UniformSampleCone 2

Percentage Accurate: 98.9% → 98.9%

Time: 42.9s

Alternatives: 20

Speedup: 1.0×

Specification

\[\left(\left(\left(\left(\left(-10000 \leq xi \land xi \leq 10000\right) \land \left(-10000 \leq yi \land yi \leq 10000\right)\right) \land \left(-10000 \leq zi \land zi \leq 10000\right)\right) \land \left(2.328306437 \cdot 10^{-10} \leq ux \land ux \leq 1\right)\right) \land \left(2.328306437 \cdot 10^{-10} \leq uy \land uy \leq 1\right)\right) \land \left(0 \leq maxCos \land maxCos \leq 1\right)\]

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\\ t_1 := \sqrt{1 - t\_0 \cdot t\_0}\\ t_2 := \left(uy \cdot 2\right) \cdot \pi\\ \left(\left(\cos t\_2 \cdot t\_1\right) \cdot xi + \left(\sin t\_2 \cdot t\_1\right) \cdot yi\right) + t\_0 \cdot zi \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* (* (- 1.0 ux) maxCos) ux))
        (t_1 (sqrt (- 1.0 (* t_0 t_0))))
        (t_2 (* (* uy 2.0) PI)))
   (+ (+ (* (* (cos t_2) t_1) xi) (* (* (sin t_2) t_1) yi)) (* t_0 zi))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = ((1.0f - ux) * maxCos) * ux;
	float t_1 = sqrtf((1.0f - (t_0 * t_0)));
	float t_2 = (uy * 2.0f) * ((float) M_PI);
	return (((cosf(t_2) * t_1) * xi) + ((sinf(t_2) * t_1) * yi)) + (t_0 * zi);
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(Float32(Float32(1.0) - ux) * maxCos) * ux)
	t_1 = sqrt(Float32(Float32(1.0) - Float32(t_0 * t_0)))
	t_2 = Float32(Float32(uy * Float32(2.0)) * Float32(pi))
	return Float32(Float32(Float32(Float32(cos(t_2) * t_1) * xi) + Float32(Float32(sin(t_2) * t_1) * yi)) + Float32(t_0 * zi))
end

MATLAB

function tmp = code(xi, yi, zi, ux, uy, maxCos)
	t_0 = ((single(1.0) - ux) * maxCos) * ux;
	t_1 = sqrt((single(1.0) - (t_0 * t_0)));
	t_2 = (uy * single(2.0)) * single(pi);
	tmp = (((cos(t_2) * t_1) * xi) + ((sin(t_2) * t_1) * yi)) + (t_0 * zi);
end

Initial Program: 98.9% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\\ t_1 := \sqrt{1 - t\_0 \cdot t\_0}\\ t_2 := \left(uy \cdot 2\right) \cdot \pi\\ \left(\left(\cos t\_2 \cdot t\_1\right) \cdot xi + \left(\sin t\_2 \cdot t\_1\right) \cdot yi\right) + t\_0 \cdot zi \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* (* (- 1.0 ux) maxCos) ux))
        (t_1 (sqrt (- 1.0 (* t_0 t_0))))
        (t_2 (* (* uy 2.0) PI)))
   (+ (+ (* (* (cos t_2) t_1) xi) (* (* (sin t_2) t_1) yi)) (* t_0 zi))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = ((1.0f - ux) * maxCos) * ux;
	float t_1 = sqrtf((1.0f - (t_0 * t_0)));
	float t_2 = (uy * 2.0f) * ((float) M_PI);
	return (((cosf(t_2) * t_1) * xi) + ((sinf(t_2) * t_1) * yi)) + (t_0 * zi);
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(Float32(Float32(1.0) - ux) * maxCos) * ux)
	t_1 = sqrt(Float32(Float32(1.0) - Float32(t_0 * t_0)))
	t_2 = Float32(Float32(uy * Float32(2.0)) * Float32(pi))
	return Float32(Float32(Float32(Float32(cos(t_2) * t_1) * xi) + Float32(Float32(sin(t_2) * t_1) * yi)) + Float32(t_0 * zi))
end

MATLAB

function tmp = code(xi, yi, zi, ux, uy, maxCos)
	t_0 = ((single(1.0) - ux) * maxCos) * ux;
	t_1 = sqrt((single(1.0) - (t_0 * t_0)));
	t_2 = (uy * single(2.0)) * single(pi);
	tmp = (((cos(t_2) * t_1) * xi) + ((sin(t_2) * t_1) * yi)) + (t_0 * zi);
end

Alternative 1: 98.9% accurate, 0.9× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(yi \cdot \left(\sin \left(uy \cdot \pi\right) \cdot 2\right), \sin \left(\mathsf{fma}\left(-\pi, uy, \frac{\pi}{2}\right)\right), \cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (- 1.0 ux)
  (* (* maxCos zi) ux)
  (*
   (sqrt (fma (* (* maxCos ux) (- ux 1.0)) (* (* maxCos ux) (- 1.0 ux)) 1.0))
   (fma
    (* yi (* (sin (* uy PI)) 2.0))
    (sin (fma (- PI) uy (/ PI 2.0)))
    (* (cos (* (+ PI PI) uy)) xi)))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f - ux), ((maxCos * zi) * ux), (sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * fmaf((yi * (sinf((uy * ((float) M_PI))) * 2.0f)), sinf(fmaf(-((float) M_PI), uy, (((float) M_PI) / 2.0f))), (cosf(((((float) M_PI) + ((float) M_PI)) * uy)) * xi))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * fma(Float32(yi * Float32(sin(Float32(uy * Float32(pi))) * Float32(2.0))), sin(fma(Float32(-Float32(pi)), uy, Float32(Float32(pi) / Float32(2.0)))), Float32(cos(Float32(Float32(Float32(pi) + Float32(pi)) * uy)) * xi))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

