math.sin on complex, imaginary part

Percentage Accurate: 52.9% → 99.8%

Time: 4.7s

Alternatives: 10

Speedup: 1.3×

• 48.3% of points produce a very large (infinite) output. You may want to add a precondition.

Specification

Math

\[\begin{array}{l} \\ \left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \end{array} \]

FPCore

(FPCore (re im)
 :precision binary64
 (* (* 0.5 (cos re)) (- (exp (- 0.0 im)) (exp im))))

C

double code(double re, double im) {
	return (0.5 * cos(re)) * (exp((0.0 - im)) - exp(im));
}

Fortran

module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(re, im)
use fmin_fmax_functions
    real(8), intent (in) :: re
    real(8), intent (in) :: im
    code = (0.5d0 * cos(re)) * (exp((0.0d0 - im)) - exp(im))
end function

Java

public static double code(double re, double im) {
	return (0.5 * Math.cos(re)) * (Math.exp((0.0 - im)) - Math.exp(im));
}

Python

def code(re, im):
	return (0.5 * math.cos(re)) * (math.exp((0.0 - im)) - math.exp(im))

Julia

function code(re, im)
	return Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im)) - exp(im)))
end

MATLAB

function tmp = code(re, im)
	tmp = (0.5 * cos(re)) * (exp((0.0 - im)) - exp(im));
end

Wolfram

code[re_, im_] := N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im), $MachinePrecision]], $MachinePrecision] - N[Exp[im], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Initial Program: 52.9% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \end{array} \]

FPCore

(FPCore (re im)
 :precision binary64
 (* (* 0.5 (cos re)) (- (exp (- 0.0 im)) (exp im))))

C

double code(double re, double im) {
	return (0.5 * cos(re)) * (exp((0.0 - im)) - exp(im));
}

Fortran

module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(re, im)
use fmin_fmax_functions
    real(8), intent (in) :: re
    real(8), intent (in) :: im
    code = (0.5d0 * cos(re)) * (exp((0.0d0 - im)) - exp(im))
end function

Java

public static double code(double re, double im) {
	return (0.5 * Math.cos(re)) * (Math.exp((0.0 - im)) - Math.exp(im));
}

Python

def code(re, im):
	return (0.5 * math.cos(re)) * (math.exp((0.0 - im)) - math.exp(im))

Julia

function code(re, im)
	return Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im)) - exp(im)))
end

MATLAB

function tmp = code(re, im)
	tmp = (0.5 * cos(re)) * (exp((0.0 - im)) - exp(im));
end

Wolfram

code[re_, im_] := N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im), $MachinePrecision]], $MachinePrecision] - N[Exp[im], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]

Alternative 1: 99.8% accurate, 0.9× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ im\_s \cdot \begin{array}{l} \mathbf{if}\;im\_m \leq 6 \cdot 10^{-6}:\\ \;\;\;\;\left(-im\_m\right) \cdot \cos re\\ \mathbf{else}:\\ \;\;\;\;\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right)\\ \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (*
  im_s
  (if (<= im_m 6e-6)
    (* (- im_m) (cos re))
    (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m))))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double tmp;
	if (im_m <= 6e-6) {
		tmp = -im_m * cos(re);
	} else {
		tmp = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: tmp
    if (im_m <= 6d-6) then
        tmp = -im_m * cos(re)
    else
        tmp = (0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double tmp;
	if (im_m <= 6e-6) {
		tmp = -im_m * Math.cos(re);
	} else {
		tmp = (0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m));
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	tmp = 0
	if im_m <= 6e-6:
		tmp = -im_m * math.cos(re)
	else:
		tmp = (0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	tmp = 0.0
	if (im_m <= 6e-6)
		tmp = Float64(Float64(-im_m) * cos(re));
	else
		tmp = Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m)));
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	tmp = 0.0;
	if (im_m <= 6e-6)
		tmp = -im_m * cos(re);
	else
		tmp = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := N[(im$95$s * If[LessEqual[im$95$m, 6e-6], N[((-im$95$m) * N[Cos[re], $MachinePrecision]), $MachinePrecision], N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]), $MachinePrecision]

Derivation

1. Split input into 2 regimes

2. if im < 6.0000000000000002e-6

   1. Initial program 7.1%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in im around 0

      \[\leadsto \color{blue}{-1 \cdot \left(im \cdot \cos re\right)} \]
   3. Step-by-step derivation

      1. associate-*r* N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      3. mul-1-neg N/A

         \[\leadsto \left(\mathsf{neg}\left(im\right)\right) \cdot \cos \color{blue}{re} \]

      4. lower-neg.f64 N/A

         \[\leadsto \left(-im\right) \cdot \cos \color{blue}{re} \]

      5. lift-cos.f64 99.7

         \[\leadsto \left(-im\right) \cdot \cos re \]

   4. Applied rewrites 99.7%

      \[\leadsto \color{blue}{\left(-im\right) \cdot \cos re} \]

   if 6.0000000000000002e-6 < im

   1. Initial program 99.9%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 2: 98.6% accurate, 0.4× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ \begin{array}{l} t_0 := \left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right)\\ im\_s \cdot \begin{array}{l} \mathbf{if}\;t\_0 \leq -100000:\\ \;\;\;\;\left(-2 \cdot \sinh im\_m\right) \cdot 0.5\\ \mathbf{elif}\;t\_0 \leq 0:\\ \;\;\;\;\left(-im\_m\right) \cdot \cos re\\ \mathbf{else}:\\ \;\;\;\;\left(\left(re \cdot re\right) \cdot -0.25\right) \cdot \left(1 - e^{im\_m}\right)\\ \end{array} \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (let* ((t_0 (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m)))))
   (*
    im_s
    (if (<= t_0 -100000.0)
      (* (* -2.0 (sinh im_m)) 0.5)
      (if (<= t_0 0.0)
        (* (- im_m) (cos re))
        (* (* (* re re) -0.25) (- 1.0 (exp im_m))))))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = (-2.0 * sinh(im_m)) * 0.5;
	} else if (t_0 <= 0.0) {
		tmp = -im_m * cos(re);
	} else {
		tmp = ((re * re) * -0.25) * (1.0 - exp(im_m));
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: t_0
    real(8) :: tmp
    t_0 = (0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))
    if (t_0 <= (-100000.0d0)) then
        tmp = ((-2.0d0) * sinh(im_m)) * 0.5d0
    else if (t_0 <= 0.0d0) then
        tmp = -im_m * cos(re)
    else
        tmp = ((re * re) * (-0.25d0)) * (1.0d0 - exp(im_m))
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = (-2.0 * Math.sinh(im_m)) * 0.5;
	} else if (t_0 <= 0.0) {
		tmp = -im_m * Math.cos(re);
	} else {
		tmp = ((re * re) * -0.25) * (1.0 - Math.exp(im_m));
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	t_0 = (0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))
	tmp = 0
	if t_0 <= -100000.0:
		tmp = (-2.0 * math.sinh(im_m)) * 0.5
	elif t_0 <= 0.0:
		tmp = -im_m * math.cos(re)
	else:
		tmp = ((re * re) * -0.25) * (1.0 - math.exp(im_m))
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	t_0 = Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m)))
	tmp = 0.0
	if (t_0 <= -100000.0)
		tmp = Float64(Float64(-2.0 * sinh(im_m)) * 0.5);
	elseif (t_0 <= 0.0)
		tmp = Float64(Float64(-im_m) * cos(re));
	else
		tmp = Float64(Float64(Float64(re * re) * -0.25) * Float64(1.0 - exp(im_m)));
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	tmp = 0.0;
	if (t_0 <= -100000.0)
		tmp = (-2.0 * sinh(im_m)) * 0.5;
	elseif (t_0 <= 0.0)
		tmp = -im_m * cos(re);
	else
		tmp = ((re * re) * -0.25) * (1.0 - exp(im_m));
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := Block[{t$95$0 = N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]}, N[(im$95$s * If[LessEqual[t$95$0, -100000.0], N[(N[(-2.0 * N[Sinh[im$95$m], $MachinePrecision]), $MachinePrecision] * 0.5), $MachinePrecision], If[LessEqual[t$95$0, 0.0], N[((-im$95$m) * N[Cos[re], $MachinePrecision]), $MachinePrecision], N[(N[(N[(re * re), $MachinePrecision] * -0.25), $MachinePrecision] * N[(1.0 - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]]), $MachinePrecision]]

Derivation

1. Split input into 3 regimes

2. if (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < -1e5

   1. Initial program 100.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2}} \cdot \left(e^{0 - im} - e^{im}\right) \]
   3. Step-by-step derivation

   4. Applied rewrites 99.9%

      \[\leadsto \color{blue}{0.5} \cdot \left(e^{0 - im} - e^{im}\right) \]
   5. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{0 - im} - e^{im}\right)} \]

      2. lift--.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \color{blue}{\left(e^{0 - im} - e^{im}\right)} \]

      3. lift--.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(e^{\color{blue}{0 - im}} - e^{im}\right) \]

      4. lift-exp.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(\color{blue}{e^{0 - im}} - e^{im}\right) \]

      5. lift-exp.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(e^{0 - im} - \color{blue}{e^{im}}\right) \]

      6. *-commutative N/A

         \[\leadsto \color{blue}{\left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2}} \]

      7. sub0-neg N/A

         \[\leadsto \left(e^{\color{blue}{\mathsf{neg}\left(im\right)}} - e^{im}\right) \cdot \frac{1}{2} \]

