void MethodHandles::verify_klass(MacroAssembler* _masm,
Register obj, SystemDictionary::WKID klass_id,
const char* error_message) {
Klass** klass_addr = SystemDictionary::well_known_klass_addr(klass_id);
KlassHandle klass = SystemDictionary::well_known_klass(klass_id);
Register temp = rdi;
Register temp2 = noreg;
LP64_ONLY(temp2 = rscratch1); // used by MacroAssembler::cmpptr
Label L_ok, L_bad;
BLOCK_COMMENT("verify_klass {");
__ verify_oop(obj);
__ testptr(obj, obj);
__ jcc(Assembler::zero, L_bad);
__ push(temp); if (temp2 != noreg) __ push(temp2);
#define UNPUSH { if (temp2 != noreg) __ pop(temp2); __ pop(temp); }
__ load_klass(temp, obj);
__ cmpptr(temp, ExternalAddress((address) klass_addr));
__ jcc(Assembler::equal, L_ok);
intptr_t super_check_offset = klass->super_check_offset();
__ movptr(temp, Address(temp, super_check_offset));
__ cmpptr(temp, ExternalAddress((address) klass_addr));
__ jcc(Assembler::equal, L_ok);
UNPUSH;
__ bind(L_bad);
__ STOP(error_message);
__ BIND(L_ok);
UNPUSH;
BLOCK_COMMENT("} verify_klass");
}
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) {
// load object
movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
biased_locking_exit(obj, hdr, done);
}
// load displaced header
movptr(hdr, Address(disp_hdr, 0));
// if the loaded hdr is NULL we had recursive locking
testptr(hdr, hdr);
// if we had recursive locking, we are done
jcc(Assembler::zero, done);
if (!UseBiasedLocking) {
// load object
movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes()));
}
verify_oop(obj);
// test if object header is pointing to the displaced header, and if so, restore
// the displaced header in the object - if the object header is not pointing to
// the displaced header, get the object header instead
if (os::is_MP()) MacroAssembler::lock(); // must be immediately before cmpxchg!
cmpxchgptr(hdr, Address(obj, hdr_offset));
// if the object header was not pointing to the displaced header,
// we do unlocking via runtime call
jcc(Assembler::notEqual, slow_case);
// done
bind(done);
}
void CompilerStubs::generate_compiler_new_object() {
comment_section("Compiler new object (any size)");
comment("Register edx holds the instance size, register ebx holds the prototypical near of the instance class");
Label slow_case;
entry("compiler_new_object");
comment("Get _inline_allocation_top");
movl(eax, Address(Constant("_inline_allocation_top")));
comment("Compute new top");
leal(ecx, Address(eax, edx, times_1));
if (GenerateDebugAssembly) {
comment("Check ExcessiveGC");
testl(Address(Constant("ExcessiveGC")), Constant(0));
jcc(not_zero, Constant(slow_case));
}
comment("Compare against _inline_allocation_end");
cmpl(ecx, Address(Constant("_inline_allocation_end")));
jcc(above, Constant(slow_case));
comment("Allocation succeeded, set _inline_allocation_top");
movl(Address(Constant("_inline_allocation_top")), ecx);
comment("Set prototypical near in object; no need for write barrier");
movl(Address(eax), ebx);
comment("Compute remaining size");
decrement(edx, oopSize);
comment("One-word object?");
Label init_done;
jcc(zero, Constant(init_done));
comment("Zero object fields");
xorl(ecx, ecx);
Label init_loop;
bind(init_loop);
movl(Address(eax, edx, times_1), ecx);
decrement(edx, oopSize);
jcc(not_zero, Constant(init_loop));
bind(init_done);
comment("The newly allocated object is in register eax");
ret();
comment("Slow case - call the VM runtime system");
bind(slow_case);
leal(eax, Address(Constant("newobject")));
goto_shared_call_vm(T_OBJECT);
entry_end(); // compiler_new_object
}
address generate_forward_exception() {
StubCodeMark mark(this, "StubRoutines", "forward exception");
address start = __ pc();
// Upon entry, the sp points to the return address returning into Java
// (interpreted or compiled) code; i.e., the return address becomes the
// throwing pc.
//
// Arguments pushed before the runtime call are still on the stack but
// the exception handler will reset the stack pointer -> ignore them.
