Hi,

A couple of comments from me while testing this trying to build a
cross-toolchain with llvm-13.  Hope this is helpful!

Julien Lepiller <julien@lepiller.eu> writes:

> * gnu/packages/llvm.scm (cross-clang): New variable.
> ---
>  gnu/packages/llvm.scm | 144 +++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 143 insertions(+), 1 deletion(-)
>
> diff --git a/gnu/packages/llvm.scm b/gnu/packages/llvm.scm
> index d6e9846699..c277e2ac35 100644
> --- a/gnu/packages/llvm.scm
> +++ b/gnu/packages/llvm.scm
> @@ -54,6 +54,7 @@ (define-module (gnu packages llvm)
>    #:use-module (guix build-system trivial)
>    #:use-module (gnu packages)
>    #:use-module (gnu packages base)
> +  #:use-module (gnu packages cross-base)
>    #:use-module (gnu packages gcc)
>    #:use-module (gnu packages bootstrap)           ;glibc-dynamic-linker
>    #:use-module (gnu packages check)               ;python-lit
> @@ -75,7 +76,9 @@ (define-module (gnu packages llvm)
>    #:use-module (ice-9 match)
>    #:export (make-lld-wrapper
>              system->llvm-target
> -            cross-llvm))
> +            cross-llvm
> +            cross-clang
> +            clang-for-target))
>  
>  (define* (system->llvm-target #:optional
>                                (system (or (and=> (%current-target-system)
> @@ -1023,6 +1026,145 @@ (define-public clang-runtime clang-runtime-13)
>  (define-public clang clang-13)
>  (define-public clang-toolchain clang-toolchain-13)
>  
> +(define* (cross-clang target
> +                      #:key
> +                      (libc (cross-libc target))
> +                      (xgcc (cross-gcc target
> +                              #:xbinutils (cross-binutils target)
> +                              #:libc (cross-libc target)))
> +                      (clang clang))
> +  "Return a cross-clang compiler for target."
> +  (define cross-clang-aux
> +    (mlambda (target libc xgcc clang)
> +      (package
> +        (inherit clang)
> +        (name (string-append "clang-cross-" target))
> +        (version (package-version clang))
> +        ;; Support the same variables as clang, even in cross-compilation context.
> +        ;; Clang does not make a difference between native and cross-compilation.
> +        (search-paths
> +          (list
> +            (search-path-specification
> +              (variable "CROSS_LIBRARY_PATH")
> +              (files '("lib")))
> +            (search-path-specification
> +              (variable "CROSS_C_INCLUDE_PATH")
> +              (files '("include")))
> +            (search-path-specification
> +              (variable "CROSS_CPLUS_INCLUDE_PATH")
> +              (files '("include/c++" "include")))))
> +        (native-search-paths '())
> +        (arguments
> +         (substitute-keyword-arguments (package-arguments clang)
> +           ((#:configure-flags _)
> +            `(list "-DCLANG_INCLUDE_TESTS=True"
> +                   (string-append "-DGCC_INSTALL_PREFIX="
> +                                  (assoc-ref %build-inputs "cross-gcc-lib"))
> +                   (string-append "-DC_INCLUDE_DIRS="
> +                                  (assoc-ref %build-inputs "target-libc")
> +                                  "/include")))
> +           ((#:phases phases)
> +            `(modify-phases ,phases
> +               (add-after 'unpack 'add-missing-libdir
> +                 (lambda _
> +                   ;; cross-gcc installs its libraries in <target>/lib instead of
> +                   ;; lib.
> +                   (substitute* "lib/Driver/ToolChain.cpp"
> +                     (("\"-L\"\\) \\+ LibPath\\)\\);")
> +                      ,(string-append "\"-L\") + LibPath));
> +  CmdArgs.push_back(Args.MakeArgString(StringRef(\"-L\") + "
> +                                     "StringRef(GCC_INSTALL_PREFIX) + StringRef(\"/"
> +                                     target "/lib\")));
> +  CmdArgs.push_back(Args.MakeArgString(StringRef(\"-rpath=\") + "
> +                                     "StringRef(GCC_INSTALL_PREFIX) + StringRef(\"/"
> +                                     target "/lib\")));")))))
> +               (add-after 'unpack 'support-cross-include-path
> +                 (lambda _
> +                   (substitute* "lib/Driver/ToolChains/Clang.cpp"
> +                     (("C_INCLUDE_PATH") "CROSS_C_INCLUDE_PATH")
> +                     (("CPLUS_INCLUDE_PATH") "CROSS_CPLUS_INCLUDE_PATH"))))
> +               (add-after 'unpack 'support-cross-library-path
> +                 (lambda _
> +                   ;; LIBRARY_PATH is only supported for native builds, but we still
> +                   ;; need it (or CROSS_LIBRARY_PATH to be precise) when
> +                   ;; cross-compiling
> +                   (substitute* "lib/Driver/ToolChains/CommonArgs.cpp"
> +                     (("LIBRARY_PATH\"")
> +                      "LIBRARY_PATH\");
> +  } else {
> +    addDirectoryList(Args, CmdArgs, \"-L\", \"CROSS_LIBRARY_PATH\""))))

