Generators: Avoid calling into runtime if operand stack is empty

This patch makes yield sites save the resume continuation and context
inline.  If the operand stack is empty, we can avoid a call into the
runtime.  This also makes the SuspendJSGeneratorObject runtime function
less magical: it just has to save the operand stack and stack handlers.

This speeds up the following case by a factor of 3 or so:

  function* until(n) {
    for (var i = 0; i < n; i++)
      yield i;
  }

  function sum(iter) {
    var sum = 0;
    for (var x of iter) sum += x;
    return sum;
  }

  for (var i = 0; i < 10000; i++) sum(until(1000))

Also, there is no more sentinel value as the generators will resume in
the right place already, allowing me to remove the hack added to the
--debug-code check in r14437.

R=mstarzinger@chromium.org
BUG=

Review URL: https://codereview.chromium.org/15990004

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@15240 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/arm/full-codegen-arm.cc

@@ -1994,14 +1994,29 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       __ push(result_register());
       // Fall through.
     case Yield::INITIAL: {
-      VisitForStackValue(expr->generator_object());
-      __ CallRuntime(Runtime::kSuspendJSGeneratorObject, 1);
-      __ ldr(context_register(),
-             MemOperand(fp, StandardFrameConstants::kContextOffset));
+      Label suspend, continuation, post_runtime, resume;
+
+      __ jmp(&suspend);
 
-      Label resume;
-      __ CompareRoot(result_register(), Heap::kTheHoleValueRootIndex);
-      __ b(ne, &resume);
+      __ bind(&continuation);
+      __ jmp(&resume);
+
+      __ bind(&suspend);
+      VisitForAccumulatorValue(expr->generator_object());
+      ASSERT(continuation.pos() > 0 && Smi::IsValid(continuation.pos()));
+      __ mov(r1, Operand(Smi::FromInt(continuation.pos())));
+      __ str(r1, FieldMemOperand(r0, JSGeneratorObject::kContinuationOffset));
+      __ str(cp, FieldMemOperand(r0, JSGeneratorObject::kContextOffset));
+      __ mov(r1, cp);
+      __ RecordWriteField(r0, JSGeneratorObject::kContextOffset, r1, r2,
+                          kLRHasBeenSaved, kDontSaveFPRegs);
+      __ add(r1, fp, Operand(StandardFrameConstants::kExpressionsOffset));
+      __ cmp(sp, r1);
+      __ b(eq, &post_runtime);
+      __ push(r0);  // generator object
+      __ CallRuntime(Runtime::kSuspendJSGeneratorObject, 1);
+      __ ldr(cp, MemOperand(fp, StandardFrameConstants::kContextOffset));
+      __ bind(&post_runtime);
       __ pop(result_register());
       EmitReturnSequence();
 
@@ -2029,7 +2044,8 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       // [sp + 1 * kPointerSize] iter
       // [sp + 0 * kPointerSize] g
 