5. Applied rewrites 98.9%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\mathsf{fma}\left(yi \cdot \left(\sin \left(uy \cdot \pi\right) \cdot 2\right), \cos \left(uy \cdot \pi\right), \cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi\right)}\right) \]

7. Applied rewrites 98.9%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(yi \cdot \left(\sin \left(uy \cdot \pi\right) \cdot 2\right), \color{blue}{\sin \left(\mathsf{fma}\left(-\pi, uy, \frac{\pi}{2}\right)\right)}, \cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi\right)\right) \]
8. Add Preprocessing

Alternative 2: 98.9% accurate, 0.9× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(yi \cdot \left(\sin \left(uy \cdot \pi\right) \cdot 2\right), \cos \left(uy \cdot \pi\right), \cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (- 1.0 ux)
  (* (* maxCos zi) ux)
  (*
   (sqrt (fma (* (* maxCos ux) (- ux 1.0)) (* (* maxCos ux) (- 1.0 ux)) 1.0))
   (fma
    (* yi (* (sin (* uy PI)) 2.0))
    (cos (* uy PI))
    (* (cos (* (+ PI PI) uy)) xi)))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f - ux), ((maxCos * zi) * ux), (sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * fmaf((yi * (sinf((uy * ((float) M_PI))) * 2.0f)), cosf((uy * ((float) M_PI))), (cosf(((((float) M_PI) + ((float) M_PI)) * uy)) * xi))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * fma(Float32(yi * Float32(sin(Float32(uy * Float32(pi))) * Float32(2.0))), cos(Float32(uy * Float32(pi))), Float32(cos(Float32(Float32(Float32(pi) + Float32(pi)) * uy)) * xi))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

5. Applied rewrites 98.9%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\mathsf{fma}\left(yi \cdot \left(\sin \left(uy \cdot \pi\right) \cdot 2\right), \cos \left(uy \cdot \pi\right), \cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi\right)}\right) \]
6. Add Preprocessing

Alternative 3: 98.9% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\\ t_1 := \left(\pi + \pi\right) \cdot uy\\ t_2 := \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), t\_0, 1\right)}\\ \mathsf{fma}\left(t\_2 \cdot yi, \sin t\_1, \mathsf{fma}\left(\cos t\_1 \cdot xi, t\_2, t\_0 \cdot zi\right)\right) \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* (* maxCos ux) (- 1.0 ux)))
        (t_1 (* (+ PI PI) uy))
        (t_2 (sqrt (fma (* (* maxCos ux) (- ux 1.0)) t_0 1.0))))
   (fma (* t_2 yi) (sin t_1) (fma (* (cos t_1) xi) t_2 (* t_0 zi)))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = (maxCos * ux) * (1.0f - ux);
	float t_1 = (((float) M_PI) + ((float) M_PI)) * uy;
	float t_2 = sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), t_0, 1.0f));
	return fmaf((t_2 * yi), sinf(t_1), fmaf((cosf(t_1) * xi), t_2, (t_0 * zi)));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux))
	t_1 = Float32(Float32(Float32(pi) + Float32(pi)) * uy)
	t_2 = sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), t_0, Float32(1.0)))
	return fma(Float32(t_2 * yi), sin(t_1), fma(Float32(cos(t_1) * xi), t_2, Float32(t_0 * zi)))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right)} \]
4. Add Preprocessing

Alternative 4: 98.9% accurate, 1.2× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(\pi + \pi\right) \cdot uy\\ \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos t\_0, xi, \sin t\_0 \cdot yi\right)\right) \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* (+ PI PI) uy)))
   (fma
    (- 1.0 ux)
    (* (* maxCos zi) ux)
    (*
     (sqrt (fma (* (* maxCos ux) (- ux 1.0)) (* (* maxCos ux) (- 1.0 ux)) 1.0))
     (fma (cos t_0) xi (* (sin t_0) yi))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = (((float) M_PI) + ((float) M_PI)) * uy;
	return fmaf((1.0f - ux), ((maxCos * zi) * ux), (sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * fmaf(cosf(t_0), xi, (sinf(t_0) * yi))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(Float32(pi) + Float32(pi)) * uy)
	return fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * fma(cos(t_0), xi, Float32(sin(t_0) * yi))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]
4. Add Preprocessing

Alternative 5: 98.9% accurate, 1.2× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(\pi + \pi\right) \cdot uy\\ t_1 := \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\\ \mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), t\_1, 1\right)}, \mathsf{fma}\left(\sin t\_0, yi, \cos t\_0 \cdot xi\right), t\_1 \cdot zi\right) \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* (+ PI PI) uy)) (t_1 (* (* maxCos ux) (- 1.0 ux))))
   (fma
    (sqrt (fma (* (* maxCos ux) (- ux 1.0)) t_1 1.0))
    (fma (sin t_0) yi (* (cos t_0) xi))
    (* t_1 zi))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = (((float) M_PI) + ((float) M_PI)) * uy;
	float t_1 = (maxCos * ux) * (1.0f - ux);
	return fmaf(sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), t_1, 1.0f)), fmaf(sinf(t_0), yi, (cosf(t_0) * xi)), (t_1 * zi));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(Float32(pi) + Float32(pi)) * uy)
	t_1 = Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux))
	return fma(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), t_1, Float32(1.0))), fma(sin(t_0), yi, Float32(cos(t_0) * xi)), Float32(t_1 * zi))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \mathsf{fma}\left(\sin \left(\left(\pi + \pi\right) \cdot uy\right), yi, \cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi\right), \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)} \]
4. Add Preprocessing

