) t_1))
        (t_4 (+ t_0 t_2)))
   (if (<= t_3 -2.5e-24)
     (* (* (sqrt (/ (+ t_3 t_2) t_0)) t_0) -2.0)
     (if (<= t_3 2e+33)
       (* 2.0 (sqrt (* t_3 t_4)))
       (* 2.0 (* t_3 (sqrt (/ t_4 t_3))))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double t_4 = t_0 + t_2;
	double tmp;
	if (t_3 <= -2.5e-24) {
		tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	} else if (t_3 <= 2e+33) {
		tmp = 2.0 * sqrt((t_3 * t_4));
	} else {
		tmp = 2.0 * (t_3 * sqrt((t_4 / t_3)));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: t_2
    real(8) :: t_3
    real(8) :: t_4
    real(8) :: tmp
    t_0 = fmin(fmin(x, y), z)
    t_1 = fmax(fmin(x, y), z)
    t_2 = fmax(fmax(x, y), t_1)
    t_3 = fmin(fmax(x, y), t_1)
    t_4 = t_0 + t_2
    if (t_3 <= (-2.5d-24)) then
        tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * (-2.0d0)
    else if (t_3 <= 2d+33) then
        tmp = 2.0d0 * sqrt((t_3 * t_4))
    else
        tmp = 2.0d0 * (t_3 * sqrt((t_4 / t_3)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double t_4 = t_0 + t_2;
	double tmp;
	if (t_3 <= -2.5e-24) {
		tmp = (Math.sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	} else if (t_3 <= 2e+33) {
		tmp = 2.0 * Math.sqrt((t_3 * t_4));
	} else {
		tmp = 2.0 * (t_3 * Math.sqrt((t_4 / t_3)));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	t_4 = t_0 + t_2
	tmp = 0
	if t_3 <= -2.5e-24:
		tmp = (math.sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0
	elif t_3 <= 2e+33:
		tmp = 2.0 * math.sqrt((t_3 * t_4))
	else:
		tmp = 2.0 * (t_3 * math.sqrt((t_4 / t_3)))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	t_4 = Float64(t_0 + t_2)
	tmp = 0.0
	if (t_3 <= -2.5e-24)
		tmp = Float64(Float64(sqrt(Float64(Float64(t_3 + t_2) / t_0)) * t_0) * -2.0);
	elseif (t_3 <= 2e+33)
		tmp = Float64(2.0 * sqrt(Float64(t_3 * t_4)));
	else
		tmp = Float64(2.0 * Float64(t_3 * sqrt(Float64(t_4 / t_3))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = min(min(x, y), z);
	t_1 = max(min(x, y), z);
	t_2 = max(max(x, y), t_1);
	t_3 = min(max(x, y), t_1);
	t_4 = t_0 + t_2;
	tmp = 0.0;
	if (t_3 <= -2.5e-24)
		tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	elseif (t_3 <= 2e+33)
		tmp = 2.0 * sqrt((t_3 * t_4));
	else
		tmp = 2.0 * (t_3 * sqrt((t_4 / t_3)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$3 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$4 = N[(t$95$0 + t$95$2), $MachinePrecision]}, If[LessEqual[t$95$3, -2.5e-24], N[(N[(N[Sqrt[N[(N[(t$95$3 + t$95$2), $MachinePrecision] / t$95$0), $MachinePrecision]], $MachinePrecision] * t$95$0), $MachinePrecision] * -2.0), $MachinePrecision], If[LessEqual[t$95$3, 2e+33], N[(2.0 * N[Sqrt[N[(t$95$3 * t$95$4), $MachinePrecision]], $MachinePrecision]), $MachinePrecision], N[(2.0 * N[(t$95$3 * N[Sqrt[N[(t$95$4 / t$95$3), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]]]]]]

Derivation

1. Split input into 3 regimes

2. if y < -2.4999999999999999e-24

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around -inf

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]

      2. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \color{blue}{\sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}}\right) \]

      3. lower-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      4. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      5. lower-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      7. lower-*.f64 30.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

   4. Applied rewrites 30.0%

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]
   5. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]

      2. *-commutative N/A

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

      3. lower-*.f64 30.0%

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

   6. Applied rewrites 30.0%

      \[\leadsto \left(\sqrt{\frac{y + z}{x}} \cdot x\right) \cdot \color{blue}{-2} \]

   if -2.4999999999999999e-24 < y < 1.9999999999999999e33

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in y around inf

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \sqrt{y \cdot \color{blue}{\left(x + z\right)}} \]

