HairBSDF, sample_f, cosTheta

Percentage Accurate: 99.5% → 99.5%

Time: 9.5s

Alternatives: 16

Speedup: 1.0×

Specification

\[\left(10^{-5} \leq u \land u \leq 1\right) \land \left(0 \leq v \land v \leq 109.746574\right)\]

Math

\[\begin{array}{l} \\ 1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v))))))))

C

float code(float u, float v) {
	return 1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))));
}

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(4) function code(u, v)
use fmin_fmax_functions
    real(4), intent (in) :: u
    real(4), intent (in) :: v
    code = 1.0e0 + (v * log((u + ((1.0e0 - u) * exp(((-2.0e0) / v))))))
end function

Julia

function code(u, v)
	return Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v)))))))
end

MATLAB

function tmp = code(u, v)
	tmp = single(1.0) + (v * log((u + ((single(1.0) - u) * exp((single(-2.0) / v))))));
end

Initial Program: 99.5% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ 1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v))))))))

C

float code(float u, float v) {
	return 1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))));
}

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(4) function code(u, v)
use fmin_fmax_functions
    real(4), intent (in) :: u
    real(4), intent (in) :: v
    code = 1.0e0 + (v * log((u + ((1.0e0 - u) * exp(((-2.0e0) / v))))))
end function

Julia

function code(u, v)
	return Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v)))))))
end

MATLAB

function tmp = code(u, v)
	tmp = single(1.0) + (v * log((u + ((single(1.0) - u) * exp((single(-2.0) / v))))));
end

Alternative 1: 99.5% accurate, 0.7× speedup

Math

\[\begin{array}{l} \\ 1 + \frac{v}{2} \cdot \log \left({\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)}^{2}\right) \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (+ 1.0 (* (/ v 2.0) (log (pow (fma (exp (/ -2.0 v)) (- 1.0 u) u) 2.0)))))

C

float code(float u, float v) {
	return 1.0f + ((v / 2.0f) * logf(powf(fmaf(expf((-2.0f / v)), (1.0f - u), u), 2.0f)));
}

Julia

function code(u, v)
	return Float32(Float32(1.0) + Float32(Float32(v / Float32(2.0)) * log((fma(exp(Float32(Float32(-2.0) / v)), Float32(Float32(1.0) - u), u) ^ Float32(2.0)))))
end

Derivation

1. Initial program 99.4%

   \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
2. Add Preprocessing
3. Step-by-step derivation

   1. lift-*.f32 N/A

      \[\leadsto 1 + \color{blue}{v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} \]

   2. lift-log.f32 N/A

      \[\leadsto 1 + v \cdot \color{blue}{\log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} \]

   3. log-pow-rev N/A

      \[\leadsto 1 + \color{blue}{\log \left({\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{v}\right)} \]

   4. sqr-pow N/A

      \[\leadsto 1 + \log \color{blue}{\left({\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{\left(\frac{v}{2}\right)} \cdot {\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{\left(\frac{v}{2}\right)}\right)} \]

   5. pow-prod-down N/A

      \[\leadsto 1 + \log \color{blue}{\left({\left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)}^{\left(\frac{v}{2}\right)}\right)} \]

   6. log-pow N/A

      \[\leadsto 1 + \color{blue}{\frac{v}{2} \cdot \log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)} \]

   7. lower-*.f32 N/A

      \[\leadsto 1 + \color{blue}{\frac{v}{2} \cdot \log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)} \]

   8. metadata-eval N/A

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

   9. lower-/.f32 N/A

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

   10. metadata-eval N/A

       \[\leadsto 1 + \frac{v}{\color{blue}{2}} \cdot \log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right) \]

   11. lower-log.f32 N/A

       \[\leadsto 1 + \frac{v}{2} \cdot \color{blue}{\log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)} \]

   12. pow2 N/A

       \[\leadsto 1 + \frac{v}{2} \cdot \log \color{blue}{\left({\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{2}\right)} \]

   13. metadata-eval N/A

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

4. Applied rewrites 99.5%

   \[\leadsto 1 + \color{blue}{\frac{v}{2} \cdot \log \left({\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)}^{2}\right)} \]
5. Add Preprocessing

Alternative 2: 97.6% accurate, 0.5× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq -0.30000001192092896:\\ \;\;\;\;\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1\\ \mathbf{else}:\\ \;\;\;\;\mathsf{fma}\left(\log \left(\left(-u\right) \cdot \mathsf{expm1}\left(\frac{-2}{v}\right)\right), v, 1\right)\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      -0.30000001192092896)
   (- (* (* u v) (expm1 (/ 2.0 v))) 1.0)
   (fma (log (* (- u) (expm1 (/ -2.0 v)))) v 1.0)))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= -0.30000001192092896f) {
		tmp = ((u * v) * expm1f((2.0f / v))) - 1.0f;
	} else {
		tmp = fmaf(logf((-u * expm1f((-2.0f / v)))), v, 1.0f);
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(-0.30000001192092896))
		tmp = Float32(Float32(Float32(u * v) * expm1(Float32(Float32(2.0) / v))) - Float32(1.0));
	else
		tmp = fma(log(Float32(Float32(-u) * expm1(Float32(Float32(-2.0) / v)))), v, Float32(1.0));
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < -0.300000012

   1. Initial program 94.4%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

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

      1. lower--.f32 N/A

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

      2. associate-*r* N/A

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

      3. lower-*.f32 N/A

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

      4. lower-*.f32 N/A

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

      5. rec-exp N/A

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

      6. distribute-neg-frac N/A

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

      7. metadata-eval N/A

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

      8. metadata-eval N/A

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

      9. associate-*r/ N/A

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

      10. lower-expm1.f32 N/A

          \[\leadsto \left(u \cdot v\right) \cdot \color{blue}{\mathsf{expm1}\left(2 \cdot \frac{1}{v}\right)} - 1 \]

