Beckmann Sample, near normal, slope_x

Percentage Accurate: 57.8% → 91.4%
Time: 8.6s
Alternatives: 7
Speedup: 11.6×

Specification

?
\[\left(\left(cosTheta\_i > 0.9999 \land cosTheta\_i \leq 1\right) \land \left(2.328306437 \cdot 10^{-10} \leq u1 \land u1 \leq 1\right)\right) \land \left(2.328306437 \cdot 10^{-10} \leq u2 \land u2 \leq 1\right)\]
\[\begin{array}{l} \\ \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \end{array} \]
(FPCore (cosTheta_i u1 u2)
 :precision binary32
 (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 (PI)) u2))))
\begin{array}{l}

\\
\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)
\end{array}

Sampling outcomes in binary32 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 7 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: 57.8% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \end{array} \]
(FPCore (cosTheta_i u1 u2)
 :precision binary32
 (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 (PI)) u2))))
\begin{array}{l}

\\
\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)
\end{array}

Alternative 1: 91.4% accurate, 0.4× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := -\log \left(1 - u1\right)\\ t_1 := \sqrt[3]{\mathsf{PI}\left(\right)}\\ \mathbf{if}\;t\_0 \leq 0.00013080000644549727:\\ \;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}\\ \mathbf{else}:\\ \;\;\;\;\cos \left(\left(\left(t\_1 \cdot u2\right) \cdot 2\right) \cdot {t\_1}^{2}\right) \cdot \sqrt{t\_0}\\ \end{array} \end{array} \]
(FPCore (cosTheta_i u1 u2)
 :precision binary32
 (let* ((t_0 (- (log (- 1.0 u1)))) (t_1 (cbrt (PI))))
   (if (<= t_0 0.00013080000644549727)
     (* (cos (* (* 2.0 (PI)) u2)) (sqrt u1))
     (* (cos (* (* (* t_1 u2) 2.0) (pow t_1 2.0))) (sqrt t_0)))))
\begin{array}{l}

\\
\begin{array}{l}
t_0 := -\log \left(1 - u1\right)\\
t_1 := \sqrt[3]{\mathsf{PI}\left(\right)}\\
\mathbf{if}\;t\_0 \leq 0.00013080000644549727:\\
\;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}\\

\mathbf{else}:\\
\;\;\;\;\cos \left(\left(\left(t\_1 \cdot u2\right) \cdot 2\right) \cdot {t\_1}^{2}\right) \cdot \sqrt{t\_0}\\


\end{array}
\end{array}
Derivation
  1. Split input into 2 regimes
  2. if (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1))) < 1.30800006e-4

    1. Initial program 39.7%

      \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. Add Preprocessing
    3. Taylor expanded in u1 around 0

      \[\leadsto \color{blue}{\left(\sqrt{u1} \cdot {\left(\sqrt{-1}\right)}^{2}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    4. Step-by-step derivation
      1. *-commutativeN/A

        \[\leadsto \color{blue}{\left({\left(\sqrt{-1}\right)}^{2} \cdot \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      2. unpow2N/A

        \[\leadsto \left(\color{blue}{\left(\sqrt{-1} \cdot \sqrt{-1}\right)} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      3. rem-square-sqrtN/A

        \[\leadsto \left(\color{blue}{-1} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      4. mul-1-negN/A

        \[\leadsto \color{blue}{\left(\mathsf{neg}\left(\sqrt{u1}\right)\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      5. lower-neg.f32N/A

        \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      6. lower-sqrt.f323.6

        \[\leadsto \left(-\color{blue}{\sqrt{u1}}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    5. Applied rewrites3.6%

      \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    6. Applied rewrites91.9%

      \[\leadsto \color{blue}{\cos \left(u2 \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)\right) \cdot \sqrt{u1}} \]

    if 1.30800006e-4 < (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1)))