      2. lower-+.f64 47.8%

         \[\leadsto 2 \cdot \sqrt{y \cdot \left(x + \color{blue}{z}\right)} \]

   4. Applied rewrites 47.8%

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]

   if 1.9999999999999999e33 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in y around inf

      \[\leadsto 2 \cdot \color{blue}{\left(y \cdot \sqrt{\frac{x + z}{y}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \left(y \cdot \color{blue}{\sqrt{\frac{x + z}{y}}}\right) \]

      2. lower-sqrt.64 N/A

         \[\leadsto 2 \cdot \left(y \cdot \sqrt{\frac{x + z}{y}}\right) \]

      3. lower-/.f64 N/A

         \[\leadsto 2 \cdot \left(y \cdot \sqrt{\frac{x + z}{y}}\right) \]

      4. lower-+.f64 29.8%

         \[\leadsto 2 \cdot \left(y \cdot \sqrt{\frac{x + z}{y}}\right) \]

   4. Applied rewrites 29.8%

      \[\leadsto 2 \cdot \color{blue}{\left(y \cdot \sqrt{\frac{x + z}{y}}\right)} \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 6: 81.7% accurate, 0.3× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_3 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ \mathbf{if}\;t\_3 \leq -2.5 \cdot 10^{-24}:\\ \;\;\;\;\left(\sqrt{\frac{t\_3 + t\_2}{t\_0}} \cdot t\_0\right) \cdot -2\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \sqrt{t\_3 \cdot \left(t\_0 + t\_2\right)}\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmax (fmax x y) t_1))
        (t_3 (fmin (fmax x y) t_1)))
   (if (<= t_3 -2.5e-24)
     (* (* (sqrt (/ (+ t_3 t_2) t_0)) t_0) -2.0)
     (* 2.0 (sqrt (* t_3 (+ t_0 t_2)))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_3 <= -2.5e-24) {
		tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	} else {
		tmp = 2.0 * sqrt((t_3 * (t_0 + t_2)));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: t_2
    real(8) :: t_3
    real(8) :: tmp
    t_0 = fmin(fmin(x, y), z)
    t_1 = fmax(fmin(x, y), z)
    t_2 = fmax(fmax(x, y), t_1)
    t_3 = fmin(fmax(x, y), t_1)
    if (t_3 <= (-2.5d-24)) then
        tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * (-2.0d0)
    else
        tmp = 2.0d0 * sqrt((t_3 * (t_0 + t_2)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_3 <= -2.5e-24) {
		tmp = (Math.sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	} else {
		tmp = 2.0 * Math.sqrt((t_3 * (t_0 + t_2)));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	tmp = 0
	if t_3 <= -2.5e-24:
		tmp = (math.sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0
	else:
		tmp = 2.0 * math.sqrt((t_3 * (t_0 + t_2)))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	tmp = 0.0
	if (t_3 <= -2.5e-24)
		tmp = Float64(Float64(sqrt(Float64(Float64(t_3 + t_2) / t_0)) * t_0) * -2.0);
	else
		tmp = Float64(2.0 * sqrt(Float64(t_3 * Float64(t_0 + t_2))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = min(min(x, y), z);
	t_1 = max(min(x, y), z);
	t_2 = max(max(x, y), t_1);
	t_3 = min(max(x, y), t_1);
	tmp = 0.0;
	if (t_3 <= -2.5e-24)
		tmp = (sqrt(((t_3 + t_2) / t_0)) * t_0) * -2.0;
	else
		tmp = 2.0 * sqrt((t_3 * (t_0 + t_2)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$3 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, If[LessEqual[t$95$3, -2.5e-24], N[(N[(N[Sqrt[N[(N[(t$95$3 + t$95$2), $MachinePrecision] / t$95$0), $MachinePrecision]], $MachinePrecision] * t$95$0), $MachinePrecision] * -2.0), $MachinePrecision], N[(2.0 * N[Sqrt[N[(t$95$3 * N[(t$95$0 + t$95$2), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]]]]]]

Derivation

1. Split input into 2 regimes

2. if y < -2.4999999999999999e-24

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around -inf

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]

      2. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \color{blue}{\sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}}\right) \]

