dart-lang/sdk
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity
Files
57b27a7b44a9112c7cb3a1cb3c73636fc149bd16
sdk/runtime/vm/compiler/stub_code_compiler_x64.cc
T
Tony 57b27a7b44 Refactor Dart runtime to replace DART_DYNAMIC_MODULES with DART_BYTECODE_INTERPRETER 
- Updated conditional compilation flags throughout the runtime codebase to transition from DART_DYNAMIC_MODULES to DART_BYTECODE_INTERPRETER.
- Adjusted logic in various files including object_graph_copy.cc, object_reload.cc, profiler.cc, and others to ensure compatibility with the new interpreter model.
- Ensured that all references to dynamic modules are replaced with bytecode interpreter checks, maintaining functionality for interpreted code execution.
- Modified stack frame handling and service-related code to align with the new interpreter architecture.
- Updated tests and service implementations to reflect the changes in the runtime environment.

Signed-off-by: Tony <tonylu@tony-cloud.com>
2026-06-25 01:58:41 +08:00

3884 lines
140 KiB
C++
Raw Blame History

// Copyright (c) 2019, the Dart project authors. Please see the AUTHORS file
// for details. All rights reserved. Use of this source code is governed by a
// BSD-style license that can be found in the LICENSE file.
#include <setjmp.h>
#include "vm/compiler/runtime_api.h"
#include "vm/globals.h"
// For `AllocateObjectInstr::WillAllocateNewOrRemembered`
// For `GenericCheckBoundInstr::UseUnboxedRepresentation`
#include "vm/compiler/backend/il.h"
#define SHOULD_NOT_INCLUDE_RUNTIME
#include "vm/compiler/backend/locations.h"
#include "vm/compiler/stub_code_compiler.h"
#if defined(TARGET_ARCH_X64)
#include "vm/class_id.h"
#include "vm/code_entry_kind.h"
#include "vm/compiler/api/type_check_mode.h"
#include "vm/compiler/assembler/assembler.h"
#include "vm/constants.h"
#include "vm/ffi_callback_metadata.h"
#include "vm/instructions.h"
#include "vm/static_type_exactness_state.h"
#include "vm/tags.h"
#define __ assembler->
namespace dart {
#ifdef DART_TARGET_SUPPORTS_PROBE_POINTS
DECLARE_FLAG(bool, generate_probe_points);
#endif
namespace compiler {
// Ensures that [RAX] is a new object, if not it will be added to the remembered
// set via a leaf runtime call.
//
// WARNING: This might clobber all registers except for [RAX], [THR] and [FP].
// The caller should simply call LeaveStubFrame() and return.
void StubCodeCompiler::EnsureIsNewOrRemembered() {
// If the object is not in an active TLAB, we call a leaf-runtime to add it to
// the remembered set and/or deferred marking worklist. This test assumes a
// Page's TLAB use is always ascending.
Label done;
__ AndImmediate(TMP, RAX, target::Page::kPageMask);
__ LoadFromOffset(TMP, TMP, target::Page::original_top_offset());
__ CompareRegisters(RAX, TMP);
__ BranchIf(UNSIGNED_GREATER_EQUAL, &done);
{
LeafRuntimeScope rt(assembler, /*frame_size=*/0,
/*preserve_registers=*/false);
__ movq(CallingConventions::kArg1Reg, RAX);
__ movq(CallingConventions::kArg2Reg, THR);
rt.Call(kEnsureRememberedAndMarkingDeferredRuntimeEntry, 2);
}
__ Bind(&done);
}
// In TSAN mode the runtime will throw an exception using an intermediary
// longjmp() call to unwind the C frames in a way that TSAN can understand.
//
// This wrapper will setup a [jmp_buf] on the stack and initialize it to be a
// target for a possible longjmp(). In the exceptional case we'll forward
// control of execution to the usual JumpToFrame stub.
//
// In non-TSAN mode this will do nothing and the runtime will call the
// JumpToFrame stub directly.
//
// The callback [fun] may be invoked with a modified [RSP] due to allocating
// a [jmp_buf] allocating structure on the stack (as well as the saved old
// [Thread::tsan_utils_->setjmp_buffer_]).
static void WithExceptionCatchingTrampoline(Assembler* assembler,
std::function<void()> fun) {
#if !defined(DART_INCLUDE_SIMULATOR)
const Register kTsanUtilsReg = RAX;
// Reserve space for arguments and align frame before entering C++ world.
const intptr_t kJumpBufferSize = sizeof(jmp_buf);
// Save & Restore the volatile CPU registers across the setjmp() call.
const RegisterSet volatile_registers(
CallingConventions::kVolatileCpuRegisters & ~(1 << RAX),
/*fpu_register_mask=*/0);
const Register kSavedRspReg = R12;
COMPILE_ASSERT(IsCalleeSavedRegister(kSavedRspReg));
// We rely on THR being preserved across the setjmp() call.
COMPILE_ASSERT(IsCalleeSavedRegister(THR));
if (FLAG_target_thread_sanitizer) {
Label do_native_call;
// Save old jmp_buf.
__ movq(kTsanUtilsReg, Address(THR, target::Thread::tsan_utils_offset()));
__ pushq(Address(kTsanUtilsReg, target::TsanUtils::setjmp_buffer_offset()));
// Allocate jmp_buf struct on stack & remember pointer to it on the
// [Thread::tsan_utils_->setjmp_buffer] (which exceptions.cc will longjmp()
// to)
__ AddImmediate(RSP, Immediate(-kJumpBufferSize));
__ movq(Address(kTsanUtilsReg, target::TsanUtils::setjmp_buffer_offset()),
RSP);
// Call setjmp() with a pointer to the allocated jmp_buf struct.
__ MoveRegister(CallingConventions::kArg1Reg, RSP);
__ PushRegisters(volatile_registers);
if (OS::ActivationFrameAlignment() > 1) {
__ MoveRegister(kSavedRspReg, RSP);
__ andq(RSP, Immediate(~(OS::ActivationFrameAlignment() - 1)));
}
__ movq(kTsanUtilsReg, Address(THR, target::Thread::tsan_utils_offset()));
__ CallCFunction(
Address(kTsanUtilsReg, target::TsanUtils::setjmp_function_offset()),
/*restore_rsp=*/true);
if (OS::ActivationFrameAlignment() > 1) {
__ MoveRegister(RSP, kSavedRspReg);
}
__ PopRegisters(volatile_registers);
// We are the target of a longjmp() iff setjmp() returns non-0.
__ CompareImmediate(RAX, 0);
__ BranchIf(EQUAL, &do_native_call);
// We are the target of a longjmp: Cleanup the stack and tail-call the
// JumpToFrame stub which will take care of unwinding the stack and hand
// execution to the catch entry.
__ AddImmediate(RSP, Immediate(kJumpBufferSize));
__ movq(kTsanUtilsReg, Address(THR, target::Thread::tsan_utils_offset()));
__ popq(Address(kTsanUtilsReg, target::TsanUtils::setjmp_buffer_offset()));
__ movq(CallingConventions::kArg1Reg,
Address(kTsanUtilsReg, target::TsanUtils::exception_pc_offset()));
__ movq(CallingConventions::kArg2Reg,
Address(kTsanUtilsReg, target::TsanUtils::exception_sp_offset()));
__ movq(CallingConventions::kArg3Reg,
Address(kTsanUtilsReg, target::TsanUtils::exception_fp_offset()));
__ MoveRegister(CallingConventions::kArg4Reg, THR);
__ jmp(Address(THR, target::Thread::jump_to_frame_entry_point_offset()));
// We leave the created [jump_buf] structure on the stack as well as the
// pushed old [Thread::tsan_utils_->setjmp_buffer_].
__ Bind(&do_native_call);
__ MoveRegister(kSavedRspReg, RSP);
}
#endif // !defined(DART_INCLUDE_SIMULATOR)
fun();
#if !defined(DART_INCLUDE_SIMULATOR)
if (FLAG_target_thread_sanitizer) {
__ MoveRegister(RSP, kSavedRspReg);
__ AddImmediate(RSP, Immediate(kJumpBufferSize));
const Register kTsanUtilsReg2 = kSavedRspReg;
__ movq(kTsanUtilsReg2, Address(THR, target::Thread::tsan_utils_offset()));
__ popq(Address(kTsanUtilsReg2, target::TsanUtils::setjmp_buffer_offset()));
}
#endif // !defined(DART_INCLUDE_SIMULATOR)
}
// Input parameters:
// RSP : points to return address.
// RSP + 8 : address of last argument in argument array.
// RSP + 8*R10 : address of first argument in argument array.
// RSP + 8*R10 + 8 : address of return value.
// RBX : address of the runtime function to call.
// R10 : number of arguments to the call.
// Must preserve callee saved registers R12 and R13.
void StubCodeCompiler::GenerateCallToRuntimeStub() {
const intptr_t thread_offset = target::NativeArguments::thread_offset();
const intptr_t argc_tag_offset = target::NativeArguments::argc_tag_offset();
const intptr_t argv_offset = target::NativeArguments::argv_offset();
const intptr_t retval_offset = target::NativeArguments::retval_offset();
__ movq(CODE_REG,
Address(THR, target::Thread::call_to_runtime_stub_offset()));
__ EnterStubFrame();
// Save exit frame information to enable stack walking as we are about
// to transition to Dart VM C++ code.
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()), RBP);
// Mark that the thread exited generated code through a runtime call.
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(target::Thread::exit_through_runtime_call()));
#if defined(DEBUG)
{
Label ok;
// Check that we are always entering from Dart code.
__ movq(RAX, Immediate(VMTag::kDartTagId));
__ cmpq(RAX, Assembler::VMTagAddress());
__ j(EQUAL, &ok, Assembler::kNearJump);
__ Stop("Not coming from Dart code.");
__ Bind(&ok);
}
#endif
// Mark that the thread is executing VM code.
__ movq(Assembler::VMTagAddress(), RBX);
WithExceptionCatchingTrampoline(assembler, [&]() {
// Reserve space for arguments and align frame before entering C++ world.
__ subq(RSP, Immediate(target::NativeArguments::StructSize()));
if (OS::ActivationFrameAlignment() > 1) {
__ andq(RSP, Immediate(~(OS::ActivationFrameAlignment() - 1)));
}
// Pass target::NativeArguments structure by value and call runtime.
__ movq(Address(RSP, thread_offset), THR); // Set thread in NativeArgs.
__ movq(Address(RSP, argc_tag_offset),
R10); // Set argc in target::NativeArguments.
// Compute argv.
__ leaq(RAX, Address(RBP, R10, TIMES_8,
target::frame_layout.param_end_from_fp *
target::kWordSize));
__ movq(Address(RSP, argv_offset),
RAX); // Set argv in target::NativeArguments.
__ addq(
RAX,
Immediate(1 * target::kWordSize)); // Retval is next to 1st argument.
__ movq(Address(RSP, retval_offset),
RAX); // Set retval in target::NativeArguments.
#if defined(DART_TARGET_OS_WINDOWS)
ASSERT(target::NativeArguments::StructSize() >
CallingConventions::kRegisterTransferLimit);
__ movq(CallingConventions::kArg1Reg, RSP);
#endif
__ CallCFunction(RBX);
// Mark that the thread is executing Dart code.
__ movq(Assembler::VMTagAddress(), Immediate(VMTag::kDartTagId));
// Mark that the thread has not exited generated Dart code.
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(0));
// Reset exit frame information in Isolate's mutator thread structure.
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()),
Immediate(0));
// Restore the global object pool after returning from runtime (old space is
// moving, so the GOP could have been relocated).
if (FLAG_precompiled_mode) {
__ movq(PP, Address(THR, target::Thread::global_object_pool_offset()));
}
});
__ LeaveStubFrame();
// The following return can jump to a lazy-deopt stub, which assumes RAX
// contains a return value and will save it in a GC-visible way. We therefore
// have to ensure RAX does not contain any garbage value left from the C
// function we called (which has return type "void").
// (See GenerateDeoptimizationSequence::saved_result_slot_from_fp.)
__ xorq(RAX, RAX);
__ ret();
}
void StubCodeCompiler::GenerateSharedStubGeneric(
bool save_fpu_registers,
intptr_t self_code_stub_offset_from_thread,
bool allow_return,
std::function<void()> perform_runtime_call) {
// We want the saved registers to appear like part of the caller's frame, so
// we push them before calling EnterStubFrame.
const RegisterSet saved_registers(
kDartAvailableCpuRegs, save_fpu_registers ? kAllFpuRegistersList : 0);
__ PushRegisters(saved_registers);
const intptr_t kSavedCpuRegisterSlots =
Utils::CountOneBitsWord(kDartAvailableCpuRegs);
const intptr_t kSavedFpuRegisterSlots =
save_fpu_registers
? kNumberOfFpuRegisters * kFpuRegisterSize / target::kWordSize
: 0;
const intptr_t kAllSavedRegistersSlots =
kSavedCpuRegisterSlots + kSavedFpuRegisterSlots;
// Copy down the return address so the stack layout is correct.
__ pushq(Address(RSP, kAllSavedRegistersSlots * target::kWordSize));
__ movq(CODE_REG, Address(THR, self_code_stub_offset_from_thread));
__ EnterStubFrame();
perform_runtime_call();
if (!allow_return) {
__ Breakpoint();
return;
}
__ LeaveStubFrame();
// Copy up the return address (in case it was changed).
__ popq(TMP);
__ movq(Address(RSP, kAllSavedRegistersSlots * target::kWordSize), TMP);
__ PopRegisters(saved_registers);
__ ret();
}
void StubCodeCompiler::GenerateSharedStub(
bool save_fpu_registers,
const RuntimeEntry* target,
intptr_t self_code_stub_offset_from_thread,
bool allow_return,
bool store_runtime_result_in_result_register) {
auto perform_runtime_call = [&]() {
if (store_runtime_result_in_result_register) {
__ PushImmediate(Immediate(0));
}
__ CallRuntime(*target, /*argument_count=*/0);
if (store_runtime_result_in_result_register) {
__ PopRegister(RAX);
__ movq(Address(RBP, target::kWordSize *
StubCodeCompiler::WordOffsetFromFpToCpuRegister(
SharedSlowPathStubABI::kResultReg)),
RAX);
}
};
GenerateSharedStubGeneric(save_fpu_registers,
self_code_stub_offset_from_thread, allow_return,
perform_runtime_call);
}
void StubCodeCompiler::GenerateEnterSafepointStub() {
RegisterSet all_registers;
all_registers.AddAllGeneralRegisters();
__ PushRegisters(all_registers);
__ EnterFrame(0);
__ ReserveAlignedFrameSpace(0);
__ movq(RAX, Address(THR, kEnterSafepointRuntimeEntry.OffsetFromThread()));
__ CallCFunction(RAX);
__ LeaveFrame();
__ PopRegisters(all_registers);
__ ret();
}
void StubCodeCompiler::GenerateExitSafepointStub() {
RegisterSet all_registers;
all_registers.AddAllGeneralRegisters();
__ PushRegisters(all_registers);
__ EnterFrame(0);
__ ReserveAlignedFrameSpace(0);
__ VerifyNotInGenerated(RAX);
__ movq(RAX, Address(THR, kExitSafepointRuntimeEntry.OffsetFromThread()));
__ CallCFunction(RAX);
__ LeaveFrame();
__ PopRegisters(all_registers);
__ ret();
}
// Calls native code within a safepoint.
//
// On entry:
// Stack: arguments set up and aligned for native call, excl. shadow space
// RBX = target address to call
//
// On exit:
// Stack pointer lowered by shadow space
// RBX, R12 clobbered
void StubCodeCompiler::GenerateCallNativeThroughSafepointStub() {
__ movq(R12, compiler::Immediate(target::Thread::exit_through_ffi()));
__ TransitionGeneratedToNative(RBX, FPREG, R12,
/*enter_safepoint=*/true);
__ popq(R12);
__ CallCFunction(RBX, /*restore_rsp=*/true);
__ TransitionNativeToGenerated(/*exit_safepoint=*/true);
// Faster than jmp because it doesn't confuse the branch predictor.
__ pushq(R12);
__ ret();
}
void StubCodeCompiler::GenerateFfiCallTrampolineStub() {
__ EnterFrame(0);
__ PushRegister(CallingConventions::kArg1Reg);
const Register args = TMP;
const Register src = RSI;
const Register dst = RDI;
const Register size = RCX;
#if defined(DART_TARGET_OS_WINDOWS)
__ PushRegistersInOrder({src, dst});
#endif
__ movq(args, CallingConventions::kArg1Reg); // FfiCallArguments*
__ movq(src, Address(args, 0 * target::kWordSize)); // stack_area
__ movq(size, Address(args, 1 * target::kWordSize)); // stack_area_end
__ subq(size, src); // size = stack_area_end - stack_area
__ subq(RSP, size);
if (OS::ActivationFrameAlignment() > 1) {
__ andq(RSP, Immediate(~(OS::ActivationFrameAlignment() - 1)));
}
__ movq(dst, RSP);
__ CopyMemoryWords(src, dst, size, RAX);
for (intptr_t i = 0; i < CallingConventions::kNumArgRegs; ++i) {
Register reg = CallingConventions::ArgumentRegisters[i];
ASSERT(reg != args);
__ movq(reg, Address(args, (2 + reg) * target::kWordSize));
}
if (CallingConventions::kVarArgFpuRegisterCount != kNoRegister) {
Register reg = CallingConventions::kVarArgFpuRegisterCount;
ASSERT(reg != args);
__ movq(reg, Address(args, (2 + reg) * target::kWordSize));
}
for (intptr_t i = 0; i < CallingConventions::kNumFpuArgRegs; ++i) {
XmmRegister reg = CallingConventions::FpuArgumentRegisters[i];
__ movsd(reg, Address(args, (2 + kNumberOfCpuRegisters + reg) *
target::kWordSize));
}
__ call(Address(args, (2 + kNumberOfCpuRegisters + kNumberOfFpuRegisters) *
target::kWordSize));
__ movq(args, Address(RBP, -8));
{
const Register reg = CallingConventions::kReturnReg;
ASSERT(reg != args);
__ movq(Address(args, (2 + reg) * target::kWordSize), reg);
}
{
const Register reg = CallingConventions::kSecondReturnReg;
ASSERT(reg != args);
__ movq(Address(args, (2 + reg) * target::kWordSize), reg);
}
{
const XmmRegister reg = CallingConventions::kReturnFpuReg;
__ movsd(
Address(args, (2 + kNumberOfCpuRegisters + reg) * target::kWordSize),
reg);
}
#if !defined(DART_TARGET_OS_WINDOWS)
{
const XmmRegister reg = CallingConventions::kSecondReturnFpuReg;
__ movsd(
Address(args, (2 + kNumberOfCpuRegisters + reg) * target::kWordSize),
reg);
}
#endif
#if defined(DART_TARGET_OS_WINDOWS)
__ leaq(RSP, Address(RBP, -24));
__ popq(dst);
__ popq(src);
#endif
__ LeaveFrame();
__ Ret();
}
void StubCodeCompiler::GenerateLoadFfiCallbackMetadataRuntimeFunction(
uword function_index,
Register dst) {
// Keep in sync with FfiCallbackMetadata::EnsureFirstTrampolinePageLocked.
