Jaedan
Jaedan
I will be releasing my implementation soon
Updated with a modern approach to bloom that actually looks good.
Hi - I was sent this direction because I've been peripherally involved in SDL_Gpu (doing reviews focused on performance and answering questions about PCI and DMA). I happen to build...
The shift to support "task groups" is an excellent and necessary change. However, I do still have a few pieces of feedback on the group task group design. 1) One...
> As for tasks vs task groups: presumably with io_uring (and Win32 i/o completion ports), we'll have a single internal thread that lives just to deal with moving data from...
> Because of this behavior the typical procedure is to still use a threadpool to issue overlapped I/O. On Linux for many many years (via Linux aio) and on Windows...
> We could create a heavy weight async I/O context when a thread first uses the async API. This could be stored in TLS and be automatically cleaned up, so...
> The only thing that affects the ABI is the question of how asynchronous I/O interacts with the storage API, and it's becoming clear that it doesn't. My primary motivation...
> To be clear, the goal here was to make something that doesn't have to block on i/o, and is easy to use, not to make something that moves data...