[turbofan] Better and more sane support for tail calls

src/compiler/instruction-selector.cc

 // Copyright 2014 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/instruction-selector.h"
 
 #include <limits>
 
 #include "src/base/adapters.h"
 #include "src/compiler/instruction-selector-impl.h"
@@ skipping 357 matching lines @@
                : (frame_state_descriptor->GetTotalSize() +
                   1);  // Include deopt id.
   }
 };
 
 
 // TODO(bmeurer): Get rid of the CallBuffer business and make
 // InstructionSelector::VisitCall platform independent instead.
 void InstructionSelector::InitializeCallBuffer(Node* call, CallBuffer* buffer,
                                                bool call_code_immediate,
-                                               bool call_address_immediate) {
+                                               bool call_address_immediate,
+                                               bool call_tail,
+                                               int stack_param_delta) {
   OperandGenerator g(this);
   DCHECK_LE(call->op()->ValueOutputCount(),
             static_cast<int>(buffer->descriptor->ReturnCount()));
   DCHECK_EQ(
       call->op()->ValueInputCount(),
       static_cast<int>(buffer->input_count() + buffer->frame_state_count()));
 
   if (buffer->descriptor->ReturnCount() > 0) {
     // Collect the projections that represent multiple outputs from this call.
     if (buffer->descriptor->ReturnCount() == 1) {
@@ skipping 86 matching lines @@
   // Split the arguments into pushed_nodes and instruction_args. Pushed
   // arguments require an explicit push instruction before the call and do
   // not appear as arguments to the call. Everything else ends up
   // as an InstructionOperand argument to the call.
   auto iter(call->inputs().begin());
   size_t pushed_count = 0;
   for (size_t index = 0; index < input_count; ++iter, ++index) {
     DCHECK(iter != call->inputs().end());
     DCHECK((*iter)->op()->opcode() != IrOpcode::kFrameState);
     if (index == 0) continue;  // The first argument (callee) is already done.
+
+    LinkageLocation location = buffer->descriptor->GetInputLocation(index);
+    if (call_tail) {
+      location = LinkageLocation::ConvertToTailCallerLocation(
+          buffer->descriptor->GetInputLocation(index), stack_param_delta);

[Jarin, 2015/11/12 13:45:12]

Replace 'buffer->descriptor->GetInputLocation(index)' with 'location' here.

[danno, 2015/11/13 10:04:17]

Done.

+    }
     InstructionOperand op =
-        g.UseLocation(*iter, buffer->descriptor->GetInputLocation(index),
-                      buffer->descriptor->GetInputType(index));
-    if (UnallocatedOperand::cast(op).HasFixedSlotPolicy()) {
+        g.UseLocation(*iter, location, buffer->descriptor->GetInputType(index));
+    if (UnallocatedOperand::cast(op).HasFixedSlotPolicy() && !call_tail) {
       int stack_index = -UnallocatedOperand::cast(op).fixed_slot_index() - 1;
       if (static_cast<size_t>(stack_index) >= buffer->pushed_nodes.size()) {
         buffer->pushed_nodes.resize(stack_index + 1, NULL);
       }
       DCHECK(!buffer->pushed_nodes[stack_index]);
       buffer->pushed_nodes[stack_index] = *iter;
       pushed_count++;
     } else {
       buffer->instruction_args.push_back(op);
     }
@@ skipping 660 matching lines @@
         node->InputAt(static_cast<int>(descriptor->InputCount())));
   }
 
   CallBuffer buffer(zone(), descriptor, frame_state_descriptor);
 
   // Compute InstructionOperands for inputs and outputs.
   // TODO(turbofan): on some architectures it's probably better to use
   // the code object in a register if there are multiple uses of it.
   // Improve constant pool and the heuristics in the register allocator
   // for where to emit constants.
-  InitializeCallBuffer(node, &buffer, true, true);
+  InitializeCallBuffer(node, &buffer, true, true, false);

[Jarin, 2015/11/12 13:45:12]

Hmm, the bool parameters are getting a bit out of hand, perhaps it is time to
introduce enums or flags?