// Note: If the stub was aligned, this could be a single PC relative load.
// Load a pointer to the beginning of the stub into dst.
const intptr_t kLeaqLength = 7;
const intptr_t code_size = __ CodeSize();
__ leaq(dst, Address::AddressRIPRelative(-kLeaqLength - code_size));
// Round dst down to the page size.
__ andq(dst, Immediate(FfiCallbackMetadata::kPageMask));
// Load the function from the function table.
__ LoadFromOffset(dst, dst,
FfiCallbackMetadata::RuntimeFunctionOffset(function_index));
}
static const RegisterSet kArgumentRegisterSet(
CallingConventions::kArgumentRegisters,
CallingConventions::kFpuArgumentRegisters);
void StubCodeCompiler::GenerateFfiCallbackTrampolineStub() {
// RAX is volatile and not used for passing any arguments.
COMPILE_ASSERT(!IsCalleeSavedRegister(RAX) && !IsArgumentRegister(RAX));
Label body;
for (intptr_t i = 0; i < FfiCallbackMetadata::NumCallbackTrampolinesPerPage();
++i) {
// The FfiCallbackMetadata table is keyed by the trampoline entry point. So
// look up the current PC, then jump to the shared section. RIP gives us the
// address of the next instruction, so to get the true entry point, we have
// to subtract the size of the leaq instruction.
const intptr_t kLeaqLength = 7;
const intptr_t size_before = __ CodeSize();
__ leaq(RAX, Address::AddressRIPRelative(-kLeaqLength));
const intptr_t size_after = __ CodeSize();
ASSERT_EQUAL(size_after - size_before, kLeaqLength);
__ jmp(&body);
}
ASSERT_EQUAL(__ CodeSize(),
FfiCallbackMetadata::kNativeCallbackTrampolineSize *
FfiCallbackMetadata::NumCallbackTrampolinesPerPage());
__ Bind(&body);
const intptr_t shared_stub_start = __ CodeSize();
// Save THR which is callee-saved.
__ EnterFrame(0);
__ pushq(THR);
__ pushq(R12);
// 4 = return address, FP, THR, R12
COMPILE_ASSERT(4 == FfiCallbackMetadata::kNativeCallbackTrampolineStackDelta);
// Load the thread, verify the callback ID and exit the safepoint.
//
// We exit the safepoint inside DLRT_GetFfiCallbackMetadata in order to save
// code size of this shared stub.
{
// Save all registers which might hold arguments.
__ PushRegistersAligned(kArgumentRegisterSet, 3 * target::kWordSize);
COMPILE_ASSERT(RAX != CallingConventions::kArg1Reg);
__ movq(CallingConventions::kArg1Reg, RAX);
__ movq(CallingConventions::kArg2Reg, RSP);
GenerateLoadFfiCallbackMetadataRuntimeFunction(
FfiCallbackMetadata::kGetFfiCallbackMetadata, RAX);
__ CallCFunction(RAX, /*restore_rsp=*/true);
__ movq(THR, RAX);
__ movq(RAX, Address(RSP, 0 * target::kWordSize)); // entry_point
__ movq(TMP, Address(RSP, 1 * target::kWordSize)); // is_tail
__ movq(R12, Address(RSP, 2 * target::kWordSize)); // epilogue
// Restore the arguments.
__ PopRegistersAligned(kArgumentRegisterSet, 3 * target::kWordSize);
}
Label tail;
__ cmpq(TMP, Immediate(0));
__ j(NOT_ZERO, &tail);
const RegisterSet return_registers(
(1 << CallingConventions::kReturnReg) |
(1 << CallingConventions::kSecondReturnReg),
(1 << CallingConventions::kReturnFpuReg) |
(1 << CallingConventions::kSecondReturnFpuReg));
{
__ call(RAX); // entry_point
__ PushRegistersAligned(return_registers, 0);
__ movq(CallingConventions::kArg1Reg, THR);
__ CallCFunction(R12, /*restore_rsp=*/true); // DLRT_ExitSyncCallback, etc
if (FLAG_target_memory_sanitizer) {
__ CallCFunction(RAX, /*restore_rsp=*/true); // dart_msan_unpoison_retval
}
__ PopRegistersAligned(return_registers, 0);
__ popq(R12);
__ popq(THR);
__ LeaveFrame();
__ ret();
}
{
__ Bind(&tail);
__ call(RAX); // entry_point
__ movq(CallingConventions::kArg1Reg, THR);
__ movq(RAX, R12);
__ popq(R12);
__ popq(THR);
__ LeaveFrame();
// Tail-call DLRT_ExitTemporaryIsolate. It is not safe to return to this
// stub, since it might be deleted once DLRT_ExitTemporaryIsolate proceeds
// enough for VM shutdown.
__ jmp(RAX); // DLRT_ExitTemporaryIsolate
__ int3();
}
// 'kNativeCallbackSharedStubSize' is an upper bound because the exact
// instruction size can vary slightly based on OS calling conventions.
ASSERT_LESS_OR_EQUAL(__ CodeSize() - shared_stub_start,
FfiCallbackMetadata::kNativeCallbackSharedStubSize);
ASSERT_LESS_OR_EQUAL(__ CodeSize(), FfiCallbackMetadata::kPageSize);
#if defined(DEBUG)
while (__ CodeSize() < FfiCallbackMetadata::kPageSize) {
__ Breakpoint();
}
#endif
}
void StubCodeCompiler::GenerateDispatchTableNullErrorStub() {
__ EnterStubFrame();
__ SmiTag(DispatchTableNullErrorABI::kClassIdReg);
__ PushRegister(DispatchTableNullErrorABI::kClassIdReg);
__ CallRuntime(kDispatchTableNullErrorRuntimeEntry, /*argument_count=*/1);
// The NullError runtime entry does not return.
__ Breakpoint();
}
void StubCodeCompiler::GenerateRangeError(bool with_fpu_regs) {
auto perform_runtime_call = [&]() {
// If the generated code has unboxed index/length we need to box them before
// calling the runtime entry.
if (GenericCheckBoundInstr::UseUnboxedRepresentation()) {
Label length, smi_case;
// The user-controlled index might not fit into a Smi.
#if !defined(DART_COMPRESSED_POINTERS)
__ addq(RangeErrorABI::kIndexReg, RangeErrorABI::kIndexReg);
__ BranchIf(NO_OVERFLOW, &length);
#else
__ movq(TMP, RangeErrorABI::kIndexReg);
__ SmiTag(RangeErrorABI::kIndexReg);
__ sarq(TMP, Immediate(30));
__ addq(TMP, Immediate(1));
__ cmpq(TMP, Immediate(2));
__ j(BELOW, &length);
#endif
{
// Allocate a mint, reload the two registers and populate the mint.
__ PushImmediate(Immediate(0));
__ CallRuntime(kAllocateMintRuntimeEntry, /*argument_count=*/0);
__ PopRegister(RangeErrorABI::kIndexReg);
__ movq(
TMP,
Address(RBP, target::kWordSize *
StubCodeCompiler::WordOffsetFromFpToCpuRegister(
RangeErrorABI::kIndexReg)));
__ movq(FieldAddress(RangeErrorABI::kIndexReg,
target::Mint::value_offset()),
TMP);
__ movq(
RangeErrorABI::kLengthReg,
Address(RBP, target::kWordSize *
StubCodeCompiler::WordOffsetFromFpToCpuRegister(
RangeErrorABI::kLengthReg)));
}
// Length is guaranteed to be in positive Smi range (it comes from a load
// of a vm recognized array).
__ Bind(&length);
__ SmiTag(RangeErrorABI::kLengthReg);
}
__ PushRegistersInOrder(
{RangeErrorABI::kLengthReg, RangeErrorABI::kIndexReg});
__ CallRuntime(kRangeErrorRuntimeEntry, /*argument_count=*/2);
__ Breakpoint();
};
GenerateSharedStubGeneric(
/*save_fpu_registers=*/with_fpu_regs,
with_fpu_regs
? target::Thread::range_error_shared_with_fpu_regs_stub_offset()
: target::Thread::range_error_shared_without_fpu_regs_stub_offset(),
/*allow_return=*/false, perform_runtime_call);
}
void StubCodeCompiler::GenerateWriteError(bool with_fpu_regs) {
auto perform_runtime_call = [&]() {
__ CallRuntime(kWriteErrorRuntimeEntry, /*argument_count=*/2);
__ Breakpoint();
};
GenerateSharedStubGeneric(
/*save_fpu_registers=*/with_fpu_regs,
with_fpu_regs
? target::Thread::write_error_shared_with_fpu_regs_stub_offset()
: target::Thread::write_error_shared_without_fpu_regs_stub_offset(),
/*allow_return=*/false, perform_runtime_call);
}
// Input parameters:
// RSP : points to return address.
// RSP + 8 : address of return value.
// R13 : address of first argument in argument array.
// RBX : address of the native function to call.
// R10 : argc_tag including number of arguments and function kind.
static void GenerateCallNativeWithWrapperStub(Assembler* assembler,
Address wrapper_address) {
const intptr_t native_args_struct_offset = 0;
const intptr_t thread_offset =
target::NativeArguments::thread_offset() + native_args_struct_offset;
const intptr_t argc_tag_offset =
target::NativeArguments::argc_tag_offset() + native_args_struct_offset;
const intptr_t argv_offset =
target::NativeArguments::argv_offset() + native_args_struct_offset;
const intptr_t retval_offset =
target::NativeArguments::retval_offset() + native_args_struct_offset;
__ EnterStubFrame();
// Save exit frame information to enable stack walking as we are about
// to transition to native code.
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()), RBP);
// Mark that the thread exited generated code through a runtime call.
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(target::Thread::exit_through_runtime_call()));
#if defined(DEBUG)
{
Label ok;
// Check that we are always entering from Dart code.
__ movq(R8, Immediate(VMTag::kDartTagId));
__ cmpq(R8, Assembler::VMTagAddress());
__ j(EQUAL, &ok, Assembler::kNearJump);
__ Stop("Not coming from Dart code.");
__ Bind(&ok);
}
#endif
// Mark that the thread is executing native code.
__ movq(Assembler::VMTagAddress(), RBX);
WithExceptionCatchingTrampoline(assembler, [&]() {
// Reserve space for the native arguments structure passed on the stack (the
// outgoing pointer parameter to the native arguments structure is passed in
// RDI) and align frame before entering the C++ world.
__ subq(RSP, Immediate(target::NativeArguments::StructSize()));
if (OS::ActivationFrameAlignment() > 1) {
__ andq(RSP, Immediate(~(OS::ActivationFrameAlignment() - 1)));
}
// Pass target::NativeArguments structure by value and call native function.
// Set thread in NativeArgs.
__ movq(Address(RSP, thread_offset), THR);
// Set argc in target::NativeArguments.
__ movq(Address(RSP, argc_tag_offset), R10);
// Set argv in target::NativeArguments.
__ movq(Address(RSP, argv_offset), R13);
// Compute return value addr.
__ leaq(RAX, Address(RBP, (target::frame_layout.param_end_from_fp + 1) *
target::kWordSize));
// Set retval in target::NativeArguments.
__ movq(Address(RSP, retval_offset), RAX);
// Pass the pointer to the target::NativeArguments.
__ movq(CallingConventions::kArg1Reg, RSP);
// Pass pointer to function entrypoint.
__ movq(CallingConventions::kArg2Reg, RBX);
__ movq(RAX, wrapper_address);
__ CallCFunction(RAX);
// Mark that the thread is executing Dart code.
__ movq(Assembler::VMTagAddress(), Immediate(VMTag::kDartTagId));
// Mark that the thread has not exited generated Dart code.
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(0));
// Reset exit frame information in Isolate's mutator thread structure.
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()),
Immediate(0));
// Restore the global object pool after returning from runtime (old space is
// moving, so the GOP could have been relocated).
if (FLAG_precompiled_mode) {
__ movq(PP, Address(THR, target::Thread::global_object_pool_offset()));
}
});
__ LeaveStubFrame();
__ ret();
}
void StubCodeCompiler::GenerateCallNoScopeNativeStub() {
GenerateCallNativeWithWrapperStub(
assembler,
Address(THR,
target::Thread::no_scope_native_wrapper_entry_point_offset()));
}
void StubCodeCompiler::GenerateCallAutoScopeNativeStub() {
GenerateCallNativeWithWrapperStub(
assembler,
Address(THR,
target::Thread::auto_scope_native_wrapper_entry_point_offset()));
}
// Input parameters:
// RSP : points to return address.
// RSP + 8 : address of return value.
// RAX : address of first argument in argument array.
// RBX : address of the native function to call.
// R10 : argc_tag including number of arguments and function kind.
void StubCodeCompiler::GenerateCallBootstrapNativeStub() {
GenerateCallNativeWithWrapperStub(
assembler,
Address(THR,
target::Thread::bootstrap_native_wrapper_entry_point_offset()));
}
// Input parameters:
// ARGS_DESC_REG: arguments descriptor array.
void StubCodeCompiler::GenerateCallStaticFunctionStub() {
__ EnterStubFrame();
__ pushq(ARGS_DESC_REG); // Preserve arguments descriptor array.
// Setup space on stack for return value.
__ pushq(Immediate(0));
__ CallRuntime(kPatchStaticCallRuntimeEntry, 0);
#if defined(DART_PRECOMPILED_RUNTIME) && defined(DART_SHOREBIRD_INTERPRETER)
__ popq(FUNCTION_REG); // Get Function object result.
__ popq(ARGS_DESC_REG); // Restore arguments descriptor array.
// Remove the stub frame as we are about to jump to the dart function.
__ LeaveStubFrame();
__ LoadCompressed(
CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
__ jmp(FieldAddress(FUNCTION_REG, target::Function::entry_point_offset()));
#else
__ popq(CODE_REG); // Get Code object result.
__ popq(ARGS_DESC_REG); // Restore arguments descriptor array.
// Remove the stub frame as we are about to jump to the dart function.
__ LeaveStubFrame();
__ movq(RBX, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
__ jmp(RBX);
#endif
}
// Called from a static call only when an invalid code has been entered
// (invalid because its function was optimized or deoptimized).
// ARGS_DESC_REG: arguments descriptor array.
void StubCodeCompiler::GenerateFixCallersTargetStub() {
Label monomorphic;
__ BranchOnMonomorphicCheckedEntryJIT(&monomorphic);
// This was a static call.
// Load code pointer to this stub from the thread:
// The one that is passed in, is not correct - it points to the code object
// that needs to be replaced.
__ movq(CODE_REG,
Address(THR, target::Thread::fix_callers_target_code_offset()));
__ EnterStubFrame();
__ pushq(ARGS_DESC_REG); // Preserve arguments descriptor array.
// Setup space on stack for return value.
__ pushq(Immediate(0));
__ CallRuntime(kFixCallersTargetRuntimeEntry, 0);
__ popq(CODE_REG); // Get Code object.
__ popq(ARGS_DESC_REG); // Restore arguments descriptor array.
__ movq(RAX, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
__ LeaveStubFrame();
__ jmp(RAX);
__ int3();
__ Bind(&monomorphic);
// This was a switchable call.
// Load code pointer to this stub from the thread:
// The one that is passed in, is not correct - it points to the code object
// that needs to be replaced.
__ movq(CODE_REG,
Address(THR, target::Thread::fix_callers_target_code_offset()));
__ EnterStubFrame();
__ pushq(Immediate(0)); // Result slot.
__ pushq(RDX); // Preserve receiver.
__ pushq(RBX); // Old cache value (also 2nd return value).
__ CallRuntime(kFixCallersTargetMonomorphicRuntimeEntry, 2);
__ popq(RBX); // Get target cache object.
__ popq(RDX); // Restore receiver.
__ popq(CODE_REG); // Get target Code object.
__ movq(RAX, FieldAddress(CODE_REG, target::Code::entry_point_offset(
CodeEntryKind::kMonomorphic)));
__ LeaveStubFrame();
__ jmp(RAX);
__ int3();
}
// Called from object allocate instruction when the allocation stub has been
// disabled.
void StubCodeCompiler::GenerateFixAllocationStubTargetStub() {
// Load code pointer to this stub from the thread:
// The one that is passed in, is not correct - it points to the code object
// that needs to be replaced.
__ movq(CODE_REG,
Address(THR, target::Thread::fix_allocation_stub_code_offset()));
__ EnterStubFrame();
// Setup space on stack for return value.
__ pushq(Immediate(0));
__ CallRuntime(kFixAllocationStubTargetRuntimeEntry, 0);
__ popq(CODE_REG); // Get Code object.
__ movq(RAX, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
__ LeaveStubFrame();
__ jmp(RAX);
__ int3();
}
// Called from object allocate instruction when the allocation stub for a
// generic class has been disabled.
void StubCodeCompiler::GenerateFixParameterizedAllocationStubTargetStub() {
// Load code pointer to this stub from the thread:
// The one that is passed in, is not correct - it points to the code object
// that needs to be replaced.
__ movq(CODE_REG,
Address(THR, target::Thread::fix_allocation_stub_code_offset()));
__ EnterStubFrame();
// Setup space on stack for return value.
__ pushq(AllocateObjectABI::kTypeArgumentsReg);
__ pushq(Immediate(0));
__ CallRuntime(kFixAllocationStubTargetRuntimeEntry, 0);
__ popq(CODE_REG); // Get Code object.
__ popq(AllocateObjectABI::kTypeArgumentsReg);
__ movq(RAX, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
__ LeaveStubFrame();
__ jmp(RAX);
__ int3();
}
// Input parameters:
// R10: smi-tagged argument count, may be zero.
// RBP[target::frame_layout.param_end_from_fp + 1]: last argument.
static void PushArrayOfArguments(Assembler* assembler) {
__ LoadObject(R12, NullObject());
// Allocate array to store arguments of caller.
__ movq(RBX, R12); // Null element type for raw Array.
__ Call(StubCodeAllocateArray());
__ SmiUntag(R10);
// RAX: newly allocated array.
// R10: length of the array (was preserved by the stub).
__ pushq(RAX); // Array is in RAX and on top of stack.
__ leaq(R12,
Address(RBP, R10, TIMES_8,
target::frame_layout.param_end_from_fp * target::kWordSize));
__ leaq(RBX, FieldAddress(RAX, target::Array::data_offset()));
// R12: address of first argument on stack.
// RBX: address of first argument in array.
Label loop, loop_condition;
#if defined(DEBUG)
static auto const kJumpLength = Assembler::kFarJump;
#else
static auto const kJumpLength = Assembler::kNearJump;
#endif // DEBUG
__ jmp(&loop_condition, kJumpLength);
__ Bind(&loop);
__ movq(RDI, Address(R12, 0));
// Generational barrier is needed, array is not necessarily in new space.
