void LIRGenerator::do_CompareAndSwap(Intrinsic* x, ValueType* type) {
assert(x->number_of_arguments() == 4, "wrong type");
LIRItem obj (x->argument_at(0), this); // object
LIRItem offset(x->argument_at(1), this); // offset of field
LIRItem cmp (x->argument_at(2), this); // value to compare with field
LIRItem val (x->argument_at(3), this); // replace field with val if matches cmp
assert(obj.type()->tag() == objectTag, "invalid type");
assert(cmp.type()->tag() == type->tag(), "invalid type");
assert(val.type()->tag() == type->tag(), "invalid type");
obj.load_item();
offset.load_nonconstant();
if (type == objectType) {
cmp.load_item_force(FrameMap::rax_oop_opr);
val.load_item();
} else if (type == intType) {
cmp.load_item_force(FrameMap::rax_opr);
val.load_item();
} else if (type == longType) {
cmp.load_item_force(FrameMap::long0_opr);
val.load_item_force(FrameMap::long1_opr);
} else {
ShouldNotReachHere();
}
LIR_Opr addr = new_pointer_register();
LIR_Address* a;
if(offset.result()->is_constant()) {
#ifdef _LP64
jlong c = offset.result()->as_jlong();
if ((jlong)((jint)c) == c) {
a = new LIR_Address(obj.result(),
(jint)c,
as_BasicType(type));
} else {
LIR_Opr tmp = new_register(T_LONG);
__ move(offset.result(), tmp);
a = new LIR_Address(obj.result(),
tmp,
as_BasicType(type));
}
#else
a = new LIR_Address(obj.result(),
offset.result()->as_jint(),
as_BasicType(type));
#endif
} else {
a = new LIR_Address(obj.result(),
offset.result(),
LIR_Address::times_1,
0,
as_BasicType(type));
}
__ leal(LIR_OprFact::address(a), addr);
if (type == objectType) { // Write-barrier needed for Object fields.
pre_barrier(addr, LIR_OprFact::illegalOpr /* pre_val */,
true /* do_load */, false /* patch */, NULL);
}
LIR_Opr ill = LIR_OprFact::illegalOpr; // for convenience
if (type == objectType)
__ cas_obj(addr, cmp.result(), val.result(), ill, ill);
else if (type == intType)
__ cas_int(addr, cmp.result(), val.result(), ill, ill);
else if (type == longType)
__ cas_long(addr, cmp.result(), val.result(), ill, ill);
else {
ShouldNotReachHere();
}
LIR_Opr result = rlock_result(x);
__ cmove(lir_cond_equal, LIR_OprFact::intConst(1), LIR_OprFact::intConst(0),
result, as_BasicType(type));
if (type == objectType) { // Write-barrier needed for Object fields.
post_barrier(addr, val.result());
}
}
void LIRGenerator::do_MathIntrinsic(Intrinsic* x) {
assert(x->number_of_arguments() == 1 || (x->number_of_arguments() == 2 && x->id() == vmIntrinsics::_dpow), "wrong type");
LIRItem value(x->argument_at(0), this);
bool use_fpu = false;
if (UseSSE >= 2) {
switch(x->id()) {
case vmIntrinsics::_dsin:
case vmIntrinsics::_dcos:
case vmIntrinsics::_dtan:
case vmIntrinsics::_dlog:
case vmIntrinsics::_dlog10:
case vmIntrinsics::_dexp:
case vmIntrinsics::_dpow:
use_fpu = true;
}
} else {
value.set_destroys_register();
}
value.load_item();
LIR_Opr calc_input = value.result();
LIR_Opr calc_input2 = NULL;
if (x->id() == vmIntrinsics::_dpow) {
LIRItem extra_arg(x->argument_at(1), this);
if (UseSSE < 2) {
extra_arg.set_destroys_register();
}
extra_arg.load_item();
calc_input2 = extra_arg.result();
}
LIR_Opr calc_result = rlock_result(x);
LIR_Opr tmp1 = FrameMap::caller_save_fpu_reg_at(0);
LIR_Opr tmp2 = FrameMap::caller_save_fpu_reg_at(1);
if (use_fpu) {
LIR_Opr tmp = FrameMap::fpu0_double_opr;
int tmp_start = 1;
if (calc_input2 != NULL) {
__ move(calc_input2, tmp);
tmp_start = 2;
calc_input2 = tmp;
}
__ move(calc_input, tmp);
calc_input = tmp;
calc_result = tmp;
tmp1 = FrameMap::caller_save_fpu_reg_at(tmp_start);
tmp2 = FrameMap::caller_save_fpu_reg_at(tmp_start + 1);
}
switch(x->id()) {
case vmIntrinsics::_dabs: __ abs (calc_input, calc_result, LIR_OprFact::illegalOpr); break;
case vmIntrinsics::_dsqrt: __ sqrt (calc_input, calc_result, LIR_OprFact::illegalOpr); break;
case vmIntrinsics::_dsin: __ sin (calc_input, calc_result, tmp1, tmp2); break;
case vmIntrinsics::_dcos: __ cos (calc_input, calc_result, tmp1, tmp2); break;
case vmIntrinsics::_dtan: __ tan (calc_input, calc_result, tmp1, tmp2); break;
case vmIntrinsics::_dlog: __ log (calc_input, calc_result, tmp1); break;
case vmIntrinsics::_dlog10: __ log10(calc_input, calc_result, tmp1); break;
case vmIntrinsics::_dexp: __ exp (calc_input, calc_result, tmp1, tmp2, FrameMap::rax_opr, FrameMap::rcx_opr, FrameMap::rdx_opr); break;
case vmIntrinsics::_dpow: __ pow (calc_input, calc_input2, calc_result, tmp1, tmp2, FrameMap::rax_opr, FrameMap::rcx_opr, FrameMap::rdx_opr); break;
default: ShouldNotReachHere();
}
if (use_fpu) {
__ move(calc_result, x->operand());
}
}
void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
assert(x->number_of_arguments() == 5, "wrong type");
CodeEmitInfo* info = state_for(x, x->state());
LIRItem src(x->argument_at(0), this);
LIRItem src_pos(x->argument_at(1), this);
LIRItem dst(x->argument_at(2), this);
LIRItem dst_pos(x->argument_at(3), this);
LIRItem length(x->argument_at(4), this);
#ifndef _LP64
src.load_item_force (FrameMap::rcx_oop_opr);
src_pos.load_item_force (FrameMap::rdx_opr);
dst.load_item_force (FrameMap::rax_oop_opr);
dst_pos.load_item_force (FrameMap::rbx_opr);
length.load_item_force (FrameMap::rdi_opr);
LIR_Opr tmp = (FrameMap::rsi_opr);
#else
src.load_item_force (FrameMap::as_oop_opr(j_rarg0));
src_pos.load_item_force (FrameMap::as_opr(j_rarg1));
dst.load_item_force (FrameMap::as_oop_opr(j_rarg2));
dst_pos.load_item_force (FrameMap::as_opr(j_rarg3));
length.load_item_force (FrameMap::as_opr(j_rarg4));
LIR_Opr tmp = FrameMap::as_opr(j_rarg5);
#endif // LP64
set_no_result(x);
int flags;
ciArrayKlass* expected_type;
arraycopy_helper(x, &flags, &expected_type);
__ arraycopy(src.result(), src_pos.result(), dst.result(), dst_pos.result(), length.result(), tmp, expected_type, flags, info); // does add_safepoint
}
void LIRGenerator::do_update_CRC32(Intrinsic* x) {
assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
LIR_Opr result = rlock_result(x);
int flags = 0;
switch (x->id()) {
case vmIntrinsics::_updateCRC32: {
LIRItem crc(x->argument_at(0), this);
LIRItem val(x->argument_at(1), this);
val.set_destroys_register();
crc.load_item();
val.load_item();
__ update_crc32(crc.result(), val.result(), result);
break;
}
case vmIntrinsics::_updateBytesCRC32:
case vmIntrinsics::_updateByteBufferCRC32: {
bool is_updateBytes = (x->id() == vmIntrinsics::_updateBytesCRC32);
LIRItem crc(x->argument_at(0), this);
LIRItem buf(x->argument_at(1), this);
LIRItem off(x->argument_at(2), this);
LIRItem len(x->argument_at(3), this);
buf.load_item();
off.load_nonconstant();
LIR_Opr index = off.result();
int offset = is_updateBytes ? arrayOopDesc::base_offset_in_bytes(T_BYTE) : 0;
if(off.result()->is_constant()) {
index = LIR_OprFact::illegalOpr;
offset += off.result()->as_jint();
}
LIR_Opr base_op = buf.result();
#ifndef _LP64
if (!is_updateBytes) { // long b raw address
base_op = new_register(T_INT);
__ convert(Bytecodes::_l2i, buf.result(), base_op);
}
#else
if (index->is_valid()) {
LIR_Opr tmp = new_register(T_LONG);
__ convert(Bytecodes::_i2l, index, tmp);
index = tmp;
}
#endif
LIR_Address* a = new LIR_Address(base_op,
index,
LIR_Address::times_1,
offset,
T_BYTE);
BasicTypeList signature(3);
signature.append(T_INT);
signature.append(T_ADDRESS);
signature.append(T_INT);
CallingConvention* cc = frame_map()->c_calling_convention(&signature);
const LIR_Opr result_reg = result_register_for(x->type());
LIR_Opr addr = new_pointer_register();
__ leal(LIR_OprFact::address(a), addr);
crc.load_item_force(cc->at(0));
__ move(addr, cc->at(1));
len.load_item_force(cc->at(2));
__ call_runtime_leaf(StubRoutines::updateBytesCRC32(), getThreadTemp(), result_reg, cc->args());
__ move(result_reg, result);
break;
}
default: {
ShouldNotReachHere();
}
}
}
LIR_Opr fixed_register_for(BasicType type) {
switch (type) {
case T_FLOAT: return FrameMap::fpu0_float_opr;
case T_DOUBLE: return FrameMap::fpu0_double_opr;
case T_INT: return FrameMap::rax_opr;
case T_LONG: return FrameMap::long0_opr;
default: ShouldNotReachHere(); return LIR_OprFact::illegalOpr;
}
}
void LIRGenerator::do_Convert(Convert* x) {
bool fixed_input = false, fixed_result = false, round_result = false, needs_stub = false;
switch (x->op()) {
case Bytecodes::_i2l: // fall through
case Bytecodes::_l2i: // fall through
case Bytecodes::_i2b: // fall through
case Bytecodes::_i2c: // fall through
case Bytecodes::_i2s: fixed_input = false; fixed_result = false; round_result = false; needs_stub = false; break;
case Bytecodes::_f2d: fixed_input = UseSSE == 1; fixed_result = false; round_result = false; needs_stub = false; break;
case Bytecodes::_d2f: fixed_input = false; fixed_result = UseSSE == 1; round_result = UseSSE < 1; needs_stub = false; break;
case Bytecodes::_i2f: fixed_input = false; fixed_result = false; round_result = UseSSE < 1; needs_stub = false; break;
case Bytecodes::_i2d: fixed_input = false; fixed_result = false; round_result = false; needs_stub = false; break;
case Bytecodes::_f2i: fixed_input = false; fixed_result = false; round_result = false; needs_stub = true; break;
case Bytecodes::_d2i: fixed_input = false; fixed_result = false; round_result = false; needs_stub = true; break;
case Bytecodes::_l2f: fixed_input = false; fixed_result = UseSSE >= 1; round_result = UseSSE < 1; needs_stub = false; break;
case Bytecodes::_l2d: fixed_input = false; fixed_result = UseSSE >= 2; round_result = UseSSE < 2; needs_stub = false; break;
case Bytecodes::_f2l: fixed_input = true; fixed_result = true; round_result = false; needs_stub = false; break;
case Bytecodes::_d2l: fixed_input = true; fixed_result = true; round_result = false; needs_stub = false; break;
default: ShouldNotReachHere();
}
LIRItem value(x->value(), this);
value.load_item();
LIR_Opr input = value.result();
LIR_Opr result = rlock(x);
LIR_Opr conv_input = input;
LIR_Opr conv_result = result;
ConversionStub* stub = NULL;
if (fixed_input) {
conv_input = fixed_register_for(input->type());
__ move(input, conv_input);
}
assert(fixed_result == false || round_result == false, "cannot set both");
if (fixed_result) {
conv_result = fixed_register_for(result->type());
} else if (round_result) {
result = new_register(result->type());
set_vreg_flag(result, must_start_in_memory);
}
if (needs_stub) {
stub = new ConversionStub(x->op(), conv_input, conv_result);
}
__ convert(x->op(), conv_input, conv_result, stub);
if (result != conv_result) {
__ move(conv_result, result);
}
assert(result->is_virtual(), "result must be virtual register");
set_result(x, result);
}
void LIRGenerator::do_NewInstance(NewInstance* x) {
print_if_not_loaded(x);
CodeEmitInfo* info = state_for(x, x->state());
LIR_Opr reg = result_register_for(x->type());
new_instance(reg, x->klass(), x->is_unresolved(),
FrameMap::rcx_oop_opr,
FrameMap::rdi_oop_opr,
FrameMap::rsi_oop_opr,
LIR_OprFact::illegalOpr,
FrameMap::rdx_metadata_opr, info);
LIR_Opr result = rlock_result(x);
__ move(reg, result);
}
void LIRGenerator::do_NewTypeArray(NewTypeArray* x) {
CodeEmitInfo* info = state_for(x, x->state());
LIRItem length(x->length(), this);
length.load_item_force(FrameMap::rbx_opr);
LIR_Opr reg = result_register_for(x->type());
LIR_Opr tmp1 = FrameMap::rcx_oop_opr;
LIR_Opr tmp2 = FrameMap::rsi_oop_opr;
LIR_Opr tmp3 = FrameMap::rdi_oop_opr;
LIR_Opr tmp4 = reg;
LIR_Opr klass_reg = FrameMap::rdx_metadata_opr;
LIR_Opr len = length.result();
BasicType elem_type = x->elt_type();
__ metadata2reg(ciTypeArrayKlass::make(elem_type)->constant_encoding(), klass_reg);
CodeStub* slow_path = new NewTypeArrayStub(klass_reg, len, reg, info);
__ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, elem_type, klass_reg, slow_path);
LIR_Opr result = rlock_result(x);
__ move(reg, result);
}
void LIRGenerator::do_NewObjectArray(NewObjectArray* x) {
LIRItem length(x->length(), this);
CodeEmitInfo* patching_info = NULL;
if (!x->klass()->is_loaded() || PatchALot) {
patching_info = state_for(x, x->state_before());
}
CodeEmitInfo* info = state_for(x, x->state());
const LIR_Opr reg = result_register_for(x->type());
LIR_Opr tmp1 = FrameMap::rcx_oop_opr;
LIR_Opr tmp2 = FrameMap::rsi_oop_opr;
LIR_Opr tmp3 = FrameMap::rdi_oop_opr;
LIR_Opr tmp4 = reg;
LIR_Opr klass_reg = FrameMap::rdx_metadata_opr;
length.load_item_force(FrameMap::rbx_opr);
LIR_Opr len = length.result();
CodeStub* slow_path = new NewObjectArrayStub(klass_reg, len, reg, info);
ciKlass* obj = (ciKlass*) ciObjArrayKlass::make(x->klass());
if (obj == ciEnv::unloaded_ciobjarrayklass()) {
BAILOUT("encountered unloaded_ciobjarrayklass due to out of memory error");
}
klass2reg_with_patching(klass_reg, obj, patching_info);
__ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, T_OBJECT, klass_reg, slow_path);
LIR_Opr result = rlock_result(x);
__ move(reg, result);
}
void LIRGenerator::do_NewMultiArray(NewMultiArray* x) {
Values* dims = x->dims();
int i = dims->length();
LIRItemList* items = new LIRItemList(dims->length(), NULL);
while (i-- > 0) {
LIRItem* size = new LIRItem(dims->at(i), this);
items->at_put(i, size);
}
CodeEmitInfo* patching_info = NULL;
if (!x->klass()->is_loaded() || PatchALot) {
patching_info = state_for(x, x->state_before());
x->set_exception_handlers(new XHandlers(x->exception_handlers()));
}
CodeEmitInfo* info = state_for(x, x->state());
i = dims->length();
while (i-- > 0) {
LIRItem* size = items->at(i);
size->load_nonconstant();
store_stack_parameter(size->result(), in_ByteSize(i*4));
}
LIR_Opr klass_reg = FrameMap::rax_metadata_opr;
klass2reg_with_patching(klass_reg, x->klass(), patching_info);
LIR_Opr rank = FrameMap::rbx_opr;
__ move(LIR_OprFact::intConst(x->rank()), rank);
LIR_Opr varargs = FrameMap::rcx_opr;
__ move(FrameMap::rsp_opr, varargs);
LIR_OprList* args = new LIR_OprList(3);
args->append(klass_reg);
args->append(rank);
args->append(varargs);
LIR_Opr reg = result_register_for(x->type());
__ call_runtime(Runtime1::entry_for(Runtime1::new_multi_array_id),
LIR_OprFact::illegalOpr,
reg, args, info);
LIR_Opr result = rlock_result(x);
__ move(reg, result);
}
void LIRGenerator::do_BlockBegin(BlockBegin* x) {
}
void LIRGenerator::do_CheckCast(CheckCast* x) {
LIRItem obj(x->obj(), this);
CodeEmitInfo* patching_info = NULL;
if (!x->klass()->is_loaded() || (PatchALot && !x->is_incompatible_class_change_check())) {
patching_info = state_for(x, x->state_before());
}
obj.load_item();
CodeEmitInfo* info_for_exception =
(x->needs_exception_state() ? state_for(x) :
state_for(x, x->state_before(), true /*ignore_xhandler*/));
CodeStub* stub;
if (x->is_incompatible_class_change_check()) {
assert(patching_info == NULL, "can't patch this");
stub = new SimpleExceptionStub(Runtime1::throw_incompatible_class_change_error_id, LIR_OprFact::illegalOpr, info_for_exception);
} else if (x->is_invokespecial_receiver_check()) {
assert(patching_info == NULL, "can't patch this");
stub = new DeoptimizeStub(info_for_exception);
} else {
stub = new SimpleExceptionStub(Runtime1::throw_class_cast_exception_id, obj.result(), info_for_exception);
}
LIR_Opr reg = rlock_result(x);
LIR_Opr tmp3 = LIR_OprFact::illegalOpr;
if (!x->klass()->is_loaded() || UseCompressedClassPointers) {
tmp3 = new_register(objectType);
}
__ checkcast(reg, obj.result(), x->klass(),
new_register(objectType), new_register(objectType), tmp3,
x->direct_compare(), info_for_exception, patching_info, stub,
x->profiled_method(), x->profiled_bci());
}
void LIRGenerator::do_InstanceOf(InstanceOf* x) {
LIRItem obj(x->obj(), this);
LIR_Opr reg = rlock_result(x);
CodeEmitInfo* patching_info = NULL;
if ((!x->klass()->is_loaded() || PatchALot)) {
patching_info = state_for(x, x->state_before());
}
obj.load_item();
LIR_Opr tmp3 = LIR_OprFact::illegalOpr;
if (!x->klass()->is_loaded() || UseCompressedClassPointers) {
tmp3 = new_register(objectType);
}
__ instanceof(reg, obj.result(), x->klass(),
new_register(objectType), new_register(objectType), tmp3,
x->direct_compare(), patching_info, x->profiled_method(), x->profiled_bci());
}
void LIRGenerator::do_If(If* x) {
assert(x->number_of_sux() == 2, "inconsistency");
ValueTag tag = x->x()->type()->tag();
bool is_safepoint = x->is_safepoint();
If::Condition cond = x->cond();
LIRItem xitem(x->x(), this);
LIRItem yitem(x->y(), this);
LIRItem* xin = &xitem;
LIRItem* yin = &yitem;
if (tag == longTag) {
if (cond == If::gtr || cond == If::leq) {
cond = Instruction::mirror(cond);
xin = &yitem;
yin = &xitem;
}
xin->set_destroys_register();
}
xin->load_item();
if (tag == longTag && yin->is_constant() && yin->get_jlong_constant() == 0 && (cond == If::eql || cond == If::neq)) {
yin->dont_load_item();
} else if (tag == longTag || tag == floatTag || tag == doubleTag) {
yin->load_item();
} else {
yin->dont_load_item();
}
if (x->is_safepoint()) {
increment_backedge_counter(state_for(x, x->state_before()), x->profiled_bci());
__ safepoint(LIR_OprFact::illegalOpr, state_for(x, x->state_before()));
}
set_no_result(x);
LIR_Opr left = xin->result();
LIR_Opr right = yin->result();
__ cmp(lir_cond(cond), left, right);
profile_branch(x, cond);
move_to_phi(x->state());
if (x->x()->type()->is_float_kind()) {
__ branch(lir_cond(cond), right->type(), x->tsux(), x->usux());
} else {
__ branch(lir_cond(cond), right->type(), x->tsux());
}
assert(x->default_sux() == x->fsux(), "wrong destination above");
__ jump(x->default_sux());
}
LIR_Opr LIRGenerator::getThreadPointer() {
#ifdef _LP64
return FrameMap::as_pointer_opr(r15_thread);
#else
LIR_Opr result = new_register(T_INT);
__ get_thread(result);
return result;
#endif //
}
void LIRGenerator::trace_block_entry(BlockBegin* block) {
store_stack_parameter(LIR_OprFact::intConst(block->block_id()), in_ByteSize(0));
LIR_OprList* args = new LIR_OprList();
address func = CAST_FROM_FN_PTR(address, Runtime1::trace_block_entry);
__ call_runtime_leaf(func, LIR_OprFact::illegalOpr, LIR_OprFact::illegalOpr, args);
}
void LIRGenerator::volatile_field_store(LIR_Opr value, LIR_Address* address,
CodeEmitInfo* info) {
if (address->type() == T_LONG) {
address = new LIR_Address(address->base(),
address->index(), address->scale(),
address->disp(), T_DOUBLE);
LIR_Opr temp_double = new_register(T_DOUBLE);
LIR_Opr spill = new_register(T_LONG);
set_vreg_flag(spill, must_start_in_memory);
__ move(value, spill);
__ volatile_move(spill, temp_double, T_LONG);
__ volatile_move(temp_double, LIR_OprFact::address(address), T_LONG, info);
} else {
__ store(value, address, info);
}
}
void LIRGenerator::volatile_field_load(LIR_Address* address, LIR_Opr result,
CodeEmitInfo* info) {
if (address->type() == T_LONG) {
address = new LIR_Address(address->base(),
address->index(), address->scale(),
address->disp(), T_DOUBLE);
LIR_Opr temp_double = new_register(T_DOUBLE);
__ volatile_move(LIR_OprFact::address(address), temp_double, T_LONG, info);
__ volatile_move(temp_double, result, T_LONG);
if (UseSSE < 2) {
set_vreg_flag(result, must_start_in_memory);
}
} else {
__ load(address, result, info);
}
}
void LIRGenerator::get_Object_unsafe(LIR_Opr dst, LIR_Opr src, LIR_Opr offset,
BasicType type, bool is_volatile) {
if (is_volatile && type == T_LONG) {
LIR_Address* addr = new LIR_Address(src, offset, T_DOUBLE);
LIR_Opr tmp = new_register(T_DOUBLE);
__ load(addr, tmp);
LIR_Opr spill = new_register(T_LONG);
set_vreg_flag(spill, must_start_in_memory);
__ move(tmp, spill);
__ move(spill, dst);
} else {
LIR_Address* addr = new LIR_Address(src, offset, type);
__ load(addr, dst);
}
}
void LIRGenerator::put_Object_unsafe(LIR_Opr src, LIR_Opr offset, LIR_Opr data,
BasicType type, bool is_volatile) {
if (is_volatile && type == T_LONG) {
LIR_Address* addr = new LIR_Address(src, offset, T_DOUBLE);
LIR_Opr tmp = new_register(T_DOUBLE);
LIR_Opr spill = new_register(T_DOUBLE);
set_vreg_flag(spill, must_start_in_memory);
__ move(data, spill);
__ move(spill, tmp);
__ move(tmp, addr);
} else {
LIR_Address* addr = new LIR_Address(src, offset, type);
bool is_obj = (type == T_ARRAY || type == T_OBJECT);
if (is_obj) {
pre_barrier(LIR_OprFact::address(addr), LIR_OprFact::illegalOpr /* pre_val */,
true /* do_load */, false /* patch */, NULL);
__ move(data, addr);
assert(src->is_register(), "must be register");
post_barrier(LIR_OprFact::address(addr), data);
} else {
__ move(data, addr);
}
}
}
void LIRGenerator::do_UnsafeGetAndSetObject(UnsafeGetAndSetObject* x) {
BasicType type = x->basic_type();
LIRItem src(x->object(), this);
LIRItem off(x->offset(), this);
LIRItem value(x->value(), this);
src.load_item();
value.load_item();
off.load_nonconstant();
LIR_Opr dst = rlock_result(x, type);
LIR_Opr data = value.result();
bool is_obj = (type == T_ARRAY || type == T_OBJECT);
LIR_Opr offset = off.result();
assert (type == T_INT || (!x->is_add() && is_obj) LP64_ONLY( || type == T_LONG ), "unexpected type");
LIR_Address* addr;
if (offset->is_constant()) {
#ifdef _LP64
jlong c = offset->as_jlong();
if ((jlong)((jint)c) == c) {
addr = new LIR_Address(src.result(), (jint)c, type);
} else {
LIR_Opr tmp = new_register(T_LONG);
__ move(offset, tmp);
addr = new LIR_Address(src.result(), tmp, type);
}
#else
addr = new LIR_Address(src.result(), offset->as_jint(), type);
#endif
} else {
addr = new LIR_Address(src.result(), offset, type);
}
__ move(data, dst);
if (x->is_add()) {
__ xadd(LIR_OprFact::address(addr), dst, dst, LIR_OprFact::illegalOpr);
} else {
if (is_obj) {
pre_barrier(LIR_OprFact::address(addr), LIR_OprFact::illegalOpr /* pre_val */,
true /* do_load */, false /* patch */, NULL);
}
__ xchg(LIR_OprFact::address(addr), dst, dst, LIR_OprFact::illegalOpr);
if (is_obj) {
post_barrier(LIR_OprFact::address(addr), data);
}
}
}
C:\hotspot-69087d08d473\src\cpu\x86\vm/c1_MacroAssembler_x86.cpp
#include "precompiled.hpp"
#include "c1/c1_MacroAssembler.hpp"
#include "c1/c1_Runtime1.hpp"
#include "classfile/systemDictionary.hpp"
#include "gc_interface/collectedHeap.hpp"
#include "interpreter/interpreter.hpp"
#include "oops/arrayOop.hpp"
#include "oops/markOop.hpp"
#include "runtime/basicLock.hpp"
#include "runtime/biasedLocking.hpp"
#include "runtime/os.hpp"
#include "runtime/stubRoutines.hpp"
int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register scratch, Label& slow_case) {
const int aligned_mask = BytesPerWord -1;
const int hdr_offset = oopDesc::mark_offset_in_bytes();
assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
Label done;
int null_check_offset = -1;
verify_oop(obj);
movptr(Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()), obj);
if (UseBiasedLocking) {
assert(scratch != noreg, "should have scratch register at this point");
null_check_offset = biased_locking_enter(disp_hdr, obj, hdr, scratch, false, done, &slow_case);
} else {
null_check_offset = offset();
}
movptr(hdr, Address(obj, hdr_offset));
orptr(hdr, markOopDesc::unlocked_value);
movptr(Address(disp_hdr, 0), hdr);
if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
if (PrintBiasedLockingStatistics) {
cond_inc32(Assembler::equal,
ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr()));
}
jcc(Assembler::equal, done);
subptr(hdr, rsp);
andptr(hdr, aligned_mask - os::vm_page_size());
movptr(Address(disp_hdr, 0), hdr);
jcc(Assembler::notZero, slow_case);
bind(done);
return null_check_offset;
}
void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
const int aligned_mask = BytesPerWord -1;
const int hdr_offset = oopDesc::mark_offset_in_bytes();
assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
Label done;
if (UseBiasedLocking) {
movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
biased_locking_exit(obj, hdr, done);
}
movptr(hdr, Address(disp_hdr, 0));
testptr(hdr, hdr);
jcc(Assembler::zero, done);
if (!UseBiasedLocking) {
movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
}
verify_oop(obj);
if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
cmpxchgptr(hdr, Address(obj, hdr_offset));
jcc(Assembler::notEqual, slow_case);
bind(done);
}
void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
if (UseTLAB) {
tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
} else {
eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
incr_allocated_bytes(noreg, var_size_in_bytes, con_size_in_bytes, t1);
}
}
void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
assert_different_registers(obj, klass, len);
if (UseBiasedLocking && !len->is_valid()) {
assert_different_registers(obj, klass, len, t1, t2);
movptr(t1, Address(klass, Klass::prototype_header_offset()));
movptr(Address(obj, oopDesc::mark_offset_in_bytes()), t1);
} else {
movptr(Address(obj, oopDesc::mark_offset_in_bytes ()), (int32_t)(intptr_t)markOopDesc::prototype());
}
#ifdef _LP64
if (UseCompressedClassPointers) { // Take care not to kill klass
movptr(t1, klass);
encode_klass_not_null(t1);
movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
} else
#endif
{
movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
}
if (len->is_valid()) {
movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
}
#ifdef _LP64
else if (UseCompressedClassPointers) {
xorptr(t1, t1);
store_klass_gap(obj, t1);
}
#endif
}
void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
Label done;
assert(obj != len_in_bytes && obj != t1 && t1 != len_in_bytes, "registers must be different");
assert((hdr_size_in_bytes & (BytesPerWord - 1)) == 0, "header size is not a multiple of BytesPerWord");
Register index = len_in_bytes;
subptr(index, hdr_size_in_bytes);
jcc(Assembler::zero, done);
#ifdef ASSERT
{ Label L;
testptr(index, BytesPerWord - 1);
jcc(Assembler::zero, L);
stop("index is not a multiple of BytesPerWord");
bind(L);
}
#endif
xorptr(t1, t1); // use _zero reg to clear memory (shorter code)
if (UseIncDec) {
shrptr(index, 3); // divide by 8/16 and set carry flag if bit 2 was set
} else {
shrptr(index, 2); // use 2 instructions to avoid partial flag stall
shrptr(index, 1);
}
#ifndef _LP64
{ Label even;
jcc(Assembler::carryClear, even);
movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 0*BytesPerWord), t1);
jcc(Assembler::zero, done);
bind(even);
}
#endif // !_LP64
{ Label loop;
bind(loop);
movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 1*BytesPerWord), t1);
NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 2*BytesPerWord), t1);)
decrement(index);
jcc(Assembler::notZero, loop);
}
bind(done);
}
void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
assert(obj == rax, "obj must be in rax, for cmpxchg");
assert_different_registers(obj, t1, t2); // XXX really?
assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2);
}
void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2) {
assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
"con_size_in_bytes is not multiple of alignment");
const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
initialize_header(obj, klass, noreg, t1, t2);
const Register t1_zero = t1;
const Register index = t2;
const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
if (var_size_in_bytes != noreg) {
mov(index, var_size_in_bytes);
initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
} else if (con_size_in_bytes <= threshold) {
xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
movptr(Address(obj, i), t1_zero);
} else if (con_size_in_bytes > hdr_size_in_bytes) {
xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
{ Label loop;
bind(loop);
movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
t1_zero);
NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
t1_zero);)
decrement(index);
jcc(Assembler::notZero, loop);
}
}
if (CURRENT_ENV->dtrace_alloc_probes()) {
assert(obj == rax, "must be");
call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
}
verify_oop(obj);
}
void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
assert(obj == rax, "obj must be in rax, for cmpxchg");
assert_different_registers(obj, len, t1, t2, klass);
assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
cmpptr(len, (int32_t)max_array_allocation_length);
jcc(Assembler::above, slow_case);
const Register arr_size = t2; // okay to be the same
movptr(arr_size, (int32_t)header_size * BytesPerWord + MinObjAlignmentInBytesMask);
lea(arr_size, Address(arr_size, len, f));
andptr(arr_size, ~MinObjAlignmentInBytesMask);
try_allocate(obj, arr_size, 0, t1, t2, slow_case);
initialize_header(obj, klass, len, t1, t2);
const Register len_zero = len;
initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
if (CURRENT_ENV->dtrace_alloc_probes()) {
assert(obj == rax, "must be");
call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
}
verify_oop(obj);
}
void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
verify_oop(receiver);
assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
int start_offset = offset();
if (UseCompressedClassPointers) {
load_klass(rscratch1, receiver);
cmpptr(rscratch1, iCache);
} else {
cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
}
jump_cc(Assembler::notEqual,
RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
}
void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
generate_stack_overflow_check(bang_size_in_bytes);
push(rbp);
if (PreserveFramePointer) {
mov(rbp, rsp);
}
#ifdef TIERED
if (UseSSE < 2 ) {
empty_FPU_stack();
}
#endif // TIERED
decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
}
void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
pop(rbp);
}
void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) {
if (C1Breakpoint) int3();
inline_cache_check(receiver, ic_klass);
}
void C1_MacroAssembler::verified_entry() {
if (C1Breakpoint || VerifyFPU || !UseStackBanging) {
fat_nop();
}
if (C1Breakpoint)int3();
verify_FPU(0, "method_entry");
}
#ifndef PRODUCT
void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
if (!VerifyOops) return;
verify_oop_addr(Address(rsp, stack_offset));
}
void C1_MacroAssembler::verify_not_null_oop(Register r) {
if (!VerifyOops) return;
Label not_null;
testptr(r, r);
jcc(Assembler::notZero, not_null);
stop("non-null oop required");
bind(not_null);
verify_oop(r);
}
void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
#ifdef ASSERT
if (inv_rax) movptr(rax, 0xDEAD);
if (inv_rbx) movptr(rbx, 0xDEAD);
if (inv_rcx) movptr(rcx, 0xDEAD);
if (inv_rdx) movptr(rdx, 0xDEAD);
if (inv_rsi) movptr(rsi, 0xDEAD);
if (inv_rdi) movptr(rdi, 0xDEAD);
#endif
}
#endif // ifndef PRODUCT
C:\hotspot-69087d08d473\src\cpu\x86\vm/c1_MacroAssembler_x86.hpp
#ifndef CPU_X86_VM_C1_MACROASSEMBLER_X86_HPP
#define CPU_X86_VM_C1_MACROASSEMBLER_X86_HPP
private:
int _rsp_offset; // track rsp changes
void pd_init() { _rsp_offset = 0; }
public:
void try_allocate(
Register obj, // result: pointer to object after successful allocation
Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
int con_size_in_bytes, // object size in bytes if known at compile time
Register t1, // temp register
Register t2, // temp register
Label& slow_case // continuation point if fast allocation fails
);
void initialize_header(Register obj, Register klass, Register len, Register t1, Register t2);
void initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1);
int lock_object (Register swap, Register obj, Register disp_hdr, Register scratch, Label& slow_case);
void unlock_object(Register swap, Register obj, Register lock, Label& slow_case);
void initialize_object(
Register obj, // result: pointer to object after successful allocation
Register klass, // object klass
Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
int con_size_in_bytes, // object size in bytes if known at compile time
Register t1, // temp register
Register t2 // temp register
);
void allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case);
enum {
max_array_allocation_length = 0x00FFFFFF
};
void allocate_array(Register obj, Register len, Register t, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case);
int rsp_offset() const { return _rsp_offset; }
void set_rsp_offset(int n) { _rsp_offset = n; }
void push_jint (jint i) { _rsp_offset++; push(i); }
void push_oop (jobject o) { _rsp_offset++; pushoop(o); }
void push_addr (Address a) { _rsp_offset++; pushptr(a); }
void push_reg (Register r) { _rsp_offset++; push(r); }
void pop_reg (Register r) { _rsp_offset--; pop(r); assert(_rsp_offset >= 0, "stack offset underflow"); }
void dec_stack (int nof_words) {
_rsp_offset -= nof_words;
assert(_rsp_offset >= 0, "stack offset underflow");
addptr(rsp, wordSize * nof_words);
}
void dec_stack_after_call (int nof_words) {
_rsp_offset -= nof_words;
assert(_rsp_offset >= 0, "stack offset underflow");
}
void invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) PRODUCT_RETURN;
#endif // CPU_X86_VM_C1_MACROASSEMBLER_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/c1_Runtime1_x86.cpp
#include "precompiled.hpp"
#include "asm/assembler.hpp"
#include "c1/c1_Defs.hpp"
#include "c1/c1_MacroAssembler.hpp"
#include "c1/c1_Runtime1.hpp"
#include "interpreter/interpreter.hpp"
#include "nativeInst_x86.hpp"
#include "oops/compiledICHolder.hpp"
#include "oops/oop.inline.hpp"
#include "prims/jvmtiExport.hpp"
#include "register_x86.hpp"
#include "runtime/sharedRuntime.hpp"
#include "runtime/signature.hpp"
#include "runtime/vframeArray.hpp"
#include "utilities/macros.hpp"
#include "vmreg_x86.inline.hpp"
#if INCLUDE_ALL_GCS
#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
#endif
int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, int args_size) {
const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread); // is callee-saved register (Visual C++ calling conventions)
assert(!(oop_result1->is_valid() || metadata_result->is_valid()) || oop_result1 != metadata_result, "registers must be different");
assert(oop_result1 != thread && metadata_result != thread, "registers must be different");
assert(args_size >= 0, "illegal args_size");
bool align_stack = false;
#ifdef _LP64
align_stack = (stub_id() == Runtime1::handle_exception_from_callee_id);
#endif
#ifdef _LP64
mov(c_rarg0, thread);
set_num_rt_args(0); // Nothing on stack
#else
set_num_rt_args(1 + args_size);
get_thread(thread);
push(thread);
#endif // _LP64
int call_offset;
if (!align_stack) {
set_last_Java_frame(thread, noreg, rbp, NULL);
} else {
address the_pc = pc();
call_offset = offset();
set_last_Java_frame(thread, noreg, rbp, the_pc);
andptr(rsp, -(StackAlignmentInBytes)); // Align stack
}
call(RuntimeAddress(entry));
if (!align_stack) {
call_offset = offset();
}
#ifdef ASSERT
guarantee(thread != rax, "change this code");
push(rax);
{ Label L;
get_thread(rax);
cmpptr(thread, rax);
jcc(Assembler::equal, L);
int3();
stop("StubAssembler::call_RT: rdi not callee saved?");
bind(L);
}
pop(rax);
#endif
reset_last_Java_frame(thread, true);
NOT_LP64(addptr(rsp, num_rt_args()*BytesPerWord));
{ Label L;
cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
jcc(Assembler::equal, L);
movptr(rax, Address(thread, Thread::pending_exception_offset()));
if (oop_result1->is_valid()) {
movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
}
if (metadata_result->is_valid()) {
movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD);
}
if (frame_size() == no_frame_size) {
leave();
jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
} else if (_stub_id == Runtime1::forward_exception_id) {
should_not_reach_here();
} else {
jump(RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
}
bind(L);
}
if (oop_result1->is_valid()) {
get_vm_result(oop_result1, thread);
}
if (metadata_result->is_valid()) {
get_vm_result_2(metadata_result, thread);
}
return call_offset;
}
int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1) {
#ifdef _LP64
mov(c_rarg1, arg1);
#else
push(arg1);
#endif // _LP64
return call_RT(oop_result1, metadata_result, entry, 1);
}
int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2) {
#ifdef _LP64
if (c_rarg1 == arg2) {
if (c_rarg2 == arg1) {
xchgq(arg1, arg2);
} else {
mov(c_rarg2, arg2);
mov(c_rarg1, arg1);
}
} else {
mov(c_rarg1, arg1);
mov(c_rarg2, arg2);
}
#else
push(arg2);
push(arg1);
#endif // _LP64
return call_RT(oop_result1, metadata_result, entry, 2);
}
int StubAssembler::call_RT(Register oop_result1, Register metadata_result, address entry, Register arg1, Register arg2, Register arg3) {
#ifdef _LP64
if (arg1 == c_rarg2 || arg1 == c_rarg3 ||
arg2 == c_rarg1 || arg2 == c_rarg3 ||
arg3 == c_rarg1 || arg3 == c_rarg2) {
push(arg3);
push(arg2);
push(arg1);
pop(c_rarg1);
pop(c_rarg2);
pop(c_rarg3);
} else {
mov(c_rarg1, arg1);
mov(c_rarg2, arg2);
mov(c_rarg3, arg3);
}
#else
push(arg3);
push(arg2);
push(arg1);
#endif // _LP64
return call_RT(oop_result1, metadata_result, entry, 3);
}
class StubFrame: public StackObj {
private:
StubAssembler* _sasm;
public:
StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments);
void load_argument(int offset_in_words, Register reg);
~StubFrame();
};
#define __ _sasm->
StubFrame::StubFrame(StubAssembler* sasm, const char* name, bool must_gc_arguments) {
_sasm = sasm;
__ set_info(name, must_gc_arguments);
__ enter();
}
void StubFrame::load_argument(int offset_in_words, Register reg) {
__ movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
}
StubFrame::~StubFrame() {
__ leave();
__ ret(0);
}
#undef __
#define __ sasm->
const int float_regs_as_doubles_size_in_slots = pd_nof_fpu_regs_frame_map * 2;
const int xmm_regs_as_doubles_size_in_slots = FrameMap::nof_xmm_regs * 2;
#ifdef _LP64
#define SLOT2(x) x,
#define SLOT_PER_WORD 2
#else
#define SLOT2(x)
#define SLOT_PER_WORD 1
#endif // _LP64
enum reg_save_layout {
#ifdef _LP64
align_dummy_0, align_dummy_1,
#endif // _LP64
#ifdef _WIN64
arg_reg_save_1, arg_reg_save_1H, // 0, 4
arg_reg_save_2, arg_reg_save_2H, // 8, 12
arg_reg_save_3, arg_reg_save_3H, // 16, 20
arg_reg_save_4, arg_reg_save_4H, // 24, 28
#endif // _WIN64
xmm_regs_as_doubles_off, // 32
float_regs_as_doubles_off = xmm_regs_as_doubles_off + xmm_regs_as_doubles_size_in_slots, // 160
fpu_state_off = float_regs_as_doubles_off + float_regs_as_doubles_size_in_slots, // 224
fpu_state_end_off = fpu_state_off + (FPUStateSizeInWords / SLOT_PER_WORD), // 352
marker = fpu_state_end_off, SLOT2(markerH) // 352, 356
extra_space_offset, // 360
#ifdef _LP64
r15_off = extra_space_offset, r15H_off, // 360, 364
r14_off, r14H_off, // 368, 372
r13_off, r13H_off, // 376, 380
r12_off, r12H_off, // 384, 388
r11_off, r11H_off, // 392, 396
r10_off, r10H_off, // 400, 404
r9_off, r9H_off, // 408, 412
r8_off, r8H_off, // 416, 420
rdi_off, rdiH_off, // 424, 428
#else
rdi_off = extra_space_offset,
#endif // _LP64
rsi_off, SLOT2(rsiH_off) // 432, 436
rbp_off, SLOT2(rbpH_off) // 440, 444
rsp_off, SLOT2(rspH_off) // 448, 452
rbx_off, SLOT2(rbxH_off) // 456, 460
rdx_off, SLOT2(rdxH_off) // 464, 468
rcx_off, SLOT2(rcxH_off) // 472, 476
rax_off, SLOT2(raxH_off) // 480, 484
saved_rbp_off, SLOT2(saved_rbpH_off) // 488, 492
return_off, SLOT2(returnH_off) // 496, 500
reg_save_frame_size // As noted: neglects any parameters to runtime // 504
};
static OopMap* generate_oop_map(StubAssembler* sasm, int num_rt_args,
bool save_fpu_registers = true) {
LP64_ONLY(num_rt_args = 0);
LP64_ONLY(assert((reg_save_frame_size * VMRegImpl::stack_slot_size) % 16 == 0, "must be 16 byte aligned");)
int frame_size_in_slots = reg_save_frame_size + num_rt_args; // args + thread
sasm->set_frame_size(frame_size_in_slots / VMRegImpl::slots_per_word );
OopMap* map = new OopMap(frame_size_in_slots, 0);
map->set_callee_saved(VMRegImpl::stack2reg(rax_off + num_rt_args), rax->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(rcx_off + num_rt_args), rcx->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(rdx_off + num_rt_args), rdx->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(rbx_off + num_rt_args), rbx->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(rsi_off + num_rt_args), rsi->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(rdi_off + num_rt_args), rdi->as_VMReg());
#ifdef _LP64
map->set_callee_saved(VMRegImpl::stack2reg(r8_off + num_rt_args), r8->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(r9_off + num_rt_args), r9->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(r10_off + num_rt_args), r10->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(r11_off + num_rt_args), r11->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(r12_off + num_rt_args), r12->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(r13_off + num_rt_args), r13->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(r14_off + num_rt_args), r14->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(r15_off + num_rt_args), r15->as_VMReg());
map->set_callee_saved(VMRegImpl::stack2reg(raxH_off + num_rt_args), rax->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(rcxH_off + num_rt_args), rcx->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(rdxH_off + num_rt_args), rdx->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(rbxH_off + num_rt_args), rbx->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(rsiH_off + num_rt_args), rsi->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(rdiH_off + num_rt_args), rdi->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r8H_off + num_rt_args), r8->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r9H_off + num_rt_args), r9->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r10H_off + num_rt_args), r10->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r11H_off + num_rt_args), r11->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r12H_off + num_rt_args), r12->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r13H_off + num_rt_args), r13->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r14H_off + num_rt_args), r14->as_VMReg()->next());
map->set_callee_saved(VMRegImpl::stack2reg(r15H_off + num_rt_args), r15->as_VMReg()->next());
#endif // _LP64
if (save_fpu_registers) {
if (UseSSE < 2) {
int fpu_off = float_regs_as_doubles_off;
for (int n = 0; n < FrameMap::nof_fpu_regs; n++) {
VMReg fpu_name_0 = FrameMap::fpu_regname(n);
map->set_callee_saved(VMRegImpl::stack2reg(fpu_off + num_rt_args), fpu_name_0);
if (true) {
map->set_callee_saved(VMRegImpl::stack2reg(fpu_off + 1 + num_rt_args), fpu_name_0->next());
}
fpu_off += 2;
}
assert(fpu_off == fpu_state_off, "incorrect number of fpu stack slots");
}
if (UseSSE >= 2) {
int xmm_off = xmm_regs_as_doubles_off;
for (int n = 0; n < FrameMap::nof_xmm_regs; n++) {
VMReg xmm_name_0 = as_XMMRegister(n)->as_VMReg();
map->set_callee_saved(VMRegImpl::stack2reg(xmm_off + num_rt_args), xmm_name_0);
if (true) {
map->set_callee_saved(VMRegImpl::stack2reg(xmm_off + 1 + num_rt_args), xmm_name_0->next());
}
xmm_off += 2;
}
assert(xmm_off == float_regs_as_doubles_off, "incorrect number of xmm registers");
} else if (UseSSE == 1) {
int xmm_off = xmm_regs_as_doubles_off;
for (int n = 0; n < FrameMap::nof_xmm_regs; n++) {
VMReg xmm_name_0 = as_XMMRegister(n)->as_VMReg();
map->set_callee_saved(VMRegImpl::stack2reg(xmm_off + num_rt_args), xmm_name_0);
xmm_off += 2;
}
assert(xmm_off == float_regs_as_doubles_off, "incorrect number of xmm registers");
}
}
return map;
}
static OopMap* save_live_registers(StubAssembler* sasm, int num_rt_args,
bool save_fpu_registers = true) {
__ block_comment("save_live_registers");
__ pusha(); // integer registers
__ subptr(rsp, extra_space_offset * VMRegImpl::stack_slot_size);
#ifdef ASSERT
__ movptr(Address(rsp, marker * VMRegImpl::stack_slot_size), (int32_t)0xfeedbeef);
#endif
if (save_fpu_registers) {
if (UseSSE < 2) {
__ fnsave(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size));
__ fwait();
#ifdef ASSERT
Label ok;
__ cmpw(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size), StubRoutines::fpu_cntrl_wrd_std());
__ jccb(Assembler::equal, ok);
__ stop("corrupted control word detected");
__ bind(ok);
#endif
__ movw(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size), StubRoutines::fpu_cntrl_wrd_std());
__ frstor(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 0));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 8));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48));
__ fstp_d(Address(rsp, float_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56));
}
if (UseSSE >= 2) {
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 0), xmm0);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 8), xmm1);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16), xmm2);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24), xmm3);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32), xmm4);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40), xmm5);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48), xmm6);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56), xmm7);
#ifdef _LP64
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 64), xmm8);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 72), xmm9);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 80), xmm10);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 88), xmm11);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 96), xmm12);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 104), xmm13);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 112), xmm14);
__ movdbl(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 120), xmm15);
#endif // _LP64
} else if (UseSSE == 1) {
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 0), xmm0);
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 8), xmm1);
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16), xmm2);
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24), xmm3);
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32), xmm4);
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40), xmm5);
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48), xmm6);
__ movflt(Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56), xmm7);
}
}
__ verify_FPU(0, "save_live_registers");
return generate_oop_map(sasm, num_rt_args, save_fpu_registers);
}
static void restore_fpu(StubAssembler* sasm, bool restore_fpu_registers = true) {
if (restore_fpu_registers) {
if (UseSSE >= 2) {
__ movdbl(xmm0, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 0));
__ movdbl(xmm1, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 8));
__ movdbl(xmm2, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16));
__ movdbl(xmm3, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24));
__ movdbl(xmm4, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32));
__ movdbl(xmm5, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40));
__ movdbl(xmm6, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48));
__ movdbl(xmm7, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56));
#ifdef _LP64
__ movdbl(xmm8, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 64));
__ movdbl(xmm9, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 72));
__ movdbl(xmm10, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 80));
__ movdbl(xmm11, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 88));
__ movdbl(xmm12, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 96));
__ movdbl(xmm13, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 104));
__ movdbl(xmm14, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 112));
__ movdbl(xmm15, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 120));
#endif // _LP64
} else if (UseSSE == 1) {
__ movflt(xmm0, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 0));
__ movflt(xmm1, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 8));
__ movflt(xmm2, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 16));
__ movflt(xmm3, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 24));
__ movflt(xmm4, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 32));
__ movflt(xmm5, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 40));
__ movflt(xmm6, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 48));
__ movflt(xmm7, Address(rsp, xmm_regs_as_doubles_off * VMRegImpl::stack_slot_size + 56));
}
if (UseSSE < 2) {
__ frstor(Address(rsp, fpu_state_off * VMRegImpl::stack_slot_size));
} else {
__ verify_FPU(0, "restore_live_registers");
}
} else {
__ verify_FPU(0, "restore_live_registers");
}
#ifdef ASSERT
{
Label ok;
__ cmpptr(Address(rsp, marker * VMRegImpl::stack_slot_size), (int32_t)0xfeedbeef);
__ jcc(Assembler::equal, ok);
__ stop("bad offsets in frame");
__ bind(ok);
}
#endif // ASSERT
__ addptr(rsp, extra_space_offset * VMRegImpl::stack_slot_size);
}
static void restore_live_registers(StubAssembler* sasm, bool restore_fpu_registers = true) {
__ block_comment("restore_live_registers");
restore_fpu(sasm, restore_fpu_registers);
__ popa();
}
static void restore_live_registers_except_rax(StubAssembler* sasm, bool restore_fpu_registers = true) {
__ block_comment("restore_live_registers_except_rax");
restore_fpu(sasm, restore_fpu_registers);
#ifdef _LP64
__ movptr(r15, Address(rsp, 0));
__ movptr(r14, Address(rsp, wordSize));
__ movptr(r13, Address(rsp, 2 * wordSize));
__ movptr(r12, Address(rsp, 3 * wordSize));
__ movptr(r11, Address(rsp, 4 * wordSize));
__ movptr(r10, Address(rsp, 5 * wordSize));
__ movptr(r9, Address(rsp, 6 * wordSize));
__ movptr(r8, Address(rsp, 7 * wordSize));
__ movptr(rdi, Address(rsp, 8 * wordSize));
__ movptr(rsi, Address(rsp, 9 * wordSize));
__ movptr(rbp, Address(rsp, 10 * wordSize));
__ movptr(rbx, Address(rsp, 12 * wordSize));
__ movptr(rdx, Address(rsp, 13 * wordSize));
__ movptr(rcx, Address(rsp, 14 * wordSize));
__ addptr(rsp, 16 * wordSize);
#else
__ pop(rdi);
__ pop(rsi);
__ pop(rbp);
__ pop(rbx); // skip this value
__ pop(rbx);
__ pop(rdx);
__ pop(rcx);
__ addptr(rsp, BytesPerWord);
#endif // _LP64
}
void Runtime1::initialize_pd() {
}
OopMapSet* Runtime1::generate_exception_throw(StubAssembler* sasm, address target, bool has_argument) {
int num_rt_args = has_argument ? 2 : 1;
OopMap* oop_map = save_live_registers(sasm, num_rt_args);
__ invalidate_registers(true, true, true, true, true, true);
const Register temp_reg = rbx;
if (has_argument) {
#ifdef _LP64
__ movptr(c_rarg1, Address(rbp, 2*BytesPerWord));
#else
__ movptr(temp_reg, Address(rbp, 2*BytesPerWord));
__ push(temp_reg);
#endif // _LP64
}
int call_offset = __ call_RT(noreg, noreg, target, num_rt_args - 1);
OopMapSet* oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, oop_map);
__ stop("should not reach here");
return oop_maps;
}
OopMapSet* Runtime1::generate_handle_exception(StubID id, StubAssembler *sasm) {
__ block_comment("generate_handle_exception");
const Register exception_oop = rax;
const Register exception_pc = rdx;
const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
OopMapSet* oop_maps = new OopMapSet();
OopMap* oop_map = NULL;
switch (id) {
case forward_exception_id:
oop_map = generate_oop_map(sasm, 1 /*thread*/);
__ movptr(exception_oop, Address(thread, Thread::pending_exception_offset()));
__ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
__ movptr(exception_pc, Address(rbp, 1*BytesPerWord));
__ movptr(Address(thread, JavaThread::vm_result_offset()), NULL_WORD);
__ movptr(Address(thread, JavaThread::vm_result_2_offset()), NULL_WORD);
break;
case handle_exception_nofpu_id:
case handle_exception_id:
oop_map = save_live_registers(sasm, 1 /*thread*/, id != handle_exception_nofpu_id);
break;
case handle_exception_from_callee_id: {
const int frame_size = 2 /*BP, return address*/ NOT_LP64(+ 1 /*thread*/) WIN64_ONLY(+ frame::arg_reg_save_area_bytes / BytesPerWord);
oop_map = new OopMap(frame_size * VMRegImpl::slots_per_word, 0);
sasm->set_frame_size(frame_size);
WIN64_ONLY(__ subq(rsp, frame::arg_reg_save_area_bytes));
break;
}
default: ShouldNotReachHere();
}
#ifdef TIERED
if (UseSSE < 2) {
__ empty_FPU_stack();
}
#endif // TIERED
__ invalidate_registers(false, true, true, false, true, true);
__ verify_not_null_oop(exception_oop);
NOT_LP64(__ get_thread(thread);)
#ifdef ASSERT
Label oop_empty;
__ cmpptr(Address(thread, JavaThread::exception_oop_offset()), (int32_t) NULL_WORD);
__ jcc(Assembler::equal, oop_empty);
__ stop("exception oop already set");
__ bind(oop_empty);
Label pc_empty;
__ cmpptr(Address(thread, JavaThread::exception_pc_offset()), 0);
__ jcc(Assembler::equal, pc_empty);
__ stop("exception pc already set");
__ bind(pc_empty);
#endif
__ movptr(Address(thread, JavaThread::exception_oop_offset()), exception_oop);
__ movptr(Address(thread, JavaThread::exception_pc_offset()), exception_pc);
__ movptr(Address(rbp, 1*BytesPerWord), exception_pc);
int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, exception_handler_for_pc));
oop_maps->add_gc_map(call_offset, oop_map);
__ invalidate_registers(false, true, true, true, true, true);
__ movptr(Address(rbp, 1*BytesPerWord), rax);
switch (id) {
case forward_exception_id:
case handle_exception_nofpu_id:
case handle_exception_id:
restore_live_registers(sasm, id != handle_exception_nofpu_id);
break;
case handle_exception_from_callee_id:
__ leave();
__ pop(rcx);
__ jmp(rcx); // jump to exception handler
break;
default: ShouldNotReachHere();
}
return oop_maps;
}
void Runtime1::generate_unwind_exception(StubAssembler *sasm) {
const Register exception_oop = rax;
const Register exception_oop_callee_saved = NOT_LP64(rsi) LP64_ONLY(r14);
const Register exception_pc = rdx;
const Register handler_addr = rbx;
const Register thread = NOT_LP64(rdi) LP64_ONLY(r15_thread);
__ invalidate_registers(false, true, true, true, true, true);
#ifdef ASSERT
NOT_LP64(__ get_thread(thread);)
Label oop_empty;
__ cmpptr(Address(thread, JavaThread::exception_oop_offset()), 0);
__ jcc(Assembler::equal, oop_empty);
__ stop("exception oop must be empty");
__ bind(oop_empty);
Label pc_empty;
__ cmpptr(Address(thread, JavaThread::exception_pc_offset()), 0);
__ jcc(Assembler::equal, pc_empty);
__ stop("exception pc must be empty");
__ bind(pc_empty);
#endif
__ empty_FPU_stack();
__ verify_not_null_oop(exception_oop);
__ movptr(exception_oop_callee_saved, exception_oop);
NOT_LP64(__ get_thread(thread);)
__ movptr(exception_pc, Address(rsp, 0));
__ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc);
__ invalidate_registers(false, true, true, true, false, true);
__ movptr(handler_addr, rax);
__ movptr(exception_oop, exception_oop_callee_saved);
__ verify_not_null_oop(exception_oop);
__ pop(exception_pc);
__ jmp(handler_addr);
}
OopMapSet* Runtime1::generate_patching(StubAssembler* sasm, address target) {
const int num_rt_args = 2; // thread + dummy
DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
assert(deopt_blob != NULL, "deoptimization blob must have been created");
OopMap* oop_map = save_live_registers(sasm, num_rt_args);
#ifdef _LP64
const Register thread = r15_thread;
__ mov(c_rarg0, thread);
#else
__ push(rax); // push dummy
const Register thread = rdi; // is callee-saved register (Visual C++ calling conventions)
__ get_thread(thread);
__ push(thread);
#endif // _LP64
__ set_last_Java_frame(thread, noreg, rbp, NULL);
__ call(RuntimeAddress(target));
OopMapSet* oop_maps = new OopMapSet();
oop_maps->add_gc_map(__ offset(), oop_map);
#ifdef ASSERT
guarantee(thread != rax, "change this code");
__ push(rax);
{ Label L;
__ get_thread(rax);
__ cmpptr(thread, rax);
__ jcc(Assembler::equal, L);
__ stop("StubAssembler::call_RT: rdi/r15 not callee saved?");
__ bind(L);
}
__ pop(rax);
#endif
__ reset_last_Java_frame(thread, true);
#ifndef _LP64
__ pop(rcx); // discard thread arg
__ pop(rcx); // discard dummy
#endif // _LP64
{ Label L;
__ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
__ jcc(Assembler::equal, L);
__ testptr(rax, rax); // have we deoptimized?
__ jump_cc(Assembler::equal,
RuntimeAddress(Runtime1::entry_for(Runtime1::forward_exception_id)));
__ movptr(rax, Address(thread, Thread::pending_exception_offset()));
__ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD);
__ verify_not_null_oop(rax);
__ movptr(rdx, Address(rsp, return_off * VMRegImpl::stack_slot_size));
#ifdef ASSERT
Label oop_empty;
__ cmpptr(Address(thread, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD);
__ jcc(Assembler::equal, oop_empty);
__ stop("exception oop must be empty");
__ bind(oop_empty);
Label pc_empty;
__ cmpptr(Address(thread, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
__ jcc(Assembler::equal, pc_empty);
__ stop("exception pc must be empty");
__ bind(pc_empty);
#endif
__ movptr(Address(thread, JavaThread::exception_oop_offset()), rax);
__ movptr(Address(thread, JavaThread::exception_pc_offset()), rdx);
restore_live_registers(sasm);
__ leave();
__ addptr(rsp, BytesPerWord); // remove return address from stack
__ jump(RuntimeAddress(deopt_blob->unpack_with_exception_in_tls()));
__ bind(L);
}
Label reexecuteEntry, cont;
__ testptr(rax, rax); // have we deoptimized?
__ jcc(Assembler::equal, cont); // no
restore_live_registers(sasm);
__ leave();
__ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
__ bind(cont);
restore_live_registers(sasm);
__ leave();
__ ret(0);
return oop_maps;
}
OopMapSet* Runtime1::generate_code_for(StubID id, StubAssembler* sasm) {
const bool must_gc_arguments = true;
const bool dont_gc_arguments = false;
bool save_fpu_registers = true;
OopMapSet* oop_maps = NULL;
switch (id) {
case forward_exception_id:
{
oop_maps = generate_handle_exception(id, sasm);
__ leave();
__ ret(0);
}
break;
case new_instance_id:
case fast_new_instance_id:
case fast_new_instance_init_check_id:
{
Register klass = rdx; // Incoming
Register obj = rax; // Result
if (id == new_instance_id) {
__ set_info("new_instance", dont_gc_arguments);
} else if (id == fast_new_instance_id) {
__ set_info("fast new_instance", dont_gc_arguments);
} else {
assert(id == fast_new_instance_init_check_id, "bad StubID");
__ set_info("fast new_instance init check", dont_gc_arguments);
}
if ((id == fast_new_instance_id || id == fast_new_instance_init_check_id) &&
UseTLAB && FastTLABRefill) {
Label slow_path;
Register obj_size = rcx;
Register t1 = rbx;
Register t2 = rsi;
assert_different_registers(klass, obj, obj_size, t1, t2);
__ push(rdi);
__ push(rbx);
if (id == fast_new_instance_init_check_id) {
__ cmpb(Address(klass, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
__ jcc(Assembler::notEqual, slow_path);
}
#ifdef ASSERT
{
Label ok, not_ok;
__ movl(obj_size, Address(klass, Klass::layout_helper_offset()));
__ cmpl(obj_size, 0); // make sure it's an instance (LH > 0)
__ jcc(Assembler::lessEqual, not_ok);
__ testl(obj_size, Klass::_lh_instance_slow_path_bit);
__ jcc(Assembler::zero, ok);
__ bind(not_ok);
__ stop("assert(can be fast path allocated)");
__ should_not_reach_here();
__ bind(ok);
}
#endif // ASSERT
Label retry_tlab, try_eden;
const Register thread =
__ tlab_refill(retry_tlab, try_eden, slow_path); // does not destroy rdx (klass), returns rdi
__ bind(retry_tlab);
__ movl(obj_size, Address(klass, Klass::layout_helper_offset()));
__ tlab_allocate(obj, obj_size, 0, t1, t2, slow_path);
__ initialize_object(obj, klass, obj_size, 0, t1, t2);
__ verify_oop(obj);
__ pop(rbx);
__ pop(rdi);
__ ret(0);
__ bind(try_eden);
__ movl(obj_size, Address(klass, Klass::layout_helper_offset()));
__ eden_allocate(obj, obj_size, 0, t1, slow_path);
__ incr_allocated_bytes(thread, obj_size, 0);
__ initialize_object(obj, klass, obj_size, 0, t1, t2);
__ verify_oop(obj);
__ pop(rbx);
__ pop(rdi);
__ ret(0);
__ bind(slow_path);
__ pop(rbx);
__ pop(rdi);
}
__ enter();
OopMap* map = save_live_registers(sasm, 2);
int call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_instance), klass);
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, map);
restore_live_registers_except_rax(sasm);
__ verify_oop(obj);
__ leave();
__ ret(0);
}
break;
case counter_overflow_id:
{
Register bci = rax, method = rbx;
__ enter();
OopMap* map = save_live_registers(sasm, 3);
__ movl(bci, Address(rbp, 2*BytesPerWord));
__ movptr(method, Address(rbp, 3*BytesPerWord));
int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, counter_overflow), bci, method);
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, map);
restore_live_registers(sasm);
__ leave();
__ ret(0);
}
break;
case new_type_array_id:
case new_object_array_id:
{
Register length = rbx; // Incoming
Register klass = rdx; // Incoming
Register obj = rax; // Result
if (id == new_type_array_id) {
__ set_info("new_type_array", dont_gc_arguments);
} else {
__ set_info("new_object_array", dont_gc_arguments);
}
#ifdef ASSERT
{
Label ok;
Register t0 = obj;
__ movl(t0, Address(klass, Klass::layout_helper_offset()));
__ sarl(t0, Klass::_lh_array_tag_shift);
int tag = ((id == new_type_array_id)
? Klass::_lh_array_tag_type_value
: Klass::_lh_array_tag_obj_value);
__ cmpl(t0, tag);
__ jcc(Assembler::equal, ok);
__ stop("assert(is an array klass)");
__ should_not_reach_here();
__ bind(ok);
}
#endif // ASSERT
if (UseTLAB && FastTLABRefill) {
Register arr_size = rsi;
Register t1 = rcx; // must be rcx for use as shift count
Register t2 = rdi;
Label slow_path;
assert_different_registers(length, klass, obj, arr_size, t1, t2);
__ cmpl(length, C1_MacroAssembler::max_array_allocation_length);
__ jcc(Assembler::above, slow_path);
Label retry_tlab, try_eden;
const Register thread =
__ tlab_refill(retry_tlab, try_eden, slow_path); // preserves rbx & rdx, returns rdi
__ bind(retry_tlab);
__ movl(t1, Address(klass, Klass::layout_helper_offset()));
__ movl(arr_size, length);
assert(t1 == rcx, "fixed register usage");
__ shlptr(arr_size /* by t1=rcx, mod 32 */);
__ shrptr(t1, Klass::_lh_header_size_shift);
__ andptr(t1, Klass::_lh_header_size_mask);
__ addptr(arr_size, t1);
__ addptr(arr_size, MinObjAlignmentInBytesMask); // align up
__ andptr(arr_size, ~MinObjAlignmentInBytesMask);
__ tlab_allocate(obj, arr_size, 0, t1, t2, slow_path); // preserves arr_size
__ initialize_header(obj, klass, length, t1, t2);
__ movb(t1, Address(klass, in_bytes(Klass::layout_helper_offset()) + (Klass::_lh_header_size_shift / BitsPerByte)));
assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
__ andptr(t1, Klass::_lh_header_size_mask);
__ subptr(arr_size, t1); // body length
__ addptr(t1, obj); // body start
__ initialize_body(t1, arr_size, 0, t2);
__ verify_oop(obj);
__ ret(0);
__ bind(try_eden);
__ movl(t1, Address(klass, Klass::layout_helper_offset()));
__ movl(arr_size, length);
assert(t1 == rcx, "fixed register usage");
__ shlptr(arr_size /* by t1=rcx, mod 32 */);
__ shrptr(t1, Klass::_lh_header_size_shift);
__ andptr(t1, Klass::_lh_header_size_mask);
__ addptr(arr_size, t1);
__ addptr(arr_size, MinObjAlignmentInBytesMask); // align up
__ andptr(arr_size, ~MinObjAlignmentInBytesMask);
__ eden_allocate(obj, arr_size, 0, t1, slow_path); // preserves arr_size
__ incr_allocated_bytes(thread, arr_size, 0);
__ initialize_header(obj, klass, length, t1, t2);
__ movb(t1, Address(klass, in_bytes(Klass::layout_helper_offset()) + (Klass::_lh_header_size_shift / BitsPerByte)));
assert(Klass::_lh_header_size_shift % BitsPerByte == 0, "bytewise");
assert(Klass::_lh_header_size_mask <= 0xFF, "bytewise");
__ andptr(t1, Klass::_lh_header_size_mask);
__ subptr(arr_size, t1); // body length
__ addptr(t1, obj); // body start
__ initialize_body(t1, arr_size, 0, t2);
__ verify_oop(obj);
__ ret(0);
__ bind(slow_path);
}
__ enter();
OopMap* map = save_live_registers(sasm, 3);
int call_offset;
if (id == new_type_array_id) {
call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_type_array), klass, length);
} else {
call_offset = __ call_RT(obj, noreg, CAST_FROM_FN_PTR(address, new_object_array), klass, length);
}
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, map);
restore_live_registers_except_rax(sasm);
__ verify_oop(obj);
__ leave();
__ ret(0);
}
break;
case new_multi_array_id:
{ StubFrame f(sasm, "new_multi_array", dont_gc_arguments);
OopMap* map = save_live_registers(sasm, 4);
int call_offset = __ call_RT(rax, noreg, CAST_FROM_FN_PTR(address, new_multi_array), rax, rbx, rcx);
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, map);
restore_live_registers_except_rax(sasm);
__ verify_oop(rax);
}
break;
case register_finalizer_id:
{
__ set_info("register_finalizer", dont_gc_arguments);
#ifdef _LP64
__ verify_oop(c_rarg0);
__ mov(rax, c_rarg0);
#else
__ movptr(rax, Address(rsp, 1 * BytesPerWord));
__ verify_oop(rax);
#endif // _LP64
Label register_finalizer;
Register t = rsi;
__ load_klass(t, rax);
__ movl(t, Address(t, Klass::access_flags_offset()));
__ testl(t, JVM_ACC_HAS_FINALIZER);
__ jcc(Assembler::notZero, register_finalizer);
__ ret(0);
__ bind(register_finalizer);
__ enter();
OopMap* oop_map = save_live_registers(sasm, 2 /*num_rt_args */);
int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, SharedRuntime::register_finalizer), rax);
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, oop_map);
restore_live_registers(sasm);
__ leave();
__ ret(0);
}
break;
case throw_range_check_failed_id:
{ StubFrame f(sasm, "range_check_failed", dont_gc_arguments);
oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_range_check_exception), true);
}
break;
case throw_index_exception_id:
{ StubFrame f(sasm, "index_range_check_failed", dont_gc_arguments);
oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_index_exception), true);
}
break;
case throw_div0_exception_id:
{ StubFrame f(sasm, "throw_div0_exception", dont_gc_arguments);
oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_div0_exception), false);
}
break;
case throw_null_pointer_exception_id:
{ StubFrame f(sasm, "throw_null_pointer_exception", dont_gc_arguments);
oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_null_pointer_exception), false);
}
break;
case handle_exception_nofpu_id:
case handle_exception_id:
{ StubFrame f(sasm, "handle_exception", dont_gc_arguments);
oop_maps = generate_handle_exception(id, sasm);
}
break;
case handle_exception_from_callee_id:
{ StubFrame f(sasm, "handle_exception_from_callee", dont_gc_arguments);
oop_maps = generate_handle_exception(id, sasm);
}
break;
case unwind_exception_id:
{ __ set_info("unwind_exception", dont_gc_arguments);
generate_unwind_exception(sasm);
}
break;
case throw_array_store_exception_id:
{ StubFrame f(sasm, "throw_array_store_exception", dont_gc_arguments);
oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_array_store_exception), true);
}
break;
case throw_class_cast_exception_id:
{ StubFrame f(sasm, "throw_class_cast_exception", dont_gc_arguments);
oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_class_cast_exception), true);
}
break;
case throw_incompatible_class_change_error_id:
{ StubFrame f(sasm, "throw_incompatible_class_cast_exception", dont_gc_arguments);
oop_maps = generate_exception_throw(sasm, CAST_FROM_FN_PTR(address, throw_incompatible_class_change_error), false);
}
break;
case slow_subtype_check_id:
{
enum layout {
rax_off, SLOT2(raxH_off)
rcx_off, SLOT2(rcxH_off)
rsi_off, SLOT2(rsiH_off)
rdi_off, SLOT2(rdiH_off)
return_off, SLOT2(returnH_off)
sup_k_off, SLOT2(sup_kH_off)
klass_off, SLOT2(superH_off)
framesize,
result_off = klass_off // deepest argument is also the return value
};
__ set_info("slow_subtype_check", dont_gc_arguments);
__ push(rdi);
__ push(rsi);
__ push(rcx);
__ push(rax);
__ movptr(rsi, Address(rsp, (klass_off) * VMRegImpl::stack_slot_size)); // subclass
__ movptr(rax, Address(rsp, (sup_k_off) * VMRegImpl::stack_slot_size)); // superclass
Label miss;
__ check_klass_subtype_slow_path(rsi, rax, rcx, rdi, NULL, &miss);
__ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), 1); // result
__ pop(rax);
__ pop(rcx);
__ pop(rsi);
__ pop(rdi);
__ ret(0);
__ bind(miss);
__ movptr(Address(rsp, (result_off) * VMRegImpl::stack_slot_size), NULL_WORD); // result
__ pop(rax);
__ pop(rcx);
__ pop(rsi);
__ pop(rdi);
__ ret(0);
}
break;
case monitorenter_nofpu_id:
save_fpu_registers = false;
case monitorenter_id:
{
StubFrame f(sasm, "monitorenter", dont_gc_arguments);
OopMap* map = save_live_registers(sasm, 3, save_fpu_registers);
f.load_argument(1, rax); // rax,: object
f.load_argument(0, rbx); // rbx,: lock address
int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorenter), rax, rbx);
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, map);
restore_live_registers(sasm, save_fpu_registers);
}
break;
case monitorexit_nofpu_id:
save_fpu_registers = false;
case monitorexit_id:
{
StubFrame f(sasm, "monitorexit", dont_gc_arguments);
OopMap* map = save_live_registers(sasm, 2, save_fpu_registers);
f.load_argument(0, rax); // rax,: lock address
int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, monitorexit), rax);
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, map);
restore_live_registers(sasm, save_fpu_registers);
}
break;
case deoptimize_id:
{
StubFrame f(sasm, "deoptimize", dont_gc_arguments);
const int num_rt_args = 1; // thread
OopMap* oop_map = save_live_registers(sasm, num_rt_args);
int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, deoptimize));
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, oop_map);
restore_live_registers(sasm);
DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
assert(deopt_blob != NULL, "deoptimization blob must have been created");
__ leave();
__ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
}
break;
case access_field_patching_id:
{ StubFrame f(sasm, "access_field_patching", dont_gc_arguments);
oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, access_field_patching));
}
break;
case load_klass_patching_id:
{ StubFrame f(sasm, "load_klass_patching", dont_gc_arguments);
oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_klass_patching));
}
break;
case load_mirror_patching_id:
{ StubFrame f(sasm, "load_mirror_patching", dont_gc_arguments);
oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_mirror_patching));
}
break;
case load_appendix_patching_id:
{ StubFrame f(sasm, "load_appendix_patching", dont_gc_arguments);
oop_maps = generate_patching(sasm, CAST_FROM_FN_PTR(address, move_appendix_patching));
}
break;
case dtrace_object_alloc_id:
{ // rax,: object
StubFrame f(sasm, "dtrace_object_alloc", dont_gc_arguments);
save_live_registers(sasm, 1);
__ NOT_LP64(push(rax)) LP64_ONLY(mov(c_rarg0, rax));
__ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc)));
NOT_LP64(__ pop(rax));
restore_live_registers(sasm);
}
break;
case fpu2long_stub_id:
{
__ push(rsi);
__ push(rcx);
LP64_ONLY(__ push(rdx);)
Label return0, do_return, return_min_jlong, do_convert;
Address value_high_word(rsp, wordSize + 4);
Address value_low_word(rsp, wordSize);
Address result_high_word(rsp, 3*wordSize + 4);
Address result_low_word(rsp, 3*wordSize);
__ subptr(rsp, 32); // more than enough on 32bit
__ fst_d(value_low_word);
__ movl(rax, value_high_word);
__ andl(rax, 0x7ff00000);
__ cmpl(rax, 0x7ff00000);
__ jcc(Assembler::notEqual, do_convert);
__ movl(rax, value_high_word);
__ andl(rax, 0xfffff);
__ orl(rax, value_low_word);
__ jcc(Assembler::notZero, return0);
__ bind(do_convert);
__ fnstcw(Address(rsp, 0));
__ movzwl(rax, Address(rsp, 0));
__ orl(rax, 0xc00);
__ movw(Address(rsp, 2), rax);
__ fldcw(Address(rsp, 2));
__ fwait();
__ fistp_d(result_low_word);
__ fldcw(Address(rsp, 0));
__ fwait();
__ movptr(rax, result_low_word);
__ movl(rdx, result_high_word);
__ mov(rcx, rax);
__ xorl(rcx, 0x0);
__ movl(rsi, 0x80000000);
__ xorl(rsi, rdx);
__ orl(rcx, rsi);
__ jcc(Assembler::notEqual, do_return);
__ fldz();
__ fcomp_d(value_low_word);
__ fnstsw_ax();
#ifdef _LP64
__ testl(rax, 0x4100); // ZF & CF == 0
__ jcc(Assembler::equal, return_min_jlong);
#else
__ sahf();
__ jcc(Assembler::above, return_min_jlong);
#endif // _LP64
#ifndef _LP64
__ movl(rdx, 0x7fffffff);
__ movl(rax, 0xffffffff);
#else
__ mov64(rax, CONST64(0x7fffffffffffffff));
#endif // _LP64
__ jmp(do_return);
__ bind(return_min_jlong);
#ifndef _LP64
__ movl(rdx, 0x80000000);
__ xorl(rax, rax);
#else
__ mov64(rax, CONST64(0x8000000000000000));
#endif // _LP64
__ jmp(do_return);
__ bind(return0);
__ fpop();
#ifndef _LP64
__ xorptr(rdx,rdx);
__ xorptr(rax,rax);
#else
__ xorptr(rax, rax);
#endif // _LP64
__ bind(do_return);
__ addptr(rsp, 32);
LP64_ONLY(__ pop(rdx);)
__ pop(rcx);
__ pop(rsi);
__ ret(0);
}
break;
#if INCLUDE_ALL_GCS
case g1_pre_barrier_slow_id:
{
StubFrame f(sasm, "g1_pre_barrier", dont_gc_arguments);
BarrierSet* bs = Universe::heap()->barrier_set();
if (bs->kind() != BarrierSet::G1SATBCTLogging) {
__ movptr(rax, (int)id);
__ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), rax);
__ should_not_reach_here();
break;
}
__ push(rax);
__ push(rdx);
const Register pre_val = rax;
const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
const Register tmp = rdx;
NOT_LP64(__ get_thread(thread);)
Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
PtrQueue::byte_offset_of_active()));
Address queue_index(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
PtrQueue::byte_offset_of_index()));
Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() +
PtrQueue::byte_offset_of_buf()));
Label done;
Label runtime;
#ifdef _LP64
__ movslq(tmp, queue_index);
__ cmpq(tmp, 0);
#else
__ cmpl(queue_index, 0);
#endif
__ jcc(Assembler::equal, runtime);
#ifdef _LP64
__ subq(tmp, wordSize);
__ movl(queue_index, tmp);
__ addq(tmp, buffer);
#else
__ subl(queue_index, wordSize);
__ movl(tmp, buffer);
__ addl(tmp, queue_index);
#endif
f.load_argument(0, pre_val);
__ movptr(Address(tmp, 0), pre_val);
__ jmp(done);
__ bind(runtime);
save_live_registers(sasm, 3);
f.load_argument(0, rcx);
__ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), rcx, thread);
restore_live_registers(sasm);
__ bind(done);
__ pop(rdx);
__ pop(rax);
}
break;
case g1_post_barrier_slow_id:
{
StubFrame f(sasm, "g1_post_barrier", dont_gc_arguments);
Address store_addr(rbp, 2*BytesPerWord);
BarrierSet* bs = Universe::heap()->barrier_set();
CardTableModRefBS* ct = (CardTableModRefBS*)bs;
assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
Label done;
Label runtime;
const Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);
Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
PtrQueue::byte_offset_of_index()));
Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() +
PtrQueue::byte_offset_of_buf()));
__ push(rax);
__ push(rcx);
const Register cardtable = rax;
const Register card_addr = rcx;
f.load_argument(0, card_addr);
__ shrptr(card_addr, CardTableModRefBS::card_shift);
__ movptr(cardtable, (intptr_t)ct->byte_map_base);
__ addptr(card_addr, cardtable);
NOT_LP64(__ get_thread(thread);)
__ cmpb(Address(card_addr, 0), (int)G1SATBCardTableModRefBS::g1_young_card_val());
__ jcc(Assembler::equal, done);
__ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
__ cmpb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
__ jcc(Assembler::equal, done);
__ movb(Address(card_addr, 0), (int)CardTableModRefBS::dirty_card_val());
__ cmpl(queue_index, 0);
__ jcc(Assembler::equal, runtime);
__ subl(queue_index, wordSize);
const Register buffer_addr = rbx;
__ push(rbx);
__ movptr(buffer_addr, buffer);
#ifdef _LP64
__ movslq(rscratch1, queue_index);
__ addptr(buffer_addr, rscratch1);
#else
__ addptr(buffer_addr, queue_index);
#endif
__ movptr(Address(buffer_addr, 0), card_addr);
__ pop(rbx);
__ jmp(done);
__ bind(runtime);
__ push(rdx);
save_live_registers(sasm, 3);
__ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread);
restore_live_registers(sasm);
__ pop(rdx);
__ bind(done);
__ pop(rcx);
__ pop(rax);
}
break;
#endif // INCLUDE_ALL_GCS
case predicate_failed_trap_id:
{
StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments);
OopMap* map = save_live_registers(sasm, 1);
int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap));
oop_maps = new OopMapSet();
oop_maps->add_gc_map(call_offset, map);
restore_live_registers(sasm);
__ leave();
DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob();
assert(deopt_blob != NULL, "deoptimization blob must have been created");
__ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution()));
}
break;
default:
{ StubFrame f(sasm, "unimplemented entry", dont_gc_arguments);
__ movptr(rax, (int)id);
__ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, unimplemented_entry), rax);
__ should_not_reach_here();
}
break;
}
return oop_maps;
}
#undef __
const char *Runtime1::pd_name_for_address(address entry) {
return "<unknown function>";
}
C:\hotspot-69087d08d473\src\cpu\x86\vm/c2_globals_x86.hpp
#ifndef CPU_X86_VM_C2_GLOBALS_X86_HPP
#define CPU_X86_VM_C2_GLOBALS_X86_HPP
#include "utilities/globalDefinitions.hpp"
#include "utilities/macros.hpp"
define_pd_global(bool, BackgroundCompilation, true);
define_pd_global(bool, UseTLAB, true);
define_pd_global(bool, ResizeTLAB, true);
define_pd_global(bool, CICompileOSR, true);
define_pd_global(bool, InlineIntrinsics, true);
define_pd_global(bool, PreferInterpreterNativeStubs, false);
define_pd_global(bool, ProfileTraps, true);
define_pd_global(bool, UseOnStackReplacement, true);
#ifdef CC_INTERP
define_pd_global(bool, ProfileInterpreter, false);
#else
define_pd_global(bool, ProfileInterpreter, true);
#endif // CC_INTERP
define_pd_global(bool, TieredCompilation, trueInTiered);
define_pd_global(intx, CompileThreshold, 10000);
define_pd_global(intx, BackEdgeThreshold, 100000);
define_pd_global(intx, OnStackReplacePercentage, 140);
define_pd_global(intx, ConditionalMoveLimit, 3);
define_pd_global(intx, FLOATPRESSURE, 6);
define_pd_global(intx, FreqInlineSize, 325);
define_pd_global(intx, MinJumpTableSize, 10);
#ifdef AMD64
define_pd_global(intx, INTPRESSURE, 13);
define_pd_global(intx, InteriorEntryAlignment, 16);
define_pd_global(intx, NewSizeThreadIncrease, ScaleForWordSize(4*K));
define_pd_global(intx, LoopUnrollLimit, 60);
define_pd_global(intx, InitialCodeCacheSize, 2496*K); // Integral multiple of CodeCacheExpansionSize
define_pd_global(intx, CodeCacheExpansionSize, 64*K);
define_pd_global(uint64_t,MaxRAM, 128ULL*G);
#else
define_pd_global(intx, INTPRESSURE, 6);
define_pd_global(intx, InteriorEntryAlignment, 4);
define_pd_global(intx, NewSizeThreadIncrease, 4*K);
define_pd_global(intx, LoopUnrollLimit, 50); // Design center runs on 1.3.1
define_pd_global(intx, InitialCodeCacheSize, 2304*K); // Integral multiple of CodeCacheExpansionSize
define_pd_global(intx, CodeCacheExpansionSize, 32*K);
define_pd_global(uint64_t,MaxRAM, 4ULL*G);
#endif // AMD64
define_pd_global(intx, RegisterCostAreaRatio, 16000);
define_pd_global(bool, OptoPeephole, true);
define_pd_global(bool, UseCISCSpill, true);
define_pd_global(bool, OptoScheduling, false);
define_pd_global(bool, OptoBundling, false);
define_pd_global(intx, ReservedCodeCacheSize, 48*M);
define_pd_global(uintx, CodeCacheMinBlockLength, 4);
define_pd_global(uintx, CodeCacheMinimumUseSpace, 400*K);
define_pd_global(bool, TrapBasedRangeChecks, false); // Not needed on x86.
