Beckmann Sample, normalization factor

Percentage Accurate: 97.8% → 98.4%
Time: 14.3s
Alternatives: 14
Speedup: 1.0×

Specification

?
\[\left(0 < cosTheta \land cosTheta < 0.9999\right) \land \left(-1 < c \land c < 1\right)\]
\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + \left(\frac{1}{\sqrt{\pi}} \cdot \frac{\sqrt{\left(1 - cosTheta\right) - cosTheta}}{cosTheta}\right) \cdot e^{\left(-cosTheta\right) \cdot cosTheta}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (* (/ 1.0 (sqrt PI)) (/ (sqrt (- (- 1.0 cosTheta) cosTheta)) cosTheta))
    (exp (* (- cosTheta) cosTheta))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (((1.0f / sqrtf(((float) M_PI))) * (sqrtf(((1.0f - cosTheta) - cosTheta)) / cosTheta)) * expf((-cosTheta * cosTheta))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(Float32(Float32(Float32(1.0) / sqrt(Float32(pi))) * Float32(sqrt(Float32(Float32(Float32(1.0) - cosTheta) - cosTheta)) / cosTheta)) * exp(Float32(Float32(-cosTheta) * cosTheta)))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + (((single(1.0) / sqrt(single(pi))) * (sqrt(((single(1.0) - cosTheta) - cosTheta)) / cosTheta)) * exp((-cosTheta * cosTheta))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + \left(\frac{1}{\sqrt{\pi}} \cdot \frac{\sqrt{\left(1 - cosTheta\right) - cosTheta}}{cosTheta}\right) \cdot e^{\left(-cosTheta\right) \cdot cosTheta}}
\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 14 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: 97.8% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + \left(\frac{1}{\sqrt{\pi}} \cdot \frac{\sqrt{\left(1 - cosTheta\right) - cosTheta}}{cosTheta}\right) \cdot e^{\left(-cosTheta\right) \cdot cosTheta}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (* (/ 1.0 (sqrt PI)) (/ (sqrt (- (- 1.0 cosTheta) cosTheta)) cosTheta))
    (exp (* (- cosTheta) cosTheta))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (((1.0f / sqrtf(((float) M_PI))) * (sqrtf(((1.0f - cosTheta) - cosTheta)) / cosTheta)) * expf((-cosTheta * cosTheta))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(Float32(Float32(Float32(1.0) / sqrt(Float32(pi))) * Float32(sqrt(Float32(Float32(Float32(1.0) - cosTheta) - cosTheta)) / cosTheta)) * exp(Float32(Float32(-cosTheta) * cosTheta)))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + (((single(1.0) / sqrt(single(pi))) * (sqrt(((single(1.0) - cosTheta) - cosTheta)) / cosTheta)) * exp((-cosTheta * cosTheta))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + \left(\frac{1}{\sqrt{\pi}} \cdot \frac{\sqrt{\left(1 - cosTheta\right) - cosTheta}}{cosTheta}\right) \cdot e^{\left(-cosTheta\right) \cdot cosTheta}}
\end{array}