__ StoreCompressedIntoObject(RAX, Address(RBX, 0), RDI);
__ addq(RBX, Immediate(target::kCompressedWordSize));
__ subq(R12, Immediate(target::kWordSize));
__ Bind(&loop_condition);
__ subq(R10, Immediate(1));
__ j(POSITIVE, &loop, Assembler::kNearJump);
}
// Used by eager and lazy deoptimization. Preserve result in RAX if necessary.
// This stub translates optimized frame into unoptimized frame. The optimized
// frame can contain values in registers and on stack, the unoptimized
// frame contains all values on stack.
// Deoptimization occurs in following steps:
// - Push all registers that can contain values.
// - Call C routine to copy the stack and saved registers into temporary buffer.
// - Adjust caller's frame to correct unoptimized frame size.
// - Fill the unoptimized frame.
// - Materialize objects that require allocation (e.g. Double instances).
// GC can occur only after frame is fully rewritten.
// Stack after EnterDartFrame(0, PP, kNoRegister) below:
// +------------------+
// | Saved PP | <- PP
// +------------------+
// | PC marker | <- TOS
// +------------------+
// | Saved FP | <- FP of stub
// +------------------+
// | return-address | (deoptimization point)
// +------------------+
// | Saved CODE_REG |
// +------------------+
// | ... | <- SP of optimized frame
//
// Parts of the code cannot GC, part of the code can GC.
static void GenerateDeoptimizationSequence(Assembler* assembler,
DeoptStubKind kind) {
// DeoptimizeCopyFrame expects a Dart frame, i.e. EnterDartFrame(0), but there
// is no need to set the correct PC marker or load PP, since they get patched.
__ EnterStubFrame();
// The code in this frame may not cause GC. kDeoptimizeCopyFrameRuntimeEntry
// and kDeoptimizeFillFrameRuntimeEntry are leaf runtime calls.
const intptr_t saved_result_slot_from_fp =
target::frame_layout.first_local_from_fp + 1 -
(kNumberOfCpuRegisters - RAX);
const intptr_t saved_exception_slot_from_fp =
target::frame_layout.first_local_from_fp + 1 -
(kNumberOfCpuRegisters - RAX);
const intptr_t saved_stacktrace_slot_from_fp =
target::frame_layout.first_local_from_fp + 1 -
(kNumberOfCpuRegisters - RDX);
// Result in RAX is preserved as part of pushing all registers below.
// Push registers in their enumeration order: lowest register number at
// lowest address.
for (intptr_t i = kNumberOfCpuRegisters - 1; i >= 0; i--) {
if (i == CODE_REG) {
// Save the original value of CODE_REG pushed before invoking this stub
// instead of the value used to call this stub.
__ pushq(Address(RBP, 2 * target::kWordSize));
} else {
__ pushq(static_cast<Register>(i));
}
}
__ subq(RSP, Immediate(kNumberOfXmmRegisters * kFpuRegisterSize));
intptr_t offset = 0;
for (intptr_t reg_idx = 0; reg_idx < kNumberOfXmmRegisters; ++reg_idx) {
XmmRegister xmm_reg = static_cast<XmmRegister>(reg_idx);
__ movups(Address(RSP, offset), xmm_reg);
offset += kFpuRegisterSize;
}
{
// Pass address of saved registers block.
__ movq(CallingConventions::kArg1Reg, RSP);
LeafRuntimeScope rt(assembler,
/*frame_size=*/0,
/*preserve_registers=*/false);
bool is_lazy =
(kind == kLazyDeoptFromReturn) || (kind == kLazyDeoptFromThrow);
__ movq(CallingConventions::kArg2Reg, Immediate(is_lazy ? 1 : 0));
rt.Call(kDeoptimizeCopyFrameRuntimeEntry, 2);
// Result (RAX) is stack-size (FP - SP) in bytes.
}
if (kind == kLazyDeoptFromReturn) {
// Restore result into RBX temporarily.
__ movq(RBX, Address(RBP, saved_result_slot_from_fp * target::kWordSize));
} else if (kind == kLazyDeoptFromThrow) {
// Restore result into RBX temporarily.
__ movq(RBX,
Address(RBP, saved_exception_slot_from_fp * target::kWordSize));
__ movq(RDX,
Address(RBP, saved_stacktrace_slot_from_fp * target::kWordSize));
}
// There is a Dart Frame on the stack. We must restore PP and leave frame.
__ RestoreCodePointer();
__ LeaveStubFrame();
__ popq(RCX); // Preserve return address.
__ movq(RSP, RBP); // Discard optimized frame.
__ subq(RSP, RAX); // Reserve space for deoptimized frame.
__ pushq(RCX); // Restore return address.
// DeoptimizeFillFrame expects a Dart frame, i.e. EnterDartFrame(0), but there
// is no need to set the correct PC marker or load PP, since they get patched.
__ EnterStubFrame();
if (kind == kLazyDeoptFromReturn) {
__ pushq(RBX); // Preserve result as first local.
} else if (kind == kLazyDeoptFromThrow) {
__ pushq(RBX); // Preserve exception as first local.
__ pushq(RDX); // Preserve stacktrace as second local.
}
{
__ movq(CallingConventions::kArg1Reg, RBP); // Pass last FP as a parameter.
LeafRuntimeScope rt(assembler,
/*frame_size=*/0,
/*preserve_registers=*/false);
rt.Call(kDeoptimizeFillFrameRuntimeEntry, 1);
}
if (FLAG_target_thread_sanitizer) {
__ movq(CALLEE_SAVED_TEMP, RAX); // Frames created
__ TsanFuncExit(/*preserve_registers=*/false); // Optimized frame.
Label loop;
__ Bind(&loop);
__ movq(CallingConventions::kArg1Reg, Immediate(42));
__ TsanFuncEntry(/*preserve_registers=*/false); // Unoptimized frames.
__ subq(CALLEE_SAVED_TEMP, Immediate(1));
__ j(NOT_ZERO, &loop);
}
if (kind == kLazyDeoptFromReturn) {
// Restore result into RBX.
__ movq(RBX, Address(RBP, target::frame_layout.first_local_from_fp *
target::kWordSize));
} else if (kind == kLazyDeoptFromThrow) {
// Restore exception into RBX.
__ movq(RBX, Address(RBP, target::frame_layout.first_local_from_fp *
target::kWordSize));
// Restore stacktrace into RDX.
__ movq(RDX, Address(RBP, (target::frame_layout.first_local_from_fp - 1) *
target::kWordSize));
}
// Code above cannot cause GC.
// There is a Dart Frame on the stack. We must restore PP and leave frame.
__ RestoreCodePointer();
__ LeaveStubFrame();
// Frame is fully rewritten at this point and it is safe to perform a GC.
// Materialize any objects that were deferred by FillFrame because they
// require allocation.
// Enter stub frame with loading PP. The caller's PP is not materialized yet.
__ EnterStubFrame();
if (kind == kLazyDeoptFromReturn) {
__ pushq(RBX); // Preserve result, it will be GC-d here.
} else if (kind == kLazyDeoptFromThrow) {
// Preserve CODE_REG for one more runtime call.
__ pushq(CODE_REG);
__ pushq(RBX); // Preserve exception.
__ pushq(RDX); // Preserve stacktrace.
}
__ pushq(Immediate(target::ToRawSmi(0))); // Space for the result.
__ CallRuntime(kDeoptimizeMaterializeRuntimeEntry, 0);
// Result tells stub how many bytes to remove from the expression stack
// of the bottom-most frame. They were used as materialization arguments.
__ popq(RBX);
__ SmiUntag(RBX);
if (kind == kLazyDeoptFromReturn) {
__ popq(RAX); // Restore result.
} else if (kind == kLazyDeoptFromThrow) {
__ popq(RDX); // Restore stacktrace.
__ popq(RAX); // Restore exception.
__ popq(CODE_REG);
}
__ LeaveStubFrame();
__ popq(RCX); // Pop return address.
__ addq(RSP, RBX); // Remove materialization arguments.
__ pushq(RCX); // Push return address.
// The caller is responsible for emitting the return instruction.
if (kind == kLazyDeoptFromThrow) {
// Unoptimized frame is now ready to accept the exception. Rethrow it to
// find the right handler.
__ EnterStubFrame();
__ pushq(Immediate(target::ToRawSmi(0))); // Space for the result.
__ pushq(RAX); // Exception
__ pushq(RDX); // Stacktrace
__ pushq(Immediate(target::ToRawSmi(1))); // Bypass debugger.
__ CallRuntime(kReThrowRuntimeEntry, 3);
__ LeaveStubFrame();
}
}
// RAX: result, must be preserved
void StubCodeCompiler::GenerateDeoptimizeLazyFromReturnStub() {
// Push zap value instead of CODE_REG for lazy deopt.
__ pushq(Immediate(kZapCodeReg));
// Return address for "call" to deopt stub.
__ pushq(Immediate(kZapReturnAddress));
__ movq(CODE_REG,
Address(THR, target::Thread::lazy_deopt_from_return_stub_offset()));
GenerateDeoptimizationSequence(assembler, kLazyDeoptFromReturn);
__ ret();
}
// RAX: exception, must be preserved
// RDX: stacktrace, must be preserved
void StubCodeCompiler::GenerateDeoptimizeLazyFromThrowStub() {
// Push zap value instead of CODE_REG for lazy deopt.
__ pushq(Immediate(kZapCodeReg));
// Return address for "call" to deopt stub.
__ pushq(Immediate(kZapReturnAddress));
__ movq(CODE_REG,
Address(THR, target::Thread::lazy_deopt_from_throw_stub_offset()));
GenerateDeoptimizationSequence(assembler, kLazyDeoptFromThrow);
__ ret();
}
void StubCodeCompiler::GenerateDeoptimizeStub() {
__ popq(TMP);
__ pushq(CODE_REG);
__ pushq(TMP);
__ movq(CODE_REG, Address(THR, target::Thread::deoptimize_stub_offset()));
GenerateDeoptimizationSequence(assembler, kEagerDeopt);
__ ret();
}
// Input:
// IC_DATA_REG - icdata/megamorphic_cache
// RDI - arguments descriptor size
static void GenerateNoSuchMethodDispatcherBody(Assembler* assembler,
Register receiver_reg) {
__ pushq(Immediate(0)); // Setup space on stack for result.
__ pushq(receiver_reg); // Receiver.
__ pushq(IC_DATA_REG); // ICData/MegamorphicCache.
__ pushq(ARGS_DESC_REG); // Arguments descriptor array.
// Adjust arguments count.
__ OBJ(cmp)(FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::type_args_len_offset()),
Immediate(0));
__ OBJ(mov)(R10, RDI);
Label args_count_ok;
__ j(EQUAL, &args_count_ok, Assembler::kNearJump);
// Include the type arguments.
__ OBJ(add)(R10, Immediate(target::ToRawSmi(1)));
__ Bind(&args_count_ok);
// R10: Smi-tagged arguments array length.
PushArrayOfArguments(assembler);
const intptr_t kNumArgs = 4;
__ CallRuntime(kNoSuchMethodFromCallStubRuntimeEntry, kNumArgs);
__ Drop(4);
__ popq(RAX); // Return value.
__ LeaveStubFrame();
__ ret();
}
// Input:
// IC_DATA_REG - icdata/megamorphic_cache
// ARGS_DESC_REG - argument descriptor
static void GenerateDispatcherCode(Assembler* assembler,
Label* call_target_function) {
__ Comment("NoSuchMethodDispatch");
// When lazily generated invocation dispatchers are disabled, the
// miss-handler may return null.
__ CompareObject(RAX, NullObject());
__ j(NOT_EQUAL, call_target_function);
__ EnterStubFrame();
// Load the receiver.
__ OBJ(mov)(RDI, FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::size_offset()));
__ movq(RAX,
Address(RBP, RDI, TIMES_HALF_WORD_SIZE,
target::frame_layout.param_end_from_fp * target::kWordSize));
GenerateNoSuchMethodDispatcherBody(assembler, /*receiver_reg=*/RAX);
}
// Input:
// IC_DATA_REG - icdata/megamorphic_cache
// RDX - receiver
void StubCodeCompiler::GenerateNoSuchMethodDispatcherStub() {
__ EnterStubFrame();
__ movq(ARGS_DESC_REG,
FieldAddress(IC_DATA_REG,
target::CallSiteData::arguments_descriptor_offset()));
__ OBJ(mov)(RDI, FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::size_offset()));
GenerateNoSuchMethodDispatcherBody(assembler, /*receiver_reg=*/RDX);
}
#ifdef DART_TARGET_SUPPORTS_PROBE_POINTS
void StubCodeCompiler::GenerateAllocationProbePointStub() {
if (!FLAG_generate_probe_points) {
__ Stop("unexpected invocation of an allocation probe");
return;
}
__ EnterStubFrame();
// Probe will be placed here by `runtime/tools/profiling/bin/set_uprobe.dart`.
const intptr_t probe_offset = __ CodeSize();
__ LeaveStubFrame();
__ Ret();
// This dummy instruction is encoding offset to the probe point. It will be
// used by set_uprobe.dart script.
__ TestImmediate(RAX, Immediate(probe_offset));
}
#endif
static void InvokeAllocationProbePoint(Assembler* assembler) {
#ifdef DART_TARGET_SUPPORTS_PROBE_POINTS
if (FLAG_precompiled_mode && FLAG_generate_probe_points) {
__ EnterStubFrame();
__ Call(StubCode::AllocationProbePoint());
__ LeaveStubFrame();
}
#endif
}
// Called for inline allocation of arrays.
// Input registers (preserved):
// AllocateArrayABI::kLengthReg: array length as Smi.
// AllocateArrayABI::kTypeArgumentsReg: type arguments of array.
// Output registers:
// AllocateArrayABI::kResultReg: newly allocated array.
// Clobbered:
// RCX, RDI, R12
void StubCodeCompiler::GenerateAllocateArrayStub() {
if (!FLAG_use_slow_path && FLAG_inline_alloc) {
Label slow_case;
// Compute the size to be allocated, it is based on the array length
// and is computed as:
// RoundedAllocationSize(
// (array_length * target::kCompressedWordSize) +
// target::Array::header_size()).
__ movq(RDI, AllocateArrayABI::kLengthReg); // Array Length.
// Check that length is Smi.
__ testq(RDI, Immediate(kSmiTagMask));
__ j(NOT_ZERO, &slow_case);
// Check length >= 0 && length <= kMaxNewSpaceElements
const Immediate& max_len =
Immediate(target::ToRawSmi(target::Array::kMaxNewSpaceElements));
__ OBJ(cmp)(RDI, max_len);
__ j(ABOVE, &slow_case);
// Check for allocation tracing.
NOT_IN_PRODUCT(__ MaybeTraceAllocation(kArrayCid, &slow_case));
const intptr_t fixed_size_plus_alignment_padding =
target::Array::header_size() +
target::ObjectAlignment::kObjectAlignment - 1;
// RDI is a Smi.
__ OBJ(lea)(RDI, Address(RDI, TIMES_COMPRESSED_HALF_WORD_SIZE,
fixed_size_plus_alignment_padding));
ASSERT(kSmiTagShift == 1);
__ andq(RDI, Immediate(-target::ObjectAlignment::kObjectAlignment));
const intptr_t cid = kArrayCid;
__ movq(AllocateArrayABI::kResultReg,
Address(THR, target::Thread::top_offset()));
// RDI: allocation size.
__ movq(RCX, AllocateArrayABI::kResultReg);
__ addq(RCX, RDI);
__ j(CARRY, &slow_case);
// Check if the allocation fits into the remaining space.
// AllocateArrayABI::kResultReg: potential new object start.
// RCX: potential next object start.
// RDI: allocation size.
__ cmpq(RCX, Address(THR, target::Thread::end_offset()));
__ j(ABOVE_EQUAL, &slow_case);
__ CheckAllocationCanary(AllocateArrayABI::kResultReg);
// Successfully allocated the object(s), now update top to point to
// next object start and initialize the object.
__ movq(Address(THR, target::Thread::top_offset()), RCX);
__ addq(AllocateArrayABI::kResultReg, Immediate(kHeapObjectTag));
// Initialize the tags.
// AllocateArrayABI::kResultReg: new object start as a tagged pointer.
// RDI: allocation size.
{
Label size_tag_overflow, done;
__ cmpq(RDI, Immediate(target::UntaggedObject::kSizeTagMaxSizeTag));
__ j(ABOVE, &size_tag_overflow, Assembler::kNearJump);
__ shlq(RDI, Immediate(target::UntaggedObject::kSizeTagPos -
target::ObjectAlignment::kObjectAlignmentLog2));
__ jmp(&done, Assembler::kNearJump);
__ Bind(&size_tag_overflow);
__ LoadImmediate(RDI, Immediate(0));
__ Bind(&done);
// Get the class index and insert it into the tags.
uword tags = target::MakeTagWordForNewSpaceObject(cid, 0);
__ orq(RDI, Immediate(tags));
__ InitializeHeader(RDI, RAX);
}
// AllocateArrayABI::kResultReg: new object start as a tagged pointer.
// Store the type argument field.
// No generational barrier needed, since we store into a new object.
__ StoreCompressedIntoObjectNoBarrier(
AllocateArrayABI::kResultReg,
FieldAddress(AllocateArrayABI::kResultReg,
target::Array::type_arguments_offset()),
AllocateArrayABI::kTypeArgumentsReg);
// Set the length field.
__ StoreCompressedIntoObjectNoBarrier(
AllocateArrayABI::kResultReg,
FieldAddress(AllocateArrayABI::kResultReg,
target::Array::length_offset()),
AllocateArrayABI::kLengthReg);
// Initialize all array elements to raw_null.
// AllocateArrayABI::kResultReg: new object start as a tagged pointer.
// RCX: new object end address.
// RDI: iterator which initially points to the start of the variable
// data area to be initialized.
__ LoadObject(R12, NullObject());
__ leaq(RDI, FieldAddress(AllocateArrayABI::kResultReg,
target::Array::header_size()));
Label loop;
__ Bind(&loop);
for (intptr_t offset = 0; offset < target::kObjectAlignment;
offset += target::kCompressedWordSize) {
// No generational barrier needed, since we are storing null.
__ StoreCompressedIntoObjectNoBarrier(AllocateArrayABI::kResultReg,
Address(RDI, offset), R12);
}
// Safe to only check every kObjectAlignment bytes instead of each word.
ASSERT(kAllocationRedZoneSize >= target::kObjectAlignment);
__ addq(RDI, Immediate(target::kObjectAlignment));
__ cmpq(RDI, RCX);
__ j(UNSIGNED_LESS, &loop);
__ WriteAllocationCanary(RCX);
InvokeAllocationProbePoint(assembler);
__ ret();
// Unable to allocate the array using the fast inline code, just call
// into the runtime.
__ Bind(&slow_case);
}
// Create a stub frame as we are pushing some objects on the stack before
// calling into the runtime.
