HairBSDF, Mp, lower

Percentage Accurate: 99.6% → 99.6%
Time: 29.2s
Alternatives: 13
Speedup: 1.9×

Specification

?
\[\left(\left(\left(\left(-1 \leq cosTheta_i \land cosTheta_i \leq 1\right) \land \left(-1 \leq cosTheta_O \land cosTheta_O \leq 1\right)\right) \land \left(-1 \leq sinTheta_i \land sinTheta_i \leq 1\right)\right) \land \left(-1 \leq sinTheta_O \land sinTheta_O \leq 1\right)\right) \land \left(-1.5707964 \leq v \land v \leq 0.1\right)\]
\[\begin{array}{l} \\ e^{\left(\left(\left(\frac{cosTheta_i \cdot cosTheta_O}{v} - \frac{sinTheta_i \cdot sinTheta_O}{v}\right) - \frac{1}{v}\right) + 0.6931\right) + \log \left(\frac{1}{2 \cdot v}\right)} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (exp
  (+
   (+
    (-
     (- (/ (* cosTheta_i cosTheta_O) v) (/ (* sinTheta_i sinTheta_O) v))
     (/ 1.0 v))
    0.6931)
   (log (/ 1.0 (* 2.0 v))))))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return expf(((((((cosTheta_i * cosTheta_O) / v) - ((sinTheta_i * sinTheta_O) / v)) - (1.0f / v)) + 0.6931f) + logf((1.0f / (2.0f * v)))));
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = exp(((((((costheta_i * costheta_o) / v) - ((sintheta_i * sintheta_o) / v)) - (1.0e0 / v)) + 0.6931e0) + log((1.0e0 / (2.0e0 * v)))))
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return exp(Float32(Float32(Float32(Float32(Float32(Float32(cosTheta_i * cosTheta_O) / v) - Float32(Float32(sinTheta_i * sinTheta_O) / v)) - Float32(Float32(1.0) / v)) + Float32(0.6931)) + log(Float32(Float32(1.0) / Float32(Float32(2.0) * v)))))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = exp(((((((cosTheta_i * cosTheta_O) / v) - ((sinTheta_i * sinTheta_O) / v)) - (single(1.0) / v)) + single(0.6931)) + log((single(1.0) / (single(2.0) * v)))));
end

\begin{array}{l}

\\
e^{\left(\left(\left(\frac{cosTheta_i \cdot cosTheta_O}{v} - \frac{sinTheta_i \cdot sinTheta_O}{v}\right) - \frac{1}{v}\right) + 0.6931\right) + \log \left(\frac{1}{2 \cdot v}\right)}
\end{array}

Sampling outcomes in binary32 precision:

Local Percentage Accuracy vs ?

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

Accuracy vs Speed?

Herbie found 13 alternatives:

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

Initial Program: 99.6% accurate, 1.0× speedup?

\[\begin{array}{l} \\ e^{\left(\left(\left(\frac{cosTheta_i \cdot cosTheta_O}{v} - \frac{sinTheta_i \cdot sinTheta_O}{v}\right) - \frac{1}{v}\right) + 0.6931\right) + \log \left(\frac{1}{2 \cdot v}\right)} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (exp
  (+
   (+
    (-
     (- (/ (* cosTheta_i cosTheta_O) v) (/ (* sinTheta_i sinTheta_O) v))
     (/ 1.0 v))
    0.6931)
   (log (/ 1.0 (* 2.0 v))))))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return expf(((((((cosTheta_i * cosTheta_O) / v) - ((sinTheta_i * sinTheta_O) / v)) - (1.0f / v)) + 0.6931f) + logf((1.0f / (2.0f * v)))));
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = exp(((((((costheta_i * costheta_o) / v) - ((sintheta_i * sintheta_o) / v)) - (1.0e0 / v)) + 0.6931e0) + log((1.0e0 / (2.0e0 * v)))))
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return exp(Float32(Float32(Float32(Float32(Float32(Float32(cosTheta_i * cosTheta_O) / v) - Float32(Float32(sinTheta_i * sinTheta_O) / v)) - Float32(Float32(1.0) / v)) + Float32(0.6931)) + log(Float32(Float32(1.0) / Float32(Float32(2.0) * v)))))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = exp(((((((cosTheta_i * cosTheta_O) / v) - ((sinTheta_i * sinTheta_O) / v)) - (single(1.0) / v)) + single(0.6931)) + log((single(1.0) / (single(2.0) * v)))));
end

