WARPO uses shadow stack GC algorithm.

In frontend, WARPO will introduce 2 functions to identify shadow stack mapping.

 (import "as-builtin-fn" "~lib/rt/__localtostack" (func $~lib/rt/__localtostack (param i32) (result i32)))
 (import "as-builtin-fn" "~lib/rt/__tmptostack" (func $~lib/rt/__tmptostack (param i32) (result i32)))

In optimzation lowering passes, WARPO will convert these 2 functions to the store operation in shadow stack.

The shadow stack area is static allocated in linear memory which is from __stack_pointer to __heap_base. In the prelogue of each GC related function, __stack_pointer will decrease to allocate some space of shadow stack. In the epilogue, __stack_pointer will increase back. Then the items stored in new allocated area won't be tracked any more, in another word, these objects will be freed in next time GC.

 (global $~lib/memory/__stack_pointer (mut i32) (i32.const ...))
 (global $~lib/memory/__heap_base i32 (i32.const ...))