Alternative 1: 98.4% accurate, 0.8× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + \frac{1}{\frac{\sqrt{\pi}}{\frac{\sqrt{\mathsf{fma}\left(cosTheta, -2, 1\right)}}{cosTheta}}} \cdot e^{cosTheta \cdot \left(-cosTheta\right)}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (/ 1.0 (/ (sqrt PI) (/ (sqrt (fma cosTheta -2.0 1.0)) cosTheta)))
    (exp (* cosTheta (- cosTheta)))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + ((1.0f / (sqrtf(((float) M_PI)) / (sqrtf(fmaf(cosTheta, -2.0f, 1.0f)) / cosTheta))) * expf((cosTheta * -cosTheta))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(Float32(Float32(1.0) / Float32(sqrt(Float32(pi)) / Float32(sqrt(fma(cosTheta, Float32(-2.0), Float32(1.0))) / cosTheta))) * exp(Float32(cosTheta * Float32(-cosTheta))))))
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + \frac{1}{\frac{\sqrt{\pi}}{\frac{\sqrt{\mathsf{fma}\left(cosTheta, -2, 1\right)}}{cosTheta}}} \cdot e^{cosTheta \cdot \left(-cosTheta\right)}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 2: 98.5% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{cosTheta \cdot \sqrt{\frac{\pi}{\mathsf{fma}\left(cosTheta, -2, 1\right)}}}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (exp (* cosTheta (- cosTheta)))
    (/ 1.0 (* cosTheta (sqrt (/ PI (fma cosTheta -2.0 1.0)))))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (expf((cosTheta * -cosTheta)) * (1.0f / (cosTheta * sqrtf((((float) M_PI) / fmaf(cosTheta, -2.0f, 1.0f)))))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(exp(Float32(cosTheta * Float32(-cosTheta))) * Float32(Float32(1.0) / Float32(cosTheta * sqrt(Float32(Float32(pi) / fma(cosTheta, Float32(-2.0), Float32(1.0)))))))))
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{cosTheta \cdot \sqrt{\frac{\pi}{\mathsf{fma}\left(cosTheta, -2, 1\right)}}}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 3: 98.5% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + \frac{\frac{\sqrt{1 + cosTheta \cdot -2}}{\sqrt{\pi} \cdot cosTheta}}{e^{cosTheta \cdot cosTheta}}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (/
    (/ (sqrt (+ 1.0 (* cosTheta -2.0))) (* (sqrt PI) cosTheta))
    (exp (* cosTheta cosTheta))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + ((sqrtf((1.0f + (cosTheta * -2.0f))) / (sqrtf(((float) M_PI)) * cosTheta)) / expf((cosTheta * cosTheta))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(Float32(sqrt(Float32(Float32(1.0) + Float32(cosTheta * Float32(-2.0)))) / Float32(sqrt(Float32(pi)) * cosTheta)) / exp(Float32(cosTheta * cosTheta)))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + ((sqrt((single(1.0) + (cosTheta * single(-2.0)))) / (sqrt(single(pi)) * cosTheta)) / exp((cosTheta * cosTheta))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + \frac{\frac{\sqrt{1 + cosTheta \cdot -2}}{\sqrt{\pi} \cdot cosTheta}}{e^{cosTheta \cdot cosTheta}}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 4: 98.0% accurate, 1.0× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + \frac{\frac{\sqrt{\frac{\mathsf{fma}\left(cosTheta, -2, 1\right)}{\pi}}}{cosTheta}}{e^{cosTheta \cdot cosTheta}}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (/
    (/ (sqrt (/ (fma cosTheta -2.0 1.0) PI)) cosTheta)
    (exp (* cosTheta cosTheta))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + ((sqrtf((fmaf(cosTheta, -2.0f, 1.0f) / ((float) M_PI))) / cosTheta) / expf((cosTheta * cosTheta))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(Float32(sqrt(Float32(fma(cosTheta, Float32(-2.0), Float32(1.0)) / Float32(pi))) / cosTheta) / exp(Float32(cosTheta * cosTheta)))))
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + \frac{\frac{\sqrt{\frac{\mathsf{fma}\left(cosTheta, -2, 1\right)}{\pi}}}{cosTheta}}{e^{cosTheta \cdot cosTheta}}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 5: 96.8% accurate, 1.3× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{\frac{\sqrt{\pi}}{\left(cosTheta \cdot -0.5 + \frac{1}{cosTheta}\right) + -1}}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (exp (* cosTheta (- cosTheta)))