// A potential result in registers can be ignored as well.
#ifdef ASSERT
// make sure this code is only executed if there is a pending exception
{ Label L;
__ get_thread(ecx);
__ cmpl(Address(ecx, Thread::pending_exception_offset()), (int)NULL);
__ jcc(Assembler::notEqual, L);
__ stop("StubRoutines::forward exception: no pending exception (1)");
__ bind(L);
}
#endif
// compute exception handler into ebx
__ movl(eax, Address(esp));
__ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), eax);
__ movl(ebx, eax);
// setup eax & edx, remove return address & clear pending exception
__ get_thread(ecx);
__ popl(edx);
__ movl(eax, Address(ecx, Thread::pending_exception_offset()));
__ movl(Address(ecx, Thread::pending_exception_offset()), (int)NULL);
#ifdef ASSERT
// make sure exception is set
{ Label L;
__ testl(eax, eax);
__ jcc(Assembler::notEqual, L);
__ stop("StubRoutines::forward exception: no pending exception (2)");
__ bind(L);
}
#endif
// continue at exception handler (return address removed)
// eax: exception
// ebx: exception handler
// edx: throwing pc
__ verify_oop(eax);
__ jmp(ebx);
return start;
}
/**
* Method entry for static native methods:
* int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
* int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
*/
address InterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
if (UseCRC32Intrinsics) {
address entry = __ pc();
// rbx,: Method*
// rsi: senderSP must preserved for slow path, set SP to it on fast path
// rdx: scratch
// rdi: scratch
Label slow_path;
// If we need a safepoint check, generate full interpreter entry.
ExternalAddress state(SafepointSynchronize::address_of_state());
__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
SafepointSynchronize::_not_synchronized);
__ jcc(Assembler::notEqual, slow_path);
// We don't generate local frame and don't align stack because
// we call stub code and there is no safepoint on this path.
// Load parameters
const Register crc = rax; // crc
const Register buf = rdx; // source java byte array address
const Register len = rdi; // length
// value x86_32
// interp. arg ptr ESP + 4
// int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
// 3 2 1 0
// int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
// 4 2,3 1 0
// Arguments are reversed on java expression stack
__ movl(len, Address(rsp, 4 + 0)); // Length
// Calculate address of start element
if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
__ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // long buf
__ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset
__ movl(crc, Address(rsp, 4 + 4 * wordSize)); // Initial CRC
} else {
__ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // byte[] array
__ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
__ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset
__ movl(crc, Address(rsp, 4 + 3 * wordSize)); // Initial CRC
}
__ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
// result in rax
// _areturn
__ pop(rdi); // get return address
__ mov(rsp, rsi); // set sp to sender sp
__ jmp(rdi);
// generate a vanilla native entry as the slow path
__ bind(slow_path);
__ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
return entry;
}
return NULL;
}
address InterpreterGenerator::generate_empty_entry(void) {
// rbx,: methodOop
// rcx: receiver (unused)
// rsi: previous interpreter state (C++ interpreter) must preserve
// rsi: sender sp must set sp to this value on return
if (!UseFastEmptyMethods) return NULL;
address entry_point = __ pc();
// If we need a safepoint check, generate full interpreter entry.
Label slow_path;
ExternalAddress state(SafepointSynchronize::address_of_state());
__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
SafepointSynchronize::_not_synchronized);
__ jcc(Assembler::notEqual, slow_path);
// do nothing for empty methods (do not even increment invocation counter)
// Code: _return
// _return
// return w/o popping parameters
__ pop(rax);
__ mov(rsp, rsi);
__ jmp(rax);
__ bind(slow_path);
(void) generate_normal_entry(false);
return entry_point;
}
address generate_catch_exception() {
StubCodeMark mark(this, "StubRoutines", "catch_exception");
const Address esp_after_call(ebp, -4 * wordSize); // same as in generate_call_stub()!
const Address thread (ebp, 9 * wordSize); // same as in generate_call_stub()!
