1 /*
2 * Copyright 2000-2008 Sun Microsystems, Inc. All Rights Reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
24
25 # include "incls/_precompiled.incl"
26 # include "incls/_c1_LIRAssembler_x86.cpp.incl"
27
28
29 // These masks are used to provide 128-bit aligned bitmasks to the XMM
30 // instructions, to allow sign-masking or sign-bit flipping. They allow
31 // fast versions of NegF/NegD and AbsF/AbsD.
32
33 // Note: 'double' and 'long long' have 32-bits alignment on x86.
34 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
35 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
36 // of 128-bits operands for SSE instructions.
37 jlong *operand = (jlong*)(((long)adr)&((long)(~0xF)));
38 // Store the value to a 128-bits operand.
39 operand[0] = lo;
40 operand[1] = hi;
41 return operand;
42 }
43
44 // Buffer for 128-bits masks used by SSE instructions.
45 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
46
47 // Static initialization during VM startup.
48 static jlong *float_signmask_pool = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
49 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
50 static jlong *float_signflip_pool = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
51 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
52
53
54
55 NEEDS_CLEANUP // remove this definitions ?
56 const Register IC_Klass = rax; // where the IC klass is cached
57 const Register SYNC_header = rax; // synchronization header
58 const Register SHIFT_count = rcx; // where count for shift operations must be
59
60 #define __ _masm->
61
62
63 static void select_different_registers(Register preserve,
64 Register extra,
65 Register &tmp1,
66 Register &tmp2) {
67 if (tmp1 == preserve) {
68 assert_different_registers(tmp1, tmp2, extra);
69 tmp1 = extra;
70 } else if (tmp2 == preserve) {
71 assert_different_registers(tmp1, tmp2, extra);
72 tmp2 = extra;
73 }
74 assert_different_registers(preserve, tmp1, tmp2);
75 }
76
77
78
79 static void select_different_registers(Register preserve,
80 Register extra,
81 Register &tmp1,
82 Register &tmp2,
83 Register &tmp3) {
84 if (tmp1 == preserve) {
85 assert_different_registers(tmp1, tmp2, tmp3, extra);
86 tmp1 = extra;
87 } else if (tmp2 == preserve) {
88 assert_different_registers(tmp1, tmp2, tmp3, extra);
89 tmp2 = extra;
90 } else if (tmp3 == preserve) {
91 assert_different_registers(tmp1, tmp2, tmp3, extra);
92 tmp3 = extra;
93 }
94 assert_different_registers(preserve, tmp1, tmp2, tmp3);
95 }
96
97
98
99 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
100 if (opr->is_constant()) {
101 LIR_Const* constant = opr->as_constant_ptr();
102 switch (constant->type()) {
103 case T_INT: {
104 return true;
105 }
106
107 default:
108 return false;
109 }
110 }
111 return false;
112 }
113
114
115 LIR_Opr LIR_Assembler::receiverOpr() {
116 return FrameMap::receiver_opr;
117 }
118
119 LIR_Opr LIR_Assembler::incomingReceiverOpr() {
120 return receiverOpr();
121 }
122
123 LIR_Opr LIR_Assembler::osrBufferPointer() {
124 return FrameMap::as_pointer_opr(receiverOpr()->as_register());
125 }
126
127 //--------------fpu register translations-----------------------
128
129
130 address LIR_Assembler::float_constant(float f) {
131 address const_addr = __ float_constant(f);
132 if (const_addr == NULL) {
133 bailout("const section overflow");
134 return __ code()->consts()->start();
135 } else {
136 return const_addr;
137 }
138 }
139
140
141 address LIR_Assembler::double_constant(double d) {
142 address const_addr = __ double_constant(d);
143 if (const_addr == NULL) {
144 bailout("const section overflow");
145 return __ code()->consts()->start();
146 } else {
147 return const_addr;
148 }
149 }
150
151
152 void LIR_Assembler::set_24bit_FPU() {
153 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
154 }
155
156 void LIR_Assembler::reset_FPU() {
157 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
158 }
159
160 void LIR_Assembler::fpop() {
161 __ fpop();
162 }
163
164 void LIR_Assembler::fxch(int i) {
165 __ fxch(i);
166 }
167
168 void LIR_Assembler::fld(int i) {
169 __ fld_s(i);
170 }
171
172 void LIR_Assembler::ffree(int i) {
173 __ ffree(i);
174 }
175
176 void LIR_Assembler::breakpoint() {
177 __ int3();
178 }
179
180 void LIR_Assembler::push(LIR_Opr opr) {
181 if (opr->is_single_cpu()) {
182 __ push_reg(opr->as_register());
183 } else if (opr->is_double_cpu()) {
184 NOT_LP64(__ push_reg(opr->as_register_hi()));
185 __ push_reg(opr->as_register_lo());
186 } else if (opr->is_stack()) {
187 __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
188 } else if (opr->is_constant()) {
189 LIR_Const* const_opr = opr->as_constant_ptr();
190 if (const_opr->type() == T_OBJECT) {
191 __ push_oop(const_opr->as_jobject());
192 } else if (const_opr->type() == T_INT) {
193 __ push_jint(const_opr->as_jint());
194 } else {
195 ShouldNotReachHere();
196 }
197
198 } else {
199 ShouldNotReachHere();
200 }
201 }
202
203 void LIR_Assembler::pop(LIR_Opr opr) {
204 if (opr->is_single_cpu()) {
205 __ pop_reg(opr->as_register());
206 } else {
207 ShouldNotReachHere();
208 }
209 }
210
211 bool LIR_Assembler::is_literal_address(LIR_Address* addr) {
212 return addr->base()->is_illegal() && addr->index()->is_illegal();
213 }
214
215 //-------------------------------------------
216
217 Address LIR_Assembler::as_Address(LIR_Address* addr) {
218 return as_Address(addr, rscratch1);
219 }
220
221 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
222 if (addr->base()->is_illegal()) {
223 assert(addr->index()->is_illegal(), "must be illegal too");
224 AddressLiteral laddr((address)addr->disp(), relocInfo::none);
225 if (! __ reachable(laddr)) {
226 __ movptr(tmp, laddr.addr());
227 Address res(tmp, 0);
228 return res;
229 } else {
230 return __ as_Address(laddr);
231 }
232 }
233
234 Register base = addr->base()->as_pointer_register();
235
236 if (addr->index()->is_illegal()) {
237 return Address( base, addr->disp());
238 } else if (addr->index()->is_cpu_register()) {
239 Register index = addr->index()->as_pointer_register();
240 return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
241 } else if (addr->index()->is_constant()) {
242 intptr_t addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
243 assert(Assembler::is_simm32(addr_offset), "must be");
244
245 return Address(base, addr_offset);
246 } else {
247 Unimplemented();
248 return Address();
249 }
250 }
251
252
253 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
254 Address base = as_Address(addr);
255 return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
256 }
257
258
259 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
260 return as_Address(addr);
261 }
262
263
264 void LIR_Assembler::osr_entry() {
265 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
266 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
267 ValueStack* entry_state = osr_entry->state();
268 int number_of_locks = entry_state->locks_size();
269
270 // we jump here if osr happens with the interpreter
271 // state set up to continue at the beginning of the
272 // loop that triggered osr - in particular, we have
273 // the following registers setup:
274 //
275 // rcx: osr buffer
276 //
277
278 // build frame
279 ciMethod* m = compilation()->method();
280 __ build_frame(initial_frame_size_in_bytes());
281
282 // OSR buffer is
283 //
284 // locals[nlocals-1..0]
285 // monitors[0..number_of_locks]
286 //
287 // locals is a direct copy of the interpreter frame so in the osr buffer
288 // so first slot in the local array is the last local from the interpreter
289 // and last slot is local[0] (receiver) from the interpreter
290 //
291 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
292 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
293 // in the interpreter frame (the method lock if a sync method)
294
295 // Initialize monitors in the compiled activation.
296 // rcx: pointer to osr buffer
297 //
298 // All other registers are dead at this point and the locals will be
299 // copied into place by code emitted in the IR.
300
301 Register OSR_buf = osrBufferPointer()->as_pointer_register();
302 { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
303 int monitor_offset = BytesPerWord * method()->max_locals() +
304 (BasicObjectLock::size() * BytesPerWord) * (number_of_locks - 1);
305 for (int i = 0; i < number_of_locks; i++) {
306 int slot_offset = monitor_offset - ((i * BasicObjectLock::size()) * BytesPerWord);
307 #ifdef ASSERT
308 // verify the interpreter's monitor has a non-null object
309 {
310 Label L;
311 __ cmpptr(Address(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);
312 __ jcc(Assembler::notZero, L);
313 __ stop("locked object is NULL");
314 __ bind(L);
315 }
316 #endif
317 __ movptr(rbx, Address(OSR_buf, slot_offset + BasicObjectLock::lock_offset_in_bytes()));
318 __ movptr(frame_map()->address_for_monitor_lock(i), rbx);
319 __ movptr(rbx, Address(OSR_buf, slot_offset + BasicObjectLock::obj_offset_in_bytes()));
320 __ movptr(frame_map()->address_for_monitor_object(i), rbx);
321 }
322 }
323 }
324
325
326 // inline cache check; done before the frame is built.
327 int LIR_Assembler::check_icache() {
328 Register receiver = FrameMap::receiver_opr->as_register();
329 Register ic_klass = IC_Klass;
330 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
331
332 if (!VerifyOops) {
333 // insert some nops so that the verified entry point is aligned on CodeEntryAlignment
334 while ((__ offset() + ic_cmp_size) % CodeEntryAlignment != 0) {
335 __ nop();
336 }
337 }
338 int offset = __ offset();
339 __ inline_cache_check(receiver, IC_Klass);
340 assert(__ offset() % CodeEntryAlignment == 0 || VerifyOops, "alignment must be correct");
341 if (VerifyOops) {
342 // force alignment after the cache check.
343 // It's been verified to be aligned if !VerifyOops
344 __ align(CodeEntryAlignment);
345 }
346 return offset;
347 }
348
349
350 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
351 jobject o = NULL;
352 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
353 __ movoop(reg, o);
354 patching_epilog(patch, lir_patch_normal, reg, info);
355 }
356
357
358 void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register new_hdr, int monitor_no, Register exception) {
359 if (exception->is_valid()) {
360 // preserve exception
361 // note: the monitor_exit runtime call is a leaf routine
362 // and cannot block => no GC can happen
363 // The slow case (MonitorAccessStub) uses the first two stack slots
364 // ([esp+0] and [esp+4]), therefore we store the exception at [esp+8]
365 __ movptr (Address(rsp, 2*wordSize), exception);
366 }
367
368 Register obj_reg = obj_opr->as_register();
369 Register lock_reg = lock_opr->as_register();
370
371 // setup registers (lock_reg must be rax, for lock_object)
372 assert(obj_reg != SYNC_header && lock_reg != SYNC_header, "rax, must be available here");
373 Register hdr = lock_reg;
374 assert(new_hdr == SYNC_header, "wrong register");
375 lock_reg = new_hdr;
376 // compute pointer to BasicLock
377 Address lock_addr = frame_map()->address_for_monitor_lock(monitor_no);
378 __ lea(lock_reg, lock_addr);
379 // unlock object
380 MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, true, monitor_no);
381 // _slow_case_stubs->append(slow_case);
382 // temporary fix: must be created after exceptionhandler, therefore as call stub
383 _slow_case_stubs->append(slow_case);
384 if (UseFastLocking) {
385 // try inlined fast unlocking first, revert to slow locking if it fails
386 // note: lock_reg points to the displaced header since the displaced header offset is 0!
387 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
388 __ unlock_object(hdr, obj_reg, lock_reg, *slow_case->entry());
389 } else {
390 // always do slow unlocking
391 // note: the slow unlocking code could be inlined here, however if we use
392 // slow unlocking, speed doesn't matter anyway and this solution is
393 // simpler and requires less duplicated code - additionally, the
394 // slow unlocking code is the same in either case which simplifies
395 // debugging
396 __ jmp(*slow_case->entry());
397 }
398 // done
399 __ bind(*slow_case->continuation());
400
401 if (exception->is_valid()) {
402 // restore exception
403 __ movptr (exception, Address(rsp, 2 * wordSize));
404 }
405 }
406
407 // This specifies the rsp decrement needed to build the frame
408 int LIR_Assembler::initial_frame_size_in_bytes() {
409 // if rounding, must let FrameMap know!
