https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90883
--- Comment #18 from Jeffrey A. Law <law at redhat dot com> --- It's the aarch64's definition of CLEAR_RATIO that's coming into play here. The hole in the structure is critical here to show the DSE pessimization. Changing the size of the object to something "nice" and eliminating the hole defeats the whole purpose here. he fact that the aarch64 port is inlining the expansion essentially invalidates the test, hence if there isn't some guidance from the aarch64 folks, I'm just going to xfail it.