[danno, 2015/11/13 10:04:17]

Done.

 
   EmitPrepareArguments(&(buffer.pushed_nodes), descriptor, node);
 
   // Pass label of exception handler block.
   CallDescriptor::Flags flags = descriptor->flags();
   if (handler) {
     DCHECK_EQ(IrOpcode::kIfException, handler->front()->opcode());
     IfExceptionHint hint = OpParameter<IfExceptionHint>(handler->front());
     if (hint == IfExceptionHint::kLocallyCaught) {
       flags |= CallDescriptor::kHasLocalCatchHandler;
@@ skipping 32 matching lines @@
 
 void InstructionSelector::VisitTailCall(Node* node) {
   OperandGenerator g(this);
   CallDescriptor const* descriptor = OpParameter<CallDescriptor const*>(node);
   DCHECK_NE(0, descriptor->flags() & CallDescriptor::kSupportsTailCalls);
   DCHECK_EQ(0, descriptor->flags() & CallDescriptor::kPatchableCallSite);
   DCHECK_EQ(0, descriptor->flags() & CallDescriptor::kNeedsNopAfterCall);
 
   // TODO(turbofan): Relax restriction for stack parameters.
 
-  if (linkage()->GetIncomingDescriptor()->CanTailCall(node)) {
+  int stack_param_delta = 0;
+  if (linkage()->GetIncomingDescriptor()->CanTailCall(node,
+                                                      &stack_param_delta)) {
     CallBuffer buffer(zone(), descriptor, nullptr);
 
     // Compute InstructionOperands for inputs and outputs.
-    InitializeCallBuffer(node, &buffer, true, IsTailCallAddressImmediate());
+    InitializeCallBuffer(node, &buffer, true, IsTailCallAddressImmediate(),
+                         true, stack_param_delta);
 
     // Select the appropriate opcode based on the call type.
     InstructionCode opcode;
     switch (descriptor->kind()) {
       case CallDescriptor::kCallCodeObject:
         opcode = kArchTailCallCodeObject;
         break;
       case CallDescriptor::kCallJSFunction:
         opcode = kArchTailCallJSFunction;
         break;
       default:
         UNREACHABLE();
         return;
     }
     opcode |= MiscField::encode(descriptor->flags());
 
+    buffer.instruction_args.push_back(g.TempImmediate(stack_param_delta));
+
     // Emit the tailcall instruction.
     Emit(opcode, 0, nullptr, buffer.instruction_args.size(),
          &buffer.instruction_args.front());
   } else {
     FrameStateDescriptor* frame_state_descriptor =
         descriptor->NeedsFrameState()
             ? GetFrameStateDescriptor(
                   node->InputAt(static_cast<int>(descriptor->InputCount())))
             : nullptr;
 
     CallBuffer buffer(zone(), descriptor, frame_state_descriptor);
 
     // Compute InstructionOperands for inputs and outputs.
-    InitializeCallBuffer(node, &buffer, true, IsTailCallAddressImmediate());
+    InitializeCallBuffer(node, &buffer, true, IsTailCallAddressImmediate(),
+                         false);
 
     EmitPrepareArguments(&(buffer.pushed_nodes), descriptor, node);
 
     // Select the appropriate opcode based on the call type.
     InstructionCode opcode;
     switch (descriptor->kind()) {
       case CallDescriptor::kCallCodeObject:
         opcode = kArchCallCodeObject;
         break;
       case CallDescriptor::kCallJSFunction:
@@ skipping 93 matching lines @@
   return new (instruction_zone()) FrameStateDescriptor(
       instruction_zone(), state_info.type(), state_info.bailout_id(),
       state_info.state_combine(), parameters, locals, stack,
       state_info.shared_info(), outer_state);
 }
 
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8