UniformSampleCone, x

Percentage Accurate: 57.5% → 99.0%
Time: 8.1s
Alternatives: 13
Speedup: 5.8×

Specification

?
\[\left(\left(2.328306437 \cdot 10^{-10} \leq ux \land ux \leq 1\right) \land \left(2.328306437 \cdot 10^{-10} \leq uy \land uy \leq 1\right)\right) \land \left(0 \leq maxCos \land maxCos \leq 1\right)\]
\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(1 - ux\right) + ux \cdot maxCos\\ \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \end{array} \end{array} \]
(FPCore (ux uy maxCos)
 :precision binary32
 (let* ((t_0 (+ (- 1.0 ux) (* ux maxCos))))
   (* (cos (* (* uy 2.0) (PI))) (sqrt (- 1.0 (* t_0 t_0))))))
\begin{array}{l}

\\
\begin{array}{l}
t_0 := \left(1 - ux\right) + ux \cdot maxCos\\
\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0}
\end{array}
\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 13 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: 57.5% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(1 - ux\right) + ux \cdot maxCos\\ \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \end{array} \end{array} \]
(FPCore (ux uy maxCos)
 :precision binary32
 (let* ((t_0 (+ (- 1.0 ux) (* ux maxCos))))
   (* (cos (* (* uy 2.0) (PI))) (sqrt (- 1.0 (* t_0 t_0))))))
\begin{array}{l}

\\
\begin{array}{l}
t_0 := \left(1 - ux\right) + ux \cdot maxCos\\
\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0}
\end{array}
\end{array}

Alternative 1: 99.0% accurate, 0.6× speedup?

\[\begin{array}{l} \\ \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(2 - {\left(maxCos - 1\right)}^{2} \cdot ux, ux, \left(-2 \cdot maxCos\right) \cdot ux\right)} \end{array} \]
(FPCore (ux uy maxCos)
 :precision binary32
 (*
  (cos (* (* uy 2.0) (PI)))
  (sqrt
   (fma (- 2.0 (* (pow (- maxCos 1.0) 2.0) ux)) ux (* (* -2.0 maxCos) ux)))))
\begin{array}{l}

\\
\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(2 - {\left(maxCos - 1\right)}^{2} \cdot ux, ux, \left(-2 \cdot maxCos\right) \cdot ux\right)}
\end{array}
Derivation
  1. Initial program 54.0%

    \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
  2. Add Preprocessing
  3. Taylor expanded in ux around 0

    \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(\left(2 + -1 \cdot \left(ux \cdot {\left(maxCos - 1\right)}^{2}\right)\right) - 2 \cdot maxCos\right)}} \]
  4. Step-by-step derivation
    1. Applied rewrites99.1%

      \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux}} \]
    2. Step-by-step derivation
      1. Applied rewrites99.1%

        \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(2 - {\left(maxCos - 1\right)}^{2} \cdot ux, \color{blue}{ux}, \left(-2 \cdot maxCos\right) \cdot ux\right)} \]
      2. Add Preprocessing

      Alternative 2: 99.0% accurate, 0.6× speedup?

      \[\begin{array}{l} \\ \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \end{array} \]
      (FPCore (ux uy maxCos)
       :precision binary32
       (*
        (cos (* (* uy 2.0) (PI)))
        (sqrt (* (fma -2.0 maxCos (- 2.0 (* (pow (- maxCos 1.0) 2.0) ux))) ux))))
      \begin{array}{l}
      
      \\
      \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux}
      \end{array}
      
      Derivation
      1. Initial program 54.0%

        \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
      2. Add Preprocessing
      3. Taylor expanded in ux around 0

        \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(\left(2 + -1 \cdot \left(ux \cdot {\left(maxCos - 1\right)}^{2}\right)\right) - 2 \cdot maxCos\right)}} \]
      4. Step-by-step derivation
        1. Applied rewrites99.1%

          \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux}} \]
        2. Add Preprocessing

        Alternative 3: 83.9% accurate, 0.7× speedup?

        \[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(1 - ux\right) + ux \cdot maxCos\\ \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.004000000189989805:\\ \;\;\;\;\mathsf{fma}\left(-2 \cdot \left(uy \cdot uy\right), \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), 1\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}\\ \mathbf{else}:\\ \;\;\;\;1 \cdot \sqrt{\left(\left(\mathsf{fma}\left(\left(1 - maxCos\right) \cdot \left(maxCos - 1\right), ux, \left(-maxCos\right) - -1\right) - -1\right) - maxCos\right) \cdot ux}\\ \end{array} \end{array} \]
        (FPCore (ux uy maxCos)
         :precision binary32
         (let* ((t_0 (+ (- 1.0 ux) (* ux maxCos))))
           (if (<=
                (* (cos (* (* uy 2.0) (PI))) (sqrt (- 1.0 (* t_0 t_0))))
                0.004000000189989805)
             (*
              (fma (* -2.0 (* uy uy)) (* (PI) (PI)) 1.0)
              (sqrt (* (fma -2.0 maxCos 2.0) ux)))
             (*
              1.0
              (sqrt
               (*
                (-
                 (-
                  (fma (* (- 1.0 maxCos) (- maxCos 1.0)) ux (- (- maxCos) -1.0))
                  -1.0)
                 maxCos)
                ux))))))
        \begin{array}{l}
        
        \\
        \begin{array}{l}
        t_0 := \left(1 - ux\right) + ux \cdot maxCos\\
        \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.004000000189989805:\\
        \;\;\;\;\mathsf{fma}\left(-2 \cdot \left(uy \cdot uy\right), \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), 1\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}\\
        
        \mathbf{else}:\\
        \;\;\;\;1 \cdot \sqrt{\left(\left(\mathsf{fma}\left(\left(1 - maxCos\right) \cdot \left(maxCos - 1\right), ux, \left(-maxCos\right) - -1\right) - -1\right) - maxCos\right) \cdot ux}\\
        
        
        \end{array}
        \end{array}
        
        Derivation
        1. Split input into 2 regimes
        2. if (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)))))) < 0.00400000019

          1. Initial program 29.2%

            \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
          2. Add Preprocessing
          3. Taylor expanded in ux around 0

