jyknight added a comment.
Note that this change caused LLVM to no longer be aware that a TLS variable
cannot be NULL. Thus, code like:
__thread int x;
int main() {
int* y = &x;
return *y;
}
built with `clang -O -fsanitize=null` emits a null-pointer check, when it
wouldn't previously. I think llvm.threadlocal.address's return value probably
ought to be given the nonnull attribute. We also might want to infer other
properties of the returned pointer based on knowledge of the global value
parameter, such as alignment.
Furthermore, while the above problem should simply be a very minor optimization
degradation, in fact it caused miscompiles, due to a pre-existing bug in the
X86 backend. I've sent https://reviews.llvm.org/D131716 to fix //that// part of
the problem.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D129833/new/
https://reviews.llvm.org/D129833
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