Alternative 6: 98.8% accurate, 1.4× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\sin \left(\mathsf{fma}\left(\pi + \pi, -uy, \frac{\pi}{2}\right)\right) \cdot xi, 1, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (* 1.0 yi)
  (sin (* (+ PI PI) uy))
  (fma
   (* (sin (fma (+ PI PI) (- uy) (/ PI 2.0))) xi)
   1.0
   (* (* (* maxCos ux) (- 1.0 ux)) zi))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f * yi), sinf(((((float) M_PI) + ((float) M_PI)) * uy)), fmaf((sinf(fmaf((((float) M_PI) + ((float) M_PI)), -uy, (((float) M_PI) / 2.0f))) * xi), 1.0f, (((maxCos * ux) * (1.0f - ux)) * zi)));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) * yi), sin(Float32(Float32(Float32(pi) + Float32(pi)) * uy)), fma(Float32(sin(fma(Float32(Float32(pi) + Float32(pi)), Float32(-uy), Float32(Float32(pi) / Float32(2.0)))) * xi), Float32(1.0), Float32(Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)) * zi)))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right)} \]

4. Taylor expanded in ux around 0

   \[\leadsto \mathsf{fma}\left(\color{blue}{1} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

6. Applied rewrites 98.9%

   \[\leadsto \mathsf{fma}\left(\color{blue}{1} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

7. Taylor expanded in ux around 0

   \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \color{blue}{1}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

9. Applied rewrites 98.8%

   \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \color{blue}{1}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

11. Applied rewrites 98.8%

    \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\color{blue}{\sin \left(\mathsf{fma}\left(\pi + \pi, -uy, \frac{\pi}{2}\right)\right)} \cdot xi, 1, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]
12. Add Preprocessing

Alternative 7: 98.8% accurate, 1.5× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(\pi + \pi\right) \cdot uy\\ \mathsf{fma}\left(1 \cdot yi, \sin t\_0, \mathsf{fma}\left(\cos t\_0 \cdot xi, 1, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* (+ PI PI) uy)))
   (fma
    (* 1.0 yi)
    (sin t_0)
    (fma (* (cos t_0) xi) 1.0 (* (* (* maxCos ux) (- 1.0 ux)) zi)))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = (((float) M_PI) + ((float) M_PI)) * uy;
	return fmaf((1.0f * yi), sinf(t_0), fmaf((cosf(t_0) * xi), 1.0f, (((maxCos * ux) * (1.0f - ux)) * zi)));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(Float32(pi) + Float32(pi)) * uy)
	return fma(Float32(Float32(1.0) * yi), sin(t_0), fma(Float32(cos(t_0) * xi), Float32(1.0), Float32(Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)) * zi)))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right)} \]

4. Taylor expanded in ux around 0

   \[\leadsto \mathsf{fma}\left(\color{blue}{1} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

6. Applied rewrites 98.9%

   \[\leadsto \mathsf{fma}\left(\color{blue}{1} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

7. Taylor expanded in ux around 0

   \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \color{blue}{1}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

9. Applied rewrites 98.8%

   \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \color{blue}{1}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]
10. Add Preprocessing

Alternative 8: 98.8% accurate, 1.6× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := 2 \cdot \left(uy \cdot \pi\right)\\ \mathsf{fma}\left(maxCos, ux \cdot \left(zi \cdot \left(1 - ux\right)\right), \mathsf{fma}\left(xi, \cos t\_0, yi \cdot \sin t\_0\right)\right) \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* 2.0 (* uy PI))))
   (fma maxCos (* ux (* zi (- 1.0 ux))) (fma xi (cos t_0) (* yi (sin t_0))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = 2.0f * (uy * ((float) M_PI));
	return fmaf(maxCos, (ux * (zi * (1.0f - ux))), fmaf(xi, cosf(t_0), (yi * sinf(t_0))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(2.0) * Float32(uy * Float32(pi)))
	return fma(maxCos, Float32(ux * Float32(zi * Float32(Float32(1.0) - ux))), fma(xi, cos(t_0), Float32(yi * sin(t_0))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

2. Taylor expanded in maxCos around 0

   \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) + \left(xi \cdot \cos \left(2 \cdot \left(uy \cdot \mathsf{PI}\left(\right)\right)\right) + yi \cdot \sin \left(2 \cdot \left(uy \cdot \mathsf{PI}\left(\right)\right)\right)\right)} \]

4. Applied rewrites 98.8%

   \[\leadsto \color{blue}{\mathsf{fma}\left(maxCos, ux \cdot \left(zi \cdot \left(1 - ux\right)\right), \mathsf{fma}\left(xi, \cos \left(2 \cdot \left(uy \cdot \pi\right)\right), yi \cdot \sin \left(2 \cdot \left(uy \cdot \pi\right)\right)\right)\right)} \]
5. Add Preprocessing

