The structure of a branch instruction

Question

I am currently building a static analyzer for 'llvm-ir' and working on branch instructions. The 'condition' of a branch instruction seems to be either a result of a compare instruction, or a result of some logical operations of a compare instruction, which looks like

 %cmp1377 = icmp sgt i32 %length, 0
 br i1 %cmp1377, label %for.cond14.preheader.lr.ph, label %for.cond.cleanup

or probably like

%cmp2 = icmp slt i32 %rem, %div1
%cmp3 = icmp slt i32 %div, %div1
%or.cond = or i1 %cmp2, %cmp3
br i1 %or.cond, label %if.then9, label %lor.lhs.false4

But I'm not sure whether if LLVM-IR only uses compared values or it might also just use raw integer values. Is there any documentation that I can find how it constructs the condition of the branch instructions?

Answer 1

Condition for br instruction can be any value of type i1, no matter it comes from a comparison, logical operation or load from a global variable.

It is possible for clang to generate either of such instructions. For example, for code:

int f1(bool* a){
  if ( *a ){
    return 12;
  }else{
    return 15;
  }
}

It generates the following llvm-ir:

define i32 @f(bool*)(i8* %a) #0 {
  %1 = alloca i32, align 4
  %2 = alloca i8*, align 8
  store i8* %a, i8** %2, align 8
  %3 = load i8*, i8** %2, align 8
  %4 = load i8, i8* %3, align 1
  %5 = trunc i8 %4 to i1                  ;;condition value
  br i1 %5, label %6, label %7            ;;br instruction

; <label>:6                                      
  store i32 12, i32* %1
  br label %8

; <label>:7                                      
  store i32 15, i32* %1
  br label %8

; <label>:8                                       
  %9 = load i32, i32* %1
  ret i32 %9
}