define_pd_global(uintx,MetaspaceSize, ScaleForWordSize(16*M));
define_pd_global(bool, NeverActAsServerClassMachine, false);
#endif // CPU_X86_VM_C2_GLOBALS_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/c2_init_x86.cpp
#include "precompiled.hpp"
#include "opto/compile.hpp"
#include "opto/node.hpp"
void Compile::pd_compiler2_init() {
guarantee(CodeEntryAlignment >= InteriorEntryAlignment, "" );
#ifndef AMD64
if (!VM_Version::supports_cmov()) {
ConditionalMoveLimit = 0;
}
#endif // AMD64
}
C:\hotspot-69087d08d473\src\cpu\x86\vm/codeBuffer_x86.hpp
#ifndef CPU_X86_VM_CODEBUFFER_X86_HPP
#define CPU_X86_VM_CODEBUFFER_X86_HPP
private:
void pd_initialize() {}
public:
void flush_bundle(bool start_new_bundle) {}
#endif // CPU_X86_VM_CODEBUFFER_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/compiledIC_x86.cpp
#include "precompiled.hpp"
#include "asm/macroAssembler.inline.hpp"
#include "code/compiledIC.hpp"
#include "code/icBuffer.hpp"
#include "code/nmethod.hpp"
#include "memory/resourceArea.hpp"
#include "runtime/mutexLocker.hpp"
#include "runtime/safepoint.hpp"
void CompiledIC::cleanup_call_site(virtual_call_Relocation* call_site) {
NativeCall* call = nativeCall_at(call_site->addr());
if (is_icholder_entry(call->destination())) {
NativeMovConstReg* value = nativeMovConstReg_at(call_site->cached_value());
InlineCacheBuffer::queue_for_release((CompiledICHolder*)value->data());
}
}
bool CompiledIC::is_icholder_call_site(virtual_call_Relocation* call_site) {
NativeCall* call = nativeCall_at(call_site->addr());
return is_icholder_entry(call->destination());
}
#define __ _masm.
address CompiledStaticCall::emit_to_interp_stub(CodeBuffer &cbuf) {
address mark = cbuf.insts_mark(); // Get mark within main instrs section.
MacroAssembler _masm(&cbuf);
address base = __ start_a_stub(to_interp_stub_size());
if (base == NULL) {
return NULL; // CodeBuffer::expand failed.
}
__ relocate(static_stub_Relocation::spec(mark), Assembler::imm_operand);
__ mov_metadata(rbx, (Metadata*) NULL); // Method is zapped till fixup time.
__ jump(RuntimeAddress(__ pc()));
__ end_a_stub();
return base;
}
#undef __
int CompiledStaticCall::to_interp_stub_size() {
return NOT_LP64(10) // movl; jmp
LP64_ONLY(15); // movq (1+1+8); jmp (1+4)
}
int CompiledStaticCall::reloc_to_interp_stub() {
return 4; // 3 in emit_to_interp_stub + 1 in emit_call
}
void CompiledStaticCall::set_to_interpreted(methodHandle callee, address entry) {
address stub = find_stub();
guarantee(stub != NULL, "stub not found");
if (TraceICs) {
ResourceMark rm;
tty->print_cr("CompiledStaticCall@" INTPTR_FORMAT ": set_to_interpreted %s",
p2i(instruction_address()),
callee->name_and_sig_as_C_string());
}
NativeMovConstReg* method_holder = nativeMovConstReg_at(stub);
NativeJump* jump = nativeJump_at(method_holder->next_instruction_address());
assert(method_holder->data() == 0 || method_holder->data() == (intptr_t)callee(),
"a) MT-unsafe modification of inline cache");
assert(jump->jump_destination() == (address)-1 || jump->jump_destination() == entry,
"b) MT-unsafe modification of inline cache");
method_holder->set_data((intptr_t)callee());
jump->set_jump_destination(entry);
set_destination_mt_safe(stub);
}
void CompiledStaticCall::set_stub_to_clean(static_stub_Relocation* static_stub) {
assert (CompiledIC_lock->is_locked() || SafepointSynchronize::is_at_safepoint(), "mt unsafe call");
address stub = static_stub->addr();
assert(stub != NULL, "stub not found");
NativeMovConstReg* method_holder = nativeMovConstReg_at(stub);
NativeJump* jump = nativeJump_at(method_holder->next_instruction_address());
method_holder->set_data(0);
jump->set_jump_destination((address)-1);
}
#ifndef PRODUCT
void CompiledStaticCall::verify() {
NativeCall::verify();
if (os::is_MP()) {
verify_alignment();
}
address stub = find_stub();
assert(stub != NULL, "no stub found for static call");
NativeMovConstReg* method_holder = nativeMovConstReg_at(stub);
NativeJump* jump = nativeJump_at(method_holder->next_instruction_address());
assert(is_clean() || is_call_to_compiled() || is_call_to_interpreted(), "sanity check");
}
#endif // !PRODUCT
C:\hotspot-69087d08d473\src\cpu\x86\vm/copy_x86.hpp
#ifndef CPU_X86_VM_COPY_X86_HPP
#define CPU_X86_VM_COPY_X86_HPP
#ifdef TARGET_OS_ARCH_linux_x86
# include "copy_linux_x86.inline.hpp"
#endif
#ifdef TARGET_OS_ARCH_solaris_x86
# include "copy_solaris_x86.inline.hpp"
#endif
#ifdef TARGET_OS_ARCH_windows_x86
# include "copy_windows_x86.inline.hpp"
#endif
#ifdef TARGET_OS_ARCH_bsd_x86
# include "copy_bsd_x86.inline.hpp"
#endif
static void pd_fill_to_words(HeapWord* tohw, size_t count, juint value) {
#ifdef AMD64
julong* to = (julong*) tohw;
julong v = ((julong) value << 32) | value;
while (count-- > 0) {
}
#else
juint* to = (juint*)tohw;
count *= HeapWordSize / BytesPerInt;
while (count-- > 0) {
}
#endif // AMD64
}
static void pd_fill_to_aligned_words(HeapWord* tohw, size_t count, juint value) {
pd_fill_to_words(tohw, count, value);
}
static void pd_fill_to_bytes(void* to, size_t count, jubyte value) {
(void)memset(to, value, count);
}
static void pd_zero_to_words(HeapWord* tohw, size_t count) {
pd_fill_to_words(tohw, count, 0);
}
static void pd_zero_to_bytes(void* to, size_t count) {
(void)memset(to, 0, count);
}
#endif // CPU_X86_VM_COPY_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/cppInterpreterGenerator_x86.hpp
#ifndef CPU_X86_VM_CPPINTERPRETERGENERATOR_X86_HPP
#define CPU_X86_VM_CPPINTERPRETERGENERATOR_X86_HPP
protected:
#if 0
address generate_asm_interpreter_entry(bool synchronized);
address generate_native_entry(bool synchronized);
address generate_abstract_entry(void);
address generate_math_entry(AbstractInterpreter::MethodKind kind);
address generate_empty_entry(void);
address generate_accessor_entry(void);
address generate_Reference_get_entry(void);
void lock_method(void);
void generate_stack_overflow_check(void);
void generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue);
void generate_counter_overflow(Label* do_continue);
#endif
void generate_more_monitors();
void generate_deopt_handling();
address generate_interpreter_frame_manager(bool synchronized); // C++ interpreter only
void generate_compute_interpreter_state(const Register state,
const Register prev_state,
const Register sender_sp,
bool native); // C++ interpreter only
#endif // CPU_X86_VM_CPPINTERPRETERGENERATOR_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/cppInterpreter_x86.cpp
#include "precompiled.hpp"
#include "asm/macroAssembler.hpp"
#include "interpreter/bytecodeHistogram.hpp"
#include "interpreter/cppInterpreter.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interpreterGenerator.hpp"
#include "interpreter/interpreterRuntime.hpp"
#include "oops/arrayOop.hpp"
#include "oops/methodData.hpp"
#include "oops/method.hpp"
#include "oops/oop.inline.hpp"
#include "prims/jvmtiExport.hpp"
#include "prims/jvmtiThreadState.hpp"
#include "runtime/arguments.hpp"
#include "runtime/deoptimization.hpp"
#include "runtime/frame.inline.hpp"
#include "runtime/interfaceSupport.hpp"
#include "runtime/sharedRuntime.hpp"
#include "runtime/stubRoutines.hpp"
#include "runtime/synchronizer.hpp"
#include "runtime/timer.hpp"
#include "runtime/vframeArray.hpp"
#include "utilities/debug.hpp"
#include "utilities/macros.hpp"
#ifdef SHARK
#include "shark/shark_globals.hpp"
#endif
#ifdef CC_INTERP
extern "C" void RecursiveInterpreterActivation(interpreterState istate )
{
ShouldNotReachHere();
}
#define __ _masm->
#define STATE(field_name) (Address(state, byte_offset_of(BytecodeInterpreter, field_name)))
Label fast_accessor_slow_entry_path; // fast accessor methods need to be able to jmp to unsynchronized
const Register state = NOT_LP64(rsi) LP64_ONLY(r13);
const Register sender_sp_on_entry = NOT_LP64(rsi) LP64_ONLY(r13);
static address unctrap_frame_manager_entry = NULL;
static address deopt_frame_manager_return_atos = NULL;
static address deopt_frame_manager_return_btos = NULL;
static address deopt_frame_manager_return_itos = NULL;
static address deopt_frame_manager_return_ltos = NULL;
static address deopt_frame_manager_return_ftos = NULL;
static address deopt_frame_manager_return_dtos = NULL;
static address deopt_frame_manager_return_vtos = NULL;
int AbstractInterpreter::BasicType_as_index(BasicType type) {
int i = 0;
switch (type) {
case T_BOOLEAN: i = 0; break;
case T_CHAR : i = 1; break;
case T_BYTE : i = 2; break;
case T_SHORT : i = 3; break;
case T_INT : i = 4; break;
case T_VOID : i = 5; break;
case T_FLOAT : i = 8; break;
case T_LONG : i = 9; break;
case T_DOUBLE : i = 6; break;
case T_OBJECT : // fall through
case T_ARRAY : i = 7; break;
default : ShouldNotReachHere();
}
assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
return i;
}
bool CppInterpreter::contains(address pc) {
return (_code->contains(pc) ||
pc == CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
}
address CppInterpreterGenerator::generate_result_handler_for(BasicType type) {
address entry = __ pc();
switch (type) {
case T_BOOLEAN: __ c2bool(rax); break;
case T_CHAR : __ andl(rax, 0xFFFF); break;
case T_BYTE : __ sign_extend_byte (rax); break;
case T_SHORT : __ sign_extend_short(rax); break;
case T_VOID : // fall thru
case T_LONG : // fall thru
case T_INT : /* nothing to do */ break;
case T_DOUBLE :
case T_FLOAT :
{
const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
__ pop(t); // remove return address first
if (type == T_FLOAT && UseSSE >= 1) {
#ifndef _LP64
__ fld_d(Address(rsp, 0));
__ fstp_s(Address(rsp, 0));
#endif // !_LP64
__ movflt(xmm0, Address(rsp, 0));
} else if (type == T_DOUBLE && UseSSE >= 2 ) {
__ movdbl(xmm0, Address(rsp, 0));
} else {
__ fld_d(Address(rsp, 0));
}
__ addptr(rsp, 2 * wordSize);
__ push(t); // restore return address
}
break;
case T_OBJECT :
__ movptr(rax, STATE(_oop_temp));
__ verify_oop(rax);
break;
default : ShouldNotReachHere();
}
__ ret(0); // return from result handler
return entry;
}
#undef EXTEND // SHOULD NOT BE NEEDED
address CppInterpreterGenerator::generate_tosca_to_stack_converter(BasicType type) {
address entry = __ pc();
const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
__ pop(t); // remove return address first
switch (type) {
case T_VOID:
break;
case T_BOOLEAN:
#ifdef EXTEND
__ c2bool(rax);
#endif
__ push(rax);
break;
case T_CHAR :
#ifdef EXTEND
__ andl(rax, 0xFFFF);
#endif
__ push(rax);
break;
case T_BYTE :
#ifdef EXTEND
__ sign_extend_byte (rax);
#endif
__ push(rax);
break;
case T_SHORT :
#ifdef EXTEND
__ sign_extend_short(rax);
#endif
__ push(rax);
break;
case T_LONG :
__ push(rdx); // pushes useless junk on 64bit
__ push(rax);
break;
case T_INT :
__ push(rax);
break;
case T_FLOAT :
__ subptr(rsp, wordSize);
if ( UseSSE < 1) {
__ fstp_s(Address(rsp, 0));
} else {
__ movflt(Address(rsp, 0), xmm0);
}
break;
case T_DOUBLE :
__ subptr(rsp, 2*wordSize);
if ( UseSSE < 2 ) {
__ fstp_d(Address(rsp, 0));
} else {
__ movdbl(Address(rsp, 0), xmm0);
}
break;
case T_OBJECT :
__ verify_oop(rax); // verify it
__ push(rax);
break;
default : ShouldNotReachHere();
}
__ jmp(t); // return from result handler
return entry;
}
address CppInterpreterGenerator::generate_stack_to_stack_converter(BasicType type) {
address entry = __ pc();
const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp();
switch (type) {
case T_VOID:
__ movptr(rax, STATE(_locals)); // pop parameters get new stack value
__ addptr(rax, wordSize); // account for prepush before we return
break;
case T_FLOAT :
case T_BOOLEAN:
case T_CHAR :
case T_BYTE :
case T_SHORT :
case T_INT :
__ movptr(rdx, STATE(_stack));
__ movptr(rax, STATE(_locals)); // address for result
__ movl(rdx, Address(rdx, wordSize)); // get result
__ movptr(Address(rax, 0), rdx); // and store it
break;
case T_LONG :
case T_DOUBLE :
__ movptr(rax, STATE(_locals)); // address for result
__ movptr(rcx, STATE(_stack));
__ subptr(rax, wordSize); // need addition word besides locals[0]
__ movptr(rdx, Address(rcx, 2*wordSize)); // get result word (junk in 64bit)
__ movptr(Address(rax, wordSize), rdx); // and store it
__ movptr(rdx, Address(rcx, wordSize)); // get result word
__ movptr(Address(rax, 0), rdx); // and store it
break;
case T_OBJECT :
__ movptr(rdx, STATE(_stack));
__ movptr(rax, STATE(_locals)); // address for result
__ movptr(rdx, Address(rdx, wordSize)); // get result
__ verify_oop(rdx); // verify it
__ movptr(Address(rax, 0), rdx); // and store it
break;
default : ShouldNotReachHere();
}
__ ret(0);
return entry;
}
address CppInterpreterGenerator::generate_stack_to_native_abi_converter(BasicType type) {
address entry = __ pc();
switch (type) {
case T_VOID:
break;
case T_BOOLEAN:
case T_CHAR :
case T_BYTE :
case T_SHORT :
case T_INT :
__ movptr(rdx, STATE(_stack)); // get top of stack
__ movl(rax, Address(rdx, wordSize)); // get result word 1
break;
case T_LONG :
__ movptr(rdx, STATE(_stack)); // get top of stack
__ movptr(rax, Address(rdx, wordSize)); // get result low word
NOT_LP64(__ movl(rdx, Address(rdx, 2*wordSize));) // get result high word
break;
case T_FLOAT :
__ movptr(rdx, STATE(_stack)); // get top of stack
if ( UseSSE >= 1) {
__ movflt(xmm0, Address(rdx, wordSize));
} else {
__ fld_s(Address(rdx, wordSize)); // pushd float result
}
break;
case T_DOUBLE :
__ movptr(rdx, STATE(_stack)); // get top of stack
if ( UseSSE > 1) {
__ movdbl(xmm0, Address(rdx, wordSize));
} else {
__ fld_d(Address(rdx, wordSize)); // push double result
}
break;
case T_OBJECT :
__ movptr(rdx, STATE(_stack)); // get top of stack
__ movptr(rax, Address(rdx, wordSize)); // get result word 1
__ verify_oop(rax); // verify it
break;
default : ShouldNotReachHere();
}
__ ret(0);
return entry;
}
address CppInterpreter::return_entry(TosState state, int length, Bytecodes::Code code) {
return CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation);
}
address CppInterpreter::deopt_entry(TosState state, int length) {
address ret = NULL;
if (length != 0) {
switch (state) {
case atos: ret = deopt_frame_manager_return_atos; break;
case btos: ret = deopt_frame_manager_return_btos; break;
case ctos:
case stos:
case itos: ret = deopt_frame_manager_return_itos; break;
case ltos: ret = deopt_frame_manager_return_ltos; break;
case ftos: ret = deopt_frame_manager_return_ftos; break;
case dtos: ret = deopt_frame_manager_return_dtos; break;
case vtos: ret = deopt_frame_manager_return_vtos; break;
}
} else {
ret = unctrap_frame_manager_entry; // re-execute the bytecode ( e.g. uncommon trap)
}
assert(ret != NULL, "Not initialized");
return ret;
}
void CppInterpreterGenerator::generate_compute_interpreter_state(const Register state,
const Register locals,
const Register sender_sp,
bool native) {
const Address const_offset (rbx, Method::const_offset());
if (!native) {
#ifdef PRODUCT
__ subptr(rsp, 2*wordSize);
#else /* PRODUCT */
__ push((int32_t)NULL_WORD);
__ push(state); // make it look like a real argument
#endif /* PRODUCT */
}
__ push(rax);
__ enter();
__ mov(rax, state); // save current state
__ lea(rsp, Address(rsp, -(int)sizeof(BytecodeInterpreter)));
__ mov(state, rsp);
__ movptr(STATE(_locals), locals); // state->_locals = locals()
__ movptr(STATE(_self_link), state); // point to self
__ movptr(STATE(_prev_link), rax); // state->_link = state on entry (NULL or previous state)
__ movptr(STATE(_sender_sp), sender_sp); // state->_sender_sp = sender_sp
#ifdef _LP64
__ movptr(STATE(_thread), r15_thread); // state->_bcp = codes()
#else
__ get_thread(rax); // get vm's javathread*
__ movptr(STATE(_thread), rax); // state->_bcp = codes()
#endif // _LP64
__ movptr(rdx, Address(rbx, Method::const_offset())); // get constantMethodOop
__ lea(rdx, Address(rdx, ConstMethod::codes_offset())); // get code base
if (native) {
__ movptr(STATE(_bcp), (int32_t)NULL_WORD); // state->_bcp = NULL
} else {
__ movptr(STATE(_bcp), rdx); // state->_bcp = codes()
}
__ xorptr(rdx, rdx);
__ movptr(STATE(_oop_temp), rdx); // state->_oop_temp = NULL (only really needed for native)
__ movptr(STATE(_mdx), rdx); // state->_mdx = NULL
__ movptr(rdx, Address(rbx, Method::const_offset()));
__ movptr(rdx, Address(rdx, ConstMethod::constants_offset()));
__ movptr(rdx, Address(rdx, ConstantPool::cache_offset_in_bytes()));
__ movptr(STATE(_constants), rdx); // state->_constants = constants()
__ movptr(STATE(_method), rbx); // state->_method = method()
__ movl(STATE(_msg), (int32_t) BytecodeInterpreter::method_entry); // state->_msg = initial method entry
__ movptr(STATE(_result._to_call._callee), (int32_t) NULL_WORD); // state->_result._to_call._callee_callee = NULL
__ movptr(STATE(_monitor_base), rsp); // set monitor block bottom (grows down) this would point to entry [0]
{
const Address access_flags (rbx, Method::access_flags_offset());
const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
Label not_synced;
__ movl(rax, access_flags);
__ testl(rax, JVM_ACC_SYNCHRONIZED);
__ jcc(Assembler::zero, not_synced);
Label done;
const int mirror_offset = in_bytes(Klass::java_mirror_offset());
__ movl(rax, access_flags);
__ testl(rax, JVM_ACC_STATIC);
__ movptr(rax, Address(locals, 0)); // get receiver (assume this is frequent case)
__ jcc(Assembler::zero, done);
__ movptr(rax, Address(rbx, Method::const_offset()));
__ movptr(rax, Address(rax, ConstMethod::constants_offset()));
__ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
__ movptr(rax, Address(rax, mirror_offset));
__ bind(done);
__ subptr(rsp, entry_size); // add space for a monitor entry
__ movptr(Address(rsp, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
__ bind(not_synced);
}
__ movptr(STATE(_stack_base), rsp); // set expression stack base ( == &monitors[-count])
if (native) {
__ movptr(STATE(_stack), rsp); // set current expression stack tos
__ movptr(STATE(_stack_limit), rsp);
} else {
__ subptr(rsp, wordSize); // pre-push stack
__ movptr(STATE(_stack), rsp); // set current expression stack tos
__ movptr(rdx, Address(rbx, Method::const_offset()));
__ load_unsigned_short(rdx, Address(rdx, ConstMethod::max_stack_offset())); // get size of expression stack in words
__ negptr(rdx); // so we can subtract in next step
__ lea(rsp, Address(rsp, rdx, Address::times_ptr, -Method::extra_stack_words()));
__ movptr(STATE(_stack_limit), rsp);
}
#ifdef _LP64
__ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
__ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
#endif // _LP64
}
void InterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) {
Label done;
const Address invocation_counter(rax,
MethodCounters::invocation_counter_offset() +
InvocationCounter::counter_offset());
const Address backedge_counter (rax,
MethodCounter::backedge_counter_offset() +
InvocationCounter::counter_offset());
__ get_method_counters(rbx, rax, done);
if (ProfileInterpreter) {
__ incrementl(Address(rax,
MethodCounters::interpreter_invocation_counter_offset()));
}
__ movl(rcx, invocation_counter);
__ increment(rcx, InvocationCounter::count_increment);
__ movl(invocation_counter, rcx); // save invocation count
__ movl(rax, backedge_counter); // load backedge counter
__ andl(rax, InvocationCounter::count_mask_value); // mask out the status bits
__ addl(rcx, rax); // add both counters
__ cmp32(rcx,
ExternalAddress((address)&InvocationCounter::InterpreterInvocationLimit));
__ jcc(Assembler::aboveEqual, *overflow);
__ bind(done);
}
void InterpreterGenerator::generate_counter_overflow(Label* do_continue) {
__ movptr(rax, (int32_t)false);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rax);
__ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter))); // restore state
__ movptr(rbx, Address(state, byte_offset_of(BytecodeInterpreter, _method))); // restore method
__ movptr(rdi, Address(state, byte_offset_of(BytecodeInterpreter, _locals))); // get locals pointer
__ jmp(*do_continue, relocInfo::none);
}
void InterpreterGenerator::generate_stack_overflow_check(void) {
const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
const int overhead_size = (int)sizeof(BytecodeInterpreter);
const int page_size = os::vm_page_size();
Label after_frame_check;
Label after_frame_check_pop;
__ push(state);
const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rsi);
NOT_LP64(__ get_thread(thread));
const Address stack_base(thread, Thread::stack_base_offset());
const Address stack_size(thread, Thread::stack_size_offset());
const int one_monitor = frame::interpreter_frame_monitor_size() * wordSize;
__ movptr(rax, Address(rbx, Method::const_offset()));
__ load_unsigned_short(rax, Address(rax, ConstMethod::max_stack_offset())); // get size of expression stack in words
__ lea(rax, Address(noreg, rax, Interpreter::stackElementScale(), one_monitor+Method::extra_stack_words()));
__ lea(rax, Address(rax, rdx, Interpreter::stackElementScale(), overhead_size));
#ifdef ASSERT
Label stack_base_okay, stack_size_okay;
__ cmpptr(stack_base, (int32_t)0);
__ jcc(Assembler::notEqual, stack_base_okay);
__ stop("stack base is zero");
__ bind(stack_base_okay);
__ cmpptr(stack_size, (int32_t)0);
__ jcc(Assembler::notEqual, stack_size_okay);
__ stop("stack size is zero");
__ bind(stack_size_okay);
#endif
__ addptr(rax, stack_base);
__ subptr(rax, stack_size);
const int slop = 6 * K;
const int max_pages = StackShadowPages > (StackRedPages+StackYellowPages) ? StackShadowPages :
(StackRedPages+StackYellowPages);
