Transpose the calculation of spill weights such that we are calculating one

register at a time. This turns out to be slightly faster than iterating over
instructions, but more importantly, it allows us to compute spill weights for
new registers created after the spill weight pass has run.

Also compute the allocation hint at the same time as the spill weight. This
allows us to use the spill weight as a cost metric for copies, and choose the
most profitable hint if there is more than one possibility.

The new hints provide a very small (< 0.1%) but universal code size improvement.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110631 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 6 insertions, 0 deletions

diff --git a/include/llvm/CodeGen/LiveIntervalAnalysis.h b/include/llvm/CodeGen/LiveIntervalAnalysis.h
index b154bf1b05..04477acfaa 100644
--- a/include/llvm/CodeGen/LiveIntervalAnalysis.h
+++ b/include/llvm/CodeGen/LiveIntervalAnalysis.h
@@ -105,6 +105,12 @@ namespace llvm {
       return r2iMap_.count(reg);
     }
 
+    /// isAllocatable - is the physical register reg allocatable in the current
+    /// function?
+    bool isAllocatable(unsigned reg) const {
+      return allocatableRegs_.test(reg);
+    }
+
     /// getScaledIntervalSize - get the size of an interval in "units,"
     /// where every function is composed of one thousand units.  This
     /// measure scales properly with empty index slots in the function.