    1. Initial program 89.3%

      \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. Add Preprocessing
    3. Step-by-step derivation
      1. lift-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)} \]
      2. lift-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{\left(2 \cdot \mathsf{PI}\left(\right)\right)} \cdot u2\right) \]
      3. associate-*l*N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(2 \cdot \left(\mathsf{PI}\left(\right) \cdot u2\right)\right)} \]
      4. *-commutativeN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(\mathsf{PI}\left(\right) \cdot u2\right) \cdot 2\right)} \]
      5. lift-PI.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\color{blue}{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right) \]
      6. add-cube-cbrtN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\color{blue}{\left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right) \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right)} \cdot u2\right) \cdot 2\right) \]
      7. associate-*l*N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{\left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right) \cdot \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right)\right)} \cdot 2\right) \]
      8. associate-*l*N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right) \cdot \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right)} \]
      9. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right) \cdot \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right)} \]
      10. pow2N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{{\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2}} \cdot \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right) \]
      11. lower-pow.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{{\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2}} \cdot \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right) \]
      12. lift-PI.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\left(\sqrt[3]{\color{blue}{\mathsf{PI}\left(\right)}}\right)}^{2} \cdot \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right) \]
      13. lower-cbrt.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\color{blue}{\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}}^{2} \cdot \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right) \]
      14. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2} \cdot \color{blue}{\left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)}\right) \]
      15. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2} \cdot \left(\color{blue}{\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right)} \cdot 2\right)\right) \]
      16. lift-PI.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2} \cdot \left(\left(\sqrt[3]{\color{blue}{\mathsf{PI}\left(\right)}} \cdot u2\right) \cdot 2\right)\right) \]
      17. lower-cbrt.f3289.4

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2} \cdot \left(\left(\color{blue}{\sqrt[3]{\mathsf{PI}\left(\right)}} \cdot u2\right) \cdot 2\right)\right) \]
    4. Applied rewrites89.4%

      \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left({\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2} \cdot \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right)} \]
  3. Recombined 2 regimes into one program.
  4. Final simplification90.9%

    \[\leadsto \begin{array}{l} \mathbf{if}\;-\log \left(1 - u1\right) \leq 0.00013080000644549727:\\ \;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}\\ \mathbf{else}:\\ \;\;\;\;\cos \left(\left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right) \cdot {\left(\sqrt[3]{\mathsf{PI}\left(\right)}\right)}^{2}\right) \cdot \sqrt{-\log \left(1 - u1\right)}\\ \end{array} \]
  5. Add Preprocessing

Alternative 2: 91.4% accurate, 0.4× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := -\log \left(1 - u1\right)\\ \mathbf{if}\;t\_0 \leq 0.00013080000644549727:\\ \;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}\\ \mathbf{else}:\\ \;\;\;\;\cos \left(\left(\left(\left(\sqrt{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right) \cdot {\mathsf{PI}\left(\right)}^{0.16666666666666666}\right) \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{t\_0}\\ \end{array} \end{array} \]
(FPCore (cosTheta_i u1 u2)
 :precision binary32
 (let* ((t_0 (- (log (- 1.0 u1)))))
   (if (<= t_0 0.00013080000644549727)
     (* (cos (* (* 2.0 (PI)) u2)) (sqrt u1))
     (*
      (cos
       (*
        (* (* (* (sqrt (PI)) u2) 2.0) (pow (PI) 0.16666666666666666))
        (cbrt (PI))))
      (sqrt t_0)))))
\begin{array}{l}

\\
\begin{array}{l}
t_0 := -\log \left(1 - u1\right)\\
\mathbf{if}\;t\_0 \leq 0.00013080000644549727:\\
\;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}\\

\mathbf{else}:\\
\;\;\;\;\cos \left(\left(\left(\left(\sqrt{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right) \cdot {\mathsf{PI}\left(\right)}^{0.16666666666666666}\right) \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{t\_0}\\


\end{array}
\end{array}
Derivation
  1. Split input into 2 regimes
  2. if (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1))) < 1.30800006e-4

