[turbofan]: Support using push instructions for setting up tail call parameters

This optimizes the passing of stack parameters in function calls.

For some architectures (ia32/x64), using pushes when possible instead
of bumping the stack and then storing parameters generates much
smaller code, and in some cases is faster (e.g. when a push of a memory
location can implement a memory-to-memory copy and thus elide an
intermediate load. On others (e.g. ARM), the benefit is smaller, where
it's only possible to elide direct stack pointer adjustment in certain cases
or combine multiple register stores into a single instruction in other limited
situations. On yet other platforms (ARM64, MIPS), there are no push instructions,
and this optimization isn't used at all.

Ideally, this mechanism would be used for both tail calls and normal calls,
but "normal" calls are currently pretty efficient, and tail calls are very
inefficient, so this CL sets the bar low for building a new mechanism to
handle parameter pushing that only needs to raise the bar on tail calls for now.

The key aspect of this change is that adjustment to the stack pointer
for tail calls (and perhaps later real calls) is an explicit step separate from
instruction selection and gap resolution, but aware of both, making it possible
to safely recognize gap moves that are actually pushes.

Review-Url: https://codereview.chromium.org/2082263002
Cr-Commit-Position: refs/heads/master@{#37477}

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

@@ -415,21 +415,7 @@ void CodeGenerator::AssembleDeconstructFrame() {
   __ LeaveFrame(StackFrame::MANUAL);
 }
 
-void CodeGenerator::AssembleDeconstructActivationRecord(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta > 0) {
-    __ add(sp, sp, Operand(sp_slot_delta * kPointerSize));
-  }
-  frame_access_state()->SetFrameAccessToDefault();
-}
-
-
-void CodeGenerator::AssemblePrepareTailCall(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta < 0) {
-    __ sub(sp, sp, Operand(-sp_slot_delta * kPointerSize));
-    frame_access_state()->IncreaseSPDelta(-sp_slot_delta);
-  }
+void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
     if (FLAG_enable_embedded_constant_pool) {
       __ ldr(cp, MemOperand(fp, StandardFrameConstants::kConstantPoolOffset));
@@ -465,6 +451,115 @@ void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
   __ bind(&done);
 }
 
+namespace {
+
+void FlushPendingPushRegisters(MacroAssembler* masm,
+                               FrameAccessState* frame_access_state,
+                               ZoneVector<Register>* pending_pushes) {
+  switch (pending_pushes->size()) {
+    case 0:
+      break;
+    case 1:
+      masm->push((*pending_pushes)[0]);
+      break;
+    case 2:
+      masm->Push((*pending_pushes)[0], (*pending_pushes)[1]);
+      break;
+    case 3:
+      masm->Push((*pending_pushes)[0], (*pending_pushes)[1],
+                 (*pending_pushes)[2]);
+      break;
+    default:
+      UNREACHABLE();
+      break;
+  }
+  frame_access_state->IncreaseSPDelta(pending_pushes->size());
+  pending_pushes->resize(0);
+}
+
+void AddPendingPushRegister(MacroAssembler* masm,
+                            FrameAccessState* frame_access_state,
+                            ZoneVector<Register>* pending_pushes,
+                            Register reg) {
+  pending_pushes->push_back(reg);
+  if (pending_pushes->size() == 3 || reg.is(ip)) {
+    FlushPendingPushRegisters(masm, frame_access_state, pending_pushes);
+  }
+}
+
+void AdjustStackPointerForTailCall(
+    MacroAssembler* masm, FrameAccessState* state, int new_slot_above_sp,
+    ZoneVector<Register>* pending_pushes = nullptr,
+    bool allow_shrinkage = true) {
+  int current_sp_offset = state->GetSPToFPSlotCount() +
+                          StandardFrameConstants::kFixedSlotCountAboveFp;
+  int stack_slot_delta = new_slot_above_sp - current_sp_offset;
+  if (stack_slot_delta > 0) {
+    if (pending_pushes != nullptr) {
+      FlushPendingPushRegisters(masm, state, pending_pushes);
+    }
+    masm->sub(sp, sp, Operand(stack_slot_delta * kPointerSize));
+    state->IncreaseSPDelta(stack_slot_delta);
+  } else if (allow_shrinkage && stack_slot_delta < 0) {
+    if (pending_pushes != nullptr) {
+      FlushPendingPushRegisters(masm, state, pending_pushes);
+    }
+    masm->add(sp, sp, Operand(-stack_slot_delta * kPointerSize));
+    state->IncreaseSPDelta(stack_slot_delta);
+  }
+}
+
+}  // namespace
+
+void CodeGenerator::AssembleTailCallBeforeGap(Instruction* instr,
+                                              int first_unused_stack_slot) {
+  CodeGenerator::PushTypeFlags flags(kImmediatePush | kScalarPush);
+  ZoneVector<MoveOperands*> pushes(zone());
+  GetPushCompatibleMoves(instr, flags, &pushes);
+
+  if (!pushes.empty() &&
+      (LocationOperand::cast(pushes.back()->destination()).index() + 1 ==
+       first_unused_stack_slot)) {
+    ArmOperandConverter g(this, instr);
+    ZoneVector<Register> pending_pushes(zone());
+    for (auto move : pushes) {
+      LocationOperand destination_location(
+          LocationOperand::cast(move->destination()));
+      InstructionOperand source(move->source());
+      AdjustStackPointerForTailCall(
+          masm(), frame_access_state(),
+          destination_location.index() - pending_pushes.size(),
+          &pending_pushes);
+      if (source.IsStackSlot()) {
+        LocationOperand source_location(LocationOperand::cast(source));
+        __ ldr(ip, g.SlotToMemOperand(source_location.index()));
+        AddPendingPushRegister(masm(), frame_access_state(), &pending_pushes,
+                               ip);
+      } else if (source.IsRegister()) {
+        LocationOperand source_location(LocationOperand::cast(source));
+        AddPendingPushRegister(masm(), frame_access_state(), &pending_pushes,
+                               source_location.GetRegister());
+      } else if (source.IsImmediate()) {
+        AddPendingPushRegister(masm(), frame_access_state(), &pending_pushes,
+                               ip);
+      } else {
+        // Pushes of non-scalar data types is not supported.
+        UNIMPLEMENTED();
+      }
+      move->Eliminate();
+    }
+    FlushPendingPushRegisters(masm(), frame_access_state(), &pending_pushes);
+  }
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot, nullptr, false);
+}
+
+void CodeGenerator::AssembleTailCallAfterGap(Instruction* instr,
+                                             int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot);
+}
+
 // Assembles an instruction after register allocation, producing machine code.
 CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     Instruction* instr) {
@@ -491,8 +586,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     }
     case kArchTailCallCodeObjectFromJSFunction:
     case kArchTailCallCodeObject: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallCodeObjectFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -508,14 +601,14 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       }
       DCHECK_EQ(LeaveCC, i.OutputSBit());
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchTailCallAddress: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       CHECK(!instr->InputAt(0)->IsImmediate());
       __ Jump(i.InputRegister(0));
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchCallJSFunction: {
@@ -543,8 +636,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ cmp(cp, kScratchReg);
         __ Assert(eq, kWrongFunctionContext);
       }
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallJSFunctionFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -554,6 +645,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ Jump(ip);
       DCHECK_EQ(LeaveCC, i.OutputSBit());
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchPrepareCallCFunction: {
@@ -564,7 +656,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kArchPrepareTailCall:
-      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());

