
(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 10 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
(pow (* alpha alpha) PI)
(fma (* cosTheta cosTheta) (fma alpha alpha -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), fmaf(alpha, alpha, -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), fma(alpha, alpha, 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, \mathsf{fma}\left(\alpha, \alpha, -1\right), 1\right)\right)}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
*-commutativeN/A
log-pow-revN/A
pow2N/A
log-pow-revN/A
lower-log.f32N/A
Applied rewrites98.6%
(FPCore (cosTheta alpha)
:precision binary32
(/
(- (* alpha alpha) 1.0)
(log
(pow
alpha
(* (* 2.0 PI) (fma (* cosTheta cosTheta) (fma alpha alpha -1.0) 1.0))))))
float code(float cosTheta, float alpha) {
return ((alpha * alpha) - 1.0f) / logf(powf(alpha, ((2.0f * ((float) M_PI)) * fmaf((cosTheta * cosTheta), fmaf(alpha, alpha, -1.0f), 1.0f))));
}
function code(cosTheta, alpha) return Float32(Float32(Float32(alpha * alpha) - Float32(1.0)) / log((alpha ^ Float32(Float32(Float32(2.0) * Float32(pi)) * fma(Float32(cosTheta * cosTheta), fma(alpha, alpha, Float32(-1.0)), Float32(1.0)))))) end
\begin{array}{l}
\\
\frac{\alpha \cdot \alpha - 1}{\log \left({\alpha}^{\left(\left(2 \cdot \pi\right) \cdot \mathsf{fma}\left(cosTheta \cdot cosTheta, \mathsf{fma}\left(\alpha, \alpha, -1\right), 1\right)\right)}\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
*-commutativeN/A
log-pow-revN/A
pow2N/A
log-pow-revN/A
lower-log.f32N/A
Applied rewrites98.6%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
pow-powN/A
pow2N/A
+-commutativeN/A
difference-of-sqr--1N/A
difference-of-sqr-1N/A
associate-*l*N/A
*-commutativeN/A
*-commutativeN/A
pow-powN/A
Applied rewrites98.6%
(FPCore (cosTheta alpha) :precision binary32 (/ (fma alpha alpha -1.0) (* (log alpha) (* (fma (* cosTheta cosTheta) (fma alpha alpha -1.0) 1.0) (+ PI PI)))))
float code(float cosTheta, float alpha) {
return fmaf(alpha, alpha, -1.0f) / (logf(alpha) * (fmaf((cosTheta * cosTheta), fmaf(alpha, alpha, -1.0f), 1.0f) * (((float) M_PI) + ((float) M_PI))));
}
function code(cosTheta, alpha) return Float32(fma(alpha, alpha, Float32(-1.0)) / Float32(log(alpha) * Float32(fma(Float32(cosTheta * cosTheta), fma(alpha, alpha, Float32(-1.0)), Float32(1.0)) * Float32(Float32(pi) + Float32(pi))))) end
\begin{array}{l}
\\
\frac{\mathsf{fma}\left(\alpha, \alpha, -1\right)}{\log \alpha \cdot \left(\mathsf{fma}\left(cosTheta \cdot cosTheta, \mathsf{fma}\left(\alpha, \alpha, -1\right), 1\right) \cdot \left(\pi + \pi\right)\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
*-commutativeN/A
log-pow-revN/A
pow2N/A
log-pow-revN/A
lower-log.f32N/A
Applied rewrites98.6%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
pow-powN/A
pow2N/A
+-commutativeN/A
difference-of-sqr--1N/A
difference-of-sqr-1N/A
associate-*l*N/A
*-commutativeN/A
*-commutativeN/A
pow-powN/A
Applied rewrites98.6%
lift-*.f32N/A
lift--.f32N/A
difference-of-sqr-1-revN/A
difference-of-sqr--1-revN/A
lift-fma.f3298.6
lift-log.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
pow-to-expN/A
rem-log-expN/A
lower-*.f32N/A
Applied rewrites98.5%
(FPCore (cosTheta alpha) :precision binary32 (/ (fma alpha alpha -1.0) (* (log alpha) (* (- 1.0 (* cosTheta cosTheta)) (+ PI PI)))))
float code(float cosTheta, float alpha) {
return fmaf(alpha, alpha, -1.0f) / (logf(alpha) * ((1.0f - (cosTheta * cosTheta)) * (((float) M_PI) + ((float) M_PI))));
}
function code(cosTheta, alpha) return Float32(fma(alpha, alpha, Float32(-1.0)) / Float32(log(alpha) * Float32(Float32(Float32(1.0) - Float32(cosTheta * cosTheta)) * Float32(Float32(pi) + Float32(pi))))) end
\begin{array}{l}
\\
\frac{\mathsf{fma}\left(\alpha, \alpha, -1\right)}{\log \alpha \cdot \left(\left(1 - cosTheta \cdot cosTheta\right) \cdot \left(\pi + \pi\right)\right)}
\end{array}
