tanhf (example 3.4)

Percentage Accurate: 53.8% → 100.0%

Time: 8.2s

Alternatives: 5

Speedup: 2.0×

Specification

Math

\[\begin{array}{l} \\ \frac{1 - \cos x}{\sin x} \end{array} \]

FPCore

(FPCore (x) :precision binary64 (/ (- 1.0 (cos x)) (sin x)))

C

double code(double x) {
	return (1.0 - cos(x)) / sin(x);
}

Fortran

real(8) function code(x)
    real(8), intent (in) :: x
    code = (1.0d0 - cos(x)) / sin(x)
end function

Java

public static double code(double x) {
	return (1.0 - Math.cos(x)) / Math.sin(x);
}

Python

def code(x):
	return (1.0 - math.cos(x)) / math.sin(x)

Julia

function code(x)
	return Float64(Float64(1.0 - cos(x)) / sin(x))
end

MATLAB

function tmp = code(x)
	tmp = (1.0 - cos(x)) / sin(x);
end

Wolfram

code[x_] := N[(N[(1.0 - N[Cos[x], $MachinePrecision]), $MachinePrecision] / N[Sin[x], $MachinePrecision]), $MachinePrecision]

Initial Program: 53.8% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \frac{1 - \cos x}{\sin x} \end{array} \]

FPCore

(FPCore (x) :precision binary64 (/ (- 1.0 (cos x)) (sin x)))

C

double code(double x) {
	return (1.0 - cos(x)) / sin(x);
}

Fortran

real(8) function code(x)
    real(8), intent (in) :: x
    code = (1.0d0 - cos(x)) / sin(x)
end function

Java

public static double code(double x) {
	return (1.0 - Math.cos(x)) / Math.sin(x);
}

Python

def code(x):
	return (1.0 - math.cos(x)) / math.sin(x)

Julia

function code(x)
	return Float64(Float64(1.0 - cos(x)) / sin(x))
end

MATLAB

function tmp = code(x)
	tmp = (1.0 - cos(x)) / sin(x);
end

Wolfram

code[x_] := N[(N[(1.0 - N[Cos[x], $MachinePrecision]), $MachinePrecision] / N[Sin[x], $MachinePrecision]), $MachinePrecision]

Alternative 1: 100.0% accurate, 2.0× speedup

Math

\[\begin{array}{l} \\ \tan \left(x \cdot 0.5\right) \end{array} \]

FPCore

(FPCore (x) :precision binary64 (tan (* x 0.5)))

C

double code(double x) {
	return tan((x * 0.5));
}

Fortran

real(8) function code(x)
    real(8), intent (in) :: x
    code = tan((x * 0.5d0))
end function

Java

public static double code(double x) {
	return Math.tan((x * 0.5));
}

Python

def code(x):
	return math.tan((x * 0.5))

Julia

function code(x)
	return tan(Float64(x * 0.5))
end

MATLAB

function tmp = code(x)
	tmp = tan((x * 0.5));
end

Wolfram

code[x_] := N[Tan[N[(x * 0.5), $MachinePrecision]], $MachinePrecision]

Derivation

1. Initial program 51.7%

   \[\frac{1 - \cos x}{\sin x} \]
2. Add Preprocessing
3. Step-by-step derivation

   1. hang-p0-tan N/A

      \[\leadsto \color{blue}{\tan \left(\frac{x}{2}\right)} \]

   2. lower-tan.f64 N/A

      \[\leadsto \color{blue}{\tan \left(\frac{x}{2}\right)} \]

   3. div-inv N/A

      \[\leadsto \tan \color{blue}{\left(x \cdot \frac{1}{2}\right)} \]

   4. metadata-eval N/A

      \[\leadsto \tan \left(x \cdot \color{blue}{\frac{1}{2}}\right) \]

   5. lower-*.f64 100.0

      \[\leadsto \tan \color{blue}{\left(x \cdot 0.5\right)} \]

