2cbrt (problem 3.3.4)

Percentage Accurate: 6.9% → 50.5%
Time: 8.2s
Alternatives: 6
Speedup: 1.7×

Specification

?
\[x > 1 \land x < 10^{+308}\]
\[\begin{array}{l} \\ \sqrt[3]{x + 1} - \sqrt[3]{x} \end{array} \]
(FPCore (x) :precision binary64 (- (cbrt (+ x 1.0)) (cbrt x)))
double code(double x) {
	return cbrt((x + 1.0)) - cbrt(x);
}
public static double code(double x) {
	return Math.cbrt((x + 1.0)) - Math.cbrt(x);
}
function code(x)
	return Float64(cbrt(Float64(x + 1.0)) - cbrt(x))
end
code[x_] := N[(N[Power[N[(x + 1.0), $MachinePrecision], 1/3], $MachinePrecision] - N[Power[x, 1/3], $MachinePrecision]), $MachinePrecision]
\begin{array}{l}

\\
\sqrt[3]{x + 1} - \sqrt[3]{x}
\end{array}

Sampling outcomes in binary64 precision:

Local Percentage Accuracy vs ?

The average percentage accuracy by input value. Horizontal axis shows value of an input variable; the variable is choosen in the title. Vertical axis is accuracy; higher is better. Red represent the original program, while blue represents Herbie's suggestion. These can be toggled with buttons below the plot. The line is an average while dots represent individual samples.

Accuracy vs Speed?

Herbie found 6 alternatives:

AlternativeAccuracySpeedup
The accuracy (vertical axis) and speed (horizontal axis) of each alternatives. Up and to the right is better. The red square shows the initial program, and each blue circle shows an alternative.The line shows the best available speed-accuracy tradeoffs.

Initial Program: 6.9% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \sqrt[3]{x + 1} - \sqrt[3]{x} \end{array} \]
(FPCore (x) :precision binary64 (- (cbrt (+ x 1.0)) (cbrt x)))
double code(double x) {
	return cbrt((x + 1.0)) - cbrt(x);
}
public static double code(double x) {
	return Math.cbrt((x + 1.0)) - Math.cbrt(x);
}
function code(x)
	return Float64(cbrt(Float64(x + 1.0)) - cbrt(x))
end
code[x_] := N[(N[Power[N[(x + 1.0), $MachinePrecision], 1/3], $MachinePrecision] - N[Power[x, 1/3], $MachinePrecision]), $MachinePrecision]
\begin{array}{l}

\\
\sqrt[3]{x + 1} - \sqrt[3]{x}
\end{array}

Alternative 1: 50.5% accurate, 0.6× speedup?

\[\begin{array}{l} \\ \mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, -0.1111111111111111, \sqrt[3]{\frac{1}{x \cdot x}} \cdot 0.3333333333333333\right) \end{array} \]
(FPCore (x)
 :precision binary64
 (fma
  (cbrt (/ 1.0 (pow x 5.0)))
  -0.1111111111111111
  (* (cbrt (/ 1.0 (* x x))) 0.3333333333333333)))
double code(double x) {
	return fma(cbrt((1.0 / pow(x, 5.0))), -0.1111111111111111, (cbrt((1.0 / (x * x))) * 0.3333333333333333));
}
function code(x)
	return fma(cbrt(Float64(1.0 / (x ^ 5.0))), -0.1111111111111111, Float64(cbrt(Float64(1.0 / Float64(x * x))) * 0.3333333333333333))
end
code[x_] := N[(N[Power[N[(1.0 / N[Power[x, 5.0], $MachinePrecision]), $MachinePrecision], 1/3], $MachinePrecision] * -0.1111111111111111 + N[(N[Power[N[(1.0 / N[(x * x), $MachinePrecision]), $MachinePrecision], 1/3], $MachinePrecision] * 0.3333333333333333), $MachinePrecision]), $MachinePrecision]
\begin{array}{l}