            \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(2 - 2 \cdot maxCos\right)}} \]
          4. Step-by-step derivation
            1. Applied rewrites94.6%

              \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}} \]
            2. Taylor expanded in uy around 0

              \[\leadsto \color{blue}{\left(1 + -2 \cdot \left({uy}^{2} \cdot {\mathsf{PI}\left(\right)}^{2}\right)\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux} \]
            3. Step-by-step derivation
              1. Applied rewrites80.6%

                \[\leadsto \color{blue}{\mathsf{fma}\left(-2 \cdot \left(uy \cdot uy\right), \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), 1\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux} \]

              if 0.00400000019 < (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos))))))

              1. Initial program 82.2%

                \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
              2. Add Preprocessing
              3. Taylor expanded in uy around 0

                \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
              4. Step-by-step derivation
                1. Applied rewrites72.4%

                  \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                2. Taylor expanded in maxCos around inf

                  \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(maxCos \cdot \left(\left(ux + \frac{1}{maxCos}\right) - \frac{ux}{maxCos}\right)\right)}} \]
                3. Step-by-step derivation
                  1. Applied rewrites71.3%

                    \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                  2. Step-by-step derivation
                    1. lift--.f32N/A

                      \[\leadsto 1 \cdot \sqrt{\color{blue}{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                    2. lift-*.f32N/A

                      \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                    3. fp-cancel-sub-sign-invN/A

                      \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                    4. +-commutativeN/A

                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right) + 1}} \]
                    5. lower-fma.f32N/A

                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)}} \]
                    6. lower-neg.f3271.3

                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(\color{blue}{-\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                    7. lift-+.f32N/A

                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                    8. +-commutativeN/A

                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(ux \cdot maxCos + \left(1 - ux\right)\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                    9. lift-*.f32N/A

                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\left(\color{blue}{ux \cdot maxCos} + \left(1 - ux\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                    10. lower-fma.f3271.3

                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\mathsf{fma}\left(ux, maxCos, 1 - ux\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                  3. Applied rewrites71.3%

                    \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-\mathsf{fma}\left(ux, maxCos, 1 - ux\right), \left(\frac{1 - ux}{maxCos} + ux\right) \cdot maxCos, 1\right)}} \]
                  4. Taylor expanded in ux around 0

                    \[\leadsto 1 \cdot \sqrt{\color{blue}{ux \cdot \left(\left(1 + \left(-1 \cdot \left(maxCos - 1\right) + ux \cdot \left(\left(1 - maxCos\right) \cdot \left(maxCos - 1\right)\right)\right)\right) - maxCos\right)}} \]
                  5. Step-by-step derivation
                    1. Applied rewrites84.7%

                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\left(\mathsf{fma}\left(\left(1 - maxCos\right) \cdot \left(maxCos - 1\right), ux, -\left(maxCos - 1\right)\right) + 1\right) - maxCos\right) \cdot ux}} \]
                  6. Recombined 2 regimes into one program.
                  7. Final simplification82.5%

                    \[\leadsto \begin{array}{l} \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \leq 0.004000000189989805:\\ \;\;\;\;\mathsf{fma}\left(-2 \cdot \left(uy \cdot uy\right), \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), 1\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}\\ \mathbf{else}:\\ \;\;\;\;1 \cdot \sqrt{\left(\left(\mathsf{fma}\left(\left(1 - maxCos\right) \cdot \left(maxCos - 1\right), ux, \left(-maxCos\right) - -1\right) - -1\right) - maxCos\right) \cdot ux}\\ \end{array} \]
                  8. Add Preprocessing

                  Alternative 4: 80.3% accurate, 0.7× speedup?

                  \[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(1 - ux\right) + ux \cdot maxCos\\ \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.02199999988079071:\\ \;\;\;\;\mathsf{fma}\left(-2 \cdot \left(uy \cdot uy\right), \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), 1\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}\\ \mathbf{else}:\\ \;\;\;\;\sqrt{\mathsf{fma}\left(ux - \mathsf{fma}\left(maxCos, ux, 1\right), \mathsf{fma}\left(maxCos, ux, 1 - ux\right), 1\right)}\\ \end{array} \end{array} \]
                  (FPCore (ux uy maxCos)
                   :precision binary32
                   (let* ((t_0 (+ (- 1.0 ux) (* ux maxCos))))
                     (if (<=
                          (* (cos (* (* uy 2.0) (PI))) (sqrt (- 1.0 (* t_0 t_0))))
                          0.02199999988079071)
                       (*
                        (fma (* -2.0 (* uy uy)) (* (PI) (PI)) 1.0)
                        (sqrt (* (fma -2.0 maxCos 2.0) ux)))
                       (sqrt (fma (- ux (fma maxCos ux 1.0)) (fma maxCos ux (- 1.0 ux)) 1.0)))))
                  \begin{array}{l}
                  
                  \\
                  \begin{array}{l}
                  t_0 := \left(1 - ux\right) + ux \cdot maxCos\\
                  \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.02199999988079071:\\
                  \;\;\;\;\mathsf{fma}\left(-2 \cdot \left(uy \cdot uy\right), \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), 1\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}\\
                  
                  \mathbf{else}:\\
                  \;\;\;\;\sqrt{\mathsf{fma}\left(ux - \mathsf{fma}\left(maxCos, ux, 1\right), \mathsf{fma}\left(maxCos, ux, 1 - ux\right), 1\right)}\\
                  
                  
                  \end{array}
                  \end{array}
                  
                  Derivation
                  1. Split input into 2 regimes
                  2. if (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)))))) < 0.0219999999

                    1. Initial program 37.3%

                      \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                    2. Add Preprocessing
                    3. Taylor expanded in ux around 0

                      \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(2 - 2 \cdot maxCos\right)}} \]
                    4. Step-by-step derivation
                      1. Applied rewrites90.9%

                        \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}} \]
                      2. Taylor expanded in uy around 0

                        \[\leadsto \color{blue}{\left(1 + -2 \cdot \left({uy}^{2} \cdot {\mathsf{PI}\left(\right)}^{2}\right)\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux} \]
                      3. Step-by-step derivation
                        1. Applied rewrites78.6%

                          \[\leadsto \color{blue}{\mathsf{fma}\left(-2 \cdot \left(uy \cdot uy\right), \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), 1\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux} \]