4. Applied egg-rr 100.0%

   \[\leadsto \color{blue}{\tan \left(x \cdot 0.5\right)} \]
5. Add Preprocessing

Alternative 2: 53.8% accurate, 0.7× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;\frac{1 - \cos x}{\sin x} \leq 10^{-5}:\\ \;\;\;\;x \cdot \mathsf{fma}\left(-0.006944444444444444 \cdot \left(x \cdot \left(x \cdot x\right)\right), x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right), \frac{-0.25}{\mathsf{fma}\left(x, x \cdot 0.041666666666666664, -0.5\right)}\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (x)
 :precision binary64
 (if (<= (/ (- 1.0 (cos x)) (sin x)) 1e-5)
   (*
    x
    (fma
     (* -0.006944444444444444 (* x (* x x)))
     (* x (fma x (* x 0.041666666666666664) 0.5))
     (/ (- 0.25) (fma x (* x 0.041666666666666664) -0.5))))
   1.0))

C

double code(double x) {
	double tmp;
	if (((1.0 - cos(x)) / sin(x)) <= 1e-5) {
		tmp = x * fma((-0.006944444444444444 * (x * (x * x))), (x * fma(x, (x * 0.041666666666666664), 0.5)), (-0.25 / fma(x, (x * 0.041666666666666664), -0.5)));
	} else {
		tmp = 1.0;
	}
	return tmp;
}

Julia

function code(x)
	tmp = 0.0
	if (Float64(Float64(1.0 - cos(x)) / sin(x)) <= 1e-5)
		tmp = Float64(x * fma(Float64(-0.006944444444444444 * Float64(x * Float64(x * x))), Float64(x * fma(x, Float64(x * 0.041666666666666664), 0.5)), Float64(Float64(-0.25) / fma(x, Float64(x * 0.041666666666666664), -0.5))));
	else
		tmp = 1.0;
	end
	return tmp
end

Wolfram

code[x_] := If[LessEqual[N[(N[(1.0 - N[Cos[x], $MachinePrecision]), $MachinePrecision] / N[Sin[x], $MachinePrecision]), $MachinePrecision], 1e-5], N[(x * N[(N[(-0.006944444444444444 * N[(x * N[(x * x), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * N[(x * N[(x * N[(x * 0.041666666666666664), $MachinePrecision] + 0.5), $MachinePrecision]), $MachinePrecision] + N[((-0.25) / N[(x * N[(x * 0.041666666666666664), $MachinePrecision] + -0.5), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], 1.0]

Derivation

1. Split input into 2 regimes

2. if (/.f64 (-.f64 #s(literal 1 binary64) (cos.f64 x)) (sin.f64 x)) < 1.00000000000000008e-5

   1. Initial program 35.0%

      \[\frac{1 - \cos x}{\sin x} \]
   2. Add Preprocessing

   3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{x \cdot \left(\frac{1}{2} + \frac{1}{24} \cdot {x}^{2}\right)} \]
   4. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto \color{blue}{x \cdot \left(\frac{1}{2} + \frac{1}{24} \cdot {x}^{2}\right)} \]

      2. +-commutative N/A

         \[\leadsto x \cdot \color{blue}{\left(\frac{1}{24} \cdot {x}^{2} + \frac{1}{2}\right)} \]

      3. unpow2 N/A

         \[\leadsto x \cdot \left(\frac{1}{24} \cdot \color{blue}{\left(x \cdot x\right)} + \frac{1}{2}\right) \]

      4. associate-*r* N/A

         \[\leadsto x \cdot \left(\color{blue}{\left(\frac{1}{24} \cdot x\right) \cdot x} + \frac{1}{2}\right) \]

      5. *-commutative N/A

         \[\leadsto x \cdot \left(\color{blue}{x \cdot \left(\frac{1}{24} \cdot x\right)} + \frac{1}{2}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto x \cdot \color{blue}{\mathsf{fma}\left(x, \frac{1}{24} \cdot x, \frac{1}{2}\right)} \]

      7. *-commutative N/A

         \[\leadsto x \cdot \mathsf{fma}\left(x, \color{blue}{x \cdot \frac{1}{24}}, \frac{1}{2}\right) \]

      8. lower-*.f64 69.4

         \[\leadsto x \cdot \mathsf{fma}\left(x, \color{blue}{x \cdot 0.041666666666666664}, 0.5\right) \]

   5. Simplified 69.4%

      \[\leadsto \color{blue}{x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right)} \]

