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

src/ia32/disasm-ia32.cc

Index: src/ia32/disasm-ia32.cc
diff --git a/src/ia32/disasm-ia32.cc b/src/ia32/disasm-ia32.cc
index 5a432806592f7c03903e66c99e629d8f87ab70f7..dc64887c8254efbd6e82350e582a55cd77601707 100644
--- a/src/ia32/disasm-ia32.cc
+++ b/src/ia32/disasm-ia32.cc
@@ -1484,6 +1484,29 @@ int DisassemblerIA32::InstructionDecode(v8::internal::Vector<char> out_buffer,
             } else {
               AppendToBuffer(",%s,cl", NameOfCPURegister(regop));
             }
+          } else if (f0byte == 0xAE) {
+            // ldmxcsr and stmxcsr
+            data += 2;
+            byte modrm = *data;
+            int mod, regop, rm;
+            get_modrm(modrm, &mod, &regop, &rm);
+            regop &= 0x7;  // The REX.R bit does not affect the operation.
+            const char* mnem = NULL;
+            switch (regop) {
+              case 2:
+                mnem = "ldmxcsr";
+                break;
+              case 3:
+                mnem = "stmxcsr";
+                break;
+              default:
+                UnimplementedInstruction();
+                return 2;
+            }
+            DCHECK_NOT_NULL(mnem);
+            AppendToBuffer("%s ", mnem);
+            data += PrintRightOperandHelper(
+                data, &DisassemblerIA32::NameOfCPURegister);
           } else if (f0byte == 0xBC) {
             data += 2;
             int mod, regop, rm;