// Copyright 2012 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.

#if V8_TARGET_ARCH_ARM

#include "src/api/api-arguments.h"
#include "src/codegen/code-factory.h"
#include "src/debug/debug.h"
#include "src/deoptimizer/deoptimizer.h"
#include "src/execution/frame-constants.h"
#include "src/execution/frames.h"
#include "src/logging/counters.h"
// For interpreter_entry_return_pc_offset. TODO(jkummerow): Drop.
#include "src/codegen/macro-assembler-inl.h"
#include "src/codegen/register-configuration.h"
#include "src/heap/heap-inl.h"
#include "src/objects/cell.h"
#include "src/objects/foreign.h"
#include "src/objects/heap-number.h"
#include "src/objects/js-generator.h"
#include "src/objects/objects-inl.h"
#include "src/objects/smi.h"
#include "src/runtime/runtime.h"
#include "src/wasm/wasm-objects.h"

namespace v8 {
namespace internal {

#define __ ACCESS_MASM(masm)

void Builtins::Generate_Adaptor(MacroAssembler* masm, Address address) {
#if defined(__thumb__)
  // Thumb mode builtin.
  DCHECK_EQ(1, reinterpret_cast<uintptr_t>(
                   ExternalReference::Create(address).address()) &
                   1);
#endif
  __ Move(kJavaScriptCallExtraArg1Register, ExternalReference::Create(address));
  __ Jump(BUILTIN_CODE(masm->isolate(), AdaptorWithBuiltinExitFrame),
          RelocInfo::CODE_TARGET);
}

void Builtins::Generate_InternalArrayConstructor(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0     : number of arguments
  //  -- lr     : return address
  //  -- sp[...]: constructor arguments
  // -----------------------------------

  if (FLAG_debug_code) {
    // Initial map for the builtin InternalArray functions should be maps.
    __ ldr(r2, FieldMemOperand(r1, JSFunction::kPrototypeOrInitialMapOffset));
    __ SmiTst(r2);
    __ Assert(ne, AbortReason::kUnexpectedInitialMapForInternalArrayFunction);
    __ CompareObjectType(r2, r3, r4, MAP_TYPE);
    __ Assert(eq, AbortReason::kUnexpectedInitialMapForInternalArrayFunction);
  }

  // Run the native code for the InternalArray function called as a normal
  // function.
  __ Jump(BUILTIN_CODE(masm->isolate(), InternalArrayConstructorImpl),
          RelocInfo::CODE_TARGET);
}

static void GenerateTailCallToReturnedCode(MacroAssembler* masm,
                                           Runtime::FunctionId function_id) {
  // ----------- S t a t e -------------
  //  -- r1 : target function (preserved for callee)
  //  -- r3 : new target (preserved for callee)
  // -----------------------------------
  {
    FrameAndConstantPoolScope scope(masm, StackFrame::INTERNAL);
    // Push a copy of the target function and the new target.
    __ push(r1);
    __ push(r3);
    // Push function as parameter to the runtime call.
    __ Push(r1);

    __ CallRuntime(function_id, 1);
    __ mov(r2, r0);

    // Restore target function and new target.
    __ pop(r3);
    __ pop(r1);
  }
  static_assert(kJavaScriptCallCodeStartRegister == r2, "ABI mismatch");
  __ JumpCodeObject(r2);
}

namespace {

void Generate_StackOverflowCheck(MacroAssembler* masm, Register num_args,
                                 Register scratch, Label* stack_overflow) {
  // Check the stack for overflow. We are not trying to catch
  // interruptions (e.g. debug break and preemption) here, so the "real stack
  // limit" is checked.
  __ LoadRoot(scratch, RootIndex::kRealStackLimit);
  // Make scratch the space we have left. The stack might already be overflowed
  // here which will cause scratch to become negative.
  __ sub(scratch, sp, scratch);
  // Check if the arguments will overflow the stack.
  __ cmp(scratch, Operand(num_args, LSL, kPointerSizeLog2));
  __ b(le, stack_overflow);  // Signed comparison.
}

void Generate_JSBuiltinsConstructStubHelper(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0     : number of arguments
  //  -- r1     : constructor function
  //  -- r3     : new target
  //  -- cp     : context
  //  -- lr     : return address
  //  -- sp[...]: constructor arguments
  // -----------------------------------

  Register scratch = r2;

  Label stack_overflow;

  Generate_StackOverflowCheck(masm, r0, scratch, &stack_overflow);

  // Enter a construct frame.
  {
    FrameAndConstantPoolScope scope(masm, StackFrame::CONSTRUCT);

    // Preserve the incoming parameters on the stack.
    __ SmiTag(r0);
    __ Push(cp, r0);
    __ SmiUntag(r0);

    // The receiver for the builtin/api call.
    __ PushRoot(RootIndex::kTheHoleValue);

    // Set up pointer to last argument.
    __ add(r4, fp, Operand(StandardFrameConstants::kCallerSPOffset));

    // Copy arguments and receiver to the expression stack.
    Label loop, entry;
    __ mov(r5, r0);
    // ----------- S t a t e -------------
    //  --                 r0: number of arguments (untagged)
    //  --                 r1: constructor function
    //  --                 r3: new target
    //  --                 r4: pointer to last argument
    //  --                 r5: counter
    //  -- sp[0*kPointerSize]: the hole (receiver)
    //  -- sp[1*kPointerSize]: number of arguments (tagged)
    //  -- sp[2*kPointerSize]: context
    // -----------------------------------
    __ b(&entry);
    __ bind(&loop);
    __ ldr(scratch, MemOperand(r4, r5, LSL, kPointerSizeLog2));
    __ push(scratch);
    __ bind(&entry);
    __ sub(r5, r5, Operand(1), SetCC);
    __ b(ge, &loop);

    // Call the function.
    // r0: number of arguments (untagged)
    // r1: constructor function
    // r3: new target
    ParameterCount actual(r0);
    __ InvokeFunction(r1, r3, actual, CALL_FUNCTION);

    // Restore context from the frame.
    __ ldr(cp, MemOperand(fp, ConstructFrameConstants::kContextOffset));
    // Restore smi-tagged arguments count from the frame.
    __ ldr(scratch, MemOperand(fp, ConstructFrameConstants::kLengthOffset));
    // Leave construct frame.
  }

  // Remove caller arguments from the stack and return.
  STATIC_ASSERT(kSmiTagSize == 1 && kSmiTag == 0);
  __ add(sp, sp, Operand(scratch, LSL, kPointerSizeLog2 - kSmiTagSize));
  __ add(sp, sp, Operand(kPointerSize));
  __ Jump(lr);

  __ bind(&stack_overflow);
  {
    FrameScope scope(masm, StackFrame::INTERNAL);
    __ CallRuntime(Runtime::kThrowStackOverflow);
    __ bkpt(0);  // Unreachable code.
  }
}

}  // namespace

// The construct stub for ES5 constructor functions and ES6 class constructors.
void Builtins::Generate_JSConstructStubGeneric(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  --      r0: number of arguments (untagged)
  //  --      r1: constructor function
  //  --      r3: new target
  //  --      cp: context
  //  --      lr: return address
  //  -- sp[...]: constructor arguments
  // -----------------------------------

  // Enter a construct frame.
  {
    FrameAndConstantPoolScope scope(masm, StackFrame::CONSTRUCT);
    Label post_instantiation_deopt_entry, not_create_implicit_receiver;

    // Preserve the incoming parameters on the stack.
    __ LoadRoot(r4, RootIndex::kTheHoleValue);
    __ SmiTag(r0);
    __ Push(cp, r0, r1, r4, r3);

    // ----------- S t a t e -------------
    //  --        sp[0*kPointerSize]: new target
    //  --        sp[1*kPointerSize]: padding
    //  -- r1 and sp[2*kPointerSize]: constructor function
    //  --        sp[3*kPointerSize]: number of arguments (tagged)
    //  --        sp[4*kPointerSize]: context
    // -----------------------------------

    __ ldr(r4, FieldMemOperand(r1, JSFunction::kSharedFunctionInfoOffset));
    __ ldr(r4, FieldMemOperand(r4, SharedFunctionInfo::kFlagsOffset));
    __ DecodeField<SharedFunctionInfo::FunctionKindBits>(r4);
    __ JumpIfIsInRange(r4, kDefaultDerivedConstructor, kDerivedConstructor,
                       &not_create_implicit_receiver);

    // If not derived class constructor: Allocate the new receiver object.
    __ IncrementCounter(masm->isolate()->counters()->constructed_objects(), 1,
                        r4, r5);
    __ Call(BUILTIN_CODE(masm->isolate(), FastNewObject),
            RelocInfo::CODE_TARGET);
    __ b(&post_instantiation_deopt_entry);

    // Else: use TheHoleValue as receiver for constructor call
    __ bind(&not_create_implicit_receiver);
    __ LoadRoot(r0, RootIndex::kTheHoleValue);

    // ----------- S t a t e -------------
    //  --                          r0: receiver
    //  -- Slot 3 / sp[0*kPointerSize]: new target
    //  -- Slot 2 / sp[1*kPointerSize]: constructor function
    //  -- Slot 1 / sp[2*kPointerSize]: number of arguments (tagged)
    //  -- Slot 0 / sp[3*kPointerSize]: context
    // -----------------------------------
    // Deoptimizer enters here.
    masm->isolate()->heap()->SetConstructStubCreateDeoptPCOffset(
        masm->pc_offset());
    __ bind(&post_instantiation_deopt_entry);

    // Restore new target.
    __ Pop(r3);
    // Push the allocated receiver to the stack. We need two copies
    // because we may have to return the original one and the calling
    // conventions dictate that the called function pops the receiver.
    __ Push(r0, r0);

    // ----------- S t a t e -------------
    //  --                 r3: new target
    //  -- sp[0*kPointerSize]: implicit receiver
    //  -- sp[1*kPointerSize]: implicit receiver
    //  -- sp[2*kPointerSize]: padding
    //  -- sp[3*kPointerSize]: constructor function
    //  -- sp[4*kPointerSize]: number of arguments (tagged)
    //  -- sp[5*kPointerSize]: context
    // -----------------------------------

    // Restore constructor function and argument count.
    __ ldr(r1, MemOperand(fp, ConstructFrameConstants::kConstructorOffset));
    __ ldr(r0, MemOperand(fp, ConstructFrameConstants::kLengthOffset));
    __ SmiUntag(r0);

    // Set up pointer to last argument.
    __ add(r4, fp, Operand(StandardFrameConstants::kCallerSPOffset));

    Label enough_stack_space, stack_overflow;
    Generate_StackOverflowCheck(masm, r0, r5, &stack_overflow);
    __ b(&enough_stack_space);

    __ bind(&stack_overflow);
    // Restore the context from the frame.
    __ ldr(cp, MemOperand(fp, ConstructFrameConstants::kContextOffset));
    __ CallRuntime(Runtime::kThrowStackOverflow);
    // Unreachable code.
    __ bkpt(0);

    __ bind(&enough_stack_space);

    // Copy arguments and receiver to the expression stack.
    Label loop, entry;
    __ mov(r5, r0);
    // ----------- S t a t e -------------
    //  --                        r0: number of arguments (untagged)
    //  --                        r3: new target
    //  --                        r4: pointer to last argument
    //  --                        r5: counter
    //  --        sp[0*kPointerSize]: implicit receiver
    //  --        sp[1*kPointerSize]: implicit receiver
    //  --        sp[2*kPointerSize]: padding
    //  -- r1 and sp[3*kPointerSize]: constructor function
    //  --        sp[4*kPointerSize]: number of arguments (tagged)
    //  --        sp[5*kPointerSize]: context
    // -----------------------------------
    __ b(&entry);

    __ bind(&loop);
    __ ldr(r6, MemOperand(r4, r5, LSL, kPointerSizeLog2));
    __ push(r6);
    __ bind(&entry);
    __ sub(r5, r5, Operand(1), SetCC);
    __ b(ge, &loop);

    // Call the function.
    ParameterCount actual(r0);
    __ InvokeFunction(r1, r3, actual, CALL_FUNCTION);

    // ----------- S t a t e -------------
    //  --                 r0: constructor result
    //  -- sp[0*kPointerSize]: implicit receiver
    //  -- sp[1*kPointerSize]: padding
    //  -- sp[2*kPointerSize]: constructor function
    //  -- sp[3*kPointerSize]: number of arguments
    //  -- sp[4*kPointerSize]: context
    // -----------------------------------

    // Store offset of return address for deoptimizer.
    masm->isolate()->heap()->SetConstructStubInvokeDeoptPCOffset(
        masm->pc_offset());

    // Restore the context from the frame.
    __ ldr(cp, MemOperand(fp, ConstructFrameConstants::kContextOffset));

    // If the result is an object (in the ECMA sense), we should get rid
    // of the receiver and use the result; see ECMA-262 section 13.2.2-7
    // on page 74.
    Label use_receiver, do_throw, leave_frame;

    // If the result is undefined, we jump out to using the implicit receiver.
    __ JumpIfRoot(r0, RootIndex::kUndefinedValue, &use_receiver);

    // Otherwise we do a smi check and fall through to check if the return value
    // is a valid receiver.

    // If the result is a smi, it is *not* an object in the ECMA sense.
    __ JumpIfSmi(r0, &use_receiver);

    // If the type of the result (stored in its map) is less than
    // FIRST_JS_RECEIVER_TYPE, it is not an object in the ECMA sense.
    STATIC_ASSERT(LAST_JS_RECEIVER_TYPE == LAST_TYPE);
    __ CompareObjectType(r0, r4, r5, FIRST_JS_RECEIVER_TYPE);
    __ b(ge, &leave_frame);
    __ b(&use_receiver);

    __ bind(&do_throw);
    __ CallRuntime(Runtime::kThrowConstructorReturnedNonObject);

    // Throw away the result of the constructor invocation and use the
    // on-stack receiver as the result.
    __ bind(&use_receiver);
    __ ldr(r0, MemOperand(sp, 0 * kPointerSize));
    __ JumpIfRoot(r0, RootIndex::kTheHoleValue, &do_throw);

    __ bind(&leave_frame);
    // Restore smi-tagged arguments count from the frame.
    __ ldr(r1, MemOperand(fp, ConstructFrameConstants::kLengthOffset));
    // Leave construct frame.
  }
  // Remove caller arguments from the stack and return.
  STATIC_ASSERT(kSmiTagSize == 1 && kSmiTag == 0);
  __ add(sp, sp, Operand(r1, LSL, kPointerSizeLog2 - kSmiTagSize));
  __ add(sp, sp, Operand(kPointerSize));
  __ Jump(lr);
}

void Builtins::Generate_JSBuiltinsConstructStub(MacroAssembler* masm) {
  Generate_JSBuiltinsConstructStubHelper(masm);
}

static void GetSharedFunctionInfoBytecode(MacroAssembler* masm,
                                          Register sfi_data,
                                          Register scratch1) {
  Label done;

  __ CompareObjectType(sfi_data, scratch1, scratch1, INTERPRETER_DATA_TYPE);
  __ b(ne, &done);
  __ ldr(sfi_data,
         FieldMemOperand(sfi_data, InterpreterData::kBytecodeArrayOffset));

  __ bind(&done);
}

// static
void Builtins::Generate_ResumeGeneratorTrampoline(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : the value to pass to the generator
  //  -- r1 : the JSGeneratorObject to resume
  //  -- lr : return address
  // -----------------------------------
  __ AssertGeneratorObject(r1);

  // Store input value into generator object.
  __ str(r0, FieldMemOperand(r1, JSGeneratorObject::kInputOrDebugPosOffset));
  __ RecordWriteField(r1, JSGeneratorObject::kInputOrDebugPosOffset, r0,
                      kLRHasNotBeenSaved, kDontSaveFPRegs);

  // Load suspended function and context.
  __ ldr(r4, FieldMemOperand(r1, JSGeneratorObject::kFunctionOffset));
  __ ldr(cp, FieldMemOperand(r4, JSFunction::kContextOffset));

  Label prepare_step_in_if_stepping, prepare_step_in_suspended_generator;
  Label stepping_prepared;
  Register scratch = r5;

  // Flood function if we are stepping.
  ExternalReference debug_hook =
      ExternalReference::debug_hook_on_function_call_address(masm->isolate());
  __ Move(scratch, debug_hook);
  __ ldrsb(scratch, MemOperand(scratch));
  __ cmp(scratch, Operand(0));
  __ b(ne, &prepare_step_in_if_stepping);