      8. flip-- N/A

         \[\leadsto \color{blue}{\frac{e^{\mathsf{neg}\left(im\right)} \cdot e^{\mathsf{neg}\left(im\right)} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}}} \cdot \frac{1}{2} \]

      9. lift-neg.f64 N/A

         \[\leadsto \frac{e^{\color{blue}{-im}} \cdot e^{\mathsf{neg}\left(im\right)} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      10. lift-neg.f64 N/A

          \[\leadsto \frac{e^{-im} \cdot e^{\color{blue}{-im}} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      11. exp-lft-sqr N/A

          \[\leadsto \frac{\color{blue}{e^{\left(-im\right) \cdot 2}} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      12. lift-neg.f64 N/A

          \[\leadsto \frac{e^{\color{blue}{\left(\mathsf{neg}\left(im\right)\right)} \cdot 2} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      13. exp-sum N/A

          \[\leadsto \frac{e^{\left(\mathsf{neg}\left(im\right)\right) \cdot 2} - \color{blue}{e^{im + im}}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

   6. Applied rewrites 99.9%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   if -1e5 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < 0.0

   1. Initial program 8.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in im around 0

      \[\leadsto \color{blue}{-1 \cdot \left(im \cdot \cos re\right)} \]
   3. Step-by-step derivation

      1. associate-*r* N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      3. mul-1-neg N/A

         \[\leadsto \left(\mathsf{neg}\left(im\right)\right) \cdot \cos \color{blue}{re} \]

      4. lower-neg.f64 N/A

         \[\leadsto \left(-im\right) \cdot \cos \color{blue}{re} \]

      5. lift-cos.f64 98.9

         \[\leadsto \left(-im\right) \cdot \cos re \]

   4. Applied rewrites 98.9%

      \[\leadsto \color{blue}{\left(-im\right) \cdot \cos re} \]

   if 0.0 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im)))

   1. Initial program 98.1%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\left(\frac{1}{2} + \frac{-1}{4} \cdot {re}^{2}\right)} \cdot \left(e^{0 - im} - e^{im}\right) \]
   3. Step-by-step derivation

      1. +-commutative N/A

         \[\leadsto \left(\frac{-1}{4} \cdot {re}^{2} + \color{blue}{\frac{1}{2}}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      2. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot \frac{-1}{4} + \frac{1}{2}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      3. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left({re}^{2}, \color{blue}{\frac{-1}{4}}, \frac{1}{2}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      4. unpow2 N/A

         \[\leadsto \mathsf{fma}\left(re \cdot re, \frac{-1}{4}, \frac{1}{2}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      5. lower-*.f64 94.0

         \[\leadsto \mathsf{fma}\left(re \cdot re, -0.25, 0.5\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   4. Applied rewrites 94.0%

      \[\leadsto \color{blue}{\mathsf{fma}\left(re \cdot re, -0.25, 0.5\right)} \cdot \left(e^{0 - im} - e^{im}\right) \]

   5. Taylor expanded in im around 0

      \[\leadsto \mathsf{fma}\left(re \cdot re, \frac{-1}{4}, \frac{1}{2}\right) \cdot \left(\color{blue}{1} - e^{im}\right) \]
   6. Step-by-step derivation

      1. sub0-neg 94.0

         \[\leadsto \mathsf{fma}\left(re \cdot re, -0.25, 0.5\right) \cdot \left(1 - e^{im}\right) \]

   7. Applied rewrites 94.0%

      \[\leadsto \mathsf{fma}\left(re \cdot re, -0.25, 0.5\right) \cdot \left(\color{blue}{1} - e^{im}\right) \]

   8. Taylor expanded in re around inf

      \[\leadsto \left(\frac{-1}{4} \cdot \color{blue}{{re}^{2}}\right) \cdot \left(1 - e^{im}\right) \]
   9. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot \frac{-1}{4}\right) \cdot \left(1 - e^{im}\right) \]

      2. lower-*.f64 N/A

         \[\leadsto \left({re}^{2} \cdot \frac{-1}{4}\right) \cdot \left(1 - e^{im}\right) \]

      3. pow2 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot \frac{-1}{4}\right) \cdot \left(1 - e^{im}\right) \]

      4. lift-*.f64 94.0

         \[\leadsto \left(\left(re \cdot re\right) \cdot -0.25\right) \cdot \left(1 - e^{im}\right) \]

   10. Applied rewrites 94.0%

       \[\leadsto \left(\left(re \cdot re\right) \cdot \color{blue}{-0.25}\right) \cdot \left(1 - e^{im}\right) \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 3: 77.2% accurate, 0.7× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ im\_s \cdot \begin{array}{l} \mathbf{if}\;\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right) \leq 0:\\ \;\;\;\;\left(-2 \cdot \sinh im\_m\right) \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;\left(\left(re \cdot re\right) \cdot -0.25\right) \cdot \left(1 - e^{im\_m}\right)\\ \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (*
  im_s
  (if (<= (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m))) 0.0)
    (* (* -2.0 (sinh im_m)) 0.5)
    (* (* (* re re) -0.25) (- 1.0 (exp im_m))))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double tmp;
	if (((0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m))) <= 0.0) {
		tmp = (-2.0 * sinh(im_m)) * 0.5;
	} else {
		tmp = ((re * re) * -0.25) * (1.0 - exp(im_m));
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: tmp
    if (((0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))) <= 0.0d0) then
        tmp = ((-2.0d0) * sinh(im_m)) * 0.5d0
    else
        tmp = ((re * re) * (-0.25d0)) * (1.0d0 - exp(im_m))
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double tmp;
	if (((0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m))) <= 0.0) {
		tmp = (-2.0 * Math.sinh(im_m)) * 0.5;
	} else {
		tmp = ((re * re) * -0.25) * (1.0 - Math.exp(im_m));
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	tmp = 0
	if ((0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))) <= 0.0:
		tmp = (-2.0 * math.sinh(im_m)) * 0.5
	else:
		tmp = ((re * re) * -0.25) * (1.0 - math.exp(im_m))
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	tmp = 0.0
	if (Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m))) <= 0.0)
		tmp = Float64(Float64(-2.0 * sinh(im_m)) * 0.5);
	else
		tmp = Float64(Float64(Float64(re * re) * -0.25) * Float64(1.0 - exp(im_m)));
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	tmp = 0.0;
	if (((0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m))) <= 0.0)
		tmp = (-2.0 * sinh(im_m)) * 0.5;
	else
		tmp = ((re * re) * -0.25) * (1.0 - exp(im_m));
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := N[(im$95$s * If[LessEqual[N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], 0.0], N[(N[(-2.0 * N[Sinh[im$95$m], $MachinePrecision]), $MachinePrecision] * 0.5), $MachinePrecision], N[(N[(N[(re * re), $MachinePrecision] * -0.25), $MachinePrecision] * N[(1.0 - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]]), $MachinePrecision]

Derivation

1. Split input into 2 regimes

2. if (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < 0.0

   1. Initial program 46.4%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2}} \cdot \left(e^{0 - im} - e^{im}\right) \]
   3. Step-by-step derivation

   4. Applied rewrites 46.0%

      \[\leadsto \color{blue}{0.5} \cdot \left(e^{0 - im} - e^{im}\right) \]
   5. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{0 - im} - e^{im}\right)} \]

      2. lift--.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \color{blue}{\left(e^{0 - im} - e^{im}\right)} \]

      3. lift--.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(e^{\color{blue}{0 - im}} - e^{im}\right) \]

      4. lift-exp.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(\color{blue}{e^{0 - im}} - e^{im}\right) \]

      5. lift-exp.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(e^{0 - im} - \color{blue}{e^{im}}\right) \]

      6. *-commutative N/A

         \[\leadsto \color{blue}{\left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2}} \]

      7. sub0-neg N/A

         \[\leadsto \left(e^{\color{blue}{\mathsf{neg}\left(im\right)}} - e^{im}\right) \cdot \frac{1}{2} \]

      8. flip-- N/A

         \[\leadsto \color{blue}{\frac{e^{\mathsf{neg}\left(im\right)} \cdot e^{\mathsf{neg}\left(im\right)} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}}} \cdot \frac{1}{2} \]

      9. lift-neg.f64 N/A

         \[\leadsto \frac{e^{\color{blue}{-im}} \cdot e^{\mathsf{neg}\left(im\right)} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      10. lift-neg.f64 N/A

          \[\leadsto \frac{e^{-im} \cdot e^{\color{blue}{-im}} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      11. exp-lft-sqr N/A

          \[\leadsto \frac{\color{blue}{e^{\left(-im\right) \cdot 2}} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      12. lift-neg.f64 N/A

          \[\leadsto \frac{e^{\color{blue}{\left(\mathsf{neg}\left(im\right)\right)} \cdot 2} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      13. exp-sum N/A

          \[\leadsto \frac{e^{\left(\mathsf{neg}\left(im\right)\right) \cdot 2} - \color{blue}{e^{im + im}}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

   6. Applied rewrites 74.8%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   if 0.0 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im)))

   1. Initial program 98.1%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\left(\frac{1}{2} + \frac{-1}{4} \cdot {re}^{2}\right)} \cdot \left(e^{0 - im} - e^{im}\right) \]
   3. Step-by-step derivation

