
(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 16 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)))) (cos (* (* u2 PI) 2.0))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-log1pf(-u1)) * cosf(((u2 * ((float) M_PI)) * 2.0f));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(-log1p(Float32(-u1)))) * cos(Float32(Float32(u2 * Float32(pi)) * Float32(2.0)))) end
\begin{array}{l}
\\
\sqrt{-\mathsf{log1p}\left(-u1\right)} \cdot \cos \left(\left(u2 \cdot \pi\right) \cdot 2\right)
\end{array}
Initial program 54.6%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3294.5
Applied rewrites94.5%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites94.6%
Taylor expanded in u2 around inf
associate-*l*N/A
*-commutativeN/A
sqrt-prodN/A
lower-*.f32N/A
Applied rewrites99.0%
count-2-rev99.0
lift-*.f32N/A
lift-cos.f32N/A
lift-*.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
lift-sqrt.f32N/A
lift-*.f32N/A
lift-neg.f32N/A
lift-log1p.f32N/A
*-commutativeN/A
Applied rewrites99.0%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (sqrt (- (log (- 1.0 u1))))) (t_1 (cos (* (* 2.0 PI) u2))))
(if (<= (* t_0 t_1) 0.17000000178813934)
(* (sqrt (fma (* (fma 0.3333333333333333 u1 0.5) u1) u1 u1)) t_1)
(* t_0 (fma (* (* PI PI) (* u2 u2)) -2.0 1.0)))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = sqrtf(-logf((1.0f - u1)));
float t_1 = cosf(((2.0f * ((float) M_PI)) * u2));
float tmp;
if ((t_0 * t_1) <= 0.17000000178813934f) {
tmp = sqrtf(fmaf((fmaf(0.3333333333333333f, u1, 0.5f) * u1), u1, u1)) * t_1;
} else {
tmp = t_0 * fmaf(((((float) M_PI) * ((float) M_PI)) * (u2 * u2)), -2.0f, 1.0f);
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) t_1 = cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2)) tmp = Float32(0.0) if (Float32(t_0 * t_1) <= Float32(0.17000000178813934)) tmp = Float32(sqrt(fma(Float32(fma(Float32(0.3333333333333333), u1, Float32(0.5)) * u1), u1, u1)) * t_1); else tmp = Float32(t_0 * fma(Float32(Float32(Float32(pi) * Float32(pi)) * Float32(u2 * u2)), Float32(-2.0), Float32(1.0))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{-\log \left(1 - u1\right)}\\
t_1 := \cos \left(\left(2 \cdot \pi\right) \cdot u2\right)\\
\mathbf{if}\;t\_0 \cdot t\_1 \leq 0.17000000178813934:\\
\;\;\;\;\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(0.3333333333333333, u1, 0.5\right) \cdot u1, u1, u1\right)} \cdot t\_1\\
\mathbf{else}:\\
\;\;\;\;t\_0 \cdot \mathsf{fma}\left(\left(\pi \cdot \pi\right) \cdot \left(u2 \cdot u2\right), -2, 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))) < 0.170000002Initial program 48.4%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3298.1
Applied rewrites98.1%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites98.2%
lift-fma.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
*-commutativeN/A
Applied rewrites98.2%
Taylor expanded in u1 around 0
Applied rewrites97.6%
if 0.170000002 < (*.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 96.8%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
pow-prod-downN/A
lower-pow.f32N/A
*-commutativeN/A
lower-*.f32N/A
lift-PI.f3290.7
Applied rewrites90.7%
lift-pow.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
unpow-prod-downN/A
lower-*.f32N/A
unpow2N/A
lower-*.f32N/A
lift-PI.f32N/A
lift-PI.f32N/A
