Use CFG connectedness as a secondary sort key when deciding the order of copy coalescing.

This means that well connected blocks are copy coalesced before the less connected blocks. Connected blocks are more difficult to
coalesce because intervals are more complicated, so handling them first gives a greater chance of success.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90194 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/SimpleRegisterCoalescing.cpp

@@ -2371,9 +2371,19 @@ namespace {
   struct DepthMBBCompare {
     typedef std::pair<unsigned, MachineBasicBlock*> DepthMBBPair;
     bool operator()(const DepthMBBPair &LHS, const DepthMBBPair &RHS) const {
-      if (LHS.first > RHS.first) return true;   // Deeper loops first
-      return LHS.first == RHS.first &&
-        LHS.second->getNumber() < RHS.second->getNumber();
+      // Deeper loops first
+      if (LHS.first != RHS.first)
+        return LHS.first > RHS.first;
+
+      // Prefer blocks that are more connected in the CFG. This takes care of
+      // the most difficult copies first while intervals are short.
+      unsigned cl = LHS.second->pred_size() + LHS.second->succ_size();
+      unsigned cr = RHS.second->pred_size() + RHS.second->succ_size();
+      if (cl != cr)
+        return cl > cr;
+
+      // As a last resort, sort by block number.
+      return LHS.second->getNumber() < RHS.second->getNumber();
     }
   };
 }

test/CodeGen/X86/2009-09-19-SchedCustomLoweringBug.ll

@@ -10,6 +10,7 @@ entry:
 
 bb:                                               ; preds = %bb1, %entry
 ; CHECK:      addl $1
+; CHECK-NEXT: movl %e
 ; CHECK-NEXT: adcl $0
   %i.0 = phi i64 [ 0, %entry ], [ %0, %bb1 ]      ; <i64> [#uses=1]
   %0 = add nsw i64 %i.0, 1                        ; <i64> [#uses=2]