      3. lower-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      4. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      5. lower-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      7. lower-*.f64 30.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

   4. Applied rewrites 30.0%

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]
   5. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]

      2. *-commutative N/A

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

      3. lower-*.f64 30.0%

         \[\leadsto \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \cdot \color{blue}{-2} \]

   6. Applied rewrites 30.0%

      \[\leadsto \left(\sqrt{\frac{y + z}{x}} \cdot x\right) \cdot \color{blue}{-2} \]

   if -2.4999999999999999e-24 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in y around inf

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \sqrt{y \cdot \color{blue}{\left(x + z\right)}} \]

      2. lower-+.f64 47.8%

         \[\leadsto 2 \cdot \sqrt{y \cdot \left(x + \color{blue}{z}\right)} \]

   4. Applied rewrites 47.8%

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 7: 81.3% accurate, 0.3× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ \mathbf{if}\;t\_2 \leq -4.5 \cdot 10^{-26}:\\ \;\;\;\;-2 \cdot \left(t\_0 \cdot \sqrt{\frac{t\_2}{t\_0}}\right)\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \sqrt{t\_2 \cdot \left(t\_0 + \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\right)}\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmin (fmax x y) t_1)))
   (if (<= t_2 -4.5e-26)
     (* -2.0 (* t_0 (sqrt (/ t_2 t_0))))
     (* 2.0 (sqrt (* t_2 (+ t_0 (fmax (fmax x y) t_1))))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_2 <= -4.5e-26) {
		tmp = -2.0 * (t_0 * sqrt((t_2 / t_0)));
	} else {
		tmp = 2.0 * sqrt((t_2 * (t_0 + fmax(fmax(x, y), t_1))));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: t_2
    real(8) :: tmp
    t_0 = fmin(fmin(x, y), z)
    t_1 = fmax(fmin(x, y), z)
    t_2 = fmin(fmax(x, y), t_1)
    if (t_2 <= (-4.5d-26)) then
        tmp = (-2.0d0) * (t_0 * sqrt((t_2 / t_0)))
    else
        tmp = 2.0d0 * sqrt((t_2 * (t_0 + fmax(fmax(x, y), t_1))))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_2 <= -4.5e-26) {
		tmp = -2.0 * (t_0 * Math.sqrt((t_2 / t_0)));
	} else {
		tmp = 2.0 * Math.sqrt((t_2 * (t_0 + fmax(fmax(x, y), t_1))));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmin(fmax(x, y), t_1)
	tmp = 0
	if t_2 <= -4.5e-26:
		tmp = -2.0 * (t_0 * math.sqrt((t_2 / t_0)))
	else:
		tmp = 2.0 * math.sqrt((t_2 * (t_0 + fmax(fmax(x, y), t_1))))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmin(fmax(x, y), t_1)
	tmp = 0.0
	if (t_2 <= -4.5e-26)
		tmp = Float64(-2.0 * Float64(t_0 * sqrt(Float64(t_2 / t_0))));
	else
		tmp = Float64(2.0 * sqrt(Float64(t_2 * Float64(t_0 + fmax(fmax(x, y), t_1)))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = min(min(x, y), z);
	t_1 = max(min(x, y), z);
	t_2 = min(max(x, y), t_1);
	tmp = 0.0;
	if (t_2 <= -4.5e-26)
		tmp = -2.0 * (t_0 * sqrt((t_2 / t_0)));
	else
		tmp = 2.0 * sqrt((t_2 * (t_0 + max(max(x, y), t_1))));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, If[LessEqual[t$95$2, -4.5e-26], N[(-2.0 * N[(t$95$0 * N[Sqrt[N[(t$95$2 / t$95$0), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], N[(2.0 * N[Sqrt[N[(t$95$2 * N[(t$95$0 + N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]]]]]

Derivation

1. Split input into 2 regimes

2. if y < -4.4999999999999999e-26

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around -inf

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto -2 \cdot \color{blue}{\left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right)} \]

      2. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \color{blue}{\sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}}\right) \]

      3. lower-sqrt.64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      4. lower-*.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      5. lower-/.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{-1 \cdot y + -1 \cdot z}{x}}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

      7. lower-*.f64 30.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right) \]

   4. Applied rewrites 30.0%

      \[\leadsto \color{blue}{-2 \cdot \left(x \cdot \sqrt{-1 \cdot \frac{\mathsf{fma}\left(-1, y, -1 \cdot z\right)}{x}}\right)} \]