__ EnterStubFrame();
__ pushq(Immediate(0)); // Space for return value.
__ pushq(AllocateArrayABI::kLengthReg); // Array length as Smi.
__ pushq(AllocateArrayABI::kTypeArgumentsReg); // Element type.
__ CallRuntime(kAllocateArrayRuntimeEntry, 2);
// Write-barrier elimination might be enabled for this array (depending on the
// array length). To be sure we will check if the allocated object is in old
// space and if so call a leaf runtime to add it to the remembered set.
__ movq(AllocateArrayABI::kResultReg, Address(RSP, 2 * target::kWordSize));
EnsureIsNewOrRemembered();
__ popq(AllocateArrayABI::kTypeArgumentsReg); // Pop element type argument.
__ popq(AllocateArrayABI::kLengthReg); // Pop array length argument.
__ popq(AllocateArrayABI::kResultReg); // Pop allocated object.
__ LeaveStubFrame();
__ ret();
}
void StubCodeCompiler::GenerateAllocateMintSharedWithFPURegsStub() {
// For test purpose call allocation stub without inline allocation attempt.
if (!FLAG_use_slow_path && FLAG_inline_alloc) {
Label slow_case;
__ TryAllocate(compiler::MintClass(), &slow_case, Assembler::kNearJump,
AllocateMintABI::kResultReg, AllocateMintABI::kTempReg);
InvokeAllocationProbePoint(assembler);
__ Ret();
__ Bind(&slow_case);
}
COMPILE_ASSERT(AllocateMintABI::kResultReg ==
SharedSlowPathStubABI::kResultReg);
GenerateSharedStub(/*save_fpu_registers=*/true, &kAllocateMintRuntimeEntry,
target::Thread::allocate_mint_with_fpu_regs_stub_offset(),
/*allow_return=*/true,
/*store_runtime_result_in_result_register=*/true);
}
void StubCodeCompiler::GenerateAllocateMintSharedWithoutFPURegsStub() {
// For test purpose call allocation stub without inline allocation attempt.
if (!FLAG_use_slow_path && FLAG_inline_alloc) {
Label slow_case;
__ TryAllocate(compiler::MintClass(), &slow_case, Assembler::kNearJump,
AllocateMintABI::kResultReg, AllocateMintABI::kTempReg);
InvokeAllocationProbePoint(assembler);
__ Ret();
__ Bind(&slow_case);
}
COMPILE_ASSERT(AllocateMintABI::kResultReg ==
SharedSlowPathStubABI::kResultReg);
GenerateSharedStub(
/*save_fpu_registers=*/false, &kAllocateMintRuntimeEntry,
target::Thread::allocate_mint_without_fpu_regs_stub_offset(),
/*allow_return=*/true,
/*store_runtime_result_in_result_register=*/true);
}
static const RegisterSet kCalleeSavedRegisterSet(
CallingConventions::kCalleeSaveCpuRegisters,
CallingConventions::kCalleeSaveXmmRegisters);
// Called when invoking Dart code from C++ (VM code).
// Input parameters:
// RSP : points to return address.
// RDI : target code or entry point (in AOT mode).
// RSI : arguments descriptor array.
// RDX : arguments array.
// RCX : current thread.
void StubCodeCompiler::GenerateInvokeDartCodeStub() {
__ EnterFrame(0);
const Register kTargetReg = CallingConventions::kArg1Reg;
const Register kArgDescReg = CallingConventions::kArg2Reg;
const Register kArgsReg = CallingConventions::kArg3Reg;
const Register kThreadReg = CallingConventions::kArg4Reg;
// Push code object to PC marker slot.
__ pushq(Address(kThreadReg, target::Thread::invoke_dart_code_stub_offset()));
// At this point, the stack looks like:
// | stub code object
// | saved RBP | <-- RBP
// | saved PC (return to DartEntry::InvokeFunction) |
const intptr_t kInitialOffset = 2;
// Save arguments descriptor array, later replaced by Smi argument count.
const intptr_t kArgumentsDescOffset = -(kInitialOffset)*target::kWordSize;
__ pushq(kArgDescReg);
// Save C++ ABI callee-saved registers.
__ PushRegisters(kCalleeSavedRegisterSet);
// If any additional (or fewer) values are pushed, the offsets in
// target::frame_layout.exit_link_slot_from_entry_fp will need to be changed.
// Set up THR, which caches the current thread in Dart code.
if (THR != kThreadReg) {
__ movq(THR, kThreadReg);
}
#if defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif
// Save the current VMTag on the stack.
__ movq(RAX, Assembler::VMTagAddress());
__ pushq(RAX);
// Save top resource and top exit frame info. Use RAX as a temporary register.
// StackFrameIterator reads the top exit frame info saved in this frame.
__ movq(RAX, Address(THR, target::Thread::top_resource_offset()));
__ pushq(RAX);
__ movq(Address(THR, target::Thread::top_resource_offset()), Immediate(0));
__ movq(RAX, Address(THR, target::Thread::exit_through_ffi_offset()));
__ pushq(RAX);
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(0));
__ movq(RAX, Address(THR, target::Thread::top_exit_frame_info_offset()));
__ pushq(RAX);
// The constant target::frame_layout.exit_link_slot_from_entry_fp must be kept
// in sync with the code above.
__ EmitEntryFrameVerification();
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()),
Immediate(0));
// Mark that the thread is executing Dart code. Do this after initializing the
// exit link for the profiler.
__ movq(Assembler::VMTagAddress(), Immediate(VMTag::kDartTagId));
// Load arguments descriptor array into R10, which is passed to Dart code.
__ movq(R10, kArgDescReg);
// Push arguments. At this point we only need to preserve kTargetReg.
ASSERT(kTargetReg != RDX);
// Load number of arguments into RBX and adjust count for type arguments.
__ OBJ(mov)(RBX,
FieldAddress(R10, target::ArgumentsDescriptor::count_offset()));
__ OBJ(cmp)(
FieldAddress(R10, target::ArgumentsDescriptor::type_args_len_offset()),
Immediate(0));
Label args_count_ok;
__ j(EQUAL, &args_count_ok, Assembler::kNearJump);
__ addq(RBX, Immediate(target::ToRawSmi(1))); // Include the type arguments.
__ Bind(&args_count_ok);
// Save number of arguments as Smi on stack, replacing saved ArgumentsDesc.
__ movq(Address(RBP, kArgumentsDescOffset), RBX);
__ SmiUntag(RBX);
// Compute address of 'arguments array' data area into RDX.
__ leaq(RDX, FieldAddress(kArgsReg, target::Array::data_offset()));
// Set up arguments for the Dart call.
Label push_arguments;
Label done_push_arguments;
__ j(ZERO, &done_push_arguments, Assembler::kNearJump);
__ LoadImmediate(RAX, Immediate(0));
__ Bind(&push_arguments);
#if defined(DART_COMPRESSED_POINTERS)
__ LoadCompressed(TMP, Address(RDX, RAX, TIMES_COMPRESSED_WORD_SIZE, 0));
__ pushq(TMP);
#else
__ pushq(Address(RDX, RAX, TIMES_8, 0));
#endif
__ addq(RAX, Immediate(1));
__ cmpq(RAX, RBX);
__ j(LESS, &push_arguments, Assembler::kNearJump);
__ Bind(&done_push_arguments);
// Call the Dart code entrypoint.
if (FLAG_precompiled_mode) {
__ movq(PP, Address(THR, target::Thread::global_object_pool_offset()));
__ xorq(CODE_REG, CODE_REG); // GC-safe value into CODE_REG.
} else {
__ xorq(PP, PP); // GC-safe value into PP.
__ movq(CODE_REG, kTargetReg);
__ movq(kTargetReg,
FieldAddress(CODE_REG, target::Code::entry_point_offset()));
}
__ call(kTargetReg); // R10 is the arguments descriptor array.
// Read the saved number of passed arguments as Smi.
__ movq(RDX, Address(RBP, kArgumentsDescOffset));
// Get rid of arguments pushed on the stack.
__ leaq(RSP, Address(RSP, RDX, TIMES_4, 0)); // RDX is a Smi.
// Restore the saved top exit frame info and top resource back into the
// Isolate structure.
__ popq(Address(THR, target::Thread::top_exit_frame_info_offset()));
__ popq(Address(THR, target::Thread::exit_through_ffi_offset()));
__ popq(Address(THR, target::Thread::top_resource_offset()));
// Restore the current VMTag from the stack.
__ popq(Assembler::VMTagAddress());
#if defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif
// Restore C++ ABI callee-saved registers.
__ PopRegisters(kCalleeSavedRegisterSet);
__ set_constant_pool_allowed(false);
// Restore the frame pointer.
__ LeaveFrame();
__ ret();
}
// Called when invoking compiled Dart code from interpreted Dart code.
// Input parameters:
// RSP : points to return address.
// RDI : target code or entry point (in AOT mode).
// RSI : arguments descriptor array.
// RDX : address of first argument.
// RCX : current thread.
void StubCodeCompiler::GenerateInvokeDartCodeFromBytecodeStub() {
#if defined(DART_BYTECODE_INTERPRETER)
__ EnterFrame(0);
const Register kTargetReg = CallingConventions::kArg1Reg;
const Register kArgDescReg = CallingConventions::kArg2Reg;
const Register kArg0Reg = CallingConventions::kArg3Reg;
const Register kThreadReg = CallingConventions::kArg4Reg;
// Push code object to PC marker slot.
__ pushq(
Address(kThreadReg,
target::Thread::invoke_dart_code_from_bytecode_stub_offset()));
// At this point, the stack looks like:
// | stub code object
// | saved RBP | <-- RBP
// | saved PC (return to interpreter's InvokeCompiled) |
const intptr_t kInitialOffset = 2;
// Save arguments descriptor array, later replaced by Smi argument count.
const intptr_t kArgumentsDescOffset = -(kInitialOffset)*target::kWordSize;
__ pushq(kArgDescReg);
// Save C++ ABI callee-saved registers.
__ PushRegisters(kCalleeSavedRegisterSet);
// If any additional (or fewer) values are pushed, the offsets in
// target::frame_layout.exit_link_slot_from_entry_fp will need to be changed.
// Set up THR, which caches the current thread in Dart code.
if (THR != kThreadReg) {
__ movq(THR, kThreadReg);
}
#if defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif
// Save the current VMTag on the stack.
__ movq(RAX, Assembler::VMTagAddress());
__ pushq(RAX);
// Save top resource and top exit frame info. Use RAX as a temporary register.
// StackFrameIterator reads the top exit frame info saved in this frame.
__ movq(RAX, Address(THR, target::Thread::top_resource_offset()));
__ pushq(RAX);
__ movq(Address(THR, target::Thread::top_resource_offset()), Immediate(0));
__ movq(RAX, Address(THR, target::Thread::exit_through_ffi_offset()));
__ pushq(RAX);
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(0));
__ movq(RAX, Address(THR, target::Thread::top_exit_frame_info_offset()));
__ pushq(RAX);
// The constant target::frame_layout.exit_link_slot_from_entry_fp must be kept
// in sync with the code above.
__ EmitEntryFrameVerification();
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()),
Immediate(0));
// Mark that the thread is executing Dart code. Do this after initializing the
// exit link for the profiler.
__ movq(Assembler::VMTagAddress(), Immediate(VMTag::kDartTagId));
// Load arguments descriptor array into R10, which is passed to Dart code.
__ movq(R10, kArgDescReg);
// Push arguments. At this point we only need to preserve kTargetReg.
ASSERT(kTargetReg != RDX);
// Load number of arguments into RBX and adjust count for type arguments.
__ OBJ(mov)(RBX,
FieldAddress(R10, target::ArgumentsDescriptor::count_offset()));
__ OBJ(cmp)(
FieldAddress(R10, target::ArgumentsDescriptor::type_args_len_offset()),
Immediate(0));
Label args_count_ok;
__ j(EQUAL, &args_count_ok, Assembler::kNearJump);
__ addq(RBX, Immediate(target::ToRawSmi(1))); // Include the type arguments.
__ Bind(&args_count_ok);
// Save number of arguments as Smi on stack, replacing saved ArgumentsDesc.
__ movq(Address(RBP, kArgumentsDescOffset), RBX);
__ SmiUntag(RBX);
// Compute address of first argument into RDX.
__ MoveRegister(RDX, kArg0Reg);
// Set up arguments for the Dart call.
Label push_arguments;
Label done_push_arguments;
__ j(ZERO, &done_push_arguments, Assembler::kNearJump);
__ LoadImmediate(RAX, Immediate(0));
__ Bind(&push_arguments);
__ pushq(Address(RDX, RAX, TIMES_8, 0));
__ addq(RAX, Immediate(1));
__ cmpq(RAX, RBX);
__ j(LESS, &push_arguments, Assembler::kNearJump);
__ Bind(&done_push_arguments);
// Call the Dart code entrypoint.
if (FLAG_precompiled_mode) {
__ movq(PP, Address(THR, target::Thread::global_object_pool_offset()));
__ xorq(CODE_REG, CODE_REG); // GC-safe value into CODE_REG.
} else {
__ xorq(PP, PP); // GC-safe value into PP.
__ movq(CODE_REG, kTargetReg);
__ movq(kTargetReg,
FieldAddress(CODE_REG, target::Code::entry_point_offset()));
}
__ call(kTargetReg); // R10 is the arguments descriptor array.
// Read the saved number of passed arguments as Smi.
__ movq(RDX, Address(RBP, kArgumentsDescOffset));
// Get rid of arguments pushed on the stack.
__ leaq(RSP, Address(RSP, RDX, TIMES_4, 0)); // RDX is a Smi.
// Restore the saved top exit frame info and top resource back into the
// Isolate structure.
__ popq(Address(THR, target::Thread::top_exit_frame_info_offset()));
__ popq(Address(THR, target::Thread::exit_through_ffi_offset()));
__ popq(Address(THR, target::Thread::top_resource_offset()));
// Restore the current VMTag from the stack.
__ popq(Assembler::VMTagAddress());
#if defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif
// Restore C++ ABI callee-saved registers.
__ PopRegisters(kCalleeSavedRegisterSet);
__ set_constant_pool_allowed(false);
// Restore the frame pointer.
__ LeaveFrame();
__ ret();
#else
__ Stop("Not using Dart dynamic modules");
#endif // defined(DART_BYTECODE_INTERPRETER)
}
// Helper to generate space allocation of context stub.
// This does not initialize the fields of the context.
// Input:
// R10: number of context variables.
// Output:
// RAX: new, uninitialized allocated Context object.
// Clobbered:
// R13
static void GenerateAllocateContextSpaceStub(Assembler* assembler,
Label* slow_case) {
// First compute the rounded instance size.
// R10: number of context variables.
intptr_t fixed_size_plus_alignment_padding =
(target::Context::header_size() +
target::ObjectAlignment::kObjectAlignment - 1);
__ leaq(R13, Address(R10, TIMES_COMPRESSED_WORD_SIZE,
fixed_size_plus_alignment_padding));
__ andq(R13, Immediate(-target::ObjectAlignment::kObjectAlignment));
// Check for allocation tracing.
NOT_IN_PRODUCT(__ MaybeTraceAllocation(kContextCid, slow_case));
// Now allocate the object.
// R10: number of context variables.
__ movq(RAX, Address(THR, target::Thread::top_offset()));
__ addq(R13, RAX);
// Check if the allocation fits into the remaining space.
// RAX: potential new object.
// R13: potential next object start.
// R10: number of context variables.
__ cmpq(R13, Address(THR, target::Thread::end_offset()));
__ j(ABOVE_EQUAL, slow_case);
__ CheckAllocationCanary(RAX);
// Successfully allocated the object, now update top to point to
// next object start and initialize the object.
// RAX: new object.
// R13: next object start.
// R10: number of context variables.
__ movq(Address(THR, target::Thread::top_offset()), R13);
// R13: Size of allocation in bytes.
__ subq(R13, RAX);
__ addq(RAX, Immediate(kHeapObjectTag));
// Generate isolate-independent code to allow sharing between isolates.
// Calculate the size tag.
// RAX: new object.
// R10: number of context variables.
{
Label size_tag_overflow, done;
__ leaq(R13, Address(R10, TIMES_COMPRESSED_WORD_SIZE,
fixed_size_plus_alignment_padding));
__ andq(R13, Immediate(-target::ObjectAlignment::kObjectAlignment));
__ cmpq(R13, Immediate(target::UntaggedObject::kSizeTagMaxSizeTag));
__ j(ABOVE, &size_tag_overflow, Assembler::kNearJump);
__ shlq(R13, Immediate(target::UntaggedObject::kSizeTagPos -
target::ObjectAlignment::kObjectAlignmentLog2));
__ jmp(&done);
__ Bind(&size_tag_overflow);
// Set overflow size tag value.
__ LoadImmediate(R13, Immediate(0));
__ Bind(&done);
// RAX: new object.
// R10: number of context variables.
// R13: size and bit tags.
uword tags = target::MakeTagWordForNewSpaceObject(kContextCid, 0);
__ orq(R13, Immediate(tags));
__ InitializeHeader(R13, RAX);
}
// Setup up number of context variables field.
// RAX: new object.
// R10: number of context variables as integer value (not object).
__ movl(FieldAddress(RAX, target::Context::num_variables_offset()), R10);
}
// Called for inline allocation of contexts.
// Input:
// R10: number of context variables.
// Output:
// RAX: new allocated Context object.
// Clobbered:
// R9, R13
void StubCodeCompiler::GenerateAllocateContextStub() {
__ LoadObject(R9, NullObject());
if (!FLAG_use_slow_path && FLAG_inline_alloc) {
Label slow_case;
GenerateAllocateContextSpaceStub(assembler, &slow_case);
// Setup the parent field.
// RAX: new object.
// R9: Parent object, initialized to null.
// No generational barrier needed, since we are storing null.
__ StoreCompressedIntoObjectNoBarrier(
RAX, FieldAddress(RAX, target::Context::parent_offset()), R9);
// Initialize the context variables.
// RAX: new object.
// R10: number of context variables.
{
Label loop, entry;
__ leaq(R13, FieldAddress(RAX, target::Context::variable_offset(0)));
#if defined(DEBUG)
static auto const kJumpLength = Assembler::kFarJump;
#else
static auto const kJumpLength = Assembler::kNearJump;
#endif // DEBUG
__ jmp(&entry, kJumpLength);
__ Bind(&loop);
__ subq(R10, Immediate(1));
// No generational barrier needed, since we are storing null.
__ StoreCompressedIntoObjectNoBarrier(
RAX, Address(R13, R10, TIMES_COMPRESSED_WORD_SIZE, 0), R9);
__ Bind(&entry);
__ cmpq(R10, Immediate(0));
__ j(NOT_EQUAL, &loop, Assembler::kNearJump);
}
// Done allocating and initializing the context.
// RAX: new object.
InvokeAllocationProbePoint(assembler);
__ ret();
__ Bind(&slow_case);
}
// Create a stub frame.
