[llvm-branch-commits] [clang] [flang] [lld] [llvm] [Flang] LLVM_ENABLE_RUNTIMES=flang-rt (PR #110217)

[via llvm-branch-commits]

================
@@ -15,6 +15,7 @@
 # used for building this FortranFloat128Math library.
 
 include(CheckLibraryExists)
+include(CheckIncludeFile)
----------------
jeanPerier wrote:

So my issue is indeed related to the bootsrap build of the runtime: the gcc compiler I use to build flang/llvm (the in my cmake command) knows how to find quadmath.h without hints because this header sits in this gcc compiler toochain include directory. When reaching the flang-rt build step, the build switch to using the just built clang that does not know how to find it.

That explains why I never had the issue in out-of-tree builds where I controlled the runtime compiler and flags.

Three questions:
- Is it possible to tell llvm to build flang-rt with the same compiler and options than the one from flang (instead of bootstraping)?
- In bootstrap builds, how can one specify compiler options for the runtime build steps, but not the compiler build? That way I could point the gcc toolchain I want to use to clang for the second step.
- Is there a drawback in always building the runtime out-of-tree instead of using bootstrap builds, apart from the extra cmake command (i.e., are there some features/driver comfiguration step that only happens in bootstrapped builds)?

https://github.com/llvm/llvm-project/pull/110217