How Clang generates code for function parameters? - llvm

Question

In a function , I want to know how the parameters is passed into the function body, so that to track the flow of the parameters. I tried a simple code, and find there seems to be an alloc-store pattern for every parameter, I wonder whether it is true or not?
A demo code is
int add(int x, int y){
return x+y;
}
The llvm ir it generated is:
; Function Attrs: nounwind uwtable
define i32 #add(i32 %x, i32 %y) #0 {
%1 = alloca i32, align 4
%2 = alloca i32, align 4
store i32 %x, i32* %1, align 4
store i32 %y, i32* %2, align 4
%3 = load i32, i32* %1, align 4
%4 = load i32, i32* %2, align 4
%5 = add nsw i32 %3, %4
ret i32 %5
}
In the example we can see that,
For every parameters, the Clang use a alloc instruction to define a
local variable
Following the alloc instruction, store instructions is used to
assign values?
My questions are:
Is all the function LLVM IR are generated in this alloc and store patterns? Or what exactly LLVM do with the parameters?
The order of parameters is determined by the convention it used?

Answer 1

I think this pattern holds for code that has no compile-time optimizations; however, if you instead compile the code with -O3 (or anything that applies the mem2reg optimization), this pattern is optimized out:
(clang -emit-llvm -S -O0 add.c)
define i32 #add(i32 %x, i32 %y) #0 {
%1 = alloca i32, align 4
%2 = alloca i32, align 4
store i32 %x, i32* %1, align 4
store i32 %y, i32* %2, align 4
%3 = load i32, i32* %1, align 4
%4 = load i32, i32* %2, align 4
%5 = add nsw i32 %3, %4
ret i32 %5
}
(opt -mem2reg add.ll -o add_m.ll)
define i32 #add(i32 %x, i32 %y) #0 {
%1 = add nsw i32 %y, %x
ret i32 %1
}
So if you are controlling all of the code that you are analyzing, then you can rely on this pattern. I would instead recommend that you use the LLVM APIs to get the function arguments. The following code iterates through the arguments to a function F and prints them after casting to values.
for (auto AI = F->arg_begin(), AE = F->arg_end(); AI != AE; ++AI)
{
Value* v = &*AI;
errs() << *v << "\n";
}
The values in the above sample are usable in the same way as any other value in the IR.