    (/ 1.0 (/ (sqrt PI) (+ (+ (* cosTheta -0.5) (/ 1.0 cosTheta)) -1.0)))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (expf((cosTheta * -cosTheta)) * (1.0f / (sqrtf(((float) M_PI)) / (((cosTheta * -0.5f) + (1.0f / cosTheta)) + -1.0f)))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(exp(Float32(cosTheta * Float32(-cosTheta))) * Float32(Float32(1.0) / Float32(sqrt(Float32(pi)) / Float32(Float32(Float32(cosTheta * Float32(-0.5)) + Float32(Float32(1.0) / cosTheta)) + Float32(-1.0)))))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + (exp((cosTheta * -cosTheta)) * (single(1.0) / (sqrt(single(pi)) / (((cosTheta * single(-0.5)) + (single(1.0) / cosTheta)) + single(-1.0))))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{\frac{\sqrt{\pi}}{\left(cosTheta \cdot -0.5 + \frac{1}{cosTheta}\right) + -1}}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 6: 95.5% accurate, 1.3× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{\sqrt{\pi} \cdot \left(cosTheta \cdot \left(1 + cosTheta\right)\right)}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (exp (* cosTheta (- cosTheta)))
    (/ 1.0 (* (sqrt PI) (* cosTheta (+ 1.0 cosTheta))))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (expf((cosTheta * -cosTheta)) * (1.0f / (sqrtf(((float) M_PI)) * (cosTheta * (1.0f + cosTheta))))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(exp(Float32(cosTheta * Float32(-cosTheta))) * Float32(Float32(1.0) / Float32(sqrt(Float32(pi)) * Float32(cosTheta * Float32(Float32(1.0) + cosTheta)))))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + (exp((cosTheta * -cosTheta)) * (single(1.0) / (sqrt(single(pi)) * (cosTheta * (single(1.0) + cosTheta))))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{\sqrt{\pi} \cdot \left(cosTheta \cdot \left(1 + cosTheta\right)\right)}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 7: 95.8% accurate, 1.3× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{\frac{\sqrt{\pi}}{\frac{1}{cosTheta} + -1}}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (exp (* cosTheta (- cosTheta)))
    (/ 1.0 (/ (sqrt PI) (+ (/ 1.0 cosTheta) -1.0)))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (expf((cosTheta * -cosTheta)) * (1.0f / (sqrtf(((float) M_PI)) / ((1.0f / cosTheta) + -1.0f)))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(exp(Float32(cosTheta * Float32(-cosTheta))) * Float32(Float32(1.0) / Float32(sqrt(Float32(pi)) / Float32(Float32(Float32(1.0) / cosTheta) + Float32(-1.0)))))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + (exp((cosTheta * -cosTheta)) * (single(1.0) / (sqrt(single(pi)) / ((single(1.0) / cosTheta) + single(-1.0))))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \frac{1}{\frac{\sqrt{\pi}}{\frac{1}{cosTheta} + -1}}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 8: 95.3% accurate, 1.3× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \left({\pi}^{-0.5} \cdot \left(\frac{1}{cosTheta} + -1\right)\right)} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (*
    (exp (* cosTheta (- cosTheta)))
    (* (pow PI -0.5) (+ (/ 1.0 cosTheta) -1.0))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (expf((cosTheta * -cosTheta)) * (powf(((float) M_PI), -0.5f) * ((1.0f / cosTheta) + -1.0f))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(exp(Float32(cosTheta * Float32(-cosTheta))) * Float32((Float32(pi) ^ Float32(-0.5)) * Float32(Float32(Float32(1.0) / cosTheta) + Float32(-1.0))))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + (exp((cosTheta * -cosTheta)) * ((single(pi) ^ single(-0.5)) * ((single(1.0) / cosTheta) + single(-1.0)))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + e^{cosTheta \cdot \left(-cosTheta\right)} \cdot \left({\pi}^{-0.5} \cdot \left(\frac{1}{cosTheta} + -1\right)\right)}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 9: 95.3% accurate, 1.3× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + \frac{\sqrt{\frac{1}{\pi}} \cdot \left(\frac{1}{cosTheta} + -1\right)}{e^{cosTheta \cdot cosTheta}}} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/
  1.0
  (+
   (+ 1.0 c)
   (/
    (* (sqrt (/ 1.0 PI)) (+ (/ 1.0 cosTheta) -1.0))
    (exp (* cosTheta cosTheta))))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + ((sqrtf((1.0f / ((float) M_PI))) * ((1.0f / cosTheta) + -1.0f)) / expf((cosTheta * cosTheta))));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(Float32(sqrt(Float32(Float32(1.0) / Float32(pi))) * Float32(Float32(Float32(1.0) / cosTheta) + Float32(-1.0))) / exp(Float32(cosTheta * cosTheta)))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + ((sqrt((single(1.0) / single(pi))) * ((single(1.0) / cosTheta) + single(-1.0))) / exp((cosTheta * cosTheta))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + \frac{\sqrt{\frac{1}{\pi}} \cdot \left(\frac{1}{cosTheta} + -1\right)}{e^{cosTheta \cdot cosTheta}}}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 10: 94.9% accurate, 2.0× speedup?