                        if 0.0219999999 < (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos))))))

                        1. Initial program 89.0%

                          \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                        2. Add Preprocessing
                        3. Taylor expanded in uy around 0

                          \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                        4. Step-by-step derivation
                          1. Applied rewrites77.6%

                            \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                          2. Taylor expanded in maxCos around inf

                            \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(maxCos \cdot \left(\left(ux + \frac{1}{maxCos}\right) - \frac{ux}{maxCos}\right)\right)}} \]
                          3. Step-by-step derivation
                            1. Applied rewrites77.2%

                              \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                            2. Step-by-step derivation
                              1. lift--.f32N/A

                                \[\leadsto 1 \cdot \sqrt{\color{blue}{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                              2. lift-*.f32N/A

                                \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                              3. fp-cancel-sub-sign-invN/A

                                \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                              4. +-commutativeN/A

                                \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right) + 1}} \]
                              5. lower-fma.f32N/A

                                \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)}} \]
                              6. lower-neg.f3277.2

                                \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(\color{blue}{-\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                              7. lift-+.f32N/A

                                \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                              8. +-commutativeN/A

                                \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(ux \cdot maxCos + \left(1 - ux\right)\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                              9. lift-*.f32N/A

                                \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\left(\color{blue}{ux \cdot maxCos} + \left(1 - ux\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                              10. lower-fma.f3277.2

                                \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\mathsf{fma}\left(ux, maxCos, 1 - ux\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                            3. Applied rewrites77.2%

                              \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-\mathsf{fma}\left(ux, maxCos, 1 - ux\right), \left(\frac{1 - ux}{maxCos} + ux\right) \cdot maxCos, 1\right)}} \]
                            4. Taylor expanded in uy around 0

                              \[\leadsto \color{blue}{\sqrt{1 + \left(ux - \left(1 + maxCos \cdot ux\right)\right) \cdot \left(\left(1 + maxCos \cdot ux\right) - ux\right)}} \]
                            5. Step-by-step derivation
                              1. Applied rewrites78.2%

                                \[\leadsto \color{blue}{\sqrt{\mathsf{fma}\left(ux - \mathsf{fma}\left(maxCos, ux, 1\right), \mathsf{fma}\left(maxCos, ux, 1 - ux\right), 1\right)}} \]
                            6. Recombined 2 regimes into one program.
                            7. Add Preprocessing

                            Alternative 5: 74.2% accurate, 0.8× speedup?

                            \[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(1 - ux\right) + ux \cdot maxCos\\ \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.025200000032782555:\\ \;\;\;\;1 \cdot \sqrt{\left(\mathsf{fma}\left(-1, maxCos - 1, 1\right) - maxCos\right) \cdot ux}\\ \mathbf{else}:\\ \;\;\;\;1 \cdot \sqrt{\mathsf{fma}\left(ux - 1, 1 - ux, 1\right)}\\ \end{array} \end{array} \]
                            (FPCore (ux uy maxCos)
                             :precision binary32
                             (let* ((t_0 (+ (- 1.0 ux) (* ux maxCos))))
                               (if (<=
                                    (* (cos (* (* uy 2.0) (PI))) (sqrt (- 1.0 (* t_0 t_0))))
                                    0.025200000032782555)
                                 (* 1.0 (sqrt (* (- (fma -1.0 (- maxCos 1.0) 1.0) maxCos) ux)))
                                 (* 1.0 (sqrt (fma (- ux 1.0) (- 1.0 ux) 1.0))))))
                            \begin{array}{l}
                            
                            \\
                            \begin{array}{l}
                            t_0 := \left(1 - ux\right) + ux \cdot maxCos\\
                            \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.025200000032782555:\\
                            \;\;\;\;1 \cdot \sqrt{\left(\mathsf{fma}\left(-1, maxCos - 1, 1\right) - maxCos\right) \cdot ux}\\
                            
                            \mathbf{else}:\\
                            \;\;\;\;1 \cdot \sqrt{\mathsf{fma}\left(ux - 1, 1 - ux, 1\right)}\\
                            
                            
                            \end{array}
                            \end{array}
                            
                            Derivation
                            1. Split input into 2 regimes
                            2. if (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)))))) < 0.0252

                              1. Initial program 37.7%

                                \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                              2. Add Preprocessing
                              3. Taylor expanded in uy around 0

                                \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                              4. Step-by-step derivation
                                1. Applied rewrites30.3%

                                  \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                2. Taylor expanded in maxCos around inf

                                  \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(maxCos \cdot \left(\left(ux + \frac{1}{maxCos}\right) - \frac{ux}{maxCos}\right)\right)}} \]
                                3. Step-by-step derivation
                                  1. Applied rewrites30.9%

                                    \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                  2. Step-by-step derivation
                                    1. lift--.f32N/A

                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                    2. lift-*.f32N/A

                                      \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                    3. fp-cancel-sub-sign-invN/A

                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                    4. +-commutativeN/A

                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right) + 1}} \]
                                    5. lower-fma.f32N/A

                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)}} \]
                                    6. lower-neg.f3230.9

                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(\color{blue}{-\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                    7. lift-+.f32N/A

                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                    8. +-commutativeN/A

                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(ux \cdot maxCos + \left(1 - ux\right)\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                    9. lift-*.f32N/A

                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\left(\color{blue}{ux \cdot maxCos} + \left(1 - ux\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                    10. lower-fma.f3230.9

                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\mathsf{fma}\left(ux, maxCos, 1 - ux\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                  3. Applied rewrites30.9%

                                    \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-\mathsf{fma}\left(ux, maxCos, 1 - ux\right), \left(\frac{1 - ux}{maxCos} + ux\right) \cdot maxCos, 1\right)}} \]
                                  4. Taylor expanded in ux around 0

                                    \[\leadsto 1 \cdot \sqrt{\color{blue}{ux \cdot \left(\left(1 + -1 \cdot \left(maxCos - 1\right)\right) - maxCos\right)}} \]
                                  5. Step-by-step derivation
                                    1. Applied rewrites69.6%

                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{fma}\left(-1, maxCos - 1, 1\right) - maxCos\right) \cdot ux}} \]

