[PATCH] [OPENMP] Loop canonical form analysis (Sema)

[Richard Smith]

LGTM

================
Comment at: lib/Sema/SemaOpenMP.cpp:1269-1270
@@ +1268,4 @@
+
+  if (ISC.Dependent())
+    return HasErrors;
+
----------------
You could remove this, and the function `Dependent` itself. I guess maybe you're keeping it because you'll want this check once you implement the FIXME below? If so, this is fine to leave here.

================
Comment at: test/OpenMP/simd_loop_messages.cpp:474-475
@@ +473,4 @@
+
+  // expected-note at +3 1 {{loop step is expected to be negative due to this condition}}
+  // expected-error at +2 1 {{increment expression must cause 'I' to decrease on each iteration of OpenMP for loop}}
+  #pragma omp simd
----------------
You can drop the `1`s here.

http://reviews.llvm.org/D3778