Alternative 9: 96.2% accurate, 1.7× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := 2 \cdot \left(uy \cdot \pi\right)\\ \mathbf{if}\;uy \leq 0.00139999995008111:\\ \;\;\;\;\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + \mathsf{fma}\left(uy \cdot \left(yi + yi\right), \pi, \left(\left(\pi \cdot \pi\right) \cdot \left(uy \cdot xi\right)\right) \cdot \left(uy \cdot -2\right)\right)\right)\right)\\ \mathbf{else}:\\ \;\;\;\;\mathsf{fma}\left(xi, \cos t\_0, yi \cdot \sin t\_0\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* 2.0 (* uy PI))))
   (if (<= uy 0.00139999995008111)
     (fma
      (- 1.0 ux)
      (* (* maxCos zi) ux)
      (*
       (sqrt
        (fma (* (* maxCos ux) (- ux 1.0)) (* (* maxCos ux) (- 1.0 ux)) 1.0))
       (+
        xi
        (fma (* uy (+ yi yi)) PI (* (* (* PI PI) (* uy xi)) (* uy -2.0))))))
     (fma xi (cos t_0) (* yi (sin t_0))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = 2.0f * (uy * ((float) M_PI));
	float tmp;
	if (uy <= 0.00139999995008111f) {
		tmp = fmaf((1.0f - ux), ((maxCos * zi) * ux), (sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * (xi + fmaf((uy * (yi + yi)), ((float) M_PI), (((((float) M_PI) * ((float) M_PI)) * (uy * xi)) * (uy * -2.0f))))));
	} else {
		tmp = fmaf(xi, cosf(t_0), (yi * sinf(t_0)));
	}
	return tmp;
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(2.0) * Float32(uy * Float32(pi)))
	tmp = Float32(0.0)
	if (uy <= Float32(0.00139999995008111))
		tmp = fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * Float32(xi + fma(Float32(uy * Float32(yi + yi)), Float32(pi), Float32(Float32(Float32(Float32(pi) * Float32(pi)) * Float32(uy * xi)) * Float32(uy * Float32(-2.0)))))));
	else
		tmp = fma(xi, cos(t_0), Float32(yi * sin(t_0)));
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if uy < 0.00139999995

   1. Initial program 98.9%

      \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

   3. Applied rewrites 98.9%

      \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

   4. Taylor expanded in uy around 0

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

   6. Applied rewrites 85.4%

      \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\left(xi + uy \cdot \mathsf{fma}\left(-2, uy \cdot \left(xi \cdot {\pi}^{2}\right), 2 \cdot \left(yi \cdot \pi\right)\right)\right)}\right) \]

   8. Applied rewrites 85.4%

      \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + \mathsf{fma}\left(uy \cdot \left(yi + yi\right), \color{blue}{\pi}, \left(\left(\pi \cdot \pi\right) \cdot \left(uy \cdot xi\right)\right) \cdot \left(uy \cdot -2\right)\right)\right)\right) \]

   if 0.00139999995 < uy

   1. Initial program 98.9%

      \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

   2. Taylor expanded in ux around 0

      \[\leadsto \color{blue}{xi \cdot \cos \left(2 \cdot \left(uy \cdot \mathsf{PI}\left(\right)\right)\right) + yi \cdot \sin \left(2 \cdot \left(uy \cdot \mathsf{PI}\left(\right)\right)\right)} \]

   4. Applied rewrites 90.1%

      \[\leadsto \color{blue}{\mathsf{fma}\left(xi, \cos \left(2 \cdot \left(uy \cdot \pi\right)\right), yi \cdot \sin \left(2 \cdot \left(uy \cdot \pi\right)\right)\right)} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 10: 95.7% accurate, 1.6× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(maxCos, ux \cdot zi, xi \cdot \cos \left(2 \cdot \left(uy \cdot \pi\right)\right)\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (* 1.0 yi)
  (sin (* (+ PI PI) uy))
  (fma maxCos (* ux zi) (* xi (cos (* 2.0 (* uy PI)))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f * yi), sinf(((((float) M_PI) + ((float) M_PI)) * uy)), fmaf(maxCos, (ux * zi), (xi * cosf((2.0f * (uy * ((float) M_PI)))))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) * yi), sin(Float32(Float32(Float32(pi) + Float32(pi)) * uy)), fma(maxCos, Float32(ux * zi), Float32(xi * cos(Float32(Float32(2.0) * Float32(uy * Float32(pi)))))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right)} \]

4. Taylor expanded in ux around 0

   \[\leadsto \mathsf{fma}\left(\color{blue}{1} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

6. Applied rewrites 98.9%

   \[\leadsto \mathsf{fma}\left(\color{blue}{1} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

7. Taylor expanded in ux around 0

   \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \color{blue}{1}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

9. Applied rewrites 98.8%

   \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \color{blue}{1}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right) \]

10. Taylor expanded in ux around 0

    \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \color{blue}{maxCos \cdot \left(ux \cdot zi\right) + xi \cdot \cos \left(2 \cdot \left(uy \cdot \mathsf{PI}\left(\right)\right)\right)}\right) \]

12. Applied rewrites 95.7%

    \[\leadsto \mathsf{fma}\left(1 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \color{blue}{\mathsf{fma}\left(maxCos, ux \cdot zi, xi \cdot \cos \left(2 \cdot \left(uy \cdot \pi\right)\right)\right)}\right) \]
13. Add Preprocessing

Alternative 11: 95.7% accurate, 1.6× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := 2 \cdot \left(uy \cdot \pi\right)\\ \mathsf{fma}\left(maxCos, ux \cdot zi, \mathsf{fma}\left(xi, \cos t\_0, yi \cdot \sin t\_0\right)\right) \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* 2.0 (* uy PI))))
   (fma maxCos (* ux zi) (fma xi (cos t_0) (* yi (sin t_0))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = 2.0f * (uy * ((float) M_PI));
	return fmaf(maxCos, (ux * zi), fmaf(xi, cosf(t_0), (yi * sinf(t_0))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(2.0) * Float32(uy * Float32(pi)))
	return fma(maxCos, Float32(ux * zi), fma(xi, cos(t_0), Float32(yi * sin(t_0))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