__ addptr(rax, slop + 2*max_pages * page_size);
__ cmpptr(rsp, rax);
__ jcc(Assembler::above, after_frame_check_pop);
__ pop(state); // get c++ prev state.
__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_StackOverflowError));
__ bind(after_frame_check_pop);
__ pop(state);
__ bind(after_frame_check);
}
void InterpreterGenerator::lock_method(void) {
const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
const Address access_flags (rbx, Method::access_flags_offset());
const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
__ movptr(monitor, STATE(_monitor_base)); // get monitor bottom limit
__ subptr(monitor, entry_size); // point to initial monitor
#ifdef ASSERT
{ Label L;
__ movl(rax, access_flags);
__ testl(rax, JVM_ACC_SYNCHRONIZED);
__ jcc(Assembler::notZero, L);
__ stop("method doesn't need synchronization");
__ bind(L);
}
#endif // ASSERT
{ Label done;
const int mirror_offset = in_bytes(Klass::java_mirror_offset());
__ movl(rax, access_flags);
__ movptr(rdi, STATE(_locals)); // prepare to get receiver (assume common case)
__ testl(rax, JVM_ACC_STATIC);
__ movptr(rax, Address(rdi, 0)); // get receiver (assume this is frequent case)
__ jcc(Assembler::zero, done);
__ movptr(rax, Address(rbx, Method::const_offset()));
__ movptr(rax, Address(rax, ConstMethod::constants_offset()));
__ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
__ movptr(rax, Address(rax, mirror_offset));
__ bind(done);
}
#ifdef ASSERT
{ Label L;
__ cmpptr(rax, Address(monitor, BasicObjectLock::obj_offset_in_bytes())); // correct object?
__ jcc(Assembler::equal, L);
__ stop("wrong synchronization lobject");
__ bind(L);
}
#endif // ASSERT
__ lock_object(monitor);
}
address InterpreterGenerator::generate_accessor_entry(void) {
Label xreturn_path;
if (UseFastAccessorMethods) {
address entry_point = __ pc();
Label slow_path;
ExternalAddress state(SafepointSynchronize::address_of_state());
__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
SafepointSynchronize::_not_synchronized);
__ jcc(Assembler::notEqual, slow_path);
__ movptr(rax, Address(rsp, wordSize));
__ testptr(rax, rax);
__ jcc(Assembler::zero, slow_path);
__ movptr(rdx, Address(rbx, Method::const_offset()));
__ movptr(rdi, Address(rdx, ConstMethod::constants_offset()));
__ movl(rdx, Address(rdx, ConstMethod::codes_offset()));
__ shrl(rdx, 2*BitsPerByte);
__ shll(rdx, exact_log2(in_words(ConstantPoolCacheEntry::size())));
__ movptr(rdi, Address(rdi, ConstantPool::cache_offset_in_bytes()));
assert(in_words(ConstantPoolCacheEntry::size()) == 4, "adjust shift below");
__ movl(rcx,
Address(rdi,
rdx,
Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
__ shrl(rcx, 2*BitsPerByte);
__ andl(rcx, 0xFF);
__ cmpl(rcx, Bytecodes::_getfield);
__ jcc(Assembler::notEqual, slow_path);
__ movptr(rcx,
Address(rdi,
rdx,
Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset()));
__ movl(rdx,
Address(rdi,
rdx,
Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()));
Label notByte, notBool, notShort, notChar;
const Address field_address (rax, rcx, Address::times_1);
__ shrl(rdx, ConstantPoolCacheEntry::tos_state_shift);
ConstantPoolCacheEntry::verify_tos_state_shift();
#ifdef _LP64
Label notObj;
__ cmpl(rdx, atos);
__ jcc(Assembler::notEqual, notObj);
__ movptr(rax, field_address);
__ jmp(xreturn_path);
__ bind(notObj);
#endif // _LP64
__ cmpl(rdx, ztos);
__ jcc(Assembler::notEqual, notBool);
__ load_signed_byte(rax, field_address);
__ jmp(xreturn_path);
__ cmpl(rdx, btos);
__ jcc(Assembler::notEqual, notByte);
__ load_signed_byte(rax, field_address);
__ jmp(xreturn_path);
__ bind(notByte);
__ cmpl(rdx, stos);
__ jcc(Assembler::notEqual, notShort);
__ load_signed_short(rax, field_address);
__ jmp(xreturn_path);
__ bind(notShort);
__ cmpl(rdx, ctos);
__ jcc(Assembler::notEqual, notChar);
__ load_unsigned_short(rax, field_address);
__ jmp(xreturn_path);
__ bind(notChar);
#ifdef ASSERT
Label okay;
#ifndef _LP64
__ cmpl(rdx, atos);
__ jcc(Assembler::equal, okay);
#endif // _LP64
__ cmpl(rdx, itos);
__ jcc(Assembler::equal, okay);
__ stop("what type is this?");
__ bind(okay);
#endif // ASSERT
__ movl(rax, field_address);
__ bind(xreturn_path);
__ pop(rdi); // get return address
__ mov(rsp, sender_sp_on_entry); // set sp to sender sp
__ jmp(rdi);
__ bind(slow_path);
__ jmp(fast_accessor_slow_entry_path);
return entry_point;
} else {
return NULL;
}
}
address InterpreterGenerator::generate_Reference_get_entry(void) {
#if INCLUDE_ALL_GCS
if (UseG1GC) {
Unimplemented();
}
#endif // INCLUDE_ALL_GCS
return generate_accessor_entry();
}
address InterpreterGenerator::generate_native_entry(bool synchronized) {
bool inc_counter = UseCompiler || CountCompiledCalls;
address entry_point = __ pc();
const Address constMethod (rbx, Method::const_offset());
const Address access_flags (rbx, Method::access_flags_offset());
const Address size_of_parameters(rcx, ConstMethod::size_of_parameters_offset());
const Register locals = rdi;
__ movptr(rcx, constMethod);
__ load_unsigned_short(rcx, size_of_parameters);
__ pop(rax); // get return address
__ lea(locals, Address(rsp, rcx, Address::times_ptr, -wordSize));
__ mov(rcx, sender_sp_on_entry);
__ movptr(state, (int32_t)NULL_WORD);
generate_compute_interpreter_state(state, locals, rcx, true);
#ifdef ASSERT
{ Label L;
__ movptr(rax, STATE(_stack_base));
#ifdef _LP64
__ subptr(rax, frame::arg_reg_save_area_bytes); // windows
__ andptr(rax, -16); // must be 16 byte boundary (see amd64 ABI)
#endif // _LP64
__ cmpptr(rax, rsp);
__ jcc(Assembler::equal, L);
__ stop("broken stack frame setup in interpreter");
__ bind(L);
}
#endif
const Register unlock_thread = LP64_ONLY(r15_thread) NOT_LP64(rax);
NOT_LP64(__ movptr(unlock_thread, STATE(_thread));) // get thread
const Address do_not_unlock_if_synchronized(unlock_thread,
in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
__ movbool(do_not_unlock_if_synchronized, true);
#ifdef ASSERT
__ movl(rax, access_flags);
{
Label L;
__ testl(rax, JVM_ACC_NATIVE);
__ jcc(Assembler::notZero, L);
__ stop("tried to execute non-native method as native");
__ bind(L);
}
{ Label L;
__ testl(rax, JVM_ACC_ABSTRACT);
__ jcc(Assembler::zero, L);
__ stop("tried to execute abstract method in interpreter");
__ bind(L);
}
#endif
Label invocation_counter_overflow;
if (inc_counter) {
generate_counter_incr(&invocation_counter_overflow, NULL, NULL);
}
Label continue_after_compile;
__ bind(continue_after_compile);
bang_stack_shadow_pages(true);
NOT_LP64(__ movl(rax, STATE(_thread));) // get thread
__ movbool(do_not_unlock_if_synchronized, false);
if (synchronized) {
lock_method();
} else {
#ifdef ASSERT
{ Label L;
__ movl(rax, access_flags);
__ testl(rax, JVM_ACC_SYNCHRONIZED);
__ jcc(Assembler::zero, L);
__ stop("method needs synchronization");
__ bind(L);
}
#endif
}
__ notify_method_entry();
const Register method = rbx;
const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rdi);
const Register t = InterpreterRuntime::SignatureHandlerGenerator::temp(); // rcx|rscratch1
const Address constMethod (method, Method::const_offset());
const Address size_of_parameters(t, ConstMethod::size_of_parameters_offset());
__ movptr(method, STATE(_method));
__ verify_method_ptr(method);
__ movptr(t, constMethod);
__ load_unsigned_short(t, size_of_parameters);
__ shll(t, 2);
#ifdef _LP64
__ subptr(rsp, t);
__ subptr(rsp, frame::arg_reg_save_area_bytes); // windows
__ andptr(rsp, -16); // must be 16 byte boundary (see amd64 ABI)
#else
__ addptr(t, 2*wordSize); // allocate two more slots for JNIEnv and possible mirror
__ subptr(rsp, t);
__ andptr(rsp, -(StackAlignmentInBytes)); // gcc needs 16 byte aligned stacks to do XMM intrinsics
#endif // _LP64
Label pending_exception_present;
{ Label L;
__ movptr(t, Address(method, Method::signature_handler_offset()));
__ testptr(t, t);
__ jcc(Assembler::notZero, L);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method, false);
__ movptr(method, STATE(_method));
__ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
__ jcc(Assembler::notEqual, pending_exception_present);
__ verify_method_ptr(method);
__ movptr(t, Address(method, Method::signature_handler_offset()));
__ bind(L);
}
#ifdef ASSERT
{
Label L;
__ push(t);
__ get_thread(t); // get vm's javathread*
__ cmpptr(t, STATE(_thread));
__ jcc(Assembler::equal, L);
__ int3();
__ bind(L);
__ pop(t);
}
#endif //
const Register from_ptr = InterpreterRuntime::SignatureHandlerGenerator::from();
assert(InterpreterRuntime::SignatureHandlerGenerator::to () == rsp, "adjust this code");
__ movptr(from_ptr, STATE(_locals)); // get the from pointer
__ call(t);
__ movptr(method, STATE(_method));
__ verify_method_ptr(method);
__ movptr(STATE(_result_handler), rax);
{ Label L;
__ movptr(rax, Address(method, Method::native_function_offset()));
__ testptr(rax, rax);
__ jcc(Assembler::notZero, L);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), method);
__ movptr(method, STATE(_method));
__ verify_method_ptr(method);
__ movptr(rax, Address(method, Method::native_function_offset()));
__ bind(L);
}
{ Label L;
const int mirror_offset = in_bytes(Klass::java_mirror_offset());
__ movl(t, Address(method, Method::access_flags_offset()));
__ testl(t, JVM_ACC_STATIC);
__ jcc(Assembler::zero, L);
__ movptr(t, Address(method, Method:: const_offset()));
__ movptr(t, Address(t, ConstMethod::constants_offset()));
__ movptr(t, Address(t, ConstantPool::pool_holder_offset_in_bytes()));
__ movptr(t, Address(t, mirror_offset));
__ movptr(STATE(_oop_temp), t);
#ifdef _LP64
__ lea(c_rarg1, STATE(_oop_temp));
#else
__ lea(t, STATE(_oop_temp));
__ movptr(Address(rsp, wordSize), t);
#endif // _LP64
__ bind(L);
}
#ifdef ASSERT
{
Label L;
__ push(t);
__ get_thread(t); // get vm's javathread*
__ cmpptr(t, STATE(_thread));
__ jcc(Assembler::equal, L);
__ int3();
__ bind(L);
__ pop(t);
}
#endif //
#ifdef _LP64
__ lea(c_rarg0, Address(thread, JavaThread::jni_environment_offset()));
#else
__ movptr(thread, STATE(_thread)); // get thread
__ lea(t, Address(thread, JavaThread::jni_environment_offset()));
__ movptr(Address(rsp, 0), t);
#endif // _LP64
#ifdef ASSERT
{
Label L;
__ push(t);
__ get_thread(t); // get vm's javathread*
__ cmpptr(t, STATE(_thread));
__ jcc(Assembler::equal, L);
__ int3();
__ bind(L);
__ pop(t);
}
#endif //
#ifdef ASSERT
{ Label L;
__ movl(t, Address(thread, JavaThread::thread_state_offset()));
__ cmpl(t, _thread_in_Java);
__ jcc(Assembler::equal, L);
__ stop("Wrong thread state in native stub");
__ bind(L);
}
#endif
__ set_last_Java_frame(thread, noreg, rbp, __ pc());
__ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native);
__ call(rax);
__ movptr(method, STATE(_method));
NOT_LP64(__ movptr(thread, STATE(_thread));) // get thread
{ Label Lpush, Lskip;
ExternalAddress float_handler(AbstractInterpreter::result_handler(T_FLOAT));
ExternalAddress double_handler(AbstractInterpreter::result_handler(T_DOUBLE));
__ cmpptr(STATE(_result_handler), float_handler.addr());
__ jcc(Assembler::equal, Lpush);
__ cmpptr(STATE(_result_handler), double_handler.addr());
__ jcc(Assembler::notEqual, Lskip);
__ bind(Lpush);
__ subptr(rsp, 2*wordSize);
if ( UseSSE < 2 ) {
__ fstp_d(Address(rsp, 0));
} else {
__ movdbl(Address(rsp, 0), xmm0);
}
__ bind(Lskip);
}
__ push(rax);
#ifndef _LP64
__ push(rdx);
#endif // _LP64
__ restore_cpu_control_state_after_jni();
__ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_native_trans);
if(os::is_MP()) {
__ serialize_memory(thread, rcx);
}
{ Label Continue;
__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
SafepointSynchronize::_not_synchronized);
Label L;
__ jcc(Assembler::notEqual, L);
__ cmpl(Address(thread, JavaThread::suspend_flags_offset()), 0);
__ jcc(Assembler::equal, Continue);
__ bind(L);
((MacroAssembler*)_masm)->call_VM_leaf(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans),
thread);
__ increment(rsp, wordSize);
__ movptr(method, STATE(_method));
__ verify_method_ptr(method);
__ movptr(thread, STATE(_thread)); // get thread
__ bind(Continue);
}
__ movl(Address(thread, JavaThread::thread_state_offset()), _thread_in_Java);
__ reset_last_Java_frame(thread, true, true);
__ movptr(t, Address(thread, JavaThread::active_handles_offset()));
__ movl(Address(t, JNIHandleBlock::top_offset_in_bytes()), (int32_t)NULL_WORD);
{ Label L;
Label no_oop, store_result;
ExternalAddress oop_handler(AbstractInterpreter::result_handler(T_OBJECT));
__ cmpptr(STATE(_result_handler), oop_handler.addr());
__ jcc(Assembler::notEqual, no_oop);
#ifndef _LP64
__ pop(rdx);
#endif // _LP64
__ pop(rax);
__ testptr(rax, rax);
__ jcc(Assembler::zero, store_result);
__ movptr(rax, Address(rax, 0));
__ bind(store_result);
__ movptr(STATE(_oop_temp), rax);
__ push(rax);
#ifndef _LP64
__ push(rdx);
#endif // _LP64
__ bind(no_oop);
}
{
Label no_reguard;
__ cmpl(Address(thread, JavaThread::stack_guard_state_offset()), JavaThread::stack_guard_yellow_disabled);
__ jcc(Assembler::notEqual, no_reguard);
__ pusha();
__ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::reguard_yellow_pages)));
__ popa();
__ bind(no_reguard);
}
{ Label L;
__ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
__ jcc(Assembler::zero, L);
__ bind(pending_exception_present);
__ movptr(t, STATE(_sender_sp));
__ leave(); // remove frame anchor
__ pop(rdi); // get return address
__ movptr(state, STATE(_prev_link)); // get previous state for return
__ mov(rsp, t); // set sp to sender sp
__ push(rdi); // push throwing pc
__ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
__ bind(L);
}
{ Label L;
__ movl(t, Address(method, Method::access_flags_offset()));
__ testl(t, JVM_ACC_SYNCHRONIZED);
__ jcc(Assembler::zero, L);
{ Label unlock;
const Register monitor = NOT_LP64(rdx) LP64_ONLY(c_rarg1);
__ movptr(monitor, STATE(_monitor_base));
__ subptr(monitor, frame::interpreter_frame_monitor_size() * wordSize); // address of initial monitor
__ movptr(t, Address(monitor, BasicObjectLock::obj_offset_in_bytes()));
__ testptr(t, t);
__ jcc(Assembler::notZero, unlock);
__ MacroAssembler::call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
__ should_not_reach_here();
__ bind(unlock);
__ unlock_object(monitor);
__ movptr(method, STATE(_method));
}
__ bind(L);
}
__ notify_method_exit(vtos, InterpreterMacroAssembler::NotifyJVMTI);
#ifndef _LP64
__ pop(rdx);
#endif // _LP64
__ pop(rax);
__ movptr(t, STATE(_result_handler)); // get result handler
__ call(t); // call result handler to convert to tosca form
__ movptr(t, STATE(_sender_sp));
__ leave(); // remove frame anchor
__ pop(rdi); // get return address
__ movptr(state, STATE(_prev_link)); // get previous state for return (if c++ interpreter was caller)
__ mov(rsp, t); // set sp to sender sp
__ jmp(rdi);
if (inc_counter) {
__ bind(invocation_counter_overflow);
generate_counter_overflow(&continue_after_compile);
}
return entry_point;
}
void CppInterpreterGenerator::generate_deopt_handling() {
Label return_from_deopt_common;
deopt_frame_manager_return_atos = __ pc();
__ movl(rcx, AbstractInterpreter::BasicType_as_index(T_OBJECT)); // Result stub address array index
__ jmp(return_from_deopt_common);
deopt_frame_manager_return_btos = __ pc();
__ movl(rcx, AbstractInterpreter::BasicType_as_index(T_BOOLEAN)); // Result stub address array index
__ jmp(return_from_deopt_common);
deopt_frame_manager_return_itos = __ pc();
__ movl(rcx, AbstractInterpreter::BasicType_as_index(T_INT)); // Result stub address array index
__ jmp(return_from_deopt_common);
deopt_frame_manager_return_ltos = __ pc();
__ movl(rcx, AbstractInterpreter::BasicType_as_index(T_LONG)); // Result stub address array index
__ jmp(return_from_deopt_common);
deopt_frame_manager_return_ftos = __ pc();
__ movl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
__ jmp(return_from_deopt_common);
deopt_frame_manager_return_dtos = __ pc();
__ movl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
__ jmp(return_from_deopt_common);
deopt_frame_manager_return_vtos = __ pc();
__ movl(rcx, AbstractInterpreter::BasicType_as_index(T_VOID));
__ bind(return_from_deopt_common);
__ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
__ movptr(rsp, STATE(_stack)); // trim stack (is prepushed)
__ addptr(rsp, wordSize); // undo prepush
ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
__ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr)));
__ call(rcx); // call result converter
__ movl(STATE(_msg), (int)BytecodeInterpreter::deopt_resume);
__ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
__ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
__ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
}
void CppInterpreterGenerator::generate_more_monitors() {
Label entry, loop;
const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
__ movptr(rdx, STATE(_stack_base)); // rdx: old expression stack bottom
__ subptr(rsp, entry_size); // move expression stack top limit
__ subptr(STATE(_stack), entry_size); // update interpreter stack top
__ subptr(STATE(_stack_limit), entry_size); // inform interpreter
__ subptr(rdx, entry_size); // move expression stack bottom
__ movptr(STATE(_stack_base), rdx); // inform interpreter
__ movptr(rcx, STATE(_stack)); // set start value for copy loop
__ jmp(entry);
__ bind(loop);
__ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
__ movptr(Address(rcx, 0), rbx); // and store it at new location
__ addptr(rcx, wordSize); // advance to next word
__ bind(entry);
__ cmpptr(rcx, rdx); // check if bottom reached
__ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
__ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
__ movl(STATE(_msg), (int)BytecodeInterpreter::got_monitors);
}
static address interpreter_frame_manager = NULL;
address InterpreterGenerator::generate_normal_entry(bool synchronized) {
if (interpreter_frame_manager) return interpreter_frame_manager;
address entry_point = __ pc();
if (UseFastAccessorMethods && !synchronized) __ bind(fast_accessor_slow_entry_path);
Label dispatch_entry_2;
__ movptr(rcx, sender_sp_on_entry);
__ movptr(state, (int32_t)NULL_WORD); // no current activation
__ jmp(dispatch_entry_2);
const Register locals = rdi;
Label re_dispatch;
__ bind(re_dispatch);
__ lea(rcx, Address(rsp, wordSize));
__ bind(dispatch_entry_2);
__ push(rcx);
const Address constMethod (rbx, Method::const_offset());
const Address access_flags (rbx, Method::access_flags_offset());
const Address size_of_parameters(rdx, ConstMethod::size_of_parameters_offset());
const Address size_of_locals (rdx, ConstMethod::size_of_locals_offset());
__ movptr(rdx, constMethod);
__ load_unsigned_short(rcx, size_of_parameters);
__ load_unsigned_short(rdx, size_of_locals); // get size of locals in words
__ subptr(rdx, rcx); // rdx = no. of additional locals
generate_stack_overflow_check(); // C++
bang_stack_shadow_pages(false);
__ lea(locals, Address(rsp, rcx, Address::times_ptr, wordSize));
__ pop(rcx);
__ pop(rax);
{
Label exit, loop;
__ testl(rdx, rdx); // (32bit ok)
__ jcc(Assembler::lessEqual, exit); // do nothing if rdx <= 0
__ bind(loop);
__ push((int32_t)NULL_WORD); // initialize local variables
__ decrement(rdx); // until everything initialized
__ jcc(Assembler::greater, loop);
__ bind(exit);
}
generate_compute_interpreter_state(state, locals, rcx, false);
Label call_interpreter;
__ bind(call_interpreter);
bang_stack_shadow_pages(false);
Label call_interpreter_2;
__ bind(call_interpreter_2);
{
const Register thread = NOT_LP64(rcx) LP64_ONLY(r15_thread);
#ifdef _LP64
__ mov(c_rarg0, state);
#else
__ push(state); // push arg to interpreter
__ movptr(thread, STATE(_thread));
#endif // _LP64
__ movptr(Address(thread, JavaThread::last_Java_fp_offset()), rbp);
__ movptr(Address(thread, JavaThread::last_Java_sp_offset()), rsp);
RuntimeAddress normal(CAST_FROM_FN_PTR(address, BytecodeInterpreter::run));
RuntimeAddress checking(CAST_FROM_FN_PTR(address, BytecodeInterpreter::runWithChecks));
__ call(JvmtiExport::can_post_interpreter_events() ? checking : normal);
NOT_LP64(__ pop(rax);) // discard parameter to run
NOT_LP64(__ movl(thread, STATE(_thread));)
__ reset_last_Java_frame(thread, true, true);
}
__ movl(rdx, STATE(_msg)); // Get new message
Label call_method;
Label return_from_interpreted_method;
Label throw_exception;
Label bad_msg;
Label do_OSR;
__ cmpl(rdx, (int32_t)BytecodeInterpreter::call_method);
__ jcc(Assembler::equal, call_method);
__ cmpl(rdx, (int32_t)BytecodeInterpreter::return_from_method);
__ jcc(Assembler::equal, return_from_interpreted_method);
__ cmpl(rdx, (int32_t)BytecodeInterpreter::do_osr);
__ jcc(Assembler::equal, do_OSR);
__ cmpl(rdx, (int32_t)BytecodeInterpreter::throwing_exception);
__ jcc(Assembler::equal, throw_exception);
__ cmpl(rdx, (int32_t)BytecodeInterpreter::more_monitors);
__ jcc(Assembler::notEqual, bad_msg);
generate_more_monitors();
__ jmp(call_interpreter);
unctrap_frame_manager_entry = __ pc();
__ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
__ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
__ jmp(call_interpreter_2);
generate_deopt_handling();
__ jmp(call_interpreter);
Interpreter::_rethrow_exception_entry = __ pc();
Label return_with_exception;
Label unwind_and_forward;
__ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
__ movptr(rbx, STATE(_method)); // get method
#ifdef _LP64
__ movptr(Address(r15_thread, Thread::pending_exception_offset()), rax);
#else
__ movl(rcx, STATE(_thread)); // get thread
__ movptr(Address(rcx, Thread::pending_exception_offset()), rax);
#endif // _LP64
__ movl(rdx, access_flags);
__ testl(rdx, JVM_ACC_NATIVE);
__ jcc(Assembler::zero, return_with_exception); // vanilla interpreted frame, handle directly
__ bind(unwind_and_forward);
__ movptr(rcx, STATE(_sender_sp));
__ leave();
__ pop(rdx);
__ mov(rsp, rcx);
__ push(rdx);
__ jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
Label resume_interpreter;
Label do_float;
Label do_double;
Label done_conv;
if (UseSSE < 2) {
__ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
__ movptr(rbx, STATE(_result._to_call._callee)); // get method just executed
__ movl(rcx, Address(rbx, Method::result_index_offset()));
__ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_FLOAT)); // Result stub address array index
__ jcc(Assembler::equal, do_float);
__ cmpl(rcx, AbstractInterpreter::BasicType_as_index(T_DOUBLE)); // Result stub address array index
__ jcc(Assembler::equal, do_double);
#if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2)
__ empty_FPU_stack();
#endif // COMPILER2
__ jmp(done_conv);
__ bind(do_float);
#ifdef COMPILER2
for (int i = 1; i < 8; i++) {
__ ffree(i);
}
#endif // COMPILER2
__ jmp(done_conv);
__ bind(do_double);
#ifdef COMPILER2
for (int i = 1; i < 8; i++) {
__ ffree(i);
}
#endif // COMPILER2
__ jmp(done_conv);
} else {
__ MacroAssembler::verify_FPU(0, "generate_return_entry_for compiled");
__ jmp(done_conv);
}
InternalAddress return_from_native_method(__ pc());
__ bind(done_conv);
__ lea(state, Address(rbp, -(int)sizeof(BytecodeInterpreter)));
__ movptr(rsp, STATE(_stack));
__ lea(rsp, Address(rsp, wordSize));
#ifdef _LP64
__ cmpptr(Address(r15_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
#else
__ movptr(rcx, STATE(_thread)); // get thread
__ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD);
#endif // _LP64
__ jcc(Assembler::notZero, return_with_exception);
__ movptr(rbx, STATE(_result._to_call._callee));
__ movptr(rcx, constMethod);
__ load_unsigned_short(rcx, Address(rcx, ConstMethod::size_of_parameters_offset()));
__ lea(rsp, Address(rsp, rcx, Address::times_ptr));
__ movl(rcx, Address(rbx, Method::result_index_offset()));
ExternalAddress tosca_to_stack((address)CppInterpreter::_tosca_to_stack);
__ movptr(rcx, ArrayAddress(tosca_to_stack, Address(noreg, rcx, Address::times_ptr)));
__ call(rcx); // call result converter
__ jmp(resume_interpreter);
__ bind(return_with_exception);
__ movptr(rsp, STATE(_stack_base));
__ jmp(resume_interpreter);
__ bind(return_from_interpreted_method);
Label return_to_initial_caller;
