Yes, globals are shared between threads (technically, you could even access the stack of the thread if you have its address, or anything else). That's one reason they are bad "mult-threading" practice, since they result in subtle bugs.

A way to use "globals" without putting them on the stack (local vars) is to have only one "local var" for each thread (or TLS, but that's more complicated) that is a pointer to all the "globals". So instead of keeping your vars as globals, keep them in a struct -- and pass the address of the struct to any function that needs the globals. Then each thread would have its own "globals". Of course you'll have to make a new struct for every thread, but that's the point of threads.

fork creates a new process so there's no sharing. It's easier in your case, but it has far more overhead.

I'm using sys_clone to spawn CPU threads here: https://board.flatassembler.net/topic.php?t=13676

Thanks, I always see your code and try to resist the urge to copy and paste large portions of it!