1 /*
2 * Copyright (c) 2000, 2011, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "c1/c1_Compilation.hpp"
28 #include "c1/c1_LIRAssembler.hpp"
29 #include "c1/c1_MacroAssembler.hpp"
30 #include "c1/c1_Runtime1.hpp"
31 #include "c1/c1_ValueStack.hpp"
32 #include "ci/ciArrayKlass.hpp"
33 #include "ci/ciInstance.hpp"
34 #include "gc_interface/collectedHeap.hpp"
35 #include "memory/barrierSet.hpp"
36 #include "memory/cardTableModRefBS.hpp"
37 #include "nativeInst_x86.hpp"
38 #include "oops/objArrayKlass.hpp"
39 #include "runtime/sharedRuntime.hpp"
40
41
42 // These masks are used to provide 128-bit aligned bitmasks to the XMM
43 // instructions, to allow sign-masking or sign-bit flipping. They allow
44 // fast versions of NegF/NegD and AbsF/AbsD.
45
46 // Note: 'double' and 'long long' have 32-bits alignment on x86.
47 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
48 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
49 // of 128-bits operands for SSE instructions.
50 jlong *operand = (jlong*)(((intptr_t)adr) & ((intptr_t)(~0xF)));
51 // Store the value to a 128-bits operand.
52 operand[0] = lo;
53 operand[1] = hi;
54 return operand;
55 }
56
57 // Buffer for 128-bits masks used by SSE instructions.
58 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
59
60 // Static initialization during VM startup.
61 static jlong *float_signmask_pool = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
62 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
63 static jlong *float_signflip_pool = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
64 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
65
66
67
68 NEEDS_CLEANUP // remove this definitions ?
69 const Register IC_Klass = rax; // where the IC klass is cached
70 const Register SYNC_header = rax; // synchronization header
71 const Register SHIFT_count = rcx; // where count for shift operations must be
72
73 #define __ _masm->
74
75
76 static void select_different_registers(Register preserve,
77 Register extra,
78 Register &tmp1,
79 Register &tmp2) {
80 if (tmp1 == preserve) {
81 assert_different_registers(tmp1, tmp2, extra);
82 tmp1 = extra;
83 } else if (tmp2 == preserve) {
84 assert_different_registers(tmp1, tmp2, extra);
85 tmp2 = extra;
86 }
87 assert_different_registers(preserve, tmp1, tmp2);
88 }
89
90
91
92 static void select_different_registers(Register preserve,
93 Register extra,
94 Register &tmp1,
95 Register &tmp2,
96 Register &tmp3) {
97 if (tmp1 == preserve) {
98 assert_different_registers(tmp1, tmp2, tmp3, extra);
99 tmp1 = extra;
100 } else if (tmp2 == preserve) {
101 assert_different_registers(tmp1, tmp2, tmp3, extra);
102 tmp2 = extra;
103 } else if (tmp3 == preserve) {
104 assert_different_registers(tmp1, tmp2, tmp3, extra);
105 tmp3 = extra;
106 }
107 assert_different_registers(preserve, tmp1, tmp2, tmp3);
108 }
109
110
111
112 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
113 if (opr->is_constant()) {
114 LIR_Const* constant = opr->as_constant_ptr();
115 switch (constant->type()) {
116 case T_INT: {
117 return true;
118 }
119
120 default:
121 return false;
122 }
123 }
124 return false;
125 }
126
127
128 LIR_Opr LIR_Assembler::receiverOpr() {
129 return FrameMap::receiver_opr;
130 }
131
132 LIR_Opr LIR_Assembler::incomingReceiverOpr() {
133 return receiverOpr();
134 }
135
136 LIR_Opr LIR_Assembler::osrBufferPointer() {
137 return FrameMap::as_pointer_opr(receiverOpr()->as_register());
138 }
139
140 //--------------fpu register translations-----------------------
141
142
143 address LIR_Assembler::float_constant(float f) {
144 address const_addr = __ float_constant(f);
145 if (const_addr == NULL) {
146 bailout("const section overflow");
147 return __ code()->consts()->start();
148 } else {
149 return const_addr;
150 }
151 }
152
153
154 address LIR_Assembler::double_constant(double d) {
155 address const_addr = __ double_constant(d);
156 if (const_addr == NULL) {
157 bailout("const section overflow");
158 return __ code()->consts()->start();
159 } else {
160 return const_addr;
161 }
162 }
163
164
165 void LIR_Assembler::set_24bit_FPU() {
166 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
167 }
168
169 void LIR_Assembler::reset_FPU() {
170 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
171 }
172
173 void LIR_Assembler::fpop() {
174 __ fpop();
175 }
176
177 void LIR_Assembler::fxch(int i) {
178 __ fxch(i);
179 }
180
181 void LIR_Assembler::fld(int i) {
182 __ fld_s(i);
183 }
184
185 void LIR_Assembler::ffree(int i) {
186 __ ffree(i);
187 }
188
189 void LIR_Assembler::breakpoint() {
190 __ int3();
191 }
192
193 void LIR_Assembler::push(LIR_Opr opr) {
194 if (opr->is_single_cpu()) {
195 __ push_reg(opr->as_register());
196 } else if (opr->is_double_cpu()) {
197 NOT_LP64(__ push_reg(opr->as_register_hi()));
198 __ push_reg(opr->as_register_lo());
199 } else if (opr->is_stack()) {
200 __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
201 } else if (opr->is_constant()) {
202 LIR_Const* const_opr = opr->as_constant_ptr();
203 if (const_opr->type() == T_OBJECT) {
204 __ push_oop(const_opr->as_jobject());
205 } else if (const_opr->type() == T_INT) {
206 __ push_jint(const_opr->as_jint());
207 } else {
208 ShouldNotReachHere();
209 }
210
211 } else {
212 ShouldNotReachHere();
213 }
214 }
215
216 void LIR_Assembler::pop(LIR_Opr opr) {
217 if (opr->is_single_cpu()) {
218 __ pop_reg(opr->as_register());
219 } else {
220 ShouldNotReachHere();
221 }
222 }
223
224 bool LIR_Assembler::is_literal_address(LIR_Address* addr) {
225 return addr->base()->is_illegal() && addr->index()->is_illegal();
226 }
227
228 //-------------------------------------------
229
230 Address LIR_Assembler::as_Address(LIR_Address* addr) {
231 return as_Address(addr, rscratch1);
232 }
233
234 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
235 if (addr->base()->is_illegal()) {
236 assert(addr->index()->is_illegal(), "must be illegal too");
237 AddressLiteral laddr((address)addr->disp(), relocInfo::none);
238 if (! __ reachable(laddr)) {
239 __ movptr(tmp, laddr.addr());
240 Address res(tmp, 0);
241 return res;
242 } else {
243 return __ as_Address(laddr);
244 }
245 }
246
247 Register base = addr->base()->as_pointer_register();
248
249 if (addr->index()->is_illegal()) {
250 return Address( base, addr->disp());
251 } else if (addr->index()->is_cpu_register()) {
252 Register index = addr->index()->as_pointer_register();
253 return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
254 } else if (addr->index()->is_constant()) {
255 intptr_t addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
256 assert(Assembler::is_simm32(addr_offset), "must be");
257
258 return Address(base, addr_offset);
259 } else {
260 Unimplemented();
261 return Address();
262 }
263 }
264
265
266 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
267 Address base = as_Address(addr);
268 return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
269 }
270
271
272 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
273 return as_Address(addr);
274 }
275
276
277 void LIR_Assembler::osr_entry() {
278 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
279 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
280 ValueStack* entry_state = osr_entry->state();
281 int number_of_locks = entry_state->locks_size();
282
283 // we jump here if osr happens with the interpreter
284 // state set up to continue at the beginning of the
285 // loop that triggered osr - in particular, we have
286 // the following registers setup:
287 //
288 // rcx: osr buffer
289 //
290
291 // build frame
292 ciMethod* m = compilation()->method();
293 __ build_frame(initial_frame_size_in_bytes());
294
295 // OSR buffer is
296 //
297 // locals[nlocals-1..0]
298 // monitors[0..number_of_locks]
299 //
300 // locals is a direct copy of the interpreter frame so in the osr buffer
301 // so first slot in the local array is the last local from the interpreter
302 // and last slot is local[0] (receiver) from the interpreter
303 //
304 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
305 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
306 // in the interpreter frame (the method lock if a sync method)
307
308 // Initialize monitors in the compiled activation.
309 // rcx: pointer to osr buffer
310 //
311 // All other registers are dead at this point and the locals will be
312 // copied into place by code emitted in the IR.
313
314 Register OSR_buf = osrBufferPointer()->as_pointer_register();
315 { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
316 int monitor_offset = BytesPerWord * method()->max_locals() +
317 (2 * BytesPerWord) * (number_of_locks - 1);
318 // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
319 // the OSR buffer using 2 word entries: first the lock and then
320 // the oop.
321 for (int i = 0; i < number_of_locks; i++) {
322 int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
323 #ifdef ASSERT
324 // verify the interpreter's monitor has a non-null object
325 {
326 Label L;
327 __ cmpptr(Address(OSR_buf, slot_offset + 1*BytesPerWord), (int32_t)NULL_WORD);
328 __ jcc(Assembler::notZero, L);
329 __ stop("locked object is NULL");
330 __ bind(L);
331 }
332 #endif
333 __ movptr(rbx, Address(OSR_buf, slot_offset + 0));
334 __ movptr(frame_map()->address_for_monitor_lock(i), rbx);
335 __ movptr(rbx, Address(OSR_buf, slot_offset + 1*BytesPerWord));
336 __ movptr(frame_map()->address_for_monitor_object(i), rbx);
337 }
338 }
339 }
340
341
342 // inline cache check; done before the frame is built.