410
411 // The frame_map records size in slots (32bit word)
412
413 // subtract two words to account for return address and link
414 return (frame_map()->framesize() - (2*VMRegImpl::slots_per_word)) * VMRegImpl::stack_slot_size;
415 }
416
417
418 void LIR_Assembler::emit_exception_handler() {
419 // if the last instruction is a call (typically to do a throw which
420 // is coming at the end after block reordering) the return address
421 // must still point into the code area in order to avoid assertion
422 // failures when searching for the corresponding bci => add a nop
423 // (was bug 5/14/1999 - gri)
424
425 __ nop();
426
427 // generate code for exception handler
428 address handler_base = __ start_a_stub(exception_handler_size);
429 if (handler_base == NULL) {
430 // not enough space left for the handler
431 bailout("exception handler overflow");
432 return;
433 }
434 #ifdef ASSERT
435 int offset = code_offset();
436 #endif // ASSERT
437
438 compilation()->offsets()->set_value(CodeOffsets::Exceptions, code_offset());
439
440 // if the method does not have an exception handler, then there is
441 // no reason to search for one
442 if (compilation()->has_exception_handlers() || JvmtiExport::can_post_exceptions()) {
443 // the exception oop and pc are in rax, and rdx
444 // no other registers need to be preserved, so invalidate them
445 __ invalidate_registers(false, true, true, false, true, true);
446
447 // check that there is really an exception
448 __ verify_not_null_oop(rax);
449
450 // search an exception handler (rax: exception oop, rdx: throwing pc)
451 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_nofpu_id)));
452
453 // if the call returns here, then the exception handler for particular
454 // exception doesn't exist -> unwind activation and forward exception to caller
455 }
456
457 // the exception oop is in rax,
458 // no other registers need to be preserved, so invalidate them
459 __ invalidate_registers(false, true, true, true, true, true);
460
461 // check that there is really an exception
462 __ verify_not_null_oop(rax);
463
464 // unlock the receiver/klass if necessary
465 // rax,: exception
466 ciMethod* method = compilation()->method();
467 if (method->is_synchronized() && GenerateSynchronizationCode) {
468 monitorexit(FrameMap::rbx_oop_opr, FrameMap::rcx_opr, SYNC_header, 0, rax);
469 }
470
471 // unwind activation and forward exception to caller
472 // rax,: exception
473 __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
474
475 assert(code_offset() - offset <= exception_handler_size, "overflow");
476
477 __ end_a_stub();
478 }
479
480 void LIR_Assembler::emit_deopt_handler() {
481 // if the last instruction is a call (typically to do a throw which
482 // is coming at the end after block reordering) the return address
483 // must still point into the code area in order to avoid assertion
484 // failures when searching for the corresponding bci => add a nop
485 // (was bug 5/14/1999 - gri)
486
487 __ nop();
488
489 // generate code for exception handler
490 address handler_base = __ start_a_stub(deopt_handler_size);
491 if (handler_base == NULL) {
492 // not enough space left for the handler
493 bailout("deopt handler overflow");
494 return;
495 }
496 #ifdef ASSERT
497 int offset = code_offset();
498 #endif // ASSERT
499
500 compilation()->offsets()->set_value(CodeOffsets::Deopt, code_offset());
501
502 InternalAddress here(__ pc());
503 __ pushptr(here.addr());
504
505 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
506
507 assert(code_offset() - offset <= deopt_handler_size, "overflow");
508
509 __ end_a_stub();
510
511 }
512
513
514 // This is the fast version of java.lang.String.compare; it has not
515 // OSR-entry and therefore, we generate a slow version for OSR's
516 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
517 __ movptr (rbx, rcx); // receiver is in rcx
518 __ movptr (rax, arg1->as_register());
519
520 // Get addresses of first characters from both Strings
521 __ movptr (rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
522 __ movptr (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
523 __ lea (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
524
525
526 // rbx, may be NULL
527 add_debug_info_for_null_check_here(info);
528 __ movptr (rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
529 __ movptr (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
530 __ lea (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
531
532 // compute minimum length (in rax) and difference of lengths (on top of stack)
533 if (VM_Version::supports_cmov()) {
534 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
535 __ movl (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
536 __ mov (rcx, rbx);
537 __ subptr (rbx, rax); // subtract lengths
538 __ push (rbx); // result
539 __ cmov (Assembler::lessEqual, rax, rcx);
540 } else {
541 Label L;
542 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
543 __ movl (rcx, Address(rax, java_lang_String::count_offset_in_bytes()));
544 __ mov (rax, rbx);
545 __ subptr (rbx, rcx);
546 __ push (rbx);
547 __ jcc (Assembler::lessEqual, L);
548 __ mov (rax, rcx);
549 __ bind (L);
550 }
551 // is minimum length 0?
552 Label noLoop, haveResult;
553 __ testptr (rax, rax);
554 __ jcc (Assembler::zero, noLoop);
555
556 // compare first characters
557 __ load_unsigned_word(rcx, Address(rdi, 0));
558 __ load_unsigned_word(rbx, Address(rsi, 0));
559 __ subl(rcx, rbx);
560 __ jcc(Assembler::notZero, haveResult);
561 // starting loop
562 __ decrement(rax); // we already tested index: skip one
563 __ jcc(Assembler::zero, noLoop);
564
565 // set rsi.edi to the end of the arrays (arrays have same length)
566 // negate the index
567
568 __ lea(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
569 __ lea(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
570 __ negptr(rax);
571
572 // compare the strings in a loop
573
574 Label loop;
575 __ align(wordSize);
576 __ bind(loop);
577 __ load_unsigned_word(rcx, Address(rdi, rax, Address::times_2, 0));
578 __ load_unsigned_word(rbx, Address(rsi, rax, Address::times_2, 0));
579 __ subl(rcx, rbx);
580 __ jcc(Assembler::notZero, haveResult);
581 __ increment(rax);
582 __ jcc(Assembler::notZero, loop);
583
584 // strings are equal up to min length
585
586 __ bind(noLoop);
587 __ pop(rax);
588 return_op(LIR_OprFact::illegalOpr);
589
590 __ bind(haveResult);
591 // leave instruction is going to discard the TOS value
592 __ mov (rax, rcx); // result of call is in rax,
593 }
594
595
596 void LIR_Assembler::return_op(LIR_Opr result) {
597 assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
598 if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
599 assert(result->fpu() == 0, "result must already be on TOS");
600 }
601
602 // Pop the stack before the safepoint code
603 __ leave();
604
605 bool result_is_oop = result->is_valid() ? result->is_oop() : false;
606
607 // Note: we do not need to round double result; float result has the right precision
608 // the poll sets the condition code, but no data registers
609 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
610 relocInfo::poll_return_type);
611
612 // NOTE: the requires that the polling page be reachable else the reloc
613 // goes to the movq that loads the address and not the faulting instruction
614 // which breaks the signal handler code
615
616 __ test32(rax, polling_page);
617
618 __ ret(0);
619 }
620
621
622 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
623 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
624 relocInfo::poll_type);
625
626 if (info != NULL) {
627 add_debug_info_for_branch(info);
628 } else {
629 ShouldNotReachHere();
630 }
631
632 int offset = __ offset();
633
634 // NOTE: the requires that the polling page be reachable else the reloc
635 // goes to the movq that loads the address and not the faulting instruction
636 // which breaks the signal handler code
637
638 __ test32(rax, polling_page);
639 return offset;
640 }
641
642
643 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
644 if (from_reg != to_reg) __ mov(to_reg, from_reg);
645 }
646
647 void LIR_Assembler::swap_reg(Register a, Register b) {
648 __ xchgptr(a, b);
649 }
650
651
652 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
653 assert(src->is_constant(), "should not call otherwise");
654 assert(dest->is_register(), "should not call otherwise");
655 LIR_Const* c = src->as_constant_ptr();
656
657 switch (c->type()) {
658 case T_INT: {
659 assert(patch_code == lir_patch_none, "no patching handled here");
660 __ movl(dest->as_register(), c->as_jint());
661 break;
662 }
663
664 case T_LONG: {
665 assert(patch_code == lir_patch_none, "no patching handled here");
666 #ifdef _LP64
667 __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
668 #else
669 __ movptr(dest->as_register_lo(), c->as_jint_lo());
670 __ movptr(dest->as_register_hi(), c->as_jint_hi());
671 #endif // _LP64
672 break;
673 }
674
675 case T_OBJECT: {
676 if (patch_code != lir_patch_none) {
677 jobject2reg_with_patching(dest->as_register(), info);
678 } else {
679 __ movoop(dest->as_register(), c->as_jobject());
680 }
681 break;
682 }
683
684 case T_FLOAT: {
685 if (dest->is_single_xmm()) {
686 if (c->is_zero_float()) {
687 __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
688 } else {
689 __ movflt(dest->as_xmm_float_reg(),
690 InternalAddress(float_constant(c->as_jfloat())));
691 }
692 } else {
693 assert(dest->is_single_fpu(), "must be");
694 assert(dest->fpu_regnr() == 0, "dest must be TOS");
695 if (c->is_zero_float()) {
696 __ fldz();
697 } else if (c->is_one_float()) {
698 __ fld1();
699 } else {
700 __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
701 }
702 }
703 break;
704 }
705
706 case T_DOUBLE: {
707 if (dest->is_double_xmm()) {
708 if (c->is_zero_double()) {
709 __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
710 } else {
711 __ movdbl(dest->as_xmm_double_reg(),
712 InternalAddress(double_constant(c->as_jdouble())));
713 }
714 } else {
715 assert(dest->is_double_fpu(), "must be");
716 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
717 if (c->is_zero_double()) {
718 __ fldz();
719 } else if (c->is_one_double()) {
720 __ fld1();
721 } else {
722 __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
723 }
724 }
725 break;
726 }
727
728 default:
729 ShouldNotReachHere();
730 }
731 }
732
733 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
734 assert(src->is_constant(), "should not call otherwise");
735 assert(dest->is_stack(), "should not call otherwise");
736 LIR_Const* c = src->as_constant_ptr();
737
738 switch (c->type()) {
739 case T_INT: // fall through
740 case T_FLOAT:
741 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
742 break;
743
744 case T_OBJECT:
745 __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
746 break;
747
748 case T_LONG: // fall through
749 case T_DOUBLE:
750 #ifdef _LP64
751 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
752 lo_word_offset_in_bytes), (intptr_t)c->as_jlong_bits());
753 #else
754 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
755 lo_word_offset_in_bytes), c->as_jint_lo_bits());
756 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
757 hi_word_offset_in_bytes), c->as_jint_hi_bits());
758 #endif // _LP64
759 break;
760
761 default:
762 ShouldNotReachHere();
763 }
764 }
765
766 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info ) {
767 assert(src->is_constant(), "should not call otherwise");
768 assert(dest->is_address(), "should not call otherwise");
769 LIR_Const* c = src->as_constant_ptr();
770 LIR_Address* addr = dest->as_address_ptr();
771
772 int null_check_here = code_offset();
773 switch (type) {
774 case T_INT: // fall through
775 case T_FLOAT:
776 __ movl(as_Address(addr), c->as_jint_bits());
777 break;
778
779 case T_OBJECT: // fall through
780 case T_ARRAY:
781 if (c->as_jobject() == NULL) {
782 __ movptr(as_Address(addr), (int32_t)NULL_WORD);
783 } else {
784 if (is_literal_address(addr)) {
785 ShouldNotReachHere();
786 __ movoop(as_Address(addr, noreg), c->as_jobject());
787 } else {
788 __ movoop(as_Address(addr), c->as_jobject());
789 }
790 }
791 break;
792
793 case T_LONG: // fall through
794 case T_DOUBLE:
795 #ifdef _LP64
796 if (is_literal_address(addr)) {
797 ShouldNotReachHere();
798 __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
799 } else {
800 __ movptr(r10, (intptr_t)c->as_jlong_bits());
801 null_check_here = code_offset();
802 __ movptr(as_Address_lo(addr), r10);
803 }
804 #else
805 // Always reachable in 32bit so this doesn't produce useless move literal
806 __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
807 __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
808 #endif // _LP64
809 break;
810
811 case T_BOOLEAN: // fall through
812 case T_BYTE:
813 __ movb(as_Address(addr), c->as_jint() & 0xFF);
814 break;
815
816 case T_CHAR: // fall through
817 case T_SHORT:
818 __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
819 break;
820
821 default:
822 ShouldNotReachHere();
823 };
824
825 if (info != NULL) {
826 add_debug_info_for_null_check(null_check_here, info);
827 }
828 }
829
830
831 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
832 assert(src->is_register(), "should not call otherwise");
833 assert(dest->is_register(), "should not call otherwise");
834
835 // move between cpu-registers
836 if (dest->is_single_cpu()) {
837 #ifdef _LP64
838 if (src->type() == T_LONG) {
839 // Can do LONG -> OBJECT
840 move_regs(src->as_register_lo(), dest->as_register());
841 return;
842 }
843 #endif
844 assert(src->is_single_cpu(), "must match");
845 if (src->type() == T_OBJECT) {
846 __ verify_oop(src->as_register());
847 }
848 move_regs(src->as_register(), dest->as_register());
849
850 } else if (dest->is_double_cpu()) {
851 #ifdef _LP64
852 if (src->type() == T_OBJECT || src->type() == T_ARRAY) {
853 // Surprising to me but we can see move of a long to t_object
854 __ verify_oop(src->as_register());
855 move_regs(src->as_register(), dest->as_register_lo());
856 return;
857 }
858 #endif
859 assert(src->is_double_cpu(), "must match");
860 Register f_lo = src->as_register_lo();
861 Register f_hi = src->as_register_hi();
862 Register t_lo = dest->as_register_lo();
863 Register t_hi = dest->as_register_hi();
864 #ifdef _LP64
865 assert(f_hi == f_lo, "must be same");
866 assert(t_hi == t_lo, "must be same");
867 move_regs(f_lo, t_lo);
868 #else
869 assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
870
871
872 if (f_lo == t_hi && f_hi == t_lo) {
873 swap_reg(f_lo, f_hi);
874 } else if (f_hi == t_lo) {
875 assert(f_lo != t_hi, "overwriting register");
876 move_regs(f_hi, t_hi);
877 move_regs(f_lo, t_lo);
878 } else {
879 assert(f_hi != t_lo, "overwriting register");
880 move_regs(f_lo, t_lo);
881 move_regs(f_hi, t_hi);
882 }
883 #endif // LP64
884
885 // special moves from fpu-register to xmm-register
886 // necessary for method results
887 } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