\\
\mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, -0.1111111111111111, \sqrt[3]{\frac{1}{x \cdot x}} \cdot 0.3333333333333333\right)
\end{array}
Derivation
  1. Initial program 7.4%

    \[\sqrt[3]{x + 1} - \sqrt[3]{x} \]
  2. Add Preprocessing
  3. Taylor expanded in x around inf

    \[\leadsto \color{blue}{\frac{\frac{-1}{9} \cdot \sqrt[3]{x} + \frac{1}{3} \cdot \sqrt[3]{{x}^{4}}}{{x}^{2}}} \]
  4. Step-by-step derivation
    1. lower-/.f64N/A

      \[\leadsto \color{blue}{\frac{\frac{-1}{9} \cdot \sqrt[3]{x} + \frac{1}{3} \cdot \sqrt[3]{{x}^{4}}}{{x}^{2}}} \]
    2. +-commutativeN/A

      \[\leadsto \frac{\color{blue}{\frac{1}{3} \cdot \sqrt[3]{{x}^{4}} + \frac{-1}{9} \cdot \sqrt[3]{x}}}{{x}^{2}} \]
    3. lower-fma.f64N/A

      \[\leadsto \frac{\color{blue}{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{{x}^{4}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}}{{x}^{2}} \]
    4. metadata-evalN/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{{x}^{\color{blue}{\left(2 \cdot 2\right)}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    5. pow-sqrN/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{{x}^{2} \cdot {x}^{2}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    6. lower-cbrt.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \color{blue}{\sqrt[3]{{x}^{2} \cdot {x}^{2}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    7. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{{x}^{2} \cdot {x}^{2}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    8. unpow2N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{\left(x \cdot x\right)} \cdot {x}^{2}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    9. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{\left(x \cdot x\right)} \cdot {x}^{2}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    10. unpow2N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \color{blue}{\left(x \cdot x\right)}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    11. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \color{blue}{\left(x \cdot x\right)}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    12. *-commutativeN/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \color{blue}{\sqrt[3]{x} \cdot \frac{-1}{9}}\right)}{{x}^{2}} \]
    13. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \color{blue}{\sqrt[3]{x} \cdot \frac{-1}{9}}\right)}{{x}^{2}} \]
    14. lower-cbrt.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \color{blue}{\sqrt[3]{x}} \cdot \frac{-1}{9}\right)}{{x}^{2}} \]
    15. unpow2N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \sqrt[3]{x} \cdot \frac{-1}{9}\right)}{\color{blue}{x \cdot x}} \]
    16. lower-*.f6424.8

      \[\leadsto \frac{\mathsf{fma}\left(0.3333333333333333, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \sqrt[3]{x} \cdot -0.1111111111111111\right)}{\color{blue}{x \cdot x}} \]
  5. Simplified24.8%

    \[\leadsto \color{blue}{\frac{\mathsf{fma}\left(0.3333333333333333, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \sqrt[3]{x} \cdot -0.1111111111111111\right)}{x \cdot x}} \]
  6. Taylor expanded in x around inf

    \[\leadsto \color{blue}{\frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}} + \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}} \]
  7. Step-by-step derivation
    1. +-commutativeN/A

      \[\leadsto \color{blue}{\frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}} + \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}} \]
    2. *-commutativeN/A

      \[\leadsto \color{blue}{\sqrt[3]{\frac{1}{{x}^{2}}} \cdot \frac{1}{3}} + \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}} \]
    3. lower-fma.f64N/A

      \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{2}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right)} \]
    4. lower-cbrt.f64N/A

      \[\leadsto \mathsf{fma}\left(\color{blue}{\sqrt[3]{\frac{1}{{x}^{2}}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    5. lower-/.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\color{blue}{\frac{1}{{x}^{2}}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    6. unpow2N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    7. lower-*.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    8. lower-*.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, \frac{1}{3}, \color{blue}{\frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}}\right) \]
    9. lower-cbrt.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, \frac{1}{3}, \frac{-1}{9} \cdot \color{blue}{\sqrt[3]{\frac{1}{{x}^{5}}}}\right) \]
    10. lower-/.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\color{blue}{\frac{1}{{x}^{5}}}}\right) \]
    11. lower-pow.f6452.2

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, 0.3333333333333333, -0.1111111111111111 \cdot \sqrt[3]{\frac{1}{\color{blue}{{x}^{5}}}}\right) \]
  8. Simplified52.2%

    \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, 0.3333333333333333, -0.1111111111111111 \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right)} \]
  9. Taylor expanded in x around inf

    \[\leadsto \color{blue}{\frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}} + \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}} \]
  10. Step-by-step derivation
    1. *-commutativeN/A

      \[\leadsto \color{blue}{\sqrt[3]{\frac{1}{{x}^{5}}} \cdot \frac{-1}{9}} + \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}} \]
    2. lower-fma.f64N/A

