[turbofan] Two simplifications in TF handling of generators

Mark Runtime::kInlineGeneratorGetContext as not needing a FrameState
(matching the other Generator field-loading intrinsics) and avoid
a call to PrepareEagerCheckpoint() in VisitResumeGenerator() (since
there should never be a deopt during resume).

Change-Id: I03a2d89914bc7de27bbfe6228ca115e635ea4c4e
Reviewed-on: https://chromium-review.googlesource.com/506815Reviewed-by: Jaroslav Sevcik <jarin@chromium.org>
Commit-Queue: Adam Klein <adamk@chromium.org>
Cr-Commit-Position: refs/heads/master@{#45382}

src/compiler/bytecode-graph-builder.cc

@@ -2295,8 +2295,6 @@ void BytecodeGraphBuilder::VisitSuspendGenerator() {
 }
 
 void BytecodeGraphBuilder::VisitResumeGenerator() {
-  PrepareEagerCheckpoint();
-
   Node* generator = environment()->LookupRegister(
       bytecode_iterator().GetRegisterOperand(0));
 
@@ -2310,7 +2308,7 @@ void BytecodeGraphBuilder::VisitResumeGenerator() {
   Node* state =
       NewNode(javascript()->GeneratorRestoreContinuation(), generator);
 
-  environment()->BindAccumulator(state, Environment::kAttachFrameState);
+  environment()->BindAccumulator(state);
 }
 
 void BytecodeGraphBuilder::VisitWide() {

src/compiler/linkage.cc

@@ -176,6 +176,7 @@ bool Linkage::NeedsFrameStateInput(Runtime::FunctionId function) {
     case Runtime::kInlineFixedArrayGet:
     case Runtime::kInlineFixedArraySet:
     case Runtime::kInlineGeneratorClose:
+    case Runtime::kInlineGeneratorGetContext:
     case Runtime::kInlineGeneratorGetInputOrDebugPos:
     case Runtime::kInlineGeneratorGetResumeMode:
     case Runtime::kInlineCreateJSGeneratorObject: