few questions about work stealing queue implementation

[marakew]

1. why tokio use unsync_load instead atomic Relaxed access ?

https://github.com/tokio-rs/tokio/blob/6fcd9c02176bf3cd570bc7de88edaa3b95ea480a/tokio/src/runtime/scheduler/multi_thread/queue.rs#L149C49-L149C56

i think you are missed while look at golang imlementaton
because golang have not such memory model like C++ or Rust

https://github.com/golang/go/blob/07fc59199b9522bfe0d14f35c4391394efc336c9/src/runtime/proc.go#L6571

all unsync_load can be just atomic Relaxed access

you can simplify check this by godbolt for golang and Rust or C++
few access for week memory cpu like mips or powerpc

2. why you not use cache padding for head,tail index and buffer ?
   spmc seems require cache pad for prevent false sharing and impove perfomance

[Darksonn]

all unsync_load can be just atomic Relaxed access

Unsynchronized loads are more relaxed than relaxed loads.

[marakew]

what mean more relaxed ?
for all CPU platform "atomic load relaxed" just volatile access
that why C++ no need raw access inside atomic implementation