  // Flood function if we need to continue stepping in the suspended
  // generator.
  ExternalReference debug_suspended_generator =
      ExternalReference::debug_suspended_generator_address(masm->isolate());
  __ Move(scratch, debug_suspended_generator);
  __ ldr(scratch, MemOperand(scratch));
  __ cmp(scratch, Operand(r1));
  __ b(eq, &prepare_step_in_suspended_generator);
  __ bind(&stepping_prepared);

  // Check the stack for overflow. We are not trying to catch interruptions
  // (i.e. debug break and preemption) here, so check the "real stack limit".
  Label stack_overflow;
  __ CompareRoot(sp, RootIndex::kRealStackLimit);
  __ b(lo, &stack_overflow);

  // Push receiver.
  __ ldr(scratch, FieldMemOperand(r1, JSGeneratorObject::kReceiverOffset));
  __ Push(scratch);

  // ----------- S t a t e -------------
  //  -- r1    : the JSGeneratorObject to resume
  //  -- r4    : generator function
  //  -- cp    : generator context
  //  -- lr    : return address
  //  -- sp[0] : generator receiver
  // -----------------------------------

  // Copy the function arguments from the generator object's register file.
  __ ldr(r3, FieldMemOperand(r4, JSFunction::kSharedFunctionInfoOffset));
  __ ldrh(r3,
          FieldMemOperand(r3, SharedFunctionInfo::kFormalParameterCountOffset));
  __ ldr(r2,
         FieldMemOperand(r1, JSGeneratorObject::kParametersAndRegistersOffset));
  {
    Label done_loop, loop;
    __ mov(r6, Operand(0));

    __ bind(&loop);
    __ cmp(r6, r3);
    __ b(ge, &done_loop);
    __ add(scratch, r2, Operand(r6, LSL, kPointerSizeLog2));
    __ ldr(scratch, FieldMemOperand(scratch, FixedArray::kHeaderSize));
    __ Push(scratch);
    __ add(r6, r6, Operand(1));
    __ b(&loop);

    __ bind(&done_loop);
  }

  // Underlying function needs to have bytecode available.
  if (FLAG_debug_code) {
    __ ldr(r3, FieldMemOperand(r4, JSFunction::kSharedFunctionInfoOffset));
    __ ldr(r3, FieldMemOperand(r3, SharedFunctionInfo::kFunctionDataOffset));
    GetSharedFunctionInfoBytecode(masm, r3, r0);
    __ CompareObjectType(r3, r3, r3, BYTECODE_ARRAY_TYPE);
    __ Assert(eq, AbortReason::kMissingBytecodeArray);
  }

  // Resume (Ignition/TurboFan) generator object.
  {
    __ ldr(r0, FieldMemOperand(r4, JSFunction::kSharedFunctionInfoOffset));
    __ ldrh(r0, FieldMemOperand(
                    r0, SharedFunctionInfo::kFormalParameterCountOffset));
    // We abuse new.target both to indicate that this is a resume call and to
    // pass in the generator object.  In ordinary calls, new.target is always
    // undefined because generator functions are non-constructable.
    __ Move(r3, r1);
    __ Move(r1, r4);
    static_assert(kJavaScriptCallCodeStartRegister == r2, "ABI mismatch");
    __ ldr(r2, FieldMemOperand(r1, JSFunction::kCodeOffset));
    __ JumpCodeObject(r2);
  }

  __ bind(&prepare_step_in_if_stepping);
  {
    FrameAndConstantPoolScope scope(masm, StackFrame::INTERNAL);
    __ Push(r1, r4);
    // Push hole as receiver since we do not use it for stepping.
    __ PushRoot(RootIndex::kTheHoleValue);
    __ CallRuntime(Runtime::kDebugOnFunctionCall);
    __ Pop(r1);
    __ ldr(r4, FieldMemOperand(r1, JSGeneratorObject::kFunctionOffset));
  }
  __ b(&stepping_prepared);

  __ bind(&prepare_step_in_suspended_generator);
  {
    FrameAndConstantPoolScope scope(masm, StackFrame::INTERNAL);
    __ Push(r1);
    __ CallRuntime(Runtime::kDebugPrepareStepInSuspendedGenerator);
    __ Pop(r1);
    __ ldr(r4, FieldMemOperand(r1, JSGeneratorObject::kFunctionOffset));
  }
  __ b(&stepping_prepared);

  __ bind(&stack_overflow);
  {
    FrameScope scope(masm, StackFrame::INTERNAL);
    __ CallRuntime(Runtime::kThrowStackOverflow);
    __ bkpt(0);  // This should be unreachable.
  }
}

void Builtins::Generate_ConstructedNonConstructable(MacroAssembler* masm) {
  FrameScope scope(masm, StackFrame::INTERNAL);
  __ push(r1);
  __ CallRuntime(Runtime::kThrowConstructedNonConstructable);
}

namespace {

// Total size of the stack space pushed by JSEntryVariant.
// JSEntryTrampoline uses this to access on stack arguments passed to
// JSEntryVariant.
constexpr int kPushedStackSpace = kNumCalleeSaved * kPointerSize +
                                  kPointerSize /* LR */ +
                                  kNumDoubleCalleeSaved * kDoubleSize +
                                  4 * kPointerSize /* r5, r6, r7, scratch */ +
                                  EntryFrameConstants::kCallerFPOffset;

// Called with the native C calling convention. The corresponding function
// signature is either:
//
//   using JSEntryFunction = GeneratedCode<Address(
//       Address root_register_value, Address new_target, Address target,
//       Address receiver, intptr_t argc, Address** argv)>;
// or
//   using JSEntryFunction = GeneratedCode<Address(
//       Address root_register_value, MicrotaskQueue* microtask_queue)>;
void Generate_JSEntryVariant(MacroAssembler* masm, StackFrame::Type type,
                             Builtins::Name entry_trampoline) {
  // The register state is either:
  //   r0:                            root_register_value
  //   r1:                            code entry
  //   r2:                            function
  //   r3:                            receiver
  //   [sp + 0 * kSystemPointerSize]: argc
  //   [sp + 1 * kSystemPointerSize]: argv
  // or
  //   r0: root_register_value
  //   r1: microtask_queue
  // Preserve all but r0 and pass them to entry_trampoline.
  Label invoke, handler_entry, exit;

  // Update |pushed_stack_space| when we manipulate the stack.
  int pushed_stack_space = EntryFrameConstants::kCallerFPOffset;
  {
    NoRootArrayScope no_root_array(masm);

    // Called from C, so do not pop argc and args on exit (preserve sp)
    // No need to save register-passed args
    // Save callee-saved registers (incl. cp and fp), sp, and lr
    __ stm(db_w, sp, kCalleeSaved | lr.bit());
    pushed_stack_space +=
        kNumCalleeSaved * kPointerSize + kPointerSize /* LR */;

    // Save callee-saved vfp registers.
    __ vstm(db_w, sp, kFirstCalleeSavedDoubleReg, kLastCalleeSavedDoubleReg);
    pushed_stack_space += kNumDoubleCalleeSaved * kDoubleSize;

    // Set up the reserved register for 0.0.
    __ vmov(kDoubleRegZero, Double(0.0));

    // Initialize the root register.
    // C calling convention. The first argument is passed in r0.
    __ mov(kRootRegister, r0);
  }

  // Push a frame with special values setup to mark it as an entry frame.
  // r0: root_register_value
  __ mov(r7, Operand(StackFrame::TypeToMarker(type)));
  __ mov(r6, Operand(StackFrame::TypeToMarker(type)));
  __ Move(r5, ExternalReference::Create(IsolateAddressId::kCEntryFPAddress,
                                        masm->isolate()));
  __ ldr(r5, MemOperand(r5));
  {
    UseScratchRegisterScope temps(masm);
    Register scratch = temps.Acquire();

    // Push a bad frame pointer to fail if it is used.
    __ mov(scratch, Operand(-1));
    __ stm(db_w, sp, r5.bit() | r6.bit() | r7.bit() | scratch.bit());
    pushed_stack_space += 4 * kPointerSize /* r5, r6, r7, scratch */;
  }

  Register scratch = r6;

  // Set up frame pointer for the frame to be pushed.
  __ add(fp, sp, Operand(-EntryFrameConstants::kCallerFPOffset));

  // If this is the outermost JS call, set js_entry_sp value.
  Label non_outermost_js;
  ExternalReference js_entry_sp = ExternalReference::Create(
      IsolateAddressId::kJSEntrySPAddress, masm->isolate());
  __ Move(r5, js_entry_sp);
  __ ldr(scratch, MemOperand(r5));
  __ cmp(scratch, Operand::Zero());
  __ b(ne, &non_outermost_js);
  __ str(fp, MemOperand(r5));
  __ mov(scratch, Operand(StackFrame::OUTERMOST_JSENTRY_FRAME));
  Label cont;
  __ b(&cont);
  __ bind(&non_outermost_js);
  __ mov(scratch, Operand(StackFrame::INNER_JSENTRY_FRAME));
  __ bind(&cont);
  __ push(scratch);

  // Jump to a faked try block that does the invoke, with a faked catch
  // block that sets the pending exception.
  __ jmp(&invoke);

  // Block literal pool emission whilst taking the position of the handler
  // entry. This avoids making the assumption that literal pools are always
  // emitted after an instruction is emitted, rather than before.
  {
    Assembler::BlockConstPoolScope block_const_pool(masm);
    __ bind(&handler_entry);

    // Store the current pc as the handler offset. It's used later to create the
    // handler table.
    masm->isolate()->builtins()->SetJSEntryHandlerOffset(handler_entry.pos());

    // Caught exception: Store result (exception) in the pending exception
    // field in the JSEnv and return a failure sentinel.  Coming in here the
    // fp will be invalid because the PushStackHandler below sets it to 0 to
    // signal the existence of the JSEntry frame.
    __ Move(scratch,
            ExternalReference::Create(
                IsolateAddressId::kPendingExceptionAddress, masm->isolate()));
  }
  __ str(r0, MemOperand(scratch));
  __ LoadRoot(r0, RootIndex::kException);
  __ b(&exit);

  // Invoke: Link this frame into the handler chain.
  __ bind(&invoke);
  // Must preserve r0-r4, r5-r6 are available.
  __ PushStackHandler();
  // If an exception not caught by another handler occurs, this handler
  // returns control to the code after the bl(&invoke) above, which
  // restores all kCalleeSaved registers (including cp and fp) to their
  // saved values before returning a failure to C.
  //
  // Invoke the function by calling through JS entry trampoline builtin and
  // pop the faked function when we return.
  Handle<Code> trampoline_code =
      masm->isolate()->builtins()->builtin_handle(entry_trampoline);
  DCHECK_EQ(kPushedStackSpace, pushed_stack_space);
  __ Call(trampoline_code, RelocInfo::CODE_TARGET);

  // Unlink this frame from the handler chain.
  __ PopStackHandler();

  __ bind(&exit);  // r0 holds result
  // Check if the current stack frame is marked as the outermost JS frame.
  Label non_outermost_js_2;
  __ pop(r5);
  __ cmp(r5, Operand(StackFrame::OUTERMOST_JSENTRY_FRAME));
  __ b(ne, &non_outermost_js_2);
  __ mov(r6, Operand::Zero());
  __ Move(r5, js_entry_sp);
  __ str(r6, MemOperand(r5));
  __ bind(&non_outermost_js_2);

  // Restore the top frame descriptors from the stack.
  __ pop(r3);
  __ Move(scratch, ExternalReference::Create(IsolateAddressId::kCEntryFPAddress,
                                             masm->isolate()));
  __ str(r3, MemOperand(scratch));

  // Reset the stack to the callee saved registers.
  __ add(sp, sp, Operand(-EntryFrameConstants::kCallerFPOffset));

  // Restore callee-saved registers and return.
#ifdef DEBUG
  if (FLAG_debug_code) {
    __ mov(lr, Operand(pc));
  }
#endif

  // Restore callee-saved vfp registers.
  __ vldm(ia_w, sp, kFirstCalleeSavedDoubleReg, kLastCalleeSavedDoubleReg);

  __ ldm(ia_w, sp, kCalleeSaved | pc.bit());
}

}  // namespace

void Builtins::Generate_JSEntry(MacroAssembler* masm) {
  Generate_JSEntryVariant(masm, StackFrame::ENTRY,
                          Builtins::kJSEntryTrampoline);
}

void Builtins::Generate_JSConstructEntry(MacroAssembler* masm) {
  Generate_JSEntryVariant(masm, StackFrame::CONSTRUCT_ENTRY,
                          Builtins::kJSConstructEntryTrampoline);
}

void Builtins::Generate_JSRunMicrotasksEntry(MacroAssembler* masm) {
  Generate_JSEntryVariant(masm, StackFrame::ENTRY,
                          Builtins::kRunMicrotasksTrampoline);
}

static void Generate_JSEntryTrampolineHelper(MacroAssembler* masm,
                                             bool is_construct) {
  // Called from Generate_JS_Entry
  // r0:                                                root_register_value
  // r1:                                                new.target
  // r2:                                                function
  // r3:                                                receiver
  // [fp + kPushedStackSpace + 0 * kSystemPointerSize]: argc
  // [fp + kPushedStackSpace + 1 * kSystemPointerSize]: argv
  // r5-r6, r8 and cp may be clobbered

  __ ldr(r0,
         MemOperand(fp, kPushedStackSpace + EntryFrameConstants::kArgcOffset));
  __ ldr(r4,
         MemOperand(fp, kPushedStackSpace + EntryFrameConstants::kArgvOffset));

  // r1: new.target
  // r2: function
  // r3: receiver
  // r0: argc
  // r4: argv

  // Enter an internal frame.
  {
    FrameScope scope(masm, StackFrame::INTERNAL);

    // Setup the context (we need to use the caller context from the isolate).
    ExternalReference context_address = ExternalReference::Create(
        IsolateAddressId::kContextAddress, masm->isolate());
    __ Move(cp, context_address);
    __ ldr(cp, MemOperand(cp));

    // Push the function and the receiver onto the stack.
    __ Push(r2, r3);

    // Check if we have enough stack space to push all arguments.
    // Clobbers r3.
    Label enough_stack_space, stack_overflow;
    Generate_StackOverflowCheck(masm, r0, r3, &stack_overflow);
    __ b(&enough_stack_space);
    __ bind(&stack_overflow);
    __ CallRuntime(Runtime::kThrowStackOverflow);
    // Unreachable code.
    __ bkpt(0);

    __ bind(&enough_stack_space);

    // Copy arguments to the stack in a loop.
    // r1: new.target
    // r2: function
    // r0: argc
    // r4: argv, i.e. points to first arg
    Label loop, entry;
    __ add(r3, r4, Operand(r0, LSL, kPointerSizeLog2));
    // r1 points past last arg.
    __ b(&entry);
    __ bind(&loop);
    __ ldr(r5, MemOperand(r4, kPointerSize, PostIndex));  // read next parameter
    __ ldr(r5, MemOperand(r5));                           // dereference handle
    __ push(r5);                                          // push parameter
    __ bind(&entry);
    __ cmp(r4, r3);
    __ b(ne, &loop);

    // Setup new.target and function.
    __ mov(r3, r1);
    __ mov(r1, r2);
    // r0: argc
    // r1: function
    // r3: new.target

    // Initialize all JavaScript callee-saved registers, since they will be seen
    // by the garbage collector as part of handlers.
    __ LoadRoot(r4, RootIndex::kUndefinedValue);
    __ mov(r2, r4);
    __ mov(r5, r4);
    __ mov(r6, r4);
    __ mov(r8, r4);
    if (kR9Available == 1) {
      __ mov(r9, r4);
    }

    // Invoke the code.
    Handle<Code> builtin = is_construct
                               ? BUILTIN_CODE(masm->isolate(), Construct)
                               : masm->isolate()->builtins()->Call();
    __ Call(builtin, RelocInfo::CODE_TARGET);

    // Exit the JS frame and remove the parameters (except function), and
    // return.
    // Respect ABI stack constraint.
  }
  __ Jump(lr);

  // r0: result
}

void Builtins::Generate_JSEntryTrampoline(MacroAssembler* masm) {
  Generate_JSEntryTrampolineHelper(masm, false);
}

void Builtins::Generate_JSConstructEntryTrampoline(MacroAssembler* masm) {
  Generate_JSEntryTrampolineHelper(masm, true);
}

void Builtins::Generate_RunMicrotasksTrampoline(MacroAssembler* masm) {
  // This expects two C++ function parameters passed by Invoke() in
  // execution.cc.
  //   r0: root_register_value
  //   r1: microtask_queue