      \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, \frac{-1}{9}, \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}\right)} \]
    3. lower-cbrt.f64N/A

      \[\leadsto \mathsf{fma}\left(\color{blue}{\sqrt[3]{\frac{1}{{x}^{5}}}}, \frac{-1}{9}, \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}\right) \]
    4. lower-/.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\color{blue}{\frac{1}{{x}^{5}}}}, \frac{-1}{9}, \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}\right) \]
    5. lower-pow.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{\color{blue}{{x}^{5}}}}, \frac{-1}{9}, \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}\right) \]
    6. *-commutativeN/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, \frac{-1}{9}, \color{blue}{\sqrt[3]{\frac{1}{{x}^{2}}} \cdot \frac{1}{3}}\right) \]
    7. lower-*.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, \frac{-1}{9}, \color{blue}{\sqrt[3]{\frac{1}{{x}^{2}}} \cdot \frac{1}{3}}\right) \]
    8. lower-cbrt.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, \frac{-1}{9}, \color{blue}{\sqrt[3]{\frac{1}{{x}^{2}}}} \cdot \frac{1}{3}\right) \]
    9. lower-/.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, \frac{-1}{9}, \sqrt[3]{\color{blue}{\frac{1}{{x}^{2}}}} \cdot \frac{1}{3}\right) \]
    10. unpow2N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, \frac{-1}{9}, \sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}} \cdot \frac{1}{3}\right) \]
    11. lower-*.f6452.2

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, -0.1111111111111111, \sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}} \cdot 0.3333333333333333\right) \]
  11. Simplified52.2%

    \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{5}}}, -0.1111111111111111, \sqrt[3]{\frac{1}{x \cdot x}} \cdot 0.3333333333333333\right)} \]
  12. Add Preprocessing

Alternative 2: 50.5% accurate, 0.6× speedup?

\[\begin{array}{l} \\ \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, 0.3333333333333333, \sqrt[3]{\frac{1}{{x}^{5}}} \cdot -0.1111111111111111\right) \end{array} \]
(FPCore (x)
 :precision binary64
 (fma
  (cbrt (/ 1.0 (* x x)))
  0.3333333333333333
  (* (cbrt (/ 1.0 (pow x 5.0))) -0.1111111111111111)))
double code(double x) {
	return fma(cbrt((1.0 / (x * x))), 0.3333333333333333, (cbrt((1.0 / pow(x, 5.0))) * -0.1111111111111111));
}
function code(x)
	return fma(cbrt(Float64(1.0 / Float64(x * x))), 0.3333333333333333, Float64(cbrt(Float64(1.0 / (x ^ 5.0))) * -0.1111111111111111))
end
code[x_] := N[(N[Power[N[(1.0 / N[(x * x), $MachinePrecision]), $MachinePrecision], 1/3], $MachinePrecision] * 0.3333333333333333 + N[(N[Power[N[(1.0 / N[Power[x, 5.0], $MachinePrecision]), $MachinePrecision], 1/3], $MachinePrecision] * -0.1111111111111111), $MachinePrecision]), $MachinePrecision]
\begin{array}{l}

\\
\mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, 0.3333333333333333, \sqrt[3]{\frac{1}{{x}^{5}}} \cdot -0.1111111111111111\right)
\end{array}
Derivation
  1. Initial program 7.4%

    \[\sqrt[3]{x + 1} - \sqrt[3]{x} \]
  2. Add Preprocessing
  3. Taylor expanded in x around inf

    \[\leadsto \color{blue}{\frac{\frac{-1}{9} \cdot \sqrt[3]{x} + \frac{1}{3} \cdot \sqrt[3]{{x}^{4}}}{{x}^{2}}} \]
  4. Step-by-step derivation
    1. lower-/.f64N/A

      \[\leadsto \color{blue}{\frac{\frac{-1}{9} \cdot \sqrt[3]{x} + \frac{1}{3} \cdot \sqrt[3]{{x}^{4}}}{{x}^{2}}} \]
    2. +-commutativeN/A