343 int LIR_Assembler::check_icache() {
344 Register receiver = FrameMap::receiver_opr->as_register();
345 Register ic_klass = IC_Klass;
346 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
347 const bool do_post_padding = VerifyOops || UseCompressedOops;
348 if (!do_post_padding) {
349 // insert some nops so that the verified entry point is aligned on CodeEntryAlignment
350 while ((__ offset() + ic_cmp_size) % CodeEntryAlignment != 0) {
351 __ nop();
352 }
353 }
354 int offset = __ offset();
355 __ inline_cache_check(receiver, IC_Klass);
356 assert(__ offset() % CodeEntryAlignment == 0 || do_post_padding, "alignment must be correct");
357 if (do_post_padding) {
358 // force alignment after the cache check.
359 // It's been verified to be aligned if !VerifyOops
360 __ align(CodeEntryAlignment);
361 }
362 return offset;
363 }
364
365
366 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
367 jobject o = NULL;
368 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
369 __ movoop(reg, o);
370 patching_epilog(patch, lir_patch_normal, reg, info);
371 }
372
373
374 void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register new_hdr, int monitor_no, Register exception) {
375 if (exception->is_valid()) {
376 // preserve exception
377 // note: the monitor_exit runtime call is a leaf routine
378 // and cannot block => no GC can happen
379 // The slow case (MonitorAccessStub) uses the first two stack slots
380 // ([esp+0] and [esp+4]), therefore we store the exception at [esp+8]
381 __ movptr (Address(rsp, 2*wordSize), exception);
382 }
383
384 Register obj_reg = obj_opr->as_register();
385 Register lock_reg = lock_opr->as_register();
386
387 // setup registers (lock_reg must be rax, for lock_object)
388 assert(obj_reg != SYNC_header && lock_reg != SYNC_header, "rax, must be available here");
389 Register hdr = lock_reg;
390 assert(new_hdr == SYNC_header, "wrong register");
391 lock_reg = new_hdr;
392 // compute pointer to BasicLock
393 Address lock_addr = frame_map()->address_for_monitor_lock(monitor_no);
394 __ lea(lock_reg, lock_addr);
395 // unlock object
396 MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, true, monitor_no);
397 // _slow_case_stubs->append(slow_case);
398 // temporary fix: must be created after exceptionhandler, therefore as call stub
399 _slow_case_stubs->append(slow_case);
400 if (UseFastLocking) {
401 // try inlined fast unlocking first, revert to slow locking if it fails
402 // note: lock_reg points to the displaced header since the displaced header offset is 0!
403 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
404 __ unlock_object(hdr, obj_reg, lock_reg, *slow_case->entry());
405 } else {
406 // always do slow unlocking
407 // note: the slow unlocking code could be inlined here, however if we use
408 // slow unlocking, speed doesn't matter anyway and this solution is
409 // simpler and requires less duplicated code - additionally, the
410 // slow unlocking code is the same in either case which simplifies
411 // debugging
412 __ jmp(*slow_case->entry());
413 }
414 // done
415 __ bind(*slow_case->continuation());
416
417 if (exception->is_valid()) {
418 // restore exception
419 __ movptr (exception, Address(rsp, 2 * wordSize));
420 }
421 }
422
423 // This specifies the rsp decrement needed to build the frame
424 int LIR_Assembler::initial_frame_size_in_bytes() {
425 // if rounding, must let FrameMap know!
426
427 // The frame_map records size in slots (32bit word)
428
429 // subtract two words to account for return address and link
430 return (frame_map()->framesize() - (2*VMRegImpl::slots_per_word)) * VMRegImpl::stack_slot_size;
431 }
432
433
434 int LIR_Assembler::emit_exception_handler() {
435 // if the last instruction is a call (typically to do a throw which
436 // is coming at the end after block reordering) the return address
437 // must still point into the code area in order to avoid assertion
438 // failures when searching for the corresponding bci => add a nop
439 // (was bug 5/14/1999 - gri)
440 __ nop();
441
442 // generate code for exception handler
443 address handler_base = __ start_a_stub(exception_handler_size);
444 if (handler_base == NULL) {
445 // not enough space left for the handler
446 bailout("exception handler overflow");
447 return -1;
448 }
449
450 int offset = code_offset();
451
452 // the exception oop and pc are in rax, and rdx
453 // no other registers need to be preserved, so invalidate them
454 __ invalidate_registers(false, true, true, false, true, true);
455
456 // check that there is really an exception
457 __ verify_not_null_oop(rax);
458
459 // search an exception handler (rax: exception oop, rdx: throwing pc)
460 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id)));
461 __ should_not_reach_here();
462 assert(code_offset() - offset <= exception_handler_size, "overflow");
463 __ end_a_stub();
464
465 return offset;
466 }
467
468
469 // Emit the code to remove the frame from the stack in the exception
470 // unwind path.
471 int LIR_Assembler::emit_unwind_handler() {
472 #ifndef PRODUCT
473 if (CommentedAssembly) {
474 _masm->block_comment("Unwind handler");
475 }
476 #endif
477
478 int offset = code_offset();
479
480 // Fetch the exception from TLS and clear out exception related thread state
481 __ get_thread(rsi);
482 __ movptr(rax, Address(rsi, JavaThread::exception_oop_offset()));
483 __ movptr(Address(rsi, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD);
484 __ movptr(Address(rsi, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
485
486 __ bind(_unwind_handler_entry);
487 __ verify_not_null_oop(rax);
488 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
489 __ mov(rsi, rax); // Preserve the exception
490 }
491
492 // Preform needed unlocking
493 MonitorExitStub* stub = NULL;
494 if (method()->is_synchronized()) {
495 monitor_address(0, FrameMap::rax_opr);
496 stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
497 __ unlock_object(rdi, rbx, rax, *stub->entry());
498 __ bind(*stub->continuation());
499 }
500
501 if (compilation()->env()->dtrace_method_probes()) {
502 __ get_thread(rax);
503 __ movptr(Address(rsp, 0), rax);
504 __ movoop(Address(rsp, sizeof(void*)), method()->constant_encoding());
505 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
506 }
507
508 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
509 __ mov(rax, rsi); // Restore the exception
510 }
511
512 // remove the activation and dispatch to the unwind handler
513 __ remove_frame(initial_frame_size_in_bytes());
514 __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
515
516 // Emit the slow path assembly
517 if (stub != NULL) {
518 stub->emit_code(this);
519 }
520
521 return offset;
522 }
523
524
525 int LIR_Assembler::emit_deopt_handler() {
526 // if the last instruction is a call (typically to do a throw which
527 // is coming at the end after block reordering) the return address
528 // must still point into the code area in order to avoid assertion
529 // failures when searching for the corresponding bci => add a nop
530 // (was bug 5/14/1999 - gri)
531 __ nop();
532
533 // generate code for exception handler
534 address handler_base = __ start_a_stub(deopt_handler_size);
535 if (handler_base == NULL) {
536 // not enough space left for the handler
537 bailout("deopt handler overflow");
538 return -1;
539 }
540
541 int offset = code_offset();
542 InternalAddress here(__ pc());
543
544 __ pushptr(here.addr());
545 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
546
547 assert(code_offset() - offset <= deopt_handler_size, "overflow");
548 __ end_a_stub();
549
550 return offset;
551 }
552
553
554 // This is the fast version of java.lang.String.compare; it has not
555 // OSR-entry and therefore, we generate a slow version for OSR's
556 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
557 __ movptr (rbx, rcx); // receiver is in rcx
558 __ movptr (rax, arg1->as_register());
559
560 // Get addresses of first characters from both Strings
561 __ load_heap_oop(rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
562 __ movptr (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
563 __ lea (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
564
565
566 // rbx, may be NULL
567 add_debug_info_for_null_check_here(info);
568 __ load_heap_oop(rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
569 __ movptr (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
570 __ lea (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
571
572 // compute minimum length (in rax) and difference of lengths (on top of stack)
573 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
574 __ movl (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
575 __ mov (rcx, rbx);
576 __ subptr(rbx, rax); // subtract lengths
577 __ push (rbx); // result
578 __ cmov (Assembler::lessEqual, rax, rcx);
579
580 // is minimum length 0?