unpow2N/A
lower-*.f3290.7
Applied rewrites90.7%
Final simplification96.7%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (sqrt (- (log (- 1.0 u1))))))
(if (<= (* t_0 (cos (* (* 2.0 PI) u2))) 0.17000000178813934)
(*
(sqrt (* (fma (fma 0.3333333333333333 u1 0.5) u1 1.0) u1))
(cos (* (+ PI PI) u2)))
(* t_0 (fma (* (* PI PI) (* u2 u2)) -2.0 1.0)))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = sqrtf(-logf((1.0f - u1)));
float tmp;
if ((t_0 * cosf(((2.0f * ((float) M_PI)) * u2))) <= 0.17000000178813934f) {
tmp = sqrtf((fmaf(fmaf(0.3333333333333333f, u1, 0.5f), u1, 1.0f) * u1)) * cosf(((((float) M_PI) + ((float) M_PI)) * u2));
} else {
tmp = t_0 * fmaf(((((float) M_PI) * ((float) M_PI)) * (u2 * u2)), -2.0f, 1.0f);
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) tmp = Float32(0.0) if (Float32(t_0 * cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2))) <= Float32(0.17000000178813934)) tmp = Float32(sqrt(Float32(fma(fma(Float32(0.3333333333333333), u1, Float32(0.5)), u1, Float32(1.0)) * u1)) * cos(Float32(Float32(Float32(pi) + Float32(pi)) * u2))); else tmp = Float32(t_0 * fma(Float32(Float32(Float32(pi) * Float32(pi)) * Float32(u2 * u2)), Float32(-2.0), Float32(1.0))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{-\log \left(1 - u1\right)}\\
\mathbf{if}\;t\_0 \cdot \cos \left(\left(2 \cdot \pi\right) \cdot u2\right) \leq 0.17000000178813934:\\
\;\;\;\;\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(0.3333333333333333, u1, 0.5\right), u1, 1\right) \cdot u1} \cdot \cos \left(\left(\pi + \pi\right) \cdot u2\right)\\
\mathbf{else}:\\
\;\;\;\;t\_0 \cdot \mathsf{fma}\left(\left(\pi \cdot \pi\right) \cdot \left(u2 \cdot u2\right), -2, 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))) < 0.170000002Initial program 48.4%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
lower-fma.f3295.5
Applied rewrites95.5%
lift-PI.f32N/A
lift-*.f32N/A
count-2-revN/A
lower-+.f32N/A
lift-PI.f32N/A
lift-PI.f3295.5
Applied rewrites95.5%
Taylor expanded in u1 around 0
neg-logN/A
flip--N/A
metadata-evalN/A
pow2N/A
pow2N/A
neg-logN/A
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3297.5
Applied rewrites97.5%
if 0.170000002 < (*.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 96.8%
Taylor expanded in u2 around 0
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
pow-prod-downN/A
lower-pow.f32N/A
*-commutativeN/A
lower-*.f32N/A
lift-PI.f3290.7
Applied rewrites90.7%
lift-pow.f32N/A
lift-PI.f32N/A
lift-*.f32N/A
unpow-prod-downN/A
lower-*.f32N/A
unpow2N/A
lower-*.f32N/A
lift-PI.f32N/A
lift-PI.f32N/A
unpow2N/A
lower-*.f3290.7
Applied rewrites90.7%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (cos (* (* 2.0 PI) u2))))
(if (<= (* (sqrt (- (log (- 1.0 u1)))) t_0) 0.029500000178813934)
(* (sqrt (fma (* 0.5 u1) u1 u1)) t_0)
(sqrt (- (log1p (- u1)))))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = cosf(((2.0f * ((float) M_PI)) * u2));
float tmp;
if ((sqrtf(-logf((1.0f - u1))) * t_0) <= 0.029500000178813934f) {
tmp = sqrtf(fmaf((0.5f * u1), u1, u1)) * t_0;
} else {
tmp = sqrtf(-log1pf(-u1));
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2)) tmp = Float32(0.0) if (Float32(sqrt(Float32(-log(Float32(Float32(1.0) - u1)))) * t_0) <= Float32(0.029500000178813934)) tmp = Float32(sqrt(fma(Float32(Float32(0.5) * u1), u1, u1)) * t_0); else tmp = sqrt(Float32(-log1p(Float32(-u1)))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos \left(\left(2 \cdot \pi\right) \cdot u2\right)\\
