Conversation (7)
Open reply Open reply 
Greg Bell
@ferrix@mastodon.online · Mar 18
@dramforever @gsuberland processors were not meant to synchronize. You've made a race condition with anxiety. Absolute fools, etc
Open reply 
bartholin (4 arc)
@bartholin@fops.cloud · Mar 18
@dramforever isn't this about gcc removing these instructions during optimization, because it thinks they serve no purpose?
Open reply 
jan-leila
@jan_leila@mspsocial.net · Mar 18
@dramforever I believe that the CPU itself sometimes will schedule actions out if order so even if you have 0 optimizations compile side the CPU will still do its own
Open reply
dram🎀
@dramforever@mastodon.social · Mar 18
Thread reply
@jan_leila Fortunately I specifically excluded that by saying "acquire/release (RCpc)"
Open reply 
👽🐦🦇🐉💻 Θὡθ
@loganer@mastodon.social · Mar 18
@dramforever myth?
just do it over the network, I'm sure it will be all fine. :)
Open reply
dram🎀
@dramforever@mastodon.social · Mar 20
Thread reply
@regehr i don't have a concrete one that i can share immediately, but i had someone ask me "why are atomics needed? this works perfectly fine" on a concurrent spsc circular buffer queue that worked entirely relying on compilation unit barriers acting as compiler barriers, word sized load and store being atomic, and x86 tso behavior. running the same program with tests on arm immediately breaks, and so does turning on -flto