
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 PI) u2))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-logf((1.0f - u1))) * cosf(((2.0f * ((float) M_PI)) * u2));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) * cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2))) end
function tmp = code(cosTheta_i, u1, u2) tmp = sqrt(-log((single(1.0) - u1))) * cos(((single(2.0) * single(pi)) * u2)); end
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \pi\right) \cdot u2\right)
\end{array}
Sampling outcomes in binary32 precision:
Herbie found 17 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log (- 1.0 u1)))) (cos (* (* 2.0 PI) u2))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-logf((1.0f - u1))) * cosf(((2.0f * ((float) M_PI)) * u2));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) * cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2))) end
function tmp = code(cosTheta_i, u1, u2) tmp = sqrt(-log((single(1.0) - u1))) * cos(((single(2.0) * single(pi)) * u2)); end
\begin{array}{l}
\\
\sqrt{-\log \left(1 - u1\right)} \cdot \cos \left(\left(2 \cdot \pi\right) \cdot u2\right)
\end{array}
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log1p (- u1)))) (+ (+ 0.5 (* 0.5 (cos (* 2.0 (* (pow (cbrt PI) 3.0) u2))))) (- (* 0.5 (cos (* 2.0 (* PI u2)))) 0.5))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-log1pf(-u1)) * ((0.5f + (0.5f * cosf((2.0f * (powf(cbrtf(((float) M_PI)), 3.0f) * u2))))) + ((0.5f * cosf((2.0f * (((float) M_PI) * u2)))) - 0.5f));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(-log1p(Float32(-u1)))) * Float32(Float32(Float32(0.5) + Float32(Float32(0.5) * cos(Float32(Float32(2.0) * Float32((cbrt(Float32(pi)) ^ Float32(3.0)) * u2))))) + Float32(Float32(Float32(0.5) * cos(Float32(Float32(2.0) * Float32(Float32(pi) * u2)))) - Float32(0.5)))) end
\begin{array}{l}
\\
\sqrt{-\mathsf{log1p}\left(-u1\right)} \cdot \left(\left(0.5 + 0.5 \cdot \cos \left(2 \cdot \left({\left(\sqrt[3]{\pi}\right)}^{3} \cdot u2\right)\right)\right) + \left(0.5 \cdot \cos \left(2 \cdot \left(\pi \cdot u2\right)\right) - 0.5\right)\right)
\end{array}
Initial program 58.5%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3298.9
Applied egg-rr98.9%
associate-*l*N/A
cos-2N/A
--lowering--.f32N/A
Applied egg-rr98.7%
rem-cube-cbrtN/A
pow-lowering-pow.f32N/A
cbrt-lowering-cbrt.f32N/A
PI-lowering-PI.f3298.9
Applied egg-rr98.9%
Final simplification98.9%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (cos (* u2 (* 2.0 PI)))))
(if (<= (* t_0 (sqrt (- (log (- 1.0 u1))))) 0.16200000047683716)
(*
t_0
(sqrt (fma (* u1 u1) (fma u1 (fma u1 0.25 0.3333333333333333) 0.5) u1)))
(* (sqrt (- (log1p (- u1)))) (fma u2 (* u2 (* PI (* PI -2.0))) 1.0)))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = cosf((u2 * (2.0f * ((float) M_PI))));
float tmp;
if ((t_0 * sqrtf(-logf((1.0f - u1)))) <= 0.16200000047683716f) {
tmp = t_0 * sqrtf(fmaf((u1 * u1), fmaf(u1, fmaf(u1, 0.25f, 0.3333333333333333f), 0.5f), u1));
} else {
tmp = sqrtf(-log1pf(-u1)) * fmaf(u2, (u2 * (((float) M_PI) * (((float) M_PI) * -2.0f))), 1.0f);
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = cos(Float32(u2 * Float32(Float32(2.0) * Float32(pi)))) tmp = Float32(0.0) if (Float32(t_0 * sqrt(Float32(-log(Float32(Float32(1.0) - u1))))) <= Float32(0.16200000047683716)) tmp = Float32(t_0 * sqrt(fma(Float32(u1 * u1), fma(u1, fma(u1, Float32(0.25), Float32(0.3333333333333333)), Float32(0.5)), u1))); else tmp = Float32(sqrt(Float32(-log1p(Float32(-u1)))) * fma(u2, Float32(u2 * Float32(Float32(pi) * Float32(Float32(pi) * Float32(-2.0)))), Float32(1.0))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos \left(u2 \cdot \left(2 \cdot \pi\right)\right)\\