src/compiler/arm64/code-generator-arm64.cc

@@ -541,21 +541,7 @@ void CodeGenerator::AssembleDeconstructFrame() {
   __ Pop(fp, lr);
 }
 
-void CodeGenerator::AssembleDeconstructActivationRecord(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta > 0) {
-    __ Drop(sp_slot_delta);
-  }
-  frame_access_state()->SetFrameAccessToDefault();
-}
-
-
-void CodeGenerator::AssemblePrepareTailCall(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta < 0) {
-    __ Claim(-sp_slot_delta);
-    frame_access_state()->IncreaseSPDelta(-sp_slot_delta);
-  }
+void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
     __ Ldr(lr, MemOperand(fp, StandardFrameConstants::kCallerPCOffset));
     __ Ldr(fp, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
@@ -588,6 +574,38 @@ void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
   __ bind(&done);
 }
 
+namespace {
+
+void AdjustStackPointerForTailCall(MacroAssembler* masm,
+                                   FrameAccessState* state,
+                                   int new_slot_above_sp,
+                                   bool allow_shrinkage = true) {
+  int current_sp_offset = state->GetSPToFPSlotCount() +
+                          StandardFrameConstants::kFixedSlotCountAboveFp;
+  int stack_slot_delta = new_slot_above_sp - current_sp_offset;
+  if (stack_slot_delta > 0) {
+    masm->Claim(stack_slot_delta);
+    state->IncreaseSPDelta(stack_slot_delta);
+  } else if (allow_shrinkage && stack_slot_delta < 0) {
+    masm->Drop(-stack_slot_delta);
+    state->IncreaseSPDelta(stack_slot_delta);
+  }
+}
+
+}  // namespace
+
+void CodeGenerator::AssembleTailCallBeforeGap(Instruction* instr,
+                                              int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot, false);
+}
+
+void CodeGenerator::AssembleTailCallAfterGap(Instruction* instr,
+                                             int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot);
+}
+
 // Assembles an instruction after register allocation, producing machine code.
 CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     Instruction* instr) {
@@ -623,8 +641,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     }
     case kArchTailCallCodeObjectFromJSFunction:
     case kArchTailCallCodeObject: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallCodeObjectFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -639,14 +655,14 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ Jump(target);
       }
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchTailCallAddress: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       CHECK(!instr->InputAt(0)->IsImmediate());
       __ Jump(i.InputRegister(0));
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchCallJSFunction: {
@@ -689,8 +705,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ cmp(cp, temp);
         __ Assert(eq, kWrongFunctionContext);
       }
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallJSFunctionFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -699,6 +713,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ Ldr(x10, FieldMemOperand(func, JSFunction::kCodeEntryOffset));
       __ Jump(x10);
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchPrepareCallCFunction:
@@ -708,7 +723,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       UNREACHABLE();
       break;
     case kArchPrepareTailCall:
-      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());

src/compiler/code-generator.cc

@@ -5,6 +5,7 @@
 #include "src/compiler/code-generator.h"
 
