address AbstractInterpreterGenerator::generate_result_handler_for(BasicType type) {
//
// Registers alive
// R3_RET
// LR
//
// Registers updated
// R3_RET
//
Label done;
address entry = __ pc();
switch (type) {
case T_BOOLEAN:
// convert !=0 to 1
__ neg(R0, R3_RET);
__ orr(R0, R3_RET, R0);
__ srwi(R3_RET, R0, 31);
break;
case T_BYTE:
// sign extend 8 bits
__ extsb(R3_RET, R3_RET);
break;
case T_CHAR:
// zero extend 16 bits
__ clrldi(R3_RET, R3_RET, 48);
break;
case T_SHORT:
// sign extend 16 bits
__ extsh(R3_RET, R3_RET);
break;
case T_INT:
// sign extend 32 bits
__ extsw(R3_RET, R3_RET);
break;
case T_LONG:
break;
case T_OBJECT:
// unbox result if not null
__ cmpdi(CCR0, R3_RET, 0);
__ beq(CCR0, done);
__ ld(R3_RET, 0, R3_RET);
__ verify_oop(R3_RET);
break;
case T_FLOAT:
break;
case T_DOUBLE:
break;
case T_VOID:
break;
default: ShouldNotReachHere();
}
__ BIND(done);
__ blr();
return entry;
}
void C1_MacroAssembler::verify_not_null_oop(Register r) {
Label not_null;
cmpdi(CCR0, r, 0);
bne(CCR0, not_null);
stop("non-null oop required");
bind(not_null);
if (!VerifyOops) return;
verify_oop(r);
}
void C1_MacroAssembler::null_check(Register r, Label* Lnull) {
if (TrapBasedNullChecks) { // SIGTRAP based
trap_null_check(r);
} else { // explicit
//const address exception_entry = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id);
assert(Lnull != NULL, "must have Label for explicit check");
cmpdi(CCR0, r, 0);
bc_far_optimized(Assembler::bcondCRbiIs1, bi0(CCR0, Assembler::equal), *Lnull);
}
}
// Do an explicit null check if access to a+offset will not raise a SIGSEGV.
// Either issue a trap instruction that raises SIGTRAP, or do a compare that
// branches to exception_entry.
// No support for compressed oops (base page of heap). Does not distinguish
// loads and stores.
inline void MacroAssembler::null_check_throw(Register a, int offset, Register temp_reg,
address exception_entry) {
if (!ImplicitNullChecks || needs_explicit_null_check(offset) || !os::zero_page_read_protected()) {
if (TrapBasedNullChecks) {
assert(UseSIGTRAP, "sanity");
trap_null_check(a);
} else {
Label ok;
cmpdi(CCR0, a, 0);
bne(CCR0, ok);
load_const_optimized(temp_reg, exception_entry);
mtctr(temp_reg);
bctr();
bind(ok);
}
}
}
void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) {
assert_different_registers(Rmark, Roop, Rbox);
Label slow_int, done;
Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
assert(mark_addr.disp() == 0, "cas must take a zero displacement");
if (UseBiasedLocking) {
// Load the object out of the BasicObjectLock.
ld(Roop, BasicObjectLock::obj_offset_in_bytes(), Rbox);
verify_oop(Roop);
biased_locking_exit(CCR0, Roop, R0, done);
}
// Test first it it is a fast recursive unlock.
ld(Rmark, BasicLock::displaced_header_offset_in_bytes(), Rbox);
cmpdi(CCR0, Rmark, 0);
beq(CCR0, done);
if (!UseBiasedLocking) {
// Load object.
ld(Roop, BasicObjectLock::obj_offset_in_bytes(), Rbox);
verify_oop(Roop);
}
// Check if it is still a light weight lock, this is is true if we see
// the stack address of the basicLock in the markOop of the object.
cmpxchgd(/*flag=*/CCR0,
/*current_value=*/R0,
/*compare_value=*/Rbox,
/*exchange_value=*/Rmark,
/*where=*/Roop,
MacroAssembler::MemBarRel,
MacroAssembler::cmpxchgx_hint_release_lock(),
noreg,
&slow_int);
b(done);
bind(slow_int);
b(slow_case); // far
// Done
bind(done);
}
void MethodHandles::verify_klass(MacroAssembler* _masm,
Register obj_reg, SystemDictionary::WKID klass_id,
Register temp_reg, Register temp2_reg,
const char* error_message) {
Klass** klass_addr = SystemDictionary::well_known_klass_addr(klass_id);
KlassHandle klass = SystemDictionary::well_known_klass(klass_id);
Label L_ok, L_bad;
BLOCK_COMMENT("verify_klass {");
__ verify_oop(obj_reg);
__ cmpdi(CCR0, obj_reg, 0);
__ beq(CCR0, L_bad);
__ load_klass(temp_reg, obj_reg);
__ load_const_optimized(temp2_reg, (address) klass_addr);
__ ld(temp2_reg, 0, temp2_reg);
__ cmpd(CCR0, temp_reg, temp2_reg);
__ beq(CCR0, L_ok);
__ ld(temp_reg, klass->super_check_offset(), temp_reg);
__ cmpd(CCR0, temp_reg, temp2_reg);
__ beq(CCR0, L_ok);
__ BIND(L_bad);
__ stop(error_message);
__ BIND(L_ok);
BLOCK_COMMENT("} verify_klass");
}
void MethodHandles::generate_method_handle_dispatch(MacroAssembler* _masm,
vmIntrinsics::ID iid,
Register receiver_reg,
Register member_reg,
bool for_compiler_entry) {
assert(is_signature_polymorphic(iid), "expected invoke iid");
Register temp1 = (for_compiler_entry ? R25_tmp5 : R7);
Register temp2 = (for_compiler_entry ? R22_tmp2 : R8);
Register temp3 = (for_compiler_entry ? R23_tmp3 : R9);
Register temp4 = (for_compiler_entry ? R24_tmp4 : R10);
if (receiver_reg != noreg) assert_different_registers(temp1, temp2, temp3, temp4, receiver_reg);
if (member_reg != noreg) assert_different_registers(temp1, temp2, temp3, temp4, member_reg);
if (iid == vmIntrinsics::_invokeBasic) {
// indirect through MH.form.vmentry.vmtarget
jump_to_lambda_form(_masm, receiver_reg, R19_method, temp1, temp2, for_compiler_entry);
} else {
// The method is a member invoker used by direct method handles.
