int main ()
{
testc (0, 0);
testc (1, 1);
testc (SCHAR_MAX, 1);
testc (SCHAR_MAX+1, 0);
testc (UCHAR_MAX, 0);
tests (0, 0);
tests (1, 1);
tests (SHRT_MAX, 1);
tests (SHRT_MAX+1, 0);
tests (USHRT_MAX, 0);
testi (0, 0);
testi (1, 1);
testi (INT_MAX, 1);
testi (INT_MAX+1U, 0);
testi (UINT_MAX, 0);
testl (0, 0);
testl (1, 1);
testl (LONG_MAX, 1);
testl (LONG_MAX+1UL, 0);
testl (ULONG_MAX, 0);
return 0;
}
int
main(int argc, char *argv[])
{
printf("1..3\n");
test_invalid(0.0, 0.0);
test_invalid(1.0, 0.0);
test_invalid(INFINITY, 0.0);
test_invalid(INFINITY, 1.0);
test_invalid(-INFINITY, 1.0);
test_invalid(NAN, 1.0);
test_invalid(1.0, NAN);
test(4, 4, 0, 1);
test(0, 3.0, 0, 0);
testd(0x1p-1074, 1, 0x1p-1074, 0);
testf(0x1p-149, 1, 0x1p-149, 0);
test(3.0, 4, -1, 1);
test(3.0, -4, -1, -1);
testd(275 * 1193040, 275, 0, 1193040);
test(4.5 * 7.5, 4.5, -2.25, 8); /* we should get the even one */
testf(0x1.9044f6p-1, 0x1.ce662ep-1, -0x1.f109cp-4, 1);
#if LDBL_MANT_DIG > 53
testl(-0x1.23456789abcdefp-2000L, 0x1.fedcba987654321p-2000L,
0x1.b72ea61d950c862p-2001L, -1);
#endif
printf("ok 1 - rem\n");
/*
* The actual quotient here is 864062210.50000003..., but
* double-precision division gets -8.64062210.5, which rounds
* the wrong way. This test ensures that remquo() is smart
* enough to get the low-order bit right.
*/
testd(-0x1.98260f22fc6dep-302, 0x1.fb3167c430a13p-332,
0x1.fb3165b82de72p-333, -864062211);
/* Even harder cases with greater exponent separation */
test(0x1.fp100, 0x1.ep-40, -0x1.cp-41, 143165577);
testd(-0x1.abcdefp120, 0x1.87654321p-120,
-0x1.69c78ec4p-121, -63816414);
printf("ok 2 - rem\n");
test(0x1.66666cp+120, 0x1p+71, 0.0, 1476395008);
testd(-0x1.0000000000003p+0, 0x1.0000000000003p+0, -0.0, -1);
testl(-0x1.0000000000003p+0, 0x1.0000000000003p+0, -0.0, -1);
testd(-0x1.0000000000001p-749, 0x1.4p-1072, 0x1p-1074, -1288490189);
testl(-0x1.0000000000001p-749, 0x1.4p-1072, 0x1p-1074, -1288490189);
printf("ok 3 - rem\n");
return (0);
}
int main()
{
test(0.0);
testf(0.0);
testl(0.0);
return 0;
}
int main ()
{
test (1.0);
testf (1.0f);
testl (1.0l);
return 0;
}
mnist_dataset(const std::string& path, const bool out_of_n_coding = true){
std::cout << "Reading MNIST dataset..."<<std::flush;
std::ifstream ftraind((path + "/train-images.idx3-ubyte").c_str(),std::ios::in | std::ios::binary); // image data
std::ifstream ftrainl((path + "/train-labels.idx1-ubyte").c_str(),std::ios::in | std::ios::binary); // label data
std::ifstream ftestd ((path + "/t10k-images.idx3-ubyte").c_str(),std::ios::in | std::ios::binary); // image data
std::ifstream ftestl ((path + "/t10k-labels.idx1-ubyte").c_str(),std::ios::in | std::ios::binary); // label data
assert(ftraind.is_open());
assert(ftrainl.is_open());
assert(ftestd.is_open());
assert(ftestl.is_open());
char buf[16];
ftraind.read(buf,16);
ftrainl.read(buf, 8);
ftestd.read(buf,16);
ftestl.read(buf, 8);
cuv::tensor<unsigned char,cuv::host_memory_space> traind(cuv::extents[60000][784]);
cuv::tensor<unsigned char,cuv::host_memory_space> trainl(cuv::extents[60000]);
cuv::tensor<unsigned char,cuv::host_memory_space> testd(cuv::extents[10000][784]);
cuv::tensor<unsigned char,cuv::host_memory_space> testl(cuv::extents[10000]);
ftraind.read((char*)traind.ptr(), traind.size()); assert(ftraind.good());
ftrainl.read((char*)trainl.ptr(), trainl.size()); assert(ftrainl.good());
ftestd.read((char*)testd.ptr(), testd.size()); assert(ftestd.good());
ftestl.read((char*)testl.ptr(), testl.size()); assert(ftestl.good());
train_data.resize(traind.shape());
test_data.resize(testd.shape());
convert(train_data , traind); // convert data type
convert(test_data , testd); // convert data type
if (out_of_n_coding){
train_labels.resize(cuv::extents[60000][10]);
test_labels.resize(cuv::extents[10000][10]);
train_labels = 0.f;
test_labels = 0.f;
for (unsigned int i = 0; i < trainl.size(); ++i){
train_labels(i, trainl[i]) = 1.f;
}
for (unsigned int i = 0; i < testl.size(); ++i){
test_labels(i, testl[i]) = 1.f;
}
} else {
train_labels.resize(cuv::extents[60000]);
test_labels.resize(cuv::extents[10000]);
for (unsigned int i = 0; i < trainl.size(); ++i){
train_labels(i) = trainl[i];
}
for (unsigned int i = 0; i < testl.size(); ++i){
test_labels(i) = testl[i];
}
}
//train_data = train_data[cuv::indices[cuv::index_range(0,5000)][cuv::index_range()]];
//train_labels = train_labels[cuv::indices[cuv::index_range(0,5000)][cuv::index_range()]];
binary = true;
channels = 1;
image_size = 28;
std::cout << "done."<<std::endl;
}
// Helper to insert argument slots into the stack.
// arg_slots must be a multiple of stack_move_unit() and <= 0
void MethodHandles::insert_arg_slots(MacroAssembler* _masm,
RegisterOrConstant arg_slots,
int arg_mask,
Register rax_argslot,
Register rbx_temp, Register rdx_temp, Register temp3_reg) {
assert(temp3_reg == noreg, "temp3 not required");
assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
(!arg_slots.is_register() ? rsp : arg_slots.as_register()));
#ifdef ASSERT
verify_argslot(_masm, rax_argslot, "insertion point must fall within current frame");
if (arg_slots.is_register()) {
Label L_ok, L_bad;
__ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
__ jccb(Assembler::greater, L_bad);
__ testl(arg_slots.as_register(), -stack_move_unit() - 1);
__ jccb(Assembler::zero, L_ok);
__ bind(L_bad);
__ stop("assert arg_slots <= 0 and clear low bits");
__ bind(L_ok);
} else {
assert(arg_slots.as_constant() <= 0, "");
assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
}
#endif //ASSERT
#ifdef _LP64
if (arg_slots.is_register()) {
// clean high bits of stack motion register (was loaded as an int)
__ movslq(arg_slots.as_register(), arg_slots.as_register());
}
#endif
// Make space on the stack for the inserted argument(s).
// Then pull down everything shallower than rax_argslot.
// The stacked return address gets pulled down with everything else.
// That is, copy [rsp, argslot) downward by -size words. In pseudo-code:
// rsp -= size;
// for (rdx = rsp + size; rdx < argslot; rdx++)
// rdx[-size] = rdx[0]
// argslot -= size;
BLOCK_COMMENT("insert_arg_slots {");
__ mov(rdx_temp, rsp); // source pointer for copy
__ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
{
Label loop;
__ BIND(loop);
// pull one word down each time through the loop
__ movptr(rbx_temp, Address(rdx_temp, 0));
__ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
__ addptr(rdx_temp, wordSize);
__ cmpptr(rdx_temp, rax_argslot);
__ jccb(Assembler::less, loop);
}
// Now move the argslot down, to point to the opened-up space.
__ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
BLOCK_COMMENT("} insert_arg_slots");
}
int
main ()
{
int res[6];
testf (f, res);
test (d, res);
testl (ld, res);
return 0;
}
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;
}
int main()
{
if (testf (__FLT_MAX__) < 1)
abort ();
if (test (__DBL_MAX__) < 1)
abort ();
if (testl (__LDBL_MAX__) < 1)
abort ();
return 0;
}
void MnistDataSet<M>::read() {
std::string path = "/home/local/datasets/MNIST";
ifstream ftraind((path + "/train-images.idx3-ubyte").c_str());
ifstream ftrainl((path + "/train-labels.idx1-ubyte").c_str());
ifstream ftestd ((path + "/t10k-images.idx3-ubyte").c_str());
ifstream ftestl ((path + "/t10k-labels.idx1-ubyte").c_str());
char buf[16];
ftraind.read(buf,16); ftrainl.read(buf, 8);
ftestd.read(buf,16); ftestl.read(buf, 8);
tensor<unsigned char, host_memory_space> traind(extents[n_trainData][n_dim]);
tensor<unsigned char, host_memory_space> trainl(extents[n_trainData]);
tensor<unsigned char, host_memory_space> testd(extents[n_testData][n_dim]);
tensor<unsigned char, host_memory_space> testl(extents[n_testData]);
ftraind.read((char*)traind.ptr(), traind.size());
assert(ftraind.good());
ftrainl.read((char*)trainl.ptr(), trainl.size());
assert(ftrainl.good());
ftestd.read((char*)testd.ptr(), testd.size());
assert(ftestd.good());
ftestl.read((char*)testl.ptr(), testl.size());
assert(ftestl.good());
tensor<unsigned char, M> train_d(extents[n_trainData][n_dim]);
tensor<unsigned char, M> train_l(extents[n_trainData]);
tensor<unsigned char, M> test_d(extents[n_testData][n_dim]);
tensor<unsigned char, M> test_l(extents[n_testData]);
train_d = traind;
train_l = trainl;
test_d = testd;
test_l = testl;
// conversion to float:
convert(this->X_train, train_d);
convert(this->Y_train, train_l);
convert(this->X_test, test_d);
convert(this->Y_test, test_l);
}
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 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
}
void InterpreterStubs::generate_interpreter_fill_in_tags() {
comment_section("Interpreter fill in tags");
entry("interpreter_fill_in_tags");
comment("eax: return address of method");
comment("ebx: method");
comment("ecx: size of parameters. Guaranteed to be >= 1");
comment("edx: call info from call site");
comment("Must preserve eax, ebx, ecx");
// stack layout:
// sp return address of caller
// --> argument n
// --> ...