      1. +-commutative N/A

         \[\leadsto \left(\frac{-1}{4} \cdot {re}^{2} + \color{blue}{\frac{1}{2}}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      2. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot \frac{-1}{4} + \frac{1}{2}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      3. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left({re}^{2}, \color{blue}{\frac{-1}{4}}, \frac{1}{2}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      4. unpow2 N/A

         \[\leadsto \mathsf{fma}\left(re \cdot re, \frac{-1}{4}, \frac{1}{2}\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

      5. lower-*.f64 94.0

         \[\leadsto \mathsf{fma}\left(re \cdot re, -0.25, 0.5\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   4. Applied rewrites 94.0%

      \[\leadsto \color{blue}{\mathsf{fma}\left(re \cdot re, -0.25, 0.5\right)} \cdot \left(e^{0 - im} - e^{im}\right) \]

   5. Taylor expanded in im around 0

      \[\leadsto \mathsf{fma}\left(re \cdot re, \frac{-1}{4}, \frac{1}{2}\right) \cdot \left(\color{blue}{1} - e^{im}\right) \]
   6. Step-by-step derivation

      1. sub0-neg 94.0

         \[\leadsto \mathsf{fma}\left(re \cdot re, -0.25, 0.5\right) \cdot \left(1 - e^{im}\right) \]

   7. Applied rewrites 94.0%

      \[\leadsto \mathsf{fma}\left(re \cdot re, -0.25, 0.5\right) \cdot \left(\color{blue}{1} - e^{im}\right) \]

   8. Taylor expanded in re around inf

      \[\leadsto \left(\frac{-1}{4} \cdot \color{blue}{{re}^{2}}\right) \cdot \left(1 - e^{im}\right) \]
   9. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot \frac{-1}{4}\right) \cdot \left(1 - e^{im}\right) \]

      2. lower-*.f64 N/A

         \[\leadsto \left({re}^{2} \cdot \frac{-1}{4}\right) \cdot \left(1 - e^{im}\right) \]

      3. pow2 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot \frac{-1}{4}\right) \cdot \left(1 - e^{im}\right) \]

      4. lift-*.f64 94.0

         \[\leadsto \left(\left(re \cdot re\right) \cdot -0.25\right) \cdot \left(1 - e^{im}\right) \]

   10. Applied rewrites 94.0%

       \[\leadsto \left(\left(re \cdot re\right) \cdot \color{blue}{-0.25}\right) \cdot \left(1 - e^{im}\right) \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 4: 75.3% accurate, 0.7× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ im\_s \cdot \begin{array}{l} \mathbf{if}\;\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right) \leq 0:\\ \;\;\;\;\left(-2 \cdot \sinh im\_m\right) \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;\left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\_m\right) \cdot 0.5\\ \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (*
  im_s
  (if (<= (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m))) 0.0)
    (* (* -2.0 (sinh im_m)) 0.5)
    (* (* (sqrt (* (* (* re re) re) re)) im_m) 0.5))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double tmp;
	if (((0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m))) <= 0.0) {
		tmp = (-2.0 * sinh(im_m)) * 0.5;
	} else {
		tmp = (sqrt((((re * re) * re) * re)) * im_m) * 0.5;
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: tmp
    if (((0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))) <= 0.0d0) then
        tmp = ((-2.0d0) * sinh(im_m)) * 0.5d0
    else
        tmp = (sqrt((((re * re) * re) * re)) * im_m) * 0.5d0
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double tmp;
	if (((0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m))) <= 0.0) {
		tmp = (-2.0 * Math.sinh(im_m)) * 0.5;
	} else {
		tmp = (Math.sqrt((((re * re) * re) * re)) * im_m) * 0.5;
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	tmp = 0
	if ((0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))) <= 0.0:
		tmp = (-2.0 * math.sinh(im_m)) * 0.5
	else:
		tmp = (math.sqrt((((re * re) * re) * re)) * im_m) * 0.5
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	tmp = 0.0
	if (Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m))) <= 0.0)
		tmp = Float64(Float64(-2.0 * sinh(im_m)) * 0.5);
	else
		tmp = Float64(Float64(sqrt(Float64(Float64(Float64(re * re) * re) * re)) * im_m) * 0.5);
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	tmp = 0.0;
	if (((0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m))) <= 0.0)
		tmp = (-2.0 * sinh(im_m)) * 0.5;
	else
		tmp = (sqrt((((re * re) * re) * re)) * im_m) * 0.5;
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := N[(im$95$s * If[LessEqual[N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], 0.0], N[(N[(-2.0 * N[Sinh[im$95$m], $MachinePrecision]), $MachinePrecision] * 0.5), $MachinePrecision], N[(N[(N[Sqrt[N[(N[(N[(re * re), $MachinePrecision] * re), $MachinePrecision] * re), $MachinePrecision]], $MachinePrecision] * im$95$m), $MachinePrecision] * 0.5), $MachinePrecision]]), $MachinePrecision]

Derivation

1. Split input into 2 regimes

2. if (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < 0.0

   1. Initial program 46.4%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2}} \cdot \left(e^{0 - im} - e^{im}\right) \]
   3. Step-by-step derivation

   4. Applied rewrites 46.0%

      \[\leadsto \color{blue}{0.5} \cdot \left(e^{0 - im} - e^{im}\right) \]
   5. Step-by-step derivation

      1. lift-*.f64 N/A

         \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{0 - im} - e^{im}\right)} \]

      2. lift--.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \color{blue}{\left(e^{0 - im} - e^{im}\right)} \]

      3. lift--.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(e^{\color{blue}{0 - im}} - e^{im}\right) \]

      4. lift-exp.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(\color{blue}{e^{0 - im}} - e^{im}\right) \]

      5. lift-exp.f64 N/A

         \[\leadsto \frac{1}{2} \cdot \left(e^{0 - im} - \color{blue}{e^{im}}\right) \]

      6. *-commutative N/A

         \[\leadsto \color{blue}{\left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2}} \]

      7. sub0-neg N/A

         \[\leadsto \left(e^{\color{blue}{\mathsf{neg}\left(im\right)}} - e^{im}\right) \cdot \frac{1}{2} \]

      8. flip-- N/A

         \[\leadsto \color{blue}{\frac{e^{\mathsf{neg}\left(im\right)} \cdot e^{\mathsf{neg}\left(im\right)} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}}} \cdot \frac{1}{2} \]

      9. lift-neg.f64 N/A

         \[\leadsto \frac{e^{\color{blue}{-im}} \cdot e^{\mathsf{neg}\left(im\right)} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      10. lift-neg.f64 N/A

          \[\leadsto \frac{e^{-im} \cdot e^{\color{blue}{-im}} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      11. exp-lft-sqr N/A

          \[\leadsto \frac{\color{blue}{e^{\left(-im\right) \cdot 2}} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      12. lift-neg.f64 N/A

          \[\leadsto \frac{e^{\color{blue}{\left(\mathsf{neg}\left(im\right)\right)} \cdot 2} - e^{im} \cdot e^{im}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

      13. exp-sum N/A

          \[\leadsto \frac{e^{\left(\mathsf{neg}\left(im\right)\right) \cdot 2} - \color{blue}{e^{im + im}}}{e^{\mathsf{neg}\left(im\right)} + e^{im}} \cdot \frac{1}{2} \]

   6. Applied rewrites 74.8%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   if 0.0 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im)))

   1. Initial program 98.1%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in im around 0

      \[\leadsto \color{blue}{-1 \cdot \left(im \cdot \cos re\right)} \]
   3. Step-by-step derivation

      1. associate-*r* N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      3. mul-1-neg N/A

         \[\leadsto \left(\mathsf{neg}\left(im\right)\right) \cdot \cos \color{blue}{re} \]

      4. lower-neg.f64 N/A

         \[\leadsto \left(-im\right) \cdot \cos \color{blue}{re} \]

      5. lift-cos.f64 9.6

         \[\leadsto \left(-im\right) \cdot \cos re \]

   4. Applied rewrites 9.6%

      \[\leadsto \color{blue}{\left(-im\right) \cdot \cos re} \]

   5. Taylor expanded in re around 0

      \[\leadsto -1 \cdot im + \color{blue}{\frac{1}{2} \cdot \left(im \cdot {re}^{2}\right)} \]
   6. Step-by-step derivation

      1. +-commutative N/A

         \[\leadsto \frac{1}{2} \cdot \left(im \cdot {re}^{2}\right) + -1 \cdot \color{blue}{im} \]

      2. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} + -1 \cdot im \]

      3. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left(im \cdot {re}^{2}, \frac{1}{2}, -1 \cdot im\right) \]

      4. *-commutative N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      5. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      6. pow2 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      7. lift-*.f64 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      8. mul-1-neg N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, \mathsf{neg}\left(im\right)\right) \]

      9. lift-neg.f64 74.9

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, 0.5, -im\right) \]

   7. Applied rewrites 74.9%

      \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \color{blue}{0.5}, -im\right) \]

   8. Taylor expanded in re around inf

      \[\leadsto \frac{1}{2} \cdot \left(im \cdot \color{blue}{{re}^{2}}\right) \]
   9. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      3. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot im\right) \cdot \frac{1}{2} \]

      4. pow2 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      5. lift-*.f64 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      6. lift-*.f64 74.9

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]

   10. Applied rewrites 74.9%

       \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]
   11. Step-by-step derivation

       1. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       2. fabs-sqr N/A

          \[\leadsto \left(\left|re \cdot re\right| \cdot im\right) \cdot \frac{1}{2} \]

       3. pow2 N/A

          \[\leadsto \left(\left|{re}^{2}\right| \cdot im\right) \cdot \frac{1}{2} \]