   5. Taylor expanded in y around inf

      \[\leadsto -2 \cdot \left(x \cdot \sqrt{\frac{y}{x}}\right) \]
   6. Step-by-step derivation

      1. lower-/.f64 16.0%

         \[\leadsto -2 \cdot \left(x \cdot \sqrt{\frac{y}{x}}\right) \]

   7. Applied rewrites 16.0%

      \[\leadsto -2 \cdot \left(x \cdot \sqrt{\frac{y}{x}}\right) \]

   if -4.4999999999999999e-26 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in y around inf

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \sqrt{y \cdot \color{blue}{\left(x + z\right)}} \]

      2. lower-+.f64 47.8%

         \[\leadsto 2 \cdot \sqrt{y \cdot \left(x + \color{blue}{z}\right)} \]

   4. Applied rewrites 47.8%

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 8: 70.4% accurate, 0.3× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_2 := \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ t_3 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_1\right)\\ \mathbf{if}\;t\_3 \leq -5 \cdot 10^{-285}:\\ \;\;\;\;\sqrt{t\_0 \cdot \left(t\_3 + t\_2\right)} \cdot 2\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \sqrt{t\_2 \cdot \left(t\_0 + t\_3\right)}\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmin (fmin x y) z))
        (t_1 (fmax (fmin x y) z))
        (t_2 (fmax (fmax x y) t_1))
        (t_3 (fmin (fmax x y) t_1)))
   (if (<= t_3 -5e-285)
     (* (sqrt (* t_0 (+ t_3 t_2))) 2.0)
     (* 2.0 (sqrt (* t_2 (+ t_0 t_3)))))))

C

double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_3 <= -5e-285) {
		tmp = sqrt((t_0 * (t_3 + t_2))) * 2.0;
	} else {
		tmp = 2.0 * sqrt((t_2 * (t_0 + t_3)));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: t_2
    real(8) :: t_3
    real(8) :: tmp
    t_0 = fmin(fmin(x, y), z)
    t_1 = fmax(fmin(x, y), z)
    t_2 = fmax(fmax(x, y), t_1)
    t_3 = fmin(fmax(x, y), t_1)
    if (t_3 <= (-5d-285)) then
        tmp = sqrt((t_0 * (t_3 + t_2))) * 2.0d0
    else
        tmp = 2.0d0 * sqrt((t_2 * (t_0 + t_3)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmin(fmin(x, y), z);
	double t_1 = fmax(fmin(x, y), z);
	double t_2 = fmax(fmax(x, y), t_1);
	double t_3 = fmin(fmax(x, y), t_1);
	double tmp;
	if (t_3 <= -5e-285) {
		tmp = Math.sqrt((t_0 * (t_3 + t_2))) * 2.0;
	} else {
		tmp = 2.0 * Math.sqrt((t_2 * (t_0 + t_3)));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	tmp = 0
	if t_3 <= -5e-285:
		tmp = math.sqrt((t_0 * (t_3 + t_2))) * 2.0
	else:
		tmp = 2.0 * math.sqrt((t_2 * (t_0 + t_3)))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmin(fmin(x, y), z)
	t_1 = fmax(fmin(x, y), z)
	t_2 = fmax(fmax(x, y), t_1)
	t_3 = fmin(fmax(x, y), t_1)
	tmp = 0.0
	if (t_3 <= -5e-285)
		tmp = Float64(sqrt(Float64(t_0 * Float64(t_3 + t_2))) * 2.0);
	else
		tmp = Float64(2.0 * sqrt(Float64(t_2 * Float64(t_0 + t_3))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = min(min(x, y), z);
	t_1 = max(min(x, y), z);
	t_2 = max(max(x, y), t_1);
	t_3 = min(max(x, y), t_1);
	tmp = 0.0;
	if (t_3 <= -5e-285)
		tmp = sqrt((t_0 * (t_3 + t_2))) * 2.0;
	else
		tmp = 2.0 * sqrt((t_2 * (t_0 + t_3)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$2 = N[Max[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, Block[{t$95$3 = N[Min[N[Max[x, y], $MachinePrecision], t$95$1], $MachinePrecision]}, If[LessEqual[t$95$3, -5e-285], N[(N[Sqrt[N[(t$95$0 * N[(t$95$3 + t$95$2), $MachinePrecision]), $MachinePrecision]], $MachinePrecision] * 2.0), $MachinePrecision], N[(2.0 * N[Sqrt[N[(t$95$2 * N[(t$95$0 + t$95$3), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]]]]]]