888 __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
889 __ fld_s(Address(rsp, 0));
890 } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
891 __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
892 __ fld_d(Address(rsp, 0));
893 } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
894 __ fstp_s(Address(rsp, 0));
895 __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
896 } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
897 __ fstp_d(Address(rsp, 0));
898 __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
899
900 // move between xmm-registers
901 } else if (dest->is_single_xmm()) {
902 assert(src->is_single_xmm(), "must match");
903 __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
904 } else if (dest->is_double_xmm()) {
905 assert(src->is_double_xmm(), "must match");
906 __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
907
908 // move between fpu-registers (no instruction necessary because of fpu-stack)
909 } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
910 assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
911 assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
912 } else {
913 ShouldNotReachHere();
914 }
915 }
916
917 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
918 assert(src->is_register(), "should not call otherwise");
919 assert(dest->is_stack(), "should not call otherwise");
920
921 if (src->is_single_cpu()) {
922 Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
923 if (type == T_OBJECT || type == T_ARRAY) {
924 __ verify_oop(src->as_register());
925 __ movptr (dst, src->as_register());
926 } else {
927 __ movl (dst, src->as_register());
928 }
929
930 } else if (src->is_double_cpu()) {
931 Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
932 Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
933 __ movptr (dstLO, src->as_register_lo());
934 NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
935
936 } else if (src->is_single_xmm()) {
937 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
938 __ movflt(dst_addr, src->as_xmm_float_reg());
939
940 } else if (src->is_double_xmm()) {
941 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
942 __ movdbl(dst_addr, src->as_xmm_double_reg());
943
944 } else if (src->is_single_fpu()) {
945 assert(src->fpu_regnr() == 0, "argument must be on TOS");
946 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
947 if (pop_fpu_stack) __ fstp_s (dst_addr);
948 else __ fst_s (dst_addr);
949
950 } else if (src->is_double_fpu()) {
951 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
952 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
953 if (pop_fpu_stack) __ fstp_d (dst_addr);
954 else __ fst_d (dst_addr);
955
956 } else {
957 ShouldNotReachHere();
958 }
959 }
960
961
962 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack, bool /* unaligned */) {
963 LIR_Address* to_addr = dest->as_address_ptr();
964 PatchingStub* patch = NULL;
965
966 if (type == T_ARRAY || type == T_OBJECT) {
967 __ verify_oop(src->as_register());
968 }
969 if (patch_code != lir_patch_none) {
970 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
971 Address toa = as_Address(to_addr);
972 assert(toa.disp() != 0, "must have");
973 }
974 if (info != NULL) {
975 add_debug_info_for_null_check_here(info);
976 }
977
978 switch (type) {
979 case T_FLOAT: {
980 if (src->is_single_xmm()) {
981 __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
982 } else {
983 assert(src->is_single_fpu(), "must be");
984 assert(src->fpu_regnr() == 0, "argument must be on TOS");
985 if (pop_fpu_stack) __ fstp_s(as_Address(to_addr));
986 else __ fst_s (as_Address(to_addr));
987 }
988 break;
989 }
990
991 case T_DOUBLE: {
992 if (src->is_double_xmm()) {
993 __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
994 } else {
995 assert(src->is_double_fpu(), "must be");
996 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
997 if (pop_fpu_stack) __ fstp_d(as_Address(to_addr));
998 else __ fst_d (as_Address(to_addr));
999 }
1000 break;
1001 }
1002
1003 case T_ADDRESS: // fall through
1004 case T_ARRAY: // fall through
1005 case T_OBJECT: // fall through
1006 #ifdef _LP64
1007 __ movptr(as_Address(to_addr), src->as_register());
1008 break;
1009 #endif // _LP64
1010 case T_INT:
1011 __ movl(as_Address(to_addr), src->as_register());
1012 break;
1013
1014 case T_LONG: {
1015 Register from_lo = src->as_register_lo();
1016 Register from_hi = src->as_register_hi();
1017 #ifdef _LP64
1018 __ movptr(as_Address_lo(to_addr), from_lo);
1019 #else
1020 Register base = to_addr->base()->as_register();
1021 Register index = noreg;
1022 if (to_addr->index()->is_register()) {
1023 index = to_addr->index()->as_register();
1024 }
1025 if (base == from_lo || index == from_lo) {
1026 assert(base != from_hi, "can't be");
1027 assert(index == noreg || (index != base && index != from_hi), "can't handle this");
1028 __ movl(as_Address_hi(to_addr), from_hi);
1029 if (patch != NULL) {
1030 patching_epilog(patch, lir_patch_high, base, info);
1031 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1032 patch_code = lir_patch_low;
1033 }
1034 __ movl(as_Address_lo(to_addr), from_lo);
1035 } else {
1036 assert(index == noreg || (index != base && index != from_lo), "can't handle this");
1037 __ movl(as_Address_lo(to_addr), from_lo);
1038 if (patch != NULL) {
1039 patching_epilog(patch, lir_patch_low, base, info);
1040 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1041 patch_code = lir_patch_high;
1042 }
1043 __ movl(as_Address_hi(to_addr), from_hi);
1044 }
1045 #endif // _LP64
1046 break;
1047 }
1048
1049 case T_BYTE: // fall through
1050 case T_BOOLEAN: {
1051 Register src_reg = src->as_register();
1052 Address dst_addr = as_Address(to_addr);
1053 assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
1054 __ movb(dst_addr, src_reg);
1055 break;
1056 }
1057
1058 case T_CHAR: // fall through
1059 case T_SHORT:
1060 __ movw(as_Address(to_addr), src->as_register());
1061 break;
1062
1063 default:
1064 ShouldNotReachHere();
1065 }
1066
1067 if (patch_code != lir_patch_none) {
1068 patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
1069 }
1070 }
1071
1072
1073 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1074 assert(src->is_stack(), "should not call otherwise");
1075 assert(dest->is_register(), "should not call otherwise");
1076
1077 if (dest->is_single_cpu()) {
1078 if (type == T_ARRAY || type == T_OBJECT) {
1079 __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1080 __ verify_oop(dest->as_register());
1081 } else {
1082 __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1083 }
1084
1085 } else if (dest->is_double_cpu()) {
1086 Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
1087 Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
1088 __ movptr(dest->as_register_lo(), src_addr_LO);
1089 NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
1090
1091 } else if (dest->is_single_xmm()) {
1092 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1093 __ movflt(dest->as_xmm_float_reg(), src_addr);
1094
1095 } else if (dest->is_double_xmm()) {
1096 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1097 __ movdbl(dest->as_xmm_double_reg(), src_addr);
1098
1099 } else if (dest->is_single_fpu()) {
1100 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1101 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1102 __ fld_s(src_addr);
1103
1104 } else if (dest->is_double_fpu()) {
1105 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1106 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1107 __ fld_d(src_addr);
1108
1109 } else {
1110 ShouldNotReachHere();
1111 }
1112 }
1113
1114
1115 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1116 if (src->is_single_stack()) {
1117 if (type == T_OBJECT || type == T_ARRAY) {
1118 __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
1119 __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
1120 } else {
1121 __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
1122 __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
1123 }
1124
1125 } else if (src->is_double_stack()) {
1126 #ifdef _LP64
1127 __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
1128 __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
1129 #else
1130 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
1131 // push and pop the part at src + wordSize, adding wordSize for the previous push
1132 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), wordSize));
1133 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), wordSize));
1134 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
1135 #endif // _LP64
1136
1137 } else {
1138 ShouldNotReachHere();
1139 }
1140 }
1141
1142
1143 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool /* unaligned */) {
1144 assert(src->is_address(), "should not call otherwise");
1145 assert(dest->is_register(), "should not call otherwise");
1146
1147 LIR_Address* addr = src->as_address_ptr();
1148 Address from_addr = as_Address(addr);
1149
1150 switch (type) {
1151 case T_BOOLEAN: // fall through
1152 case T_BYTE: // fall through
1153 case T_CHAR: // fall through
1154 case T_SHORT:
1155 if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
1156 // on pre P6 processors we may get partial register stalls
1157 // so blow away the value of to_rinfo before loading a
1158 // partial word into it. Do it here so that it precedes
1159 // the potential patch point below.
1160 __ xorptr(dest->as_register(), dest->as_register());
1161 }
1162 break;
1163 }
1164
1165 PatchingStub* patch = NULL;
1166 if (patch_code != lir_patch_none) {
1167 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1168 assert(from_addr.disp() != 0, "must have");
1169 }
1170 if (info != NULL) {
1171 add_debug_info_for_null_check_here(info);
1172 }
1173
1174 switch (type) {
1175 case T_FLOAT: {
1176 if (dest->is_single_xmm()) {
1177 __ movflt(dest->as_xmm_float_reg(), from_addr);
1178 } else {
1179 assert(dest->is_single_fpu(), "must be");
1180 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1181 __ fld_s(from_addr);
1182 }
1183 break;
1184 }
1185
1186 case T_DOUBLE: {
1187 if (dest->is_double_xmm()) {
1188 __ movdbl(dest->as_xmm_double_reg(), from_addr);
1189 } else {
1190 assert(dest->is_double_fpu(), "must be");
1191 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1192 __ fld_d(from_addr);
1193 }
1194 break;
1195 }
1196
1197 case T_ADDRESS: // fall through
1198 case T_OBJECT: // fall through
1199 case T_ARRAY: // fall through
1200 #ifdef _LP64
1201 __ movptr(dest->as_register(), from_addr);
1202 break;
1203 #endif // _L64
1204 case T_INT:
1205 // %%% could this be a movl? this is safer but longer instruction
1206 __ movl2ptr(dest->as_register(), from_addr);
1207 break;
1208
1209 case T_LONG: {
1210 Register to_lo = dest->as_register_lo();
1211 Register to_hi = dest->as_register_hi();
1212 #ifdef _LP64
1213 __ movptr(to_lo, as_Address_lo(addr));
1214 #else
1215 Register base = addr->base()->as_register();
1216 Register index = noreg;
1217 if (addr->index()->is_register()) {
1218 index = addr->index()->as_register();
1219 }
1220 if ((base == to_lo && index == to_hi) ||
1221 (base == to_hi && index == to_lo)) {
1222 // addresses with 2 registers are only formed as a result of
1223 // array access so this code will never have to deal with
1224 // patches or null checks.
1225 assert(info == NULL && patch == NULL, "must be");
1226 __ lea(to_hi, as_Address(addr));
1227 __ movl(to_lo, Address(to_hi, 0));
1228 __ movl(to_hi, Address(to_hi, BytesPerWord));
1229 } else if (base == to_lo || index == to_lo) {
1230 assert(base != to_hi, "can't be");
1231 assert(index == noreg || (index != base && index != to_hi), "can't handle this");
1232 __ movl(to_hi, as_Address_hi(addr));
1233 if (patch != NULL) {
1234 patching_epilog(patch, lir_patch_high, base, info);
1235 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1236 patch_code = lir_patch_low;
1237 }
1238 __ movl(to_lo, as_Address_lo(addr));
1239 } else {
1240 assert(index == noreg || (index != base && index != to_lo), "can't handle this");
1241 __ movl(to_lo, as_Address_lo(addr));
1242 if (patch != NULL) {
1243 patching_epilog(patch, lir_patch_low, base, info);
1244 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1245 patch_code = lir_patch_high;
1246 }
1247 __ movl(to_hi, as_Address_hi(addr));
1248 }
1249 #endif // _LP64
1250 break;
1251 }
1252
1253 case T_BOOLEAN: // fall through
1254 case T_BYTE: {
1255 Register dest_reg = dest->as_register();
1256 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1257 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1258 __ movsbl(dest_reg, from_addr);
1259 } else {
1260 __ movb(dest_reg, from_addr);
1261 __ shll(dest_reg, 24);
1262 __ sarl(dest_reg, 24);
1263 }
1264 // These are unsigned so the zero extension on 64bit is just what we need
1265 break;
1266 }
1267
1268 case T_CHAR: {
1269 Register dest_reg = dest->as_register();
1270 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1271 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1272 __ movzwl(dest_reg, from_addr);
1273 } else {
1274 __ movw(dest_reg, from_addr);
1275 }
1276 // This is unsigned so the zero extension on 64bit is just what we need
1277 // __ movl2ptr(dest_reg, dest_reg);
1278 break;
1279 }
1280
1281 case T_SHORT: {
1282 Register dest_reg = dest->as_register();
1283 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1284 __ movswl(dest_reg, from_addr);
1285 } else {
1286 __ movw(dest_reg, from_addr);
1287 __ shll(dest_reg, 16);
1288 __ sarl(dest_reg, 16);
1289 }
1290 // Might not be needed in 64bit but certainly doesn't hurt (except for code size)
1291 __ movl2ptr(dest_reg, dest_reg);
1292 break;
1293 }
1294
1295 default:
1296 ShouldNotReachHere();
1297 }
1298
1299 if (patch != NULL) {
1300 patching_epilog(patch, patch_code, addr->base()->as_register(), info);
1301 }
1302
1303 if (type == T_ARRAY || type == T_OBJECT) {
1304 __ verify_oop(dest->as_register());
1305 }
1306 }
1307
1308
1309 void LIR_Assembler::prefetchr(LIR_Opr src) {
1310 LIR_Address* addr = src->as_address_ptr();
1311 Address from_addr = as_Address(addr);
1312
1313 if (VM_Version::supports_sse()) {
1314 switch (ReadPrefetchInstr) {
1315 case 0:
1316 __ prefetchnta(from_addr); break;
1317 case 1:
1318 __ prefetcht0(from_addr); break;
1319 case 2:
1320 __ prefetcht2(from_addr); break;
1321 default:
1322 ShouldNotReachHere(); break;
1323 }
1324 } else if (VM_Version::supports_3dnow()) {
1325 __ prefetchr(from_addr);
1326 }
1327 }
1328
1329
1330 void LIR_Assembler::prefetchw(LIR_Opr src) {
1331 LIR_Address* addr = src->as_address_ptr();
1332 Address from_addr = as_Address(addr);
1333
1334 if (VM_Version::supports_sse()) {
1335 switch (AllocatePrefetchInstr) {
1336 case 0:
1337 __ prefetchnta(from_addr); break;
1338 case 1:
1339 __ prefetcht0(from_addr); break;
1340 case 2:
1341 __ prefetcht2(from_addr); break;
1342 case 3:
1343 __ prefetchw(from_addr); break;
1344 default:
1345 ShouldNotReachHere(); break;
1346 }
1347 } else if (VM_Version::supports_3dnow()) {
1348 __ prefetchw(from_addr);