\mathbf{if}\;\sqrt{-\log \left(1 - u1\right)} \cdot t\_0 \leq 0.029500000178813934:\\
\;\;\;\;\sqrt{\mathsf{fma}\left(0.5 \cdot u1, u1, u1\right)} \cdot t\_0\\
\mathbf{else}:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\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))) < 0.0295000002Initial program 43.0%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3298.1
Applied rewrites98.1%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites98.2%
lift-fma.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
*-commutativeN/A
Applied rewrites98.2%
Taylor expanded in u1 around 0
Applied rewrites97.7%
if 0.0295000002 < (*.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 93.4%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3282.5
Applied rewrites82.5%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites82.6%
Taylor expanded in u2 around 0
sqrt-prodN/A
lower-sqrt.f32N/A
*-commutativeN/A
lower-*.f32N/A
*-rgt-identityN/A
*-commutativeN/A
metadata-evalN/A
fp-cancel-sign-sub-invN/A
mul-1-negN/A
lower-log1p.f32N/A
lower-neg.f3283.0
Applied rewrites83.0%
Final simplification94.3%
(FPCore (cosTheta_i u1 u2)
:precision binary32
(let* ((t_0 (log (- 1.0 u1))))
(if (<= t_0 -0.03799999877810478)
(* (sqrt (- t_0)) (cos (* (+ PI PI) u2)))
(*
(sqrt (fma (* (fma (fma 0.25 u1 0.3333333333333333) u1 0.5) u1) u1 u1))
(cos (* (* 2.0 PI) u2))))))
float code(float cosTheta_i, float u1, float u2) {
float t_0 = logf((1.0f - u1));
float tmp;
if (t_0 <= -0.03799999877810478f) {
tmp = sqrtf(-t_0) * cosf(((((float) M_PI) + ((float) M_PI)) * u2));
} else {
tmp = sqrtf(fmaf((fmaf(fmaf(0.25f, u1, 0.3333333333333333f), u1, 0.5f) * u1), u1, u1)) * cosf(((2.0f * ((float) M_PI)) * u2));
}
return tmp;
}
function code(cosTheta_i, u1, u2) t_0 = log(Float32(Float32(1.0) - u1)) tmp = Float32(0.0) if (t_0 <= Float32(-0.03799999877810478)) tmp = Float32(sqrt(Float32(-t_0)) * cos(Float32(Float32(Float32(pi) + Float32(pi)) * u2))); else tmp = Float32(sqrt(fma(Float32(fma(fma(Float32(0.25), u1, Float32(0.3333333333333333)), u1, Float32(0.5)) * u1), u1, u1)) * cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \log \left(1 - u1\right)\\
\mathbf{if}\;t\_0 \leq -0.03799999877810478:\\
\;\;\;\;\sqrt{-t\_0} \cdot \cos \left(\left(\pi + \pi\right) \cdot u2\right)\\
\mathbf{else}:\\
\;\;\;\;\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(\mathsf{fma}\left(0.25, u1, 0.3333333333333333\right), u1, 0.5\right) \cdot u1, u1, u1\right)} \cdot \cos \left(\left(2 \cdot \pi\right) \cdot u2\right)\\
\end{array}
\end{array}
if (log.f32 (-.f32 #s(literal 1 binary32) u1)) < -0.0379999988Initial program 97.2%
lift-PI.f32N/A
lift-*.f32N/A
count-2-revN/A
lower-+.f32N/A
lift-PI.f32N/A
lift-PI.f3297.2
Applied rewrites97.2%
if -0.0379999988 < (log.f32 (-.f32 #s(literal 1 binary32) u1)) Initial program 47.9%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3298.9
Applied rewrites98.9%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites98.9%
lift-fma.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
*-commutativeN/A
Applied rewrites98.9%
Final simplification98.7%
