We will continue our reading of the WRL Research report on Swift Java compiler. We now look at the organization of Swift and its optimizations. Specifically we look at the sequence of passes that convert the Java bytecode to Alpha machine code.The IR for a method is built based on the bytecode and annotated with the available profile information. A variety of interprocedural optimizations such as method inlining are applied. This is then followed by machine independent optimizations  A tree matching algorithm is then used to convert many operations to a lower and/or machine depenent form and to do peephole optimizations. Then further optimizations are applied to the machine dependent form.  These mostly include global CSE and code motion. Finally alpha code is generated by a sequence of passes that do instruction scheduling, register allocation, and code generation. All of these passes are applicable to the representation of the method in Swift IR.To summarize, the method bytecode is translated to IR and annotated with profile information, then with the information available from other methods and classes, interprocedural analysis and optimizations are carried out, then the machine independent and machine dependent optimizations follow  in that order. These include global CSE and code motion and finally instruction scheduling, register allocation and code generation in terms of alpha code is done. As we discuss each of these passes in details below, perhaps a list of all the optimizations in each stage could be enumerated so we know which pass and at what stage are we discussing the concerned optimization benefits. Interprocedural analyses involve alias analyses, class hierarchy analyses, escape analysis, field analysis, type propagation etc. Inter procedural optimizations  include method resolution, method inlining, method splitting, object inlining, stack allocation, synchronization removal. Intraprocedural optimizations involve bounds check removal, branch removal, conditional constant propagation, dead code elimination, global CSE, global code motion, loop peeling, null check removal, peephole optimizations,strength reduction, type test elimination. Machine dependent passes include conversion to machine dependent operations, peephole optimizations, sign-extension elimination. trace scheduling, register allocation, - live range splitting, biased graph coloring with rematerialization, block layout and code generation.