  __ mov(RunMicrotasksDescriptor::MicrotaskQueueRegister(), r1);
  __ Jump(BUILTIN_CODE(masm->isolate(), RunMicrotasks), RelocInfo::CODE_TARGET);
}

static void ReplaceClosureCodeWithOptimizedCode(MacroAssembler* masm,
                                                Register optimized_code,
                                                Register closure) {
  // Store code entry in the closure.
  __ str(optimized_code, FieldMemOperand(closure, JSFunction::kCodeOffset));
  __ RecordWriteField(closure, JSFunction::kCodeOffset, optimized_code,
                      kLRHasNotBeenSaved, kDontSaveFPRegs, OMIT_REMEMBERED_SET,
                      OMIT_SMI_CHECK);
}

static void LeaveInterpreterFrame(MacroAssembler* masm, Register scratch) {
  Register args_count = scratch;

  // Get the arguments + receiver count.
  __ ldr(args_count,
         MemOperand(fp, InterpreterFrameConstants::kBytecodeArrayFromFp));
  __ ldr(args_count,
         FieldMemOperand(args_count, BytecodeArray::kParameterSizeOffset));

  // Leave the frame (also dropping the register file).
  __ LeaveFrame(StackFrame::INTERPRETED);

  // Drop receiver + arguments.
  __ add(sp, sp, args_count, LeaveCC);
}

// Tail-call |function_id| if |smi_entry| == |marker|
static void TailCallRuntimeIfMarkerEquals(MacroAssembler* masm,
                                          Register smi_entry,
                                          OptimizationMarker marker,
                                          Runtime::FunctionId function_id) {
  Label no_match;
  __ cmp(smi_entry, Operand(Smi::FromEnum(marker)));
  __ b(ne, &no_match);
  GenerateTailCallToReturnedCode(masm, function_id);
  __ bind(&no_match);
}

static void MaybeTailCallOptimizedCodeSlot(MacroAssembler* masm,
                                           Register feedback_vector,
                                           Register scratch1,
                                           Register scratch2) {
  // ----------- S t a t e -------------
  //  -- r3 : new target (preserved for callee if needed, and caller)
  //  -- r1 : target function (preserved for callee if needed, and caller)
  //  -- feedback vector (preserved for caller if needed)
  // -----------------------------------
  DCHECK(!AreAliased(feedback_vector, r1, r3, scratch1, scratch2));

  Label optimized_code_slot_is_weak_ref, fallthrough;

  Register closure = r1;
  Register optimized_code_entry = scratch1;

  __ ldr(
      optimized_code_entry,
      FieldMemOperand(feedback_vector,
                      FeedbackVector::kOptimizedCodeWeakOrSmiOffset));

  // Check if the code entry is a Smi. If yes, we interpret it as an
  // optimisation marker. Otherwise, interpret it as a weak reference to a code
  // object.
  __ JumpIfNotSmi(optimized_code_entry, &optimized_code_slot_is_weak_ref);

  {
    // Optimized code slot is a Smi optimization marker.

    // Fall through if no optimization trigger.
    __ cmp(optimized_code_entry,
           Operand(Smi::FromEnum(OptimizationMarker::kNone)));
    __ b(eq, &fallthrough);

    // TODO(v8:8394): The logging of first execution will break if
    // feedback vectors are not allocated. We need to find a different way of
    // logging these events if required.
    TailCallRuntimeIfMarkerEquals(masm, optimized_code_entry,
                                  OptimizationMarker::kLogFirstExecution,
                                  Runtime::kFunctionFirstExecution);
    TailCallRuntimeIfMarkerEquals(masm, optimized_code_entry,
                                  OptimizationMarker::kCompileOptimized,
                                  Runtime::kCompileOptimized_NotConcurrent);
    TailCallRuntimeIfMarkerEquals(
        masm, optimized_code_entry,
        OptimizationMarker::kCompileOptimizedConcurrent,
        Runtime::kCompileOptimized_Concurrent);

    {
      // Otherwise, the marker is InOptimizationQueue, so fall through hoping
      // that an interrupt will eventually update the slot with optimized code.
      if (FLAG_debug_code) {
        __ cmp(
            optimized_code_entry,
            Operand(Smi::FromEnum(OptimizationMarker::kInOptimizationQueue)));
        __ Assert(eq, AbortReason::kExpectedOptimizationSentinel);
      }
      __ jmp(&fallthrough);
    }
  }

  {
    // Optimized code slot is a weak reference.
    __ bind(&optimized_code_slot_is_weak_ref);

    __ LoadWeakValue(optimized_code_entry, optimized_code_entry, &fallthrough);

    // Check if the optimized code is marked for deopt. If it is, call the
    // runtime to clear it.
    Label found_deoptimized_code;
    __ ldr(scratch2, FieldMemOperand(optimized_code_entry,
                                     Code::kCodeDataContainerOffset));
    __ ldr(
        scratch2,
        FieldMemOperand(scratch2, CodeDataContainer::kKindSpecificFlagsOffset));
    __ tst(scratch2, Operand(1 << Code::kMarkedForDeoptimizationBit));
    __ b(ne, &found_deoptimized_code);

    // Optimized code is good, get it into the closure and link the closure into
    // the optimized functions list, then tail call the optimized code.
    // The feedback vector is no longer used, so re-use it as a scratch
    // register.
    ReplaceClosureCodeWithOptimizedCode(masm, optimized_code_entry, closure);
    static_assert(kJavaScriptCallCodeStartRegister == r2, "ABI mismatch");
    __ LoadCodeObjectEntry(r2, optimized_code_entry);
    __ Jump(r2);

    // Optimized code slot contains deoptimized code, evict it and re-enter the
    // closure's code.
    __ bind(&found_deoptimized_code);
    GenerateTailCallToReturnedCode(masm, Runtime::kEvictOptimizedCodeSlot);
  }

  // Fall-through if the optimized code cell is clear and there is no
  // optimization marker.
  __ bind(&fallthrough);
}

// Advance the current bytecode offset. This simulates what all bytecode
// handlers do upon completion of the underlying operation. Will bail out to a
// label if the bytecode (without prefix) is a return bytecode.
static void AdvanceBytecodeOffsetOrReturn(MacroAssembler* masm,
                                          Register bytecode_array,
                                          Register bytecode_offset,
                                          Register bytecode, Register scratch1,
                                          Label* if_return) {
  Register bytecode_size_table = scratch1;
  DCHECK(!AreAliased(bytecode_array, bytecode_offset, bytecode_size_table,
                     bytecode));

  __ Move(bytecode_size_table,
          ExternalReference::bytecode_size_table_address());

  // Check if the bytecode is a Wide or ExtraWide prefix bytecode.
  Label process_bytecode, extra_wide;
  STATIC_ASSERT(0 == static_cast<int>(interpreter::Bytecode::kWide));
  STATIC_ASSERT(1 == static_cast<int>(interpreter::Bytecode::kExtraWide));
  STATIC_ASSERT(2 == static_cast<int>(interpreter::Bytecode::kDebugBreakWide));
  STATIC_ASSERT(3 ==
                static_cast<int>(interpreter::Bytecode::kDebugBreakExtraWide));
  __ cmp(bytecode, Operand(0x3));
  __ b(hi, &process_bytecode);
  __ tst(bytecode, Operand(0x1));
  __ b(ne, &extra_wide);

  // Load the next bytecode and update table to the wide scaled table.
  __ add(bytecode_offset, bytecode_offset, Operand(1));
  __ ldrb(bytecode, MemOperand(bytecode_array, bytecode_offset));
  __ add(bytecode_size_table, bytecode_size_table,
         Operand(kIntSize * interpreter::Bytecodes::kBytecodeCount));
  __ jmp(&process_bytecode);

  __ bind(&extra_wide);
  // Load the next bytecode and update table to the extra wide scaled table.
  __ add(bytecode_offset, bytecode_offset, Operand(1));
  __ ldrb(bytecode, MemOperand(bytecode_array, bytecode_offset));
  __ add(bytecode_size_table, bytecode_size_table,
         Operand(2 * kIntSize * interpreter::Bytecodes::kBytecodeCount));

  __ bind(&process_bytecode);

// Bailout to the return label if this is a return bytecode.
#define JUMP_IF_EQUAL(NAME)                                                    \
  __ cmp(bytecode, Operand(static_cast<int>(interpreter::Bytecode::k##NAME))); \
  __ b(if_return, eq);
  RETURN_BYTECODE_LIST(JUMP_IF_EQUAL)
#undef JUMP_IF_EQUAL

  // Otherwise, load the size of the current bytecode and advance the offset.
  __ ldr(scratch1, MemOperand(bytecode_size_table, bytecode, LSL, 2));
  __ add(bytecode_offset, bytecode_offset, scratch1);
}

// Generate code for entering a JS function with the interpreter.
// On entry to the function the receiver and arguments have been pushed on the
// stack left to right.  The actual argument count matches the formal parameter
// count expected by the function.
//
// The live registers are:
//   o r1: the JS function object being called.
//   o r3: the incoming new target or generator object
//   o cp: our context
//   o fp: the caller's frame pointer
//   o sp: stack pointer
//   o lr: return address
//
// The function builds an interpreter frame.  See InterpreterFrameConstants in
// frames.h for its layout.
void Builtins::Generate_InterpreterEntryTrampoline(MacroAssembler* masm) {
  Register closure = r1;
  Register feedback_vector = r2;

  // Get the bytecode array from the function object and load it into
  // kInterpreterBytecodeArrayRegister.
  __ ldr(r0, FieldMemOperand(closure, JSFunction::kSharedFunctionInfoOffset));
  __ ldr(kInterpreterBytecodeArrayRegister,
         FieldMemOperand(r0, SharedFunctionInfo::kFunctionDataOffset));
  GetSharedFunctionInfoBytecode(masm, kInterpreterBytecodeArrayRegister, r4);

  // The bytecode array could have been flushed from the shared function info,
  // if so, call into CompileLazy.
  Label compile_lazy;
  __ CompareObjectType(kInterpreterBytecodeArrayRegister, r0, no_reg,
                       BYTECODE_ARRAY_TYPE);
  __ b(ne, &compile_lazy);

  // Load the feedback vector from the closure.
  __ ldr(feedback_vector,
         FieldMemOperand(closure, JSFunction::kFeedbackCellOffset));
  __ ldr(feedback_vector, FieldMemOperand(feedback_vector, Cell::kValueOffset));

  Label push_stack_frame;
  // Check if feedback vector is valid. If valid, check for optimized code
  // and update invocation count. Otherwise, setup the stack frame.
  __ ldr(r4, FieldMemOperand(feedback_vector, HeapObject::kMapOffset));
  __ ldrh(r4, FieldMemOperand(r4, Map::kInstanceTypeOffset));
  __ cmp(r4, Operand(FEEDBACK_VECTOR_TYPE));
  __ b(ne, &push_stack_frame);

  // Read off the optimized code slot in the feedback vector, and if there
  // is optimized code or an optimization marker, call that instead.
  MaybeTailCallOptimizedCodeSlot(masm, feedback_vector, r4, r6);

  // Increment invocation count for the function.
  __ ldr(r9, FieldMemOperand(feedback_vector,
                             FeedbackVector::kInvocationCountOffset));
  __ add(r9, r9, Operand(1));
  __ str(r9, FieldMemOperand(feedback_vector,
                             FeedbackVector::kInvocationCountOffset));

  // Open a frame scope to indicate that there is a frame on the stack.  The
  // MANUAL indicates that the scope shouldn't actually generate code to set up
  // the frame (that is done below).
  __ bind(&push_stack_frame);
  FrameScope frame_scope(masm, StackFrame::MANUAL);
  __ PushStandardFrame(closure);

  // Reset code age and the OSR arming. The OSR field and BytecodeAgeOffset are
  // 8-bit fields next to each other, so we could just optimize by writing a
  // 16-bit. These static asserts guard our assumption is valid.
  STATIC_ASSERT(BytecodeArray::kBytecodeAgeOffset ==
                BytecodeArray::kOsrNestingLevelOffset + kCharSize);
  STATIC_ASSERT(BytecodeArray::kNoAgeBytecodeAge == 0);
  __ mov(r9, Operand(0));
  __ strh(r9, FieldMemOperand(kInterpreterBytecodeArrayRegister,
                              BytecodeArray::kOsrNestingLevelOffset));

  // Load the initial bytecode offset.
  __ mov(kInterpreterBytecodeOffsetRegister,
         Operand(BytecodeArray::kHeaderSize - kHeapObjectTag));

  // Push bytecode array and Smi tagged bytecode array offset.
  __ SmiTag(r0, kInterpreterBytecodeOffsetRegister);
  __ Push(kInterpreterBytecodeArrayRegister, r0);

  // Allocate the local and temporary register file on the stack.
  {
    // Load frame size from the BytecodeArray object.
    __ ldr(r4, FieldMemOperand(kInterpreterBytecodeArrayRegister,
                               BytecodeArray::kFrameSizeOffset));

    // Do a stack check to ensure we don't go over the limit.
    Label ok;
    __ sub(r9, sp, Operand(r4));
    __ LoadRoot(r2, RootIndex::kRealStackLimit);
    __ cmp(r9, Operand(r2));
    __ b(hs, &ok);
    __ CallRuntime(Runtime::kThrowStackOverflow);
    __ bind(&ok);

    // If ok, push undefined as the initial value for all register file entries.
    Label loop_header;
    Label loop_check;
    __ LoadRoot(r9, RootIndex::kUndefinedValue);
    __ b(&loop_check, al);
    __ bind(&loop_header);
    // TODO(rmcilroy): Consider doing more than one push per loop iteration.
    __ push(r9);
    // Continue loop if not done.
    __ bind(&loop_check);
    __ sub(r4, r4, Operand(kPointerSize), SetCC);
    __ b(&loop_header, ge);
  }

  // If the bytecode array has a valid incoming new target or generator object
  // register, initialize it with incoming value which was passed in r3.
  __ ldr(r9, FieldMemOperand(
                 kInterpreterBytecodeArrayRegister,
                 BytecodeArray::kIncomingNewTargetOrGeneratorRegisterOffset));
  __ cmp(r9, Operand::Zero());
  __ str(r3, MemOperand(fp, r9, LSL, kPointerSizeLog2), ne);

  // Load accumulator with undefined.
  __ LoadRoot(kInterpreterAccumulatorRegister, RootIndex::kUndefinedValue);

  // Load the dispatch table into a register and dispatch to the bytecode
  // handler at the current bytecode offset.
  Label do_dispatch;
  __ bind(&do_dispatch);
  __ Move(
      kInterpreterDispatchTableRegister,
      ExternalReference::interpreter_dispatch_table_address(masm->isolate()));
  __ ldrb(r4, MemOperand(kInterpreterBytecodeArrayRegister,
                         kInterpreterBytecodeOffsetRegister));
  __ ldr(
      kJavaScriptCallCodeStartRegister,
      MemOperand(kInterpreterDispatchTableRegister, r4, LSL, kPointerSizeLog2));
  __ Call(kJavaScriptCallCodeStartRegister);
  masm->isolate()->heap()->SetInterpreterEntryReturnPCOffset(masm->pc_offset());

  // Any returns to the entry trampoline are either due to the return bytecode
  // or the interpreter tail calling a builtin and then a dispatch.

