[X86][FastISel] Teach how to select SSE4A nontemporal stores.

[oota-llvm.git] / lib / Target / X86 / X86FastISel.cpp

diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index 0771cbeb4a267b5a390a5fcb3cc6e5efb0c4caa3..263c133698cda1f47343a0d8fb3b6aeb4839e9e1 100644 (file)
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -434,6 +434,7 @@ bool X86FastISel::X86FastEmitStore(EVT VT, unsigned ValReg, bool ValIsKill,
                                    X86AddressMode &AM,
                                    MachineMemOperand *MMO, bool Aligned) {
   bool HasSSE2 = Subtarget->hasSSE2();
+  bool HasSSE4A = Subtarget->hasSSE4A();
   bool HasAVX = Subtarget->hasAVX();
   bool IsNonTemporal = MMO && MMO->isNonTemporal();
 
@@ -461,12 +462,22 @@ bool X86FastISel::X86FastEmitStore(EVT VT, unsigned ValReg, bool ValIsKill,
     Opc = (IsNonTemporal && HasSSE2) ? X86::MOVNTI_64mr : X86::MOV64mr;
     break;
   case MVT::f32:
-    Opc = X86ScalarSSEf32 ? 
-          (HasAVX ? X86::VMOVSSmr : X86::MOVSSmr) : X86::ST_Fp32m;
+    if (X86ScalarSSEf32) {
+      if (IsNonTemporal && HasSSE4A)
+        Opc = X86::MOVNTSS;
+      else
+        Opc = HasAVX ? X86::VMOVSSmr : X86::MOVSSmr;
+    } else
+      Opc = X86::ST_Fp32m;
     break;
   case MVT::f64:
-    Opc = X86ScalarSSEf64 ?
-          (HasAVX ? X86::VMOVSDmr : X86::MOVSDmr) : X86::ST_Fp64m;
+    if (X86ScalarSSEf32) {
+      if (IsNonTemporal && HasSSE4A)
+        Opc = X86::MOVNTSD;
+      else
+        Opc = HasAVX ? X86::VMOVSDmr : X86::MOVSDmr;
+    } else
+      Opc = X86::ST_Fp64m;
     break;
   case MVT::v4f32:
     if (Aligned) {