      11. associate-*r/ N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2 \cdot 1}{v}}\right) - 1 \]

      12. metadata-eval N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{\color{blue}{2}}{v}\right) - 1 \]

      13. lower-/.f32 75.8

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2}{v}}\right) - 1 \]

   5. Applied rewrites 75.8%

      \[\leadsto \color{blue}{\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1} \]

   if -0.300000012 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing
   3. Step-by-step derivation

      1. lift-+.f32 N/A

         \[\leadsto \color{blue}{1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} \]

      2. +-commutative N/A

         \[\leadsto \color{blue}{v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) + 1} \]

      3. lift-*.f32 N/A

         \[\leadsto \color{blue}{v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} + 1 \]

      4. *-commutative N/A

         \[\leadsto \color{blue}{\log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot v} + 1 \]

      5. lower-fma.f32 100.0

         \[\leadsto \color{blue}{\mathsf{fma}\left(\log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right), v, 1\right)} \]

      6. lift-+.f32 N/A

         \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}, v, 1\right) \]

      7. +-commutative N/A

         \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right)}, v, 1\right) \]

      8. lift-*.f32 N/A

         \[\leadsto \mathsf{fma}\left(\log \left(\color{blue}{\left(1 - u\right) \cdot e^{\frac{-2}{v}}} + u\right), v, 1\right) \]

      9. *-commutative N/A

         \[\leadsto \mathsf{fma}\left(\log \left(\color{blue}{e^{\frac{-2}{v}} \cdot \left(1 - u\right)} + u\right), v, 1\right) \]

      10. lower-fma.f32 100.0

          \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)}, v, 1\right) \]

   4. Applied rewrites 100.0%

      \[\leadsto \color{blue}{\mathsf{fma}\left(\log \left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right), v, 1\right)} \]

   5. Taylor expanded in u around -inf

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

      1. associate-*r* N/A

         \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(\left(-1 \cdot u\right) \cdot \left(e^{\frac{-2}{v}} - 1\right)\right)}, v, 1\right) \]

      2. mul-1-neg N/A

         \[\leadsto \mathsf{fma}\left(\log \left(\color{blue}{\left(\mathsf{neg}\left(u\right)\right)} \cdot \left(e^{\frac{-2}{v}} - 1\right)\right), v, 1\right) \]

      3. lower-*.f32 N/A

         \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(\left(\mathsf{neg}\left(u\right)\right) \cdot \left(e^{\frac{-2}{v}} - 1\right)\right)}, v, 1\right) \]

      4. lower-neg.f32 N/A

         \[\leadsto \mathsf{fma}\left(\log \left(\color{blue}{\left(-u\right)} \cdot \left(e^{\frac{-2}{v}} - 1\right)\right), v, 1\right) \]

      5. lower-expm1.f32 N/A

         \[\leadsto \mathsf{fma}\left(\log \left(\left(-u\right) \cdot \color{blue}{\mathsf{expm1}\left(\frac{-2}{v}\right)}\right), v, 1\right) \]

      6. lower-/.f32 99.1

         \[\leadsto \mathsf{fma}\left(\log \left(\left(-u\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{-2}{v}}\right)\right), v, 1\right) \]

   7. Applied rewrites 99.1%

      \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(\left(-u\right) \cdot \mathsf{expm1}\left(\frac{-2}{v}\right)\right)}, v, 1\right) \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 3: 91.2% accurate, 0.6× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq -0.20000000298023224:\\ \;\;\;\;\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      -0.20000000298023224)
   (- (* (* u v) (expm1 (/ 2.0 v))) 1.0)
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= -0.20000000298023224f) {
		tmp = ((u * v) * expm1f((2.0f / v))) - 1.0f;
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(-0.20000000298023224))
		tmp = Float32(Float32(Float32(u * v) * expm1(Float32(Float32(2.0) / v))) - Float32(1.0));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < -0.200000003

   1. Initial program 94.3%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

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

      1. lower--.f32 N/A

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

      2. associate-*r* N/A

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

      3. lower-*.f32 N/A

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

      4. lower-*.f32 N/A

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

      5. rec-exp N/A

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

      6. distribute-neg-frac N/A

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

      7. metadata-eval N/A

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

      8. metadata-eval N/A

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

      9. associate-*r/ N/A

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

      10. lower-expm1.f32 N/A

          \[\leadsto \left(u \cdot v\right) \cdot \color{blue}{\mathsf{expm1}\left(2 \cdot \frac{1}{v}\right)} - 1 \]

      11. associate-*r/ N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2 \cdot 1}{v}}\right) - 1 \]

      12. metadata-eval N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{\color{blue}{2}}{v}\right) - 1 \]

      13. lower-/.f32 72.9

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2}{v}}\right) - 1 \]

   5. Applied rewrites 72.9%

      \[\leadsto \color{blue}{\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1} \]

   if -0.200000003 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 92.7%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 4: 91.3% accurate, 0.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq 0.5:\\ \;\;\;\;1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\left(\mathsf{fma}\left(-16, u, 24\right) \cdot u - 8\right) \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<= (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v))))))) 0.5)
   (+
    1.0
    (fma
     (- 1.0 u)
     -2.0
     (/
      (fma
       -0.16666666666666666
       (/ (* (- (* (fma -16.0 u 24.0) u) 8.0) u) v)
       (* (fma -2.0 u 2.0) u))
      v)))
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= 0.5f) {
		tmp = 1.0f + fmaf((1.0f - u), -2.0f, (fmaf(-0.16666666666666666f, ((((fmaf(-16.0f, u, 24.0f) * u) - 8.0f) * u) / v), (fmaf(-2.0f, u, 2.0f) * u)) / v));
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(0.5))
		tmp = Float32(Float32(1.0) + fma(Float32(Float32(1.0) - u), Float32(-2.0), Float32(fma(Float32(-0.16666666666666666), Float32(Float32(Float32(Float32(fma(Float32(-16.0), u, Float32(24.0)) * u) - Float32(8.0)) * u) / v), Float32(fma(Float32(-2.0), u, Float32(2.0)) * u)) / v)));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < 0.5

