
(FPCore (kx ky th) :precision binary64 (* (/ (sin ky) (sqrt (+ (pow (sin kx) 2.0) (pow (sin ky) 2.0)))) (sin th)))
double code(double kx, double ky, double th) {
return (sin(ky) / sqrt((pow(sin(kx), 2.0) + pow(sin(ky), 2.0)))) * sin(th);
}
real(8) function code(kx, ky, th)
real(8), intent (in) :: kx
real(8), intent (in) :: ky
real(8), intent (in) :: th
code = (sin(ky) / sqrt(((sin(kx) ** 2.0d0) + (sin(ky) ** 2.0d0)))) * sin(th)
end function
public static double code(double kx, double ky, double th) {
return (Math.sin(ky) / Math.sqrt((Math.pow(Math.sin(kx), 2.0) + Math.pow(Math.sin(ky), 2.0)))) * Math.sin(th);
}
def code(kx, ky, th): return (math.sin(ky) / math.sqrt((math.pow(math.sin(kx), 2.0) + math.pow(math.sin(ky), 2.0)))) * math.sin(th)
function code(kx, ky, th) return Float64(Float64(sin(ky) / sqrt(Float64((sin(kx) ^ 2.0) + (sin(ky) ^ 2.0)))) * sin(th)) end
function tmp = code(kx, ky, th) tmp = (sin(ky) / sqrt(((sin(kx) ^ 2.0) + (sin(ky) ^ 2.0)))) * sin(th); end
code[kx_, ky_, th_] := N[(N[(N[Sin[ky], $MachinePrecision] / N[Sqrt[N[(N[Power[N[Sin[kx], $MachinePrecision], 2.0], $MachinePrecision] + N[Power[N[Sin[ky], $MachinePrecision], 2.0], $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[Sin[th], $MachinePrecision]), $MachinePrecision]
\begin{array}{l}
\\
\frac{\sin ky}{\sqrt{{\sin kx}^{2} + {\sin ky}^{2}}} \cdot \sin th
\end{array}
Sampling outcomes in binary64 precision:
Herbie found 28 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (kx ky th) :precision binary64 (* (/ (sin ky) (sqrt (+ (pow (sin kx) 2.0) (pow (sin ky) 2.0)))) (sin th)))
double code(double kx, double ky, double th) {
return (sin(ky) / sqrt((pow(sin(kx), 2.0) + pow(sin(ky), 2.0)))) * sin(th);
}
real(8) function code(kx, ky, th)
real(8), intent (in) :: kx
real(8), intent (in) :: ky
real(8), intent (in) :: th
code = (sin(ky) / sqrt(((sin(kx) ** 2.0d0) + (sin(ky) ** 2.0d0)))) * sin(th)
end function
public static double code(double kx, double ky, double th) {
return (Math.sin(ky) / Math.sqrt((Math.pow(Math.sin(kx), 2.0) + Math.pow(Math.sin(ky), 2.0)))) * Math.sin(th);
}
def code(kx, ky, th): return (math.sin(ky) / math.sqrt((math.pow(math.sin(kx), 2.0) + math.pow(math.sin(ky), 2.0)))) * math.sin(th)
function code(kx, ky, th) return Float64(Float64(sin(ky) / sqrt(Float64((sin(kx) ^ 2.0) + (sin(ky) ^ 2.0)))) * sin(th)) end
function tmp = code(kx, ky, th) tmp = (sin(ky) / sqrt(((sin(kx) ^ 2.0) + (sin(ky) ^ 2.0)))) * sin(th); end