__ EnterStubFrame();
__ pushq(R9); // Setup space on stack for the return value.
__ SmiTag(R10);
__ pushq(R10); // Push number of context variables.
__ CallRuntime(kAllocateContextRuntimeEntry, 1); // Allocate context.
__ popq(RAX); // Pop number of context variables argument.
__ popq(RAX); // Pop the new context object.
// Write-barrier elimination might be enabled for this context (depending on
// the size). To be sure we will check if the allocated object is in old
// space and if so call a leaf runtime to add it to the remembered set.
EnsureIsNewOrRemembered();
// RAX: new object
// Restore the frame pointer.
__ LeaveStubFrame();
__ ret();
}
// Called for inline clone of contexts.
// Input:
// R9: context to clone.
// Output:
// RAX: new allocated Context object.
// Clobbered:
// R10, R13
void StubCodeCompiler::GenerateCloneContextStub() {
if (!FLAG_use_slow_path && FLAG_inline_alloc) {
Label slow_case;
// Load num. variable (int32_t) in the existing context.
__ movsxd(R10, FieldAddress(R9, target::Context::num_variables_offset()));
// Allocate new context of same size.
GenerateAllocateContextSpaceStub(assembler, &slow_case);
// Load parent in the existing context.
__ LoadCompressed(R13, FieldAddress(R9, target::Context::parent_offset()));
// Setup the parent field.
// RAX: new object.
// R9: Old parent object.
__ StoreCompressedIntoObjectNoBarrier(
RAX, FieldAddress(RAX, target::Context::parent_offset()), R13);
// Clone the context variables.
// RAX: new context clone.
// R10: number of context variables.
{
Label loop, entry;
__ jmp(&entry, Assembler::kNearJump);
__ Bind(&loop);
__ subq(R10, Immediate(1));
__ LoadCompressed(R13, FieldAddress(R9, R10, TIMES_COMPRESSED_WORD_SIZE,
target::Context::variable_offset(0)));
__ StoreCompressedIntoObjectNoBarrier(
RAX,
FieldAddress(RAX, R10, TIMES_COMPRESSED_WORD_SIZE,
target::Context::variable_offset(0)),
R13);
__ Bind(&entry);
__ cmpq(R10, Immediate(0));
__ j(NOT_EQUAL, &loop, Assembler::kNearJump);
}
// Done allocating and initializing the context.
// RAX: new object.
InvokeAllocationProbePoint(assembler);
__ ret();
__ Bind(&slow_case);
}
// Create a stub frame.
__ EnterStubFrame();
__ PushObject(NullObject()); // Make space on stack for the return value.
__ pushq(R9); // Push context.
__ CallRuntime(kCloneContextRuntimeEntry, 1); // Clone context.
__ popq(RAX); // Pop context argument.
__ popq(RAX); // Pop the new context object.
// Write-barrier elimination might be enabled for this context (depending on
// the size). To be sure we will check if the allocated object is in old
// space and if so call a leaf runtime to add it to the remembered set.
EnsureIsNewOrRemembered();
// RAX: new object
// Restore the frame pointer.
__ LeaveStubFrame();
__ ret();
}
void StubCodeCompiler::GenerateWriteBarrierWrappersStub() {
for (intptr_t i = 0; i < kNumberOfCpuRegisters; ++i) {
if ((kDartAvailableCpuRegs & (1 << i)) == 0) continue;
Register reg = static_cast<Register>(i);
intptr_t start = __ CodeSize();
__ pushq(kWriteBarrierObjectReg);
__ movq(kWriteBarrierObjectReg, reg);
__ call(Address(THR, target::Thread::write_barrier_entry_point_offset()));
__ popq(kWriteBarrierObjectReg);
__ ret();
intptr_t end = __ CodeSize();
RELEASE_ASSERT(end - start == kStoreBufferWrapperSize);
}
}
// Helper stub to implement Assembler::StoreIntoObject/Array.
// Input parameters:
// RDX: Object (old)
// RAX: Value (old or new)
// R13: Slot
// If RAX is new, add RDX to the store buffer. Otherwise RAX is old, mark RAX
// and add it to the mark list.
COMPILE_ASSERT(kWriteBarrierObjectReg == RDX);
COMPILE_ASSERT(kWriteBarrierValueReg == RAX);
COMPILE_ASSERT(kWriteBarrierSlotReg == R13);
static void GenerateWriteBarrierStubHelper(Assembler* assembler, bool cards) {
Label skip_marking;
__ movq(TMP, FieldAddress(RAX, target::Object::tags_offset()));
__ andq(TMP, Address(THR, target::Thread::write_barrier_mask_offset()));
__ testq(TMP, Immediate(target::UntaggedObject::kIncrementalBarrierMask));
__ j(ZERO, &skip_marking);
{
// Atomically clear kNotMarkedBit.
Label is_new, done;
__ pushq(RCX); // Spill.
__ lock();
__ btrq(FieldAddress(RAX, target::Object::tags_offset()),
target::UntaggedObject::kNotMarkedBit);
__ j(NOT_CARRY, &done); // Marked by another thread.
__ testq(RAX,
Immediate(1 << target::ObjectAlignment::kNewObjectBitPosition));
__ j(NOT_ZERO, &is_new);
auto mark_stack_push = [&](intptr_t offset, const RuntimeEntry& entry) {
__ movq(TMP, Address(THR, offset));
__ movl(RCX, Address(TMP, target::MarkingStackBlock::top_offset()));
__ movq(Address(TMP, RCX, TIMES_8,
target::MarkingStackBlock::pointers_offset()),
RAX);
__ addq(RCX, Immediate(1));
__ movl(Address(TMP, target::MarkingStackBlock::top_offset()), RCX);
__ cmpl(RCX, Immediate(target::MarkingStackBlock::kSize));
__ j(NOT_EQUAL, &done);
{
LeafRuntimeScope rt(assembler,
/*frame_size=*/0,
/*preserve_registers=*/true);
__ movq(CallingConventions::kArg1Reg, THR);
rt.Call(entry, 1);
}
};
mark_stack_push(target::Thread::old_marking_stack_block_offset(),
kOldMarkingStackBlockProcessRuntimeEntry);
__ jmp(&done);
__ Bind(&is_new);
mark_stack_push(target::Thread::new_marking_stack_block_offset(),
kNewMarkingStackBlockProcessRuntimeEntry);
__ Bind(&done);
__ popq(RCX); // Unspill.
}
Label add_to_remembered_set, remember_card;
__ Bind(&skip_marking);
__ movq(TMP, FieldAddress(RDX, target::Object::tags_offset()));
__ shrl(TMP, Immediate(target::UntaggedObject::kBarrierOverlapShift));
__ andq(TMP, FieldAddress(RAX, target::Object::tags_offset()));
__ testq(TMP, Immediate(target::UntaggedObject::kGenerationalBarrierMask));
__ j(NOT_ZERO, &add_to_remembered_set, Assembler::kNearJump);
__ ret();
__ Bind(&add_to_remembered_set);
if (cards) {
__ movl(TMP, FieldAddress(RDX, target::Object::tags_offset()));
__ testl(TMP, Immediate(1 << target::UntaggedObject::kCardRememberedBit));
__ j(NOT_ZERO, &remember_card, Assembler::kFarJump);
} else {
#if defined(DEBUG)
Label ok;
__ movl(TMP, FieldAddress(RDX, target::Object::tags_offset()));
__ testl(TMP, Immediate(1 << target::UntaggedObject::kCardRememberedBit));
__ j(ZERO, &ok, Assembler::kFarJump);
__ Stop("Wrong barrier");
__ Bind(&ok);
#endif
}
{
// Atomically clear kOldAndNotRemembered.
Label done;
__ pushq(RCX); // Spill.
__ lock();
__ btrq(FieldAddress(RDX, target::Object::tags_offset()),
target::UntaggedObject::kOldAndNotRememberedBit);
__ j(NOT_CARRY, &done); // Remembered by another thread.
// Load the StoreBuffer block out of the thread. Then load top_ out of the
// StoreBufferBlock and add the address to the pointers_.
// RDX: Address being stored
__ movq(TMP, Address(THR, target::Thread::store_buffer_block_offset()));
__ movl(RCX, Address(TMP, target::StoreBufferBlock::top_offset()));
__ movq(
Address(TMP, RCX, TIMES_8, target::StoreBufferBlock::pointers_offset()),
RDX);
// Increment top_ and check for overflow.
// RCX: top_
// TMP: StoreBufferBlock
__ addq(RCX, Immediate(1));
__ movl(Address(TMP, target::StoreBufferBlock::top_offset()), RCX);
__ cmpl(RCX, Immediate(target::StoreBufferBlock::kSize));
__ j(NOT_EQUAL, &done);
{
LeafRuntimeScope rt(assembler,
/*frame_size=*/0,
/*preserve_registers=*/true);
__ movq(CallingConventions::kArg1Reg, THR);
rt.Call(kStoreBufferBlockProcessRuntimeEntry, 1);
}
__ Bind(&done);
__ popq(RCX); // Unspill.
__ ret();
}
if (cards) {
// Get card table.
__ Bind(&remember_card);
__ movq(TMP, RDX); // Object.
__ andq(TMP, Immediate(target::Page::kPageMask)); // Page.
// Atomically dirty the card.
__ pushq(RAX);
__ pushq(RCX);
__ subq(R13, TMP); // Offset in page.
__ movq(TMP,
Address(TMP, target::Page::card_table_offset())); // Card table.
__ shrq(R13, Immediate(target::Page::kBytesPerCardLog2)); // Card index.
__ movq(RCX, R13);
__ shrq(R13, Immediate(target::kBitsPerWordLog2)); // Word offset.
__ movq(RAX, Immediate(1));
__ shlq(RAX, RCX); // Bit mask. (Shift amount is mod 63.)
__ lock();
__ orq(Address(TMP, R13, TIMES_8, 0), RAX);
__ popq(RCX);
__ popq(RAX);
__ ret();
}
}
void StubCodeCompiler::GenerateWriteBarrierStub() {
GenerateWriteBarrierStubHelper(assembler, false);
}
void StubCodeCompiler::GenerateArrayWriteBarrierStub() {
GenerateWriteBarrierStubHelper(assembler, true);
}
static void GenerateAllocateObjectHelper(Assembler* assembler,
bool is_cls_parameterized) {
// Note: Keep in sync with calling function.
const Register kTagsReg = AllocateObjectABI::kTagsReg;
{
Label slow_case;
const Register kNewTopReg = R9;
#if !defined(PRODUCT)
{
const Register kCidRegister = RSI;
__ ExtractClassIdFromTags(kCidRegister, AllocateObjectABI::kTagsReg);
__ MaybeTraceAllocation(kCidRegister, &slow_case, TMP);
}
#endif
// Allocate the object and update top to point to
// next object start and initialize the allocated object.
{
const Register kInstanceSizeReg = RSI;
__ ExtractInstanceSizeFromTags(kInstanceSizeReg, kTagsReg);
__ movq(AllocateObjectABI::kResultReg,
Address(THR, target::Thread::top_offset()));
__ leaq(kNewTopReg, Address(AllocateObjectABI::kResultReg,
kInstanceSizeReg, TIMES_1, 0));
// Check if the allocation fits into the remaining space.
__ cmpq(kNewTopReg, Address(THR, target::Thread::end_offset()));
__ j(ABOVE_EQUAL, &slow_case);
__ CheckAllocationCanary(AllocateObjectABI::kResultReg);
__ movq(Address(THR, target::Thread::top_offset()), kNewTopReg);
} // kInstanceSizeReg = RSI
// Set the tags.
// 64 bit store also zeros the identity hash field.
__ InitializeHeaderUntagged(kTagsReg, AllocateObjectABI::kResultReg);
__ addq(AllocateObjectABI::kResultReg, Immediate(kHeapObjectTag));
// Initialize the remaining words of the object.
{
const Register kNextFieldReg = RDI;
__ leaq(kNextFieldReg,
FieldAddress(AllocateObjectABI::kResultReg,
target::Instance::first_field_offset()));
const Register kNullReg = R10;
__ LoadObject(kNullReg, NullObject());
// Loop until the whole object is initialized.
Label loop;
__ Bind(&loop);
for (intptr_t offset = 0; offset < target::kObjectAlignment;
offset += target::kCompressedWordSize) {
__ StoreCompressedIntoObjectNoBarrier(AllocateObjectABI::kResultReg,
Address(kNextFieldReg, offset),
kNullReg);
}
// Safe to only check every kObjectAlignment bytes instead of each word.
ASSERT(kAllocationRedZoneSize >= target::kObjectAlignment);
__ addq(kNextFieldReg, Immediate(target::kObjectAlignment));
__ cmpq(kNextFieldReg, kNewTopReg);
__ j(UNSIGNED_LESS, &loop);
} // kNextFieldReg = RDI, kNullReg = R10
__ WriteAllocationCanary(kNewTopReg); // Fix overshoot.
if (is_cls_parameterized) {
Label not_parameterized_case;
const Register kClsIdReg = R9;
const Register kTypeOffsetReg = RDI;
__ ExtractClassIdFromTags(kClsIdReg, kTagsReg);
// Load class' type_arguments_field offset in words.
__ LoadClassById(kTypeOffsetReg, kClsIdReg);
__ movl(
kTypeOffsetReg,
FieldAddress(kTypeOffsetReg,
target::Class::
host_type_arguments_field_offset_in_words_offset()));
// Set the type arguments in the new object.
__ StoreCompressedIntoObject(
AllocateObjectABI::kResultReg,
FieldAddress(AllocateObjectABI::kResultReg, kTypeOffsetReg,
TIMES_COMPRESSED_WORD_SIZE, 0),
AllocateObjectABI::kTypeArgumentsReg);
__ Bind(&not_parameterized_case);
} // kTypeOffsetReg = RDI;
InvokeAllocationProbePoint(assembler);
__ ret();
__ Bind(&slow_case);
} // kNewTopReg = R9;
// Fall back on slow case:
if (!is_cls_parameterized) {
__ LoadObject(AllocateObjectABI::kTypeArgumentsReg, NullObject());
}
// Tail call to generic allocation stub.
__ jmp(
Address(THR, target::Thread::allocate_object_slow_entry_point_offset()));
}
// Called for inline allocation of objects (any class).
void StubCodeCompiler::GenerateAllocateObjectStub() {
GenerateAllocateObjectHelper(assembler, /*is_cls_parameterized=*/false);
}
void StubCodeCompiler::GenerateAllocateObjectParameterizedStub() {
GenerateAllocateObjectHelper(assembler, /*is_cls_parameterized=*/true);
}
void StubCodeCompiler::GenerateAllocateObjectSlowStub() {
if (!FLAG_precompiled_mode) {
__ movq(CODE_REG,
Address(THR, target::Thread::call_to_runtime_stub_offset()));
}
__ ExtractClassIdFromTags(AllocateObjectABI::kTagsReg,
AllocateObjectABI::kTagsReg);
// Create a stub frame.
// Ensure constant pool is allowed so we can e.g. load class object.
__ EnterStubFrame();
// Setup space on stack for return value.
__ LoadObject(AllocateObjectABI::kResultReg, NullObject());
__ pushq(AllocateObjectABI::kResultReg);
// Push class of object to be allocated.
__ LoadClassById(AllocateObjectABI::kResultReg, AllocateObjectABI::kTagsReg);
__ pushq(AllocateObjectABI::kResultReg);
// Must be Object::null() if non-parameterized class.
__ pushq(AllocateObjectABI::kTypeArgumentsReg);
__ CallRuntime(kAllocateObjectRuntimeEntry, 2);
__ popq(AllocateObjectABI::kResultReg); // Drop type arguments.
__ popq(AllocateObjectABI::kResultReg); // Drop class.
__ popq(AllocateObjectABI::kResultReg); // Pop newly allocated object.
// Write-barrier elimination is enabled for [cls] and we therefore need to
// ensure that the object is in new-space or has remembered bit set.
EnsureIsNewOrRemembered();
// AllocateObjectABI::kResultReg: new object
// Restore the frame pointer.
__ LeaveStubFrame();
__ ret();
}
// Called for inline allocation of objects.
void StubCodeCompiler::GenerateAllocationStubForClass(
UnresolvedPcRelativeCalls* unresolved_calls,
const Class& cls,
const Code& allocate_object,
const Code& allocat_object_parametrized) {
classid_t cls_id = target::Class::GetId(cls);
ASSERT(cls_id != kIllegalCid);
const intptr_t cls_type_arg_field_offset =
target::Class::TypeArgumentsFieldOffset(cls);
// The generated code is different if the class is parameterized.
const bool is_cls_parameterized = target::Class::NumTypeArguments(cls) > 0;
ASSERT(!is_cls_parameterized ||
cls_type_arg_field_offset != target::Class::kNoTypeArguments);
const intptr_t instance_size = target::Class::GetInstanceSize(cls);
ASSERT(instance_size > 0);
const uword tags =
target::MakeTagWordForNewSpaceObject(cls_id, instance_size);
const Register kTagsReg = AllocateObjectABI::kTagsReg;
__ movq(kTagsReg, Immediate(tags));
// Load the appropriate generic alloc. stub.
if (!FLAG_use_slow_path && FLAG_inline_alloc &&
!target::Class::TraceAllocation(cls) &&
target::SizeFitsInSizeTag(instance_size)) {
RELEASE_ASSERT(AllocateObjectInstr::WillAllocateNewOrRemembered(cls));
RELEASE_ASSERT(target::Heap::IsAllocatableInNewSpace(instance_size));
if (is_cls_parameterized) {
if (!IsSameObject(NullObject(),
CastHandle<Object>(allocat_object_parametrized))) {
__ GenerateUnRelocatedPcRelativeTailCall();
unresolved_calls->Add(new UnresolvedPcRelativeCall(
__ CodeSize(), allocat_object_parametrized, /*is_tail_call=*/true));
} else {
__ jmp(Address(THR,
target::Thread::
allocate_object_parameterized_entry_point_offset()));
}
} else {
if (!IsSameObject(NullObject(), CastHandle<Object>(allocate_object))) {
__ GenerateUnRelocatedPcRelativeTailCall();
unresolved_calls->Add(new UnresolvedPcRelativeCall(
__ CodeSize(), allocate_object, /*is_tail_call=*/true));
} else {
__ jmp(
Address(THR, target::Thread::allocate_object_entry_point_offset()));
}
}
} else {
if (!is_cls_parameterized) {
__ LoadObject(AllocateObjectABI::kTypeArgumentsReg, NullObject());
}
__ jmp(Address(THR,
target::Thread::allocate_object_slow_entry_point_offset()));
}
}
// Called for invoking "dynamic noSuchMethod(Invocation invocation)" function
// from the entry code of a dart function after an error in passed argument
// name or number is detected.
// Input parameters:
// RSP : points to return address.
// RSP + 8 : address of last argument.
// R10 : arguments descriptor array.
void StubCodeCompiler::GenerateCallClosureNoSuchMethodStub() {
__ EnterStubFrame();
// Load the receiver.
