ENA, Section 1.4, Exercise 4b, n=5

Percentage Accurate: 88.3% → 99.4%

Time: 6.6s

Alternatives: 3

Speedup: N/A×

Specification

\[\left(-1000000000 \leq x \land x \leq 1000000000\right) \land \left(-1 \leq \varepsilon \land \varepsilon \leq 1\right)\]

Math

\[\begin{array}{l} \\ {\left(x + \varepsilon\right)}^{5} - {x}^{5} \end{array} \]

FPCore

(FPCore (x eps) :precision binary64 (- (pow (+ x eps) 5.0) (pow x 5.0)))

C

double code(double x, double eps) {
	return pow((x + eps), 5.0) - pow(x, 5.0);
}

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, eps)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: eps
    code = ((x + eps) ** 5.0d0) - (x ** 5.0d0)
end function

Java

public static double code(double x, double eps) {
	return Math.pow((x + eps), 5.0) - Math.pow(x, 5.0);
}

Python

def code(x, eps):
	return math.pow((x + eps), 5.0) - math.pow(x, 5.0)

Julia

function code(x, eps)
	return Float64((Float64(x + eps) ^ 5.0) - (x ^ 5.0))
end

MATLAB

function tmp = code(x, eps)
	tmp = ((x + eps) ^ 5.0) - (x ^ 5.0);
end

Wolfram

code[x_, eps_] := N[(N[Power[N[(x + eps), $MachinePrecision], 5.0], $MachinePrecision] - N[Power[x, 5.0], $MachinePrecision]), $MachinePrecision]

Initial Program: 88.3% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ {\left(x + \varepsilon\right)}^{5} - {x}^{5} \end{array} \]

FPCore

(FPCore (x eps) :precision binary64 (- (pow (+ x eps) 5.0) (pow x 5.0)))

C

double code(double x, double eps) {
	return pow((x + eps), 5.0) - pow(x, 5.0);
}

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, eps)
use fmin_fmax_functions
    real(8), intent (in) :: x
    real(8), intent (in) :: eps
    code = ((x + eps) ** 5.0d0) - (x ** 5.0d0)
end function

Java

public static double code(double x, double eps) {
	return Math.pow((x + eps), 5.0) - Math.pow(x, 5.0);
}

Python

def code(x, eps):
	return math.pow((x + eps), 5.0) - math.pow(x, 5.0)

Julia

function code(x, eps)
	return Float64((Float64(x + eps) ^ 5.0) - (x ^ 5.0))
end

MATLAB

function tmp = code(x, eps)
	tmp = ((x + eps) ^ 5.0) - (x ^ 5.0);
end

Wolfram

code[x_, eps_] := N[(N[Power[N[(x + eps), $MachinePrecision], 5.0], $MachinePrecision] - N[Power[x, 5.0], $MachinePrecision]), $MachinePrecision]

Alternative 1: 99.4% accurate, N/A× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := {\left(x + \varepsilon\right)}^{5} - {x}^{5}\\ \mathbf{if}\;t\_0 \leq -2 \cdot 10^{-308}:\\ \;\;\;\;t\_0\\ \mathbf{elif}\;t\_0 \leq 0:\\ \;\;\;\;\mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \mathsf{fma}\left(10, \varepsilon, x\right)\right)\right)\right) \cdot \varepsilon\\ \mathbf{else}:\\ \;\;\;\;t\_0\\ \end{array} \end{array} \]

FPCore

(FPCore (x eps)
 :precision binary64
 (let* ((t_0 (- (pow (+ x eps) 5.0) (pow x 5.0))))
   (if (<= t_0 -2e-308)
     t_0
     (if (<= t_0 0.0)
       (*
        (fma
         (pow x 4.0)
         4.0
         (*
          x
          (fma
           5.0
           (pow eps 3.0)
           (* x (fma 10.0 (* eps eps) (* x (fma 10.0 eps x)))))))
        eps)
       t_0))))

C

double code(double x, double eps) {
	double t_0 = pow((x + eps), 5.0) - pow(x, 5.0);
	double tmp;
	if (t_0 <= -2e-308) {
		tmp = t_0;
	} else if (t_0 <= 0.0) {
		tmp = fma(pow(x, 4.0), 4.0, (x * fma(5.0, pow(eps, 3.0), (x * fma(10.0, (eps * eps), (x * fma(10.0, eps, x))))))) * eps;
	} else {
		tmp = t_0;
	}
	return tmp;
}

