
(FPCore (a1 a2 th) :precision binary64 (let* ((t_1 (/ (cos th) (sqrt 2.0)))) (+ (* t_1 (* a1 a1)) (* t_1 (* a2 a2)))))
double code(double a1, double a2, double th) {
double t_1 = cos(th) / sqrt(2.0);
return (t_1 * (a1 * a1)) + (t_1 * (a2 * a2));
}
real(8) function code(a1, a2, th)
real(8), intent (in) :: a1
real(8), intent (in) :: a2
real(8), intent (in) :: th
real(8) :: t_1
t_1 = cos(th) / sqrt(2.0d0)
code = (t_1 * (a1 * a1)) + (t_1 * (a2 * a2))
end function
public static double code(double a1, double a2, double th) {
double t_1 = Math.cos(th) / Math.sqrt(2.0);
return (t_1 * (a1 * a1)) + (t_1 * (a2 * a2));
}
def code(a1, a2, th): t_1 = math.cos(th) / math.sqrt(2.0) return (t_1 * (a1 * a1)) + (t_1 * (a2 * a2))
function code(a1, a2, th) t_1 = Float64(cos(th) / sqrt(2.0)) return Float64(Float64(t_1 * Float64(a1 * a1)) + Float64(t_1 * Float64(a2 * a2))) end
function tmp = code(a1, a2, th) t_1 = cos(th) / sqrt(2.0); tmp = (t_1 * (a1 * a1)) + (t_1 * (a2 * a2)); end
code[a1_, a2_, th_] := Block[{t$95$1 = N[(N[Cos[th], $MachinePrecision] / N[Sqrt[2.0], $MachinePrecision]), $MachinePrecision]}, N[(N[(t$95$1 * N[(a1 * a1), $MachinePrecision]), $MachinePrecision] + N[(t$95$1 * N[(a2 * a2), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]
\begin{array}{l}
\\
\begin{array}{l}
t_1 := \frac{\cos th}{\sqrt{2}}\\
t\_1 \cdot \left(a1 \cdot a1\right) + t\_1 \cdot \left(a2 \cdot a2\right)
\end{array}
\end{array}
Sampling outcomes in binary64 precision:
Herbie found 7 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (a1 a2 th) :precision binary64 (let* ((t_1 (/ (cos th) (sqrt 2.0)))) (+ (* t_1 (* a1 a1)) (* t_1 (* a2 a2)))))
double code(double a1, double a2, double th) {
double t_1 = cos(th) / sqrt(2.0);
return (t_1 * (a1 * a1)) + (t_1 * (a2 * a2));
}
real(8) function code(a1, a2, th)
real(8), intent (in) :: a1
real(8), intent (in) :: a2
real(8), intent (in) :: th
real(8) :: t_1
t_1 = cos(th) / sqrt(2.0d0)
code = (t_1 * (a1 * a1)) + (t_1 * (a2 * a2))
end function
public static double code(double a1, double a2, double th) {
double t_1 = Math.cos(th) / Math.sqrt(2.0);
return (t_1 * (a1 * a1)) + (t_1 * (a2 * a2));
}
def code(a1, a2, th): t_1 = math.cos(th) / math.sqrt(2.0) return (t_1 * (a1 * a1)) + (t_1 * (a2 * a2))
function code(a1, a2, th) t_1 = Float64(cos(th) / sqrt(2.0)) return Float64(Float64(t_1 * Float64(a1 * a1)) + Float64(t_1 * Float64(a2 * a2))) end
function tmp = code(a1, a2, th) t_1 = cos(th) / sqrt(2.0); tmp = (t_1 * (a1 * a1)) + (t_1 * (a2 * a2)); end
code[a1_, a2_, th_] := Block[{t$95$1 = N[(N[Cos[th], $MachinePrecision] / N[Sqrt[2.0], $MachinePrecision]), $MachinePrecision]}, N[(N[(t$95$1 * N[(a1 * a1), $MachinePrecision]), $MachinePrecision] + N[(t$95$1 * N[(a2 * a2), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]
\begin{array}{l}
