Faster 64-bit right-shift for the ia32 compiler.

runtime/vm/intermediate_language_ia32.cc

 // Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"  // Needed here to get TARGET_ARCH_IA32.
 #if defined(TARGET_ARCH_IA32)
 
 #include "vm/intermediate_language.h"
 
 #include "lib/error.h"
@@ skipping 2355 matching lines @@
     case Token::kSUB: {
       Label* deopt  = compiler->AddDeoptStub(deopt_id(),
                                              kDeoptBinaryMintOp);
       Label done, overflow;
       __ pextrd(EAX, right, Immediate(0));  // Lower half left
       __ pextrd(EDX, right, Immediate(1));  // Upper half left
       __ subl(ESP, Immediate(2 * kWordSize));
       __ movq(Address(ESP, 0), left);
       if (op_kind() == Token::kADD) {
         __ addl(Address(ESP, 0), EAX);
-        __ adcl(Address(ESP, 4), EDX);
+        __ adcl(Address(ESP, 1 * kWordSize), EDX);
       } else {
         __ subl(Address(ESP, 0), EAX);
-        __ sbbl(Address(ESP, 4), EDX);
+        __ sbbl(Address(ESP, 1 * kWordSize), EDX);
       }
       __ j(OVERFLOW, &overflow);
       __ movq(left, Address(ESP, 0));
       __ addl(ESP, Immediate(2 * kWordSize));
       __ jmp(&done);
       __ Bind(&overflow);
       __ addl(ESP, Immediate(2 * kWordSize));
       __ jmp(deopt);
       __ Bind(&done);
       break;
     }
     default: UNREACHABLE();
   }
 }
 
 
+LocationSummary* ShiftMintOpInstr::MakeLocationSummary() const {
+  const intptr_t kNumInputs = 2;
+  const intptr_t kNumTemps = 1;
+  LocationSummary* summary =
+      new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
+  summary->set_in(0, Location::RequiresXmmRegister());
+  summary->set_in(1, Location::RegisterLocation(ECX));
+  summary->set_temp(0, Location::RequiresRegister());
+  summary->set_out(Location::SameAsFirstInput());
+  return summary;
+}
+
+
+void ShiftMintOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
+  XmmRegister left = locs()->in(0).xmm_reg();
+  Register temp = locs()->temp(0).reg();
+  ASSERT(locs()->in(1).reg() == ECX);
+  ASSERT(locs()->out().xmm_reg() == left);
+
+  switch (op_kind()) {
+    case Token::kSHR: {
+      Label* deopt  = compiler->AddDeoptStub(deopt_id(),
+                                             kDeoptShiftMintOp);
+      __ subl(ESP, Immediate(2 * kWordSize));
+      __ movq(Address(ESP, 0), left);
+      // Deoptimize if shift count is > 31.
+      // sarl operation masks the count to 5 bits and
+      // shrd is undefined with count > operand size (32)
+      // TODO(fschneider): Support shift counts > 31 without deoptimization.
+      __ SmiUntag(ECX);
+      const Immediate kCountLimit = Immediate(31);
+      __ cmpl(ECX, kCountLimit);
+      __ j(ABOVE, deopt);
+      __ movl(temp, Address(ESP, 1 * kWordSize));
+      __ shrd(Address(ESP, 0), temp);  // Shift count in CL.
+      __ sarl(Address(ESP, 1 * kWordSize), ECX);  // Shift count in CL.
+      __ movq(left, Address(ESP, 0));
+      __ addl(ESP, Immediate(2 * kWordSize));
+      break;
+    }
+    case Token::kSHL:
+      UNIMPLEMENTED();
+      break;
+    default:
+      UNREACHABLE();
+      break;
+  }
+}
+
+
 LocationSummary* UnaryMintOpInstr::MakeLocationSummary() const {
   const intptr_t kNumInputs = 1;
   const intptr_t kNumTemps = 0;
   LocationSummary* summary =
       new LocationSummary(kNumInputs, kNumTemps, LocationSummary::kNoCall);
   summary->set_in(0, Location::RequiresXmmRegister());
   summary->set_out(Location::SameAsFirstInput());
   return summary;
 }
 
 
 void UnaryMintOpInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
   ASSERT(op_kind() == Token::kBIT_NOT);
   XmmRegister value = locs()->in(0).xmm_reg();
   ASSERT(value == locs()->out().xmm_reg());
   __ pcmpeqq(XMM0, XMM0);  // Generate all 1's.
   __ pxor(value, XMM0);
 }
 
 
 }  // namespace dart
 
 #undef __
 
 #endif  // defined TARGET_ARCH_X64