Build and use stack maps in the SSA compiler.

runtime/vm/flow_graph_compiler_x64.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_X64.
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/flow_graph_compiler.h"
 
 #include "lib/error.h"
@@ skipping 441 matching lines @@
 // - RAX: object.
 // - RDX: instantiator type arguments or raw_null.
 // - RCX: instantiator or raw_null.
 // Clobbers RCX and RDX.
 // Returns:
 // - true or false in RAX.
 void FlowGraphCompiler::GenerateInstanceOf(intptr_t deopt_id,
                                            intptr_t token_pos,
                                            intptr_t try_index,
                                            const AbstractType& type,
-                                           bool negate_result) {
+                                           bool negate_result,
+                                           BitmapBuilder* stack_bitmap) {
   ASSERT(type.IsFinalized() && !type.IsMalformed());
 
   const Immediate raw_null =
       Immediate(reinterpret_cast<intptr_t>(Object::null()));
   Label is_instance, is_not_instance;
   __ pushq(RCX);  // Store instantiator on stack.
   __ pushq(RDX);  // Store instantiator type arguments.
   // If type is instantiated and non-parameterized, we can inline code
   // checking whether the tested instance is a Smi.
   if (type.IsInstantiated()) {
@@ skipping 20 matching lines @@
     __ movq(RDX, Address(RSP, 0));  // Get instantiator type arguments.
     __ movq(RCX, Address(RSP, kWordSize));  // Get instantiator.
     __ PushObject(Object::ZoneHandle());  // Make room for the result.
     __ pushq(RAX);  // Push the instance.
     __ PushObject(type);  // Push the type.
     __ pushq(RCX);  // TODO(srdjan): Pass instantiator instead of null.
     __ pushq(RDX);  // Instantiator type arguments.
     __ LoadObject(RAX, test_cache);
     __ pushq(RAX);
     GenerateCallRuntime(deopt_id, token_pos, try_index,
-                        kInstanceofRuntimeEntry);
+                        kInstanceofRuntimeEntry, stack_bitmap);
     // Pop the parameters supplied to the runtime entry. The result of the
     // instanceof runtime call will be left as the result of the operation.
     __ Drop(5);
     if (negate_result) {
       __ popq(RDX);
       __ LoadObject(RAX, bool_true());
       __ cmpq(RDX, RAX);
       __ j(NOT_EQUAL, &done, Assembler::kNearJump);
       __ LoadObject(RAX, bool_false());
     } else {
@@ skipping 22 matching lines @@
 // - RDX: instantiator type arguments or raw_null.
 // - RCX: instantiator or raw_null.
 // Returns:
 // - object in RAX for successful assignable check (or throws TypeError).
 // Performance notes: positive checks must be quick, negative checks can be slow
 // as they throw an exception.
 void FlowGraphCompiler::GenerateAssertAssignable(intptr_t deopt_id,
                                                  intptr_t token_pos,
                                                  intptr_t try_index,
                                                  const AbstractType& dst_type,
-                                                 const String& dst_name) {
+                                                 const String& dst_name,
+                                                 BitmapBuilder* stack_bitmap) {
   ASSERT(token_pos >= 0);
   ASSERT(!dst_type.IsNull());
   ASSERT(dst_type.IsFinalized());
   // Assignable check is skipped in FlowGraphBuilder, not here.
   ASSERT(dst_type.IsMalformed() ||
          (!dst_type.IsDynamicType() && !dst_type.IsObjectType()));
   __ pushq(RCX);  // Store instantiator.
   __ pushq(RDX);  // Store instantiator type arguments.
   // A null object is always assignable and is returned as result.
   const Immediate raw_null =
       Immediate(reinterpret_cast<intptr_t>(Object::null()));
   Label is_assignable, runtime_call;
   __ cmpq(RAX, raw_null);
   __ j(EQUAL, &is_assignable);
 