Julia

function code(x, eps)
	t_0 = Float64((Float64(x + eps) ^ 5.0) - (x ^ 5.0))
	tmp = 0.0
	if (t_0 <= -2e-308)
		tmp = t_0;
	elseif (t_0 <= 0.0)
		tmp = Float64(fma((x ^ 4.0), 4.0, Float64(x * fma(5.0, (eps ^ 3.0), Float64(x * fma(10.0, Float64(eps * eps), Float64(x * fma(10.0, eps, x))))))) * eps);
	else
		tmp = t_0;
	end
	return tmp
end

Wolfram

code[x_, eps_] := Block[{t$95$0 = N[(N[Power[N[(x + eps), $MachinePrecision], 5.0], $MachinePrecision] - N[Power[x, 5.0], $MachinePrecision]), $MachinePrecision]}, If[LessEqual[t$95$0, -2e-308], t$95$0, If[LessEqual[t$95$0, 0.0], N[(N[(N[Power[x, 4.0], $MachinePrecision] * 4.0 + N[(x * N[(5.0 * N[Power[eps, 3.0], $MachinePrecision] + N[(x * N[(10.0 * N[(eps * eps), $MachinePrecision] + N[(x * N[(10.0 * eps + x), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * eps), $MachinePrecision], t$95$0]]]

Derivation

1. Split input into 2 regimes

2. if (-.f64 (pow.f64 (+.f64 x eps) #s(literal 5 binary64)) (pow.f64 x #s(literal 5 binary64))) < -1.9999999999999998e-308 or 0.0 < (-.f64 (pow.f64 (+.f64 x eps) #s(literal 5 binary64)) (pow.f64 x #s(literal 5 binary64)))

   1. Initial program 97.9%

      \[{\left(x + \varepsilon\right)}^{5} - {x}^{5} \]

   if -1.9999999999999998e-308 < (-.f64 (pow.f64 (+.f64 x eps) #s(literal 5 binary64)) (pow.f64 x #s(literal 5 binary64))) < 0.0

   1. Initial program 86.0%

      \[{\left(x + \varepsilon\right)}^{5} - {x}^{5} \]

   2. Taylor expanded in eps around 0

      \[\leadsto \color{blue}{\varepsilon \cdot \left(4 \cdot {x}^{4} + \left(\varepsilon \cdot \left(4 \cdot {x}^{3} + \left(\varepsilon \cdot \left(2 \cdot {x}^{2} + \left(8 \cdot {x}^{2} + \varepsilon \cdot \left(x + 4 \cdot x\right)\right)\right) + x \cdot \left(2 \cdot {x}^{2} + 4 \cdot {x}^{2}\right)\right)\right) + {x}^{4}\right)\right)} \]

   4. Applied rewrites 99.7%

      \[\leadsto \color{blue}{\mathsf{fma}\left({x}^{4}, 4, \mathsf{fma}\left(\mathsf{fma}\left(\left(x \cdot x\right) \cdot x, 4, \mathsf{fma}\left(\mathsf{fma}\left(x \cdot x, 10, \left(5 \cdot x\right) \cdot \varepsilon\right), \varepsilon, \left(\left(x \cdot x\right) \cdot 6\right) \cdot x\right)\right), \varepsilon, {x}^{4}\right)\right) \cdot \varepsilon} \]

   5. Taylor expanded in x around 0

      \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \left(5 \cdot {\varepsilon}^{3} + x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]
   6. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \left(5 \cdot {\varepsilon}^{3} + x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      2. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      3. lift-pow.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      4. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      5. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, {\varepsilon}^{2}, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      6. pow2 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      7. lift-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      8. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      9. +-commutative N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(10 \cdot \varepsilon + x\right)\right)\right)\right) \cdot \varepsilon \]

      10. lower-fma.f64 99.7

          \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \mathsf{fma}\left(10, \varepsilon, x\right)\right)\right)\right) \cdot \varepsilon \]

