What exactly PHI instruction does and how to use it in LLVM

A phi node is an instruction used to select a value depending on the predecessor of the current block (Look here to see the full hierarchy - it's also used as a value, which is one of the classes which it inherits from).

Phi nodes are necessary due to the structure of the SSA (static single assignment) style of the LLVM code - for example, the following C++ function

void m(bool r, bool y){
    bool l = y || r ;
}

gets translated into the following IR: (created through clang -c -emit-llvm file.c -o out.bc - and then viewed through llvm-dis)

define void @_Z1mbb(i1 zeroext %r, i1 zeroext %y) nounwind {
entry:
  %r.addr = alloca i8, align 1
  %y.addr = alloca i8, align 1
  %l = alloca i8, align 1
  %frombool = zext i1 %r to i8
  store i8 %frombool, i8* %r.addr, align 1
  %frombool1 = zext i1 %y to i8
  store i8 %frombool1, i8* %y.addr, align 1
  %0 = load i8* %y.addr, align 1
  %tobool = trunc i8 %0 to i1
  br i1 %tobool, label %lor.end, label %lor.rhs

lor.rhs:                                          ; preds = %entry
  %1 = load i8* %r.addr, align 1
  %tobool2 = trunc i8 %1 to i1
  br label %lor.end

lor.end:                                          ; preds = %lor.rhs, %entry
  %2 = phi i1 [ true, %entry ], [ %tobool2, %lor.rhs ]
  %frombool3 = zext i1 %2 to i8
  store i8 %frombool3, i8* %l, align 1
  ret void
}

So what happens here? Unlike the C++ code, where the variable bool l could be either 0 or 1, in the LLVM IR it has to be defined once. So we check if %tobool is true, and then jump to lor.end or lor.rhs.

In lor.end we finally have the value of the || operator. If we arrived from the entry block - then it's just true. Otherwise, it is equal to the value of %tobool2 - and that's exactly what we get from the following IR line:

%2 = phi i1 [ true, %entry ], [ %tobool2, %lor.rhs ]

You don't need to use phi at all. Just create bunch of temporary variables. LLVM optimization passes will take care of optimizing temporary variables away and will use phi node for that automatically.

For example, if you want to do this:

x = 4;
if (something) x = x + 2;
print(x);

You can use phi node for that (in pseudocode):

assign 4 to x1 if (!something) branch to 4 calculate x2 from x1 by adding 2 assign x3 phi from x1 and x2 call print with x3

But you can do without phi node (in pseudocode):

allocate local variable on stack called x load into temp x1 value 4 store x1 to x if (!something) branch to 8 load x to temp x2 add x2 with 4 to temp x3 store x3 to x load x to temp x4 call print with x4

By running optimization passes with llvm this second code will get optimized to first code.

The existing answers are good. But, I want to make it even simpler and shorter.

block3:
    %result = phi i32 [%a, %block1], [%b, %block2]

This means that if the previous block was block1, choose value a. If the previous block was block2, choose value b.

Why do we write like this? This is to prevent assignign result in two different blocks such as if block and else block. Because, we do not want to violate SSA principle. SSA helps compilers to apply variety of optimizations and it is a de-facto standard for the intermediate codes. For more information, refer to this reference.