address start = __ pc();
// get thread directly
__ movl(ecx, thread);
#ifdef ASSERT
// verify that threads correspond
{ Label L;
__ get_thread(ebx);
__ cmpl(ebx, ecx);
__ jcc(Assembler::equal, L);
__ stop("StubRoutines::catch_exception: threads must correspond");
__ bind(L);
}
#endif
// set pending exception
__ verify_oop(eax);
__ movl(Address(ecx, Thread::pending_exception_offset()), eax );
__ movl(Address(ecx, Thread::exception_file_offset ()), (int)__FILE__);
__ movl(Address(ecx, Thread::exception_line_offset ()), __LINE__);
// complete return to VM
assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before");
__ jmp(StubRoutines::_call_stub_return_address, relocInfo::none);
return start;
}
// preserves obj, destroys len_in_bytes
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;
// index is positive and ptr sized
subptr(index, hdr_size_in_bytes);
jcc(Assembler::zero, done);
// initialize topmost word, divide index by 2, check if odd and test if zero
// note: for the remaining code to work, index must be a multiple of BytesPerWord
#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
// index could have been not a multiple of 8 (i.e., bit 2 was set)
{ Label even;
// note: if index was a multiple of 8, than it cannot
// be 0 now otherwise it must have been 0 before
// => if it is even, we don't need to check for 0 again
jcc(Assembler::carryClear, even);
// clear topmost word (no jump needed if conditional assignment would work here)
movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - 0*BytesPerWord), t1);
// index could be 0 now, need to check again
jcc(Assembler::zero, done);
bind(even);
}
#endif // !_LP64
// initialize remaining object fields: rdx is a multiple of 2 now
{ 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);
}
void InterpreterRuntime::SignatureHandlerGenerator::box(int from_offset, int to_offset) {
__ leal(temp(), Address(from(), from_offset * wordSize));
__ cmpl(Address(from(), from_offset * wordSize), 0); // do not use temp() to avoid AGI
Label L;
__ jcc(Assembler::notZero, L);
__ movl(temp(), 0);
__ bind(L);
__ movl(Address(to(), to_offset * wordSize), temp());
}
void check_cycles_int(cpu_options *opts, uint32_t address)
{
code_info *code = &opts->code;
cmp_rr(code, opts->cycles, opts->limit, SZ_D);
code_ptr jmp_off = code->cur+1;
jcc(code, CC_A, jmp_off+1);
mov_ir(code, address, opts->scratch1, SZ_D);
call(code, opts->handle_cycle_limit_int);
*jmp_off = code->cur - (jmp_off+1);
}
void check_cycles(cpu_options * opts)
{
code_info *code = &opts->code;
cmp_rr(code, opts->cycles, opts->limit, SZ_D);
check_alloc_code(code, MAX_INST_LEN*2);
code_ptr jmp_off = code->cur+1;
jcc(code, CC_A, jmp_off+1);
call(code, opts->handle_cycle_limit);
*jmp_off = code->cur - (jmp_off+1);
}
// preserves obj, destroys len_in_bytes
void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
Label done;
// len_in_bytes is positive and ptr sized
subptr(len_in_bytes, hdr_size_in_bytes);
jcc(Assembler::zero, done);
zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
bind(done);
}
void InterpreterStubs::generate_interpreter_rethrow_exception() {
comment_section("Interpreter rethrow exception");
comment("Register eax holds the exception; Interpreter state is not in registers");
entry("interpreter_rethrow_exception");
comment("Restore bytecode and locals pointers");
movl(esi, Address(ebp, Constant(JavaFrame::bcp_store_offset())));
movl(edi, Address(ebp, Constant(JavaFrame::locals_pointer_offset())));
comment("Mark the bytecode pointer as being inside an exception");
addl(esi, Constant(JavaFrame::exception_frame_flag));
comment("Clear the expression stack");
movl(esp, Address(ebp, Constant(JavaFrame::stack_bottom_pointer_offset())));
comment("Push the exception on the expression stack");
push_obj(eax);
comment("Get exception handler bci for exception");
interpreter_call_vm(Constant("exception_handler_bci_for_exception"), T_INT);
comment("Check if we got a bci - otherwise unwind the activation");
cmpl(eax, Constant(-1));
jcc(equal, Constant("interpreter_unwind_activation"));
#if ENABLE_JAVA_DEBUGGER
Label skip;
cmpb(Address(Constant("_debugger_active")), Constant(0));
jcc(equal, Constant(skip));
movl(edx, eax);
interpreter_call_vm(Constant("handle_caught_exception"), T_VOID);
comment("Re-get exception handler bci for exception");
interpreter_call_vm(Constant("exception_handler_bci_for_exception"), T_INT);
bind(skip);
#endif
comment("Convert the bytecode index into a bytecode pointer");
movl(ecx, Address(ebp, Constant(JavaFrame::method_offset())));
leal(esi, Address(ecx, eax, times_1, Constant(Method::base_offset())));
// Dispatch to the exception handler.