       4. rem-sqrt-square-rev N/A

          \[\leadsto \left(\sqrt{{re}^{2} \cdot {re}^{2}} \cdot im\right) \cdot \frac{1}{2} \]

       5. pow-prod-up N/A

          \[\leadsto \left(\sqrt{{re}^{\left(2 + 2\right)}} \cdot im\right) \cdot \frac{1}{2} \]

       6. metadata-eval N/A

          \[\leadsto \left(\sqrt{{re}^{4}} \cdot im\right) \cdot \frac{1}{2} \]

       7. lower-sqrt.f64 N/A

          \[\leadsto \left(\sqrt{{re}^{4}} \cdot im\right) \cdot \frac{1}{2} \]

       8. metadata-eval N/A

          \[\leadsto \left(\sqrt{{re}^{\left(2 + 2\right)}} \cdot im\right) \cdot \frac{1}{2} \]

       9. pow-prod-up N/A

          \[\leadsto \left(\sqrt{{re}^{2} \cdot {re}^{2}} \cdot im\right) \cdot \frac{1}{2} \]

       10. pow2 N/A

           \[\leadsto \left(\sqrt{{re}^{2} \cdot \left(re \cdot re\right)} \cdot im\right) \cdot \frac{1}{2} \]

       11. associate-*r* N/A

           \[\leadsto \left(\sqrt{\left({re}^{2} \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       12. pow-plus N/A

           \[\leadsto \left(\sqrt{{re}^{\left(2 + 1\right)} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       13. metadata-eval N/A

           \[\leadsto \left(\sqrt{{re}^{3} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       14. lower-*.f64 N/A

           \[\leadsto \left(\sqrt{{re}^{3} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       15. metadata-eval N/A

           \[\leadsto \left(\sqrt{{re}^{\left(2 + 1\right)} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       16. pow-plus N/A

           \[\leadsto \left(\sqrt{\left({re}^{2} \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       17. lower-*.f64 N/A

           \[\leadsto \left(\sqrt{\left({re}^{2} \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       18. pow2 N/A

           \[\leadsto \left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       19. lift-*.f64 79.2

           \[\leadsto \left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\right) \cdot 0.5 \]

   12. Applied rewrites 79.2%

       \[\leadsto \left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\right) \cdot 0.5 \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 5: 75.0% accurate, 0.4× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ \begin{array}{l} t_0 := \left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right)\\ im\_s \cdot \begin{array}{l} \mathbf{if}\;t\_0 \leq -100000:\\ \;\;\;\;0.5 \cdot \left(1 - e^{im\_m}\right)\\ \mathbf{elif}\;t\_0 \leq 0:\\ \;\;\;\;-im\_m\\ \mathbf{else}:\\ \;\;\;\;\left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\_m\right) \cdot 0.5\\ \end{array} \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (let* ((t_0 (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m)))))
   (*
    im_s
    (if (<= t_0 -100000.0)
      (* 0.5 (- 1.0 (exp im_m)))
      (if (<= t_0 0.0)
        (- im_m)
        (* (* (sqrt (* (* (* re re) re) re)) im_m) 0.5))))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = 0.5 * (1.0 - exp(im_m));
	} else if (t_0 <= 0.0) {
		tmp = -im_m;
	} else {
		tmp = (sqrt((((re * re) * re) * re)) * im_m) * 0.5;
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: t_0
    real(8) :: tmp
    t_0 = (0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))
    if (t_0 <= (-100000.0d0)) then
        tmp = 0.5d0 * (1.0d0 - exp(im_m))
    else if (t_0 <= 0.0d0) then
        tmp = -im_m
    else
        tmp = (sqrt((((re * re) * re) * re)) * im_m) * 0.5d0
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = 0.5 * (1.0 - Math.exp(im_m));
	} else if (t_0 <= 0.0) {
		tmp = -im_m;
	} else {
		tmp = (Math.sqrt((((re * re) * re) * re)) * im_m) * 0.5;
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	t_0 = (0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))
	tmp = 0
	if t_0 <= -100000.0:
		tmp = 0.5 * (1.0 - math.exp(im_m))
	elif t_0 <= 0.0:
		tmp = -im_m
	else:
		tmp = (math.sqrt((((re * re) * re) * re)) * im_m) * 0.5
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	t_0 = Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m)))
	tmp = 0.0
	if (t_0 <= -100000.0)
		tmp = Float64(0.5 * Float64(1.0 - exp(im_m)));
	elseif (t_0 <= 0.0)
		tmp = Float64(-im_m);
	else
		tmp = Float64(Float64(sqrt(Float64(Float64(Float64(re * re) * re) * re)) * im_m) * 0.5);
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	tmp = 0.0;
	if (t_0 <= -100000.0)
		tmp = 0.5 * (1.0 - exp(im_m));
	elseif (t_0 <= 0.0)
		tmp = -im_m;
	else
		tmp = (sqrt((((re * re) * re) * re)) * im_m) * 0.5;
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := Block[{t$95$0 = N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]}, N[(im$95$s * If[LessEqual[t$95$0, -100000.0], N[(0.5 * N[(1.0 - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$0, 0.0], (-im$95$m), N[(N[(N[Sqrt[N[(N[(N[(re * re), $MachinePrecision] * re), $MachinePrecision] * re), $MachinePrecision]], $MachinePrecision] * im$95$m), $MachinePrecision] * 0.5), $MachinePrecision]]]), $MachinePrecision]]

Derivation

1. Split input into 3 regimes

2. if (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < -1e5

   1. Initial program 100.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2}} \cdot \left(e^{0 - im} - e^{im}\right) \]
   3. Step-by-step derivation

   4. Applied rewrites 99.9%

      \[\leadsto \color{blue}{0.5} \cdot \left(e^{0 - im} - e^{im}\right) \]

   5. Taylor expanded in im around 0

      \[\leadsto \frac{1}{2} \cdot \left(\color{blue}{1} - e^{im}\right) \]
   6. Step-by-step derivation

      1. sub0-neg 99.9

         \[\leadsto 0.5 \cdot \left(1 - e^{im}\right) \]

   7. Applied rewrites 99.9%

      \[\leadsto 0.5 \cdot \left(\color{blue}{1} - e^{im}\right) \]

   if -1e5 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < 0.0

   1. Initial program 8.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      3. sub0-neg N/A

         \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

      4. sub-negate N/A

         \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

      5. lower-neg.f64 N/A

         \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

      6. sub0-neg N/A

         \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

      7. sinh-undef N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      8. lower-*.f64 N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      9. lower-sinh.f64 56.7

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

   4. Applied rewrites 56.7%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   5. Taylor expanded in im around 0

      \[\leadsto -1 \cdot \color{blue}{im} \]
   6. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. lift-neg.f64 56.1

         \[\leadsto -im \]

   7. Applied rewrites 56.1%

      \[\leadsto -im \]

   if 0.0 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im)))

   1. Initial program 98.1%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in im around 0

      \[\leadsto \color{blue}{-1 \cdot \left(im \cdot \cos re\right)} \]
   3. Step-by-step derivation

      1. associate-*r* N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      3. mul-1-neg N/A

         \[\leadsto \left(\mathsf{neg}\left(im\right)\right) \cdot \cos \color{blue}{re} \]

      4. lower-neg.f64 N/A

         \[\leadsto \left(-im\right) \cdot \cos \color{blue}{re} \]

      5. lift-cos.f64 9.6

         \[\leadsto \left(-im\right) \cdot \cos re \]

   4. Applied rewrites 9.6%

      \[\leadsto \color{blue}{\left(-im\right) \cdot \cos re} \]

   5. Taylor expanded in re around 0

      \[\leadsto -1 \cdot im + \color{blue}{\frac{1}{2} \cdot \left(im \cdot {re}^{2}\right)} \]
   6. Step-by-step derivation

      1. +-commutative N/A

         \[\leadsto \frac{1}{2} \cdot \left(im \cdot {re}^{2}\right) + -1 \cdot \color{blue}{im} \]

      2. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} + -1 \cdot im \]

      3. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left(im \cdot {re}^{2}, \frac{1}{2}, -1 \cdot im\right) \]

      4. *-commutative N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      5. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      6. pow2 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      7. lift-*.f64 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      8. mul-1-neg N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, \mathsf{neg}\left(im\right)\right) \]

      9. lift-neg.f64 74.9

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, 0.5, -im\right) \]

   7. Applied rewrites 74.9%

      \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \color{blue}{0.5}, -im\right) \]

   8. Taylor expanded in re around inf

      \[\leadsto \frac{1}{2} \cdot \left(im \cdot \color{blue}{{re}^{2}}\right) \]
   9. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      3. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot im\right) \cdot \frac{1}{2} \]

      4. pow2 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      5. lift-*.f64 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      6. lift-*.f64 74.9

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]

   10. Applied rewrites 74.9%

       \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]
   11. Step-by-step derivation

       1. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       2. fabs-sqr N/A

          \[\leadsto \left(\left|re \cdot re\right| \cdot im\right) \cdot \frac{1}{2} \]

       3. pow2 N/A

          \[\leadsto \left(\left|{re}^{2}\right| \cdot im\right) \cdot \frac{1}{2} \]

       4. rem-sqrt-square-rev N/A

          \[\leadsto \left(\sqrt{{re}^{2} \cdot {re}^{2}} \cdot im\right) \cdot \frac{1}{2} \]