\mathbf{if}\;t\_0 \cdot \sqrt{-\log \left(1 - u1\right)} \leq 0.16200000047683716:\\
\;\;\;\;t\_0 \cdot \sqrt{\mathsf{fma}\left(u1 \cdot u1, \mathsf{fma}\left(u1, \mathsf{fma}\left(u1, 0.25, 0.3333333333333333\right), 0.5\right), u1\right)}\\
\mathbf{else}:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\right)} \cdot \mathsf{fma}\left(u2, u2 \cdot \left(\pi \cdot \left(\pi \cdot -2\right)\right), 1\right)\\
\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))) < 0.162Initial program 50.4%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
associate-*r*N/A
unpow2N/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
+-commutativeN/A
accelerator-lowering-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
accelerator-lowering-fma.f3298.5
Simplified98.5%
if 0.162 < (*.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 97.4%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3299.1
Applied egg-rr99.1%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
*-lowering-*.f32N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
PI-lowering-PI.f3292.7
Simplified92.7%
Final simplification97.5%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (cos (* u2 (* 2.0 PI)))))
(if (<= (* t_0 (sqrt (- (log (- 1.0 u1))))) 0.0820000022649765)
(* t_0 (sqrt (fma (* u1 u1) (fma u1 0.3333333333333333 0.5) u1)))
(* (sqrt (- (log1p (- u1)))) (fma u2 (* u2 (* PI (* PI -2.0))) 1.0)))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = cosf((u2 * (2.0f * ((float) M_PI))));
float tmp;
if ((t_0 * sqrtf(-logf((1.0f - u1)))) <= 0.0820000022649765f) {
tmp = t_0 * sqrtf(fmaf((u1 * u1), fmaf(u1, 0.3333333333333333f, 0.5f), u1));
} else {
tmp = sqrtf(-log1pf(-u1)) * fmaf(u2, (u2 * (((float) M_PI) * (((float) M_PI) * -2.0f))), 1.0f);
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = cos(Float32(u2 * Float32(Float32(2.0) * Float32(pi)))) tmp = Float32(0.0) if (Float32(t_0 * sqrt(Float32(-log(Float32(Float32(1.0) - u1))))) <= Float32(0.0820000022649765)) tmp = Float32(t_0 * sqrt(fma(Float32(u1 * u1), fma(u1, Float32(0.3333333333333333), Float32(0.5)), u1))); else tmp = Float32(sqrt(Float32(-log1p(Float32(-u1)))) * fma(u2, Float32(u2 * Float32(Float32(pi) * Float32(Float32(pi) * Float32(-2.0)))), Float32(1.0))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos \left(u2 \cdot \left(2 \cdot \pi\right)\right)\\
\mathbf{if}\;t\_0 \cdot \sqrt{-\log \left(1 - u1\right)} \leq 0.0820000022649765:\\
\;\;\;\;t\_0 \cdot \sqrt{\mathsf{fma}\left(u1 \cdot u1, \mathsf{fma}\left(u1, 0.3333333333333333, 0.5\right), u1\right)}\\
\mathbf{else}:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\right)} \cdot \mathsf{fma}\left(u2, u2 \cdot \left(\pi \cdot \left(\pi \cdot -2\right)\right), 1\right)\\
\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))) < 0.0820000023Initial program 46.0%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
associate-*r*N/A
unpow2N/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
+-commutativeN/A
*-commutativeN/A
accelerator-lowering-fma.f3298.2
Simplified98.2%
if 0.0820000023 < (*.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 95.8%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3299.2