  // Get bytecode array and bytecode offset from the stack frame.
  __ ldr(kInterpreterBytecodeArrayRegister,
         MemOperand(fp, InterpreterFrameConstants::kBytecodeArrayFromFp));
  __ ldr(kInterpreterBytecodeOffsetRegister,
         MemOperand(fp, InterpreterFrameConstants::kBytecodeOffsetFromFp));
  __ SmiUntag(kInterpreterBytecodeOffsetRegister);

  // Either return, or advance to the next bytecode and dispatch.
  Label do_return;
  __ ldrb(r1, MemOperand(kInterpreterBytecodeArrayRegister,
                         kInterpreterBytecodeOffsetRegister));
  AdvanceBytecodeOffsetOrReturn(masm, kInterpreterBytecodeArrayRegister,
                                kInterpreterBytecodeOffsetRegister, r1, r2,
                                &do_return);
  __ jmp(&do_dispatch);

  __ bind(&do_return);
  // The return value is in r0.
  LeaveInterpreterFrame(masm, r2);
  __ Jump(lr);

  __ bind(&compile_lazy);
  GenerateTailCallToReturnedCode(masm, Runtime::kCompileLazy);
  __ bkpt(0);  // Should not return.
}

static void Generate_InterpreterPushArgs(MacroAssembler* masm,
                                         Register num_args, Register index,
                                         Register limit, Register scratch) {
  // Find the address of the last argument.
  __ mov(limit, num_args);
  __ mov(limit, Operand(limit, LSL, kPointerSizeLog2));
  __ sub(limit, index, limit);

  Label loop_header, loop_check;
  __ b(al, &loop_check);
  __ bind(&loop_header);
  __ ldr(scratch, MemOperand(index, -kPointerSize, PostIndex));
  __ push(scratch);
  __ bind(&loop_check);
  __ cmp(index, limit);
  __ b(gt, &loop_header);
}

// static
void Builtins::Generate_InterpreterPushArgsThenCallImpl(
    MacroAssembler* masm, ConvertReceiverMode receiver_mode,
    InterpreterPushArgsMode mode) {
  DCHECK(mode != InterpreterPushArgsMode::kArrayFunction);
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r2 : the address of the first argument to be pushed. Subsequent
  //          arguments should be consecutive above this, in the same order as
  //          they are to be pushed onto the stack.
  //  -- r1 : the target to call (can be any Object).
  // -----------------------------------
  Label stack_overflow;

  __ add(r3, r0, Operand(1));  // Add one for receiver.

  Generate_StackOverflowCheck(masm, r3, r4, &stack_overflow);

  // Push "undefined" as the receiver arg if we need to.
  if (receiver_mode == ConvertReceiverMode::kNullOrUndefined) {
    __ PushRoot(RootIndex::kUndefinedValue);
    __ mov(r3, r0);  // Argument count is correct.
  }

  // Push the arguments. r2, r4, r5 will be modified.
  Generate_InterpreterPushArgs(masm, r3, r2, r4, r5);

  if (mode == InterpreterPushArgsMode::kWithFinalSpread) {
    __ Pop(r2);                  // Pass the spread in a register
    __ sub(r0, r0, Operand(1));  // Subtract one for spread
  }

  // Call the target.
  if (mode == InterpreterPushArgsMode::kWithFinalSpread) {
    __ Jump(BUILTIN_CODE(masm->isolate(), CallWithSpread),
            RelocInfo::CODE_TARGET);
  } else {
    __ Jump(masm->isolate()->builtins()->Call(ConvertReceiverMode::kAny),
            RelocInfo::CODE_TARGET);
  }

  __ bind(&stack_overflow);
  {
    __ TailCallRuntime(Runtime::kThrowStackOverflow);
    // Unreachable code.
    __ bkpt(0);
  }
}

// static
void Builtins::Generate_InterpreterPushArgsThenConstructImpl(
    MacroAssembler* masm, InterpreterPushArgsMode mode) {
  // ----------- S t a t e -------------
  // -- r0 : argument count (not including receiver)
  // -- r3 : new target
  // -- r1 : constructor to call
  // -- r2 : allocation site feedback if available, undefined otherwise.
  // -- r4 : address of the first argument
  // -----------------------------------
  Label stack_overflow;

  // Push a slot for the receiver to be constructed.
  __ mov(r5, Operand::Zero());
  __ push(r5);

  Generate_StackOverflowCheck(masm, r0, r5, &stack_overflow);

  // Push the arguments. r5, r4, r6 will be modified.
  Generate_InterpreterPushArgs(masm, r0, r4, r5, r6);

  if (mode == InterpreterPushArgsMode::kWithFinalSpread) {
    __ Pop(r2);                  // Pass the spread in a register
    __ sub(r0, r0, Operand(1));  // Subtract one for spread
  } else {
    __ AssertUndefinedOrAllocationSite(r2, r5);
  }

  if (mode == InterpreterPushArgsMode::kArrayFunction) {
    __ AssertFunction(r1);

    // Tail call to the array construct stub (still in the caller
    // context at this point).
    Handle<Code> code = BUILTIN_CODE(masm->isolate(), ArrayConstructorImpl);
    __ Jump(code, RelocInfo::CODE_TARGET);
  } else if (mode == InterpreterPushArgsMode::kWithFinalSpread) {
    // Call the constructor with r0, r1, and r3 unmodified.
    __ Jump(BUILTIN_CODE(masm->isolate(), ConstructWithSpread),
            RelocInfo::CODE_TARGET);
  } else {
    DCHECK_EQ(InterpreterPushArgsMode::kOther, mode);
    // Call the constructor with r0, r1, and r3 unmodified.
    __ Jump(BUILTIN_CODE(masm->isolate(), Construct), RelocInfo::CODE_TARGET);
  }

  __ bind(&stack_overflow);
  {
    __ TailCallRuntime(Runtime::kThrowStackOverflow);
    // Unreachable code.
    __ bkpt(0);
  }
}

static void Generate_InterpreterEnterBytecode(MacroAssembler* masm) {
  // Set the return address to the correct point in the interpreter entry
  // trampoline.
  Label builtin_trampoline, trampoline_loaded;
  Smi interpreter_entry_return_pc_offset(
      masm->isolate()->heap()->interpreter_entry_return_pc_offset());
  DCHECK_NE(interpreter_entry_return_pc_offset, Smi::kZero);

  // If the SFI function_data is an InterpreterData, the function will have a
  // custom copy of the interpreter entry trampoline for profiling. If so,
  // get the custom trampoline, otherwise grab the entry address of the global
  // trampoline.
  __ ldr(r2, MemOperand(fp, StandardFrameConstants::kFunctionOffset));
  __ ldr(r2, FieldMemOperand(r2, JSFunction::kSharedFunctionInfoOffset));
  __ ldr(r2, FieldMemOperand(r2, SharedFunctionInfo::kFunctionDataOffset));
  __ CompareObjectType(r2, kInterpreterDispatchTableRegister,
                       kInterpreterDispatchTableRegister,
                       INTERPRETER_DATA_TYPE);
  __ b(ne, &builtin_trampoline);

  __ ldr(r2,
         FieldMemOperand(r2, InterpreterData::kInterpreterTrampolineOffset));
  __ add(r2, r2, Operand(Code::kHeaderSize - kHeapObjectTag));
  __ b(&trampoline_loaded);

  __ bind(&builtin_trampoline);
  __ Move(r2, ExternalReference::
                  address_of_interpreter_entry_trampoline_instruction_start(
                      masm->isolate()));
  __ ldr(r2, MemOperand(r2));

  __ bind(&trampoline_loaded);
  __ add(lr, r2, Operand(interpreter_entry_return_pc_offset.value()));

  // Initialize the dispatch table register.
  __ Move(
      kInterpreterDispatchTableRegister,
      ExternalReference::interpreter_dispatch_table_address(masm->isolate()));

  // Get the bytecode array pointer from the frame.
  __ ldr(kInterpreterBytecodeArrayRegister,
         MemOperand(fp, InterpreterFrameConstants::kBytecodeArrayFromFp));

  if (FLAG_debug_code) {
    // Check function data field is actually a BytecodeArray object.
    __ SmiTst(kInterpreterBytecodeArrayRegister);
    __ Assert(
        ne, AbortReason::kFunctionDataShouldBeBytecodeArrayOnInterpreterEntry);
    __ CompareObjectType(kInterpreterBytecodeArrayRegister, r1, no_reg,
                         BYTECODE_ARRAY_TYPE);
    __ Assert(
        eq, AbortReason::kFunctionDataShouldBeBytecodeArrayOnInterpreterEntry);
  }

  // Get the target bytecode offset from the frame.
  __ ldr(kInterpreterBytecodeOffsetRegister,
         MemOperand(fp, InterpreterFrameConstants::kBytecodeOffsetFromFp));
  __ SmiUntag(kInterpreterBytecodeOffsetRegister);

  // Dispatch to the target bytecode.
  UseScratchRegisterScope temps(masm);
  Register scratch = temps.Acquire();
  __ ldrb(scratch, MemOperand(kInterpreterBytecodeArrayRegister,
                              kInterpreterBytecodeOffsetRegister));
  __ ldr(kJavaScriptCallCodeStartRegister,
         MemOperand(kInterpreterDispatchTableRegister, scratch, LSL,
                    kPointerSizeLog2));
  __ Jump(kJavaScriptCallCodeStartRegister);
}

void Builtins::Generate_InterpreterEnterBytecodeAdvance(MacroAssembler* masm) {
  // Get bytecode array and bytecode offset from the stack frame.
  __ ldr(kInterpreterBytecodeArrayRegister,
         MemOperand(fp, InterpreterFrameConstants::kBytecodeArrayFromFp));
  __ ldr(kInterpreterBytecodeOffsetRegister,
         MemOperand(fp, InterpreterFrameConstants::kBytecodeOffsetFromFp));
  __ SmiUntag(kInterpreterBytecodeOffsetRegister);

  // Load the current bytecode.
  __ ldrb(r1, MemOperand(kInterpreterBytecodeArrayRegister,
                         kInterpreterBytecodeOffsetRegister));

  // Advance to the next bytecode.
  Label if_return;
  AdvanceBytecodeOffsetOrReturn(masm, kInterpreterBytecodeArrayRegister,
                                kInterpreterBytecodeOffsetRegister, r1, r2,
                                &if_return);

  // Convert new bytecode offset to a Smi and save in the stackframe.
  __ SmiTag(r2, kInterpreterBytecodeOffsetRegister);
  __ str(r2, MemOperand(fp, InterpreterFrameConstants::kBytecodeOffsetFromFp));

  Generate_InterpreterEnterBytecode(masm);

  // We should never take the if_return path.
  __ bind(&if_return);
  __ Abort(AbortReason::kInvalidBytecodeAdvance);
}

void Builtins::Generate_InterpreterEnterBytecodeDispatch(MacroAssembler* masm) {
  Generate_InterpreterEnterBytecode(masm);
}

void Builtins::Generate_InstantiateAsmJs(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : argument count (preserved for callee)
  //  -- r1 : new target (preserved for callee)
  //  -- r3 : target function (preserved for callee)
  // -----------------------------------
  Label failed;
  {
    FrameScope scope(masm, StackFrame::INTERNAL);
    // Preserve argument count for later compare.
    __ Move(r4, r0);
    // Push the number of arguments to the callee.
    __ SmiTag(r0);
    __ push(r0);
    // Push a copy of the target function and the new target.
    __ push(r1);
    __ push(r3);

    // The function.
    __ push(r1);
    // Copy arguments from caller (stdlib, foreign, heap).
    Label args_done;
    for (int j = 0; j < 4; ++j) {
      Label over;
      if (j < 3) {
        __ cmp(r4, Operand(j));
        __ b(ne, &over);
      }
      for (int i = j - 1; i >= 0; --i) {
        __ ldr(r4, MemOperand(fp, StandardFrameConstants::kCallerSPOffset +
                                      i * kPointerSize));
        __ push(r4);
      }
      for (int i = 0; i < 3 - j; ++i) {
        __ PushRoot(RootIndex::kUndefinedValue);
      }
      if (j < 3) {
        __ jmp(&args_done);
        __ bind(&over);
      }
    }
    __ bind(&args_done);

    // Call runtime, on success unwind frame, and parent frame.
    __ CallRuntime(Runtime::kInstantiateAsmJs, 4);
    // A smi 0 is returned on failure, an object on success.
    __ JumpIfSmi(r0, &failed);

    __ Drop(2);
    __ pop(r4);
    __ SmiUntag(r4);
    scope.GenerateLeaveFrame();

    __ add(r4, r4, Operand(1));
    __ Drop(r4);
    __ Ret();

    __ bind(&failed);
    // Restore target function and new target.
    __ pop(r3);
    __ pop(r1);
    __ pop(r0);
    __ SmiUntag(r0);
  }
  // On failure, tail call back to regular js by re-calling the function
  // which has be reset to the compile lazy builtin.
  static_assert(kJavaScriptCallCodeStartRegister == r2, "ABI mismatch");
  __ ldr(r2, FieldMemOperand(r1, JSFunction::kCodeOffset));
  __ JumpCodeObject(r2);
}

namespace {
void Generate_ContinueToBuiltinHelper(MacroAssembler* masm,
                                      bool java_script_builtin,
                                      bool with_result) {
  const RegisterConfiguration* config(RegisterConfiguration::Default());
  int allocatable_register_count = config->num_allocatable_general_registers();
  if (with_result) {
    // Overwrite the hole inserted by the deoptimizer with the return value from
    // the LAZY deopt point.
    __ str(r0,
           MemOperand(
               sp, config->num_allocatable_general_registers() * kPointerSize +
                       BuiltinContinuationFrameConstants::kFixedFrameSize));
  }
  for (int i = allocatable_register_count - 1; i >= 0; --i) {
    int code = config->GetAllocatableGeneralCode(i);
    __ Pop(Register::from_code(code));
    if (java_script_builtin && code == kJavaScriptCallArgCountRegister.code()) {
      __ SmiUntag(Register::from_code(code));
    }
  }
  __ ldr(fp, MemOperand(
                 sp, BuiltinContinuationFrameConstants::kFixedFrameSizeFromFp));

  // Load builtin index (stored as a Smi) and use it to get the builtin start
  // address from the builtins table.
  UseScratchRegisterScope temps(masm);
  Register builtin = temps.Acquire();
  __ Pop(builtin);
  __ add(sp, sp,
         Operand(BuiltinContinuationFrameConstants::kFixedFrameSizeFromFp));
  __ Pop(lr);
  __ LoadEntryFromBuiltinIndex(builtin);
  __ bx(builtin);
}
}  // namespace

void Builtins::Generate_ContinueToCodeStubBuiltin(MacroAssembler* masm) {
  Generate_ContinueToBuiltinHelper(masm, false, false);
}

void Builtins::Generate_ContinueToCodeStubBuiltinWithResult(
    MacroAssembler* masm) {
  Generate_ContinueToBuiltinHelper(masm, false, true);
}

void Builtins::Generate_ContinueToJavaScriptBuiltin(MacroAssembler* masm) {
  Generate_ContinueToBuiltinHelper(masm, true, false);
}

void Builtins::Generate_ContinueToJavaScriptBuiltinWithResult(
    MacroAssembler* masm) {
  Generate_ContinueToBuiltinHelper(masm, true, true);
}

void Builtins::Generate_NotifyDeoptimized(MacroAssembler* masm) {
  {
    FrameAndConstantPoolScope scope(masm, StackFrame::INTERNAL);
    __ CallRuntime(Runtime::kNotifyDeoptimized);
  }

  DCHECK_EQ(kInterpreterAccumulatorRegister.code(), r0.code());
  __ pop(r0);
  __ Ret();
}

void Builtins::Generate_InterpreterOnStackReplacement(MacroAssembler* masm) {
  // Lookup the function in the JavaScript frame.
  __ ldr(r0, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
  __ ldr(r0, MemOperand(r0, JavaScriptFrameConstants::kFunctionOffset));

  {
    FrameAndConstantPoolScope scope(masm, StackFrame::INTERNAL);
    // Pass function as argument.
    __ push(r0);
    __ CallRuntime(Runtime::kCompileForOnStackReplacement);
  }

  // If the code object is null, just return to the caller.
  Label skip;
  __ cmp(r0, Operand(Smi::zero()));
  __ b(ne, &skip);
  __ Ret();

  __ bind(&skip);

  // Drop the handler frame that is be sitting on top of the actual
  // JavaScript frame. This is the case then OSR is triggered from bytecode.
  __ LeaveFrame(StackFrame::STUB);

  // Load deoptimization data from the code object.
  // <deopt_data> = <code>[#deoptimization_data_offset]
  __ ldr(r1, FieldMemOperand(r0, Code::kDeoptimizationDataOffset));

  {
    ConstantPoolUnavailableScope constant_pool_unavailable(masm);
    __ add(r0, r0, Operand(Code::kHeaderSize - kHeapObjectTag));  // Code start

    // Load the OSR entrypoint offset from the deoptimization data.
    // <osr_offset> = <deopt_data>[#header_size + #osr_pc_offset]
    __ ldr(r1, FieldMemOperand(r1, FixedArray::OffsetOfElementAt(
                                       DeoptimizationData::kOsrPcOffsetIndex)));

    // Compute the target address = code start + osr_offset
    __ add(lr, r0, Operand::SmiUntag(r1));

    // And "return" to the OSR entry point of the function.
    __ Ret();
  }
}

// static
void Builtins::Generate_FunctionPrototypeApply(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0    : argc
  //  -- sp[0] : argArray
  //  -- sp[4] : thisArg
  //  -- sp[8] : receiver
  // -----------------------------------

  // 1. Load receiver into r1, argArray into r2 (if present), remove all
  // arguments from the stack (including the receiver), and push thisArg (if
  // present) instead.
  {
    __ LoadRoot(r5, RootIndex::kUndefinedValue);
    __ mov(r2, r5);
    __ ldr(r1, MemOperand(sp, r0, LSL, kPointerSizeLog2));  // receiver
    __ sub(r4, r0, Operand(1), SetCC);
    __ ldr(r5, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // thisArg
    __ sub(r4, r4, Operand(1), SetCC, ge);
    __ ldr(r2, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // argArray
    __ add(sp, sp, Operand(r0, LSL, kPointerSizeLog2));
    __ str(r5, MemOperand(sp, 0));
  }

  // ----------- S t a t e -------------
  //  -- r2    : argArray
  //  -- r1    : receiver
  //  -- sp[0] : thisArg
  // -----------------------------------

  // 2. We don't need to check explicitly for callable receiver here,
  // since that's the first thing the Call/CallWithArrayLike builtins
  // will do.