    1. Initial program 39.7%

      \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. Add Preprocessing
    3. Taylor expanded in u1 around 0

      \[\leadsto \color{blue}{\left(\sqrt{u1} \cdot {\left(\sqrt{-1}\right)}^{2}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    4. Step-by-step derivation
      1. *-commutativeN/A

        \[\leadsto \color{blue}{\left({\left(\sqrt{-1}\right)}^{2} \cdot \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      2. unpow2N/A

        \[\leadsto \left(\color{blue}{\left(\sqrt{-1} \cdot \sqrt{-1}\right)} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      3. rem-square-sqrtN/A

        \[\leadsto \left(\color{blue}{-1} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      4. mul-1-negN/A

        \[\leadsto \color{blue}{\left(\mathsf{neg}\left(\sqrt{u1}\right)\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      5. lower-neg.f32N/A

        \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      6. lower-sqrt.f323.6

        \[\leadsto \left(-\color{blue}{\sqrt{u1}}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    5. Applied rewrites3.6%

      \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    6. Applied rewrites91.9%

      \[\leadsto \color{blue}{\cos \left(u2 \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)\right) \cdot \sqrt{u1}} \]

    if 1.30800006e-4 < (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1)))

    1. Initial program 89.3%

      \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. Add Preprocessing
    3. Step-by-step derivation
      1. lift-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)} \]
      2. *-commutativeN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(u2 \cdot \left(2 \cdot \mathsf{PI}\left(\right)\right)\right)} \]
      3. lift-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(u2 \cdot \color{blue}{\left(2 \cdot \mathsf{PI}\left(\right)\right)}\right) \]
      4. associate-*r*N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(u2 \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right)} \]
      5. lift-PI.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(u2 \cdot 2\right) \cdot \color{blue}{\mathsf{PI}\left(\right)}\right) \]
      6. add-sqr-sqrtN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(u2 \cdot 2\right) \cdot \color{blue}{\left(\sqrt{\mathsf{PI}\left(\right)} \cdot \sqrt{\mathsf{PI}\left(\right)}\right)}\right) \]
      7. associate-*r*N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)} \]
      8. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)} \]
      9. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)} \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \]
      10. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\color{blue}{\left(u2 \cdot 2\right)} \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \]
      11. lift-PI.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\left(u2 \cdot 2\right) \cdot \sqrt{\color{blue}{\mathsf{PI}\left(\right)}}\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \]
      12. lower-sqrt.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\left(u2 \cdot 2\right) \cdot \color{blue}{\sqrt{\mathsf{PI}\left(\right)}}\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \]
      13. lift-PI.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{\color{blue}{\mathsf{PI}\left(\right)}}\right) \]
      14. lower-sqrt.f3289.2

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \cdot \color{blue}{\sqrt{\mathsf{PI}\left(\right)}}\right) \]
    4. Applied rewrites89.2%