\\
\begin{array}{l}
t_1 := \frac{\cos th}{\sqrt{2}}\\
t\_1 \cdot \left(a1 \cdot a1\right) + t\_1 \cdot \left(a2 \cdot a2\right)
\end{array}
\end{array}
(FPCore (a1 a2 th) :precision binary64 (* (* (sqrt 2.0) (* (cos th) (fma a2 a2 (* a1 a1)))) 0.5))
double code(double a1, double a2, double th) {
return (sqrt(2.0) * (cos(th) * fma(a2, a2, (a1 * a1)))) * 0.5;
}
function code(a1, a2, th) return Float64(Float64(sqrt(2.0) * Float64(cos(th) * fma(a2, a2, Float64(a1 * a1)))) * 0.5) end
code[a1_, a2_, th_] := N[(N[(N[Sqrt[2.0], $MachinePrecision] * N[(N[Cos[th], $MachinePrecision] * N[(a2 * a2 + N[(a1 * a1), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision] * 0.5), $MachinePrecision]
\begin{array}{l}
\\
\left(\sqrt{2} \cdot \left(\cos th \cdot \mathsf{fma}\left(a2, a2, a1 \cdot a1\right)\right)\right) \cdot 0.5
\end{array}
Initial program 99.6%
lift-cos.f64N/A
lift-sqrt.f64N/A
lift-*.f64N/A
associate-*l/N/A
lift-cos.f64N/A
lift-sqrt.f64N/A
lift-*.f64N/A
associate-*l/N/A
frac-addN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
rem-square-sqrtN/A
div-invN/A
metadata-evalN/A
lower-*.f64N/A
Applied egg-rr99.7%
lift-cos.f64N/A
lift-*.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-sqrt.f64N/A
lift-cos.f64N/A
lift-*.f64N/A
lift-*.f64N/A
*-commutativeN/A
distribute-rgt-outN/A
lift-*.f64N/A
distribute-lft-inN/A
+-commutativeN/A
lift-*.f64N/A
lift-fma.f64N/A
lower-*.f64N/A
lower-*.f6499.7
Applied egg-rr99.7%
(FPCore (a1 a2 th)
:precision binary64
(let* ((t_1 (/ (cos th) (sqrt 2.0))))
(if (<= (+ (* (* a1 a1) t_1) (* (* a2 a2) t_1)) -5e-134)
(* (* a2 a2) (* (sqrt 2.0) (fma -0.25 (* th th) 0.5)))
(* 0.5 (* (sqrt 2.0) (fma a1 a1 (* a2 a2)))))))
double code(double a1, double a2, double th) {
double t_1 = cos(th) / sqrt(2.0);
double tmp;
if ((((a1 * a1) * t_1) + ((a2 * a2) * t_1)) <= -5e-134) {
tmp = (a2 * a2) * (sqrt(2.0) * fma(-0.25, (th * th), 0.5));
} else {
tmp = 0.5 * (sqrt(2.0) * fma(a1, a1, (a2 * a2)));
}
return tmp;
}
function code(a1, a2, th) t_1 = Float64(cos(th) / sqrt(2.0)) tmp = 0.0 if (Float64(Float64(Float64(a1 * a1) * t_1) + Float64(Float64(a2 * a2) * t_1)) <= -5e-134) tmp = Float64(Float64(a2 * a2) * Float64(sqrt(2.0) * fma(-0.25, Float64(th * th), 0.5))); else tmp = Float64(0.5 * Float64(sqrt(2.0) * fma(a1, a1, Float64(a2 * a2)))); end return tmp end
code[a1_, a2_, th_] := Block[{t$95$1 = N[(N[Cos[th], $MachinePrecision] / N[Sqrt[2.0], $MachinePrecision]), $MachinePrecision]}, If[LessEqual[N[(N[(N[(a1 * a1), $MachinePrecision] * t$95$1), $MachinePrecision] + N[(N[(a2 * a2), $MachinePrecision] * t$95$1), $MachinePrecision]), $MachinePrecision], -5e-134], N[(N[(a2 * a2), $MachinePrecision] * N[(N[Sqrt[2.0], $MachinePrecision] * N[(-0.25 * N[(th * th), $MachinePrecision] + 0.5), $MachinePrecision]), $MachinePrecision]), $MachinePrecision], N[(0.5 * N[(N[Sqrt[2.0], $MachinePrecision] * N[(a1 * a1 + N[(a2 * a2), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]