  // 3. Tail call with no arguments if argArray is null or undefined.
  Label no_arguments;
  __ JumpIfRoot(r2, RootIndex::kNullValue, &no_arguments);
  __ JumpIfRoot(r2, RootIndex::kUndefinedValue, &no_arguments);

  // 4a. Apply the receiver to the given argArray.
  __ Jump(BUILTIN_CODE(masm->isolate(), CallWithArrayLike),
          RelocInfo::CODE_TARGET);

  // 4b. The argArray is either null or undefined, so we tail call without any
  // arguments to the receiver.
  __ bind(&no_arguments);
  {
    __ mov(r0, Operand(0));
    __ Jump(masm->isolate()->builtins()->Call(), RelocInfo::CODE_TARGET);
  }
}

// static
void Builtins::Generate_FunctionPrototypeCall(MacroAssembler* masm) {
  // 1. Make sure we have at least one argument.
  // r0: actual number of arguments
  {
    Label done;
    __ cmp(r0, Operand::Zero());
    __ b(ne, &done);
    __ PushRoot(RootIndex::kUndefinedValue);
    __ add(r0, r0, Operand(1));
    __ bind(&done);
  }

  // 2. Get the callable to call (passed as receiver) from the stack.
  // r0: actual number of arguments
  __ ldr(r1, MemOperand(sp, r0, LSL, kPointerSizeLog2));

  // 3. Shift arguments and return address one slot down on the stack
  //    (overwriting the original receiver).  Adjust argument count to make
  //    the original first argument the new receiver.
  // r0: actual number of arguments
  // r1: callable
  {
    Register scratch = r3;
    Label loop;
    // Calculate the copy start address (destination). Copy end address is sp.
    __ add(r2, sp, Operand(r0, LSL, kPointerSizeLog2));

    __ bind(&loop);
    __ ldr(scratch, MemOperand(r2, -kPointerSize));
    __ str(scratch, MemOperand(r2));
    __ sub(r2, r2, Operand(kPointerSize));
    __ cmp(r2, sp);
    __ b(ne, &loop);
    // Adjust the actual number of arguments and remove the top element
    // (which is a copy of the last argument).
    __ sub(r0, r0, Operand(1));
    __ pop();
  }

  // 4. Call the callable.
  __ Jump(masm->isolate()->builtins()->Call(), RelocInfo::CODE_TARGET);
}

void Builtins::Generate_ReflectApply(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0     : argc
  //  -- sp[0]  : argumentsList
  //  -- sp[4]  : thisArgument
  //  -- sp[8]  : target
  //  -- sp[12] : receiver
  // -----------------------------------

  // 1. Load target into r1 (if present), argumentsList into r2 (if present),
  // remove all arguments from the stack (including the receiver), and push
  // thisArgument (if present) instead.
  {
    __ LoadRoot(r1, RootIndex::kUndefinedValue);
    __ mov(r5, r1);
    __ mov(r2, r1);
    __ sub(r4, r0, Operand(1), SetCC);
    __ ldr(r1, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // target
    __ sub(r4, r4, Operand(1), SetCC, ge);
    __ ldr(r5, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // thisArgument
    __ sub(r4, r4, Operand(1), SetCC, ge);
    __ ldr(r2, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // argumentsList
    __ add(sp, sp, Operand(r0, LSL, kPointerSizeLog2));
    __ str(r5, MemOperand(sp, 0));
  }

  // ----------- S t a t e -------------
  //  -- r2    : argumentsList
  //  -- r1    : target
  //  -- sp[0] : thisArgument
  // -----------------------------------

  // 2. We don't need to check explicitly for callable target here,
  // since that's the first thing the Call/CallWithArrayLike builtins
  // will do.

  // 3. Apply the target to the given argumentsList.
  __ Jump(BUILTIN_CODE(masm->isolate(), CallWithArrayLike),
          RelocInfo::CODE_TARGET);
}

void Builtins::Generate_ReflectConstruct(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0     : argc
  //  -- sp[0]  : new.target (optional)
  //  -- sp[4]  : argumentsList
  //  -- sp[8]  : target
  //  -- sp[12] : receiver
  // -----------------------------------

  // 1. Load target into r1 (if present), argumentsList into r2 (if present),
  // new.target into r3 (if present, otherwise use target), remove all
  // arguments from the stack (including the receiver), and push thisArgument
  // (if present) instead.
  {
    __ LoadRoot(r1, RootIndex::kUndefinedValue);
    __ mov(r2, r1);
    __ str(r2, MemOperand(sp, r0, LSL, kPointerSizeLog2));  // receiver
    __ sub(r4, r0, Operand(1), SetCC);
    __ ldr(r1, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // target
    __ mov(r3, r1);  // new.target defaults to target
    __ sub(r4, r4, Operand(1), SetCC, ge);
    __ ldr(r2, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // argumentsList
    __ sub(r4, r4, Operand(1), SetCC, ge);
    __ ldr(r3, MemOperand(sp, r4, LSL, kPointerSizeLog2), ge);  // new.target
    __ add(sp, sp, Operand(r0, LSL, kPointerSizeLog2));
  }

  // ----------- S t a t e -------------
  //  -- r2    : argumentsList
  //  -- r3    : new.target
  //  -- r1    : target
  //  -- sp[0] : receiver (undefined)
  // -----------------------------------

  // 2. We don't need to check explicitly for constructor target here,
  // since that's the first thing the Construct/ConstructWithArrayLike
  // builtins will do.

  // 3. We don't need to check explicitly for constructor new.target here,
  // since that's the second thing the Construct/ConstructWithArrayLike
  // builtins will do.

  // 4. Construct the target with the given new.target and argumentsList.
  __ Jump(BUILTIN_CODE(masm->isolate(), ConstructWithArrayLike),
          RelocInfo::CODE_TARGET);
}

static void EnterArgumentsAdaptorFrame(MacroAssembler* masm) {
  __ SmiTag(r0);
  __ mov(r4, Operand(StackFrame::TypeToMarker(StackFrame::ARGUMENTS_ADAPTOR)));
  __ stm(db_w, sp, r0.bit() | r1.bit() | r4.bit() |
                       fp.bit() | lr.bit());
  __ Push(Smi::zero());  // Padding.
  __ add(fp, sp,
         Operand(ArgumentsAdaptorFrameConstants::kFixedFrameSizeFromFp));
}

static void LeaveArgumentsAdaptorFrame(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : result being passed through
  // -----------------------------------
  // Get the number of arguments passed (as a smi), tear down the frame and
  // then tear down the parameters.
  __ ldr(r1, MemOperand(fp, ArgumentsAdaptorFrameConstants::kLengthOffset));

  __ LeaveFrame(StackFrame::ARGUMENTS_ADAPTOR);
  __ add(sp, sp, Operand::PointerOffsetFromSmiKey(r1));
  __ add(sp, sp, Operand(kPointerSize));  // adjust for receiver
}

// static
void Builtins::Generate_CallOrConstructVarargs(MacroAssembler* masm,
                                               Handle<Code> code) {
  // ----------- S t a t e -------------
  //  -- r1 : target
  //  -- r0 : number of parameters on the stack (not including the receiver)
  //  -- r2 : arguments list (a FixedArray)
  //  -- r4 : len (number of elements to push from args)
  //  -- r3 : new.target (for [[Construct]])
  // -----------------------------------
  Register scratch = r8;

  if (masm->emit_debug_code()) {
    // Allow r2 to be a FixedArray, or a FixedDoubleArray if r4 == 0.
    Label ok, fail;
    __ AssertNotSmi(r2);
    __ ldr(scratch, FieldMemOperand(r2, HeapObject::kMapOffset));
    __ ldrh(r6, FieldMemOperand(scratch, Map::kInstanceTypeOffset));
    __ cmp(r6, Operand(FIXED_ARRAY_TYPE));
    __ b(eq, &ok);
    __ cmp(r6, Operand(FIXED_DOUBLE_ARRAY_TYPE));
    __ b(ne, &fail);
    __ cmp(r4, Operand(0));
    __ b(eq, &ok);
    // Fall through.
    __ bind(&fail);
    __ Abort(AbortReason::kOperandIsNotAFixedArray);

    __ bind(&ok);
  }

  Label stack_overflow;
  Generate_StackOverflowCheck(masm, r4, scratch, &stack_overflow);

  // Push arguments onto the stack (thisArgument is already on the stack).
  {
    __ mov(r6, Operand(0));
    __ LoadRoot(r5, RootIndex::kTheHoleValue);
    Label done, loop;
    __ bind(&loop);
    __ cmp(r6, r4);
    __ b(eq, &done);
    __ add(scratch, r2, Operand(r6, LSL, kPointerSizeLog2));
    __ ldr(scratch, FieldMemOperand(scratch, FixedArray::kHeaderSize));
    __ cmp(scratch, r5);
    __ LoadRoot(scratch, RootIndex::kUndefinedValue, eq);
    __ Push(scratch);
    __ add(r6, r6, Operand(1));
    __ b(&loop);
    __ bind(&done);
    __ add(r0, r0, r6);
  }

  // Tail-call to the actual Call or Construct builtin.
  __ Jump(code, RelocInfo::CODE_TARGET);

  __ bind(&stack_overflow);
  __ TailCallRuntime(Runtime::kThrowStackOverflow);
}

// static
void Builtins::Generate_CallOrConstructForwardVarargs(MacroAssembler* masm,
                                                      CallOrConstructMode mode,
                                                      Handle<Code> code) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r3 : the new.target (for [[Construct]] calls)
  //  -- r1 : the target to call (can be any Object)
  //  -- r2 : start index (to support rest parameters)
  // -----------------------------------

  Register scratch = r6;

  // Check if new.target has a [[Construct]] internal method.
  if (mode == CallOrConstructMode::kConstruct) {
    Label new_target_constructor, new_target_not_constructor;
    __ JumpIfSmi(r3, &new_target_not_constructor);
    __ ldr(scratch, FieldMemOperand(r3, HeapObject::kMapOffset));
    __ ldrb(scratch, FieldMemOperand(scratch, Map::kBitFieldOffset));
    __ tst(scratch, Operand(Map::IsConstructorBit::kMask));
    __ b(ne, &new_target_constructor);
    __ bind(&new_target_not_constructor);
    {
      FrameScope scope(masm, StackFrame::MANUAL);
      __ EnterFrame(StackFrame::INTERNAL);
      __ Push(r3);
      __ CallRuntime(Runtime::kThrowNotConstructor);
    }
    __ bind(&new_target_constructor);
  }

  // Check if we have an arguments adaptor frame below the function frame.
  Label arguments_adaptor, arguments_done;
  __ ldr(r4, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
  __ ldr(scratch,
         MemOperand(r4, CommonFrameConstants::kContextOrFrameTypeOffset));
  __ cmp(scratch,
         Operand(StackFrame::TypeToMarker(StackFrame::ARGUMENTS_ADAPTOR)));
  __ b(eq, &arguments_adaptor);
  {
    __ ldr(r5, MemOperand(fp, JavaScriptFrameConstants::kFunctionOffset));
    __ ldr(r5, FieldMemOperand(r5, JSFunction::kSharedFunctionInfoOffset));
    __ ldrh(r5, FieldMemOperand(
                    r5, SharedFunctionInfo::kFormalParameterCountOffset));
    __ mov(r4, fp);
  }
  __ b(&arguments_done);
  __ bind(&arguments_adaptor);
  {
    // Load the length from the ArgumentsAdaptorFrame.
    __ ldr(r5, MemOperand(r4, ArgumentsAdaptorFrameConstants::kLengthOffset));
    __ SmiUntag(r5);
  }
  __ bind(&arguments_done);

  Label stack_done, stack_overflow;
  __ sub(r5, r5, r2, SetCC);
  __ b(le, &stack_done);
  {
    // Check for stack overflow.
    Generate_StackOverflowCheck(masm, r5, r2, &stack_overflow);

    // Forward the arguments from the caller frame.
    {
      Label loop;
      __ add(r4, r4, Operand(kPointerSize));
      __ add(r0, r0, r5);
      __ bind(&loop);
      {
        __ ldr(scratch, MemOperand(r4, r5, LSL, kPointerSizeLog2));
        __ push(scratch);
        __ sub(r5, r5, Operand(1), SetCC);
        __ b(ne, &loop);
      }
    }
  }
  __ b(&stack_done);
  __ bind(&stack_overflow);
  __ TailCallRuntime(Runtime::kThrowStackOverflow);
  __ bind(&stack_done);

  // Tail-call to the {code} handler.
  __ Jump(code, RelocInfo::CODE_TARGET);
}

// static
void Builtins::Generate_CallFunction(MacroAssembler* masm,
                                     ConvertReceiverMode mode) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : the function to call (checked to be a JSFunction)
  // -----------------------------------
  __ AssertFunction(r1);

  // See ES6 section 9.2.1 [[Call]] ( thisArgument, argumentsList)
  // Check that the function is not a "classConstructor".
  Label class_constructor;
  __ ldr(r2, FieldMemOperand(r1, JSFunction::kSharedFunctionInfoOffset));
  __ ldr(r3, FieldMemOperand(r2, SharedFunctionInfo::kFlagsOffset));
  __ tst(r3, Operand(SharedFunctionInfo::IsClassConstructorBit::kMask));
  __ b(ne, &class_constructor);

  // Enter the context of the function; ToObject has to run in the function
  // context, and we also need to take the global proxy from the function
  // context in case of conversion.
  __ ldr(cp, FieldMemOperand(r1, JSFunction::kContextOffset));
  // We need to convert the receiver for non-native sloppy mode functions.
  Label done_convert;
  __ ldr(r3, FieldMemOperand(r2, SharedFunctionInfo::kFlagsOffset));
  __ tst(r3, Operand(SharedFunctionInfo::IsNativeBit::kMask |
                     SharedFunctionInfo::IsStrictBit::kMask));
  __ b(ne, &done_convert);
  {
    // ----------- S t a t e -------------
    //  -- r0 : the number of arguments (not including the receiver)
    //  -- r1 : the function to call (checked to be a JSFunction)
    //  -- r2 : the shared function info.
    //  -- cp : the function context.
    // -----------------------------------

    if (mode == ConvertReceiverMode::kNullOrUndefined) {
      // Patch receiver to global proxy.
      __ LoadGlobalProxy(r3);
    } else {
      Label convert_to_object, convert_receiver;
      __ ldr(r3, MemOperand(sp, r0, LSL, kPointerSizeLog2));
      __ JumpIfSmi(r3, &convert_to_object);
      STATIC_ASSERT(LAST_JS_RECEIVER_TYPE == LAST_TYPE);
      __ CompareObjectType(r3, r4, r4, FIRST_JS_RECEIVER_TYPE);
      __ b(hs, &done_convert);
      if (mode != ConvertReceiverMode::kNotNullOrUndefined) {
        Label convert_global_proxy;
        __ JumpIfRoot(r3, RootIndex::kUndefinedValue, &convert_global_proxy);
        __ JumpIfNotRoot(r3, RootIndex::kNullValue, &convert_to_object);
        __ bind(&convert_global_proxy);
        {
          // Patch receiver to global proxy.
          __ LoadGlobalProxy(r3);
        }
        __ b(&convert_receiver);
      }
      __ bind(&convert_to_object);
      {
        // Convert receiver using ToObject.
        // TODO(bmeurer): Inline the allocation here to avoid building the frame
        // in the fast case? (fall back to AllocateInNewSpace?)
        FrameAndConstantPoolScope scope(masm, StackFrame::INTERNAL);
        __ SmiTag(r0);
        __ Push(r0, r1);
        __ mov(r0, r3);
        __ Push(cp);
        __ Call(BUILTIN_CODE(masm->isolate(), ToObject),
                RelocInfo::CODE_TARGET);
        __ Pop(cp);
        __ mov(r3, r0);
        __ Pop(r0, r1);
        __ SmiUntag(r0);
      }
      __ ldr(r2, FieldMemOperand(r1, JSFunction::kSharedFunctionInfoOffset));
      __ bind(&convert_receiver);
    }
    __ str(r3, MemOperand(sp, r0, LSL, kPointerSizeLog2));
  }
  __ bind(&done_convert);

