Twitter | Pretraživanje | |
Corentin
A Universal I/O Abstraction for C++ New blog post about executors, asynchronous I/O, io_uring, coroutines and more ! ➡️ ⬅️
Reply Retweet Označi sa "sviđa mi se" More
Jens Axboe 31. sij
Odgovor korisniku/ci @Cor3ntin
A few questions/comments: "a single system io_uring_enter" - missing "call"? And any reason for the dummy read rather than a NOP?
Reply Retweet Označi sa "sviđa mi se"
Corentin 31. sij
Odgovor korisniku/ci @axboe
Missing call indeed the read reads from an eventfd descriptor, which is written to from an arbitrary thread - it's just to avoid a lock around the sqe handling - as apparently eventfd is a bit cheaper than a futex.
Reply Retweet Označi sa "sviđa mi se"
Nagy-Egri Máté 31. sij
Odgovor korisniku/ci @Cor3ntin @ericniebler @blelbach
I cannot stop but think how this would translate to and . Draw calls and kernels as I/O. How would cancellation work? (Need API extensions?) Command lists playing the role of tasks... how does batch execution come into play?
Reply Retweet Označi sa "sviđa mi se"
Corentin 31. sij
Odgovor korisniku/ci @nagyegrimate @ericniebler i 2 ostali
There would be a bulk_execute that runs an operation n times. But I conviniently did not talk about the things I do not understand... I know that ultimately wants GPUs to just work with regular c++ but I have no clue how.
Reply Retweet Označi sa "sviđa mi se"
Matheus Catarino 1. velj
Odgovor korisniku/ci @Cor3ntin
Interesting, although I was not able to completely abstract the information on this subject. Thank you for sharing this very robust and also complex explanation.
Reply Retweet Označi sa "sviđa mi se"