       5. pow-prod-up N/A

          \[\leadsto \left(\sqrt{{re}^{\left(2 + 2\right)}} \cdot im\right) \cdot \frac{1}{2} \]

       6. metadata-eval N/A

          \[\leadsto \left(\sqrt{{re}^{4}} \cdot im\right) \cdot \frac{1}{2} \]

       7. lower-sqrt.f64 N/A

          \[\leadsto \left(\sqrt{{re}^{4}} \cdot im\right) \cdot \frac{1}{2} \]

       8. metadata-eval N/A

          \[\leadsto \left(\sqrt{{re}^{\left(2 + 2\right)}} \cdot im\right) \cdot \frac{1}{2} \]

       9. pow-prod-up N/A

          \[\leadsto \left(\sqrt{{re}^{2} \cdot {re}^{2}} \cdot im\right) \cdot \frac{1}{2} \]

       10. pow2 N/A

           \[\leadsto \left(\sqrt{{re}^{2} \cdot \left(re \cdot re\right)} \cdot im\right) \cdot \frac{1}{2} \]

       11. associate-*r* N/A

           \[\leadsto \left(\sqrt{\left({re}^{2} \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       12. pow-plus N/A

           \[\leadsto \left(\sqrt{{re}^{\left(2 + 1\right)} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       13. metadata-eval N/A

           \[\leadsto \left(\sqrt{{re}^{3} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       14. lower-*.f64 N/A

           \[\leadsto \left(\sqrt{{re}^{3} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       15. metadata-eval N/A

           \[\leadsto \left(\sqrt{{re}^{\left(2 + 1\right)} \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       16. pow-plus N/A

           \[\leadsto \left(\sqrt{\left({re}^{2} \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       17. lower-*.f64 N/A

           \[\leadsto \left(\sqrt{\left({re}^{2} \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       18. pow2 N/A

           \[\leadsto \left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\right) \cdot \frac{1}{2} \]

       19. lift-*.f64 79.2

           \[\leadsto \left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\right) \cdot 0.5 \]

   12. Applied rewrites 79.2%

       \[\leadsto \left(\sqrt{\left(\left(re \cdot re\right) \cdot re\right) \cdot re} \cdot im\right) \cdot 0.5 \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 6: 74.4% accurate, 0.4× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ \begin{array}{l} t_0 := \left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right)\\ im\_s \cdot \begin{array}{l} \mathbf{if}\;t\_0 \leq -100000:\\ \;\;\;\;0.5 \cdot \left(1 - e^{im\_m}\right)\\ \mathbf{elif}\;t\_0 \leq 0:\\ \;\;\;\;-im\_m\\ \mathbf{else}:\\ \;\;\;\;\left(re \cdot \left(re \cdot im\_m\right)\right) \cdot 0.5\\ \end{array} \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (let* ((t_0 (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m)))))
   (*
    im_s
    (if (<= t_0 -100000.0)
      (* 0.5 (- 1.0 (exp im_m)))
      (if (<= t_0 0.0) (- im_m) (* (* re (* re im_m)) 0.5))))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = 0.5 * (1.0 - exp(im_m));
	} else if (t_0 <= 0.0) {
		tmp = -im_m;
	} else {
		tmp = (re * (re * im_m)) * 0.5;
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: t_0
    real(8) :: tmp
    t_0 = (0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))
    if (t_0 <= (-100000.0d0)) then
        tmp = 0.5d0 * (1.0d0 - exp(im_m))
    else if (t_0 <= 0.0d0) then
        tmp = -im_m
    else
        tmp = (re * (re * im_m)) * 0.5d0
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = 0.5 * (1.0 - Math.exp(im_m));
	} else if (t_0 <= 0.0) {
		tmp = -im_m;
	} else {
		tmp = (re * (re * im_m)) * 0.5;
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	t_0 = (0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))
	tmp = 0
	if t_0 <= -100000.0:
		tmp = 0.5 * (1.0 - math.exp(im_m))
	elif t_0 <= 0.0:
		tmp = -im_m
	else:
		tmp = (re * (re * im_m)) * 0.5
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	t_0 = Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m)))
	tmp = 0.0
	if (t_0 <= -100000.0)
		tmp = Float64(0.5 * Float64(1.0 - exp(im_m)));
	elseif (t_0 <= 0.0)
		tmp = Float64(-im_m);
	else
		tmp = Float64(Float64(re * Float64(re * im_m)) * 0.5);
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	tmp = 0.0;
	if (t_0 <= -100000.0)
		tmp = 0.5 * (1.0 - exp(im_m));
	elseif (t_0 <= 0.0)
		tmp = -im_m;
	else
		tmp = (re * (re * im_m)) * 0.5;
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := Block[{t$95$0 = N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]}, N[(im$95$s * If[LessEqual[t$95$0, -100000.0], N[(0.5 * N[(1.0 - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], If[LessEqual[t$95$0, 0.0], (-im$95$m), N[(N[(re * N[(re * im$95$m), $MachinePrecision]), $MachinePrecision] * 0.5), $MachinePrecision]]]), $MachinePrecision]]

Derivation

1. Split input into 3 regimes

2. if (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < -1e5

   1. Initial program 100.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2}} \cdot \left(e^{0 - im} - e^{im}\right) \]
   3. Step-by-step derivation

   4. Applied rewrites 99.9%

      \[\leadsto \color{blue}{0.5} \cdot \left(e^{0 - im} - e^{im}\right) \]

   5. Taylor expanded in im around 0

      \[\leadsto \frac{1}{2} \cdot \left(\color{blue}{1} - e^{im}\right) \]
   6. Step-by-step derivation

      1. sub0-neg 99.9

         \[\leadsto 0.5 \cdot \left(1 - e^{im}\right) \]

   7. Applied rewrites 99.9%

      \[\leadsto 0.5 \cdot \left(\color{blue}{1} - e^{im}\right) \]

   if -1e5 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < 0.0

   1. Initial program 8.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      3. sub0-neg N/A

         \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

      4. sub-negate N/A

         \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

      5. lower-neg.f64 N/A

         \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

      6. sub0-neg N/A

         \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

      7. sinh-undef N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      8. lower-*.f64 N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      9. lower-sinh.f64 56.7

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

   4. Applied rewrites 56.7%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   5. Taylor expanded in im around 0

      \[\leadsto -1 \cdot \color{blue}{im} \]
   6. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. lift-neg.f64 56.1

         \[\leadsto -im \]

   7. Applied rewrites 56.1%

      \[\leadsto -im \]

   if 0.0 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im)))

   1. Initial program 98.1%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in im around 0

      \[\leadsto \color{blue}{-1 \cdot \left(im \cdot \cos re\right)} \]
   3. Step-by-step derivation

      1. associate-*r* N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      3. mul-1-neg N/A

         \[\leadsto \left(\mathsf{neg}\left(im\right)\right) \cdot \cos \color{blue}{re} \]

      4. lower-neg.f64 N/A

         \[\leadsto \left(-im\right) \cdot \cos \color{blue}{re} \]

      5. lift-cos.f64 9.6

         \[\leadsto \left(-im\right) \cdot \cos re \]

   4. Applied rewrites 9.6%

      \[\leadsto \color{blue}{\left(-im\right) \cdot \cos re} \]

   5. Taylor expanded in re around 0

      \[\leadsto -1 \cdot im + \color{blue}{\frac{1}{2} \cdot \left(im \cdot {re}^{2}\right)} \]
   6. Step-by-step derivation

      1. +-commutative N/A

         \[\leadsto \frac{1}{2} \cdot \left(im \cdot {re}^{2}\right) + -1 \cdot \color{blue}{im} \]

      2. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} + -1 \cdot im \]

      3. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left(im \cdot {re}^{2}, \frac{1}{2}, -1 \cdot im\right) \]

      4. *-commutative N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      5. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      6. pow2 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      7. lift-*.f64 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      8. mul-1-neg N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, \mathsf{neg}\left(im\right)\right) \]

      9. lift-neg.f64 74.9

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, 0.5, -im\right) \]

   7. Applied rewrites 74.9%

      \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \color{blue}{0.5}, -im\right) \]

   8. Taylor expanded in re around inf

      \[\leadsto \frac{1}{2} \cdot \left(im \cdot \color{blue}{{re}^{2}}\right) \]
   9. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      3. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot im\right) \cdot \frac{1}{2} \]

      4. pow2 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      5. lift-*.f64 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      6. lift-*.f64 74.9

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]

   10. Applied rewrites 74.9%

       \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]
   11. Step-by-step derivation

       1. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       2. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       3. associate-*l* N/A

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot \frac{1}{2} \]

       4. lower-*.f64 N/A

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot \frac{1}{2} \]

       5. lower-*.f64 74.9

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot 0.5 \]

   12. Applied rewrites 74.9%

       \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot 0.5 \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 7: 63.4% accurate, 1.3× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ im\_s \cdot \begin{array}{l} \mathbf{if}\;0.5 \cdot \cos re \leq -0.005:\\ \;\;\;\;\left(re \cdot \left(re \cdot im\_m\right)\right) \cdot 0.5\\ \mathbf{else}:\\ \;\;\;\;\mathsf{fma}\left(im\_m \cdot im\_m, -0.16666666666666666, -1\right) \cdot im\_m\\ \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (*
  im_s
  (if (<= (* 0.5 (cos re)) -0.005)
    (* (* re (* re im_m)) 0.5)
    (* (fma (* im_m im_m) -0.16666666666666666 -1.0) im_m))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double tmp;
	if ((0.5 * cos(re)) <= -0.005) {
		tmp = (re * (re * im_m)) * 0.5;
	} else {
		tmp = fma((im_m * im_m), -0.16666666666666666, -1.0) * im_m;
	}
	return im_s * tmp;
}