                                    if 0.0252 < (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos))))))

                                    1. Initial program 89.4%

                                      \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                    2. Add Preprocessing
                                    3. Taylor expanded in uy around 0

                                      \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                    4. Step-by-step derivation
                                      1. Applied rewrites77.7%

                                        \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                      2. Taylor expanded in maxCos around inf

                                        \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(maxCos \cdot \left(\left(ux + \frac{1}{maxCos}\right) - \frac{ux}{maxCos}\right)\right)}} \]
                                      3. Step-by-step derivation
                                        1. Applied rewrites77.3%

                                          \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                        2. Step-by-step derivation
                                          1. lift--.f32N/A

                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                          2. lift-*.f32N/A

                                            \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                          3. fp-cancel-sub-sign-invN/A

                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                          4. +-commutativeN/A

                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right) + 1}} \]
                                          5. lower-fma.f32N/A

                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)}} \]
                                          6. lower-neg.f3277.4

                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(\color{blue}{-\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                          7. lift-+.f32N/A

                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                          8. +-commutativeN/A

                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(ux \cdot maxCos + \left(1 - ux\right)\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                          9. lift-*.f32N/A

                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\left(\color{blue}{ux \cdot maxCos} + \left(1 - ux\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                          10. lower-fma.f3277.4

                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\mathsf{fma}\left(ux, maxCos, 1 - ux\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                        3. Applied rewrites77.4%

                                          \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-\mathsf{fma}\left(ux, maxCos, 1 - ux\right), \left(\frac{1 - ux}{maxCos} + ux\right) \cdot maxCos, 1\right)}} \]
                                        4. Taylor expanded in maxCos around 0

                                          \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(1 - ux\right) \cdot \left(ux - 1\right)}} \]
                                        5. Step-by-step derivation
                                          1. Applied rewrites75.2%

                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(ux - 1, 1 - ux, 1\right)}} \]
                                        6. Recombined 2 regimes into one program.
                                        7. Add Preprocessing

                                        Alternative 6: 74.2% accurate, 0.8× speedup?

                                        \[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(1 - ux\right) + ux \cdot maxCos\\ \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.025200000032782555:\\ \;\;\;\;1 \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}\\ \mathbf{else}:\\ \;\;\;\;1 \cdot \sqrt{\mathsf{fma}\left(ux - 1, 1 - ux, 1\right)}\\ \end{array} \end{array} \]
                                        (FPCore (ux uy maxCos)
                                         :precision binary32
                                         (let* ((t_0 (+ (- 1.0 ux) (* ux maxCos))))
                                           (if (<=
                                                (* (cos (* (* uy 2.0) (PI))) (sqrt (- 1.0 (* t_0 t_0))))
                                                0.025200000032782555)
                                             (* 1.0 (sqrt (* (fma -2.0 maxCos 2.0) ux)))
                                             (* 1.0 (sqrt (fma (- ux 1.0) (- 1.0 ux) 1.0))))))
                                        \begin{array}{l}
                                        
                                        \\
                                        \begin{array}{l}
                                        t_0 := \left(1 - ux\right) + ux \cdot maxCos\\
                                        \mathbf{if}\;\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - t\_0 \cdot t\_0} \leq 0.025200000032782555:\\
                                        \;\;\;\;1 \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}\\
                                        
                                        \mathbf{else}:\\
                                        \;\;\;\;1 \cdot \sqrt{\mathsf{fma}\left(ux - 1, 1 - ux, 1\right)}\\
                                        
                                        
                                        \end{array}
                                        \end{array}
                                        
                                        Derivation
                                        1. Split input into 2 regimes
                                        2. if (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)))))) < 0.0252

                                          1. Initial program 37.7%

                                            \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                          2. Add Preprocessing
                                          3. Taylor expanded in uy around 0

                                            \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                          4. Step-by-step derivation
                                            1. Applied rewrites30.3%

                                              \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                            2. Taylor expanded in ux around 0

                                              \[\leadsto 1 \cdot \sqrt{\color{blue}{ux \cdot \left(2 - 2 \cdot maxCos\right)}} \]
                                            3. Step-by-step derivation
                                              1. Applied rewrites69.6%

                                                \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}} \]

                                              if 0.0252 < (*.f32 (cos.f32 (*.f32 (*.f32 uy #s(literal 2 binary32)) (PI.f32))) (sqrt.f32 (-.f32 #s(literal 1 binary32) (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos))))))

                                              1. Initial program 89.4%

                                                \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                              2. Add Preprocessing
                                              3. Taylor expanded in uy around 0

                                                \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                              4. Step-by-step derivation
                                                1. Applied rewrites77.7%

                                                  \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                2. Taylor expanded in maxCos around inf

                                                  \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(maxCos \cdot \left(\left(ux + \frac{1}{maxCos}\right) - \frac{ux}{maxCos}\right)\right)}} \]
                                                3. Step-by-step derivation
                                                  1. Applied rewrites77.3%

                                                    \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                  2. Step-by-step derivation
                                                    1. lift--.f32N/A

                                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                    2. lift-*.f32N/A

                                                      \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                    3. fp-cancel-sub-sign-invN/A

                                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                    4. +-commutativeN/A

                                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right) + 1}} \]
                                                    5. lower-fma.f32N/A

                                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)}} \]
                                                    6. lower-neg.f3277.4

                                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(\color{blue}{-\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                    7. lift-+.f32N/A

                                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                    8. +-commutativeN/A

                                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(ux \cdot maxCos + \left(1 - ux\right)\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                    9. lift-*.f32N/A

                                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\left(\color{blue}{ux \cdot maxCos} + \left(1 - ux\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                    10. lower-fma.f3277.4

                                                      \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\mathsf{fma}\left(ux, maxCos, 1 - ux\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                  3. Applied rewrites77.4%

                                                    \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-\mathsf{fma}\left(ux, maxCos, 1 - ux\right), \left(\frac{1 - ux}{maxCos} + ux\right) \cdot maxCos, 1\right)}} \]
                                                  4. Taylor expanded in maxCos around 0

                                                    \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(1 - ux\right) \cdot \left(ux - 1\right)}} \]
                                                  5. Step-by-step derivation
                                                    1. Applied rewrites75.2%

                                                      \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(ux - 1, 1 - ux, 1\right)}} \]
                                                  6. Recombined 2 regimes into one program.
                                                  7. Add Preprocessing

                                                  Alternative 7: 99.0% accurate, 1.0× speedup?