-      Label l_catch, l_try, l_resume, l_next, l_call, l_loop;
+      Label l_catch, l_try, l_suspend, l_continuation, l_resume;
+      Label l_next, l_call, l_loop;
       // Initial send value is undefined.
       __ LoadRoot(r0, Heap::kUndefinedValueRootIndex);
       __ b(&l_next);
@@ -2051,13 +2067,22 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       __ PushTryHandler(StackHandler::CATCH, expr->index());
       const int handler_size = StackHandlerConstants::kSize;
       __ push(r0);                                       // result
-      __ ldr(r3, MemOperand(sp, (0 + 1) * kPointerSize + handler_size));  // g
-      __ push(r3);                                       // g
+      __ jmp(&l_suspend);
+      __ bind(&l_continuation);
+      __ jmp(&l_resume);
+      __ bind(&l_suspend);
+      const int generator_object_depth = kPointerSize + handler_size;
+      __ ldr(r0, MemOperand(sp, generator_object_depth));
+      __ push(r0);                                       // g
+      ASSERT(l_continuation.pos() > 0 && Smi::IsValid(l_continuation.pos()));
+      __ mov(r1, Operand(Smi::FromInt(l_continuation.pos())));
+      __ str(r1, FieldMemOperand(r0, JSGeneratorObject::kContinuationOffset));
+      __ str(cp, FieldMemOperand(r0, JSGeneratorObject::kContextOffset));
+      __ mov(r1, cp);
+      __ RecordWriteField(r0, JSGeneratorObject::kContextOffset, r1, r2,
+                          kLRHasBeenSaved, kDontSaveFPRegs);
       __ CallRuntime(Runtime::kSuspendJSGeneratorObject, 1);
-      __ ldr(context_register(),
-             MemOperand(fp, StandardFrameConstants::kContextOffset));
-      __ CompareRoot(r0, Heap::kTheHoleValueRootIndex);
-      __ b(ne, &l_resume);
+      __ ldr(cp, MemOperand(fp, StandardFrameConstants::kContextOffset));
       __ pop(r0);                                        // result
       EmitReturnSequence();
       __ bind(&l_resume);                                // received in r0

src/ia32/code-stubs-ia32.cc

@@ -5050,11 +5050,6 @@ void CEntryStub::GenerateCore(MacroAssembler* masm,
     Label okay;
     __ cmp(eax, masm->isolate()->factory()->the_hole_value());
     __ j(not_equal, &okay, Label::kNear);
-    // TODO(wingo): Currently SuspendJSGeneratorObject returns the hole.  Change
-    // to return another sentinel like a harmony symbol.
-    __ cmp(ebx, Immediate(ExternalReference(
-        Runtime::kSuspendJSGeneratorObject, masm->isolate())));
-    __ j(equal, &okay, Label::kNear);
     __ int3();
     __ bind(&okay);
   }

src/ia32/full-codegen-ia32.cc

@@ -1953,14 +1953,30 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       __ push(result_register());
       // Fall through.
     case Yield::INITIAL: {
-      VisitForStackValue(expr->generator_object());
+      Label suspend, continuation, post_runtime, resume;
+
+      __ jmp(&suspend);
+
+      __ bind(&continuation);
+      __ jmp(&resume);
+
+      __ bind(&suspend);
+      VisitForAccumulatorValue(expr->generator_object());
+      ASSERT(continuation.pos() > 0 && Smi::IsValid(continuation.pos()));
+      __ mov(FieldOperand(eax, JSGeneratorObject::kContinuationOffset),
+             Immediate(Smi::FromInt(continuation.pos())));
+      __ mov(FieldOperand(eax, JSGeneratorObject::kContextOffset), esi);
+      __ mov(ecx, esi);
+      __ RecordWriteField(eax, JSGeneratorObject::kContextOffset, ecx, edx,
+                          kDontSaveFPRegs);
+      __ lea(ebx, Operand(ebp, StandardFrameConstants::kExpressionsOffset));
+      __ cmp(esp, ebx);
+      __ j(equal, &post_runtime);
+      __ push(eax);  // generator object
       __ CallRuntime(Runtime::kSuspendJSGeneratorObject, 1);
       __ mov(context_register(),
              Operand(ebp, StandardFrameConstants::kContextOffset));
-
-      Label resume;
-      __ CompareRoot(result_register(), Heap::kTheHoleValueRootIndex);
-      __ j(not_equal, &resume);
+      __ bind(&post_runtime);
       __ pop(result_register());
       EmitReturnSequence();
 
@@ -1988,7 +2004,8 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       // [sp + 1 * kPointerSize] iter
       // [sp + 0 * kPointerSize] g
 