code[kx_, ky_, th_] := N[(N[(N[Sin[ky], $MachinePrecision] / N[Sqrt[N[(N[Power[N[Sin[kx], $MachinePrecision], 2.0], $MachinePrecision] + N[Power[N[Sin[ky], $MachinePrecision], 2.0], $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * N[Sin[th], $MachinePrecision]), $MachinePrecision]
\begin{array}{l}
\\
\frac{\sin ky}{\sqrt{{\sin kx}^{2} + {\sin ky}^{2}}} \cdot \sin th
\end{array}
(FPCore (kx ky th) :precision binary64 (* (/ (sin ky) (hypot (sin ky) (sin kx))) (sin th)))
double code(double kx, double ky, double th) {
return (sin(ky) / hypot(sin(ky), sin(kx))) * sin(th);
}
public static double code(double kx, double ky, double th) {
return (Math.sin(ky) / Math.hypot(Math.sin(ky), Math.sin(kx))) * Math.sin(th);
}
def code(kx, ky, th): return (math.sin(ky) / math.hypot(math.sin(ky), math.sin(kx))) * math.sin(th)
function code(kx, ky, th) return Float64(Float64(sin(ky) / hypot(sin(ky), sin(kx))) * sin(th)) end
function tmp = code(kx, ky, th) tmp = (sin(ky) / hypot(sin(ky), sin(kx))) * sin(th); end
code[kx_, ky_, th_] := N[(N[(N[Sin[ky], $MachinePrecision] / N[Sqrt[N[Sin[ky], $MachinePrecision] ^ 2 + N[Sin[kx], $MachinePrecision] ^ 2], $MachinePrecision]), $MachinePrecision] * N[Sin[th], $MachinePrecision]), $MachinePrecision]
\begin{array}{l}
\\
\frac{\sin ky}{\mathsf{hypot}\left(\sin ky, \sin kx\right)} \cdot \sin th
\end{array}
Initial program 94.2%
lift-sqrt.f64N/A
lift-+.f64N/A
+-commutativeN/A
lift-pow.f64N/A
unpow2N/A
lift-pow.f64N/A
unpow2N/A
lower-hypot.f6499.7
Applied rewrites99.7%
(FPCore (kx ky th)
:precision binary64
(let* ((t_1 (- 1.0 (cos (* kx -2.0))))
(t_2 (/ (sin ky) (sqrt (+ (pow (sin kx) 2.0) (pow (sin ky) 2.0)))))
(t_3 (fma -0.5 (cos (* ky -2.0)) 0.5)))
(if (<= t_2 -1.0)
(* (sin th) (/ (sin ky) (sqrt t_3)))
(if (<= t_2 -0.2)
(*
th
(*
(sqrt (/ 1.0 (fma 0.5 t_1 t_3)))
(* (sin ky) (fma -0.16666666666666666 (* th th) 1.0))))
(if (<= t_2 4e-144)
(* (sin th) (/ (sin ky) (* (sqrt t_1) (sqrt 0.5))))
(if (<= t_2 0.1)
(* (sin th) (/ (sin ky) (sin kx)))
(if (<= t_2 0.996)
(/
1.0
(/
(sqrt
(fma
(- 1.0 (cos (+ kx kx)))
0.5
(+ 0.5 (* -0.5 (cos (+ ky ky))))))
(*
(sin ky)
(fma
th
(*
(* th th)
(fma 0.008333333333333333 (* th th) -0.16666666666666666))
th))))
(*
(sin th)
(/ (sin ky) (fma kx (* kx (/ 0.5 ky)) (sin ky)))))))))))
double code(double kx, double ky, double th) {
double t_1 = 1.0 - cos((kx * -2.0));
double t_2 = sin(ky) / sqrt((pow(sin(kx), 2.0) + pow(sin(ky), 2.0)));
double t_3 = fma(-0.5, cos((ky * -2.0)), 0.5);
double tmp;
if (t_2 <= -1.0) {
tmp = sin(th) * (sin(ky) / sqrt(t_3));
} else if (t_2 <= -0.2) {
tmp = th * (sqrt((1.0 / fma(0.5, t_1, t_3))) * (sin(ky) * fma(-0.16666666666666666, (th * th), 1.0)));
} else if (t_2 <= 4e-144) {
tmp = sin(th) * (sin(ky) / (sqrt(t_1) * sqrt(0.5)));
