Performance
2-3x faster execution compared to standard scalar functions
Optimized for hot code paths and performance-critical sections
Minimal overhead and streamlined implementation
Accuracy
Typically within 2-3 ULP (Units in the Last Place)
Sufficient accuracy for most engineering and scientific applications
Trade-off between precision and speed
Limitations
Does NOT handle special cases (NaN, Inf)
Not IEEE 754 compliant
Assumes valid, finite input values within normal ranges
Undefined behavior for exceptional inputs
Precision
Single precision (
float/ 32-bit): Functions ending withfDouble precision (
double/ 64-bit): Functions without thefsuffix