if (VerifyMethodHandles) {
// make sure the trailing argument really is a MemberName (caller responsibility)
verify_klass(_masm, member_reg, SystemDictionary::WK_KLASS_ENUM_NAME(MemberName_klass),
temp1, temp2,
"MemberName required for invokeVirtual etc.");
}
Register temp1_recv_klass = temp1;
if (iid != vmIntrinsics::_linkToStatic) {
__ verify_oop(receiver_reg);
if (iid == vmIntrinsics::_linkToSpecial) {
// Don't actually load the klass; just null-check the receiver.
__ null_check_throw(receiver_reg, -1, temp1, EXCEPTION_ENTRY);
} else {
// load receiver klass itself
__ null_check_throw(receiver_reg, oopDesc::klass_offset_in_bytes(), temp1, EXCEPTION_ENTRY);
__ load_klass(temp1_recv_klass, receiver_reg);
__ verify_klass_ptr(temp1_recv_klass);
}
BLOCK_COMMENT("check_receiver {");
// The receiver for the MemberName must be in receiver_reg.
// Check the receiver against the MemberName.clazz
if (VerifyMethodHandles && iid == vmIntrinsics::_linkToSpecial) {
// Did not load it above...
__ load_klass(temp1_recv_klass, receiver_reg);
__ verify_klass_ptr(temp1_recv_klass);
}
if (VerifyMethodHandles && iid != vmIntrinsics::_linkToInterface) {
Label L_ok;
Register temp2_defc = temp2;
__ load_heap_oop_not_null(temp2_defc, NONZERO(java_lang_invoke_MemberName::clazz_offset_in_bytes()), member_reg, temp3);
load_klass_from_Class(_masm, temp2_defc, temp3, temp4);
__ verify_klass_ptr(temp2_defc);
__ check_klass_subtype(temp1_recv_klass, temp2_defc, temp3, temp4, L_ok);
// If we get here, the type check failed!
__ stop("receiver class disagrees with MemberName.clazz");
__ BIND(L_ok);
}
BLOCK_COMMENT("} check_receiver");
}
if (iid == vmIntrinsics::_linkToSpecial ||
iid == vmIntrinsics::_linkToStatic) {
DEBUG_ONLY(temp1_recv_klass = noreg); // these guys didn't load the recv_klass
}
// Live registers at this point:
// member_reg - MemberName that was the trailing argument
// temp1_recv_klass - klass of stacked receiver, if needed
// O5_savedSP - interpreter linkage (if interpreted)
// O0..O5 - compiler arguments (if compiled)
Label L_incompatible_class_change_error;
switch (iid) {
case vmIntrinsics::_linkToSpecial:
if (VerifyMethodHandles) {
verify_ref_kind(_masm, JVM_REF_invokeSpecial, member_reg, temp2);
}
__ ld(R19_method, NONZERO(java_lang_invoke_MemberName::vmtarget_offset_in_bytes()), member_reg);
break;
case vmIntrinsics::_linkToStatic:
if (VerifyMethodHandles) {
verify_ref_kind(_masm, JVM_REF_invokeStatic, member_reg, temp2);
}
__ ld(R19_method, NONZERO(java_lang_invoke_MemberName::vmtarget_offset_in_bytes()), member_reg);
break;
case vmIntrinsics::_linkToVirtual:
{
// same as TemplateTable::invokevirtual,
// minus the CP setup and profiling:
if (VerifyMethodHandles) {
verify_ref_kind(_masm, JVM_REF_invokeVirtual, member_reg, temp2);
}
// pick out the vtable index from the MemberName, and then we can discard it:
Register temp2_index = temp2;
__ ld(temp2_index, NONZERO(java_lang_invoke_MemberName::vmindex_offset_in_bytes()), member_reg);
if (VerifyMethodHandles) {
Label L_index_ok;
__ cmpdi(CCR1, temp2_index, 0);
__ bge(CCR1, L_index_ok);
__ stop("no virtual index");
__ BIND(L_index_ok);
}
// Note: The verifier invariants allow us to ignore MemberName.clazz and vmtarget
// at this point. And VerifyMethodHandles has already checked clazz, if needed.
