Hi Oliver, Oliver Kowalke wrote:
I've uploaded a new version of boost.fiber to boost vault.
It provides assembler for x86 (32/64bit) - PowerPC and ARM are planed. Performance of the raw context-switch is 6-12 times faster than swapcontext(POSIX).
Have you implemented this in a way that is interchangeable with swapcontext ? For example, could I take your optimised implementations and use them with e.g. Boost.Coroutine ? If this part of your work could be factored out and used independently, rather than being left as an implementation detail, that would potentially increase the usefulness of the contribution. Also, do you have any observations on why swapcontext is slow? On what platforms have you measured this? If you need any help with ARM I may be able to assist. Regards, Phil.