[turbofan] Stand-alone deferred block splitting - full.

src/compiler/move-optimizer.cc

Index: src/compiler/move-optimizer.cc
diff --git a/src/compiler/move-optimizer.cc b/src/compiler/move-optimizer.cc
index b869185e60ed2aed9dbc939878d1874e92e4eda6..6d71ef2e1b6b0098bb8ee1f26979ffaaf3dbfafb 100644
--- a/src/compiler/move-optimizer.cc
+++ b/src/compiler/move-optimizer.cc
@@ -59,6 +59,14 @@ void MoveOptimizer::Run() {
   }
   for (auto block : code()->instruction_blocks()) {
     if (block->PredecessorCount() <= 1) continue;
+    bool has_only_deferred = true;
+    for (RpoNumber pred_id : block->predecessors()) {
+      if (!code()->InstructionBlockAt(pred_id)->IsDeferred()) {
+        has_only_deferred = false;
+        break;
+      }
+    }
+    if (has_only_deferred) continue;

[Jarin, 2015/08/04 19:39:43]

Could you explain why we skip the optimization here?

[Mircea Trofin, 2015/08/04 20:34:46]

This would pull down common fills. If the fills occur in deferred blocks, and
the closest common successor is not deferred, we lose the optimization of
spilling/filling just in deferred blocks.

[Mircea Trofin, 2015/08/04 20:39:04]

Should have added: the last statement of the MultipleDeferredBlockSpills unit
test (EXPECT_EQ...) would fail without this logic.

     OptimizeMerge(block);
   }
   for (auto gap : to_finalize_) {