Applied egg-rr99.2%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
*-lowering-*.f32N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
PI-lowering-PI.f3294.0
Simplified94.0%
Final simplification97.2%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (cos (* u2 (* 2.0 PI)))))
(if (<= t_0 0.9999997615814209)
(* t_0 (sqrt (fma u1 (* u1 0.5) u1)))
(sqrt (- (log1p (- u1)))))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = cosf((u2 * (2.0f * ((float) M_PI))));
float tmp;
if (t_0 <= 0.9999997615814209f) {
tmp = t_0 * sqrtf(fmaf(u1, (u1 * 0.5f), u1));
} else {
tmp = sqrtf(-log1pf(-u1));
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = cos(Float32(u2 * Float32(Float32(2.0) * Float32(pi)))) tmp = Float32(0.0) if (t_0 <= Float32(0.9999997615814209)) tmp = Float32(t_0 * sqrt(fma(u1, Float32(u1 * Float32(0.5)), u1))); else tmp = sqrt(Float32(-log1p(Float32(-u1)))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos \left(u2 \cdot \left(2 \cdot \pi\right)\right)\\
\mathbf{if}\;t\_0 \leq 0.9999997615814209:\\
\;\;\;\;t\_0 \cdot \sqrt{\mathsf{fma}\left(u1, u1 \cdot 0.5, u1\right)}\\
\mathbf{else}:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\right)}\\
\end{array}
\end{array}
if (cos.f32 (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2)) < 0.999999762Initial program 60.4%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
*-commutativeN/A
*-lowering-*.f3284.7
Simplified84.7%
if 0.999999762 < (cos.f32 (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2)) Initial program 57.4%
Taylor expanded in u2 around 0
Simplified57.4%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3299.5
Applied egg-rr99.5%
Final simplification94.2%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (- (log1p (- u1)))) (cos (* u2 (* 2.0 PI)))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-log1pf(-u1)) * cosf((u2 * (2.0f * ((float) M_PI))));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(-log1p(Float32(-u1)))) * cos(Float32(u2 * Float32(Float32(2.0) * Float32(pi))))) end
\begin{array}{l}
\\
\sqrt{-\mathsf{log1p}\left(-u1\right)} \cdot \cos \left(u2 \cdot \left(2 \cdot \pi\right)\right)
\end{array}
Initial program 58.5%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3298.9
Applied egg-rr98.9%
Final simplification98.9%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (* u2 (* 2.0 PI))))
(if (<= t_0 0.00039999998989515007)
(sqrt (- (log1p (- u1))))
(if (<= t_0 0.20000000298023224)
(*
(/
(sqrt
(*
(- u1)
(fma (fma u1 -0.3333333333333333 -0.5) (* u1 (* u1 -0.5)) -1.0)))
(sqrt (fma u1 (fma u1 -0.3333333333333333 -0.5) 1.0)))
(fma (* PI PI) (* -2.0 (* u2 u2)) 1.0))
(* (cos t_0) (sqrt u1))))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = u2 * (2.0f * ((float) M_PI));
float tmp;
if (t_0 <= 0.00039999998989515007f) {
tmp = sqrtf(-log1pf(-u1));
} else if (t_0 <= 0.20000000298023224f) {
tmp = (sqrtf((-u1 * fmaf(fmaf(u1, -0.3333333333333333f, -0.5f), (u1 * (u1 * -0.5f)), -1.0f))) / sqrtf(fmaf(u1, fmaf(u1, -0.3333333333333333f, -0.5f), 1.0f))) * fmaf((((float) M_PI) * ((float) M_PI)), (-2.0f * (u2 * u2)), 1.0f);
} else {
tmp = cosf(t_0) * sqrtf(u1);