dispatch_next();
entry_end(); // interpreter_rethrow_exception
}
/**
* Method entry for static native methods:
* int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len)
* int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
*/
address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) {
if (UseCRC32Intrinsics) {
address entry = __ pc();
// rbx,: Method*
// r13: senderSP must preserved for slow path, set SP to it on fast path
Label slow_path;
// If we need a safepoint check, generate full interpreter entry.
ExternalAddress state(SafepointSynchronize::address_of_state());
__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
SafepointSynchronize::_not_synchronized);
__ jcc(Assembler::notEqual, slow_path);
// We don't generate local frame and don't align stack because
// we call stub code and there is no safepoint on this path.
// Load parameters
const Register crc = c_rarg0; // crc
const Register buf = c_rarg1; // source java byte array address
const Register len = c_rarg2; // length
const Register off = len; // offset (never overlaps with 'len')
// Arguments are reversed on java expression stack
// Calculate address of start element
if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) {
__ movptr(buf, Address(rsp, 3*wordSize)); // long buf
__ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
__ addq(buf, off); // + offset
__ movl(crc, Address(rsp, 5*wordSize)); // Initial CRC
} else {
__ movptr(buf, Address(rsp, 3*wordSize)); // byte[] array
__ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size
__ movl2ptr(off, Address(rsp, 2*wordSize)); // offset
__ addq(buf, off); // + offset
__ movl(crc, Address(rsp, 4*wordSize)); // Initial CRC
}
// Can now load 'len' since we're finished with 'off'
__ movl(len, Address(rsp, wordSize)); // Length
__ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len);
// result in rax
// _areturn
__ pop(rdi); // get return address
__ mov(rsp, r13); // set sp to sender sp
__ jmp(rdi);
// generate a vanilla native entry as the slow path
__ bind(slow_path);
__ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
return entry;
}
return NULL;
}
void G1PostBarrierStub::emit_code(LIR_Assembler* ce) {
__ bind(_entry);
assert(addr()->is_register(), "Precondition.");
assert(new_val()->is_register(), "Precondition.");
Register new_val_reg = new_val()->as_register();
__ cmpptr(new_val_reg, (int32_t) NULL_WORD);
__ jcc(Assembler::equal, _continuation);
ce->store_parameter(addr()->as_pointer_register(), 0);
__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::g1_post_barrier_slow_id)));
__ jmp(_continuation);
}
void InterpreterStubs::generate_interpreter_call_vm_redo() {
#if ENABLE_JAVA_DEBUGGER
Label check_breakpoint, no_breakpoint;
#endif
comment_section("Interpreter call VM - and repeat current bytecode upon termination");
entry("interpreter_call_vm_redo");
comment("Save bytecode pointer");
movl(Address(ebp, Constant(JavaFrame::bcp_store_offset())), esi);
comment("Call the shared call vm and disregard any return value");
call_shared_call_vm(T_VOID);
comment("Restore bytecode pointer");
movl(esi, Address(ebp, Constant(JavaFrame::bcp_store_offset())));
comment("Restore locals pointer");
movl(edi, Address(ebp, Constant(JavaFrame::locals_pointer_offset())));
#if ENABLE_JAVA_DEBUGGER
comment("Check to see if we are connected to a debugger");
cmpb(Address(Constant("_debugger_active")), Constant(0));
jcc(not_zero, Constant(check_breakpoint));
bind(no_breakpoint);
comment("Not debugging, so just dispatch");
dispatch_next(0);
bind(check_breakpoint);
comment("We are debugging, so let's see if we just replaced a breakpoint opcode");
cmpb(Address(esi), Constant(Bytecodes::_breakpoint));
jcc(not_zero, Constant(no_breakpoint));
comment("There is a breakpoint in the code, so that means that eax has the correct opcode");
comment("So just jmp directly without using esi");
andl(eax, Constant(0xFF));
movl(ebx, eax);
jmp(Address(no_reg, ebx, times_4, Constant("interpreter_dispatch_table")));
#else
dispatch_next(0);
#endif
entry_end(); // interpreter_call_vm_redo
}
int main()
{
JamCoinCreator jcc(50,16);
std::cout << "Case #1:" << std::endl;
for (int i = 0; i < jcc.jamcoins.size(); i++) {
std::cout << jcc.jamcoins[i].value;
for (int j = 0; j < jcc.jamcoins[i].proofs.size(); j++) {
std::cout << " " << jcc.jamcoins[i].proofs[j];
}
std::cout << std::endl;
}
//delete[] jcc.composites;
}
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);
// Force ordering wrt cflush.