      \[\leadsto \frac{\color{blue}{\frac{1}{3} \cdot \sqrt[3]{{x}^{4}} + \frac{-1}{9} \cdot \sqrt[3]{x}}}{{x}^{2}} \]
    3. lower-fma.f64N/A

      \[\leadsto \frac{\color{blue}{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{{x}^{4}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}}{{x}^{2}} \]
    4. metadata-evalN/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{{x}^{\color{blue}{\left(2 \cdot 2\right)}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    5. pow-sqrN/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{{x}^{2} \cdot {x}^{2}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    6. lower-cbrt.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \color{blue}{\sqrt[3]{{x}^{2} \cdot {x}^{2}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    7. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{{x}^{2} \cdot {x}^{2}}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    8. unpow2N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{\left(x \cdot x\right)} \cdot {x}^{2}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    9. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\color{blue}{\left(x \cdot x\right)} \cdot {x}^{2}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    10. unpow2N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \color{blue}{\left(x \cdot x\right)}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    11. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \color{blue}{\left(x \cdot x\right)}}, \frac{-1}{9} \cdot \sqrt[3]{x}\right)}{{x}^{2}} \]
    12. *-commutativeN/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \color{blue}{\sqrt[3]{x} \cdot \frac{-1}{9}}\right)}{{x}^{2}} \]
    13. lower-*.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \color{blue}{\sqrt[3]{x} \cdot \frac{-1}{9}}\right)}{{x}^{2}} \]
    14. lower-cbrt.f64N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \color{blue}{\sqrt[3]{x}} \cdot \frac{-1}{9}\right)}{{x}^{2}} \]
    15. unpow2N/A

      \[\leadsto \frac{\mathsf{fma}\left(\frac{1}{3}, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \sqrt[3]{x} \cdot \frac{-1}{9}\right)}{\color{blue}{x \cdot x}} \]
    16. lower-*.f6424.8

      \[\leadsto \frac{\mathsf{fma}\left(0.3333333333333333, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \sqrt[3]{x} \cdot -0.1111111111111111\right)}{\color{blue}{x \cdot x}} \]
  5. Simplified24.8%

    \[\leadsto \color{blue}{\frac{\mathsf{fma}\left(0.3333333333333333, \sqrt[3]{\left(x \cdot x\right) \cdot \left(x \cdot x\right)}, \sqrt[3]{x} \cdot -0.1111111111111111\right)}{x \cdot x}} \]
  6. Taylor expanded in x around inf

    \[\leadsto \color{blue}{\frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}} + \frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}} \]
  7. Step-by-step derivation
    1. +-commutativeN/A

      \[\leadsto \color{blue}{\frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}} + \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}} \]
    2. *-commutativeN/A

      \[\leadsto \color{blue}{\sqrt[3]{\frac{1}{{x}^{2}}} \cdot \frac{1}{3}} + \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}} \]
    3. lower-fma.f64N/A

      \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt[3]{\frac{1}{{x}^{2}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right)} \]
    4. lower-cbrt.f64N/A

      \[\leadsto \mathsf{fma}\left(\color{blue}{\sqrt[3]{\frac{1}{{x}^{2}}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    5. lower-/.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\color{blue}{\frac{1}{{x}^{2}}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    6. unpow2N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    7. lower-*.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right) \]
    8. lower-*.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, \frac{1}{3}, \color{blue}{\frac{-1}{9} \cdot \sqrt[3]{\frac{1}{{x}^{5}}}}\right) \]
    9. lower-cbrt.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, \frac{1}{3}, \frac{-1}{9} \cdot \color{blue}{\sqrt[3]{\frac{1}{{x}^{5}}}}\right) \]
    10. lower-/.f64N/A

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, \frac{1}{3}, \frac{-1}{9} \cdot \sqrt[3]{\color{blue}{\frac{1}{{x}^{5}}}}\right) \]
    11. lower-pow.f6452.2

      \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, 0.3333333333333333, -0.1111111111111111 \cdot \sqrt[3]{\frac{1}{\color{blue}{{x}^{5}}}}\right) \]
  8. Simplified52.2%

    \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, 0.3333333333333333, -0.1111111111111111 \cdot \sqrt[3]{\frac{1}{{x}^{5}}}\right)} \]
  9. Final simplification52.2%

    \[\leadsto \mathsf{fma}\left(\sqrt[3]{\frac{1}{x \cdot x}}, 0.3333333333333333, \sqrt[3]{\frac{1}{{x}^{5}}} \cdot -0.1111111111111111\right) \]
  10. Add Preprocessing