581 Label noLoop, haveResult;
582 __ testptr (rax, rax);
583 __ jcc (Assembler::zero, noLoop);
584
585 // compare first characters
586 __ load_unsigned_short(rcx, Address(rdi, 0));
587 __ load_unsigned_short(rbx, Address(rsi, 0));
588 __ subl(rcx, rbx);
589 __ jcc(Assembler::notZero, haveResult);
590 // starting loop
591 __ decrement(rax); // we already tested index: skip one
592 __ jcc(Assembler::zero, noLoop);
593
594 // set rsi.edi to the end of the arrays (arrays have same length)
595 // negate the index
596
597 __ lea(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
598 __ lea(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
599 __ negptr(rax);
600
601 // compare the strings in a loop
602
603 Label loop;
604 __ align(wordSize);
605 __ bind(loop);
606 __ load_unsigned_short(rcx, Address(rdi, rax, Address::times_2, 0));
607 __ load_unsigned_short(rbx, Address(rsi, rax, Address::times_2, 0));
608 __ subl(rcx, rbx);
609 __ jcc(Assembler::notZero, haveResult);
610 __ increment(rax);
611 __ jcc(Assembler::notZero, loop);
612
613 // strings are equal up to min length
614
615 __ bind(noLoop);
616 __ pop(rax);
617 return_op(LIR_OprFact::illegalOpr);
618
619 __ bind(haveResult);
620 // leave instruction is going to discard the TOS value
621 __ mov (rax, rcx); // result of call is in rax,
622 }
623
624
625 void LIR_Assembler::return_op(LIR_Opr result) {
626 assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
627 if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
628 assert(result->fpu() == 0, "result must already be on TOS");
629 }
630
631 // Pop the stack before the safepoint code
632 __ remove_frame(initial_frame_size_in_bytes());
633
634 bool result_is_oop = result->is_valid() ? result->is_oop() : false;
635
636 // Note: we do not need to round double result; float result has the right precision
637 // the poll sets the condition code, but no data registers
638 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
639 relocInfo::poll_return_type);
640
641 if (Assembler::is_polling_page_far()) {
642 __ lea(rscratch1, polling_page);
643 __ relocate(relocInfo::poll_return_type);
644 __ testl(rax, Address(rscratch1, 0));
645 } else {
646 __ testl(rax, polling_page);
647 }
648 __ ret(0);
649 }
650
651
652 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
653 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
654 relocInfo::poll_type);
655 guarantee(info != NULL, "Shouldn't be NULL");
656 int offset = __ offset();
657 if (Assembler::is_polling_page_far()) {
658 __ lea(rscratch1, polling_page);
659 offset = __ offset();
660 add_debug_info_for_branch(info);
661 __ testl(rax, Address(rscratch1, 0));
662 } else {
663 add_debug_info_for_branch(info);
664 __ testl(rax, polling_page);
665 }
666 return offset;
667 }
668
669
670 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
671 if (from_reg != to_reg) __ mov(to_reg, from_reg);
672 }
673
674 void LIR_Assembler::swap_reg(Register a, Register b) {
675 __ xchgptr(a, b);
676 }
677
678
679 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
680 assert(src->is_constant(), "should not call otherwise");
681 assert(dest->is_register(), "should not call otherwise");
682 LIR_Const* c = src->as_constant_ptr();
683
684 switch (c->type()) {
685 case T_INT: {
686 assert(patch_code == lir_patch_none, "no patching handled here");
687 __ movl(dest->as_register(), c->as_jint());
688 break;
689 }
690
691 case T_ADDRESS: {
692 assert(patch_code == lir_patch_none, "no patching handled here");
693 __ movptr(dest->as_register(), c->as_jint());
694 break;
695 }
696
697 case T_LONG: {
698 assert(patch_code == lir_patch_none, "no patching handled here");
699 #ifdef _LP64
700 __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
701 #else
702 __ movptr(dest->as_register_lo(), c->as_jint_lo());
703 __ movptr(dest->as_register_hi(), c->as_jint_hi());
704 #endif // _LP64
705 break;
706 }
707
708 case T_OBJECT: {
709 if (patch_code != lir_patch_none) {
710 jobject2reg_with_patching(dest->as_register(), info);
711 } else {
712 __ movoop(dest->as_register(), c->as_jobject());
713 }
714 break;
715 }
716
717 case T_FLOAT: {
718 if (dest->is_single_xmm()) {
719 if (c->is_zero_float()) {
720 __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
721 } else {
722 __ movflt(dest->as_xmm_float_reg(),
723 InternalAddress(float_constant(c->as_jfloat())));
724 }
725 } else {
726 assert(dest->is_single_fpu(), "must be");
727 assert(dest->fpu_regnr() == 0, "dest must be TOS");
728 if (c->is_zero_float()) {
729 __ fldz();
730 } else if (c->is_one_float()) {
731 __ fld1();
732 } else {
733 __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
734 }
735 }
736 break;
737 }
738
739 case T_DOUBLE: {
740 if (dest->is_double_xmm()) {
741 if (c->is_zero_double()) {
742 __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
743 } else {
744 __ movdbl(dest->as_xmm_double_reg(),
745 InternalAddress(double_constant(c->as_jdouble())));
746 }
747 } else {
748 assert(dest->is_double_fpu(), "must be");
749 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
750 if (c->is_zero_double()) {
751 __ fldz();
752 } else if (c->is_one_double()) {
753 __ fld1();
754 } else {
755 __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
756 }
757 }
758 break;
759 }
760
761 default:
762 ShouldNotReachHere();
763 }
764 }
765
766 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
767 assert(src->is_constant(), "should not call otherwise");
768 assert(dest->is_stack(), "should not call otherwise");
769 LIR_Const* c = src->as_constant_ptr();
770
771 switch (c->type()) {
772 case T_INT: // fall through
773 case T_FLOAT:
774 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
775 break;
776
777 case T_ADDRESS:
778 __ movptr(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
779 break;
780
781 case T_OBJECT:
782 __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
783 break;
784
785 case T_LONG: // fall through
786 case T_DOUBLE:
787 #ifdef _LP64
788 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
789 lo_word_offset_in_bytes), (intptr_t)c->as_jlong_bits());
790 #else
791 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
792 lo_word_offset_in_bytes), c->as_jint_lo_bits());
793 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
794 hi_word_offset_in_bytes), c->as_jint_hi_bits());
795 #endif // _LP64
796 break;
797
798 default:
799 ShouldNotReachHere();
800 }
801 }
802
803 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
804 assert(src->is_constant(), "should not call otherwise");
805 assert(dest->is_address(), "should not call otherwise");
806 LIR_Const* c = src->as_constant_ptr();
807 LIR_Address* addr = dest->as_address_ptr();
808
809 int null_check_here = code_offset();
810 switch (type) {
811 case T_INT: // fall through
812 case T_FLOAT:
813 __ movl(as_Address(addr), c->as_jint_bits());
814 break;
815
816 case T_ADDRESS:
817 __ movptr(as_Address(addr), c->as_jint_bits());
818 break;
819
820 case T_OBJECT: // fall through
821 case T_ARRAY:
822 if (c->as_jobject() == NULL) {
823 if (UseCompressedOops && !wide) {
824 __ movl(as_Address(addr), (int32_t)NULL_WORD);
825 } else {
826 __ movptr(as_Address(addr), NULL_WORD);
827 }
828 } else {
829 if (is_literal_address(addr)) {
830 ShouldNotReachHere();
831 __ movoop(as_Address(addr, noreg), c->as_jobject());
832 } else {
833 #ifdef _LP64
834 __ movoop(rscratch1, c->as_jobject());
835 if (UseCompressedOops && !wide) {
836 __ encode_heap_oop(rscratch1);
837 null_check_here = code_offset();
838 __ movl(as_Address_lo(addr), rscratch1);
839 } else {
840 null_check_here = code_offset();
841 __ movptr(as_Address_lo(addr), rscratch1);
842 }
843 #else
844 __ movoop(as_Address(addr), c->as_jobject());
845 #endif
846 }
847 }
848 break;
849
850 case T_LONG: // fall through
851 case T_DOUBLE:
852 #ifdef _LP64
853 if (is_literal_address(addr)) {
854 ShouldNotReachHere();
855 __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
856 } else {
857 __ movptr(r10, (intptr_t)c->as_jlong_bits());
858 null_check_here = code_offset();
859 __ movptr(as_Address_lo(addr), r10);
860 }
861 #else
862 // Always reachable in 32bit so this doesn't produce useless move literal
863 __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
864 __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
865 #endif // _LP64
866 break;
867
868 case T_BOOLEAN: // fall through
869 case T_BYTE:
870 __ movb(as_Address(addr), c->as_jint() & 0xFF);
871 break;
872
873 case T_CHAR: // fall through
874 case T_SHORT:
875 __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
876 break;
877
878 default:
879 ShouldNotReachHere();
880 };
881
882 if (info != NULL) {
883 add_debug_info_for_null_check(null_check_here, info);
884 }
885 }
886
887
888 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
889 assert(src->is_register(), "should not call otherwise");
890 assert(dest->is_register(), "should not call otherwise");
891
892 // move between cpu-registers
893 if (dest->is_single_cpu()) {
894 #ifdef _LP64
895 if (src->type() == T_LONG) {
896 // Can do LONG -> OBJECT
897 move_regs(src->as_register_lo(), dest->as_register());
898 return;
899 }
900 #endif
901 assert(src->is_single_cpu(), "must match");
902 if (src->type() == T_OBJECT) {
903 __ verify_oop(src->as_register());
904 }
905 move_regs(src->as_register(), dest->as_register());
906
907 } else if (dest->is_double_cpu()) {
908 #ifdef _LP64
909 if (src->type() == T_OBJECT || src->type() == T_ARRAY) {
910 // Surprising to me but we can see move of a long to t_object
911 __ verify_oop(src->as_register());
912 move_regs(src->as_register(), dest->as_register_lo());
913 return;
914 }
915 #endif
916 assert(src->is_double_cpu(), "must match");
917 Register f_lo = src->as_register_lo();
918 Register f_hi = src->as_register_hi();
919 Register t_lo = dest->as_register_lo();
920 Register t_hi = dest->as_register_hi();
921 #ifdef _LP64
922 assert(f_hi == f_lo, "must be same");
923 assert(t_hi == t_lo, "must be same");
924 move_regs(f_lo, t_lo);
925 #else
926 assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
927
928
929 if (f_lo == t_hi && f_hi == t_lo) {
930 swap_reg(f_lo, f_hi);
931 } else if (f_hi == t_lo) {
932 assert(f_lo != t_hi, "overwriting register");
933 move_regs(f_hi, t_hi);
934 move_regs(f_lo, t_lo);
935 } else {
936 assert(f_hi != t_lo, "overwriting register");
937 move_regs(f_lo, t_lo);
938 move_regs(f_hi, t_hi);
939 }
940 #endif // LP64
941
942 // special moves from fpu-register to xmm-register
943 // necessary for method results
944 } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
945 __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
946 __ fld_s(Address(rsp, 0));
947 } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
948 __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
949 __ fld_d(Address(rsp, 0));
950 } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
951 __ fstp_s(Address(rsp, 0));
952 __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
953 } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
954 __ fstp_d(Address(rsp, 0));
955 __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
956
957 // move between xmm-registers
958 } else if (dest->is_single_xmm()) {
959 assert(src->is_single_xmm(), "must match");
960 __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
