
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 (PI)) u2))))
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)
\end{array}
Sampling outcomes in binary32 precision:
Herbie found 7 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 (PI)) u2))))
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)
\end{array}
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 (PI)) u2))))
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)
\end{array}
Initial program 55.3%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (sqrt (- (log (- 1.0 u1)))))
(t_1 (* (* 2.0 (PI)) u2))
(t_2 (cos t_1)))
(if (<= (* t_0 t_2) 0.0) (* t_1 t_2) t_0)))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{-\log \left(1 - u1\right)}\\
t_1 := \left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\\
t_2 := \cos t\_1\\
\mathbf{if}\;t\_0 \cdot t\_2 \leq 0:\\
\;\;\;\;t\_1 \cdot t\_2\\
\mathbf{else}:\\
\;\;\;\;t\_0\\
\end{array}
\end{array}
if (*.f32 (sqrt.f32 (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1)))) (cos.f32 (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2))) < -0.0Initial program 12.6%
Taylor expanded in u1 around 0
Applied rewrites19.0%
if -0.0 < (*.f32 (sqrt.f32 (neg.f32 (log.f32 (-.f32 #s(literal 1 binary32) u1)))) (cos.f32 (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2))) Initial program 71.0%
Taylor expanded in u1 around 0
Applied rewrites62.5%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (- (log (- 1.0 u1)))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-logf((1.0f - u1)));
}
real(4) function code(costheta_i, u1, u2)
real(4), intent (in) :: costheta_i
real(4), intent (in) :: u1
real(4), intent (in) :: u2
code = sqrt(-log((1.0e0 - u1)))
end function
function code(cosTheta_i, u1, u2) return sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) end
function tmp = code(cosTheta_i, u1, u2) tmp = sqrt(-log((single(1.0) - u1))); end
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)}
\end{array}
Initial program 55.3%
Taylor expanded in u1 around 0
Applied rewrites47.4%
(FPCore (cosTheta_i u1 u2) :precision binary32 (cos (* (* 2.0 (PI)) u2)))
\begin{array}{l}
\\
\cos \left(\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2\right)
\end{array}
Initial program 55.3%
Taylor expanded in u1 around 0
Applied rewrites17.2%
Taylor expanded in u1 around 0
Applied rewrites19.8%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (* 2.0 (PI)) u2))
\begin{array}{l}
\\
\left(2 \cdot \mathsf{PI}\left(\right)\right) \cdot u2
\end{array}
Initial program 55.3%
Taylor expanded in u1 around 0
Applied rewrites17.2%
Taylor expanded in u1 around 0
Applied rewrites19.8%
Taylor expanded in u1 around 0
Applied rewrites19.2%
(FPCore (cosTheta_i u1 u2) :precision binary32 (- 1.0 u1))
float code(float cosTheta_i, float u1, float u2) {
return 1.0f - u1;
}
real(4) function code(costheta_i, u1, u2)
real(4), intent (in) :: costheta_i
real(4), intent (in) :: u1
real(4), intent (in) :: u2
code = 1.0e0 - u1
end function
function code(cosTheta_i, u1, u2) return Float32(Float32(1.0) - u1) end
function tmp = code(cosTheta_i, u1, u2) tmp = single(1.0) - u1; end
\begin{array}{l}
\\
1 - u1
\end{array}
Initial program 55.3%
Taylor expanded in u1 around 0
Applied rewrites4.8%
Taylor expanded in u1 around 0
Applied rewrites19.2%
(FPCore (cosTheta_i u1 u2) :precision binary32 (PI))
\begin{array}{l}
\\
\mathsf{PI}\left(\right)
\end{array}
Initial program 55.3%
Taylor expanded in u1 around 0
Applied rewrites17.2%
Taylor expanded in u1 around 0
Applied rewrites19.8%
Taylor expanded in u1 around 0
Applied rewrites19.2%
Taylor expanded in u1 around 0
Applied rewrites18.0%
herbie shell --seed 2024321
(FPCore (cosTheta_i u1 u2)
:name "Beckmann Sample, near normal, slope_x"
:precision binary32
:pre (and (and (and (> cosTheta_i 0.9999) (<= cosTheta_i 1.0)) (and (<= 2.328306437e-10 u1) (<= u1 1.0))) (and (<= 2.328306437e-10 u2) (<= u2 1.0)))
(* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 (PI)) u2))))