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = Float32(u2 * Float32(Float32(2.0) * Float32(pi))) tmp = Float32(0.0) if (t_0 <= Float32(0.00039999998989515007)) tmp = sqrt(Float32(-log1p(Float32(-u1)))); elseif (t_0 <= Float32(0.20000000298023224)) tmp = Float32(Float32(sqrt(Float32(Float32(-u1) * fma(fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(u1 * Float32(u1 * Float32(-0.5))), Float32(-1.0)))) / sqrt(fma(u1, fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(1.0)))) * fma(Float32(Float32(pi) * Float32(pi)), Float32(Float32(-2.0) * Float32(u2 * u2)), Float32(1.0))); else tmp = Float32(cos(t_0) * sqrt(u1)); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := u2 \cdot \left(2 \cdot \pi\right)\\
\mathbf{if}\;t\_0 \leq 0.00039999998989515007:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\right)}\\
\mathbf{elif}\;t\_0 \leq 0.20000000298023224:\\
\;\;\;\;\frac{\sqrt{\left(-u1\right) \cdot \mathsf{fma}\left(\mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), u1 \cdot \left(u1 \cdot -0.5\right), -1\right)}}{\sqrt{\mathsf{fma}\left(u1, \mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), 1\right)}} \cdot \mathsf{fma}\left(\pi \cdot \pi, -2 \cdot \left(u2 \cdot u2\right), 1\right)\\
\mathbf{else}:\\
\;\;\;\;\cos t\_0 \cdot \sqrt{u1}\\
\end{array}
\end{array}
if (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2) < 3.9999999e-4Initial program 58.3%
Taylor expanded in u2 around 0
Simplified58.2%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3299.5
Applied egg-rr99.5%
if 3.9999999e-4 < (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2) < 0.200000003Initial program 57.5%
Taylor expanded in u1 around 0
*-lowering-*.f32N/A
sub-negN/A
metadata-evalN/A
accelerator-lowering-fma.f32N/A
sub-negN/A
*-commutativeN/A
metadata-evalN/A
accelerator-lowering-fma.f3291.7
Simplified91.7%
distribute-lft-neg-inN/A
flip-+N/A
associate-*r/N/A
sqrt-divN/A
/-lowering-/.f32N/A
Applied egg-rr91.5%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
unpow2N/A
*-lowering-*.f3289.7
Simplified89.7%
Taylor expanded in u1 around 0
*-commutativeN/A
*-lowering-*.f3290.9
Simplified90.9%
if 0.200000003 < (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2) Initial program 61.1%
Taylor expanded in u1 around 0
Simplified70.1%
Final simplification93.5%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (* u2 (* 2.0 PI))))
(if (<= t_0 0.15000000596046448)
(* (sqrt (- (log1p (- u1)))) (fma u2 (* u2 (* PI (* PI -2.0))) 1.0))
(* (cos t_0) (sqrt (fma u1 (* u1 0.5) u1))))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = u2 * (2.0f * ((float) M_PI));
float tmp;
if (t_0 <= 0.15000000596046448f) {
tmp = sqrtf(-log1pf(-u1)) * fmaf(u2, (u2 * (((float) M_PI) * (((float) M_PI) * -2.0f))), 1.0f);
} else {
tmp = cosf(t_0) * sqrtf(fmaf(u1, (u1 * 0.5f), u1));
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = Float32(u2 * Float32(Float32(2.0) * Float32(pi))) tmp = Float32(0.0) if (t_0 <= Float32(0.15000000596046448)) tmp = Float32(sqrt(Float32(-log1p(Float32(-u1)))) * fma(u2, Float32(u2 * Float32(Float32(pi) * Float32(Float32(pi) * Float32(-2.0)))), Float32(1.0))); else tmp = Float32(cos(t_0) * sqrt(fma(u1, Float32(u1 * Float32(0.5)), u1))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := u2 \cdot \left(2 \cdot \pi\right)\\
\mathbf{if}\;t\_0 \leq 0.15000000596046448:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\right)} \cdot \mathsf{fma}\left(u2, u2 \cdot \left(\pi \cdot \left(\pi \cdot -2\right)\right), 1\right)\\
\mathbf{else}:\\
\;\;\;\;\cos t\_0 \cdot \sqrt{\mathsf{fma}\left(u1, u1 \cdot 0.5, u1\right)}\\
\end{array}
\end{array}