// get target Method* & entry point
__ lookup_virtual_method(temp1_recv_klass, temp2_index, R19_method);
break;
}
case vmIntrinsics::_linkToInterface:
{
// same as TemplateTable::invokeinterface
// (minus the CP setup and profiling, with different argument motion)
if (VerifyMethodHandles) {
verify_ref_kind(_masm, JVM_REF_invokeInterface, member_reg, temp2);
}
Register temp2_intf = temp2;
__ load_heap_oop_not_null(temp2_intf, NONZERO(java_lang_invoke_MemberName::clazz_offset_in_bytes()), member_reg, temp3);
load_klass_from_Class(_masm, temp2_intf, temp3, temp4);
__ verify_klass_ptr(temp2_intf);
Register vtable_index = R19_method;
__ ld(vtable_index, NONZERO(java_lang_invoke_MemberName::vmindex_offset_in_bytes()), member_reg);
if (VerifyMethodHandles) {
Label L_index_ok;
__ cmpdi(CCR1, vtable_index, 0);
__ bge(CCR1, L_index_ok);
__ stop("invalid vtable index for MH.invokeInterface");
__ BIND(L_index_ok);
}
// given intf, index, and recv klass, dispatch to the implementation method
__ lookup_interface_method(temp1_recv_klass, temp2_intf,
// note: next two args must be the same:
vtable_index, R19_method,
temp3, temp4,
L_incompatible_class_change_error);
break;
}
default:
fatal(err_msg_res("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid)));
break;
}
// Live at this point:
// R19_method
// O5_savedSP (if interpreted)
// After figuring out which concrete method to call, jump into it.
// Note that this works in the interpreter with no data motion.
// But the compiled version will require that rcx_recv be shifted out.
__ verify_method_ptr(R19_method);
jump_from_method_handle(_masm, R19_method, temp1, temp2, for_compiler_entry);
if (iid == vmIntrinsics::_linkToInterface) {
__ BIND(L_incompatible_class_change_error);
__ load_const_optimized(temp1, StubRoutines::throw_IncompatibleClassChangeError_entry());
__ mtctr(temp1);
__ bctr();
}
}
}
// Used by compiler only; may use only caller saved, non-argument
// registers.
VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
// PPC port: use fixed size.
const int code_length = VtableStub::pd_code_size_limit(true);
VtableStub* s = new (code_length) VtableStub(true, vtable_index);
// Can be NULL if there is no free space in the code cache.
if (s == NULL) {
return NULL;
}
ResourceMark rm;
CodeBuffer cb(s->entry_point(), code_length);
MacroAssembler* masm = new MacroAssembler(&cb);
#ifndef PRODUCT
if (CountCompiledCalls) {
int offs = __ load_const_optimized(R11_scratch1, SharedRuntime::nof_megamorphic_calls_addr(), R12_scratch2, true);
__ lwz(R12_scratch2, offs, R11_scratch1);
__ addi(R12_scratch2, R12_scratch2, 1);
__ stw(R12_scratch2, offs, R11_scratch1);
}
#endif
assert(VtableStub::receiver_location() == R3_ARG1->as_VMReg(), "receiver expected in R3_ARG1");
// Get receiver klass.
const Register rcvr_klass = R11_scratch1;
// We might implicit NULL fault here.
address npe_addr = __ pc(); // npe = null pointer exception
__ load_klass_with_trap_null_check(rcvr_klass, R3);
// Set method (in case of interpreted method), and destination address.
int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
#ifndef PRODUCT
if (DebugVtables) {
Label L;
// Check offset vs vtable length.
const Register vtable_len = R12_scratch2;
__ lwz(vtable_len, InstanceKlass::vtable_length_offset()*wordSize, rcvr_klass);
__ cmpwi(CCR0, vtable_len, vtable_index*vtableEntry::size());
__ bge(CCR0, L);
__ li(R12_scratch2, vtable_index);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), R3_ARG1, R12_scratch2, false);
__ bind(L);
}
#endif
int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
__ ld(R19_method, v_off, rcvr_klass);
#ifndef PRODUCT
if (DebugVtables) {
Label L;
__ cmpdi(CCR0, R19_method, 0);
__ bne(CCR0, L);
__ stop("Vtable entry is ZERO", 102);
__ bind(L);
}
#endif
// If the vtable entry is null, the method is abstract.
address ame_addr = __ pc(); // ame = abstract method error
__ load_with_trap_null_check(R12_scratch2, in_bytes(Method::from_compiled_offset()), R19_method);
__ mtctr(R12_scratch2);
__ bctr();
masm->flush();
guarantee(__ pc() <= s->code_end(), "overflowed buffer");
s->set_exception_points(npe_addr, ame_addr);
return s;
}
VtableStub* VtableStubs::create_itable_stub(int itable_index) {
// PPC port: use fixed size.
const int code_length = VtableStub::pd_code_size_limit(false);
VtableStub* s = new (code_length) VtableStub(false, itable_index);
// Can be NULL if there is no free space in the code cache.
if (s == NULL) {
return NULL;
}
ResourceMark rm;
CodeBuffer cb(s->entry_point(), code_length);
MacroAssembler* masm = new MacroAssembler(&cb);
address start_pc;
#ifndef PRODUCT
if (CountCompiledCalls) {
int offs = __ load_const_optimized(R11_scratch1, SharedRuntime::nof_megamorphic_calls_addr(), R12_scratch2, true);
__ lwz(R12_scratch2, offs, R11_scratch1);
__ addi(R12_scratch2, R12_scratch2, 1);
__ stw(R12_scratch2, offs, R11_scratch1);
}
#endif
assert(VtableStub::receiver_location() == R3_ARG1->as_VMReg(), "receiver expected in R3_ARG1");
// Entry arguments:
// R19_method: Interface
// R3_ARG1: Receiver
Label L_no_such_interface;
const Register rcvr_klass = R11_scratch1,
interface = R12_scratch2,
tmp1 = R21_tmp1,
tmp2 = R22_tmp2;
address npe_addr = __ pc(); // npe = null pointer exception
__ load_klass_with_trap_null_check(rcvr_klass, R3_ARG1);
// Receiver subtype check against REFC.