// Note: In compressed pointer mode LoadCompressedSmi zero extends R13,
// rather than sign extending it. This is ok since it's an unsigned value.
__ LoadCompressedSmi(
R13, FieldAddress(R10, target::ArgumentsDescriptor::size_offset()));
__ movq(RAX,
Address(RBP, R13, TIMES_4,
target::frame_layout.param_end_from_fp * target::kWordSize));
// Load the function.
__ LoadCompressed(RBX, FieldAddress(RAX, target::Closure::function_offset()));
__ pushq(Immediate(0)); // Result slot.
__ pushq(RAX); // Receiver.
__ pushq(RBX); // Function.
__ pushq(R10); // Arguments descriptor array.
// Adjust arguments count.
__ OBJ(cmp)(
FieldAddress(R10, target::ArgumentsDescriptor::type_args_len_offset()),
Immediate(0));
__ movq(R10, R13);
Label args_count_ok;
__ j(EQUAL, &args_count_ok, Assembler::kNearJump);
__ addq(R10, Immediate(target::ToRawSmi(1))); // Include the type arguments.
__ Bind(&args_count_ok);
// R10: Smi-tagged arguments array length.
PushArrayOfArguments(assembler);
const intptr_t kNumArgs = 4;
__ CallRuntime(kNoSuchMethodFromPrologueRuntimeEntry, kNumArgs);
// noSuchMethod on closures always throws an error, so it will never return.
__ int3();
}
// Cannot use function object from ICData as it may be the inlined
// function and not the top-scope function.
void StubCodeCompiler::GenerateOptimizedUsageCounterIncrement() {
if (FLAG_precompiled_mode) {
__ Breakpoint();
return;
}
Register func_reg = RDI;
__ addl(FieldAddress(func_reg, target::Function::usage_counter_offset()),
Immediate(1));
}
// Loads function into 'temp_reg', preserves IC_DATA_REG.
void StubCodeCompiler::GenerateUsageCounterIncrement(Register temp_reg) {
if (FLAG_precompiled_mode) {
__ Breakpoint();
return;
}
if (FLAG_optimization_counter_threshold >= 0) {
Register func_reg = temp_reg;
ASSERT(func_reg != IC_DATA_REG);
__ Comment("Increment function counter");
__ movq(func_reg,
FieldAddress(IC_DATA_REG, target::ICData::owner_offset()));
__ addl(FieldAddress(func_reg, target::Function::usage_counter_offset()),
Immediate(1));
}
}
// Note: RBX must be preserved.
// Attempt a quick Smi operation for known operations ('kind'). The ICData
// must have been primed with a Smi/Smi check that will be used for counting
// the invocations.
static void EmitFastSmiOp(Assembler* assembler,
Token::Kind kind,
intptr_t num_args,
Label* not_smi_or_overflow) {
__ Comment("Fast Smi op");
ASSERT(num_args == 2);
__ movq(RAX, Address(RSP, +2 * target::kWordSize)); // Left.
__ movq(RCX, Address(RSP, +1 * target::kWordSize)); // Right
__ movq(R13, RCX);
__ orq(R13, RAX);
__ testq(R13, Immediate(kSmiTagMask));
__ j(NOT_ZERO, not_smi_or_overflow);
switch (kind) {
case Token::kADD: {
__ OBJ(add)(RAX, RCX);
__ j(OVERFLOW, not_smi_or_overflow);
break;
}
case Token::kLT: {
__ OBJ(cmp)(RAX, RCX);
__ setcc(GREATER_EQUAL, ByteRegisterOf(RAX));
__ movzxb(RAX, RAX); // RAX := RAX < RCX ? 0 : 1
__ movq(RAX,
Address(THR, RAX, TIMES_8, target::Thread::bool_true_offset()));
ASSERT(target::Thread::bool_true_offset() + 8 ==
target::Thread::bool_false_offset());
break;
}
case Token::kEQ: {
__ OBJ(cmp)(RAX, RCX);
__ setcc(NOT_EQUAL, ByteRegisterOf(RAX));
__ movzxb(RAX, RAX); // RAX := RAX == RCX ? 0 : 1
__ movq(RAX,
Address(THR, RAX, TIMES_8, target::Thread::bool_true_offset()));
ASSERT(target::Thread::bool_true_offset() + 8 ==
target::Thread::bool_false_offset());
break;
}
default:
UNIMPLEMENTED();
}
// RBX: IC data object (preserved).
__ movq(R13, FieldAddress(RBX, target::ICData::entries_offset()));
// R13: ic_data_array with check entries: classes and target functions.
__ leaq(R13, FieldAddress(R13, target::Array::data_offset()));
// R13: points directly to the first ic data array element.
#if defined(DEBUG)
// Check that first entry is for Smi/Smi.
Label error, ok;
const Immediate& imm_smi_cid = Immediate(target::ToRawSmi(kSmiCid));
__ OBJ(cmp)(Address(R13, 0 * target::kCompressedWordSize), imm_smi_cid);
__ j(NOT_EQUAL, &error, Assembler::kNearJump);
__ OBJ(cmp)(Address(R13, 1 * target::kCompressedWordSize), imm_smi_cid);
__ j(EQUAL, &ok, Assembler::kNearJump);
__ Bind(&error);
__ Stop("Incorrect IC data");
__ Bind(&ok);
#endif
if (FLAG_optimization_counter_threshold >= 0) {
const intptr_t count_offset =
target::ICData::CountIndexFor(num_args) * target::kCompressedWordSize;
// Update counter, ignore overflow.
__ OBJ(add)(Address(R13, count_offset), Immediate(target::ToRawSmi(1)));
}
__ ret();
}
// Saves the offset of the target entry-point (from the Function) into R8.
//
// Must be the first code generated, since any code before will be skipped in
// the unchecked entry-point.
static void GenerateRecordEntryPoint(Assembler* assembler) {
Label done;
__ movq(R8,
Immediate(target::Function::entry_point_offset() - kHeapObjectTag));
__ jmp(&done);
__ BindUncheckedEntryPoint();
__ movq(R8, Immediate(target::Function::entry_point_offset(
CodeEntryKind::kUnchecked) -
kHeapObjectTag));
__ Bind(&done);
}
// Generate inline cache check for 'num_args'.
// RDX: receiver (if instance call)
// RBX: ICData
// RSP[0]: return address
// Control flow:
// - If receiver is null -> jump to IC miss.
// - If receiver is Smi -> load Smi class.
// - If receiver is not-Smi -> load receiver's class.
// - Check if 'num_args' (including receiver) match any IC data group.
// - Match found -> jump to target.
// - Match not found -> jump to IC miss.
void StubCodeCompiler::GenerateNArgsCheckInlineCacheStub(
intptr_t num_args,
const RuntimeEntry& handle_ic_miss,
Token::Kind kind,
Optimized optimized,
CallType type,
Exactness exactness) {
if (FLAG_precompiled_mode) {
__ Breakpoint();
return;
}
const bool save_entry_point = kind == Token::kILLEGAL;
if (save_entry_point) {
GenerateRecordEntryPoint(assembler);
}
if (optimized == kOptimized) {
GenerateOptimizedUsageCounterIncrement();
} else {
GenerateUsageCounterIncrement(/* scratch */ RCX);
}
ASSERT(num_args == 1 || num_args == 2);
#if defined(DEBUG)
{
Label ok;
// Check that the IC data array has NumArgsTested() == num_args.
// 'NumArgsTested' is stored in the least significant bits of 'state_bits'.
__ movl(RCX, FieldAddress(RBX, target::ICData::state_bits_offset()));
ASSERT(target::ICData::NumArgsTestedShift() == 0); // No shift needed.
__ andq(RCX, Immediate(target::ICData::NumArgsTestedMask()));
__ cmpq(RCX, Immediate(num_args));
__ j(EQUAL, &ok, Assembler::kNearJump);
__ Stop("Incorrect stub for IC data");
__ Bind(&ok);
}
#endif // DEBUG
#if !defined(PRODUCT)
Label stepping, done_stepping;
if (optimized == kUnoptimized) {
__ Comment("Check single stepping");
__ cmpb(Address(THR, target::Thread::single_step_offset()), Immediate(0));
__ j(NOT_EQUAL, &stepping);
__ Bind(&done_stepping);
}
#endif
Label not_smi_or_overflow;
if (kind != Token::kILLEGAL) {
EmitFastSmiOp(assembler, kind, num_args, &not_smi_or_overflow);
}
__ Bind(&not_smi_or_overflow);
__ Comment("Extract ICData initial values and receiver cid");
// RBX: IC data object (preserved).
__ movq(R13, FieldAddress(RBX, target::ICData::entries_offset()));
// R13: ic_data_array with check entries: classes and target functions.
__ leaq(R13, FieldAddress(R13, target::Array::data_offset()));
// R13: points directly to the first ic data array element.
if (type == kInstanceCall) {
__ LoadTaggedClassIdMayBeSmi(RAX, RDX);
__ movq(
ARGS_DESC_REG,
FieldAddress(RBX, target::CallSiteData::arguments_descriptor_offset()));
if (num_args == 2) {
__ OBJ(mov)(RCX,
FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::count_offset()));
__ movq(R9, Address(RSP, RCX, TIMES_4, -target::kWordSize));
__ LoadTaggedClassIdMayBeSmi(RCX, R9);
}
} else {
__ movq(
ARGS_DESC_REG,
FieldAddress(RBX, target::CallSiteData::arguments_descriptor_offset()));
__ OBJ(mov)(RCX, FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::count_offset()));
__ movq(RDX, Address(RSP, RCX, TIMES_4, 0));
__ LoadTaggedClassIdMayBeSmi(RAX, RDX);
if (num_args == 2) {
__ movq(R9, Address(RSP, RCX, TIMES_4, -target::kWordSize));
__ LoadTaggedClassIdMayBeSmi(RCX, R9);
}
}
// RAX: first argument class ID as Smi.
// RCX: second argument class ID as Smi.
// R10: args descriptor
// Loop that checks if there is an IC data match.
Label loop, found, miss;
__ Comment("ICData loop");
// We unroll the generic one that is generated once more than the others.
const bool optimize = kind == Token::kILLEGAL;
const intptr_t target_offset =
target::ICData::TargetIndexFor(num_args) * target::kCompressedWordSize;
const intptr_t count_offset =
target::ICData::CountIndexFor(num_args) * target::kCompressedWordSize;
const intptr_t exactness_offset =
target::ICData::ExactnessIndexFor(num_args) * target::kCompressedWordSize;
__ Bind(&loop);
for (int unroll = optimize ? 4 : 2; unroll >= 0; unroll--) {
Label update;
__ OBJ(mov)(R9, Address(R13, 0));
__ cmpq(RAX, R9); // Class id match?
if (num_args == 2) {
__ j(NOT_EQUAL, &update); // Continue.
__ OBJ(mov)(R9, Address(R13, target::kCompressedWordSize));
// R9: next class ID to check (smi).
__ cmpq(RCX, R9); // Class id match?
}
__ j(EQUAL, &found); // Break.
__ Bind(&update);
const intptr_t entry_size = target::ICData::TestEntryLengthFor(
num_args, exactness == kCheckExactness) *
target::kCompressedWordSize;
__ addq(R13, Immediate(entry_size)); // Next entry.
__ cmpq(R9, Immediate(target::ToRawSmi(kIllegalCid))); // Done?
if (unroll == 0) {
__ j(NOT_EQUAL, &loop);
} else {
__ j(EQUAL, &miss);
}
}
__ Bind(&miss);
__ Comment("IC miss");
// Compute address of arguments (first read number of arguments from
// arguments descriptor array and then compute address on the stack).
__ OBJ(mov)(RAX, FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::count_offset()));
__ leaq(RAX, Address(RSP, RAX, TIMES_4, 0)); // RAX is Smi.
__ EnterStubFrame();
if (save_entry_point) {
__ SmiTag(R8); // Entry-point offset is not Smi.
__ pushq(R8); // Preserve entry point.
}
__ pushq(ARGS_DESC_REG); // Preserve arguments descriptor array.
__ pushq(RBX); // Preserve IC data object.
__ pushq(Immediate(0)); // Result slot.
// Push call arguments.
for (intptr_t i = 0; i < num_args; i++) {
__ movq(RCX, Address(RAX, -target::kWordSize * i));
__ pushq(RCX);
}
__ pushq(RBX); // Pass IC data object.
__ CallRuntime(handle_ic_miss, num_args + 1);
// Remove the call arguments pushed earlier, including the IC data object.
for (intptr_t i = 0; i < num_args + 1; i++) {
__ popq(RAX);
}
__ popq(FUNCTION_REG); // Pop returned function object into RAX.
__ popq(RBX); // Restore IC data array.
__ popq(ARGS_DESC_REG); // Restore arguments descriptor array.
if (save_entry_point) {
__ popq(R8); // Restore entry point.
__ SmiUntag(R8); // Entry-point offset is not Smi.
}
__ RestoreCodePointer();
__ LeaveStubFrame();
Label call_target_function;
if (FLAG_precompiled_mode) {
GenerateDispatcherCode(assembler, &call_target_function);
} else {
__ jmp(&call_target_function);
}
__ Bind(&found);
// R13: Pointer to an IC data check group.
Label call_target_function_through_unchecked_entry;
if (exactness == kCheckExactness) {
Label exactness_ok;
ASSERT(num_args == 1);
__ OBJ(mov)(RAX, Address(R13, exactness_offset));
__ OBJ(cmp)(RAX,
Immediate(target::ToRawSmi(
StaticTypeExactnessState::HasExactSuperType().Encode())));
__ j(LESS, &exactness_ok);
__ j(EQUAL, &call_target_function_through_unchecked_entry);
// Check trivial exactness.
// Note: UntaggedICData::receivers_static_type_ is guaranteed to be not null
// because we only emit calls to this stub when it is not null.
__ movq(RCX,
FieldAddress(RBX, target::ICData::receivers_static_type_offset()));
__ LoadCompressed(RCX, FieldAddress(RCX, target::Type::arguments_offset()));
// RAX contains an offset to type arguments in words as a smi,
// hence TIMES_4. RDX is guaranteed to be non-smi because it is expected
// to have type arguments.
#if defined(DART_COMPRESSED_POINTERS)
__ movsxd(RAX, RAX);
#endif
__ OBJ(cmp)(RCX,
FieldAddress(RDX, RAX, TIMES_COMPRESSED_HALF_WORD_SIZE, 0));
__ j(EQUAL, &call_target_function_through_unchecked_entry);
// Update exactness state (not-exact anymore).
__ OBJ(mov)(Address(R13, exactness_offset),
Immediate(target::ToRawSmi(
StaticTypeExactnessState::NotExact().Encode())));
__ Bind(&exactness_ok);
}
__ LoadCompressed(FUNCTION_REG, Address(R13, target_offset));
if (FLAG_optimization_counter_threshold >= 0) {
__ Comment("Update ICData counter");
// Ignore overflow.
__ OBJ(add)(Address(R13, count_offset), Immediate(target::ToRawSmi(1)));
}
__ Comment("Call target (via specified entry point)");
__ Bind(&call_target_function);
// RAX: Target function.
__ LoadCompressed(
CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
if (save_entry_point) {
__ addq(R8, RAX);
__ jmp(Address(R8, 0));
} else {
__ jmp(FieldAddress(FUNCTION_REG, target::Function::entry_point_offset()));
}
if (exactness == kCheckExactness) {
__ Bind(&call_target_function_through_unchecked_entry);
if (FLAG_optimization_counter_threshold >= 0) {
__ Comment("Update ICData counter");
// Ignore overflow.
__ addq(Address(R13, count_offset), Immediate(target::ToRawSmi(1)));
}
__ Comment("Call target (via unchecked entry point)");
__ LoadCompressed(FUNCTION_REG, Address(R13, target_offset));
__ LoadCompressed(
CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
__ jmp(FieldAddress(FUNCTION_REG, target::Function::entry_point_offset(
CodeEntryKind::kUnchecked)));
}
#if !defined(PRODUCT)
if (optimized == kUnoptimized) {
__ Bind(&stepping);
__ EnterStubFrame();
if (type == kInstanceCall) {
__ pushq(RDX); // Preserve receiver.
}
__ pushq(RBX); // Preserve ICData.
if (save_entry_point) {
__ SmiTag(R8); // Entry-point offset is not Smi.
__ pushq(R8); // Preserve entry point.
}
__ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
if (save_entry_point) {
__ popq(R8); // Restore entry point.
__ SmiUntag(R8);
}
__ popq(RBX); // Restore ICData.
if (type == kInstanceCall) {
__ popq(RDX); // Restore receiver.
}
__ RestoreCodePointer();
__ LeaveStubFrame();
__ jmp(&done_stepping);
}
#endif
}
// RDX: receiver
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateOneArgCheckInlineCacheStub() {
GenerateNArgsCheckInlineCacheStub(
1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL,
kUnoptimized, kInstanceCall, kIgnoreExactness);
}
// RDX: receiver
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateOneArgCheckInlineCacheWithExactnessCheckStub() {
GenerateNArgsCheckInlineCacheStub(
1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL,
kUnoptimized, kInstanceCall, kCheckExactness);
}
// RDX: receiver
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateTwoArgsCheckInlineCacheStub() {
GenerateNArgsCheckInlineCacheStub(
2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kILLEGAL,
kUnoptimized, kInstanceCall, kIgnoreExactness);
}
// RDX: receiver
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateSmiAddInlineCacheStub() {
GenerateNArgsCheckInlineCacheStub(
2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kADD, kUnoptimized,
kInstanceCall, kIgnoreExactness);
}
// RDX: receiver
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateSmiLessInlineCacheStub() {
GenerateNArgsCheckInlineCacheStub(
2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kLT, kUnoptimized,
kInstanceCall, kIgnoreExactness);
}
// RDX: receiver
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateSmiEqualInlineCacheStub() {
GenerateNArgsCheckInlineCacheStub(
2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kEQ, kUnoptimized,
kInstanceCall, kIgnoreExactness);
}
// RDX: receiver
// RBX: ICData
// RDI: Function
// RSP[0]: return address
void StubCodeCompiler::GenerateOneArgOptimizedCheckInlineCacheStub() {
GenerateNArgsCheckInlineCacheStub(
1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL, kOptimized,
kInstanceCall, kIgnoreExactness);
}
// RDX: receiver
// RBX: ICData
// RDI: Function
// RSP[0]: return address
void StubCodeCompiler::
GenerateOneArgOptimizedCheckInlineCacheWithExactnessCheckStub() {
GenerateNArgsCheckInlineCacheStub(
1, kInlineCacheMissHandlerOneArgRuntimeEntry, Token::kILLEGAL, kOptimized,
kInstanceCall, kCheckExactness);
}
// RDX: receiver
// RBX: ICData
// RDI: Function
// RSP[0]: return address
void StubCodeCompiler::GenerateTwoArgsOptimizedCheckInlineCacheStub() {
GenerateNArgsCheckInlineCacheStub(
2, kInlineCacheMissHandlerTwoArgsRuntimeEntry, Token::kILLEGAL,
kOptimized, kInstanceCall, kIgnoreExactness);
}
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateZeroArgsUnoptimizedStaticCallStub() {
GenerateRecordEntryPoint(assembler);
GenerateUsageCounterIncrement(/* scratch */ RCX);
#if defined(DEBUG)
{
Label ok;
// Check that the IC data array has NumArgsTested() == 0.