   1. Initial program 94.5%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around -inf

      \[\leadsto 1 + \color{blue}{\left(-2 \cdot \left(1 - u\right) + -1 \cdot \frac{\frac{-1}{2} \cdot \left(-4 \cdot {\left(1 - u\right)}^{2} + 4 \cdot \left(1 - u\right)\right) + \frac{1}{6} \cdot \frac{-24 \cdot {\left(1 - u\right)}^{2} + \left(8 \cdot \left(1 - u\right) + 16 \cdot {\left(1 - u\right)}^{3}\right)}{v}}{v}\right)} \]

   5. Applied rewrites 69.0%

      \[\leadsto 1 + \color{blue}{\mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\mathsf{fma}\left(-24, {\left(1 - u\right)}^{2}, \mathsf{fma}\left(8, 1 - u, 16 \cdot {\left(1 - u\right)}^{3}\right)\right)}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right)} \]

   6. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]
   7. Step-by-step derivation

   8. Applied rewrites 67.7%

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]

   9. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, u \cdot \left(2 + -2 \cdot u\right)\right)}{v}\right) \]
   10. Step-by-step derivation

   11. Applied rewrites 67.7%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]

   12. Taylor expanded in u around 0

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{u \cdot \left(u \cdot \left(24 + -16 \cdot u\right) - 8\right)}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]
   13. Step-by-step derivation

   14. Applied rewrites 69.0%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\left(\mathsf{fma}\left(-16, u, 24\right) \cdot u - 8\right) \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]

   if 0.5 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 93.0%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 5: 91.2% accurate, 0.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq 0.550000011920929:\\ \;\;\;\;1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\mathsf{fma}\left(\frac{4}{v}, -1, -2\right), u, \frac{1.3333333333333333}{v} + 2\right) \cdot u}{v}\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      0.550000011920929)
   (+
    1.0
    (fma
     (- 1.0 u)
     -2.0
     (/
      (* (fma (fma (/ 4.0 v) -1.0 -2.0) u (+ (/ 1.3333333333333333 v) 2.0)) u)
      v)))
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= 0.550000011920929f) {
		tmp = 1.0f + fmaf((1.0f - u), -2.0f, ((fmaf(fmaf((4.0f / v), -1.0f, -2.0f), u, ((1.3333333333333333f / v) + 2.0f)) * u) / v));
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(0.550000011920929))
		tmp = Float32(Float32(1.0) + fma(Float32(Float32(1.0) - u), Float32(-2.0), Float32(Float32(fma(fma(Float32(Float32(4.0) / v), Float32(-1.0), Float32(-2.0)), u, Float32(Float32(Float32(1.3333333333333333) / v) + Float32(2.0))) * u) / v)));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < 0.550000012

   1. Initial program 94.6%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around -inf

      \[\leadsto 1 + \color{blue}{\left(-2 \cdot \left(1 - u\right) + -1 \cdot \frac{\frac{-1}{2} \cdot \left(-4 \cdot {\left(1 - u\right)}^{2} + 4 \cdot \left(1 - u\right)\right) + \frac{1}{6} \cdot \frac{-24 \cdot {\left(1 - u\right)}^{2} + \left(8 \cdot \left(1 - u\right) + 16 \cdot {\left(1 - u\right)}^{3}\right)}{v}}{v}\right)} \]

   5. Applied rewrites 67.5%

      \[\leadsto 1 + \color{blue}{\mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\mathsf{fma}\left(-24, {\left(1 - u\right)}^{2}, \mathsf{fma}\left(8, 1 - u, 16 \cdot {\left(1 - u\right)}^{3}\right)\right)}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right)} \]

   6. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]
   7. Step-by-step derivation

   8. Applied rewrites 66.1%

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]

   9. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, u \cdot \left(2 + -2 \cdot u\right)\right)}{v}\right) \]
   10. Step-by-step derivation

   11. Applied rewrites 66.1%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]

   12. Taylor expanded in u around 0

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{u \cdot \left(2 + \left(-1 \cdot \left(u \cdot \left(2 + 4 \cdot \frac{1}{v}\right)\right) + \frac{4}{3} \cdot \frac{1}{v}\right)\right)}{v}\right) \]
   13. Step-by-step derivation

   14. Applied rewrites 67.1%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\mathsf{fma}\left(\frac{4}{v}, -1, -2\right), u, \frac{1.3333333333333333}{v} + 2\right) \cdot u}{v}\right) \]

   if 0.550000012 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 93.3%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 6: 91.2% accurate, 0.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq 0.550000011920929:\\ \;\;\;\;1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\left(24 \cdot u - 8\right) \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      0.550000011920929)
   (+
    1.0
    (fma
     (- 1.0 u)
     -2.0
     (/
      (fma
       -0.16666666666666666
       (/ (* (- (* 24.0 u) 8.0) u) v)
       (* (fma -2.0 u 2.0) u))
      v)))
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= 0.550000011920929f) {
		tmp = 1.0f + fmaf((1.0f - u), -2.0f, (fmaf(-0.16666666666666666f, ((((24.0f * u) - 8.0f) * u) / v), (fmaf(-2.0f, u, 2.0f) * u)) / v));
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(0.550000011920929))
		tmp = Float32(Float32(1.0) + fma(Float32(Float32(1.0) - u), Float32(-2.0), Float32(fma(Float32(-0.16666666666666666), Float32(Float32(Float32(Float32(Float32(24.0) * u) - Float32(8.0)) * u) / v), Float32(fma(Float32(-2.0), u, Float32(2.0)) * u)) / v)));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < 0.550000012