      \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)} \]
    5. Step-by-step derivation
      1. lift-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)} \]
      2. *-commutativeN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\sqrt{\mathsf{PI}\left(\right)} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)} \]
      3. lift-sqrt.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{\sqrt{\mathsf{PI}\left(\right)}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      4. pow1/2N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{{\mathsf{PI}\left(\right)}^{\frac{1}{2}}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      5. metadata-evalN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\mathsf{PI}\left(\right)}^{\color{blue}{\left(\frac{1}{3} + \frac{1}{6}\right)}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      6. metadata-evalN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left({\mathsf{PI}\left(\right)}^{\left(\frac{1}{3} + \color{blue}{\frac{\frac{1}{3}}{2}}\right)} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      7. pow-prod-upN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\color{blue}{\left({\mathsf{PI}\left(\right)}^{\frac{1}{3}} \cdot {\mathsf{PI}\left(\right)}^{\left(\frac{\frac{1}{3}}{2}\right)}\right)} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      8. pow1/3N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\color{blue}{\sqrt[3]{\mathsf{PI}\left(\right)}} \cdot {\mathsf{PI}\left(\right)}^{\left(\frac{\frac{1}{3}}{2}\right)}\right) \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      9. lift-cbrt.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\color{blue}{\sqrt[3]{\mathsf{PI}\left(\right)}} \cdot {\mathsf{PI}\left(\right)}^{\left(\frac{\frac{1}{3}}{2}\right)}\right) \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      10. sqrt-pow2N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \color{blue}{{\left(\sqrt{\mathsf{PI}\left(\right)}\right)}^{\frac{1}{3}}}\right) \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      11. lift-sqrt.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot {\color{blue}{\left(\sqrt{\mathsf{PI}\left(\right)}\right)}}^{\frac{1}{3}}\right) \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right) \]
      12. associate-*l*N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left({\left(\sqrt{\mathsf{PI}\left(\right)}\right)}^{\frac{1}{3}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)\right)} \]
      13. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left({\left(\sqrt{\mathsf{PI}\left(\right)}\right)}^{\frac{1}{3}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)\right)} \]
      14. lower-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \color{blue}{\left({\left(\sqrt{\mathsf{PI}\left(\right)}\right)}^{\frac{1}{3}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)}\right) \]
      15. lift-sqrt.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left({\color{blue}{\left(\sqrt{\mathsf{PI}\left(\right)}\right)}}^{\frac{1}{3}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)\right) \]
      16. sqrt-pow2N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left(\color{blue}{{\mathsf{PI}\left(\right)}^{\left(\frac{\frac{1}{3}}{2}\right)}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)\right) \]
      17. lower-pow.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left(\color{blue}{{\mathsf{PI}\left(\right)}^{\left(\frac{\frac{1}{3}}{2}\right)}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)\right) \]
      18. metadata-eval89.3

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left({\mathsf{PI}\left(\right)}^{\color{blue}{0.16666666666666666}} \cdot \left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)\right)\right) \]
      19. lift-*.f32N/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left({\mathsf{PI}\left(\right)}^{\frac{1}{6}} \cdot \color{blue}{\left(\left(u2 \cdot 2\right) \cdot \sqrt{\mathsf{PI}\left(\right)}\right)}\right)\right) \]
      20. *-commutativeN/A

        \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left({\mathsf{PI}\left(\right)}^{\frac{1}{6}} \cdot \color{blue}{\left(\sqrt{\mathsf{PI}\left(\right)} \cdot \left(u2 \cdot 2\right)\right)}\right)\right) \]
    6. Applied rewrites89.3%

      \[\leadsto \sqrt{-\log \left(1 - u1\right)} \cdot \cos \color{blue}{\left(\sqrt[3]{\mathsf{PI}\left(\right)} \cdot \left({\mathsf{PI}\left(\right)}^{0.16666666666666666} \cdot \left(\left(\sqrt{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right)\right)\right)} \]
  3. Recombined 2 regimes into one program.
  4. Final simplification90.8%

    \[\leadsto \begin{array}{l} \mathbf{if}\;-\log \left(1 - u1\right) \leq 0.00013080000644549727:\\ \;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}\\ \mathbf{else}:\\ \;\;\;\;\cos \left(\left(\left(\left(\sqrt{\mathsf{PI}\left(\right)} \cdot u2\right) \cdot 2\right) \cdot {\mathsf{PI}\left(\right)}^{0.16666666666666666}\right) \cdot \sqrt[3]{\mathsf{PI}\left(\right)}\right) \cdot \sqrt{-\log \left(1 - u1\right)}\\ \end{array} \]
  5. Add Preprocessing