  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : the function to call (checked to be a JSFunction)
  //  -- r2 : the shared function info.
  //  -- cp : the function context.
  // -----------------------------------

  __ ldrh(r2,
          FieldMemOperand(r2, SharedFunctionInfo::kFormalParameterCountOffset));
  ParameterCount actual(r0);
  ParameterCount expected(r2);
  __ InvokeFunctionCode(r1, no_reg, expected, actual, JUMP_FUNCTION);

  // The function is a "classConstructor", need to raise an exception.
  __ bind(&class_constructor);
  {
    FrameScope frame(masm, StackFrame::INTERNAL);
    __ push(r1);
    __ CallRuntime(Runtime::kThrowConstructorNonCallableError);
  }
}

namespace {

void Generate_PushBoundArguments(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : target (checked to be a JSBoundFunction)
  //  -- r3 : new.target (only in case of [[Construct]])
  // -----------------------------------

  // Load [[BoundArguments]] into r2 and length of that into r4.
  Label no_bound_arguments;
  __ ldr(r2, FieldMemOperand(r1, JSBoundFunction::kBoundArgumentsOffset));
  __ ldr(r4, FieldMemOperand(r2, FixedArray::kLengthOffset));
  __ SmiUntag(r4);
  __ cmp(r4, Operand(0));
  __ b(eq, &no_bound_arguments);
  {
    // ----------- S t a t e -------------
    //  -- r0 : the number of arguments (not including the receiver)
    //  -- r1 : target (checked to be a JSBoundFunction)
    //  -- r2 : the [[BoundArguments]] (implemented as FixedArray)
    //  -- r3 : new.target (only in case of [[Construct]])
    //  -- r4 : the number of [[BoundArguments]]
    // -----------------------------------

    Register scratch = r6;

    {
      // Check the stack for overflow. We are not trying to catch interruptions
      // (i.e. debug break and preemption) here, so check the "real stack
      // limit".
      Label done;
      __ mov(scratch, Operand(r4, LSL, kPointerSizeLog2));
      {
        UseScratchRegisterScope temps(masm);
        Register remaining_stack_size = temps.Acquire();

        // Compute the space we have left. The stack might already be overflowed
        // here which will cause remaining_stack_size to become negative.
        __ LoadRoot(remaining_stack_size, RootIndex::kRealStackLimit);
        __ sub(remaining_stack_size, sp, remaining_stack_size);

        // Check if the arguments will overflow the stack.
        __ cmp(remaining_stack_size, scratch);
      }
      __ b(gt, &done);
      {
        FrameScope scope(masm, StackFrame::MANUAL);
        __ EnterFrame(StackFrame::INTERNAL);
        __ CallRuntime(Runtime::kThrowStackOverflow);
      }
      __ bind(&done);
    }

    // Reserve stack space for the [[BoundArguments]].
    __ AllocateStackSpace(scratch);

    // Relocate arguments down the stack.
    {
      Label loop, done_loop;
      __ mov(r5, Operand(0));
      __ bind(&loop);
      __ cmp(r5, r0);
      __ b(gt, &done_loop);
      __ ldr(scratch, MemOperand(sp, r4, LSL, kPointerSizeLog2));
      __ str(scratch, MemOperand(sp, r5, LSL, kPointerSizeLog2));
      __ add(r4, r4, Operand(1));
      __ add(r5, r5, Operand(1));
      __ b(&loop);
      __ bind(&done_loop);
    }

    // Copy [[BoundArguments]] to the stack (below the arguments).
    {
      Label loop;
      __ ldr(r4, FieldMemOperand(r2, FixedArray::kLengthOffset));
      __ SmiUntag(r4);
      __ add(r2, r2, Operand(FixedArray::kHeaderSize - kHeapObjectTag));
      __ bind(&loop);
      __ sub(r4, r4, Operand(1), SetCC);
      __ ldr(scratch, MemOperand(r2, r4, LSL, kPointerSizeLog2));
      __ str(scratch, MemOperand(sp, r0, LSL, kPointerSizeLog2));
      __ add(r0, r0, Operand(1));
      __ b(gt, &loop);
    }
  }
  __ bind(&no_bound_arguments);
}

}  // namespace

// static
void Builtins::Generate_CallBoundFunctionImpl(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : the function to call (checked to be a JSBoundFunction)
  // -----------------------------------
  __ AssertBoundFunction(r1);

  // Patch the receiver to [[BoundThis]].
  __ ldr(r3, FieldMemOperand(r1, JSBoundFunction::kBoundThisOffset));
  __ str(r3, MemOperand(sp, r0, LSL, kPointerSizeLog2));

  // Push the [[BoundArguments]] onto the stack.
  Generate_PushBoundArguments(masm);

  // Call the [[BoundTargetFunction]] via the Call builtin.
  __ ldr(r1, FieldMemOperand(r1, JSBoundFunction::kBoundTargetFunctionOffset));
  __ Jump(BUILTIN_CODE(masm->isolate(), Call_ReceiverIsAny),
          RelocInfo::CODE_TARGET);
}

// static
void Builtins::Generate_Call(MacroAssembler* masm, ConvertReceiverMode mode) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : the target to call (can be any Object).
  // -----------------------------------

  Label non_callable, non_function, non_smi;
  __ JumpIfSmi(r1, &non_callable);
  __ bind(&non_smi);
  __ CompareObjectType(r1, r4, r5, JS_FUNCTION_TYPE);
  __ Jump(masm->isolate()->builtins()->CallFunction(mode),
          RelocInfo::CODE_TARGET, eq);
  __ cmp(r5, Operand(JS_BOUND_FUNCTION_TYPE));
  __ Jump(BUILTIN_CODE(masm->isolate(), CallBoundFunction),
          RelocInfo::CODE_TARGET, eq);

  // Check if target has a [[Call]] internal method.
  __ ldrb(r4, FieldMemOperand(r4, Map::kBitFieldOffset));
  __ tst(r4, Operand(Map::IsCallableBit::kMask));
  __ b(eq, &non_callable);

  // Check if target is a proxy and call CallProxy external builtin
  __ cmp(r5, Operand(JS_PROXY_TYPE));
  __ b(ne, &non_function);
  __ Jump(BUILTIN_CODE(masm->isolate(), CallProxy), RelocInfo::CODE_TARGET);

  // 2. Call to something else, which might have a [[Call]] internal method (if
  // not we raise an exception).
  __ bind(&non_function);
  // Overwrite the original receiver the (original) target.
  __ str(r1, MemOperand(sp, r0, LSL, kPointerSizeLog2));
  // Let the "call_as_function_delegate" take care of the rest.
  __ LoadNativeContextSlot(Context::CALL_AS_FUNCTION_DELEGATE_INDEX, r1);
  __ Jump(masm->isolate()->builtins()->CallFunction(
              ConvertReceiverMode::kNotNullOrUndefined),
          RelocInfo::CODE_TARGET);

  // 3. Call to something that is not callable.
  __ bind(&non_callable);
  {
    FrameAndConstantPoolScope scope(masm, StackFrame::INTERNAL);
    __ Push(r1);
    __ CallRuntime(Runtime::kThrowCalledNonCallable);
  }
}

// static
void Builtins::Generate_ConstructFunction(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : the constructor to call (checked to be a JSFunction)
  //  -- r3 : the new target (checked to be a constructor)
  // -----------------------------------
  __ AssertConstructor(r1);
  __ AssertFunction(r1);

  // Calling convention for function specific ConstructStubs require
  // r2 to contain either an AllocationSite or undefined.
  __ LoadRoot(r2, RootIndex::kUndefinedValue);

  Label call_generic_stub;

  // Jump to JSBuiltinsConstructStub or JSConstructStubGeneric.
  __ ldr(r4, FieldMemOperand(r1, JSFunction::kSharedFunctionInfoOffset));
  __ ldr(r4, FieldMemOperand(r4, SharedFunctionInfo::kFlagsOffset));
  __ tst(r4, Operand(SharedFunctionInfo::ConstructAsBuiltinBit::kMask));
  __ b(eq, &call_generic_stub);

  __ Jump(BUILTIN_CODE(masm->isolate(), JSBuiltinsConstructStub),
          RelocInfo::CODE_TARGET);

  __ bind(&call_generic_stub);
  __ Jump(BUILTIN_CODE(masm->isolate(), JSConstructStubGeneric),
          RelocInfo::CODE_TARGET);
}

// static
void Builtins::Generate_ConstructBoundFunction(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : the function to call (checked to be a JSBoundFunction)
  //  -- r3 : the new target (checked to be a constructor)
  // -----------------------------------
  __ AssertConstructor(r1);
  __ AssertBoundFunction(r1);

  // Push the [[BoundArguments]] onto the stack.
  Generate_PushBoundArguments(masm);

  // Patch new.target to [[BoundTargetFunction]] if new.target equals target.
  __ cmp(r1, r3);
  __ ldr(r3, FieldMemOperand(r1, JSBoundFunction::kBoundTargetFunctionOffset),
         eq);

  // Construct the [[BoundTargetFunction]] via the Construct builtin.
  __ ldr(r1, FieldMemOperand(r1, JSBoundFunction::kBoundTargetFunctionOffset));
  __ Jump(BUILTIN_CODE(masm->isolate(), Construct), RelocInfo::CODE_TARGET);
}

// static
void Builtins::Generate_Construct(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : the number of arguments (not including the receiver)
  //  -- r1 : the constructor to call (can be any Object)
  //  -- r3 : the new target (either the same as the constructor or
  //          the JSFunction on which new was invoked initially)
  // -----------------------------------

  // Check if target is a Smi.
  Label non_constructor, non_proxy;
  __ JumpIfSmi(r1, &non_constructor);

  // Check if target has a [[Construct]] internal method.
  __ ldr(r4, FieldMemOperand(r1, HeapObject::kMapOffset));
  __ ldrb(r2, FieldMemOperand(r4, Map::kBitFieldOffset));
  __ tst(r2, Operand(Map::IsConstructorBit::kMask));
  __ b(eq, &non_constructor);

  // Dispatch based on instance type.
  __ CompareInstanceType(r4, r5, JS_FUNCTION_TYPE);
  __ Jump(BUILTIN_CODE(masm->isolate(), ConstructFunction),
          RelocInfo::CODE_TARGET, eq);

  // Only dispatch to bound functions after checking whether they are
  // constructors.
  __ cmp(r5, Operand(JS_BOUND_FUNCTION_TYPE));
  __ Jump(BUILTIN_CODE(masm->isolate(), ConstructBoundFunction),
          RelocInfo::CODE_TARGET, eq);

  // Only dispatch to proxies after checking whether they are constructors.
  __ cmp(r5, Operand(JS_PROXY_TYPE));
  __ b(ne, &non_proxy);
  __ Jump(BUILTIN_CODE(masm->isolate(), ConstructProxy),
          RelocInfo::CODE_TARGET);

  // Called Construct on an exotic Object with a [[Construct]] internal method.
  __ bind(&non_proxy);
  {
    // Overwrite the original receiver with the (original) target.
    __ str(r1, MemOperand(sp, r0, LSL, kPointerSizeLog2));
    // Let the "call_as_constructor_delegate" take care of the rest.
    __ LoadNativeContextSlot(Context::CALL_AS_CONSTRUCTOR_DELEGATE_INDEX, r1);
    __ Jump(masm->isolate()->builtins()->CallFunction(),
            RelocInfo::CODE_TARGET);
  }

  // Called Construct on an Object that doesn't have a [[Construct]] internal
  // method.
  __ bind(&non_constructor);
  __ Jump(BUILTIN_CODE(masm->isolate(), ConstructedNonConstructable),
          RelocInfo::CODE_TARGET);
}

void Builtins::Generate_ArgumentsAdaptorTrampoline(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : actual number of arguments
  //  -- r1 : function (passed through to callee)
  //  -- r2 : expected number of arguments
  //  -- r3 : new target (passed through to callee)
  // -----------------------------------

  Label dont_adapt_arguments, stack_overflow, skip_adapt_arguments;
  __ cmp(r2, Operand(SharedFunctionInfo::kDontAdaptArgumentsSentinel));
  __ b(eq, &dont_adapt_arguments);
  __ ldr(r4, FieldMemOperand(r1, JSFunction::kSharedFunctionInfoOffset));
  __ ldr(r4, FieldMemOperand(r4, SharedFunctionInfo::kFlagsOffset));
  __ tst(r4,
         Operand(SharedFunctionInfo::IsSafeToSkipArgumentsAdaptorBit::kMask));
  __ b(ne, &skip_adapt_arguments);

  // -------------------------------------------
  // Adapt arguments.
  // -------------------------------------------
  {
    Label under_application, over_application, invoke;
    __ cmp(r0, r2);
    __ b(lt, &under_application);

    // Enough parameters: actual >= expected
    __ bind(&over_application);
    {
      EnterArgumentsAdaptorFrame(masm);
      Generate_StackOverflowCheck(masm, r2, r5, &stack_overflow);

      // Calculate copy start address into r0 and copy end address into r4.
      // r0: actual number of arguments as a smi
      // r1: function
      // r2: expected number of arguments
      // r3: new target (passed through to callee)
      __ add(r0, fp, Operand::PointerOffsetFromSmiKey(r0));
      // adjust for return address and receiver
      __ add(r0, r0, Operand(2 * kPointerSize));
      __ sub(r4, r0, Operand(r2, LSL, kPointerSizeLog2));

      // Copy the arguments (including the receiver) to the new stack frame.
      // r0: copy start address
      // r1: function
      // r2: expected number of arguments
      // r3: new target (passed through to callee)
      // r4: copy end address

      Label copy;
      __ bind(&copy);
      __ ldr(r5, MemOperand(r0, 0));
      __ push(r5);
      __ cmp(r0, r4);  // Compare before moving to next argument.
      __ sub(r0, r0, Operand(kPointerSize));
      __ b(ne, &copy);

      __ b(&invoke);
    }

    // Too few parameters: Actual < expected
    __ bind(&under_application);
    {
      EnterArgumentsAdaptorFrame(masm);
      Generate_StackOverflowCheck(masm, r2, r5, &stack_overflow);

      // Calculate copy start address into r0 and copy end address is fp.
      // r0: actual number of arguments as a smi
      // r1: function
      // r2: expected number of arguments
      // r3: new target (passed through to callee)
      __ add(r0, fp, Operand::PointerOffsetFromSmiKey(r0));

      // Copy the arguments (including the receiver) to the new stack frame.
      // r0: copy start address
      // r1: function
      // r2: expected number of arguments
      // r3: new target (passed through to callee)
      Label copy;
      __ bind(&copy);

      // Adjust load for return address and receiver.
      __ ldr(r5, MemOperand(r0, 2 * kPointerSize));
      __ push(r5);

      __ cmp(r0, fp);  // Compare before moving to next argument.
      __ sub(r0, r0, Operand(kPointerSize));
      __ b(ne, &copy);

      // Fill the remaining expected arguments with undefined.
      // r1: function
      // r2: expected number of arguments
      // r3: new target (passed through to callee)
      __ LoadRoot(r5, RootIndex::kUndefinedValue);
      __ sub(r4, fp, Operand(r2, LSL, kPointerSizeLog2));
      // Adjust for frame.
      __ sub(r4, r4,
             Operand(ArgumentsAdaptorFrameConstants::kFixedFrameSizeFromFp +
                     kPointerSize));

      Label fill;
      __ bind(&fill);
      __ push(r5);
      __ cmp(sp, r4);
      __ b(ne, &fill);
    }

    // Call the entry point.
    __ bind(&invoke);
    __ mov(r0, r2);
    // r0 : expected number of arguments
    // r1 : function (passed through to callee)
    // r3 : new target (passed through to callee)
    static_assert(kJavaScriptCallCodeStartRegister == r2, "ABI mismatch");
    __ ldr(r2, FieldMemOperand(r1, JSFunction::kCodeOffset));
    __ CallCodeObject(r2);