 #include "src/address-map.h"
+#include "src/base/adapters.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/linkage.h"
 #include "src/compiler/pipeline.h"
@@ -312,9 +313,95 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleBlock(
   return kSuccess;
 }
 
+bool CodeGenerator::IsValidPush(InstructionOperand source,
+                                CodeGenerator::PushTypeFlags push_type) {
+  if (source.IsImmediate() &&
+      ((push_type & CodeGenerator::kImmediatePush) != 0)) {
+    return true;
+  }
+  if ((source.IsRegister() || source.IsStackSlot()) &&
+      ((push_type & CodeGenerator::kScalarPush) != 0)) {
+    return true;
+  }
+  if ((source.IsFloatRegister() || source.IsFloatStackSlot()) &&
+      ((push_type & CodeGenerator::kFloat32Push) != 0)) {
+    return true;
+  }
+  if ((source.IsDoubleRegister() || source.IsFloatStackSlot()) &&
+      ((push_type & CodeGenerator::kFloat64Push) != 0)) {
+    return true;
+  }
+  return false;
+}
+
+void CodeGenerator::GetPushCompatibleMoves(Instruction* instr,
+                                           PushTypeFlags push_type,
+                                           ZoneVector<MoveOperands*>* pushes) {
+  pushes->clear();
+  for (int i = Instruction::FIRST_GAP_POSITION;
+       i <= Instruction::LAST_GAP_POSITION; ++i) {
+    Instruction::GapPosition inner_pos =
+        static_cast<Instruction::GapPosition>(i);
+    ParallelMove* parallel_move = instr->GetParallelMove(inner_pos);
+    if (parallel_move != nullptr) {
+      for (auto move : *parallel_move) {
+        InstructionOperand source = move->source();
+        InstructionOperand destination = move->destination();
+        int first_push_compatible_index =
+            V8_TARGET_ARCH_STORES_RETURN_ADDRESS_ON_STACK ? 1 : 0;
+        // If there are any moves from slots that will be overridden by pushes,
+        // then the full gap resolver must be used since optimization with
+        // pushes don't participate in the parallel move and might clobber
+        // values needed for the gap resolve.
+        if (source.IsStackSlot() &&
+            LocationOperand::cast(source).index() >=
+                first_push_compatible_index) {
+          pushes->clear();
+          return;
+        }
+        // TODO(danno): Right now, only consider moves from the FIRST gap for
+        // pushes. Theoretically, we could extract pushes for both gaps (there
+        // are cases where this happens), but the logic for that would also have
+        // to check to make sure that non-memory inputs to the pushes from the
+        // LAST gap don't get clobbered in the FIRST gap.
+        if (i == Instruction::FIRST_GAP_POSITION) {
+          if (destination.IsStackSlot() &&
+              LocationOperand::cast(destination).index() >=
+                  first_push_compatible_index) {
+            int index = LocationOperand::cast(destination).index();
+            if (IsValidPush(source, push_type)) {
+              if (index >= static_cast<int>(pushes->size())) {
+                pushes->resize(index + 1);
+              }
+              (*pushes)[index] = move;
+            }
+          }
+        }
+      }
+    }
+  }
+
+  // For now, only support a set of continuous pushes at the end of the list.
+  size_t push_count_upper_bound = pushes->size();
+  size_t push_begin = push_count_upper_bound;
+  for (auto move : base::Reversed(*pushes)) {
+    if (move == nullptr) break;
+    push_begin--;
+  }
+  size_t push_count = pushes->size() - push_begin;
+  std::copy(pushes->begin() + push_begin,
+            pushes->begin() + push_begin + push_count, pushes->begin());
+  pushes->resize(push_count);
+}
+
 CodeGenerator::CodeGenResult CodeGenerator::AssembleInstruction(
     Instruction* instr, const InstructionBlock* block) {
+  int first_unused_stack_slot;
+  bool adjust_stack =
+      GetSlotAboveSPBeforeTailCall(instr, &first_unused_stack_slot);
+  if (adjust_stack) AssembleTailCallBeforeGap(instr, first_unused_stack_slot);
   AssembleGaps(instr);
+  if (adjust_stack) AssembleTailCallAfterGap(instr, first_unused_stack_slot);
   DCHECK_IMPLIES(
       block->must_deconstruct_frame(),
       instr != code()->InstructionAt(block->last_instruction_index()) ||
@@ -414,6 +501,16 @@ void CodeGenerator::AssembleSourcePosition(Instruction* instr) {
   }
 }
 
+bool CodeGenerator::GetSlotAboveSPBeforeTailCall(Instruction* instr,
+                                                 int* slot) {
+  if (instr->IsTailCall()) {
+    InstructionOperandConverter g(this, instr);
+    *slot = g.InputInt32(instr->InputCount() - 1);
+    return true;
+  } else {
+    return false;
+  }
+}
 
 void CodeGenerator::AssembleGaps(Instruction* instr) {
   for (int i = Instruction::FIRST_GAP_POSITION;
@@ -796,18 +893,6 @@ DeoptimizationExit* CodeGenerator::AddDeoptimizationExit(
   return exit;
 }
 
