
(FPCore (cosTheta alpha)
:precision binary32
(let* ((t_0 (- (* alpha alpha) 1.0)))
(/
t_0
(* (* PI (log (* alpha alpha))) (+ 1.0 (* (* t_0 cosTheta) cosTheta))))))
float code(float cosTheta, float alpha) {
float t_0 = (alpha * alpha) - 1.0f;
return t_0 / ((((float) M_PI) * logf((alpha * alpha))) * (1.0f + ((t_0 * cosTheta) * cosTheta)));
}
function code(cosTheta, alpha) t_0 = Float32(Float32(alpha * alpha) - Float32(1.0)) return Float32(t_0 / Float32(Float32(Float32(pi) * log(Float32(alpha * alpha))) * Float32(Float32(1.0) + Float32(Float32(t_0 * cosTheta) * cosTheta)))) end
function tmp = code(cosTheta, alpha) t_0 = (alpha * alpha) - single(1.0); tmp = t_0 / ((single(pi) * log((alpha * alpha))) * (single(1.0) + ((t_0 * cosTheta) * cosTheta))); end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \alpha \cdot \alpha - 1\\
\frac{t\_0}{\left(\pi \cdot \log \left(\alpha \cdot \alpha\right)\right) \cdot \left(1 + \left(t\_0 \cdot cosTheta\right) \cdot cosTheta\right)}
\end{array}
\end{array}
Herbie found 11 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (cosTheta alpha)
:precision binary32
(let* ((t_0 (- (* alpha alpha) 1.0)))
(/
t_0
(* (* PI (log (* alpha alpha))) (+ 1.0 (* (* t_0 cosTheta) cosTheta))))))
float code(float cosTheta, float alpha) {
float t_0 = (alpha * alpha) - 1.0f;
return t_0 / ((((float) M_PI) * logf((alpha * alpha))) * (1.0f + ((t_0 * cosTheta) * cosTheta)));
}
function code(cosTheta, alpha) t_0 = Float32(Float32(alpha * alpha) - Float32(1.0)) return Float32(t_0 / Float32(Float32(Float32(pi) * log(Float32(alpha * alpha))) * Float32(Float32(1.0) + Float32(Float32(t_0 * cosTheta) * cosTheta)))) end
function tmp = code(cosTheta, alpha) t_0 = (alpha * alpha) - single(1.0); tmp = t_0 / ((single(pi) * log((alpha * alpha))) * (single(1.0) + ((t_0 * cosTheta) * cosTheta))); end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \alpha \cdot \alpha - 1\\
\frac{t\_0}{\left(\pi \cdot \log \left(\alpha \cdot \alpha\right)\right) \cdot \left(1 + \left(t\_0 \cdot cosTheta\right) \cdot cosTheta\right)}
\end{array}
\end{array}
(FPCore (cosTheta alpha)
:precision binary32
(/
(- (* alpha alpha) 1.0)
(log
(pow
(* alpha alpha)
(fma (* (* (fma alpha alpha -1.0) PI) cosTheta) cosTheta PI)))))
float code(float cosTheta, float alpha) {
return ((alpha * alpha) - 1.0f) / logf(powf((alpha * alpha), fmaf(((fmaf(alpha, alpha, -1.0f) * ((float) M_PI)) * cosTheta), cosTheta, ((float) M_PI))));
}
function code(cosTheta, alpha) return Float32(Float32(Float32(alpha * alpha) - Float32(1.0)) / log((Float32(alpha * alpha) ^ fma(Float32(Float32(fma(alpha, alpha, Float32(-1.0)) * Float32(pi)) * cosTheta), cosTheta, Float32(pi))))) end
\begin{array}{l}
\\
\frac{\alpha \cdot \alpha - 1}{\log \left({\left(\alpha \cdot \alpha\right)}^{\left(\mathsf{fma}\left(\left(\mathsf{fma}\left(\alpha, \alpha, -1\right) \cdot \pi\right) \cdot cosTheta, cosTheta, \pi\right)\right)}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-*.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
*-commutativeN/A
lift-*.f32N/A
log-pow-revN/A
Applied rewrites98.7%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift--.f32N/A
