Date: Thu, 24 Aug 2023 01:25:46 -0600
> On Aug 24, 2023, at 00:20, Thiago Macieira via Std-Discussion <std-discussion_at_[hidden]> wrote:
>
> Don't forget that the moves between register files (the MOVD instructions) are
> also non-trivial and cost 3 cycles each. So there's an argument that locking a
> mutex and performing this addition a single time can be less expensive than
> iterating multiple times, because you're paying the same cost for CMPXCHG, but
> you're definitely doing a single addition and no register file movements.
By saying “locking a mutex” you’re proposing that std::atomic<float> etc should no longer be lock free on this platform. That means that every single std::atomic<float> in every program now needs an associated mutex, which now has to be locked and unlocked around every single operation, including plain loads and stores which currently are a single cheap instruction.
>
> Don't forget that the moves between register files (the MOVD instructions) are
> also non-trivial and cost 3 cycles each. So there's an argument that locking a
> mutex and performing this addition a single time can be less expensive than
> iterating multiple times, because you're paying the same cost for CMPXCHG, but
> you're definitely doing a single addition and no register file movements.
By saying “locking a mutex” you’re proposing that std::atomic<float> etc should no longer be lock free on this platform. That means that every single std::atomic<float> in every program now needs an associated mutex, which now has to be locked and unlocked around every single operation, including plain loads and stores which currently are a single cheap instruction.
Received on 2023-08-24 07:25:59