[PATCH] Port memory barriers intrinsics to AArch64

[Tim Northover]

Hi Yi,

Are you sure this is going to interact well with D4516? The builtins you're adding are in the "__builtin_arm" space but that's (correctly, IMO) changing the prefix to "aarch64". Either way, we probably want a Clang-side test for that GCCBuiltin too.

One other thing I noticed was some extra complexity that shouldn't be needed (see end).

Cheers.

Tim.

================
Comment at: lib/Target/AArch64/AArch64InstrFormats.td:546
@@ +545,3 @@
+}]> {
+  let ParserMatchClass = Imm0_15Operand;
+}
----------------
Something weird is going on here. You're adding a ParserMatchClass but not using this operand in any assembly context so it's almost certainly ignored. You probably only need the "ImmLeaf" part of this definition.

http://reviews.llvm.org/D4520