Octave 3.8, jcobi/2

Average Error: 23.8 → 1.2

Time: 9.7s

Precision: binary64

\[\left(\alpha > -1 \land \beta > -1\right) \land i > 0\]

Math

\[\frac{\frac{\frac{\left(\alpha + \beta\right) \cdot \left(\beta - \alpha\right)}{\left(\alpha + \beta\right) + 2 \cdot i}}{\left(\left(\alpha + \beta\right) + 2 \cdot i\right) + 2} + 1}{2} \]

↓

\[\begin{array}{l} t_0 := \mathsf{fma}\left(2, i, \alpha + \beta\right)\\ t_1 := \left(\alpha + \beta\right) + 2 \cdot i\\ \mathbf{if}\;\frac{\frac{\left(\alpha + \beta\right) \cdot \left(\beta - \alpha\right)}{t_1}}{2 + t_1} \leq -0.9999999999929182:\\ \;\;\;\;\frac{2 \cdot \frac{1}{\alpha} + \left(4 \cdot \frac{i}{\alpha} + 2 \cdot \frac{\beta}{\alpha}\right)}{2}\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{fma}\left(\alpha + \beta, \mathsf{expm1}\left(\mathsf{log1p}\left(\frac{\frac{\beta - \alpha}{t_0}}{2 + t_0}\right)\right), 1\right)}{2}\\ \end{array} \]

FPCore

(FPCore (alpha beta i)
 :precision binary64
 (/
  (+
   (/
    (/ (* (+ alpha beta) (- beta alpha)) (+ (+ alpha beta) (* 2.0 i)))
    (+ (+ (+ alpha beta) (* 2.0 i)) 2.0))
   1.0)
  2.0))

↓

(FPCore (alpha beta i)
 :precision binary64
 (let* ((t_0 (fma 2.0 i (+ alpha beta))) (t_1 (+ (+ alpha beta) (* 2.0 i))))
   (if (<=
        (/ (/ (* (+ alpha beta) (- beta alpha)) t_1) (+ 2.0 t_1))
        -0.9999999999929182)
     (/
      (+ (* 2.0 (/ 1.0 alpha)) (+ (* 4.0 (/ i alpha)) (* 2.0 (/ beta alpha))))
      2.0)
     (/
      (fma
       (+ alpha beta)
       (expm1 (log1p (/ (/ (- beta alpha) t_0) (+ 2.0 t_0))))
       1.0)
      2.0))))

C

double code(double alpha, double beta, double i) {
	return (((((alpha + beta) * (beta - alpha)) / ((alpha + beta) + (2.0 * i))) / (((alpha + beta) + (2.0 * i)) + 2.0)) + 1.0) / 2.0;
}

↓

double code(double alpha, double beta, double i) {
	double t_0 = fma(2.0, i, (alpha + beta));
	double t_1 = (alpha + beta) + (2.0 * i);
	double tmp;
	if (((((alpha + beta) * (beta - alpha)) / t_1) / (2.0 + t_1)) <= -0.9999999999929182) {
		tmp = ((2.0 * (1.0 / alpha)) + ((4.0 * (i / alpha)) + (2.0 * (beta / alpha)))) / 2.0;
	} else {
		tmp = fma((alpha + beta), expm1(log1p((((beta - alpha) / t_0) / (2.0 + t_0)))), 1.0) / 2.0;
	}
	return tmp;
}

Error

Bits error versus alpha

Bits error versus beta

Bits error versus i

Derivation

1. Split input into 2 regimes

2. if (/.f64 (/.f64 (*.f64 (+.f64 alpha beta) (-.f64 beta alpha)) (+.f64 (+.f64 alpha beta) (*.f64 2 i))) (+.f64 (+.f64 (+.f64 alpha beta) (*.f64 2 i)) 2)) < -0.99999999999291822

   1. Initial program 62.9

      \[\frac{\frac{\frac{\left(\alpha + \beta\right) \cdot \left(\beta - \alpha\right)}{\left(\alpha + \beta\right) + 2 \cdot i}}{\left(\left(\alpha + \beta\right) + 2 \cdot i\right) + 2} + 1}{2} \]

   2. Simplified 56.4

      \[\leadsto \color{blue}{\frac{\mathsf{fma}\left(\alpha + \beta, \frac{\beta - \alpha}{\mathsf{fma}\left(2, i, \alpha + \beta\right) \cdot \left(2 + \mathsf{fma}\left(2, i, \alpha + \beta\right)\right)}, 1\right)}{2}} \]

