[turbofan]: Implement tail calls with more callee than caller parameters

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

 // Copyright 2013 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/code-generator.h"
 
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
 #include "src/compiler/osr.h"
@@ skipping 556 matching lines @@
       Register value = i.InputRegister(4);                       \
       ASSEMBLE_CHECKED_STORE_INTEGER_IMPL(asm_instr, Register);  \
     } else {                                                     \
       Immediate value = i.InputImmediate(4);                     \
       ASSEMBLE_CHECKED_STORE_INTEGER_IMPL(asm_instr, Immediate); \
     }                                                            \
   } while (false)
 
 
 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));
+  }
   CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
-  int stack_slots = frame()->GetSpillSlotCount();
-  if (descriptor->IsJSFunctionCall() || stack_slots > 0) {
-    __ movq(rsp, rbp);
+  int spill_slots = frame()->GetSpillSlotCount();
+  bool has_frame = descriptor->IsJSFunctionCall() || spill_slots > 0;
+  if (has_frame) {
     __ popq(rbp);
   }
-  if (stack_param_delta < 0) {
-    int offset = -(stack_param_delta + 1) * kPointerSize;
-    __ popq(Operand(rsp, offset));
-    if (offset != 0) {
-      __ addq(rsp, Immediate(offset));
-    }
-  }
 }
 
+
+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));
+  }
+}
+
 
 // Assembles an instruction after register allocation, producing machine code.
 void CodeGenerator::AssembleArchInstruction(Instruction* instr) {
   X64OperandConverter i(this, instr);
 
   switch (ArchOpcodeField::decode(instr->opcode())) {
     case kArchCallCodeObject: {
       EnsureSpaceForLazyDeopt();
       if (HasImmediateInput(instr, 0)) {
         Handle<Code> code = Handle<Code>::cast(i.InputHeapObject(0));
@@ skipping 46 matching lines @@
     case kArchLazyBailout: {
       EnsureSpaceForLazyDeopt();
       RecordCallPosition(instr);
       break;
     }
     case kArchPrepareCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());
       __ PrepareCallCFunction(num_parameters);
       break;
     }
+    case kArchPrepareTailCall:
+      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      break;
     case kArchCallCFunction: {
       int const num_parameters = MiscField::decode(instr->opcode());
       if (HasImmediateInput(instr, 0)) {
         ExternalReference ref = i.InputExternalReference(0);
         __ CallCFunction(ref, num_parameters);
       } else {
         Register func = i.InputRegister(0);
         __ CallCFunction(func, num_parameters);
       }
       break;
@@ skipping 1272 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8