[turbofan] Implement rounding of floats on x64 and ia32 without sse4.1.

src/snapshot/serialize.cc

Index: src/snapshot/serialize.cc
diff --git a/src/snapshot/serialize.cc b/src/snapshot/serialize.cc
index 58a80940ee51b8d8403b7daa156bb19e3c022e04..7fcdca4c601ff322e556d763fab5a7422149d347 100644
--- a/src/snapshot/serialize.cc
+++ b/src/snapshot/serialize.cc
@@ -109,6 +109,8 @@ ExternalReferenceTable::ExternalReferenceTable(Isolate* isolate) {
   Add(ExternalReference::allocation_sites_list_address(isolate).address(),
       "Heap::allocation_sites_list_address()");
   Add(ExternalReference::address_of_uint32_bias().address(), "uint32_bias");
+  Add(ExternalReference::address_of_rounding_limit().address(),
+      "rounding_limit");
   Add(ExternalReference::get_mark_code_as_executed_function(isolate).address(),
       "Code::MarkCodeAsExecuted");
   Add(ExternalReference::is_profiling_address(isolate).address(),