2. Taylor expanded in ux around 0

   \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot zi\right) + \left(xi \cdot \cos \left(2 \cdot \left(uy \cdot \mathsf{PI}\left(\right)\right)\right) + yi \cdot \sin \left(2 \cdot \left(uy \cdot \mathsf{PI}\left(\right)\right)\right)\right)} \]

4. Applied rewrites 95.7%

   \[\leadsto \color{blue}{\mathsf{fma}\left(maxCos, ux \cdot zi, \mathsf{fma}\left(xi, \cos \left(2 \cdot \left(uy \cdot \pi\right)\right), yi \cdot \sin \left(2 \cdot \left(uy \cdot \pi\right)\right)\right)\right)} \]
5. Add Preprocessing

Alternative 12: 88.3% accurate, 1.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}\\ \mathbf{if}\;uy \leq 0.02490999922156334:\\ \;\;\;\;\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, t\_0 \cdot \left(xi + \mathsf{fma}\left(uy \cdot \left(yi + yi\right), \pi, \left(\left(\pi \cdot \pi\right) \cdot \left(uy \cdot xi\right)\right) \cdot \left(uy \cdot -2\right)\right)\right)\right)\\ \mathbf{else}:\\ \;\;\;\;\mathsf{fma}\left(t\_0 \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right)\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0
         (sqrt
          (fma
           (* (* maxCos ux) (- ux 1.0))
           (* (* maxCos ux) (- 1.0 ux))
           1.0))))
   (if (<= uy 0.02490999922156334)
     (fma
      (- 1.0 ux)
      (* (* maxCos zi) ux)
      (*
       t_0
       (+
        xi
        (fma (* uy (+ yi yi)) PI (* (* (* PI PI) (* uy xi)) (* uy -2.0))))))
     (fma
      (* t_0 yi)
      (sin (* (+ PI PI) uy))
      (* maxCos (* ux (* zi (- 1.0 ux))))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f));
	float tmp;
	if (uy <= 0.02490999922156334f) {
		tmp = fmaf((1.0f - ux), ((maxCos * zi) * ux), (t_0 * (xi + fmaf((uy * (yi + yi)), ((float) M_PI), (((((float) M_PI) * ((float) M_PI)) * (uy * xi)) * (uy * -2.0f))))));
	} else {
		tmp = fmaf((t_0 * yi), sinf(((((float) M_PI) + ((float) M_PI)) * uy)), (maxCos * (ux * (zi * (1.0f - ux)))));
	}
	return tmp;
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0)))
	tmp = Float32(0.0)
	if (uy <= Float32(0.02490999922156334))
		tmp = fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(t_0 * Float32(xi + fma(Float32(uy * Float32(yi + yi)), Float32(pi), Float32(Float32(Float32(Float32(pi) * Float32(pi)) * Float32(uy * xi)) * Float32(uy * Float32(-2.0)))))));
	else
		tmp = fma(Float32(t_0 * yi), sin(Float32(Float32(Float32(pi) + Float32(pi)) * uy)), Float32(maxCos * Float32(ux * Float32(zi * Float32(Float32(1.0) - ux)))));
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if uy < 0.0249099992

   1. Initial program 98.9%

      \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

   3. Applied rewrites 98.9%

      \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

   4. Taylor expanded in uy around 0

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

   6. Applied rewrites 85.4%

      \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\left(xi + uy \cdot \mathsf{fma}\left(-2, uy \cdot \left(xi \cdot {\pi}^{2}\right), 2 \cdot \left(yi \cdot \pi\right)\right)\right)}\right) \]

   8. Applied rewrites 85.4%

      \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + \mathsf{fma}\left(uy \cdot \left(yi + yi\right), \color{blue}{\pi}, \left(\left(\pi \cdot \pi\right) \cdot \left(uy \cdot xi\right)\right) \cdot \left(uy \cdot -2\right)\right)\right)\right) \]

   if 0.0249099992 < uy

   1. Initial program 98.9%

      \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

   3. Applied rewrites 98.9%

      \[\leadsto \color{blue}{\mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right) \cdot xi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)}, \left(\left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\right) \cdot zi\right)\right)} \]

   4. Taylor expanded in xi around 0

      \[\leadsto \mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right)}\right) \]

   6. Applied rewrites 44.9%

      \[\leadsto \mathsf{fma}\left(\sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot yi, \sin \left(\left(\pi + \pi\right) \cdot uy\right), \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right)}\right) \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 13: 85.5% accurate, 2.2× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + \mathsf{fma}\left(uy \cdot \left(yi + yi\right), \pi, \left(\left(\pi \cdot \pi\right) \cdot \left(uy \cdot xi\right)\right) \cdot \left(uy \cdot -2\right)\right)\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (- 1.0 ux)
  (* (* maxCos zi) ux)
  (*
   (sqrt (fma (* (* maxCos ux) (- ux 1.0)) (* (* maxCos ux) (- 1.0 ux)) 1.0))
   (+ xi (fma (* uy (+ yi yi)) PI (* (* (* PI PI) (* uy xi)) (* uy -2.0)))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f - ux), ((maxCos * zi) * ux), (sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * (xi + fmaf((uy * (yi + yi)), ((float) M_PI), (((((float) M_PI) * ((float) M_PI)) * (uy * xi)) * (uy * -2.0f))))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * Float32(xi + fma(Float32(uy * Float32(yi + yi)), Float32(pi), Float32(Float32(Float32(Float32(pi) * Float32(pi)) * Float32(uy * xi)) * Float32(uy * Float32(-2.0)))))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