\begin{array}{l}

\\
e^{\left(\left(\left(\frac{cosTheta_i \cdot cosTheta_O}{v} - \frac{sinTheta_i \cdot sinTheta_O}{v}\right) - \frac{1}{v}\right) + 0.6931\right) + \log \left(\frac{1}{2 \cdot v}\right)}
\end{array}

Alternative 1: 99.6% accurate, 0.4× speedup?

\[\begin{array}{l} \\ {\left({\left({\left(e^{\mathsf{fma}\left(\frac{cosTheta_i}{v}, cosTheta_O, -\mathsf{fma}\left(\frac{sinTheta_i}{v}, sinTheta_O, \frac{1}{v}\right)\right) + 0.6931} \cdot \frac{0.5}{v}\right)}^{0.16666666666666666}\right)}^{2}\right)}^{3} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (pow
  (pow
   (pow
    (*
     (exp
      (+
       (fma
        (/ cosTheta_i v)
        cosTheta_O
        (- (fma (/ sinTheta_i v) sinTheta_O (/ 1.0 v))))
       0.6931))
     (/ 0.5 v))
    0.16666666666666666)
   2.0)
  3.0))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return powf(powf(powf((expf((fmaf((cosTheta_i / v), cosTheta_O, -fmaf((sinTheta_i / v), sinTheta_O, (1.0f / v))) + 0.6931f)) * (0.5f / v)), 0.16666666666666666f), 2.0f), 3.0f);
}

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return ((Float32(exp(Float32(fma(Float32(cosTheta_i / v), cosTheta_O, Float32(-fma(Float32(sinTheta_i / v), sinTheta_O, Float32(Float32(1.0) / v)))) + Float32(0.6931))) * Float32(Float32(0.5) / v)) ^ Float32(0.16666666666666666)) ^ Float32(2.0)) ^ Float32(3.0)
end

\begin{array}{l}

\\
{\left({\left({\left(e^{\mathsf{fma}\left(\frac{cosTheta_i}{v}, cosTheta_O, -\mathsf{fma}\left(\frac{sinTheta_i}{v}, sinTheta_O, \frac{1}{v}\right)\right) + 0.6931} \cdot \frac{0.5}{v}\right)}^{0.16666666666666666}\right)}^{2}\right)}^{3}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 2: 99.7% accurate, 0.5× speedup?

\[\begin{array}{l} \\ {\left({\left({\left(\frac{0.5 \cdot e^{0.6931 + \frac{-1}{v}}}{v}\right)}^{0.16666666666666666}\right)}^{2}\right)}^{3} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (pow
  (pow (pow (/ (* 0.5 (exp (+ 0.6931 (/ -1.0 v)))) v) 0.16666666666666666) 2.0)
  3.0))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return powf(powf(powf(((0.5f * expf((0.6931f + (-1.0f / v)))) / v), 0.16666666666666666f), 2.0f), 3.0f);
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = ((((0.5e0 * exp((0.6931e0 + ((-1.0e0) / v)))) / v) ** 0.16666666666666666e0) ** 2.0e0) ** 3.0e0
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return ((Float32(Float32(Float32(0.5) * exp(Float32(Float32(0.6931) + Float32(Float32(-1.0) / v)))) / v) ^ Float32(0.16666666666666666)) ^ Float32(2.0)) ^ Float32(3.0)
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = ((((single(0.5) * exp((single(0.6931) + (single(-1.0) / v)))) / v) ^ single(0.16666666666666666)) ^ single(2.0)) ^ single(3.0);
end

\begin{array}{l}

\\
{\left({\left({\left(\frac{0.5 \cdot e^{0.6931 + \frac{-1}{v}}}{v}\right)}^{0.16666666666666666}\right)}^{2}\right)}^{3}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 3: 99.6% accurate, 0.7× speedup?