// Other fence and sync instructions won't do the job.
__ 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);
// Must be set here so StubCodeMark destructor can call the flush stub.
*flush_icache_stub = (ICache::flush_icache_stub_t)start;
}
void InterpreterMacroAssembler::narrow(Register result) {
// Get method->_constMethod->_result_type
movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
movptr(rcx, Address(rcx, Method::const_offset()));
load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
Label done, notBool, notByte, notChar;
// common case first
cmpl(rcx, T_INT);
jcc(Assembler::equal, done);
// mask integer result to narrower return type.
cmpl(rcx, T_BOOLEAN);
jcc(Assembler::notEqual, notBool);
andl(result, 0x1);
jmp(done);
bind(notBool);
cmpl(rcx, T_BYTE);
jcc(Assembler::notEqual, notByte);
LP64_ONLY(movsbl(result, result);)
void CompilerStubs::generate_compiler_idiv_irem() {
comment_section("Compiler integer divide and remainder");
comment("Register eax holds the dividend, register ebx holds the divisor");
entry("compiler_idiv_irem");
const int min_int = 0x80000000;
Label normal_case, special_case;
Label throw_exception;
testl(ebx,ebx);
jcc(equal, Constant(throw_exception));
// Check for special case
cmpl(eax, Constant(min_int));
jcc(not_equal, Constant(normal_case));
xorl(edx, edx); // Prepare edx for possible special case (where remainder = 0)
cmpl(ebx, Constant(-1));
jcc(equal, Constant(special_case));
// Handle normal case
bind(normal_case);
cdql();
idivl(ebx);
// Normal and special case exit
bind(special_case);
ret();
bind(throw_exception);
comment("Throw a DivisionByZeroException");
leal(eax, Address(Constant("division_by_zero_exception")));
goto_shared_call_vm(T_VOID);
entry_end(); // compiler_idiv_irem
}
/**
* Method entry for static native methods:
* int java.util.zip.CRC32.update(int crc, int b)
*/
address TemplateInterpreterGenerator::generate_CRC32_update_entry() {
if (UseCRC32Intrinsics) {
address entry = __ pc();
// rbx,: Method*
// r13: senderSP must preserved for slow path, set SP to it on fast path
// c_rarg0: scratch (rdi on non-Win64, rcx on Win64)
// c_rarg1: scratch (rsi on non-Win64, rdx on Win64)
Label slow_path;
// If we need a safepoint check, generate full interpreter entry.
ExternalAddress state(SafepointSynchronize::address_of_state());
__ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()),
SafepointSynchronize::_not_synchronized);
__ jcc(Assembler::notEqual, slow_path);
// We don't generate local frame and don't align stack because
// we call stub code and there is no safepoint on this path.
