VM: Fix FP math discrepancies by adding -mfpmath=sse to 32-bit simulator and cross host builds.

tests/language/vm/regress_24517_test.dart

Index: tests/language/vm/regress_24517_test.dart
diff --git a/tests/language/vm/regress_24517_test.dart b/tests/language/vm/regress_24517_test.dart
new file mode 100644
index 0000000000000000000000000000000000000000..654caf9956310266bda4bd65749f9440d3480b6f
--- /dev/null
+++ b/tests/language/vm/regress_24517_test.dart
@@ -0,0 +1,16 @@
+// Copyright (c) 2015, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+
+// VMOptions=--no-intrinsify
+
+// Test that math runtime function (non-intrinsified) produce the expected
+// result and don't deviate due to double-rounding when using 80-bit FP ops.
+
+import "dart:math";
+import "package:expect/expect.dart";
+
+main() {
+  var x = 2.028240960366921e+31;
+  Expect.equals(4503599627372443.0, sqrt(x));
+}