[PATCH] D102096: [DAGCombiner] Fix DAG combine store elimination, different address space.

[Hendrik Greving via Phabricator via llvm-commits]

hgreving added inline comments.


================
Comment at: llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp:17920
+        Ld->getPointerInfo().getAddrSpace() ==
+            ST->getPointerInfo().getAddrSpace() &&
         // There can't be any side effects between the load and store, such as
----------------
myhsu wrote:
> just curious, is the code here formatted by clang-format?
Yes.


================
Comment at: llvm/test/CodeGen/X86/fs_gs_dag_combine.ll:18
+; CHECK-NEXT: movl %eax, %fs:2
+define i32 @copy_fs_diff() {
+entry:
----------------
myhsu wrote:
> copy_fs_diff and copy_gs_diff seem a little redundant to me: The address spaces are already different, why should we care about address? Can you elaborate a little more on the purpose of these two functions?
I've added a comment. I thought the two more tests won't hurt, I'm fine removing the :gs versions if considered unnecessary.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D102096/new/

https://reviews.llvm.org/D102096