// --> argument 0
Label extended_call_info;
comment("Compact call info or normal call info?");
testl(edx, edx);
jcc(positive, Constant(extended_call_info));
Label loop_entry, loop_condition;
comment("We have a compact call info");
movl(edi, ecx);
bind(loop_entry);
decl(edi);
comment("Store int tag");
movl(Address(esp, edi, times_8, Constant(BytesPerWord)), Constant(int_tag));
comment("Test the bit in the call info");
GUARANTEE(CallInfo::format1_tag_start == 0, "Tag must start at bit position 0 for this code to work");
btl(edx, edi);
jcc(carry_clear, Constant(loop_condition));
comment("Store obj tag");
movl(Address(esp, edi, times_8, Constant(BytesPerWord)), Constant(obj_tag));
bind(loop_condition);
testl(edi, edi);
jcc(not_zero, Constant(loop_entry));
ret();
bind(extended_call_info);
comment("Normal call info");
// The following code is slightly complicated. "Bit offset" below
// pretends like the callinfo's are in a bit array, as follows:
// Callinfo describing bci and offset
// Size [16 bits] and stack info 0-3
// Stack info 4-11
// We ignore the fact that each of these words is preceded by a byte
// that makes it look like an instruction.
pushl(ecx);
pushl(ebx);
Label loopx_entry, loopx_done;
comment("Bit offset of first argument in CallInfo array");
movzxw(edx, Address(eax, Constant(5 + 1))); // total number of locals/expr
subl(edx, ecx); // number of locals/expr belonging to callee
shll(edx, Constant(2)); // number of bits per nybble
addl(edx, Constant(32 + 16)); // 48 bits is the 32 bit callinfo and 16bit size info
comment("Decrement argument count; move to more convenient register");
leal(esi, Address(ecx, Constant(-1)));
comment("Location of tag of esi-th local");
leal(ebx, Address(esp, Constant(3 * BytesPerWord)));
bind(loopx_entry);
comment("eax holds the return address");
comment("ebx holds address of the esi-th tag");
comment("esi is the local whose tag we are setting");
comment("edx contains the bit offset of Local 0 in the CallInfo array");
comment("Get bit offset of esi-th local");
leal(ecx, Address(edx, esi, times_4));
comment("From bit offset, get word offset, then multiply by 5");
movl(edi, ecx);
shrl(edi, Constant(5));
leal(edi, Address(edi, edi, times_4));
comment("Get the appropriate CallInfo word; extract the nybble");
movl(edi, Address(eax, edi, times_1, Constant(1)));
shrl(edi);
andl(edi, Constant(0xF));
comment("Tag is (1 << value) >> 1. This is 0 when value == 0");
movl(ecx, edi);
movl(edi, Constant(1));
shll(edi);
shrl(edi, Constant(1));
comment("Store the tag");
movl(Address(ebx), edi);
comment("Are we done?");
decl(esi);
addl(ebx, Constant(8));
testl(esi, esi);
jcc(greater_equal, Constant(loopx_entry));
bind(loopx_done);
popl(ebx);
popl(ecx);
ret();
entry_end(); // interpreter_fill_in_tags
}
int
main(int argc, char *argv[])
{
static const int ex_under = FE_UNDERFLOW | FE_INEXACT; /* shorthand */
static const int ex_over = FE_OVERFLOW | FE_INEXACT;
long double ldbl_small, ldbl_eps, ldbl_max;
printf("1..5\n");
#ifdef __i386__
fpsetprec(FP_PE);
#endif
/*
* We can't use a compile-time constant here because gcc on
* FreeBSD/i386 assumes long doubles are truncated to the
* double format.
*/
ldbl_small = ldexpl(1.0, LDBL_MIN_EXP - LDBL_MANT_DIG);
ldbl_eps = LDBL_EPSILON;
ldbl_max = ldexpl(1.0 - ldbl_eps / 2, LDBL_MAX_EXP);
/*
* Special cases involving zeroes.
*/
#define ztest(prec) \
test##prec(copysign##prec(1.0, nextafter##prec(0.0, -0.0)), -1.0, 0); \
test##prec(copysign##prec(1.0, nextafter##prec(-0.0, 0.0)), 1.0, 0); \
test##prec(copysign##prec(1.0, nexttoward##prec(0.0, -0.0)), -1.0, 0);\
test##prec(copysign##prec(1.0, nexttoward##prec(-0.0, 0.0)), 1.0, 0)
ztest();
ztest(f);
ztest(l);
#undef ztest
#define stest(next, eps, prec) \
test##prec(next(-0.0, 42.0), eps, ex_under); \
test##prec(next(0.0, -42.0), -eps, ex_under); \
test##prec(next(0.0, INFINITY), eps, ex_under); \
test##prec(next(-0.0, -INFINITY), -eps, ex_under)
stest(nextafter, 0x1p-1074, );
stest(nextafterf, 0x1p-149f, f);
stest(nextafterl, ldbl_small, l);
stest(nexttoward, 0x1p-1074, );
stest(nexttowardf, 0x1p-149f, f);
stest(nexttowardl, ldbl_small, l);
#undef stest
printf("ok 1 - next\n");
/*
* `x == y' and NaN tests
*/
testall(42.0, 42.0, 42.0, 0);
testall(-42.0, -42.0, -42.0, 0);
testall(INFINITY, INFINITY, INFINITY, 0);
testall(-INFINITY, -INFINITY, -INFINITY, 0);
testall(NAN, 42.0, NAN, 0);
testall(42.0, NAN, NAN, 0);
testall(NAN, NAN, NAN, 0);
printf("ok 2 - next\n");
/*
* Tests where x is an ordinary normalized number
*/
testboth(1.0, 2.0, 1.0 + DBL_EPSILON, 0, );
testboth(1.0, -INFINITY, 1.0 - DBL_EPSILON/2, 0, );
testboth(1.0, 2.0, 1.0 + FLT_EPSILON, 0, f);
testboth(1.0, -INFINITY, 1.0 - FLT_EPSILON/2, 0, f);
testboth(1.0, 2.0, 1.0 + ldbl_eps, 0, l);
testboth(1.0, -INFINITY, 1.0 - ldbl_eps/2, 0, l);
testboth(-1.0, 2.0, -1.0 + DBL_EPSILON/2, 0, );
testboth(-1.0, -INFINITY, -1.0 - DBL_EPSILON, 0, );
testboth(-1.0, 2.0, -1.0 + FLT_EPSILON/2, 0, f);
testboth(-1.0, -INFINITY, -1.0 - FLT_EPSILON, 0, f);
testboth(-1.0, 2.0, -1.0 + ldbl_eps/2, 0, l);
testboth(-1.0, -INFINITY, -1.0 - ldbl_eps, 0, l);
/* Cases where nextafter(...) != nexttoward(...) */
test(nexttoward(1.0, 1.0 + ldbl_eps), 1.0 + DBL_EPSILON, 0);
testf(nexttowardf(1.0, 1.0 + ldbl_eps), 1.0 + FLT_EPSILON, 0);
testl(nexttowardl(1.0, 1.0 + ldbl_eps), 1.0 + ldbl_eps, 0);
printf("ok 3 - next\n");
/*
* Tests at word boundaries, normalization boundaries, etc.
*/
testboth(0x1.87654ffffffffp+0, INFINITY, 0x1.87655p+0, 0, );
testboth(0x1.87655p+0, -INFINITY, 0x1.87654ffffffffp+0, 0, );
testboth(0x1.fffffffffffffp+0, INFINITY, 0x1p1, 0, );
testboth(0x1p1, -INFINITY, 0x1.fffffffffffffp+0, 0, );
testboth(0x0.fffffffffffffp-1022, INFINITY, 0x1p-1022, 0, );
testboth(0x1p-1022, -INFINITY, 0x0.fffffffffffffp-1022, ex_under, );
testboth(0x1.fffffep0f, INFINITY, 0x1p1, 0, f);
testboth(0x1p1, -INFINITY, 0x1.fffffep0f, 0, f);
testboth(0x0.fffffep-126f, INFINITY, 0x1p-126f, 0, f);
testboth(0x1p-126f, -INFINITY, 0x0.fffffep-126f, ex_under, f);
#if LDBL_MANT_DIG == 53
testboth(0x1.87654ffffffffp+0L, INFINITY, 0x1.87655p+0L, 0, l);
testboth(0x1.87655p+0L, -INFINITY, 0x1.87654ffffffffp+0L, 0, l);
testboth(0x1.fffffffffffffp+0L, INFINITY, 0x1p1L, 0, l);
testboth(0x1p1L, -INFINITY, 0x1.fffffffffffffp+0L, 0, l);
testboth(0x0.fffffffffffffp-1022L, INFINITY, 0x1p-1022L, 0, l);
testboth(0x1p-1022L, -INFINITY, 0x0.fffffffffffffp-1022L, ex_under, l);
#elif LDBL_MANT_DIG == 64 && !defined(__i386)
testboth(0x1.87654321fffffffep+0L, INFINITY, 0x1.87654322p+0L, 0, l);
testboth(0x1.87654322p+0L, -INFINITY, 0x1.87654321fffffffep+0L, 0, l);
testboth(0x1.fffffffffffffffep0L, INFINITY, 0x1p1L, 0, l);
testboth(0x1p1L, -INFINITY, 0x1.fffffffffffffffep0L, 0, l);
testboth(0x0.fffffffffffffffep-16382L, INFINITY, 0x1p-16382L, 0, l);
testboth(0x1p-16382L, -INFINITY,
0x0.fffffffffffffffep-16382L, ex_under, l);
#elif LDBL_MANT_DIG == 113
testboth(0x1.876543210987ffffffffffffffffp+0L, INFINITY,
0x1.876543210988p+0, 0, l);
testboth(0x1.876543210988p+0L, -INFINITY,
0x1.876543210987ffffffffffffffffp+0L, 0, l);
testboth(0x1.ffffffffffffffffffffffffffffp0L, INFINITY, 0x1p1L, 0, l);
testboth(0x1p1L, -INFINITY, 0x1.ffffffffffffffffffffffffffffp0L, 0, l);
testboth(0x0.ffffffffffffffffffffffffffffp-16382L, INFINITY,
0x1p-16382L, 0, l);
testboth(0x1p-16382L, -INFINITY,
0x0.ffffffffffffffffffffffffffffp-16382L, ex_under, l);
#endif
printf("ok 4 - next\n");
/*
* Overflow tests
*/
test(idd(nextafter(DBL_MAX, INFINITY)), INFINITY, ex_over);
test(idd(nextafter(INFINITY, 0.0)), DBL_MAX, 0);
test(idd(nexttoward(DBL_MAX, DBL_MAX * 2.0L)), INFINITY, ex_over);
#if LDBL_MANT_DIG > 53
test(idd(nexttoward(INFINITY, DBL_MAX * 2.0L)), DBL_MAX, 0);
#endif
testf(idf(nextafterf(FLT_MAX, INFINITY)), INFINITY, ex_over);
testf(idf(nextafterf(INFINITY, 0.0)), FLT_MAX, 0);
testf(idf(nexttowardf(FLT_MAX, FLT_MAX * 2.0)), INFINITY, ex_over);
testf(idf(nexttowardf(INFINITY, FLT_MAX * 2.0)), FLT_MAX, 0);
testboth(ldbl_max, INFINITY, INFINITY, ex_over, l);
testboth(INFINITY, 0.0, ldbl_max, 0, l);
printf("ok 5 - next\n");
return (0);
}
void CompilerStubs::generate_compiler_new_obj_array() {
comment_section("Compiler stub: new object array");
comment("- ebx holds the prototypical near of the array class");
entry("compiler_new_obj_array");
comment("Get array length");
// add BytesPerWord for return address
movl(edx, Address(esp, Constant(BytesPerWord + JavaFrame::arg_offset_from_sp(0))));
Label slow_case;
comment("Check if the array length is too large or negative");
cmpl(edx, Constant(maximum_safe_array_length));
jcc(above, Constant(slow_case));
comment("Get _inline_allocation_top");
movl(eax, Address(Constant("_inline_allocation_top")));
if (GenerateDebugAssembly) {
comment("Check ExcessiveGC");
testl(Address(Constant("ExcessiveGC")), Constant(0));
jcc(not_zero, Constant(slow_case));
}
comment("Compute new top");
leal(ecx, Address(eax, edx, times_4, Constant(Array::base_offset())));
comment("Check for overflow");
cmpl(ecx, eax);
jcc(below, 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("Set the length");
movl(Address(eax, Constant(Array::length_offset())), edx);
comment("Compute remaining size");
testl(edx, edx);
comment("Empty array?");
Label init_done;
jcc(equal, Constant(init_done));
comment("Zero array elements");
xorl(ecx, ecx);
Label init_loop;
bind(init_loop);
movl(Address(eax, edx, times_4, Constant(Array::base_offset() - oopSize)), ecx);
decrement(edx, 1);
jcc(not_zero, Constant(init_loop));
bind(init_done);
comment("The newly allocated array is in register eax");
ret();
comment("Slow case - call the VM runtime system");
bind(slow_case);
leal(eax, Address(Constant("anewarray")));
goto_shared_call_vm(T_ARRAY);
entry_end(); // compiler_new_obj_array
}
address generate_call_stub(address& return_address) {
StubCodeMark mark(this, "StubRoutines", "call_stub");
address start = __ pc();
// stub code parameters / addresses
assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code");
bool sse_save = false;
const Address esp_after_call(ebp, -4 * wordSize); // same as in generate_catch_exception()!
