Incorrect transformation for loops by optimization passes

[ararmine]

When compiling the following example with clang -O3 opt level, the program does not print the expected output and the returned status is 48.
Steps to reproduce:

1. cat bug.cpp
#include <stdio.h>
#include <stdlib.h>

struct C {
  int i[100000000];
};

int main()
{
  // Eat all available chunks of memory that are big enough for
  // struct C so that the next allocation request will fail.
  while (malloc(sizeof(struct C)));
  printf("End\n");
  return 0;
}

2. clang++ bug.cpp -O3
3. ./a.out
4. echo $?
    48

This is the generated LLVM IR after the optimizations. The loop is optimized away and so is 'printf':

define dso_local noundef i32 @main() local_unnamed_addr #0 !dbg !267 {
  unreachable, !dbg !268
}

Compiling the same example with GCC, the program produces the expected output:

1. g++ bug.cpp -O3
2. ./a.out 
End
3. echo $?
    0