Testing this with llvm-13, this substitution doesn't look quite right
and causes the build to fail, they might have removed braces between
versions:
https://github.com/llvm/llvm-project/blob/release/13.x/clang/lib/Driver/ToolChains/CommonArgs.cpp#L236

Doing a blame points to this commit, which suggest it's changed with
LLVM 10 release I believe.

https://github.com/llvm/llvm-project/commit/7dbdc8de183e7aa2dc0a9c3a4f473142ddd460bf,

> +               (replace 'set-glibc-file-names
> +                 (lambda* (#:key inputs #:allow-other-keys)
> +                   (let ((libc (assoc-ref inputs "target-libc"))
> +                         (compiler-rt (assoc-ref inputs "clang-runtime"))

Quick question, compiler-rt here is the host's runtime right?  Do we not
need to cross-compile it as well?

It's not always clear to me what is part of compiler-rt vs libc vs
libgcc, it could be target-dependent.  For example when trying to target
WebAssembly, I needed a cross-compiled compiler-rt to provide the
libclang_rt.builtins-wasm32.a library.

> +                         (gcc (assoc-ref inputs "cross-gcc")))
> +                     (setenv "LIBRARY_PATH"
> +                             (string-append
> +                               (assoc-ref inputs "libc") "/lib:" (getenv "LIBRARY_PATH")))
> +                     ,@(cond
> +                        ((version>=? version "6.0")
> +                         `(;; Link to libclang_rt files from clang-runtime.
> +                           (substitute* "lib/Driver/ToolChain.cpp"
> +                             (("getDriver\\(\\)\\.ResourceDir")
> +                              (string-append "\"" compiler-rt "\"")))
> +
> +                           ;; Make "LibDir" refer to <glibc>/lib so that it
> +                           ;; uses the right dynamic linker file name.
> +                           (substitute* "lib/Driver/ToolChains/Linux.cpp"
> +                             (("(^[[:blank:]]+LibDir = ).*" _ declaration)
> +                              (string-append declaration "\"" libc "/lib\";\n"))
> +
> +                             ;; Make clang look for libstdc++ in the right
> +                             ;; location.
> +                             (("LibStdCXXIncludePathCandidates\\[\\] = \\{")
> +                              (string-append
> +                               "LibStdCXXIncludePathCandidates[] = { \"" gcc
> +                               "/include/c++\","))
> +
> +                             ;; Make sure libc's libdir is on the search path, to
> +                             ;; allow crt1.o & co. to be found.
> +                             (("@GLIBC_LIBDIR@")
> +                              (string-append libc "/lib")))))
> +                        (else
> +                         `((substitute* "lib/Driver/Tools.cpp"
> +                             ;; Patch the 'getLinuxDynamicLinker' function so that
> +                             ;; it uses the right dynamic linker file name.
> +                             (("/lib64/ld-linux-x86-64.so.2")
> +                              (string-append libc ,(glibc-dynamic-linker))))
> +
> +                           ;; Link to libclang_rt files from clang-runtime.
> +                           ;; This substitution needed slight adjustment in 3.8.
> +                           ,@(if (version>=? version "3.8")
> +                                 '((substitute* "lib/Driver/Tools.cpp"
> +                                     (("TC\\.getDriver\\(\\)\\.ResourceDir")
> +                                      (string-append "\"" compiler-rt "\""))))
> +                                 '((substitute* "lib/Driver/ToolChain.cpp"
> +                                     (("getDriver\\(\\)\\.ResourceDir")
> +                                      (string-append "\"" compiler-rt "\"")))))
> +
> +                           ;; Make sure libc's libdir is on the search path, to
> +                           ;; allow crt1.o & co. to be found.
> +                           (substitute* "lib/Driver/ToolChains.cpp"
> +                             (("@GLIBC_LIBDIR@")
> +                              (string-append libc "/lib")))))))))))))
> +        (inputs
> +         `(("target-libc" ,libc)
> +           ("cross-gcc-lib" ,xgcc "lib")
> +           ("cross-gcc" ,xgcc)
> +           ,@(package-inputs clang)))
> +        (propagated-inputs
> +          (modify-inputs (package-propagated-inputs clang)
> +            (replace "llvm"
> +              (cross-llvm
> +                (car (assoc-ref (package-propagated-inputs clang) "llvm"))
> +                target)))))))
> +  (cross-clang-aux target libc xgcc clang))
> +
> +(define* (clang-for-target #:optional (clang clang))
> +  (if (%current-target-system)
> +      (cross-clang (%current-target-system) #:clang clang)
> +      clang))
> +
>  (define-public llvm-for-rocm
>    (package
>      ;; Actually based on LLVM 13 as of v4.3, but llvm-12 works just fine.