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

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

Index: src/compiler/arm/code-generator-arm.cc
diff --git a/src/compiler/arm/code-generator-arm.cc b/src/compiler/arm/code-generator-arm.cc
index 347079c31f0dfb61c4dc27c1bcd5095f0f66ce77..b7c3d9128c7563f99a1aeb39ddfc452324aab216 100644
--- a/src/compiler/arm/code-generator-arm.cc
+++ b/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());