The Compiler
Overview
The core translator produces ANSI standard C99 C code which uses the Message Passing Interface (version 2) for communication. Therefore, on the target machine, an implementation of MPI, such as OpenMPI, MPICH or a vendor specific MPI is required and will all work with the generated code. Additionally our runtime library (known as Idaho) needs to be also linked in. The runtime library performs three roles - firstly it is architecture specific (and versions exist for different flavours of Linux) as it contains any none portable code which is needed and is also optimised for specific platforms. Secondly the runtime library contains functions which are often called and would increase the size of generated C code. Lastly, by placing certain functionality in this library means that if one wishes to tune or modify behaviour for a specific platform then it can be done at the library level rather than having to recompile all existing Mesham codes. The standard runtime library requires the Boehm-Demers-Weiser conservative garbage collector (libgc).
The resulting executable can be thought of as any normal executable, and can be run in a number of ways. In order to allow for simplicity the user can run their program just with one process, the program will automatically spawn the number of processors required. Secondly the executable can be run with the exact number of processes needed and this may be instigated via a process file or queue submission program. It should be noted that, as long as your MPI implementation supports multi-core (and the majority of them do) then the code can be executed properly on a multi core machine, often with the processes wrapping around the cores (for instance 2 processes on 2 cores is 1 process on each, 6 processes on 2 cores is 3 processes on each etc...)
Whilst earlier versions of the MPICH daemon allowed for the user to simply run their executable and the daemon would pick it up, Hydra which is the latest MPICH process manager, requires you to run it via the mpiexec command. We suggest mpiexec -np 1 ./name, where name is the name of your executable and the code will spawn the necessary number of processes.
Compilation in more detail
The compiler itself is contained within a number of different phases. Firstly, your Mesham code goes through a preprocessor which will expand the directives (such as include) into Mesham code. It is at the preprocessor stage that the standard function libraries are made available to the code if the programmer has included them. The code is then fed into the core compiler which contains the keywords and general rules of the language but does not contain any types. These types exist in a separate library and behaviour is called via an API, from the core compiler into the appropriate types.
The Oubliette core produces non human readable ANSI C99 code as an intermediate representation (IR), which is then fed into an applicable C compiler. This stage is also performed by the compiler - although it is possible to dump out this C code and manually compiler if desired.