// Load parameters
const Register crc = rax; // crc
const Register val = c_rarg0; // source java byte value
const Register tbl = c_rarg1; // scratch
// Arguments are reversed on java expression stack
__ movl(val, Address(rsp, wordSize)); // byte value
__ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC
__ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr()));
__ notl(crc); // ~crc
__ update_byte_crc32(crc, val, tbl);
__ notl(crc); // ~crc
// result in rax
// _areturn
__ pop(rdi); // get return address
__ mov(rsp, r13); // set sp to sender sp
__ jmp(rdi);
// generate a vanilla native entry as the slow path
__ bind(slow_path);
__ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native));
return entry;
}
return NULL;
}
void G1PreBarrierStub::emit_code(LIR_Assembler* ce) {
// At this point we know that marking is in progress
__ bind(_entry);
assert(pre_val()->is_register(), "Precondition.");
Register pre_val_reg = pre_val()->as_register();
ce->mem2reg(addr(), pre_val(), T_OBJECT, patch_code(), info(), false /*wide*/, false /*unaligned*/);
__ cmpptr(pre_val_reg, (int32_t) NULL_WORD);
__ jcc(Assembler::equal, _continuation);
ce->store_parameter(pre_val()->as_register(), 0);
__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::g1_pre_barrier_slow_id)));
__ jmp(_continuation);
}
void MethodHandles::verify_ref_kind(MacroAssembler* _masm, int ref_kind, Register member_reg, Register temp) {
Label L;
BLOCK_COMMENT("verify_ref_kind {");
__ movl(temp, Address(member_reg, NONZERO(java_lang_invoke_MemberName::flags_offset_in_bytes())));
__ shrl(temp, java_lang_invoke_MemberName::MN_REFERENCE_KIND_SHIFT);
__ andl(temp, java_lang_invoke_MemberName::MN_REFERENCE_KIND_MASK);
__ cmpl(temp, ref_kind);
__ jcc(Assembler::equal, L);
{ char* buf = NEW_C_HEAP_ARRAY(char, 100, mtInternal);
jio_snprintf(buf, 100, "verify_ref_kind expected %x", ref_kind);
if (ref_kind == JVM_REF_invokeVirtual ||
ref_kind == JVM_REF_invokeSpecial)
// could do this for all ref_kinds, but would explode assembly code size
trace_method_handle(_masm, buf);
__ STOP(buf);
}
BLOCK_COMMENT("} verify_ref_kind");
__ bind(L);
}
void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
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);
if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
// clear rest of allocated space
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) {
// use explicit null stores
// code size = 2 + 3*n bytes (n = number of fields to clear)
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) {
// use loop to null out the fields
// code size = 16 bytes for even n (n = number of fields to clear)
// initialize last object field first if odd number of fields
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);
// initialize last object field if constant size is odd
if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
// initialize remaining object fields: rdx is a multiple of 2
{ 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);
}
}
void MethodHandles::jump_to_lambda_form(MacroAssembler* _masm,
Register recv, Register method_temp,
Register temp2,
bool for_compiler_entry) {
BLOCK_COMMENT("jump_to_lambda_form {");
// This is the initial entry point of a lazy method handle.
// After type checking, it picks up the invoker from the LambdaForm.
assert_different_registers(recv, method_temp, temp2);
assert(recv != noreg, "required register");
assert(method_temp == rbx, "required register for loading method");
//NOT_PRODUCT({ FlagSetting fs(TraceMethodHandles, true); trace_method_handle(_masm, "LZMH"); });
// Load the invoker, as MH -> MH.form -> LF.vmentry
__ verify_oop(recv);
__ load_heap_oop(method_temp, Address(recv, NONZERO(java_lang_invoke_MethodHandle::form_offset_in_bytes())));
__ verify_oop(method_temp);
__ load_heap_oop(method_temp, Address(method_temp, NONZERO(java_lang_invoke_LambdaForm::vmentry_offset_in_bytes())));
__ verify_oop(method_temp);
// the following assumes that a Method* is normally compressed in the vmtarget field:
__ movptr(method_temp, Address(method_temp, NONZERO(java_lang_invoke_MemberName::vmtarget_offset_in_bytes())));
if (VerifyMethodHandles && !for_compiler_entry) {
// make sure recv is already on stack
__ movptr(temp2, Address(method_temp, Method::const_offset()));
__ load_sized_value(temp2,
Address(temp2, ConstMethod::size_of_parameters_offset()),
sizeof(u2), /*is_signed*/ false);
// assert(sizeof(u2) == sizeof(Method::_size_of_parameters), "");
Label L;
__ cmpptr(recv, __ argument_address(temp2, -1));
__ jcc(Assembler::equal, L);
__ movptr(rax, __ argument_address(temp2, -1));
__ STOP("receiver not on stack");
__ BIND(L);
}
jump_from_method_handle(_masm, method_temp, temp2, for_compiler_entry);
BLOCK_COMMENT("} jump_to_lambda_form");
}
void InterpreterStubs::generate_interpreter_rethrow_exception_init() {
comment_section("Interpreter rethrow exception init");
comment("Register eax holds the exception; Interpreter state is not in registers");
entry("interpreter_rethrow_exception_init");
#if ENABLE_JAVA_DEBUGGER
Label skip;
cmpb(Address(Constant("_debugger_active")), Constant(0));
jcc(equal, Constant(skip));
comment("push the exception object so we don't nuke it");
push_obj(eax);
comment("call debugger code to store relevant info about where exception happened");
interpreter_call_vm(Constant("handle_exception_info"), T_VOID);
pop_obj(eax, edx);
bind(skip);
#endif
if (GenerateInlineAsm)
jmp(Constant("interpreter_rethrow_exception_init"));
else
comment("fall through to rethrow_exception"); // IMPL_NOTE: FALLTHROUGH
entry_end(); // interpreter_rethrow_exception_init
}
void G1PreBarrierStub::emit_code(LIR_Assembler* ce) {
// At this point we know that marking is in progress.