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	tmp = 0.0
	if (Float64(0.5 * cos(re)) <= -0.005)
		tmp = Float64(Float64(re * Float64(re * im_m)) * 0.5);
	else
		tmp = Float64(fma(Float64(im_m * im_m), -0.16666666666666666, -1.0) * im_m);
	end
	return Float64(im_s * tmp)
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := N[(im$95$s * If[LessEqual[N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision], -0.005], N[(N[(re * N[(re * im$95$m), $MachinePrecision]), $MachinePrecision] * 0.5), $MachinePrecision], N[(N[(N[(im$95$m * im$95$m), $MachinePrecision] * -0.16666666666666666 + -1.0), $MachinePrecision] * im$95$m), $MachinePrecision]]), $MachinePrecision]

Derivation

1. Split input into 2 regimes

2. if (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) < -0.0050000000000000001

   1. Initial program 53.4%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in im around 0

      \[\leadsto \color{blue}{-1 \cdot \left(im \cdot \cos re\right)} \]
   3. Step-by-step derivation

      1. associate-*r* N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      3. mul-1-neg N/A

         \[\leadsto \left(\mathsf{neg}\left(im\right)\right) \cdot \cos \color{blue}{re} \]

      4. lower-neg.f64 N/A

         \[\leadsto \left(-im\right) \cdot \cos \color{blue}{re} \]

      5. lift-cos.f64 53.0

         \[\leadsto \left(-im\right) \cdot \cos re \]

   4. Applied rewrites 53.0%

      \[\leadsto \color{blue}{\left(-im\right) \cdot \cos re} \]

   5. Taylor expanded in re around 0

      \[\leadsto -1 \cdot im + \color{blue}{\frac{1}{2} \cdot \left(im \cdot {re}^{2}\right)} \]
   6. Step-by-step derivation

      1. +-commutative N/A

         \[\leadsto \frac{1}{2} \cdot \left(im \cdot {re}^{2}\right) + -1 \cdot \color{blue}{im} \]

      2. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} + -1 \cdot im \]

      3. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left(im \cdot {re}^{2}, \frac{1}{2}, -1 \cdot im\right) \]

      4. *-commutative N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      5. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      6. pow2 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      7. lift-*.f64 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      8. mul-1-neg N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, \mathsf{neg}\left(im\right)\right) \]

      9. lift-neg.f64 40.6

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, 0.5, -im\right) \]

   7. Applied rewrites 40.6%

      \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \color{blue}{0.5}, -im\right) \]

   8. Taylor expanded in re around inf

      \[\leadsto \frac{1}{2} \cdot \left(im \cdot \color{blue}{{re}^{2}}\right) \]
   9. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      3. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot im\right) \cdot \frac{1}{2} \]

      4. pow2 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      5. lift-*.f64 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      6. lift-*.f64 40.6

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]

   10. Applied rewrites 40.6%

       \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]
   11. Step-by-step derivation

       1. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       2. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       3. associate-*l* N/A

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot \frac{1}{2} \]

       4. lower-*.f64 N/A

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot \frac{1}{2} \]

       5. lower-*.f64 40.7

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot 0.5 \]

   12. Applied rewrites 40.7%

       \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot 0.5 \]

   if -0.0050000000000000001 < (*.f64 #s(literal 1/2 binary64) (cos.f64 re))

   1. Initial program 52.8%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      3. sub0-neg N/A

         \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

      4. sub-negate N/A

         \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

      5. lower-neg.f64 N/A

         \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

      6. sub0-neg N/A

         \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

      7. sinh-undef N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      8. lower-*.f64 N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      9. lower-sinh.f64 85.7

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

   4. Applied rewrites 85.7%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   5. Taylor expanded in im around 0

      \[\leadsto im \cdot \color{blue}{\left(\frac{-1}{6} \cdot {im}^{2} - 1\right)} \]
   6. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(\frac{-1}{6} \cdot {im}^{2} - 1\right) \cdot im \]

      2. lower-*.f64 N/A

         \[\leadsto \left(\frac{-1}{6} \cdot {im}^{2} - 1\right) \cdot im \]

      3. sub-flip N/A

         \[\leadsto \left(\frac{-1}{6} \cdot {im}^{2} + \left(\mathsf{neg}\left(1\right)\right)\right) \cdot im \]

      4. *-commutative N/A

         \[\leadsto \left({im}^{2} \cdot \frac{-1}{6} + \left(\mathsf{neg}\left(1\right)\right)\right) \cdot im \]

      5. metadata-eval N/A

         \[\leadsto \left({im}^{2} \cdot \frac{-1}{6} + -1\right) \cdot im \]

      6. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left({im}^{2}, \frac{-1}{6}, -1\right) \cdot im \]

      7. unpow2 N/A

         \[\leadsto \mathsf{fma}\left(im \cdot im, \frac{-1}{6}, -1\right) \cdot im \]

      8. lower-*.f64 70.8

         \[\leadsto \mathsf{fma}\left(im \cdot im, -0.16666666666666666, -1\right) \cdot im \]

   7. Applied rewrites 70.8%

      \[\leadsto \mathsf{fma}\left(im \cdot im, -0.16666666666666666, -1\right) \cdot \color{blue}{im} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 8: 57.5% accurate, 0.4× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ \begin{array}{l} t_0 := \left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right)\\ im\_s \cdot \begin{array}{l} \mathbf{if}\;t\_0 \leq -100000:\\ \;\;\;\;\frac{im\_m \cdot im\_m}{-im\_m}\\ \mathbf{elif}\;t\_0 \leq 0:\\ \;\;\;\;-im\_m\\ \mathbf{else}:\\ \;\;\;\;\left(re \cdot \left(re \cdot im\_m\right)\right) \cdot 0.5\\ \end{array} \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (let* ((t_0 (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m)))))
   (*
    im_s
    (if (<= t_0 -100000.0)
      (/ (* im_m im_m) (- im_m))
      (if (<= t_0 0.0) (- im_m) (* (* re (* re im_m)) 0.5))))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = (im_m * im_m) / -im_m;
	} else if (t_0 <= 0.0) {
		tmp = -im_m;
	} else {
		tmp = (re * (re * im_m)) * 0.5;
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: t_0
    real(8) :: tmp
    t_0 = (0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))
    if (t_0 <= (-100000.0d0)) then
        tmp = (im_m * im_m) / -im_m
    else if (t_0 <= 0.0d0) then
        tmp = -im_m
    else
        tmp = (re * (re * im_m)) * 0.5d0
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double t_0 = (0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m));
	double tmp;
	if (t_0 <= -100000.0) {
		tmp = (im_m * im_m) / -im_m;
	} else if (t_0 <= 0.0) {
		tmp = -im_m;
	} else {
		tmp = (re * (re * im_m)) * 0.5;
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	t_0 = (0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))
	tmp = 0
	if t_0 <= -100000.0:
		tmp = (im_m * im_m) / -im_m
	elif t_0 <= 0.0:
		tmp = -im_m
	else:
		tmp = (re * (re * im_m)) * 0.5
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	t_0 = Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m)))
	tmp = 0.0
	if (t_0 <= -100000.0)
		tmp = Float64(Float64(im_m * im_m) / Float64(-im_m));
	elseif (t_0 <= 0.0)
		tmp = Float64(-im_m);
	else
		tmp = Float64(Float64(re * Float64(re * im_m)) * 0.5);
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	t_0 = (0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m));
	tmp = 0.0;
	if (t_0 <= -100000.0)
		tmp = (im_m * im_m) / -im_m;
	elseif (t_0 <= 0.0)
		tmp = -im_m;
	else
		tmp = (re * (re * im_m)) * 0.5;
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := Block[{t$95$0 = N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision]}, N[(im$95$s * If[LessEqual[t$95$0, -100000.0], N[(N[(im$95$m * im$95$m), $MachinePrecision] / (-im$95$m)), $MachinePrecision], If[LessEqual[t$95$0, 0.0], (-im$95$m), N[(N[(re * N[(re * im$95$m), $MachinePrecision]), $MachinePrecision] * 0.5), $MachinePrecision]]]), $MachinePrecision]]

Derivation

1. Split input into 3 regimes

2. if (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < -1e5

   1. Initial program 100.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      3. sub0-neg N/A

         \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

      4. sub-negate N/A

         \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

      5. lower-neg.f64 N/A

         \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

      6. sub0-neg N/A

         \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

      7. sinh-undef N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      8. lower-*.f64 N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      9. lower-sinh.f64 99.9

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

   4. Applied rewrites 99.9%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   5. Taylor expanded in im around 0

      \[\leadsto -1 \cdot \color{blue}{im} \]
   6. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. lift-neg.f64 5.5

         \[\leadsto -im \]

   7. Applied rewrites 5.5%

      \[\leadsto -im \]
   8. Step-by-step derivation

      1. lift-neg.f64 N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. sub0-neg N/A

         \[\leadsto 0 - im \]

      3. flip-- N/A

         \[\leadsto \frac{0 \cdot 0 - im \cdot im}{0 + \color{blue}{im}} \]