\[\begin{array}{l} \\ \frac{0.5}{v} \cdot \left(e^{0.6931} \cdot e^{\frac{-1 + cosTheta_i \cdot cosTheta_O}{v}} - \frac{e^{0.6931 + \frac{-1}{v}} \cdot \left(sinTheta_i \cdot sinTheta_O\right)}{v}\right) \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (*
  (/ 0.5 v)
  (-
   (* (exp 0.6931) (exp (/ (+ -1.0 (* cosTheta_i cosTheta_O)) v)))
   (/ (* (exp (+ 0.6931 (/ -1.0 v))) (* sinTheta_i sinTheta_O)) v))))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return (0.5f / v) * ((expf(0.6931f) * expf(((-1.0f + (cosTheta_i * cosTheta_O)) / v))) - ((expf((0.6931f + (-1.0f / v))) * (sinTheta_i * sinTheta_O)) / v));
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = (0.5e0 / v) * ((exp(0.6931e0) * exp((((-1.0e0) + (costheta_i * costheta_o)) / v))) - ((exp((0.6931e0 + ((-1.0e0) / v))) * (sintheta_i * sintheta_o)) / v))
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return Float32(Float32(Float32(0.5) / v) * Float32(Float32(exp(Float32(0.6931)) * exp(Float32(Float32(Float32(-1.0) + Float32(cosTheta_i * cosTheta_O)) / v))) - Float32(Float32(exp(Float32(Float32(0.6931) + Float32(Float32(-1.0) / v))) * Float32(sinTheta_i * sinTheta_O)) / v)))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = (single(0.5) / v) * ((exp(single(0.6931)) * exp(((single(-1.0) + (cosTheta_i * cosTheta_O)) / v))) - ((exp((single(0.6931) + (single(-1.0) / v))) * (sinTheta_i * sinTheta_O)) / v));
end

\begin{array}{l}

\\
\frac{0.5}{v} \cdot \left(e^{0.6931} \cdot e^{\frac{-1 + cosTheta_i \cdot cosTheta_O}{v}} - \frac{e^{0.6931 + \frac{-1}{v}} \cdot \left(sinTheta_i \cdot sinTheta_O\right)}{v}\right)
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 4: 99.7% accurate, 0.7× speedup?

\[\begin{array}{l} \\ 0.5 \cdot \frac{e^{0.6931} \cdot e^{\frac{-1}{v}} - \frac{sinTheta_O \cdot \left(sinTheta_i \cdot e^{0.6931 + \frac{-1}{v}}\right)}{v}}{v} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (*
  0.5
  (/
   (-
    (* (exp 0.6931) (exp (/ -1.0 v)))
    (/ (* sinTheta_O (* sinTheta_i (exp (+ 0.6931 (/ -1.0 v))))) v))
   v)))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return 0.5f * (((expf(0.6931f) * expf((-1.0f / v))) - ((sinTheta_O * (sinTheta_i * expf((0.6931f + (-1.0f / v))))) / v)) / v);
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = 0.5e0 * (((exp(0.6931e0) * exp(((-1.0e0) / v))) - ((sintheta_o * (sintheta_i * exp((0.6931e0 + ((-1.0e0) / v))))) / v)) / v)
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return Float32(Float32(0.5) * Float32(Float32(Float32(exp(Float32(0.6931)) * exp(Float32(Float32(-1.0) / v))) - Float32(Float32(sinTheta_O * Float32(sinTheta_i * exp(Float32(Float32(0.6931) + Float32(Float32(-1.0) / v))))) / v)) / v))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = single(0.5) * (((exp(single(0.6931)) * exp((single(-1.0) / v))) - ((sinTheta_O * (sinTheta_i * exp((single(0.6931) + (single(-1.0) / v))))) / v)) / v);
end

\begin{array}{l}

\\
0.5 \cdot \frac{e^{0.6931} \cdot e^{\frac{-1}{v}} - \frac{sinTheta_O \cdot \left(sinTheta_i \cdot e^{0.6931 + \frac{-1}{v}}\right)}{v}}{v}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 5: 99.6% accurate, 0.7× speedup?

