Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/davidar/lljvm

Low Level Java Virtual Machine
https://github.com/davidar/lljvm

Last synced: 4 days ago
JSON representation

Low Level Java Virtual Machine

Awesome Lists containing this project

README 

        
# LLJVM



**LLJVM** provides a set of tools and libraries for running comparatively low level

languages (such as C) on the JVM.



The C to JVM bytecode compilation provided by LLJVM involves several steps. Source code is first compiled to [LLVM][llvm] intermediate representation (IR) by a frontend such as `llvm-gcc` or [clang][clang]. LLVM IR is then translated to [Jasmin][jasmin] assembly code, linked against other Java classes, and then assembled to JVM bytecode.



The use of LLVM IR as the intermediate representation allows more information about the source program to be preserved, compared to other methods which use MIPS binary as the intermediate representation (such as [NestedVM][nestedvm] and [Cibyl][cibyl]). For example, functions are mapped to individual JVM methods, and all function calls are made with native JVM invocation instructions. This allows compiled code to be linked against arbitrary Java classes, and Java programs to natively call individual functions in the compiled code. It also allows programs to be split across multiple classes (comparable to dynamic linking), rather than statically linking everything into a single class.



Also note that while C is currently the only supported input language, any language with a compiler targeting LLVM IR could potentially be supported.



This project is currently unmaintained. You may be interested in:



- [one of the forks](https://github.com/davidar/lljvm/network)

- [LLJVM Translator](https://github.com/maropu/lljvm-translator)



## QUICKSTART



For a **quick demonstration** of some pre-compiled classes, download the [runtime library][lljvm-jar] and the [demo package][lljvm-demo-jar] to a directory on your machine, and run `java -jar lljvm-demo-0.2.jar`.



To **quickly start** translating your own programs you might want to use [Docker][docker]. Getting LLJVM to run on modern machines (and keeping it running) can be quite tricky as it depends heavily on an old version of LLVM. As an alternative, Martin Haye has set up a virtual machine image [here][mhaye-lljvm] that is pre-configured for and includes LLJVM. Install Docker (you may need Boot2Docker) and start it this way: `docker run -i -v /your/project/dir:/project -t mhaye/lljvm /bin/bash`



To compile LLJVM from source, follow the instructions below. If you have problems compiling from source and are running Linux on an i386-compatible platform, download the [binary release][lljvm-bin], extract it, and download the [runtime library][lljvm-jar] into the resulting directory.



- [HN comments][hn-lljvm]

- 2009-2015 [David A Roberts](https://davidar.io)



## INSTALLATION

[LLVM 2.7][llvm] must be installed to compile and run LLJVM. [Jasmin][jasmin] is needed for

assembling JVM bytecode.



To compile LLJVM, simply call `make` in the project root directory, and call

`make check` to run the testsuite.



To generate the the documentation for the java packages and the backend, call

`make doc`, which will generate HTML documentation in `doc/java/` and

`doc/backend/` respectively. PDF documentation for the backend can be obtained by

calling `make doc-pdf` (requires PDFLaTeX), which will generate

`doc/backend.pdf`. A zip archive containing all of the documentation can be

created by calling `make doc-zip`.



## C COMPILER

The `lljvm-cc` frontend ties together several components to provide a C source

code to JVM class file compiler.



In order to use the frontend, either `llvm-gcc` or `clang` must be installed.

`llvm-gcc` is recommended, and is used instead of `clang` if both are available.

It is also required that the classpath contain [jasmin.jar][jasmin-jar].



To compile a small number of source files to a class file, the following style

of command can be used:



    lljvm-cc foo.c bar.c -o foo



This will generate `foo.class`, as well as a shell script `foo`, which sets the

classpath appropriately and executes the classfile with the arguments passed to

it. This allows the class to be called in the same way as any other executable.

If calling the class without the shell script, note that the entire list of

arguments must be passed to the class, including argument 0 (the name of the

command).



For a larger number of source files, it is often more efficient to separately

compile each file to an object file as needed:



    lljvm-cc -c foo.c

then link the object files together into a class file:



    lljvm-cc -link foo.o bar.o baz.o -o foo

Note that the -link flag must be used if object files are being supplied

instead of source files.



To link the object files together as a library instead of an executable,

`-link-as-library` can be passed instead of `-link`. In this case the shell script

will not be generated.



To link against a library generated by `lljvm-cc`, the `-l` flag can be used.

This will search for a library in the classpath called `lib.class`.

Additional directories can be added to the classpath with the `-L` flag.

Such additions to the classpath will also be added to the shell script, so

when linking an executable, the `-L` flag must be passed for each directory

containing libraries needed by the executable (except the current directory and

directories in the system classpath).