   7. Applied egg-rr 69.0%

      \[\leadsto x \cdot \color{blue}{\mathsf{fma}\left(\frac{x \cdot \left(\left(x \cdot x\right) \cdot 0.001736111111111111\right)}{\mathsf{fma}\left(x \cdot x, \left(x \cdot x\right) \cdot 0.001736111111111111, -0.25\right)}, x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right), -\frac{0.25}{\mathsf{fma}\left(x, x \cdot 0.041666666666666664, -0.5\right)}\right)} \]

   8. Taylor expanded in x around 0

      \[\leadsto x \cdot \mathsf{fma}\left(\color{blue}{\frac{-1}{144} \cdot {x}^{3}}, x \cdot \mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{1}{2}\right), \mathsf{neg}\left(\frac{\frac{1}{4}}{\mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{-1}{2}\right)}\right)\right) \]
   9. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto x \cdot \mathsf{fma}\left(\color{blue}{\frac{-1}{144} \cdot {x}^{3}}, x \cdot \mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{1}{2}\right), \mathsf{neg}\left(\frac{\frac{1}{4}}{\mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{-1}{2}\right)}\right)\right) \]

      2. cube-mult N/A

         \[\leadsto x \cdot \mathsf{fma}\left(\frac{-1}{144} \cdot \color{blue}{\left(x \cdot \left(x \cdot x\right)\right)}, x \cdot \mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{1}{2}\right), \mathsf{neg}\left(\frac{\frac{1}{4}}{\mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{-1}{2}\right)}\right)\right) \]

      3. unpow2 N/A

         \[\leadsto x \cdot \mathsf{fma}\left(\frac{-1}{144} \cdot \left(x \cdot \color{blue}{{x}^{2}}\right), x \cdot \mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{1}{2}\right), \mathsf{neg}\left(\frac{\frac{1}{4}}{\mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{-1}{2}\right)}\right)\right) \]

      4. lower-*.f64 N/A

         \[\leadsto x \cdot \mathsf{fma}\left(\frac{-1}{144} \cdot \color{blue}{\left(x \cdot {x}^{2}\right)}, x \cdot \mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{1}{2}\right), \mathsf{neg}\left(\frac{\frac{1}{4}}{\mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{-1}{2}\right)}\right)\right) \]

      5. unpow2 N/A

         \[\leadsto x \cdot \mathsf{fma}\left(\frac{-1}{144} \cdot \left(x \cdot \color{blue}{\left(x \cdot x\right)}\right), x \cdot \mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{1}{2}\right), \mathsf{neg}\left(\frac{\frac{1}{4}}{\mathsf{fma}\left(x, x \cdot \frac{1}{24}, \frac{-1}{2}\right)}\right)\right) \]

      6. lower-*.f64 69.4

         \[\leadsto x \cdot \mathsf{fma}\left(-0.006944444444444444 \cdot \left(x \cdot \color{blue}{\left(x \cdot x\right)}\right), x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right), -\frac{0.25}{\mathsf{fma}\left(x, x \cdot 0.041666666666666664, -0.5\right)}\right) \]

   10. Simplified 69.4%

       \[\leadsto x \cdot \mathsf{fma}\left(\color{blue}{-0.006944444444444444 \cdot \left(x \cdot \left(x \cdot x\right)\right)}, x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right), -\frac{0.25}{\mathsf{fma}\left(x, x \cdot 0.041666666666666664, -0.5\right)}\right) \]

   if 1.00000000000000008e-5 < (/.f64 (-.f64 #s(literal 1 binary64) (cos.f64 x)) (sin.f64 x))

   1. Initial program 98.9%

      \[\frac{1 - \cos x}{\sin x} \]
   2. Add Preprocessing

   4. Applied egg-rr 18.8%

      \[\leadsto \color{blue}{{1}^{-0.5}} \]
   5. Step-by-step derivation

      1. pow-base-1 18.8

         \[\leadsto \color{blue}{1} \]

   6. Applied egg-rr 18.8%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.

4. Final simplification 56.2%

   \[\leadsto \begin{array}{l} \mathbf{if}\;\frac{1 - \cos x}{\sin x} \leq 10^{-5}:\\ \;\;\;\;x \cdot \mathsf{fma}\left(-0.006944444444444444 \cdot \left(x \cdot \left(x \cdot x\right)\right), x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right), \frac{-0.25}{\mathsf{fma}\left(x, x \cdot 0.041666666666666664, -0.5\right)}\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \]
5. Add Preprocessing

