[nwallin]

A dedicated absolute value instruction would save exactly one instruction every time you do absolute value, and both of the instructions for naive absolute value are already pretty fast. Dedicated L/TZCNT instructions save a dozen or so instructions every time you do that operation, and some 'clever' ways to do it might use fewer instructions but some are fairly slow instructions.

Amdahl's law. You don't gain a lot of benefit by improving something which is already fast.