\[\begin{array}{l} \\ {\left(\sqrt[3]{\frac{0.5}{v} \cdot e^{\left(0.6931 + \frac{cosTheta_i \cdot cosTheta_O}{v}\right) + \frac{-1}{v}}}\right)}^{3} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (pow
  (cbrt
   (*
    (/ 0.5 v)
    (exp (+ (+ 0.6931 (/ (* cosTheta_i cosTheta_O) v)) (/ -1.0 v)))))
  3.0))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return powf(cbrtf(((0.5f / v) * expf(((0.6931f + ((cosTheta_i * cosTheta_O) / v)) + (-1.0f / v))))), 3.0f);
}

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return cbrt(Float32(Float32(Float32(0.5) / v) * exp(Float32(Float32(Float32(0.6931) + Float32(Float32(cosTheta_i * cosTheta_O) / v)) + Float32(Float32(-1.0) / v))))) ^ Float32(3.0)
end

\begin{array}{l}

\\
{\left(\sqrt[3]{\frac{0.5}{v} \cdot e^{\left(0.6931 + \frac{cosTheta_i \cdot cosTheta_O}{v}\right) + \frac{-1}{v}}}\right)}^{3}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 6: 99.6% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \begin{array}{l} t_0 := e^{0.6931 + \frac{-1}{v}}\\ \frac{0.5}{v} \cdot \left(t_0 - \frac{t_0 \cdot \left(sinTheta_i \cdot sinTheta_O\right)}{v}\right) \end{array} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (let* ((t_0 (exp (+ 0.6931 (/ -1.0 v)))))
   (* (/ 0.5 v) (- t_0 (/ (* t_0 (* sinTheta_i sinTheta_O)) v)))))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	float t_0 = expf((0.6931f + (-1.0f / v)));
	return (0.5f / v) * (t_0 - ((t_0 * (sinTheta_i * sinTheta_O)) / v));
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    real(4) :: t_0
    t_0 = exp((0.6931e0 + ((-1.0e0) / v)))
    code = (0.5e0 / v) * (t_0 - ((t_0 * (sintheta_i * sintheta_o)) / v))
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	t_0 = exp(Float32(Float32(0.6931) + Float32(Float32(-1.0) / v)))
	return Float32(Float32(Float32(0.5) / v) * Float32(t_0 - Float32(Float32(t_0 * Float32(sinTheta_i * sinTheta_O)) / v)))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	t_0 = exp((single(0.6931) + (single(-1.0) / v)));
	tmp = (single(0.5) / v) * (t_0 - ((t_0 * (sinTheta_i * sinTheta_O)) / v));
end

\begin{array}{l}

\\
\begin{array}{l}
t_0 := e^{0.6931 + \frac{-1}{v}}\\
\frac{0.5}{v} \cdot \left(t_0 - \frac{t_0 \cdot \left(sinTheta_i \cdot sinTheta_O\right)}{v}\right)
\end{array}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 7: 99.7% accurate, 1.0× speedup?

\[\begin{array}{l} \\ 0.5 \cdot \frac{e^{0.6931} \cdot e^{\frac{-1 + cosTheta_i \cdot cosTheta_O}{v}}}{v} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (* 0.5 (/ (* (exp 0.6931) (exp (/ (+ -1.0 (* cosTheta_i cosTheta_O)) v))) v)))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return 0.5f * ((expf(0.6931f) * expf(((-1.0f + (cosTheta_i * cosTheta_O)) / v))) / v);
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = 0.5e0 * ((exp(0.6931e0) * exp((((-1.0e0) + (costheta_i * costheta_o)) / v))) / v)
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return Float32(Float32(0.5) * Float32(Float32(exp(Float32(0.6931)) * exp(Float32(Float32(Float32(-1.0) + Float32(cosTheta_i * cosTheta_O)) / v))) / v))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = single(0.5) * ((exp(single(0.6931)) * exp(((single(-1.0) + (cosTheta_i * cosTheta_O)) / v))) / v);
end

