glaubitz added inline comments.
================
Comment at: clang/lib/Driver/ToolChains/Solaris.cpp:135
}
+ // LLVM lacks atomics support on 32-bit SPARC, so forcibly link with
+ // libatomic as a workaround.
----------------
ro wrote:
> glaubitz wrote:
> > joerg wrote:
> > > This comment is misleading. It's not so much that LLVM doesn't support
> > > them, but that SPARCv8 doesn't have the necessary hardware support. The
> > > v8+ support is incomplete, which is a related problem though.
> > As far as I know, 64-bit atomics are supported if you enable V8+ in GCC -
> > without linking against `libatomic`:
> >
> > ```
> > glaubitz@gcc202:~$ cat atomic.c
> > #include <stdint.h>
> >
> > int main()
> > {
> > int64_t x = 0, y = 1;
> > y = __sync_val_compare_and_swap(&x, x, y);
> > return 0;
> > }
> > glaubitz@gcc202:~$ gcc -m32 -mv8plus atomic.c -o atomic
> > glaubitz@gcc202:~$
> > ```
> I know, that's why I referred to my patch to default `clang` on
> Solaris/sparc to V8+. I'll update the comment.
>
> I'd tried to actually fix the underlying issue (`clang` not emitting
> `casx` with `-m32 -mcpu=v9`), but ran into internal errors and
> areas of LLVM I know nothing about. I might post a WIP patch
> for reference since there are several issues there.
I think @jrtc27 might be able to give advise here having the knowledge on the
Tablegen stuff (if my mind serves me right).
The disassembly shows in any case that `casx` is being emitted as you say:
```
69c: c7 f0 50 02 casx [ %g1 ], %g2, %g3
```
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D118021/new/
https://reviews.llvm.org/D118021
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
