In the joiner, merge the small interval into the large interval. This restores

us back to taking about 10.5s on gcc, instead of taking 15.6s!  The net result
is that my big patches have hand no significant effect on compile time or code
quality.  heh.

llvm-svn: 15156

1 files changed, 9 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/LiveInterval.cpp b/llvm/lib/CodeGen/LiveInterval.cpp
index 1cd9fb1a4a8..79aa710c03b 100644
--- a/llvm/lib/CodeGen/LiveInterval.cpp
+++ b/llvm/lib/CodeGen/LiveInterval.cpp
@@ -260,6 +260,15 @@ void LiveInterval::join(LiveInterval &Other, unsigned CopyIdx) {
   unsigned MergedSrcValIdx = SourceLR->ValId;
   unsigned MergedDstValIdx = DestLR->ValId;
 
+  // Try to do the least amount of work possible.  In particular, if there are
+  // more liverange chunks in the other set than there are in the 'this' set,
+  // swap sets to merge the fewest chunks in possible.
+  if (Other.ranges.size() > ranges.size()) {
+    std::swap(MergedSrcValIdx, MergedDstValIdx);
+    std::swap(ranges, Other.ranges);
+    std::swap(NumValues, Other.NumValues);
+  }
+
   // Join the ranges of other into the ranges of this interval.
   Ranges::iterator InsertPos = ranges.begin();
   std::map<unsigned, unsigned> Dst2SrcIdxMap;