// 'NumArgsTested' is stored in the least significant bits of 'state_bits'.
__ movl(RCX, FieldAddress(RBX, target::ICData::state_bits_offset()));
ASSERT(target::ICData::NumArgsTestedShift() == 0); // No shift needed.
__ andq(RCX, Immediate(target::ICData::NumArgsTestedMask()));
__ cmpq(RCX, Immediate(0));
__ j(EQUAL, &ok, Assembler::kNearJump);
__ Stop("Incorrect IC data for unoptimized static call");
__ Bind(&ok);
}
#endif // DEBUG
#if !defined(PRODUCT)
// Check single stepping.
Label stepping, done_stepping;
__ cmpb(Address(THR, target::Thread::single_step_offset()), Immediate(0));
#if defined(DEBUG)
static auto const kJumpLength = Assembler::kFarJump;
#else
static auto const kJumpLength = Assembler::kNearJump;
#endif // DEBUG
__ j(NOT_EQUAL, &stepping, kJumpLength);
__ Bind(&done_stepping);
#endif
// RBX: IC data object (preserved).
__ movq(R12, FieldAddress(RBX, target::ICData::entries_offset()));
// R12: ic_data_array with entries: target functions and count.
__ leaq(R12, FieldAddress(R12, target::Array::data_offset()));
// R12: points directly to the first ic data array element.
const intptr_t target_offset =
target::ICData::TargetIndexFor(0) * target::kCompressedWordSize;
const intptr_t count_offset =
target::ICData::CountIndexFor(0) * target::kCompressedWordSize;
if (FLAG_optimization_counter_threshold >= 0) {
// Increment count for this call, ignore overflow.
__ OBJ(add)(Address(R12, count_offset), Immediate(target::ToRawSmi(1)));
}
// Load arguments descriptor into R10.
__ movq(
ARGS_DESC_REG,
FieldAddress(RBX, target::CallSiteData::arguments_descriptor_offset()));
// Get function and call it, if possible.
__ LoadCompressed(FUNCTION_REG, Address(R12, target_offset));
__ LoadCompressed(
CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
__ addq(R8, FUNCTION_REG);
__ jmp(Address(R8, 0));
#if !defined(PRODUCT)
__ Bind(&stepping);
__ EnterStubFrame();
__ pushq(RBX); // Preserve IC data object.
__ SmiTag(R8); // Entry-point is not Smi.
__ pushq(R8); // Preserve entry-point.
__ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
__ popq(R8); // Restore entry-point.
__ SmiUntag(R8);
__ popq(RBX);
__ RestoreCodePointer();
__ LeaveStubFrame();
__ jmp(&done_stepping, Assembler::kNearJump);
#endif
}
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateOneArgUnoptimizedStaticCallStub() {
GenerateNArgsCheckInlineCacheStub(1, kStaticCallMissHandlerOneArgRuntimeEntry,
Token::kILLEGAL, kUnoptimized, kStaticCall,
kIgnoreExactness);
}
// RBX: ICData
// RSP[0]: return address
void StubCodeCompiler::GenerateTwoArgsUnoptimizedStaticCallStub() {
GenerateNArgsCheckInlineCacheStub(
2, kStaticCallMissHandlerTwoArgsRuntimeEntry, Token::kILLEGAL,
kUnoptimized, kStaticCall, kIgnoreExactness);
}
// Stub for compiling a function and jumping to the compiled code.
// ARGS_DESC_REG: Arguments descriptor.
// FUNCTION_REG: Function.
void StubCodeCompiler::GenerateLazyCompileStub() {
__ EnterStubFrame();
__ pushq(ARGS_DESC_REG); // Preserve arguments descriptor array.
__ pushq(FUNCTION_REG); // Pass function.
__ CallRuntime(kCompileFunctionRuntimeEntry, 1);
__ popq(FUNCTION_REG); // Restore function.
__ popq(ARGS_DESC_REG); // Restore arguments descriptor array.
__ LeaveStubFrame();
__ LoadCompressed(
CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
__ movq(RCX,
FieldAddress(FUNCTION_REG, target::Function::entry_point_offset()));
__ jmp(RCX);
}
// Stub for interpreting a function call.
// ARGS_DESC_REG: Arguments descriptor.
// FUNCTION_REG: Function.
void StubCodeCompiler::GenerateInterpretCallStub() {
#if defined(DART_BYTECODE_INTERPRETER)
__ EnterStubFrame();
#if defined(DEBUG)
{
Label ok;
// Check that we are always entering from Dart code.
__ movq(R8, Immediate(VMTag::kDartTagId));
__ cmpq(R8, Assembler::VMTagAddress());
__ j(EQUAL, &ok, Assembler::kNearJump);
__ Stop("Not coming from Dart code.");
__ Bind(&ok);
}
#endif
// Adjust arguments count for type arguments vector.
__ OBJ(mov)(R11, FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::count_offset()));
__ SmiUntag(R11);
__ OBJ(cmp)(FieldAddress(ARGS_DESC_REG,
target::ArgumentsDescriptor::type_args_len_offset()),
Immediate(0));
Label args_count_ok;
__ j(EQUAL, &args_count_ok, Assembler::kNearJump);
__ addq(R11, Immediate(1));
__ Bind(&args_count_ok);
// Compute argv.
__ leaq(R12,
Address(RBP, R11, TIMES_8,
target::frame_layout.param_end_from_fp * target::kWordSize));
// Indicate decreasing memory addresses of arguments with negative argc.
__ negq(R11);
// Reserve shadow space for args and align frame before entering C++ world.
__ subq(RSP, Immediate(5 * target::kWordSize));
if (OS::ActivationFrameAlignment() > 1) {
__ andq(RSP, Immediate(~(OS::ActivationFrameAlignment() - 1)));
}
__ movq(CallingConventions::kArg1Reg, FUNCTION_REG); // Function.
__ movq(CallingConventions::kArg2Reg,
ARGS_DESC_REG); // Arguments descriptor.
__ movq(CallingConventions::kArg3Reg, R11); // Negative argc.
__ movq(CallingConventions::kArg4Reg, R12); // Argv.
#if defined(DART_TARGET_OS_WINDOWS)
__ movq(Address(RSP, 0 * target::kWordSize), THR); // Thread.
#else
__ movq(CallingConventions::kArg5Reg, THR); // Thread.
#endif
// Save exit frame information to enable stack walking as we are about
// to transition to Dart VM C++ code.
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()), RBP);
// Mark that the thread exited generated code through a runtime call.
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(target::Thread::exit_through_runtime_call()));
// Mark that the thread is executing VM code.
__ movq(RAX,
Address(THR, target::Thread::interpret_call_entry_point_offset()));
__ movq(Assembler::VMTagAddress(), RAX);
__ call(RAX);
// Mark that the thread is executing Dart code.
__ movq(Assembler::VMTagAddress(), Immediate(VMTag::kDartTagId));
// Mark that the thread has not exited generated Dart code.
__ movq(Address(THR, target::Thread::exit_through_ffi_offset()),
Immediate(0));
// Reset exit frame information in Isolate's mutator thread structure.
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()),
Immediate(0));
__ LeaveStubFrame();
__ ret();
#else
__ Stop("Not using Dart dynamic modules");
#endif // defined(DART_BYTECODE_INTERPRETER)
}
// RBX: Contains an ICData.
// TOS(0): return address (Dart code).
void StubCodeCompiler::GenerateICCallBreakpointStub() {
#if defined(PRODUCT)
__ Stop("No debugging in PRODUCT mode");
#else
__ EnterStubFrame();
__ pushq(RDX); // Preserve receiver.
__ pushq(RBX); // Preserve IC data.
__ pushq(Immediate(0)); // Result slot.
__ CallRuntime(kBreakpointRuntimeHandlerRuntimeEntry, 0);
__ popq(CODE_REG); // Original stub.
__ popq(RBX); // Restore IC data.
__ popq(RDX); // Restore receiver.
__ LeaveStubFrame();
__ movq(RAX, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
__ jmp(RAX); // Jump to original stub.
#endif // defined(PRODUCT)
}
void StubCodeCompiler::GenerateUnoptStaticCallBreakpointStub() {
#if defined(PRODUCT)
__ Stop("No debugging in PRODUCT mode");
#else
__ EnterStubFrame();
__ pushq(RBX); // Preserve IC data.
__ pushq(Immediate(0)); // Result slot.
__ CallRuntime(kBreakpointRuntimeHandlerRuntimeEntry, 0);
__ popq(CODE_REG); // Original stub.
__ popq(RBX); // Restore IC data.
__ LeaveStubFrame();
__ movq(RAX, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
__ jmp(RAX); // Jump to original stub.
#endif // defined(PRODUCT)
}
// TOS(0): return address (Dart code).
void StubCodeCompiler::GenerateRuntimeCallBreakpointStub() {
#if defined(PRODUCT)
__ Stop("No debugging in PRODUCT mode");
#else
__ EnterStubFrame();
__ pushq(Immediate(0)); // Result slot.
__ CallRuntime(kBreakpointRuntimeHandlerRuntimeEntry, 0);
__ popq(CODE_REG); // Original stub.
__ LeaveStubFrame();
__ movq(RAX, FieldAddress(CODE_REG, target::Code::entry_point_offset()));
__ jmp(RAX); // Jump to original stub.
#endif // defined(PRODUCT)
}
// Called only from unoptimized code.
void StubCodeCompiler::GenerateDebugStepCheckStub() {
#if defined(PRODUCT)
__ Stop("No debugging in PRODUCT mode");
#else
// Check single stepping.
Label stepping, done_stepping;
__ cmpb(Address(THR, target::Thread::single_step_offset()), Immediate(0));
__ j(NOT_EQUAL, &stepping, Assembler::kNearJump);
__ Bind(&done_stepping);
__ ret();
__ Bind(&stepping);
__ EnterStubFrame();
__ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
__ LeaveStubFrame();
__ jmp(&done_stepping, Assembler::kNearJump);
#endif // defined(PRODUCT)
}
// Used to check class and type arguments. Arguments passed in registers:
//
// Input registers (all preserved, from TypeTestABI struct):
// - kSubtypeTestCacheReg: UntaggedSubtypeTestCache
// - kInstanceReg: instance to test against (must be preserved).
// - kDstTypeReg: destination type (for n>=7).
// - kInstantiatorTypeArgumentsReg : instantiator type arguments (for n>=3).
// - kFunctionTypeArgumentsReg : function type arguments (for n>=4).
// Inputs from stack:
// - TOS + 0: return address.
//
// Outputs (from TypeTestABI struct):
// - kSubtypeTestCacheResultReg: the cached result, or null if not found.
void StubCodeCompiler::GenerateSubtypeNTestCacheStub(Assembler* assembler,
int n) {
ASSERT(n >= 1);
ASSERT(n <= SubtypeTestCache::kMaxInputs);
// If we need the parent function type arguments for a closure, we also need
// the delayed type arguments, so this case will never happen.
ASSERT(n != 5);
RegisterSet saved_registers;
// Until we have the result, we use the result register to store the null
// value for quick access. This has the side benefit of initializing the
// result to null, so it only needs to be changed if found.
const Register kNullReg = TypeTestABI::kSubtypeTestCacheResultReg;
__ LoadObject(kNullReg, NullObject());
// Free up additional registers needed for checks in the loop. Initially
// define them as kNoRegister so any unexpected uses are caught.
Register kInstanceParentFunctionTypeArgumentsReg = kNoRegister;
if (n >= 5) {
kInstanceParentFunctionTypeArgumentsReg = PP;
saved_registers.AddRegister(kInstanceParentFunctionTypeArgumentsReg);
}
Register kInstanceDelayedFunctionTypeArgumentsReg = kNoRegister;
if (n >= 6) {
kInstanceDelayedFunctionTypeArgumentsReg = CODE_REG;
saved_registers.AddRegister(kInstanceDelayedFunctionTypeArgumentsReg);
}
// We'll replace these with actual registers if possible, but fall back to
// the stack if register pressure is too great. The last two values are
// used in every loop iteration, and so are more important to put in
// registers if possible, whereas the first is used only when we go off
// the end of the backing array (usually at most once per check).
Register kCacheContentsSizeReg = kNoRegister;
if (n < 5) {
// Use the register we would have used for the parent function type args.
kCacheContentsSizeReg = PP;
saved_registers.AddRegister(kCacheContentsSizeReg);
}
Register kProbeDistanceReg = kNoRegister;
if (n < 6) {
// Use the register we would have used for the delayed type args.
kProbeDistanceReg = CODE_REG;
saved_registers.AddRegister(kProbeDistanceReg);
}
Register kCacheEntryEndReg = kNoRegister;
if (n < 2) {
// This register isn't in use and doesn't require saving/restoring.
kCacheEntryEndReg = STCInternalRegs::kInstanceInstantiatorTypeArgumentsReg;
} else if (n < 7) {
// Use the destination type, as that is the last input that might be unused.
kCacheEntryEndReg = TypeTestABI::kDstTypeReg;
saved_registers.AddRegister(TypeTestABI::kDstTypeReg);
}
__ PushRegisters(saved_registers);
Label done;
GenerateSubtypeTestCacheSearch(
assembler, n, kNullReg, STCInternalRegs::kCacheEntryReg,
STCInternalRegs::kInstanceCidOrSignatureReg,
STCInternalRegs::kInstanceInstantiatorTypeArgumentsReg,
kInstanceParentFunctionTypeArgumentsReg,
kInstanceDelayedFunctionTypeArgumentsReg, kCacheEntryEndReg,
kCacheContentsSizeReg, kProbeDistanceReg,
[&](Assembler* assembler, int n) {
__ LoadCompressed(TypeTestABI::kSubtypeTestCacheResultReg,
Address(STCInternalRegs::kCacheEntryReg,
target::kCompressedWordSize *
target::SubtypeTestCache::kTestResult));
__ PopRegisters(saved_registers);
__ Ret();
},
[&](Assembler* assembler, int n) {
// We initialize kSubtypeTestCacheResultReg to null so it can be used
// for null checks, so the result value is already set.
__ PopRegisters(saved_registers);
__ Ret();
});
}
// Jump to a frame on the call stack.
// TOS + 0: return address
// Arg1: program counter
// Arg2: stack pointer
// Arg3: frame_pointer
// Arg4: thread
// No Result.
void StubCodeCompiler::GenerateJumpToFrameStub() {
__ movq(THR, CallingConventions::kArg4Reg);
if (FLAG_target_thread_sanitizer) {
Label again, done;
__ movq(CALLEE_SAVED_TEMP,
Address(THR, target::Thread::top_exit_frame_info_offset()));
// Skip CallToRuntime/CallNativeWithWrapper stub frame, which did not call
// __tsan_func_entry.
__ movq(CALLEE_SAVED_TEMP,
Address(CALLEE_SAVED_TEMP,
target::frame_layout.saved_caller_fp_from_fp *
target::kWordSize));
__ Bind(&again);
__ cmpq(CALLEE_SAVED_TEMP, CallingConventions::kArg3Reg);
__ j(EQUAL, &done);
__ TsanFuncExit();
__ movq(CALLEE_SAVED_TEMP,
Address(CALLEE_SAVED_TEMP,
target::frame_layout.saved_caller_fp_from_fp *
target::kWordSize));
__ jmp(&again);
__ Bind(&done);
}
__ movq(RBP, CallingConventions::kArg3Reg);
__ movq(RSP, CallingConventions::kArg2Reg);
#if defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif
// Set the tag.
__ movq(Assembler::VMTagAddress(), Immediate(VMTag::kDartTagId));
// Clear top exit frame.
__ movq(Address(THR, target::Thread::top_exit_frame_info_offset()),
Immediate(0));
// Restore the pool pointer.
__ RestoreCodePointer();
if (FLAG_precompiled_mode) {
__ movq(PP, Address(THR, target::Thread::global_object_pool_offset()));
} else {
__ LoadPoolPointer(PP);
}
__ jmp(CallingConventions::kArg1Reg); // Jump to program counter.
}
// Run an exception handler. Execution comes from JumpToFrame stub.
//
// The arguments are stored in the Thread object.
// No result.
static void GenerateRunExceptionHandler(Assembler* assembler,
bool unbox_exception) {
ASSERT(kExceptionObjectReg == RAX);
ASSERT(kStackTraceObjectReg == RDX);
__ movq(CallingConventions::kArg1Reg,
Address(THR, target::Thread::resume_pc_offset()));
word offset_from_thread = 0;
bool ok = target::CanLoadFromThread(NullObject(), &offset_from_thread);
ASSERT(ok);
__ movq(TMP, Address(THR, offset_from_thread));
// Load the exception from the current thread.
Address exception_addr(THR, target::Thread::active_exception_offset());
__ movq(kExceptionObjectReg, exception_addr);
__ movq(exception_addr, TMP);
if (unbox_exception) {
compiler::Label not_smi, done;
__ BranchIfNotSmi(kExceptionObjectReg, &not_smi,
compiler::Assembler::kNearJump);
__ SmiUntagAndSignExtend(kExceptionObjectReg);
__ jmp(&done, compiler::Assembler::kNearJump);
__ Bind(&not_smi);
__ movq(kExceptionObjectReg,
compiler::FieldAddress(kExceptionObjectReg, Mint::value_offset()));
__ Bind(&done);
}
// Load the stacktrace from the current thread.
Address stacktrace_addr(THR, target::Thread::active_stacktrace_offset());
__ movq(kStackTraceObjectReg, stacktrace_addr);
__ movq(stacktrace_addr, TMP);
__ jmp(CallingConventions::kArg1Reg); // Jump to continuation point.
}
void StubCodeCompiler::GenerateRunExceptionHandlerStub() {
GenerateRunExceptionHandler(assembler, false);
}
void StubCodeCompiler::GenerateRunExceptionHandlerUnboxStub() {
GenerateRunExceptionHandler(assembler, true);
}
// Deoptimize a frame on the call stack before rewinding.
// The arguments are stored in the Thread object.
// No result.
void StubCodeCompiler::GenerateDeoptForRewindStub() {
// Push zap value instead of CODE_REG.
__ pushq(Immediate(kZapCodeReg));
// Push the deopt pc.
__ pushq(Address(THR, target::Thread::resume_pc_offset()));
#if defined(USING_SHADOW_CALL_STACK)
#error Unimplemented
#endif
GenerateDeoptimizationSequence(assembler, kEagerDeopt);
// After we have deoptimized, jump to the correct frame.
__ EnterStubFrame();
__ CallRuntime(kRewindPostDeoptRuntimeEntry, 0);
__ LeaveStubFrame();
__ int3();
}
// Calls to the runtime to optimize the given function.
