"Favor the one with lower register pressure" - my mental model says that register pressure is not a property of a single instruction, but rather depends on when successor instructions will be scheduled.
For example, if an instruction defines a value which has 8 uses, then the "register pressure of the instruction" depends on when those 8 uses will be scheduled.
I guess there is a heuristic to compute the register pressure of an instruction based on a guess of when successors will be scheduled?
"Favor the one with lower register pressure" - my mental model says that register pressure is not a property of a single instruction, but rather depends on when successor instructions will be scheduled.
For example, if an instruction defines a value which has 8 uses, then the "register pressure of the instruction" depends on when those 8 uses will be scheduled.
I guess there is a heuristic to compute the register pressure of an instruction based on a guess of when successors will be scheduled?