Alternative 3: 51.8% accurate, 9.3× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;x \leq 3.1:\\ \;\;\;\;x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (x)
 :precision binary64
 (if (<= x 3.1) (* x (fma x (* x 0.041666666666666664) 0.5)) 1.0))

C

double code(double x) {
	double tmp;
	if (x <= 3.1) {
		tmp = x * fma(x, (x * 0.041666666666666664), 0.5);
	} else {
		tmp = 1.0;
	}
	return tmp;
}

Julia

function code(x)
	tmp = 0.0
	if (x <= 3.1)
		tmp = Float64(x * fma(x, Float64(x * 0.041666666666666664), 0.5));
	else
		tmp = 1.0;
	end
	return tmp
end

Wolfram

code[x_] := If[LessEqual[x, 3.1], N[(x * N[(x * N[(x * 0.041666666666666664), $MachinePrecision] + 0.5), $MachinePrecision]), $MachinePrecision], 1.0]

Derivation

1. Split input into 2 regimes

2. if x < 3.10000000000000009

   1. Initial program 34.2%

      \[\frac{1 - \cos x}{\sin x} \]
   2. Add Preprocessing

   3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{x \cdot \left(\frac{1}{2} + \frac{1}{24} \cdot {x}^{2}\right)} \]
   4. Step-by-step derivation

      1. lower-*.f64 N/A

         \[\leadsto \color{blue}{x \cdot \left(\frac{1}{2} + \frac{1}{24} \cdot {x}^{2}\right)} \]

      2. +-commutative N/A

         \[\leadsto x \cdot \color{blue}{\left(\frac{1}{24} \cdot {x}^{2} + \frac{1}{2}\right)} \]

      3. unpow2 N/A

         \[\leadsto x \cdot \left(\frac{1}{24} \cdot \color{blue}{\left(x \cdot x\right)} + \frac{1}{2}\right) \]

      4. associate-*r* N/A

         \[\leadsto x \cdot \left(\color{blue}{\left(\frac{1}{24} \cdot x\right) \cdot x} + \frac{1}{2}\right) \]

      5. *-commutative N/A

         \[\leadsto x \cdot \left(\color{blue}{x \cdot \left(\frac{1}{24} \cdot x\right)} + \frac{1}{2}\right) \]

      6. lower-fma.f64 N/A

         \[\leadsto x \cdot \color{blue}{\mathsf{fma}\left(x, \frac{1}{24} \cdot x, \frac{1}{2}\right)} \]

      7. *-commutative N/A

         \[\leadsto x \cdot \mathsf{fma}\left(x, \color{blue}{x \cdot \frac{1}{24}}, \frac{1}{2}\right) \]

      8. lower-*.f64 70.2

         \[\leadsto x \cdot \mathsf{fma}\left(x, \color{blue}{x \cdot 0.041666666666666664}, 0.5\right) \]

   5. Simplified 70.2%

      \[\leadsto \color{blue}{x \cdot \mathsf{fma}\left(x, x \cdot 0.041666666666666664, 0.5\right)} \]

   if 3.10000000000000009 < x

   1. Initial program 99.1%

      \[\frac{1 - \cos x}{\sin x} \]
   2. Add Preprocessing

   4. Applied egg-rr 11.2%

      \[\leadsto \color{blue}{{1}^{-0.5}} \]
   5. Step-by-step derivation

      1. pow-base-1 11.2

         \[\leadsto \color{blue}{1} \]

   6. Applied egg-rr 11.2%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 4: 51.8% accurate, 17.9× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;x \leq 3.1:\\ \;\;\;\;x \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (x) :precision binary64 (if (<= x 3.1) (* x 0.5) 1.0))

C

double code(double x) {
	double tmp;
	if (x <= 3.1) {
		tmp = x * 0.5;
	} else {
		tmp = 1.0;
	}
	return tmp;
}

Fortran

real(8) function code(x)
    real(8), intent (in) :: x
    real(8) :: tmp
    if (x <= 3.1d0) then
        tmp = x * 0.5d0
    else
        tmp = 1.0d0
    end if
    code = tmp
end function