Alternative 3: 49.5% accurate, 1.7× speedup?

\[\begin{array}{l} \\ \sqrt[3]{\frac{1}{x \cdot x}} \cdot 0.3333333333333333 \end{array} \]
(FPCore (x) :precision binary64 (* (cbrt (/ 1.0 (* x x))) 0.3333333333333333))
double code(double x) {
	return cbrt((1.0 / (x * x))) * 0.3333333333333333;
}
public static double code(double x) {
	return Math.cbrt((1.0 / (x * x))) * 0.3333333333333333;
}
function code(x)
	return Float64(cbrt(Float64(1.0 / Float64(x * x))) * 0.3333333333333333)
end
code[x_] := N[(N[Power[N[(1.0 / N[(x * x), $MachinePrecision]), $MachinePrecision], 1/3], $MachinePrecision] * 0.3333333333333333), $MachinePrecision]
\begin{array}{l}

\\
\sqrt[3]{\frac{1}{x \cdot x}} \cdot 0.3333333333333333
\end{array}
Derivation
  1. Initial program 7.4%

    \[\sqrt[3]{x + 1} - \sqrt[3]{x} \]
  2. Add Preprocessing
  3. Taylor expanded in x around inf

    \[\leadsto \color{blue}{\frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}} \]
  4. Step-by-step derivation
    1. lower-*.f64N/A

      \[\leadsto \color{blue}{\frac{1}{3} \cdot \sqrt[3]{\frac{1}{{x}^{2}}}} \]
    2. metadata-evalN/A

      \[\leadsto \frac{1}{3} \cdot \sqrt[3]{\frac{\color{blue}{-1 \cdot -1}}{{x}^{2}}} \]
    3. associate-*r/N/A

      \[\leadsto \frac{1}{3} \cdot \sqrt[3]{\color{blue}{-1 \cdot \frac{-1}{{x}^{2}}}} \]
    4. lower-cbrt.f64N/A

      \[\leadsto \frac{1}{3} \cdot \color{blue}{\sqrt[3]{-1 \cdot \frac{-1}{{x}^{2}}}} \]
    5. associate-*r/N/A

      \[\leadsto \frac{1}{3} \cdot \sqrt[3]{\color{blue}{\frac{-1 \cdot -1}{{x}^{2}}}} \]
    6. metadata-evalN/A

      \[\leadsto \frac{1}{3} \cdot \sqrt[3]{\frac{\color{blue}{1}}{{x}^{2}}} \]
    7. lower-/.f64N/A

      \[\leadsto \frac{1}{3} \cdot \sqrt[3]{\color{blue}{\frac{1}{{x}^{2}}}} \]
    8. unpow2N/A

      \[\leadsto \frac{1}{3} \cdot \sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}} \]
    9. lower-*.f6451.1

      \[\leadsto 0.3333333333333333 \cdot \sqrt[3]{\frac{1}{\color{blue}{x \cdot x}}} \]
  5. Simplified51.1%

    \[\leadsto \color{blue}{0.3333333333333333 \cdot \sqrt[3]{\frac{1}{x \cdot x}}} \]
  6. Final simplification51.1%

    \[\leadsto \sqrt[3]{\frac{1}{x \cdot x}} \cdot 0.3333333333333333 \]
  7. Add Preprocessing

Alternative 4: 4.2% accurate, 1.9× speedup?

\[\begin{array}{l} \\ \mathsf{fma}\left(x, 0.3333333333333333, -\sqrt[3]{x}\right) \end{array} \]
(FPCore (x) :precision binary64 (fma x 0.3333333333333333 (- (cbrt x))))
double code(double x) {
	return fma(x, 0.3333333333333333, -cbrt(x));
}
function code(x)
	return fma(x, 0.3333333333333333, Float64(-cbrt(x)))
end
code[x_] := N[(x * 0.3333333333333333 + (-N[Power[x, 1/3], $MachinePrecision])), $MachinePrecision]
\begin{array}{l}