\begin{array}{l}

\\
0.5 \cdot \frac{e^{0.6931} \cdot e^{\frac{-1 + cosTheta_i \cdot cosTheta_O}{v}}}{v}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 8: 99.6% accurate, 1.9× speedup?

\[\begin{array}{l} \\ \frac{0.5}{v} \cdot e^{\left(0.6931 + \frac{cosTheta_i \cdot cosTheta_O}{v}\right) + \frac{-1}{v}} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (* (/ 0.5 v) (exp (+ (+ 0.6931 (/ (* cosTheta_i cosTheta_O) v)) (/ -1.0 v)))))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return (0.5f / v) * expf(((0.6931f + ((cosTheta_i * cosTheta_O) / v)) + (-1.0f / v)));
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = (0.5e0 / v) * exp(((0.6931e0 + ((costheta_i * costheta_o) / v)) + ((-1.0e0) / v)))
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return Float32(Float32(Float32(0.5) / v) * exp(Float32(Float32(Float32(0.6931) + Float32(Float32(cosTheta_i * cosTheta_O) / v)) + Float32(Float32(-1.0) / v))))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = (single(0.5) / v) * exp(((single(0.6931) + ((cosTheta_i * cosTheta_O) / v)) + (single(-1.0) / v)));
end

\begin{array}{l}

\\
\frac{0.5}{v} \cdot e^{\left(0.6931 + \frac{cosTheta_i \cdot cosTheta_O}{v}\right) + \frac{-1}{v}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 9: 37.6% accurate, 2.1× speedup?

\[\begin{array}{l} \\ \begin{array}{l} \mathbf{if}\;sinTheta_i \leq 5.219999894930545 \cdot 10^{-24}:\\ \;\;\;\;\frac{sinTheta_i \cdot \left(-sinTheta_O\right)}{v}\\ \mathbf{else}:\\ \;\;\;\;e^{\frac{sinTheta_i}{v} \cdot \left(-sinTheta_O\right)}\\ \end{array} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (if (<= sinTheta_i 5.219999894930545e-24)
   (/ (* sinTheta_i (- sinTheta_O)) v)
   (exp (* (/ sinTheta_i v) (- sinTheta_O)))))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	float tmp;
	if (sinTheta_i <= 5.219999894930545e-24f) {
		tmp = (sinTheta_i * -sinTheta_O) / v;
	} else {
		tmp = expf(((sinTheta_i / v) * -sinTheta_O));
	}
	return tmp;
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    real(4) :: tmp
    if (sintheta_i <= 5.219999894930545e-24) then
        tmp = (sintheta_i * -sintheta_o) / v
    else
        tmp = exp(((sintheta_i / v) * -sintheta_o))
    end if
    code = tmp
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = Float32(0.0)
	if (sinTheta_i <= Float32(5.219999894930545e-24))
		tmp = Float32(Float32(sinTheta_i * Float32(-sinTheta_O)) / v);
	else
		tmp = exp(Float32(Float32(sinTheta_i / v) * Float32(-sinTheta_O)));
	end
	return tmp
end

function tmp_2 = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = single(0.0);
	if (sinTheta_i <= single(5.219999894930545e-24))
		tmp = (sinTheta_i * -sinTheta_O) / v;
	else
		tmp = exp(((sinTheta_i / v) * -sinTheta_O));
	end
	tmp_2 = tmp;
end

\begin{array}{l}

\\
\begin{array}{l}
\mathbf{if}\;sinTheta_i \leq 5.219999894930545 \cdot 10^{-24}:\\
\;\;\;\;\frac{sinTheta_i \cdot \left(-sinTheta_O\right)}{v}\\