// RDI: function to be reoptimized.
// ARGS_DESC_REG: argument descriptor (preserved).
void StubCodeCompiler::GenerateOptimizeFunctionStub() {
__ movq(CODE_REG, Address(THR, target::Thread::optimize_stub_offset()));
__ EnterStubFrame();
__ pushq(ARGS_DESC_REG); // Preserve args descriptor.
__ pushq(Immediate(0)); // Result slot.
__ pushq(RDI); // Arg0: function to optimize
__ CallRuntime(kOptimizeInvokedFunctionRuntimeEntry, 1);
__ popq(RAX); // Discard argument.
__ popq(FUNCTION_REG); // Get Function object.
__ popq(ARGS_DESC_REG); // Restore argument descriptor.
__ LeaveStubFrame();
__ LoadCompressed(
CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
__ movq(RCX,
FieldAddress(FUNCTION_REG, target::Function::entry_point_offset()));
__ jmp(RCX);
__ int3();
}
// Does identical check (object references are equal or not equal) with special
// checks for boxed numbers.
// Left and right are pushed on stack.
// Return ZF set.
// Note: A Mint cannot contain a value that would fit in Smi.
static void GenerateIdenticalWithNumberCheckStub(Assembler* assembler,
const Register left,
const Register right) {
Label reference_compare, done, check_mint;
// If any of the arguments is Smi do reference compare.
__ testq(left, Immediate(kSmiTagMask));
__ j(ZERO, &reference_compare);
__ testq(right, Immediate(kSmiTagMask));
__ j(ZERO, &reference_compare);
// Value compare for two doubles.
__ CompareClassId(left, kDoubleCid);
__ j(NOT_EQUAL, &check_mint, Assembler::kNearJump);
__ CompareClassId(right, kDoubleCid);
__ j(NOT_EQUAL, &done, Assembler::kFarJump);
// Double values bitwise compare.
__ movq(left, FieldAddress(left, target::Double::value_offset()));
__ cmpq(left, FieldAddress(right, target::Double::value_offset()));
__ jmp(&done, Assembler::kFarJump);
__ Bind(&check_mint);
__ CompareClassId(left, kMintCid);
__ j(NOT_EQUAL, &reference_compare, Assembler::kNearJump);
__ CompareClassId(right, kMintCid);
__ j(NOT_EQUAL, &done, Assembler::kFarJump);
__ movq(left, FieldAddress(left, target::Mint::value_offset()));
__ cmpq(left, FieldAddress(right, target::Mint::value_offset()));
__ jmp(&done, Assembler::kFarJump);
__ Bind(&reference_compare);
__ CompareObjectRegisters(left, right);
__ Bind(&done);
}
// Called only from unoptimized code. All relevant registers have been saved.
// TOS + 0: return address
// TOS + 1: right argument.
// TOS + 2: left argument.
// Returns ZF set.
void StubCodeCompiler::GenerateUnoptimizedIdenticalWithNumberCheckStub() {
#if !defined(PRODUCT)
// Check single stepping.
Label stepping, done_stepping;
__ cmpb(Address(THR, target::Thread::single_step_offset()), Immediate(0));
__ j(NOT_EQUAL, &stepping);
__ Bind(&done_stepping);
#endif
const Register left = RAX;
const Register right = RDX;
__ movq(left, Address(RSP, 2 * target::kWordSize));
__ movq(right, Address(RSP, 1 * target::kWordSize));
GenerateIdenticalWithNumberCheckStub(assembler, left, right);
__ ret();
#if !defined(PRODUCT)
__ Bind(&stepping);
__ EnterStubFrame();
__ CallRuntime(kSingleStepHandlerRuntimeEntry, 0);
__ RestoreCodePointer();
__ LeaveStubFrame();
__ jmp(&done_stepping);
#endif
}
// Called from optimized code only.
// TOS + 0: return address
// TOS + 1: right argument.
// TOS + 2: left argument.
// Returns ZF set.
void StubCodeCompiler::GenerateOptimizedIdenticalWithNumberCheckStub() {
const Register left = RAX;
const Register right = RDX;
__ movq(left, Address(RSP, 2 * target::kWordSize));
__ movq(right, Address(RSP, 1 * target::kWordSize));
GenerateIdenticalWithNumberCheckStub(assembler, left, right);
__ ret();
}
// Called from megamorphic calls.
// RDX: receiver (passed to target)
// IC_DATA_REG: target::MegamorphicCache (preserved)
// Passed to target:
// FUNCTION_REG: target function
// CODE_REG: target Code
// ARGS_DESC_REG: arguments descriptor
void StubCodeCompiler::GenerateMegamorphicCallStub() {
// Jump if receiver is a smi.
Label smi_case;
__ testq(RDX, Immediate(kSmiTagMask));
// Jump out of line for smi case.
__ j(ZERO, &smi_case, Assembler::kNearJump);
// Loads the cid of the object.
__ LoadClassId(RAX, RDX);
Label cid_loaded;
__ Bind(&cid_loaded);
__ movq(R9,
FieldAddress(IC_DATA_REG, target::MegamorphicCache::mask_offset()));
__ movq(RDI, FieldAddress(IC_DATA_REG,
target::MegamorphicCache::buckets_offset()));
// R9: mask as a smi - load first to support insertion w/o stopping Dart code.
// RDI: cache buckets array.
// Tag cid as a smi.
__ addq(RAX, RAX);
// Compute the table index.
ASSERT(target::MegamorphicCache::kSpreadFactor == 7);
// Use leaq and subq multiply with 7 == 8 - 1.
__ leaq(RCX, Address(RAX, TIMES_8, 0));
__ subq(RCX, RAX);
Label loop;
__ Bind(&loop);
__ andq(RCX, R9);
const intptr_t base = target::Array::data_offset();
// RCX is smi tagged, but table entries are two words, so TIMES_8.
Label probe_failed;
__ OBJ(cmp)(RAX, FieldAddress(RDI, RCX, TIMES_COMPRESSED_WORD_SIZE, base));
__ j(NOT_EQUAL, &probe_failed, Assembler::kNearJump);
Label load_target;
__ Bind(&load_target);
// Call the target found in the cache. For a class id match, this is a
// proper target for the given name and arguments descriptor. If the
// illegal class id was found, the target is a cache miss handler that can
// be invoked as a normal Dart function.
__ LoadCompressed(FUNCTION_REG,
FieldAddress(RDI, RCX, TIMES_COMPRESSED_WORD_SIZE,
base + target::kCompressedWordSize));
__ movq(ARGS_DESC_REG,
FieldAddress(IC_DATA_REG,
target::CallSiteData::arguments_descriptor_offset()));
__ movq(RCX,
FieldAddress(FUNCTION_REG, target::Function::entry_point_offset()));
if (!FLAG_precompiled_mode) {
__ LoadCompressed(
CODE_REG, FieldAddress(FUNCTION_REG, target::Function::code_offset()));
}
__ jmp(RCX);
// Probe failed, check if it is a miss.
__ Bind(&probe_failed);
__ OBJ(cmp)(FieldAddress(RDI, RCX, TIMES_COMPRESSED_WORD_SIZE, base),
Immediate(target::ToRawSmi(kIllegalCid)));
Label miss;
__ j(ZERO, &miss, Assembler::kNearJump);
// Try next entry in the table.
__ AddImmediate(RCX, Immediate(target::ToRawSmi(1)));
__ jmp(&loop);
// Load cid for the Smi case.
__ Bind(&smi_case);
__ movq(RAX, Immediate(kSmiCid));
__ jmp(&cid_loaded);
__ Bind(&miss);
GenerateSwitchableCallMissStub();
}
// Input:
// IC_DATA_REG - icdata
// RDX - receiver object
void StubCodeCompiler::GenerateICCallThroughCodeStub() {
Label loop, found, miss;
__ movq(R13, FieldAddress(IC_DATA_REG, target::ICData::entries_offset()));
__ movq(ARGS_DESC_REG,
FieldAddress(IC_DATA_REG,
target::CallSiteData::arguments_descriptor_offset()));
__ leaq(R13, FieldAddress(R13, target::Array::data_offset()));
// R13: first IC entry
__ LoadTaggedClassIdMayBeSmi(RAX, RDX);
// RAX: receiver cid as Smi
__ Bind(&loop);
__ OBJ(mov)(R9, Address(R13, 0));
__ OBJ(cmp)(RAX, R9);
__ j(EQUAL, &found, Assembler::kNearJump);
ASSERT(target::ToRawSmi(kIllegalCid) == 0);
__ OBJ(test)(R9, R9);
__ j(ZERO, &miss, Assembler::kNearJump);
const intptr_t entry_length =
target::ICData::TestEntryLengthFor(1, /*exactness_check=*/false) *
target::kCompressedWordSize;
__ addq(R13, Immediate(entry_length)); // Next entry.
__ jmp(&loop);
__ Bind(&found);
if (FLAG_precompiled_mode) {
const intptr_t entry_offset =
target::ICData::EntryPointIndexFor(1) * target::kCompressedWordSize;
__ LoadCompressed(FUNCTION_REG, Address(R13, entry_offset));
__ jmp(FieldAddress(FUNCTION_REG, target::Function::entry_point_offset()));
} else {
const intptr_t code_offset =
target::ICData::CodeIndexFor(1) * target::kCompressedWordSize;
__ LoadCompressed(CODE_REG, Address(R13, code_offset));
__ jmp(FieldAddress(CODE_REG, target::Code::entry_point_offset()));
}
__ Bind(&miss);
__ jmp(Address(THR, target::Thread::switchable_call_miss_entry_offset()));
}
void StubCodeCompiler::GenerateMonomorphicSmiableCheckStub() {
Label have_cid, miss;
__ movq(RAX, Immediate(kSmiCid));
__ movzxw(
RCX,
FieldAddress(RBX, target::MonomorphicSmiableCall::expected_cid_offset()));
__ testq(RDX, Immediate(kSmiTagMask));
__ j(ZERO, &have_cid, Assembler::kNearJump);
__ LoadClassId(RAX, RDX);
__ Bind(&have_cid);
__ cmpq(RAX, RCX);
__ j(NOT_EQUAL, &miss, Assembler::kNearJump);
// Note: this stub is only used in AOT mode, hence the direct (bare) call.
__ jmp(
FieldAddress(RBX, target::MonomorphicSmiableCall::entrypoint_offset()));
__ Bind(&miss);
__ jmp(Address(THR, target::Thread::switchable_call_miss_entry_offset()));
}
// Called from switchable IC calls.
// RDX: receiver
void StubCodeCompiler::GenerateSwitchableCallMissStub() {
__ movq(CODE_REG,
Address(THR, target::Thread::switchable_call_miss_stub_offset()));
__ EnterStubFrame();
__ pushq(RDX); // Preserve receiver.
__ pushq(Immediate(0)); // Result slot.
__ pushq(Immediate(0)); // Arg0: stub out.
__ pushq(RDX); // Arg1: Receiver
__ CallRuntime(kSwitchableCallMissRuntimeEntry, 2);
__ popq(RBX);
__ popq(CODE_REG); // result = stub
__ popq(RBX); // result = IC
__ popq(RDX); // Restore receiver.
__ LeaveStubFrame();
__ movq(RCX, FieldAddress(CODE_REG, target::Code::entry_point_offset(
CodeEntryKind::kNormal)));
__ jmp(RCX);
}
// Called from switchable IC calls.
// RDX: receiver
// RBX: SingleTargetCache
// Passed to target::
// CODE_REG: target Code object
void StubCodeCompiler::GenerateSingleTargetCallStub() {
Label miss;
__ LoadClassIdMayBeSmi(RAX, RDX);
__ movzxw(R9,
FieldAddress(RBX, target::SingleTargetCache::lower_limit_offset()));
__ movzxw(R10,
FieldAddress(RBX, target::SingleTargetCache::upper_limit_offset()));
__ cmpq(RAX, R9);
__ j(LESS, &miss, Assembler::kNearJump);
__ cmpq(RAX, R10);
__ j(GREATER, &miss, Assembler::kNearJump);
__ movq(RCX,
FieldAddress(RBX, target::SingleTargetCache::entry_point_offset()));
__ movq(CODE_REG,
FieldAddress(RBX, target::SingleTargetCache::target_offset()));
__ jmp(RCX);
__ Bind(&miss);
__ EnterStubFrame();
__ pushq(RDX); // Preserve receiver.
__ pushq(Immediate(0)); // Result slot.
__ pushq(Immediate(0)); // Arg0: stub out
__ pushq(RDX); // Arg1: Receiver
__ CallRuntime(kSwitchableCallMissRuntimeEntry, 2);
__ popq(RBX);
__ popq(CODE_REG); // result = stub
__ popq(RBX); // result = IC
__ popq(RDX); // Restore receiver.
__ LeaveStubFrame();
__ movq(RCX, FieldAddress(CODE_REG, target::Code::entry_point_offset(
CodeEntryKind::kMonomorphic)));
__ jmp(RCX);
}
static ScaleFactor GetScaleFactor(intptr_t size) {
switch (size) {
case 1:
return TIMES_1;
case 2:
return TIMES_2;
case 4:
return TIMES_4;
case 8:
return TIMES_8;
case 16:
return TIMES_16;
}
UNREACHABLE();
return static_cast<ScaleFactor>(0);
}
void StubCodeCompiler::GenerateAllocateTypedDataArrayStub(intptr_t cid) {
const intptr_t element_size = TypedDataElementSizeInBytes(cid);
const intptr_t max_len = TypedDataMaxNewSpaceElements(cid);
ScaleFactor scale_factor = GetScaleFactor(element_size);
COMPILE_ASSERT(AllocateTypedDataArrayABI::kLengthReg == RAX);
COMPILE_ASSERT(AllocateTypedDataArrayABI::kResultReg == RAX);
if (!FLAG_use_slow_path && FLAG_inline_alloc) {
// Save length argument for possible runtime call, as
// RAX is clobbered.
Label call_runtime;
__ pushq(AllocateTypedDataArrayABI::kLengthReg);
NOT_IN_PRODUCT(__ MaybeTraceAllocation(cid, &call_runtime));
__ movq(RDI, AllocateTypedDataArrayABI::kLengthReg);
/* Check that length is a positive Smi. */
/* RDI: requested array length argument. */
__ testq(RDI, Immediate(kSmiTagMask));
__ j(NOT_ZERO, &call_runtime);
__ SmiUntag(RDI);
/* Check for length >= 0 && length <= max_len. */
/* RDI: untagged array length. */
__ cmpq(RDI, Immediate(max_len));
__ j(ABOVE, &call_runtime);
/* Special case for scaling by 16. */
if (scale_factor == TIMES_16) {
/* double length of array. */
__ addq(RDI, RDI);
/* only scale by 8. */
scale_factor = TIMES_8;
}
const intptr_t fixed_size_plus_alignment_padding =
target::TypedData::HeaderSize() +
target::ObjectAlignment::kObjectAlignment - 1;
__ leaq(RDI, Address(RDI, scale_factor, fixed_size_plus_alignment_padding));
__ andq(RDI, Immediate(-target::ObjectAlignment::kObjectAlignment));
__ movq(RAX, Address(THR, target::Thread::top_offset()));
__ movq(RCX, RAX);
/* RDI: allocation size. */
__ addq(RCX, RDI);
__ j(CARRY, &call_runtime);
/* Check if the allocation fits into the remaining space. */
/* RAX: potential new object start. */
/* RCX: potential next object start. */
/* RDI: allocation size. */
__ cmpq(RCX, Address(THR, target::Thread::end_offset()));
__ j(ABOVE_EQUAL, &call_runtime);
__ CheckAllocationCanary(RAX);
/* Successfully allocated the object(s), now update top to point to */
/* next object start and initialize the object. */
__ movq(Address(THR, target::Thread::top_offset()), RCX);
__ addq(RAX, Immediate(kHeapObjectTag));
/* Initialize the tags. */
/* RAX: new object start as a tagged pointer. */
/* RCX: new object end address. */
/* RDI: allocation size. */
/* R13: scratch register. */
{
Label size_tag_overflow, done;
__ cmpq(RDI, Immediate(target::UntaggedObject::kSizeTagMaxSizeTag));
__ j(ABOVE, &size_tag_overflow, Assembler::kNearJump);
__ shlq(RDI, Immediate(target::UntaggedObject::kSizeTagPos -
target::ObjectAlignment::kObjectAlignmentLog2));
__ jmp(&done, Assembler::kNearJump);
__ Bind(&size_tag_overflow);
__ LoadImmediate(RDI, Immediate(0));
__ Bind(&done);
/* Get the class index and insert it into the tags. */
uword tags =
target::MakeTagWordForNewSpaceObject(cid, /*instance_size=*/0);
__ orq(RDI, Immediate(tags));
__ InitializeHeader(RDI, RAX);
}
/* Set the length field. */
/* RAX: new object start as a tagged pointer. */
/* RCX: new object end address. */
__ popq(RDI); /* Array length. */
__ StoreCompressedIntoObjectNoBarrier(
RAX, FieldAddress(RAX, target::TypedDataBase::length_offset()), RDI);
/* Initialize all array elements to 0. */
/* RAX: new object start as a tagged pointer. */
/* RCX: new object end address. */
/* RDI: iterator which initially points to the start of the variable */
/* RBX: scratch register. */
/* data area to be initialized. */
__ pxor(XMM0, XMM0); /* Zero. */
__ leaq(RDI, FieldAddress(RAX, target::TypedData::HeaderSize()));
__ StoreInternalPointer(
RAX, FieldAddress(RAX, target::PointerBase::data_offset()), RDI);
Label loop;
__ Bind(&loop);
ASSERT(target::kObjectAlignment == kFpuRegisterSize);
__ movups(Address(RDI, 0), XMM0);
// Safe to only check every kObjectAlignment bytes instead of each word.
ASSERT(kAllocationRedZoneSize >= target::kObjectAlignment);
__ addq(RDI, Immediate(target::kObjectAlignment));
__ cmpq(RDI, RCX);
__ j(UNSIGNED_LESS, &loop, Assembler::kNearJump);
__ WriteAllocationCanary(RCX); // Fix overshoot.
InvokeAllocationProbePoint(assembler);
__ ret();
__ Bind(&call_runtime);
__ popq(AllocateTypedDataArrayABI::kLengthReg);
}
__ EnterStubFrame();
__ PushObject(Object::null_object()); // Make room for the result.
__ PushImmediate(Immediate(target::ToRawSmi(cid)));
__ pushq(AllocateTypedDataArrayABI::kLengthReg);
__ CallRuntime(kAllocateTypedDataRuntimeEntry, 2);
__ Drop(2); // Drop arguments.
__ popq(AllocateTypedDataArrayABI::kResultReg);
__ LeaveStubFrame();
__ ret();
}
} // namespace compiler
} // namespace dart
#endif // defined(TARGET_ARCH_X64)
Reference in New Issue View Git Blame Copy Permalink