4. Taylor expanded in uy around 0

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

6. Applied rewrites 85.4%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\left(xi + uy \cdot \mathsf{fma}\left(-2, uy \cdot \left(xi \cdot {\pi}^{2}\right), 2 \cdot \left(yi \cdot \pi\right)\right)\right)}\right) \]

8. Applied rewrites 85.4%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + \mathsf{fma}\left(uy \cdot \left(yi + yi\right), \color{blue}{\pi}, \left(\left(\pi \cdot \pi\right) \cdot \left(uy \cdot xi\right)\right) \cdot \left(uy \cdot -2\right)\right)\right)\right) \]
9. Add Preprocessing

Alternative 14: 85.4% accurate, 2.4× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\mathsf{fma}\left(yi, \pi + \pi, \left(xi \cdot \left(\pi \cdot \pi\right)\right) \cdot \left(uy \cdot -2\right)\right), uy, xi\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (- 1.0 ux)
  (* (* maxCos zi) ux)
  (*
   (sqrt (fma (* (* maxCos ux) (- ux 1.0)) (* (* maxCos ux) (- 1.0 ux)) 1.0))
   (fma (fma yi (+ PI PI) (* (* xi (* PI PI)) (* uy -2.0))) uy xi))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f - ux), ((maxCos * zi) * ux), (sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * fmaf(fmaf(yi, (((float) M_PI) + ((float) M_PI)), ((xi * (((float) M_PI) * ((float) M_PI))) * (uy * -2.0f))), uy, xi)));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * fma(fma(yi, Float32(Float32(pi) + Float32(pi)), Float32(Float32(xi * Float32(Float32(pi) * Float32(pi))) * Float32(uy * Float32(-2.0)))), uy, xi)))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

4. Taylor expanded in uy around 0

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

6. Applied rewrites 85.4%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\left(xi + uy \cdot \mathsf{fma}\left(-2, uy \cdot \left(xi \cdot {\pi}^{2}\right), 2 \cdot \left(yi \cdot \pi\right)\right)\right)}\right) \]

8. Applied rewrites 85.5%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\mathsf{fma}\left(yi, \pi + \pi, \left(xi \cdot \left(\pi \cdot \pi\right)\right) \cdot \left(uy \cdot -2\right)\right), \color{blue}{uy}, xi\right)\right) \]
9. Add Preprocessing

Alternative 15: 81.2% accurate, 2.9× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 - ux, \left(maxCos \cdot ux\right) \cdot zi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + uy \cdot \left(2 \cdot \left(yi \cdot \pi\right)\right)\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (- 1.0 ux)
  (* (* maxCos ux) zi)
  (*
   (sqrt (fma (* (* maxCos ux) (- ux 1.0)) (* (* maxCos ux) (- 1.0 ux)) 1.0))
   (+ xi (* uy (* 2.0 (* yi PI)))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f - ux), ((maxCos * ux) * zi), (sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * (xi + (uy * (2.0f * (yi * ((float) M_PI)))))));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * ux) * zi), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * Float32(xi + Float32(uy * Float32(Float32(2.0) * Float32(yi * Float32(pi)))))))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

3. Applied rewrites 98.9%

   \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

4. Taylor expanded in uy around 0

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

6. Applied rewrites 85.4%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\left(xi + uy \cdot \mathsf{fma}\left(-2, uy \cdot \left(xi \cdot {\pi}^{2}\right), 2 \cdot \left(yi \cdot \pi\right)\right)\right)}\right) \]

8. Applied rewrites 85.4%

   \[\leadsto \mathsf{fma}\left(1 - ux, \color{blue}{\left(maxCos \cdot ux\right) \cdot zi}, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + uy \cdot \mathsf{fma}\left(-2, uy \cdot \left(xi \cdot {\pi}^{2}\right), 2 \cdot \left(yi \cdot \pi\right)\right)\right)\right) \]

9. Taylor expanded in xi around 0

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

11. Applied rewrites 81.2%

    \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot ux\right) \cdot zi, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \left(xi + uy \cdot \left(2 \cdot \color{blue}{\left(yi \cdot \pi\right)}\right)\right)\right) \]
12. Add Preprocessing

