
(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 12 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 (pow t_0 2.0))
(t_2 (/ (PI) 2.0))
(t_3 (pow t_2 2.0))
(t_4 (fma t_0 (- t_0 t_2) t_3))
(t_5 (cbrt (- (sqrt (PI))))))
(/
(/
(fma
(* (+ (pow t_0 3.0) (pow t_2 3.0)) (PI))
(+ t_1 t_3)
(* t_4 (* -2.0 (- (pow t_2 4.0) (pow t_0 4.0)))))
(* t_4 (+ t_1 (pow (* (pow (* t_5 t_5) 2.0) (/ (cbrt (PI)) 2.0)) 2.0))))
(* 2.0 (+ t_2 t_0)))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos^{-1} \left(1 - x\right)\\
t_1 := {t\_0}^{2}\\
t_2 := \frac{\mathsf{PI}\left(\right)}{2}\\
t_3 := {t\_2}^{2}\\
t_4 := \mathsf{fma}\left(t\_0, t\_0 - t\_2, t\_3\right)\\
t_5 := \sqrt[3]{-\sqrt{\mathsf{PI}\left(\right)}}\\
\frac{\frac{\mathsf{fma}\left(\left({t\_0}^{3} + {t\_2}^{3}\right) \cdot \mathsf{PI}\left(\right), t\_1 + t\_3, t\_4 \cdot \left(-2 \cdot \left({t\_2}^{4} - {t\_0}^{4}\right)\right)\right)}{t\_4 \cdot \left(t\_1 + {\left({\left(t\_5 \cdot t\_5\right)}^{2} \cdot \frac{\sqrt[3]{\mathsf{PI}\left(\right)}}{2}\right)}^{2}\right)}}{2 \cdot \left(t\_2 + t\_0\right)}
\end{array}
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
asin-acosN/A
lift-acos.f64N/A
flip--N/A
frac-subN/A
lower-/.f64N/A
Applied rewrites6.5%
Applied rewrites10.1%
lift-/.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-/l*N/A
lower-*.f64N/A
lower-/.f6410.1
Applied rewrites10.1%
lift-cbrt.f64N/A
lift-PI.f64N/A
add-sqr-sqrtN/A
sqr-neg-revN/A
cbrt-prodN/A
lower-*.f64N/A
lower-cbrt.f64N/A
lower-neg.f64N/A
lift-PI.f64N/A
lower-sqrt.f64N/A
lower-cbrt.f64N/A
lower-neg.f64N/A
lift-PI.f64N/A
lower-sqrt.f6410.1
Applied rewrites10.1%
(FPCore (x)
:precision binary64
(let* ((t_0 (cbrt (PI)))
(t_1 (acos (- 1.0 x)))
(t_2 (/ (PI) 2.0))
(t_3 (pow t_2 2.0))
(t_4 (- t_1 t_2))
(t_5 (pow t_1 2.0))
(t_6 (sqrt (PI))))
(/
(/
(fma
(* (+ (pow t_1 3.0) (pow t_2 3.0)) (PI))
(+ t_5 t_3)
(* (fma t_1 t_4 t_3) (* -2.0 (- (pow t_2 4.0) (pow t_1 4.0)))))
(*
(fma t_1 t_4 (pow (/ (* t_6 t_6) 2.0) 2.0))
(+ t_5 (pow (* (pow t_0 2.0) (/ t_0 2.0)) 2.0))))
(* 2.0 (+ t_2 t_1)))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sqrt[3]{\mathsf{PI}\left(\right)}\\
t_1 := \cos^{-1} \left(1 - x\right)\\
t_2 := \frac{\mathsf{PI}\left(\right)}{2}\\
t_3 := {t\_2}^{2}\\
t_4 := t\_1 - t\_2\\
t_5 := {t\_1}^{2}\\
t_6 := \sqrt{\mathsf{PI}\left(\right)}\\
\frac{\frac{\mathsf{fma}\left(\left({t\_1}^{3} + {t\_2}^{3}\right) \cdot \mathsf{PI}\left(\right), t\_5 + t\_3, \mathsf{fma}\left(t\_1, t\_4, t\_3\right) \cdot \left(-2 \cdot \left({t\_2}^{4} - {t\_1}^{4}\right)\right)\right)}{\mathsf{fma}\left(t\_1, t\_4, {\left(\frac{t\_6 \cdot t\_6}{2}\right)}^{2}\right) \cdot \left(t\_5 + {\left({t\_0}^{2} \cdot \frac{t\_0}{2}\right)}^{2}\right)}}{2 \cdot \left(t\_2 + t\_1\right)}
\end{array}
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
asin-acosN/A
lift-acos.f64N/A
flip--N/A
frac-subN/A
lower-/.f64N/A
Applied rewrites6.5%
Applied rewrites10.1%
lift-/.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-/l*N/A