   // Generate throw new TypeError() if the type is malformed.
   if (dst_type.IsMalformed()) {
     const Error& error = Error::Handle(dst_type.malformed_error());
     const String& error_message = String::ZoneHandle(
         Symbols::New(error.ToErrorCString()));
     __ PushObject(Object::ZoneHandle());  // Make room for the result.
     __ pushq(RAX);  // Push the source object.
     __ PushObject(dst_name);  // Push the name of the destination.
     __ PushObject(error_message);
     GenerateCallRuntime(deopt_id,
                         token_pos,
                         try_index,
-                        kMalformedTypeErrorRuntimeEntry);
+                        kMalformedTypeErrorRuntimeEntry,
+                        stack_bitmap);
     // We should never return here.
     __ int3();
 
     __ Bind(&is_assignable);  // For a null object.
     __ popq(RDX);  // Remove pushed instantiator type arguments.
     __ popq(RCX);  // Remove pushed instantiator.
     return;
   }
 
   // Generate inline type check, linking to runtime call if not assignable.
   SubtypeTestCache& test_cache = SubtypeTestCache::ZoneHandle();
   test_cache = GenerateInlineInstanceof(token_pos, dst_type,
                                         &is_assignable, &runtime_call);
 
   __ Bind(&runtime_call);
   __ movq(RDX, Address(RSP, 0));  // Get instantiator type arguments.
   __ movq(RCX, Address(RSP, kWordSize));  // Get instantiator.
   __ PushObject(Object::ZoneHandle());  // Make room for the result.
   __ pushq(RAX);  // Push the source object.
   __ PushObject(dst_type);  // Push the type of the destination.
   __ pushq(RCX);  // Instantiator.
   __ pushq(RDX);  // Instantiator type arguments.
   __ PushObject(dst_name);  // Push the name of the destination.
   __ LoadObject(RAX, test_cache);
   __ pushq(RAX);
   GenerateCallRuntime(deopt_id,
                       token_pos,
                       try_index,
-                      kTypeCheckRuntimeEntry);
+                      kTypeCheckRuntimeEntry,
+                      stack_bitmap);
   // Pop the parameters supplied to the runtime entry. The result of the
   // type check runtime call is the checked value.
   __ Drop(6);
   __ popq(RAX);
 
   __ Bind(&is_assignable);
   __ popq(RDX);  // Remove pushed instantiator type arguments.
   __ popq(RCX);  // Remove pushed instantiator.
 }
 
@@ skipping 161 matching lines @@
   if (StackSize() != 0) {
     // We need to unwind the space we reserved for locals and copied parameters.
     // The NoSuchMethodFunction stub does not expect to see that area on the
     // stack.
     __ addq(RSP, Immediate(StackSize() * kWordSize));
   }
   if (function.IsClosureFunction()) {
     GenerateCallRuntime(Isolate::kNoDeoptId,
                         0,
                         CatchClauseNode::kInvalidTryIndex,
-                        kClosureArgumentMismatchRuntimeEntry);
+                        kClosureArgumentMismatchRuntimeEntry,
+                        NULL);

[Vyacheslav Egorov (Google), 2012/08/13 12:54:46]

This NULL looks suspicious to me because we already expanded the stack above and
now we can have some garbage in out frame . 

If this call causes GC it might see all kinds of wierdness in the frame.

   } else {
     ASSERT(!IsLeaf());
     // Invoke noSuchMethod function.
     const int kNumArgsChecked = 1;
     ICData& ic_data = ICData::ZoneHandle();
     ic_data = ICData::New(function,
                           String::Handle(function.name()),
                           Isolate::kNoDeoptId,
                           kNumArgsChecked);
     __ LoadObject(RBX, ic_data);
     // RBP - 8 : PC marker, allows easy identification of RawInstruction obj.
     // RBP : points to previous frame pointer.
     // RBP + 8 : points to return address.
     // RBP + 16 : address of last argument (arg n-1).
     // RSP + 16 + 8*(n-1) : address of first argument (arg 0).
     // RBX : ic-data.
     // R10 : arguments descriptor array.
     __ call(&StubCode::CallNoSuchMethodFunctionLabel());
   }
 