   1. Initial program 94.6%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around -inf

      \[\leadsto 1 + \color{blue}{\left(-2 \cdot \left(1 - u\right) + -1 \cdot \frac{\frac{-1}{2} \cdot \left(-4 \cdot {\left(1 - u\right)}^{2} + 4 \cdot \left(1 - u\right)\right) + \frac{1}{6} \cdot \frac{-24 \cdot {\left(1 - u\right)}^{2} + \left(8 \cdot \left(1 - u\right) + 16 \cdot {\left(1 - u\right)}^{3}\right)}{v}}{v}\right)} \]

   5. Applied rewrites 67.5%

      \[\leadsto 1 + \color{blue}{\mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\mathsf{fma}\left(-24, {\left(1 - u\right)}^{2}, \mathsf{fma}\left(8, 1 - u, 16 \cdot {\left(1 - u\right)}^{3}\right)\right)}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right)} \]

   6. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]
   7. Step-by-step derivation

   8. Applied rewrites 66.1%

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]

   9. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, u \cdot \left(2 + -2 \cdot u\right)\right)}{v}\right) \]
   10. Step-by-step derivation

   11. Applied rewrites 66.1%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]

   12. Taylor expanded in u around 0

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{u \cdot \left(24 \cdot u - 8\right)}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]
   13. Step-by-step derivation

   14. Applied rewrites 67.1%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\left(24 \cdot u - 8\right) \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]

   if 0.550000012 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 93.3%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 7: 91.0% accurate, 0.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq 0.5:\\ \;\;\;\;\mathsf{fma}\left(\frac{1.3333333333333333}{v}, \frac{u}{v}, 2 \cdot \left(\frac{u}{v} + u\right)\right) - 1\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<= (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v))))))) 0.5)
   (- (fma (/ 1.3333333333333333 v) (/ u v) (* 2.0 (+ (/ u v) u))) 1.0)
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= 0.5f) {
		tmp = fmaf((1.3333333333333333f / v), (u / v), (2.0f * ((u / v) + u))) - 1.0f;
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(0.5))
		tmp = Float32(fma(Float32(Float32(1.3333333333333333) / v), Float32(u / v), Float32(Float32(2.0) * Float32(Float32(u / v) + u))) - Float32(1.0));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < 0.5

   1. Initial program 94.5%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

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

      1. lower--.f32 N/A

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

      2. associate-*r* N/A

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

      3. lower-*.f32 N/A

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

      4. lower-*.f32 N/A

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

      5. rec-exp N/A

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

      6. distribute-neg-frac N/A

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

      7. metadata-eval N/A

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

      8. metadata-eval N/A

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

      9. associate-*r/ N/A

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

      10. lower-expm1.f32 N/A

          \[\leadsto \left(u \cdot v\right) \cdot \color{blue}{\mathsf{expm1}\left(2 \cdot \frac{1}{v}\right)} - 1 \]

      11. associate-*r/ N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2 \cdot 1}{v}}\right) - 1 \]

      12. metadata-eval N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{\color{blue}{2}}{v}\right) - 1 \]

      13. lower-/.f32 70.8

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2}{v}}\right) - 1 \]

   5. Applied rewrites 70.8%

      \[\leadsto \color{blue}{\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1} \]

   6. Taylor expanded in v around inf

      \[\leadsto \left(\frac{4}{3} \cdot \frac{u}{{v}^{2}} + \left(2 \cdot u + 2 \cdot \frac{u}{v}\right)\right) - 1 \]
   7. Step-by-step derivation

   8. Applied rewrites 67.8%

      \[\leadsto \mathsf{fma}\left(\frac{1.3333333333333333}{v}, \frac{u}{v}, 2 \cdot \left(\frac{u}{v} + u\right)\right) - 1 \]

   if 0.5 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 93.0%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 8: 91.0% accurate, 0.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq 0.5:\\ \;\;\;\;\mathsf{fma}\left(2, u, \frac{\mathsf{fma}\left(\frac{u}{v}, -1.3333333333333333, -2 \cdot u\right)}{-v}\right) - 1\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<= (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v))))))) 0.5)
   (- (fma 2.0 u (/ (fma (/ u v) -1.3333333333333333 (* -2.0 u)) (- v))) 1.0)
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= 0.5f) {
		tmp = fmaf(2.0f, u, (fmaf((u / v), -1.3333333333333333f, (-2.0f * u)) / -v)) - 1.0f;
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(0.5))
		tmp = Float32(fma(Float32(2.0), u, Float32(fma(Float32(u / v), Float32(-1.3333333333333333), Float32(Float32(-2.0) * u)) / Float32(-v))) - Float32(1.0));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < 0.5

   1. Initial program 94.5%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

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

      1. lower--.f32 N/A

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

      2. associate-*r* N/A

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

      3. lower-*.f32 N/A

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

      4. lower-*.f32 N/A

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

      5. rec-exp N/A

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

      6. distribute-neg-frac N/A

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

      7. metadata-eval N/A

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

      8. metadata-eval N/A

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

      9. associate-*r/ N/A

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

      10. lower-expm1.f32 N/A

          \[\leadsto \left(u \cdot v\right) \cdot \color{blue}{\mathsf{expm1}\left(2 \cdot \frac{1}{v}\right)} - 1 \]

      11. associate-*r/ N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2 \cdot 1}{v}}\right) - 1 \]

      12. metadata-eval N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{\color{blue}{2}}{v}\right) - 1 \]

      13. lower-/.f32 70.8

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2}{v}}\right) - 1 \]

   5. Applied rewrites 70.8%

      \[\leadsto \color{blue}{\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1} \]

   6. Taylor expanded in v around -inf

      \[\leadsto \left(-1 \cdot \frac{-2 \cdot u + \frac{-4}{3} \cdot \frac{u}{v}}{v} + 2 \cdot u\right) - 1 \]
   7. Step-by-step derivation