Alternative 3: 91.5% accurate, 0.7× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := -\log \left(1 - u1\right)\\ t_1 := \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)\\ \mathbf{if}\;t\_0 \leq 0.00013080000644549727:\\ \;\;\;\;t\_1 \cdot \sqrt{u1}\\ \mathbf{else}:\\ \;\;\;\;t\_1 \cdot \sqrt{t\_0}\\ \end{array} \end{array} \]
(FPCore (cosTheta_i u1 u2)
 :precision binary32
 (let* ((t_0 (- (log (- 1.0 u1)))) (t_1 (cos (* (* 2.0 (PI)) u2))))
   (if (<= t_0 0.00013080000644549727) (* t_1 (sqrt u1)) (* t_1 (sqrt t_0)))))
\begin{array}{l}

\\
\begin{array}{l}
t_0 := -\log \left(1 - u1\right)\\
t_1 := \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)\\
\mathbf{if}\;t\_0 \leq 0.00013080000644549727:\\
\;\;\;\;t\_1 \cdot \sqrt{u1}\\

\mathbf{else}:\\
\;\;\;\;t\_1 \cdot \sqrt{t\_0}\\


\end{array}
\end{array}
Derivation
  1. Split input into 2 regimes
  2. if (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1))) < 1.30800006e-4

    1. Initial program 39.7%

      \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. Add Preprocessing
    3. Taylor expanded in u1 around 0

      \[\leadsto \color{blue}{\left(\sqrt{u1} \cdot {\left(\sqrt{-1}\right)}^{2}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    4. Step-by-step derivation
      1. *-commutativeN/A

        \[\leadsto \color{blue}{\left({\left(\sqrt{-1}\right)}^{2} \cdot \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      2. unpow2N/A

        \[\leadsto \left(\color{blue}{\left(\sqrt{-1} \cdot \sqrt{-1}\right)} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      3. rem-square-sqrtN/A

        \[\leadsto \left(\color{blue}{-1} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      4. mul-1-negN/A

        \[\leadsto \color{blue}{\left(\mathsf{neg}\left(\sqrt{u1}\right)\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      5. lower-neg.f32N/A

        \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      6. lower-sqrt.f323.6

        \[\leadsto \left(-\color{blue}{\sqrt{u1}}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    5. Applied rewrites3.6%

      \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    6. Applied rewrites91.9%

      \[\leadsto \color{blue}{\cos \left(u2 \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)\right) \cdot \sqrt{u1}} \]

    if 1.30800006e-4 < (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1)))

    1. Initial program 89.3%

      \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. Add Preprocessing
  3. Recombined 2 regimes into one program.
  4. Final simplification90.8%

    \[\leadsto \begin{array}{l} \mathbf{if}\;-\log \left(1 - u1\right) \leq 0.00013080000644549727:\\ \;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}\\ \mathbf{else}:\\ \;\;\;\;\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{-\log \left(1 - u1\right)}\\ \end{array} \]
  5. Add Preprocessing

Alternative 4: 76.4% accurate, 1.5× speedup?

\[\begin{array}{l} \\ \left(\sqrt{u1} + \left(\left(0.16666666666666666 \cdot \sqrt{u1}\right) \cdot u1\right) \cdot u1\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \end{array} \]
(FPCore (cosTheta_i u1 u2)
 :precision binary32
 (*
  (+ (sqrt u1) (* (* (* 0.16666666666666666 (sqrt u1)) u1) u1))
  (cos (* (* 2.0 (PI)) u2))))
\begin{array}{l}

\\
\left(\sqrt{u1} + \left(\left(0.16666666666666666 \cdot \sqrt{u1}\right) \cdot u1\right) \cdot u1\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)
\end{array}
Derivation
  1. Initial program 59.8%

