Diagrams.ThreeD.Transform:aboutX from diagrams-lib-1.3.0.3, B

Percentage Accurate: 99.8% → 99.8%
Time: 12.3s
Alternatives: 8
Speedup: 1.0×

Specification

?
\[\begin{array}{l} \\ x \cdot \sin y + z \cdot \cos y \end{array} \]
(FPCore (x y z) :precision binary64 (+ (* x (sin y)) (* z (cos y))))
double code(double x, double y, double z) {
	return (x * sin(y)) + (z * cos(y));
}

real(8) function code(x, y, z)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    code = (x * sin(y)) + (z * cos(y))
end function

public static double code(double x, double y, double z) {
	return (x * Math.sin(y)) + (z * Math.cos(y));
}

def code(x, y, z):
	return (x * math.sin(y)) + (z * math.cos(y))

function code(x, y, z)
	return Float64(Float64(x * sin(y)) + Float64(z * cos(y)))
end

function tmp = code(x, y, z)
	tmp = (x * sin(y)) + (z * cos(y));
end

code[x_, y_, z_] := N[(N[(x * N[Sin[y], $MachinePrecision]), $MachinePrecision] + N[(z * N[Cos[y], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

\begin{array}{l}

\\
x \cdot \sin y + z \cdot \cos y
\end{array}

Sampling outcomes in binary64 precision:

Local Percentage Accuracy vs ?

The average percentage accuracy by input value. Horizontal axis shows value of an input variable; the variable is choosen in the title. Vertical axis is accuracy; higher is better. Red represent the original program, while blue represents Herbie's suggestion. These can be toggled with buttons below the plot. The line is an average while dots represent individual samples.

Accuracy vs Speed?

Herbie found 8 alternatives:

AlternativeAccuracySpeedup
The accuracy (vertical axis) and speed (horizontal axis) of each alternatives. Up and to the right is better. The red square shows the initial program, and each blue circle shows an alternative.The line shows the best available speed-accuracy tradeoffs.

Initial Program: 99.8% accurate, 1.0× speedup?

\[\begin{array}{l} \\ x \cdot \sin y + z \cdot \cos y \end{array} \]
(FPCore (x y z) :precision binary64 (+ (* x (sin y)) (* z (cos y))))
double code(double x, double y, double z) {
	return (x * sin(y)) + (z * cos(y));
}

real(8) function code(x, y, z)
    real(8), intent (in) :: x
    real(8), intent (in) :: y
    real(8), intent (in) :: z
    code = (x * sin(y)) + (z * cos(y))
end function

public static double code(double x, double y, double z) {
	return (x * Math.sin(y)) + (z * Math.cos(y));
}

def code(x, y, z):
	return (x * math.sin(y)) + (z * math.cos(y))

function code(x, y, z)
	return Float64(Float64(x * sin(y)) + Float64(z * cos(y)))
end

function tmp = code(x, y, z)
	tmp = (x * sin(y)) + (z * cos(y));
end

code[x_, y_, z_] := N[(N[(x * N[Sin[y], $MachinePrecision]), $MachinePrecision] + N[(z * N[Cos[y], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

\begin{array}{l}

\\
x \cdot \sin y + z \cdot \cos y
\end{array}

Alternative 1: 99.8% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \mathsf{fma}\left(\sin y, x, z \cdot \cos y\right) \end{array} \]
(FPCore (x y z) :precision binary64 (fma (sin y) x (* z (cos y))))
double code(double x, double y, double z) {
	return fma(sin(y), x, (z * cos(y)));
}

function code(x, y, z)
	return fma(sin(y), x, Float64(z * cos(y)))
end

code[x_, y_, z_] := N[(N[Sin[y], $MachinePrecision] * x + N[(z * N[Cos[y], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

\begin{array}{l}

\\
\mathsf{fma}\left(\sin y, x, z \cdot \cos y\right)
\end{array}
Derivation

  1. Initial program 99.8%

    \[x \cdot \sin y + z \cdot \cos y \]
  2. Add Preprocessing
  3. Step-by-step derivation

    1. lift-+.f64N/A

      \[\leadsto \color{blue}{x \cdot \sin y + z \cdot \cos y} \]

    2. lift-*.f64N/A

      \[\leadsto \color{blue}{x \cdot \sin y} + z \cdot \cos y \]

    3. *-commutativeN/A

      \[\leadsto \color{blue}{\sin y \cdot x} + z \cdot \cos y \]

    4. lower-fma.f6499.8

      \[\leadsto \color{blue}{\mathsf{fma}\left(\sin y, x, z \cdot \cos y\right)} \]

  4. Applied rewrites99.8%

    \[\leadsto \color{blue}{\mathsf{fma}\left(\sin y, x, z \cdot \cos y\right)} \]
  5. Add Preprocessing