if (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2) < 0.150000006Initial program 58.0%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3299.4
Applied egg-rr99.4%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
*-lowering-*.f32N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
PI-lowering-PI.f3298.7
Simplified98.7%
if 0.150000006 < (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2) Initial program 61.6%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
*-commutativeN/A
*-lowering-*.f3281.7
Simplified81.7%
Final simplification96.3%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(if (<= (* u2 (* 2.0 PI)) 0.00039999998989515007)
(sqrt (- (log1p (- u1))))
(*
(/
(sqrt
(*
(- u1)
(fma (fma u1 -0.3333333333333333 -0.5) (* u1 (* u1 -0.5)) -1.0)))
(sqrt (fma u1 (fma u1 -0.3333333333333333 -0.5) 1.0)))
(fma (* PI PI) (* -2.0 (* u2 u2)) 1.0))))
float code(float cosTheta_i, float u1, float u2) {
float tmp;
if ((u2 * (2.0f * ((float) M_PI))) <= 0.00039999998989515007f) {
tmp = sqrtf(-log1pf(-u1));
} else {
tmp = (sqrtf((-u1 * fmaf(fmaf(u1, -0.3333333333333333f, -0.5f), (u1 * (u1 * -0.5f)), -1.0f))) / sqrtf(fmaf(u1, fmaf(u1, -0.3333333333333333f, -0.5f), 1.0f))) * fmaf((((float) M_PI) * ((float) M_PI)), (-2.0f * (u2 * u2)), 1.0f);
}
return tmp;
}
function code(cosTheta_i, u1, u2) tmp = Float32(0.0) if (Float32(u2 * Float32(Float32(2.0) * Float32(pi))) <= Float32(0.00039999998989515007)) tmp = sqrt(Float32(-log1p(Float32(-u1)))); else tmp = Float32(Float32(sqrt(Float32(Float32(-u1) * fma(fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(u1 * Float32(u1 * Float32(-0.5))), Float32(-1.0)))) / sqrt(fma(u1, fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(1.0)))) * fma(Float32(Float32(pi) * Float32(pi)), Float32(Float32(-2.0) * Float32(u2 * u2)), Float32(1.0))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
\mathbf{if}\;u2 \cdot \left(2 \cdot \pi\right) \leq 0.00039999998989515007:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\right)}\\
\mathbf{else}:\\
\;\;\;\;\frac{\sqrt{\left(-u1\right) \cdot \mathsf{fma}\left(\mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), u1 \cdot \left(u1 \cdot -0.5\right), -1\right)}}{\sqrt{\mathsf{fma}\left(u1, \mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), 1\right)}} \cdot \mathsf{fma}\left(\pi \cdot \pi, -2 \cdot \left(u2 \cdot u2\right), 1\right)\\
\end{array}
\end{array}
if (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2) < 3.9999999e-4Initial program 58.3%
Taylor expanded in u2 around 0
Simplified58.2%
sub-negN/A
accelerator-lowering-log1p.f32N/A
neg-lowering-neg.f3299.5
Applied egg-rr99.5%
if 3.9999999e-4 < (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2) Initial program 58.8%
Taylor expanded in u1 around 0
*-lowering-*.f32N/A
sub-negN/A
metadata-evalN/A
accelerator-lowering-fma.f32N/A
sub-negN/A
*-commutativeN/A
metadata-evalN/A
accelerator-lowering-fma.f3289.8
Simplified89.8%
distribute-lft-neg-inN/A
flip-+N/A
associate-*r/N/A
sqrt-divN/A
/-lowering-/.f32N/A
Applied egg-rr89.7%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
unpow2N/A
*-lowering-*.f3271.3
Simplified71.3%
Taylor expanded in u1 around 0
*-commutativeN/A
*-lowering-*.f3272.1
Simplified72.1%
Final simplification89.3%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(*
(/
(sqrt
(* (- u1) (fma (fma u1 -0.3333333333333333 -0.5) (* u1 (* u1 -0.5)) -1.0)))