Initial program 98.5%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-+.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift--.f32N/A
*-commutativeN/A
log-pow-revN/A
pow2N/A
log-pow-revN/A
lower-log.f32N/A
Applied rewrites98.6%
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
pow-powN/A
pow2N/A
+-commutativeN/A
difference-of-sqr--1N/A
difference-of-sqr-1N/A
associate-*l*N/A
*-commutativeN/A
*-commutativeN/A
pow-powN/A
Applied rewrites98.6%
lift-*.f32N/A
lift--.f32N/A
difference-of-sqr-1-revN/A
difference-of-sqr--1-revN/A
lift-fma.f3298.6
lift-log.f32N/A
lift-pow.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
pow-to-expN/A
rem-log-expN/A
lower-*.f32N/A
Applied rewrites98.5%
Taylor expanded in alpha around 0
mul-1-negN/A
negate-subN/A
pow2N/A
lift--.f32N/A
lift-*.f3297.6
Applied rewrites97.6%
(FPCore (cosTheta alpha) :precision binary32 (/ (fma alpha alpha -1.0) (* (+ PI PI) (* (- 1.0 (* cosTheta cosTheta)) (log alpha)))))
float code(float cosTheta, float alpha) {
return fmaf(alpha, alpha, -1.0f) / ((((float) M_PI) + ((float) M_PI)) * ((1.0f - (cosTheta * cosTheta)) * logf(alpha)));
}
function code(cosTheta, alpha) return Float32(fma(alpha, alpha, Float32(-1.0)) / Float32(Float32(Float32(pi) + Float32(pi)) * Float32(Float32(Float32(1.0) - Float32(cosTheta * cosTheta)) * log(alpha)))) end
\begin{array}{l}
\\
\frac{\mathsf{fma}\left(\alpha, \alpha, -1\right)}{\left(\pi + \pi\right) \cdot \left(\left(1 - cosTheta \cdot cosTheta\right) \cdot \log \alpha\right)}
\end{array}
Initial program 98.5%
Taylor expanded in alpha around 0
*-commutativeN/A
lower-*.f32N/A
associate-*r*N/A
lower-*.f32N/A
*-commutativeN/A
lower-*.f32N/A
lower-log.f32N/A
lift-PI.f32N/A
+-commutativeN/A
lower-+.f32N/A
mul-1-negN/A
lower-neg.f32N/A
unpow2N/A
lower-*.f3297.5
Applied rewrites97.5%
Applied rewrites97.6%
(FPCore (cosTheta alpha) :precision binary32 (/ (- (* alpha alpha) 1.0) (* (* PI (log (* alpha alpha))) 1.0)))
float code(float cosTheta, float alpha) {
return ((alpha * alpha) - 1.0f) / ((((float) M_PI) * logf((alpha * alpha))) * 1.0f);
}
function code(cosTheta, alpha) return Float32(Float32(Float32(alpha * alpha) - Float32(1.0)) / Float32(Float32(Float32(pi) * log(Float32(alpha * alpha))) * Float32(1.0))) end
function tmp = code(cosTheta, alpha) tmp = ((alpha * alpha) - single(1.0)) / ((single(pi) * log((alpha * alpha))) * single(1.0)); end
\begin{array}{l}
\\
\frac{\alpha \cdot \alpha - 1}{\left(\pi \cdot \log \left(\alpha \cdot \alpha\right)\right) \cdot 1}
\end{array}
Initial program 98.5%
Taylor expanded in cosTheta around 0
Applied rewrites95.2%
(FPCore (cosTheta alpha) :precision binary32 (/ (fma alpha alpha -1.0) (* (+ PI PI) (log alpha))))
float code(float cosTheta, float alpha) {
return fmaf(alpha, alpha, -1.0f) / ((((float) M_PI) + ((float) M_PI)) * logf(alpha));
}
function code(cosTheta, alpha) return Float32(fma(alpha, alpha, Float32(-1.0)) / Float32(Float32(Float32(pi) + Float32(pi)) * log(alpha))) end
\begin{array}{l}
\\
\frac{\mathsf{fma}\left(\alpha, \alpha, -1\right)}{\left(\pi + \pi\right) \cdot \log \alpha}
\end{array}
Initial program 98.5%
Taylor expanded in alpha around inf
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
lower-*.f32N/A
unpow2N/A
lower-*.f32N/A
lift-PI.f32N/A
log-recN/A
lower-neg.f32N/A
lower-log.f321.8
Applied rewrites1.8%
lift-/.f32N/A
metadata-evalN/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-neg.f32N/A
distribute-rgt-neg-outN/A
pow2N/A
associate-*r*N/A
frac-2negN/A
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
pow2N/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-log.f321.8
Applied rewrites1.8%
Taylor expanded in cosTheta around 0
Applied rewrites95.2%
(FPCore (cosTheta alpha) :precision binary32 (/ 0.5 (* (* (log alpha) PI) (* cosTheta cosTheta))))
float code(float cosTheta, float alpha) {
return 0.5f / ((logf(alpha) * ((float) M_PI)) * (cosTheta * cosTheta));
}