__ ld(interface, CompiledICHolder::holder_klass_offset(), R19_method);
__ lookup_interface_method(rcvr_klass, interface, noreg,
R0, tmp1, tmp2,
L_no_such_interface, /*return_method=*/ false);
// Get Method* and entrypoint for compiler
__ ld(interface, CompiledICHolder::holder_metadata_offset(), R19_method);
__ lookup_interface_method(rcvr_klass, interface, itable_index,
R19_method, tmp1, tmp2,
L_no_such_interface, /*return_method=*/ true);
#ifndef PRODUCT
if (DebugVtables) {
Label ok;
__ cmpd(CCR0, R19_method, 0);
__ bne(CCR0, ok);
__ stop("method is null", 103);
__ bind(ok);
}
#endif
// If the vtable entry is null, the method is abstract.
address ame_addr = __ pc(); // ame = abstract method error
// Must do an explicit check if implicit checks are disabled.
assert(!MacroAssembler::needs_explicit_null_check(in_bytes(Method::from_compiled_offset())), "sanity");
if (!ImplicitNullChecks || !os::zero_page_read_protected()) {
if (TrapBasedNullChecks) {
__ trap_null_check(R19_method);
} else {
__ cmpdi(CCR0, R19_method, 0);
__ beq(CCR0, L_no_such_interface);
}
}
__ ld(R12_scratch2, in_bytes(Method::from_compiled_offset()), R19_method);
__ mtctr(R12_scratch2);
__ bctr();
// Handle IncompatibleClassChangeError in itable stubs.
// More detailed error message.
// We force resolving of the call site by jumping to the "handle
// wrong method" stub, and so let the interpreter runtime do all the
// dirty work.
__ bind(L_no_such_interface);
__ load_const_optimized(R11_scratch1, SharedRuntime::get_handle_wrong_method_stub(), R12_scratch2);
__ mtctr(R11_scratch1);
__ bctr();
masm->flush();
guarantee(__ pc() <= s->code_end(), "overflowed buffer");
s->set_exception_points(npe_addr, ame_addr);
return s;
}
address AbstractInterpreterGenerator::generate_slow_signature_handler() {
// Slow_signature handler that respects the PPC C calling conventions.
//
// We get called by the native entry code with our output register
// area == 8. First we call InterpreterRuntime::get_result_handler
// to copy the pointer to the signature string temporarily to the
// first C-argument and to return the result_handler in
// R3_RET. Since native_entry will copy the jni-pointer to the
// first C-argument slot later on, it is OK to occupy this slot
// temporarilly. Then we copy the argument list on the java
// expression stack into native varargs format on the native stack
// and load arguments into argument registers. Integer arguments in
// the varargs vector will be sign-extended to 8 bytes.
//
// On entry:
// R3_ARG1 - intptr_t* Address of java argument list in memory.
// R15_prev_state - BytecodeInterpreter* Address of interpreter state for
// this method
// R19_method
//
// On exit (just before return instruction):
// R3_RET - contains the address of the result_handler.
// R4_ARG2 - is not updated for static methods and contains "this" otherwise.
// R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double,
// ARGi contains this argument. Otherwise, ARGi is not updated.
// F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double.
const int LogSizeOfTwoInstructions = 3;
// FIXME: use Argument:: GL: Argument names different numbers!
const int max_fp_register_arguments = 13;
const int max_int_register_arguments = 6; // first 2 are reserved
const Register arg_java = R21_tmp1;
const Register arg_c = R22_tmp2;
const Register signature = R23_tmp3; // is string
const Register sig_byte = R24_tmp4;
const Register fpcnt = R25_tmp5;
const Register argcnt = R26_tmp6;
const Register intSlot = R27_tmp7;
const Register target_sp = R28_tmp8;
const FloatRegister floatSlot = F0;
address entry = __ function_entry();
__ save_LR_CR(R0);
__ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
// We use target_sp for storing arguments in the C frame.
__ mr(target_sp, R1_SP);
__ push_frame_reg_args_nonvolatiles(0, R11_scratch1);
__ mr(arg_java, R3_ARG1);
__ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method);
// Signature is in R3_RET. Signature is callee saved.
__ mr(signature, R3_RET);
// Get the result handler.
__ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method);
{
Label L;
// test if static
// _access_flags._flags must be at offset 0.
// TODO PPC port: requires change in shared code.
//assert(in_bytes(AccessFlags::flags_offset()) == 0,
// "MethodDesc._access_flags == MethodDesc._access_flags._flags");
// _access_flags must be a 32 bit value.
assert(sizeof(AccessFlags) == 4, "wrong size");
__ lwa(R11_scratch1/*access_flags*/, method_(access_flags));
// testbit with condition register.
__ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT);
__ btrue(CCR0, L);
// For non-static functions, pass "this" in R4_ARG2 and copy it
// to 2nd C-arg slot.
// We need to box the Java object here, so we use arg_java
// (address of current Java stack slot) as argument and don't
// dereference it as in case of ints, floats, etc.
__ mr(R4_ARG2, arg_java);
__ addi(arg_java, arg_java, -BytesPerWord);
__ std(R4_ARG2, _abi(carg_2), target_sp);
__ bind(L);
}
// Will be incremented directly after loop_start. argcnt=0
// corresponds to 3rd C argument.
__ li(argcnt, -1);
// arg_c points to 3rd C argument
__ addi(arg_c, target_sp, _abi(carg_3));
// no floating-point args parsed so far
__ li(fpcnt, 0);
Label move_intSlot_to_ARG, move_floatSlot_to_FARG;
Label loop_start, loop_end;
Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed;
// signature points to '(' at entry
#ifdef ASSERT
__ lbz(sig_byte, 0, signature);
__ cmplwi(CCR0, sig_byte, '(');
__ bne(CCR0, do_dontreachhere);
#endif
__ bind(loop_start);
__ addi(argcnt, argcnt, 1);
__ lbzu(sig_byte, 1, signature);
__ cmplwi(CCR0, sig_byte, ')'); // end of signature
__ beq(CCR0, loop_end);
__ cmplwi(CCR0, sig_byte, 'B'); // byte
__ beq(CCR0, do_int);
__ cmplwi(CCR0, sig_byte, 'C'); // char
__ beq(CCR0, do_int);
__ cmplwi(CCR0, sig_byte, 'D'); // double
__ beq(CCR0, do_double);
__ cmplwi(CCR0, sig_byte, 'F'); // float
__ beq(CCR0, do_float);
__ cmplwi(CCR0, sig_byte, 'I'); // int
__ beq(CCR0, do_int);
__ cmplwi(CCR0, sig_byte, 'J'); // long
__ beq(CCR0, do_long);
__ cmplwi(CCR0, sig_byte, 'S'); // short
__ beq(CCR0, do_int);
__ cmplwi(CCR0, sig_byte, 'Z'); // boolean
__ beq(CCR0, do_int);
__ cmplwi(CCR0, sig_byte, 'L'); // object
__ beq(CCR0, do_object);
__ cmplwi(CCR0, sig_byte, '['); // array
__ beq(CCR0, do_array);
// __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type
// __ beq(CCR0, do_void);
__ bind(do_dontreachhere);
__ unimplemented("ShouldNotReachHere in slow_signature_handler", 120);
__ bind(do_array);
{
Label start_skip, end_skip;
__ bind(start_skip);
__ lbzu(sig_byte, 1, signature);
__ cmplwi(CCR0, sig_byte, '[');
__ beq(CCR0, start_skip); // skip further brackets
__ cmplwi(CCR0, sig_byte, '9');
__ bgt(CCR0, end_skip); // no optional size
__ cmplwi(CCR0, sig_byte, '0');
__ bge(CCR0, start_skip); // skip optional size
__ bind(end_skip);
__ cmplwi(CCR0, sig_byte, 'L');
__ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped
__ b(do_boxed); // otherwise, go directly to do_boxed
}
__ bind(do_object);
{
Label L;
__ bind(L);
__ lbzu(sig_byte, 1, signature);
__ cmplwi(CCR0, sig_byte, ';');
__ bne(CCR0, L);
}
// Need to box the Java object here, so we use arg_java (address of
// current Java stack slot) as argument and don't dereference it as
// in case of ints, floats, etc.
Label do_null;
__ bind(do_boxed);
__ ld(R0,0, arg_java);
__ cmpdi(CCR0, R0, 0);
__ li(intSlot,0);
__ beq(CCR0, do_null);
__ mr(intSlot, arg_java);
__ bind(do_null);
__ std(intSlot, 0, arg_c);
__ addi(arg_java, arg_java, -BytesPerWord);
__ addi(arg_c, arg_c, BytesPerWord);
__ cmplwi(CCR0, argcnt, max_int_register_arguments);
__ blt(CCR0, move_intSlot_to_ARG);
__ b(loop_start);
__ bind(do_int);
__ lwa(intSlot, 0, arg_java);
__ std(intSlot, 0, arg_c);
__ addi(arg_java, arg_java, -BytesPerWord);
__ addi(arg_c, arg_c, BytesPerWord);
__ cmplwi(CCR0, argcnt, max_int_register_arguments);
__ blt(CCR0, move_intSlot_to_ARG);
__ b(loop_start);
__ bind(do_long);
__ ld(intSlot, -BytesPerWord, arg_java);
__ std(intSlot, 0, arg_c);
__ addi(arg_java, arg_java, - 2 * BytesPerWord);
__ addi(arg_c, arg_c, BytesPerWord);
__ cmplwi(CCR0, argcnt, max_int_register_arguments);
__ blt(CCR0, move_intSlot_to_ARG);
__ b(loop_start);
__ bind(do_float);
__ lfs(floatSlot, 0, arg_java);
#if defined(LINUX)
// Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float
// in the least significant word of an argument slot.