\begin{array}{l}
\\
\begin{array}{l}
t_1 := \frac{\cos th}{\sqrt{2}}\\
\mathbf{if}\;\left(a1 \cdot a1\right) \cdot t\_1 + \left(a2 \cdot a2\right) \cdot t\_1 \leq -5 \cdot 10^{-134}:\\
\;\;\;\;\left(a2 \cdot a2\right) \cdot \left(\sqrt{2} \cdot \mathsf{fma}\left(-0.25, th \cdot th, 0.5\right)\right)\\
\mathbf{else}:\\
\;\;\;\;0.5 \cdot \left(\sqrt{2} \cdot \mathsf{fma}\left(a1, a1, a2 \cdot a2\right)\right)\\
\end{array}
\end{array}
if (+.f64 (*.f64 (/.f64 (cos.f64 th) (sqrt.f64 #s(literal 2 binary64))) (*.f64 a1 a1)) (*.f64 (/.f64 (cos.f64 th) (sqrt.f64 #s(literal 2 binary64))) (*.f64 a2 a2))) < -5.0000000000000003e-134Initial program 99.6%
lift-cos.f64N/A
lift-sqrt.f64N/A
lift-*.f64N/A
associate-*l/N/A
lift-cos.f64N/A
lift-sqrt.f64N/A
lift-*.f64N/A
associate-*l/N/A
frac-addN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
rem-square-sqrtN/A
div-invN/A
metadata-evalN/A
lower-*.f64N/A
Applied egg-rr99.5%
lift-cos.f64N/A
lift-*.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-sqrt.f64N/A
lift-cos.f64N/A
lift-*.f64N/A
lift-*.f64N/A
*-commutativeN/A
distribute-rgt-outN/A
lift-*.f64N/A
distribute-lft-inN/A
+-commutativeN/A
lift-*.f64N/A
lift-fma.f64N/A
lower-*.f64N/A
lower-*.f6499.6
Applied egg-rr99.6%
Taylor expanded in a2 around inf
associate-*r*N/A
*-commutativeN/A
lower-*.f64N/A
*-commutativeN/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-cos.f64N/A
lower-*.f64N/A
unpow2N/A
lower-*.f6455.5
Simplified55.5%
Taylor expanded in th around 0
*-commutativeN/A
associate-*l*N/A
*-commutativeN/A
associate-*l*N/A
*-commutativeN/A
distribute-lft-outN/A
lower-*.f64N/A
unpow2N/A
lower-*.f64N/A
*-commutativeN/A
associate-*r*N/A
*-commutativeN/A
distribute-rgt-outN/A
lower-*.f64N/A
lower-sqrt.f64N/A
*-commutativeN/A
lower-fma.f64N/A
unpow2N/A
lower-*.f6440.4
Simplified40.4%
if -5.0000000000000003e-134 < (+.f64 (*.f64 (/.f64 (cos.f64 th) (sqrt.f64 #s(literal 2 binary64))) (*.f64 a1 a1)) (*.f64 (/.f64 (cos.f64 th) (sqrt.f64 #s(literal 2 binary64))) (*.f64 a2 a2))) Initial program 99.5%
lift-cos.f64N/A
lift-sqrt.f64N/A
lift-*.f64N/A
associate-*l/N/A
lift-cos.f64N/A
lift-sqrt.f64N/A
lift-*.f64N/A
associate-*l/N/A
frac-addN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
rem-square-sqrtN/A
div-invN/A
metadata-evalN/A
lower-*.f64N/A
Applied egg-rr99.6%
Taylor expanded in th around 0
distribute-rgt-outN/A
lower-*.f64N/A
lower-sqrt.f64N/A
unpow2N/A
lower-fma.f64N/A
unpow2N/A
lower-*.f6484.1
Simplified84.1%
Final simplification75.0%
herbie shell --seed 2024218
(FPCore (a1 a2 th)
:name "Migdal et al, Equation (64)"
:precision binary64
(+ (* (/ (cos th) (sqrt 2.0)) (* a1 a1)) (* (/ (cos th) (sqrt 2.0)) (* a2 a2))))