[turbofan] Frame elision for code stubs

src/compiler/arm64/code-generator-arm64.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/code-generator.h"
 
 #include "src/arm64/frames-arm64.h"
 #include "src/arm64/macro-assembler-arm64.h"
 #include "src/ast/scopes.h"
 #include "src/compiler/code-generator-impl.h"
@@ skipping 261 matching lines @@
  public:
   OutOfLineRecordWrite(CodeGenerator* gen, Register object, Operand index,
                        Register value, Register scratch0, Register scratch1,
                        RecordWriteMode mode)
       : OutOfLineCode(gen),
         object_(object),
         index_(index),
         value_(value),
         scratch0_(scratch0),
         scratch1_(scratch1),
-        mode_(mode) {}
+        mode_(mode),
+        must_save_lr_(!gen->frame_access_state()->has_frame()) {}
 
   void Generate() final {
     if (mode_ > RecordWriteMode::kValueIsPointer) {
       __ JumpIfSmi(value_, exit());
     }
     __ CheckPageFlagClear(value_, scratch0_,
                           MemoryChunk::kPointersToHereAreInterestingMask,
                           exit());
     RememberedSetAction const remembered_set_action =
         mode_ > RecordWriteMode::kValueIsMap ? EMIT_REMEMBERED_SET
                                              : OMIT_REMEMBERED_SET;
     SaveFPRegsMode const save_fp_mode =
         frame()->DidAllocateDoubleRegisters() ? kSaveFPRegs : kDontSaveFPRegs;
-    if (!frame()->needs_frame()) {
+    if (must_save_lr_) {
       // We need to save and restore lr if the frame was elided.
       __ Push(lr);
     }
     RecordWriteStub stub(isolate(), object_, scratch0_, scratch1_,
                          remembered_set_action, save_fp_mode);
     __ Add(scratch1_, object_, index_);
     __ CallStub(&stub);
-    if (!frame()->needs_frame()) {
+    if (must_save_lr_) {
       __ Pop(lr);
     }
   }
 
  private:
   Register const object_;
   Operand const index_;
   Register const value_;
   Register const scratch0_;
   Register const scratch1_;
   RecordWriteMode const mode_;
+  bool must_save_lr_;
 };
 
 
 Condition FlagsConditionToCondition(FlagsCondition condition) {
   switch (condition) {
     case kEqual:
       return eq;
     case kNotEqual:
       return ne;
     case kSignedLessThan:
@@ skipping 133 matching lines @@
       __ asm_instr(i.OutputRegister##width(), i.InputRegister##width(0),    \
                    i.InputRegister##width(1));                              \
     } else {                                                                \
       uint32_t imm =                                                        \
           static_cast<uint32_t>(i.InputOperand##width(1).ImmediateValue()); \
       __ asm_instr(i.OutputRegister##width(), i.InputRegister##width(0),    \
                    imm % (width));                                          \
     }                                                                       \
   } while (0)
 
+void CodeGenerator::AssembleDeconstructFrame() {
+  const CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
+  if (descriptor->IsCFunctionCall() || descriptor->UseNativeStack()) {
+    __ Mov(csp, fp);
+  } else {
+    __ Mov(jssp, fp);
+  }
+  __ 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);
   }
-  if (frame()->needs_frame()) {
+  if (frame_access_state()->has_frame()) {
     __ Ldr(lr, MemOperand(fp, StandardFrameConstants::kCallerPCOffset));
     __ Ldr(fp, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
   }
   frame_access_state()->SetFrameAccessToSP();
 }
 
 void CodeGenerator::AssemblePopArgumentsAdaptorFrame(Register args_reg,
                                                      Register scratch1,
                                                      Register scratch2,
                                                      Register scratch3) {
@@ skipping 161 matching lines @@
     case kArchRet:
       AssembleReturn();
       break;
     case kArchStackPointer:
       __ mov(i.OutputRegister(), masm()->StackPointer());
       break;
     case kArchFramePointer:
       __ mov(i.OutputRegister(), fp);
       break;
     case kArchParentFramePointer:
-      if (frame_access_state()->frame()->needs_frame()) {
+      if (frame_access_state()->has_frame()) {
         __ ldr(i.OutputRegister(), MemOperand(fp, 0));
       } else {
         __ mov(i.OutputRegister(), fp);
       }
       break;
     case kArchTruncateDoubleToI:
       __ TruncateDoubleToI(i.OutputRegister(), i.InputDoubleRegister(0));
       break;
     case kArchStoreWithWriteBarrier: {
       RecordWriteMode mode =
@@ skipping 779 matching lines @@
 }
 
 
 void CodeGenerator::AssembleDeoptimizerCall(
     int deoptimization_id, Deoptimizer::BailoutType bailout_type) {
   Address deopt_entry = Deoptimizer::GetDeoptimizationEntry(
       isolate(), deoptimization_id, bailout_type);
   __ Call(deopt_entry, RelocInfo::RUNTIME_ENTRY);
 }
 