Alternative 2: 86.8% accurate, 1.7× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := z \cdot \cos y\\ \mathbf{if}\;z \leq -5.9 \cdot 10^{+66}:\\ \;\;\;\;t\_0\\ \mathbf{elif}\;z \leq 7.2 \cdot 10^{+95}:\\ \;\;\;\;\mathsf{fma}\left(\sin y, x, z \cdot 1\right)\\ \mathbf{else}:\\ \;\;\;\;t\_0\\ \end{array} \end{array} \]
(FPCore (x y z)
 :precision binary64
 (let* ((t_0 (* z (cos y))))
   (if (<= z -5.9e+66) t_0 (if (<= z 7.2e+95) (fma (sin y) x (* z 1.0)) t_0))))
double code(double x, double y, double z) {
	double t_0 = z * cos(y);
	double tmp;
	if (z <= -5.9e+66) {
		tmp = t_0;
	} else if (z <= 7.2e+95) {
		tmp = fma(sin(y), x, (z * 1.0));
	} else {
		tmp = t_0;
	}
	return tmp;
}

function code(x, y, z)
	t_0 = Float64(z * cos(y))
	tmp = 0.0
	if (z <= -5.9e+66)
		tmp = t_0;
	elseif (z <= 7.2e+95)
		tmp = fma(sin(y), x, Float64(z * 1.0));
	else
		tmp = t_0;
	end
	return tmp
end

code[x_, y_, z_] := Block[{t$95$0 = N[(z * N[Cos[y], $MachinePrecision]), $MachinePrecision]}, If[LessEqual[z, -5.9e+66], t$95$0, If[LessEqual[z, 7.2e+95], N[(N[Sin[y], $MachinePrecision] * x + N[(z * 1.0), $MachinePrecision]), $MachinePrecision], t$95$0]]]

\begin{array}{l}

\\
\begin{array}{l}
t_0 := z \cdot \cos y\\
\mathbf{if}\;z \leq -5.9 \cdot 10^{+66}:\\
\;\;\;\;t\_0\\

\mathbf{elif}\;z \leq 7.2 \cdot 10^{+95}:\\
\;\;\;\;\mathsf{fma}\left(\sin y, x, z \cdot 1\right)\\

\mathbf{else}:\\
\;\;\;\;t\_0\\


\end{array}
\end{array}
Derivation
  1. Split input into 2 regimes

  2. if z < -5.89999999999999988e66 or 7.19999999999999955e95 < z

    1. Initial program 99.8%

      \[x \cdot \sin y + z \cdot \cos y \]
    2. Add Preprocessing

    3. Taylor expanded in x around 0

      \[\leadsto \color{blue}{z \cdot \cos y} \]
    4. Step-by-step derivation

      1. lower-*.f64N/A

        \[\leadsto \color{blue}{z \cdot \cos y} \]

      2. lower-cos.f6488.8

        \[\leadsto z \cdot \color{blue}{\cos y} \]

    5. Applied rewrites88.8%

      \[\leadsto \color{blue}{z \cdot \cos y} \]

    if -5.89999999999999988e66 < z < 7.19999999999999955e95

    1. Initial program 99.8%

      \[x \cdot \sin y + z \cdot \cos y \]
    2. Add Preprocessing
    3. Step-by-step derivation

      1. lift-+.f64N/A

        \[\leadsto \color{blue}{x \cdot \sin y + z \cdot \cos y} \]

      2. lift-*.f64N/A

        \[\leadsto \color{blue}{x \cdot \sin y} + z \cdot \cos y \]

      3. *-commutativeN/A

        \[\leadsto \color{blue}{\sin y \cdot x} + z \cdot \cos y \]

      4. lower-fma.f6499.8

        \[\leadsto \color{blue}{\mathsf{fma}\left(\sin y, x, z \cdot \cos y\right)} \]

    4. Applied rewrites99.8%

      \[\leadsto \color{blue}{\mathsf{fma}\left(\sin y, x, z \cdot \cos y\right)} \]

    5. Taylor expanded in y around 0

      \[\leadsto \mathsf{fma}\left(\sin y, x, z \cdot \color{blue}{1}\right) \]
    6. Step-by-step derivation

      1. Applied rewrites85.5%

        \[\leadsto \mathsf{fma}\left(\sin y, x, z \cdot \color{blue}{1}\right) \]
    7. Recombined 2 regimes into one program.
    8. Add Preprocessing

    Reproduce

    ?
    herbie shell --seed 2024223 
(FPCore (x y z)
  :name "Diagrams.ThreeD.Transform:aboutX from diagrams-lib-1.3.0.3, B"
  :precision binary64
  (+ (* x (sin y)) (* z (cos y))))