   7. Applied rewrites 99.7%

      \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \mathsf{fma}\left(10, \varepsilon, x\right)\right)\right)\right) \cdot \varepsilon \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 2: 98.7% accurate, N/A× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := {\left(x + \varepsilon\right)}^{5} - {x}^{5}\\ t_1 := \mathsf{fma}\left(\mathsf{fma}\left({\varepsilon}^{4}, 4, \mathsf{fma}\left(\mathsf{fma}\left({\varepsilon}^{3}, 4, \mathsf{fma}\left(\left(\varepsilon \cdot \varepsilon\right) \cdot 10, x, \left(\left(\varepsilon \cdot \varepsilon\right) \cdot 6\right) \cdot \varepsilon\right)\right), x, {\varepsilon}^{4}\right)\right), x, {\varepsilon}^{5}\right)\\ \mathbf{if}\;t\_0 \leq -2 \cdot 10^{-308}:\\ \;\;\;\;t\_1\\ \mathbf{elif}\;t\_0 \leq 0:\\ \;\;\;\;\mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \mathsf{fma}\left(10, \varepsilon, x\right)\right)\right)\right) \cdot \varepsilon\\ \mathbf{else}:\\ \;\;\;\;t\_1\\ \end{array} \end{array} \]

FPCore

(FPCore (x eps)
 :precision binary64
 (let* ((t_0 (- (pow (+ x eps) 5.0) (pow x 5.0)))
        (t_1
         (fma
          (fma
           (pow eps 4.0)
           4.0
           (fma
            (fma
             (pow eps 3.0)
             4.0
             (fma (* (* eps eps) 10.0) x (* (* (* eps eps) 6.0) eps)))
            x
            (pow eps 4.0)))
          x
          (pow eps 5.0))))
   (if (<= t_0 -2e-308)
     t_1
     (if (<= t_0 0.0)
       (*
        (fma
         (pow x 4.0)
         4.0
         (*
          x
          (fma
           5.0
           (pow eps 3.0)
           (* x (fma 10.0 (* eps eps) (* x (fma 10.0 eps x)))))))
        eps)
       t_1))))

C

double code(double x, double eps) {
	double t_0 = pow((x + eps), 5.0) - pow(x, 5.0);
	double t_1 = fma(fma(pow(eps, 4.0), 4.0, fma(fma(pow(eps, 3.0), 4.0, fma(((eps * eps) * 10.0), x, (((eps * eps) * 6.0) * eps))), x, pow(eps, 4.0))), x, pow(eps, 5.0));
	double tmp;
	if (t_0 <= -2e-308) {
		tmp = t_1;
	} else if (t_0 <= 0.0) {
		tmp = fma(pow(x, 4.0), 4.0, (x * fma(5.0, pow(eps, 3.0), (x * fma(10.0, (eps * eps), (x * fma(10.0, eps, x))))))) * eps;
	} else {
		tmp = t_1;
	}
	return tmp;
}

Julia

function code(x, eps)
	t_0 = Float64((Float64(x + eps) ^ 5.0) - (x ^ 5.0))
	t_1 = fma(fma((eps ^ 4.0), 4.0, fma(fma((eps ^ 3.0), 4.0, fma(Float64(Float64(eps * eps) * 10.0), x, Float64(Float64(Float64(eps * eps) * 6.0) * eps))), x, (eps ^ 4.0))), x, (eps ^ 5.0))
	tmp = 0.0
	if (t_0 <= -2e-308)
		tmp = t_1;
	elseif (t_0 <= 0.0)
		tmp = Float64(fma((x ^ 4.0), 4.0, Float64(x * fma(5.0, (eps ^ 3.0), Float64(x * fma(10.0, Float64(eps * eps), Float64(x * fma(10.0, eps, x))))))) * eps);
	else
		tmp = t_1;
	end
	return tmp
end

Wolfram

code[x_, eps_] := Block[{t$95$0 = N[(N[Power[N[(x + eps), $MachinePrecision], 5.0], $MachinePrecision] - N[Power[x, 5.0], $MachinePrecision]), $MachinePrecision]}, Block[{t$95$1 = N[(N[(N[Power[eps, 4.0], $MachinePrecision] * 4.0 + N[(N[(N[Power[eps, 3.0], $MachinePrecision] * 4.0 + N[(N[(N[(eps * eps), $MachinePrecision] * 10.0), $MachinePrecision] * x + N[(N[(N[(eps * eps), $MachinePrecision] * 6.0), $MachinePrecision] * eps), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * x + N[Power[eps, 4.0], $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * x + N[Power[eps, 5.0], $MachinePrecision]), $MachinePrecision]}, If[LessEqual[t$95$0, -2e-308], t$95$1, If[LessEqual[t$95$0, 0.0], N[(N[(N[Power[x, 4.0], $MachinePrecision] * 4.0 + N[(x * N[(5.0 * N[Power[eps, 3.0], $MachinePrecision] + N[(x * N[(10.0 * N[(eps * eps), $MachinePrecision] + N[(x * N[(10.0 * eps + x), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * eps), $MachinePrecision], t$95$1]]]]