Derivation

1. Split input into 2 regimes

2. if y < -5.00000000000000018e-285

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]
   2. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto \color{blue}{2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z}} \]

      2. *-commutative N/A

         \[\leadsto \color{blue}{\sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \cdot 2} \]

      3. lower-*.f64 70.2%

         \[\leadsto \color{blue}{\sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \cdot 2} \]

   3. Applied rewrites 70.3%

      \[\leadsto \color{blue}{\sqrt{\mathsf{fma}\left(z + x, y, z \cdot x\right)} \cdot 2} \]

   4. Taylor expanded in y around inf

      \[\leadsto \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \cdot 2 \]
   5. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto \sqrt{y \cdot \color{blue}{\left(x + z\right)}} \cdot 2 \]

      2. lower-+.f64 47.8%

         \[\leadsto \sqrt{y \cdot \left(x + \color{blue}{z}\right)} \cdot 2 \]

   6. Applied rewrites 47.8%

      \[\leadsto \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \cdot 2 \]

   7. Taylor expanded in x around inf

      \[\leadsto \sqrt{\color{blue}{x \cdot \left(y + z\right)}} \cdot 2 \]
   8. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto \sqrt{x \cdot \color{blue}{\left(y + z\right)}} \cdot 2 \]

      2. lower-+.f64 47.3%

         \[\leadsto \sqrt{x \cdot \left(y + \color{blue}{z}\right)} \cdot 2 \]

   9. Applied rewrites 47.3%

      \[\leadsto \sqrt{\color{blue}{x \cdot \left(y + z\right)}} \cdot 2 \]

   if -5.00000000000000018e-285 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in z around inf

      \[\leadsto 2 \cdot \sqrt{\color{blue}{z \cdot \left(x + y\right)}} \]
   3. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto 2 \cdot \sqrt{z \cdot \color{blue}{\left(x + y\right)}} \]

      2. lower-+.f64 47.2%

         \[\leadsto 2 \cdot \sqrt{z \cdot \left(x + \color{blue}{y}\right)} \]

   4. Applied rewrites 47.2%

      \[\leadsto 2 \cdot \sqrt{\color{blue}{z \cdot \left(x + y\right)}} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 9: 68.3% accurate, 0.5× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ 2 \cdot \sqrt{\mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_0\right) \cdot \left(\mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right) + \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_0\right)\right)} \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmax (fmin x y) z)))
   (*
    2.0
    (sqrt
     (*
      (fmin (fmax x y) t_0)
      (+ (fmin (fmin x y) z) (fmax (fmax x y) t_0)))))))

C

double code(double x, double y, double z) {
	double t_0 = fmax(fmin(x, y), z);
	return 2.0 * sqrt((fmin(fmax(x, y), t_0) * (fmin(fmin(x, y), z) + fmax(fmax(x, y), t_0))));
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    t_0 = fmax(fmin(x, y), z)
    code = 2.0d0 * sqrt((fmin(fmax(x, y), t_0) * (fmin(fmin(x, y), z) + fmax(fmax(x, y), t_0))))
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmax(fmin(x, y), z);
	return 2.0 * Math.sqrt((fmin(fmax(x, y), t_0) * (fmin(fmin(x, y), z) + fmax(fmax(x, y), t_0))));
}

Python

def code(x, y, z):
	t_0 = fmax(fmin(x, y), z)
	return 2.0 * math.sqrt((fmin(fmax(x, y), t_0) * (fmin(fmin(x, y), z) + fmax(fmax(x, y), t_0))))

Julia

function code(x, y, z)
	t_0 = fmax(fmin(x, y), z)
	return Float64(2.0 * sqrt(Float64(fmin(fmax(x, y), t_0) * Float64(fmin(fmin(x, y), z) + fmax(fmax(x, y), t_0)))))
end

MATLAB

function tmp = code(x, y, z)
	t_0 = max(min(x, y), z);
	tmp = 2.0 * sqrt((min(max(x, y), t_0) * (min(min(x, y), z) + max(max(x, y), t_0))));
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, N[(2.0 * N[Sqrt[N[(N[Min[N[Max[x, y], $MachinePrecision], t$95$0], $MachinePrecision] * N[(N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision] + N[Max[N[Max[x, y], $MachinePrecision], t$95$0], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]]