__ movptr(rbx, STATE(_method)); // get method just executed
__ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call?
__ movl(rax, Address(rbx, Method::result_index_offset())); // get result type index
__ jcc(Assembler::equal, return_to_initial_caller); // back to native code (call_stub/c1/c2)
ExternalAddress stack_to_stack((address)CppInterpreter::_stack_to_stack);
__ movptr(rax, ArrayAddress(stack_to_stack, Address(noreg, rax, Address::times_ptr)));
__ call(rax); // call result converter
Label unwind_recursive_activation;
__ bind(unwind_recursive_activation);
__ movptr(state, STATE(_prev_link)); // unwind state
__ leave(); // pop the frame
__ mov(rsp, rax); // unwind stack to remove args
__ bind(resume_interpreter);
__ movl(STATE(_msg), (int)BytecodeInterpreter::method_resume);
__ lea(rsp, Address(rsp, -wordSize)); // prepush stack (result if any already present)
__ movptr(STATE(_stack), rsp); // inform interpreter of new stack depth (parameters removed,
__ movptr(rsp, STATE(_stack_limit)); // restore expression stack to full depth
__ jmp(call_interpreter_2); // No need to bang
__ bind(return_to_initial_caller);
ExternalAddress stack_to_native((address)CppInterpreter::_stack_to_native_abi);
__ movptr(rax, ArrayAddress(stack_to_native, Address(noreg, rax, Address::times_ptr)));
__ call(rax); // call result converter
Label unwind_initial_activation;
__ bind(unwind_initial_activation);
[ incoming parameters ]
[ extra locals ]
[ return address to CALL_STUB/C1/C2]
fp -> [ CALL_STUB/C1/C2 fp ]
BytecodeInterpreter object
expression stack
sp ->
__ movptr(rcx, STATE(_sender_sp));
__ leave();
__ pop(rdi); // get return address
__ mov(rsp, rcx);
__ jmp(rdi); // return to call_stub
__ bind(do_OSR);
Label remove_initial_frame;
__ movptr(rcx, STATE(_result._osr._osr_buf));
__ movptr(rax, STATE(_result._osr._osr_entry));
__ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from "recursive" interpreter call?
__ jcc(Assembler::equal, remove_initial_frame); // back to native code (call_stub/c1/c2)
__ movptr(sender_sp_on_entry, STATE(_sender_sp)); // get sender's sp in expected register
__ leave(); // pop the frame
__ mov(rsp, sender_sp_on_entry); // trim any stack expansion
__ pushptr(return_from_native_method.addr());
__ jmp(rax);
__ bind(remove_initial_frame);
__ movptr(rdx, STATE(_sender_sp));
__ leave();
__ pop(rsi);
__ mov(rsp, rdx);
__ push(rsi);
__ jmp(rax);
__ bind(call_method);
__ movptr(rsp, STATE(_stack)); // pop args to c++ interpreter, set sp to java stack top
__ lea(rsp, Address(rsp, wordSize));
__ movptr(rbx, STATE(_result._to_call._callee)); // get method to execute
__ movptr(rcx, constMethod);
__ load_unsigned_short(rcx, Address(rcx, ConstMethod::size_of_parameters_offset()));
ExternalAddress recursive(CAST_FROM_FN_PTR(address, RecursiveInterpreterActivation));
__ pushptr(recursive.addr()); // make it look good in the debugger
InternalAddress entry(entry_point);
__ cmpptr(STATE(_result._to_call._callee_entry_point), entry.addr()); // returning to interpreter?
__ jcc(Assembler::equal, re_dispatch); // yes
__ pop(rax); // pop dummy address
__ movptr(rax, STATE(_result._to_call._callee_entry_point));
__ mov(sender_sp_on_entry, rsp);
__ pushptr(return_from_native_method.addr());
__ jmp(rax);
__ bind(bad_msg);
__ stop("Bad message from interpreter");
Label unwind_initial_with_pending_exception;
__ bind(throw_exception);
__ cmpptr(STATE(_prev_link), (int32_t)NULL_WORD); // returning from recursive interpreter call?
__ jcc(Assembler::equal, unwind_initial_with_pending_exception); // no, back to native code (call_stub/c1/c2)
__ movptr(rax, STATE(_locals)); // pop parameters get new stack value
__ addptr(rax, wordSize); // account for prepush before we return
__ jmp(unwind_recursive_activation);
__ bind(unwind_initial_with_pending_exception);
__ jmp(unwind_and_forward);
interpreter_frame_manager = entry_point;
return entry_point;
}
address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
bool synchronized = false;
address entry_point = NULL;
switch (kind) {
case Interpreter::zerolocals : break;
case Interpreter::zerolocals_synchronized: synchronized = true; break;
case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
case Interpreter::java_lang_math_sin : // fall thru
case Interpreter::java_lang_math_cos : // fall thru
case Interpreter::java_lang_math_tan : // fall thru
case Interpreter::java_lang_math_abs : // fall thru
case Interpreter::java_lang_math_log : // fall thru
case Interpreter::java_lang_math_log10 : // fall thru
case Interpreter::java_lang_math_sqrt : entry_point = ((InterpreterGenerator*)this)->generate_math_entry(kind); break;
case Interpreter::java_lang_ref_reference_get
: entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
default : ShouldNotReachHere(); break;
}
if (entry_point) return entry_point;
return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
}
InterpreterGenerator::InterpreterGenerator(StubQueue* code)
: CppInterpreterGenerator(code) {
generate_all(); // down here so it can be "virtual"
}
int AbstractInterpreter::size_top_interpreter_activation(Method* method) {
const int stub_code = 4; // see generate_call_stub
int monitor_size = method->is_synchronized() ?
1*frame::interpreter_frame_monitor_size() : 0;
const int overhead_size = sizeof(BytecodeInterpreter)/wordSize +
( frame::sender_sp_offset - frame::link_offset) + 2;
const int method_stack = (method->max_locals() + method->max_stack()) *
Interpreter::stackElementWords;
return overhead_size + method_stack + stub_code;
}
static int size_activation_helper(int extra_locals_size, int monitor_size) {
return (extra_locals_size + // the addition space for locals
2*BytesPerWord + // return address and saved rbp
2*BytesPerWord + // "static long no_params() method" issue
sizeof(BytecodeInterpreter) + // interpreterState
monitor_size); // monitors
}
void BytecodeInterpreter::layout_interpreterState(interpreterState to_fill,
frame* caller,
frame* current,
Method* method,
intptr_t* locals,
intptr_t* stack,
intptr_t* stack_base,
intptr_t* monitor_base,
intptr_t* frame_bottom,
bool is_top_frame
)
{
to_fill->_thread = JavaThread::current();
to_fill->_bcp = method->code_base();
to_fill->_locals = locals;
to_fill->_constants = method->constants()->cache();
to_fill->_method = method;
to_fill->_mdx = NULL;
to_fill->_stack = stack;
if (is_top_frame && JavaThread::current()->popframe_forcing_deopt_reexecution() ) {
to_fill->_msg = deopt_resume2;
} else {
to_fill->_msg = method_resume;
}
to_fill->_result._to_call._bcp_advance = 0;
to_fill->_result._to_call._callee_entry_point = NULL; // doesn't matter to anyone
to_fill->_result._to_call._callee = NULL; // doesn't matter to anyone
to_fill->_prev_link = NULL;
to_fill->_sender_sp = caller->unextended_sp();
if (caller->is_interpreted_frame()) {
interpreterState prev = caller->get_interpreterState();
to_fill->_prev_link = prev;
prev->_result._to_call._callee = method;
if (*prev->_bcp == Bytecodes::_invokeinterface) {
prev->_result._to_call._bcp_advance = 5;
} else {
prev->_result._to_call._bcp_advance = 3;
}
}
to_fill->_oop_temp = NULL;
to_fill->_stack_base = stack_base;
to_fill->_stack_limit = stack_base - (method->max_stack() + 1);
to_fill->_monitor_base = (BasicObjectLock*) monitor_base;
to_fill->_self_link = to_fill;
assert(stack >= to_fill->_stack_limit && stack < to_fill->_stack_base,
"Stack top out of range");
}
static int frame_size_helper(int max_stack,
int tempcount,
int moncount,
int callee_param_count,
int callee_locals,
bool is_top_frame,
int& monitor_size,
int& full_frame_size) {
int extra_locals_size = (callee_locals - callee_param_count) * BytesPerWord;
monitor_size = sizeof(BasicObjectLock) * moncount;
int short_frame_size = size_activation_helper(extra_locals_size,
monitor_size);
full_frame_size = short_frame_size + max_stack * BytesPerWord;
short_frame_size = short_frame_size + tempcount * BytesPerWord;
int frame_size = (is_top_frame ? full_frame_size : short_frame_size);
return frame_size;
}
int AbstractInterpreter::size_activation(int max_stack,
int tempcount,
int extra_args,
int moncount,
int callee_param_count,
int callee_locals,
bool is_top_frame) {
assert(extra_args == 0, "FIX ME");
int unused_monitor_size = 0;
int unused_full_frame_size = 0;
return frame_size_helper(max_stack, tempcount, moncount, callee_param_count, callee_locals,
is_top_frame, unused_monitor_size, unused_full_frame_size)/BytesPerWord;
}
void AbstractInterpreter::layout_activation(Method* method,
int tempcount, //
int popframe_extra_args,
int moncount,
int caller_actual_parameters,
int callee_param_count,
int callee_locals,
frame* caller,
frame* interpreter_frame,
bool is_top_frame,
bool is_bottom_frame) {
assert(popframe_extra_args == 0, "FIX ME");
int monitor_size = 0;
int full_frame_size = 0;
int frame_size = frame_size_helper(method->max_stack(), tempcount, moncount, callee_param_count, callee_locals,
is_top_frame, monitor_size, full_frame_size);
#ifdef ASSERT
assert(caller->unextended_sp() == interpreter_frame->interpreter_frame_sender_sp(), "Frame not properly walkable");
#endif
intptr_t* frame_bottom = (intptr_t*) ((intptr_t)interpreter_frame->sp() - (full_frame_size - frame_size));
interpreterState cur_state = (interpreterState) ((intptr_t)interpreter_frame->fp() - sizeof(BytecodeInterpreter));
intptr_t* locals;
if (caller->is_interpreted_frame()) {
interpreterState prev = caller->get_interpreterState();
locals = prev->stack() + method->size_of_parameters();
if (locals != interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2) {
}
} else {
locals = interpreter_frame->fp() + frame::sender_sp_offset + (method->max_locals() - 1) + 2;
}
intptr_t* monitor_base = (intptr_t*) cur_state;
intptr_t* stack_base = (intptr_t*) ((intptr_t) monitor_base - monitor_size);
intptr_t* stack = (intptr_t*) ((intptr_t) stack_base - (tempcount + 1) * BytesPerWord);
BytecodeInterpreter::layout_interpreterState(cur_state,
caller,
interpreter_frame,
method,
locals,
stack,
stack_base,
monitor_base,
frame_bottom,
is_top_frame);
}
#endif // CC_INTERP (all)
C:\hotspot-69087d08d473\src\cpu\x86\vm/cppInterpreter_x86.hpp
#ifndef CPU_X86_VM_CPPINTERPRETER_X86_HPP
#define CPU_X86_VM_CPPINTERPRETER_X86_HPP
protected:
const static int InterpreterCodeSize = 168 * 1024;
#endif // CPU_X86_VM_CPPINTERPRETER_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/debug_x86.cpp
#include "precompiled.hpp"
#include "code/codeCache.hpp"
#include "code/nmethod.hpp"
#include "runtime/frame.hpp"
#include "runtime/init.hpp"
#include "runtime/os.hpp"
#include "utilities/debug.hpp"
#include "utilities/top.hpp"
void pd_ps(frame f) {}
C:\hotspot-69087d08d473\src\cpu\x86\vm/depChecker_x86.cpp
#include "precompiled.hpp"
#include "compiler/disassembler.hpp"
#include "depChecker_x86.hpp"
C:\hotspot-69087d08d473\src\cpu\x86\vm/depChecker_x86.hpp
#ifndef CPU_X86_VM_DEPCHECKER_X86_HPP
#define CPU_X86_VM_DEPCHECKER_X86_HPP
#endif // CPU_X86_VM_DEPCHECKER_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/disassembler_x86.hpp
#ifndef CPU_X86_VM_DISASSEMBLER_X86_HPP
#define CPU_X86_VM_DISASSEMBLER_X86_HPP
static int pd_instruction_alignment() {
return 1;
}
static const char* pd_cpu_opts() {
return "";
}
#endif // CPU_X86_VM_DISASSEMBLER_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/frame_x86.cpp
#include "precompiled.hpp"
#include "interpreter/interpreter.hpp"
#include "memory/resourceArea.hpp"
#include "oops/markOop.hpp"
#include "oops/method.hpp"
#include "oops/oop.inline.hpp"
#include "prims/methodHandles.hpp"
#include "runtime/frame.inline.hpp"
#include "runtime/handles.inline.hpp"
#include "runtime/javaCalls.hpp"
#include "runtime/monitorChunk.hpp"
#include "runtime/os.hpp"
#include "runtime/signature.hpp"
#include "runtime/stubCodeGenerator.hpp"
#include "runtime/stubRoutines.hpp"
#include "vmreg_x86.inline.hpp"
#ifdef COMPILER1
#include "c1/c1_Runtime1.hpp"
#include "runtime/vframeArray.hpp"
#endif
#ifdef ASSERT
void RegisterMap::check_location_valid() {
}
#endif
PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
bool frame::safe_for_sender(JavaThread *thread) {
address sp = (address)_sp;
address fp = (address)_fp;
address unextended_sp = (address)_unextended_sp;
static size_t stack_guard_size = os::uses_stack_guard_pages() ? (StackYellowPages + StackRedPages) * os::vm_page_size() : 0;
size_t usable_stack_size = thread->stack_size() - stack_guard_size;
bool sp_safe = (sp < thread->stack_base()) &&
(sp >= thread->stack_base() - usable_stack_size);
if (!sp_safe) {
return false;
}
bool unextended_sp_safe = (unextended_sp < thread->stack_base()) &&
(unextended_sp >= sp);
if (!unextended_sp_safe) {
return false;
}
bool fp_safe = (fp < thread->stack_base() && (fp > sp) && (((fp + (return_addr_offset * sizeof(void*))) < thread->stack_base())));
if (_cb != NULL ) {
if (!_cb->is_frame_complete_at(_pc)) {
if (_cb->is_nmethod() || _cb->is_adapter_blob() || _cb->is_runtime_stub()) {
return false;
}
}
if (!_cb->code_contains(_pc)) {
return false;
}
if (is_entry_frame()) {
if (!fp_safe) return false;
address jcw = (address)entry_frame_call_wrapper();
bool jcw_safe = (jcw < thread->stack_base()) && ( jcw > fp);
return jcw_safe;
}
intptr_t* sender_sp = NULL;
address sender_pc = NULL;
if (is_interpreted_frame()) {
if (!fp_safe) {
return false;
}
sender_pc = (address) this->fp()[return_addr_offset];
sender_sp = (intptr_t*) addr_at(sender_sp_offset);
} else {
if (_cb->frame_size() <= 0) {
return false;
}
sender_sp = _unextended_sp + _cb->frame_size();
sender_pc = (address) *(sender_sp-1);
}
if (Interpreter::contains(sender_pc)) {
intptr_t *saved_fp = (intptr_t*)*(sender_sp - frame::sender_sp_offset);
bool saved_fp_safe = ((address)saved_fp < thread->stack_base()) && (saved_fp > sender_sp);
if (!saved_fp_safe) {
return false;
}
frame sender(sender_sp, saved_fp, sender_pc);
return sender.is_interpreted_frame_valid(thread);
}
CodeBlob* sender_blob = CodeCache::find_blob_unsafe(sender_pc);
if (sender_pc == NULL || sender_blob == NULL) {
return false;
}
if (sender_blob->is_zombie() || sender_blob->is_unloaded()) {
return false;
}
if (!sender_blob->code_contains(sender_pc)) {
return false;
}
if (sender_blob->is_adapter_blob()) {
return false;
}
if (StubRoutines::returns_to_call_stub(sender_pc)) {
intptr_t *saved_fp = (intptr_t*)*(sender_sp - frame::sender_sp_offset);
bool saved_fp_safe = ((address)saved_fp < thread->stack_base()) && (saved_fp > sender_sp);
if (!saved_fp_safe) {
return false;
}
frame sender(sender_sp, saved_fp, sender_pc);
address jcw = (address)sender.entry_frame_call_wrapper();
bool jcw_safe = (jcw < thread->stack_base()) && ( jcw > (address)sender.fp());
return jcw_safe;
}
if (sender_blob->is_nmethod()) {
nmethod* nm = sender_blob->as_nmethod_or_null();
if (nm != NULL) {
if (nm->is_deopt_mh_entry(sender_pc) || nm->is_deopt_entry(sender_pc) ||
nm->method()->is_method_handle_intrinsic()) {
return false;
}
}
}
if (sender_blob->frame_size() <= 0) {
assert(!sender_blob->is_nmethod(), "should count return address at least");
return false;
}
if (!sender_blob->is_nmethod()) {
return false;
}
return true;
}
if (!fp_safe) {
return false;
}
if ( (address) this->fp()[return_addr_offset] == NULL) return false;
return true;
}
void frame::patch_pc(Thread* thread, address pc) {
address* pc_addr = &(((address*) sp())[-1]);
if (TracePcPatching) {
tty->print_cr("patch_pc at address " INTPTR_FORMAT " [" INTPTR_FORMAT " -> " INTPTR_FORMAT "]",
pc_addr, *pc_addr, pc);
}
assert(_pc == *pc_addr || pc == *pc_addr, "must be");
_cb = CodeCache::find_blob(pc);
address original_pc = nmethod::get_deopt_original_pc(this);
if (original_pc != NULL) {
assert(original_pc == _pc, "expected original PC to be stored before patching");
_deopt_state = is_deoptimized;
} else {
_deopt_state = not_deoptimized;
_pc = pc;
}
}
bool frame::is_interpreted_frame() const {
return Interpreter::contains(pc());
}
int frame::frame_size(RegisterMap* map) const {
frame sender = this->sender(map);
return sender.sp() - sp();
}
intptr_t* frame::entry_frame_argument_at(int offset) const {
int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize);
return &unextended_sp()[index];
}
#ifdef CC_INTERP
intptr_t* frame::interpreter_frame_sender_sp() const {
assert(is_interpreted_frame(), "interpreted frame expected");
return get_interpreterState()->sender_sp();
}
BasicObjectLock* frame::interpreter_frame_monitor_begin() const {
return get_interpreterState()->monitor_base();
}
BasicObjectLock* frame::interpreter_frame_monitor_end() const {
return (BasicObjectLock*) get_interpreterState()->stack_base();
}
#else // CC_INTERP
intptr_t* frame::interpreter_frame_sender_sp() const {
assert(is_interpreted_frame(), "interpreted frame expected");
return (intptr_t*) at(interpreter_frame_sender_sp_offset);
}
void frame::set_interpreter_frame_sender_sp(intptr_t* sender_sp) {
assert(is_interpreted_frame(), "interpreted frame expected");
ptr_at_put(interpreter_frame_sender_sp_offset, (intptr_t) sender_sp);
}
BasicObjectLock* frame::interpreter_frame_monitor_begin() const {
return (BasicObjectLock*) addr_at(interpreter_frame_monitor_block_bottom_offset);
}
BasicObjectLock* frame::interpreter_frame_monitor_end() const {
BasicObjectLock* result = (BasicObjectLock*) *addr_at(interpreter_frame_monitor_block_top_offset);
assert(sp() <= (intptr_t*) result, "monitor end should be above the stack pointer");
assert((intptr_t*) result < fp(), "monitor end should be strictly below the frame pointer");
return result;
}
void frame::interpreter_frame_set_monitor_end(BasicObjectLock* value) {
}
void frame::interpreter_frame_set_last_sp(intptr_t* sp) {
}
#endif // CC_INTERP
frame frame::sender_for_entry_frame(RegisterMap* map) const {
assert(map != NULL, "map must be set");
JavaFrameAnchor* jfa = entry_frame_call_wrapper()->anchor();
assert(!entry_frame_is_first(), "next Java fp must be non zero");
assert(jfa->last_Java_sp() > sp(), "must be above this frame on stack");
if (!jfa->walkable()) {
jfa->capture_last_Java_pc();
}
map->clear();
assert(map->include_argument_oops(), "should be set by clear");
assert(jfa->last_Java_pc() != NULL, "not walkable");
frame fr(jfa->last_Java_sp(), jfa->last_Java_fp(), jfa->last_Java_pc());
return fr;
}
#ifdef ASSERT
void frame::verify_deopt_original_pc(nmethod* nm, intptr_t* unextended_sp) {
frame fr;
fr._unextended_sp = unextended_sp;
address original_pc = nm->get_original_pc(&fr);
assert(nm->insts_contains(original_pc), "original PC must be in nmethod");
}
#endif
void frame::adjust_unextended_sp() {
nmethod* sender_nm = (_cb == NULL) ? NULL : _cb->as_nmethod_or_null();
if (sender_nm != NULL) {
if (sender_nm->is_deopt_entry(_pc) ||
sender_nm->is_deopt_mh_entry(_pc)) {
DEBUG_ONLY(verify_deopt_original_pc(sender_nm, _unextended_sp));
}
}
}
void frame::update_map_with_saved_link(RegisterMap* map, intptr_t** link_addr) {
map->set_location(rbp->as_VMReg(), (address) link_addr);
#ifdef AMD64
if (true) {
map->set_location(rbp->as_VMReg()->next(), (address) link_addr);
}
#endif // AMD64
}
frame frame::sender_for_interpreter_frame(RegisterMap* map) const {
intptr_t* sender_sp = this->sender_sp();
intptr_t* unextended_sp = interpreter_frame_sender_sp();
#ifdef COMPILER2
if (map->update_map()) {
update_map_with_saved_link(map, (intptr_t**) addr_at(link_offset));
}
#endif // COMPILER2
return frame(sender_sp, unextended_sp, link(), sender_pc());
}
frame frame::sender_for_compiled_frame(RegisterMap* map) const {
assert(map != NULL, "map must be set");
assert(_cb->frame_size() >= 0, "must have non-zero frame size");
intptr_t* sender_sp = unextended_sp() + _cb->frame_size();
intptr_t* unextended_sp = sender_sp;
address sender_pc = (address) *(sender_sp-1);
intptr_t** saved_fp_addr = (intptr_t**) (sender_sp - frame::sender_sp_offset);
if (map->update_map()) {
map->set_include_argument_oops(_cb->caller_must_gc_arguments(map->thread()));
if (_cb->oop_maps() != NULL) {
OopMapSet::update_register_map(this, map);
}
update_map_with_saved_link(map, saved_fp_addr);
}
assert(sender_sp != sp(), "must have changed");
return frame(sender_sp, unextended_sp, *saved_fp_addr, sender_pc);
}
frame frame::sender(RegisterMap* map) const {
map->set_include_argument_oops(false);
if (is_entry_frame()) return sender_for_entry_frame(map);
if (is_interpreted_frame()) return sender_for_interpreter_frame(map);
assert(_cb == CodeCache::find_blob(pc()),"Must be the same");
if (_cb != NULL) {
return sender_for_compiled_frame(map);
}
return frame(sender_sp(), link(), sender_pc());
}