Derivation

1. Split input into 2 regimes

2. if (-.f64 (pow.f64 (+.f64 x eps) #s(literal 5 binary64)) (pow.f64 x #s(literal 5 binary64))) < -1.9999999999999998e-308 or 0.0 < (-.f64 (pow.f64 (+.f64 x eps) #s(literal 5 binary64)) (pow.f64 x #s(literal 5 binary64)))

   1. Initial program 97.9%

      \[{\left(x + \varepsilon\right)}^{5} - {x}^{5} \]

   2. Taylor expanded in x around 0

      \[\leadsto \color{blue}{x \cdot \left(4 \cdot {\varepsilon}^{4} + \left(x \cdot \left(4 \cdot {\varepsilon}^{3} + \left(\varepsilon \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(2 \cdot {\varepsilon}^{2} + 8 \cdot {\varepsilon}^{2}\right)\right)\right) + {\varepsilon}^{4}\right)\right) + {\varepsilon}^{5}} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(4 \cdot {\varepsilon}^{4} + \left(x \cdot \left(4 \cdot {\varepsilon}^{3} + \left(\varepsilon \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(2 \cdot {\varepsilon}^{2} + 8 \cdot {\varepsilon}^{2}\right)\right)\right) + {\varepsilon}^{4}\right)\right) \cdot x + {\color{blue}{\varepsilon}}^{5} \]

      2. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left(4 \cdot {\varepsilon}^{4} + \left(x \cdot \left(4 \cdot {\varepsilon}^{3} + \left(\varepsilon \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(2 \cdot {\varepsilon}^{2} + 8 \cdot {\varepsilon}^{2}\right)\right)\right) + {\varepsilon}^{4}\right), \color{blue}{x}, {\varepsilon}^{5}\right) \]

   4. Applied rewrites 94.3%

      \[\leadsto \color{blue}{\mathsf{fma}\left(\mathsf{fma}\left({\varepsilon}^{4}, 4, \mathsf{fma}\left(\mathsf{fma}\left({\varepsilon}^{3}, 4, \mathsf{fma}\left(\left(\varepsilon \cdot \varepsilon\right) \cdot 10, x, \left(\left(\varepsilon \cdot \varepsilon\right) \cdot 6\right) \cdot \varepsilon\right)\right), x, {\varepsilon}^{4}\right)\right), x, {\varepsilon}^{5}\right)} \]

   if -1.9999999999999998e-308 < (-.f64 (pow.f64 (+.f64 x eps) #s(literal 5 binary64)) (pow.f64 x #s(literal 5 binary64))) < 0.0

   1. Initial program 86.0%

      \[{\left(x + \varepsilon\right)}^{5} - {x}^{5} \]

   2. Taylor expanded in eps around 0

      \[\leadsto \color{blue}{\varepsilon \cdot \left(4 \cdot {x}^{4} + \left(\varepsilon \cdot \left(4 \cdot {x}^{3} + \left(\varepsilon \cdot \left(2 \cdot {x}^{2} + \left(8 \cdot {x}^{2} + \varepsilon \cdot \left(x + 4 \cdot x\right)\right)\right) + x \cdot \left(2 \cdot {x}^{2} + 4 \cdot {x}^{2}\right)\right)\right) + {x}^{4}\right)\right)} \]

   4. Applied rewrites 99.7%

      \[\leadsto \color{blue}{\mathsf{fma}\left({x}^{4}, 4, \mathsf{fma}\left(\mathsf{fma}\left(\left(x \cdot x\right) \cdot x, 4, \mathsf{fma}\left(\mathsf{fma}\left(x \cdot x, 10, \left(5 \cdot x\right) \cdot \varepsilon\right), \varepsilon, \left(\left(x \cdot x\right) \cdot 6\right) \cdot x\right)\right), \varepsilon, {x}^{4}\right)\right) \cdot \varepsilon} \]