(FPCore (cosTheta_i u1 u2) :precision binary32 (if (<= (cos (* (* 2.0 PI) u2)) 0.999983012676239) (* (sqrt u1) (cos (* (+ PI PI) u2))) (sqrt (- (log1p (- u1))))))
float code(float cosTheta_i, float u1, float u2) {
float tmp;
if (cosf(((2.0f * ((float) M_PI)) * u2)) <= 0.999983012676239f) {
tmp = sqrtf(u1) * cosf(((((float) M_PI) + ((float) M_PI)) * u2));
} else {
tmp = sqrtf(-log1pf(-u1));
}
return tmp;
}
function code(cosTheta_i, u1, u2) tmp = Float32(0.0) if (cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2)) <= Float32(0.999983012676239)) tmp = Float32(sqrt(u1) * cos(Float32(Float32(Float32(pi) + Float32(pi)) * u2))); else tmp = sqrt(Float32(-log1p(Float32(-u1)))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
\mathbf{if}\;\cos \left(\left(2 \cdot \pi\right) \cdot u2\right) \leq 0.999983012676239:\\
\;\;\;\;\sqrt{u1} \cdot \cos \left(\left(\pi + \pi\right) \cdot u2\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)) < 0.999983013Initial program 53.9%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
lower-fma.f3287.1
Applied rewrites87.1%
lift-PI.f32N/A
lift-*.f32N/A
count-2-revN/A
lower-+.f32N/A
lift-PI.f32N/A
lift-PI.f3287.1
Applied rewrites87.1%
Taylor expanded in u1 around 0
neg-log78.0
flip--78.0
metadata-eval78.0
pow278.0
pow278.0
neg-log78.0
Applied rewrites78.0%
if 0.999983013 < (cos.f32 (*.f32 (*.f32 #s(literal 2 binary32) (PI.f32)) u2)) Initial program 54.9%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3295.4
Applied rewrites95.4%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites95.5%
Taylor expanded in u2 around 0
sqrt-prodN/A
lower-sqrt.f32N/A
*-commutativeN/A
lower-*.f32N/A
*-rgt-identityN/A
*-commutativeN/A
metadata-evalN/A
fp-cancel-sign-sub-invN/A
mul-1-negN/A
lower-log1p.f32N/A
lower-neg.f3296.9
Applied rewrites96.9%
Final simplification90.9%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (fma (* (fma (fma 0.25 u1 0.3333333333333333) u1 0.5) u1) u1 u1)) (cos (* (* 2.0 PI) u2))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(fmaf((fmaf(fmaf(0.25f, u1, 0.3333333333333333f), u1, 0.5f) * u1), u1, u1)) * cosf(((2.0f * ((float) M_PI)) * u2));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(fma(Float32(fma(fma(Float32(0.25), u1, Float32(0.3333333333333333)), u1, Float32(0.5)) * u1), u1, u1)) * cos(Float32(Float32(Float32(2.0) * Float32(pi)) * u2))) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(\mathsf{fma}\left(0.25, u1, 0.3333333333333333\right), u1, 0.5\right) \cdot u1, u1, u1\right)} \cdot \cos \left(\left(2 \cdot \pi\right) \cdot u2\right)
\end{array}
Initial program 54.6%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3294.5
Applied rewrites94.5%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites94.6%
lift-fma.f32N/A
lift-*.f32N/A
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
*-commutativeN/A
Applied rewrites94.6%
Final simplification94.6%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (* (fma (fma (fma 0.25 u1 0.3333333333333333) u1 0.5) u1 1.0) u1)) (cos (* (+ PI PI) u2))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf((fmaf(fmaf(fmaf(0.25f, u1, 0.3333333333333333f), u1, 0.5f), u1, 1.0f) * u1)) * cosf(((((float) M_PI) + ((float) M_PI)) * u2));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(fma(fma(fma(Float32(0.25), u1, Float32(0.3333333333333333)), u1, Float32(0.5)), u1, Float32(1.0)) * u1)) * cos(Float32(Float32(Float32(pi) + Float32(pi)) * u2))) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(\mathsf{fma}\left(0.25, u1, 0.3333333333333333\right), u1, 0.5\right), u1, 1\right) \cdot u1} \cdot \cos \left(\left(\pi + \pi\right) \cdot u2\right)
