test.ll

; ModuleID = "test"
target triple = "x86_64-unknown-linux-gnu"
target datalayout = ""

declare i8* @malloc(i32)
declare void @"llvm.init.trampoline"(i8*, i8*, i8*)
declare i8* @"llvm.adjust.trampoline"(i8*)
declare void @printf(i8*, ...)

@format = global [4 x i8] c"%d\0A\00"

define i32 @loader(i32* nest %a) {
entry:
  %res = load i32, i32* %a
  ret i32 %res
}

define i32 @main(i32 %argc, i8** %argv) {
entry:
  %env_raw_ptr = call i8* (i32) @malloc(i32 4)
  %env_ptr = bitcast i8* %env_raw_ptr to i32*
  store i32 5, i32* %env_ptr

  %tramp_buf = call i8* (i32) @malloc(i32 72)
  %raw_loader = bitcast i32 (i32*)* @loader to i8*

  call void (i8*, i8*, i8*) @"llvm.init.trampoline"(i8* %tramp_buf, i8* %raw_loader, i8* %env_raw_ptr)
  %adj_tramp_buf = call i8* (i8*) @"llvm.adjust.trampoline"(i8* %tramp_buf)
  %bound_loader = bitcast i8* %adj_tramp_buf to i32 ()*

  %format_ptr = getelementptr [4 x i8], [4 x i8]* @format, i32 0, i32 0
  %loader_ret = call i32 () %bound_loader()
  ; %loader_ret = call i32 (i32*) @loader(i32* %env_ptr)

  call void(i8*, ...) @printf(i8* %format_ptr, i32 %loader_ret)
  ret i32 0
}

test.py

from ctypes import CFUNCTYPE
from ctypes import byref
from ctypes import c_int
from ctypes import c_char_p
from ctypes import POINTER
import llvmlite.binding as llvm

with open('test.ll') as tmp:
  llvm_ir = tmp.read()

llvm.initialize()
llvm.initialize_native_target()
llvm.initialize_native_asmprinter()

target = llvm.Target.from_default_triple()
target_machine = target.create_target_machine()

mod = llvm.parse_assembly(llvm_ir)
mod.verify()

engine = llvm.create_mcjit_compiler(mod, target_machine)
engine.finalize_object()
print('finalized')

main_ptr = engine.get_function_address('main')
main = CFUNCTYPE(c_int, c_int, POINTER(c_char_p))(main_ptr)

argc = c_int(1)
argv = c_char_p("test".encode("utf-8"))

res = main(argc, byref(argv))
print(res)