diff --git a/lib/Target/TargetSelectionDAG.td b/lib/Target/TargetSelectionDAG.td
index e0585b7..193a5c3 100644
--- a/lib/Target/TargetSelectionDAG.td
+++ b/lib/Target/TargetSelectionDAG.td
@@ -137,6 +137,10 @@
   SDTCisPtrTy<1>  
 ]>;
 
+def SDTStore : SDTypeProfile<0, 2, [ // store
+  SDTCisInt<1>  
+]>;
+
 //===----------------------------------------------------------------------===//
 // Selection DAG Node Properties.
 //
@@ -227,6 +231,7 @@
 def writeport  : SDNode<"ISD::WRITEPORT"  , SDTWritePort, [SDNPHasChain]>;
 
 def load       : SDNode<"ISD::LOAD"       , SDTLoad, [SDNPHasChain]>;
+def store      : SDNode<"ISD::STORE"      , SDTStore, [SDNPHasChain]>;
 
 //===----------------------------------------------------------------------===//
 // Selection DAG Condition Codes
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 57baee2..6c32308 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -387,39 +387,6 @@
                               getI16Imm(X86Lowering.getBytesToPopOnReturn()),
                                     Chain);
     }
-
-    case ISD::STORE: {
-      SDOperand Chain = Select(N->getOperand(0));     // Token chain.
-      SDOperand Tmp1 = Select(N->getOperand(1));
-      Opc = 0;
-      if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N->getOperand(1))) {
-        switch (CN->getValueType(0)) {
-          default: assert(0 && "Invalid type for operation!");
-          case MVT::i1:
-          case MVT::i8:  Opc = X86::MOV8mi;  break;
-          case MVT::i16: Opc = X86::MOV16mi; break;
-          case MVT::i32: Opc = X86::MOV32mi; break;
-        }
-      }
-
-      if (!Opc) {
-        switch (N->getOperand(1).getValueType()) {
-          default: assert(0 && "Cannot store this type!");
-          case MVT::i1:
-          case MVT::i8:  Opc = X86::MOV8mr;  break;
-          case MVT::i16: Opc = X86::MOV16mr; break;
-          case MVT::i32: Opc = X86::MOV32mr; break;
-          case MVT::f32: Opc = X86::MOVSSmr; break;
-          case MVT::f64: Opc = X86::FST64m;  break;
-        }
-      }
-
-      SDOperand Base, Scale, Index, Disp;
-      SelectAddr(N->getOperand(2), Base, Scale, Index, Disp);
-      return CurDAG->SelectNodeTo(N, Opc, MVT::Other,
-                                  Base, Scale, Index, Disp, Tmp1, Chain)
-        .getValue(Op.ResNo);
-    }
   }
 
   return SelectCode(Op);
diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index bba3cb2..a695e84 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -378,11 +378,14 @@
                    "mov{l} {$src, $dst|$dst, $src}",
                    [(set R32:$dst, imm:$src)]>;
 def MOV8mi  : Ii8 <0xC6, MRM0m, (ops i8mem :$dst, i8imm :$src),
-                   "mov{b} {$src, $dst|$dst, $src}", []>;
+                   "mov{b} {$src, $dst|$dst, $src}",
+                   [(store (i8 imm:$src), addr:$dst)]>;
 def MOV16mi : Ii16<0xC7, MRM0m, (ops i16mem:$dst, i16imm:$src),
-                   "mov{w} {$src, $dst|$dst, $src}", []>, OpSize;
+                   "mov{w} {$src, $dst|$dst, $src}",
+                   [(store (i16 imm:$src), addr:$dst)]>, OpSize;
 def MOV32mi : Ii32<0xC7, MRM0m, (ops i32mem:$dst, i32imm:$src),
-                   "mov{l} {$src, $dst|$dst, $src}", []>;
+                   "mov{l} {$src, $dst|$dst, $src}",
+                   [(store (i32 imm:$src), addr:$dst)]>;
 
 def MOV8rm  : I<0x8A, MRMSrcMem, (ops R8 :$dst, i8mem :$src),
                 "mov{b} {$src, $dst|$dst, $src}",
@@ -395,11 +398,14 @@
                 [(set R32:$dst, (load addr:$src))]>;
 
 def MOV8mr  : I<0x88, MRMDestMem, (ops i8mem :$dst, R8 :$src),
-                "mov{b} {$src, $dst|$dst, $src}", []>;
+                "mov{b} {$src, $dst|$dst, $src}",
+                [(store R8:$src, addr:$dst)]>;
 def MOV16mr : I<0x89, MRMDestMem, (ops i16mem:$dst, R16:$src),
-                "mov{w} {$src, $dst|$dst, $src}", []>, OpSize;
+                "mov{w} {$src, $dst|$dst, $src}",
+                [(store R16:$src, addr:$dst)]>, OpSize;
 def MOV32mr : I<0x89, MRMDestMem, (ops i32mem:$dst, R32:$src),
-                "mov{l} {$src, $dst|$dst, $src}", []>;
+                "mov{l} {$src, $dst|$dst, $src}",
+                [(store R32:$src, addr:$dst)]>;
                 
 //===----------------------------------------------------------------------===//
 //  Fixed-Register Multiplication and Division Instructions...
@@ -687,6 +693,7 @@
   def NOT32m : I<0xF7, MRM2m, (ops i32mem:$dst), "not{l} $dst", []>;
 }
 
+// TODO: inc/dec is slow for P4, but fast for Pentium-M.
 def INC8r  : I<0xFE, MRM0r, (ops R8 :$dst, R8 :$src), "inc{b} $dst",
                [(set R8:$dst, (add R8:$src, 1))]>;
 let isConvertibleToThreeAddress = 1 in {   // Can transform into LEA.
@@ -701,11 +708,13 @@
   def INC32m : I<0xFF, MRM0m, (ops i32mem:$dst), "inc{l} $dst", []>;
 }
 
-def DEC8r  : I<0xFE, MRM1r, (ops R8 :$dst, R8 :$src), "dec{b} $dst", []>;
+def DEC8r  : I<0xFE, MRM1r, (ops R8 :$dst, R8 :$src), "dec{b} $dst",
+               [(set R8:$dst, (add R8:$src, -1))]>;
 let isConvertibleToThreeAddress = 1 in {   // Can transform into LEA.
-def DEC16r : I<0xFF, MRM1r, (ops R16:$dst, R16:$src), "dec{w} $dst", []>,
-             OpSize;
-def DEC32r : I<0xFF, MRM1r, (ops R32:$dst, R32:$src), "dec{l} $dst", []>;
+def DEC16r : I<0xFF, MRM1r, (ops R16:$dst, R16:$src), "dec{w} $dst",
+               [(set R16:$dst, (add R16:$src, -1))]>, OpSize;
+def DEC32r : I<0xFF, MRM1r, (ops R32:$dst, R32:$src), "dec{l} $dst",
+               [(set R32:$dst, (add R32:$src, -1))]>;
 }
 
 let isTwoAddress = 0 in {