                                                  \[\begin{array}{l} \\ \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(-maxCos, \mathsf{fma}\left(\mathsf{fma}\left(-2, ux, 2\right), ux, \left(maxCos \cdot ux\right) \cdot ux\right), \left(2 - ux\right) \cdot ux\right)} \end{array} \]
                                                  (FPCore (ux uy maxCos)
                                                   :precision binary32
                                                   (*
                                                    (cos (* (* uy 2.0) (PI)))
                                                    (sqrt
                                                     (fma
                                                      (- maxCos)
                                                      (fma (fma -2.0 ux 2.0) ux (* (* maxCos ux) ux))
                                                      (* (- 2.0 ux) ux)))))
                                                  \begin{array}{l}
                                                  
                                                  \\
                                                  \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(-maxCos, \mathsf{fma}\left(\mathsf{fma}\left(-2, ux, 2\right), ux, \left(maxCos \cdot ux\right) \cdot ux\right), \left(2 - ux\right) \cdot ux\right)}
                                                  \end{array}
                                                  
                                                  Derivation
                                                  1. Initial program 54.0%

                                                    \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                  2. Add Preprocessing
                                                  3. Taylor expanded in ux around 0

                                                    \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(\left(2 + -1 \cdot \left(ux \cdot {\left(maxCos - 1\right)}^{2}\right)\right) - 2 \cdot maxCos\right)}} \]
                                                  4. Step-by-step derivation
                                                    1. Applied rewrites99.1%

                                                      \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux}} \]
                                                    2. Taylor expanded in maxCos around 0

                                                      \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{maxCos \cdot \left(-1 \cdot \left(maxCos \cdot {ux}^{2}\right) + -1 \cdot \left(ux \cdot \left(2 + -2 \cdot ux\right)\right)\right) + \color{blue}{ux \cdot \left(2 - ux\right)}} \]
                                                    3. Step-by-step derivation
                                                      1. Applied rewrites99.1%

                                                        \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(-maxCos, \color{blue}{\mathsf{fma}\left(\mathsf{fma}\left(-2, ux, 2\right), ux, \left(maxCos \cdot ux\right) \cdot ux\right)}, \left(2 - ux\right) \cdot ux\right)} \]
                                                      2. Add Preprocessing

                                                      Alternative 8: 98.4% accurate, 1.0× speedup?

                                                      \[\begin{array}{l} \\ \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(2 - ux, ux, \left(\left(\left(1 - ux\right) \cdot ux\right) \cdot maxCos\right) \cdot -2\right)} \end{array} \]
                                                      (FPCore (ux uy maxCos)
                                                       :precision binary32
                                                       (*
                                                        (cos (* (* uy 2.0) (PI)))
                                                        (sqrt (fma (- 2.0 ux) ux (* (* (* (- 1.0 ux) ux) maxCos) -2.0)))))
                                                      \begin{array}{l}
                                                      
                                                      \\
                                                      \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(2 - ux, ux, \left(\left(\left(1 - ux\right) \cdot ux\right) \cdot maxCos\right) \cdot -2\right)}
                                                      \end{array}
                                                      
                                                      Derivation
                                                      1. Initial program 54.0%

                                                        \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                      2. Add Preprocessing
                                                      3. Taylor expanded in ux around 0

                                                        \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(\left(2 + -1 \cdot \left(ux \cdot {\left(maxCos - 1\right)}^{2}\right)\right) - 2 \cdot maxCos\right)}} \]
                                                      4. Step-by-step derivation
                                                        1. Applied rewrites99.1%

                                                          \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux}} \]
                                                        2. Taylor expanded in maxCos around 0

                                                          \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{-1 \cdot \left(maxCos \cdot \left(ux \cdot \left(2 + -2 \cdot ux\right)\right)\right) + \color{blue}{ux \cdot \left(2 - ux\right)}} \]
                                                        3. Step-by-step derivation
                                                          1. Applied rewrites98.3%

                                                            \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\mathsf{fma}\left(2 - ux, \color{blue}{ux}, \left(\left(\left(1 - ux\right) \cdot ux\right) \cdot maxCos\right) \cdot -2\right)} \]
                                                          2. Add Preprocessing

                                                          Alternative 9: 93.1% accurate, 1.1× speedup?

                                                          \[\begin{array}{l} \\ \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), 2 \cdot uy, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{\left(2 - ux\right) \cdot ux} \end{array} \]
                                                          (FPCore (ux uy maxCos)
                                                           :precision binary32
                                                           (* (sin (fma (- (PI)) (* 2.0 uy) (/ (PI) 2.0))) (sqrt (* (- 2.0 ux) ux))))
                                                          \begin{array}{l}
                                                          
                                                          \\
                                                          \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), 2 \cdot uy, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{\left(2 - ux\right) \cdot ux}
                                                          \end{array}
                                                          
                                                          Derivation
                                                          1. Initial program 54.0%

                                                            \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                          2. Add Preprocessing
                                                          3. Taylor expanded in ux around 0

                                                            \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(\left(2 + -1 \cdot \left(ux \cdot {\left(maxCos - 1\right)}^{2}\right)\right) - 2 \cdot maxCos\right)}} \]
                                                          4. Step-by-step derivation
                                                            1. Applied rewrites99.1%

                                                              \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux}} \]
                                                            2. Step-by-step derivation
                                                              1. lift-cos.f32N/A

                                                                \[\leadsto \color{blue}{\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              2. cos-neg-revN/A

                                                                \[\leadsto \color{blue}{\cos \left(\mathsf{neg}\left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right)\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              3. sin-+PI/2-revN/A

                                                                \[\leadsto \color{blue}{\sin \left(\left(\mathsf{neg}\left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right)\right) + \frac{\mathsf{PI}\left(\right)}{2}\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              4. lower-sin.f32N/A

                                                                \[\leadsto \color{blue}{\sin \left(\left(\mathsf{neg}\left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right)\right) + \frac{\mathsf{PI}\left(\right)}{2}\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              5. lift-*.f32N/A

