HairBSDF, sample_f, cosTheta

Percentage Accurate: 99.5% → 99.5%

Time: 3.6s

Alternatives: 9

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, 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

Derivation

1. Initial program 99.5%

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

Alternative 2: 99.5% accurate, 1.0× speedup

Math

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

FPCore

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

C

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

Julia

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

Derivation

1. Initial program 99.5%

   \[1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \]
2. 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. lift-*.f32 N/A

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

   3. lift-log.f32 N/A

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

   4. lift-+.f32 N/A

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

   5. lift-*.f32 N/A

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

   6. lift--.f32 N/A

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

   7. lift-/.f32 N/A

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

   8. lift-exp.f32 N/A

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

   9. +-commutative N/A

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

   10. lower-fma.f32 N/A

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

3. Applied rewrites 99.5%

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

Alternative 3: 96.1% accurate, 1.2× speedup

Math

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

FPCore

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

C

float code(float u, float v) {
	return 1.0f + (v * logf((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 + exp(((-2.0e0) / v)))))
end function

Julia

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

MATLAB

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

Derivation

1. Initial program 99.5%

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

2. Taylor expanded in u around 0

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

   1. lift-exp.f32 N/A

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

   2. lift-/.f32 96.1

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

4. Applied rewrites 96.1%

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

Alternative 4: 94.4% accurate, 1.3× speedup

Math

\[\begin{array}{l} \\ \mathsf{fma}\left(\log \left(\mathsf{expm1}\left(\frac{-2}{v}\right) \cdot \left(-u\right)\right), v, 1\right) \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (fma (log (* (expm1 (/ -2.0 v)) (- u))) v 1.0))

C

float code(float u, float v) {
	return fmaf(logf((expm1f((-2.0f / v)) * -u)), v, 1.0f);
}

Julia

function code(u, v)
	return fma(log(Float32(expm1(Float32(Float32(-2.0) / v)) * Float32(-u))), v, Float32(1.0))
end

Derivation

1. Initial program 99.5%

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

2. Taylor expanded in u around -inf

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

   1. +-commutative N/A

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

   2. *-commutative N/A

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

   3. lower-fma.f32 N/A

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

4. Applied rewrites 94.4%

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

Alternative 5: 91.0% accurate, 0.6× speedup

Math

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

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      0.20000000298023224)
   (fma (* (expm1 (/ 2.0 v)) 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))))))) <= 0.20000000298023224f) {
		tmp = fmaf((expm1f((2.0f / v)) * 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(0.20000000298023224))
		tmp = fma(Float32(expm1(Float32(Float32(2.0) / v)) * 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))))))) < 0.200000003

   1. Initial program 93.1%

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

   2. Taylor expanded in u around 0

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

      1. negate-sub N/A

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

      2. *-commutative N/A

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

      3. metadata-eval N/A

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

      4. lower-fma.f32 N/A

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

      5. *-commutative N/A

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

      6. lower-*.f32 N/A

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

      7. rec-exp N/A

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

      8. lower-expm1.f32 N/A

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

      9. lower-neg.f32 N/A

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

      10. lift-/.f32 65.6

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

   4. Applied rewrites 65.6%

      \[\leadsto \color{blue}{\mathsf{fma}\left(\mathsf{expm1}\left(-\frac{-2}{v}\right) \cdot v, u, -1\right)} \]
   5. Step-by-step derivation

      1. lift-/.f32 N/A

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

      2. lift-neg.f32 N/A

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

      3. distribute-neg-frac N/A

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

      4. metadata-eval N/A

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

      5. lower-/.f32 65.6

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

   6. Applied rewrites 65.6%

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

   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. Taylor expanded in v around 0

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

   4. Applied rewrites 92.9%

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

Alternative 6: 90.8% accurate, 0.6× speedup

Math

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;1 + v \cdot \log \left(u + \left(1 - u\right) \cdot e^{\frac{-2}{v}}\right) \leq -0.20000000298023224:\\ \;\;\;\;\mathsf{fma}\left(\frac{2 - \frac{-1.3333333333333333}{v}}{v} + 2, u, -1\right)\\ \mathbf{else}:\\ \;\;\;\;1\\ \end{array} \end{array} \]

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      -0.20000000298023224)
   (fma (+ (/ (- 2.0 (/ -1.3333333333333333 v)) v) 2.0) 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 = fmaf((((2.0f - (-1.3333333333333333f / v)) / v) + 2.0f), 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(-0.20000000298023224))
		tmp = fma(Float32(Float32(Float32(Float32(2.0) - Float32(Float32(-1.3333333333333333) / v)) / v) + Float32(2.0)), 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))))))) < -0.200000003

