Average Error: 0.5 → 0.3
Time: 8.4s
Precision: 64
\[\frac{\left(\frac{\left(d1 \cdot \left(3\right)\right)}{\left(d1 \cdot d2\right)}\right)}{\left(d1 \cdot d3\right)}\]
↓
\[\left(\left(3 + d2\right) + d3\right) \cdot d1\]
\frac{\left(\frac{\left(d1 \cdot \left(3\right)\right)}{\left(d1 \cdot d2\right)}\right)}{\left(d1 \cdot d3\right)}
↓
\left(\left(3 + d2\right) + d3\right) \cdot d1
double f(double d1, double d2, double d3) {
        double r3083219 = d1;
        double r3083220 = 3.0;
        double r3083221 = /* ERROR: no posit support in C */;
        double r3083222 = r3083219 * r3083221;
        double r3083223 = d2;
        double r3083224 = r3083219 * r3083223;
        double r3083225 = r3083222 + r3083224;
        double r3083226 = d3;
        double r3083227 = r3083219 * r3083226;
        double r3083228 = r3083225 + r3083227;
        return r3083228;
}
↓
double f(double d1, double d2, double d3) {
        double r3083229 = 3.0;
        double r3083230 = d2;
        double r3083231 = r3083229 + r3083230;
        double r3083232 = d3;
        double r3083233 = r3083231 + r3083232;
        double r3083234 = d1;
        double r3083235 = r3083233 * r3083234;
        return r3083235;
}

Error

Bits error versus d1

Bits error versus d2

Bits error versus d3

Derivation

  1. Initial program 0.5

    \[\frac{\left(\frac{\left(d1 \cdot \left(3\right)\right)}{\left(d1 \cdot d2\right)}\right)}{\left(d1 \cdot d3\right)}\]

  2. Simplified0.3

    \[\leadsto \color{blue}{\left(\frac{\left(3\right)}{\left(\frac{d2}{d3}\right)}\right) \cdot d1}\]
  3. Using strategy rm

  4. Applied associate-+r+0.3

    \[\leadsto \color{blue}{\left(\frac{\left(\frac{\left(3\right)}{d2}\right)}{d3}\right)} \cdot d1\]

  5. Final simplification0.3

    \[\leadsto \left(\left(3 + d2\right) + d3\right) \cdot d1\]

Reproduce

herbie shell --seed 2019128 +o rules:numerics
(FPCore (d1 d2 d3)
  :name "FastMath test3"
  (+.p16 (+.p16 (*.p16 d1 (real->posit16 3)) (*.p16 d1 d2)) (*.p16 d1 d3)))