void CppInterpreterGenerator::generate_adjust_callers_stack()
{
StackFrame frame;
const int frame_header_size = frame.unaligned_size() + slop_factor;
const Address param_words_addr(
Rmethod, methodOopDesc::size_of_parameters_offset());
const Address local_words_addr(
Rmethod, methodOopDesc::size_of_locals_offset());
const Register param_words = r3;
const Register local_words = r4;
Label loop, done;
// Check whether extra locals are actually required
__ lhz (param_words, param_words_addr);
__ lhz (local_words, local_words_addr);
__ compare (param_words, local_words);
__ beq (done);
// Extend the frame if necessary
const Register required_bytes = r5;
const Register available_bytes = r6;
__ shift_left (required_bytes, local_words, LogBytesPerWord);
__ sub (available_bytes, Rlocals, r1);
__ subi (available_bytes, available_bytes, frame_header_size);
__ maybe_extend_frame (required_bytes, available_bytes);
// Zero the extra locals
const Register dst = r7;
__ shift_left (dst, param_words, LogBytesPerWord);
__ sub (dst, Rlocals, dst);
__ sub (r0, local_words, param_words);
__ mtctr (r0);
__ load (r0, 0);
__ bind (loop);
__ store (r0, Address(dst, 0));
__ subi (dst, dst, wordSize);
__ bdnz (loop);
__ bind (done);
}
address generate_call_stub(address& return_address)
{
assert (!TaggedStackInterpreter, "not supported");
StubCodeMark mark(this, "StubRoutines", "call_stub");
address start = __ enter();
const Register call_wrapper = r3;
const Register result = r4;
const Register result_type = r5;
const Register method = r6;
const Register entry_point = r7;
const Register parameters = r8;
const Register parameter_words = r9;
const Register thread = r10;
#ifdef ASSERT
// Make sure we have no pending exceptions
{
StackFrame frame;
Label label;
__ load (r0, Address(thread, Thread::pending_exception_offset()));
__ compare (r0, 0);
__ beq (label);
__ prolog (frame);
__ should_not_reach_here (__FILE__, __LINE__);
__ epilog (frame);
__ blr ();
__ bind (label);
}
#endif // ASSERT
// Calculate the frame size
StackFrame frame;
for (int i = 0; i < StackFrame::max_crfs; i++)
frame.get_cr_field();
for (int i = 0; i < StackFrame::max_gprs; i++)
frame.get_register();
StubRoutines::set_call_stub_base_size(frame.unaligned_size() + 3*wordSize);
// the 3 extra words are for call_wrapper, result and result_type
const Register parameter_bytes = parameter_words;
__ shift_left (parameter_bytes, parameter_words, LogBytesPerWord);
const Register frame_size = r11;
const Register padding = r12;
__ addi (frame_size, parameter_bytes, StubRoutines::call_stub_base_size());
__ calc_padding_for_alignment (padding, frame_size, StackAlignmentInBytes);
__ add (frame_size, frame_size, padding);
// Save the link register and create the new frame
__ mflr (r0);
__ store (r0, Address(r1, StackFrame::lr_save_offset * wordSize));
__ neg (r0, frame_size);
__ store_update_indexed (r1, r1, r0);
#ifdef PPC64
__ mfcr (r0);
__ store (r0, Address(r1, StackFrame::cr_save_offset * wordSize));
#endif // PPC64
// Calculate the address of the interpreter's local variables
const Register locals = frame_size;
__ addi (locals, r1, frame.start_of_locals() - wordSize);
__ add (locals, locals, padding);
__ add (locals, locals, parameter_bytes);
// Store the call wrapper address and the result stuff
const int initial_offset = 1;
int offset = initial_offset;
__ store (call_wrapper, Address(locals, offset++ * wordSize));
__ store (result, Address(locals, offset++ * wordSize));
__ store (result_type, Address(locals, offset++ * wordSize));
// Store the registers
#ifdef PPC32
__ mfcr (r0);
__ store (r0, Address(locals, offset++ * wordSize));
#endif // PPC32
for (int i = 14; i < 32; i++) {
__ store (as_Register(i), Address(locals, offset++ * wordSize));
}
const int final_offset = offset;
// Store the location of call_wrapper
frame::set_call_wrapper_offset((final_offset - initial_offset) * wordSize);
#ifdef ASSERT
// Check that we wrote all the way to the end of the frame.
// The frame may have been resized when we return from the
// interpreter, so the start of the frame may have moved
// but the end will be where we left it and we rely on this
// to find our stuff.
{
StackFrame frame;
Label label;
__ load (r3, Address(r1, 0));
__ subi (r3, r3, final_offset * wordSize);
__ compare (r3, locals);
__ beq (label);
__ prolog (frame);
__ should_not_reach_here (__FILE__, __LINE__);
__ epilog (frame);
__ blr ();
__ bind (label);
}
#endif // ASSERT
// Pass parameters if any
{
Label loop, done;
__ compare (parameter_bytes, 0);
__ ble (done);
const Register src = parameters;
const Register dst = padding;
__ mr (dst, locals);
__ shift_right (r0, parameter_bytes, LogBytesPerWord);
__ mtctr (r0);
__ bind (loop);
__ load (r0, Address(src, 0));
__ store (r0, Address(dst, 0));
__ addi (src, src, wordSize);
__ subi (dst, dst, wordSize);
__ bdnz (loop);
__ bind (done);
}
// Make the call
__ mr (Rmethod, method);
__ mr (Rlocals, locals);
__ mr (Rthread, thread);
__ mtctr (entry_point);
__ bctrl();
// This is used to identify call_stub stack frames
return_address = __ pc();
// Figure out where our stuff is stored
__ load (locals, Address(r1, 0));
__ subi (locals, locals, final_offset * wordSize);
#ifdef ASSERT
// Rlocals should contain the address we just calculated.
