Optimize double.floor and double.ceil down to roundsd when SSE4.1 is available.

runtime/vm/intermediate_language_x64.cc

Index: runtime/vm/intermediate_language_x64.cc
diff --git a/runtime/vm/intermediate_language_x64.cc b/runtime/vm/intermediate_language_x64.cc
index 0027008c8b7fef19e3e32e5562ca7613d6115a88..b524282db17e56999b4da1fb6f5068f57176076c 100644
--- a/runtime/vm/intermediate_language_x64.cc
+++ b/runtime/vm/intermediate_language_x64.cc
@@ -2341,11 +2341,23 @@ LocationSummary* DoubleToDoubleInstr::MakeLocationSummary() const {
 void DoubleToDoubleInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   XmmRegister value = locs()->in(0).fpu_reg();
   XmmRegister result = locs()->out().fpu_reg();
-  if (recognized_kind() == MethodRecognizer::kDoubleTruncate) {
-    __ roundsd(result, value,  Assembler::kRoundToZero);
-  } else {
-    XmmRegister temp = locs()->temp(0).fpu_reg();
-    __ DoubleRound(result, value, temp);
+  switch (recognized_kind()) {
+    case MethodRecognizer::kDoubleTruncate:
+      __ roundsd(result, value,  Assembler::kRoundToZero);
+      break;
+    case MethodRecognizer::kDoubleFloor:
+      __ roundsd(result, value,  Assembler::kRoundDown);
+      break;
+    case MethodRecognizer::kDoubleCeil:
+      __ roundsd(result, value,  Assembler::kRoundUp);
+      break;
+    case MethodRecognizer::kDoubleRound: {
+      XmmRegister temp = locs()->temp(0).fpu_reg();
+      __ DoubleRound(result, value, temp);
+      break;
+    }
+    default:
+      UNREACHABLE();
   }
 }