\mathbf{else}:\\
\;\;\;\;e^{\frac{sinTheta_i}{v} \cdot \left(-sinTheta_O\right)}\\


\end{array}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 10: 97.8% accurate, 2.1× speedup?

\[\begin{array}{l} \\ e^{\frac{-1 + cosTheta_i \cdot cosTheta_O}{v}} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (exp (/ (+ -1.0 (* cosTheta_i cosTheta_O)) v)))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return expf(((-1.0f + (cosTheta_i * cosTheta_O)) / v));
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = exp((((-1.0e0) + (costheta_i * costheta_o)) / v))
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return exp(Float32(Float32(Float32(-1.0) + Float32(cosTheta_i * cosTheta_O)) / v))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = exp(((single(-1.0) + (cosTheta_i * cosTheta_O)) / v));
end

\begin{array}{l}

\\
e^{\frac{-1 + cosTheta_i \cdot cosTheta_O}{v}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 11: 20.6% accurate, 37.2× speedup?

\[\begin{array}{l} \\ \frac{sinTheta_i}{v} \cdot \left(-sinTheta_O\right) \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (* (/ sinTheta_i v) (- sinTheta_O)))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return (sinTheta_i / v) * -sinTheta_O;
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = (sintheta_i / v) * -sintheta_o
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return Float32(Float32(sinTheta_i / v) * Float32(-sinTheta_O))
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = (sinTheta_i / v) * -sinTheta_O;
end

\begin{array}{l}

\\
\frac{sinTheta_i}{v} \cdot \left(-sinTheta_O\right)
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 12: 38.9% accurate, 37.2× speedup?

\[\begin{array}{l} \\ \frac{sinTheta_i \cdot \left(-sinTheta_O\right)}{v} \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 (/ (* sinTheta_i (- sinTheta_O)) v))
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return (sinTheta_i * -sinTheta_O) / v;
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = (sintheta_i * -sintheta_o) / v
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return Float32(Float32(sinTheta_i * Float32(-sinTheta_O)) / v)
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = (sinTheta_i * -sinTheta_O) / v;
end

\begin{array}{l}

\\
\frac{sinTheta_i \cdot \left(-sinTheta_O\right)}{v}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 13: 6.4% accurate, 223.0× speedup?

\[\begin{array}{l} \\ 1 \end{array} \]
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
 :precision binary32
 1.0)
float code(float cosTheta_i, float cosTheta_O, float sinTheta_i, float sinTheta_O, float v) {
	return 1.0f;
}

real(4) function code(costheta_i, costheta_o, sintheta_i, sintheta_o, v)
    real(4), intent (in) :: costheta_i
    real(4), intent (in) :: costheta_o
    real(4), intent (in) :: sintheta_i
    real(4), intent (in) :: sintheta_o
    real(4), intent (in) :: v
    code = 1.0e0
end function

function code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	return Float32(1.0)
end

function tmp = code(cosTheta_i, cosTheta_O, sinTheta_i, sinTheta_O, v)
	tmp = single(1.0);
end

\begin{array}{l}

\\
1
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Reproduce

?
herbie shell --seed 2023347 
(FPCore (cosTheta_i cosTheta_O sinTheta_i sinTheta_O v)
  :name "HairBSDF, Mp, lower"
  :precision binary32
  :pre (and (and (and (and (and (<= -1.0 cosTheta_i) (<= cosTheta_i 1.0)) (and (<= -1.0 cosTheta_O) (<= cosTheta_O 1.0))) (and (<= -1.0 sinTheta_i) (<= sinTheta_i 1.0))) (and (<= -1.0 sinTheta_O) (<= sinTheta_O 1.0))) (and (<= -1.5707964 v) (<= v 0.1)))
  (exp (+ (+ (- (- (/ (* cosTheta_i cosTheta_O) v) (/ (* sinTheta_i sinTheta_O) v)) (/ 1.0 v)) 0.6931) (log (/ 1.0 (* 2.0 v))))))