const Address mxcsr_save (ebp, -4 * wordSize);
const Address result (ebp, 3 * wordSize);
const Address result_type (ebp, 4 * wordSize);
const Address method (ebp, 5 * wordSize);
const Address entry_point (ebp, 6 * wordSize);
const Address parameters (ebp, 7 * wordSize);
const Address parameter_size(ebp, 8 * wordSize);
const Address thread (ebp, 9 * wordSize); // same as in generate_catch_exception()!
#ifdef COMPILER2
sse_save = VM_Version::supports_sse();
#endif
// stub code
__ enter();
// save edi, esi, & ebx, according to C calling conventions
__ pushl(edi);
__ pushl(esi);
__ pushl(ebx);
__ subl(esp, wordSize); // space for %mxcsr save
// save and initialize %mxcsr
if (sse_save) {
__ stmxcsr(mxcsr_save);
__ ldmxcsr(Address((int) StubRoutines::addr_mxcsr_std(), relocInfo::none));
}
#ifdef ASSERT
// make sure we have no pending exceptions
{ Label L;
__ movl(ecx, thread);
__ cmpl(Address(ecx, Thread::pending_exception_offset()), (int)NULL);
__ jcc(Assembler::equal, L);
__ stop("StubRoutines::call_stub: entered with pending exception");
__ bind(L);
}
#endif
// pass parameters if any
Label parameters_done;
__ movl(ecx, parameter_size); // parameter counter
__ testl(ecx, ecx);
__ jcc(Assembler::zero, parameters_done);
// parameter passing loop
Label loop;
__ movl(edx, parameters); // parameter pointer
__ movl(esi, ecx); // parameter counter is in esi now
__ movl(ecx, Address(edx)); // get first parameter in case it is a receiver
__ bind(loop);
__ movl(eax, Address(edx)); // get parameter
__ addl(edx, wordSize); // advance to next parameter
__ decl(esi); // decrement counter
__ pushl(eax); // pass parameter
__ jcc(Assembler::notZero, loop);
// call Java function
__ bind(parameters_done);
__ movl(ebx, method); // get methodOop
__ movl(esi, entry_point); // get entry_point
__ call(esi, relocInfo::none);
return_address = __ pc();
// store result depending on type
// (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT)
__ movl(edi, result);
Label is_long, is_float, is_double, exit;
__ movl(esi, result_type);
__ cmpl(esi, T_LONG);
__ jcc(Assembler::equal, is_long);
__ cmpl(esi, T_FLOAT);
__ jcc(Assembler::equal, is_float);
__ cmpl(esi, T_DOUBLE);
__ jcc(Assembler::equal, is_double);
// handle T_INT case
__ movl(Address(edi), eax);
__ bind(exit);
// pop parameters
__ movl(ecx, parameter_size);
__ leal(esp, Address(esp, ecx, Address::times_4));
// check if parameters have been popped correctly
#ifdef ASSERT
Label esp_wrong;
__ leal(edi, esp_after_call);
__ cmpl(esp, edi);
__ jcc(Assembler::notEqual, esp_wrong);
#endif
// restore %mxcsr
if (sse_save) {
__ ldmxcsr(mxcsr_save);
}
// restore edi & esi
__ addl(esp, wordSize); // remove %mxcsr save area
__ popl(ebx);
__ popl(esi);
__ popl(edi);
// return
__ popl(ebp);
__ ret(0);
// handle return types different from T_INT
__ bind(is_long);
__ movl(Address(edi, 0 * wordSize), eax);
__ movl(Address(edi, 1 * wordSize), edx);
__ jmp(exit);
__ bind(is_float);
__ fstp_s(Address(edi));
__ jmp(exit);
__ bind(is_double);
__ fstp_d(Address(edi));
__ jmp(exit);
#ifdef ASSERT
// stack pointer misadjusted
__ bind(esp_wrong);
__ stop("esp wrong after Java call");
#endif
return start;
}
address generate_verify_oop() {
StubCodeMark mark(this, "StubRoutines", "verify_oop");
address start = __ pc();
// Incoming arguments on stack after saving eax:
//
// [tos ]: saved edx
// [tos + 1]: saved EFLAGS
// [tos + 2]: return address
// [tos + 3]: char* error message
// [tos + 4]: oop object to verify
// [tos + 5]: saved eax - saved by caller and bashed
Label exit, error;
__ pushfd();
__ incl(Address((int)StubRoutines::verify_oop_count_addr(), relocInfo::none));
__ pushl(edx); // save edx
// make sure object is 'reasonable'
__ movl(eax, Address(esp, 4 * wordSize)); // get object
__ testl(eax, eax);
__ jcc(Assembler::zero, exit); // if obj is NULL it is ok
// Check if the oop is in the right area of memory
const int oop_mask = Universe::verify_oop_mask();
const int oop_bits = Universe::verify_oop_bits();
__ movl(edx, eax);
__ andl(edx, oop_mask);
__ cmpl(edx, oop_bits);
__ jcc(Assembler::notZero, error);
// make sure klass is 'reasonable'
__ movl(eax, Address(eax, oopDesc::klass_offset_in_bytes())); // get klass
__ testl(eax, eax);
__ jcc(Assembler::zero, error); // if klass is NULL it is broken
// Check if the klass is in the right area of memory
const int klass_mask = Universe::verify_klass_mask();
const int klass_bits = Universe::verify_klass_bits();
__ movl(edx, eax);
__ andl(edx, klass_mask);
__ cmpl(edx, klass_bits);
__ jcc(Assembler::notZero, error);
// make sure klass' klass is 'reasonable'
__ movl(eax, Address(eax, oopDesc::klass_offset_in_bytes())); // get klass' klass
__ testl(eax, eax);
__ jcc(Assembler::zero, error); // if klass' klass is NULL it is broken
__ movl(edx, eax);
__ andl(edx, klass_mask);
__ cmpl(edx, klass_bits);
__ jcc(Assembler::notZero, error); // if klass not in right area
// of memory it is broken too.