-int CodeGenerator::TailCallFrameStackSlotDelta(int stack_param_delta) {
-  // Leave the PC on the stack on platforms that have that as part of their ABI
-  int pc_slots = V8_TARGET_ARCH_STORES_RETURN_ADDRESS_ON_STACK ? 1 : 0;
-  int sp_slot_delta = frame_access_state()->has_frame()
-                          ? (frame()->GetTotalFrameSlotCount() - pc_slots)
-                          : 0;
-  // Discard only slots that won't be used by new parameters.
-  sp_slot_delta += stack_param_delta;
-  return sp_slot_delta;
-}
-
-
 OutOfLineCode::OutOfLineCode(CodeGenerator* gen)
     : frame_(gen->frame()), masm_(gen->masm()), next_(gen->ools_) {
   gen->ools_ = this;

src/compiler/code-generator.h

@@ -102,6 +102,11 @@ class CodeGenerator final : public GapResolver::Assembler {
   void AssembleSourcePosition(Instruction* instr);
   void AssembleGaps(Instruction* instr);
 
+  // Returns true if a instruction is a tail call that needs to adjust the stack
+  // pointer before execution. The stack slot index to the empty slot above the
+  // adjusted stack pointer is returned in |slot|.
+  bool GetSlotAboveSPBeforeTailCall(Instruction* instr, int* slot);
+
   // ===========================================================================
   // ============= Architecture-specific code generation methods. ==============
   // ===========================================================================
@@ -124,18 +129,49 @@ class CodeGenerator final : public GapResolver::Assembler {
   // to tear down a stack frame.
   void AssembleReturn();
 
-  // Generates code to deconstruct a the caller's frame, including arguments.
-  void AssembleDeconstructActivationRecord(int stack_param_delta);
-
   void AssembleDeconstructFrame();
 
   // Generates code to manipulate the stack in preparation for a tail call.
-  void AssemblePrepareTailCall(int stack_param_delta);
+  void AssemblePrepareTailCall();
 
   // Generates code to pop current frame if it is an arguments adaptor frame.
   void AssemblePopArgumentsAdaptorFrame(Register args_reg, Register scratch1,
                                         Register scratch2, Register scratch3);
 
+  enum PushTypeFlag {
+    kImmediatePush = 0x1,
+    kScalarPush = 0x2,
+    kFloat32Push = 0x4,
+    kFloat64Push = 0x8,
+    kFloatPush = kFloat32Push | kFloat64Push
+  };
+
+  typedef base::Flags<PushTypeFlag> PushTypeFlags;
+
+  static bool IsValidPush(InstructionOperand source, PushTypeFlags push_type);
+
+  // Generate a list moves from an instruction that are candidates to be turned
+  // into push instructions on platforms that support them. In general, the list
+  // of push candidates are moves to a set of contiguous destination
+  // InstructionOperand locations on the stack that don't clobber values that
+  // are needed for resolve the gap or use values generated by the gap,
+  // i.e. moves that can be hoisted together before the actual gap and assembled
+  // together.
+  static void GetPushCompatibleMoves(Instruction* instr,
+                                     PushTypeFlags push_type,
+                                     ZoneVector<MoveOperands*>* pushes);
+
+  // Called before a tail call |instr|'s gap moves are assembled and allows
+  // gap-specific pre-processing, e.g. adjustment of the sp for tail calls that
+  // need it before gap moves or conversion of certain gap moves into pushes.
+  void AssembleTailCallBeforeGap(Instruction* instr,
+                                 int first_unused_stack_slot);
+  // Called after a tail call |instr|'s gap moves are assembled and allows
+  // gap-specific post-processing, e.g. adjustment of the sp for tail calls that
+  // need it after gap moves.
+  void AssembleTailCallAfterGap(Instruction* instr,
+                                int first_unused_stack_slot);
+
   // ===========================================================================
   // ============== Architecture-specific gap resolver methods. ================
   // ===========================================================================
@@ -188,12 +224,6 @@ class CodeGenerator final : public GapResolver::Assembler {
   DeoptimizationExit* AddDeoptimizationExit(Instruction* instr,
                                             size_t frame_state_offset);
 
-  // Converts the delta in the number of stack parameter passed from a tail
-  // caller to the callee into the distance (in pointers) the SP must be
-  // adjusted, taking frame elision and other relevant factors into
-  // consideration.
-  int TailCallFrameStackSlotDelta(int stack_param_delta);
-
   // ===========================================================================
 
   struct DeoptimizationState : ZoneObject {

src/compiler/gap-resolver.cc

@@ -34,7 +34,6 @@ void GapResolver::Resolve(ParallelMove* moves) const {
   }
 }
 
-
 void GapResolver::PerformMove(ParallelMove* moves, MoveOperands* move) const {
   // Each call to this function performs a move and deletes it from the move
   // graph.  We first recursively perform any move blocking this one.  We mark a

