
(FPCore (x) :precision binary64 (acos (- 1.0 x)))
double code(double x) {
return acos((1.0 - x));
}
real(8) function code(x)
real(8), intent (in) :: x
code = acos((1.0d0 - x))
end function
public static double code(double x) {
return Math.acos((1.0 - x));
}
def code(x): return math.acos((1.0 - x))
function code(x) return acos(Float64(1.0 - x)) end
function tmp = code(x) tmp = acos((1.0 - x)); end
code[x_] := N[ArcCos[N[(1.0 - x), $MachinePrecision]], $MachinePrecision]
\begin{array}{l}
\\
\cos^{-1} \left(1 - x\right)
\end{array}
Sampling outcomes in binary64 precision:
Herbie found 9 alternatives:
| Alternative | Accuracy | Speedup |
|---|
(FPCore (x) :precision binary64 (acos (- 1.0 x)))
double code(double x) {
return acos((1.0 - x));
}
real(8) function code(x)
real(8), intent (in) :: x
code = acos((1.0d0 - x))
end function
public static double code(double x) {
return Math.acos((1.0 - x));
}
def code(x): return math.acos((1.0 - x))
function code(x) return acos(Float64(1.0 - x)) end
function tmp = code(x) tmp = acos((1.0 - x)); end
code[x_] := N[ArcCos[N[(1.0 - x), $MachinePrecision]], $MachinePrecision]
\begin{array}{l}
\\
\cos^{-1} \left(1 - x\right)
\end{array}
(FPCore (x)
:precision binary64
(let* ((t_0 (acos (- 1.0 x))) (t_1 (sqrt (PI))))
(fma
(* (sqrt (* 2.0 t_1)) (sqrt (* 0.5 t_1)))
(* t_1 0.5)
(/ (+ (* (- 0.25) (* (PI) (PI))) (pow t_0 2.0)) (fma 0.5 (PI) t_0)))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos^{-1} \left(1 - x\right)\\
t_1 := \sqrt{\mathsf{PI}\left(\right)}\\
\mathsf{fma}\left(\sqrt{2 \cdot t\_1} \cdot \sqrt{0.5 \cdot t\_1}, t\_1 \cdot 0.5, \frac{\left(-0.25\right) \cdot \left(\mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right)\right) + {t\_0}^{2}}{\mathsf{fma}\left(0.5, \mathsf{PI}\left(\right), t\_0\right)}\right)
\end{array}
\end{array}
Initial program 5.4%
Applied rewrites3.6%
lift-sqrt.f64N/A
*-lft-identityN/A
metadata-evalN/A
associate-*r*N/A
*-commutativeN/A
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
associate-*r*N/A
sqrt-prodN/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-*.f64N/A
lower-sqrt.f649.1
lift-*.f64N/A
*-commutativeN/A
lower-*.f649.1
Applied rewrites9.1%
lift-asin.f64N/A
asin-acosN/A
lift-PI.f64N/A
div-invN/A
metadata-evalN/A
rem-cube-cbrtN/A
lift-cbrt.f64N/A
pow3N/A
unpow2N/A
lift-pow.f64N/A
associate-*r*N/A
lift-*.f64N/A
lift-acos.f64N/A
flip--N/A
Applied rewrites9.1%
Final simplification9.1%
(FPCore (x)
:precision binary64
(let* ((t_0 (sqrt (PI))))
(fma
(* (sqrt (* 2.0 t_0)) (sqrt (* 0.5 t_0)))
(* t_0 0.5)
(fma (PI) -0.5 (acos (- 1.0 x))))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{\mathsf{PI}\left(\right)}\\
\mathsf{fma}\left(\sqrt{2 \cdot t\_0} \cdot \sqrt{0.5 \cdot t\_0}, t\_0 \cdot 0.5, \mathsf{fma}\left(\mathsf{PI}\left(\right), -0.5, \cos^{-1} \left(1 - x\right)\right)\right)
\end{array}
\end{array}
Initial program 5.4%
Applied rewrites3.6%
lift-sqrt.f64N/A
*-lft-identityN/A
metadata-evalN/A
associate-*r*N/A
*-commutativeN/A
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
associate-*r*N/A
sqrt-prodN/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-*.f64N/A