961 } else if (dest->is_double_xmm()) {
962 assert(src->is_double_xmm(), "must match");
963 __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
964
965 // move between fpu-registers (no instruction necessary because of fpu-stack)
966 } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
967 assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
968 assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
969 } else {
970 ShouldNotReachHere();
971 }
972 }
973
974 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
975 assert(src->is_register(), "should not call otherwise");
976 assert(dest->is_stack(), "should not call otherwise");
977
978 if (src->is_single_cpu()) {
979 Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
980 if (type == T_OBJECT || type == T_ARRAY) {
981 __ verify_oop(src->as_register());
982 __ movptr (dst, src->as_register());
983 } else {
984 __ movl (dst, src->as_register());
985 }
986
987 } else if (src->is_double_cpu()) {
988 Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
989 Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
990 __ movptr (dstLO, src->as_register_lo());
991 NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
992
993 } else if (src->is_single_xmm()) {
994 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
995 __ movflt(dst_addr, src->as_xmm_float_reg());
996
997 } else if (src->is_double_xmm()) {
998 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
999 __ movdbl(dst_addr, src->as_xmm_double_reg());
1000
1001 } else if (src->is_single_fpu()) {
1002 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1003 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1004 if (pop_fpu_stack) __ fstp_s (dst_addr);
1005 else __ fst_s (dst_addr);
1006
1007 } else if (src->is_double_fpu()) {
1008 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1009 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1010 if (pop_fpu_stack) __ fstp_d (dst_addr);
1011 else __ fst_d (dst_addr);
1012
1013 } else {
1014 ShouldNotReachHere();
1015 }
1016 }
1017
1018
1019 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack, bool wide, bool /* unaligned */) {
1020 LIR_Address* to_addr = dest->as_address_ptr();
1021 PatchingStub* patch = NULL;
1022 Register compressed_src = rscratch1;
1023
1024 if (type == T_ARRAY || type == T_OBJECT) {
1025 __ verify_oop(src->as_register());
1026 #ifdef _LP64
1027 if (UseCompressedOops && !wide) {
1028 __ movptr(compressed_src, src->as_register());
1029 __ encode_heap_oop(compressed_src);
1030 }
1031 #endif
1032 }
1033
1034 if (patch_code != lir_patch_none) {
1035 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1036 Address toa = as_Address(to_addr);
1037 assert(toa.disp() != 0, "must have");
1038 }
1039
1040 int null_check_here = code_offset();
1041 switch (type) {
1042 case T_FLOAT: {
1043 if (src->is_single_xmm()) {
1044 __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
1045 } else {
1046 assert(src->is_single_fpu(), "must be");
1047 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1048 if (pop_fpu_stack) __ fstp_s(as_Address(to_addr));
1049 else __ fst_s (as_Address(to_addr));
1050 }
1051 break;
1052 }
1053
1054 case T_DOUBLE: {
1055 if (src->is_double_xmm()) {
1056 __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
1057 } else {
1058 assert(src->is_double_fpu(), "must be");
1059 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1060 if (pop_fpu_stack) __ fstp_d(as_Address(to_addr));
1061 else __ fst_d (as_Address(to_addr));
1062 }
1063 break;
1064 }
1065
1066 case T_ARRAY: // fall through
1067 case T_OBJECT: // fall through
1068 if (UseCompressedOops && !wide) {
1069 __ movl(as_Address(to_addr), compressed_src);
1070 } else {
1071 __ movptr(as_Address(to_addr), src->as_register());
1072 }
1073 break;
1074 case T_ADDRESS:
1075 __ movptr(as_Address(to_addr), src->as_register());
1076 break;
1077 case T_INT:
1078 __ movl(as_Address(to_addr), src->as_register());
1079 break;
1080
1081 case T_LONG: {
1082 Register from_lo = src->as_register_lo();
1083 Register from_hi = src->as_register_hi();
1084 #ifdef _LP64
1085 __ movptr(as_Address_lo(to_addr), from_lo);
1086 #else
1087 Register base = to_addr->base()->as_register();
1088 Register index = noreg;
1089 if (to_addr->index()->is_register()) {
1090 index = to_addr->index()->as_register();
1091 }
1092 if (base == from_lo || index == from_lo) {
1093 assert(base != from_hi, "can't be");
1094 assert(index == noreg || (index != base && index != from_hi), "can't handle this");
1095 __ movl(as_Address_hi(to_addr), from_hi);
1096 if (patch != NULL) {
1097 patching_epilog(patch, lir_patch_high, base, info);
1098 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1099 patch_code = lir_patch_low;
1100 }
1101 __ movl(as_Address_lo(to_addr), from_lo);
1102 } else {
1103 assert(index == noreg || (index != base && index != from_lo), "can't handle this");
1104 __ movl(as_Address_lo(to_addr), from_lo);
1105 if (patch != NULL) {
1106 patching_epilog(patch, lir_patch_low, base, info);
1107 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1108 patch_code = lir_patch_high;
1109 }
1110 __ movl(as_Address_hi(to_addr), from_hi);
1111 }
1112 #endif // _LP64
1113 break;
1114 }
1115
1116 case T_BYTE: // fall through
1117 case T_BOOLEAN: {
1118 Register src_reg = src->as_register();
1119 Address dst_addr = as_Address(to_addr);
1120 assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
1121 __ movb(dst_addr, src_reg);
1122 break;
1123 }
1124
1125 case T_CHAR: // fall through
1126 case T_SHORT:
1127 __ movw(as_Address(to_addr), src->as_register());
1128 break;
1129
1130 default:
1131 ShouldNotReachHere();
1132 }
1133 if (info != NULL) {
1134 add_debug_info_for_null_check(null_check_here, info);
1135 }
1136
1137 if (patch_code != lir_patch_none) {
1138 patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
1139 }
1140 }
1141
1142
1143 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1144 assert(src->is_stack(), "should not call otherwise");
1145 assert(dest->is_register(), "should not call otherwise");
1146
1147 if (dest->is_single_cpu()) {
1148 if (type == T_ARRAY || type == T_OBJECT) {
1149 __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1150 __ verify_oop(dest->as_register());
1151 } else {
1152 __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1153 }
1154
1155 } else if (dest->is_double_cpu()) {
1156 Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
1157 Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
1158 __ movptr(dest->as_register_lo(), src_addr_LO);
1159 NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
1160
1161 } else if (dest->is_single_xmm()) {
1162 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1163 __ movflt(dest->as_xmm_float_reg(), src_addr);
1164
1165 } else if (dest->is_double_xmm()) {
1166 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1167 __ movdbl(dest->as_xmm_double_reg(), src_addr);
1168
1169 } else if (dest->is_single_fpu()) {
1170 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1171 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1172 __ fld_s(src_addr);
1173
1174 } else if (dest->is_double_fpu()) {
1175 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1176 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1177 __ fld_d(src_addr);
1178
1179 } else {
1180 ShouldNotReachHere();
1181 }
1182 }
1183
1184
1185 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1186 if (src->is_single_stack()) {
1187 if (type == T_OBJECT || type == T_ARRAY) {
1188 __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
1189 __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
1190 } else {
1191 #ifndef _LP64
1192 __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
1193 __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
1194 #else
1195 //no pushl on 64bits
1196 __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix()));
1197 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1);
1198 #endif
1199 }
1200
1201 } else if (src->is_double_stack()) {
1202 #ifdef _LP64
1203 __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
1204 __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
1205 #else
1206 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
1207 // push and pop the part at src + wordSize, adding wordSize for the previous push
1208 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 2 * wordSize));
1209 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 2 * wordSize));
1210 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
1211 #endif // _LP64
1212
1213 } else {
1214 ShouldNotReachHere();
1215 }
1216 }
1217
1218
1219 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool /* unaligned */) {
1220 assert(src->is_address(), "should not call otherwise");
1221 assert(dest->is_register(), "should not call otherwise");
1222
1223 LIR_Address* addr = src->as_address_ptr();
1224 Address from_addr = as_Address(addr);
1225
1226 switch (type) {
1227 case T_BOOLEAN: // fall through
1228 case T_BYTE: // fall through
1229 case T_CHAR: // fall through
1230 case T_SHORT:
1231 if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
1232 // on pre P6 processors we may get partial register stalls
1233 // so blow away the value of to_rinfo before loading a
1234 // partial word into it. Do it here so that it precedes
1235 // the potential patch point below.
1236 __ xorptr(dest->as_register(), dest->as_register());
1237 }
1238 break;
1239 }
1240
1241 PatchingStub* patch = NULL;
1242 if (patch_code != lir_patch_none) {
1243 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1244 assert(from_addr.disp() != 0, "must have");
1245 }
1246 if (info != NULL) {
1247 add_debug_info_for_null_check_here(info);
1248 }
1249
1250 switch (type) {
1251 case T_FLOAT: {
1252 if (dest->is_single_xmm()) {
1253 __ movflt(dest->as_xmm_float_reg(), from_addr);
1254 } else {
1255 assert(dest->is_single_fpu(), "must be");
1256 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1257 __ fld_s(from_addr);
1258 }
1259 break;
1260 }
1261
1262 case T_DOUBLE: {
1263 if (dest->is_double_xmm()) {
1264 __ movdbl(dest->as_xmm_double_reg(), from_addr);
1265 } else {
1266 assert(dest->is_double_fpu(), "must be");
1267 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1268 __ fld_d(from_addr);
1269 }
1270 break;
1271 }
1272
1273 case T_OBJECT: // fall through
1274 case T_ARRAY: // fall through
1275 if (UseCompressedOops && !wide) {
1276 __ movl(dest->as_register(), from_addr);
1277 } else {
1278 __ movptr(dest->as_register(), from_addr);
1279 }
1280 break;
1281
1282 case T_ADDRESS:
1283 __ movptr(dest->as_register(), from_addr);
1284 break;
1285 case T_INT:
1286 __ movl(dest->as_register(), from_addr);
1287 break;
1288
1289 case T_LONG: {
1290 Register to_lo = dest->as_register_lo();
1291 Register to_hi = dest->as_register_hi();
1292 #ifdef _LP64
1293 __ movptr(to_lo, as_Address_lo(addr));
1294 #else
1295 Register base = addr->base()->as_register();
1296 Register index = noreg;
1297 if (addr->index()->is_register()) {
1298 index = addr->index()->as_register();
1299 }
1300 if ((base == to_lo && index == to_hi) ||
1301 (base == to_hi && index == to_lo)) {
1302 // addresses with 2 registers are only formed as a result of
1303 // array access so this code will never have to deal with
1304 // patches or null checks.
