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

src/compiler/ia32/code-generator-ia32.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/ast/scopes.h"
 #include "src/compiler/code-generator-impl.h"
 #include "src/compiler/gap-resolver.h"
 #include "src/compiler/node-matchers.h"
@@ skipping 382 matching lines @@
     __ fstp_d(Operand(esp, 0));                                               \
     __ movsd(i.OutputDoubleRegister(), Operand(esp, 0));                      \
     __ add(esp, Immediate(kDoubleSize));                                      \
   } while (false)
 
 void CodeGenerator::AssembleDeconstructFrame() {
   __ mov(esp, ebp);
   __ 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));
   }
   frame_access_state()->SetFrameAccessToSP();
 }
 
 void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
                                                      Register, Register,
                                                      Register) {
   // There are not enough temp registers left on ia32 for a call instruction
@@ skipping 24 matching lines @@
   ParameterCount callee_args_count(args_reg);
   __ PrepareForTailCall(callee_args_count, caller_args_count_reg, scratch2,
                         scratch3, ReturnAddressState::kOnStack, scratch_count);
   __ pop(scratch3);
   __ pop(scratch2);
   __ pop(scratch1);
 
   __ 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) {
   IA32OperandConverter i(this, instr);
   InstructionCode opcode = instr->opcode();
   ArchOpcode arch_opcode = ArchOpcodeField::decode(opcode);
   switch (arch_opcode) {
     case kArchCallCodeObject: {
       EnsureSpaceForLazyDeopt();
       if (HasImmediateInput(instr, 0)) {
         Handle<Code> code = Handle<Code>::cast(i.InputHeapObject(0));
         __ call(code, RelocInfo::CODE_TARGET);
       } else {
         Register reg = i.InputRegister(0);
         __ add(reg, Immediate(Code::kHeaderSize - kHeapObjectTag));
         __ call(reg);
       }
       RecordCallPosition(instr);
       frame_access_state()->ClearSPDelta();
       break;
     }
     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);
       }
       if (HasImmediateInput(instr, 0)) {
         Handle<Code> code = Handle<Code>::cast(i.InputHeapObject(0));
         __ jmp(code, RelocInfo::CODE_TARGET);
       } else {
         Register reg = i.InputRegister(0);
         __ add(reg, Immediate(Code::kHeaderSize - kHeapObjectTag));
         __ 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: {
       EnsureSpaceForLazyDeopt();
       Register func = i.InputRegister(0);
       if (FLAG_debug_code) {
         // Check the function's context matches the context argument.
         __ cmp(esi, FieldOperand(func, JSFunction::kContextOffset));
         __ Assert(equal, kWrongFunctionContext);
       }
       __ call(FieldOperand(func, JSFunction::kCodeEntryOffset));
       RecordCallPosition(instr);
       frame_access_state()->ClearSPDelta();
       break;
     }
     case kArchTailCallJSFunctionFromJSFunction:
     case kArchTailCallJSFunction: {
       Register func = i.InputRegister(0);
       if (FLAG_debug_code) {
         // Check the function's context matches the context argument.
         __ 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: {
       // Frame alignment requires using FP-relative frame addressing.
       frame_access_state()->SetFrameAccessToFP();
       int const num_parameters = MiscField::decode(instr->opcode());
       __ PrepareCallCFunction(num_parameters, i.TempRegister(0));
       break;
     }
     case kArchPrepareTailCall:
-      AssemblePrepareTailCall(i.InputInt32(instr->InputCount() - 1));
+      AssemblePrepareTailCall();
       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);
       }
@@ skipping 1500 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8