-      Label l_catch, l_try, l_resume, l_next, l_call, l_loop;
+      Label l_catch, l_try, l_suspend, l_continuation, l_resume;
+      Label l_next, l_call, l_loop;
       // Initial send value is undefined.
       __ mov(eax, isolate()->factory()->undefined_value());
       __ jmp(&l_next);
@@ -2010,12 +2027,23 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       __ PushTryHandler(StackHandler::CATCH, expr->index());
       const int handler_size = StackHandlerConstants::kSize;
       __ push(eax);                                      // result
-      __ push(Operand(esp, (0 + 1) * kPointerSize + handler_size));  // g
+      __ jmp(&l_suspend);
+      __ bind(&l_continuation);
+      __ jmp(&l_resume);
+      __ bind(&l_suspend);
+      const int generator_object_depth = kPointerSize + handler_size;
+      __ mov(eax, Operand(esp, generator_object_depth));
+      __ push(eax);                                      // g
+      ASSERT(l_continuation.pos() > 0 && Smi::IsValid(l_continuation.pos()));
+      __ mov(FieldOperand(eax, JSGeneratorObject::kContinuationOffset),
+             Immediate(Smi::FromInt(l_continuation.pos())));
+      __ mov(FieldOperand(eax, JSGeneratorObject::kContextOffset), esi);
+      __ mov(ecx, esi);
+      __ RecordWriteField(eax, JSGeneratorObject::kContextOffset, ecx, edx,
+                          kDontSaveFPRegs);
       __ CallRuntime(Runtime::kSuspendJSGeneratorObject, 1);
       __ mov(context_register(),
              Operand(ebp, StandardFrameConstants::kContextOffset));
-      __ CompareRoot(eax, Heap::kTheHoleValueRootIndex);
-      __ j(not_equal, &l_resume);
       __ pop(eax);                                       // result
       EmitReturnSequence();
       __ bind(&l_resume);                                // received in eax

src/runtime.cc

@@ -2583,18 +2583,23 @@ RUNTIME_FUNCTION(MaybeObject*, Runtime_SuspendJSGeneratorObject) {
 
   JavaScriptFrameIterator stack_iterator(isolate);
   JavaScriptFrame* frame = stack_iterator.frame();
-  JSFunction* function = JSFunction::cast(frame->function());
-  RUNTIME_ASSERT(function->shared()->is_generator());
-  ASSERT_EQ(function, generator_object->function());
+  RUNTIME_ASSERT(JSFunction::cast(frame->function())->shared()->is_generator());
+  ASSERT_EQ(JSFunction::cast(frame->function()), generator_object->function());
+
+  // The caller should have saved the context and continuation already.
+  ASSERT_EQ(generator_object->context(), Context::cast(frame->context()));
+  ASSERT_LT(0, generator_object->continuation());
 
   // We expect there to be at least two values on the operand stack: the return
   // value of the yield expression, and the argument to this runtime call.
   // Neither of those should be saved.
   int operands_count = frame->ComputeOperandsCount();
-  ASSERT(operands_count >= 2);
+  ASSERT_GE(operands_count, 2);
   operands_count -= 2;
 
   if (operands_count == 0) {
+    // Although it's semantically harmless to call this function with an
+    // operands_count of zero, it is also unnecessary.
     ASSERT_EQ(generator_object->operand_stack(),
               isolate->heap()->empty_fixed_array());
     ASSERT_EQ(generator_object->stack_handler_index(), -1);
@@ -2611,20 +2616,7 @@ RUNTIME_FUNCTION(MaybeObject*, Runtime_SuspendJSGeneratorObject) {
     generator_object->set_stack_handler_index(stack_handler_index);
   }
 
-  // Set continuation down here to avoid side effects if the operand stack
-  // allocation fails.
-  intptr_t offset = frame->pc() - function->code()->instruction_start();
-  ASSERT(offset > 0 && Smi::IsValid(offset));
-  generator_object->set_continuation(static_cast<int>(offset));
-
-  // It's possible for the context to be other than the initial context even if
-  // there is no stack handler active.  For example, this is the case in the
-  // body of a "with" statement.  Therefore we always save the context.
-  generator_object->set_context(Context::cast(frame->context()));
-
-  // The return value is the hole for a suspend return, and anything else for a
-  // resume return.
-  return isolate->heap()->the_hole_value();
+  return isolate->heap()->undefined_value();
 }
 