1305 assert(info == NULL && patch == NULL, "must be");
1306 __ lea(to_hi, as_Address(addr));
1307 __ movl(to_lo, Address(to_hi, 0));
1308 __ movl(to_hi, Address(to_hi, BytesPerWord));
1309 } else if (base == to_lo || index == to_lo) {
1310 assert(base != to_hi, "can't be");
1311 assert(index == noreg || (index != base && index != to_hi), "can't handle this");
1312 __ movl(to_hi, as_Address_hi(addr));
1313 if (patch != NULL) {
1314 patching_epilog(patch, lir_patch_high, base, info);
1315 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1316 patch_code = lir_patch_low;
1317 }
1318 __ movl(to_lo, as_Address_lo(addr));
1319 } else {
1320 assert(index == noreg || (index != base && index != to_lo), "can't handle this");
1321 __ movl(to_lo, as_Address_lo(addr));
1322 if (patch != NULL) {
1323 patching_epilog(patch, lir_patch_low, base, info);
1324 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1325 patch_code = lir_patch_high;
1326 }
1327 __ movl(to_hi, as_Address_hi(addr));
1328 }
1329 #endif // _LP64
1330 break;
1331 }
1332
1333 case T_BOOLEAN: // fall through
1334 case T_BYTE: {
1335 Register dest_reg = dest->as_register();
1336 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1337 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1338 __ movsbl(dest_reg, from_addr);
1339 } else {
1340 __ movb(dest_reg, from_addr);
1341 __ shll(dest_reg, 24);
1342 __ sarl(dest_reg, 24);
1343 }
1344 break;
1345 }
1346
1347 case T_CHAR: {
1348 Register dest_reg = dest->as_register();
1349 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1350 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1351 __ movzwl(dest_reg, from_addr);
1352 } else {
1353 __ movw(dest_reg, from_addr);
1354 }
1355 break;
1356 }
1357
1358 case T_SHORT: {
1359 Register dest_reg = dest->as_register();
1360 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1361 __ movswl(dest_reg, from_addr);
1362 } else {
1363 __ movw(dest_reg, from_addr);
1364 __ shll(dest_reg, 16);
1365 __ sarl(dest_reg, 16);
1366 }
1367 break;
1368 }
1369
1370 default:
1371 ShouldNotReachHere();
1372 }
1373
1374 if (patch != NULL) {
1375 patching_epilog(patch, patch_code, addr->base()->as_register(), info);
1376 }
1377
1378 if (type == T_ARRAY || type == T_OBJECT) {
1379 #ifdef _LP64
1380 if (UseCompressedOops && !wide) {
1381 __ decode_heap_oop(dest->as_register());
1382 }
1383 #endif
1384 __ verify_oop(dest->as_register());
1385 }
1386 }
1387
1388
1389 void LIR_Assembler::prefetchr(LIR_Opr src) {
1390 LIR_Address* addr = src->as_address_ptr();
1391 Address from_addr = as_Address(addr);
1392
1393 if (VM_Version::supports_sse()) {
1394 switch (ReadPrefetchInstr) {
1395 case 0:
1396 __ prefetchnta(from_addr); break;
1397 case 1:
1398 __ prefetcht0(from_addr); break;
1399 case 2:
1400 __ prefetcht2(from_addr); break;
1401 default:
1402 ShouldNotReachHere(); break;
1403 }
1404 } else if (VM_Version::supports_3dnow_prefetch()) {
1405 __ prefetchr(from_addr);
1406 }
1407 }
1408
1409
1410 void LIR_Assembler::prefetchw(LIR_Opr src) {
1411 LIR_Address* addr = src->as_address_ptr();
1412 Address from_addr = as_Address(addr);
1413
1414 if (VM_Version::supports_sse()) {
1415 switch (AllocatePrefetchInstr) {
1416 case 0:
1417 __ prefetchnta(from_addr); break;
1418 case 1:
1419 __ prefetcht0(from_addr); break;
1420 case 2:
1421 __ prefetcht2(from_addr); break;
1422 case 3:
1423 __ prefetchw(from_addr); break;
1424 default:
1425 ShouldNotReachHere(); break;
1426 }
1427 } else if (VM_Version::supports_3dnow_prefetch()) {
1428 __ prefetchw(from_addr);
1429 }
1430 }
1431
1432
1433 NEEDS_CLEANUP; // This could be static?
1434 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
1435 int elem_size = type2aelembytes(type);
1436 switch (elem_size) {
1437 case 1: return Address::times_1;
1438 case 2: return Address::times_2;
1439 case 4: return Address::times_4;
1440 case 8: return Address::times_8;
1441 }
1442 ShouldNotReachHere();
1443 return Address::no_scale;
1444 }
1445
1446
1447 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1448 switch (op->code()) {
1449 case lir_idiv:
1450 case lir_irem:
1451 arithmetic_idiv(op->code(),
1452 op->in_opr1(),
1453 op->in_opr2(),
1454 op->in_opr3(),
1455 op->result_opr(),
1456 op->info());
1457 break;
1458 default: ShouldNotReachHere(); break;
1459 }
1460 }
1461
1462 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1463 #ifdef ASSERT
1464 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
1465 if (op->block() != NULL) _branch_target_blocks.append(op->block());
1466 if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
1467 #endif
1468
1469 if (op->cond() == lir_cond_always) {
1470 if (op->info() != NULL) add_debug_info_for_branch(op->info());
1471 __ jmp (*(op->label()));
1472 } else {
1473 Assembler::Condition acond = Assembler::zero;
1474 if (op->code() == lir_cond_float_branch) {
1475 assert(op->ublock() != NULL, "must have unordered successor");
1476 __ jcc(Assembler::parity, *(op->ublock()->label()));
1477 switch(op->cond()) {
1478 case lir_cond_equal: acond = Assembler::equal; break;
1479 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1480 case lir_cond_less: acond = Assembler::below; break;
1481 case lir_cond_lessEqual: acond = Assembler::belowEqual; break;
1482 case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
1483 case lir_cond_greater: acond = Assembler::above; break;
1484 default: ShouldNotReachHere();
1485 }
1486 } else {
1487 switch (op->cond()) {
1488 case lir_cond_equal: acond = Assembler::equal; break;
1489 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1490 case lir_cond_less: acond = Assembler::less; break;
1491 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
1492 case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
1493 case lir_cond_greater: acond = Assembler::greater; break;
1494 case lir_cond_belowEqual: acond = Assembler::belowEqual; break;
1495 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break;
1496 default: ShouldNotReachHere();
1497 }
1498 }
1499 __ jcc(acond,*(op->label()));
1500 }
1501 }
1502
1503 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1504 LIR_Opr src = op->in_opr();
1505 LIR_Opr dest = op->result_opr();
1506
1507 switch (op->bytecode()) {
1508 case Bytecodes::_i2l:
1509 #ifdef _LP64
1510 __ movl2ptr(dest->as_register_lo(), src->as_register());
1511 #else
1512 move_regs(src->as_register(), dest->as_register_lo());
1513 move_regs(src->as_register(), dest->as_register_hi());
1514 __ sarl(dest->as_register_hi(), 31);
1515 #endif // LP64
1516 break;
1517
1518 case Bytecodes::_l2i:
1519 move_regs(src->as_register_lo(), dest->as_register());
1520 break;
1521
1522 case Bytecodes::_i2b:
1523 move_regs(src->as_register(), dest->as_register());
1524 __ sign_extend_byte(dest->as_register());
1525 break;
1526
1527 case Bytecodes::_i2c:
1528 move_regs(src->as_register(), dest->as_register());
1529 __ andl(dest->as_register(), 0xFFFF);
1530 break;
1531
1532 case Bytecodes::_i2s:
1533 move_regs(src->as_register(), dest->as_register());
1534 __ sign_extend_short(dest->as_register());
1535 break;
1536
1537
1538 case Bytecodes::_f2d:
1539 case Bytecodes::_d2f:
1540 if (dest->is_single_xmm()) {
1541 __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1542 } else if (dest->is_double_xmm()) {
1543 __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1544 } else {
1545 assert(src->fpu() == dest->fpu(), "register must be equal");
1546 // do nothing (float result is rounded later through spilling)
1547 }
1548 break;
1549
1550 case Bytecodes::_i2f:
1551 case Bytecodes::_i2d:
1552 if (dest->is_single_xmm()) {
1553 __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1554 } else if (dest->is_double_xmm()) {
1555 __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1556 } else {
1557 assert(dest->fpu() == 0, "result must be on TOS");
1558 __ movl(Address(rsp, 0), src->as_register());
1559 __ fild_s(Address(rsp, 0));
1560 }
1561 break;
1562
1563 case Bytecodes::_f2i:
1564 case Bytecodes::_d2i:
1565 if (src->is_single_xmm()) {
1566 __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
1567 } else if (src->is_double_xmm()) {
1568 __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
1569 } else {
1570 assert(src->fpu() == 0, "input must be on TOS");
1571 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
1572 __ fist_s(Address(rsp, 0));
1573 __ movl(dest->as_register(), Address(rsp, 0));
1574 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1575 }
1576
1577 // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
1578 assert(op->stub() != NULL, "stub required");
1579 __ cmpl(dest->as_register(), 0x80000000);
1580 __ jcc(Assembler::equal, *op->stub()->entry());
1581 __ bind(*op->stub()->continuation());
1582 break;
1583
1584 case Bytecodes::_l2f:
1585 case Bytecodes::_l2d:
1586 assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
1587 assert(dest->fpu() == 0, "result must be on TOS");
1588
1589 __ movptr(Address(rsp, 0), src->as_register_lo());
1590 NOT_LP64(__ movl(Address(rsp, BytesPerWord), src->as_register_hi()));
1591 __ fild_d(Address(rsp, 0));
1592 // float result is rounded later through spilling
1593 break;
1594
1595 case Bytecodes::_f2l:
1596 case Bytecodes::_d2l:
1597 assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
1598 assert(src->fpu() == 0, "input must be on TOS");
1599 assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
1600
1601 // instruction sequence too long to inline it here
1602 {
1603 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
1604 }
1605 break;
1606
1607 default: ShouldNotReachHere();
1608 }
1609 }
1610
1611 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1612 if (op->init_check()) {
1613 __ cmpl(Address(op->klass()->as_register(),
1614 instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)),
1615 instanceKlass::fully_initialized);
1616 add_debug_info_for_null_check_here(op->stub()->info());
1617 __ jcc(Assembler::notEqual, *op->stub()->entry());
1618 }
1619 __ allocate_object(op->obj()->as_register(),
1620 op->tmp1()->as_register(),
1621 op->tmp2()->as_register(),
1622 op->header_size(),
1623 op->object_size(),
1624 op->klass()->as_register(),
1625 *op->stub()->entry());
1626 __ bind(*op->stub()->continuation());
1627 }
1628
1629 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1630 Register len = op->len()->as_register();
1631 LP64_ONLY( __ movslq(len, len); )
1632
1633 if (UseSlowPath ||
1634 (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
1635 (!UseFastNewTypeArray && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
1636 __ jmp(*op->stub()->entry());
1637 } else {
1638 Register tmp1 = op->tmp1()->as_register();
1639 Register tmp2 = op->tmp2()->as_register();
1640 Register tmp3 = op->tmp3()->as_register();
1641 if (len == tmp1) {
1642 tmp1 = tmp3;
1643 } else if (len == tmp2) {
1644 tmp2 = tmp3;
1645 } else if (len == tmp3) {
1646 // everything is ok
1647 } else {
1648 __ mov(tmp3, len);
1649 }
1650 __ allocate_array(op->obj()->as_register(),
1651 len,
1652 tmp1,
1653 tmp2,
1654 arrayOopDesc::header_size(op->type()),
1655 array_element_size(op->type()),
1656 op->klass()->as_register(),
1657 *op->stub()->entry());
1658 }
1659 __ bind(*op->stub()->continuation());
1660 }
1661
1662 void LIR_Assembler::type_profile_helper(Register mdo,
1663 ciMethodData *md, ciProfileData *data,
1664 Register recv, Label* update_done) {
1665 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1666 Label next_test;
1667 // See if the receiver is receiver[n].
