[PATCH] D59287: [X86] Only define __GCC_HAVE_SYNC_COMPARE_AND_SWAP_16 in 64-bit mode.
JF Bastien via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Wed Mar 13 08:37:56 PDT 2019
jfb added a comment.
In D59287#1427485 <https://reviews.llvm.org/D59287#1427485>, @craig.topper wrote:
> Isn’t that setMaxAtomicWidth in the x86-64 derived class?
Right you are!
> As far as preventing “cx16” from being set in 32-bit mode, we’ll need to check the behavior of CPUID in 32-bit mode or -march=native might still end up setting it.
What I want to make sure is that your fix to the macro (which seems correct!) doesn't diverge from what the rest LLVM ends up doing with atomics. i.e. the macro says "no cmpxchg16b" but somehow codegen does generate it.
CHANGES SINCE LAST ACTION
More information about the cfe-commits