bool frame::interpreter_frame_equals_unpacked_fp(intptr_t* fp) {
assert(is_interpreted_frame(), "must be interpreter frame");
Method* method = interpreter_frame_method();
int diff = (method->max_locals() - method->size_of_parameters()) *
Interpreter::stackElementWords;
return _fp == (fp - diff);
}
void frame::pd_gc_epilog() {
}
bool frame::is_interpreted_frame_valid(JavaThread* thread) const {
#ifdef CC_INTERP
#else
assert(is_interpreted_frame(), "Not an interpreted frame");
if (fp() == 0 || (intptr_t(fp()) & (wordSize-1)) != 0) {
return false;
}
if (sp() == 0 || (intptr_t(sp()) & (wordSize-1)) != 0) {
return false;
}
if (fp() + interpreter_frame_initial_sp_offset < sp()) {
return false;
}
if (fp() <= sp()) { // this attempts to deal with unsigned comparison above
return false;
}
Method* m = *interpreter_frame_method_addr();
if (!m->is_valid_method()) return false;
if (fp() - sp() > 1024 + m->max_stack()*Interpreter::stackElementSize) {
return false;
}
intptr_t bcx = interpreter_frame_bcx();
if (m->validate_bci_from_bcx(bcx) < 0) {
return false;
}
ConstantPoolCache* cp = *interpreter_frame_cache_addr();
if (cp == NULL || !cp->is_metaspace_object()) return false;
address locals = (address) *interpreter_frame_locals_addr();
if (locals > thread->stack_base() || locals < (address) fp()) return false;
#endif // CC_INTERP
return true;
}
BasicType frame::interpreter_frame_result(oop* oop_result, jvalue* value_result) {
#ifdef CC_INTERP
interpreterState istate = get_interpreterState();
#endif // CC_INTERP
assert(is_interpreted_frame(), "interpreted frame expected");
Method* method = interpreter_frame_method();
BasicType type = method->result_type();
intptr_t* tos_addr;
if (method->is_native()) {
tos_addr = (intptr_t*)sp();
if (type == T_FLOAT || type == T_DOUBLE) {
#ifdef AMD64
tos_addr += 2 * Interpreter::stackElementWords;
#else
tos_addr += 2;
#endif // AMD64
}
} else {
tos_addr = (intptr_t*)interpreter_frame_tos_address();
}
switch (type) {
case T_OBJECT :
case T_ARRAY : {
oop obj;
if (method->is_native()) {
#ifdef CC_INTERP
obj = istate->_oop_temp;
#else
obj = cast_to_oop(at(interpreter_frame_oop_temp_offset));
#endif // CC_INTERP
} else {
oop* obj_p = (oop*)tos_addr;
obj = (obj_p == NULL) ? (oop)NULL : *obj_p;
}
assert(obj == NULL || Universe::heap()->is_in(obj), "sanity check");
break;
}
case T_BOOLEAN : value_result->z = *(jboolean*)tos_addr; break;
case T_BYTE : value_result->b = *(jbyte*)tos_addr; break;
case T_CHAR : value_result->c = *(jchar*)tos_addr; break;
case T_SHORT : value_result->s = *(jshort*)tos_addr; break;
case T_INT : value_result->i = *(jint*)tos_addr; break;
case T_LONG : value_result->j = *(jlong*)tos_addr; break;
case T_FLOAT : {
#ifdef AMD64
value_result->f = *(jfloat*)tos_addr;
#else
if (method->is_native()) {
jdouble d = *(jdouble*)tos_addr; // Result was in ST0 so need to convert to jfloat
value_result->f = (jfloat)d;
} else {
value_result->f = *(jfloat*)tos_addr;
}
#endif // AMD64
break;
}
case T_DOUBLE : value_result->d = *(jdouble*)tos_addr; break;
case T_VOID : /* Nothing to do */ break;
default : ShouldNotReachHere();
}
return type;
}
intptr_t* frame::interpreter_frame_tos_at(jint offset) const {
int index = (Interpreter::expr_offset_in_bytes(offset)/wordSize);
return &interpreter_frame_tos_address()[index];
}
#ifndef PRODUCT
#define DESCRIBE_FP_OFFSET(name) \
values.describe(frame_no, fp() + frame::name##_offset, #name)
void frame::describe_pd(FrameValues& values, int frame_no) {
if (is_interpreted_frame()) {
DESCRIBE_FP_OFFSET(interpreter_frame_sender_sp);
DESCRIBE_FP_OFFSET(interpreter_frame_last_sp);
DESCRIBE_FP_OFFSET(interpreter_frame_method);
DESCRIBE_FP_OFFSET(interpreter_frame_mdx);
DESCRIBE_FP_OFFSET(interpreter_frame_cache);
DESCRIBE_FP_OFFSET(interpreter_frame_locals);
DESCRIBE_FP_OFFSET(interpreter_frame_bcx);
DESCRIBE_FP_OFFSET(interpreter_frame_initial_sp);
}
}
#endif
intptr_t *frame::initial_deoptimization_info() {
return fp();
}
intptr_t* frame::real_fp() const {
if (_cb != NULL) {
int size = _cb->frame_size();
if (size > 0) {
return unextended_sp() + size;
}
}
assert(! is_compiled_frame(), "unknown compiled frame size");
return fp();
}
#ifndef PRODUCT
frame::frame(void* sp, void* fp, void* pc) {
init((intptr_t*)sp, (intptr_t*)fp, (address)pc);
}
#endif
void JavaFrameAnchor::make_walkable(JavaThread* thread) {
if (last_Java_sp() == NULL) return;
if (walkable()) return;
assert(Thread::current() == (Thread*)thread, "not current thread");
assert(last_Java_sp() != NULL, "not called from Java code?");
assert(last_Java_pc() == NULL, "already walkable");
capture_last_Java_pc();
assert(walkable(), "something went wrong");
}
void JavaFrameAnchor::capture_last_Java_pc() {
assert(_last_Java_sp != NULL, "no last frame set");
assert(_last_Java_pc == NULL, "already walkable");
_last_Java_pc = (address)_last_Java_sp[-1];
}
C:\hotspot-69087d08d473\src\cpu\x86\vm/frame_x86.hpp
#ifndef CPU_X86_VM_FRAME_X86_HPP
#define CPU_X86_VM_FRAME_X86_HPP
#include "runtime/synchronizer.hpp"
#include "utilities/top.hpp"
public:
enum {
pc_return_offset = 0,
link_offset = 0,
return_addr_offset = 1,
sender_sp_offset = 2,
#ifndef CC_INTERP
interpreter_frame_result_handler_offset = 3, // for native calls only
interpreter_frame_oop_temp_offset = 2, // for native calls only
interpreter_frame_sender_sp_offset = -1,
interpreter_frame_last_sp_offset = interpreter_frame_sender_sp_offset - 1,
interpreter_frame_method_offset = interpreter_frame_last_sp_offset - 1,
interpreter_frame_mdx_offset = interpreter_frame_method_offset - 1,
interpreter_frame_cache_offset = interpreter_frame_mdx_offset - 1,
interpreter_frame_locals_offset = interpreter_frame_cache_offset - 1,
interpreter_frame_bcx_offset = interpreter_frame_locals_offset - 1,
interpreter_frame_initial_sp_offset = interpreter_frame_bcx_offset - 1,
interpreter_frame_monitor_block_top_offset = interpreter_frame_initial_sp_offset,
interpreter_frame_monitor_block_bottom_offset = interpreter_frame_initial_sp_offset,
#endif // CC_INTERP
#ifdef AMD64
#ifdef _WIN64
entry_frame_after_call_words = 28,
entry_frame_call_wrapper_offset = 2,
arg_reg_save_area_bytes = 32, // Register argument save area
#else
entry_frame_after_call_words = 13,
entry_frame_call_wrapper_offset = -6,
arg_reg_save_area_bytes = 0,
#endif // _WIN64
#else
entry_frame_call_wrapper_offset = 2,
#endif // AMD64
native_frame_initial_param_offset = 2
};
intptr_t ptr_at(int offset) const {
return *ptr_at_addr(offset);
}
void ptr_at_put(int offset, intptr_t value) {
}
private:
intptr_t* _fp; // frame pointer
intptr_t* _unextended_sp;
void adjust_unextended_sp();
intptr_t* ptr_at_addr(int offset) const {
return (intptr_t*) addr_at(offset);
}
#ifdef ASSERT
static void verify_deopt_original_pc(nmethod* nm, intptr_t* unextended_sp);
#endif
public:
frame(intptr_t* sp, intptr_t* fp, address pc);
frame(intptr_t* sp, intptr_t* unextended_sp, intptr_t* fp, address pc);
frame(intptr_t* sp, intptr_t* fp);
void init(intptr_t* sp, intptr_t* fp, address pc);
intptr_t* fp() const { return _fp; }
inline address* sender_pc_addr() const;
inline address* native_param_addr(int idx) const;
intptr_t* interpreter_frame_last_sp() const;
static void update_map_with_saved_link(RegisterMap* map, intptr_t** link_addr);
#ifndef CC_INTERP
void interpreter_frame_set_last_sp(intptr_t* sp);
#endif // CC_INTERP
#ifdef CC_INTERP
inline interpreterState get_interpreterState() const;
#endif // CC_INTERP
#endif // CPU_X86_VM_FRAME_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/frame_x86.inline.hpp
#ifndef CPU_X86_VM_FRAME_X86_INLINE_HPP
#define CPU_X86_VM_FRAME_X86_INLINE_HPP
#include "code/codeCache.hpp"
inline frame::frame() {
_pc = NULL;
_sp = NULL;
_unextended_sp = NULL;
_fp = NULL;
_cb = NULL;
_deopt_state = unknown;
}
inline void frame::init(intptr_t* sp, intptr_t* fp, address pc) {
_sp = sp;
_unextended_sp = sp;
_fp = fp;
_pc = pc;
assert(pc != NULL, "no pc?");
_cb = CodeCache::find_blob(pc);
adjust_unextended_sp();
address original_pc = nmethod::get_deopt_original_pc(this);
if (original_pc != NULL) {
_pc = original_pc;
_deopt_state = is_deoptimized;
} else {
_deopt_state = not_deoptimized;
}
}
inline frame::frame(intptr_t* sp, intptr_t* fp, address pc) {
init(sp, fp, pc);
}
inline frame::frame(intptr_t* sp, intptr_t* unextended_sp, intptr_t* fp, address pc) {
_sp = sp;
_unextended_sp = unextended_sp;
_fp = fp;
_pc = pc;
assert(pc != NULL, "no pc?");
_cb = CodeCache::find_blob(pc);
adjust_unextended_sp();
address original_pc = nmethod::get_deopt_original_pc(this);
if (original_pc != NULL) {
_pc = original_pc;
assert(((nmethod*)_cb)->insts_contains(_pc), "original PC must be in nmethod");
_deopt_state = is_deoptimized;
} else {
_deopt_state = not_deoptimized;
}
}
inline frame::frame(intptr_t* sp, intptr_t* fp) {
_sp = sp;
_unextended_sp = sp;
_fp = fp;
_pc = (address)(sp[-1]);
_cb = CodeCache::find_blob(_pc);
adjust_unextended_sp();
address original_pc = nmethod::get_deopt_original_pc(this);
if (original_pc != NULL) {
_pc = original_pc;
_deopt_state = is_deoptimized;
} else {
_deopt_state = not_deoptimized;
}
}
inline bool frame::equal(frame other) const {
bool ret = sp() == other.sp()
&& unextended_sp() == other.unextended_sp()
&& fp() == other.fp()
&& pc() == other.pc();
assert(!ret || ret && cb() == other.cb() && _deopt_state == other._deopt_state, "inconsistent construction");
return ret;
}
inline intptr_t* frame::id(void) const { return unextended_sp(); }
inline bool frame::is_younger(intptr_t* id) const { assert(this->id() != NULL && id != NULL, "NULL frame id");
return this->id() < id ; }
inline bool frame::is_older(intptr_t* id) const { assert(this->id() != NULL && id != NULL, "NULL frame id");
return this->id() > id ; }
inline intptr_t* frame::link() const { return (intptr_t*) *(intptr_t **)addr_at(link_offset); }
inline void frame::set_link(intptr_t* addr) { *(intptr_t **)addr_at(link_offset) = addr; }
inline intptr_t* frame::unextended_sp() const { return _unextended_sp; }
inline address* frame::sender_pc_addr() const { return (address*) addr_at( return_addr_offset); }
inline address frame::sender_pc() const { return *sender_pc_addr(); }
inline address* frame::native_param_addr(int idx) const { return (address*) addr_at( native_frame_initial_param_offset+idx); }
#ifdef CC_INTERP
inline interpreterState frame::get_interpreterState() const {
return ((interpreterState)addr_at( -((int)sizeof(BytecodeInterpreter))/wordSize ));
}
inline intptr_t* frame::sender_sp() const {
if (is_interpreted_frame()) {
assert(false, "should never happen");
return get_interpreterState()->sender_sp();
} else {
return addr_at(sender_sp_offset);
}
}
inline intptr_t** frame::interpreter_frame_locals_addr() const {
assert(is_interpreted_frame(), "must be interpreted");
return &(get_interpreterState()->_locals);
}
inline intptr_t* frame::interpreter_frame_bcx_addr() const {
assert(is_interpreted_frame(), "must be interpreted");
return (intptr_t*) &(get_interpreterState()->_bcp);
}
inline ConstantPoolCache** frame::interpreter_frame_cache_addr() const {
assert(is_interpreted_frame(), "must be interpreted");
return &(get_interpreterState()->_constants);
}
inline Method** frame::interpreter_frame_method_addr() const {
assert(is_interpreted_frame(), "must be interpreted");
return &(get_interpreterState()->_method);
}
inline intptr_t* frame::interpreter_frame_mdx_addr() const {
assert(is_interpreted_frame(), "must be interpreted");
return (intptr_t*) &(get_interpreterState()->_mdx);
}
inline intptr_t* frame::interpreter_frame_tos_address() const {
assert(is_interpreted_frame(), "wrong frame type");
return get_interpreterState()->_stack + 1;
}
#else /* asm interpreter */
inline intptr_t* frame::sender_sp() const { return addr_at( sender_sp_offset); }
inline intptr_t** frame::interpreter_frame_locals_addr() const {
return (intptr_t**)addr_at(interpreter_frame_locals_offset);
}
inline intptr_t* frame::interpreter_frame_last_sp() const {
return *(intptr_t**)addr_at(interpreter_frame_last_sp_offset);
}
inline intptr_t* frame::interpreter_frame_bcx_addr() const {
return (intptr_t*)addr_at(interpreter_frame_bcx_offset);
}
inline intptr_t* frame::interpreter_frame_mdx_addr() const {
return (intptr_t*)addr_at(interpreter_frame_mdx_offset);
}
inline ConstantPoolCache** frame::interpreter_frame_cache_addr() const {
return (ConstantPoolCache**)addr_at(interpreter_frame_cache_offset);
}
inline Method** frame::interpreter_frame_method_addr() const {
return (Method**)addr_at(interpreter_frame_method_offset);
}
inline intptr_t* frame::interpreter_frame_tos_address() const {
intptr_t* last_sp = interpreter_frame_last_sp();
if (last_sp == NULL) {
return sp();
} else {
assert(last_sp <= (intptr_t*) interpreter_frame_monitor_end(), "bad tos");
return last_sp;
}
}
inline oop* frame::interpreter_frame_temp_oop_addr() const {
return (oop *)(fp() + interpreter_frame_oop_temp_offset);
}
#endif /* CC_INTERP */
inline int frame::pd_oop_map_offset_adjustment() const {
return 0;
}
inline int frame::interpreter_frame_monitor_size() {
return BasicObjectLock::size();
}
inline intptr_t* frame::interpreter_frame_expression_stack() const {
intptr_t* monitor_end = (intptr_t*) interpreter_frame_monitor_end();
return monitor_end-1;
}
inline jint frame::interpreter_frame_expression_stack_direction() { return -1; }
inline JavaCallWrapper** frame::entry_frame_call_wrapper_addr() const {
return (JavaCallWrapper**)addr_at(entry_frame_call_wrapper_offset);
}
inline int frame::local_offset_for_compiler(int local_index, int nof_args, int max_nof_locals, int max_nof_monitors) {
return (nof_args - local_index + (local_index < nof_args ? 1: -1));
}
inline int frame::monitor_offset_for_compiler(int local_index, int nof_args, int max_nof_locals, int max_nof_monitors) {
return local_offset_for_compiler(local_index, nof_args, max_nof_locals, max_nof_monitors);
}
inline int frame::min_local_offset_for_compiler(int nof_args, int max_nof_locals, int max_nof_monitors) {
return (nof_args - (max_nof_locals + max_nof_monitors*2) - 1);
}
inline bool frame::volatile_across_calls(Register reg) {
return true;
}
inline oop frame::saved_oop_result(RegisterMap* map) const {
oop* result_adr = (oop *)map->location(rax->as_VMReg());
guarantee(result_adr != NULL, "bad register save location");
return (*result_adr);
}
inline void frame::set_saved_oop_result(RegisterMap* map, oop obj) {
oop* result_adr = (oop *)map->location(rax->as_VMReg());
guarantee(result_adr != NULL, "bad register save location");
}
#endif // CPU_X86_VM_FRAME_X86_INLINE_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/globalDefinitions_x86.hpp
#ifndef CPU_X86_VM_GLOBALDEFINITIONS_X86_HPP
#define CPU_X86_VM_GLOBALDEFINITIONS_X86_HPP
const int StackAlignmentInBytes = 16;
const bool CCallingConventionRequiresIntsAsLongs = false;
#define SUPPORTS_NATIVE_CX8
#endif // CPU_X86_VM_GLOBALDEFINITIONS_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/globals_x86.hpp
#ifndef CPU_X86_VM_GLOBALS_X86_HPP
#define CPU_X86_VM_GLOBALS_X86_HPP
#include "utilities/globalDefinitions.hpp"
#include "utilities/macros.hpp"
define_pd_global(bool, ConvertSleepToYield, true);
define_pd_global(bool, CountInterpCalls, true);
define_pd_global(bool, NeedsDeoptSuspend, false); // only register window machines need this
define_pd_global(bool, ImplicitNullChecks, true); // Generate code for implicit null checks
define_pd_global(bool, TrapBasedNullChecks, false); // Not needed on x86.
define_pd_global(bool, UncommonNullCast, true); // Uncommon-trap NULLs passed to check cast
#ifdef COMPILER2
define_pd_global(intx, CodeEntryAlignment, 32);
#else
define_pd_global(intx, CodeEntryAlignment, 16);
#endif // COMPILER2
define_pd_global(intx, OptoLoopAlignment, 16);
define_pd_global(intx, InlineFrequencyCount, 100);
define_pd_global(intx, InlineSmallCode, 1000);
define_pd_global(intx, StackYellowPages, NOT_WINDOWS(2) WINDOWS_ONLY(3));
define_pd_global(intx, StackRedPages, 1);
#ifdef AMD64
define_pd_global(intx, StackShadowPages, NOT_WIN64(20) WIN64_ONLY(6) DEBUG_ONLY(+2));
#else
define_pd_global(intx, StackShadowPages, 4 DEBUG_ONLY(+5));
#endif // AMD64
define_pd_global(intx, PreInflateSpin, 10);
define_pd_global(bool, RewriteBytecodes, true);
define_pd_global(bool, RewriteFrequentPairs, true);
#ifdef _ALLBSD_SOURCE
define_pd_global(bool, UseMembar, true);
#else
define_pd_global(bool, UseMembar, false);
#endif
define_pd_global(uintx, CMSYoungGenPerWorker, 64*M); // default max size of CMS young gen, per GC worker thread
define_pd_global(uintx, TypeProfileLevel, 111);
define_pd_global(bool, PreserveFramePointer, false);
#define ARCH_FLAGS(develop, product, diagnostic, experimental, notproduct) \
\
develop(bool, IEEEPrecision, true, \
"Enables IEEE precision (for INTEL only)") \
\
product(intx, FenceInstruction, 0, \
"(Unsafe,Unstable) Experimental") \
\
product(intx, ReadPrefetchInstr, 0, \
"Prefetch instruction to prefetch ahead") \
\
product(bool, UseStoreImmI16, true, \
"Use store immediate 16-bits value instruction on x86") \
\
product(intx, UseAVX, 99, \
"Highest supported AVX instructions set on x86/x64") \
\
product(bool, UseCLMUL, false, \
"Control whether CLMUL instructions can be used on x86/x64") \
\
diagnostic(bool, UseIncDec, true, \
"Use INC, DEC instructions on x86") \
\
product(bool, UseNewLongLShift, false, \
"Use optimized bitwise shift left") \
\
product(bool, UseAddressNop, false, \
"Use '0F 1F [addr]' NOP instructions on x86 cpus") \
\
product(bool, UseXmmLoadAndClearUpper, true, \
"Load low part of XMM register and clear upper part") \
\
product(bool, UseXmmRegToRegMoveAll, false, \
"Copy all XMM register bits when moving value between registers") \
\
product(bool, UseXmmI2D, false, \
"Use SSE2 CVTDQ2PD instruction to convert Integer to Double") \
\
product(bool, UseXmmI2F, false, \
"Use SSE2 CVTDQ2PS instruction to convert Integer to Float") \
\
product(bool, UseUnalignedLoadStores, false, \
"Use SSE2 MOVDQU instruction for Arraycopy") \
\
product(bool, UseFastStosb, false, \
"Use fast-string operation for zeroing: rep stosb") \
\
product(bool, UseRTMLocking, false, \
"Enable RTM lock eliding for inflated locks in compiled code") \
\
experimental(bool, UseRTMForStackLocks, false, \
"Enable RTM lock eliding for stack locks in compiled code") \
\
product(bool, UseRTMDeopt, false, \
"Perform deopt and recompilation based on RTM abort ratio") \
\
product(uintx, RTMRetryCount, 5, \
"Number of RTM retries on lock abort or busy") \
\
experimental(intx, RTMSpinLoopCount, 100, \
"Spin count for lock to become free before RTM retry") \
\
experimental(intx, RTMAbortThreshold, 1000, \
"Calculate abort ratio after this number of aborts") \
\
experimental(intx, RTMLockingThreshold, 10000, \
"Lock count at which to do RTM lock eliding without " \
"abort ratio calculation") \
\
experimental(intx, RTMAbortRatio, 50, \
"Lock abort ratio at which to stop use RTM lock eliding") \
\
experimental(intx, RTMTotalCountIncrRate, 64, \
"Increment total RTM attempted lock count once every n times") \
\
experimental(intx, RTMLockingCalculationDelay, 0, \
"Number of milliseconds to wait before start calculating aborts " \
"for RTM locking") \
\
experimental(bool, UseRTMXendForLockBusy, true, \
"Use RTM Xend instead of Xabort when lock busy") \
\
product(bool, Use486InstrsOnly, false, \
"Use 80486 Compliant instruction subset") \
\
product(bool, UseCountLeadingZerosInstruction, false, \
"Use count leading zeros instruction") \
\
product(bool, UseCountTrailingZerosInstruction, false, \
"Use count trailing zeros instruction") \
\
product(bool, UseBMI1Instructions, false, \
"Use BMI1 instructions") \
\
product(bool, UseBMI2Instructions, false, \
"Use BMI2 instructions")
#endif // CPU_X86_VM_GLOBALS_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/icache_x86.cpp
#include "precompiled.hpp"
#include "asm/macroAssembler.hpp"
#include "runtime/icache.hpp"
#define __ _masm->
void ICacheStubGenerator::generate_icache_flush(ICache::flush_icache_stub_t* flush_icache_stub) {
StubCodeMark mark(this, "ICache", "flush_icache_stub");
address start = __ pc();
#ifdef AMD64
const Register addr = c_rarg0;
const Register lines = c_rarg1;
const Register magic = c_rarg2;
Label flush_line, done;
__ testl(lines, lines);
__ jcc(Assembler::zero, done);
__ mfence();
__ bind(flush_line);
__ clflush(Address(addr, 0));
__ addptr(addr, ICache::line_size);
__ decrementl(lines);
__ jcc(Assembler::notZero, flush_line);
__ mfence();
__ bind(done);
#else
const Address magic(rsp, 3*wordSize);