lower-sqrt.f649.1
lift-*.f64N/A
*-commutativeN/A
lower-*.f649.1
Applied rewrites9.1%
lift-neg.f64N/A
neg-sub0N/A
lift-asin.f64N/A
asin-acosN/A
lift-PI.f64N/A
div-invN/A
metadata-evalN/A
rem-cube-cbrtN/A
lift-cbrt.f64N/A
pow3N/A
unpow2N/A
lift-pow.f64N/A
associate-*r*N/A
lift-*.f64N/A
lift-acos.f64N/A
associate--r-N/A
Applied rewrites9.1%
(FPCore (x)
:precision binary64
(let* ((t_0 (sqrt (PI))))
(fma
(* (sqrt (* 2.0 t_0)) (sqrt (* 0.5 t_0)))
(* t_0 0.5)
(- (asin (- 1.0 x))))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{\mathsf{PI}\left(\right)}\\
\mathsf{fma}\left(\sqrt{2 \cdot t\_0} \cdot \sqrt{0.5 \cdot t\_0}, t\_0 \cdot 0.5, -\sin^{-1} \left(1 - x\right)\right)
\end{array}
\end{array}
Initial program 5.4%
Applied rewrites3.6%
lift-sqrt.f64N/A
*-lft-identityN/A
metadata-evalN/A
associate-*r*N/A
*-commutativeN/A
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
associate-*r*N/A
sqrt-prodN/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-*.f64N/A
lower-sqrt.f649.1
lift-*.f64N/A
*-commutativeN/A
lower-*.f649.1
Applied rewrites9.1%
(FPCore (x) :precision binary64 (let* ((t_0 (sqrt (PI)))) (fma t_0 (* t_0 0.5) (- (fma (* 0.5 t_0) t_0 (- (acos (- 1.0 x))))))))
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{\mathsf{PI}\left(\right)}\\
\mathsf{fma}\left(t\_0, t\_0 \cdot 0.5, -\mathsf{fma}\left(0.5 \cdot t\_0, t\_0, -\cos^{-1} \left(1 - x\right)\right)\right)
\end{array}
\end{array}
Initial program 5.4%
Applied rewrites3.6%
lift-asin.f64N/A
asin-acosN/A
lift-PI.f64N/A
div-invN/A
metadata-evalN/A
*-commutativeN/A
lift-acos.f64N/A
sub-negN/A
*-commutativeN/A
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
*-commutativeN/A
lower-fma.f64N/A
lift-*.f64N/A
*-commutativeN/A
lower-*.f64N/A
lower-neg.f649.0
Applied rewrites9.0%
(FPCore (x) :precision binary64 (- (fma (* -0.5 (sqrt 0.5)) (* (sqrt 2.0) (PI)) (asin (- 1.0 x)))))
\begin{array}{l}
\\
-\mathsf{fma}\left(-0.5 \cdot \sqrt{0.5}, \sqrt{2} \cdot \mathsf{PI}\left(\right), \sin^{-1} \left(1 - x\right)\right)
\end{array}
Initial program 5.4%
Applied rewrites3.6%
lift-sqrt.f64N/A
*-lft-identityN/A
metadata-evalN/A
associate-*r*N/A
*-commutativeN/A
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
associate-*r*N/A
sqrt-prodN/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-*.f64N/A
lower-sqrt.f649.1
lift-*.f64N/A
*-commutativeN/A
lower-*.f649.1
Applied rewrites9.1%
Taylor expanded in x around 0
sub-negN/A
mul-1-negN/A
sub-negN/A
remove-double-negN/A
distribute-neg-inN/A
+-commutativeN/A
sub-negN/A
lower-neg.f64N/A
sub-negN/A
+-commutativeN/A
Applied rewrites9.0%
(FPCore (x) :precision binary64 (let* ((t_0 (sqrt (PI)))) (fma (* -0.5 t_0) t_0 (fma 0.5 (PI) (acos (- 1.0 x))))))
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{\mathsf{PI}\left(\right)}\\
\mathsf{fma}\left(-0.5 \cdot t\_0, t\_0, \mathsf{fma}\left(0.5, \mathsf{PI}\left(\right), \cos^{-1} \left(1 - x\right)\right)\right)
\end{array}
\end{array}
Initial program 5.4%
Applied rewrites5.4%
lift-neg.f64N/A
neg-sub0N/A
lift-asin.f64N/A
asin-acosN/A
lift-PI.f64N/A
div-invN/A
metadata-evalN/A
rem-cube-cbrtN/A