    \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
  2. Add Preprocessing
  3. Applied rewrites43.9%

    \[\leadsto \color{blue}{e^{\log \left(\mathsf{log1p}\left(u1\right)\right) \cdot 0.5}} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
  4. Taylor expanded in u1 around 0

    \[\leadsto \color{blue}{\left(\sqrt{u1} + {u1}^{2} \cdot \left(\frac{-1}{4} \cdot \sqrt{\frac{1}{u1}} + \frac{1}{6} \cdot \sqrt{u1}\right)\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
  5. Step-by-step derivation
    1. +-commutativeN/A

      \[\leadsto \color{blue}{\left({u1}^{2} \cdot \left(\frac{-1}{4} \cdot \sqrt{\frac{1}{u1}} + \frac{1}{6} \cdot \sqrt{u1}\right) + \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. *-commutativeN/A

      \[\leadsto \left(\color{blue}{\left(\frac{-1}{4} \cdot \sqrt{\frac{1}{u1}} + \frac{1}{6} \cdot \sqrt{u1}\right) \cdot {u1}^{2}} + \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    3. lower-fma.f32N/A

      \[\leadsto \color{blue}{\mathsf{fma}\left(\frac{-1}{4} \cdot \sqrt{\frac{1}{u1}} + \frac{1}{6} \cdot \sqrt{u1}, {u1}^{2}, \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    4. *-commutativeN/A

      \[\leadsto \mathsf{fma}\left(\color{blue}{\sqrt{\frac{1}{u1}} \cdot \frac{-1}{4}} + \frac{1}{6} \cdot \sqrt{u1}, {u1}^{2}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    5. lower-fma.f32N/A

      \[\leadsto \mathsf{fma}\left(\color{blue}{\mathsf{fma}\left(\sqrt{\frac{1}{u1}}, \frac{-1}{4}, \frac{1}{6} \cdot \sqrt{u1}\right)}, {u1}^{2}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    6. lower-sqrt.f32N/A

      \[\leadsto \mathsf{fma}\left(\mathsf{fma}\left(\color{blue}{\sqrt{\frac{1}{u1}}}, \frac{-1}{4}, \frac{1}{6} \cdot \sqrt{u1}\right), {u1}^{2}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    7. lower-/.f32N/A

      \[\leadsto \mathsf{fma}\left(\mathsf{fma}\left(\sqrt{\color{blue}{\frac{1}{u1}}}, \frac{-1}{4}, \frac{1}{6} \cdot \sqrt{u1}\right), {u1}^{2}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    8. lower-*.f32N/A

      \[\leadsto \mathsf{fma}\left(\mathsf{fma}\left(\sqrt{\frac{1}{u1}}, \frac{-1}{4}, \color{blue}{\frac{1}{6} \cdot \sqrt{u1}}\right), {u1}^{2}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    9. lower-sqrt.f32N/A

      \[\leadsto \mathsf{fma}\left(\mathsf{fma}\left(\sqrt{\frac{1}{u1}}, \frac{-1}{4}, \frac{1}{6} \cdot \color{blue}{\sqrt{u1}}\right), {u1}^{2}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    10. unpow2N/A

      \[\leadsto \mathsf{fma}\left(\mathsf{fma}\left(\sqrt{\frac{1}{u1}}, \frac{-1}{4}, \frac{1}{6} \cdot \sqrt{u1}\right), \color{blue}{u1 \cdot u1}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    11. lower-*.f32N/A

      \[\leadsto \mathsf{fma}\left(\mathsf{fma}\left(\sqrt{\frac{1}{u1}}, \frac{-1}{4}, \frac{1}{6} \cdot \sqrt{u1}\right), \color{blue}{u1 \cdot u1}, \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    12. lower-sqrt.f3275.9

      \[\leadsto \mathsf{fma}\left(\mathsf{fma}\left(\sqrt{\frac{1}{u1}}, -0.25, 0.16666666666666666 \cdot \sqrt{u1}\right), u1 \cdot u1, \color{blue}{\sqrt{u1}}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
  6. Applied rewrites75.9%

    \[\leadsto \color{blue}{\mathsf{fma}\left(\mathsf{fma}\left(\sqrt{\frac{1}{u1}}, -0.25, 0.16666666666666666 \cdot \sqrt{u1}\right), u1 \cdot u1, \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
  7. Step-by-step derivation
    1. Applied rewrites75.7%