1349 }
1350 }
1351
1352
1353 NEEDS_CLEANUP; // This could be static?
1354 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
1355 int elem_size = type2aelembytes(type);
1356 switch (elem_size) {
1357 case 1: return Address::times_1;
1358 case 2: return Address::times_2;
1359 case 4: return Address::times_4;
1360 case 8: return Address::times_8;
1361 }
1362 ShouldNotReachHere();
1363 return Address::no_scale;
1364 }
1365
1366
1367 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1368 switch (op->code()) {
1369 case lir_idiv:
1370 case lir_irem:
1371 arithmetic_idiv(op->code(),
1372 op->in_opr1(),
1373 op->in_opr2(),
1374 op->in_opr3(),
1375 op->result_opr(),
1376 op->info());
1377 break;
1378 default: ShouldNotReachHere(); break;
1379 }
1380 }
1381
1382 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1383 #ifdef ASSERT
1384 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
1385 if (op->block() != NULL) _branch_target_blocks.append(op->block());
1386 if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
1387 #endif
1388
1389 if (op->cond() == lir_cond_always) {
1390 if (op->info() != NULL) add_debug_info_for_branch(op->info());
1391 __ jmp (*(op->label()));
1392 } else {
1393 Assembler::Condition acond = Assembler::zero;
1394 if (op->code() == lir_cond_float_branch) {
1395 assert(op->ublock() != NULL, "must have unordered successor");
1396 __ jcc(Assembler::parity, *(op->ublock()->label()));
1397 switch(op->cond()) {
1398 case lir_cond_equal: acond = Assembler::equal; break;
1399 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1400 case lir_cond_less: acond = Assembler::below; break;
1401 case lir_cond_lessEqual: acond = Assembler::belowEqual; break;
1402 case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
1403 case lir_cond_greater: acond = Assembler::above; break;
1404 default: ShouldNotReachHere();
1405 }
1406 } else {
1407 switch (op->cond()) {
1408 case lir_cond_equal: acond = Assembler::equal; break;
1409 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1410 case lir_cond_less: acond = Assembler::less; break;
1411 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
1412 case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
1413 case lir_cond_greater: acond = Assembler::greater; break;
1414 case lir_cond_belowEqual: acond = Assembler::belowEqual; break;
1415 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break;
1416 default: ShouldNotReachHere();
1417 }
1418 }
1419 __ jcc(acond,*(op->label()));
1420 }
1421 }
1422
1423 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1424 LIR_Opr src = op->in_opr();
1425 LIR_Opr dest = op->result_opr();
1426
1427 switch (op->bytecode()) {
1428 case Bytecodes::_i2l:
1429 #ifdef _LP64
1430 __ movl2ptr(dest->as_register_lo(), src->as_register());
1431 #else
1432 move_regs(src->as_register(), dest->as_register_lo());
1433 move_regs(src->as_register(), dest->as_register_hi());
1434 __ sarl(dest->as_register_hi(), 31);
1435 #endif // LP64
1436 break;
1437
1438 case Bytecodes::_l2i:
1439 move_regs(src->as_register_lo(), dest->as_register());
1440 break;
1441
1442 case Bytecodes::_i2b:
1443 move_regs(src->as_register(), dest->as_register());
1444 __ sign_extend_byte(dest->as_register());
1445 break;
1446
1447 case Bytecodes::_i2c:
1448 move_regs(src->as_register(), dest->as_register());
1449 __ andl(dest->as_register(), 0xFFFF);
1450 break;
1451
1452 case Bytecodes::_i2s:
1453 move_regs(src->as_register(), dest->as_register());
1454 __ sign_extend_short(dest->as_register());
1455 break;
1456
1457
1458 case Bytecodes::_f2d:
1459 case Bytecodes::_d2f:
1460 if (dest->is_single_xmm()) {
1461 __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1462 } else if (dest->is_double_xmm()) {
1463 __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1464 } else {
1465 assert(src->fpu() == dest->fpu(), "register must be equal");
1466 // do nothing (float result is rounded later through spilling)
1467 }
1468 break;
1469
1470 case Bytecodes::_i2f:
1471 case Bytecodes::_i2d:
1472 if (dest->is_single_xmm()) {
1473 __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1474 } else if (dest->is_double_xmm()) {
1475 __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1476 } else {
1477 assert(dest->fpu() == 0, "result must be on TOS");
1478 __ movl(Address(rsp, 0), src->as_register());
1479 __ fild_s(Address(rsp, 0));
1480 }
1481 break;
1482
1483 case Bytecodes::_f2i:
1484 case Bytecodes::_d2i:
1485 if (src->is_single_xmm()) {
1486 __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
1487 } else if (src->is_double_xmm()) {
1488 __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
1489 } else {
1490 assert(src->fpu() == 0, "input must be on TOS");
1491 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
1492 __ fist_s(Address(rsp, 0));
1493 __ movl(dest->as_register(), Address(rsp, 0));
1494 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1495 }
1496
1497 // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
1498 assert(op->stub() != NULL, "stub required");
1499 __ cmpl(dest->as_register(), 0x80000000);
1500 __ jcc(Assembler::equal, *op->stub()->entry());
1501 __ bind(*op->stub()->continuation());
1502 break;
1503
1504 case Bytecodes::_l2f:
1505 case Bytecodes::_l2d:
1506 assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
1507 assert(dest->fpu() == 0, "result must be on TOS");
1508
1509 __ movptr(Address(rsp, 0), src->as_register_lo());
1510 NOT_LP64(__ movl(Address(rsp, BytesPerWord), src->as_register_hi()));
1511 __ fild_d(Address(rsp, 0));
1512 // float result is rounded later through spilling
1513 break;
1514
1515 case Bytecodes::_f2l:
1516 case Bytecodes::_d2l:
1517 assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
1518 assert(src->fpu() == 0, "input must be on TOS");
1519 assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
1520
1521 // instruction sequence too long to inline it here
1522 {
1523 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
1524 }
1525 break;
1526
1527 default: ShouldNotReachHere();
1528 }
1529 }
1530
1531 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1532 if (op->init_check()) {
1533 __ cmpl(Address(op->klass()->as_register(),
1534 instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)),
1535 instanceKlass::fully_initialized);
1536 add_debug_info_for_null_check_here(op->stub()->info());
1537 __ jcc(Assembler::notEqual, *op->stub()->entry());
1538 }
1539 __ allocate_object(op->obj()->as_register(),
1540 op->tmp1()->as_register(),
1541 op->tmp2()->as_register(),
1542 op->header_size(),
1543 op->object_size(),
1544 op->klass()->as_register(),
1545 *op->stub()->entry());
1546 __ bind(*op->stub()->continuation());
1547 }
1548
1549 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1550 if (UseSlowPath ||
1551 (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
1552 (!UseFastNewTypeArray && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
1553 __ jmp(*op->stub()->entry());
1554 } else {
1555 Register len = op->len()->as_register();
1556 Register tmp1 = op->tmp1()->as_register();
1557 Register tmp2 = op->tmp2()->as_register();
1558 Register tmp3 = op->tmp3()->as_register();
1559 if (len == tmp1) {
1560 tmp1 = tmp3;
1561 } else if (len == tmp2) {
1562 tmp2 = tmp3;
1563 } else if (len == tmp3) {
1564 // everything is ok
1565 } else {
1566 __ mov(tmp3, len);
1567 }
1568 __ allocate_array(op->obj()->as_register(),
1569 len,
1570 tmp1,
1571 tmp2,
1572 arrayOopDesc::header_size(op->type()),
1573 array_element_size(op->type()),
1574 op->klass()->as_register(),
1575 *op->stub()->entry());
1576 }
1577 __ bind(*op->stub()->continuation());
1578 }
1579
1580
1581
1582 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1583 LIR_Code code = op->code();
1584 if (code == lir_store_check) {
1585 Register value = op->object()->as_register();
1586 Register array = op->array()->as_register();
1587 Register k_RInfo = op->tmp1()->as_register();
1588 Register klass_RInfo = op->tmp2()->as_register();
1589 Register Rtmp1 = op->tmp3()->as_register();
1590
1591 CodeStub* stub = op->stub();
1592 Label done;
1593 __ cmpptr(value, (int32_t)NULL_WORD);
1594 __ jcc(Assembler::equal, done);
1595 add_debug_info_for_null_check_here(op->info_for_exception());
1596 __ movptr(k_RInfo, Address(array, oopDesc::klass_offset_in_bytes()));
1597 __ movptr(klass_RInfo, Address(value, oopDesc::klass_offset_in_bytes()));
1598
1599 // get instance klass
1600 __ movptr(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
1601 // get super_check_offset
1602 __ movl(Rtmp1, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes()));
1603 // See if we get an immediate positive hit
1604 __ cmpptr(k_RInfo, Address(klass_RInfo, Rtmp1, Address::times_1));
1605 __ jcc(Assembler::equal, done);
1606 // check for immediate negative hit
1607 __ cmpl(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
1608 __ jcc(Assembler::notEqual, *stub->entry());
1609 // check for self
1610 __ cmpptr(klass_RInfo, k_RInfo);
1611 __ jcc(Assembler::equal, done);
1612
1613 __ push(klass_RInfo);
1614 __ push(k_RInfo);
1615 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1616 __ pop(klass_RInfo);
1617 __ pop(k_RInfo);
1618 // result is a boolean
1619 __ cmpl(k_RInfo, 0);
1620 __ jcc(Assembler::equal, *stub->entry());
1621 __ bind(done);
1622 } else if (op->code() == lir_checkcast) {
1623 // we always need a stub for the failure case.
1624 CodeStub* stub = op->stub();
1625 Register obj = op->object()->as_register();
1626 Register k_RInfo = op->tmp1()->as_register();
1627 Register klass_RInfo = op->tmp2()->as_register();
1628 Register dst = op->result_opr()->as_register();
1629 ciKlass* k = op->klass();
1630 Register Rtmp1 = noreg;
1631
1632 Label done;
1633 if (obj == k_RInfo) {
1634 k_RInfo = dst;
1635 } else if (obj == klass_RInfo) {
1636 klass_RInfo = dst;
1637 }
1638 if (k->is_loaded()) {
1639 select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1640 } else {
1641 Rtmp1 = op->tmp3()->as_register();
1642 select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1643 }
1644
1645 assert_different_registers(obj, k_RInfo, klass_RInfo);
1646 if (!k->is_loaded()) {
1647 jobject2reg_with_patching(k_RInfo, op->info_for_patch());
1648 } else {
1649 #ifdef _LP64
1650 __ movoop(k_RInfo, k->encoding());
1651 #else
1652 k_RInfo = noreg;
1653 #endif // _LP64
1654 }
1655 assert(obj != k_RInfo, "must be different");
1656 __ cmpptr(obj, (int32_t)NULL_WORD);
1657 if (op->profiled_method() != NULL) {
1658 ciMethod* method = op->profiled_method();
1659 int bci = op->profiled_bci();
1660
1661 Label profile_done;
1662 __ jcc(Assembler::notEqual, profile_done);
1663 // Object is null; update methodDataOop
1664 ciMethodData* md = method->method_data();
1665 if (md == NULL) {
1666 bailout("out of memory building methodDataOop");
1667 return;
1668 }
1669 ciProfileData* data = md->bci_to_data(bci);
1670 assert(data != NULL, "need data for checkcast");
1671 assert(data->is_BitData(), "need BitData for checkcast");
1672 Register mdo = klass_RInfo;
1673 __ movoop(mdo, md->encoding());
1674 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1675 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1676 __ orl(data_addr, header_bits);
1677 __ jmp(done);
1678 __ bind(profile_done);
1679 } else {
1680 __ jcc(Assembler::equal, done);
1681 }
1682 __ verify_oop(obj);
1683
1684 if (op->fast_check()) {
1685 // get object classo
1686 // not a safepoint as obj null check happens earlier
1687 if (k->is_loaded()) {
1688 #ifdef _LP64
1689 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1690 #else
1691 __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->encoding());
1692 #endif // _LP64
1693 } else {
1694 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1695
1696 }
1697 __ jcc(Assembler::notEqual, *stub->entry());
1698 __ bind(done);
1699 } else {
1700 // get object class
1701 // not a safepoint as obj null check happens earlier
1702 __ movptr(klass_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1703 if (k->is_loaded()) {
1704 // See if we get an immediate positive hit
1705 #ifdef _LP64
1706 __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
1707 #else
1708 __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->encoding());
1709 #endif // _LP64
1710 if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() != k->super_check_offset()) {
1711 __ jcc(Assembler::notEqual, *stub->entry());
1712 } else {
1713 // See if we get an immediate positive hit
1714 __ jcc(Assembler::equal, done);
1715 // check for self
1716 #ifdef _LP64
1717 __ cmpptr(klass_RInfo, k_RInfo);
1718 #else
1719 __ cmpoop(klass_RInfo, k->encoding());
1720 #endif // _LP64
1721 __ jcc(Assembler::equal, done);
1722
1723 __ push(klass_RInfo);
1724 #ifdef _LP64
1725 __ push(k_RInfo);
1726 #else
1727 __ pushoop(k->encoding());
1728 #endif // _LP64
1729 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1730 __ pop(klass_RInfo);
1731 __ pop(klass_RInfo);
1732 // result is a boolean
1733 __ cmpl(klass_RInfo, 0);
1734 __ jcc(Assembler::equal, *stub->entry());
1735 }
1736 __ bind(done);
1737 } else {
1738 __ movl(Rtmp1, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes()));
1739 // See if we get an immediate positive hit
1740 __ cmpptr(k_RInfo, Address(klass_RInfo, Rtmp1, Address::times_1));
1741 __ jcc(Assembler::equal, done);
1742 // check for immediate negative hit
1743 __ cmpl(Rtmp1, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
1744 __ jcc(Assembler::notEqual, *stub->entry());
1745 // check for self
1746 __ cmpptr(klass_RInfo, k_RInfo);
1747 __ jcc(Assembler::equal, done);
1748
1749 __ push(klass_RInfo);
1750 __ push(k_RInfo);
1751 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1752 __ pop(klass_RInfo);
1753 __ pop(k_RInfo);
1754 // result is a boolean
1755 __ cmpl(k_RInfo, 0);
1756 __ jcc(Assembler::equal, *stub->entry());
1757 __ bind(done);
1758 }
1759
1760 }
1761 if (dst != obj) {
1762 __ mov(dst, obj);
1763 }
1764 } else if (code == lir_instanceof) {
1765 Register obj = op->object()->as_register();
1766 Register k_RInfo = op->tmp1()->as_register();
1767 Register klass_RInfo = op->tmp2()->as_register();
1768 Register dst = op->result_opr()->as_register();