\end{array}
Initial program 54.6%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3294.5
Applied rewrites94.5%
lift-PI.f32N/A
lift-*.f32N/A
count-2-revN/A
lower-+.f32N/A
lift-PI.f32N/A
lift-PI.f3294.5
Applied rewrites94.5%
Final simplification94.5%
(FPCore (cosTheta_i u1 u2) :precision binary32 (if (<= u2 7.999999797903001e-5) (sqrt (- (log1p (- u1)))) (* (sqrt (* (fma 0.5 u1 1.0) u1)) (cos (* (+ PI PI) u2)))))
float code(float cosTheta_i, float u1, float u2) {
float tmp;
if (u2 <= 7.999999797903001e-5f) {
tmp = sqrtf(-log1pf(-u1));
} else {
tmp = sqrtf((fmaf(0.5f, u1, 1.0f) * u1)) * cosf(((((float) M_PI) + ((float) M_PI)) * u2));
}
return tmp;
}
function code(cosTheta_i, u1, u2) tmp = Float32(0.0) if (u2 <= Float32(7.999999797903001e-5)) tmp = sqrt(Float32(-log1p(Float32(-u1)))); else tmp = Float32(sqrt(Float32(fma(Float32(0.5), u1, Float32(1.0)) * u1)) * cos(Float32(Float32(Float32(pi) + Float32(pi)) * u2))); end return tmp end
\begin{array}{l}
\\
\begin{array}{l}
\mathbf{if}\;u2 \leq 7.999999797903001 \cdot 10^{-5}:\\
\;\;\;\;\sqrt{-\mathsf{log1p}\left(-u1\right)}\\
\mathbf{else}:\\
\;\;\;\;\sqrt{\mathsf{fma}\left(0.5, u1, 1\right) \cdot u1} \cdot \cos \left(\left(\pi + \pi\right) \cdot u2\right)\\
\end{array}
\end{array}
if u2 < 7.9999998e-5Initial program 53.2%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3295.6
Applied rewrites95.6%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites95.7%
Taylor expanded in u2 around 0
sqrt-prodN/A
lower-sqrt.f32N/A
*-commutativeN/A
lower-*.f32N/A
*-rgt-identityN/A
*-commutativeN/A
metadata-evalN/A
fp-cancel-sign-sub-invN/A
mul-1-negN/A
lower-log1p.f32N/A
lower-neg.f3299.8
Applied rewrites99.8%
if 7.9999998e-5 < u2 Initial program 56.3%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
lower-fma.f3287.9
Applied rewrites87.9%
lift-PI.f32N/A
lift-*.f32N/A
count-2-revN/A
lower-+.f32N/A
lift-PI.f32N/A
lift-PI.f3287.9
Applied rewrites87.9%
Final simplification94.2%
(FPCore (cosTheta_i u1 u2) :precision binary32 (* (sqrt (* (fma (fma 0.3333333333333333 u1 0.5) u1 1.0) u1)) (cos (* (+ PI PI) u2))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf((fmaf(fmaf(0.3333333333333333f, u1, 0.5f), u1, 1.0f) * u1)) * cosf(((((float) M_PI) + ((float) M_PI)) * u2));
}
function code(cosTheta_i, u1, u2) return Float32(sqrt(Float32(fma(fma(Float32(0.3333333333333333), u1, Float32(0.5)), u1, Float32(1.0)) * u1)) * cos(Float32(Float32(Float32(pi) + Float32(pi)) * u2))) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(0.3333333333333333, u1, 0.5\right), u1, 1\right) \cdot u1} \cdot \cos \left(\left(\pi + \pi\right) \cdot u2\right)
\end{array}
Initial program 54.6%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
lower-fma.f3289.7
Applied rewrites89.7%
lift-PI.f32N/A
lift-*.f32N/A
count-2-revN/A
lower-+.f32N/A
lift-PI.f32N/A
lift-PI.f3289.7
Applied rewrites89.7%
Taylor expanded in u1 around 0
neg-logN/A
flip--N/A
metadata-evalN/A