lift-*.f32N/A
pow-powN/A
pow2N/A
lower-pow.f32N/A
pow2N/A
lift-*.f32N/A
lower-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-fma.f32N/A
Applied rewrites98.7%
Taylor expanded in cosTheta around 0
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
*-commutativeN/A
lower-*.f32N/A
pow2N/A
difference-of-sqr-1-revN/A
difference-of-sqr--1-revN/A
lower-fma.f32N/A
lift-PI.f32N/A
pow2N/A
lift-*.f32N/A
lift-PI.f3298.7
Applied rewrites98.7%
lift-*.f32N/A
lift-PI.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
pow2N/A
lower-fma.f32N/A
pow2N/A
associate-*r*N/A
lower-fma.f32N/A
lower-*.f32N/A
lift-fma.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-PI.f3298.7
Applied rewrites98.7%
(FPCore (cosTheta alpha)
:precision binary32
(/
(fma alpha alpha -1.0)
(log
(pow
(* alpha alpha)
(fma (* (fma alpha alpha -1.0) PI) (* cosTheta cosTheta) PI)))))
float code(float cosTheta, float alpha) {
return fmaf(alpha, alpha, -1.0f) / logf(powf((alpha * alpha), fmaf((fmaf(alpha, alpha, -1.0f) * ((float) M_PI)), (cosTheta * cosTheta), ((float) M_PI))));
}
function code(cosTheta, alpha) return Float32(fma(alpha, alpha, Float32(-1.0)) / log((Float32(alpha * alpha) ^ fma(Float32(fma(alpha, alpha, Float32(-1.0)) * Float32(pi)), Float32(cosTheta * cosTheta), Float32(pi))))) end
\begin{array}{l}
\\
\frac{\mathsf{fma}\left(\alpha, \alpha, -1\right)}{\log \left({\left(\alpha \cdot \alpha\right)}^{\left(\mathsf{fma}\left(\mathsf{fma}\left(\alpha, \alpha, -1\right) \cdot \pi, cosTheta \cdot cosTheta, \pi\right)\right)}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-*.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
*-commutativeN/A
lift-*.f32N/A
log-pow-revN/A
Applied rewrites98.7%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift--.f32N/A
lift-*.f32N/A
pow-powN/A
pow2N/A
lower-pow.f32N/A
pow2N/A
lift-*.f32N/A
lower-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-fma.f32N/A
Applied rewrites98.7%
Taylor expanded in cosTheta around 0
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
*-commutativeN/A
lower-*.f32N/A
pow2N/A
difference-of-sqr-1-revN/A
difference-of-sqr--1-revN/A
lower-fma.f32N/A
lift-PI.f32N/A
pow2N/A
lift-*.f32N/A
lift-PI.f3298.7
Applied rewrites98.7%
lift--.f32N/A
lift-*.f32N/A
difference-of-sqr-1N/A
difference-of-sqr--1N/A
lift-fma.f3298.6
Applied rewrites98.6%
(FPCore (cosTheta alpha) :precision binary32 (/ (/ (fma alpha alpha -1.0) (fma (* cosTheta cosTheta) (fma alpha alpha -1.0) 1.0)) (log (pow (* alpha alpha) PI))))
float code(float cosTheta, float alpha) {
return (fmaf(alpha, alpha, -1.0f) / fmaf((cosTheta * cosTheta), fmaf(alpha, alpha, -1.0f), 1.0f)) / logf(powf((alpha * alpha), ((float) M_PI)));
}
function code(cosTheta, alpha) return Float32(Float32(fma(alpha, alpha, Float32(-1.0)) / fma(Float32(cosTheta * cosTheta), fma(alpha, alpha, Float32(-1.0)), Float32(1.0))) / log((Float32(alpha * alpha) ^ Float32(pi)))) end
\begin{array}{l}
\\
\frac{\frac{\mathsf{fma}\left(\alpha, \alpha, -1\right)}{\mathsf{fma}\left(cosTheta \cdot cosTheta, \mathsf{fma}\left(\alpha, \alpha, -1\right), 1\right)}}{\log \left({\left(\alpha \cdot \alpha\right)}^{\pi}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-*.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