} else if (t_2 <= 0.1) {
tmp = sin(th) * (sin(ky) / sin(kx));
} else if (t_2 <= 0.996) {
tmp = 1.0 / (sqrt(fma((1.0 - cos((kx + kx))), 0.5, (0.5 + (-0.5 * cos((ky + ky)))))) / (sin(ky) * fma(th, ((th * th) * fma(0.008333333333333333, (th * th), -0.16666666666666666)), th)));
} else {
tmp = sin(th) * (sin(ky) / fma(kx, (kx * (0.5 / ky)), sin(ky)));
}
return tmp;
}
function code(kx, ky, th) t_1 = Float64(1.0 - cos(Float64(kx * -2.0))) t_2 = Float64(sin(ky) / sqrt(Float64((sin(kx) ^ 2.0) + (sin(ky) ^ 2.0)))) t_3 = fma(-0.5, cos(Float64(ky * -2.0)), 0.5) tmp = 0.0 if (t_2 <= -1.0) tmp = Float64(sin(th) * Float64(sin(ky) / sqrt(t_3))); elseif (t_2 <= -0.2) tmp = Float64(th * Float64(sqrt(Float64(1.0 / fma(0.5, t_1, t_3))) * Float64(sin(ky) * fma(-0.16666666666666666, Float64(th * th), 1.0)))); elseif (t_2 <= 4e-144) tmp = Float64(sin(th) * Float64(sin(ky) / Float64(sqrt(t_1) * sqrt(0.5)))); elseif (t_2 <= 0.1) tmp = Float64(sin(th) * Float64(sin(ky) / sin(kx))); elseif (t_2 <= 0.996) tmp = Float64(1.0 / Float64(sqrt(fma(Float64(1.0 - cos(Float64(kx + kx))), 0.5, Float64(0.5 + Float64(-0.5 * cos(Float64(ky + ky)))))) / Float64(sin(ky) * fma(th, Float64(Float64(th * th) * fma(0.008333333333333333, Float64(th * th), -0.16666666666666666)), th)))); else tmp = Float64(sin(th) * Float64(sin(ky) / fma(kx, Float64(kx * Float64(0.5 / ky)), sin(ky)))); end return tmp end
code[kx_, ky_, th_] := Block[{t$95$1 = N[(1.0 - N[Cos[N[(kx * -2.0), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]}, Block[{t$95$2 = N[(N[Sin[ky], $MachinePrecision] / N[Sqrt[N[(N[Power[N[Sin[kx], $MachinePrecision], 2.0], $MachinePrecision] + N[Power[N[Sin[ky], $MachinePrecision], 2.0], $MachinePrecision]), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]}, Block[{t$95$3 = N[(-0.5 * N[Cos[N[(ky * -2.0), $MachinePrecision]], $MachinePrecision] + 0.5), $MachinePrecision]}, If[LessEqual[t$95$2, -1.0], N[(N[Sin[th], $MachinePrecision] * N[(N[Sin[ky], $MachinePrecision] / N[Sqrt[t$95$3], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$2, -0.2], N[(th * N[(N[Sqrt[N[(1.0 / N[(0.5 * t$95$1 + t$95$3), $MachinePrecision]), $MachinePrecision]], $MachinePrecision] * N[(N[Sin[ky], $MachinePrecision] * N[(-0.16666666666666666 * N[(th * th), $MachinePrecision] + 1.0), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$2, 4e-144], N[(N[Sin[th], $MachinePrecision] * N[(N[Sin[ky], $MachinePrecision] / N[(N[Sqrt[t$95$1], $MachinePrecision] * N[Sqrt[0.5], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$2, 