   8. Applied rewrites 67.8%

      \[\leadsto \mathsf{fma}\left(2, u, \frac{\mathsf{fma}\left(\frac{u}{v}, -1.3333333333333333, -2 \cdot u\right)}{-v}\right) - 1 \]

   if 0.5 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 93.0%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 9: 90.9% accurate, 0.8× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq 0.5:\\ \;\;\;\;1 + \mathsf{fma}\left(1 - u, -2, \frac{\left(\frac{1.3333333333333333}{v} + 2\right) \cdot u}{v}\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<= (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v))))))) 0.5)
   (+ 1.0 (fma (- 1.0 u) -2.0 (/ (* (+ (/ 1.3333333333333333 v) 2.0) u) v)))
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= 0.5f) {
		tmp = 1.0f + fmaf((1.0f - u), -2.0f, ((((1.3333333333333333f / v) + 2.0f) * u) / v));
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(0.5))
		tmp = Float32(Float32(1.0) + fma(Float32(Float32(1.0) - u), Float32(-2.0), Float32(Float32(Float32(Float32(Float32(1.3333333333333333) / v) + Float32(2.0)) * u) / v)));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < 0.5

   1. Initial program 94.5%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around -inf

      \[\leadsto 1 + \color{blue}{\left(-2 \cdot \left(1 - u\right) + -1 \cdot \frac{\frac{-1}{2} \cdot \left(-4 \cdot {\left(1 - u\right)}^{2} + 4 \cdot \left(1 - u\right)\right) + \frac{1}{6} \cdot \frac{-24 \cdot {\left(1 - u\right)}^{2} + \left(8 \cdot \left(1 - u\right) + 16 \cdot {\left(1 - u\right)}^{3}\right)}{v}}{v}\right)} \]

   5. Applied rewrites 69.0%

      \[\leadsto 1 + \color{blue}{\mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{\mathsf{fma}\left(-24, {\left(1 - u\right)}^{2}, \mathsf{fma}\left(8, 1 - u, 16 \cdot {\left(1 - u\right)}^{3}\right)\right)}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right)} \]

   6. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]
   7. Step-by-step derivation

   8. Applied rewrites 67.7%

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left({\left(1 - u\right)}^{2}, -2, \left(1 - u\right) \cdot 2\right)\right)}{v}\right) \]

   9. Taylor expanded in u around 0

      \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(\frac{-1}{6}, \frac{-8 \cdot u}{v}, u \cdot \left(2 + -2 \cdot u\right)\right)}{v}\right) \]
   10. Step-by-step derivation

   11. Applied rewrites 67.7%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\mathsf{fma}\left(-0.16666666666666666, \frac{-8 \cdot u}{v}, \mathsf{fma}\left(-2, u, 2\right) \cdot u\right)}{v}\right) \]

   12. Taylor expanded in u around 0

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{u \cdot \left(2 + \frac{4}{3} \cdot \frac{1}{v}\right)}{v}\right) \]
   13. Step-by-step derivation

   14. Applied rewrites 67.5%

       \[\leadsto 1 + \mathsf{fma}\left(1 - u, -2, \frac{\left(\frac{1.3333333333333333}{v} + 2\right) \cdot u}{v}\right) \]

   if 0.5 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 93.0%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 10: 90.2% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq -0.20000000298023224:\\ \;\;\;\;\left(u + u\right) - 1\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      -0.20000000298023224)
   (- (+ u u) 1.0)
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= -0.20000000298023224f) {
		tmp = (u + u) - 1.0f;
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

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(4) function code(u, v)
use fmin_fmax_functions
    real(4), intent (in) :: u
    real(4), intent (in) :: v
    real(4) :: tmp
    if ((1.0e0 + (v * log((u + ((1.0e0 - u) * exp(((-2.0e0) / v))))))) <= (-0.20000000298023224e0)) then
        tmp = (u + u) - 1.0e0
    else
        tmp = 1.0e0
    end if
    code = tmp
end function

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(-0.20000000298023224))
		tmp = Float32(Float32(u + u) - Float32(1.0));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

MATLAB

function tmp_2 = code(u, v)
	tmp = single(0.0);
	if ((single(1.0) + (v * log((u + ((single(1.0) - u) * exp((single(-2.0) / v))))))) <= single(-0.20000000298023224))
		tmp = (u + u) - single(1.0);
	else
		tmp = single(1.0);
	end
	tmp_2 = tmp;
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < -0.200000003

   1. Initial program 94.3%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

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

      1. lower--.f32 N/A

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

      2. associate-*r* N/A

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

      3. lower-*.f32 N/A

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

      4. lower-*.f32 N/A

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

      5. rec-exp N/A

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

      6. distribute-neg-frac N/A

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

      7. metadata-eval N/A

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

      8. metadata-eval N/A

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

      9. associate-*r/ N/A

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

      10. lower-expm1.f32 N/A

          \[\leadsto \left(u \cdot v\right) \cdot \color{blue}{\mathsf{expm1}\left(2 \cdot \frac{1}{v}\right)} - 1 \]

      11. associate-*r/ N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2 \cdot 1}{v}}\right) - 1 \]

      12. metadata-eval N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{\color{blue}{2}}{v}\right) - 1 \]

      13. lower-/.f32 72.9

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2}{v}}\right) - 1 \]

   5. Applied rewrites 72.9%

      \[\leadsto \color{blue}{\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1} \]

   6. Taylor expanded in v around inf

      \[\leadsto 2 \cdot u - \color{blue}{1} \]
   7. Step-by-step derivation

   8. Applied rewrites 54.7%

      \[\leadsto 2 \cdot u - \color{blue}{1} \]
   9. Step-by-step derivation

   10. Applied rewrites 54.7%

       \[\leadsto \left(u + u\right) - 1 \]

   if -0.200000003 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 92.7%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 11: 90.2% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq -0.20000000298023224:\\ \;\;\;\;u + \left(u - 1\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      -0.20000000298023224)
   (+ u (- u 1.0))
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= -0.20000000298023224f) {
		tmp = u + (u - 1.0f);
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