// If do_load() is true then we have to emit the
// load of the previous value; otherwise it has already
// been loaded into _pre_val.
__ bind(_entry);
assert(pre_val()->is_register(), "Precondition.");
Register pre_val_reg = pre_val()->as_register();
if (do_load()) {
ce->mem2reg(addr(), pre_val(), T_OBJECT, patch_code(), info(), false /*wide*/, false /*unaligned*/);
}
__ cmpptr(pre_val_reg, (int32_t) NULL_WORD);
__ jcc(Assembler::equal, _continuation);
ce->store_parameter(pre_val()->as_register(), 0);
__ call(RuntimeAddress(Runtime1::entry_for(Runtime1::g1_pre_barrier_slow_id)));
__ jmp(_continuation);
}
void MethodHandles::jump_from_method_handle(MacroAssembler* _masm, Register method, Register temp,
bool for_compiler_entry) {
assert(method == rbx, "interpreter calling convention");
Label L_no_such_method;
__ testptr(rbx, rbx);
__ jcc(Assembler::zero, L_no_such_method);
__ verify_method_ptr(method);
if (!for_compiler_entry && JvmtiExport::can_post_interpreter_events()) {
Label run_compiled_code;
// JVMTI events, such as single-stepping, are implemented partly by avoiding running
// compiled code in threads for which the event is enabled. Check here for
// interp_only_mode if these events CAN be enabled.
#ifdef _LP64
Register rthread = r15_thread;
#else
Register rthread = temp;
__ get_thread(rthread);
#endif
// interp_only is an int, on little endian it is sufficient to test the byte only
// Is a cmpl faster?
__ cmpb(Address(rthread, JavaThread::interp_only_mode_offset()), 0);
__ jccb(Assembler::zero, run_compiled_code);
__ jmp(Address(method, Method::interpreter_entry_offset()));
__ BIND(run_compiled_code);
}
const ByteSize entry_offset = for_compiler_entry ? Method::from_compiled_offset() :
Method::from_interpreted_offset();
__ jmp(Address(method, entry_offset));
__ bind(L_no_such_method);
__ jump(RuntimeAddress(StubRoutines::throw_AbstractMethodError_entry()));
}
// These stubs are used by the compiler only.
// Argument registers, which must be preserved:
// rcx - receiver (always first argument)
// rdx - second argument (if any)
// Other registers that might be usable:
// rax - inline cache register (is interface for itable stub)
// rbx - method (used when calling out to interpreter)
// Available now, but may become callee-save at some point:
// rsi, rdi
// Note that rax and rdx are also used for return values.