   if (FLAG_trace_functions) {
     __ pushq(RAX);  // Preserve result.
     __ PushObject(Function::ZoneHandle(function.raw()));
     GenerateCallRuntime(Isolate::kNoDeoptId,
                         0,
                         CatchClauseNode::kInvalidTryIndex,
-                        kTraceFunctionExitRuntimeEntry);
+                        kTraceFunctionExitRuntimeEntry,
+                        NULL);

[Vyacheslav Egorov (Google), 2012/08/13 12:54:46]

This NULL looks suspicious to me because we already expanded the stack above and
now we can have some garbage in out frame . 

If this call causes GC it might see all kinds of wierdness in the frame.

     __ popq(RAX);  // Remove argument.
     __ popq(RAX);  // Restore result.
   }
   __ LeaveFrame();
   __ ret();
 
   __ Bind(&all_arguments_processed);
   // Nullify originally passed arguments only after they have been copied and
   // checked, otherwise noSuchMethod would not see their original values.
   // This step can be skipped in case we decide that formal parameters are
@@ skipping 73 matching lines @@
     // code size, and nop(2) increases the minimum code size appropriately.
     __ nop(2);
     __ int3();
     __ jmp(&StubCode::FixCallersTargetLabel());
     return;
   }
   // Specialized version of entry code from CodeGenerator::GenerateEntryCode.
   const Function& function = parsed_function().function();
 
   const int parameter_count = function.num_fixed_parameters();
-  const int num_copied_params = parsed_function().copied_parameter_count();
+  const int copied_parameter_count = parsed_function().copied_parameter_count();
   const int local_count = parsed_function().stack_local_count();
   __ Comment("Enter frame");
   if (IsLeaf()) {
     AssemblerMacros::EnterDartLeafFrame(assembler(), (StackSize() * kWordSize));
   } else {
     AssemblerMacros::EnterDartFrame(assembler(), (StackSize() * kWordSize));
   }
   // We check the number of passed arguments when we have to copy them due to
   // the presence of optional named parameters.
   // No such checking code is generated if only fixed parameters are declared,
   // unless we are debug mode or unless we are compiling a closure.
-  if (num_copied_params == 0) {
+  if (copied_parameter_count == 0) {
 #ifdef DEBUG
     const bool check_arguments = true;
 #else
     const bool check_arguments = function.IsClosureFunction();
 #endif
     if (check_arguments) {
       __ Comment("Check argument count");
       // Check that num_fixed <= argc <= num_params.
       Label argc_in_range;
       // Total number of args is the first Smi in args descriptor array (R10).
       __ movq(RAX, FieldAddress(R10, Array::data_offset()));
       __ cmpq(RAX, Immediate(Smi::RawValue(parameter_count)));
       __ j(EQUAL, &argc_in_range, Assembler::kNearJump);
       if (function.IsClosureFunction()) {
         GenerateCallRuntime(Isolate::kNoDeoptId,
                             function.token_pos(),
                             CatchClauseNode::kInvalidTryIndex,
-                            kClosureArgumentMismatchRuntimeEntry);
+                            kClosureArgumentMismatchRuntimeEntry,
+                            NULL);

[Vyacheslav Egorov (Google), 2012/08/13 12:54:46]

This NULL looks suspicious to me because we already expanded the stack above and
now we can have some garbage in out frame . 

If this call causes GC it might see all kinds of wierdness in the frame.