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

@@ -400,21 +400,7 @@ void CodeGenerator::AssembleDeconstructFrame() {
   __ pop(ebp);
 }
 
-void CodeGenerator::AssembleDeconstructActivationRecord(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta > 0) {
-    __ add(esp, Immediate(sp_slot_delta * kPointerSize));
-  }
-  frame_access_state()->SetFrameAccessToDefault();
-}
-
-
-void CodeGenerator::AssemblePrepareTailCall(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta < 0) {
-    __ sub(esp, Immediate(-sp_slot_delta * kPointerSize));
-    frame_access_state()->IncreaseSPDelta(-sp_slot_delta);
-  }
+void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
     __ mov(ebp, MemOperand(ebp, 0));
   }
@@ -459,6 +445,68 @@ void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
   __ bind(&done);
 }
 
+namespace {
+
+void AdjustStackPointerForTailCall(MacroAssembler* masm,
+                                   FrameAccessState* state,
+                                   int new_slot_above_sp,
+                                   bool allow_shrinkage = true) {
+  int current_sp_offset = state->GetSPToFPSlotCount() +
+                          StandardFrameConstants::kFixedSlotCountAboveFp;
+  int stack_slot_delta = new_slot_above_sp - current_sp_offset;
+  if (stack_slot_delta > 0) {
+    masm->sub(esp, Immediate(stack_slot_delta * kPointerSize));
+    state->IncreaseSPDelta(stack_slot_delta);
+  } else if (allow_shrinkage && stack_slot_delta < 0) {
+    masm->add(esp, Immediate(-stack_slot_delta * kPointerSize));
+    state->IncreaseSPDelta(stack_slot_delta);
+  }
+}
+
+}  // namespace
+
+void CodeGenerator::AssembleTailCallBeforeGap(Instruction* instr,
+                                              int first_unused_stack_slot) {
+  CodeGenerator::PushTypeFlags flags(kImmediatePush | kScalarPush);
+  ZoneVector<MoveOperands*> pushes(zone());
+  GetPushCompatibleMoves(instr, flags, &pushes);
+
+  if (!pushes.empty() &&
+      (LocationOperand::cast(pushes.back()->destination()).index() + 1 ==
+       first_unused_stack_slot)) {
+    IA32OperandConverter g(this, instr);
+    for (auto move : pushes) {
+      LocationOperand destination_location(
+          LocationOperand::cast(move->destination()));
+      InstructionOperand source(move->source());
+      AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                    destination_location.index());
+      if (source.IsStackSlot()) {
+        LocationOperand source_location(LocationOperand::cast(source));
+        __ push(g.SlotToOperand(source_location.index()));
+      } else if (source.IsRegister()) {
+        LocationOperand source_location(LocationOperand::cast(source));
+        __ push(source_location.GetRegister());
+      } else if (source.IsImmediate()) {
+        __ push(Immediate(ImmediateOperand::cast(source).inline_value()));
+      } else {
+        // Pushes of non-scalar data types is not supported.
+        UNIMPLEMENTED();
+      }
+      frame_access_state()->IncreaseSPDelta(1);
+      move->Eliminate();
+    }
+  }
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot, false);
+}
+
+void CodeGenerator::AssembleTailCallAfterGap(Instruction* instr,
+                                             int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot);
+}
+
 // Assembles an instruction after register allocation, producing machine code.
 CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     Instruction* instr) {
@@ -482,8 +530,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     }
     case kArchTailCallCodeObjectFromJSFunction:
     case kArchTailCallCodeObject: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallCodeObjectFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          no_reg, no_reg, no_reg);
@@ -497,15 +543,15 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ jmp(reg);
       }
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchTailCallAddress: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       CHECK(!HasImmediateInput(instr, 0));
       Register reg = i.InputRegister(0);
       __ jmp(reg);
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchCallJSFunction: {
@@ -529,14 +575,13 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ cmp(esi, FieldOperand(func, JSFunction::kContextOffset));
         __ Assert(equal, kWrongFunctionContext);
       }
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallJSFunctionFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          no_reg, no_reg, no_reg);
       }
       __ jmp(FieldOperand(func, JSFunction::kCodeEntryOffset));
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchPrepareCallCFunction: {
@@ -547,7 +592,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kArchPrepareTailCall:
-      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());

src/compiler/instruction-selector.cc

@@ -1812,10 +1812,12 @@ void InstructionSelector::VisitTailCall(Node* node) {
     }
     opcode |= MiscField::encode(descriptor->flags());
 
-    buffer.instruction_args.push_back(g.TempImmediate(stack_param_delta));
+    Emit(kArchPrepareTailCall, g.NoOutput());
 
-    Emit(kArchPrepareTailCall, g.NoOutput(),
-         g.TempImmediate(stack_param_delta));
+    int first_unused_stack_slot =
+        (V8_TARGET_ARCH_STORES_RETURN_ADDRESS_ON_STACK ? 1 : 0) +
+        stack_param_delta;
+    buffer.instruction_args.push_back(g.TempImmediate(first_unused_stack_slot));
 