1769 ciKlass* k = op->klass();
1770
1771 Label done;
1772 Label zero;
1773 Label one;
1774 if (obj == k_RInfo) {
1775 k_RInfo = klass_RInfo;
1776 klass_RInfo = obj;
1777 }
1778 // patching may screw with our temporaries on sparc,
1779 // so let's do it before loading the class
1780 if (!k->is_loaded()) {
1781 jobject2reg_with_patching(k_RInfo, op->info_for_patch());
1782 } else {
1783 LP64_ONLY(__ movoop(k_RInfo, k->encoding()));
1784 }
1785 assert(obj != k_RInfo, "must be different");
1786
1787 __ verify_oop(obj);
1788 if (op->fast_check()) {
1789 __ cmpptr(obj, (int32_t)NULL_WORD);
1790 __ jcc(Assembler::equal, zero);
1791 // get object class
1792 // not a safepoint as obj null check happens earlier
1793 if (LP64_ONLY(false &&) k->is_loaded()) {
1794 NOT_LP64(__ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->encoding()));
1795 k_RInfo = noreg;
1796 } else {
1797 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1798
1799 }
1800 __ jcc(Assembler::equal, one);
1801 } else {
1802 // get object class
1803 // not a safepoint as obj null check happens earlier
1804 __ cmpptr(obj, (int32_t)NULL_WORD);
1805 __ jcc(Assembler::equal, zero);
1806 __ movptr(klass_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1807
1808 #ifndef _LP64
1809 if (k->is_loaded()) {
1810 // See if we get an immediate positive hit
1811 __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->encoding());
1812 __ jcc(Assembler::equal, one);
1813 if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() == k->super_check_offset()) {
1814 // check for self
1815 __ cmpoop(klass_RInfo, k->encoding());
1816 __ jcc(Assembler::equal, one);
1817 __ push(klass_RInfo);
1818 __ pushoop(k->encoding());
1819 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1820 __ pop(klass_RInfo);
1821 __ pop(dst);
1822 __ jmp(done);
1823 }
1824 } else {
1825 #else
1826 { // YUCK
1827 #endif // LP64
1828 assert(dst != klass_RInfo && dst != k_RInfo, "need 3 registers");
1829
1830 __ movl(dst, Address(k_RInfo, sizeof(oopDesc) + Klass::super_check_offset_offset_in_bytes()));
1831 // See if we get an immediate positive hit
1832 __ cmpptr(k_RInfo, Address(klass_RInfo, dst, Address::times_1));
1833 __ jcc(Assembler::equal, one);
1834 // check for immediate negative hit
1835 __ cmpl(dst, sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes());
1836 __ jcc(Assembler::notEqual, zero);
1837 // check for self
1838 __ cmpptr(klass_RInfo, k_RInfo);
1839 __ jcc(Assembler::equal, one);
1840
1841 __ push(klass_RInfo);
1842 __ push(k_RInfo);
1843 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1844 __ pop(klass_RInfo);
1845 __ pop(dst);
1846 __ jmp(done);
1847 }
1848 }
1849 __ bind(zero);
1850 __ xorptr(dst, dst);
1851 __ jmp(done);
1852 __ bind(one);
1853 __ movptr(dst, 1);
1854 __ bind(done);
1855 } else {
1856 ShouldNotReachHere();
1857 }
1858
1859 }
1860
1861
1862 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1863 if (LP64_ONLY(false &&) op->code() == lir_cas_long && VM_Version::supports_cx8()) {
1864 assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
1865 assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
1866 assert(op->new_value()->as_register_lo() == rbx, "wrong register");
1867 assert(op->new_value()->as_register_hi() == rcx, "wrong register");
1868 Register addr = op->addr()->as_register();
1869 if (os::is_MP()) {
1870 __ lock();
1871 }
1872 NOT_LP64(__ cmpxchg8(Address(addr, 0)));
1873
1874 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
1875 NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
1876 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1877 Register newval = op->new_value()->as_register();
1878 Register cmpval = op->cmp_value()->as_register();
1879 assert(cmpval == rax, "wrong register");
1880 assert(newval != NULL, "new val must be register");
1881 assert(cmpval != newval, "cmp and new values must be in different registers");
1882 assert(cmpval != addr, "cmp and addr must be in different registers");
1883 assert(newval != addr, "new value and addr must be in different registers");
1884 if (os::is_MP()) {
1885 __ lock();
1886 }
1887 if ( op->code() == lir_cas_obj) {
1888 __ cmpxchgptr(newval, Address(addr, 0));
1889 } else if (op->code() == lir_cas_int) {
1890 __ cmpxchgl(newval, Address(addr, 0));
1891 } else {
1892 LP64_ONLY(__ cmpxchgq(newval, Address(addr, 0)));
1893 }
1894 #ifdef _LP64
1895 } else if (op->code() == lir_cas_long) {
1896 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1897 Register newval = op->new_value()->as_register_lo();
1898 Register cmpval = op->cmp_value()->as_register_lo();
1899 assert(cmpval == rax, "wrong register");
1900 assert(newval != NULL, "new val must be register");
1901 assert(cmpval != newval, "cmp and new values must be in different registers");
1902 assert(cmpval != addr, "cmp and addr must be in different registers");
1903 assert(newval != addr, "new value and addr must be in different registers");
1904 if (os::is_MP()) {
1905 __ lock();
1906 }
1907 __ cmpxchgq(newval, Address(addr, 0));
1908 #endif // _LP64
1909 } else {
1910 Unimplemented();
1911 }
1912 }
1913
1914
1915 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result) {
1916 Assembler::Condition acond, ncond;
1917 switch (condition) {
1918 case lir_cond_equal: acond = Assembler::equal; ncond = Assembler::notEqual; break;
1919 case lir_cond_notEqual: acond = Assembler::notEqual; ncond = Assembler::equal; break;
1920 case lir_cond_less: acond = Assembler::less; ncond = Assembler::greaterEqual; break;
1921 case lir_cond_lessEqual: acond = Assembler::lessEqual; ncond = Assembler::greater; break;
1922 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less; break;
1923 case lir_cond_greater: acond = Assembler::greater; ncond = Assembler::lessEqual; break;
1924 case lir_cond_belowEqual: acond = Assembler::belowEqual; ncond = Assembler::above; break;
1925 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; ncond = Assembler::below; break;
1926 default: ShouldNotReachHere();
1927 }
1928
1929 if (opr1->is_cpu_register()) {
1930 reg2reg(opr1, result);
1931 } else if (opr1->is_stack()) {
1932 stack2reg(opr1, result, result->type());
1933 } else if (opr1->is_constant()) {
1934 const2reg(opr1, result, lir_patch_none, NULL);
1935 } else {
1936 ShouldNotReachHere();
1937 }
1938
1939 if (VM_Version::supports_cmov() && !opr2->is_constant()) {
1940 // optimized version that does not require a branch
1941 if (opr2->is_single_cpu()) {
1942 assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
1943 __ cmov(ncond, result->as_register(), opr2->as_register());
1944 } else if (opr2->is_double_cpu()) {
1945 assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1946 assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1947 __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
1948 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
1949 } else if (opr2->is_single_stack()) {
1950 __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
1951 } else if (opr2->is_double_stack()) {
1952 __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
1953 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
1954 } else {
1955 ShouldNotReachHere();
1956 }
1957
1958 } else {
1959 Label skip;
1960 __ jcc (acond, skip);
1961 if (opr2->is_cpu_register()) {
1962 reg2reg(opr2, result);
1963 } else if (opr2->is_stack()) {
1964 stack2reg(opr2, result, result->type());
1965 } else if (opr2->is_constant()) {
1966 const2reg(opr2, result, lir_patch_none, NULL);
1967 } else {
1968 ShouldNotReachHere();
1969 }
1970 __ bind(skip);
1971 }
1972 }
1973
1974
1975 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
1976 assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
1977
1978 if (left->is_single_cpu()) {
1979 assert(left == dest, "left and dest must be equal");
1980 Register lreg = left->as_register();
1981
1982 if (right->is_single_cpu()) {
1983 // cpu register - cpu register
1984 Register rreg = right->as_register();
1985 switch (code) {
1986 case lir_add: __ addl (lreg, rreg); break;
1987 case lir_sub: __ subl (lreg, rreg); break;
1988 case lir_mul: __ imull(lreg, rreg); break;
1989 default: ShouldNotReachHere();
1990 }
1991
1992 } else if (right->is_stack()) {
1993 // cpu register - stack
1994 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1995 switch (code) {
1996 case lir_add: __ addl(lreg, raddr); break;
1997 case lir_sub: __ subl(lreg, raddr); break;
1998 default: ShouldNotReachHere();
1999 }
2000
2001 } else if (right->is_constant()) {
2002 // cpu register - constant
2003 jint c = right->as_constant_ptr()->as_jint();
2004 switch (code) {
2005 case lir_add: {
2006 __ increment(lreg, c);
2007 break;
2008 }
2009 case lir_sub: {
2010 __ decrement(lreg, c);
2011 break;
2012 }
2013 default: ShouldNotReachHere();
2014 }
2015
2016 } else {
2017 ShouldNotReachHere();
2018 }
2019
2020 } else if (left->is_double_cpu()) {
2021 assert(left == dest, "left and dest must be equal");
2022 Register lreg_lo = left->as_register_lo();
2023 Register lreg_hi = left->as_register_hi();
2024
2025 if (right->is_double_cpu()) {
2026 // cpu register - cpu register
2027 Register rreg_lo = right->as_register_lo();
2028 Register rreg_hi = right->as_register_hi();
2029 NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
2030 LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
2031 switch (code) {
2032 case lir_add:
2033 __ addptr(lreg_lo, rreg_lo);
2034 NOT_LP64(__ adcl(lreg_hi, rreg_hi));
2035 break;
2036 case lir_sub:
2037 __ subptr(lreg_lo, rreg_lo);
2038 NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
2039 break;
2040 case lir_mul:
2041 #ifdef _LP64
2042 __ imulq(lreg_lo, rreg_lo);
2043 #else
2044 assert(lreg_lo == rax && lreg_hi == rdx, "must be");
2045 __ imull(lreg_hi, rreg_lo);
2046 __ imull(rreg_hi, lreg_lo);
2047 __ addl (rreg_hi, lreg_hi);
2048 __ mull (rreg_lo);
2049 __ addl (lreg_hi, rreg_hi);
2050 #endif // _LP64
2051 break;
2052 default:
2053 ShouldNotReachHere();
2054 }
2055
2056 } else if (right->is_constant()) {
2057 // cpu register - constant
2058 #ifdef _LP64
2059 jlong c = right->as_constant_ptr()->as_jlong_bits();
2060 __ movptr(r10, (intptr_t) c);
2061 switch (code) {
2062 case lir_add:
2063 __ addptr(lreg_lo, r10);
2064 break;
2065 case lir_sub:
2066 __ subptr(lreg_lo, r10);
2067 break;
2068 default:
2069 ShouldNotReachHere();
2070 }
2071 #else
2072 jint c_lo = right->as_constant_ptr()->as_jint_lo();
2073 jint c_hi = right->as_constant_ptr()->as_jint_hi();
2074 switch (code) {
2075 case lir_add:
2076 __ addptr(lreg_lo, c_lo);
2077 __ adcl(lreg_hi, c_hi);
2078 break;
2079 case lir_sub:
2080 __ subptr(lreg_lo, c_lo);
2081 __ sbbl(lreg_hi, c_hi);
2082 break;
2083 default:
2084 ShouldNotReachHere();
2085 }
2086 #endif // _LP64
2087
2088 } else {
2089 ShouldNotReachHere();
2090 }
2091
2092 } else if (left->is_single_xmm()) {
2093 assert(left == dest, "left and dest must be equal");
2094 XMMRegister lreg = left->as_xmm_float_reg();
2095
2096 if (right->is_single_xmm()) {
2097 XMMRegister rreg = right->as_xmm_float_reg();
2098 switch (code) {
2099 case lir_add: __ addss(lreg, rreg); break;
2100 case lir_sub: __ subss(lreg, rreg); break;
2101 case lir_mul_strictfp: // fall through
2102 case lir_mul: __ mulss(lreg, rreg); break;
2103 case lir_div_strictfp: // fall through
2104 case lir_div: __ divss(lreg, rreg); break;
2105 default: ShouldNotReachHere();
2106 }
2107 } else {
2108 Address raddr;
2109 if (right->is_single_stack()) {
2110 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2111 } else if (right->is_constant()) {
2112 // hack for now
2113 raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
2114 } else {
2115 ShouldNotReachHere();
2116 }
2117 switch (code) {
2118 case lir_add: __ addss(lreg, raddr); break;
2119 case lir_sub: __ subss(lreg, raddr); break;
2120 case lir_mul_strictfp: // fall through
2121 case lir_mul: __ mulss(lreg, raddr); break;
2122 case lir_div_strictfp: // fall through
2123 case lir_div: __ divss(lreg, raddr); break;
2124 default: ShouldNotReachHere();
2125 }
2126 }
2127
2128 } else if (left->is_double_xmm()) {
2129 assert(left == dest, "left and dest must be equal");
2130
2131 XMMRegister lreg = left->as_xmm_double_reg();
2132 if (right->is_double_xmm()) {
2133 XMMRegister rreg = right->as_xmm_double_reg();
2134 switch (code) {
2135 case lir_add: __ addsd(lreg, rreg); break;
2136 case lir_sub: __ subsd(lreg, rreg); break;
2137 case lir_mul_strictfp: // fall through
2138 case lir_mul: __ mulsd(lreg, rreg); break;
2139 case lir_div_strictfp: // fall through
2140 case lir_div: __ divsd(lreg, rreg); break;
2141 default: ShouldNotReachHere();
2142 }
2143 } else {
2144 Address raddr;
2145 if (right->is_double_stack()) {
2146 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2147 } else if (right->is_constant()) {
2148 // hack for now
2149 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2150 } else {
2151 ShouldNotReachHere();
2152 }
2153 switch (code) {
2154 case lir_add: __ addsd(lreg, raddr); break;
2155 case lir_sub: __ subsd(lreg, raddr); break;
2156 case lir_mul_strictfp: // fall through
2157 case lir_mul: __ mulsd(lreg, raddr); break;
2158 case lir_div_strictfp: // fall through
2159 case lir_div: __ divsd(lreg, raddr); break;
2160 default: ShouldNotReachHere();
2161 }
2162 }
2163
2164 } else if (left->is_single_fpu()) {
2165 assert(dest->is_single_fpu(), "fpu stack allocation required");
2166
2167 if (right->is_single_fpu()) {
2168 arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
2169
2170 } else {
2171 assert(left->fpu_regnr() == 0, "left must be on TOS");
2172 assert(dest->fpu_regnr() == 0, "dest must be on TOS");
2173
2174 Address raddr;
2175 if (right->is_single_stack()) {
2176 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2177 } else if (right->is_constant()) {
2178 address const_addr = float_constant(right->as_jfloat());
2179 assert(const_addr != NULL, "incorrect float/double constant maintainance");
2180 // hack for now
2181 raddr = __ as_Address(InternalAddress(const_addr));
2182 } else {
2183 ShouldNotReachHere();
2184 }
2185
2186 switch (code) {
2187 case lir_add: __ fadd_s(raddr); break;
2188 case lir_sub: __ fsub_s(raddr); break;