pow2N/A
pow2N/A
neg-logN/A
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3292.8
Applied rewrites92.8%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (- (log1p (- u1)))))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(-log1pf(-u1));
}
function code(cosTheta_i, u1, u2) return sqrt(Float32(-log1p(Float32(-u1)))) end
\begin{array}{l}
\\
\sqrt{-\mathsf{log1p}\left(-u1\right)}
\end{array}
Initial program 54.6%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3294.5
Applied rewrites94.5%
lift-*.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
distribute-lft-inN/A
lower-fma.f32N/A
lower-*.f32N/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
Applied rewrites94.6%
Taylor expanded in u2 around 0
sqrt-prodN/A
lower-sqrt.f32N/A
*-commutativeN/A
lower-*.f32N/A
*-rgt-identityN/A
*-commutativeN/A
metadata-evalN/A
fp-cancel-sign-sub-invN/A
mul-1-negN/A
lower-log1p.f32N/A
lower-neg.f3279.2
Applied rewrites79.2%
Final simplification79.2%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (fma (* (fma (fma 0.25 u1 0.3333333333333333) u1 0.5) u1) u1 u1)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(fmaf((fmaf(fmaf(0.25f, u1, 0.3333333333333333f), u1, 0.5f) * u1), u1, u1));
}
function code(cosTheta_i, u1, u2) return sqrt(fma(Float32(fma(fma(Float32(0.25), u1, Float32(0.3333333333333333)), u1, Float32(0.5)) * u1), u1, u1)) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(\mathsf{fma}\left(0.25, u1, 0.3333333333333333\right), u1, 0.5\right) \cdot u1, u1, u1\right)}
\end{array}
Initial program 54.6%
Taylor expanded in u2 around 0
sqrt-unprodN/A
lower-sqrt.f32N/A
lower-*.f32N/A
lift-log.f32N/A
lift--.f3246.8
Applied rewrites46.8%
Taylor expanded in u1 around 0
Applied rewrites66.9%
Taylor expanded in u1 around 0
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
distribute-lft-outN/A
*-rgt-identityN/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-*.f3276.6
Applied rewrites76.6%
Final simplification76.6%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (* (fma (fma (fma 0.25 u1 0.3333333333333333) u1 0.5) u1 1.0) u1)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf((fmaf(fmaf(fmaf(0.25f, u1, 0.3333333333333333f), u1, 0.5f), u1, 1.0f) * u1));
}
function code(cosTheta_i, u1, u2) return sqrt(Float32(fma(fma(fma(Float32(0.25), u1, Float32(0.3333333333333333)), u1, Float32(0.5)), u1, Float32(1.0)) * u1)) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(\mathsf{fma}\left(0.25, u1, 0.3333333333333333\right), u1, 0.5\right), u1, 1\right) \cdot u1}
\end{array}
Initial program 54.6%
Taylor expanded in u2 around 0
sqrt-unprodN/A
lower-sqrt.f32N/A
lower-*.f32N/A
lift-log.f32N/A
lift--.f3246.8
Applied rewrites46.8%
Taylor expanded in u1 around 0
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
*-commutativeN/A
+-commutativeN/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-fma.f32N/A
lift-*.f3276.5
Applied rewrites76.5%
Final simplification76.5%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (* (fma (fma 0.3333333333333333 u1 0.5) u1 1.0) u1)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf((fmaf(fmaf(0.3333333333333333f, u1, 0.5f), u1, 1.0f) * u1));
}
