Two performance improvements that yield 10% improvement on smaller buffers.
Circumvent the dispatcher on I/O calls with bypass functions.
Remove compare and swap from suspend functions. This only exists because of the Windows VMHole problem which we don't deal with, so we can save a lot of time by just changing the value instead of using CAS.
Two performance improvements that yield 10% improvement on smaller buffers.
Circumvent the dispatcher on I/O calls with bypass functions.
Remove compare and swap from suspend functions. This only exists because of the Windows VMHole problem which we don't deal with, so we can save a lot of time by just changing the value instead of using CAS.