   3. Taylor expanded in alpha around inf 5.0

      \[\leadsto \frac{\color{blue}{\frac{2 \cdot \beta + \left(2 + 4 \cdot i\right)}{\alpha}}}{2} \]

   4. Simplified 5.0

      \[\leadsto \frac{\color{blue}{\frac{\mathsf{fma}\left(2, \beta, 2 + i \cdot 4\right)}{\alpha}}}{2} \]

   5. Taylor expanded in beta around inf 5.0

      \[\leadsto \frac{\color{blue}{2 \cdot \frac{1}{\alpha} + \left(4 \cdot \frac{i}{\alpha} + 2 \cdot \frac{\beta}{\alpha}\right)}}{2} \]

   if -0.99999999999291822 < (/.f64 (/.f64 (*.f64 (+.f64 alpha beta) (-.f64 beta alpha)) (+.f64 (+.f64 alpha beta) (*.f64 2 i))) (+.f64 (+.f64 (+.f64 alpha beta) (*.f64 2 i)) 2))

   1. Initial program 12.8

      \[\frac{\frac{\frac{\left(\alpha + \beta\right) \cdot \left(\beta - \alpha\right)}{\left(\alpha + \beta\right) + 2 \cdot i}}{\left(\left(\alpha + \beta\right) + 2 \cdot i\right) + 2} + 1}{2} \]

   2. Simplified 9.3

      \[\leadsto \color{blue}{\frac{\mathsf{fma}\left(\alpha + \beta, \frac{\beta - \alpha}{\mathsf{fma}\left(2, i, \alpha + \beta\right) \cdot \left(2 + \mathsf{fma}\left(2, i, \alpha + \beta\right)\right)}, 1\right)}{2}} \]

   3. Applied egg-rr 0.2

      \[\leadsto \frac{\mathsf{fma}\left(\alpha + \beta, \color{blue}{\frac{\beta - \alpha}{\mathsf{fma}\left(2, i, \beta + \alpha\right)} \cdot \frac{1}{2 + \mathsf{fma}\left(2, i, \beta + \alpha\right)}}, 1\right)}{2} \]

   4. Applied egg-rr 0.2

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

   5. Applied egg-rr 0.2

      \[\leadsto \frac{\mathsf{fma}\left(\alpha + \beta, \color{blue}{\mathsf{expm1}\left(\mathsf{log1p}\left(\frac{\frac{\beta - \alpha}{\mathsf{fma}\left(2, i, \alpha + \beta\right)}}{2 + \mathsf{fma}\left(2, i, \alpha + \beta\right)}\right)\right)}, 1\right)}{2} \]
3. Recombined 2 regimes into one program.

4. Final simplification 1.2

   \[\leadsto \begin{array}{l} \mathbf{if}\;\frac{\frac{\left(\alpha + \beta\right) \cdot \left(\beta - \alpha\right)}{\left(\alpha + \beta\right) + 2 \cdot i}}{2 + \left(\left(\alpha + \beta\right) + 2 \cdot i\right)} \leq -0.9999999999929182:\\ \;\;\;\;\frac{2 \cdot \frac{1}{\alpha} + \left(4 \cdot \frac{i}{\alpha} + 2 \cdot \frac{\beta}{\alpha}\right)}{2}\\ \mathbf{else}:\\ \;\;\;\;\frac{\mathsf{fma}\left(\alpha + \beta, \mathsf{expm1}\left(\mathsf{log1p}\left(\frac{\frac{\beta - \alpha}{\mathsf{fma}\left(2, i, \alpha + \beta\right)}}{2 + \mathsf{fma}\left(2, i, \alpha + \beta\right)}\right)\right), 1\right)}{2}\\ \end{array} \]

Reproduce

herbie shell --seed 2022125 
(FPCore (alpha beta i)
  :name "Octave 3.8, jcobi/2"
  :precision binary64
  :pre (and (and (> alpha -1.0) (> beta -1.0)) (> i 0.0))
  (/ (+ (/ (/ (* (+ alpha beta) (- beta alpha)) (+ (+ alpha beta) (* 2.0 i))) (+ (+ (+ alpha beta) (* 2.0 i)) 2.0)) 1.0) 2.0))