      4. lower-/.f64 N/A

         \[\leadsto \frac{0 \cdot 0 - im \cdot im}{0 + \color{blue}{im}} \]

      5. metadata-eval N/A

         \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

      6. unpow2 N/A

         \[\leadsto \frac{0 - {im}^{2}}{0 + im} \]

      7. lower--.f64 N/A

         \[\leadsto \frac{0 - {im}^{2}}{0 + im} \]

      8. unpow2 N/A

         \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

      9. lower-*.f64 N/A

         \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

      10. lower-+.f64 53.6

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

   9. Applied rewrites 53.6%

      \[\leadsto \frac{0 - im \cdot im}{0 + \color{blue}{im}} \]
   10. Step-by-step derivation

       1. lift-+.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

       2. lift-/.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + \color{blue}{im}} \]

       3. lift-*.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

       4. lift--.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

       5. pow2 N/A

          \[\leadsto \frac{0 - {im}^{2}}{0 + im} \]

       6. sub0-neg N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{0 + im} \]

       7. add-flip N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(\left(\left(\mathsf{neg}\left(0\right)\right) - im\right)\right)} \]

       8. metadata-eval N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(\left(0 - im\right)\right)} \]

       9. sub0-neg N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(\left(\mathsf{neg}\left(im\right)\right)\right)} \]

       10. mul-1-neg N/A

           \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(-1 \cdot im\right)} \]

       11. frac-2neg-rev N/A

           \[\leadsto \frac{{im}^{2}}{-1 \cdot \color{blue}{im}} \]

       12. lower-/.f64 N/A

           \[\leadsto \frac{{im}^{2}}{-1 \cdot \color{blue}{im}} \]

       13. pow2 N/A

           \[\leadsto \frac{im \cdot im}{-1 \cdot im} \]

       14. lift-*.f64 N/A

           \[\leadsto \frac{im \cdot im}{-1 \cdot im} \]

       15. mul-1-neg N/A

           \[\leadsto \frac{im \cdot im}{\mathsf{neg}\left(im\right)} \]

       16. lift-neg.f64 53.6

           \[\leadsto \frac{im \cdot im}{-im} \]

   11. Applied rewrites 53.6%

       \[\leadsto \frac{im \cdot im}{-im} \]

   if -1e5 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < 0.0

   1. Initial program 8.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      3. sub0-neg N/A

         \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

      4. sub-negate N/A

         \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

      5. lower-neg.f64 N/A

         \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

      6. sub0-neg N/A

         \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

      7. sinh-undef N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      8. lower-*.f64 N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      9. lower-sinh.f64 56.7

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

   4. Applied rewrites 56.7%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   5. Taylor expanded in im around 0

      \[\leadsto -1 \cdot \color{blue}{im} \]
   6. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. lift-neg.f64 56.1

         \[\leadsto -im \]

   7. Applied rewrites 56.1%

      \[\leadsto -im \]

   if 0.0 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im)))

   1. Initial program 98.1%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in im around 0

      \[\leadsto \color{blue}{-1 \cdot \left(im \cdot \cos re\right)} \]
   3. Step-by-step derivation

      1. associate-*r* N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(-1 \cdot im\right) \cdot \color{blue}{\cos re} \]

      3. mul-1-neg N/A

         \[\leadsto \left(\mathsf{neg}\left(im\right)\right) \cdot \cos \color{blue}{re} \]

      4. lower-neg.f64 N/A

         \[\leadsto \left(-im\right) \cdot \cos \color{blue}{re} \]

      5. lift-cos.f64 9.6

         \[\leadsto \left(-im\right) \cdot \cos re \]

   4. Applied rewrites 9.6%

      \[\leadsto \color{blue}{\left(-im\right) \cdot \cos re} \]

   5. Taylor expanded in re around 0

      \[\leadsto -1 \cdot im + \color{blue}{\frac{1}{2} \cdot \left(im \cdot {re}^{2}\right)} \]
   6. Step-by-step derivation

      1. +-commutative N/A

         \[\leadsto \frac{1}{2} \cdot \left(im \cdot {re}^{2}\right) + -1 \cdot \color{blue}{im} \]

      2. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} + -1 \cdot im \]

      3. lower-fma.f64 N/A

         \[\leadsto \mathsf{fma}\left(im \cdot {re}^{2}, \frac{1}{2}, -1 \cdot im\right) \]

      4. *-commutative N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      5. lower-*.f64 N/A

         \[\leadsto \mathsf{fma}\left({re}^{2} \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      6. pow2 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      7. lift-*.f64 N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, -1 \cdot im\right) \]

      8. mul-1-neg N/A

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \frac{1}{2}, \mathsf{neg}\left(im\right)\right) \]

      9. lift-neg.f64 74.9

         \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, 0.5, -im\right) \]

   7. Applied rewrites 74.9%

      \[\leadsto \mathsf{fma}\left(\left(re \cdot re\right) \cdot im, \color{blue}{0.5}, -im\right) \]

   8. Taylor expanded in re around inf

      \[\leadsto \frac{1}{2} \cdot \left(im \cdot \color{blue}{{re}^{2}}\right) \]
   9. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(im \cdot {re}^{2}\right) \cdot \frac{1}{2} \]

      3. *-commutative N/A

         \[\leadsto \left({re}^{2} \cdot im\right) \cdot \frac{1}{2} \]

      4. pow2 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      5. lift-*.f64 N/A

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

      6. lift-*.f64 74.9

         \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]

   10. Applied rewrites 74.9%

       \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot 0.5 \]
   11. Step-by-step derivation

       1. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       2. lift-*.f64 N/A

          \[\leadsto \left(\left(re \cdot re\right) \cdot im\right) \cdot \frac{1}{2} \]

       3. associate-*l* N/A

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot \frac{1}{2} \]

       4. lower-*.f64 N/A

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot \frac{1}{2} \]

       5. lower-*.f64 74.9

          \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot 0.5 \]

   12. Applied rewrites 74.9%

       \[\leadsto \left(re \cdot \left(re \cdot im\right)\right) \cdot 0.5 \]
3. Recombined 3 regimes into one program.
4. Add Preprocessing

Alternative 9: 48.1% accurate, 0.9× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ im\_s \cdot \begin{array}{l} \mathbf{if}\;\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im\_m} - e^{im\_m}\right) \leq -100000:\\ \;\;\;\;\frac{im\_m \cdot im\_m}{-im\_m}\\ \mathbf{else}:\\ \;\;\;\;-im\_m\\ \end{array} \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m)
 :precision binary64
 (*
  im_s
  (if (<= (* (* 0.5 (cos re)) (- (exp (- 0.0 im_m)) (exp im_m))) -100000.0)
    (/ (* im_m im_m) (- im_m))
    (- im_m))))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	double tmp;
	if (((0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m))) <= -100000.0) {
		tmp = (im_m * im_m) / -im_m;
	} else {
		tmp = -im_m;
	}
	return im_s * tmp;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    real(8) :: tmp
    if (((0.5d0 * cos(re)) * (exp((0.0d0 - im_m)) - exp(im_m))) <= (-100000.0d0)) then
        tmp = (im_m * im_m) / -im_m
    else
        tmp = -im_m
    end if
    code = im_s * tmp
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	double tmp;
	if (((0.5 * Math.cos(re)) * (Math.exp((0.0 - im_m)) - Math.exp(im_m))) <= -100000.0) {
		tmp = (im_m * im_m) / -im_m;
	} else {
		tmp = -im_m;
	}
	return im_s * tmp;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	tmp = 0
	if ((0.5 * math.cos(re)) * (math.exp((0.0 - im_m)) - math.exp(im_m))) <= -100000.0:
		tmp = (im_m * im_m) / -im_m
	else:
		tmp = -im_m
	return im_s * tmp

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	tmp = 0.0
	if (Float64(Float64(0.5 * cos(re)) * Float64(exp(Float64(0.0 - im_m)) - exp(im_m))) <= -100000.0)
		tmp = Float64(Float64(im_m * im_m) / Float64(-im_m));
	else
		tmp = Float64(-im_m);
	end
	return Float64(im_s * tmp)
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp_2 = code(im_s, re, im_m)
	tmp = 0.0;
	if (((0.5 * cos(re)) * (exp((0.0 - im_m)) - exp(im_m))) <= -100000.0)
		tmp = (im_m * im_m) / -im_m;
	else
		tmp = -im_m;
	end
	tmp_2 = im_s * tmp;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := N[(im$95$s * If[LessEqual[N[(N[(0.5 * N[Cos[re], $MachinePrecision]), $MachinePrecision] * N[(N[Exp[N[(0.0 - im$95$m), $MachinePrecision]], $MachinePrecision] - N[Exp[im$95$m], $MachinePrecision]), $MachinePrecision]), $MachinePrecision], -100000.0], N[(N[(im$95$m * im$95$m), $MachinePrecision] / (-im$95$m)), $MachinePrecision], (-im$95$m)]), $MachinePrecision]

Derivation

1. Split input into 2 regimes

2. if (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im))) < -1e5

   1. Initial program 100.0%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      3. sub0-neg N/A

         \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

      4. sub-negate N/A

         \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

      5. lower-neg.f64 N/A

         \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

      6. sub0-neg N/A

         \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

      7. sinh-undef N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      8. lower-*.f64 N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      9. lower-sinh.f64 99.9

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

   4. Applied rewrites 99.9%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   5. Taylor expanded in im around 0

      \[\leadsto -1 \cdot \color{blue}{im} \]
   6. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. lift-neg.f64 5.5

         \[\leadsto -im \]