1668 __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
1669 __ jccb(Assembler::notEqual, next_test);
1670 Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
1671 __ addptr(data_addr, DataLayout::counter_increment);
1672 __ jmp(*update_done);
1673 __ bind(next_test);
1674 }
1675
1676 // Didn't find receiver; find next empty slot and fill it in
1677 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1678 Label next_test;
1679 Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
1680 __ cmpptr(recv_addr, (intptr_t)NULL_WORD);
1681 __ jccb(Assembler::notEqual, next_test);
1682 __ movptr(recv_addr, recv);
1683 __ movptr(Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))), DataLayout::counter_increment);
1684 __ jmp(*update_done);
1685 __ bind(next_test);
1686 }
1687 }
1688
1689 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
1690 // we always need a stub for the failure case.
1691 CodeStub* stub = op->stub();
1692 Register obj = op->object()->as_register();
1693 Register k_RInfo = op->tmp1()->as_register();
1694 Register klass_RInfo = op->tmp2()->as_register();
1695 Register dst = op->result_opr()->as_register();
1696 ciKlass* k = op->klass();
1697 Register Rtmp1 = noreg;
1698
1699 // check if it needs to be profiled
1700 ciMethodData* md;
1701 ciProfileData* data;
1702
1703 if (op->should_profile()) {
1704 ciMethod* method = op->profiled_method();
1705 assert(method != NULL, "Should have method");
1706 int bci = op->profiled_bci();
1707 md = method->method_data_or_null();
1708 assert(md != NULL, "Sanity");
1709 data = md->bci_to_data(bci);
1710 assert(data != NULL, "need data for type check");
1711 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1712 }
1713 Label profile_cast_success, profile_cast_failure;
1714 Label *success_target = op->should_profile() ? &profile_cast_success : success;
1715 Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
1716
1717 if (obj == k_RInfo) {
1718 k_RInfo = dst;
1719 } else if (obj == klass_RInfo) {
1720 klass_RInfo = dst;
1721 }
1722 if (k->is_loaded() && !UseCompressedOops) {
1723 select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1724 } else {
1725 Rtmp1 = op->tmp3()->as_register();
1726 select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1727 }
1728
1729 assert_different_registers(obj, k_RInfo, klass_RInfo);
1730 if (!k->is_loaded()) {
1731 jobject2reg_with_patching(k_RInfo, op->info_for_patch());
1732 } else {
1733 #ifdef _LP64
1734 __ movoop(k_RInfo, k->constant_encoding());
1735 #endif // _LP64
1736 }
1737 assert(obj != k_RInfo, "must be different");
1738
1739 __ cmpptr(obj, (int32_t)NULL_WORD);
1740 if (op->should_profile()) {
1741 Label not_null;
1742 __ jccb(Assembler::notEqual, not_null);
1743 // Object is null; update MDO and exit
1744 Register mdo = klass_RInfo;
1745 __ movoop(mdo, md->constant_encoding());
1746 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1747 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1748 __ orl(data_addr, header_bits);
1749 __ jmp(*obj_is_null);
1750 __ bind(not_null);
1751 } else {
1752 __ jcc(Assembler::equal, *obj_is_null);
1753 }
1754 __ verify_oop(obj);
1755
1756 if (op->fast_check()) {
1757 // get object class
1758 // not a safepoint as obj null check happens earlier
1759 #ifdef _LP64
1760 if (UseCompressedOops) {
1761 __ load_klass(Rtmp1, obj);
1762 __ cmpptr(k_RInfo, Rtmp1);
1763 } else {
1764 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1765 }
1766 #else
1767 if (k->is_loaded()) {
1768 __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->constant_encoding());
1769 } else {
1770 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1771 }
1772 #endif
1773 __ jcc(Assembler::notEqual, *failure_target);
1774 // successful cast, fall through to profile or jump
1775 } else {
1776 // get object class
1777 // not a safepoint as obj null check happens earlier
1778 __ load_klass(klass_RInfo, obj);
1779 if (k->is_loaded()) {
1780 // See if we get an immediate positive hit
1781 #ifdef _LP64
1782 __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
1783 #else
1784 __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->constant_encoding());
1785 #endif // _LP64
1786 if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() != k->super_check_offset()) {
1787 __ jcc(Assembler::notEqual, *failure_target);
1788 // successful cast, fall through to profile or jump
1789 } else {
1790 // See if we get an immediate positive hit
1791 __ jcc(Assembler::equal, *success_target);
1792 // check for self
1793 #ifdef _LP64
1794 __ cmpptr(klass_RInfo, k_RInfo);
1795 #else
1796 __ cmpoop(klass_RInfo, k->constant_encoding());
1797 #endif // _LP64
1798 __ jcc(Assembler::equal, *success_target);
1799
1800 __ push(klass_RInfo);
1801 #ifdef _LP64
1802 __ push(k_RInfo);
1803 #else
1804 __ pushoop(k->constant_encoding());
1805 #endif // _LP64
1806 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1807 __ pop(klass_RInfo);
1808 __ pop(klass_RInfo);
1809 // result is a boolean
1810 __ cmpl(klass_RInfo, 0);
1811 __ jcc(Assembler::equal, *failure_target);
1812 // successful cast, fall through to profile or jump
1813 }
1814 } else {
1815 // perform the fast part of the checking logic
1816 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1817 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1818 __ push(klass_RInfo);
1819 __ push(k_RInfo);
1820 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1821 __ pop(klass_RInfo);
1822 __ pop(k_RInfo);
1823 // result is a boolean
1824 __ cmpl(k_RInfo, 0);
1825 __ jcc(Assembler::equal, *failure_target);
1826 // successful cast, fall through to profile or jump
1827 }
1828 }
1829 if (op->should_profile()) {
1830 Register mdo = klass_RInfo, recv = k_RInfo;
1831 __ bind(profile_cast_success);
1832 __ movoop(mdo, md->constant_encoding());
1833 __ load_klass(recv, obj);
1834 Label update_done;
1835 type_profile_helper(mdo, md, data, recv, success);
1836 __ jmp(*success);
1837
1838 __ bind(profile_cast_failure);
1839 __ movoop(mdo, md->constant_encoding());
1840 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1841 __ subptr(counter_addr, DataLayout::counter_increment);
1842 __ jmp(*failure);
1843 }
1844 __ jmp(*success);
1845 }
1846
1847
1848 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1849 LIR_Code code = op->code();
1850 if (code == lir_store_check) {
1851 Register value = op->object()->as_register();
1852 Register array = op->array()->as_register();
1853 Register k_RInfo = op->tmp1()->as_register();
1854 Register klass_RInfo = op->tmp2()->as_register();
1855 Register Rtmp1 = op->tmp3()->as_register();
1856
1857 CodeStub* stub = op->stub();
1858
1859 // check if it needs to be profiled
1860 ciMethodData* md;
1861 ciProfileData* data;
1862
1863 if (op->should_profile()) {
1864 ciMethod* method = op->profiled_method();
1865 assert(method != NULL, "Should have method");
1866 int bci = op->profiled_bci();
1867 md = method->method_data_or_null();
1868 assert(md != NULL, "Sanity");
1869 data = md->bci_to_data(bci);
1870 assert(data != NULL, "need data for type check");
1871 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1872 }
1873 Label profile_cast_success, profile_cast_failure, done;
1874 Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1875 Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
1876
1877 __ cmpptr(value, (int32_t)NULL_WORD);
1878 if (op->should_profile()) {
1879 Label not_null;
1880 __ jccb(Assembler::notEqual, not_null);
1881 // Object is null; update MDO and exit
1882 Register mdo = klass_RInfo;
1883 __ movoop(mdo, md->constant_encoding());
1884 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1885 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1886 __ orl(data_addr, header_bits);
1887 __ jmp(done);
1888 __ bind(not_null);
1889 } else {
1890 __ jcc(Assembler::equal, done);
1891 }
1892
1893 add_debug_info_for_null_check_here(op->info_for_exception());
1894 __ load_klass(k_RInfo, array);
1895 __ load_klass(klass_RInfo, value);
1896
1897 // get instance klass (it's already uncompressed)
1898 __ movptr(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
1899 // perform the fast part of the checking logic
1900 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1901 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1902 __ push(klass_RInfo);
1903 __ push(k_RInfo);
1904 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1905 __ pop(klass_RInfo);
1906 __ pop(k_RInfo);
1907 // result is a boolean
1908 __ cmpl(k_RInfo, 0);
1909 __ jcc(Assembler::equal, *failure_target);
1910 // fall through to the success case
1911
1912 if (op->should_profile()) {
1913 Register mdo = klass_RInfo, recv = k_RInfo;
1914 __ bind(profile_cast_success);
1915 __ movoop(mdo, md->constant_encoding());
1916 __ load_klass(recv, value);
1917 Label update_done;