(sqrt (fma u1 (fma u1 -0.3333333333333333 -0.5) 1.0)))
(fma (* PI PI) (* -2.0 (* u2 u2)) 1.0)))
float code(float cosTheta_i, float u1, float u2) {
return (sqrtf((-u1 * fmaf(fmaf(u1, -0.3333333333333333f, -0.5f), (u1 * (u1 * -0.5f)), -1.0f))) / sqrtf(fmaf(u1, fmaf(u1, -0.3333333333333333f, -0.5f), 1.0f))) * fmaf((((float) M_PI) * ((float) M_PI)), (-2.0f * (u2 * u2)), 1.0f);
}
function code(cosTheta_i, u1, u2) return Float32(Float32(sqrt(Float32(Float32(-u1) * fma(fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(u1 * Float32(u1 * Float32(-0.5))), Float32(-1.0)))) / sqrt(fma(u1, fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(1.0)))) * fma(Float32(Float32(pi) * Float32(pi)), Float32(Float32(-2.0) * Float32(u2 * u2)), Float32(1.0))) end
\begin{array}{l}
\\
\frac{\sqrt{\left(-u1\right) \cdot \mathsf{fma}\left(\mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), u1 \cdot \left(u1 \cdot -0.5\right), -1\right)}}{\sqrt{\mathsf{fma}\left(u1, \mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), 1\right)}} \cdot \mathsf{fma}\left(\pi \cdot \pi, -2 \cdot \left(u2 \cdot u2\right), 1\right)
\end{array}
Initial program 58.5%
Taylor expanded in u1 around 0
*-lowering-*.f32N/A
sub-negN/A
metadata-evalN/A
accelerator-lowering-fma.f32N/A
sub-negN/A
*-commutativeN/A
metadata-evalN/A
accelerator-lowering-fma.f3290.9
Simplified90.9%
distribute-lft-neg-inN/A
flip-+N/A
associate-*r/N/A
sqrt-divN/A
/-lowering-/.f32N/A
Applied egg-rr90.6%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
unpow2N/A
*-lowering-*.f3283.8
Simplified83.8%
Taylor expanded in u1 around 0
*-commutativeN/A
*-lowering-*.f3284.9
Simplified84.9%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (fma (* PI PI) (* -2.0 (* u2 u2)) 1.0) (/ (sqrt (fma u1 (* (* u1 u1) -0.25) u1)) (sqrt (fma u1 (fma u1 -0.3333333333333333 -0.5) 1.0)))))
float code(float cosTheta_i, float u1, float u2) {
return fmaf((((float) M_PI) * ((float) M_PI)), (-2.0f * (u2 * u2)), 1.0f) * (sqrtf(fmaf(u1, ((u1 * u1) * -0.25f), u1)) / sqrtf(fmaf(u1, fmaf(u1, -0.3333333333333333f, -0.5f), 1.0f)));
}
function code(cosTheta_i, u1, u2) return Float32(fma(Float32(Float32(pi) * Float32(pi)), Float32(Float32(-2.0) * Float32(u2 * u2)), Float32(1.0)) * Float32(sqrt(fma(u1, Float32(Float32(u1 * u1) * Float32(-0.25)), u1)) / sqrt(fma(u1, fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(1.0))))) end
\begin{array}{l}
\\
\mathsf{fma}\left(\pi \cdot \pi, -2 \cdot \left(u2 \cdot u2\right), 1\right) \cdot \frac{\sqrt{\mathsf{fma}\left(u1, \left(u1 \cdot u1\right) \cdot -0.25, u1\right)}}{\sqrt{\mathsf{fma}\left(u1, \mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), 1\right)}}
\end{array}
Initial program 58.5%
Taylor expanded in u1 around 0
*-lowering-*.f32N/A
sub-negN/A
metadata-evalN/A
accelerator-lowering-fma.f32N/A
sub-negN/A
*-commutativeN/A
metadata-evalN/A
accelerator-lowering-fma.f3290.9
Simplified90.9%
distribute-lft-neg-inN/A
flip-+N/A
associate-*r/N/A
sqrt-divN/A
/-lowering-/.f32N/A
Applied egg-rr90.6%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
unpow2N/A
*-lowering-*.f3283.8
Simplified83.8%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
*-lowering-*.f32N/A
unpow2N/A
*-lowering-*.f3284.9
Simplified84.9%
Final simplification84.9%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0
(sqrt (* (- u1) (fma u1 (fma u1 -0.3333333333333333 -0.5) -1.0)))))