\\
\mathsf{fma}\left(x, 0.3333333333333333, -\sqrt[3]{x}\right)
\end{array}
Derivation
  1. Initial program 7.4%

    \[\sqrt[3]{x + 1} - \sqrt[3]{x} \]
  2. Add Preprocessing
  3. Taylor expanded in x around 0

    \[\leadsto \color{blue}{\left(1 + \frac{1}{3} \cdot x\right) - \sqrt[3]{x}} \]
  4. Step-by-step derivation
    1. +-commutativeN/A

      \[\leadsto \color{blue}{\left(\frac{1}{3} \cdot x + 1\right)} - \sqrt[3]{x} \]
    2. associate--l+N/A

      \[\leadsto \color{blue}{\frac{1}{3} \cdot x + \left(1 - \sqrt[3]{x}\right)} \]
    3. *-commutativeN/A

      \[\leadsto \color{blue}{x \cdot \frac{1}{3}} + \left(1 - \sqrt[3]{x}\right) \]
    4. lower-fma.f64N/A

      \[\leadsto \color{blue}{\mathsf{fma}\left(x, \frac{1}{3}, 1 - \sqrt[3]{x}\right)} \]
    5. lower--.f64N/A

      \[\leadsto \mathsf{fma}\left(x, \frac{1}{3}, \color{blue}{1 - \sqrt[3]{x}}\right) \]
    6. lower-cbrt.f644.3

      \[\leadsto \mathsf{fma}\left(x, 0.3333333333333333, 1 - \color{blue}{\sqrt[3]{x}}\right) \]
  5. Simplified4.3%

    \[\leadsto \color{blue}{\mathsf{fma}\left(x, 0.3333333333333333, 1 - \sqrt[3]{x}\right)} \]
  6. Taylor expanded in x around inf

    \[\leadsto \mathsf{fma}\left(x, \frac{1}{3}, \color{blue}{-1 \cdot \sqrt[3]{x}}\right) \]
  7. Step-by-step derivation
    1. mul-1-negN/A

      \[\leadsto \mathsf{fma}\left(x, \frac{1}{3}, \color{blue}{\mathsf{neg}\left(\sqrt[3]{x}\right)}\right) \]
    2. lower-neg.f64N/A

      \[\leadsto \mathsf{fma}\left(x, \frac{1}{3}, \color{blue}{\mathsf{neg}\left(\sqrt[3]{x}\right)}\right) \]
    3. lower-cbrt.f644.3

      \[\leadsto \mathsf{fma}\left(x, 0.3333333333333333, -\color{blue}{\sqrt[3]{x}}\right) \]
  8. Simplified4.3%

    \[\leadsto \mathsf{fma}\left(x, 0.3333333333333333, \color{blue}{-\sqrt[3]{x}}\right) \]
  9. Add Preprocessing

Alternative 5: 1.8% accurate, 2.0× speedup?

\[\begin{array}{l} \\ 1 - \sqrt[3]{x} \end{array} \]
(FPCore (x) :precision binary64 (- 1.0 (cbrt x)))
double code(double x) {
	return 1.0 - cbrt(x);
}
public static double code(double x) {
	return 1.0 - Math.cbrt(x);
}
function code(x)
	return Float64(1.0 - cbrt(x))
end
code[x_] := N[(1.0 - N[Power[x, 1/3], $MachinePrecision]), $MachinePrecision]
\begin{array}{l}

\\
1 - \sqrt[3]{x}
\end{array}
Derivation
  1. Initial program 7.4%

    \[\sqrt[3]{x + 1} - \sqrt[3]{x} \]
  2. Add Preprocessing
  3. Taylor expanded in x around 0

    \[\leadsto \color{blue}{1 - \sqrt[3]{x}} \]
  4. Step-by-step derivation
    1. lower--.f64N/A

      \[\leadsto \color{blue}{1 - \sqrt[3]{x}} \]
    2. lower-cbrt.f641.8

      \[\leadsto 1 - \color{blue}{\sqrt[3]{x}} \]
  5. Simplified1.8%

    \[\leadsto \color{blue}{1 - \sqrt[3]{x}} \]
  6. Add Preprocessing

Alternative 6: 1.8% accurate, 2.0× speedup?