Derivation

1. Initial program 70.2%

   \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

2. Taylor expanded in y around inf

   \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
3. Step-by-step derivation

   1. lower-*.f64 N/A

      \[\leadsto 2 \cdot \sqrt{y \cdot \color{blue}{\left(x + z\right)}} \]

   2. lower-+.f64 47.8%

      \[\leadsto 2 \cdot \sqrt{y \cdot \left(x + \color{blue}{z}\right)} \]

4. Applied rewrites 47.8%

   \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot \left(x + z\right)}} \]
5. Add Preprocessing

Alternative 10: 68.3% accurate, 0.4× speedup

Math

\[\begin{array}{l} t_0 := \mathsf{max}\left(\mathsf{min}\left(x, y\right), z\right)\\ t_1 := \mathsf{min}\left(\mathsf{max}\left(x, y\right), t\_0\right)\\ \mathbf{if}\;t\_1 \leq -4 \cdot 10^{-310}:\\ \;\;\;\;2 \cdot \sqrt{\mathsf{min}\left(\mathsf{min}\left(x, y\right), z\right) \cdot t\_1}\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \sqrt{t\_1 \cdot \mathsf{max}\left(\mathsf{max}\left(x, y\right), t\_0\right)}\\ \end{array} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (fmax (fmin x y) z)) (t_1 (fmin (fmax x y) t_0)))
   (if (<= t_1 -4e-310)
     (* 2.0 (sqrt (* (fmin (fmin x y) z) t_1)))
     (* 2.0 (sqrt (* t_1 (fmax (fmax x y) t_0)))))))

C

double code(double x, double y, double z) {
	double t_0 = fmax(fmin(x, y), z);
	double t_1 = fmin(fmax(x, y), t_0);
	double tmp;
	if (t_1 <= -4e-310) {
		tmp = 2.0 * sqrt((fmin(fmin(x, y), z) * t_1));
	} else {
		tmp = 2.0 * sqrt((t_1 * fmax(fmax(x, y), t_0)));
	}
	return tmp;
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    real(8) :: t_0
    real(8) :: t_1
    real(8) :: tmp
    t_0 = fmax(fmin(x, y), z)
    t_1 = fmin(fmax(x, y), t_0)
    if (t_1 <= (-4d-310)) then
        tmp = 2.0d0 * sqrt((fmin(fmin(x, y), z) * t_1))
    else
        tmp = 2.0d0 * sqrt((t_1 * fmax(fmax(x, y), t_0)))
    end if
    code = tmp
end function

Java

public static double code(double x, double y, double z) {
	double t_0 = fmax(fmin(x, y), z);
	double t_1 = fmin(fmax(x, y), t_0);
	double tmp;
	if (t_1 <= -4e-310) {
		tmp = 2.0 * Math.sqrt((fmin(fmin(x, y), z) * t_1));
	} else {
		tmp = 2.0 * Math.sqrt((t_1 * fmax(fmax(x, y), t_0)));
	}
	return tmp;
}

Python

def code(x, y, z):
	t_0 = fmax(fmin(x, y), z)
	t_1 = fmin(fmax(x, y), t_0)
	tmp = 0
	if t_1 <= -4e-310:
		tmp = 2.0 * math.sqrt((fmin(fmin(x, y), z) * t_1))
	else:
		tmp = 2.0 * math.sqrt((t_1 * fmax(fmax(x, y), t_0)))
	return tmp

Julia

function code(x, y, z)
	t_0 = fmax(fmin(x, y), z)
	t_1 = fmin(fmax(x, y), t_0)
	tmp = 0.0
	if (t_1 <= -4e-310)
		tmp = Float64(2.0 * sqrt(Float64(fmin(fmin(x, y), z) * t_1)));
	else
		tmp = Float64(2.0 * sqrt(Float64(t_1 * fmax(fmax(x, y), t_0))));
	end
	return tmp
end

MATLAB

function tmp_2 = code(x, y, z)
	t_0 = max(min(x, y), z);
	t_1 = min(max(x, y), t_0);
	tmp = 0.0;
	if (t_1 <= -4e-310)
		tmp = 2.0 * sqrt((min(min(x, y), z) * t_1));
	else
		tmp = 2.0 * sqrt((t_1 * max(max(x, y), t_0)));
	end
	tmp_2 = tmp;
end

