
(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 8 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 (asin (- 1.0 x))))
(/
(fma
(pow (PI) -1.0)
(* (PI) (PI))
(fma
t_0
2.0
(*
(fma
(fma t_0 2.0 (PI))
(- (acos (- 1.0 x)))
(* (fma (PI) 0.5 t_0) (PI)))
(/ -2.0 (PI)))))
(* (* (fma 0.5 (PI) t_0) 2.0) (/ 2.0 (PI))))))\begin{array}{l}
\\
\begin{array}{l}
t_0 := \sin^{-1} \left(1 - x\right)\\
\frac{\mathsf{fma}\left({\mathsf{PI}\left(\right)}^{-1}, \mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right), \mathsf{fma}\left(t\_0, 2, \mathsf{fma}\left(\mathsf{fma}\left(t\_0, 2, \mathsf{PI}\left(\right)\right), -\cos^{-1} \left(1 - x\right), \mathsf{fma}\left(\mathsf{PI}\left(\right), 0.5, t\_0\right) \cdot \mathsf{PI}\left(\right)\right) \cdot \frac{-2}{\mathsf{PI}\left(\right)}\right)\right)}{\left(\mathsf{fma}\left(0.5, \mathsf{PI}\left(\right), t\_0\right) \cdot 2\right) \cdot \frac{2}{\mathsf{PI}\left(\right)}}
\end{array}
\end{array}
Initial program 5.6%
lift-acos.f64N/A
acos-asinN/A
clear-numN/A
asin-acosN/A
acos-asinN/A
flip--N/A
frac-subN/A
frac-subN/A
lower-/.f64N/A
Applied rewrites5.6%
rem-cube-cbrtN/A
lift-PI.f64N/A
lift-PI.f64N/A
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
cbrt-prodN/A
unpow-prod-downN/A
lower-*.f64N/A
lower-pow.f64N/A
lower-cbrt.f64N/A
lower-pow.f64N/A
lower-cbrt.f643.7
Applied rewrites3.7%
Applied rewrites9.4%
Final simplification9.4%
(FPCore (x) :precision binary64 (fma (* (fabs (sqrt (PI))) 0.5) (sqrt (* (sqrt 0.5) (* (sqrt 2.0) (PI)))) (- (asin (- 1.0 x)))))
\begin{array}{l}
\\
\mathsf{fma}\left(\left|\sqrt{\mathsf{PI}\left(\right)}\right| \cdot 0.5, \sqrt{\sqrt{0.5} \cdot \left(\sqrt{2} \cdot \mathsf{PI}\left(\right)\right)}, -\sin^{-1} \left(1 - x\right)\right)
\end{array}
Initial program 5.6%
lift-acos.f64N/A
acos-asinN/A
sub-negN/A
div-invN/A
add-sqr-sqrtN/A
associate-*l*N/A
lower-fma.f64N/A
lower-sqrt.f64N/A
lower-PI.f64N/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-PI.f64N/A
metadata-evalN/A
lower-neg.f64N/A
lower-asin.f643.7
Applied rewrites3.7%
lift-sqrt.f64N/A
rem-cbrt-cubeN/A
sqr-powN/A
cbrt-prodN/A
rem-sqrt-squareN/A
lower-fabs.f64N/A
lower-cbrt.f64N/A
lower-pow.f64N/A
metadata-eval9.4
Applied rewrites9.4%
unpow1N/A
metadata-evalN/A
sqrt-pow1N/A
pow2N/A
Applied rewrites9.4%
Taylor expanded in x around 0
sub-negN/A
*-commutativeN/A
associate-*r*N/A
sub-negN/A
mul-1-negN/A
lower-fma.f64N/A
Applied rewrites9.4%
Final simplification9.4%
(FPCore (x) :precision binary64 (let* ((t_0 (sqrt (PI)))) (fma (PI) 0.5 (- (fma (* t_0 0.5) t_0 (- (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 0.5, t\_0, -\cos^{-1} \left(1 - x\right)\right)\right)
\end{array}
\end{array}
Initial program 5.6%
lift-acos.f64N/A
acos-asinN/A
sub-negN/A
div-invN/A
lower-fma.f64N/A
lower-PI.f64N/A
metadata-evalN/A
lower-neg.f64N/A
lower-asin.f645.6
Applied rewrites5.6%