2189 case lir_mul_strictfp: // fall through
2190 case lir_mul: __ fmul_s(raddr); break;
2191 case lir_div_strictfp: // fall through
2192 case lir_div: __ fdiv_s(raddr); break;
2193 default: ShouldNotReachHere();
2194 }
2195 }
2196
2197 } else if (left->is_double_fpu()) {
2198 assert(dest->is_double_fpu(), "fpu stack allocation required");
2199
2200 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2201 // Double values require special handling for strictfp mul/div on x86
2202 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
2203 __ fmulp(left->fpu_regnrLo() + 1);
2204 }
2205
2206 if (right->is_double_fpu()) {
2207 arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
2208
2209 } else {
2210 assert(left->fpu_regnrLo() == 0, "left must be on TOS");
2211 assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
2212
2213 Address raddr;
2214 if (right->is_double_stack()) {
2215 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2216 } else if (right->is_constant()) {
2217 // hack for now
2218 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2219 } else {
2220 ShouldNotReachHere();
2221 }
2222
2223 switch (code) {
2224 case lir_add: __ fadd_d(raddr); break;
2225 case lir_sub: __ fsub_d(raddr); break;
2226 case lir_mul_strictfp: // fall through
2227 case lir_mul: __ fmul_d(raddr); break;
2228 case lir_div_strictfp: // fall through
2229 case lir_div: __ fdiv_d(raddr); break;
2230 default: ShouldNotReachHere();
2231 }
2232 }
2233
2234 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2235 // Double values require special handling for strictfp mul/div on x86
2236 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
2237 __ fmulp(dest->fpu_regnrLo() + 1);
2238 }
2239
2240 } else if (left->is_single_stack() || left->is_address()) {
2241 assert(left == dest, "left and dest must be equal");
2242
2243 Address laddr;
2244 if (left->is_single_stack()) {
2245 laddr = frame_map()->address_for_slot(left->single_stack_ix());
2246 } else if (left->is_address()) {
2247 laddr = as_Address(left->as_address_ptr());
2248 } else {
2249 ShouldNotReachHere();
2250 }
2251
2252 if (right->is_single_cpu()) {
2253 Register rreg = right->as_register();
2254 switch (code) {
2255 case lir_add: __ addl(laddr, rreg); break;
2256 case lir_sub: __ subl(laddr, rreg); break;
2257 default: ShouldNotReachHere();
2258 }
2259 } else if (right->is_constant()) {
2260 jint c = right->as_constant_ptr()->as_jint();
2261 switch (code) {
2262 case lir_add: {
2263 __ incrementl(laddr, c);
2264 break;
2265 }
2266 case lir_sub: {
2267 __ decrementl(laddr, c);
2268 break;
2269 }
2270 default: ShouldNotReachHere();
2271 }
2272 } else {
2273 ShouldNotReachHere();
2274 }
2275
2276 } else {
2277 ShouldNotReachHere();
2278 }
2279 }
2280
2281 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
2282 assert(pop_fpu_stack || (left_index == dest_index || right_index == dest_index), "invalid LIR");
2283 assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
2284 assert(left_index == 0 || right_index == 0, "either must be on top of stack");
2285
2286 bool left_is_tos = (left_index == 0);
2287 bool dest_is_tos = (dest_index == 0);
2288 int non_tos_index = (left_is_tos ? right_index : left_index);
2289
2290 switch (code) {
2291 case lir_add:
2292 if (pop_fpu_stack) __ faddp(non_tos_index);
2293 else if (dest_is_tos) __ fadd (non_tos_index);
2294 else __ fadda(non_tos_index);
2295 break;
2296
2297 case lir_sub:
2298 if (left_is_tos) {
2299 if (pop_fpu_stack) __ fsubrp(non_tos_index);
2300 else if (dest_is_tos) __ fsub (non_tos_index);
2301 else __ fsubra(non_tos_index);
2302 } else {
2303 if (pop_fpu_stack) __ fsubp (non_tos_index);
2304 else if (dest_is_tos) __ fsubr (non_tos_index);
2305 else __ fsuba (non_tos_index);
2306 }
2307 break;
2308
2309 case lir_mul_strictfp: // fall through
2310 case lir_mul:
2311 if (pop_fpu_stack) __ fmulp(non_tos_index);
2312 else if (dest_is_tos) __ fmul (non_tos_index);
2313 else __ fmula(non_tos_index);
2314 break;
2315
2316 case lir_div_strictfp: // fall through
2317 case lir_div:
2318 if (left_is_tos) {
2319 if (pop_fpu_stack) __ fdivrp(non_tos_index);
2320 else if (dest_is_tos) __ fdiv (non_tos_index);
2321 else __ fdivra(non_tos_index);
2322 } else {
2323 if (pop_fpu_stack) __ fdivp (non_tos_index);
2324 else if (dest_is_tos) __ fdivr (non_tos_index);
2325 else __ fdiva (non_tos_index);
2326 }
2327 break;
2328
2329 case lir_rem:
2330 assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
2331 __ fremr(noreg);
2332 break;
2333
2334 default:
2335 ShouldNotReachHere();
2336 }
2337 }
2338
2339
2340 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
2341 if (value->is_double_xmm()) {
2342 switch(code) {
2343 case lir_abs :
2344 {
2345 if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
2346 __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
2347 }
2348 __ andpd(dest->as_xmm_double_reg(),
2349 ExternalAddress((address)double_signmask_pool));
2350 }
2351 break;
2352
2353 case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
2354 // all other intrinsics are not available in the SSE instruction set, so FPU is used
2355 default : ShouldNotReachHere();
2356 }
2357
2358 } else if (value->is_double_fpu()) {
2359 assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
2360 switch(code) {
2361 case lir_log : __ flog() ; break;
2362 case lir_log10 : __ flog10() ; break;
2363 case lir_abs : __ fabs() ; break;
2364 case lir_sqrt : __ fsqrt(); break;
2365 case lir_sin :
2366 // Should consider not saving rbx, if not necessary
2367 __ trigfunc('s', op->as_Op2()->fpu_stack_size());
2368 break;
2369 case lir_cos :
2370 // Should consider not saving rbx, if not necessary
2371 assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
2372 __ trigfunc('c', op->as_Op2()->fpu_stack_size());
2373 break;
2374 case lir_tan :
2375 // Should consider not saving rbx, if not necessary
2376 __ trigfunc('t', op->as_Op2()->fpu_stack_size());
2377 break;
2378 default : ShouldNotReachHere();
2379 }
2380 } else {
2381 Unimplemented();
2382 }
2383 }
2384
2385 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
2386 // assert(left->destroys_register(), "check");
2387 if (left->is_single_cpu()) {
2388 Register reg = left->as_register();
2389 if (right->is_constant()) {
2390 int val = right->as_constant_ptr()->as_jint();
2391 switch (code) {
2392 case lir_logic_and: __ andl (reg, val); break;
2393 case lir_logic_or: __ orl (reg, val); break;
2394 case lir_logic_xor: __ xorl (reg, val); break;
2395 default: ShouldNotReachHere();
2396 }
2397 } else if (right->is_stack()) {
2398 // added support for stack operands
2399 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2400 switch (code) {
2401 case lir_logic_and: __ andl (reg, raddr); break;
2402 case lir_logic_or: __ orl (reg, raddr); break;
2403 case lir_logic_xor: __ xorl (reg, raddr); break;
2404 default: ShouldNotReachHere();
2405 }
2406 } else {
2407 Register rright = right->as_register();
2408 switch (code) {
2409 case lir_logic_and: __ andptr (reg, rright); break;
2410 case lir_logic_or : __ orptr (reg, rright); break;
2411 case lir_logic_xor: __ xorptr (reg, rright); break;
2412 default: ShouldNotReachHere();
2413 }
2414 }
2415 move_regs(reg, dst->as_register());
2416 } else {
2417 Register l_lo = left->as_register_lo();
2418 Register l_hi = left->as_register_hi();
2419 if (right->is_constant()) {
2420 #ifdef _LP64
2421 __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
2422 switch (code) {
2423 case lir_logic_and:
2424 __ andq(l_lo, rscratch1);
2425 break;
2426 case lir_logic_or:
2427 __ orq(l_lo, rscratch1);
2428 break;
2429 case lir_logic_xor:
2430 __ xorq(l_lo, rscratch1);
2431 break;
2432 default: ShouldNotReachHere();
2433 }
2434 #else
2435 int r_lo = right->as_constant_ptr()->as_jint_lo();
2436 int r_hi = right->as_constant_ptr()->as_jint_hi();
2437 switch (code) {
2438 case lir_logic_and:
2439 __ andl(l_lo, r_lo);
2440 __ andl(l_hi, r_hi);
2441 break;
2442 case lir_logic_or:
2443 __ orl(l_lo, r_lo);
2444 __ orl(l_hi, r_hi);
2445 break;
2446 case lir_logic_xor:
2447 __ xorl(l_lo, r_lo);
2448 __ xorl(l_hi, r_hi);
2449 break;
2450 default: ShouldNotReachHere();
2451 }
2452 #endif // _LP64
2453 } else {
2454 Register r_lo = right->as_register_lo();
2455 Register r_hi = right->as_register_hi();
2456 assert(l_lo != r_hi, "overwriting registers");
2457 switch (code) {
2458 case lir_logic_and:
2459 __ andptr(l_lo, r_lo);
2460 NOT_LP64(__ andptr(l_hi, r_hi);)
2461 break;
2462 case lir_logic_or:
2463 __ orptr(l_lo, r_lo);
2464 NOT_LP64(__ orptr(l_hi, r_hi);)
2465 break;
2466 case lir_logic_xor:
2467 __ xorptr(l_lo, r_lo);
2468 NOT_LP64(__ xorptr(l_hi, r_hi);)
2469 break;
2470 default: ShouldNotReachHere();
2471 }
2472 }
2473
2474 Register dst_lo = dst->as_register_lo();
2475 Register dst_hi = dst->as_register_hi();
2476
2477 #ifdef _LP64
2478 move_regs(l_lo, dst_lo);
2479 #else
2480 if (dst_lo == l_hi) {
2481 assert(dst_hi != l_lo, "overwriting registers");
2482 move_regs(l_hi, dst_hi);
2483 move_regs(l_lo, dst_lo);
2484 } else {
2485 assert(dst_lo != l_hi, "overwriting registers");
2486 move_regs(l_lo, dst_lo);
2487 move_regs(l_hi, dst_hi);
2488 }
2489 #endif // _LP64
2490 }
2491 }
2492
2493
2494 // we assume that rax, and rdx can be overwritten
2495 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
2496
2497 assert(left->is_single_cpu(), "left must be register");
2498 assert(right->is_single_cpu() || right->is_constant(), "right must be register or constant");
2499 assert(result->is_single_cpu(), "result must be register");
2500
2501 // assert(left->destroys_register(), "check");
2502 // assert(right->destroys_register(), "check");
2503
2504 Register lreg = left->as_register();
2505 Register dreg = result->as_register();
2506
2507 if (right->is_constant()) {
2508 int divisor = right->as_constant_ptr()->as_jint();
2509 assert(divisor > 0 && is_power_of_2(divisor), "must be");
2510 if (code == lir_idiv) {
2511 assert(lreg == rax, "must be rax,");
2512 assert(temp->as_register() == rdx, "tmp register must be rdx");
2513 __ cdql(); // sign extend into rdx:rax
2514 if (divisor == 2) {
2515 __ subl(lreg, rdx);
2516 } else {
2517 __ andl(rdx, divisor - 1);
2518 __ addl(lreg, rdx);
2519 }
2520 __ sarl(lreg, log2_intptr(divisor));
2521 move_regs(lreg, dreg);
2522 } else if (code == lir_irem) {
2523 Label done;
2524 __ mov(dreg, lreg);
2525 __ andl(dreg, 0x80000000 | (divisor - 1));
2526 __ jcc(Assembler::positive, done);
2527 __ decrement(dreg);
2528 __ orl(dreg, ~(divisor - 1));
2529 __ increment(dreg);
2530 __ bind(done);
2531 } else {
2532 ShouldNotReachHere();
2533 }
2534 } else {
2535 Register rreg = right->as_register();
2536 assert(lreg == rax, "left register must be rax,");
2537 assert(rreg != rdx, "right register must not be rdx");
2538 assert(temp->as_register() == rdx, "tmp register must be rdx");
2539
2540 move_regs(lreg, rax);
2541
2542 int idivl_offset = __ corrected_idivl(rreg);
2543 add_debug_info_for_div0(idivl_offset, info);
2544 if (code == lir_irem) {
2545 move_regs(rdx, dreg); // result is in rdx
2546 } else {
2547 move_regs(rax, dreg);
2548 }
2549 }
2550 }
2551
2552
2553 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
2554 if (opr1->is_single_cpu()) {
2555 Register reg1 = opr1->as_register();
2556 if (opr2->is_single_cpu()) {
2557 // cpu register - cpu register
2558 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2559 __ cmpptr(reg1, opr2->as_register());
2560 } else {
2561 assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
2562 __ cmpl(reg1, opr2->as_register());
2563 }
2564 } else if (opr2->is_stack()) {
2565 // cpu register - stack
2566 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2567 __ cmpptr(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2568 } else {
2569 __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2570 }
2571 } else if (opr2->is_constant()) {
2572 // cpu register - constant
2573 LIR_Const* c = opr2->as_constant_ptr();
2574 if (c->type() == T_INT) {
2575 __ cmpl(reg1, c->as_jint());
2576 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2577 // In 64bit oops are single register
2578 jobject o = c->as_jobject();
2579 if (o == NULL) {
2580 __ cmpptr(reg1, (int32_t)NULL_WORD);
2581 } else {
2582 #ifdef _LP64
2583 __ movoop(rscratch1, o);
2584 __ cmpptr(reg1, rscratch1);
2585 #else
2586 __ cmpoop(reg1, c->as_jobject());
2587 #endif // _LP64
2588 }
2589 } else {
2590 ShouldNotReachHere();
2591 }
2592 // cpu register - address
2593 } else if (opr2->is_address()) {
2594 if (op->info() != NULL) {
2595 add_debug_info_for_null_check_here(op->info());
2596 }
2597 __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
2598 } else {
2599 ShouldNotReachHere();
2600 }
2601
2602 } else if(opr1->is_double_cpu()) {
2603 Register xlo = opr1->as_register_lo();
2604 Register xhi = opr1->as_register_hi();
2605 if (opr2->is_double_cpu()) {
2606 #ifdef _LP64
2607 __ cmpptr(xlo, opr2->as_register_lo());
2608 #else
2609 // cpu register - cpu register
2610 Register ylo = opr2->as_register_lo();
2611 Register yhi = opr2->as_register_hi();
2612 __ subl(xlo, ylo);
2613 __ sbbl(xhi, yhi);
2614 if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
2615 __ orl(xhi, xlo);
2616 }
2617 #endif // _LP64
2618 } else if (opr2->is_constant()) {
2619 // cpu register - constant 0
2620 assert(opr2->as_jlong() == (jlong)0, "only handles zero");
2621 #ifdef _LP64
2622 __ cmpptr(xlo, (int32_t)opr2->as_jlong());
2623 #else
2624 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
2625 __ orl(xhi, xlo);
2626 #endif // _LP64
2627 } else {
2628 ShouldNotReachHere();
2629 }
2630
2631 } else if (opr1->is_single_xmm()) {
2632 XMMRegister reg1 = opr1->as_xmm_float_reg();
2633 if (opr2->is_single_xmm()) {
2634 // xmm register - xmm register
2635 __ ucomiss(reg1, opr2->as_xmm_float_reg());
2636 } else if (opr2->is_stack()) {
2637 // xmm register - stack
2638 __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2639 } else if (opr2->is_constant()) {
2640 // xmm register - constant
2641 __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
2642 } else if (opr2->is_address()) {
2643 // xmm register - address
