ART: x86_64 RoundDouble/Float intrinsics should initialize out value.

x86_64 RoundDouble intrinsic should initialize output register for the case of
"inPlusPointFive >= maxLong" as expected. The same for the RoundFloat intrinsic.
Fixed also the out register type in CreateSSE41FPToIntLocations provoking
a DCHECK failure.

Signed-off-by: Pavel Vyssotski <pavel.n.vyssotski@intel.com>

(cherry picked from commit 9ca257196b46fd7629bce0b338580e571e4113a8)

Bug: 22973442
Change-Id: If974e79d33311587d0b541a01ca8a4c9c11b9468
diff --git a/compiler/optimizing/intrinsics_x86_64.cc b/compiler/optimizing/intrinsics_x86_64.cc
index c245cb6..450e550 100644
--- a/compiler/optimizing/intrinsics_x86_64.cc
+++ b/compiler/optimizing/intrinsics_x86_64.cc
@@ -686,7 +686,7 @@
                                                               LocationSummary::kNoCall,
                                                               kIntrinsified);
     locations->SetInAt(0, Location::RequiresFpuRegister());
-    locations->SetOut(Location::RequiresFpuRegister());
+    locations->SetOut(Location::RequiresRegister());
     locations->AddTemp(Location::RequiresFpuRegister());
     return;
   }
@@ -728,6 +728,9 @@
   // And truncate to an integer.
   __ roundss(inPlusPointFive, inPlusPointFive, Immediate(1));
 
+  // Load maxInt into out.
+  codegen_->Load64BitValue(out, kPrimIntMax);
+
   // if inPlusPointFive >= maxInt goto done
   __ comiss(inPlusPointFive, codegen_->LiteralFloatAddress(static_cast<float>(kPrimIntMax)));
   __ j(kAboveEqual, &done);
@@ -772,6 +775,9 @@
   // And truncate to an integer.
   __ roundsd(inPlusPointFive, inPlusPointFive, Immediate(1));
 
+  // Load maxLong into out.
+  codegen_->Load64BitValue(out, kPrimLongMax);
+
   // if inPlusPointFive >= maxLong goto done
   __ comisd(inPlusPointFive, codegen_->LiteralDoubleAddress(static_cast<double>(kPrimLongMax)));
   __ j(kAboveEqual, &done);