1918 type_profile_helper(mdo, md, data, recv, &done);
1919 __ jmpb(done);
1920
1921 __ bind(profile_cast_failure);
1922 __ movoop(mdo, md->constant_encoding());
1923 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1924 __ subptr(counter_addr, DataLayout::counter_increment);
1925 __ jmp(*stub->entry());
1926 }
1927
1928 __ bind(done);
1929 } else
1930 if (code == lir_checkcast) {
1931 Register obj = op->object()->as_register();
1932 Register dst = op->result_opr()->as_register();
1933 Label success;
1934 emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
1935 __ bind(success);
1936 if (dst != obj) {
1937 __ mov(dst, obj);
1938 }
1939 } else
1940 if (code == lir_instanceof) {
1941 Register obj = op->object()->as_register();
1942 Register dst = op->result_opr()->as_register();
1943 Label success, failure, done;
1944 emit_typecheck_helper(op, &success, &failure, &failure);
1945 __ bind(failure);
1946 __ xorptr(dst, dst);
1947 __ jmpb(done);
1948 __ bind(success);
1949 __ movptr(dst, 1);
1950 __ bind(done);
1951 } else {
1952 ShouldNotReachHere();
1953 }
1954
1955 }
1956
1957
1958 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1959 if (LP64_ONLY(false &&) op->code() == lir_cas_long && VM_Version::supports_cx8()) {
1960 assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
1961 assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
1962 assert(op->new_value()->as_register_lo() == rbx, "wrong register");
1963 assert(op->new_value()->as_register_hi() == rcx, "wrong register");
1964 Register addr = op->addr()->as_register();
1965 if (os::is_MP()) {
1966 __ lock();
1967 }
1968 NOT_LP64(__ cmpxchg8(Address(addr, 0)));
1969
1970 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
1971 NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
1972 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1973 Register newval = op->new_value()->as_register();
1974 Register cmpval = op->cmp_value()->as_register();
1975 assert(cmpval == rax, "wrong register");
1976 assert(newval != NULL, "new val must be register");
1977 assert(cmpval != newval, "cmp and new values must be in different registers");
1978 assert(cmpval != addr, "cmp and addr must be in different registers");
1979 assert(newval != addr, "new value and addr must be in different registers");
1980
1981 if ( op->code() == lir_cas_obj) {
1982 #ifdef _LP64
1983 if (UseCompressedOops) {
1984 __ encode_heap_oop(cmpval);
1985 __ mov(rscratch1, newval);
1986 __ encode_heap_oop(rscratch1);
1987 if (os::is_MP()) {
1988 __ lock();
1989 }
1990 // cmpval (rax) is implicitly used by this instruction
1991 __ cmpxchgl(rscratch1, Address(addr, 0));
1992 } else
1993 #endif
1994 {
1995 if (os::is_MP()) {
1996 __ lock();
1997 }
1998 __ cmpxchgptr(newval, Address(addr, 0));
1999 }
2000 } else {
2001 assert(op->code() == lir_cas_int, "lir_cas_int expected");
2002 if (os::is_MP()) {
2003 __ lock();
2004 }
2005 __ cmpxchgl(newval, Address(addr, 0));
2006 }
2007 #ifdef _LP64
2008 } else if (op->code() == lir_cas_long) {
2009 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
2010 Register newval = op->new_value()->as_register_lo();
2011 Register cmpval = op->cmp_value()->as_register_lo();
2012 assert(cmpval == rax, "wrong register");
2013 assert(newval != NULL, "new val must be register");
2014 assert(cmpval != newval, "cmp and new values must be in different registers");
2015 assert(cmpval != addr, "cmp and addr must be in different registers");
2016 assert(newval != addr, "new value and addr must be in different registers");
2017 if (os::is_MP()) {
2018 __ lock();
2019 }
2020 __ cmpxchgq(newval, Address(addr, 0));
2021 #endif // _LP64
2022 } else {
2023 Unimplemented();
2024 }
2025 }
2026
2027 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
2028 Assembler::Condition acond, ncond;
2029 switch (condition) {
2030 case lir_cond_equal: acond = Assembler::equal; ncond = Assembler::notEqual; break;
2031 case lir_cond_notEqual: acond = Assembler::notEqual; ncond = Assembler::equal; break;
2032 case lir_cond_less: acond = Assembler::less; ncond = Assembler::greaterEqual; break;
2033 case lir_cond_lessEqual: acond = Assembler::lessEqual; ncond = Assembler::greater; break;
2034 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less; break;
2035 case lir_cond_greater: acond = Assembler::greater; ncond = Assembler::lessEqual; break;
2036 case lir_cond_belowEqual: acond = Assembler::belowEqual; ncond = Assembler::above; break;
2037 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; ncond = Assembler::below; break;
2038 default: ShouldNotReachHere();
2039 }
2040
2041 if (opr1->is_cpu_register()) {
2042 reg2reg(opr1, result);
2043 } else if (opr1->is_stack()) {
2044 stack2reg(opr1, result, result->type());
2045 } else if (opr1->is_constant()) {
2046 const2reg(opr1, result, lir_patch_none, NULL);
2047 } else {
2048 ShouldNotReachHere();
2049 }
2050
2051 if (VM_Version::supports_cmov() && !opr2->is_constant()) {
2052 // optimized version that does not require a branch
2053 if (opr2->is_single_cpu()) {
2054 assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
2055 __ cmov(ncond, result->as_register(), opr2->as_register());
2056 } else if (opr2->is_double_cpu()) {
2057 assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2058 assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2059 __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
2060 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
2061 } else if (opr2->is_single_stack()) {
2062 __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
2063 } else if (opr2->is_double_stack()) {
2064 __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
2065 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
2066 } else {
2067 ShouldNotReachHere();
2068 }
2069
2070 } else {
2071 Label skip;
2072 __ jcc (acond, skip);
2073 if (opr2->is_cpu_register()) {
2074 reg2reg(opr2, result);
2075 } else if (opr2->is_stack()) {
2076 stack2reg(opr2, result, result->type());
2077 } else if (opr2->is_constant()) {
2078 const2reg(opr2, result, lir_patch_none, NULL);
2079 } else {
2080 ShouldNotReachHere();
2081 }
2082 __ bind(skip);
2083 }
2084 }
2085
2086
2087 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
2088 assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
2089
2090 if (left->is_single_cpu()) {
2091 assert(left == dest, "left and dest must be equal");
2092 Register lreg = left->as_register();
2093
2094 if (right->is_single_cpu()) {
2095 // cpu register - cpu register
2096 Register rreg = right->as_register();
2097 switch (code) {
2098 case lir_add: __ addl (lreg, rreg); break;
2099 case lir_sub: __ subl (lreg, rreg); break;
2100 case lir_mul: __ imull(lreg, rreg); break;
2101 default: ShouldNotReachHere();
2102 }
2103
2104 } else if (right->is_stack()) {
2105 // cpu register - stack
2106 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2107 switch (code) {
2108 case lir_add: __ addl(lreg, raddr); break;
2109 case lir_sub: __ subl(lreg, raddr); break;
2110 default: ShouldNotReachHere();
2111 }
2112
2113 } else if (right->is_constant()) {
2114 // cpu register - constant
2115 jint c = right->as_constant_ptr()->as_jint();
2116 switch (code) {
2117 case lir_add: {
2118 __ incrementl(lreg, c);
2119 break;
2120 }
2121 case lir_sub: {
2122 __ decrementl(lreg, c);
2123 break;
2124 }
2125 default: ShouldNotReachHere();
2126 }
2127
2128 } else {
2129 ShouldNotReachHere();
2130 }
2131
2132 } else if (left->is_double_cpu()) {
2133 assert(left == dest, "left and dest must be equal");
2134 Register lreg_lo = left->as_register_lo();
2135 Register lreg_hi = left->as_register_hi();
2136
2137 if (right->is_double_cpu()) {
2138 // cpu register - cpu register
2139 Register rreg_lo = right->as_register_lo();
2140 Register rreg_hi = right->as_register_hi();
2141 NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
2142 LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
2143 switch (code) {
2144 case lir_add:
2145 __ addptr(lreg_lo, rreg_lo);
2146 NOT_LP64(__ adcl(lreg_hi, rreg_hi));
2147 break;
2148 case lir_sub:
2149 __ subptr(lreg_lo, rreg_lo);
2150 NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
2151 break;
2152 case lir_mul:
2153 #ifdef _LP64
2154 __ imulq(lreg_lo, rreg_lo);
2155 #else
2156 assert(lreg_lo == rax && lreg_hi == rdx, "must be");
2157 __ imull(lreg_hi, rreg_lo);
2158 __ imull(rreg_hi, lreg_lo);
2159 __ addl (rreg_hi, lreg_hi);
2160 __ mull (rreg_lo);
2161 __ addl (lreg_hi, rreg_hi);
2162 #endif // _LP64
2163 break;
2164 default:
2165 ShouldNotReachHere();
2166 }
2167
2168 } else if (right->is_constant()) {
2169 // cpu register - constant
2170 #ifdef _LP64
2171 jlong c = right->as_constant_ptr()->as_jlong_bits();
2172 __ movptr(r10, (intptr_t) c);
2173 switch (code) {
2174 case lir_add:
2175 __ addptr(lreg_lo, r10);
2176 break;
2177 case lir_sub:
