LLVM fails to detect very simple loop trip count

Question

I'm trying to understand how the loop trip count calculation happens in LLVM using the Scalar Evolution analysis. However, I can't get a simple test case to work.

I have the following test program:

// bug.cpp
int main()
{
  for (int i = 0; i < 16; i++) {
    // do nothing
  }
  return 0;
}

Which I compile like this:

$ clang++ -O0 -emit-llvm -c -o bug.bc bug.cpp

And analyze like this:

$ opt -analyze -indvars -loop-simplify -scalar-evolution < bug.bc

But the output of the scalar evolution is this:

...
Determining loop execution counts for: @main
Loop %for.cond: Unpredictable backedge-taken count.
Loop %for.cond: Unpredictable max backedge-taken count.

Why is this very simple loop have an unpredictable trip count? Am I doing something wrong in invoking the analysis? This is LLVM 3.4.

Answer 1

Solved with the help of this related question:

The use of getSmallConstantTripCount method of Loop in LLVM

Add mem2reg as an optimization pass and the trip count is correctly computed.

$ opt -analyze -mem2reg -indvars -loop-simplify -scalar-evolution < bug.bc