// return if everything seems ok
__ bind(exit);
__ movl(eax, Address(esp, 5 * wordSize)); // get saved eax back
__ popl(edx); // restore edx
__ popfd(); // restore EFLAGS
__ ret(3 * wordSize); // pop arguments
// handle errors
__ bind(error);
__ movl(eax, Address(esp, 5 * wordSize)); // get saved eax back
__ popl(edx); // get saved edx back
__ popfd(); // get saved EFLAGS off stack -- will be ignored
__ pushad(); // push registers (eip = return address & msg are already pushed)
__ call(CAST_FROM_FN_PTR(address, MacroAssembler::debug), relocInfo::runtime_call_type);
__ popad();
__ ret(3 * wordSize); // pop arguments
return start;
}
void NativeGenerator::generate_native_system_entries() {
comment_section("Native entry points for system functions");
rom_linkable_entry("native_jvm_unchecked_byte_arraycopy_entry");
jmp(Constant("native_system_arraycopy_entry"));
rom_linkable_entry_end();
rom_linkable_entry("native_jvm_unchecked_char_arraycopy_entry");
jmp(Constant("native_system_arraycopy_entry"));
rom_linkable_entry_end();
rom_linkable_entry("native_jvm_unchecked_int_arraycopy_entry");
jmp(Constant("native_system_arraycopy_entry"));
rom_linkable_entry_end();
rom_linkable_entry("native_jvm_unchecked_long_arraycopy_entry");
jmp(Constant("native_system_arraycopy_entry"));
rom_linkable_entry_end();
rom_linkable_entry("native_jvm_unchecked_obj_arraycopy_entry");
jmp(Constant("native_system_arraycopy_entry"));
rom_linkable_entry_end();
rom_linkable_entry("native_system_arraycopy_entry");
wtk_profile_quick_call(/* param_size*/ 5);
Label bailout, cont, try_2_byte, try_4_byte, try_8_byte, do_4_byte;
// public static native void arraycopy(Object src, int src_position,
// Object dst, int dst_position,
// int length);
comment("preserve method");
pushl(ebx);
// 8 is for the preserved method and the return address
int length_offset = JavaFrame::arg_offset_from_sp(0) + 8,
dst_pos_offset = JavaFrame::arg_offset_from_sp(1) + 8,
dst_offset = JavaFrame::arg_offset_from_sp(2) + 8,
src_pos_offset = JavaFrame::arg_offset_from_sp(3) + 8,
src_offset = JavaFrame::arg_offset_from_sp(4) + 8;
comment("load arguments to registers");
movl(ecx, Address(esp, Constant(length_offset)));
movl(edi, Address(esp, Constant(dst_pos_offset)));
movl(edx, Address(esp, Constant(dst_offset)));
movl(esi, Address(esp, Constant(src_pos_offset)));
movl(eax, Address(esp, Constant(src_offset)));
// eax = src
// ebx = tmp register
// edx = dst
// ecx = length
// esi = src_pos
// edi = dst_pos
comment("if (src == NULL) goto bailout;");
testl( eax, eax );
jcc(zero, Constant(bailout));
comment("if (dst == NULL) goto bailout;");
testl( edx, edx );
jcc(zero, Constant(bailout));
comment("if (length < 0 || src_pos < 0 || dst_pos < 0) goto bailout;");
movl(ebx, ecx);
orl(ebx, esi);
orl(ebx, edi);
jcc(negative, Constant(bailout));
comment("if ((unsigned int) dst.length < (unsigned int) dst_pos + (unsigned int) length) goto bailout;");
movl(ebx, ecx);
addl(ebx, edi);
cmpl(Address(edx, Constant(Array::length_offset())), ebx);
jcc(below, Constant(bailout));
comment("if ((unsigned int) src.length < (unsigned int) src_pos + (unsigned int) length) goto bailout;");
movl(ebx, ecx);
addl(ebx, esi);
cmpl(Address(eax, Constant(Array::length_offset())), ebx);
jcc(below, Constant(bailout));
comment("Same near test");
comment("if (src.near != dst.near) goto bailout;");
movl(ebx, Address(eax, Constant(Oop::klass_offset())));
cmpl(ebx, Address(edx, Constant(Oop::klass_offset())));
jcc(not_equal, Constant(bailout));
comment("load the instance_size");
movl(ebx, Address(ebx, Constant(JavaNear::klass_offset())));
movsxw(ebx, Address(ebx, Constant(FarClass::instance_size_offset())));
comment("if (instance_size != size_type_array_1()) goto try_2_byte");
cmpl(ebx, Constant(InstanceSize::size_type_array_1));
jcc(not_equal, Constant(try_2_byte));
leal(esi, Address(eax, esi, times_1, Constant(Array::base_offset())));
leal(edi, Address(edx, edi, times_1, Constant(Array::base_offset())));
jmp(Constant(cont));
bind(try_2_byte);
comment("if (instance_size != size_type_array_2()) goto try_4_byte");
cmpl(ebx, Constant(InstanceSize::size_type_array_2));
jcc(not_equal, Constant(try_4_byte));
leal(esi, Address(eax, esi, times_2, Constant(Array::base_offset())));
leal(edi, Address(edx, edi, times_2, Constant(Array::base_offset())));
shll(ecx, Constant(1));
jmp(Constant(cont));
bind(try_4_byte);
comment("if (instance_size == size_type_array_4()) goto do_4_byte");
cmpl(ebx, Constant(InstanceSize::size_type_array_4));
jcc(equal, Constant(do_4_byte) );
comment("if (instance_size != size_obj_array()) goto bailout");
cmpl(ebx, Constant(InstanceSize::size_obj_array));
jcc(not_equal, Constant(bailout));
comment("if (dst < old_generation_end) goto bailout");
cmpl( edx, Address( Constant( "_old_generation_end" ) ) );
jcc( below, Constant(bailout));
bind(do_4_byte);
leal(esi, Address(eax, esi, times_4, Constant(Array::base_offset())));
leal(edi, Address(edx, edi, times_4, Constant(Array::base_offset())));
shll(ecx, Constant(2));
bind(cont);
comment("memmove(edi, esi, ecx);");
pushl(ecx);
pushl(esi);
pushl(edi);
call(Constant("memmove"));
addl(esp, Constant(16));
ret(Constant(5 * BytesPerStackElement));
comment("Bail out to the general arraycopy implementation");
bind(bailout);
comment("pop method");
popl(ebx);
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end(); // native_system_arraycopy_entry
}
void NativeGenerator::generate_native_string_entries() {
comment_section("Native entry points for string functions");
{
//--------------------java.lang.String.indexof0---------------------------
rom_linkable_entry("native_string_indexof0_entry");
wtk_profile_quick_call(/* param_size*/ 2);
comment("Pop the return address");
popl(edi);
comment("Push zero for fromIndex");
pushl(Constant(0));
comment("Push back the return address");
pushl(edi);
jmp(Constant("native_string_indexof_entry"));
rom_linkable_entry_end(); // native_string_indexof0_entry
//--------------------java.lang.String.indexof---------------------------
rom_linkable_entry("native_string_indexof_entry");
Label cont, loop, test, failure, success;
wtk_profile_quick_call(/* param_size*/ 3);
comment("Pop the return address");
popl(edi);
comment("Pop the argument: fromIndex");
pop_int(eax, eax);
comment("Pop the argument: ch");
pop_int(ebx, ebx);
comment("Pop the receiver");
pop_obj(ecx, ecx);
cmpl(ebx, Constant(0xFFFF));
jcc(greater, Constant(failure));
cmpl(eax, Constant(0));
jcc(greater_equal, Constant(cont));
movl(eax, Constant(0));
bind(cont);
movl(esi, Address(ecx, Constant(String::count_offset())));
comment("if (fromIndex >= count) { return -1; }");
cmpl(eax, esi);
jcc(greater_equal, Constant(failure));
movl(edx, Address(ecx, Constant(String::offset_offset())));
addl(eax, edx); // i = offset + fromIndex
addl(edx, esi); // int max = offset + count;
movl(esi, Address(ecx, Constant(String::value_offset()))); // v = value.
jmp(Constant(test));
bind(loop);
cmpw(Address(esi, eax, times_2, Constant(Array::base_offset())), ebx);
jcc(equal, Constant(success));
incl(eax);
bind(test);
cmpl(eax, edx);
jcc(less, Constant(loop));
comment("Return -1 by pushing the value and jumping to the return address");
bind(failure);
push_int(-1);
jmp(edi);
comment("Return i - offset by pushing the value and jumping to the return address");
bind(success);
movl(esi, Address(ecx, Constant(String::offset_offset()))); // i = offset + fromIndex
subl(eax, esi);
push_int(eax);
jmp(edi);
rom_linkable_entry_end(); // native_string_indexof_entry
}
//----------------------java.lang.String.charAt---------------------------
{
rom_linkable_entry("native_string_charAt_entry");
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end();
}
//----------------------java.lang.String(java.lang.StringBuffer)-------------
{
rom_linkable_entry("native_string_init_entry");
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end();
}
//----------------------java.lang.String.equals(java.lang.Object)------------
{
rom_linkable_entry("native_string_equals_entry");
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end();
}
//----------------------java.lang.String.indexOf(java.lang.String)-----------
{
rom_linkable_entry("native_string_indexof0_string_entry");
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end();
}
//----------------------java.lang.String.indexOf(java.lang.String)-----------
{
rom_linkable_entry("native_string_indexof_string_entry");
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end();
}
//----------------------java.lang.String.compareTo---------------------------
{ // java.lang.String.compareTo
// Method int compareTo(java.lang.String)
rom_linkable_entry("native_string_compareTo_entry");
wtk_profile_quick_call(/* param_size*/ 2);
comment("preserve method");
pushl(ebx);
// 8 is return address plus pushed method
int str1_offset = JavaFrame::arg_offset_from_sp(0) + 8,
str0_offset = JavaFrame::arg_offset_from_sp(1) + 8;
comment("load arguments to registers");
movl(ecx, Address(esp, Constant(str1_offset)));
movl(eax, Address(esp, Constant(str0_offset)));
// eax: str0: this String
// ebx: str1: String to compare against
Label bailout;
comment("Null check");
testl(ecx, ecx);
jcc(zero, Constant(bailout));
comment("get str0.value[]");
movl(esi, Address(eax, Constant(String::value_offset())));
comment("get str0.offset");
movl(ebx, Address(eax, Constant(String::offset_offset())));
comment("compute start of character data");
leal(esi, Address(esi, ebx, times_2, Constant(Array::base_offset())));
comment("get str0.count");
movl(eax, Address(eax, Constant(String::count_offset())));
comment("get str1.value[]");
movl(edi, Address(ecx, Constant(String::value_offset())));
comment("get str1.offset");
movl(ebx, Address(ecx, Constant(String::offset_offset())));
comment("compute start of character data");
leal(edi, Address(edi, ebx, times_2, Constant(Array::base_offset())));
comment("get str1.count");
movl(ebx, Address(ecx, Constant(String::count_offset())));
// esi = str0 start of character data
// edi = str1 start of character data
// eax = str0 length
// ebx = str1 length
Label str1_longest;
subl(eax, ebx);
jcc(greater_equal, Constant(str1_longest));
// str1 is longer than str0
addl(ebx, eax);
bind(str1_longest);
// esi = str0 start of character data
// edi = str1 start of character data
// eax = str0.count - str1.count
// ebx = min(str0.count, str1.count)
// save str0.count - str1.count, we might need it later
pushl(eax);
xorl(ecx, ecx);
Label loop, check_lengths, done;
bind(loop);
cmpl(ecx, ebx);
jcc(above_equal, Constant(check_lengths));
movzxw(eax, Address(esi, ecx, times_2));
movzxw(edx, Address(edi, ecx, times_2));
subl(eax, edx);
jcc(not_equal, Constant(done));
incl(ecx);
jmp(Constant(loop));
bind(check_lengths);
movl(eax, Address(esp));
bind(done);
popl(ebx); // remove saved length difference
// Push result on stack and return to caller
popl(ebx); // remove method
popl(edi); // pop return address
addl(esp, Constant(2 * BytesPerStackElement)); // remove arguments
push_int(eax); // push result
jmp(edi); // return
comment("Bail out to the general compareTo implementation");
bind(bailout);
comment("pop method");
popl(ebx);
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end(); // native_string_compareTo_entry
}
//----------------------java.lang.String.endsWith----------------
{
// java.lang.String.endsWith
// Method boolean endsWith(java.lang.String)
rom_linkable_entry("native_string_endsWith_entry");
wtk_profile_quick_call(/* param_size*/ 2);
Label bailout;
// 4 is return address
int suffix_offset = JavaFrame::arg_offset_from_sp(0) + 4,
