[PATCH] D12844: Support align attribute for return values

[Philip Reames via llvm-commits]

reames added inline comments.

================
Comment at: lib/AsmParser/LLParser.cpp:1382
@@ -1381,1 +1381,3 @@
     }
+    case lltok::kw_align: {
+      unsigned Alignment;
----------------
The fact your changing the IR parser and not the IR writer, bitcode parser, or bitcode writer seems mildly suspect.  Have you tested the ability to round trip through both IR and Bitcode?

================
Comment at: test/Analysis/ValueTracking/memory-dereferenceable.ll:120
@@ +119,3 @@
+    %deref_return = call dereferenceable(32) i32* @foo()
+    %deref_and_aligned_return = call dereferenceable(32) align 16 i32* @foo()
+    %load23 = load i32, i32* %no_deref_return
----------------
The syntax here feels really weird for me.  "align 16" vs "align(16)" feels very inconsistent with "dereferenceable(32)".  Not saying we need to fix that right now, just noting it.  


http://reviews.llvm.org/D12844