    // Store offset of return address for deoptimizer.
    masm->isolate()->heap()->SetArgumentsAdaptorDeoptPCOffset(
        masm->pc_offset());

    // Exit frame and return.
    LeaveArgumentsAdaptorFrame(masm);
    __ Jump(lr);
  }

  // -------------------------------------------
  // Skip adapt arguments.
  // -------------------------------------------
  __ bind(&skip_adapt_arguments);
  {
    // The callee cannot observe the actual arguments, so it's safe to just
    // pass the expected arguments by massaging the stack appropriately. See
    // http://bit.ly/v8-faster-calls-with-arguments-mismatch for details.
    Label under_application, over_application;
    __ cmp(r0, r2);
    __ b(lt, &under_application);

    __ bind(&over_application);
    {
      // Remove superfluous parameters from the stack.
      __ sub(r4, r0, r2);
      __ mov(r0, r2);
      __ add(sp, sp, Operand(r4, LSL, kPointerSizeLog2));
      __ b(&dont_adapt_arguments);
    }

    __ bind(&under_application);
    {
      // Fill remaining expected arguments with undefined values.
      Label fill;
      __ LoadRoot(r4, RootIndex::kUndefinedValue);
      __ bind(&fill);
      __ add(r0, r0, Operand(1));
      __ push(r4);
      __ cmp(r0, r2);
      __ b(lt, &fill);
      __ b(&dont_adapt_arguments);
    }
  }

  // -------------------------------------------
  // Dont adapt arguments.
  // -------------------------------------------
  __ bind(&dont_adapt_arguments);
  static_assert(kJavaScriptCallCodeStartRegister == r2, "ABI mismatch");
  __ ldr(r2, FieldMemOperand(r1, JSFunction::kCodeOffset));
  __ JumpCodeObject(r2);

  __ bind(&stack_overflow);
  {
    FrameScope frame(masm, StackFrame::MANUAL);
    __ CallRuntime(Runtime::kThrowStackOverflow);
    __ bkpt(0);
  }
}

void Builtins::Generate_WasmCompileLazy(MacroAssembler* masm) {
  // The function index was put in a register by the jump table trampoline.
  // Convert to Smi for the runtime call.
  __ SmiTag(kWasmCompileLazyFuncIndexRegister,
            kWasmCompileLazyFuncIndexRegister);
  {
    HardAbortScope hard_abort(masm);  // Avoid calls to Abort.
    FrameAndConstantPoolScope scope(masm, StackFrame::WASM_COMPILE_LAZY);

    // Save all parameter registers (see wasm-linkage.cc). They might be
    // overwritten in the runtime call below. We don't have any callee-saved
    // registers in wasm, so no need to store anything else.
    constexpr RegList gp_regs = Register::ListOf<r0, r1, r2, r3>();
    constexpr DwVfpRegister lowest_fp_reg = d0;
    constexpr DwVfpRegister highest_fp_reg = d7;

    __ stm(db_w, sp, gp_regs);
    __ vstm(db_w, sp, lowest_fp_reg, highest_fp_reg);

    // Pass instance and function index as explicit arguments to the runtime
    // function.
    __ push(kWasmInstanceRegister);
    __ push(kWasmCompileLazyFuncIndexRegister);
    // Load the correct CEntry builtin from the instance object.
    __ ldr(r2, FieldMemOperand(kWasmInstanceRegister,
                               WasmInstanceObject::kCEntryStubOffset));
    // Initialize the JavaScript context with 0. CEntry will use it to
    // set the current context on the isolate.
    __ Move(cp, Smi::zero());
    __ CallRuntimeWithCEntry(Runtime::kWasmCompileLazy, r2);
    // The entrypoint address is the return value.
    __ mov(r8, kReturnRegister0);

    // Restore registers.
    __ vldm(ia_w, sp, lowest_fp_reg, highest_fp_reg);
    __ ldm(ia_w, sp, gp_regs);
  }
  // Finally, jump to the entrypoint.
  __ Jump(r8);
}

void Builtins::Generate_CEntry(MacroAssembler* masm, int result_size,
                               SaveFPRegsMode save_doubles, ArgvMode argv_mode,
                               bool builtin_exit_frame) {
  // Called from JavaScript; parameters are on stack as if calling JS function.
  // r0: number of arguments including receiver
  // r1: pointer to builtin function
  // fp: frame pointer  (restored after C call)
  // sp: stack pointer  (restored as callee's sp after C call)
  // cp: current context  (C callee-saved)
  //
  // If argv_mode == kArgvInRegister:
  // r2: pointer to the first argument

  __ mov(r5, Operand(r1));

  if (argv_mode == kArgvInRegister) {
    // Move argv into the correct register.
    __ mov(r1, Operand(r2));
  } else {
    // Compute the argv pointer in a callee-saved register.
    __ add(r1, sp, Operand(r0, LSL, kPointerSizeLog2));
    __ sub(r1, r1, Operand(kPointerSize));
  }

  // Enter the exit frame that transitions from JavaScript to C++.
  FrameScope scope(masm, StackFrame::MANUAL);
  __ EnterExitFrame(
      save_doubles == kSaveFPRegs, 0,
      builtin_exit_frame ? StackFrame::BUILTIN_EXIT : StackFrame::EXIT);

  // Store a copy of argc in callee-saved registers for later.
  __ mov(r4, Operand(r0));

// r0, r4: number of arguments including receiver  (C callee-saved)
// r1: pointer to the first argument (C callee-saved)
// r5: pointer to builtin function  (C callee-saved)

#if V8_HOST_ARCH_ARM
  int frame_alignment = MacroAssembler::ActivationFrameAlignment();
  int frame_alignment_mask = frame_alignment - 1;
  if (FLAG_debug_code) {
    if (frame_alignment > kPointerSize) {
      Label alignment_as_expected;
      DCHECK(base::bits::IsPowerOfTwo(frame_alignment));
      __ tst(sp, Operand(frame_alignment_mask));
      __ b(eq, &alignment_as_expected);
      // Don't use Check here, as it will call Runtime_Abort re-entering here.
      __ stop();
      __ bind(&alignment_as_expected);
    }
  }
#endif

  // Call C built-in.
  // r0 = argc, r1 = argv, r2 = isolate
  __ Move(r2, ExternalReference::isolate_address(masm->isolate()));
  __ StoreReturnAddressAndCall(r5);

  // Result returned in r0 or r1:r0 - do not destroy these registers!

  // Check result for exception sentinel.
  Label exception_returned;
  __ CompareRoot(r0, RootIndex::kException);
  __ b(eq, &exception_returned);

  // Check that there is no pending exception, otherwise we
  // should have returned the exception sentinel.
  if (FLAG_debug_code) {
    Label okay;
    ExternalReference pending_exception_address = ExternalReference::Create(
        IsolateAddressId::kPendingExceptionAddress, masm->isolate());
    __ Move(r3, pending_exception_address);
    __ ldr(r3, MemOperand(r3));
    __ CompareRoot(r3, RootIndex::kTheHoleValue);
    // Cannot use check here as it attempts to generate call into runtime.
    __ b(eq, &okay);
    __ stop();
    __ bind(&okay);
  }

  // Exit C frame and return.
  // r0:r1: result
  // sp: stack pointer
  // fp: frame pointer
  Register argc = argv_mode == kArgvInRegister
                      // We don't want to pop arguments so set argc to no_reg.
                      ? no_reg
                      // Callee-saved register r4 still holds argc.
                      : r4;
  __ LeaveExitFrame(save_doubles == kSaveFPRegs, argc);
  __ mov(pc, lr);

  // Handling of exception.
  __ bind(&exception_returned);

  ExternalReference pending_handler_context_address = ExternalReference::Create(
      IsolateAddressId::kPendingHandlerContextAddress, masm->isolate());
  ExternalReference pending_handler_entrypoint_address =
      ExternalReference::Create(
          IsolateAddressId::kPendingHandlerEntrypointAddress, masm->isolate());
  ExternalReference pending_handler_fp_address = ExternalReference::Create(
      IsolateAddressId::kPendingHandlerFPAddress, masm->isolate());
  ExternalReference pending_handler_sp_address = ExternalReference::Create(
      IsolateAddressId::kPendingHandlerSPAddress, masm->isolate());

  // Ask the runtime for help to determine the handler. This will set r0 to
  // contain the current pending exception, don't clobber it.
  ExternalReference find_handler =
      ExternalReference::Create(Runtime::kUnwindAndFindExceptionHandler);
  {
    FrameScope scope(masm, StackFrame::MANUAL);
    __ PrepareCallCFunction(3, 0);
    __ mov(r0, Operand(0));
    __ mov(r1, Operand(0));
    __ Move(r2, ExternalReference::isolate_address(masm->isolate()));
    __ CallCFunction(find_handler, 3);
  }

  // Retrieve the handler context, SP and FP.
  __ Move(cp, pending_handler_context_address);
  __ ldr(cp, MemOperand(cp));
  __ Move(sp, pending_handler_sp_address);
  __ ldr(sp, MemOperand(sp));
  __ Move(fp, pending_handler_fp_address);
  __ ldr(fp, MemOperand(fp));

  // If the handler is a JS frame, restore the context to the frame. Note that
  // the context will be set to (cp == 0) for non-JS frames.
  __ cmp(cp, Operand(0));
  __ str(cp, MemOperand(fp, StandardFrameConstants::kContextOffset), ne);

  // Reset the masking register. This is done independent of the underlying
  // feature flag {FLAG_untrusted_code_mitigations} to make the snapshot work
  // with both configurations. It is safe to always do this, because the
  // underlying register is caller-saved and can be arbitrarily clobbered.
  __ ResetSpeculationPoisonRegister();

  // Compute the handler entry address and jump to it.
  ConstantPoolUnavailableScope constant_pool_unavailable(masm);
  __ Move(r1, pending_handler_entrypoint_address);
  __ ldr(r1, MemOperand(r1));
  __ Jump(r1);
}

void Builtins::Generate_DoubleToI(MacroAssembler* masm) {
  Label negate, done;

  HardAbortScope hard_abort(masm);  // Avoid calls to Abort.
  UseScratchRegisterScope temps(masm);
  Register result_reg = r7;
  Register double_low = GetRegisterThatIsNotOneOf(result_reg);
  Register double_high = GetRegisterThatIsNotOneOf(result_reg, double_low);
  LowDwVfpRegister double_scratch = temps.AcquireLowD();

  // Save the old values from these temporary registers on the stack.
  __ Push(result_reg, double_high, double_low);

  // Account for saved regs.
  const int kArgumentOffset = 3 * kPointerSize;

  MemOperand input_operand(sp, kArgumentOffset);
  MemOperand result_operand = input_operand;

  // Load double input.
  __ vldr(double_scratch, input_operand);
  __ vmov(double_low, double_high, double_scratch);
  // Try to convert with a FPU convert instruction. This handles all
  // non-saturating cases.
  __ TryInlineTruncateDoubleToI(result_reg, double_scratch, &done);

  Register scratch = temps.Acquire();
  __ Ubfx(scratch, double_high, HeapNumber::kExponentShift,
          HeapNumber::kExponentBits);
  // Load scratch with exponent - 1. This is faster than loading
  // with exponent because Bias + 1 = 1024 which is an *ARM* immediate value.
  STATIC_ASSERT(HeapNumber::kExponentBias + 1 == 1024);
  __ sub(scratch, scratch, Operand(HeapNumber::kExponentBias + 1));
  // If exponent is greater than or equal to 84, the 32 less significant
  // bits are 0s (2^84 = 1, 52 significant bits, 32 uncoded bits),
  // the result is 0.
  // Compare exponent with 84 (compare exponent - 1 with 83). If the exponent is
  // greater than this, the conversion is out of range, so return zero.
  __ cmp(scratch, Operand(83));
  __ mov(result_reg, Operand::Zero(), LeaveCC, ge);
  __ b(ge, &done);

  // If we reach this code, 30 <= exponent <= 83.
  // `TryInlineTruncateDoubleToI` above will have truncated any double with an
  // exponent lower than 30.
  if (masm->emit_debug_code()) {
    // Scratch is exponent - 1.
    __ cmp(scratch, Operand(30 - 1));
    __ Check(ge, AbortReason::kUnexpectedValue);
  }

  // We don't have to handle cases where 0 <= exponent <= 20 for which we would
  // need to shift right the high part of the mantissa.
  // Scratch contains exponent - 1.
  // Load scratch with 52 - exponent (load with 51 - (exponent - 1)).
  __ rsb(scratch, scratch, Operand(51), SetCC);

  // 52 <= exponent <= 83, shift only double_low.
  // On entry, scratch contains: 52 - exponent.
  __ rsb(scratch, scratch, Operand::Zero(), LeaveCC, ls);
  __ mov(result_reg, Operand(double_low, LSL, scratch), LeaveCC, ls);
  __ b(ls, &negate);

  // 21 <= exponent <= 51, shift double_low and double_high
  // to generate the result.
  __ mov(double_low, Operand(double_low, LSR, scratch));
  // Scratch contains: 52 - exponent.
  // We needs: exponent - 20.
  // So we use: 32 - scratch = 32 - 52 + exponent = exponent - 20.
  __ rsb(scratch, scratch, Operand(32));
  __ Ubfx(result_reg, double_high, 0, HeapNumber::kMantissaBitsInTopWord);
  // Set the implicit 1 before the mantissa part in double_high.
  __ orr(result_reg, result_reg,
         Operand(1 << HeapNumber::kMantissaBitsInTopWord));
  __ orr(result_reg, double_low, Operand(result_reg, LSL, scratch));

  __ bind(&negate);
  // If input was positive, double_high ASR 31 equals 0 and
  // double_high LSR 31 equals zero.
  // New result = (result eor 0) + 0 = result.
  // If the input was negative, we have to negate the result.
  // Input_high ASR 31 equals 0xFFFFFFFF and double_high LSR 31 equals 1.
  // New result = (result eor 0xFFFFFFFF) + 1 = 0 - result.
  __ eor(result_reg, result_reg, Operand(double_high, ASR, 31));
  __ add(result_reg, result_reg, Operand(double_high, LSR, 31));

  __ bind(&done);
  __ str(result_reg, result_operand);

  // Restore registers corrupted in this routine and return.
  __ Pop(result_reg, double_high, double_low);
  __ Ret();
}

void Builtins::Generate_InternalArrayConstructorImpl(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- r0 : argc
  //  -- r1 : constructor
  //  -- sp[0] : return address
  //  -- sp[4] : last argument
  // -----------------------------------

  if (FLAG_debug_code) {
    // The array construct code is only set for the global and natives
    // builtin Array functions which always have maps.