      \[\leadsto \left(\left(\left(\sqrt{u1} \cdot 0.16666666666666666\right) \cdot u1\right) \cdot u1 + \color{blue}{\mathsf{fma}\left(\frac{-0.25}{\sqrt{u1}} \cdot u1, u1, \sqrt{u1}\right)}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    2. Taylor expanded in u1 around 0

      \[\leadsto \left(\left(\left(\sqrt{u1} \cdot \frac{1}{6}\right) \cdot u1\right) \cdot u1 + \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
    3. Step-by-step derivation
      1. Applied rewrites76.0%

        \[\leadsto \left(\left(\left(\sqrt{u1} \cdot 0.16666666666666666\right) \cdot u1\right) \cdot u1 + \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      2. Final simplification76.0%

        \[\leadsto \left(\sqrt{u1} + \left(\left(0.16666666666666666 \cdot \sqrt{u1}\right) \cdot u1\right) \cdot u1\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      3. Add Preprocessing

      Alternative 5: 76.3% accurate, 1.8× speedup?

      \[\begin{array}{l} \\ \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1} \end{array} \]
      (FPCore (cosTheta_i u1 u2)
       :precision binary32
       (* (cos (* (* 2.0 (PI)) u2)) (sqrt u1)))
      \begin{array}{l}
      
      \\
      \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1}
      \end{array}
      
      Derivation
      1. Initial program 59.8%

        \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      2. Add Preprocessing
      3. Taylor expanded in u1 around 0

        \[\leadsto \color{blue}{\left(\sqrt{u1} \cdot {\left(\sqrt{-1}\right)}^{2}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      4. Step-by-step derivation
        1. *-commutativeN/A

          \[\leadsto \color{blue}{\left({\left(\sqrt{-1}\right)}^{2} \cdot \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        2. unpow2N/A

          \[\leadsto \left(\color{blue}{\left(\sqrt{-1} \cdot \sqrt{-1}\right)} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        3. rem-square-sqrtN/A

          \[\leadsto \left(\color{blue}{-1} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        4. mul-1-negN/A

          \[\leadsto \color{blue}{\left(\mathsf{neg}\left(\sqrt{u1}\right)\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        5. lower-neg.f32N/A

          \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        6. lower-sqrt.f323.5

          \[\leadsto \left(-\color{blue}{\sqrt{u1}}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      5. Applied rewrites3.5%

        \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      6. Applied rewrites75.9%

        \[\leadsto \color{blue}{\cos \left(u2 \cdot \left(\mathsf{PI}\left(\right) \cdot 2\right)\right) \cdot \sqrt{u1}} \]
      7. Final simplification75.9%

        \[\leadsto \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \cdot \sqrt{u1} \]
      8. Add Preprocessing

      Alternative 6: 64.6% accurate, 11.6× speedup?

      \[\begin{array}{l} \\ 1 \cdot \sqrt{-\left(-u1\right)} \end{array} \]
      (FPCore (cosTheta_i u1 u2) :precision binary32 (* 1.0 (sqrt (- (- u1)))))
      float code(float cosTheta_i, float u1, float u2) {
      	return 1.0f * sqrtf(-(-u1));
      }
      
      real(4) function code(costheta_i, u1, u2)
          real(4), intent (in) :: costheta_i
          real(4), intent (in) :: u1
          real(4), intent (in) :: u2
          code = 1.0e0 * sqrt(-(-u1))
      end function
      
      function code(cosTheta_i, u1, u2)
      	return Float32(Float32(1.0) * sqrt(Float32(-Float32(-u1))))
      end
      
      function tmp = code(cosTheta_i, u1, u2)
      	tmp = single(1.0) * sqrt(-(-u1));
      end
      
      \begin{array}{l}
      
      \\
      1 \cdot \sqrt{-\left(-u1\right)}
      \end{array}
      
      Derivation
      1. Initial program 59.8%

        \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      2. Add Preprocessing
      3. Taylor expanded in u1 around 0

        \[\leadsto \sqrt{-\color{blue}{-1 \cdot u1}} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      4. Step-by-step derivation
        1. mul-1-negN/A