     // Emit the tailcall instruction.
     Emit(opcode, 0, nullptr, buffer.instruction_args.size(),

src/compiler/instruction-selector.h

@@ -249,7 +249,7 @@ class InstructionSelector final {
   // {call_code_immediate} to generate immediate operands to calls of code.
   // {call_address_immediate} to generate immediate operands to address calls.
   void InitializeCallBuffer(Node* call, CallBuffer* buffer,
-                            CallBufferFlags flags, int stack_param_delta = 0);
+                            CallBufferFlags flags, int stack_slot_delta = 0);
   bool IsTailCallAddressImmediate();
   int GetTempsCountForTailCallFromJSFunction();
 

src/compiler/linkage.cc

@@ -97,14 +97,14 @@ bool CallDescriptor::CanTailCall(const Node* node,
   while (more_other || more_this) {
     if (other_input < other->InputCount()) {
       if (!other->GetInputLocation(other_input).IsRegister()) {
-        (*stack_param_delta)--;
+        (*stack_param_delta)++;
       }
     } else {
       more_other = false;
     }
     if (current_input < InputCount()) {
       if (!GetInputLocation(current_input).IsRegister()) {
-        (*stack_param_delta)++;
+        (*stack_param_delta)--;
       }
     } else {
       more_this = false;

src/compiler/linkage.h

@@ -86,7 +86,7 @@ class LinkageLocation {
       LinkageLocation caller_location, int stack_param_delta) {
     if (!caller_location.IsRegister()) {
       return LinkageLocation(STACK_SLOT,
-                             caller_location.GetLocation() - stack_param_delta);
+                             caller_location.GetLocation() + stack_param_delta);
     }
     return caller_location;
   }

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

@@ -513,21 +513,7 @@ void CodeGenerator::AssembleDeconstructFrame() {
   __ Pop(ra, fp);
 }
 
-void CodeGenerator::AssembleDeconstructActivationRecord(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta > 0) {
-    __ addiu(sp, sp, sp_slot_delta * kPointerSize);
-  }
-  frame_access_state()->SetFrameAccessToDefault();
-}
-
-
-void CodeGenerator::AssemblePrepareTailCall(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta < 0) {
-    __ Subu(sp, sp, Operand(-sp_slot_delta * kPointerSize));
-    frame_access_state()->IncreaseSPDelta(-sp_slot_delta);
-  }
+void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
     __ lw(ra, MemOperand(fp, StandardFrameConstants::kCallerPCOffset));
     __ lw(fp, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
@@ -560,6 +546,38 @@ void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
   __ bind(&done);
 }
 
+namespace {
+
+void AdjustStackPointerForTailCall(MacroAssembler* masm,
+                                   FrameAccessState* state,
+                                   int new_slot_above_sp,
+                                   bool allow_shrinkage = true) {
+  int current_sp_offset = state->GetSPToFPSlotCount() +
+                          StandardFrameConstants::kFixedSlotCountAboveFp;
+  int stack_slot_delta = new_slot_above_sp - current_sp_offset;
+  if (stack_slot_delta > 0) {
+    masm->Subu(sp, sp, stack_slot_delta * kPointerSize);
+    state->IncreaseSPDelta(stack_slot_delta);
+  } else if (allow_shrinkage && stack_slot_delta < 0) {
+    masm->Addu(sp, sp, stack_slot_delta * kPointerSize);
+    state->IncreaseSPDelta(stack_slot_delta);
+  }
+}
+
+}  // namespace
+
+void CodeGenerator::AssembleTailCallBeforeGap(Instruction* instr,
+                                              int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot, false);
+}
+
+void CodeGenerator::AssembleTailCallAfterGap(Instruction* instr,
+                                             int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot);
+}
+
 // Assembles an instruction after register allocation, producing machine code.
 CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     Instruction* instr) {
@@ -582,8 +600,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     }
     case kArchTailCallCodeObjectFromJSFunction:
     case kArchTailCallCodeObject: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallCodeObjectFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -597,11 +613,10 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ Jump(at);
       }
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchTailCallAddress: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       CHECK(!instr->InputAt(0)->IsImmediate());
       __ Jump(i.InputRegister(0));
       frame_access_state()->ClearSPDelta();
@@ -620,6 +635,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ Call(at);
       RecordCallPosition(instr);
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchTailCallJSFunctionFromJSFunction:
@@ -631,8 +647,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ Assert(eq, kWrongFunctionContext, cp, Operand(kScratchReg));
       }
 
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallJSFunctionFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -641,6 +655,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ lw(at, FieldMemOperand(func, JSFunction::kCodeEntryOffset));
       __ Jump(at);
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchPrepareCallCFunction: {
@@ -651,7 +666,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kArchPrepareTailCall:
-      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());

src/compiler/mips64/code-generator-mips64.cc

@@ -524,21 +524,7 @@ void CodeGenerator::AssembleDeconstructFrame() {
   __ Pop(ra, fp);
 }
 