+void CodeGenerator::SetupStackPointer() {

[danno, 2016/03/24 12:53:05]

If wonder if this should be called AssemblePrologue if what is currently
AssemblePrologue become AssembleConstructFrame

[Mircea Trofin, 2016/03/24 17:46:11]

Possibly. I want to restructure AssemblePrologue because it does a few things,
including mutating the Frame (in
some cases). I'd prefer doing this rename with that restructuring.

+  const CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
+  if (descriptor->UseNativeStack() || descriptor->IsCFunctionCall()) {
+    __ SetStackPointer(csp);
+  } else {
+    __ SetStackPointer(jssp);
+  }
+}
 
 void CodeGenerator::AssemblePrologue() {
   CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
   frame()->AlignFrame(16);

[danno, 2016/03/24 12:53:05]

Shouldn't you remove this now?

[Mircea Trofin, 2016/03/24 17:46:11]

Done.

   int stack_shrink_slots = frame()->GetSpillSlotCount();
-  if (frame()->needs_frame()) {
+  if (frame_access_state()->has_frame()) {
     if (descriptor->IsJSFunctionCall()) {
       DCHECK(!descriptor->UseNativeStack());
-      __ SetStackPointer(jssp);
       __ Prologue(this->info()->GeneratePreagedPrologue());
     } else {
-      if (descriptor->UseNativeStack() || descriptor->IsCFunctionCall()) {
-        __ SetStackPointer(csp);
-      } else {
-        __ SetStackPointer(jssp);
-      }
       if (descriptor->IsCFunctionCall()) {
         __ Push(lr, fp);
         __ Mov(fp, masm_.StackPointer());
         __ Claim(stack_shrink_slots);
       } else {
         __ StubPrologue(info()->GetOutputStackFrameType(),
                         frame()->GetTotalFrameSlotCount());
       }
     }
-  } else {
-    if (descriptor->UseNativeStack()) {
-      __ SetStackPointer(csp);
-    } else {
-      __ SetStackPointer(jssp);
-    }
-    frame()->SetElidedFrameSizeInSlots(0);
   }
-  frame_access_state()->SetFrameAccessToDefault();
+
   if (info()->is_osr()) {
     // TurboFan OSR-compiled functions cannot be entered directly.
     __ Abort(kShouldNotDirectlyEnterOsrFunction);
 
     // Unoptimized code jumps directly to this entrypoint while the unoptimized
     // frame is still on the stack. Optimized code uses OSR values directly from
     // the unoptimized frame. Thus, all that needs to be done is to allocate the
     // remaining stack slots.
     if (FLAG_code_comments) __ RecordComment("-- OSR entrypoint --");
     osr_pc_offset_ = __ pc_offset();
@@ skipping 40 matching lines @@
 
   // Restore fp registers.
   CPURegList saves_fp = CPURegList(CPURegister::kFPRegister, kDRegSizeInBits,
                                    descriptor->CalleeSavedFPRegisters());
   if (saves_fp.Count() != 0) {
     __ PopCPURegList(saves_fp);
   }
 
   int pop_count = static_cast<int>(descriptor->StackParameterCount());
   if (descriptor->IsCFunctionCall()) {
-    __ Mov(csp, fp);
-    __ Pop(fp, lr);
-  } else if (frame()->needs_frame()) {
-    // Canonicalize JSFunction return sites for now.
+    AssembleDeconstructFrame();
+  } else if (frame_access_state()->has_frame()) {
     if (return_label_.is_bound()) {
       __ B(&return_label_);
       return;
     } else {
       __ Bind(&return_label_);
-      if (descriptor->UseNativeStack()) {
-        __ Mov(csp, fp);
-        pop_count += (pop_count & 1);  // align
-      } else {
-        __ Mov(jssp, fp);
-      }
-      __ Pop(fp, lr);
+      AssembleDeconstructFrame();
     }
   } else if (descriptor->UseNativeStack()) {
     pop_count += (pop_count & 1);  // align
   }
   __ Drop(pop_count);
   __ Ret();
 }
 
 
 void CodeGenerator::AssembleMove(InstructionOperand* source,
@@ skipping 178 matching lines @@
       padding_size -= kInstructionSize;
     }
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8