0.1], N[(N[Sin[th], $MachinePrecision] * N[(N[Sin[ky], $MachinePrecision] / N[Sin[kx], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$2, 0.996], N[(1.0 / N[(N[Sqrt[N[(N[(1.0 - N[Cos[N[(kx + kx), $MachinePrecision]], $MachinePrecision]), $MachinePrecision] * 0.5 + N[(0.5 + N[(-0.5 * N[Cos[N[(ky + ky), $MachinePrecision]], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]], $MachinePrecision] / N[(N[Sin[ky], $MachinePrecision] * N[(th * N[(N[(th * th), $MachinePrecision] * N[(0.008333333333333333 * N[(th * th), $MachinePrecision] + -0.16666666666666666), $MachinePrecision]), $MachinePrecision] + th), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], N[(N[Sin[th], $MachinePrecision] * N[(N[Sin[ky], $MachinePrecision] / N[(kx * N[(kx * N[(0.5 / ky), $MachinePrecision]), $MachinePrecision] + N[Sin[ky], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]]]]]]]
\begin{array}{l}
\\
\begin{array}{l}
t_1 := 1 - \cos \left(kx \cdot -2\right)\\
t_2 := \frac{\sin ky}{\sqrt{{\sin kx}^{2} + {\sin ky}^{2}}}\\
t_3 := \mathsf{fma}\left(-0.5, \cos \left(ky \cdot -2\right), 0.5\right)\\
\mathbf{if}\;t\_2 \leq -1:\\
\;\;\;\;\sin th \cdot \frac{\sin ky}{\sqrt{t\_3}}\\
\mathbf{elif}\;t\_2 \leq -0.2:\\
\;\;\;\;th \cdot \left(\sqrt{\frac{1}{\mathsf{fma}\left(0.5, t\_1, t\_3\right)}} \cdot \left(\sin ky \cdot \mathsf{fma}\left(-0.16666666666666666, th \cdot th, 1\right)\right)\right)\\
\mathbf{elif}\;t\_2 \leq 4 \cdot 10^{-144}:\\
\;\;\;\;\sin th \cdot \frac{\sin ky}{\sqrt{t\_1} \cdot \sqrt{0.5}}\\
\mathbf{elif}\;t\_2 \leq 0.1:\\
\;\;\;\;\sin th \cdot \frac{\sin ky}{\sin kx}\\
\mathbf{elif}\;t\_2 \leq 0.996:\\
\;\;\;\;\frac{1}{\frac{\sqrt{\mathsf{fma}\left(1 - \cos \left(kx + kx\right), 0.5, 0.5 + -0.5 \cdot \cos \left(ky + ky\right)\right)}}{\sin ky \cdot \mathsf{fma}\left(th, \left(th \cdot th\right) \cdot \mathsf{fma}\left(0.008333333333333333, th \cdot th, -0.16666666666666666\right), th\right)}}\\
\mathbf{else}:\\
\;\;\;\;\sin th \cdot \frac{\sin ky}{\mathsf{fma}\left(kx, kx \cdot \frac{0.5}{ky}, \sin ky\right)}\\
\end{array}
\end{array}
if (/.f64 (sin.f64 ky) (sqrt.f64 (+.f64 (pow.f64 (sin.f64 kx) #s(literal 2 binary64)) (pow.f64 (sin.f64 ky) #s(literal 2 binary64))))) < -1Initial program 85.2%
lift-*.f64N/A
lift-/.f64N/A
associate-*l/N/A
clear-numN/A
lower-/.f64N/A
lower-/.f64N/A
Applied rewrites61.2%
Taylor expanded in kx around 0
lower-*.f64N/A
*-commutativeN/A
lower-*.f64N/A
lower-sin.f64N/A
lower-sin.f64N/A
lower-sqrt.f64N/A
lower-/.f64N/A
+-commutativeN/A
metadata-evalN/A
distribute-lft-neg-inN/A
lower-fma.f64N/A
cos-negN/A
lower-cos.f64N/A
*-commutativeN/A
lower-*.f6461.2
Applied rewrites61.2%