lower-*.f64N/A
lower-/.f6410.1
Applied rewrites10.1%
lift-PI.f64N/A
add-sqr-sqrtN/A
lower-*.f64N/A
lift-PI.f64N/A
lower-sqrt.f64N/A
lift-PI.f64N/A
lower-sqrt.f6410.1
Applied rewrites10.1%
(FPCore (x)
:precision binary64
(let* ((t_0 (acos (- 1.0 x)))
(t_1 (pow t_0 2.0))
(t_2 (/ (PI) 2.0))
(t_3 (pow t_2 2.0))
(t_4 (* (PI) 0.5))
(t_5 (- t_0 t_2))
(t_6 (cbrt (PI))))
(/
(/
(fma
(* (+ (pow t_0 3.0) (pow t_2 3.0)) (PI))
(+ t_1 t_3)
(* (fma t_0 t_5 t_3) (* -2.0 (- (pow t_2 4.0) (pow t_0 4.0)))))
(*
(fma t_0 t_5 (* t_4 t_4))
(+ t_1 (pow (* (pow t_6 2.0) (/ t_6 2.0)) 2.0))))
(* 2.0 (+ t_2 t_0)))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos^{-1} \left(1 - x\right)\\
t_1 := {t\_0}^{2}\\
t_2 := \frac{\mathsf{PI}\left(\right)}{2}\\
t_3 := {t\_2}^{2}\\
t_4 := \mathsf{PI}\left(\right) \cdot 0.5\\
t_5 := t\_0 - t\_2\\
t_6 := \sqrt[3]{\mathsf{PI}\left(\right)}\\
\frac{\frac{\mathsf{fma}\left(\left({t\_0}^{3} + {t\_2}^{3}\right) \cdot \mathsf{PI}\left(\right), t\_1 + t\_3, \mathsf{fma}\left(t\_0, t\_5, t\_3\right) \cdot \left(-2 \cdot \left({t\_2}^{4} - {t\_0}^{4}\right)\right)\right)}{\mathsf{fma}\left(t\_0, t\_5, t\_4 \cdot t\_4\right) \cdot \left(t\_1 + {\left({t\_6}^{2} \cdot \frac{t\_6}{2}\right)}^{2}\right)}}{2 \cdot \left(t\_2 + t\_0\right)}
\end{array}
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
asin-acosN/A
lift-acos.f64N/A
flip--N/A
frac-subN/A
lower-/.f64N/A
Applied rewrites6.5%
Applied rewrites10.1%
lift-/.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-/l*N/A
lower-*.f64N/A
lower-/.f6410.1
Applied rewrites10.1%
lift-pow.f64N/A
unpow2N/A
lower-*.f6410.1
lift-/.f64N/A
*-rgt-identityN/A
associate-/l*N/A
metadata-evalN/A
lower-*.f6410.1
lift-/.f64N/A
*-rgt-identityN/A
associate-/l*N/A
metadata-evalN/A
lower-*.f6410.1
Applied rewrites10.1%
(FPCore (x)
:precision binary64
(let* ((t_0 (acos (- 1.0 x)))
(t_1 (pow t_0 2.0))
(t_2 (/ (PI) 2.0))
(t_3 (pow t_2 2.0))
(t_4 (fma t_0 (- t_0 t_2) t_3)))
(/
(/
(fma
(* (+ (pow t_0 3.0) (pow t_2 3.0)) (PI))
(+ t_1 t_3)
(* t_4 (* -2.0 (- (pow t_2 4.0) (pow t_0 4.0)))))
(* t_4 (fma (PI) (pow (/ (sqrt (PI)) 2.0) 2.0) t_1)))
(* 2.0 (+ t_2 t_0)))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \cos^{-1} \left(1 - x\right)\\
t_1 := {t\_0}^{2}\\
t_2 := \frac{\mathsf{PI}\left(\right)}{2}\\
t_3 := {t\_2}^{2}\\
t_4 := \mathsf{fma}\left(t\_0, t\_0 - t\_2, t\_3\right)\\
\frac{\frac{\mathsf{fma}\left(\left({t\_0}^{3} + {t\_2}^{3}\right) \cdot \mathsf{PI}\left(\right), t\_1 + t\_3, t\_4 \cdot \left(-2 \cdot \left({t\_2}^{4} - {t\_0}^{4}\right)\right)\right)}{t\_4 \cdot \mathsf{fma}\left(\mathsf{PI}\left(\right), {\left(\frac{\sqrt{\mathsf{PI}\left(\right)}}{2}\right)}^{2}, t\_1\right)}}{2 \cdot \left(t\_2 + t\_0\right)}
\end{array}
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
asin-acosN/A
lift-acos.f64N/A
flip--N/A
frac-subN/A
lower-/.f64N/A
Applied rewrites6.5%
Applied rewrites10.1%
lift-+.f64N/A
+-commutativeN/A
lift-pow.f64N/A
lift-/.f64N/A
lift-PI.f64N/A
add-sqr-sqrtN/A