#if defined(VM_LITTLE_ENDIAN)
__ stfs(floatSlot, 0, arg_c);
#else
__ stfs(floatSlot, 4, arg_c);
#endif
#elif defined(AIX)
// Although AIX runs on big endian CPU, float is in most significant
// word of an argument slot.
__ stfs(floatSlot, 0, arg_c);
#else
#error "unknown OS"
#endif
__ addi(arg_java, arg_java, -BytesPerWord);
__ addi(arg_c, arg_c, BytesPerWord);
__ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
__ blt(CCR0, move_floatSlot_to_FARG);
__ b(loop_start);
__ bind(do_double);
__ lfd(floatSlot, - BytesPerWord, arg_java);
__ stfd(floatSlot, 0, arg_c);
__ addi(arg_java, arg_java, - 2 * BytesPerWord);
__ addi(arg_c, arg_c, BytesPerWord);
__ cmplwi(CCR0, fpcnt, max_fp_register_arguments);
__ blt(CCR0, move_floatSlot_to_FARG);
__ b(loop_start);
__ bind(loop_end);
__ pop_frame();
__ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14));
__ restore_LR_CR(R0);
__ blr();
Label move_int_arg, move_float_arg;
__ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
__ mr(R5_ARG3, intSlot); __ b(loop_start);
__ mr(R6_ARG4, intSlot); __ b(loop_start);
__ mr(R7_ARG5, intSlot); __ b(loop_start);
__ mr(R8_ARG6, intSlot); __ b(loop_start);
__ mr(R9_ARG7, intSlot); __ b(loop_start);
__ mr(R10_ARG8, intSlot); __ b(loop_start);
__ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions)
__ fmr(F1_ARG1, floatSlot); __ b(loop_start);
__ fmr(F2_ARG2, floatSlot); __ b(loop_start);
__ fmr(F3_ARG3, floatSlot); __ b(loop_start);
__ fmr(F4_ARG4, floatSlot); __ b(loop_start);
__ fmr(F5_ARG5, floatSlot); __ b(loop_start);
__ fmr(F6_ARG6, floatSlot); __ b(loop_start);
__ fmr(F7_ARG7, floatSlot); __ b(loop_start);
__ fmr(F8_ARG8, floatSlot); __ b(loop_start);
__ fmr(F9_ARG9, floatSlot); __ b(loop_start);
__ fmr(F10_ARG10, floatSlot); __ b(loop_start);
__ fmr(F11_ARG11, floatSlot); __ b(loop_start);
__ fmr(F12_ARG12, floatSlot); __ b(loop_start);
__ fmr(F13_ARG13, floatSlot); __ b(loop_start);
__ bind(move_intSlot_to_ARG);
__ sldi(R0, argcnt, LogSizeOfTwoInstructions);
__ load_const(R11_scratch1, move_int_arg); // Label must be bound here.
__ add(R11_scratch1, R0, R11_scratch1);
__ mtctr(R11_scratch1/*branch_target*/);
__ bctr();
__ bind(move_floatSlot_to_FARG);
__ sldi(R0, fpcnt, LogSizeOfTwoInstructions);
__ addi(fpcnt, fpcnt, 1);
__ load_const(R11_scratch1, move_float_arg); // Label must be bound here.
__ add(R11_scratch1, R0, R11_scratch1);
__ mtctr(R11_scratch1/*branch_target*/);
__ bctr();
return entry;
}
// Interpreter intrinsic for WeakReference.get().
// 1. Don't push a full blown frame and go on dispatching, but fetch the value
// into R8 and return quickly
// 2. If G1 is active we *must* execute this intrinsic for corrrectness:
// It contains a GC barrier which puts the reference into the satb buffer
// to indicate that someone holds a strong reference to the object the
// weak ref points to!
address InterpreterGenerator::generate_Reference_get_entry(void) {
// Code: _aload_0, _getfield, _areturn
// parameter size = 1
//
// The code that gets generated by this routine is split into 2 parts:
// 1. the "intrinsified" code for G1 (or any SATB based GC),
// 2. the slow path - which is an expansion of the regular method entry.
//
// Notes:
// * In the G1 code we do not check whether we need to block for
// a safepoint. If G1 is enabled then we must execute the specialized
// code for Reference.get (except when the Reference object is null)
// so that we can log the value in the referent field with an SATB
// update buffer.
// If the code for the getfield template is modified so that the
// G1 pre-barrier code is executed when the current method is
// Reference.get() then going through the normal method entry
// will be fine.
// * The G1 code can, however, check the receiver object (the instance
// of java.lang.Reference) and jump to the slow path if null. If the
// Reference object is null then we obviously cannot fetch the referent
// and so we don't need to call the G1 pre-barrier. Thus we can use the
// regular method entry code to generate the NPE.
//
// This code is based on generate_accessor_enty.
address entry = __ pc();
const int referent_offset = java_lang_ref_Reference::referent_offset;
guarantee(referent_offset > 0, "referent offset not initialized");
if (UseG1GC) {
Label slow_path;
// Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH);
// In the G1 code we don't check if we need to reach a safepoint. We
// continue and the thread will safepoint at the next bytecode dispatch.
// If the receiver is null then it is OK to jump to the slow path.
__ ld(R3_RET, Interpreter::stackElementSize, CC_INTERP_ONLY(R17_tos) NOT_CC_INTERP(R15_esp)); // get receiver
// Check if receiver == NULL and go the slow path.