An exception to the above is the `-lm` flag. This will not search for `libm.class`,

but will rather link against `java.lang.Math` and `lljvm.runtime.Math`. Also, `-lc`

should not be used as `libc` is linked by default. If this is not desired, then

the `-nostdlib` flag can be used.



By default all classes generated are placed in the default package. To assign

the class to a specific package, the `-classname` flag can be used. For example,

the following command:



    lljvm-cc ... -classname=com.example.foo -o bar/baz

will create a class named foo, in the package com.example, placing it in the

directory `bar/com/example/`. The shell script will be output to `bar/baz`.



If the `-g3` flag is used, then Jasmin assembly with full debugging information

will be output to `.j`.



In addition to the above flags, any flag accepted by `gcc` or `ld` can also be

used. However, sometimes these flags may not be passed to the correct

component (this is a bug with `lljvm-cc` and should be reported).



The frontend can also be used to compile `autoconf`-based projects relatively

easily (although sometimes some changes to the project's build system may be

required):



    ./configure CC=lljvm-cc LD='lljvm-cc -link'

    make CCLD='lljvm-cc -link'



## DEMO

To demonstrate the capabilities of LLJVM, several common software packages can

be compiled with `lljvm-cc` by entering the `demo/` subdirectory and calling

`make`. To verify that all of these compiled correctly, call `make check`.



A JAR archive containing all of the demo programs can be created by calling

`make demo` in the project root directory. For usage information, execute the

JAR in the project root directory with no arguments:



    java -jar lljvm-demo.jar



## BACKEND

The LLJVM Backend transforms LLVM IR into Jasmin assembly code.

It can be invoked by:



    lljvm-backend foo.bc > foo.j



There are two flags that are accepted by the backend: `-classname` and `-g`.

Run `lljvm-backend --help` for more information.



The output file should then be linked by the LLJVM Linker (see below), and

assembled into a class file by [Jasmin][jasmin]:



    java -jar jasmin.jar foo.j



## RUNTIME

The LLVJM Runtime has three components, the Core Runtime (`lljvm.runtime`), the

I/O Support Library (`lljvm.io`), and the C Standard Library (`lljvm.lib.c`). See

the JavaDoc documentation for further details on the former two. The latter is

[Newlib][newlib] compiled to JVM bytecode by `lljvm-cc`.



## TOOLS

There are two command-line tools available: the linker, and the info utility.

There are several ways to invoke these tools. The simplest way is to call them

directly from the jar archive:



    java -jar lljvm.jar  args...



Alternatively, if `lljvm.jar` is already in the classpath, one of the following

can be used:



    java lljvm.tools.Main  args...

    java lljvm.tools..Main args...



## LINKER

The LLJVM Linker qualifies references to external methods and fields in Jasmin

assembly code. At the top of the code (before any `.method` directives), external

references should be specified through the `.extern` pseudo-directive, such as:



    .extern field foo I

    .extern method bar(I)V



Then the linker can be invoked by:



    java -jar lljvm.jar ld LIBRARY... < INPUT.j > OUTPUT.j



For example, the following assembly code:



    .extern method cos(D)D

    .extern field NULL I

    ...

    invokestatic cos(D)D

    getstatic NULL I

    CLASSFORMETHOD cos(D)D

linked with the command:



    java -jar lljvm.jar ld java.lang.Math lljvm.runtime.Memory

would produce:



    ...

    invokestatic java/lang/Math/cos(D)D

    getstatic lljvm/runtime/Memory/NULL I

    ldc "java/lang/Math"



## INFO

The info utility lists the type signatures of the public static fields and

methods provided by a class. For example, to list those provided by `libc`:



    java -jar lljvm.jar info lljvm.lib.c



By default, any identifiers beginning with an underscore are omitted. To

disable this behaviour, pass the `-v` flag:



    java -jar lljvm.jar info -v lljvm.lib.c



[llvm]: http://llvm.org/

[jasmin]: https://github.com/davidar/jasmin

[jasmin-jar]: https://github.com/davidar/jasmin/raw/master/jasmin.jar

[newlib]: http://sourceware.org/newlib/

[hn-lljvm]: http://news.ycombinator.com/item?id=961834

[lljvm-jar]: https://github.com/davidar/lljvm/releases/download/0.2/lljvm-0.2.jar

[docker]: https://www.docker.com/

[mhaye-lljvm]: https://registry.hub.docker.com/u/mhaye/lljvm/

[lljvm-demo-jar]: https://github.com/davidar/lljvm/releases/download/0.2/lljvm-demo-0.2.jar

[lljvm-bin]: https://github.com/davidar/lljvm/releases/download/0.2/lljvm-bin-linux-i386-0.2.tar.gz

[clang]: http://clang.llvm.org/

[nestedvm]: http://nestedvm.ibex.org/

[cibyl]: https://github.com/SimonKagstrom/cibyl