function code(cosTheta_i, u1, u2) return sqrt(Float32(fma(fma(Float32(0.3333333333333333), u1, Float32(0.5)), u1, Float32(1.0)) * u1)) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(\mathsf{fma}\left(0.3333333333333333, u1, 0.5\right), u1, 1\right) \cdot u1}
\end{array}
Initial program 54.6%
Taylor expanded in u2 around 0
sqrt-unprodN/A
lower-sqrt.f32N/A
lower-*.f32N/A
lift-log.f32N/A
lift--.f3246.8
Applied rewrites46.8%
Taylor expanded in u1 around 0
*-commutativeN/A
lower-*.f32N/A
+-commutativeN/A
*-commutativeN/A
lower-fma.f32N/A
+-commutativeN/A
lower-fma.f3275.5
Applied rewrites75.5%
Final simplification75.5%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt (* (fma 0.5 u1 1.0) u1)))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf((fmaf(0.5f, u1, 1.0f) * u1));
}
function code(cosTheta_i, u1, u2) return sqrt(Float32(fma(Float32(0.5), u1, Float32(1.0)) * u1)) end
\begin{array}{l}
\\
\sqrt{\mathsf{fma}\left(0.5, u1, 1\right) \cdot u1}
\end{array}
Initial program 54.6%
Taylor expanded in u2 around 0
sqrt-unprodN/A
lower-sqrt.f32N/A
lower-*.f32N/A
lift-log.f32N/A
lift--.f3246.8
Applied rewrites46.8%
Taylor expanded in u1 around 0
+-commutativeN/A
*-commutativeN/A
lift-fma.f32N/A
lift-*.f3273.9
Applied rewrites73.9%
Final simplification73.9%
(FPCore (cosTheta_i u1 u2) :precision binary32 (sqrt u1))
float code(float cosTheta_i, float u1, float u2) {
return sqrtf(u1);
}
module fmin_fmax_functions
implicit none
private
public fmax
public fmin
interface fmax
module procedure fmax88
module procedure fmax44
module procedure fmax84
module procedure fmax48
end interface
interface fmin
module procedure fmin88
module procedure fmin44
module procedure fmin84
module procedure fmin48
end interface
contains
real(8) function fmax88(x, y) result (res)
real(8), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(x, max(x, y), y /= y), x /= x)
end function
real(4) function fmax44(x, y) result (res)
real(4), intent (in) :: x
real(4), intent (in) :: y
res = merge(y, merge(x, max(x, y), y /= y), x /= x)
end function
real(8) function fmax84(x, y) result(res)
real(8), intent (in) :: x
real(4), intent (in) :: y
res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
end function
real(8) function fmax48(x, y) result(res)
real(4), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
end function
real(8) function fmin88(x, y) result (res)
real(8), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(x, min(x, y), y /= y), x /= x)
end function
real(4) function fmin44(x, y) result (res)
real(4), intent (in) :: x
real(4), intent (in) :: y
res = merge(y, merge(x, min(x, y), y /= y), x /= x)
end function
real(8) function fmin84(x, y) result(res)
real(8), intent (in) :: x
real(4), intent (in) :: y
res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
end function
real(8) function fmin48(x, y) result(res)
real(4), intent (in) :: x
real(8), intent (in) :: y
res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
end function
end module
real(4) function code(costheta_i, u1, u2)
use fmin_fmax_functions
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 54.6%
Taylor expanded in u2 around 0
sqrt-unprodN/A
lower-sqrt.f32N/A
lower-*.f32N/A
lift-log.f32N/A
lift--.f3246.8
Applied rewrites46.8%
Taylor expanded in u1 around 0
Applied rewrites66.9%
Final simplification66.9%
herbie shell --seed 2025064
(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))))