__ cmpdi(CCR0, R3_RET, 0);
__ beq(CCR0, slow_path);
// Load the value of the referent field.
__ load_heap_oop(R3_RET, referent_offset, R3_RET);
// Generate the G1 pre-barrier code to log the value of
// the referent field in an SATB buffer. Note with
// these parameters the pre-barrier does not generate
// the load of the previous value.
// Restore caller sp for c2i case.
#ifdef ASSERT
__ ld(R9_ARG7, 0, R1_SP);
__ ld(R10_ARG8, 0, R21_sender_SP);
__ cmpd(CCR0, R9_ARG7, R10_ARG8);
__ asm_assert_eq("backlink", 0x544);
#endif // ASSERT
__ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
__ g1_write_barrier_pre(noreg, // obj
noreg, // offset
R3_RET, // pre_val
R11_scratch1, // tmp
R12_scratch2, // tmp
true); // needs_frame
__ blr();
// Generate regular method entry.
__ bind(slow_path);
__ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1);
__ flush();
return entry;
} else {
return generate_accessor_entry();
}
}
// Call an accessor method (assuming it is resolved, otherwise drop into
// vanilla (slow path) entry.
address InterpreterGenerator::generate_accessor_entry(void) {
if (!UseFastAccessorMethods && (!FLAG_IS_ERGO(UseFastAccessorMethods))) {
return NULL;
}
Label Lslow_path, Lacquire;
const Register
Rclass_or_obj = R3_ARG1,
Rconst_method = R4_ARG2,
Rcodes = Rconst_method,
Rcpool_cache = R5_ARG3,
Rscratch = R11_scratch1,
Rjvmti_mode = Rscratch,
Roffset = R12_scratch2,
Rflags = R6_ARG4,
Rbtable = R7_ARG5;
static address branch_table[number_of_states];
address entry = __ pc();
// Check for safepoint:
// Ditch this, real man don't need safepoint checks.
// Also check for JVMTI mode
// Check for null obj, take slow path if so.
__ ld(Rclass_or_obj, Interpreter::stackElementSize, CC_INTERP_ONLY(R17_tos) NOT_CC_INTERP(R15_esp));
__ lwz(Rjvmti_mode, thread_(interp_only_mode));
__ cmpdi(CCR1, Rclass_or_obj, 0);
__ cmpwi(CCR0, Rjvmti_mode, 0);
__ crorc(/*CCR0 eq*/2, /*CCR1 eq*/4+2, /*CCR0 eq*/2);
__ beq(CCR0, Lslow_path); // this==null or jvmti_mode!=0
// Do 2 things in parallel:
// 1. Load the index out of the first instruction word, which looks like this:
// <0x2a><0xb4><index (2 byte, native endianess)>.
// 2. Load constant pool cache base.
__ ld(Rconst_method, in_bytes(Method::const_offset()), R19_method);
__ ld(Rcpool_cache, in_bytes(ConstMethod::constants_offset()), Rconst_method);
__ lhz(Rcodes, in_bytes(ConstMethod::codes_offset()) + 2, Rconst_method); // Lower half of 32 bit field.
__ ld(Rcpool_cache, ConstantPool::cache_offset_in_bytes(), Rcpool_cache);
// Get the const pool entry by means of <index>.
const int codes_shift = exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord);
__ slwi(Rscratch, Rcodes, codes_shift); // (codes&0xFFFF)<<codes_shift
__ add(Rcpool_cache, Rscratch, Rcpool_cache);
// Check if cpool cache entry is resolved.
// We are resolved if the indices offset contains the current bytecode.
ByteSize cp_base_offset = ConstantPoolCache::base_offset();
// Big Endian:
__ lbz(Rscratch, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::indices_offset()) + 7 - 2, Rcpool_cache);
__ cmpwi(CCR0, Rscratch, Bytecodes::_getfield);
__ bne(CCR0, Lslow_path);
__ isync(); // Order succeeding loads wrt. load of _indices field from cpool_cache.
// Finally, start loading the value: Get cp cache entry into regs.
__ ld(Rflags, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::flags_offset()), Rcpool_cache);
__ ld(Roffset, in_bytes(cp_base_offset) + in_bytes(ConstantPoolCacheEntry::f2_offset()), Rcpool_cache);
// Following code is from templateTable::getfield_or_static
// Load pointer to branch table
__ load_const_optimized(Rbtable, (address)branch_table, Rscratch);
// Get volatile flag
__ rldicl(Rscratch, Rflags, 64-ConstantPoolCacheEntry::is_volatile_shift, 63); // extract volatile bit
// note: sync is needed before volatile load on PPC64
// Check field type
__ rldicl(Rflags, Rflags, 64-ConstantPoolCacheEntry::tos_state_shift, 64-ConstantPoolCacheEntry::tos_state_bits);
#ifdef ASSERT
Label LFlagInvalid;