associate-/l*N/A
unpow-prod-downN/A
pow2N/A
add-sqr-sqrtN/A
lift-PI.f64N/A
lower-fma.f64N/A
lower-pow.f64N/A
lower-/.f64N/A
lift-PI.f64N/A
lower-sqrt.f6410.1
Applied rewrites10.1%
(FPCore (x)
:precision binary64
(let* ((t_0 (/ (PI) 2.0))
(t_1 (acos (- 1.0 x)))
(t_2 (pow t_1 2.0))
(t_3 (fma (PI) (/ (PI) 4.0) t_2))
(t_4 (fma (asin (- 1.0 x)) t_0 t_2)))
(/
(/
(fma
t_3
(* (+ (pow t_0 3.0) (pow t_1 3.0)) (PI))
(* (* -2.0 t_4) (- (pow (/ (PI) -2.0) 4.0) (pow t_1 4.0))))
(* t_4 t_3))
(* 2.0 (+ t_0 t_1)))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \frac{\mathsf{PI}\left(\right)}{2}\\
t_1 := \cos^{-1} \left(1 - x\right)\\
t_2 := {t\_1}^{2}\\
t_3 := \mathsf{fma}\left(\mathsf{PI}\left(\right), \frac{\mathsf{PI}\left(\right)}{4}, t\_2\right)\\
t_4 := \mathsf{fma}\left(\sin^{-1} \left(1 - x\right), t\_0, t\_2\right)\\
\frac{\frac{\mathsf{fma}\left(t\_3, \left({t\_0}^{3} + {t\_1}^{3}\right) \cdot \mathsf{PI}\left(\right), \left(-2 \cdot t\_4\right) \cdot \left({\left(\frac{\mathsf{PI}\left(\right)}{-2}\right)}^{4} - {t\_1}^{4}\right)\right)}{t\_4 \cdot t\_3}}{2 \cdot \left(t\_0 + t\_1\right)}
\end{array}
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
asin-acosN/A
lift-acos.f64N/A
flip--N/A
frac-subN/A
lower-/.f64N/A
Applied rewrites6.5%
Applied rewrites10.1%
Applied rewrites10.1%
(FPCore (x) :precision binary64 (let* ((t_0 (asin (- 1.0 x)))) (/ (fma t_0 (asin (- x 1.0)) (* (* (PI) (PI)) 0.25)) (fma 0.5 (PI) t_0))))
\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sin^{-1} \left(1 - x\right)\\
\frac{\mathsf{fma}\left(t\_0, \sin^{-1} \left(x - 1\right), \left(\mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right)\right) \cdot 0.25\right)}{\mathsf{fma}\left(0.5, \mathsf{PI}\left(\right), t\_0\right)}
\end{array}
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
flip--N/A
lower-/.f64N/A
lower--.f64N/A
pow2N/A
lower-pow.f64N/A
lower-/.f64N/A
lower-PI.f64N/A
pow2N/A
lower-pow.f64N/A
lower-asin.f64N/A
+-commutativeN/A
lower-+.f64N/A
lower-asin.f64N/A
lower-/.f64N/A
lower-PI.f646.5
Applied rewrites6.5%
lift--.f64N/A
lift-pow.f64N/A
unpow2N/A
fp-cancel-sub-sign-invN/A
lift-pow.f64N/A
unpow2N/A
lower-fma.f64N/A
lower-*.f64N/A
lift--.f64N/A
lift-asin.f64N/A
asin-neg-revN/A
lower-asin.f64N/A
lower-neg.f64N/A
lift--.f644.7
Applied rewrites4.7%
Taylor expanded in x around 0
*-lft-identityN/A
metadata-evalN/A
fp-cancel-sign-sub-invN/A
lower-/.f64N/A
Applied rewrites10.1%
(FPCore (x) :precision binary64 (- (/ (cbrt (pow (PI) 3.0)) 2.0) (asin (- 1.0 x))))
\begin{array}{l}
\\
\frac{\sqrt[3]{{\mathsf{PI}\left(\right)}^{3}}}{2} - \sin^{-1} \left(1 - x\right)
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
lower--.f64N/A
lower-/.f64N/A
lower-PI.f64N/A
lower-asin.f646.5
Applied rewrites6.5%
unpow1N/A
metadata-evalN/A
pow-powN/A
*-rgt-identityN/A
lift-*.f64N/A
pow1/3N/A
lower-cbrt.f64N/A
lift-*.f64N/A
*-rgt-identityN/A
lower-pow.f6410.0
Applied rewrites10.0%
(FPCore (x) :precision binary64 (- (/ (PI) 2.0) (pow (sqrt (asin (- 1.0 x))) 2.0)))