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(4) function code(u, v)
use fmin_fmax_functions
    real(4), intent (in) :: u
    real(4), intent (in) :: v
    real(4) :: tmp
    if ((1.0e0 + (v * log((u + ((1.0e0 - u) * exp(((-2.0e0) / v))))))) <= (-0.20000000298023224e0)) then
        tmp = u + (u - 1.0e0)
    else
        tmp = 1.0e0
    end if
    code = tmp
end function

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(-0.20000000298023224))
		tmp = Float32(u + Float32(u - Float32(1.0)));
	else
		tmp = Float32(1.0);
	end
	return tmp
end

MATLAB

function tmp_2 = code(u, v)
	tmp = single(0.0);
	if ((single(1.0) + (v * log((u + ((single(1.0) - u) * exp((single(-2.0) / v))))))) <= single(-0.20000000298023224))
		tmp = u + (u - single(1.0));
	else
		tmp = single(1.0);
	end
	tmp_2 = tmp;
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < -0.200000003

   1. Initial program 94.3%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

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

      1. lower--.f32 N/A

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

      2. associate-*r* N/A

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

      3. lower-*.f32 N/A

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

      4. lower-*.f32 N/A

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

      5. rec-exp N/A

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

      6. distribute-neg-frac N/A

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

      7. metadata-eval N/A

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

      8. metadata-eval N/A

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

      9. associate-*r/ N/A

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

      10. lower-expm1.f32 N/A

          \[\leadsto \left(u \cdot v\right) \cdot \color{blue}{\mathsf{expm1}\left(2 \cdot \frac{1}{v}\right)} - 1 \]

      11. associate-*r/ N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2 \cdot 1}{v}}\right) - 1 \]

      12. metadata-eval N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{\color{blue}{2}}{v}\right) - 1 \]

      13. lower-/.f32 72.9

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2}{v}}\right) - 1 \]

   5. Applied rewrites 72.9%

      \[\leadsto \color{blue}{\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1} \]

   6. Taylor expanded in v around inf

      \[\leadsto 2 \cdot u - \color{blue}{1} \]
   7. Step-by-step derivation

   8. Applied rewrites 54.7%

      \[\leadsto 2 \cdot u - \color{blue}{1} \]
   9. Step-by-step derivation

   10. Applied rewrites 54.7%

       \[\leadsto u + \left(u - \color{blue}{1}\right) \]

   if -0.200000003 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 92.7%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 12: 89.7% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq -0.20000000298023224:\\ \;\;\;\;-1\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      -0.20000000298023224)
   -1.0
   1.0))

C

float code(float u, float v) {
	float tmp;
	if ((1.0f + (v * logf((u + ((1.0f - u) * expf((-2.0f / v))))))) <= -0.20000000298023224f) {
		tmp = -1.0f;
	} else {
		tmp = 1.0f;
	}
	return tmp;
}

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(4) function code(u, v)
use fmin_fmax_functions
    real(4), intent (in) :: u
    real(4), intent (in) :: v
    real(4) :: tmp
    if ((1.0e0 + (v * log((u + ((1.0e0 - u) * exp(((-2.0e0) / v))))))) <= (-0.20000000298023224e0)) then
        tmp = -1.0e0
    else
        tmp = 1.0e0
    end if
    code = tmp
end function

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (Float32(Float32(1.0) + Float32(v * log(Float32(u + Float32(Float32(Float32(1.0) - u) * exp(Float32(Float32(-2.0) / v))))))) <= Float32(-0.20000000298023224))
		tmp = Float32(-1.0);
	else
		tmp = Float32(1.0);
	end
	return tmp
end

MATLAB

function tmp_2 = code(u, v)
	tmp = single(0.0);
	if ((single(1.0) + (v * log((u + ((single(1.0) - u) * exp((single(-2.0) / v))))))) <= single(-0.20000000298023224))
		tmp = single(-1.0);
	else
		tmp = single(1.0);
	end
	tmp_2 = tmp;
end

Derivation

1. Split input into 2 regimes

2. if (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v))))))) < -0.200000003

   1. Initial program 94.3%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

      \[\leadsto \color{blue}{-1} \]
   4. Step-by-step derivation

   5. Applied rewrites 48.4%

      \[\leadsto \color{blue}{-1} \]

   if -0.200000003 < (+.f32 #s(literal 1 binary32) (*.f32 v (log.f32 (+.f32 u (*.f32 (-.f32 #s(literal 1 binary32) u) (exp.f32 (/.f32 #s(literal -2 binary32) v)))))))

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 92.7%

      \[\leadsto \color{blue}{1} \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 13: 99.5% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(\log \left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right), v, 1\right) \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (fma (log (fma (exp (/ -2.0 v)) (- 1.0 u) u)) v 1.0))

C

float code(float u, float v) {
	return fmaf(logf(fmaf(expf((-2.0f / v)), (1.0f - u), u)), v, 1.0f);
}

Julia

function code(u, v)
	return fma(log(fma(exp(Float32(Float32(-2.0) / v)), Float32(Float32(1.0) - u), u)), v, Float32(1.0))
end

Derivation

1. Initial program 99.4%

   \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
2. Add Preprocessing
3. Step-by-step derivation

   1. lift-+.f32 N/A

      \[\leadsto \color{blue}{1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} \]

   2. +-commutative N/A

      \[\leadsto \color{blue}{v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) + 1} \]

   3. lift-*.f32 N/A

      \[\leadsto \color{blue}{v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} + 1 \]

   4. *-commutative N/A

      \[\leadsto \color{blue}{\log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot v} + 1 \]

   5. lower-fma.f32 99.5

      \[\leadsto \color{blue}{\mathsf{fma}\left(\log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right), v, 1\right)} \]