-void CodeGenerator::AssembleDeconstructActivationRecord(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta > 0) {
-    __ daddiu(sp, sp, sp_slot_delta * kPointerSize);
-  }
-  frame_access_state()->SetFrameAccessToDefault();
-}
-
-
-void CodeGenerator::AssemblePrepareTailCall(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta < 0) {
-    __ Dsubu(sp, sp, Operand(-sp_slot_delta * kPointerSize));
-    frame_access_state()->IncreaseSPDelta(-sp_slot_delta);
-  }
+void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
     __ ld(ra, MemOperand(fp, StandardFrameConstants::kCallerPCOffset));
     __ ld(fp, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
@@ -571,6 +557,38 @@ void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
   __ bind(&done);
 }
 
+namespace {
+
+void AdjustStackPointerForTailCall(MacroAssembler* masm,
+                                   FrameAccessState* state,
+                                   int new_slot_above_sp,
+                                   bool allow_shrinkage = true) {
+  int current_sp_offset = state->GetSPToFPSlotCount() +
+                          StandardFrameConstants::kFixedSlotCountAboveFp;
+  int stack_slot_delta = new_slot_above_sp - current_sp_offset;
+  if (stack_slot_delta > 0) {
+    masm->Dsubu(sp, sp, stack_slot_delta * kPointerSize);
+    state->IncreaseSPDelta(stack_slot_delta);
+  } else if (allow_shrinkage && stack_slot_delta < 0) {
+    masm->Daddu(sp, sp, stack_slot_delta * kPointerSize);
+    state->IncreaseSPDelta(stack_slot_delta);
+  }
+}
+
+}  // namespace
+
+void CodeGenerator::AssembleTailCallBeforeGap(Instruction* instr,
+                                              int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot, false);
+}
+
+void CodeGenerator::AssembleTailCallAfterGap(Instruction* instr,
+                                             int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot);
+}
+
 // Assembles an instruction after register allocation, producing machine code.
 CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     Instruction* instr) {
@@ -593,8 +611,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     }
     case kArchTailCallCodeObjectFromJSFunction:
     case kArchTailCallCodeObject: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallCodeObjectFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -608,14 +624,14 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ Jump(at);
       }
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchTailCallAddress: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       CHECK(!instr->InputAt(0)->IsImmediate());
       __ Jump(i.InputRegister(0));
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchCallJSFunction: {
@@ -640,8 +656,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ ld(kScratchReg, FieldMemOperand(func, JSFunction::kContextOffset));
         __ Assert(eq, kWrongFunctionContext, cp, Operand(kScratchReg));
       }
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallJSFunctionFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -650,6 +664,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       __ ld(at, FieldMemOperand(func, JSFunction::kCodeEntryOffset));
       __ Jump(at);
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchPrepareCallCFunction: {
@@ -660,7 +675,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kArchPrepareTailCall:
-      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());

src/compiler/move-optimizer.cc

@@ -160,7 +160,7 @@ void MoveOptimizer::RemoveClobberedDestinations(Instruction* instruction) {
 
   // The ret instruction makes any assignment before it unnecessary, except for
   // the one for its input.
-  if (instruction->opcode() == ArchOpcode::kArchRet) {
+  if (instruction->IsRet() || instruction->IsTailCall()) {
     for (MoveOperands* move : *moves) {
       if (inputs.find(move->destination()) == inputs.end()) {
         move->Eliminate();

src/compiler/x64/code-generator-x64.cc

@@ -626,21 +626,7 @@ void CodeGenerator::AssembleDeconstructFrame() {
   __ popq(rbp);
 }
 
-void CodeGenerator::AssembleDeconstructActivationRecord(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta > 0) {
-    __ addq(rsp, Immediate(sp_slot_delta * kPointerSize));
-  }
-  frame_access_state()->SetFrameAccessToDefault();
-}
-
-
-void CodeGenerator::AssemblePrepareTailCall(int stack_param_delta) {
-  int sp_slot_delta = TailCallFrameStackSlotDelta(stack_param_delta);
-  if (sp_slot_delta < 0) {
-    __ subq(rsp, Immediate(-sp_slot_delta * kPointerSize));
-    frame_access_state()->IncreaseSPDelta(-sp_slot_delta);
-  }
+void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
     __ movq(rbp, MemOperand(rbp, 0));
   }
@@ -672,6 +658,68 @@ void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
   __ bind(&done);
 }
 