Alternative 16: 62.8% accurate, 2.7× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} t_0 := \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right)\\ t_1 := \left(maxCos \cdot zi\right) \cdot ux\\ t_2 := \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), t\_0, 1\right)}\\ \mathbf{if}\;xi \leq -4.999999841327613 \cdot 10^{-22}:\\ \;\;\;\;\mathsf{fma}\left(1 - ux, t\_1, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot -1, t\_0, 1\right)} \cdot xi\right)\\ \mathbf{elif}\;xi \leq 1.000000031374395 \cdot 10^{-22}:\\ \;\;\;\;\mathsf{fma}\left(1 - ux, t\_1, t\_2 \cdot \left(2 \cdot \left(uy \cdot \left(yi \cdot \pi\right)\right)\right)\right)\\ \mathbf{else}:\\ \;\;\;\;\mathsf{fma}\left(zi \cdot \left(\left(1 - ux\right) \cdot maxCos\right), ux, t\_2 \cdot xi\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (let* ((t_0 (* (* maxCos ux) (- 1.0 ux)))
        (t_1 (* (* maxCos zi) ux))
        (t_2 (sqrt (fma (* (* maxCos ux) (- ux 1.0)) t_0 1.0))))
   (if (<= xi -4.999999841327613e-22)
     (fma (- 1.0 ux) t_1 (* (sqrt (fma (* (* maxCos ux) -1.0) t_0 1.0)) xi))
     (if (<= xi 1.000000031374395e-22)
       (fma (- 1.0 ux) t_1 (* t_2 (* 2.0 (* uy (* yi PI)))))
       (fma (* zi (* (- 1.0 ux) maxCos)) ux (* t_2 xi))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	float t_0 = (maxCos * ux) * (1.0f - ux);
	float t_1 = (maxCos * zi) * ux;
	float t_2 = sqrtf(fmaf(((maxCos * ux) * (ux - 1.0f)), t_0, 1.0f));
	float tmp;
	if (xi <= -4.999999841327613e-22f) {
		tmp = fmaf((1.0f - ux), t_1, (sqrtf(fmaf(((maxCos * ux) * -1.0f), t_0, 1.0f)) * xi));
	} else if (xi <= 1.000000031374395e-22f) {
		tmp = fmaf((1.0f - ux), t_1, (t_2 * (2.0f * (uy * (yi * ((float) M_PI))))));
	} else {
		tmp = fmaf((zi * ((1.0f - ux) * maxCos)), ux, (t_2 * xi));
	}
	return tmp;
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	t_0 = Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux))
	t_1 = Float32(Float32(maxCos * zi) * ux)
	t_2 = sqrt(fma(Float32(Float32(maxCos * ux) * Float32(ux - Float32(1.0))), t_0, Float32(1.0)))
	tmp = Float32(0.0)
	if (xi <= Float32(-4.999999841327613e-22))
		tmp = fma(Float32(Float32(1.0) - ux), t_1, Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(-1.0)), t_0, Float32(1.0))) * xi));
	elseif (xi <= Float32(1.000000031374395e-22))
		tmp = fma(Float32(Float32(1.0) - ux), t_1, Float32(t_2 * Float32(Float32(2.0) * Float32(uy * Float32(yi * Float32(pi))))));
	else
		tmp = fma(Float32(zi * Float32(Float32(Float32(1.0) - ux) * maxCos)), ux, Float32(t_2 * xi));
	end
	return tmp
end

Derivation

1. Split input into 3 regimes

2. if xi < -4.9999998e-22

   1. Initial program 98.9%

      \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

   2. Taylor expanded in uy around 0

      \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) + xi \cdot \sqrt{1 - {maxCos}^{2} \cdot \left({ux}^{2} \cdot {\left(1 - ux\right)}^{2}\right)}} \]

   4. Applied rewrites 51.6%

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

   6. Applied rewrites 51.6%

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

   7. Taylor expanded in ux around 0

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

   9. Applied rewrites 51.6%

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

   if -4.9999998e-22 < xi < 1.00000003e-22

   1. Initial program 98.9%

      \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

   3. Applied rewrites 98.9%

      \[\leadsto \color{blue}{\mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \mathsf{fma}\left(\cos \left(\left(\pi + \pi\right) \cdot uy\right), xi, \sin \left(\left(\pi + \pi\right) \cdot uy\right) \cdot yi\right)\right)} \]

   4. Taylor expanded in uy around 0

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

   6. Applied rewrites 85.4%

      \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot \color{blue}{\left(xi + uy \cdot \mathsf{fma}\left(-2, uy \cdot \left(xi \cdot {\pi}^{2}\right), 2 \cdot \left(yi \cdot \pi\right)\right)\right)}\right) \]

   7. Taylor expanded in xi around 0

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

   9. Applied rewrites 37.9%

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

   if 1.00000003e-22 < xi

   1. Initial program 98.9%

      \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

   2. Taylor expanded in uy around 0

      \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) + xi \cdot \sqrt{1 - {maxCos}^{2} \cdot \left({ux}^{2} \cdot {\left(1 - ux\right)}^{2}\right)}} \]

   4. Applied rewrites 51.6%

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

   6. Applied rewrites 51.6%

      \[\leadsto \mathsf{fma}\left(zi \cdot \left(\left(1 - ux\right) \cdot maxCos\right), \color{blue}{ux}, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot \left(ux - 1\right), \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot xi\right) \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 17: 51.6% accurate, 4.0× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot -1, \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot xi\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (fma
  (- 1.0 ux)
  (* (* maxCos zi) ux)
  (* (sqrt (fma (* (* maxCos ux) -1.0) (* (* maxCos ux) (- 1.0 ux)) 1.0)) xi)))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((1.0f - ux), ((maxCos * zi) * ux), (sqrtf(fmaf(((maxCos * ux) * -1.0f), ((maxCos * ux) * (1.0f - ux)), 1.0f)) * xi));
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(Float32(1.0) - ux), Float32(Float32(maxCos * zi) * ux), Float32(sqrt(fma(Float32(Float32(maxCos * ux) * Float32(-1.0)), Float32(Float32(maxCos * ux) * Float32(Float32(1.0) - ux)), Float32(1.0))) * xi))
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

2. Taylor expanded in uy around 0

   \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) + xi \cdot \sqrt{1 - {maxCos}^{2} \cdot \left({ux}^{2} \cdot {\left(1 - ux\right)}^{2}\right)}} \]

4. Applied rewrites 51.6%

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

6. Applied rewrites 51.6%

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

7. Taylor expanded in ux around 0

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

9. Applied rewrites 51.6%

   \[\leadsto \mathsf{fma}\left(1 - ux, \left(maxCos \cdot zi\right) \cdot ux, \sqrt{\mathsf{fma}\left(\left(maxCos \cdot ux\right) \cdot -1, \left(maxCos \cdot ux\right) \cdot \left(1 - ux\right), 1\right)} \cdot xi\right) \]
10. Add Preprocessing

Alternative 18: 51.6% accurate, 10.4× speedup

Math

\[\begin{array}{l} \\ xi + maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos)
 :precision binary32
 (+ xi (* maxCos (* ux (* zi (- 1.0 ux))))))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return xi + (maxCos * (ux * (zi * (1.0f - ux))));
}