   5. Taylor expanded in x around 0

      \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \left(5 \cdot {\varepsilon}^{3} + x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]
   6. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \left(5 \cdot {\varepsilon}^{3} + x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      2. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      3. lift-pow.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      4. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \left(10 \cdot {\varepsilon}^{2} + x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      5. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, {\varepsilon}^{2}, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      6. pow2 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      7. lift-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      8. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(x + 10 \cdot \varepsilon\right)\right)\right)\right) \cdot \varepsilon \]

      9. +-commutative N/A

         \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \left(10 \cdot \varepsilon + x\right)\right)\right)\right) \cdot \varepsilon \]

      10. lower-fma.f64 99.7

          \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \mathsf{fma}\left(10, \varepsilon, x\right)\right)\right)\right) \cdot \varepsilon \]

   7. Applied rewrites 99.7%

      \[\leadsto \mathsf{fma}\left({x}^{4}, 4, x \cdot \mathsf{fma}\left(5, {\varepsilon}^{3}, x \cdot \mathsf{fma}\left(10, \varepsilon \cdot \varepsilon, x \cdot \mathsf{fma}\left(10, \varepsilon, x\right)\right)\right)\right) \cdot \varepsilon \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 3: 82.8% accurate, N/A× speedup

Math

\[\begin{array}{l} \\ \left(x \cdot x\right) \cdot \mathsf{fma}\left(4, {\varepsilon}^{3}, \mathsf{fma}\left(x, \mathsf{fma}\left(-1, \left(\varepsilon \cdot \varepsilon\right) \cdot -10, x \cdot \left(5 \cdot \varepsilon\right)\right), 6 \cdot {\varepsilon}^{3}\right)\right) \end{array} \]

FPCore

(FPCore (x eps)
 :precision binary64
 (*
  (* x x)
  (fma
   4.0
   (pow eps 3.0)
   (fma
    x
    (fma -1.0 (* (* eps eps) -10.0) (* x (* 5.0 eps)))
    (* 6.0 (pow eps 3.0))))))

C

double code(double x, double eps) {
	return (x * x) * fma(4.0, pow(eps, 3.0), fma(x, fma(-1.0, ((eps * eps) * -10.0), (x * (5.0 * eps))), (6.0 * pow(eps, 3.0))));
}

Julia

function code(x, eps)
	return Float64(Float64(x * x) * fma(4.0, (eps ^ 3.0), fma(x, fma(-1.0, Float64(Float64(eps * eps) * -10.0), Float64(x * Float64(5.0 * eps))), Float64(6.0 * (eps ^ 3.0)))))
end

Wolfram

code[x_, eps_] := N[(N[(x * x), $MachinePrecision] * N[(4.0 * N[Power[eps, 3.0], $MachinePrecision] + N[(x * N[(-1.0 * N[(N[(eps * eps), $MachinePrecision] * -10.0), $MachinePrecision] + N[(x * N[(5.0 * eps), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] + N[(6.0 * N[Power[eps, 3.0], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Derivation

1. Initial program 88.3%

   \[{\left(x + \varepsilon\right)}^{5} - {x}^{5} \]

2. Taylor expanded in x around -inf

   \[\leadsto \color{blue}{{x}^{4} \cdot \left(\varepsilon + \left(-1 \cdot \frac{-4 \cdot {\varepsilon}^{2} + \left(-1 \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right) + -1 \cdot \frac{4 \cdot {\varepsilon}^{3} + \varepsilon \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right)}{x}\right)}{x} + 4 \cdot \varepsilon\right)\right)} \]
3. Step-by-step derivation

   1. *-commutative N/A

      \[\leadsto \left(\varepsilon + \left(-1 \cdot \frac{-4 \cdot {\varepsilon}^{2} + \left(-1 \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right) + -1 \cdot \frac{4 \cdot {\varepsilon}^{3} + \varepsilon \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right)}{x}\right)}{x} + 4 \cdot \varepsilon\right)\right) \cdot \color{blue}{{x}^{4}} \]

