[turbofan] Frame elision for code stubs

src/compiler/code-generator.cc

Index: src/compiler/code-generator.cc
diff --git a/src/compiler/code-generator.cc b/src/compiler/code-generator.cc
index b091a5ceb8ffda5c142c7784197ac1e1cc2086f4..b68c83f3457f1675ec1f169cf5c0f22fe5450077 100644
--- a/src/compiler/code-generator.cc
+++ b/src/compiler/code-generator.cc
@@ -56,12 +56,8 @@ CodeGenerator::CodeGenerator(Frame* frame, Linkage* linkage,
   for (int i = 0; i < code->InstructionBlockCount(); ++i) {
     new (&labels_[i]) Label;
   }
-  if (code->ContainsCall()) {
-    frame->MarkNeedsFrame();
-  }
 }
 
-
 Handle<Code> CodeGenerator::GenerateCode() {
   CompilationInfo* info = this->info();
 
@@ -80,7 +76,6 @@ Handle<Code> CodeGenerator::GenerateCode() {
   }
   // Architecture-specific, linkage-specific prologue.
   info->set_prologue_offset(masm()->pc_offset());
-  AssemblePrologue();
   if (linkage()->GetIncomingDescriptor()->InitializeRootRegister()) {
     masm()->InitializeRootRegister();
   }
@@ -104,6 +99,9 @@ Handle<Code> CodeGenerator::GenerateCode() {
     }
   }
 
+  // Finish the Frame
+  frame()->AlignFrame(kFrameAlignmentInBytes);
+  AssembleSetupStackPointer();
   // Assemble all non-deferred blocks, followed by deferred ones.
   for (int deferred = 0; deferred < 2; ++deferred) {
     for (const InstructionBlock* block : code()->instruction_blocks()) {
@@ -143,9 +141,23 @@ Handle<Code> CodeGenerator::GenerateCode() {
         SNPrintF(buffer, " --");
         masm()->RecordComment(buffer_start);
       }
+
+      frame_access_state()->MarkHasFrame(block->needs_frame());
+
       masm()->bind(GetLabel(current_block_));
+      if (block->must_construct_frame()) {
+        AssemblePrologue();

[Benedikt Meurer, 2016/03/30 04:00:25]

Maybe we should rename AssemblePrologue to AssembleConstructFrame for
consistency? Then it would be obvious that these two are dual.

[Mircea Trofin, 2016/03/30 04:13:27]

I want to do that, but together with separating frame mutation out of it, so in
a separate CL.

[Benedikt Meurer, 2016/03/30 04:34:08]

Acknowledged.

+      }
+
       for (int i = block->code_start(); i < block->code_end(); ++i) {
-        AssembleInstruction(code()->InstructionAt(i));
+        Instruction* instr = code()->InstructionAt(i);

[Benedikt Meurer, 2016/03/30 04:00:25]

Please put all of this into AssembleInstruction below and let the for-loop body
be just the call to AssembleInstruction.

[Mircea Trofin, 2016/03/30 04:13:27]

We need the operands involved in the gap to be emitted before the frame is
deconstructed.

[Benedikt Meurer, 2016/03/30 04:34:08]

Sure, but that can all be done inside AssembleInstruction, which previously
contained the AssembleGaps call.

[Mircea Trofin, 2016/03/30 05:58:20]

Done.

+        AssembleGaps(instr);
+        if (block->must_deconstruct_frame() &&

[Benedikt Meurer, 2016/03/30 04:00:25]

This condition looks unnecessarily complex. Isn't what you actually want just

if (instr->IsJump() && block->must_deconstruct_frame()) {
  DCHECK_EQ(block->last_instruction_index(), i);
  DCHECK(!instr->IsRet());
  ...
}

or am I missing something? Also even for instr->IsRet() you probably need to
deconstruct the frame somehow (I guess that's part of AssembleReturn tho).

[Mircea Trofin, 2016/03/30 04:13:27]

That's true for ret, and I plan to address that separately. I agree with the
condition change.

[Benedikt Meurer, 2016/03/30 04:34:08]

Acknowledged.

+            i == block->last_instruction_index() && !instr->IsRet()) {
+          DCHECK(instr->IsJump());
+          AssembleDeconstructFrame();
+        }
+        AssembleInstruction(instr);
       }
     }
   }
@@ -290,9 +302,7 @@ bool CodeGenerator::IsMaterializableFromRoot(
   return false;
 }
 
-
 void CodeGenerator::AssembleInstruction(Instruction* instr) {
-  AssembleGaps(instr);
   AssembleSourcePosition(instr);
   // Assemble architecture-specific code for the instruction.
   AssembleArchInstruction(instr);
@@ -761,19 +771,16 @@ DeoptimizationExit* CodeGenerator::AddDeoptimizationExit(
 }
 
 int CodeGenerator::TailCallFrameStackSlotDelta(int stack_param_delta) {
-  CallDescriptor* descriptor = linkage()->GetIncomingDescriptor();
-  int spill_slots = frame()->GetSpillSlotCount();
-  bool has_frame = descriptor->IsJSFunctionCall() || spill_slots > 0;
   // 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 =
-      has_frame ? (frame()->GetTotalFrameSlotCount() - pc_slots) : 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;