\begin{array}{l}
\\
\frac{\mathsf{PI}\left(\right)}{2} - {\left(\sqrt{\sin^{-1} \left(1 - x\right)}\right)}^{2}
\end{array}
Initial program 6.5%
lift-acos.f64N/A
acos-asinN/A
lower--.f64N/A
lower-/.f64N/A
lower-PI.f64N/A
lower-asin.f646.5
Applied rewrites6.5%
unpow1N/A
sqr-powN/A
pow2N/A
lower-pow.f64N/A
metadata-evalN/A
unpow1/2N/A
lower-sqrt.f6410.0
Applied rewrites10.0%
(FPCore (x) :precision binary64 (if (<= x 5.6e-17) (acos (- x)) (- (/ (PI) 2.0) (- (* (PI) 0.5) (acos (- 1.0 x))))))
\begin{array}{l}
\\
\begin{array}{l}
\mathbf{if}\;x \leq 5.6 \cdot 10^{-17}:\\
\;\;\;\;\cos^{-1} \left(-x\right)\\
\mathbf{else}:\\
\;\;\;\;\frac{\mathsf{PI}\left(\right)}{2} - \left(\mathsf{PI}\left(\right) \cdot 0.5 - \cos^{-1} \left(1 - x\right)\right)\\
\end{array}
\end{array}
if x < 5.5999999999999998e-17Initial program 3.8%
Taylor expanded in x around inf
mul-1-negN/A
lower-neg.f646.6
Applied rewrites6.6%
if 5.5999999999999998e-17 < x Initial program 55.6%
lift-acos.f64N/A
acos-asinN/A
lower--.f64N/A
lower-/.f64N/A
lower-PI.f64N/A
lower-asin.f6455.6
Applied rewrites55.6%
lift-asin.f64N/A
asin-acosN/A
lift-PI.f64N/A
lift-/.f64N/A
lift-acos.f64N/A
lower--.f6455.7
lift-/.f64N/A
*-rgt-identityN/A
associate-/l*N/A
metadata-evalN/A
lower-*.f6455.7
Applied rewrites55.7%
(FPCore (x) :precision binary64 (if (<= x 5.6e-17) (acos (- x)) (acos (- 1.0 x))))
double code(double x) {
double tmp;
if (x <= 5.6e-17) {
tmp = acos(-x);
} else {
tmp = acos((1.0 - x));
}
return tmp;
}
real(8) function code(x)
real(8), intent (in) :: x
real(8) :: tmp
if (x <= 5.6d-17) then
tmp = acos(-x)
else
tmp = acos((1.0d0 - x))
end if
code = tmp
end function
public static double code(double x) {
double tmp;
if (x <= 5.6e-17) {
tmp = Math.acos(-x);
} else {
tmp = Math.acos((1.0 - x));
}
return tmp;
}
def code(x): tmp = 0 if x <= 5.6e-17: tmp = math.acos(-x) else: tmp = math.acos((1.0 - x)) return tmp
function code(x) tmp = 0.0 if (x <= 5.6e-17) tmp = acos(Float64(-x)); else tmp = acos(Float64(1.0 - x)); end return tmp end
function tmp_2 = code(x) tmp = 0.0; if (x <= 5.6e-17) tmp = acos(-x); else tmp = acos((1.0 - x)); end tmp_2 = tmp; end
code[x_] := If[LessEqual[x, 5.6e-17], N[ArcCos[(-x)], $MachinePrecision], N[ArcCos[N[(1.0 - x), $MachinePrecision]], $MachinePrecision]]
\begin{array}{l}
\\
\begin{array}{l}
\mathbf{if}\;x \leq 5.6 \cdot 10^{-17}:\\
\;\;\;\;\cos^{-1} \left(-x\right)\\
\mathbf{else}:\\
\;\;\;\;\cos^{-1} \left(1 - x\right)\\
\end{array}
\end{array}
if x < 5.5999999999999998e-17Initial program 3.8%
Taylor expanded in x around inf
mul-1-negN/A
lower-neg.f646.6
Applied rewrites6.6%
if 5.5999999999999998e-17 < x Initial program 55.6%
(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 6.5%
Taylor expanded in x around inf
mul-1-negN/A
lower-neg.f646.9
Applied rewrites6.9%
(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 6.5%
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 2024337
(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)))