lift-cbrt.f64N/A
pow3N/A
unpow2N/A
lift-pow.f64N/A
associate-*r*N/A
lift-*.f64N/A
lift-acos.f64N/A
associate--r-N/A
Applied rewrites5.4%
lift-fma.f64N/A
rem-3cbrt-lftN/A
lift-cbrt.f64N/A
lift-cbrt.f64N/A
unpow2N/A
lift-pow.f64N/A
lift-cbrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
+-commutativeN/A
lift-fma.f64N/A
associate-+l+N/A
Applied rewrites9.0%
(FPCore (x) :precision binary64 (let* ((t_0 (sqrt (PI)))) (fma (PI) 0.5 (fma (* t_0 t_0) -0.5 (acos (- 1.0 x))))))
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt{\mathsf{PI}\left(\right)}\\
\mathsf{fma}\left(\mathsf{PI}\left(\right), 0.5, \mathsf{fma}\left(t\_0 \cdot t\_0, -0.5, \cos^{-1} \left(1 - x\right)\right)\right)
\end{array}
\end{array}
Initial program 5.4%
Applied rewrites5.4%
lift-neg.f64N/A
neg-sub0N/A
lift-asin.f64N/A
asin-acosN/A
lift-PI.f64N/A
div-invN/A
metadata-evalN/A
rem-cube-cbrtN/A
lift-cbrt.f64N/A
pow3N/A
unpow2N/A
lift-pow.f64N/A
associate-*r*N/A
lift-*.f64N/A
lift-acos.f64N/A
associate--r-N/A
Applied rewrites5.4%
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
lower-*.f649.0
Applied rewrites9.0%
(FPCore (x) :precision binary64 (acos (- x)))
double code(double x) {
return acos(-x);
}
real(8) function code(x)
real(8), intent (in) :: x
code = acos(-x)
end function
public static double code(double x) {
return Math.acos(-x);
}
def code(x): return math.acos(-x)
function code(x) return acos(Float64(-x)) end
function tmp = code(x) tmp = acos(-x); end
code[x_] := N[ArcCos[(-x)], $MachinePrecision]
\begin{array}{l}
\\
\cos^{-1} \left(-x\right)
\end{array}
Initial program 5.4%
Taylor expanded in x around inf
mul-1-negN/A
lower-neg.f646.7
Applied rewrites6.7%
(FPCore (x) :precision binary64 (acos 1.0))
double code(double x) {
return acos(1.0);
}
real(8) function code(x)
real(8), intent (in) :: x
code = acos(1.0d0)
end function
public static double code(double x) {
return Math.acos(1.0);
}
def code(x): return math.acos(1.0)
function code(x) return acos(1.0) end
function tmp = code(x) tmp = acos(1.0); end
code[x_] := N[ArcCos[1.0], $MachinePrecision]
\begin{array}{l}
\\
\cos^{-1} 1
\end{array}
Initial program 5.4%
Taylor expanded in x around 0
Applied rewrites3.8%
(FPCore (x) :precision binary64 (* 2.0 (asin (sqrt (/ x 2.0)))))
double code(double x) {
return 2.0 * asin(sqrt((x / 2.0)));
}
real(8) function code(x)
real(8), intent (in) :: x
code = 2.0d0 * asin(sqrt((x / 2.0d0)))
end function
public static double code(double x) {
return 2.0 * Math.asin(Math.sqrt((x / 2.0)));
}
def code(x): return 2.0 * math.asin(math.sqrt((x / 2.0)))
function code(x) return Float64(2.0 * asin(sqrt(Float64(x / 2.0)))) end
function tmp = code(x) tmp = 2.0 * asin(sqrt((x / 2.0))); end
code[x_] := N[(2.0 * N[ArcSin[N[Sqrt[N[(x / 2.0), $MachinePrecision]], $MachinePrecision]], $MachinePrecision]), $MachinePrecision]
\begin{array}{l}
\\
2 \cdot \sin^{-1} \left(\sqrt{\frac{x}{2}}\right)
\end{array}
herbie shell --seed 2024313
(FPCore (x)
:name "bug323 (missed optimization)"
:precision binary64
:pre (and (<= 0.0 x) (<= x 0.5))
:alt
(! :herbie-platform default (* 2 (asin (sqrt (/ x 2)))))
(acos (- 1.0 x)))