this_offset = JavaFrame::arg_offset_from_sp(1) + 4;
comment("load arguments to registers");
movl(eax, Address(esp, Constant(suffix_offset)));
cmpl(eax, Constant(0));
jcc(equal, Constant(bailout));
movl(ecx, Address(esp, Constant(this_offset)));
comment("Pop the return address");
popl(edi);
movl(edx, Address(ecx, Constant(String::count_offset())));
subl(edx, Address(eax, Constant(String::count_offset())));
comment("Push (this.count - suffix.count) for toffset");
pushl(edx);
comment("Push back the return address");
pushl(edi);
jmp(Constant("native_string_startsWith_entry"));
comment("Bail out to the general startsWith implementation");
bind(bailout);
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end(); // native_string_endsWith_entry
}
//----------------------java.lang.String.startsWith----------------
{
// java.lang.String.startsWith
// Method boolean startsWith(java.lang.String)
rom_linkable_entry("native_string_startsWith0_entry");
wtk_profile_quick_call(/* param_size*/ 2);
Label bailout;
// 4 is return address
int prefix_offset = JavaFrame::arg_offset_from_sp(0) + 4;
comment("Check if prefix is null");
cmpl(Address(esp, Constant(prefix_offset)), Constant(0));
jcc(equal, Constant(bailout));
comment("Pop the return address");
popl(edi);
comment("Push zero for toffset");
pushl(Constant(0));
comment("Push back the return address");
pushl(edi);
jmp(Constant("native_string_startsWith_entry"));
comment("Bail out to the general startsWith implementation");
bind(bailout);
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end(); // native_string_startsWith0_entry
}
{
// ----------- java.lang.String.startsWith ------------------------------
// Method boolean startsWith(java.lang.String,int)
rom_linkable_entry("native_string_startsWith_entry");
wtk_profile_quick_call(/* param_size*/ 3);
Label bailout, return_false;
// 4 is return address
int prefix_offset = JavaFrame::arg_offset_from_sp(1) + 4;
comment("Check if prefix is null");
cmpl(Address(esp, Constant(prefix_offset)), Constant(0));
jcc(equal, Constant(bailout));
comment("Pop the return address");
popl(edi);
comment("Pop the argument: toffset");
pop_int(edx, edx);
comment("Pop the argument: prefix");
pop_obj(eax, eax);
comment("Pop the receiver");
pop_obj(ecx, ecx);
comment("Preserve the return address");
pushl(edi);
// ecx: this String
// eax: prefix
cmpl(edx, Constant(0));
jcc(less, Constant(return_false));
comment("if (toffset > this.count - prefix.count) return false;");
movl(ebx, Address(ecx, Constant(String::count_offset())));
subl(ebx, Address(eax, Constant(String::count_offset())));
cmpl(edx, ebx);
jcc(greater, Constant(return_false));
comment("get this.value[]");
movl(esi, Address(ecx, Constant(String::value_offset())));
comment("get this.offset");
movl(ebx, Address(ecx, Constant(String::offset_offset())));
comment("add toffset");
addl(ebx, edx);
comment("compute start of character data");
leal(esi, Address(esi, ebx, times_2, Constant(Array::base_offset())));
comment("get prefix.value[]");
movl(edi, Address(eax, Constant(String::value_offset())));
comment("get prefix.offset");
movl(ebx, Address(eax, Constant(String::offset_offset())));
comment("compute start of character data");
leal(edi, Address(edi, ebx, times_2, Constant(Array::base_offset())));
comment("get prefix.count");
movl(ecx, Address(eax, Constant(String::count_offset())));
comment("get the number of bytes to compare");
shll(ecx, Constant(1));
comment("memcmp(edi, esi, ecx);");
pushl(ecx);
pushl(esi);
pushl(edi);
if (GenerateInlineAsm) {
// VC++ treats memcmp() as an intrinsic function and would cause
// reference to memcmp in Interpreter_i386.c to fail to compile.
call(Constant("memcmp_from_interpreter"));
} else {
call(Constant("memcmp"));
}
addl(esp, Constant(12));
cmpl(eax, Constant(0));
jcc(not_equal, Constant(return_false));
// Push 1 on stack and return to caller
popl(edi); // pop return address
push_int(1); // push result
jmp(edi); // return
bind(return_false);
// Push 0 on stack and return to caller
popl(edi); // pop return address
push_int(0); // push result
jmp(edi); // return
comment("Bail out to the general startsWith implementation");
bind(bailout);
if (AddExternCUnderscore) {
emit_instruction("jmp _interpreter_method_entry");
} else {
emit_instruction("jmp interpreter_method_entry");
}
rom_linkable_entry_end(); // native_string_startsWith_entry
}
}
address TemplateInterpreterGenerator::generate_slow_signature_handler() {
address entry = __ pc();
// rbx: method
// r14: pointer to locals
// c_rarg3: first stack arg - wordSize
__ mov(c_rarg3, rsp);
// adjust rsp
__ subptr(rsp, 14 * wordSize);
__ call_VM(noreg,
CAST_FROM_FN_PTR(address,
InterpreterRuntime::slow_signature_handler),
rbx, r14, c_rarg3);
// rax: result handler
// Stack layout:
// rsp: 5 integer args (if static first is unused)
// 1 float/double identifiers
// 8 double args
// return address
// stack args
// garbage
// expression stack bottom
// bcp (NULL)
// ...
// Do FP first so we can use c_rarg3 as temp
__ 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);
}
// Now handle integrals. Only do c_rarg1 if not static.
__ 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));
// restore rsp
__ addptr(rsp, 14 * wordSize);
__ ret(0);
return entry;
}
// Helper to remove argument slots from the stack.
// arg_slots must be a multiple of stack_move_unit() and >= 0
void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
RegisterOrConstant arg_slots,
Register rax_argslot,
Register rbx_temp, Register rdx_temp, Register temp3_reg) {
assert(temp3_reg == noreg, "temp3 not required");
assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
(!arg_slots.is_register() ? rsp : arg_slots.as_register()));
#ifdef ASSERT
// Verify that [argslot..argslot+size) lies within (rsp, rbp).
__ lea(rbx_temp, Address(rax_argslot, arg_slots, Address::times_ptr));
verify_argslot(_masm, rbx_temp, "deleted argument(s) must fall within current frame");
if (arg_slots.is_register()) {
Label L_ok, L_bad;
__ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
__ jccb(Assembler::less, L_bad);
__ testl(arg_slots.as_register(), -stack_move_unit() - 1);
__ jccb(Assembler::zero, L_ok);
__ bind(L_bad);
__ stop("assert arg_slots >= 0 and clear low bits");
__ bind(L_ok);
} else {
assert(arg_slots.as_constant() >= 0, "");
assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
}
#endif //ASSERT
#ifdef _LP64
if (false) { // not needed, since register is positive
// clean high bits of stack motion register (was loaded as an int)
if (arg_slots.is_register())
__ movslq(arg_slots.as_register(), arg_slots.as_register());
}
#endif
BLOCK_COMMENT("remove_arg_slots {");
// Pull up everything shallower than rax_argslot.
// Then remove the excess space on the stack.
// The stacked return address gets pulled up with everything else.
// That is, copy [rsp, argslot) upward by size words. In pseudo-code:
// for (rdx = argslot-1; rdx >= rsp; --rdx)
// rdx[size] = rdx[0]
// argslot += size;
// rsp += size;
__ lea(rdx_temp, Address(rax_argslot, -wordSize)); // source pointer for copy
{
Label loop;
__ BIND(loop);
// pull one word up each time through the loop
__ movptr(rbx_temp, Address(rdx_temp, 0));
__ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
__ addptr(rdx_temp, -wordSize);
__ cmpptr(rdx_temp, rsp);
__ jccb(Assembler::greaterEqual, loop);
}
// Now move the argslot up, to point to the just-copied block.
__ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
// And adjust the argslot address to point at the deletion point.
__ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
BLOCK_COMMENT("} remove_arg_slots");
}
//------------------------------------------------------------------------------
// MethodHandles::generate_method_handle_stub
//
// Generate an "entry" field for a method handle.
// This determines how the method handle will respond to calls.
void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {
// Here is the register state during an interpreted call,
// as set up by generate_method_handle_interpreter_entry():
// - rbx: garbage temp (was MethodHandle.invoke methodOop, unused)
// - rcx: receiver method handle
// - rax: method handle type (only used by the check_mtype entry point)
// - rsi/r13: sender SP (must preserve; see prepare_to_jump_from_interpreted)
// - rdx: garbage temp, can blow away
const Register rcx_recv = rcx;
const Register rax_argslot = rax;
const Register rbx_temp = rbx;
const Register rdx_temp = rdx;
// This guy is set up by prepare_to_jump_from_interpreted (from interpreted calls)
// and gen_c2i_adapter (from compiled calls):
const Register saved_last_sp = LP64_ONLY(r13) NOT_LP64(rsi);
// Argument registers for _raise_exception.
// 32-bit: Pass first two oop/int args in registers ECX and EDX.
const Register rarg0_code = LP64_ONLY(j_rarg0) NOT_LP64(rcx);
const Register rarg1_actual = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
const Register rarg2_required = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
assert_different_registers(rarg0_code, rarg1_actual, rarg2_required, saved_last_sp);
guarantee(java_lang_invoke_MethodHandle::vmentry_offset_in_bytes() != 0, "must have offsets");
// some handy addresses
Address rbx_method_fie( rbx, methodOopDesc::from_interpreted_offset() );
Address rbx_method_fce( rbx, methodOopDesc::from_compiled_offset() );
Address rcx_mh_vmtarget( rcx_recv, java_lang_invoke_MethodHandle::vmtarget_offset_in_bytes() );
Address rcx_dmh_vmindex( rcx_recv, java_lang_invoke_DirectMethodHandle::vmindex_offset_in_bytes() );
Address rcx_bmh_vmargslot( rcx_recv, java_lang_invoke_BoundMethodHandle::vmargslot_offset_in_bytes() );
Address rcx_bmh_argument( rcx_recv, java_lang_invoke_BoundMethodHandle::argument_offset_in_bytes() );
Address rcx_amh_vmargslot( rcx_recv, java_lang_invoke_AdapterMethodHandle::vmargslot_offset_in_bytes() );
Address rcx_amh_argument( rcx_recv, java_lang_invoke_AdapterMethodHandle::argument_offset_in_bytes() );
Address rcx_amh_conversion( rcx_recv, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes() );
Address vmarg; // __ argument_address(vmargslot)
const int java_mirror_offset = klassOopDesc::klass_part_offset_in_bytes() + Klass::java_mirror_offset_in_bytes();
if (have_entry(ek)) {
__ nop(); // empty stubs make SG sick
return;
}
address interp_entry = __ pc();
trace_method_handle(_masm, entry_name(ek));
BLOCK_COMMENT(entry_name(ek));
switch ((int) ek) {
case _raise_exception:
{
// Not a real MH entry, but rather shared code for raising an
// exception. Since we use the compiled entry, arguments are
// expected in compiler argument registers.
assert(raise_exception_method(), "must be set");
assert(raise_exception_method()->from_compiled_entry(), "method must be linked");
const Register rdi_pc = rax;
__ pop(rdi_pc); // caller PC
__ mov(rsp, saved_last_sp); // cut the stack back to where the caller started
Register rbx_method = rbx_temp;
Label L_no_method;
// FIXME: fill in _raise_exception_method with a suitable java.lang.invoke method
__ movptr(rbx_method, ExternalAddress((address) &_raise_exception_method));
__ testptr(rbx_method, rbx_method);
__ jccb(Assembler::zero, L_no_method);
const int jobject_oop_offset = 0;
__ movptr(rbx_method, Address(rbx_method, jobject_oop_offset)); // dereference the jobject
__ testptr(rbx_method, rbx_method);
__ jccb(Assembler::zero, L_no_method);
__ verify_oop(rbx_method);
NOT_LP64(__ push(rarg2_required));
__ push(rdi_pc); // restore caller PC
__ jmp(rbx_method_fce); // jump to compiled entry
// Do something that is at least causes a valid throw from the interpreter.