       } else {
         __ Stop("Wrong number of arguments");
       }
       __ Bind(&argc_in_range);
     }
   } else {
     CopyParameters();
   }
 
+  // Initialize (non-argument) stack allocated slots to null.
+  //
   // TODO(vegorov): introduce stack maps and stop initializing all spill slots
   // with null.
-  const intptr_t stack_slot_count =
-      is_ssa_ ? block_order_[0]->AsGraphEntry()->spill_slot_count()
-              : local_count;
-
+  intptr_t uninitialized_slot_count;
+  if (is_ssa_) {
+    GraphEntryInstr* entry = block_order_[0]->AsGraphEntry();
+    uninitialized_slot_count =
+        entry->spill_slot_count() - copied_parameter_count;
+  } else {
+    uninitialized_slot_count = local_count;
+  }
   const intptr_t slot_base = parsed_function().first_stack_local_index();
 
-  // Initialize (non-argument) stack allocated locals to null.
-  if (stack_slot_count > 0) {
+  if (uninitialized_slot_count > 0) {
     __ Comment("Initialize spill slots");
     const Immediate raw_null =
         Immediate(reinterpret_cast<intptr_t>(Object::null()));
     __ movq(RAX, raw_null);
-    for (intptr_t i = 0; i < stack_slot_count; ++i) {
+    for (intptr_t i = 0; i < uninitialized_slot_count; ++i) {
       // Subtract index i (locals lie at lower addresses than RBP).
       __ movq(Address(RBP, (slot_base - i) * kWordSize), RAX);
     }
   }
 
   if (FLAG_print_scopes) {
     // Print the function scope (again) after generating the prologue in order
     // to see annotations such as allocation indices of locals.
     if (FLAG_print_ast) {
       // Second printing.
@@ skipping 13 matching lines @@
                                        Isolate::kNoDeoptId,
                                        0,
                                        -1);
   __ jmp(&StubCode::FixCallersTargetLabel());
 }
 
 
 void FlowGraphCompiler::GenerateCall(intptr_t token_pos,
                                      intptr_t try_index,
                                      const ExternalLabel* label,
-                                     PcDescriptors::Kind kind) {
+                                     PcDescriptors::Kind kind,
+                                     BitmapBuilder* stack_bitmap) {
   ASSERT(!IsLeaf());
   ASSERT(frame_register_allocator()->IsSpilled());
   __ call(label);
+  if (is_ssa() && (stack_bitmap != NULL)) {

[Vyacheslav Egorov (Google), 2012/08/13 12:54:46]

I would prefer to have stack_bitmap always not equal to NULL with assertion.

All call sites should have some kind of bitmap.

[Kevin Millikin (Google), 2012/08/13 15:10:37]

I agree completely.  With this change we fall back on the existing behavior for
those points, which at least should not introduce new bugs.

+    stackmap_table_builder_->AddEntry(assembler()->CodeSize(), stack_bitmap);
+  }
   AddCurrentDescriptor(kind, Isolate::kNoDeoptId, token_pos, try_index);
 }
 
 
 void FlowGraphCompiler::GenerateCallRuntime(intptr_t deopt_id,
                                             intptr_t token_pos,
                                             intptr_t try_index,
-                                            const RuntimeEntry& entry) {
+                                            const RuntimeEntry& entry,
+                                            BitmapBuilder* stack_bitmap) {
   ASSERT(!IsLeaf());
   ASSERT(frame_register_allocator()->IsSpilled());
   __ CallRuntime(entry);
+  if (is_ssa() && (stack_bitmap != NULL)) {
+    stackmap_table_builder_->AddEntry(assembler()->CodeSize(), stack_bitmap);
+  }
   AddCurrentDescriptor(PcDescriptors::kOther, deopt_id, token_pos, try_index);
 }
 
 
 intptr_t FlowGraphCompiler::EmitInstanceCall(ExternalLabel* target_label,
                                              const ICData& ic_data,
                                              const Array& arguments_descriptor,
                                              intptr_t argument_count) {
   ASSERT(!IsLeaf());
   __ LoadObject(RBX, ic_data);
@@ skipping 174 matching lines @@
 void ParallelMoveResolver::Exchange(const Address& mem1, const Address& mem2) {
   __ Exchange(mem1, mem2);
 }
 
 
 #undef __
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_X64