\[\begin{array}{l} \\ -\sqrt[3]{x} \end{array} \]
(FPCore (x) :precision binary64 (- (cbrt x)))
double code(double x) {
	return -cbrt(x);
}
public static double code(double x) {
	return -Math.cbrt(x);
}
function code(x)
	return Float64(-cbrt(x))
end
code[x_] := (-N[Power[x, 1/3], $MachinePrecision])
\begin{array}{l}

\\
-\sqrt[3]{x}
\end{array}
Derivation
  1. Initial program 7.4%

    \[\sqrt[3]{x + 1} - \sqrt[3]{x} \]
  2. Add Preprocessing
  3. Taylor expanded in x around 0

    \[\leadsto \color{blue}{1 - \sqrt[3]{x}} \]
  4. Step-by-step derivation
    1. lower--.f64N/A

      \[\leadsto \color{blue}{1 - \sqrt[3]{x}} \]
    2. lower-cbrt.f641.8

      \[\leadsto 1 - \color{blue}{\sqrt[3]{x}} \]
  5. Simplified1.8%

    \[\leadsto \color{blue}{1 - \sqrt[3]{x}} \]
  6. Taylor expanded in x around inf

    \[\leadsto \color{blue}{-1 \cdot \sqrt[3]{x}} \]
  7. Step-by-step derivation
    1. mul-1-negN/A

      \[\leadsto \color{blue}{\mathsf{neg}\left(\sqrt[3]{x}\right)} \]
    2. lower-neg.f64N/A

      \[\leadsto \color{blue}{\mathsf{neg}\left(\sqrt[3]{x}\right)} \]
    3. lower-cbrt.f641.8

      \[\leadsto -\color{blue}{\sqrt[3]{x}} \]
  8. Simplified1.8%

    \[\leadsto \color{blue}{-\sqrt[3]{x}} \]
  9. Add Preprocessing

Developer Target 1: 98.5% accurate, 0.3× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \sqrt[3]{x + 1}\\ \frac{1}{\left(t\_0 \cdot t\_0 + \sqrt[3]{x} \cdot t\_0\right) + \sqrt[3]{x} \cdot \sqrt[3]{x}} \end{array} \end{array} \]
(FPCore (x)
 :precision binary64
 (let* ((t_0 (cbrt (+ x 1.0))))
   (/ 1.0 (+ (+ (* t_0 t_0) (* (cbrt x) t_0)) (* (cbrt x) (cbrt x))))))
double code(double x) {
	double t_0 = cbrt((x + 1.0));
	return 1.0 / (((t_0 * t_0) + (cbrt(x) * t_0)) + (cbrt(x) * cbrt(x)));
}
public static double code(double x) {
	double t_0 = Math.cbrt((x + 1.0));
	return 1.0 / (((t_0 * t_0) + (Math.cbrt(x) * t_0)) + (Math.cbrt(x) * Math.cbrt(x)));
}
function code(x)
	t_0 = cbrt(Float64(x + 1.0))
	return Float64(1.0 / Float64(Float64(Float64(t_0 * t_0) + Float64(cbrt(x) * t_0)) + Float64(cbrt(x) * cbrt(x))))
end
code[x_] := Block[{t$95$0 = N[Power[N[(x + 1.0), $MachinePrecision], 1/3], $MachinePrecision]}, N[(1.0 / N[(N[(N[(t$95$0 * t$95$0), $MachinePrecision] + N[(N[Power[x, 1/3], $MachinePrecision] * t$95$0), $MachinePrecision]), $MachinePrecision] + N[(N[Power[x, 1/3], $MachinePrecision] * N[Power[x, 1/3], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]
\begin{array}{l}

\\
\begin{array}{l}
t_0 := \sqrt[3]{x + 1}\\
\frac{1}{\left(t\_0 \cdot t\_0 + \sqrt[3]{x} \cdot t\_0\right) + \sqrt[3]{x} \cdot \sqrt[3]{x}}
\end{array}
\end{array}

Reproduce

?
herbie shell --seed 2024215 
(FPCore (x)
  :name "2cbrt (problem 3.3.4)"
  :precision binary64
  :pre (and (> x 1.0) (< x 1e+308))

  :alt
  (! :herbie-platform default (/ 1 (+ (* (cbrt (+ x 1)) (cbrt (+ x 1))) (* (cbrt x) (cbrt (+ x 1))) (* (cbrt x) (cbrt x)))))

  (- (cbrt (+ x 1.0)) (cbrt x)))