alexander-shaposhnikov marked 2 inline comments as done.
alexander-shaposhnikov added inline comments.
================
Comment at: clang/test/CodeGenCXX/threadlocal_address.cpp:27
// CHECK-O1-NEXT: entry:
-// CHECK-O1-NEXT: %[[I_ADDR:.+]] = {{.*}}call ptr
@llvm.threadlocal.address.p0(ptr nonnull @i)
+// CHECK-O1-NEXT: %[[I_ADDR:.+]] = {{.*}}call ptr
@llvm.threadlocal.address.p0(ptr @i)
// CHECK-O1-NEXT: %[[VAL:.+]] = load i32, ptr %[[I_ADDR]]
----------------
ChuanqiXu wrote:
> Why do we move the `nonnull` specifier here? And it looks like we need a
> nonnull attribute for the return value. (Same for other tests.)
Previously (without this patch) this nonnull attribute was added by
InstCombinePass
(at the call site)
(https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp#L2949
),
this patch annotates llvm.threadlocal.address.p0 itself (see the check on line
23) thus adding the attribute at the call site becomes unnecessary (and
InstCombine skips it:
https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp#L2938)
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D131872/new/
https://reviews.llvm.org/D131872
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
