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

src/compiler/ia32/code-generator-ia32.cc

 // Copyright 2013 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 "src/compiler/code-generator.h"
 
 #include "src/ast/scopes.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
@@ skipping 636 matching lines @@
       break;
     }
     case kSSEFloat32Neg: {
       // TODO(bmeurer): Use 128-bit constants.
       __ pcmpeqd(kScratchDoubleReg, kScratchDoubleReg);
       __ psllq(kScratchDoubleReg, 31);
       __ xorps(i.OutputDoubleRegister(), kScratchDoubleReg);
       break;
     }
     case kSSEFloat32Round: {
-      CpuFeatureScope sse_scope(masm(), SSE4_1);
       RoundingMode const mode =
           static_cast<RoundingMode>(MiscField::decode(instr->opcode()));
+      if (CpuFeatures::IsSupported(SSE4_1)) {
+        CpuFeatureScope sse_scope(masm(), SSE4_1);
       __ roundss(i.OutputDoubleRegister(), i.InputDoubleRegister(0), mode);
+      } else {
+        Register scratch = i.TempRegister(0);
+        __ Roundss(i.OutputDoubleRegister(), i.InputDoubleRegister(0), scratch,
+                   kScratchDoubleReg, mode);
+      }
       break;
     }
     case kSSEFloat64Cmp:
       __ ucomisd(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat64Add:
       __ addsd(i.InputDoubleRegister(0), i.InputOperand(1));
       break;
     case kSSEFloat64Sub:
       __ subsd(i.InputDoubleRegister(0), i.InputOperand(1));
@@ skipping 49 matching lines @@
       // TODO(bmeurer): Use 128-bit constants.
       __ pcmpeqd(kScratchDoubleReg, kScratchDoubleReg);
       __ psllq(kScratchDoubleReg, 63);
       __ xorpd(i.OutputDoubleRegister(), kScratchDoubleReg);
       break;
     }
     case kSSEFloat64Sqrt:
       __ sqrtsd(i.OutputDoubleRegister(), i.InputOperand(0));
       break;
     case kSSEFloat64Round: {
-      CpuFeatureScope sse_scope(masm(), SSE4_1);
       RoundingMode const mode =
           static_cast<RoundingMode>(MiscField::decode(instr->opcode()));
-      __ roundsd(i.OutputDoubleRegister(), i.InputDoubleRegister(0), mode);
+      if (CpuFeatures::IsSupported(SSE4_1)) {
+        CpuFeatureScope sse_scope(masm(), SSE4_1);
+        __ roundsd(i.OutputDoubleRegister(), i.InputDoubleRegister(0), mode);
+      } else {
+        Register scratch = i.TempRegister(0);
+        __ Roundsd(i.OutputDoubleRegister(), i.InputDoubleRegister(0), scratch,
+                   kScratchDoubleReg, mode);
+      }
       break;
     }
     case kSSEFloat32ToFloat64:
       __ cvtss2sd(i.OutputDoubleRegister(), i.InputOperand(0));
       break;
     case kSSEFloat64ToFloat32:
       __ cvtsd2ss(i.OutputDoubleRegister(), i.InputOperand(0));
       break;
     case kSSEFloat32ToInt32:
       __ cvttss2si(i.OutputRegister(), i.InputOperand(0));
@@ skipping 959 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8