{
StackFrame frame;
Label label;
__ compare (Rlocals, locals);
__ beq (label);
__ prolog (frame);
__ should_not_reach_here (__FILE__, __LINE__);
__ epilog (frame);
__ blr ();
__ bind (label);
}
#endif // ASSERT
// Is an exception being thrown?
Label exit;
__ load (r0, Address(Rthread, Thread::pending_exception_offset()));
__ compare (r0, 0);
__ bne (exit);
// Store result depending on type
const Register result_addr = r6;
Label is_int, is_long, is_object;
offset = initial_offset + 1; // skip call_wrapper
__ load (result_addr, Address(locals, offset++ * wordSize));
__ load (result_type, Address(locals, offset++ * wordSize));
__ compare (result_type, T_INT);
__ beq (is_int);
__ compare (result_type, T_LONG);
__ beq (is_long);
__ compare (result_type, T_OBJECT);
__ beq (is_object);
__ should_not_reach_here (__FILE__, __LINE__);
__ bind (is_int);
__ stw (r3, Address(result_addr, 0));
__ b (exit);
__ bind (is_long);
#ifdef PPC32
__ store (r4, Address(result_addr, wordSize));
#endif
__ store (r3, Address(result_addr, 0));
__ b (exit);
__ bind (is_object);
__ store (r3, Address(result_addr, 0));
//__ b (exit);
// Restore the registers
__ bind (exit);
#ifdef PPC32
__ load (r0, Address(locals, offset++ * wordSize));
__ mtcr (r0);
#endif // PPC32
for (int i = 14; i < 32; i++) {
__ load (as_Register(i), Address(locals, offset++ * wordSize));
}
#ifdef PPC64
__ load (r0, Address(r1, StackFrame::cr_save_offset * wordSize));
__ mtcr (r0);
#endif // PPC64
assert (offset == final_offset, "save and restore must match");
// Unwind and return
__ load (r1, Address(r1, StackFrame::back_chain_offset * wordSize));
__ load (r0, Address(r1, StackFrame::lr_save_offset * wordSize));
__ mtlr (r0);
__ blr ();
return start;
}
void C1_MacroAssembler::initialize_body(Register obj, Register tmp1, Register tmp2,
int obj_size_in_bytes, int hdr_size_in_bytes) {
const int index = (obj_size_in_bytes - hdr_size_in_bytes) / HeapWordSize;
const int cl_size = VM_Version::L1_data_cache_line_size(),
cl_dwords = cl_size>>3,
cl_dw_addr_bits = exact_log2(cl_dwords);
const Register tmp = R0,
base_ptr = tmp1,
cnt_dwords = tmp2;
if (index <= 6) {
// Use explicit NULL stores.
if (index > 0) { li(tmp, 0); }
for (int i = 0; i < index; ++i) { std(tmp, hdr_size_in_bytes + i * HeapWordSize, obj); }
} else if (index < (2<<cl_dw_addr_bits)-1) {
// simple loop
Label loop;
li(cnt_dwords, index);
addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
li(tmp, 0);
mtctr(cnt_dwords); // Load counter.
bind(loop);
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(loop);
} else {
// like clear_memory_doubleword
Label startloop, fast, fastloop, restloop, done;
addi(base_ptr, obj, hdr_size_in_bytes); // Compute address of first element.
load_const_optimized(cnt_dwords, index);
rldicl_(tmp, base_ptr, 64-3, 64-cl_dw_addr_bits); // Extract dword offset within first cache line.
beq(CCR0, fast); // Already 128byte aligned.
subfic(tmp, tmp, cl_dwords);
mtctr(tmp); // Set ctr to hit 128byte boundary (0<ctr<cl_dwords).
subf(cnt_dwords, tmp, cnt_dwords); // rest.
li(tmp, 0);
bind(startloop); // Clear at the beginning to reach 128byte boundary.
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(startloop);
bind(fast); // Clear 128byte blocks.
srdi(tmp, cnt_dwords, cl_dw_addr_bits); // Loop count for 128byte loop (>0).
andi(cnt_dwords, cnt_dwords, cl_dwords-1); // Rest in dwords.
mtctr(tmp); // Load counter.
bind(fastloop);
dcbz(base_ptr); // Clear 128byte aligned block.
addi(base_ptr, base_ptr, cl_size);
bdnz(fastloop);
cmpdi(CCR0, cnt_dwords, 0); // size 0?
beq(CCR0, done); // rest == 0
li(tmp, 0);
mtctr(cnt_dwords); // Load counter.
bind(restloop); // Clear rest.
std(tmp, 0, base_ptr); // Clear 8byte aligned block.
addi(base_ptr, base_ptr, 8);
bdnz(restloop);
bind(done);
}
}