Java

public static double code(double x) {
	double tmp;
	if (x <= 3.1) {
		tmp = x * 0.5;
	} else {
		tmp = 1.0;
	}
	return tmp;
}

Python

def code(x):
	tmp = 0
	if x <= 3.1:
		tmp = x * 0.5
	else:
		tmp = 1.0
	return tmp

Julia

function code(x)
	tmp = 0.0
	if (x <= 3.1)
		tmp = Float64(x * 0.5);
	else
		tmp = 1.0;
	end
	return tmp
end

MATLAB

function tmp_2 = code(x)
	tmp = 0.0;
	if (x <= 3.1)
		tmp = x * 0.5;
	else
		tmp = 1.0;
	end
	tmp_2 = tmp;
end

Wolfram

code[x_] := If[LessEqual[x, 3.1], N[(x * 0.5), $MachinePrecision], 1.0]

Derivation

1. Split input into 2 regimes

2. if x < 3.10000000000000009

   1. Initial program 34.2%

      \[\frac{1 - \cos x}{\sin x} \]
   2. Add Preprocessing

   3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot x} \]
   4. Step-by-step derivation

      1. lower-*.f64 70.3

         \[\leadsto \color{blue}{0.5 \cdot x} \]

   5. Simplified 70.3%

      \[\leadsto \color{blue}{0.5 \cdot x} \]

   if 3.10000000000000009 < x

   1. Initial program 99.1%

      \[\frac{1 - \cos x}{\sin x} \]
   2. Add Preprocessing

   4. Applied egg-rr 11.2%

      \[\leadsto \color{blue}{{1}^{-0.5}} \]
   5. Step-by-step derivation

      1. pow-base-1 11.2

         \[\leadsto \color{blue}{1} \]

   6. Applied egg-rr 11.2%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.

4. Final simplification 54.3%

   \[\leadsto \begin{array}{l} \mathbf{if}\;x \leq 3.1:\\ \;\;\;\;x \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \]
5. Add Preprocessing

Alternative 5: 7.1% accurate, 215.0× speedup

Math

\[\begin{array}{l} \\ 1 \end{array} \]

FPCore

(FPCore (x) :precision binary64 1.0)

C

double code(double x) {
	return 1.0;
}

Fortran

real(8) function code(x)
    real(8), intent (in) :: x
    code = 1.0d0
end function

Java

public static double code(double x) {
	return 1.0;
}

Python

def code(x):
	return 1.0

Julia

function code(x)
	return 1.0
end

MATLAB

function tmp = code(x)
	tmp = 1.0;
end

Wolfram

code[x_] := 1.0

Derivation

1. Initial program 51.7%

   \[\frac{1 - \cos x}{\sin x} \]
2. Add Preprocessing

4. Applied egg-rr 7.1%

   \[\leadsto \color{blue}{{1}^{-0.5}} \]
5. Step-by-step derivation

   1. pow-base-1 7.1

      \[\leadsto \color{blue}{1} \]

6. Applied egg-rr 7.1%

   \[\leadsto \color{blue}{1} \]
7. Add Preprocessing

Developer Target 1: 100.0% accurate, 1.9× speedup

Math

\[\begin{array}{l} \\ \tan \left(\frac{x}{2}\right) \end{array} \]

FPCore

(FPCore (x) :precision binary64 (tan (/ x 2.0)))

C

double code(double x) {
	return tan((x / 2.0));
}

Fortran

real(8) function code(x)
    real(8), intent (in) :: x
    code = tan((x / 2.0d0))
end function

Java

public static double code(double x) {
	return Math.tan((x / 2.0));
}

Python

def code(x):
	return math.tan((x / 2.0))

Julia

function code(x)
	return tan(Float64(x / 2.0))
end

MATLAB

function tmp = code(x)
	tmp = tan((x / 2.0));
end

Wolfram

code[x_] := N[Tan[N[(x / 2.0), $MachinePrecision]], $MachinePrecision]

Reproduce

herbie shell --seed 2024215 
(FPCore (x)
  :name "tanhf (example 3.4)"
  :precision binary64

  :alt
  (! :herbie-platform default (tan (/ x 2)))

  (/ (- 1.0 (cos x)) (sin x)))