Foundation for Quick LLVM compiler

Ready for review - probably better get this cleaned up and
checked in even though much work remains.

Basic conversion from MIR to GreenlandIR and from GreenlandIR
back to LIR.  Support sufficient to run Fibonacci test.

Note some structural changes in MIR to support this work:
    o  retaining incoming label for phi nodes
    o  constant propagation
    o  include object reference detection in type inference pass

Change-Id: I8ba63c73e76d071aa40cae0f744e598b96f68699
diff --git a/src/compiler/codegen/arm/armv7-a/Codegen.cc b/src/compiler/codegen/arm/armv7-a/Codegen.cc
index 6f62ccf..ba1e7ab 100644
--- a/src/compiler/codegen/arm/armv7-a/Codegen.cc
+++ b/src/compiler/codegen/arm/armv7-a/Codegen.cc
@@ -45,6 +45,9 @@
 /* Thumb2-specific register allocation */
 #include "../Thumb2/Ralloc.cc"
 
+/* Bitcode conversion */
+#include "../../MethodBitcode.cc"
+
 /* MIR2LIR dispatcher and architectural independent codegen routines */
 #include "../../MethodCodegenDriver.cc"