Implement single-precision round intrinsic in x86

Rationale:
X86 does not provide a direct instruction for the
required rounding and NaN and large positive numbers
must be dealt with too. This CL generates code that
correctly implements SP round in a reasonably
efficient manner (I hope....)

Test: 580-checker-round

BUG=26327751

Change-Id: Ic5f4d9cff9c27c855a8ad577c51ed3ed37fb60cd
diff --git a/compiler/optimizing/pc_relative_fixups_x86.cc b/compiler/optimizing/pc_relative_fixups_x86.cc
index 921f3df..ad0921d 100644
--- a/compiler/optimizing/pc_relative_fixups_x86.cc
+++ b/compiler/optimizing/pc_relative_fixups_x86.cc
@@ -227,6 +227,7 @@
       case Intrinsics::kMathMaxFloatFloat:
       case Intrinsics::kMathMinDoubleDouble:
       case Intrinsics::kMathMinFloatFloat:
+      case Intrinsics::kMathRoundFloat:
         if (!base_added) {
           DCHECK(invoke_static_or_direct != nullptr);
           DCHECK(!invoke_static_or_direct->HasCurrentMethodInput());