Selectively invert ICMP operands for better address optimization

src/IceCfg.h

 //===- subzero/src/IceCfg.h - Control flow graph ----------------*- C++ -*-===//
 //
 //                        The Subzero Code Generator
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 ///
 /// \file
@@ skipping 184 matching lines @@
   /// instructions from the Cfg, to speed up later passes over the instructions.
   void renumberInstructions();
   void placePhiLoads();
   void placePhiStores();
   void deletePhis();
   void advancedPhiLowering();
   void reorderNodes();
   void shuffleNodes();
   void localCSE();
   void shortCircuitJumps();
+  void invertICMP();

[John, 2016/07/07 21:20:56]

This seems like X86-specific. If it is, then it should not live in IceCfg, but
in IceTargetLoweringX86Base

[manasijm, 2016/07/07 22:43:35]

Found a nice place to move it.
Added a virtual function doAddressOptOther which is only implemented for X86. It
can be used as a general visitor to modify other instructions before load/store
opt.

 
   /// Scan allocas to determine whether we need to use a frame pointer.
   /// If SortAndCombine == true, merge all the fixed-size allocas in the
   /// entry block and emit stack or frame pointer-relative addressing.
   void processAllocas(bool SortAndCombine);
   void doAddressOpt();
   /// Find clusters of insertelement/extractelement instructions that can be
   /// replaced by a shufflevector instruction.
   void materializeVectorShuffles();
   void doArgLowering();
@@ skipping 140 matching lines @@
   using OwnerType = Cfg;
   static StringPool *getStrings(const OwnerType *PoolOwner) {
     return PoolOwner->getVarStrings();
   }
 };
 using VariableString = StringID<VariableStringPoolTraits>;
 
 } // end of namespace Ice
 
 #endif // SUBZERO_SRC_ICECFG_H