Faster checks in polymorphic instance calls if Smi-s are involved

runtime/vm/flow_graph_compiler_x64.cc

 // Copyright (c) 2013, 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_X64.
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/flow_graph_compiler.h"
 
 #include "vm/ast_printer.h"
@@ skipping 1453 matching lines @@
     if (tmp.IsRegister() &&
         !locs->live_registers()->ContainsRegister(tmp.reg())) {
       __ movq(tmp.reg(), Immediate(0xf7));
     }
   }
 }
 #endif
 
 
 void FlowGraphCompiler::EmitTestAndCall(const ICData& ic_data,
-                                        Register class_id_reg,
                                         intptr_t argument_count,
                                         const Array& argument_names,
-                                        Label* deopt,
+                                        Label* failed,
+                                        Label* match_found,
                                         intptr_t deopt_id,
                                         intptr_t token_index,
                                         LocationSummary* locs) {
   ASSERT(is_optimizing());
-  ASSERT(!ic_data.IsNull() && (ic_data.NumberOfUsedChecks() > 0));
-  Label match_found;
-  const intptr_t len = ic_data.NumberOfChecks();
-  GrowableArray<CidTarget> sorted(len);
-  SortICDataByCount(ic_data, &sorted, /* drop_smi = */ false);
-  ASSERT(class_id_reg != R10);
-  ASSERT(len > 0);  // Why bother otherwise.
+
+  __ Comment("EmitTestAndCall");
   const Array& arguments_descriptor =
       Array::ZoneHandle(ArgumentsDescriptor::New(argument_count,
                                                  argument_names));
   StubCode* stub_code = isolate()->stub_code();
+  // Load receiver into RAX.
+  __ movq(RAX,
+      Address(RSP, (argument_count - 1) * kWordSize));
+  __ LoadObject(R10, arguments_descriptor, PP);
 
-  __ LoadObject(R10, arguments_descriptor, PP);
-  for (intptr_t i = 0; i < len; i++) {
-    const bool is_last_check = (i == (len - 1));
+  const bool kFirstCheckIsSmi = ic_data.GetReceiverClassIdAt(0) == kSmiCid;
+  const intptr_t kNumChecks = ic_data.NumberOfChecks();
+
+  ASSERT(!ic_data.IsNull() && (kNumChecks > 0));
+
+  Label next_test;

[zra, 2015/06/15 20:17:23]

after_smi_test

+  __ testq(RAX, Immediate(kSmiTagMask));
+  if (kFirstCheckIsSmi) {
+    // Jump if receiver is not Smi.
+    if (kNumChecks == 1) {
+      __ j(NOT_ZERO, failed);
+    } else {
+      __ j(NOT_ZERO, &next_test);
+    }
+    // Do not use the code from the function, but let the code be patched so
+    // that we can record the outgoing edges to other code.
+    GenerateDartCall(deopt_id,
+                     token_index,
+                     &stub_code->CallStaticFunctionLabel(),
+                     RawPcDescriptors::kOther,
+                     locs);
+    const Function& function = Function::Handle(ic_data.GetTargetAt(0));
+    AddStaticCallTarget(function);
+    __ Drop(argument_count, RCX);
+    if (kNumChecks > 1) {
+      __ jmp(match_found);
+    }
+  } else {
+    // It is Smi, butSmi is not handled here.
+    __ j(ZERO, failed);
+  }
+  __ Bind(&next_test);
+
+  ASSERT(!ic_data.IsNull() && (kNumChecks > 0));
+  GrowableArray<CidTarget> sorted(kNumChecks);
+  SortICDataByCount(ic_data, &sorted, /* drop_smi = */ true);
+
+  const intptr_t kSortedLen = sorted.length();
+  if (kSortedLen == 0) return;
+
+  // Value is not Smi,
+  __ LoadClassId(RDI, RAX);
+  for (intptr_t i = 0; i < kSortedLen; i++) {
+    const bool kIsLastCheck = (i == (kSortedLen - 1));
     Label next_test;
-    __ cmpl(class_id_reg, Immediate(sorted[i].cid));
-    if (is_last_check) {
-      __ j(NOT_EQUAL, deopt);
+    __ cmpl(RDI, Immediate(sorted[i].cid));
+    if (kIsLastCheck) {
+      __ j(NOT_EQUAL, failed);
     } else {
       __ j(NOT_EQUAL, &next_test);
     }
     // Do not use the code from the function, but let the code be patched so
     // that we can record the outgoing edges to other code.
     GenerateDartCall(deopt_id,
                      token_index,
                      &stub_code->CallStaticFunctionLabel(),
                      RawPcDescriptors::kOther,
                      locs);
     const Function& function = *sorted[i].target;
     AddStaticCallTarget(function);
     __ Drop(argument_count, RCX);
-    if (!is_last_check) {
-      __ jmp(&match_found);
+    if (!kIsLastCheck) {
+      __ jmp(match_found);
     }
     __ Bind(&next_test);
   }
-  __ Bind(&match_found);
 }
 
 
 #undef __
 #define __ compiler_->assembler()->
 
 
 void ParallelMoveResolver::EmitMove(int index) {
   MoveOperands* move = moves_[index];
   const Location source = move->src();
@@ skipping 221 matching lines @@
   __ movups(reg, Address(RSP, 0));
   __ AddImmediate(RSP, Immediate(kFpuRegisterSize), PP);
 }
 
 
 #undef __
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_X64