ComplexPatterns sse_load_f32 and sse_load_f64 returns in / out chain operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@30892 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index a74116f..76aa0da 100644 (file)
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -148,7 +148,8 @@ namespace {
     bool SelectLEAAddr(SDOperand N, SDOperand &Base, SDOperand &Scale,
                        SDOperand &Index, SDOperand &Disp);
     bool SelectScalarSSELoad(SDOperand N, SDOperand &Base, SDOperand &Scale,
-                             SDOperand &Index, SDOperand &Disp);
+                             SDOperand &Index, SDOperand &Disp,
+                             SDOperand &InChain, SDOperand &OutChain);
     bool TryFoldLoad(SDOperand P, SDOperand N,
                      SDOperand &Base, SDOperand &Scale,
                      SDOperand &Index, SDOperand &Disp);
@@ -781,20 +782,22 @@ bool X86DAGToDAGISel::SelectAddr(SDOperand N, SDOperand &Base, SDOperand &Scale,
 /// match a load whose top elements are either undef or zeros.  The load flavor
 /// is derived from the type of N, which is either v4f32 or v2f64.
 bool X86DAGToDAGISel::SelectScalarSSELoad(SDOperand N, SDOperand &Base,
-                                          SDOperand &Scale,
-                                          SDOperand &Index, SDOperand &Disp) {
-#if 0
+                                          SDOperand &Scale, SDOperand &Index,
+                                          SDOperand &Disp, SDOperand &InChain,
+                                          SDOperand &OutChain) {
   if (N.getOpcode() == ISD::SCALAR_TO_VECTOR) {
-    if (N.getOperand(0).getOpcode() == ISD::LOAD) {
-      SDOperand LoadAddr = N.getOperand(0).getOperand(0);
+    InChain  = N.getOperand(0);
+    if (ISD::isNON_EXTLoad(InChain.Val)) {
+      LoadSDNode *LD = cast<LoadSDNode>(InChain);
+      SDOperand LoadAddr = LD->getBasePtr();
       if (!SelectAddr(LoadAddr, Base, Scale, Index, Disp))
         return false;
+      OutChain = LD->getChain();
       return true;
     }
   }
   // TODO: Also handle the case where we explicitly require zeros in the top
   // elements.  This is a vector shuffle from the zero vector.
-#endif
   
   return false;
 }

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 81f59fb..43965dd 100644 (file)
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -39,8 +39,10 @@ def X86pinsrw  : SDNode<"X86ISD::PINSRW", SDTypeProfile<1, 3, []>, []>;
 // These are 'extloads' from a scalar to the low element of a vector, zeroing
 // the top elements.  These are used for the SSE 'ss' and 'sd' instruction
 // forms.
-def sse_load_f32 : ComplexPattern<v4f32, 4, "SelectScalarSSELoad", []>;
-def sse_load_f64 : ComplexPattern<v2f64, 4, "SelectScalarSSELoad", []>;
+def sse_load_f32 : ComplexPattern<v4f32, 4, "SelectScalarSSELoad", [],
+                                  [SDNPHasChain]>;
+def sse_load_f64 : ComplexPattern<v2f64, 4, "SelectScalarSSELoad", [],
+                                  [SDNPHasChain]>;
 
 def ssmem : Operand<v4f32> {
   let PrintMethod = "printf32mem";