
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log (- 1.0 u1)))) (sin (* (* 2.0 PI) u2))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-logf((1.0f - u1))) * sinf(((2.0f * ((float) M_PI)) * u2));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) * sin(Float32(Float32(Float32(2.0) * Float32(pi)) * u2))) end
function tmp = code(cosTheta_i, u1, u2) tmp = sqrt(-log((single(1.0) - u1))) * sin(((single(2.0) * single(pi)) * u2)); end
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)} \cdot \sin \left(\left(2 \cdot \pi\right) \cdot u2\right)
\end{array}
Sampling outcomes in binary32 precision:
Herbie found 1 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log (- 1.0 u1)))) (sin (* (* 2.0 PI) u2))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-logf((1.0f - u1))) * sinf(((2.0f * ((float) M_PI)) * u2));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) * sin(Float32(Float32(Float32(2.0) * Float32(pi)) * u2))) end
function tmp = code(cosTheta_i, u1, u2) tmp = sqrt(-log((single(1.0) - u1))) * sin(((single(2.0) * single(pi)) * u2)); end
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)} \cdot \sin \left(\left(2 \cdot \pi\right) \cdot u2\right)
\end{array}
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (* (- (* (- (* -0.25 u1) 0.3333333333333333) u1) 0.5) u1))
(t_1 (sin (* (* 2.0 PI) u2))))
(if (<= u1 0.05000000074505806)
(*
(sqrt
(-
(*
(/
(- (pow t_0 3.0) 1.0)
(fma t_0 (* (- (* -0.3333333333333333 u1) 0.5) u1) (+ 1.0 t_0)))
u1)))
t_1)
(* (sqrt (- (log (- 1.0 u1)))) t_1))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = ((((-0.25f * u1) - 0.3333333333333333f) * u1) - 0.5f) * u1;
float t_1 = sinf(((2.0f * ((float) M_PI)) * u2));
float tmp;
if (u1 <= 0.05000000074505806f) {
tmp = sqrtf(-(((powf(t_0, 3.0f) - 1.0f) / fmaf(t_0, (((-0.3333333333333333f * u1) - 0.5f) * u1), (1.0f + t_0))) * u1)) * t_1;
} else {
tmp = sqrtf(-logf((1.0f - u1))) * t_1;
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = Float32(Float32(Float32(Float32(Float32(Float32(-0.25) * u1) - Float32(0.3333333333333333)) * u1) - Float32(0.5)) * u1) t_1 = sin(Float32(Float32(Float32(2.0) * Float32(pi)) * u2)) tmp = Float32(0.0) if (u1 <= Float32(0.05000000074505806)) tmp = Float32(sqrt(Float32(-Float32(Float32(Float32((t_0 ^ Float32(3.0)) - Float32(1.0)) / fma(t_0, Float32(Float32(Float32(Float32(-0.3333333333333333) * u1) - Float32(0.5)) * u1), Float32(Float32(1.0) + t_0))) * u1))) * t_1); else tmp = Float32(sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) * t_1); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \left(\left(-0.25 \cdot u1 - 0.3333333333333333\right) \cdot u1 - 0.5\right) \cdot u1\\
t_1 := \sin \left(\left(2 \cdot \pi\right) \cdot u2\right)\\
\mathbf{if}\;u1 \leq 0.05000000074505806:\\
\;\;\;\;\sqrt{-\frac{{t\_0}^{3} - 1}{\mathsf{fma}\left(t\_0, \left(-0.3333333333333333 \cdot u1 - 0.5\right) \cdot u1, 1 + t\_0\right)} \cdot u1} \cdot t\_1\\
\mathbf{else}:\\
\;\;\;\;\sqrt{-\log \left(1 - u1\right)} \cdot t\_1\\
\end{array}
\end{array}
if u1 < 0.0500000007Initial program 52.2%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
lower--.f32N/A
*-commutativeN/A
lower-*.f32N/A
lower--.f32N/A
*-commutativeN/A
lower-*.f32N/A
lower--.f32N/A
lower-*.f3298.1
Applied rewrites98.1%
lift--.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
lift--.f32N/A
lift-*.f32N/A
flip3--N/A
lower-/.f32N/A
Applied rewrites98.1%
Taylor expanded in u1 around 0
Applied rewrites98.2%
if 0.0500000007 < u1 Initial program 97.9%
Final simplification98.2%
herbie shell --seed 2025059
(FPCore (cosTheta_i u1 u2)
:name "Beckmann Sample, near normal, slope_y"
: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)))) (sin (* (* 2.0 PI) u2))))