Home

What is MathCompile

MathCompile is a package that translates Wolfram Language functions into C++ code and generate dynamic libraries that can be called in Wolfram Language.

The primary usage of MathCompile is to speed up the execution of Wolfram Language code, similar to the built-in compiler. Compared to the built-in compiler, MathCompile supports functional programming much better, and it implements all supported functions so that the compiled functions do not depend on the Wolfram run-time library to run.

When to use MathCompile

Here are some scenarios where you should consider using MathCompile:

• This part of the code is a performance bottleneck;
• The code does numerical calculation;
• The code uses loop, such as Table, Do, or NestList;
• The code does numerous operations on small lists.

And here are some scenarios where you cannot/do not need to use MathCompile:

• The code does symbolic computation;
• The code uses functions that are not compilable, e.g. NIntegrate;
• The performance is limited by the functions highly optimized already, e.g. matrix multiplication;
• Compilation does not provide a speed-up according to tests.

How to use

See the Guide.

How is the performance

See Performance Tests on some small-scale test cases. The comparisons between MathCompile and the built-in compiler can also be found.

What to expect

MathCompile is still in a very early stage, where a lot of functionalities are to be implemented. Here is a incomplete list of what will possibly be added.

• More functions (for completeness)
  Partition, FreeQ, MemberQ, Delete, Insert, Inner, Outer, ReplacePart, and a few more.

• Helpful Run-time error messages
  Currently, all run-time errors are caught and returned as library function error. Showing a detail description will be added.

• Inlining arrays
  Array constants should be able to pulled into a compiled function. Since compiling List with numerous arguments is not ideal, special treatments are necessary.

• Import and export functions
  Import and export of binary data and tables of numbers can be built into the compiler without too much effort.

• Parallelization
  Parallelization can greatly improve the performance of listable functions and Dot, while ParallelTable and ParallelMap, etc. are harder to implement correctly.

• Interacting with the kernel
  The built-in compiler (with CompilationTarget->"C") is able to call a not-compilable function, which is useful when a necessary function happens to be missing. In addition, supporting Print or Echo is useful.

• More data structures?
  Supporting sparse array might not be worthwhile, but Association (as std::map) has the potential of performance improvement.