 

src/x64/full-codegen-x64.cc

@@ -1976,14 +1976,31 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       __ push(result_register());
       // Fall through.
     case Yield::INITIAL: {
-      VisitForStackValue(expr->generator_object());
+      Label suspend, continuation, post_runtime, resume;
+
+      __ jmp(&suspend);
+
+      __ bind(&continuation);
+      __ jmp(&resume);
+
+      __ bind(&suspend);
+      VisitForAccumulatorValue(expr->generator_object());
+      ASSERT(continuation.pos() > 0 && Smi::IsValid(continuation.pos()));
+      __ Move(FieldOperand(rax, JSGeneratorObject::kContinuationOffset),
+              Smi::FromInt(continuation.pos()));
+      __ movq(FieldOperand(rax, JSGeneratorObject::kContextOffset), rsi);
+      __ movq(rcx, rsi);
+      __ RecordWriteField(rax, JSGeneratorObject::kContextOffset, rcx, rdx,
+                          kDontSaveFPRegs);
+      __ lea(rbx, Operand(rbp, StandardFrameConstants::kExpressionsOffset));
+      __ cmpq(rsp, rbx);
+      __ j(equal, &post_runtime);
+      __ push(rax);  // generator object
       __ CallRuntime(Runtime::kSuspendJSGeneratorObject, 1);
       __ movq(context_register(),
               Operand(rbp, StandardFrameConstants::kContextOffset));
+      __ bind(&post_runtime);
 
-      Label resume;
-      __ CompareRoot(result_register(), Heap::kTheHoleValueRootIndex);
-      __ j(not_equal, &resume);
       __ pop(result_register());
       EmitReturnSequence();
 
@@ -2011,7 +2028,8 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       // [sp + 1 * kPointerSize] iter
       // [sp + 0 * kPointerSize] g
 
-      Label l_catch, l_try, l_resume, l_next, l_call, l_loop;
+      Label l_catch, l_try, l_suspend, l_continuation, l_resume;
+      Label l_next, l_call, l_loop;
       // Initial send value is undefined.
       __ LoadRoot(rax, Heap::kUndefinedValueRootIndex);
       __ jmp(&l_next);
@@ -2033,12 +2051,23 @@ void FullCodeGenerator::VisitYield(Yield* expr) {
       __ PushTryHandler(StackHandler::CATCH, expr->index());
       const int handler_size = StackHandlerConstants::kSize;
       __ push(rax);                                      // result
-      __ push(Operand(rsp, (0 + 1) * kPointerSize + handler_size));  // g
+      __ jmp(&l_suspend);
+      __ bind(&l_continuation);
+      __ jmp(&l_resume);
+      __ bind(&l_suspend);
+      const int generator_object_depth = kPointerSize + handler_size;
+      __ movq(rax, Operand(rsp, generator_object_depth));
+      __ push(rax);                                      // g
+      ASSERT(l_continuation.pos() > 0 && Smi::IsValid(l_continuation.pos()));
+      __ Move(FieldOperand(rax, JSGeneratorObject::kContinuationOffset),
+              Smi::FromInt(l_continuation.pos()));
+      __ movq(FieldOperand(rax, JSGeneratorObject::kContextOffset), rsi);
+      __ movq(rcx, rsi);
+      __ RecordWriteField(rax, JSGeneratorObject::kContextOffset, rcx, rdx,
+                          kDontSaveFPRegs);
       __ CallRuntime(Runtime::kSuspendJSGeneratorObject, 1);
       __ movq(context_register(),
               Operand(rbp, StandardFrameConstants::kContextOffset));
-      __ CompareRoot(rax, Heap::kTheHoleValueRootIndex);
-      __ j(not_equal, &l_resume);
       __ pop(rax);                                       // result
       EmitReturnSequence();
       __ bind(&l_resume);                                // received in rax