*-commutativeN/A
lift-*.f32N/A
log-pow-revN/A
Applied rewrites98.7%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift--.f32N/A
lift-*.f32N/A
pow-powN/A
pow2N/A
lower-pow.f32N/A
pow2N/A
lift-*.f32N/A
lower-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-fma.f32N/A
Applied rewrites98.7%
Applied rewrites98.6%
(FPCore (cosTheta alpha) :precision binary32 (/ (/ (/ (fma alpha alpha -1.0) PI) (log (* alpha alpha))) (fma (* cosTheta cosTheta) (fma alpha alpha -1.0) 1.0)))
float code(float cosTheta, float alpha) {
return ((fmaf(alpha, alpha, -1.0f) / ((float) M_PI)) / logf((alpha * alpha))) / fmaf((cosTheta * cosTheta), fmaf(alpha, alpha, -1.0f), 1.0f);
}
function code(cosTheta, alpha) return Float32(Float32(Float32(fma(alpha, alpha, Float32(-1.0)) / Float32(pi)) / log(Float32(alpha * alpha))) / fma(Float32(cosTheta * cosTheta), fma(alpha, alpha, Float32(-1.0)), Float32(1.0))) end
\begin{array}{l}
\\
\frac{\frac{\frac{\mathsf{fma}\left(\alpha, \alpha, -1\right)}{\pi}}{\log \left(\alpha \cdot \alpha\right)}}{\mathsf{fma}\left(cosTheta \cdot cosTheta, \mathsf{fma}\left(\alpha, \alpha, -1\right), 1\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-*.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
*-commutativeN/A
lift-*.f32N/A
log-pow-revN/A
Applied rewrites98.7%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift--.f32N/A
lift-*.f32N/A
pow-powN/A
pow2N/A
lower-pow.f32N/A
pow2N/A
lift-*.f32N/A
lower-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-fma.f32N/A
Applied rewrites98.7%
Applied rewrites98.3%
(FPCore (cosTheta alpha) :precision binary32 (/ (- (* alpha alpha) 1.0) (log (pow (pow (* alpha alpha) PI) (fma (* cosTheta cosTheta) -1.0 1.0)))))
float code(float cosTheta, float alpha) {
return ((alpha * alpha) - 1.0f) / logf(powf(powf((alpha * alpha), ((float) M_PI)), fmaf((cosTheta * cosTheta), -1.0f, 1.0f)));
}
function code(cosTheta, alpha) return Float32(Float32(Float32(alpha * alpha) - Float32(1.0)) / log(((Float32(alpha * alpha) ^ Float32(pi)) ^ fma(Float32(cosTheta * cosTheta), Float32(-1.0), Float32(1.0))))) end
\begin{array}{l}
\\
\frac{\alpha \cdot \alpha - 1}{\log \left({\left({\left(\alpha \cdot \alpha\right)}^{\pi}\right)}^{\left(\mathsf{fma}\left(cosTheta \cdot cosTheta, -1, 1\right)\right)}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-*.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
*-commutativeN/A
lift-*.f32N/A
log-pow-revN/A
Applied rewrites98.7%
Taylor expanded in alpha around 0
Applied rewrites97.7%
(FPCore (cosTheta alpha) :precision binary32 (/ (- (* alpha alpha) 1.0) (log (pow (* alpha alpha) (fma (- PI) (* cosTheta cosTheta) PI)))))
float code(float cosTheta, float alpha) {
return ((alpha * alpha) - 1.0f) / logf(powf((alpha * alpha), fmaf(-((float) M_PI), (cosTheta * cosTheta), ((float) M_PI))));
}
function code(cosTheta, alpha) return Float32(Float32(Float32(alpha * alpha) - Float32(1.0)) / log((Float32(alpha * alpha) ^ fma(Float32(-Float32(pi)), Float32(cosTheta * cosTheta), Float32(pi))))) end