Wolfram

code[x_, y_, z_] := Block[{t$95$0 = N[Max[N[Min[x, y], $MachinePrecision], z], $MachinePrecision]}, Block[{t$95$1 = N[Min[N[Max[x, y], $MachinePrecision], t$95$0], $MachinePrecision]}, If[LessEqual[t$95$1, -4e-310], N[(2.0 * N[Sqrt[N[(N[Min[N[Min[x, y], $MachinePrecision], z], $MachinePrecision] * t$95$1), $MachinePrecision]], $MachinePrecision]), $MachinePrecision], N[(2.0 * N[Sqrt[N[(t$95$1 * N[Max[N[Max[x, y], $MachinePrecision], t$95$0], $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]]]]

Derivation

1. Split input into 2 regimes

2. if y < -3.999999999999988e-310

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in z around 0

      \[\leadsto 2 \cdot \color{blue}{\sqrt{x \cdot y}} \]
   3. Step-by-step derivation

      1. lower-sqrt.64 N/A

         \[\leadsto 2 \cdot \sqrt{x \cdot y} \]

      2. lower-*.f64 25.1%

         \[\leadsto 2 \cdot \sqrt{x \cdot y} \]

   4. Applied rewrites 25.1%

      \[\leadsto 2 \cdot \color{blue}{\sqrt{x \cdot y}} \]

   if -3.999999999999988e-310 < y

   1. Initial program 70.2%

      \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

   2. Taylor expanded in x around 0

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot z}} \]
   3. Step-by-step derivation

      1. lower-*.f64 25.0%

         \[\leadsto 2 \cdot \sqrt{y \cdot \color{blue}{z}} \]

   4. Applied rewrites 25.0%

      \[\leadsto 2 \cdot \sqrt{\color{blue}{y \cdot z}} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 11: 35.6% accurate, 1.1× speedup

Math

\[2 \cdot \sqrt{\mathsf{min}\left(x, z\right) \cdot \mathsf{min}\left(y, \mathsf{max}\left(x, z\right)\right)} \]

FPCore

(FPCore (x y z)
 :precision binary64
 (* 2.0 (sqrt (* (fmin x z) (fmin y (fmax x z))))))

C

double code(double x, double y, double z) {
	return 2.0 * sqrt((fmin(x, z) * fmin(y, fmax(x, z))));
}

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(x, y, z)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    code = 2.0d0 * sqrt((fmin(x, z) * fmin(y, fmax(x, z))))
end function

Java

public static double code(double x, double y, double z) {
	return 2.0 * Math.sqrt((fmin(x, z) * fmin(y, fmax(x, z))));
}

Python

def code(x, y, z):
	return 2.0 * math.sqrt((fmin(x, z) * fmin(y, fmax(x, z))))

Julia

function code(x, y, z)
	return Float64(2.0 * sqrt(Float64(fmin(x, z) * fmin(y, fmax(x, z)))))
end

MATLAB

function tmp = code(x, y, z)
	tmp = 2.0 * sqrt((min(x, z) * min(y, max(x, z))));
end

Wolfram

code[x_, y_, z_] := N[(2.0 * N[Sqrt[N[(N[Min[x, z], $MachinePrecision] * N[Min[y, N[Max[x, z], $MachinePrecision]], $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]

Derivation

1. Initial program 70.2%

   \[2 \cdot \sqrt{\left(x \cdot y + x \cdot z\right) + y \cdot z} \]

2. Taylor expanded in z around 0

   \[\leadsto 2 \cdot \color{blue}{\sqrt{x \cdot y}} \]
3. Step-by-step derivation

   1. lower-sqrt.64 N/A

      \[\leadsto 2 \cdot \sqrt{x \cdot y} \]

   2. lower-*.f64 25.1%

      \[\leadsto 2 \cdot \sqrt{x \cdot y} \]

4. Applied rewrites 25.1%

   \[\leadsto 2 \cdot \color{blue}{\sqrt{x \cdot y}} \]
5. Add Preprocessing

Reproduce

herbie shell --seed 2025183 
(FPCore (x y z)
  :name "Diagrams.TwoD.Apollonian:descartes from diagrams-contrib-1.3.0.5"
  :precision binary64
  (* 2.0 (sqrt (+ (+ (* x y) (* x z)) (* y z)))))