[LLVMdev] Building compiler-rt for foreign target?

[Solra Bizna]

I'm developing something for an ARMv7-A system in big-endian mode. I'm
trying to use clang as the compiler for it. Things have been going
well so far, but I've hit a point where the generated code is
depending on functions which are implemented in compiler-rt. Since I'm
doing the building on an amd64 machine, the compiler-rt that I have
only targets x86/amd64. So, I have three overlapping questions:

1. How do I build a compiler-rt for my target processor?

2. Since my target processor supports the hardware divide
instructions, what specific steps do I need to take to ensure that
compiler-rt will use them? (Same applies for several other instruction
set extensions.)

3. Since (again) my target processor supports the hardware divide
instructions, is there a way to get LLVM's code generator to use them
directly instead of calling the EABI division functions? (Same applies
for several other instruction set extensions.)