> "The implementation should be simple too at less than 10K C lines, because I want wider adoption of this tool. Simple code is easier to learn and maintain. I'd like also to avoid any external dependencies for this project. At the end of the article, you can see the actual results I have now. (...) As for the generated code performance, I decided that it should be at least 70% of GCC -O2 performance. The implementation should be simple too at less than 10K C lines, because I want wider adoption of this tool. Simple code is easier to learn and maintain. (...) The biggest cons of the existing projects were their size, the fact that it would be hard for me to achieve my goals using them, and that I can not control the projects. Also, the smaller source size of the MIR project makes studying the code easier for other people and reduces the effort required to maintain it."

I can understand the point for LLVM, but I don't fully understand why QBE (https://c9x.me/compile/doc/llvm.html, which is part of his assessment) didn't fit in what he is looking for.