x86/AMD64 LLVM MinGW (minimal) toolchain

This is a recipe for reproducibly building a LLVM/Clang/LLD based mingw-w64 toolchain.

Heavily modified and trimmed down by Alec Ari.

Main changes include:

• Build everything statically! Portability FTW!
• Build everything with LLVM toolchain instead of GCC (This requires Clang)
• Bump MinGW target to 0x0A00 (Windows 10 instead of 7)
• Complete re-work of toolchain wrappers
• ThinLTO enabled build
• Polly optimizations enabled by default (-O3 -mllvm -polly)
• -march=x86-64-v3 enabled by default for x86_64-w64-mingw32
  • Requires: AVX AVX2 BMI1 BMI2 F16C FMA ABM MOVBE XSAVE
• -march=prescott enabled by default for i686-w64-mingw32
  • Requires: MMX SSE SSE2 SSE3 FXSR
• Remove Docker
• Remove CI/CD
• Remove tests
• Remove many features such as CFI, python, busybox, sanitizers (this list is very long)
• Simplify all Bash scripts and remove pretty much everything else (ARM, Mac support, etc.) This does remove a lot of flexibility, support, and control but should also help avoid a few bugs. This is a very minimalistic LLVM MinGW toolchain and should only be used if you know you do not need any additional features such as profiling. I am personally using this to simply compile Wine and DXVK.

Building

./build-all.sh "${HOME}/llvm-mingw"