    // Initial map for the builtin Array function should be a map.
    __ ldr(r3, FieldMemOperand(r1, JSFunction::kPrototypeOrInitialMapOffset));
    // Will both indicate a nullptr and a Smi.
    __ tst(r3, Operand(kSmiTagMask));
    __ Assert(ne, AbortReason::kUnexpectedInitialMapForArrayFunction);
    __ CompareObjectType(r3, r3, r4, MAP_TYPE);
    __ Assert(eq, AbortReason::kUnexpectedInitialMapForArrayFunction);

    // Figure out the right elements kind
    __ ldr(r3, FieldMemOperand(r1, JSFunction::kPrototypeOrInitialMapOffset));
    // Load the map's "bit field 2" into |result|. We only need the first byte,
    // but the following bit field extraction takes care of that anyway.
    __ ldr(r3, FieldMemOperand(r3, Map::kBitField2Offset));
    // Retrieve elements_kind from bit field 2.
    __ DecodeField<Map::ElementsKindBits>(r3);

    // Initial elements kind should be packed elements.
    __ cmp(r3, Operand(PACKED_ELEMENTS));
    __ Assert(eq, AbortReason::kInvalidElementsKindForInternalPackedArray);

    // No arguments should be passed.
    __ cmp(r0, Operand(0));
    __ Assert(eq, AbortReason::kWrongNumberOfArgumentsForInternalPackedArray);
  }

  __ Jump(
      BUILTIN_CODE(masm->isolate(), InternalArrayNoArgumentConstructor_Packed),
      RelocInfo::CODE_TARGET);
}

namespace {

int AddressOffset(ExternalReference ref0, ExternalReference ref1) {
  return ref0.address() - ref1.address();
}

// Calls an API function.  Allocates HandleScope, extracts returned value
// from handle and propagates exceptions.  Restores context.  stack_space
// - space to be unwound on exit (includes the call JS arguments space and
// the additional space allocated for the fast call).
void CallApiFunctionAndReturn(MacroAssembler* masm, Register function_address,
                              ExternalReference thunk_ref, int stack_space,
                              MemOperand* stack_space_operand,
                              MemOperand return_value_operand) {
  Isolate* isolate = masm->isolate();
  ExternalReference next_address =
      ExternalReference::handle_scope_next_address(isolate);
  const int kNextOffset = 0;
  const int kLimitOffset = AddressOffset(
      ExternalReference::handle_scope_limit_address(isolate), next_address);
  const int kLevelOffset = AddressOffset(
      ExternalReference::handle_scope_level_address(isolate), next_address);

  DCHECK(function_address == r1 || function_address == r2);

  Label profiler_enabled, end_profiler_check;
  __ Move(r9, ExternalReference::is_profiling_address(isolate));
  __ ldrb(r9, MemOperand(r9, 0));
  __ cmp(r9, Operand(0));
  __ b(ne, &profiler_enabled);
  __ Move(r9, ExternalReference::address_of_runtime_stats_flag());
  __ ldr(r9, MemOperand(r9, 0));
  __ cmp(r9, Operand(0));
  __ b(ne, &profiler_enabled);
  {
    // Call the api function directly.
    __ Move(r3, function_address);
    __ b(&end_profiler_check);
  }
  __ bind(&profiler_enabled);
  {
    // Additional parameter is the address of the actual callback.
    __ Move(r3, thunk_ref);
  }
  __ bind(&end_profiler_check);

  // Allocate HandleScope in callee-save registers.
  __ Move(r9, next_address);
  __ ldr(r4, MemOperand(r9, kNextOffset));
  __ ldr(r5, MemOperand(r9, kLimitOffset));
  __ ldr(r6, MemOperand(r9, kLevelOffset));
  __ add(r6, r6, Operand(1));
  __ str(r6, MemOperand(r9, kLevelOffset));

  __ StoreReturnAddressAndCall(r3);

  Label promote_scheduled_exception;
  Label delete_allocated_handles;
  Label leave_exit_frame;
  Label return_value_loaded;

  // load value from ReturnValue
  __ ldr(r0, return_value_operand);
  __ bind(&return_value_loaded);
  // No more valid handles (the result handle was the last one). Restore
  // previous handle scope.
  __ str(r4, MemOperand(r9, kNextOffset));
  if (__ emit_debug_code()) {
    __ ldr(r1, MemOperand(r9, kLevelOffset));
    __ cmp(r1, r6);
    __ Check(eq, AbortReason::kUnexpectedLevelAfterReturnFromApiCall);
  }
  __ sub(r6, r6, Operand(1));
  __ str(r6, MemOperand(r9, kLevelOffset));
  __ ldr(r6, MemOperand(r9, kLimitOffset));
  __ cmp(r5, r6);
  __ b(ne, &delete_allocated_handles);

  // Leave the API exit frame.
  __ bind(&leave_exit_frame);
  // LeaveExitFrame expects unwind space to be in a register.
  if (stack_space_operand == nullptr) {
    DCHECK_NE(stack_space, 0);
    __ mov(r4, Operand(stack_space));
  } else {
    DCHECK_EQ(stack_space, 0);
    __ ldr(r4, *stack_space_operand);
  }
  __ LeaveExitFrame(false, r4, stack_space_operand != nullptr);

  // Check if the function scheduled an exception.
  __ LoadRoot(r4, RootIndex::kTheHoleValue);
  __ Move(r6, ExternalReference::scheduled_exception_address(isolate));
  __ ldr(r5, MemOperand(r6));
  __ cmp(r4, r5);
  __ b(ne, &promote_scheduled_exception);

  __ mov(pc, lr);

  // Re-throw by promoting a scheduled exception.
  __ bind(&promote_scheduled_exception);
  __ TailCallRuntime(Runtime::kPromoteScheduledException);

  // HandleScope limit has changed. Delete allocated extensions.
  __ bind(&delete_allocated_handles);
  __ str(r5, MemOperand(r9, kLimitOffset));
  __ mov(r4, r0);
  __ PrepareCallCFunction(1);
  __ Move(r0, ExternalReference::isolate_address(isolate));
  __ CallCFunction(ExternalReference::delete_handle_scope_extensions(), 1);
  __ mov(r0, r4);
  __ jmp(&leave_exit_frame);
}

}  // namespace

void Builtins::Generate_CallApiCallback(MacroAssembler* masm) {
  // ----------- S t a t e -------------
  //  -- cp                  : context
  //  -- r1                  : api function address
  //  -- r2                  : arguments count (not including the receiver)
  //  -- r3                  : call data
  //  -- r0                  : holder
  //  -- sp[0]               : last argument
  //  -- ...
  //  -- sp[(argc - 1) * 4]  : first argument
  //  -- sp[(argc + 0) * 4]  : receiver
  // -----------------------------------

  Register api_function_address = r1;
  Register argc = r2;
  Register call_data = r3;
  Register holder = r0;
  Register scratch = r4;

  DCHECK(!AreAliased(api_function_address, argc, call_data, holder, scratch));

  using FCA = FunctionCallbackArguments;

  STATIC_ASSERT(FCA::kArgsLength == 6);
  STATIC_ASSERT(FCA::kNewTargetIndex == 5);
  STATIC_ASSERT(FCA::kDataIndex == 4);
  STATIC_ASSERT(FCA::kReturnValueOffset == 3);
  STATIC_ASSERT(FCA::kReturnValueDefaultValueIndex == 2);
  STATIC_ASSERT(FCA::kIsolateIndex == 1);
  STATIC_ASSERT(FCA::kHolderIndex == 0);

  // Set up FunctionCallbackInfo's implicit_args on the stack as follows:
  //
  // Target state:
  //   sp[0 * kPointerSize]: kHolder
  //   sp[1 * kPointerSize]: kIsolate
  //   sp[2 * kPointerSize]: undefined (kReturnValueDefaultValue)
  //   sp[3 * kPointerSize]: undefined (kReturnValue)
  //   sp[4 * kPointerSize]: kData
  //   sp[5 * kPointerSize]: undefined (kNewTarget)

  // Reserve space on the stack.
  __ AllocateStackSpace(FCA::kArgsLength * kPointerSize);

  // kHolder.
  __ str(holder, MemOperand(sp, 0 * kPointerSize));

  // kIsolate.
  __ Move(scratch, ExternalReference::isolate_address(masm->isolate()));
  __ str(scratch, MemOperand(sp, 1 * kPointerSize));

  // kReturnValueDefaultValue and kReturnValue.
  __ LoadRoot(scratch, RootIndex::kUndefinedValue);
  __ str(scratch, MemOperand(sp, 2 * kPointerSize));
  __ str(scratch, MemOperand(sp, 3 * kPointerSize));

  // kData.
  __ str(call_data, MemOperand(sp, 4 * kPointerSize));

  // kNewTarget.
  __ str(scratch, MemOperand(sp, 5 * kPointerSize));

  // Keep a pointer to kHolder (= implicit_args) in a scratch register.
  // We use it below to set up the FunctionCallbackInfo object.
  __ mov(scratch, sp);

  // Allocate the v8::Arguments structure in the arguments' space since
  // it's not controlled by GC.
  static constexpr int kApiStackSpace = 4;
  static constexpr bool kDontSaveDoubles = false;
  FrameScope frame_scope(masm, StackFrame::MANUAL);
  __ EnterExitFrame(kDontSaveDoubles, kApiStackSpace);

  // FunctionCallbackInfo::implicit_args_ (points at kHolder as set up above).
  // Arguments are after the return address (pushed by EnterExitFrame()).
  __ str(scratch, MemOperand(sp, 1 * kPointerSize));

  // FunctionCallbackInfo::values_ (points at the first varargs argument passed
  // on the stack).
  __ add(scratch, scratch, Operand((FCA::kArgsLength - 1) * kPointerSize));
  __ add(scratch, scratch, Operand(argc, LSL, kPointerSizeLog2));
  __ str(scratch, MemOperand(sp, 2 * kPointerSize));

  // FunctionCallbackInfo::length_.
  __ str(argc, MemOperand(sp, 3 * kPointerSize));

  // We also store the number of bytes to drop from the stack after returning
  // from the API function here.
  __ mov(scratch,
         Operand((FCA::kArgsLength + 1 /* receiver */) * kPointerSize));
  __ add(scratch, scratch, Operand(argc, LSL, kPointerSizeLog2));
  __ str(scratch, MemOperand(sp, 4 * kPointerSize));

  // v8::InvocationCallback's argument.
  __ add(r0, sp, Operand(1 * kPointerSize));

  ExternalReference thunk_ref = ExternalReference::invoke_function_callback();

  // There are two stack slots above the arguments we constructed on the stack.
  // TODO(jgruber): Document what these arguments are.
  static constexpr int kStackSlotsAboveFCA = 2;
  MemOperand return_value_operand(
      fp, (kStackSlotsAboveFCA + FCA::kReturnValueOffset) * kPointerSize);

  static constexpr int kUseStackSpaceOperand = 0;
  MemOperand stack_space_operand(sp, 4 * kPointerSize);

  AllowExternalCallThatCantCauseGC scope(masm);
  CallApiFunctionAndReturn(masm, api_function_address, thunk_ref,
                           kUseStackSpaceOperand, &stack_space_operand,
                           return_value_operand);
}

void Builtins::Generate_CallApiGetter(MacroAssembler* masm) {
  // Build v8::PropertyCallbackInfo::args_ array on the stack and push property
  // name below the exit frame to make GC aware of them.
  STATIC_ASSERT(PropertyCallbackArguments::kShouldThrowOnErrorIndex == 0);
  STATIC_ASSERT(PropertyCallbackArguments::kHolderIndex == 1);
  STATIC_ASSERT(PropertyCallbackArguments::kIsolateIndex == 2);
  STATIC_ASSERT(PropertyCallbackArguments::kReturnValueDefaultValueIndex == 3);
  STATIC_ASSERT(PropertyCallbackArguments::kReturnValueOffset == 4);
  STATIC_ASSERT(PropertyCallbackArguments::kDataIndex == 5);
  STATIC_ASSERT(PropertyCallbackArguments::kThisIndex == 6);
  STATIC_ASSERT(PropertyCallbackArguments::kArgsLength == 7);

  Register receiver = ApiGetterDescriptor::ReceiverRegister();
  Register holder = ApiGetterDescriptor::HolderRegister();
  Register callback = ApiGetterDescriptor::CallbackRegister();
  Register scratch = r4;
  DCHECK(!AreAliased(receiver, holder, callback, scratch));

  Register api_function_address = r2;

  __ push(receiver);
  // Push data from AccessorInfo.
  __ ldr(scratch, FieldMemOperand(callback, AccessorInfo::kDataOffset));
  __ push(scratch);
  __ LoadRoot(scratch, RootIndex::kUndefinedValue);
  __ Push(scratch, scratch);
  __ Move(scratch, ExternalReference::isolate_address(masm->isolate()));
  __ Push(scratch, holder);
  __ Push(Smi::zero());  // should_throw_on_error -> false
  __ ldr(scratch, FieldMemOperand(callback, AccessorInfo::kNameOffset));
  __ push(scratch);
  // v8::PropertyCallbackInfo::args_ array and name handle.
  const int kStackUnwindSpace = PropertyCallbackArguments::kArgsLength + 1;

  // Load address of v8::PropertyAccessorInfo::args_ array and name handle.
  __ mov(r0, sp);                             // r0 = Handle<Name>
  __ add(r1, r0, Operand(1 * kPointerSize));  // r1 = v8::PCI::args_

  const int kApiStackSpace = 1;
  FrameScope frame_scope(masm, StackFrame::MANUAL);
  __ EnterExitFrame(false, kApiStackSpace);

  // Create v8::PropertyCallbackInfo object on the stack and initialize
  // it's args_ field.
  __ str(r1, MemOperand(sp, 1 * kPointerSize));
  __ add(r1, sp, Operand(1 * kPointerSize));  // r1 = v8::PropertyCallbackInfo&

  ExternalReference thunk_ref =
      ExternalReference::invoke_accessor_getter_callback();

  __ ldr(scratch, FieldMemOperand(callback, AccessorInfo::kJsGetterOffset));
  __ ldr(api_function_address,
         FieldMemOperand(scratch, Foreign::kForeignAddressOffset));

  // +3 is to skip prolog, return address and name handle.
  MemOperand return_value_operand(
      fp, (PropertyCallbackArguments::kReturnValueOffset + 3) * kPointerSize);
  MemOperand* const kUseStackSpaceConstant = nullptr;
  CallApiFunctionAndReturn(masm, api_function_address, thunk_ref,
                           kStackUnwindSpace, kUseStackSpaceConstant,
                           return_value_operand);
}

void Builtins::Generate_DirectCEntry(MacroAssembler* masm) {
  // The sole purpose of DirectCEntry is for movable callers (e.g. any general
  // purpose Code object) to be able to call into C functions that may trigger
  // GC and thus move the caller.
  //
  // DirectCEntry places the return address on the stack (updated by the GC),
  // making the call GC safe. The irregexp backend relies on this.

  __ str(lr, MemOperand(sp, 0));  // Store the return address.
  __ blx(ip);                     // Call the C++ function.
  __ ldr(pc, MemOperand(sp, 0));  // Return to calling code.
}

void Builtins::Generate_MemCopyUint8Uint8(MacroAssembler* masm) {
  Register dest = r0;
  Register src = r1;
  Register chars = r2;
  Register temp1 = r3;
  Label less_4;

  {
    UseScratchRegisterScope temps(masm);
    Register temp2 = temps.Acquire();
    Label loop;

    __ bic(temp2, chars, Operand(0x3), SetCC);
    __ b(&less_4, eq);
    __ add(temp2, dest, temp2);

    __ bind(&loop);
    __ ldr(temp1, MemOperand(src, 4, PostIndex));
    __ str(temp1, MemOperand(dest, 4, PostIndex));
    __ cmp(dest, temp2);
    __ b(&loop, ne);
  }

  __ bind(&less_4);
  __ mov(chars, Operand(chars, LSL, 31), SetCC);
  // bit0 => Z (ne), bit1 => C (cs)
  __ ldrh(temp1, MemOperand(src, 2, PostIndex), cs);
  __ strh(temp1, MemOperand(dest, 2, PostIndex), cs);
  __ ldrb(temp1, MemOperand(src), ne);
  __ strb(temp1, MemOperand(dest), ne);
  __ Ret();
}

void Builtins::Generate_MemCopyUint16Uint8(MacroAssembler* masm) {
  Register dest = r0;
  Register src = r1;
  Register chars = r2;

  {
    UseScratchRegisterScope temps(masm);

    Register temp1 = r3;
    Register temp2 = temps.Acquire();
    Register temp3 = lr;
    Register temp4 = r4;
    Label loop;
    Label not_two;

    __ Push(lr, r4);
    __ bic(temp2, chars, Operand(0x3));
    __ add(temp2, dest, Operand(temp2, LSL, 1));

    __ bind(&loop);
    __ ldr(temp1, MemOperand(src, 4, PostIndex));
    __ uxtb16(temp3, temp1);
    __ uxtb16(temp4, temp1, 8);
    __ pkhbt(temp1, temp3, Operand(temp4, LSL, 16));
    __ str(temp1, MemOperand(dest));
    __ pkhtb(temp1, temp4, Operand(temp3, ASR, 16));
    __ str(temp1, MemOperand(dest, 4));
    __ add(dest, dest, Operand(8));
    __ cmp(dest, temp2);
    __ b(&loop, ne);

    __ mov(chars, Operand(chars, LSL, 31), SetCC);  // bit0 => ne, bit1 => cs
    __ b(&not_two, cc);
    __ ldrh(temp1, MemOperand(src, 2, PostIndex));
    __ uxtb(temp3, temp1, 8);
    __ mov(temp3, Operand(temp3, LSL, 16));
    __ uxtab(temp3, temp3, temp1);
    __ str(temp3, MemOperand(dest, 4, PostIndex));
    __ bind(&not_two);
    __ ldrb(temp1, MemOperand(src), ne);
    __ strh(temp1, MemOperand(dest), ne);
    __ Pop(pc, r4);
  }
}

#undef __

}  // namespace internal
}  // namespace v8

#endif  // V8_TARGET_ARCH_ARM