So, with the vendored llvm patch (it's backport will be in the next Release candidate in 2 weeks) I can now locally build libc-for-gpu, which is the last piece that I had to postpone so far. Now it finishes the llvm and ompoffload/libc build steps.

@jieyouxu I rebased one the latest llvm, which had another backport for us.
I got it to work in one config, but the main setup for local builds now still fails. That is, if someone sets clang = true and offload = true.

The problem is simply, that we want to compile libc-for-gpu with a nvptx/amdgcn as targets.
We don't want a full cross-compilation of rustc, so we have x86 as target.
That works for all other omp/offload libraries, but libc-for-gpu breaks if we try to compile it for x86, which is fair.
I tried to remove the line where we set the target in configure_cmake (cfg.target(&target.triple).host(&builder.config.host_target.triple);), so libc-for-gpu could guess its GPU target, and rust bootstrap on the other hand isn't "surprised" by a different target.

This runs into the following assertion:

Building OpenMP/Offload for x86_64-unknown-linux-gnu

thread 'main' (868937) panicked at /g/g90/drehwald1/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/cmake-0.1.54/src/lib.rs:1119:5:

environment variable `TARGET` not defined

build script failed, must exit now

I also tried to hardcode it as amdgcn-amd-amdhsa. This resulted in

Building OpenMP/Offload for x86_64-unknown-linux-gnu
CMAKE_TOOLCHAIN_FILE_x86_64-unknown-linux-gnu = None
CMAKE_TOOLCHAIN_FILE_x86_64_unknown_linux_gnu = None
TARGET_CMAKE_TOOLCHAIN_FILE = None
CMAKE_TOOLCHAIN_FILE = None

thread 'main' (831028) panicked at /g/g90/drehwald1/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/cmake-0.1.54/src/lib.rs:1119:5:

environment variable `CARGO_CFG_TARGET_OS` not defined

build script failed, must exit now

I experimented a bit with setting CARGO_CFG_TARGET_{OS|ARCH}, but that just lead to more bugs.
Do you know a somewhat clean solution to this?

(I'll try to take a look this weekend)