ILCC: A small update and working sample/benchmark for my .NET C Compiler

May 19 2012

I have reached a new milestone. I have been fixing bugs and now I can compile and run successfully one of my programs. It is a compression tool, that is great for benchmarking. So I have made a benchmark to test the performance of the .NET code in terms of the native code.

Project:https://github.com/soywiz/ilcc

You can try the sample by yourself:https://sites.google.com/site/cspspemu/downloads/comptoe_ilcc.7z?attredirects=0&d=1

Native executables are build using microsoft c/c++ compiler with “cl /Ox” (maximum optimizations).

comptoe_ilcc.exe -c3 file.7z file.7z.c3Encoding[03] file.7z -> file.7z.c3…SuccessElapsed: 1225

comptoe_native_cl_x86.exe -c3 file.7z file.7z.c3Encoding[03] file.7z -> file.7z.c3…SuccessElapsed: 810

comptoe_native_cl_x64.exe -c3 file.7z file.7z.c3Encoding[03] file.7z -> file.7z.c3…SuccessElapsed: 804

Also there are more targets:

Available Targets: yaml – Outputs YAML markup pinvoke – Outputs .NET pinvoke source with function declarations and structures (not fully implemented yet) cil – Outputs .NET IL code (not fully implemented yet) xml – Outputs YAML XML

Besides to generate IL Code, you can generate pinvoke signatures and output the AST as XML or YAML. So it can be used to analyze C code in C# using the library or in other languages outputing to XML.

Executable sizes:

25.088 comptoe_ilcc.exe16.896 ilcc.Runtime.dll

76.288 comptoe_native_cl_x64.exe66.560 comptoe_native_cl_x86.exe

At this point runtime lacks lot of stuffs, so it will get larger. For larger projects the runtime will be smaller than the executable and .NET executables get a better compression, so the executables will be smaller than the native executables most of time (at least compressed).

.NET version is just about 50% slower. That’s great! It is in the same order of magnitude and it is faster than I initially expected.So you will be able to run C code on managed .NET (but with unsafe code at this point) platform without recompiling just with a 50% tradeoff (related to the best optimization level CL can get) and without having to port C code to C# by hand.