(fma PI (* (* PI (* u2 (* u2 -2.0))) t_0) t_0)))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = sqrtf((-u1 * fmaf(u1, fmaf(u1, -0.3333333333333333f, -0.5f), -1.0f)));
return fmaf(((float) M_PI), ((((float) M_PI) * (u2 * (u2 * -2.0f))) * t_0), t_0);
}
function code(cosTheta_i, u1, u2) t_0 = sqrt(Float32(Float32(-u1) * fma(u1, fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(-1.0)))) return fma(Float32(pi), Float32(Float32(Float32(pi) * Float32(u2 * Float32(u2 * Float32(-2.0)))) * t_0), t_0) end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{\left(-u1\right) \cdot \mathsf{fma}\left(u1, \mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), -1\right)}\\
\mathsf{fma}\left(\pi, \left(\pi \cdot \left(u2 \cdot \left(u2 \cdot -2\right)\right)\right) \cdot t\_0, t\_0\right)
\end{array}
\end{array}
Initial program 58.5%
Taylor expanded in u1 around 0
*-lowering-*.f32N/A
sub-negN/A
metadata-evalN/A
accelerator-lowering-fma.f32N/A
sub-negN/A
*-commutativeN/A
metadata-evalN/A
accelerator-lowering-fma.f3290.9
Simplified90.9%
distribute-lft-neg-inN/A
flip-+N/A
associate-*r/N/A
sqrt-divN/A
/-lowering-/.f32N/A
Applied egg-rr90.6%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
unpow2N/A
*-lowering-*.f3283.8
Simplified83.8%
Applied egg-rr84.0%
Final simplification84.0%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (* (- u1) (fma u1 (fma u1 -0.3333333333333333 -0.5) -1.0))) (fma PI (* PI (* u2 (* u2 -2.0))) 1.0)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf((-u1 * fmaf(u1, fmaf(u1, -0.3333333333333333f, -0.5f), -1.0f))) * fmaf(((float) M_PI), (((float) M_PI) * (u2 * (u2 * -2.0f))), 1.0f);
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(Float32(-u1) * fma(u1, fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(-1.0)))) * fma(Float32(pi), Float32(Float32(pi) * Float32(u2 * Float32(u2 * Float32(-2.0)))), Float32(1.0))) end
\begin{array}{l}
\\
\sqrt{\left(-u1\right) \cdot \mathsf{fma}\left(u1, \mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), -1\right)} \cdot \mathsf{fma}\left(\pi, \pi \cdot \left(u2 \cdot \left(u2 \cdot -2\right)\right), 1\right)
\end{array}
Initial program 58.5%
Taylor expanded in u1 around 0
*-lowering-*.f32N/A
sub-negN/A
metadata-evalN/A
accelerator-lowering-fma.f32N/A
sub-negN/A
*-commutativeN/A
metadata-evalN/A
accelerator-lowering-fma.f3290.9
Simplified90.9%
distribute-lft-neg-inN/A
flip-+N/A
associate-*r/N/A
sqrt-divN/A
/-lowering-/.f32N/A
Applied egg-rr90.6%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
unpow2N/A
*-lowering-*.f3283.8
Simplified83.8%
Applied egg-rr83.9%
Final simplification83.9%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (fma u2 (* u2 (* PI (* PI -2.0))) 1.0) (sqrt (* (- u1) (fma u1 (fma u1 -0.3333333333333333 -0.5) -1.0)))))
float code(float cosTheta_i, float u1, float u2) {
return fmaf(u2, (u2 * (((float) M_PI) * (((float) M_PI) * -2.0f))), 1.0f) * sqrtf((-u1 * fmaf(u1, fmaf(u1, -0.3333333333333333f, -0.5f), -1.0f)));
}
function code(cosTheta_i, u1, u2) return Float32(fma(u2, Float32(u2 * Float32(Float32(pi) * Float32(Float32(pi) * Float32(-2.0)))), Float32(1.0)) * sqrt(Float32(Float32(-u1) * fma(u1, fma(u1, Float32(-0.3333333333333333), Float32(-0.5)), Float32(-1.0))))) end
\begin{array}{l}
\\
\mathsf{fma}\left(u2, u2 \cdot \left(\pi \cdot \left(\pi \cdot -2\right)\right), 1\right) \cdot \sqrt{\left(-u1\right) \cdot \mathsf{fma}\left(u1, \mathsf{fma}\left(u1, -0.3333333333333333, -0.5\right), -1\right)}