__ cmpldi(CCR0, Rflags, number_of_states);
__ bge(CCR0, LFlagInvalid);
__ ld(R9_ARG7, 0, R1_SP);
__ ld(R10_ARG8, 0, R21_sender_SP);
__ cmpd(CCR0, R9_ARG7, R10_ARG8);
__ asm_assert_eq("backlink", 0x543);
#endif // ASSERT
__ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started.
// Load from branch table and dispatch (volatile case: one instruction ahead)
__ sldi(Rflags, Rflags, LogBytesPerWord);
__ cmpwi(CCR6, Rscratch, 1); // volatile?
if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
__ sldi(Rscratch, Rscratch, exact_log2(BytesPerInstWord)); // volatile ? size of 1 instruction : 0
}
__ ldx(Rbtable, Rbtable, Rflags);
if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
__ subf(Rbtable, Rscratch, Rbtable); // point to volatile/non-volatile entry point
}
__ mtctr(Rbtable);
__ bctr();
#ifdef ASSERT
__ bind(LFlagInvalid);
__ stop("got invalid flag", 0x6541);
bool all_uninitialized = true,
all_initialized = true;
for (int i = 0; i<number_of_states; ++i) {
all_uninitialized = all_uninitialized && (branch_table[i] == NULL);
all_initialized = all_initialized && (branch_table[i] != NULL);
}
assert(all_uninitialized != all_initialized, "consistency"); // either or
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
if (branch_table[vtos] == 0) branch_table[vtos] = __ pc(); // non-volatile_entry point
if (branch_table[dtos] == 0) branch_table[dtos] = __ pc(); // non-volatile_entry point
if (branch_table[ftos] == 0) branch_table[ftos] = __ pc(); // non-volatile_entry point
__ stop("unexpected type", 0x6551);
#endif
if (branch_table[itos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[itos] = __ pc(); // non-volatile_entry point
__ lwax(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[ltos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[ltos] = __ pc(); // non-volatile_entry point
__ ldx(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[btos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[btos] = __ pc(); // non-volatile_entry point
__ lbzx(R3_RET, Rclass_or_obj, Roffset);
__ extsb(R3_RET, R3_RET);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[ctos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[ctos] = __ pc(); // non-volatile_entry point
__ lhzx(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[stos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[stos] = __ pc(); // non-volatile_entry point
__ lhax(R3_RET, Rclass_or_obj, Roffset);
__ beq(CCR6, Lacquire);
__ blr();
}
if (branch_table[atos] == 0) { // generate only once
__ align(32, 28, 28); // align load
__ fence(); // volatile entry point (one instruction before non-volatile_entry point)
branch_table[atos] = __ pc(); // non-volatile_entry point
__ load_heap_oop(R3_RET, (RegisterOrConstant)Roffset, Rclass_or_obj);
__ verify_oop(R3_RET);
//__ dcbt(R3_RET); // prefetch
__ beq(CCR6, Lacquire);
__ blr();
}
__ align(32, 12);
__ bind(Lacquire);
__ twi_0(R3_RET);
__ isync(); // acquire
__ blr();
#ifdef ASSERT
for (int i = 0; i<number_of_states; ++i) {
assert(branch_table[i], "accessor_entry initialization");
//tty->print_cr("accessor_entry: branch_table[%d] = 0x%llx (opcode 0x%llx)", i, branch_table[i], *((unsigned int*)branch_table[i]));
}
#endif
__ bind(Lslow_path);
__ branch_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), Rscratch);
__ flush();
return entry;
}
void C1_MacroAssembler::initialize_body(Register obj, Register tmp1, Register tmp2,
int obj_size_in_bytes, int hdr_size_in_bytes) {
const int index = (obj_size_in_bytes - hdr_size_in_bytes) / HeapWordSize;
const int cl_size = VM_Version::L1_data_cache_line_size(),
cl_dwords = cl_size>>3,
cl_dw_addr_bits = exact_log2(cl_dwords);
const Register tmp = R0,
base_ptr = tmp1,
cnt_dwords = tmp2;
if (index <= 6) {
// Use explicit NULL stores.
if (index > 0) { li(tmp, 0); }
for (int i = 0; i < index; ++i) { std(tmp, hdr_size_in_bytes + i * HeapWordSize, obj); }
} else if (index < (2<<cl_dw_addr_bits)-1) {
// simple loop
Label loop;
li(cnt_dwords, index);
addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
li(tmp, 0);
mtctr(cnt_dwords); // Load counter.
bind(loop);
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(loop);
} else {
// like clear_memory_doubleword
Label startloop, fast, fastloop, restloop, done;
addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
load_const_optimized(cnt_dwords, index);
rldicl_(tmp, base_ptr, 64-3, 64-cl_dw_addr_bits); // Extract dword offset within first cache line.
beq(CCR0, fast); // Already 128byte aligned.
subfic(tmp, tmp, cl_dwords);
mtctr(tmp); // Set ctr to hit 128byte boundary (0<ctr<cl_dwords).
subf(cnt_dwords, tmp, cnt_dwords); // rest.
li(tmp, 0);
bind(startloop); // Clear at the beginning to reach 128byte boundary.
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(startloop);
bind(fast); // Clear 128byte blocks.
srdi(tmp, cnt_dwords, cl_dw_addr_bits); // Loop count for 128byte loop (>0).
andi(cnt_dwords, cnt_dwords, cl_dwords-1); // Rest in dwords.
mtctr(tmp); // Load counter.
bind(fastloop);
dcbz(base_ptr); // Clear 128byte aligned block.
addi(base_ptr, base_ptr, cl_size);
bdnz(fastloop);
cmpdi(CCR0, cnt_dwords, 0); // size 0?
beq(CCR0, done); // rest == 0
li(tmp, 0);
mtctr(cnt_dwords); // Load counter.
bind(restloop); // Clear rest.
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(restloop);
bind(done);
}
}