Applied rewrites61.5%
if -1 < (/.f64 (sin.f64 ky) (sqrt.f64 (+.f64 (pow.f64 (sin.f64 kx) #s(literal 2 binary64)) (pow.f64 (sin.f64 ky) #s(literal 2 binary64))))) < -0.20000000000000001Initial program 99.2%
lift-*.f64N/A
lift-/.f64N/A
associate-*l/N/A
clear-numN/A
lower-/.f64N/A
lower-/.f64N/A
Applied rewrites96.5%
Taylor expanded in th around 0
Applied rewrites51.9%
if -0.20000000000000001 < (/.f64 (sin.f64 ky) (sqrt.f64 (+.f64 (pow.f64 (sin.f64 kx) #s(literal 2 binary64)) (pow.f64 (sin.f64 ky) #s(literal 2 binary64))))) < 3.9999999999999998e-144Initial program 99.4%
lift-pow.f64N/A
unpow2N/A
lift-sin.f64N/A
lift-sin.f64N/A
sin-multN/A
clear-numN/A
lower-/.f64N/A
lower-/.f64N/A
count-2N/A
cos-diffN/A
cos-sin-sumN/A
lower--.f64N/A
count-2N/A
lower-cos.f64N/A
lower-+.f6476.7
Applied rewrites76.7%
Taylor expanded in ky around 0
*-commutativeN/A
lower-*.f64N/A
lower-sqrt.f64N/A
metadata-evalN/A
distribute-lft-neg-inN/A
lower--.f64N/A
cos-negN/A
lower-cos.f64N/A
*-commutativeN/A
lower-*.f64N/A
lower-sqrt.f6473.0
Applied rewrites73.0%
if 3.9999999999999998e-144 < (/.f64 (sin.f64 ky) (sqrt.f64 (+.f64 (pow.f64 (sin.f64 kx) #s(literal 2 binary64)) (pow.f64 (sin.f64 ky) #s(literal 2 binary64))))) < 0.10000000000000001Initial program 98.9%
Taylor expanded in ky around 0
lower-sin.f6452.6
Applied rewrites52.6%
if 0.10000000000000001 < (/.f64 (sin.f64 ky) (sqrt.f64 (+.f64 (pow.f64 (sin.f64 kx) #s(literal 2 binary64)) (pow.f64 (sin.f64 ky) #s(literal 2 binary64))))) < 0.996Initial program 99.2%
lift-*.f64N/A
lift-/.f64N/A
associate-*l/N/A
clear-numN/A
lower-/.f64N/A
lower-/.f64N/A
Applied rewrites98.5%
Taylor expanded in th around 0
+-commutativeN/A
distribute-lft-inN/A
*-rgt-identityN/A
lower-fma.f64N/A
lower-*.f64N/A
unpow2N/A
lower-*.f64N/A
sub-negN/A
metadata-evalN/A
lower-fma.f64N/A
unpow2N/A
lower-*.f6451.2
Applied rewrites51.2%
if 0.996 < (/.f64 (sin.f64 ky) (sqrt.f64 (+.f64 (pow.f64 (sin.f64 kx) #s(literal 2 binary64)) (pow.f64 (sin.f64 ky) #s(literal 2 binary64))))) Initial program 87.7%
Taylor expanded in kx around 0
+-commutativeN/A
*-commutativeN/A
associate-*l/N/A
associate-*r/N/A
metadata-evalN/A
associate-*r/N/A
unpow2N/A
associate-*l*N/A
lower-fma.f64N/A
lower-*.f64N/A
associate-*r/N/A
metadata-evalN/A
lower-/.f64N/A
lower-sin.f64N/A
lower-sin.f6491.1
Applied rewrites91.1%
Taylor expanded in ky around 0
Applied rewrites90.9%
Final simplification67.5%
herbie shell --seed 2024226
(FPCore (kx ky th)
:name "Toniolo and Linder, Equation (3b), real"
:precision binary64
(* (/ (sin ky) (sqrt (+ (pow (sin kx) 2.0) (pow (sin ky) 2.0)))) (sin th)))