\end{array}
Initial program 58.5%
Taylor expanded in u1 around 0
*-lowering-*.f32N/A
sub-negN/A
metadata-evalN/A
accelerator-lowering-fma.f32N/A
sub-negN/A
*-commutativeN/A
metadata-evalN/A
accelerator-lowering-fma.f3290.9
Simplified90.9%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
accelerator-lowering-fma.f32N/A
*-lowering-*.f32N/A
*-commutativeN/A
unpow2N/A
associate-*l*N/A
*-lowering-*.f32N/A
PI-lowering-PI.f32N/A
*-lowering-*.f32N/A
PI-lowering-PI.f3283.9
Simplified83.9%
Final simplification83.9%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (fma (* u1 u1) (fma u1 (fma u1 0.25 0.3333333333333333) 0.5) u1)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(fmaf((u1 * u1), fmaf(u1, fmaf(u1, 0.25f, 0.3333333333333333f), 0.5f), u1));
}
function code(cosTheta_i, u1, u2) return sqrt(fma(Float32(u1 * u1), fma(u1, fma(u1, Float32(0.25), Float32(0.3333333333333333)), Float32(0.5)), u1)) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(u1 \cdot u1, \mathsf{fma}\left(u1, \mathsf{fma}\left(u1, 0.25, 0.3333333333333333\right), 0.5\right), u1\right)}
\end{array}
Initial program 58.5%
Taylor expanded in u2 around 0
Simplified50.3%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
associate-*r*N/A
unpow2N/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
unpow2N/A
*-lowering-*.f32N/A
+-commutativeN/A
accelerator-lowering-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
accelerator-lowering-fma.f3279.4
Simplified79.4%
Final simplification79.4%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (fma u1 (* u1 (fma u1 0.3333333333333333 0.5)) u1)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(fmaf(u1, (u1 * fmaf(u1, 0.3333333333333333f, 0.5f)), u1));
}
function code(cosTheta_i, u1, u2) return sqrt(fma(u1, Float32(u1 * fma(u1, Float32(0.3333333333333333), Float32(0.5))), u1)) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(u1, u1 \cdot \mathsf{fma}\left(u1, 0.3333333333333333, 0.5\right), u1\right)}
\end{array}
Initial program 58.5%
Taylor expanded in u2 around 0
Simplified50.3%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
*-lowering-*.f32N/A
+-commutativeN/A
*-commutativeN/A
accelerator-lowering-fma.f3277.8
Simplified77.8%
Final simplification77.8%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (fma u1 (* u1 0.5) u1)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(fmaf(u1, (u1 * 0.5f), u1));
}
function code(cosTheta_i, u1, u2) return sqrt(fma(u1, Float32(u1 * Float32(0.5)), u1)) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(u1, u1 \cdot 0.5, u1\right)}
\end{array}
Initial program 58.5%
Taylor expanded in u2 around 0
Simplified50.3%
Taylor expanded in u1 around 0
+-commutativeN/A
distribute-lft-inN/A
*-rgt-identityN/A
accelerator-lowering-fma.f32N/A
*-commutativeN/A
*-lowering-*.f3274.8
Simplified74.8%
Final simplification74.8%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt u1))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(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(u1)
end function
function code(cosTheta_i, u1, u2) return sqrt(u1) end
function tmp = code(cosTheta_i, u1, u2) tmp = sqrt(u1); end
\begin{array}{l}
\\
\sqrt{u1}
\end{array}
Initial program 58.5%
Taylor expanded in u1 around 0
Simplified75.4%
Taylor expanded in u2 around 0
sqrt-lowering-sqrt.f3267.0
Simplified67.0%
herbie shell --seed 2024199
(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))))