+namespace {
+
+void AdjustStackPointerForTailCall(MacroAssembler* masm,
+                                   FrameAccessState* state,
+                                   int new_slot_above_sp,
+                                   bool allow_shrinkage = true) {
+  int current_sp_offset = state->GetSPToFPSlotCount() +
+                          StandardFrameConstants::kFixedSlotCountAboveFp;
+  int stack_slot_delta = new_slot_above_sp - current_sp_offset;
+  if (stack_slot_delta > 0) {
+    masm->subq(rsp, Immediate(stack_slot_delta * kPointerSize));
+    state->IncreaseSPDelta(stack_slot_delta);
+  } else if (allow_shrinkage && stack_slot_delta < 0) {
+    masm->addq(rsp, Immediate(-stack_slot_delta * kPointerSize));
+    state->IncreaseSPDelta(stack_slot_delta);
+  }
+}
+
+}  // namespace
+
+void CodeGenerator::AssembleTailCallBeforeGap(Instruction* instr,
+                                              int first_unused_stack_slot) {
+  CodeGenerator::PushTypeFlags flags(kImmediatePush | kScalarPush);
+  ZoneVector<MoveOperands*> pushes(zone());
+  GetPushCompatibleMoves(instr, flags, &pushes);
+
+  if (!pushes.empty() &&
+      (LocationOperand::cast(pushes.back()->destination()).index() + 1 ==
+       first_unused_stack_slot)) {
+    X64OperandConverter g(this, instr);
+    for (auto move : pushes) {
+      LocationOperand destination_location(
+          LocationOperand::cast(move->destination()));
+      InstructionOperand source(move->source());
+      AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                    destination_location.index());
+      if (source.IsStackSlot()) {
+        LocationOperand source_location(LocationOperand::cast(source));
+        __ Push(g.SlotToOperand(source_location.index()));
+      } else if (source.IsRegister()) {
+        LocationOperand source_location(LocationOperand::cast(source));
+        __ Push(source_location.GetRegister());
+      } else if (source.IsImmediate()) {
+        __ Push(Immediate(ImmediateOperand::cast(source).inline_value()));
+      } else {
+        // Pushes of non-scalar data types is not supported.
+        UNIMPLEMENTED();
+      }
+      frame_access_state()->IncreaseSPDelta(1);
+      move->Eliminate();
+    }
+  }
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot, false);
+}
+
+void CodeGenerator::AssembleTailCallAfterGap(Instruction* instr,
+                                             int first_unused_stack_slot) {
+  AdjustStackPointerForTailCall(masm(), frame_access_state(),
+                                first_unused_stack_slot);
+}
+
 // Assembles an instruction after register allocation, producing machine code.
 CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     Instruction* instr) {
@@ -695,8 +743,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
     }
     case kArchTailCallCodeObjectFromJSFunction:
     case kArchTailCallCodeObject: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallCodeObjectFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -711,15 +757,15 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ jmp(reg);
       }
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchTailCallAddress: {
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       CHECK(!HasImmediateInput(instr, 0));
       Register reg = i.InputRegister(0);
       __ jmp(reg);
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchCallJSFunction: {
@@ -743,8 +789,6 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
         __ cmpp(rsi, FieldOperand(func, JSFunction::kContextOffset));
         __ Assert(equal, kWrongFunctionContext);
       }
-      int stack_param_delta = i.InputInt32(instr->InputCount() - 1);
-      AssembleDeconstructActivationRecord(stack_param_delta);
       if (arch_opcode == kArchTailCallJSFunctionFromJSFunction) {
         AssemblePopArgumentsAdaptorFrame(kJavaScriptCallArgCountRegister,
                                          i.TempRegister(0), i.TempRegister(1),
@@ -752,6 +796,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       }
       __ jmp(FieldOperand(func, JSFunction::kCodeEntryOffset));
       frame_access_state()->ClearSPDelta();
+      frame_access_state()->SetFrameAccessToDefault();
       break;
     }
     case kArchPrepareCallCFunction: {
@@ -762,7 +807,7 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
       break;
     }
     case kArchPrepareTailCall:
-      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      AssemblePrepareTailCall();
       break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());

test/unittests/compiler/linkage-tail-call-unittest.cc

@@ -160,7 +160,7 @@ TEST_F(LinkageTailCall, MoreRegisterAndStackParametersCallee) {
   Node* const node = Node::New(zone(), 1, op, 0, nullptr, false);
   int stack_param_delta = 0;
   EXPECT_TRUE(desc1->CanTailCall(node, &stack_param_delta));
-  EXPECT_EQ(-1, stack_param_delta);
+  EXPECT_EQ(1, stack_param_delta);
 }
 
 
@@ -181,7 +181,7 @@ TEST_F(LinkageTailCall, MoreRegisterAndStackParametersCaller) {
   Node* const node = Node::New(zone(), 1, op, 0, nullptr, false);
   int stack_param_delta = 0;
   EXPECT_TRUE(desc1->CanTailCall(node, &stack_param_delta));
-  EXPECT_EQ(1, stack_param_delta);
+  EXPECT_EQ(-1, stack_param_delta);
 }
 
 
@@ -316,7 +316,7 @@ TEST_F(LinkageTailCall, MatchingStackParametersExtraCallerRegistersAndStack) {
       Node::New(zone(), 1, op, arraysize(parameters), parameters, false);
   int stack_param_delta = 0;
   EXPECT_TRUE(desc1->CanTailCall(node, &stack_param_delta));
-  EXPECT_EQ(1, stack_param_delta);
+  EXPECT_EQ(-1, stack_param_delta);
 }
 
 
@@ -344,7 +344,7 @@ TEST_F(LinkageTailCall, MatchingStackParametersExtraCalleeRegistersAndStack) {
       Node::New(zone(), 1, op, arraysize(parameters), parameters, false);
   int stack_param_delta = 0;
   EXPECT_TRUE(desc1->CanTailCall(node, &stack_param_delta));
-  EXPECT_EQ(-1, stack_param_delta);
+  EXPECT_EQ(1, stack_param_delta);
 }
 
 }  // namespace compiler