[EVM] An MBB could end with FCALL, but it's not a terminator

[akiramenai]

[PavelKopyl]

I think this is OK. In this case, MF should be no-return. There are other cases. Consider the following YUL code:

object "Test" {
    code {
        {
            let size := datasize("Test_deployed")
            codecopy(0, dataoffset("Test_deployed"), size)
            return(0, size)
        }
    }

    object "Test_deployed" {
        code {
            {
              mstore(0, 0xdeadbeef)
              log0(0, 32)
            }
        }
    }
}

This gives

define void @__entry() noreturn {
entry:
  store i256 3735928559, ptr addrspace(1) null, align 4294967296
  tail call void @llvm.evm.log0(ptr addrspace(1) noalias nocapture nofree noundef nonnull align 32 null, i256 32)
  unreachable
}

solc adds STOP instruction at the end of BBs with neither branches no successors. PR #805 adds this functionality to LLVM.

UP: Actually, this is not related to the original task description. The more appropriate way to handle such YUL code is to emit stop intrinsic at the BB end. solc decides on whether it's required to emit STOP or not at very late stage (at stackification), so it's up to FE if it can properly identify places where to insert the intrinsic or not.