2644 if (op->info() != NULL) {
2645 add_debug_info_for_null_check_here(op->info());
2646 }
2647 __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
2648 } else {
2649 ShouldNotReachHere();
2650 }
2651
2652 } else if (opr1->is_double_xmm()) {
2653 XMMRegister reg1 = opr1->as_xmm_double_reg();
2654 if (opr2->is_double_xmm()) {
2655 // xmm register - xmm register
2656 __ ucomisd(reg1, opr2->as_xmm_double_reg());
2657 } else if (opr2->is_stack()) {
2658 // xmm register - stack
2659 __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
2660 } else if (opr2->is_constant()) {
2661 // xmm register - constant
2662 __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
2663 } else if (opr2->is_address()) {
2664 // xmm register - address
2665 if (op->info() != NULL) {
2666 add_debug_info_for_null_check_here(op->info());
2667 }
2668 __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
2669 } else {
2670 ShouldNotReachHere();
2671 }
2672
2673 } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
2674 assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
2675 assert(opr2->is_fpu_register(), "both must be registers");
2676 __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2677
2678 } else if (opr1->is_address() && opr2->is_constant()) {
2679 LIR_Const* c = opr2->as_constant_ptr();
2680 #ifdef _LP64
2681 if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2682 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
2683 __ movoop(rscratch1, c->as_jobject());
2684 }
2685 #endif // LP64
2686 if (op->info() != NULL) {
2687 add_debug_info_for_null_check_here(op->info());
2688 }
2689 // special case: address - constant
2690 LIR_Address* addr = opr1->as_address_ptr();
2691 if (c->type() == T_INT) {
2692 __ cmpl(as_Address(addr), c->as_jint());
2693 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2694 #ifdef _LP64
2695 // %%% Make this explode if addr isn't reachable until we figure out a
2696 // better strategy by giving noreg as the temp for as_Address
2697 __ cmpptr(rscratch1, as_Address(addr, noreg));
2698 #else
2699 __ cmpoop(as_Address(addr), c->as_jobject());
2700 #endif // _LP64
2701 } else {
2702 ShouldNotReachHere();
2703 }
2704
2705 } else {
2706 ShouldNotReachHere();
2707 }
2708 }
2709
2710 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
2711 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
2712 if (left->is_single_xmm()) {
2713 assert(right->is_single_xmm(), "must match");
2714 __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2715 } else if (left->is_double_xmm()) {
2716 assert(right->is_double_xmm(), "must match");
2717 __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2718
2719 } else {
2720 assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
2721 assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
2722
2723 assert(left->fpu() == 0, "left must be on TOS");
2724 __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
2725 op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2726 }
2727 } else {
2728 assert(code == lir_cmp_l2i, "check");
2729 #ifdef _LP64
2730 Register dest = dst->as_register();
2731 __ xorptr(dest, dest);
2732 Label high, done;
2733 __ cmpptr(left->as_register_lo(), right->as_register_lo());
2734 __ jcc(Assembler::equal, done);
2735 __ jcc(Assembler::greater, high);
2736 __ decrement(dest);
2737 __ jmp(done);
2738 __ bind(high);
2739 __ increment(dest);
2740
2741 __ bind(done);
2742
2743 #else
2744 __ lcmp2int(left->as_register_hi(),
2745 left->as_register_lo(),
2746 right->as_register_hi(),
2747 right->as_register_lo());
2748 move_regs(left->as_register_hi(), dst->as_register());
2749 #endif // _LP64
2750 }
2751 }
2752
2753
2754 void LIR_Assembler::align_call(LIR_Code code) {
2755 if (os::is_MP()) {
2756 // make sure that the displacement word of the call ends up word aligned
2757 int offset = __ offset();
2758 switch (code) {
2759 case lir_static_call:
2760 case lir_optvirtual_call:
2761 offset += NativeCall::displacement_offset;
2762 break;
2763 case lir_icvirtual_call:
2764 offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
2765 break;
2766 case lir_virtual_call: // currently, sparc-specific for niagara
2767 default: ShouldNotReachHere();
2768 }
2769 while (offset++ % BytesPerWord != 0) {
2770 __ nop();
2771 }
2772 }
2773 }
2774
2775
2776 void LIR_Assembler::call(address entry, relocInfo::relocType rtype, CodeEmitInfo* info) {
2777 assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2778 "must be aligned");
2779 __ call(AddressLiteral(entry, rtype));
2780 add_call_info(code_offset(), info);
2781 }
2782
2783
2784 void LIR_Assembler::ic_call(address entry, CodeEmitInfo* info) {
2785 RelocationHolder rh = virtual_call_Relocation::spec(pc());
2786 __ movoop(IC_Klass, (jobject)Universe::non_oop_word());
2787 assert(!os::is_MP() ||
2788 (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2789 "must be aligned");
2790 __ call(AddressLiteral(entry, rh));
2791 add_call_info(code_offset(), info);
2792 }
2793
2794
2795 /* Currently, vtable-dispatch is only enabled for sparc platforms */
2796 void LIR_Assembler::vtable_call(int vtable_offset, CodeEmitInfo* info) {
2797 ShouldNotReachHere();
2798 }
2799
2800 void LIR_Assembler::emit_static_call_stub() {
2801 address call_pc = __ pc();
2802 address stub = __ start_a_stub(call_stub_size);
2803 if (stub == NULL) {
2804 bailout("static call stub overflow");
2805 return;
2806 }
2807
2808 int start = __ offset();
2809 if (os::is_MP()) {
2810 // make sure that the displacement word of the call ends up word aligned
2811 int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
2812 while (offset++ % BytesPerWord != 0) {
2813 __ nop();
2814 }
2815 }
2816 __ relocate(static_stub_Relocation::spec(call_pc));
2817 __ movoop(rbx, (jobject)NULL);
2818 // must be set to -1 at code generation time
2819 assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
2820 // On 64bit this will die since it will take a movq & jmp, must be only a jmp
2821 __ jump(RuntimeAddress(__ pc()));
2822
2823 assert(__ offset() - start <= call_stub_size, "stub too big")
2824 __ end_a_stub();
2825 }
2826
2827
2828 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info, bool unwind) {
2829 assert(exceptionOop->as_register() == rax, "must match");
2830 assert(unwind || exceptionPC->as_register() == rdx, "must match");
2831
2832 // exception object is not added to oop map by LinearScan
2833 // (LinearScan assumes that no oops are in fixed registers)
2834 info->add_register_oop(exceptionOop);
2835 Runtime1::StubID unwind_id;
2836
2837 if (!unwind) {
2838 // get current pc information
2839 // pc is only needed if the method has an exception handler, the unwind code does not need it.
2840 int pc_for_athrow_offset = __ offset();
2841 InternalAddress pc_for_athrow(__ pc());
2842 __ lea(exceptionPC->as_register(), pc_for_athrow);
2843 add_call_info(pc_for_athrow_offset, info); // for exception handler
2844
2845 __ verify_not_null_oop(rax);
2846 // search an exception handler (rax: exception oop, rdx: throwing pc)
2847 if (compilation()->has_fpu_code()) {
2848 unwind_id = Runtime1::handle_exception_id;
2849 } else {
2850 unwind_id = Runtime1::handle_exception_nofpu_id;
2851 }
2852 } else {
2853 unwind_id = Runtime1::unwind_exception_id;
2854 }
2855 __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2856
2857 // enough room for two byte trap
2858 __ nop();
2859 }
2860
2861
2862 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2863
2864 // optimized version for linear scan:
2865 // * count must be already in ECX (guaranteed by LinearScan)
2866 // * left and dest must be equal
2867 // * tmp must be unused
2868 assert(count->as_register() == SHIFT_count, "count must be in ECX");
2869 assert(left == dest, "left and dest must be equal");
2870 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2871
2872 if (left->is_single_cpu()) {
2873 Register value = left->as_register();
2874 assert(value != SHIFT_count, "left cannot be ECX");
2875
2876 switch (code) {
2877 case lir_shl: __ shll(value); break;
2878 case lir_shr: __ sarl(value); break;
2879 case lir_ushr: __ shrl(value); break;
2880 default: ShouldNotReachHere();
2881 }
2882 } else if (left->is_double_cpu()) {
2883 Register lo = left->as_register_lo();
2884 Register hi = left->as_register_hi();
2885 assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
2886 #ifdef _LP64
2887 switch (code) {
2888 case lir_shl: __ shlptr(lo); break;
2889 case lir_shr: __ sarptr(lo); break;
2890 case lir_ushr: __ shrptr(lo); break;
2891 default: ShouldNotReachHere();
2892 }
2893 #else
2894
2895 switch (code) {
2896 case lir_shl: __ lshl(hi, lo); break;
2897 case lir_shr: __ lshr(hi, lo, true); break;
2898 case lir_ushr: __ lshr(hi, lo, false); break;
2899 default: ShouldNotReachHere();
2900 }
2901 #endif // LP64
2902 } else {
2903 ShouldNotReachHere();
2904 }
2905 }
2906
2907
2908 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2909 if (dest->is_single_cpu()) {
2910 // first move left into dest so that left is not destroyed by the shift
2911 Register value = dest->as_register();
2912 count = count & 0x1F; // Java spec
2913
2914 move_regs(left->as_register(), value);
2915 switch (code) {
2916 case lir_shl: __ shll(value, count); break;
2917 case lir_shr: __ sarl(value, count); break;
2918 case lir_ushr: __ shrl(value, count); break;
2919 default: ShouldNotReachHere();
2920 }
2921 } else if (dest->is_double_cpu()) {
2922 #ifndef _LP64
2923 Unimplemented();
2924 #else
2925 // first move left into dest so that left is not destroyed by the shift
2926 Register value = dest->as_register_lo();
2927 count = count & 0x1F; // Java spec
2928
2929 move_regs(left->as_register_lo(), value);
2930 switch (code) {
2931 case lir_shl: __ shlptr(value, count); break;
2932 case lir_shr: __ sarptr(value, count); break;
2933 case lir_ushr: __ shrptr(value, count); break;
2934 default: ShouldNotReachHere();
2935 }
2936 #endif // _LP64
2937 } else {
2938 ShouldNotReachHere();
2939 }
2940 }
2941
2942
2943 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
2944 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
2945 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
2946 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2947 __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
2948 }
2949
2950
2951 void LIR_Assembler::store_parameter(jint c, int offset_from_rsp_in_words) {
2952 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
2953 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
2954 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2955 __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
2956 }
2957
2958
2959 void LIR_Assembler::store_parameter(jobject o, int offset_from_rsp_in_words) {
2960 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
2961 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
2962 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2963 __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
2964 }
2965
2966
2967 // This code replaces a call to arraycopy; no exception may
2968 // be thrown in this code, they must be thrown in the System.arraycopy
2969 // activation frame; we could save some checks if this would not be the case
2970 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
2971 ciArrayKlass* default_type = op->expected_type();
2972 Register src = op->src()->as_register();
2973 Register dst = op->dst()->as_register();
2974 Register src_pos = op->src_pos()->as_register();
2975 Register dst_pos = op->dst_pos()->as_register();
2976 Register length = op->length()->as_register();
2977 Register tmp = op->tmp()->as_register();
2978
2979 CodeStub* stub = op->stub();
2980 int flags = op->flags();
2981 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
2982 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
2983
2984 // if we don't know anything or it's an object array, just go through the generic arraycopy
2985 if (default_type == NULL) {
2986 Label done;
2987 // save outgoing arguments on stack in case call to System.arraycopy is needed
2988 // HACK ALERT. This code used to push the parameters in a hardwired fashion
2989 // for interpreter calling conventions. Now we have to do it in new style conventions.
2990 // For the moment until C1 gets the new register allocator I just force all the
2991 // args to the right place (except the register args) and then on the back side
2992 // reload the register args properly if we go slow path. Yuck
2993
2994 // These are proper for the calling convention
2995
2996 store_parameter(length, 2);
2997 store_parameter(dst_pos, 1);
2998 store_parameter(dst, 0);
2999
3000 // these are just temporary placements until we need to reload
3001 store_parameter(src_pos, 3);
3002 store_parameter(src, 4);
3003 NOT_LP64(assert(src == rcx && src_pos == rdx, "mismatch in calling convention");)
3004
3005 address entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
3006
3007 // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
3008 #ifdef _LP64
3009 // The arguments are in java calling convention so we can trivially shift them to C
3010 // convention
3011 assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
3012 __ mov(c_rarg0, j_rarg0);
3013 assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
3014 __ mov(c_rarg1, j_rarg1);
3015 assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
3016 __ mov(c_rarg2, j_rarg2);
3017 assert_different_registers(c_rarg3, j_rarg4);
3018 __ mov(c_rarg3, j_rarg3);
3019 #ifdef _WIN64
3020 // Allocate abi space for args but be sure to keep stack aligned
3021 __ subptr(rsp, 6*wordSize);
3022 store_parameter(j_rarg4, 4);
3023 __ call(RuntimeAddress(entry));
3024 __ addptr(rsp, 6*wordSize);
3025 #else
3026 __ mov(c_rarg4, j_rarg4);
3027 __ call(RuntimeAddress(entry));
3028 #endif // _WIN64
3029 #else
3030 __ push(length);
3031 __ push(dst_pos);
3032 __ push(dst);
3033 __ push(src_pos);
3034 __ push(src);
3035 __ call_VM_leaf(entry, 5); // removes pushed parameter from the stack
3036
3037 #endif // _LP64
3038
3039 __ cmpl(rax, 0);
3040 __ jcc(Assembler::equal, *stub->continuation());
3041
3042 // Reload values from the stack so they are where the stub
3043 // expects them.
