Mr. Bandela, Your talks are very interesting, but much like last year's Pattern Matching, contain way too much code to be fully comprehensible. I, personally, would have much preferred to see the distilled core idea presented, with some nice examples, then a full implementation, which is available online anyway.
The general trend is for new instructions to be added to CPU's and GPU's to handle mutex's, barriers, sync's and do things like thread wait, flush or wait for data blocks and instructions to complete. These will get mapped onto compiler intrinsics like having keywords in front of C++ instructions blocks eg. parallel { ... } or serial { ... }
Some of those bad jokes could kill a horse. Sense of humour: extremely nerdy. Talk about channels: informative and interesting.
It's meta humor.
He had me at the Fiber joke. Love it.
Mr. Bandela, Your talks are very interesting, but much like last year's Pattern Matching, contain way too much code to be fully comprehensible. I, personally, would have much preferred to see the distilled core idea presented, with some nice examples, then a full implementation, which is available online anyway.
It's not about implementation, it's about reasoning. So personally I'm glad to see that kind of presentations.
This is great, but makes me nervous that the concurrency community is moving in the wrong direction. Thank you for doing this talk!
Could you elaborate? Are you referring to future/promise performance?
The general trend is for new instructions to be added to CPU's and GPU's to handle mutex's, barriers, sync's and do things like thread wait, flush or wait for data blocks and instructions to complete. These will get mapped onto compiler intrinsics like having keywords in front of C++ instructions blocks eg. parallel { ... } or serial { ... }
This was pretty cool, I have to admit.
Amazing talk
What caused the bad performance for future comparing to callback? Could it be the dynamic memory allocation underlying future & promise?
Bo Liu mostly. but there's also a thread created
future/promise doesn't necessarily need to create threads.
The cost of thread wakeup.
Memory allocation and locks
Well, this is some cool software engineering. But.. it's not exactly brain surgery, is it?