   6. lift-+.f32 N/A

      \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}, v, 1\right) \]

   7. +-commutative N/A

      \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right)}, v, 1\right) \]

   8. lift-*.f32 N/A

      \[\leadsto \mathsf{fma}\left(\log \left(\color{blue}{\left(1 - u\right) \cdot e^{\frac{-2}{v}}} + u\right), v, 1\right) \]

   9. *-commutative N/A

      \[\leadsto \mathsf{fma}\left(\log \left(\color{blue}{e^{\frac{-2}{v}} \cdot \left(1 - u\right)} + u\right), v, 1\right) \]

   10. lower-fma.f32 99.5

       \[\leadsto \mathsf{fma}\left(\log \color{blue}{\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)}, v, 1\right) \]

4. Applied rewrites 99.5%

   \[\leadsto \color{blue}{\mathsf{fma}\left(\log \left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right), v, 1\right)} \]
5. Add Preprocessing

Alternative 14: 96.2% accurate, 1.0× speedup

Math

\[\begin{array}{l} \\ \log \left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1, u\right)\right) \cdot v + 1 \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (+ (* (log (fma (exp (/ -2.0 v)) 1.0 u)) v) 1.0))

C

float code(float u, float v) {
	return (logf(fmaf(expf((-2.0f / v)), 1.0f, u)) * v) + 1.0f;
}

Julia

function code(u, v)
	return Float32(Float32(log(fma(exp(Float32(Float32(-2.0) / v)), Float32(1.0), u)) * v) + Float32(1.0))
end

Derivation

1. Initial program 99.4%

   \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
2. Add Preprocessing
3. Step-by-step derivation

   1. lift-*.f32 N/A

      \[\leadsto 1 + \color{blue}{v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} \]

   2. lift-log.f32 N/A

      \[\leadsto 1 + v \cdot \color{blue}{\log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)} \]

   3. log-pow-rev N/A

      \[\leadsto 1 + \color{blue}{\log \left({\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{v}\right)} \]

   4. sqr-pow N/A

      \[\leadsto 1 + \log \color{blue}{\left({\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{\left(\frac{v}{2}\right)} \cdot {\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{\left(\frac{v}{2}\right)}\right)} \]

   5. pow-prod-down N/A

      \[\leadsto 1 + \log \color{blue}{\left({\left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)}^{\left(\frac{v}{2}\right)}\right)} \]

   6. log-pow N/A

      \[\leadsto 1 + \color{blue}{\frac{v}{2} \cdot \log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)} \]

   7. lower-*.f32 N/A

      \[\leadsto 1 + \color{blue}{\frac{v}{2} \cdot \log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)} \]

   8. metadata-eval N/A

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

   9. lower-/.f32 N/A

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

   10. metadata-eval N/A

       \[\leadsto 1 + \frac{v}{\color{blue}{2}} \cdot \log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right) \]

   11. lower-log.f32 N/A

       \[\leadsto 1 + \frac{v}{2} \cdot \color{blue}{\log \left(\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \cdot \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)\right)} \]

   12. pow2 N/A

       \[\leadsto 1 + \frac{v}{2} \cdot \log \color{blue}{\left({\left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right)}^{2}\right)} \]

   13. metadata-eval N/A

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

4. Applied rewrites 99.5%

   \[\leadsto 1 + \color{blue}{\frac{v}{2} \cdot \log \left({\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)}^{2}\right)} \]
5. Step-by-step derivation

   1. lift-+.f32 N/A

      \[\leadsto \color{blue}{1 + \frac{v}{2} \cdot \log \left({\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)}^{2}\right)} \]

   2. +-commutative N/A

      \[\leadsto \color{blue}{\frac{v}{2} \cdot \log \left({\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)}^{2}\right) + 1} \]

6. Applied rewrites 99.3%

   \[\leadsto \color{blue}{\log \left({\left(\mathsf{fma}\left(1 - u, e^{\frac{-2}{v}}, u\right)\right)}^{v}\right) + 1} \]
7. Step-by-step derivation

   1. lift-log.f32 N/A

      \[\leadsto \color{blue}{\log \left({\left(\mathsf{fma}\left(1 - u, e^{\frac{-2}{v}}, u\right)\right)}^{v}\right)} + 1 \]

   2. lift-pow.f32 N/A

      \[\leadsto \log \color{blue}{\left({\left(\mathsf{fma}\left(1 - u, e^{\frac{-2}{v}}, u\right)\right)}^{v}\right)} + 1 \]

   3. log-pow N/A

      \[\leadsto \color{blue}{v \cdot \log \left(\mathsf{fma}\left(1 - u, e^{\frac{-2}{v}}, u\right)\right)} + 1 \]

   4. lift-fma.f32 N/A

      \[\leadsto v \cdot \log \color{blue}{\left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right)} + 1 \]

   5. lift-*.f32 N/A

      \[\leadsto v \cdot \log \left(\color{blue}{\left(1 - u\right) \cdot e^{\frac{-2}{v}}} + u\right) + 1 \]

   6. lift-+.f32 N/A

      \[\leadsto v \cdot \log \color{blue}{\left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right)} + 1 \]

   7. lift-log.f32 N/A

      \[\leadsto v \cdot \color{blue}{\log \left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right)} + 1 \]

   8. *-commutative N/A

      \[\leadsto \color{blue}{\log \left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right) \cdot v} + 1 \]

   9. lower-*.f32 99.4

      \[\leadsto \color{blue}{\log \left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right) \cdot v} + 1 \]

   10. lift-+.f32 N/A

       \[\leadsto \log \color{blue}{\left(\left(1 - u\right) \cdot e^{\frac{-2}{v}} + u\right)} \cdot v + 1 \]

   11. lift-*.f32 N/A

       \[\leadsto \log \left(\color{blue}{\left(1 - u\right) \cdot e^{\frac{-2}{v}}} + u\right) \cdot v + 1 \]