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
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
lift-neg.f64N/A
lower-fma.f649.4
Applied rewrites9.4%
Final simplification9.4%
(FPCore (x) :precision binary64 (fma (fma (sqrt 2.0) (sqrt 0.5) -1.0) (* 0.5 (PI)) (acos (- 1.0 x))))
\begin{array}{l}
\\
\mathsf{fma}\left(\mathsf{fma}\left(\sqrt{2}, \sqrt{0.5}, -1\right), 0.5 \cdot \mathsf{PI}\left(\right), \cos^{-1} \left(1 - x\right)\right)
\end{array}
Initial program 5.6%
lift-acos.f64N/A
acos-asinN/A
sub-negN/A
div-invN/A
add-sqr-sqrtN/A
associate-*l*N/A
lower-fma.f64N/A
lower-sqrt.f64N/A
lower-PI.f64N/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-PI.f64N/A
metadata-evalN/A
lower-neg.f64N/A
lower-asin.f643.7
Applied rewrites3.7%
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.f649.4
lift-*.f64N/A
*-commutativeN/A
lower-*.f649.4
Applied rewrites9.4%
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
rem-square-sqrtN/A
lift-sqrt.f64N/A
lift-sqrt.f64N/A
associate-*r*N/A
lift-*.f64N/A
lift-neg.f64N/A
lower-fma.f649.4
Applied rewrites9.4%
Taylor expanded in x around 0
associate--l+N/A
sub-negN/A
mul-1-negN/A
+-commutativeN/A
sub-negN/A
associate-*r*N/A
neg-mul-1N/A
*-commutativeN/A
distribute-lft-inN/A
*-commutativeN/A
lower-fma.f64N/A
Applied rewrites9.4%
Final simplification9.4%
(FPCore (x) :precision binary64 (if (<= x 5.6e-17) (fma (/ 2.0 (PI)) (* 0.25 (* (PI) (PI))) (- (asin 1.0))) (acos (- 1.0 x))))
\begin{array}{l}
\\
\begin{array}{l}
\mathbf{if}\;x \leq 5.6 \cdot 10^{-17}:\\
\;\;\;\;\mathsf{fma}\left(\frac{2}{\mathsf{PI}\left(\right)}, 0.25 \cdot \left(\mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right)\right), -\sin^{-1} 1\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 0
Applied rewrites3.8%
lift-acos.f64N/A
acos-asinN/A
sub-negN/A
Applied rewrites7.7%
if 5.5999999999999998e-17 < x Initial program 77.9%
Final simplification9.4%
(FPCore (x) :precision binary64 (fma (/ 2.0 (PI)) (* 0.25 (* (PI) (PI))) (- (asin (- 1.0 x)))))
\begin{array}{l}
\\
\mathsf{fma}\left(\frac{2}{\mathsf{PI}\left(\right)}, 0.25 \cdot \left(\mathsf{PI}\left(\right) \cdot \mathsf{PI}\left(\right)\right), -\sin^{-1} \left(1 - x\right)\right)
\end{array}
Initial program 5.6%
lift-acos.f64N/A
acos-asinN/A
sub-negN/A
div-invN/A
add-sqr-sqrtN/A
associate-*l*N/A
lower-fma.f64N/A
lower-sqrt.f64N/A
lower-PI.f64N/A
lower-*.f64N/A
lower-sqrt.f64N/A
lower-PI.f64N/A
metadata-evalN/A
lower-neg.f64N/A
lower-asin.f643.7
Applied rewrites3.7%
lift-fma.f64N/A
Applied rewrites9.4%
Final simplification9.4%
(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 5.6%
Taylor expanded in x around inf
mul-1-negN/A
lower-neg.f646.8
Applied rewrites6.8%
(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 5.6%
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 2024276
(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)))