3044 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3045 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3046 __ movptr (length, Address(rsp, 2*BytesPerWord));
3047 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3048 __ movptr (src, Address(rsp, 4*BytesPerWord));
3049 __ jmp(*stub->entry());
3050
3051 __ bind(*stub->continuation());
3052 return;
3053 }
3054
3055 assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
3056
3057 int elem_size = type2aelembytes(basic_type);
3058 int shift_amount;
3059 Address::ScaleFactor scale;
3060
3061 switch (elem_size) {
3062 case 1 :
3063 shift_amount = 0;
3064 scale = Address::times_1;
3065 break;
3066 case 2 :
3067 shift_amount = 1;
3068 scale = Address::times_2;
3069 break;
3070 case 4 :
3071 shift_amount = 2;
3072 scale = Address::times_4;
3073 break;
3074 case 8 :
3075 shift_amount = 3;
3076 scale = Address::times_8;
3077 break;
3078 default:
3079 ShouldNotReachHere();
3080 }
3081
3082 Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
3083 Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
3084 Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
3085 Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
3086
3087 // length and pos's are all sign extended at this point on 64bit
3088
3089 // test for NULL
3090 if (flags & LIR_OpArrayCopy::src_null_check) {
3091 __ testptr(src, src);
3092 __ jcc(Assembler::zero, *stub->entry());
3093 }
3094 if (flags & LIR_OpArrayCopy::dst_null_check) {
3095 __ testptr(dst, dst);
3096 __ jcc(Assembler::zero, *stub->entry());
3097 }
3098
3099 // check if negative
3100 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
3101 __ testl(src_pos, src_pos);
3102 __ jcc(Assembler::less, *stub->entry());
3103 }
3104 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
3105 __ testl(dst_pos, dst_pos);
3106 __ jcc(Assembler::less, *stub->entry());
3107 }
3108 if (flags & LIR_OpArrayCopy::length_positive_check) {
3109 __ testl(length, length);
3110 __ jcc(Assembler::less, *stub->entry());
3111 }
3112
3113 if (flags & LIR_OpArrayCopy::src_range_check) {
3114 __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
3115 __ cmpl(tmp, src_length_addr);
3116 __ jcc(Assembler::above, *stub->entry());
3117 }
3118 if (flags & LIR_OpArrayCopy::dst_range_check) {
3119 __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
3120 __ cmpl(tmp, dst_length_addr);
3121 __ jcc(Assembler::above, *stub->entry());
3122 }
3123
3124 if (flags & LIR_OpArrayCopy::type_check) {
3125 __ movptr(tmp, src_klass_addr);
3126 __ cmpptr(tmp, dst_klass_addr);
3127 __ jcc(Assembler::notEqual, *stub->entry());
3128 }
3129
3130 #ifdef ASSERT
3131 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
3132 // Sanity check the known type with the incoming class. For the
3133 // primitive case the types must match exactly with src.klass and
3134 // dst.klass each exactly matching the default type. For the
3135 // object array case, if no type check is needed then either the
3136 // dst type is exactly the expected type and the src type is a
3137 // subtype which we can't check or src is the same array as dst
3138 // but not necessarily exactly of type default_type.
3139 Label known_ok, halt;
3140 __ movoop(tmp, default_type->encoding());
3141 if (basic_type != T_OBJECT) {
3142 __ cmpptr(tmp, dst_klass_addr);
3143 __ jcc(Assembler::notEqual, halt);
3144 __ cmpptr(tmp, src_klass_addr);
3145 __ jcc(Assembler::equal, known_ok);
3146 } else {
3147 __ cmpptr(tmp, dst_klass_addr);
3148 __ jcc(Assembler::equal, known_ok);
3149 __ cmpptr(src, dst);
3150 __ jcc(Assembler::equal, known_ok);
3151 }
3152 __ bind(halt);
3153 __ stop("incorrect type information in arraycopy");
3154 __ bind(known_ok);
3155 }
3156 #endif
3157
3158 if (shift_amount > 0 && basic_type != T_OBJECT) {
3159 __ shlptr(length, shift_amount);
3160 }
3161
3162 #ifdef _LP64
3163 assert_different_registers(c_rarg0, dst, dst_pos, length);
3164 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3165 assert_different_registers(c_rarg1, length);
3166 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3167 __ mov(c_rarg2, length);
3168
3169 #else
3170 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3171 store_parameter(tmp, 0);
3172 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3173 store_parameter(tmp, 1);
3174 store_parameter(length, 2);
3175 #endif // _LP64
3176 if (basic_type == T_OBJECT) {
3177 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::oop_arraycopy), 0);
3178 } else {
3179 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::primitive_arraycopy), 0);
3180 }
3181
3182 __ bind(*stub->continuation());
3183 }
3184
3185
3186 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
3187 Register obj = op->obj_opr()->as_register(); // may not be an oop
3188 Register hdr = op->hdr_opr()->as_register();
3189 Register lock = op->lock_opr()->as_register();
3190 if (!UseFastLocking) {
3191 __ jmp(*op->stub()->entry());
3192 } else if (op->code() == lir_lock) {
3193 Register scratch = noreg;
3194 if (UseBiasedLocking) {
3195 scratch = op->scratch_opr()->as_register();
3196 }
3197 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3198 // add debug info for NullPointerException only if one is possible
3199 int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
3200 if (op->info() != NULL) {
3201 add_debug_info_for_null_check(null_check_offset, op->info());
3202 }
3203 // done
3204 } else if (op->code() == lir_unlock) {
3205 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3206 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
3207 } else {
3208 Unimplemented();
3209 }
3210 __ bind(*op->stub()->continuation());
3211 }
3212
3213
3214 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
3215 ciMethod* method = op->profiled_method();
3216 int bci = op->profiled_bci();
3217
3218 // Update counter for all call types
3219 ciMethodData* md = method->method_data();
3220 if (md == NULL) {
3221 bailout("out of memory building methodDataOop");
3222 return;
3223 }
3224 ciProfileData* data = md->bci_to_data(bci);
3225 assert(data->is_CounterData(), "need CounterData for calls");
3226 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
3227 Register mdo = op->mdo()->as_register();
3228 __ movoop(mdo, md->encoding());
3229 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
3230 __ addl(counter_addr, DataLayout::counter_increment);
3231 Bytecodes::Code bc = method->java_code_at_bci(bci);
3232 // Perform additional virtual call profiling for invokevirtual and
3233 // invokeinterface bytecodes
3234 if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3235 Tier1ProfileVirtualCalls) {
3236 assert(op->recv()->is_single_cpu(), "recv must be allocated");
3237 Register recv = op->recv()->as_register();
3238 assert_different_registers(mdo, recv);
3239 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3240 ciKlass* known_klass = op->known_holder();
3241 if (Tier1OptimizeVirtualCallProfiling && known_klass != NULL) {
3242 // We know the type that will be seen at this call site; we can
3243 // statically update the methodDataOop rather than needing to do
3244 // dynamic tests on the receiver type
3245
3246 // NOTE: we should probably put a lock around this search to
3247 // avoid collisions by concurrent compilations
3248 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3249 uint i;
3250 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3251 ciKlass* receiver = vc_data->receiver(i);
3252 if (known_klass->equals(receiver)) {
3253 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3254 __ addl(data_addr, DataLayout::counter_increment);
3255 return;
3256 }
3257 }
3258
3259 // Receiver type not found in profile data; select an empty slot
3260
3261 // Note that this is less efficient than it should be because it
3262 // always does a write to the receiver part of the
3263 // VirtualCallData rather than just the first time
3264 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3265 ciKlass* receiver = vc_data->receiver(i);
3266 if (receiver == NULL) {
3267 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3268 __ movoop(recv_addr, known_klass->encoding());
3269 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3270 __ addl(data_addr, DataLayout::counter_increment);
3271 return;
3272 }
3273 }
3274 } else {
3275 __ movptr(recv, Address(recv, oopDesc::klass_offset_in_bytes()));
3276 Label update_done;
3277 uint i;
3278 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3279 Label next_test;
3280 // See if the receiver is receiver[n].
3281 __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i))));
3282 __ jcc(Assembler::notEqual, next_test);
3283 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3284 __ addl(data_addr, DataLayout::counter_increment);
3285 __ jmp(update_done);
3286 __ bind(next_test);
3287 }
3288
3289 // Didn't find receiver; find next empty slot and fill it in
3290 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3291 Label next_test;
3292 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3293 __ cmpptr(recv_addr, (int32_t)NULL_WORD);
3294 __ jcc(Assembler::notEqual, next_test);
3295 __ movptr(recv_addr, recv);
3296 __ movl(Address(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i))), DataLayout::counter_increment);
3297 if (i < (VirtualCallData::row_limit() - 1)) {
3298 __ jmp(update_done);
3299 }
3300 __ bind(next_test);
3301 }
3302
3303 __ bind(update_done);
3304 }
3305 }
3306 }
3307
3308
3309 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
3310 Unimplemented();
3311 }
3312
3313
3314 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
3315 __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
3316 }
3317
3318
3319 void LIR_Assembler::align_backward_branch_target() {
3320 __ align(BytesPerWord);
3321 }
3322
3323
3324 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3325 if (left->is_single_cpu()) {
3326 __ negl(left->as_register());
3327 move_regs(left->as_register(), dest->as_register());
3328
3329 } else if (left->is_double_cpu()) {
3330 Register lo = left->as_register_lo();
3331 #ifdef _LP64
3332 Register dst = dest->as_register_lo();
3333 __ movptr(dst, lo);
3334 __ negptr(dst);
3335 #else
3336 Register hi = left->as_register_hi();
3337 __ lneg(hi, lo);
3338 if (dest->as_register_lo() == hi) {
3339 assert(dest->as_register_hi() != lo, "destroying register");
3340 move_regs(hi, dest->as_register_hi());
3341 move_regs(lo, dest->as_register_lo());
3342 } else {
3343 move_regs(lo, dest->as_register_lo());
3344 move_regs(hi, dest->as_register_hi());
3345 }
3346 #endif // _LP64
3347
3348 } else if (dest->is_single_xmm()) {
3349 if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
3350 __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
3351 }
3352 __ xorps(dest->as_xmm_float_reg(),
3353 ExternalAddress((address)float_signflip_pool));
3354
3355 } else if (dest->is_double_xmm()) {
3356 if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
3357 __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
3358 }
3359 __ xorpd(dest->as_xmm_double_reg(),
3360 ExternalAddress((address)double_signflip_pool));
3361
3362 } else if (left->is_single_fpu() || left->is_double_fpu()) {
3363 assert(left->fpu() == 0, "arg must be on TOS");
3364 assert(dest->fpu() == 0, "dest must be TOS");
3365 __ fchs();
3366
3367 } else {
3368 ShouldNotReachHere();
3369 }
3370 }
3371
3372
3373 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
3374 assert(addr->is_address() && dest->is_register(), "check");
3375 Register reg;
3376 reg = dest->as_pointer_register();
3377 __ lea(reg, as_Address(addr->as_address_ptr()));
3378 }
3379
3380
3381
3382 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3383 assert(!tmp->is_valid(), "don't need temporary");
3384 __ call(RuntimeAddress(dest));
3385 if (info != NULL) {
3386 add_call_info_here(info);
3387 }
3388 }
3389
3390
3391 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3392 assert(type == T_LONG, "only for volatile long fields");
3393
3394 if (info != NULL) {
3395 add_debug_info_for_null_check_here(info);
3396 }
3397
3398 if (src->is_double_xmm()) {
3399 if (dest->is_double_cpu()) {
3400 #ifdef _LP64
3401 __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
3402 #else
3403 __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
3404 __ psrlq(src->as_xmm_double_reg(), 32);
3405 __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
3406 #endif // _LP64
3407 } else if (dest->is_double_stack()) {
3408 __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
3409 } else if (dest->is_address()) {
3410 __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
3411 } else {
3412 ShouldNotReachHere();
3413 }
3414
3415 } else if (dest->is_double_xmm()) {
3416 if (src->is_double_stack()) {
3417 __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
3418 } else if (src->is_address()) {
3419 __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
3420 } else {
3421 ShouldNotReachHere();
3422 }
3423
3424 } else if (src->is_double_fpu()) {
3425 assert(src->fpu_regnrLo() == 0, "must be TOS");
3426 if (dest->is_double_stack()) {
3427 __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
3428 } else if (dest->is_address()) {
3429 __ fistp_d(as_Address(dest->as_address_ptr()));
3430 } else {
3431 ShouldNotReachHere();
3432 }
3433
3434 } else if (dest->is_double_fpu()) {
3435 assert(dest->fpu_regnrLo() == 0, "must be TOS");
3436 if (src->is_double_stack()) {
3437 __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
3438 } else if (src->is_address()) {
3439 __ fild_d(as_Address(src->as_address_ptr()));
3440 } else {
3441 ShouldNotReachHere();
3442 }
3443 } else {
3444 ShouldNotReachHere();
3445 }
3446 }
3447
3448
3449 void LIR_Assembler::membar() {
3450 // QQQ sparc TSO uses this,
3451 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
3452 }
3453
3454 void LIR_Assembler::membar_acquire() {
3455 // No x86 machines currently require load fences
3456 // __ load_fence();
3457 }
3458
3459 void LIR_Assembler::membar_release() {
3460 // No x86 machines currently require store fences
3461 // __ store_fence();
3462 }
3463
3464 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3465 assert(result_reg->is_register(), "check");
3466 #ifdef _LP64
3467 // __ get_thread(result_reg->as_register_lo());
3468 __ mov(result_reg->as_register(), r15_thread);
3469 #else
3470 __ get_thread(result_reg->as_register());
3471 #endif // _LP64
3472 }
3473
3474
3475 void LIR_Assembler::peephole(LIR_List*) {
3476 // do nothing for now
3477 }
3478
3479
3480 #undef __