__ lock(); __ addl(Address(rsp, 0), 0);
#endif // AMD64
__ movptr(rax, magic); // Handshake with caller to make sure it happened!
__ ret(0);
}
#undef __
C:\hotspot-69087d08d473\src\cpu\x86\vm/icache_x86.hpp
#ifndef CPU_X86_VM_ICACHE_X86_HPP
#define CPU_X86_VM_ICACHE_X86_HPP
class ICache : public AbstractICache {
public:
#ifdef AMD64
enum {
stub_size = 64, // Size of the icache flush stub in bytes
line_size = 64, // Icache line size in bytes
log2_line_size = 6 // log2(line_size)
};
#else
enum {
stub_size = 16, // Size of the icache flush stub in bytes
line_size = BytesPerWord, // conservative
log2_line_size = LogBytesPerWord // log2(line_size)
};
#endif // AMD64
};
#endif // CPU_X86_VM_ICACHE_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/icBuffer_x86.cpp
#include "precompiled.hpp"
#include "asm/macroAssembler.hpp"
#include "asm/macroAssembler.inline.hpp"
#include "code/icBuffer.hpp"
#include "gc_interface/collectedHeap.inline.hpp"
#include "interpreter/bytecodes.hpp"
#include "memory/resourceArea.hpp"
#include "nativeInst_x86.hpp"
#include "oops/oop.inline.hpp"
#include "oops/oop.inline2.hpp"
int InlineCacheBuffer::ic_stub_code_size() {
return NativeMovConstReg::instruction_size +
NativeJump::instruction_size +
1;
}
void InlineCacheBuffer::assemble_ic_buffer_code(address code_begin, void* cached_value, address entry_point) {
ResourceMark rm;
CodeBuffer code(code_begin, ic_stub_code_size());
MacroAssembler* masm = new MacroAssembler(&code);
masm->lea(rax, AddressLiteral((address) cached_value, relocInfo::metadata_type));
masm->jump(ExternalAddress(entry_point));
}
address InlineCacheBuffer::ic_buffer_entry_point(address code_begin) {
NativeMovConstReg* move = nativeMovConstReg_at(code_begin); // creation also verifies the object
NativeJump* jump = nativeJump_at(move->next_instruction_address());
return jump->jump_destination();
}
void* InlineCacheBuffer::ic_buffer_cached_value(address code_begin) {
NativeMovConstReg* move = nativeMovConstReg_at(code_begin);
NativeJump* jump = nativeJump_at(move->next_instruction_address());
void* o = (void*)move->data();
return o;
}
C:\hotspot-69087d08d473\src\cpu\x86\vm/interpreterGenerator_x86.hpp
#ifndef CPU_X86_VM_INTERPRETERGENERATOR_X86_HPP
#define CPU_X86_VM_INTERPRETERGENERATOR_X86_HPP
friend class AbstractInterpreterGenerator;
private:
address generate_normal_entry(bool synchronized);
address generate_native_entry(bool synchronized);
address generate_abstract_entry(void);
address generate_math_entry(AbstractInterpreter::MethodKind kind);
address generate_empty_entry(void);
address generate_accessor_entry(void);
address generate_Reference_get_entry();
address generate_CRC32_update_entry();
address generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind);
void lock_method(void);
void generate_stack_overflow_check(void);
void generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue);
void generate_counter_overflow(Label* do_continue);
#endif // CPU_X86_VM_INTERPRETERGENERATOR_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/interpreterRT_x86.hpp
#ifndef CPU_X86_VM_INTERPRETERRT_X86_HPP
#define CPU_X86_VM_INTERPRETERRT_X86_HPP
#include "memory/allocation.hpp"
class SignatureHandlerGenerator: public NativeSignatureIterator {
private:
MacroAssembler* _masm;
#ifdef AMD64
#ifdef _WIN64
unsigned int _num_args;
#else
unsigned int _num_fp_args;
unsigned int _num_int_args;
#endif // _WIN64
int _stack_offset;
#else
void move(int from_offset, int to_offset);
void box(int from_offset, int to_offset);
#endif // AMD64
void pass_int();
void pass_long();
void pass_float();
#ifdef AMD64
void pass_double();
#endif // AMD64
void pass_object();
public:
SignatureHandlerGenerator(methodHandle method, CodeBuffer* buffer) : NativeSignatureIterator(method) {
_masm = new MacroAssembler(buffer);
#ifdef AMD64
#ifdef _WIN64
_num_args = (method->is_static() ? 1 : 0);
_stack_offset = (Argument::n_int_register_parameters_c+1)* wordSize; // don't overwrite return address
#else
_num_int_args = (method->is_static() ? 1 : 0);
_num_fp_args = 0;
_stack_offset = wordSize; // don't overwrite return address
#endif // _WIN64
#endif // AMD64
}
void generate(uint64_t fingerprint);
static Register from();
static Register to();
static Register temp();
};
#endif // CPU_X86_VM_INTERPRETERRT_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/interpreterRT_x86_32.cpp
#include "precompiled.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interpreterRuntime.hpp"
#include "memory/allocation.inline.hpp"
#include "memory/universe.inline.hpp"
#include "oops/method.hpp"
#include "oops/oop.inline.hpp"
#include "runtime/handles.inline.hpp"
#include "runtime/icache.hpp"
#include "runtime/interfaceSupport.hpp"
#include "runtime/signature.hpp"
#define __ _masm->
void InterpreterRuntime::SignatureHandlerGenerator::pass_int() {
move(offset(), jni_offset() + 1);
}
void InterpreterRuntime::SignatureHandlerGenerator::pass_float() {
move(offset(), jni_offset() + 1);
}
void InterpreterRuntime::SignatureHandlerGenerator::pass_long() {
move(offset(), jni_offset() + 2);
move(offset() + 1, jni_offset() + 1);
}
void InterpreterRuntime::SignatureHandlerGenerator::pass_object() {
box (offset(), jni_offset() + 1);
}
void InterpreterRuntime::SignatureHandlerGenerator::move(int from_offset, int to_offset) {
__ movl(temp(), Address(from(), Interpreter::local_offset_in_bytes(from_offset)));
__ movl(Address(to(), to_offset * wordSize), temp());
}
void InterpreterRuntime::SignatureHandlerGenerator::box(int from_offset, int to_offset) {
__ lea(temp(), Address(from(), Interpreter::local_offset_in_bytes(from_offset)));
__ cmpptr(Address(from(), Interpreter::local_offset_in_bytes(from_offset)), (int32_t)NULL_WORD); // do not use temp() to avoid AGI
Label L;
__ jcc(Assembler::notZero, L);
__ movptr(temp(), NULL_WORD);
__ bind(L);
__ movptr(Address(to(), to_offset * wordSize), temp());
}
void InterpreterRuntime::SignatureHandlerGenerator::generate( uint64_t fingerprint) {
iterate(fingerprint);
__ lea(rax,
ExternalAddress((address)Interpreter::result_handler(method()->result_type())));
__ ret(0);
__ flush();
}
Register InterpreterRuntime::SignatureHandlerGenerator::from() { return rdi; }
Register InterpreterRuntime::SignatureHandlerGenerator::to() { return rsp; }
Register InterpreterRuntime::SignatureHandlerGenerator::temp() { return rcx; }
void SignatureHandlerLibrary::pd_set_handler(address handler) {}
class SlowSignatureHandler: public NativeSignatureIterator {
private:
address _from;
intptr_t* _to;
virtual void pass_int() {
_from -= Interpreter::stackElementSize;
}
virtual void pass_float() {
_from -= Interpreter::stackElementSize;
}
virtual void pass_long() {
_to[0] = *(intptr_t*)(_from+Interpreter::local_offset_in_bytes(1));
_to[1] = *(intptr_t*)(_from+Interpreter::local_offset_in_bytes(0));
_to += 2;
_from -= 2*Interpreter::stackElementSize;
}
virtual void pass_object() {
intptr_t from_addr = (intptr_t)(_from + Interpreter::local_offset_in_bytes(0));
_from -= Interpreter::stackElementSize;
}
public:
SlowSignatureHandler(methodHandle method, address from, intptr_t* to) :
NativeSignatureIterator(method) {
_from = from;
_to = to + (is_static() ? 2 : 1);
}
};
IRT_ENTRY(address, InterpreterRuntime::slow_signature_handler(JavaThread* thread, Method* method, intptr_t* from, intptr_t* to))
methodHandle m(thread, (Method*)method);
assert(m->is_native(), "sanity check");
SlowSignatureHandler(m, (address)from, to + 1).iterate(UCONST64(-1));
return Interpreter::result_handler(m->result_type());
IRT_END
C:\hotspot-69087d08d473\src\cpu\x86\vm/interpreterRT_x86_64.cpp
#include "precompiled.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interpreterRuntime.hpp"
#include "memory/allocation.inline.hpp"
#include "memory/universe.inline.hpp"
#include "oops/method.hpp"
#include "oops/oop.inline.hpp"
#include "runtime/handles.inline.hpp"
#include "runtime/icache.hpp"
#include "runtime/interfaceSupport.hpp"
#include "runtime/signature.hpp"
#define __ _masm->
Register InterpreterRuntime::SignatureHandlerGenerator::from() { return r14; }
Register InterpreterRuntime::SignatureHandlerGenerator::to() { return rsp; }
Register InterpreterRuntime::SignatureHandlerGenerator::temp() { return rscratch1; }
void InterpreterRuntime::SignatureHandlerGenerator::pass_int() {
const Address src(from(), Interpreter::local_offset_in_bytes(offset()));
#ifdef _WIN64
switch (_num_args) {
case 0:
__ movl(c_rarg1, src);
_num_args++;
break;
case 1:
__ movl(c_rarg2, src);
_num_args++;
break;
case 2:
__ movl(c_rarg3, src);
_num_args++;
break;
default:
__ movl(rax, src);
__ movl(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
break;
}
#else
switch (_num_int_args) {
case 0:
__ movl(c_rarg1, src);
_num_int_args++;
break;
case 1:
__ movl(c_rarg2, src);
_num_int_args++;
break;
case 2:
__ movl(c_rarg3, src);
_num_int_args++;
break;
case 3:
__ movl(c_rarg4, src);
_num_int_args++;
break;
case 4:
__ movl(c_rarg5, src);
_num_int_args++;
break;
default:
__ movl(rax, src);
__ movl(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
break;
}
#endif
}
void InterpreterRuntime::SignatureHandlerGenerator::pass_long() {
const Address src(from(), Interpreter::local_offset_in_bytes(offset() + 1));
#ifdef _WIN64
switch (_num_args) {
case 0:
__ movptr(c_rarg1, src);
_num_args++;
break;
case 1:
__ movptr(c_rarg2, src);
_num_args++;
break;
case 2:
__ movptr(c_rarg3, src);
_num_args++;
break;
case 3:
default:
__ movptr(rax, src);
__ movptr(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
break;
}
#else
switch (_num_int_args) {
case 0:
__ movptr(c_rarg1, src);
_num_int_args++;
break;
case 1:
__ movptr(c_rarg2, src);
_num_int_args++;
break;
case 2:
__ movptr(c_rarg3, src);
_num_int_args++;
break;
case 3:
__ movptr(c_rarg4, src);
_num_int_args++;
break;
case 4:
__ movptr(c_rarg5, src);
_num_int_args++;
break;
default:
__ movptr(rax, src);
__ movptr(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
break;
}
#endif
}
void InterpreterRuntime::SignatureHandlerGenerator::pass_float() {
const Address src(from(), Interpreter::local_offset_in_bytes(offset()));
#ifdef _WIN64
if (_num_args < Argument::n_float_register_parameters_c-1) {
__ movflt(as_XMMRegister(++_num_args), src);
} else {
__ movl(rax, src);
__ movl(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
}
#else
if (_num_fp_args < Argument::n_float_register_parameters_c) {
__ movflt(as_XMMRegister(_num_fp_args++), src);
} else {
__ movl(rax, src);
__ movl(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
}
#endif
}
void InterpreterRuntime::SignatureHandlerGenerator::pass_double() {
const Address src(from(), Interpreter::local_offset_in_bytes(offset() + 1));
#ifdef _WIN64
if (_num_args < Argument::n_float_register_parameters_c-1) {
__ movdbl(as_XMMRegister(++_num_args), src);
} else {
__ movptr(rax, src);
__ movptr(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
}
#else
if (_num_fp_args < Argument::n_float_register_parameters_c) {
__ movdbl(as_XMMRegister(_num_fp_args++), src);
} else {
__ movptr(rax, src);
__ movptr(Address(to(), _stack_offset), rax);
_stack_offset += wordSize;
}
#endif
}
void InterpreterRuntime::SignatureHandlerGenerator::pass_object() {
const Address src(from(), Interpreter::local_offset_in_bytes(offset()));
#ifdef _WIN64
switch (_num_args) {
case 0:
assert(offset() == 0, "argument register 1 can only be (non-null) receiver");
__ lea(c_rarg1, src);
_num_args++;
break;
case 1:
__ lea(rax, src);
__ xorl(c_rarg2, c_rarg2);
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, c_rarg2, rax);
_num_args++;
break;
case 2:
__ lea(rax, src);
__ xorl(c_rarg3, c_rarg3);
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, c_rarg3, rax);
_num_args++;
break;
default:
__ lea(rax, src);
__ xorl(temp(), temp());
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, temp(), rax);
__ movptr(Address(to(), _stack_offset), temp());
_stack_offset += wordSize;
break;
}
#else
switch (_num_int_args) {
case 0:
assert(offset() == 0, "argument register 1 can only be (non-null) receiver");
__ lea(c_rarg1, src);
_num_int_args++;
break;
case 1:
__ lea(rax, src);
__ xorl(c_rarg2, c_rarg2);
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, c_rarg2, rax);
_num_int_args++;
break;
case 2:
__ lea(rax, src);
__ xorl(c_rarg3, c_rarg3);
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, c_rarg3, rax);
_num_int_args++;
break;
case 3:
__ lea(rax, src);
__ xorl(c_rarg4, c_rarg4);
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, c_rarg4, rax);
_num_int_args++;
break;
case 4:
__ lea(rax, src);
__ xorl(c_rarg5, c_rarg5);
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, c_rarg5, rax);
_num_int_args++;
break;
default:
__ lea(rax, src);
__ xorl(temp(), temp());
__ cmpptr(src, 0);
__ cmov(Assembler::notEqual, temp(), rax);
__ movptr(Address(to(), _stack_offset), temp());
_stack_offset += wordSize;
break;
}
#endif
}
void InterpreterRuntime::SignatureHandlerGenerator::generate(uint64_t fingerprint) {
iterate(fingerprint);
__ lea(rax, ExternalAddress(Interpreter::result_handler(method()->result_type())));
__ ret(0);
__ flush();
}
void SignatureHandlerLibrary::pd_set_handler(address handler) {}
#ifdef _WIN64
class SlowSignatureHandler
: public NativeSignatureIterator {
private:
address _from;
intptr_t* _to;
intptr_t* _reg_args;
intptr_t* _fp_identifiers;
unsigned int _num_args;
virtual void pass_int()
{
jint from_obj = *(jint *)(_from+Interpreter::local_offset_in_bytes(0));
_from -= Interpreter::stackElementSize;
if (_num_args < Argument::n_int_register_parameters_c-1) {
_num_args++;
} else {
}
}
virtual void pass_long()
{
intptr_t from_obj = *(intptr_t*)(_from+Interpreter::local_offset_in_bytes(1));
_from -= 2*Interpreter::stackElementSize;
if (_num_args < Argument::n_int_register_parameters_c-1) {
_num_args++;
} else {
}
}
virtual void pass_object()
{
intptr_t *from_addr = (intptr_t*)(_from + Interpreter::local_offset_in_bytes(0));
_from -= Interpreter::stackElementSize;
if (_num_args < Argument::n_int_register_parameters_c-1) {
_num_args++;
} else {
}
}
virtual void pass_float()
{
jint from_obj = *(jint *)(_from+Interpreter::local_offset_in_bytes(0));
_from -= Interpreter::stackElementSize;
if (_num_args < Argument::n_float_register_parameters_c-1) {
assert((_num_args*2) < BitsPerWord, "_num_args*2 is out of range");
_num_args++;
} else {
}
}
virtual void pass_double()
{
intptr_t from_obj = *(intptr_t*)(_from+Interpreter::local_offset_in_bytes(1));
_from -= 2*Interpreter::stackElementSize;
if (_num_args < Argument::n_float_register_parameters_c-1) {
assert((_num_args*2) < BitsPerWord, "_num_args*2 is out of range");
_num_args++;
} else {
}
}
public:
SlowSignatureHandler(methodHandle method, address from, intptr_t* to)
: NativeSignatureIterator(method)
{
_from = from;
_to = to;
_reg_args = to - (method->is_static() ? 4 : 5);
_fp_identifiers = to - 2;
_to = _to + 4; // Windows reserves stack space for register arguments
_num_args = (method->is_static() ? 1 : 0);
}
};
#else
class SlowSignatureHandler
: public NativeSignatureIterator {
private:
address _from;
intptr_t* _to;
intptr_t* _int_args;
intptr_t* _fp_args;
intptr_t* _fp_identifiers;
unsigned int _num_int_args;
unsigned int _num_fp_args;
virtual void pass_int()
{
jint from_obj = *(jint *)(_from+Interpreter::local_offset_in_bytes(0));
_from -= Interpreter::stackElementSize;
if (_num_int_args < Argument::n_int_register_parameters_c-1) {
_num_int_args++;
} else {
}
}
virtual void pass_long()
{
intptr_t from_obj = *(intptr_t*)(_from+Interpreter::local_offset_in_bytes(1));
_from -= 2*Interpreter::stackElementSize;
if (_num_int_args < Argument::n_int_register_parameters_c-1) {
_num_int_args++;
} else {
}
}
virtual void pass_object()
{
intptr_t *from_addr = (intptr_t*)(_from + Interpreter::local_offset_in_bytes(0));
_from -= Interpreter::stackElementSize;
if (_num_int_args < Argument::n_int_register_parameters_c-1) {
_num_int_args++;
} else {
}
}
virtual void pass_float()
{
jint from_obj = *(jint*)(_from+Interpreter::local_offset_in_bytes(0));
_from -= Interpreter::stackElementSize;
if (_num_fp_args < Argument::n_float_register_parameters_c) {
_num_fp_args++;
} else {
}
}
virtual void pass_double()
{
intptr_t from_obj = *(intptr_t*)(_from+Interpreter::local_offset_in_bytes(1));
_from -= 2*Interpreter::stackElementSize;
if (_num_fp_args < Argument::n_float_register_parameters_c) {
_num_fp_args++;
} else {
}
}
public:
SlowSignatureHandler(methodHandle method, address from, intptr_t* to)
: NativeSignatureIterator(method)
{
_from = from;
_to = to;
_int_args = to - (method->is_static() ? 14 : 15);
_fp_args = to - 9;
_fp_identifiers = to - 10;
_num_int_args = (method->is_static() ? 1 : 0);
_num_fp_args = 0;
}
};
#endif
IRT_ENTRY(address,
InterpreterRuntime::slow_signature_handler(JavaThread* thread,
Method* method,
intptr_t* from,
intptr_t* to))
methodHandle m(thread, (Method*)method);
assert(m->is_native(), "sanity check");
SlowSignatureHandler(m, (address)from, to + 1).iterate(UCONST64(-1));
return Interpreter::result_handler(m->result_type());
IRT_END
C:\hotspot-69087d08d473\src\cpu\x86\vm/interpreter_x86.hpp
#ifndef CPU_X86_VM_INTERPRETER_X86_HPP
#define CPU_X86_VM_INTERPRETER_X86_HPP
public:
static Address::ScaleFactor stackElementScale() {
return NOT_LP64(Address::times_4) LP64_ONLY(Address::times_8);
}
static int expr_offset_in_bytes(int i) { return stackElementSize * i; }
static int expr_index_at(int i) { return stackElementWords * i; }
static int local_index_at(int i) {
assert(i <= 0, "local direction already negated");
return stackElementWords * i;
}
#endif // CPU_X86_VM_INTERPRETER_X86_HPP
C:\hotspot-69087d08d473\src\cpu\x86\vm/interpreter_x86_32.cpp
#include "precompiled.hpp"
#include "asm/macroAssembler.hpp"
#include "interpreter/bytecodeHistogram.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interpreterGenerator.hpp"
#include "interpreter/interpreterRuntime.hpp"
#include "interpreter/templateTable.hpp"
#include "oops/arrayOop.hpp"
#include "oops/methodData.hpp"
#include "oops/method.hpp"
#include "oops/oop.inline.hpp"
#include "prims/jvmtiExport.hpp"
#include "prims/jvmtiThreadState.hpp"
#include "prims/methodHandles.hpp"
#include "runtime/arguments.hpp"
#include "runtime/deoptimization.hpp"
#include "runtime/frame.inline.hpp"
#include "runtime/sharedRuntime.hpp"
#include "runtime/stubRoutines.hpp"
#include "runtime/synchronizer.hpp"
#include "runtime/timer.hpp"
#include "runtime/vframeArray.hpp"
#include "utilities/debug.hpp"
#ifdef COMPILER1
#include "c1/c1_Runtime1.hpp"
#endif
#define __ _masm->
address AbstractInterpreterGenerator::generate_slow_signature_handler() {
address entry = __ pc();
__ mov(rcx, rsp);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), rbx, rdi, rcx);
__ ret(0);
return entry;
}
address InterpreterGenerator::generate_empty_entry(void) {
if (!UseFastEmptyMethods) return NULL;
address entry_point = __ pc();
Label slow_path;
ExternalAddress state(SafepointSynchronize::address_of_state());
__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
SafepointSynchronize::_not_synchronized);
__ jcc(Assembler::notEqual, slow_path);
__ pop(rax);
__ mov(rsp, rsi);
__ jmp(rax);
__ bind(slow_path);
(void) generate_normal_entry(false);
return entry_point;
}
address InterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) {
if (!InlineIntrinsics) return NULL; // Generate a vanilla entry
address entry_point = __ pc();
__ fld_d(Address(rsp, 1*wordSize));
switch (kind) {
case Interpreter::java_lang_math_sin :
__ trigfunc('s');
break;
case Interpreter::java_lang_math_cos :
__ trigfunc('c');
break;
case Interpreter::java_lang_math_tan :
__ trigfunc('t');
break;
case Interpreter::java_lang_math_sqrt:
__ fsqrt();
break;
case Interpreter::java_lang_math_abs:
__ fabs();
break;
case Interpreter::java_lang_math_log:
__ flog();
__ push_fTOS();
__ pop_fTOS();
break;
case Interpreter::java_lang_math_log10:
__ flog10();
__ push_fTOS();
__ pop_fTOS();
break;
case Interpreter::java_lang_math_pow:
__ fld_d(Address(rsp, 3*wordSize)); // second argument
__ pow_with_fallback(0);
__ push_fTOS();
__ pop_fTOS();
break;
case Interpreter::java_lang_math_exp:
__ exp_with_fallback(0);
__ push_fTOS();
__ pop_fTOS();
break;
default :
ShouldNotReachHere();
}
if (UseSSE >= 2) {
__ subptr(rsp, 2*wordSize);
__ fstp_d(Address(rsp, 0));
__ movdbl(xmm0, Address(rsp, 0));
__ addptr(rsp, 2*wordSize);
}
__ pop(rdi); // get return address
__ mov(rsp, rsi); // set sp to sender sp
__ jmp(rdi);
return entry_point;
}
address InterpreterGenerator::generate_abstract_entry(void) {
address entry_point = __ pc();
__ empty_expression_stack();
__ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
__ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
__ should_not_reach_here();
return entry_point;
}
void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) {
assert(f->is_interpreted_frame(), "must be interpreted");
}
C:\hotspot-69087d08d473\src\cpu\x86\vm/interpreter_x86_64.cpp
#include "precompiled.hpp"
#include "asm/macroAssembler.hpp"
#include "interpreter/bytecodeHistogram.hpp"
#include "interpreter/interpreter.hpp"
#include "interpreter/interpreterGenerator.hpp"
#include "interpreter/interpreterRuntime.hpp"
#include "interpreter/templateTable.hpp"
#include "oops/arrayOop.hpp"
#include "oops/methodData.hpp"
#include "oops/method.hpp"
#include "oops/oop.inline.hpp"
#include "prims/jvmtiExport.hpp"
#include "prims/jvmtiThreadState.hpp"
#include "prims/methodHandles.hpp"
#include "runtime/arguments.hpp"
#include "runtime/deoptimization.hpp"
#include "runtime/frame.inline.hpp"
#include "runtime/sharedRuntime.hpp"
#include "runtime/stubRoutines.hpp"
#include "runtime/synchronizer.hpp"
#include "runtime/timer.hpp"
#include "runtime/vframeArray.hpp"
#include "utilities/debug.hpp"
#ifdef COMPILER1
#include "c1/c1_Runtime1.hpp"
#endif
#define __ _masm->
PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
#ifdef _WIN64
address AbstractInterpreterGenerator::generate_slow_signature_handler() {
address entry = __ pc();
__ mov(c_rarg3, rsp);
__ subptr(rsp, 4 * wordSize);
__ call_VM(noreg,
CAST_FROM_FN_PTR(address,
InterpreterRuntime::slow_signature_handler),
rbx, r14, c_rarg3);
__ movl(c_rarg3, Address(rsp, 3 * wordSize)); // float/double identifiers
for ( int i= 0; i < Argument::n_int_register_parameters_c-1; i++ ) {
XMMRegister floatreg = as_XMMRegister(i+1);
Label isfloatordouble, isdouble, next;
__ testl(c_rarg3, 1 << (i*2)); // Float or Double?
__ jcc(Assembler::notZero, isfloatordouble);
switch ( i ) {
case 0:
__ movl(rscratch1, Address(rbx, Method::access_flags_offset()));
__ testl(rscratch1, JVM_ACC_STATIC);
__ cmovptr(Assembler::zero, c_rarg1, Address(rsp, 0));
break;
case 1:
__ movptr(c_rarg2, Address(rsp, wordSize));
break;
case 2:
__ movptr(c_rarg3, Address(rsp, 2 * wordSize));
break;
default:
break;
}
__ jmp (next);
__ bind(isfloatordouble);
__ testl(c_rarg3, 1 << ((i*2)+1)); // Double?
__ jcc(Assembler::notZero, isdouble);
__ movflt(floatreg, Address(rsp, i * wordSize));
__ jmp(next);
__ bind(isdouble);
__ movdbl(floatreg, Address(rsp, i * wordSize));
__ bind(next);
}
__ addptr(rsp, 4 * wordSize);
__ ret(0);
return entry;
}
#else
address AbstractInterpreterGenerator::generate_slow_signature_handler() {
address entry = __ pc();
__ mov(c_rarg3, rsp);
__ subptr(rsp, 14 * wordSize);
__ call_VM(noreg,
CAST_FROM_FN_PTR(address,
InterpreterRuntime::slow_signature_handler),
rbx, r14, c_rarg3);
__ movl(c_rarg3, Address(rsp, 5 * wordSize)); // float/double identifiers
for (int i = 0; i < Argument::n_float_register_parameters_c; i++) {
const XMMRegister r = as_XMMRegister(i);
Label d, done;
__ testl(c_rarg3, 1 << i);
__ jcc(Assembler::notZero, d);
__ movflt(r, Address(rsp, (6 + i) * wordSize));
__ jmp(done);
__ bind(d);
__ movdbl(r, Address(rsp, (6 + i) * wordSize));
__ bind(done);
}
__ movl(c_rarg3, Address(rbx, Method::access_flags_offset()));
__ testl(c_rarg3, JVM_ACC_STATIC);
__ cmovptr(Assembler::zero, c_rarg1, Address(rsp, 0));
__ movptr(c_rarg2, Address(rsp, wordSize));
__ movptr(c_rarg3, Address(rsp, 2 * wordSize));
__ movptr(c_rarg4, Address(rsp, 3 * wordSize));
__ movptr(c_rarg5, Address(rsp, 4 * wordSize));
__ addptr(rsp, 14 * wordSize);
__ ret(0);
return entry;
}
#endif
01-23
183
183
11-06
116
116
11-06
132
132
05-08
324
324
11-28
“相关推荐”对你有帮助么？
-
非常没帮助 -
没帮助 -
一般 -
有帮助 -
非常有帮助
提交
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