   1. Initial program 93.6%

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

   2. Taylor expanded in u around 0

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

      1. negate-sub N/A

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

      2. *-commutative N/A

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

      3. metadata-eval N/A

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

      4. lower-fma.f32 N/A

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

      5. *-commutative N/A

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

      6. lower-*.f32 N/A

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

      7. rec-exp N/A

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

      8. lower-expm1.f32 N/A

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

      9. lower-neg.f32 N/A

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

      10. lift-/.f32 71.1

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

   4. Applied rewrites 71.1%

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

   5. Taylor expanded in v around -inf

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

      1. +-commutative N/A

         \[\leadsto \mathsf{fma}\left(-1 \cdot \frac{-1 \cdot \frac{\frac{4}{3} + \frac{2}{3} \cdot \frac{1}{v}}{v} - 2}{v} + 2, u, -1\right) \]

      2. lower-+.f32 N/A

         \[\leadsto \mathsf{fma}\left(-1 \cdot \frac{-1 \cdot \frac{\frac{4}{3} + \frac{2}{3} \cdot \frac{1}{v}}{v} - 2}{v} + 2, u, -1\right) \]

   7. Applied rewrites 69.6%

      \[\leadsto \mathsf{fma}\left(\left(-\frac{\left(-\frac{\frac{0.6666666666666666}{v} + 1.3333333333333333}{v}\right) - 2}{v}\right) + 2, u, -1\right) \]

   8. Taylor expanded in v around inf

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

      1. lower-/.f32 N/A

         \[\leadsto \mathsf{fma}\left(\frac{2 + \frac{4}{3} \cdot \frac{1}{v}}{v} + 2, u, -1\right) \]

      2. fp-cancel-sign-sub-inv N/A

         \[\leadsto \mathsf{fma}\left(\frac{2 - \left(\mathsf{neg}\left(\frac{4}{3}\right)\right) \cdot \frac{1}{v}}{v} + 2, u, -1\right) \]

      3. metadata-eval N/A

         \[\leadsto \mathsf{fma}\left(\frac{2 - \frac{-4}{3} \cdot \frac{1}{v}}{v} + 2, u, -1\right) \]

      4. mult-flip N/A

         \[\leadsto \mathsf{fma}\left(\frac{2 - \frac{\frac{-4}{3}}{v}}{v} + 2, u, -1\right) \]

      5. lower--.f32 N/A

         \[\leadsto \mathsf{fma}\left(\frac{2 - \frac{\frac{-4}{3}}{v}}{v} + 2, u, -1\right) \]

      6. lower-/.f32 67.8

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

   10. Applied rewrites 67.8%

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

   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 99.9%

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

   2. Taylor expanded in v around 0

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

   4. Applied rewrites 92.3%

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

Alternative 7: 90.5% accurate, 0.7× speedup

Math

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

FPCore

(FPCore (u v)
 :precision binary32
 (if (<=
      (+ 1.0 (* v (log (+ u (* (- 1.0 u) (exp (/ -2.0 v)))))))
      -0.20000000298023224)
   (fma (- 2.0 (/ -2.0 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))))))) <= -0.20000000298023224f) {
		tmp = fmaf((2.0f - (-2.0f / 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(-0.20000000298023224))
		tmp = fma(Float32(Float32(2.0) - Float32(Float32(-2.0) / 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))))))) < -0.200000003

   1. Initial program 93.6%

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

   2. Taylor expanded in u around 0

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

      1. negate-sub N/A

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

      2. *-commutative N/A

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

      3. metadata-eval N/A

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

      4. lower-fma.f32 N/A

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

      5. *-commutative N/A

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

      6. lower-*.f32 N/A

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

      7. rec-exp N/A

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

      8. lower-expm1.f32 N/A

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

      9. lower-neg.f32 N/A

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

      10. lift-/.f32 71.1

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

   4. Applied rewrites 71.1%

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

   5. Taylor expanded in v around inf

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

      1. fp-cancel-sign-sub-inv N/A

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

      2. metadata-eval N/A

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

      3. mult-flip N/A

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

      4. lower--.f32 N/A

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

      5. lift-/.f32 63.6

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

   7. Applied rewrites 63.6%

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

   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 99.9%

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

   2. Taylor expanded in v around 0

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

   4. Applied rewrites 92.3%

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

Alternative 8: 86.9% accurate, 34.9× 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 99.5%

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

2. Taylor expanded in v around 0

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

4. Applied rewrites 86.9%

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

Alternative 9: 5.9% accurate, 34.9× 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 99.5%

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

2. Taylor expanded in u around 0

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

4. Applied rewrites 5.9%

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

Reproduce

herbie shell --seed 2025120 
(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))))))))