__ bind(L_no_method);
__ push(rarg2_required);
__ push(rarg1_actual);
__ jump(ExternalAddress(Interpreter::throw_WrongMethodType_entry()));
}
break;
case _invokestatic_mh:
case _invokespecial_mh:
{
Register rbx_method = rbx_temp;
__ load_heap_oop(rbx_method, rcx_mh_vmtarget); // target is a methodOop
__ verify_oop(rbx_method);
// same as TemplateTable::invokestatic or invokespecial,
// minus the CP setup and profiling:
if (ek == _invokespecial_mh) {
// Must load & check the first argument before entering the target method.
__ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
__ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
__ null_check(rcx_recv);
__ verify_oop(rcx_recv);
}
__ jmp(rbx_method_fie);
}
break;
case _invokevirtual_mh:
{
// same as TemplateTable::invokevirtual,
// minus the CP setup and profiling:
// pick out the vtable index and receiver offset from the MH,
// and then we can discard it:
__ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
Register rbx_index = rbx_temp;
__ movl(rbx_index, rcx_dmh_vmindex);
// Note: The verifier allows us to ignore rcx_mh_vmtarget.
__ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
__ null_check(rcx_recv, oopDesc::klass_offset_in_bytes());
// get receiver klass
Register rax_klass = rax_argslot;
__ load_klass(rax_klass, rcx_recv);
__ verify_oop(rax_klass);
// get target methodOop & entry point
const int base = instanceKlass::vtable_start_offset() * wordSize;
assert(vtableEntry::size() * wordSize == wordSize, "adjust the scaling in the code below");
Address vtable_entry_addr(rax_klass,
rbx_index, Address::times_ptr,
base + vtableEntry::method_offset_in_bytes());
Register rbx_method = rbx_temp;
__ movptr(rbx_method, vtable_entry_addr);
__ verify_oop(rbx_method);
__ jmp(rbx_method_fie);
}
break;
case _invokeinterface_mh:
{
// same as TemplateTable::invokeinterface,
// minus the CP setup and profiling:
// pick out the interface and itable index from the MH.
__ load_method_handle_vmslots(rax_argslot, rcx_recv, rdx_temp);
Register rdx_intf = rdx_temp;
Register rbx_index = rbx_temp;
__ load_heap_oop(rdx_intf, rcx_mh_vmtarget);
__ movl(rbx_index, rcx_dmh_vmindex);
__ movptr(rcx_recv, __ argument_address(rax_argslot, -1));
__ null_check(rcx_recv, oopDesc::klass_offset_in_bytes());
// get receiver klass
Register rax_klass = rax_argslot;
__ load_klass(rax_klass, rcx_recv);
__ verify_oop(rax_klass);
Register rdi_temp = rdi;
Register rbx_method = rbx_index;
// get interface klass
Label no_such_interface;
__ verify_oop(rdx_intf);
__ lookup_interface_method(rax_klass, rdx_intf,
// note: next two args must be the same:
rbx_index, rbx_method,
rdi_temp,
no_such_interface);
__ verify_oop(rbx_method);
__ jmp(rbx_method_fie);
__ hlt();
__ bind(no_such_interface);
// Throw an exception.
// For historical reasons, it will be IncompatibleClassChangeError.
__ mov(rbx_temp, rcx_recv); // rarg2_required might be RCX
assert_different_registers(rarg2_required, rbx_temp);
__ movptr(rarg2_required, Address(rdx_intf, java_mirror_offset)); // required interface
__ mov( rarg1_actual, rbx_temp); // bad receiver
__ movl( rarg0_code, (int) Bytecodes::_invokeinterface); // who is complaining?
__ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
}
break;
case _bound_ref_mh:
case _bound_int_mh:
case _bound_long_mh:
case _bound_ref_direct_mh:
case _bound_int_direct_mh:
case _bound_long_direct_mh:
{
bool direct_to_method = (ek >= _bound_ref_direct_mh);
BasicType arg_type = T_ILLEGAL;
int arg_mask = _INSERT_NO_MASK;
int arg_slots = -1;
get_ek_bound_mh_info(ek, arg_type, arg_mask, arg_slots);
// make room for the new argument:
__ movl(rax_argslot, rcx_bmh_vmargslot);
__ lea(rax_argslot, __ argument_address(rax_argslot));
insert_arg_slots(_masm, arg_slots * stack_move_unit(), arg_mask, rax_argslot, rbx_temp, rdx_temp);
// store bound argument into the new stack slot:
__ load_heap_oop(rbx_temp, rcx_bmh_argument);
if (arg_type == T_OBJECT) {
__ movptr(Address(rax_argslot, 0), rbx_temp);
} else {
Address prim_value_addr(rbx_temp, java_lang_boxing_object::value_offset_in_bytes(arg_type));
const int arg_size = type2aelembytes(arg_type);
__ load_sized_value(rdx_temp, prim_value_addr, arg_size, is_signed_subword_type(arg_type), rbx_temp);
__ store_sized_value(Address(rax_argslot, 0), rdx_temp, arg_size, rbx_temp);
}
if (direct_to_method) {
Register rbx_method = rbx_temp;
__ load_heap_oop(rbx_method, rcx_mh_vmtarget);
__ verify_oop(rbx_method);
__ jmp(rbx_method_fie);
} else {
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ verify_oop(rcx_recv);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
}
break;
case _adapter_retype_only:
case _adapter_retype_raw:
// immediately jump to the next MH layer:
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ verify_oop(rcx_recv);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
// This is OK when all parameter types widen.
// It is also OK when a return type narrows.
break;
case _adapter_check_cast:
{
// temps:
Register rbx_klass = rbx_temp; // interesting AMH data
// check a reference argument before jumping to the next layer of MH:
__ movl(rax_argslot, rcx_amh_vmargslot);
vmarg = __ argument_address(rax_argslot);
// What class are we casting to?
__ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
__ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
Label done;
__ movptr(rdx_temp, vmarg);
__ testptr(rdx_temp, rdx_temp);
__ jcc(Assembler::zero, done); // no cast if null
__ load_klass(rdx_temp, rdx_temp);
// live at this point:
// - rbx_klass: klass required by the target method
// - rdx_temp: argument klass to test
// - rcx_recv: adapter method handle
__ check_klass_subtype(rdx_temp, rbx_klass, rax_argslot, done);
// If we get here, the type check failed!
// Call the wrong_method_type stub, passing the failing argument type in rax.
Register rax_mtype = rax_argslot;
__ movl(rax_argslot, rcx_amh_vmargslot); // reload argslot field
__ movptr(rdx_temp, vmarg);
assert_different_registers(rarg2_required, rdx_temp);
__ load_heap_oop(rarg2_required, rcx_amh_argument); // required class
__ mov( rarg1_actual, rdx_temp); // bad object
__ movl( rarg0_code, (int) Bytecodes::_checkcast); // who is complaining?
__ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
__ bind(done);
// get the new MH:
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
break;
case _adapter_prim_to_prim:
case _adapter_ref_to_prim:
// handled completely by optimized cases
__ stop("init_AdapterMethodHandle should not issue this");
break;
case _adapter_opt_i2i: // optimized subcase of adapt_prim_to_prim
//case _adapter_opt_f2i: // optimized subcase of adapt_prim_to_prim
case _adapter_opt_l2i: // optimized subcase of adapt_prim_to_prim
case _adapter_opt_unboxi: // optimized subcase of adapt_ref_to_prim
{
// perform an in-place conversion to int or an int subword
__ movl(rax_argslot, rcx_amh_vmargslot);
vmarg = __ argument_address(rax_argslot);
switch (ek) {
case _adapter_opt_i2i:
__ movl(rdx_temp, vmarg);
break;
case _adapter_opt_l2i:
{
// just delete the extra slot; on a little-endian machine we keep the first
__ lea(rax_argslot, __ argument_address(rax_argslot, 1));
remove_arg_slots(_masm, -stack_move_unit(),
rax_argslot, rbx_temp, rdx_temp);
vmarg = Address(rax_argslot, -Interpreter::stackElementSize);
__ movl(rdx_temp, vmarg);
}
break;
case _adapter_opt_unboxi:
{
// Load the value up from the heap.
__ movptr(rdx_temp, vmarg);
int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_INT);
#ifdef ASSERT
for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
if (is_subword_type(BasicType(bt)))
assert(value_offset == java_lang_boxing_object::value_offset_in_bytes(BasicType(bt)), "");
}
#endif
__ null_check(rdx_temp, value_offset);
__ movl(rdx_temp, Address(rdx_temp, value_offset));
// We load this as a word. Because we are little-endian,
// the low bits will be correct, but the high bits may need cleaning.
// The vminfo will guide us to clean those bits.
}
break;
default:
ShouldNotReachHere();
}
// Do the requested conversion and store the value.