\[\begin{array}{l} \\ \frac{1}{\left(1 + c\right) + \sqrt{\frac{1}{\pi}} \cdot \left(\frac{1}{cosTheta} + -1\right)} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/ 1.0 (+ (+ 1.0 c) (* (sqrt (/ 1.0 PI)) (+ (/ 1.0 cosTheta) -1.0)))))
float code(float cosTheta, float c) {
	return 1.0f / ((1.0f + c) + (sqrtf((1.0f / ((float) M_PI))) * ((1.0f / cosTheta) + -1.0f)));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(Float32(1.0) + c) + Float32(sqrt(Float32(Float32(1.0) / Float32(pi))) * Float32(Float32(Float32(1.0) / cosTheta) + Float32(-1.0)))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / ((single(1.0) + c) + (sqrt((single(1.0) / single(pi))) * ((single(1.0) / cosTheta) + single(-1.0))));
end

\begin{array}{l}

\\
\frac{1}{\left(1 + c\right) + \sqrt{\frac{1}{\pi}} \cdot \left(\frac{1}{cosTheta} + -1\right)}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 11: 94.8% accurate, 2.0× speedup?

\[\begin{array}{l} \\ \frac{1}{1 + \sqrt{\frac{1}{\pi}} \cdot \left(\frac{1}{cosTheta} + -1\right)} \end{array} \]
(FPCore (cosTheta c)
 :precision binary32
 (/ 1.0 (+ 1.0 (* (sqrt (/ 1.0 PI)) (+ (/ 1.0 cosTheta) -1.0)))))
float code(float cosTheta, float c) {
	return 1.0f / (1.0f + (sqrtf((1.0f / ((float) M_PI))) * ((1.0f / cosTheta) + -1.0f)));
}

function code(cosTheta, c)
	return Float32(Float32(1.0) / Float32(Float32(1.0) + Float32(sqrt(Float32(Float32(1.0) / Float32(pi))) * Float32(Float32(Float32(1.0) / cosTheta) + Float32(-1.0)))))
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) / (single(1.0) + (sqrt((single(1.0) / single(pi))) * ((single(1.0) / cosTheta) + single(-1.0))));
end

\begin{array}{l}

\\
\frac{1}{1 + \sqrt{\frac{1}{\pi}} \cdot \left(\frac{1}{cosTheta} + -1\right)}
\end{array}
Derivation
    &prev;&pcontext;&pcontext2;&ctx;
  1. Add Preprocessing

Alternative 12: 92.7% accurate, 2.1× speedup?

\[\begin{array}{l} \\ \sqrt{\pi} \cdot cosTheta \end{array} \]
(FPCore (cosTheta c) :precision binary32 (* (sqrt PI) cosTheta))
float code(float cosTheta, float c) {
	return sqrtf(((float) M_PI)) * cosTheta;
}

function code(cosTheta, c)
	return Float32(sqrt(Float32(pi)) * cosTheta)
end

function tmp = code(cosTheta, c)
	tmp = sqrt(single(pi)) * cosTheta;
end

\begin{array}{l}

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

Alternative 13: 10.8% accurate, 140.3× speedup?

\[\begin{array}{l} \\ 1 - c \end{array} \]
(FPCore (cosTheta c) :precision binary32 (- 1.0 c))
float code(float cosTheta, float c) {
	return 1.0f - c;
}

real(4) function code(costheta, c)
    real(4), intent (in) :: costheta
    real(4), intent (in) :: c
    code = 1.0e0 - c
end function

function code(cosTheta, c)
	return Float32(Float32(1.0) - c)
end

function tmp = code(cosTheta, c)
	tmp = single(1.0) - c;
end

\begin{array}{l}

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

Alternative 14: 10.8% accurate, 421.0× speedup?

\[\begin{array}{l} \\ 1 \end{array} \]
(FPCore (cosTheta c) :precision binary32 1.0)
float code(float cosTheta, float c) {
	return 1.0f;
}

real(4) function code(costheta, c)
    real(4), intent (in) :: costheta
    real(4), intent (in) :: c
    code = 1.0e0
end function

function code(cosTheta, c)
	return Float32(1.0)
end

function tmp = code(cosTheta, c)
	tmp = single(1.0);
end

\begin{array}{l}

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

Reproduce

?
herbie shell --seed 2023350 
(FPCore (cosTheta c)
  :name "Beckmann Sample, normalization factor"
  :precision binary32
  :pre (and (and (< 0.0 cosTheta) (< cosTheta 0.9999)) (and (< -1.0 c) (< c 1.0)))
  (/ 1.0 (+ (+ 1.0 c) (* (* (/ 1.0 (sqrt PI)) (/ (sqrt (- (- 1.0 cosTheta) cosTheta)) cosTheta)) (exp (* (- cosTheta) cosTheta))))))