                                                                \[\leadsto \sin \left(\left(\mathsf{neg}\left(\color{blue}{\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)}\right)\right) + \frac{\mathsf{PI}\left(\right)}{2}\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              6. *-commutativeN/A

                                                                \[\leadsto \sin \left(\left(\mathsf{neg}\left(\color{blue}{\mathsf{PI}\left(\right) \cdot \left(uy \cdot 2\right)}\right)\right) + \frac{\mathsf{PI}\left(\right)}{2}\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              7. distribute-lft-neg-inN/A

                                                                \[\leadsto \sin \left(\color{blue}{\left(\mathsf{neg}\left(\mathsf{PI}\left(\right)\right)\right) \cdot \left(uy \cdot 2\right)} + \frac{\mathsf{PI}\left(\right)}{2}\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              8. lower-fma.f32N/A

                                                                \[\leadsto \sin \color{blue}{\left(\mathsf{fma}\left(\mathsf{neg}\left(\mathsf{PI}\left(\right)\right), uy \cdot 2, \frac{\mathsf{PI}\left(\right)}{2}\right)\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              9. lower-neg.f32N/A

                                                                \[\leadsto \sin \left(\mathsf{fma}\left(\color{blue}{-\mathsf{PI}\left(\right)}, uy \cdot 2, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              10. lift-*.f32N/A

                                                                \[\leadsto \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), \color{blue}{uy \cdot 2}, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              11. *-commutativeN/A

                                                                \[\leadsto \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), \color{blue}{2 \cdot uy}, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              12. lower-*.f32N/A

                                                                \[\leadsto \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), \color{blue}{2 \cdot uy}, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              13. lift-PI.f32N/A

                                                                \[\leadsto \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), 2 \cdot uy, \frac{\color{blue}{\mathsf{PI}\left(\right)}}{2}\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                              14. lower-/.f3299.1

                                                                \[\leadsto \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), 2 \cdot uy, \color{blue}{\frac{\mathsf{PI}\left(\right)}{2}}\right)\right) \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                            3. Applied rewrites99.1%

                                                              \[\leadsto \color{blue}{\sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), 2 \cdot uy, \frac{\mathsf{PI}\left(\right)}{2}\right)\right)} \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux} \]
                                                            4. Taylor expanded in maxCos around 0

                                                              \[\leadsto \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), 2 \cdot uy, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{ux \cdot \color{blue}{\left(2 - ux\right)}} \]
                                                            5. Step-by-step derivation
                                                              1. Applied rewrites93.5%

                                                                \[\leadsto \sin \left(\mathsf{fma}\left(-\mathsf{PI}\left(\right), 2 \cdot uy, \frac{\mathsf{PI}\left(\right)}{2}\right)\right) \cdot \sqrt{\left(2 - ux\right) \cdot \color{blue}{ux}} \]
                                                              2. Add Preprocessing

                                                              Alternative 10: 92.9% accurate, 1.2× speedup?

                                                              \[\begin{array}{l} \\ \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\left(2 - ux\right) \cdot ux} \end{array} \]
                                                              (FPCore (ux uy maxCos)
                                                               :precision binary32
                                                               (* (cos (* (* uy 2.0) (PI))) (sqrt (* (- 2.0 ux) ux))))
                                                              \begin{array}{l}
                                                              
                                                              \\
                                                              \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\left(2 - ux\right) \cdot ux}
                                                              \end{array}
                                                              
                                                              Derivation
                                                              1. Initial program 54.0%

                                                                \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                              2. Add Preprocessing
                                                              3. Taylor expanded in ux around 0

                                                                \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{ux \cdot \left(\left(2 + -1 \cdot \left(ux \cdot {\left(maxCos - 1\right)}^{2}\right)\right) - 2 \cdot maxCos\right)}} \]
                                                              4. Step-by-step derivation
                                                                1. Applied rewrites99.1%

                                                                  \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2 - {\left(maxCos - 1\right)}^{2} \cdot ux\right) \cdot ux}} \]
                                                                2. Taylor expanded in maxCos around 0

                                                                  \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{ux \cdot \color{blue}{\left(2 - ux\right)}} \]
                                                                3. Step-by-step derivation
                                                                  1. Applied rewrites93.4%

                                                                    \[\leadsto \cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{\left(2 - ux\right) \cdot \color{blue}{ux}} \]
                                                                  2. Add Preprocessing

                                                                  Alternative 11: 75.5% accurate, 2.3× speedup?

                                                                  \[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(1 - ux\right) + ux \cdot maxCos\\ \mathbf{if}\;t\_0 \cdot t\_0 \leq 0.9996600151062012:\\ \;\;\;\;\sqrt{\mathsf{fma}\left(ux - \mathsf{fma}\left(maxCos, ux, 1\right), \mathsf{fma}\left(maxCos, ux, 1 - ux\right), 1\right)}\\ \mathbf{else}:\\ \;\;\;\;1 \cdot \sqrt{\left(\mathsf{fma}\left(-1, maxCos - 1, 1\right) - maxCos\right) \cdot ux}\\ \end{array} \end{array} \]
                                                                  (FPCore (ux uy maxCos)
                                                                   :precision binary32
                                                                   (let* ((t_0 (+ (- 1.0 ux) (* ux maxCos))))
                                                                     (if (<= (* t_0 t_0) 0.9996600151062012)
                                                                       (sqrt (fma (- ux (fma maxCos ux 1.0)) (fma maxCos ux (- 1.0 ux)) 1.0))
                                                                       (* 1.0 (sqrt (* (- (fma -1.0 (- maxCos 1.0) 1.0) maxCos) ux))))))
                                                                  float code(float ux, float uy, float maxCos) {
                                                                  	float t_0 = (1.0f - ux) + (ux * maxCos);
                                                                  	float tmp;
                                                                  	if ((t_0 * t_0) <= 0.9996600151062012f) {
                                                                  		tmp = sqrtf(fmaf((ux - fmaf(maxCos, ux, 1.0f)), fmaf(maxCos, ux, (1.0f - ux)), 1.0f));
                                                                  	} else {
                                                                  		tmp = 1.0f * sqrtf(((fmaf(-1.0f, (maxCos - 1.0f), 1.0f) - maxCos) * ux));
                                                                  	}
                                                                  	return tmp;
                                                                  }
                                                                  