\begin{array}{l}
\\
\frac{\alpha \cdot \alpha - 1}{\log \left({\left(\alpha \cdot \alpha\right)}^{\left(\mathsf{fma}\left(-\pi, cosTheta \cdot cosTheta, \pi\right)\right)}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-*.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
*-commutativeN/A
lift-*.f32N/A
log-pow-revN/A
Applied rewrites98.7%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift--.f32N/A
lift-*.f32N/A
pow-powN/A
pow2N/A
lower-pow.f32N/A
pow2N/A
lift-*.f32N/A
lower-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-fma.f32N/A
Applied rewrites98.7%
Taylor expanded in cosTheta around 0
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
*-commutativeN/A
lower-*.f32N/A
pow2N/A
difference-of-sqr-1-revN/A
difference-of-sqr--1-revN/A
lower-fma.f32N/A
lift-PI.f32N/A
pow2N/A
lift-*.f32N/A
lift-PI.f3298.7
Applied rewrites98.7%
Taylor expanded in alpha around 0
mul-1-negN/A
lower-neg.f32N/A
lift-PI.f3297.7
Applied rewrites97.7%
(FPCore (cosTheta alpha) :precision binary32 (/ (- (* alpha alpha) 1.0) (log (pow (* alpha alpha) PI))))
float code(float cosTheta, float alpha) {
return ((alpha * alpha) - 1.0f) / logf(powf((alpha * alpha), ((float) M_PI)));
}
function code(cosTheta, alpha) return Float32(Float32(Float32(alpha * alpha) - Float32(1.0)) / log((Float32(alpha * alpha) ^ Float32(pi)))) end
function tmp = code(cosTheta, alpha) tmp = ((alpha * alpha) - single(1.0)) / log(((alpha * alpha) ^ single(pi))); end
\begin{array}{l}
\\
\frac{\alpha \cdot \alpha - 1}{\log \left({\left(\alpha \cdot \alpha\right)}^{\pi}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-*.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
*-commutativeN/A
lift-*.f32N/A
log-pow-revN/A
Applied rewrites98.7%
Taylor expanded in cosTheta around 0
lower-pow.f32N/A
pow2N/A
lift-*.f32N/A
lift-PI.f3295.4
Applied rewrites95.4%
(FPCore (cosTheta alpha) :precision binary32 (/ (- (* alpha alpha) 1.0) (log (pow (pow alpha PI) 2.0))))
float code(float cosTheta, float alpha) {
return ((alpha * alpha) - 1.0f) / logf(powf(powf(alpha, ((float) M_PI)), 2.0f));
}
function code(cosTheta, alpha) return Float32(Float32(Float32(alpha * alpha) - Float32(1.0)) / log(((alpha ^ Float32(pi)) ^ Float32(2.0)))) end
function tmp = code(cosTheta, alpha) tmp = ((alpha * alpha) - single(1.0)) / log(((alpha ^ single(pi)) ^ single(2.0))); end
\begin{array}{l}
\\
\frac{\alpha \cdot \alpha - 1}{\log \left({\left({\alpha}^{\pi}\right)}^{2}\right)}
\end{array}
Initial program 98.5%
Taylor expanded in cosTheta around 0
log-pow-revN/A
pow-powN/A
log-pow-revN/A
associate-*r*N/A
log-pow-revN/A
log-pow-revN/A
lower-log.f32N/A
lower-pow.f32N/A
lower-pow.f32N/A
lift-PI.f3295.4
Applied rewrites95.4%
(FPCore (cosTheta alpha) :precision binary32 (/ (* 0.5 (fma alpha alpha -1.0)) (* (log alpha) PI)))
float code(float cosTheta, float alpha) {
return (0.5f * fmaf(alpha, alpha, -1.0f)) / (logf(alpha) * ((float) M_PI));
}
function code(cosTheta, alpha) return Float32(Float32(Float32(0.5) * fma(alpha, alpha, Float32(-1.0))) / Float32(log(alpha) * Float32(pi))) end