Register rbx_vminfo = rbx_temp;
__ movl(rbx_vminfo, rcx_amh_conversion);
assert(CONV_VMINFO_SHIFT == 0, "preshifted");
// get the new MH:
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
// (now we are done with the old MH)
// original 32-bit vmdata word must be of this form:
// | MBZ:6 | signBitCount:8 | srcDstTypes:8 | conversionOp:8 |
__ xchgptr(rcx, rbx_vminfo); // free rcx for shifts
__ shll(rdx_temp /*, rcx*/);
Label zero_extend, done;
__ testl(rcx, CONV_VMINFO_SIGN_FLAG);
__ jccb(Assembler::zero, zero_extend);
// this path is taken for int->byte, int->short
__ sarl(rdx_temp /*, rcx*/);
__ jmpb(done);
__ bind(zero_extend);
// this is taken for int->char
__ shrl(rdx_temp /*, rcx*/);
__ bind(done);
__ movl(vmarg, rdx_temp); // Store the value.
__ xchgptr(rcx, rbx_vminfo); // restore rcx_recv
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
break;
case _adapter_opt_i2l: // optimized subcase of adapt_prim_to_prim
case _adapter_opt_unboxl: // optimized subcase of adapt_ref_to_prim
{
// perform an in-place int-to-long or ref-to-long conversion
__ movl(rax_argslot, rcx_amh_vmargslot);
// on a little-endian machine we keep the first slot and add another after
__ lea(rax_argslot, __ argument_address(rax_argslot, 1));
insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
rax_argslot, rbx_temp, rdx_temp);
Address vmarg1(rax_argslot, -Interpreter::stackElementSize);
Address vmarg2 = vmarg1.plus_disp(Interpreter::stackElementSize);
switch (ek) {
case _adapter_opt_i2l:
{
#ifdef _LP64
__ movslq(rdx_temp, vmarg1); // Load sign-extended
__ movq(vmarg1, rdx_temp); // Store into first slot
#else
__ movl(rdx_temp, vmarg1);
__ sarl(rdx_temp, BitsPerInt - 1); // __ extend_sign()
__ movl(vmarg2, rdx_temp); // store second word
#endif
}
break;
case _adapter_opt_unboxl:
{
// Load the value up from the heap.
__ movptr(rdx_temp, vmarg1);
int value_offset = java_lang_boxing_object::value_offset_in_bytes(T_LONG);
assert(value_offset == java_lang_boxing_object::value_offset_in_bytes(T_DOUBLE), "");
__ null_check(rdx_temp, value_offset);
#ifdef _LP64
__ movq(rbx_temp, Address(rdx_temp, value_offset));
__ movq(vmarg1, rbx_temp);
#else
__ movl(rbx_temp, Address(rdx_temp, value_offset + 0*BytesPerInt));
__ movl(rdx_temp, Address(rdx_temp, value_offset + 1*BytesPerInt));
__ movl(vmarg1, rbx_temp);
__ movl(vmarg2, rdx_temp);
#endif
}
break;
default:
ShouldNotReachHere();
}
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
break;
case _adapter_opt_f2d: // optimized subcase of adapt_prim_to_prim
case _adapter_opt_d2f: // optimized subcase of adapt_prim_to_prim
{
// perform an in-place floating primitive conversion
__ movl(rax_argslot, rcx_amh_vmargslot);
__ lea(rax_argslot, __ argument_address(rax_argslot, 1));
if (ek == _adapter_opt_f2d) {
insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
rax_argslot, rbx_temp, rdx_temp);
}
Address vmarg(rax_argslot, -Interpreter::stackElementSize);
#ifdef _LP64
if (ek == _adapter_opt_f2d) {
__ movflt(xmm0, vmarg);
__ cvtss2sd(xmm0, xmm0);
__ movdbl(vmarg, xmm0);
} else {
__ movdbl(xmm0, vmarg);
__ cvtsd2ss(xmm0, xmm0);
__ movflt(vmarg, xmm0);
}
#else //_LP64
if (ek == _adapter_opt_f2d) {
__ fld_s(vmarg); // load float to ST0
__ fstp_s(vmarg); // store single
} else {
__ fld_d(vmarg); // load double to ST0
__ fstp_s(vmarg); // store single
}
#endif //_LP64
if (ek == _adapter_opt_d2f) {
remove_arg_slots(_masm, -stack_move_unit(),
rax_argslot, rbx_temp, rdx_temp);
}
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
break;
case _adapter_prim_to_ref:
__ unimplemented(entry_name(ek)); // %%% FIXME: NYI
break;
case _adapter_swap_args:
case _adapter_rot_args:
// handled completely by optimized cases
__ stop("init_AdapterMethodHandle should not issue this");
break;
case _adapter_opt_swap_1:
case _adapter_opt_swap_2:
case _adapter_opt_rot_1_up:
case _adapter_opt_rot_1_down:
case _adapter_opt_rot_2_up:
case _adapter_opt_rot_2_down:
{
int swap_bytes = 0, rotate = 0;
get_ek_adapter_opt_swap_rot_info(ek, swap_bytes, rotate);
// 'argslot' is the position of the first argument to swap
__ movl(rax_argslot, rcx_amh_vmargslot);
__ lea(rax_argslot, __ argument_address(rax_argslot));
// 'vminfo' is the second
Register rbx_destslot = rbx_temp;
__ movl(rbx_destslot, rcx_amh_conversion);
assert(CONV_VMINFO_SHIFT == 0, "preshifted");
__ andl(rbx_destslot, CONV_VMINFO_MASK);
__ lea(rbx_destslot, __ argument_address(rbx_destslot));
DEBUG_ONLY(verify_argslot(_masm, rbx_destslot, "swap point must fall within current frame"));
if (!rotate) {
for (int i = 0; i < swap_bytes; i += wordSize) {
__ movptr(rdx_temp, Address(rax_argslot , i));
__ push(rdx_temp);
__ movptr(rdx_temp, Address(rbx_destslot, i));
__ movptr(Address(rax_argslot, i), rdx_temp);
__ pop(rdx_temp);
__ movptr(Address(rbx_destslot, i), rdx_temp);
}
} else {
// push the first chunk, which is going to get overwritten
for (int i = swap_bytes; (i -= wordSize) >= 0; ) {
__ movptr(rdx_temp, Address(rax_argslot, i));
__ push(rdx_temp);
}
if (rotate > 0) {
// rotate upward
__ subptr(rax_argslot, swap_bytes);
#ifdef ASSERT
{
// Verify that argslot > destslot, by at least swap_bytes.
Label L_ok;
__ cmpptr(rax_argslot, rbx_destslot);
__ jccb(Assembler::aboveEqual, L_ok);
__ stop("source must be above destination (upward rotation)");
__ bind(L_ok);
}
#endif
// work argslot down to destslot, copying contiguous data upwards
// pseudo-code:
// rax = src_addr - swap_bytes
// rbx = dest_addr
// while (rax >= rbx) *(rax + swap_bytes) = *(rax + 0), rax--;
Label loop;
__ bind(loop);
__ movptr(rdx_temp, Address(rax_argslot, 0));
__ movptr(Address(rax_argslot, swap_bytes), rdx_temp);
__ addptr(rax_argslot, -wordSize);
__ cmpptr(rax_argslot, rbx_destslot);
__ jccb(Assembler::aboveEqual, loop);
} else {
__ addptr(rax_argslot, swap_bytes);
#ifdef ASSERT
{
// Verify that argslot < destslot, by at least swap_bytes.
Label L_ok;
__ cmpptr(rax_argslot, rbx_destslot);
__ jccb(Assembler::belowEqual, L_ok);
__ stop("source must be below destination (downward rotation)");
__ bind(L_ok);
}
#endif
// work argslot up to destslot, copying contiguous data downwards
// pseudo-code:
// rax = src_addr + swap_bytes
// rbx = dest_addr
// while (rax <= rbx) *(rax - swap_bytes) = *(rax + 0), rax++;
Label loop;
__ bind(loop);
__ movptr(rdx_temp, Address(rax_argslot, 0));
__ movptr(Address(rax_argslot, -swap_bytes), rdx_temp);
__ addptr(rax_argslot, wordSize);
__ cmpptr(rax_argslot, rbx_destslot);
__ jccb(Assembler::belowEqual, loop);
}
// pop the original first chunk into the destination slot, now free
for (int i = 0; i < swap_bytes; i += wordSize) {
__ pop(rdx_temp);
__ movptr(Address(rbx_destslot, i), rdx_temp);
}
}
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
break;
case _adapter_dup_args:
{
// 'argslot' is the position of the first argument to duplicate
__ movl(rax_argslot, rcx_amh_vmargslot);
__ lea(rax_argslot, __ argument_address(rax_argslot));
// 'stack_move' is negative number of words to duplicate
Register rdx_stack_move = rdx_temp;
__ movl2ptr(rdx_stack_move, rcx_amh_conversion);
__ sarptr(rdx_stack_move, CONV_STACK_MOVE_SHIFT);
int argslot0_num = 0;
Address argslot0 = __ argument_address(RegisterOrConstant(argslot0_num));
assert(argslot0.base() == rsp, "");
int pre_arg_size = argslot0.disp();
assert(pre_arg_size % wordSize == 0, "");
assert(pre_arg_size > 0, "must include PC");
// remember the old rsp+1 (argslot[0])
Register rbx_oldarg = rbx_temp;
__ lea(rbx_oldarg, argslot0);
// move rsp down to make room for dups
__ lea(rsp, Address(rsp, rdx_stack_move, Address::times_ptr));
// compute the new rsp+1 (argslot[0])
Register rdx_newarg = rdx_temp;
__ lea(rdx_newarg, argslot0);
__ push(rdi); // need a temp
// (preceding push must be done after arg addresses are taken!)
// pull down the pre_arg_size data (PC)
for (int i = -pre_arg_size; i < 0; i += wordSize) {
__ movptr(rdi, Address(rbx_oldarg, i));
__ movptr(Address(rdx_newarg, i), rdi);
}
// copy from rax_argslot[0...] down to new_rsp[1...]
// pseudo-code:
// rbx = old_rsp+1
// rdx = new_rsp+1
// rax = argslot
// while (rdx < rbx) *rdx++ = *rax++
Label loop;
__ bind(loop);
__ movptr(rdi, Address(rax_argslot, 0));
__ movptr(Address(rdx_newarg, 0), rdi);
__ addptr(rax_argslot, wordSize);
__ addptr(rdx_newarg, wordSize);
__ cmpptr(rdx_newarg, rbx_oldarg);
__ jccb(Assembler::less, loop);
__ pop(rdi); // restore temp
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
break;
case _adapter_drop_args:
{
// 'argslot' is the position of the first argument to nuke
__ movl(rax_argslot, rcx_amh_vmargslot);
__ lea(rax_argslot, __ argument_address(rax_argslot));
__ push(rdi); // need a temp
// (must do previous push after argslot address is taken)
// 'stack_move' is number of words to drop
Register rdi_stack_move = rdi;
__ movl2ptr(rdi_stack_move, rcx_amh_conversion);
__ sarptr(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
remove_arg_slots(_masm, rdi_stack_move,
rax_argslot, rbx_temp, rdx_temp);
__ pop(rdi); // restore temp
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
}
break;
case _adapter_collect_args:
__ unimplemented(entry_name(ek)); // %%% FIXME: NYI
break;
case _adapter_spread_args:
// handled completely by optimized cases
__ stop("init_AdapterMethodHandle should not issue this");
break;
case _adapter_opt_spread_0:
case _adapter_opt_spread_1:
case _adapter_opt_spread_more:
{
// spread an array out into a group of arguments
int length_constant = get_ek_adapter_opt_spread_info(ek);
// find the address of the array argument
__ movl(rax_argslot, rcx_amh_vmargslot);
__ lea(rax_argslot, __ argument_address(rax_argslot));
// grab some temps
{ __ push(rsi); __ push(rdi); }
// (preceding pushes must be done after argslot address is taken!)