   7. Applied rewrites 5.5%

      \[\leadsto -im \]
   8. Step-by-step derivation

      1. lift-neg.f64 N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. sub0-neg N/A

         \[\leadsto 0 - im \]

      3. flip-- N/A

         \[\leadsto \frac{0 \cdot 0 - im \cdot im}{0 + \color{blue}{im}} \]

      4. lower-/.f64 N/A

         \[\leadsto \frac{0 \cdot 0 - im \cdot im}{0 + \color{blue}{im}} \]

      5. metadata-eval N/A

         \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

      6. unpow2 N/A

         \[\leadsto \frac{0 - {im}^{2}}{0 + im} \]

      7. lower--.f64 N/A

         \[\leadsto \frac{0 - {im}^{2}}{0 + im} \]

      8. unpow2 N/A

         \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

      9. lower-*.f64 N/A

         \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

      10. lower-+.f64 53.6

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

   9. Applied rewrites 53.6%

      \[\leadsto \frac{0 - im \cdot im}{0 + \color{blue}{im}} \]
   10. Step-by-step derivation

       1. lift-+.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

       2. lift-/.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + \color{blue}{im}} \]

       3. lift-*.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

       4. lift--.f64 N/A

          \[\leadsto \frac{0 - im \cdot im}{0 + im} \]

       5. pow2 N/A

          \[\leadsto \frac{0 - {im}^{2}}{0 + im} \]

       6. sub0-neg N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{0 + im} \]

       7. add-flip N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(\left(\left(\mathsf{neg}\left(0\right)\right) - im\right)\right)} \]

       8. metadata-eval N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(\left(0 - im\right)\right)} \]

       9. sub0-neg N/A

          \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(\left(\mathsf{neg}\left(im\right)\right)\right)} \]

       10. mul-1-neg N/A

           \[\leadsto \frac{\mathsf{neg}\left({im}^{2}\right)}{\mathsf{neg}\left(-1 \cdot im\right)} \]

       11. frac-2neg-rev N/A

           \[\leadsto \frac{{im}^{2}}{-1 \cdot \color{blue}{im}} \]

       12. lower-/.f64 N/A

           \[\leadsto \frac{{im}^{2}}{-1 \cdot \color{blue}{im}} \]

       13. pow2 N/A

           \[\leadsto \frac{im \cdot im}{-1 \cdot im} \]

       14. lift-*.f64 N/A

           \[\leadsto \frac{im \cdot im}{-1 \cdot im} \]

       15. mul-1-neg N/A

           \[\leadsto \frac{im \cdot im}{\mathsf{neg}\left(im\right)} \]

       16. lift-neg.f64 53.6

           \[\leadsto \frac{im \cdot im}{-im} \]

   11. Applied rewrites 53.6%

       \[\leadsto \frac{im \cdot im}{-im} \]

   if -1e5 < (*.f64 (*.f64 #s(literal 1/2 binary64) (cos.f64 re)) (-.f64 (exp.f64 (-.f64 #s(literal 0 binary64) im)) (exp.f64 im)))

   1. Initial program 25.9%

      \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

   2. Taylor expanded in re around 0

      \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
   3. Step-by-step derivation

      1. *-commutative N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      2. lower-*.f64 N/A

         \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

      3. sub0-neg N/A

         \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

      4. sub-negate N/A

         \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

      5. lower-neg.f64 N/A

         \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

      6. sub0-neg N/A

         \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

      7. sinh-undef N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      8. lower-*.f64 N/A

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

      9. lower-sinh.f64 45.4

         \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

   4. Applied rewrites 45.4%

      \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

   5. Taylor expanded in im around 0

      \[\leadsto -1 \cdot \color{blue}{im} \]
   6. Step-by-step derivation

      1. mul-1-neg N/A

         \[\leadsto \mathsf{neg}\left(im\right) \]

      2. lift-neg.f64 45.0

         \[\leadsto -im \]

   7. Applied rewrites 45.0%

      \[\leadsto -im \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 10: 30.6% accurate, 32.7× speedup

Math

\[\begin{array}{l} im\_m = \left|im\right| \\ im\_s = \mathsf{copysign}\left(1, im\right) \\ im\_s \cdot \left(-im\_m\right) \end{array} \]

FPCore

im\_m = (fabs.f64 im)
im\_s = (copysign.f64 #s(literal 1 binary64) im)
(FPCore (im_s re im_m) :precision binary64 (* im_s (- im_m)))

C

im\_m = fabs(im);
im\_s = copysign(1.0, im);
double code(double im_s, double re, double im_m) {
	return im_s * -im_m;
}

Fortran

im\_m =     private
im\_s =     private
module fmin_fmax_functions
    implicit none
    private
    public fmax
    public fmin

    interface fmax
        module procedure fmax88
        module procedure fmax44
        module procedure fmax84
        module procedure fmax48
    end interface
    interface fmin
        module procedure fmin88
        module procedure fmin44
        module procedure fmin84
        module procedure fmin48
    end interface
contains
    real(8) function fmax88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(4) function fmax44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, max(x, y), y /= y), x /= x)
    end function
    real(8) function fmax84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, max(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmax48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), max(dble(x), y), y /= y), x /= x)
    end function
    real(8) function fmin88(x, y) result (res)
        real(8), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(4) function fmin44(x, y) result (res)
        real(4), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(y, merge(x, min(x, y), y /= y), x /= x)
    end function
    real(8) function fmin84(x, y) result(res)
        real(8), intent (in) :: x
        real(4), intent (in) :: y
        res = merge(dble(y), merge(x, min(x, dble(y)), y /= y), x /= x)
    end function
    real(8) function fmin48(x, y) result(res)
        real(4), intent (in) :: x
        real(8), intent (in) :: y
        res = merge(y, merge(dble(x), min(dble(x), y), y /= y), x /= x)
    end function
end module

real(8) function code(im_s, re, im_m)
use fmin_fmax_functions
    real(8), intent (in) :: im_s
    real(8), intent (in) :: re
    real(8), intent (in) :: im_m
    code = im_s * -im_m
end function

Java

im\_m = Math.abs(im);
im\_s = Math.copySign(1.0, im);
public static double code(double im_s, double re, double im_m) {
	return im_s * -im_m;
}

Python

im\_m = math.fabs(im)
im\_s = math.copysign(1.0, im)
def code(im_s, re, im_m):
	return im_s * -im_m

Julia

im\_m = abs(im)
im\_s = copysign(1.0, im)
function code(im_s, re, im_m)
	return Float64(im_s * Float64(-im_m))
end

MATLAB

im\_m = abs(im);
im\_s = sign(im) * abs(1.0);
function tmp = code(im_s, re, im_m)
	tmp = im_s * -im_m;
end

Wolfram

im\_m = N[Abs[im], $MachinePrecision]
im\_s = N[With[{TMP1 = Abs[1.0], TMP2 = Sign[im]}, TMP1 * If[TMP2 == 0, 1, TMP2]], $MachinePrecision]
code[im$95$s_, re_, im$95$m_] := N[(im$95$s * (-im$95$m)), $MachinePrecision]

Derivation

1. Initial program 52.9%

   \[\left(0.5 \cdot \cos re\right) \cdot \left(e^{0 - im} - e^{im}\right) \]

2. Taylor expanded in re around 0

   \[\leadsto \color{blue}{\frac{1}{2} \cdot \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right)} \]
3. Step-by-step derivation

   1. *-commutative N/A

      \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

   2. lower-*.f64 N/A

      \[\leadsto \left(e^{\mathsf{neg}\left(im\right)} - e^{im}\right) \cdot \color{blue}{\frac{1}{2}} \]

   3. sub0-neg N/A

      \[\leadsto \left(e^{0 - im} - e^{im}\right) \cdot \frac{1}{2} \]

   4. sub-negate N/A

      \[\leadsto \left(\mathsf{neg}\left(\left(e^{im} - e^{0 - im}\right)\right)\right) \cdot \frac{1}{2} \]

   5. lower-neg.f64 N/A

      \[\leadsto \left(-\left(e^{im} - e^{0 - im}\right)\right) \cdot \frac{1}{2} \]

   6. sub0-neg N/A

      \[\leadsto \left(-\left(e^{im} - e^{\mathsf{neg}\left(im\right)}\right)\right) \cdot \frac{1}{2} \]

   7. sinh-undef N/A

      \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

   8. lower-*.f64 N/A

      \[\leadsto \left(-2 \cdot \sinh im\right) \cdot \frac{1}{2} \]

   9. lower-sinh.f64 65.3

      \[\leadsto \left(-2 \cdot \sinh im\right) \cdot 0.5 \]

4. Applied rewrites 65.3%

   \[\leadsto \color{blue}{\left(-2 \cdot \sinh im\right) \cdot 0.5} \]

5. Taylor expanded in im around 0

   \[\leadsto -1 \cdot \color{blue}{im} \]
6. Step-by-step derivation

   1. mul-1-neg N/A

      \[\leadsto \mathsf{neg}\left(im\right) \]

   2. lift-neg.f64 30.6

      \[\leadsto -im \]

7. Applied rewrites 30.6%

   \[\leadsto -im \]
8. Add Preprocessing

Reproduce

herbie shell --seed 2025122 
(FPCore (re im)
  :name "math.sin on complex, imaginary part"
  :precision binary64
  (* (* 0.5 (cos re)) (- (exp (- 0.0 im)) (exp im))))