   2. lower-*.f64 N/A

      \[\leadsto \left(\varepsilon + \left(-1 \cdot \frac{-4 \cdot {\varepsilon}^{2} + \left(-1 \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right) + -1 \cdot \frac{4 \cdot {\varepsilon}^{3} + \varepsilon \cdot \left(2 \cdot {\varepsilon}^{2} + 4 \cdot {\varepsilon}^{2}\right)}{x}\right)}{x} + 4 \cdot \varepsilon\right)\right) \cdot \color{blue}{{x}^{4}} \]

4. Applied rewrites 81.2%

   \[\leadsto \color{blue}{\left(\mathsf{fma}\left(\frac{\mathsf{fma}\left(-4, \varepsilon \cdot \varepsilon, \mathsf{fma}\left(\frac{\mathsf{fma}\left(\left(\varepsilon \cdot \varepsilon\right) \cdot 6, \varepsilon, {\varepsilon}^{3} \cdot 4\right)}{x}, -1, -1 \cdot \left(\left(\varepsilon \cdot \varepsilon\right) \cdot 6\right)\right)\right)}{x}, -1, 4 \cdot \varepsilon\right) + \varepsilon\right) \cdot {x}^{4}} \]

5. Taylor expanded in x around 0

   \[\leadsto {x}^{2} \cdot \color{blue}{\left(4 \cdot {\varepsilon}^{3} + \left(6 \cdot {\varepsilon}^{3} + x \cdot \left(-1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right)\right)\right)\right)} \]
6. Step-by-step derivation

   1. lower-*.f64 N/A

      \[\leadsto {x}^{2} \cdot \left(4 \cdot {\varepsilon}^{3} + \color{blue}{\left(6 \cdot {\varepsilon}^{3} + x \cdot \left(-1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right)\right)\right)}\right) \]

   2. pow2 N/A

      \[\leadsto \left(x \cdot x\right) \cdot \left(4 \cdot {\varepsilon}^{3} + \left(\color{blue}{6 \cdot {\varepsilon}^{3}} + x \cdot \left(-1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right)\right)\right)\right) \]

   3. lift-*.f64 N/A

      \[\leadsto \left(x \cdot x\right) \cdot \left(4 \cdot {\varepsilon}^{3} + \left(\color{blue}{6 \cdot {\varepsilon}^{3}} + x \cdot \left(-1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right)\right)\right)\right) \]

   4. lower-fma.f64 N/A

      \[\leadsto \left(x \cdot x\right) \cdot \mathsf{fma}\left(4, {\varepsilon}^{\color{blue}{3}}, 6 \cdot {\varepsilon}^{3} + x \cdot \left(-1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right)\right)\right) \]

   5. lift-pow.f64 N/A

      \[\leadsto \left(x \cdot x\right) \cdot \mathsf{fma}\left(4, {\varepsilon}^{3}, 6 \cdot {\varepsilon}^{3} + x \cdot \left(-1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right)\right)\right) \]

   6. +-commutative N/A

      \[\leadsto \left(x \cdot x\right) \cdot \mathsf{fma}\left(4, {\varepsilon}^{3}, x \cdot \left(-1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right)\right) + 6 \cdot {\varepsilon}^{3}\right) \]

   7. lower-fma.f64 N/A

      \[\leadsto \left(x \cdot x\right) \cdot \mathsf{fma}\left(4, {\varepsilon}^{3}, \mathsf{fma}\left(x, -1 \cdot \left(-6 \cdot {\varepsilon}^{2} + -4 \cdot {\varepsilon}^{2}\right) + x \cdot \left(\varepsilon + 4 \cdot \varepsilon\right), 6 \cdot {\varepsilon}^{3}\right)\right) \]

7. Applied rewrites 82.8%

   \[\leadsto \left(x \cdot x\right) \cdot \color{blue}{\mathsf{fma}\left(4, {\varepsilon}^{3}, \mathsf{fma}\left(x, \mathsf{fma}\left(-1, \left(\varepsilon \cdot \varepsilon\right) \cdot -10, x \cdot \left(5 \cdot \varepsilon\right)\right), 6 \cdot {\varepsilon}^{3}\right)\right)} \]
8. Add Preprocessing

Reproduce

herbie shell --seed 2025093 
(FPCore (x eps)
  :name "ENA, Section 1.4, Exercise 4b, n=5"
  :precision binary64
  :pre (and (and (<= -1000000000.0 x) (<= x 1000000000.0)) (and (<= -1.0 eps) (<= eps 1.0)))
  (- (pow (+ x eps) 5.0) (pow x 5.0)))