          \[\leadsto \sqrt{-\color{blue}{\left(\mathsf{neg}\left(u1\right)\right)}} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        2. lower-neg.f3275.9

          \[\leadsto \sqrt{-\color{blue}{\left(-u1\right)}} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      5. Applied rewrites75.9%

        \[\leadsto \sqrt{-\color{blue}{\left(-u1\right)}} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
      6. Taylor expanded in u2 around 0

        \[\leadsto \sqrt{-\left(-u1\right)} \cdot \color{blue}{1} \]
      7. Step-by-step derivation
        1. Applied rewrites62.6%

          \[\leadsto \sqrt{-\left(-u1\right)} \cdot \color{blue}{1} \]
        2. Final simplification62.6%

          \[\leadsto 1 \cdot \sqrt{-\left(-u1\right)} \]
        3. Add Preprocessing

        Alternative 7: 4.9% accurate, 12.8× speedup?

        \[\begin{array}{l} \\ \left(-\sqrt{u1}\right) \cdot 1 \end{array} \]
        (FPCore (cosTheta_i u1 u2) :precision binary32 (* (- (sqrt u1)) 1.0))
        float code(float cosTheta_i, float u1, float u2) {
        	return -sqrtf(u1) * 1.0f;
        }
        
        real(4) function code(costheta_i, u1, u2)
            real(4), intent (in) :: costheta_i
            real(4), intent (in) :: u1
            real(4), intent (in) :: u2
            code = -sqrt(u1) * 1.0e0
        end function
        
        function code(cosTheta_i, u1, u2)
        	return Float32(Float32(-sqrt(u1)) * Float32(1.0))
        end
        
        function tmp = code(cosTheta_i, u1, u2)
        	tmp = -sqrt(u1) * single(1.0);
        end
        
        \begin{array}{l}
        
        \\
        \left(-\sqrt{u1}\right) \cdot 1
        \end{array}
        
        Derivation
        1. Initial program 59.8%

          \[\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        2. Add Preprocessing
        3. Taylor expanded in u1 around 0

          \[\leadsto \color{blue}{\left(\sqrt{u1} \cdot {\left(\sqrt{-1}\right)}^{2}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        4. Step-by-step derivation
          1. *-commutativeN/A

            \[\leadsto \color{blue}{\left({\left(\sqrt{-1}\right)}^{2} \cdot \sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
          2. unpow2N/A

            \[\leadsto \left(\color{blue}{\left(\sqrt{-1} \cdot \sqrt{-1}\right)} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
          3. rem-square-sqrtN/A

            \[\leadsto \left(\color{blue}{-1} \cdot \sqrt{u1}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
          4. mul-1-negN/A

            \[\leadsto \color{blue}{\left(\mathsf{neg}\left(\sqrt{u1}\right)\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
          5. lower-neg.f32N/A

            \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
          6. lower-sqrt.f323.5

            \[\leadsto \left(-\color{blue}{\sqrt{u1}}\right) \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        5. Applied rewrites3.5%

          \[\leadsto \color{blue}{\left(-\sqrt{u1}\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right) \]
        6. Taylor expanded in u2 around 0

          \[\leadsto \left(-\sqrt{u1}\right) \cdot \color{blue}{1} \]
        7. Step-by-step derivation
          1. Applied rewrites5.7%

            \[\leadsto \left(-\sqrt{u1}\right) \cdot \color{blue}{1} \]
          2. Add Preprocessing

          Reproduce

          ?
          herbie shell --seed 2024285 
          (FPCore (cosTheta_i u1 u2)
            :name "Beckmann Sample, near normal, slope_x"
            :precision binary32
            :pre (and (and (and (> cosTheta_i 0.9999) (<= cosTheta_i 1.0)) (and (<= 2.328306437e-10 u1) (<= u1 1.0))) (and (<= 2.328306437e-10 u2) (<= u2 1.0)))
            (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 (PI)) u2))))