                                                                  function code(ux, uy, maxCos)
                                                                  	t_0 = Float32(Float32(Float32(1.0) - ux) + Float32(ux * maxCos))
                                                                  	tmp = Float32(0.0)
                                                                  	if (Float32(t_0 * t_0) <= Float32(0.9996600151062012))
                                                                  		tmp = sqrt(fma(Float32(ux - fma(maxCos, ux, Float32(1.0))), fma(maxCos, ux, Float32(Float32(1.0) - ux)), Float32(1.0)));
                                                                  	else
                                                                  		tmp = Float32(Float32(1.0) * sqrt(Float32(Float32(fma(Float32(-1.0), Float32(maxCos - Float32(1.0)), Float32(1.0)) - maxCos) * ux)));
                                                                  	end
                                                                  	return tmp
                                                                  end
                                                                  
                                                                  \begin{array}{l}
                                                                  
                                                                  \\
                                                                  \begin{array}{l}
                                                                  t_0 := \left(1 - ux\right) + ux \cdot maxCos\\
                                                                  \mathbf{if}\;t\_0 \cdot t\_0 \leq 0.9996600151062012:\\
                                                                  \;\;\;\;\sqrt{\mathsf{fma}\left(ux - \mathsf{fma}\left(maxCos, ux, 1\right), \mathsf{fma}\left(maxCos, ux, 1 - ux\right), 1\right)}\\
                                                                  
                                                                  \mathbf{else}:\\
                                                                  \;\;\;\;1 \cdot \sqrt{\left(\mathsf{fma}\left(-1, maxCos - 1, 1\right) - maxCos\right) \cdot ux}\\
                                                                  
                                                                  
                                                                  \end{array}
                                                                  \end{array}
                                                                  
                                                                  Derivation
                                                                  1. Split input into 2 regimes
                                                                  2. if (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos))) < 0.999660015

                                                                    1. Initial program 87.8%

                                                                      \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                    2. Add Preprocessing
                                                                    3. Taylor expanded in uy around 0

                                                                      \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                    4. Step-by-step derivation
                                                                      1. Applied rewrites71.4%

                                                                        \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                      2. Taylor expanded in maxCos around inf

                                                                        \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(maxCos \cdot \left(\left(ux + \frac{1}{maxCos}\right) - \frac{ux}{maxCos}\right)\right)}} \]
                                                                      3. Step-by-step derivation
                                                                        1. Applied rewrites71.0%

                                                                          \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                        2. Step-by-step derivation
                                                                          1. lift--.f32N/A

                                                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                          2. lift-*.f32N/A

                                                                            \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                          3. fp-cancel-sub-sign-invN/A

                                                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                          4. +-commutativeN/A

                                                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right) + 1}} \]
                                                                          5. lower-fma.f32N/A

                                                                            \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)}} \]
                                                                          6. lower-neg.f3271.0

                                                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(\color{blue}{-\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                          7. lift-+.f32N/A

                                                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                          8. +-commutativeN/A

                                                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(ux \cdot maxCos + \left(1 - ux\right)\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                          9. lift-*.f32N/A

                                                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\left(\color{blue}{ux \cdot maxCos} + \left(1 - ux\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                          10. lower-fma.f3271.0

                                                                            \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\mathsf{fma}\left(ux, maxCos, 1 - ux\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                        3. Applied rewrites71.0%

                                                                          \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-\mathsf{fma}\left(ux, maxCos, 1 - ux\right), \left(\frac{1 - ux}{maxCos} + ux\right) \cdot maxCos, 1\right)}} \]
                                                                        4. Taylor expanded in uy around 0

                                                                          \[\leadsto \color{blue}{\sqrt{1 + \left(ux - \left(1 + maxCos \cdot ux\right)\right) \cdot \left(\left(1 + maxCos \cdot ux\right) - ux\right)}} \]
                                                                        5. Step-by-step derivation
                                                                          1. Applied rewrites71.8%

                                                                            \[\leadsto \color{blue}{\sqrt{\mathsf{fma}\left(ux - \mathsf{fma}\left(maxCos, ux, 1\right), \mathsf{fma}\left(maxCos, ux, 1 - ux\right), 1\right)}} \]

                                                                          if 0.999660015 < (*.f32 (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)) (+.f32 (-.f32 #s(literal 1 binary32) ux) (*.f32 ux maxCos)))

                                                                          1. Initial program 33.5%

                                                                            \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                          2. Add Preprocessing
                                                                          3. Taylor expanded in uy around 0

                                                                            \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                          4. Step-by-step derivation
                                                                            1. Applied rewrites29.4%

                                                                              \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                            2. Taylor expanded in maxCos around inf

                                                                              \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(maxCos \cdot \left(\left(ux + \frac{1}{maxCos}\right) - \frac{ux}{maxCos}\right)\right)}} \]
                                                                            3. Step-by-step derivation
                                                                              1. Applied rewrites30.1%

                                                                                \[\leadsto 1 \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \color{blue}{\left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                              2. Step-by-step derivation
                                                                                1. lift--.f32N/A

                                                                                  \[\leadsto 1 \cdot \sqrt{\color{blue}{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                                2. lift-*.f32N/A

                                                                                  \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                                3. fp-cancel-sub-sign-invN/A

                                                                                  \[\leadsto 1 \cdot \sqrt{\color{blue}{1 + \left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right)}} \]
                                                                                4. +-commutativeN/A

                                                                                  \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right)\right) \cdot \left(\left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos\right) + 1}} \]
                                                                                5. lower-fma.f32N/A

                                                                                  \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(\mathsf{neg}\left(\left(\left(1 - ux\right) + ux \cdot maxCos\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)}} \]
                                                                                6. lower-neg.f3230.1

                                                                                  \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(\color{blue}{-\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                                7. lift-+.f32N/A

                                                                                  \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(\left(1 - ux\right) + ux \cdot maxCos\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                                8. +-commutativeN/A

                                                                                  \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\left(ux \cdot maxCos + \left(1 - ux\right)\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                                9. lift-*.f32N/A

