[PATCH] D14412: [Polly][WIP] Emit remarks for taken assumptions

[Johannes Doerfert via llvm-commits]

jdoerfert accepted this revision.
jdoerfert added a reviewer: jdoerfert.
jdoerfert marked an inline comment as done.
jdoerfert added a comment.
This revision is now accepted and ready to land.

Thanks for the feedback.


================
Comment at: lib/Analysis/ScopInfo.cpp:102
@@ -94,1 +101,3 @@
+                     cl::cat(PollyCategory));
+
 //===----------------------------------------------------------------------===//
----------------
grosser wrote:
> The option name made me first think that this is actually about taking assumptions or not, but indeed this only affects the reporting. This caused confusion. As this is anyhow can be controlled byt the -Rpass, -pass-remarks flags I wonder if we should not just drop this option.
I was thinking it would save time in the 99% of the runs we do not activate the Rpass option but I don't need it.

================
Comment at: lib/Analysis/ScopInfo.cpp:2357
@@ +2356,3 @@
+          (*II)->getAccessInstruction()->getDebugLoc(),
+          "Possible aliasing pointer, use restrict keyword.");
+
----------------
grosser wrote:
> This seems to be an independent change and could be committed by itself.
Ok.

================
Comment at: lib/Analysis/ScopInfo.cpp:2868
@@ +2867,3 @@
+  emitOptimizationRemarkAnalysis(F.getContext(), DEBUG_TYPE, F, Loc, Msg);
+}
+
----------------
grosser wrote:
> We might want to add this logic to ScopDetectionDiagnostic.cpp
Why? It has nothing to do with ScopDetection and it cannot be reused...

================
Comment at: lib/Analysis/ScopInfo.cpp:3836
@@ -3770,1 +3835,3 @@
+                                 "SCoP ends here.");
+
   DEBUG(scop->print(dbgs()));
----------------
grosser wrote:
> These changes seem unrelated and also seem to duplicate the logic in emitValidRemarks(). 
True, I will commit it separatly and drop emitValidRemarks


http://reviews.llvm.org/D14412