\begin{array}{l}
\\
\frac{0.5 \cdot \mathsf{fma}\left(\alpha, \alpha, -1\right)}{\log \alpha \cdot \pi}
\end{array}
Initial program 98.5%
Taylor expanded in cosTheta around 0
Applied rewrites95.3%
Taylor expanded in alpha around 0
Applied rewrites65.3%
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
log-pow-revN/A
lift-*.f32N/A
unpow-prod-downN/A
unpow2N/A
log-powN/A
log-pow-revN/A
*-commutativeN/A
lower-*.f32N/A
*-commutativeN/A
lower-*.f32N/A
lower-log.f32N/A
lift-PI.f3265.3
Applied rewrites65.3%
Taylor expanded in cosTheta around 0
pow2N/A
pow-to-expN/A
*-commutativeN/A
count-2-revN/A
log-prodN/A
*-commutativeN/A
*-commutativeN/A
associate-*r*N/A
log-pow-revN/A
pow-powN/A
pow2N/A
log-pow-revN/A
associate-*r/N/A
Applied rewrites95.3%
(FPCore (cosTheta alpha) :precision binary32 (/ -1.0 (* (log (pow (pow alpha PI) 2.0)) 1.0)))
float code(float cosTheta, float alpha) {
return -1.0f / (logf(powf(powf(alpha, ((float) M_PI)), 2.0f)) * 1.0f);
}
function code(cosTheta, alpha) return Float32(Float32(-1.0) / Float32(log(((alpha ^ Float32(pi)) ^ Float32(2.0))) * Float32(1.0))) end
function tmp = code(cosTheta, alpha) tmp = single(-1.0) / (log(((alpha ^ single(pi)) ^ single(2.0))) * single(1.0)); end
\begin{array}{l}
\\
\frac{-1}{\log \left({\left({\alpha}^{\pi}\right)}^{2}\right) \cdot 1}
\end{array}
Initial program 98.5%
Taylor expanded in cosTheta around 0
Applied rewrites95.3%
Taylor expanded in alpha around 0
Applied rewrites65.3%
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
log-pow-revN/A
lift-*.f32N/A
unpow-prod-downN/A
unpow2N/A
log-powN/A
log-pow-revN/A
*-commutativeN/A
lower-*.f32N/A
*-commutativeN/A
lower-*.f32N/A
lower-log.f32N/A
lift-PI.f3265.3
Applied rewrites65.3%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
*-commutativeN/A
*-commutativeN/A
log-pow-revN/A
log-pow-revN/A
lower-log.f32N/A
lower-pow.f32N/A
lower-pow.f32N/A
lift-PI.f3265.4
Applied rewrites65.4%
(FPCore (cosTheta alpha) :precision binary32 (/ (pow cosTheta -2.0) (log (pow (pow alpha PI) 2.0))))
float code(float cosTheta, float alpha) {
return powf(cosTheta, -2.0f) / logf(powf(powf(alpha, ((float) M_PI)), 2.0f));
}
function code(cosTheta, alpha) return Float32((cosTheta ^ Float32(-2.0)) / log(((alpha ^ Float32(pi)) ^ Float32(2.0)))) end
function tmp = code(cosTheta, alpha) tmp = (cosTheta ^ single(-2.0)) / log(((alpha ^ single(pi)) ^ single(2.0))); end
\begin{array}{l}
\\
\frac{{cosTheta}^{-2}}{\log \left({\left({\alpha}^{\pi}\right)}^{2}\right)}
\end{array}
Initial program 98.5%
Taylor expanded in cosTheta around inf
associate-/r*N/A
lower-/.f32N/A
pow-flipN/A
metadata-evalN/A
lower-pow.f32N/A
log-pow-revN/A
pow-powN/A
log-pow-revN/A
associate-*r*N/A
log-pow-revN/A
log-pow-revN/A
lower-log.f32N/A
lower-pow.f32N/A
lower-pow.f32N/A
lift-PI.f321.9
Applied rewrites1.9%
herbie shell --seed 2025101
(FPCore (cosTheta alpha)
:name "GTR1 distribution"
:precision binary32
:pre (and (and (<= 0.0 cosTheta) (<= cosTheta 1.0)) (and (<= 0.0001 alpha) (<= alpha 1.0)))
(/ (- (* alpha alpha) 1.0) (* (* PI (log (* alpha alpha))) (+ 1.0 (* (* (- (* alpha alpha) 1.0) cosTheta) cosTheta)))))