Tighter read origin validation

[hai-rise]

We currently compare transaction incarnations for validation leading to over-estimation when different incarnations write the same value to the memory location (this could be common). A tradeoff we can try is to store (cloned) read values not to invalidate when the read value hasn't changed. Another heuristic to avoid expensive value comparisons is invalidating when we read from a different transaction now (there's a high chance the value has changed).

[byhashdong]

@hai-rise, could you assist by listing the code snippet? I mean which code path need to be improved.

[hai-rise]

@byhashdong Ah this one is pretty involved. I did a quick PoC in #367 and it didn't improve much if I remember correctly. I'm sure there are many improvements over a 10-min PoC but never had the time to benchmark each variation 😅.

[byhashdong]

Okay, if you need any assistance, please inform me.