//
VtableStub* VtableStubs::create_vtable_stub(int vtable_index) {
const int i486_code_length = VtableStub::pd_code_size_limit(true);
VtableStub* s = new(i486_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(), i486_code_length);
MacroAssembler* masm = new MacroAssembler(&cb);
#ifndef PRODUCT
if (CountCompiledCalls) {
__ incrementl(ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
}
#endif /* PRODUCT */
// get receiver (need to skip return address on top of stack)
assert(VtableStub::receiver_location() == rcx->as_VMReg(), "receiver expected in rcx");
// get receiver klass
address npe_addr = __ pc();
__ movptr(rax, Address(rcx, oopDesc::klass_offset_in_bytes()));
#ifndef PRODUCT
if (DebugVtables) {
Label L;
// check offset vs vtable length
__ cmpl(Address(rax, InstanceKlass::vtable_length_offset()*wordSize), vtable_index*vtableEntry::size());
__ jcc(Assembler::greater, L);
__ movl(rbx, vtable_index);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), rcx, rbx);
__ bind(L);
}
#endif // PRODUCT
const Register method = rbx;
// load Method* and target address
__ lookup_virtual_method(rax, vtable_index, method);
if (DebugVtables) {
Label L;
__ cmpptr(method, (int32_t)NULL_WORD);
__ jcc(Assembler::equal, L);
__ cmpptr(Address(method, Method::from_compiled_offset()), (int32_t)NULL_WORD);
__ jcc(Assembler::notZero, L);
__ stop("Vtable entry is NULL");
__ bind(L);
}
// rax,: receiver klass
// method (rbx): Method*
// rcx: receiver
address ame_addr = __ pc();
__ jmp( Address(method, Method::from_compiled_offset()));
masm->flush();
if (PrintMiscellaneous && (WizardMode || Verbose)) {
tty->print_cr("vtable #%d at "PTR_FORMAT"[%d] left over: %d",
vtable_index, s->entry_point(),
(int)(s->code_end() - s->entry_point()),
(int)(s->code_end() - __ pc()));
}
guarantee(__ pc() <= s->code_end(), "overflowed buffer");
// shut the door on sizing bugs
int slop = 3; // 32-bit offset is this much larger than an 8-bit one
assert(vtable_index > 10 || __ pc() + slop <= s->code_end(), "room for 32-bit offset");
s->set_exception_points(npe_addr, ame_addr);
return s;
}
VtableStub* VtableStubs::create_itable_stub(int itable_index) {
// Note well: pd_code_size_limit is the absolute minimum we can get away with. If you
// add code here, bump the code stub size returned by pd_code_size_limit!
const int i486_code_length = VtableStub::pd_code_size_limit(false);
VtableStub* s = new(i486_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(), i486_code_length);
MacroAssembler* masm = new MacroAssembler(&cb);
// Entry arguments:
// rax,: Interface
// rcx: Receiver
#ifndef PRODUCT
if (CountCompiledCalls) {
__ incrementl(ExternalAddress((address) SharedRuntime::nof_megamorphic_calls_addr()));
}
#endif /* PRODUCT */
// get receiver (need to skip return address on top of stack)
assert(VtableStub::receiver_location() == rcx->as_VMReg(), "receiver expected in rcx");
// get receiver klass (also an implicit null-check)
address npe_addr = __ pc();
__ movptr(rsi, Address(rcx, oopDesc::klass_offset_in_bytes()));
// Most registers are in use; we'll use rax, rbx, rsi, rdi
// (If we need to make rsi, rdi callee-save, do a push/pop here.)
const Register method = rbx;
Label throw_icce;
// Get Method* and entrypoint for compiler
__ lookup_interface_method(// inputs: rec. class, interface, itable index
rsi, rax, itable_index,
// outputs: method, scan temp. reg
method, rdi,
throw_icce);
// method (rbx): Method*
// rcx: receiver
#ifdef ASSERT
if (DebugVtables) {
Label L1;
__ cmpptr(method, (int32_t)NULL_WORD);
__ jcc(Assembler::equal, L1);
__ cmpptr(Address(method, Method::from_compiled_offset()), (int32_t)NULL_WORD);
__ jcc(Assembler::notZero, L1);
__ stop("Method* is null");
__ bind(L1);
}
#endif // ASSERT
address ame_addr = __ pc();
__ jmp(Address(method, Method::from_compiled_offset()));
__ bind(throw_icce);
__ jump(RuntimeAddress(StubRoutines::throw_IncompatibleClassChangeError_entry()));
masm->flush();
if (PrintMiscellaneous && (WizardMode || Verbose)) {
tty->print_cr("itable #%d at "PTR_FORMAT"[%d] left over: %d",
itable_index, s->entry_point(),
(int)(s->code_end() - s->entry_point()),
(int)(s->code_end() - __ pc()));
}
guarantee(__ pc() <= s->code_end(), "overflowed buffer");
// shut the door on sizing bugs
int slop = 3; // 32-bit offset is this much larger than an 8-bit one
assert(itable_index > 10 || __ pc() + slop <= s->code_end(), "room for 32-bit offset");
s->set_exception_points(npe_addr, ame_addr);
return s;
}
