ro marked 2 inline comments as done.
ro added inline comments.
================
Comment at: clang/lib/Driver/ToolChains/Gnu.cpp:634
+ // LLVM support for atomics on 32-bit SPARC V8+ is incomplete, so
+ // forcibly link with libatomic as a workaround.
----------------
MaskRay wrote:
> `// TODO ...` and attach a bug link
It's split between the original bug report and a patch review: I only noticed
later (when `compiler-rt` started to make use of 64-bit atomics) that those
aren't implemented by `clang -m32 -mcpu=v9`.
================
Comment at: clang/lib/Driver/ToolChains/Gnu.cpp:637
+ if (getToolChain().getTriple().getArch() == llvm::Triple::sparc) {
+ CmdArgs.push_back("--as-needed");
+ CmdArgs.push_back("-latomic");
----------------
MaskRay wrote:
> Such --as-needed usage is a bit fragile if clang driver in the future passes
> `--as-needed` in the beginning. Better to use --push-state if you have a
> not-too-old ld.
Good point: I initially fell into that trap in D130571 when I added
`--as-needed -latomic --no-as-needed` to `SANITIZER_COMMON_LINK_FLAGS`: it got
added early to the link line, before the objects using atomics, and had no
effect. Only then did Iearn that `target_link_libraries` accepts not only
library names, but also linker flags.
As for `--push-state`, it was introduced in GNU `ld` 2.25 back in 2014, to it's
save to assume it's present.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D130569/new/
https://reviews.llvm.org/D130569
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