Fortran

module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(4) function code(xi, yi, zi, ux, uy, maxcos)
use fmin_fmax_functions
    real(4), intent (in) :: xi
    real(4), intent (in) :: yi
    real(4), intent (in) :: zi
    real(4), intent (in) :: ux
    real(4), intent (in) :: uy
    real(4), intent (in) :: maxcos
    code = xi + (maxcos * (ux * (zi * (1.0e0 - ux))))
end function

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return Float32(xi + Float32(maxCos * Float32(ux * Float32(zi * Float32(Float32(1.0) - ux)))))
end

MATLAB

function tmp = code(xi, yi, zi, ux, uy, maxCos)
	tmp = xi + (maxCos * (ux * (zi * (single(1.0) - ux))));
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

2. Taylor expanded in uy around 0

   \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) + xi \cdot \sqrt{1 - {maxCos}^{2} \cdot \left({ux}^{2} \cdot {\left(1 - ux\right)}^{2}\right)}} \]

4. Applied rewrites 51.6%

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

5. Taylor expanded in maxCos around 0

   \[\leadsto xi + \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right)} \]

7. Applied rewrites 51.6%

   \[\leadsto xi + \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right)} \]
8. Add Preprocessing

Alternative 19: 49.3% accurate, 17.3× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(maxCos \cdot ux, zi, xi\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos) :precision binary32 (fma (* maxCos ux) zi xi))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return fmaf((maxCos * ux), zi, xi);
}

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return fma(Float32(maxCos * ux), zi, xi)
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

2. Taylor expanded in uy around 0

   \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) + xi \cdot \sqrt{1 - {maxCos}^{2} \cdot \left({ux}^{2} \cdot {\left(1 - ux\right)}^{2}\right)}} \]

4. Applied rewrites 51.6%

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

5. Taylor expanded in ux around 0

   \[\leadsto xi + \color{blue}{maxCos \cdot \left(ux \cdot zi\right)} \]

7. Applied rewrites 49.3%

   \[\leadsto xi + \color{blue}{maxCos \cdot \left(ux \cdot zi\right)} \]

9. Applied rewrites 49.3%

   \[\leadsto \mathsf{fma}\left(maxCos \cdot ux, zi, xi\right) \]
10. Add Preprocessing

Alternative 20: 12.2% accurate, 22.3× speedup

Math

\[\begin{array}{l} \\ maxCos \cdot \left(ux \cdot zi\right) \end{array} \]

FPCore

(FPCore (xi yi zi ux uy maxCos) :precision binary32 (* maxCos (* ux zi)))

C

float code(float xi, float yi, float zi, float ux, float uy, float maxCos) {
	return maxCos * (ux * zi);
}

Fortran

module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(4) function code(xi, yi, zi, ux, uy, maxcos)
use fmin_fmax_functions
    real(4), intent (in) :: xi
    real(4), intent (in) :: yi
    real(4), intent (in) :: zi
    real(4), intent (in) :: ux
    real(4), intent (in) :: uy
    real(4), intent (in) :: maxcos
    code = maxcos * (ux * zi)
end function

Julia

function code(xi, yi, zi, ux, uy, maxCos)
	return Float32(maxCos * Float32(ux * zi))
end

MATLAB

function tmp = code(xi, yi, zi, ux, uy, maxCos)
	tmp = maxCos * (ux * zi);
end

Derivation

1. Initial program 98.9%

   \[\left(\left(\cos \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot xi + \left(\sin \left(\left(uy \cdot 2\right) \cdot \pi\right) \cdot \sqrt{1 - \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right)}\right) \cdot yi\right) + \left(\left(\left(1 - ux\right) \cdot maxCos\right) \cdot ux\right) \cdot zi \]

2. Taylor expanded in uy around 0

   \[\leadsto \color{blue}{maxCos \cdot \left(ux \cdot \left(zi \cdot \left(1 - ux\right)\right)\right) + xi \cdot \sqrt{1 - {maxCos}^{2} \cdot \left({ux}^{2} \cdot {\left(1 - ux\right)}^{2}\right)}} \]

4. Applied rewrites 51.6%

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

5. Taylor expanded in ux around 0

   \[\leadsto xi + \color{blue}{maxCos \cdot \left(ux \cdot zi\right)} \]

7. Applied rewrites 49.3%

   \[\leadsto xi + \color{blue}{maxCos \cdot \left(ux \cdot zi\right)} \]

8. Taylor expanded in xi around 0

   \[\leadsto maxCos \cdot \left(ux \cdot \color{blue}{zi}\right) \]

10. Applied rewrites 12.2%

    \[\leadsto maxCos \cdot \left(ux \cdot \color{blue}{zi}\right) \]
11. Add Preprocessing

Reproduce

herbie shell --seed 2025153 
(FPCore (xi yi zi ux uy maxCos)
  :name "UniformSampleCone 2"
  :precision binary32
  :pre (and (and (and (and (and (and (<= -10000.0 xi) (<= xi 10000.0)) (and (<= -10000.0 yi) (<= yi 10000.0))) (and (<= -10000.0 zi) (<= zi 10000.0))) (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) maxCos) ux) (* (* (- 1.0 ux) maxCos) ux))))) xi) (* (* (sin (* (* uy 2.0) PI)) (sqrt (- 1.0 (* (* (* (- 1.0 ux) maxCos) ux) (* (* (- 1.0 ux) maxCos) ux))))) yi)) (* (* (* (- 1.0 ux) maxCos) ux) zi)))