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

src/ia32/assembler-ia32.h

Index: src/ia32/assembler-ia32.h
diff --git a/src/ia32/assembler-ia32.h b/src/ia32/assembler-ia32.h
index 0b202529f9af16c10b5d12dfc44e3153bb42b2b8..39d278f38b88e9567876eee1226137466dccefa8 100644
--- a/src/ia32/assembler-ia32.h
+++ b/src/ia32/assembler-ia32.h
@@ -954,14 +954,19 @@ class Assembler : public AssemblerBase {
   void cvttss2si(Register dst, XMMRegister src) {
     cvttss2si(dst, Operand(src));
   }
+  void cvtss2si(Register dst, const Operand& src);
+  void cvtss2si(Register dst, XMMRegister src) { cvtss2si(dst, Operand(src)); }
   void cvttsd2si(Register dst, const Operand& src);
   void cvttsd2si(Register dst, XMMRegister src) {
     cvttsd2si(dst, Operand(src));
   }
+  void cvtsd2si(Register dst, const Operand& src);
   void cvtsd2si(Register dst, XMMRegister src);
 
   void cvtsi2sd(XMMRegister dst, Register src) { cvtsi2sd(dst, Operand(src)); }
   void cvtsi2sd(XMMRegister dst, const Operand& src);
+  void cvtsi2ss(XMMRegister dst, Register src) { cvtsi2ss(dst, Operand(src)); }
+  void cvtsi2ss(XMMRegister dst, const Operand& src);
   void cvtss2sd(XMMRegister dst, const Operand& src);
   void cvtss2sd(XMMRegister dst, XMMRegister src) {
     cvtss2sd(dst, Operand(src));
@@ -991,6 +996,9 @@ class Assembler : public AssemblerBase {
   void roundss(XMMRegister dst, XMMRegister src, RoundingMode mode);
   void roundsd(XMMRegister dst, XMMRegister src, RoundingMode mode);
 
+  void ldmxcsr(const Operand& dst);
+  void stmxcsr(const Operand& dst);
+
   void movmskpd(Register dst, XMMRegister src);
   void movmskps(Register dst, XMMRegister src);
 
@@ -1493,6 +1501,7 @@ class Assembler : public AssemblerBase {
   void emit_arith(int sel, Operand dst, const Immediate& x);
 
   void emit_operand(Register reg, const Operand& adr);
+  void emit_operand(int code, const Operand& adr);
 
   void emit_label(Label* label);