                                                                                  \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\left(\color{blue}{ux \cdot maxCos} + \left(1 - ux\right)\right), \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                                10. lower-fma.f3230.1

                                                                                  \[\leadsto 1 \cdot \sqrt{\mathsf{fma}\left(-\color{blue}{\mathsf{fma}\left(ux, maxCos, 1 - ux\right)}, \left(ux + \frac{1 - ux}{maxCos}\right) \cdot maxCos, 1\right)} \]
                                                                              3. Applied rewrites30.1%

                                                                                \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-\mathsf{fma}\left(ux, maxCos, 1 - ux\right), \left(\frac{1 - ux}{maxCos} + ux\right) \cdot maxCos, 1\right)}} \]
                                                                              4. Taylor expanded in ux around 0

                                                                                \[\leadsto 1 \cdot \sqrt{\color{blue}{ux \cdot \left(\left(1 + -1 \cdot \left(maxCos - 1\right)\right) - maxCos\right)}} \]
                                                                              5. Step-by-step derivation
                                                                                1. Applied rewrites72.8%

                                                                                  \[\leadsto 1 \cdot \sqrt{\color{blue}{\left(\mathsf{fma}\left(-1, maxCos - 1, 1\right) - maxCos\right) \cdot ux}} \]
                                                                              6. Recombined 2 regimes into one program.
                                                                              7. Add Preprocessing

                                                                              Alternative 12: 65.2% accurate, 5.8× speedup?

                                                                              \[\begin{array}{l} \\ 1 \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux} \end{array} \]
                                                                              (FPCore (ux uy maxCos)
                                                                               :precision binary32
                                                                               (* 1.0 (sqrt (* (fma -2.0 maxCos 2.0) ux))))
                                                                              float code(float ux, float uy, float maxCos) {
                                                                              	return 1.0f * sqrtf((fmaf(-2.0f, maxCos, 2.0f) * ux));
                                                                              }
                                                                              
                                                                              function code(ux, uy, maxCos)
                                                                              	return Float32(Float32(1.0) * sqrt(Float32(fma(Float32(-2.0), maxCos, Float32(2.0)) * ux)))
                                                                              end
                                                                              
                                                                              \begin{array}{l}
                                                                              
                                                                              \\
                                                                              1 \cdot \sqrt{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}
                                                                              \end{array}
                                                                              
                                                                              Derivation
                                                                              1. Initial program 54.0%

                                                                                \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                              2. Add Preprocessing
                                                                              3. Taylor expanded in uy around 0

                                                                                \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                              4. Step-by-step derivation
                                                                                1. Applied rewrites45.3%

                                                                                  \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                                2. Taylor expanded in ux around 0

                                                                                  \[\leadsto 1 \cdot \sqrt{\color{blue}{ux \cdot \left(2 - 2 \cdot maxCos\right)}} \]
                                                                                3. Step-by-step derivation
                                                                                  1. Applied rewrites63.4%

                                                                                    \[\leadsto 1 \cdot \sqrt{\color{blue}{\mathsf{fma}\left(-2, maxCos, 2\right) \cdot ux}} \]
                                                                                  2. Add Preprocessing

                                                                                  Alternative 13: 6.6% accurate, 8.2× speedup?

                                                                                  \[\begin{array}{l} \\ 1 \cdot \sqrt{1 - 1} \end{array} \]
                                                                                  (FPCore (ux uy maxCos) :precision binary32 (* 1.0 (sqrt (- 1.0 1.0))))
                                                                                  float code(float ux, float uy, float maxCos) {
                                                                                  	return 1.0f * sqrtf((1.0f - 1.0f));
                                                                                  }
                                                                                  
                                                                                  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(4) function code(ux, uy, maxcos)
                                                                                  use fmin_fmax_functions
                                                                                      real(4), intent (in) :: ux
                                                                                      real(4), intent (in) :: uy
                                                                                      real(4), intent (in) :: maxcos
                                                                                      code = 1.0e0 * sqrt((1.0e0 - 1.0e0))
                                                                                  end function
                                                                                  
                                                                                  function code(ux, uy, maxCos)
                                                                                  	return Float32(Float32(1.0) * sqrt(Float32(Float32(1.0) - Float32(1.0))))
                                                                                  end
                                                                                  
                                                                                  function tmp = code(ux, uy, maxCos)
                                                                                  	tmp = single(1.0) * sqrt((single(1.0) - single(1.0)));
                                                                                  end
                                                                                  
                                                                                  \begin{array}{l}
                                                                                  
                                                                                  \\
                                                                                  1 \cdot \sqrt{1 - 1}
                                                                                  \end{array}
                                                                                  
                                                                                  Derivation
                                                                                  1. Initial program 54.0%

                                                                                    \[\cos \left(\left(uy \cdot 2\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                                  2. Add Preprocessing
                                                                                  3. Taylor expanded in uy around 0

                                                                                    \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                                  4. Step-by-step derivation
                                                                                    1. Applied rewrites45.3%

                                                                                      \[\leadsto \color{blue}{1} \cdot \sqrt{1 - \left(\left(1 - ux\right) + ux \cdot maxCos\right) \cdot \left(\left(1 - ux\right) + ux \cdot maxCos\right)} \]
                                                                                    2. Taylor expanded in ux around 0

                                                                                      \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{1}} \]
                                                                                    3. Step-by-step derivation
                                                                                      1. Applied rewrites6.6%

                                                                                        \[\leadsto 1 \cdot \sqrt{1 - \color{blue}{1}} \]
                                                                                      2. Add Preprocessing

                                                                                      Reproduce

                                                                                      ?
                                                                                      herbie shell --seed 2025026 
                                                                                      (FPCore (ux uy maxCos)
                                                                                        :name "UniformSampleCone, x"
                                                                                        :precision binary32
                                                                                        :pre (and (and (and (<= 2.328306437e-10 ux) (<= ux 1.0)) (and (<= 2.328306437e-10 uy) (<= uy 1.0))) (and (<= 0.0 maxCos) (<= maxCos 1.0)))
                                                                                        (* (cos (* (* uy 2.0) (PI))) (sqrt (- 1.0 (* (+ (- 1.0 ux) (* ux maxCos)) (+ (- 1.0 ux) (* ux maxCos)))))))