#define UNPUSH_RSI_RDI \
{ __ pop(rdi); __ pop(rsi); }
// arx_argslot points both to the array and to the first output arg
vmarg = Address(rax_argslot, 0);
// Get the array value.
Register rsi_array = rsi;
Register rdx_array_klass = rdx_temp;
BasicType elem_type = T_OBJECT;
int length_offset = arrayOopDesc::length_offset_in_bytes();
int elem0_offset = arrayOopDesc::base_offset_in_bytes(elem_type);
__ movptr(rsi_array, vmarg);
Label skip_array_check;
if (length_constant == 0) {
__ testptr(rsi_array, rsi_array);
__ jcc(Assembler::zero, skip_array_check);
}
__ null_check(rsi_array, oopDesc::klass_offset_in_bytes());
__ load_klass(rdx_array_klass, rsi_array);
// Check the array type.
Register rbx_klass = rbx_temp;
__ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
__ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
Label ok_array_klass, bad_array_klass, bad_array_length;
__ check_klass_subtype(rdx_array_klass, rbx_klass, rdi, ok_array_klass);
// If we get here, the type check failed!
__ jmp(bad_array_klass);
__ bind(ok_array_klass);
// Check length.
if (length_constant >= 0) {
__ cmpl(Address(rsi_array, length_offset), length_constant);
} else {
Register rbx_vminfo = rbx_temp;
__ movl(rbx_vminfo, rcx_amh_conversion);
assert(CONV_VMINFO_SHIFT == 0, "preshifted");
__ andl(rbx_vminfo, CONV_VMINFO_MASK);
__ cmpl(rbx_vminfo, Address(rsi_array, length_offset));
}
__ jcc(Assembler::notEqual, bad_array_length);
Register rdx_argslot_limit = rdx_temp;
// Array length checks out. Now insert any required stack slots.
if (length_constant == -1) {
// Form a pointer to the end of the affected region.
__ lea(rdx_argslot_limit, Address(rax_argslot, Interpreter::stackElementSize));
// 'stack_move' is negative number of words to insert
Register rdi_stack_move = rdi;
__ movl2ptr(rdi_stack_move, rcx_amh_conversion);
__ sarptr(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
Register rsi_temp = rsi_array; // spill this
insert_arg_slots(_masm, rdi_stack_move, -1,
rax_argslot, rbx_temp, rsi_temp);
// reload the array (since rsi was killed)
__ movptr(rsi_array, vmarg);
} else if (length_constant > 1) {
int arg_mask = 0;
int new_slots = (length_constant - 1);
for (int i = 0; i < new_slots; i++) {
arg_mask <<= 1;
arg_mask |= _INSERT_REF_MASK;
}
insert_arg_slots(_masm, new_slots * stack_move_unit(), arg_mask,
rax_argslot, rbx_temp, rdx_temp);
} else if (length_constant == 1) {
// no stack resizing required
} else if (length_constant == 0) {
remove_arg_slots(_masm, -stack_move_unit(),
rax_argslot, rbx_temp, rdx_temp);
}
// Copy from the array to the new slots.
// Note: Stack change code preserves integrity of rax_argslot pointer.
// So even after slot insertions, rax_argslot still points to first argument.
if (length_constant == -1) {
// [rax_argslot, rdx_argslot_limit) is the area we are inserting into.
Register rsi_source = rsi_array;
__ lea(rsi_source, Address(rsi_array, elem0_offset));
Label loop;
__ bind(loop);
__ movptr(rbx_temp, Address(rsi_source, 0));
__ movptr(Address(rax_argslot, 0), rbx_temp);
__ addptr(rsi_source, type2aelembytes(elem_type));
__ addptr(rax_argslot, Interpreter::stackElementSize);
__ cmpptr(rax_argslot, rdx_argslot_limit);
__ jccb(Assembler::less, loop);
} else if (length_constant == 0) {
__ bind(skip_array_check);
// nothing to copy
} else {
int elem_offset = elem0_offset;
int slot_offset = 0;
for (int index = 0; index < length_constant; index++) {
__ movptr(rbx_temp, Address(rsi_array, elem_offset));
__ movptr(Address(rax_argslot, slot_offset), rbx_temp);
elem_offset += type2aelembytes(elem_type);
slot_offset += Interpreter::stackElementSize;
}
}
// Arguments are spread. Move to next method handle.
UNPUSH_RSI_RDI;
__ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
__ jump_to_method_handle_entry(rcx_recv, rdx_temp);
__ bind(bad_array_klass);
UNPUSH_RSI_RDI;
assert(!vmarg.uses(rarg2_required), "must be different registers");
__ movptr(rarg2_required, Address(rdx_array_klass, java_mirror_offset)); // required type
__ movptr(rarg1_actual, vmarg); // bad array
__ movl( rarg0_code, (int) Bytecodes::_aaload); // who is complaining?
__ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
__ bind(bad_array_length);
UNPUSH_RSI_RDI;
assert(!vmarg.uses(rarg2_required), "must be different registers");
__ mov (rarg2_required, rcx_recv); // AMH requiring a certain length
__ movptr(rarg1_actual, vmarg); // bad array
__ movl( rarg0_code, (int) Bytecodes::_arraylength); // who is complaining?
__ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
#undef UNPUSH_RSI_RDI
}
break;
case _adapter_flyby:
case _adapter_ricochet:
__ unimplemented(entry_name(ek)); // %%% FIXME: NYI
break;
default: ShouldNotReachHere();
}
__ hlt();
address me_cookie = MethodHandleEntry::start_compiled_entry(_masm, interp_entry);
__ unimplemented(entry_name(ek)); // %%% FIXME: NYI
init_entry(ek, MethodHandleEntry::finish_compiled_entry(_masm, me_cookie));
}
address TemplateInterpreterGenerator::generate_slow_signature_handler() {
address entry = __ pc();
// rbx: method
// r14: pointer to locals
// c_rarg3: first stack arg - wordSize
__ mov(c_rarg3, rsp);
// adjust rsp
__ subptr(rsp, 4 * wordSize);
__ call_VM(noreg,
CAST_FROM_FN_PTR(address,
InterpreterRuntime::slow_signature_handler),
rbx, r14, c_rarg3);
// rax: result handler
// Stack layout:
// rsp: 3 integer or float args (if static first is unused)
// 1 float/double identifiers
// return address
// stack args
// garbage
// expression stack bottom
// bcp (NULL)
// ...
// Do FP first so we can use c_rarg3 as temp
__ 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);
// Do Int register here
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);
// Do Float Here
__ movflt(floatreg, Address(rsp, i * wordSize));
__ jmp(next);
// Do Double here
__ bind(isdouble);
__ movdbl(floatreg, Address(rsp, i * wordSize));
__ bind(next);
}
// restore rsp
__ addptr(rsp, 4 * wordSize);
__ ret(0);
return entry;
}
// used by compiler only; may use only caller saved registers eax, ebx, ecx.
// edx holds first int arg, esi, edi, ebp are callee-save & must be preserved.
// Leave reciever in ecx; required behavior when +OptoArgsInRegisters
// is modifed to put first oop in ecx.
//
// NOTE: If this code is used by the C1, the receiver_location is always 0.
VtableStub* VtableStubs::create_vtable_stub(int vtable_index, int receiver_location) {
const int i486_code_length = VtableStub::pd_code_size_limit(true);
VtableStub* s = new(i486_code_length) VtableStub(true, vtable_index, receiver_location);
ResourceMark rm;
MacroAssembler* masm = new MacroAssembler(new CodeBuffer(s->entry_point(), i486_code_length));
#ifndef PRODUCT
#ifdef COMPILER2
if (CountCompiledCalls) __ incl(Address((int)OptoRuntime::nof_megamorphic_calls_addr(), relocInfo::none));
#endif
#endif
// get receiver (need to skip return address on top of stack)
#ifdef COMPILER1
assert(receiver_location == 0, "receiver is always in ecx - no location info needed");
#else
if( receiver_location < SharedInfo::stack0 ) {
assert(receiver_location == ECX_num, "receiver expected in ecx");
} else {
__ movl(ecx, Address(esp, SharedInfo::reg2stack(OptoReg::Name(receiver_location)) * wordSize+wordSize/*skip return address*/));
}
#endif
// get receiver klass
address npe_addr = __ pc();
__ movl(eax, Address(ecx, oopDesc::klass_offset_in_bytes()));
// compute entry offset (in words)
int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
#ifndef PRODUCT
if (DebugVtables) {
Label L;
// check offset vs vtable length
__ cmpl(Address(eax, instanceKlass::vtable_length_offset()*wordSize), vtable_index*vtableEntry::size());
__ jcc(Assembler::greater, L);
__ movl(ebx, vtable_index);
__ call_VM(noreg, CAST_FROM_FN_PTR(address, bad_compiled_vtable_index), ecx, ebx);
__ bind(L);
}
#endif // PRODUCT
// load methodOop and target address
#ifdef COMPILER1
__ movl(ebx, Address(eax, entry_offset*wordSize + vtableEntry::method_offset_in_bytes()));
address ame_addr = __ pc();
__ movl(edx, Address(ebx, methodOopDesc::from_compiled_code_entry_point_offset()));
if (DebugVtables) {
Label L;
__ testl(edx, edx);
__ jcc(Assembler::notZero, L);
__ stop("Vtable entry is NULL");
__ bind(L);
}
// eax: receiver klass
// ebx: methodOop
// ecx: receiver
// edx: entry point
__ jmp(edx);
#else
__ movl(eax, Address(eax, entry_offset*wordSize + vtableEntry::method_offset_in_bytes()));
address ame_addr = __ pc();
__ movl(ebx, Address(eax, methodOopDesc::from_compiled_code_entry_point_offset()));
if (DebugVtables) {
Label L;
__ testl(ebx, ebx);
__ jcc(Assembler::notZero, L);
__ stop("Vtable entry is NULL");
__ bind(L);
}
// jump to target (either compiled code or c2iadapter)
// eax: methodOop (in case we call c2iadapter)
__ jmp(ebx);
#endif // COMPILER1
masm->flush();
s->set_exception_points(npe_addr, ame_addr);
return s;
}