I am working on a LLVM pass, the goal of which is to transform user-defined function to coroutines. For example, a simple noop function:
void f() {}
Transforms to:
define ptr @f_coro() {
init:
%promise = alloca %CoroPromise, align 8
call void @init_promise(ptr %promise) ; my own API
%id = call token @llvm.coro.id(i32 0, ptr %promise, ptr null, ptr null)
%size = call i32 @llvm.coro.size.i32()
%hdl_alloc = call ptr @malloc(i32 %size)
%hdl = call ptr @llvm.coro.begin(token %id, ptr %hdl_alloc)
%0 = call i8 @llvm.coro.suspend(token none, i1 false)
switch i8 %0, label %suspend [
i8 0, label %terminator.1
i8 1, label %cleanup
]
cleanup:
%1 = call ptr @llvm.coro.free(token %id, ptr %hdl)
call void @free(ptr %1)
br label %suspend
suspend:
%2 = call i1 @llvm.coro.end(ptr %hdl, i1 false, token none)
ret ptr %hdl
terminator.1:
call void @set_ret_val(ptr %promise, i32 0) ; my own API
%3 = call i8 @llvm.coro.suspend(token none, i1 false)
switch i8 %3, label %suspend [
i8 0, label %cleanup
i8 1, label %cleanup
]
}
However, when I run the pass using clang (version 18.0.0):
clang++ -fno-discard-value-names -std=c++2a -emit-llvm -S -fpass-plugin=Pass.so sample.cpp
It omits some of the instructions I generated (specifically, @llvm.coro.id), which causes subsequent code to fail compilation. How should I correctly perform such manipulations?"
I've tried using opt on pre-existing bytecode instead of clang. However, this complicates the pipeline.
Forgot to set presplitcoroutine
attribute for my function.