Lets analyze one cpu instruction in your scenario: the comparison used to decide whether to resize. You have to remember that PGO operates to optimize each ind

Lets analyze one cpu instruction in your scenario: the comparison used to decide whether to resize. You have to remember that PGO operates to optimize each individual branching instruction, not the whole program. So every time you add something to the hashtable, it will check the current size against the next size it needs to resize. 90% of the time, then 99% of the time, then 99.9% of the time and so on the answer will be 'no, there is space left'. This is just because the size goes up geometrically (in general) so a very low % of insert operations trigger a resize. The 'high probability branch' for "if current_size == max_size" should always be false. This is independent of workload. The compiler would not know whether that comparison is 'almost always true' or 'almost always false' which might lead the CPU's branch predictor to guess wrong, which will lead to pipeline stalls. Read up on how pipelined, superscaler cpu's work to understand why this is very important. https://en.wikipedia.org/wiki/Pipeline_stall https://en.wikipedia.org/wiki/Pipeline_stall https://en.wikipedia.org/wiki/Predication_(computer_architecture) https://en.wikipedia.org/wiki/Predication_(computer_architec...