   12. *-commutative N/A

       \[\leadsto \log \left(\color{blue}{e^{\frac{-2}{v}} \cdot \left(1 - u\right)} + u\right) \cdot v + 1 \]

   13. lower-fma.f32 99.4

       \[\leadsto \log \color{blue}{\left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right)} \cdot v + 1 \]

8. Applied rewrites 99.4%

   \[\leadsto \color{blue}{\log \left(\mathsf{fma}\left(e^{\frac{-2}{v}}, 1 - u, u\right)\right) \cdot v} + 1 \]

9. Taylor expanded in u around 0

   \[\leadsto \log \left(\mathsf{fma}\left(e^{\frac{-2}{v}}, \color{blue}{1}, u\right)\right) \cdot v + 1 \]
10. Step-by-step derivation

11. Applied rewrites 94.6%

    \[\leadsto \log \left(\mathsf{fma}\left(e^{\frac{-2}{v}}, \color{blue}{1}, u\right)\right) \cdot v + 1 \]
12. Add Preprocessing

Alternative 15: 90.7% accurate, 8.0× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;v \leq 0.20000000298023224:\\ \;\;\;\;1\\ \mathbf{else}:\\ \;\;\;\;2 \cdot \left(\frac{u}{v} + u\right) - 1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<= v 0.20000000298023224) 1.0 (- (* 2.0 (+ (/ u v) u)) 1.0)))

C

float code(float u, float v) {
	float tmp;
	if (v <= 0.20000000298023224f) {
		tmp = 1.0f;
	} else {
		tmp = (2.0f * ((u / v) + u)) - 1.0f;
	}
	return tmp;
}

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(4) function code(u, v)
use fmin_fmax_functions
    real(4), intent (in) :: u
    real(4), intent (in) :: v
    real(4) :: tmp
    if (v <= 0.20000000298023224e0) then
        tmp = 1.0e0
    else
        tmp = (2.0e0 * ((u / v) + u)) - 1.0e0
    end if
    code = tmp
end function

Julia

function code(u, v)
	tmp = Float32(0.0)
	if (v <= Float32(0.20000000298023224))
		tmp = Float32(1.0);
	else
		tmp = Float32(Float32(Float32(2.0) * Float32(Float32(u / v) + u)) - Float32(1.0));
	end
	return tmp
end

MATLAB

function tmp_2 = code(u, v)
	tmp = single(0.0);
	if (v <= single(0.20000000298023224))
		tmp = single(1.0);
	else
		tmp = (single(2.0) * ((u / v) + u)) - single(1.0);
	end
	tmp_2 = tmp;
end

Derivation

1. Split input into 2 regimes

2. if v < 0.200000003

   1. Initial program 100.0%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in v around 0

      \[\leadsto \color{blue}{1} \]
   4. Step-by-step derivation

   5. Applied rewrites 93.3%

      \[\leadsto \color{blue}{1} \]

   if 0.200000003 < v

   1. Initial program 94.6%

      \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
   2. Add Preprocessing

   3. Taylor expanded in u around 0

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

      1. lower--.f32 N/A

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

      2. associate-*r* N/A

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

      3. lower-*.f32 N/A

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

      4. lower-*.f32 N/A

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

      5. rec-exp N/A

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

      6. distribute-neg-frac N/A

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

      7. metadata-eval N/A

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

      8. metadata-eval N/A

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

      9. associate-*r/ N/A

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

      10. lower-expm1.f32 N/A

          \[\leadsto \left(u \cdot v\right) \cdot \color{blue}{\mathsf{expm1}\left(2 \cdot \frac{1}{v}\right)} - 1 \]

      11. associate-*r/ N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2 \cdot 1}{v}}\right) - 1 \]

      12. metadata-eval N/A

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{\color{blue}{2}}{v}\right) - 1 \]

      13. lower-/.f32 69.0

          \[\leadsto \left(u \cdot v\right) \cdot \mathsf{expm1}\left(\color{blue}{\frac{2}{v}}\right) - 1 \]

   5. Applied rewrites 69.0%

      \[\leadsto \color{blue}{\left(u \cdot v\right) \cdot \mathsf{expm1}\left(\frac{2}{v}\right) - 1} \]

   6. Taylor expanded in v around inf

      \[\leadsto \left(2 \cdot u + 2 \cdot \frac{u}{v}\right) - 1 \]
   7. Step-by-step derivation

   8. Applied rewrites 59.8%

      \[\leadsto 2 \cdot \left(\frac{u}{v} + u\right) - 1 \]
3. Recombined 2 regimes into one program.
4. Add Preprocessing

Alternative 16: 5.7% accurate, 231.0× speedup

Math

\[\begin{array}{l} \\ -1 \end{array} \]

FPCore

(FPCore (u v) :precision binary32 -1.0)

C

float code(float u, float v) {
	return -1.0f;
}

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(4) function code(u, v)
use fmin_fmax_functions
    real(4), intent (in) :: u
    real(4), intent (in) :: v
    code = -1.0e0
end function

Julia

function code(u, v)
	return Float32(-1.0)
end

MATLAB

function tmp = code(u, v)
	tmp = single(-1.0);
end

Derivation

1. Initial program 99.4%

   \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
2. Add Preprocessing

3. Taylor expanded in u around 0

   \[\leadsto \color{blue}{-1} \]
4. Step-by-step derivation

5. Applied rewrites 7.6%

   \[\leadsto \color{blue}{-1} \]
6. Add Preprocessing

Reproduce

herbie shell --seed 2024352 
(FPCore (u v)
  :name "HairBSDF, sample_f, cosTheta"
  :precision binary32
  :pre (and (and (<= 1e-5 u) (<= u 1.0)) (and (<= 0.0 v) (<= v 109.746574)))
  (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v))))))))