function code(cosTheta, alpha) return Float32(Float32(0.5) / Float32(Float32(log(alpha) * Float32(pi)) * Float32(cosTheta * cosTheta))) end
function tmp = code(cosTheta, alpha) tmp = single(0.5) / ((log(alpha) * single(pi)) * (cosTheta * cosTheta)); end
\begin{array}{l}
\\
\frac{0.5}{\left(\log \alpha \cdot \pi\right) \cdot \left(cosTheta \cdot cosTheta\right)}
\end{array}
Initial program 98.5%
Taylor expanded in alpha around inf
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
lower-*.f32N/A
unpow2N/A
lower-*.f32N/A
lift-PI.f32N/A
log-recN/A
lower-neg.f32N/A
lower-log.f321.8
Applied rewrites1.8%
lift-/.f32N/A
metadata-evalN/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-neg.f32N/A
distribute-rgt-neg-outN/A
pow2N/A
associate-*r*N/A
frac-2negN/A
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
pow2N/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-log.f321.8
Applied rewrites1.8%
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
pow2N/A
lift-log.f32N/A
associate-*r*N/A
*-commutativeN/A
lower-*.f32N/A
*-commutativeN/A
lift-log.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
pow2N/A
lift-*.f321.8
Applied rewrites1.8%
(FPCore (cosTheta alpha) :precision binary32 (/ 0.5 (* (* cosTheta (* cosTheta PI)) (log alpha))))
float code(float cosTheta, float alpha) {
return 0.5f / ((cosTheta * (cosTheta * ((float) M_PI))) * logf(alpha));
}
function code(cosTheta, alpha) return Float32(Float32(0.5) / Float32(Float32(cosTheta * Float32(cosTheta * Float32(pi))) * log(alpha))) end
function tmp = code(cosTheta, alpha) tmp = single(0.5) / ((cosTheta * (cosTheta * single(pi))) * log(alpha)); end
\begin{array}{l}
\\
\frac{0.5}{\left(cosTheta \cdot \left(cosTheta \cdot \pi\right)\right) \cdot \log \alpha}
\end{array}
Initial program 98.5%
Taylor expanded in alpha around inf
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
lower-*.f32N/A
unpow2N/A
lower-*.f32N/A
lift-PI.f32N/A
log-recN/A
lower-neg.f32N/A
lower-log.f321.8
Applied rewrites1.8%
lift-/.f32N/A
metadata-evalN/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-neg.f32N/A
distribute-rgt-neg-outN/A
pow2N/A
associate-*r*N/A
frac-2negN/A
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
pow2N/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-log.f321.8
Applied rewrites1.8%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
associate-*l*N/A
lower-*.f32N/A
lower-*.f32N/A
lift-PI.f321.8
Applied rewrites1.8%
(FPCore (cosTheta alpha) :precision binary32 (/ 0.5 (* cosTheta (* (* PI cosTheta) (log alpha)))))
float code(float cosTheta, float alpha) {
return 0.5f / (cosTheta * ((((float) M_PI) * cosTheta) * logf(alpha)));
}
function code(cosTheta, alpha) return Float32(Float32(0.5) / Float32(cosTheta * Float32(Float32(Float32(pi) * cosTheta) * log(alpha)))) end
function tmp = code(cosTheta, alpha) tmp = single(0.5) / (cosTheta * ((single(pi) * cosTheta) * log(alpha))); end
\begin{array}{l}
\\
\frac{0.5}{cosTheta \cdot \left(\left(\pi \cdot cosTheta\right) \cdot \log \alpha\right)}
\end{array}
Initial program 98.5%
Taylor expanded in alpha around inf
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
lower-*.f32N/A
unpow2N/A
lower-*.f32N/A
lift-PI.f32N/A
log-recN/A
lower-neg.f32N/A
lower-log.f321.8
Applied rewrites1.8%
lift-/.f32N/A
metadata-evalN/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
lift-neg.f32N/A
distribute-rgt-neg-outN/A
pow2N/A
associate-*r*N/A
frac-2negN/A
lower-/.f32N/A
associate-*r*N/A
lower-*.f32N/A
pow2N/A
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-log.f321.8
Applied rewrites1.8%
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
associate-*l*N/A
lower-*.f32N/A
lower-*.f32N/A
lift-PI.f321.8
Applied rewrites1.8%
lift-*.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-log.f32N/A
associate-*l*N/A
lower-*.f32N/A
lower-*.f32N/A
*-commutativeN/A
lower-*.f32N/A
lift-PI.f32N/A
lift-log.f321.8
Applied rewrites1.8%
herbie shell --seed 2025112
(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)))))