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

src/ia32/disasm-ia32.cc

 // Copyright 2011 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include <assert.h>
 #include <stdarg.h>
 #include <stdio.h>
 
 #if V8_TARGET_ARCH_IA32
 
@@ skipping 1466 matching lines @@
             data += 2;
             AppendToBuffer("%s ", f0mnem);
             int mod, regop, rm;
             get_modrm(*data, &mod, &regop, &rm);
             data += PrintRightOperand(data);
             if (f0byte == 0xAB) {
               AppendToBuffer(",%s", NameOfCPURegister(regop));
             } 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;
             get_modrm(*data, &mod, &regop, &rm);
             AppendToBuffer("%s %s,", f0mnem, NameOfCPURegister(regop));
             data += PrintRightOperand(data);
           } else if (f0byte == 0xBD) {
             data += 2;
             int mod, regop, rm;
             get_modrm(*data, &mod, &regop, &rm);
@@ skipping 126 matching lines @@
                              NameOfXMMRegister(rm));
               data++;
             } else if (*data == 0x2A) {
               // movntdqa
               UnimplementedInstruction();
             } else {
               UnimplementedInstruction();
             }
           } else if (*data == 0x3A) {
             data++;
-            if (*data == 0x0B) {
+            if (*data == 0x0A) {
               data++;
               int mod, regop, rm;
               get_modrm(*data, &mod, &regop, &rm);
+              int8_t imm8 = static_cast<int8_t>(data[1]);
+              AppendToBuffer("roundss %s,%s,%d", NameOfXMMRegister(regop),
+                             NameOfXMMRegister(rm), static_cast<int>(imm8));
+              data += 2;
+            } else if (*data == 0x0B) {
+              data++;
+              int mod, regop, rm;
+              get_modrm(*data, &mod, &regop, &rm);
               int8_t imm8 = static_cast<int8_t>(data[1]);
               AppendToBuffer("roundsd %s,%s,%d",
                              NameOfXMMRegister(regop),
                              NameOfXMMRegister(rm),
                              static_cast<int>(imm8));
               data += 2;
             } else if (*data == 0x16) {
               data++;
               int mod, regop, rm;
               get_modrm(*data, &mod, &rm, &regop);
@@ skipping 606 matching lines @@
       fprintf(f, "  ");
     }
     fprintf(f, "  %s\n", buffer.start());
   }
 }
 
 
 }  // namespace disasm
 
 #endif  // V8_TARGET_ARCH_IA32