2178 __ subptr(lreg_lo, r10);
2179 break;
2180 default:
2181 ShouldNotReachHere();
2182 }
2183 #else
2184 jint c_lo = right->as_constant_ptr()->as_jint_lo();
2185 jint c_hi = right->as_constant_ptr()->as_jint_hi();
2186 switch (code) {
2187 case lir_add:
2188 __ addptr(lreg_lo, c_lo);
2189 __ adcl(lreg_hi, c_hi);
2190 break;
2191 case lir_sub:
2192 __ subptr(lreg_lo, c_lo);
2193 __ sbbl(lreg_hi, c_hi);
2194 break;
2195 default:
2196 ShouldNotReachHere();
2197 }
2198 #endif // _LP64
2199
2200 } else {
2201 ShouldNotReachHere();
2202 }
2203
2204 } else if (left->is_single_xmm()) {
2205 assert(left == dest, "left and dest must be equal");
2206 XMMRegister lreg = left->as_xmm_float_reg();
2207
2208 if (right->is_single_xmm()) {
2209 XMMRegister rreg = right->as_xmm_float_reg();
2210 switch (code) {
2211 case lir_add: __ addss(lreg, rreg); break;
2212 case lir_sub: __ subss(lreg, rreg); break;
2213 case lir_mul_strictfp: // fall through
2214 case lir_mul: __ mulss(lreg, rreg); break;
2215 case lir_div_strictfp: // fall through
2216 case lir_div: __ divss(lreg, rreg); break;
2217 default: ShouldNotReachHere();
2218 }
2219 } else {
2220 Address raddr;
2221 if (right->is_single_stack()) {
2222 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2223 } else if (right->is_constant()) {
2224 // hack for now
2225 raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
2226 } else {
2227 ShouldNotReachHere();
2228 }
2229 switch (code) {
2230 case lir_add: __ addss(lreg, raddr); break;
2231 case lir_sub: __ subss(lreg, raddr); break;
2232 case lir_mul_strictfp: // fall through
2233 case lir_mul: __ mulss(lreg, raddr); break;
2234 case lir_div_strictfp: // fall through
2235 case lir_div: __ divss(lreg, raddr); break;
2236 default: ShouldNotReachHere();
2237 }
2238 }
2239
2240 } else if (left->is_double_xmm()) {
2241 assert(left == dest, "left and dest must be equal");
2242
2243 XMMRegister lreg = left->as_xmm_double_reg();
2244 if (right->is_double_xmm()) {
2245 XMMRegister rreg = right->as_xmm_double_reg();
2246 switch (code) {
2247 case lir_add: __ addsd(lreg, rreg); break;
2248 case lir_sub: __ subsd(lreg, rreg); break;
2249 case lir_mul_strictfp: // fall through
2250 case lir_mul: __ mulsd(lreg, rreg); break;
2251 case lir_div_strictfp: // fall through
2252 case lir_div: __ divsd(lreg, rreg); break;
2253 default: ShouldNotReachHere();
2254 }
2255 } else {
2256 Address raddr;
2257 if (right->is_double_stack()) {
2258 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2259 } else if (right->is_constant()) {
2260 // hack for now
2261 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2262 } else {
2263 ShouldNotReachHere();
2264 }
2265 switch (code) {
2266 case lir_add: __ addsd(lreg, raddr); break;
2267 case lir_sub: __ subsd(lreg, raddr); break;
2268 case lir_mul_strictfp: // fall through
2269 case lir_mul: __ mulsd(lreg, raddr); break;
2270 case lir_div_strictfp: // fall through
2271 case lir_div: __ divsd(lreg, raddr); break;
2272 default: ShouldNotReachHere();
2273 }
2274 }
2275
2276 } else if (left->is_single_fpu()) {
2277 assert(dest->is_single_fpu(), "fpu stack allocation required");
2278
2279 if (right->is_single_fpu()) {
2280 arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
2281
2282 } else {
2283 assert(left->fpu_regnr() == 0, "left must be on TOS");
2284 assert(dest->fpu_regnr() == 0, "dest must be on TOS");
2285
2286 Address raddr;
2287 if (right->is_single_stack()) {
2288 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2289 } else if (right->is_constant()) {
2290 address const_addr = float_constant(right->as_jfloat());
2291 assert(const_addr != NULL, "incorrect float/double constant maintainance");
2292 // hack for now
2293 raddr = __ as_Address(InternalAddress(const_addr));
2294 } else {
2295 ShouldNotReachHere();
2296 }
2297
2298 switch (code) {
2299 case lir_add: __ fadd_s(raddr); break;
2300 case lir_sub: __ fsub_s(raddr); break;
2301 case lir_mul_strictfp: // fall through
2302 case lir_mul: __ fmul_s(raddr); break;
2303 case lir_div_strictfp: // fall through
2304 case lir_div: __ fdiv_s(raddr); break;
2305 default: ShouldNotReachHere();
2306 }
2307 }
2308
2309 } else if (left->is_double_fpu()) {
2310 assert(dest->is_double_fpu(), "fpu stack allocation required");
2311
2312 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2313 // Double values require special handling for strictfp mul/div on x86
2314 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
2315 __ fmulp(left->fpu_regnrLo() + 1);
2316 }
2317
2318 if (right->is_double_fpu()) {
2319 arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
2320
2321 } else {
2322 assert(left->fpu_regnrLo() == 0, "left must be on TOS");
2323 assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
2324
2325 Address raddr;
2326 if (right->is_double_stack()) {
2327 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2328 } else if (right->is_constant()) {
2329 // hack for now
2330 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2331 } else {
2332 ShouldNotReachHere();
2333 }
2334
2335 switch (code) {
2336 case lir_add: __ fadd_d(raddr); break;
2337 case lir_sub: __ fsub_d(raddr); break;
2338 case lir_mul_strictfp: // fall through
2339 case lir_mul: __ fmul_d(raddr); break;
2340 case lir_div_strictfp: // fall through
2341 case lir_div: __ fdiv_d(raddr); break;
2342 default: ShouldNotReachHere();
2343 }
2344 }
2345
2346 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2347 // Double values require special handling for strictfp mul/div on x86
2348 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
2349 __ fmulp(dest->fpu_regnrLo() + 1);
2350 }
2351
2352 } else if (left->is_single_stack() || left->is_address()) {
2353 assert(left == dest, "left and dest must be equal");
2354
2355 Address laddr;
2356 if (left->is_single_stack()) {
2357 laddr = frame_map()->address_for_slot(left->single_stack_ix());
2358 } else if (left->is_address()) {
2359 laddr = as_Address(left->as_address_ptr());
2360 } else {
2361 ShouldNotReachHere();
2362 }
2363
2364 if (right->is_single_cpu()) {
2365 Register rreg = right->as_register();
2366 switch (code) {
2367 case lir_add: __ addl(laddr, rreg); break;
2368 case lir_sub: __ subl(laddr, rreg); break;
2369 default: ShouldNotReachHere();
2370 }
2371 } else if (right->is_constant()) {
2372 jint c = right->as_constant_ptr()->as_jint();
2373 switch (code) {
2374 case lir_add: {
2375 __ incrementl(laddr, c);
2376 break;
2377 }
2378 case lir_sub: {
2379 __ decrementl(laddr, c);
2380 break;
2381 }
2382 default: ShouldNotReachHere();
2383 }
2384 } else {
2385 ShouldNotReachHere();
2386 }
2387
2388 } else {
2389 ShouldNotReachHere();
2390 }
2391 }
2392
2393 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
2394 assert(pop_fpu_stack || (left_index == dest_index || right_index == dest_index), "invalid LIR");
2395 assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
2396 assert(left_index == 0 || right_index == 0, "either must be on top of stack");
2397
2398 bool left_is_tos = (left_index == 0);
2399 bool dest_is_tos = (dest_index == 0);
2400 int non_tos_index = (left_is_tos ? right_index : left_index);
2401
2402 switch (code) {
2403 case lir_add:
2404 if (pop_fpu_stack) __ faddp(non_tos_index);
2405 else if (dest_is_tos) __ fadd (non_tos_index);
2406 else __ fadda(non_tos_index);
2407 break;
2408
2409 case lir_sub:
2410 if (left_is_tos) {
2411 if (pop_fpu_stack) __ fsubrp(non_tos_index);
2412 else if (dest_is_tos) __ fsub (non_tos_index);
2413 else __ fsubra(non_tos_index);
2414 } else {
2415 if (pop_fpu_stack) __ fsubp (non_tos_index);
2416 else if (dest_is_tos) __ fsubr (non_tos_index);
2417 else __ fsuba (non_tos_index);
2418 }
2419 break;
2420
2421 case lir_mul_strictfp: // fall through
2422 case lir_mul:
2423 if (pop_fpu_stack) __ fmulp(non_tos_index);
2424 else if (dest_is_tos) __ fmul (non_tos_index);
2425 else __ fmula(non_tos_index);
2426 break;
2427
2428 case lir_div_strictfp: // fall through
2429 case lir_div:
2430 if (left_is_tos) {
2431 if (pop_fpu_stack) __ fdivrp(non_tos_index);
2432 else if (dest_is_tos) __ fdiv (non_tos_index);
2433 else __ fdivra(non_tos_index);
2434 } else {
2435 if (pop_fpu_stack) __ fdivp (non_tos_index);
2436 else if (dest_is_tos) __ fdivr (non_tos_index);
2437 else __ fdiva (non_tos_index);
2438 }
2439 break;
2440
2441 case lir_rem:
2442 assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
2443 __ fremr(noreg);
2444 break;
2445
2446 default:
2447 ShouldNotReachHere();
2448 }
2449 }
2450
2451
2452 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
2453 if (value->is_double_xmm()) {
2454 switch(code) {
2455 case lir_abs :
2456 {
2457 if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
2458 __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
2459 }
2460 __ andpd(dest->as_xmm_double_reg(),
2461 ExternalAddress((address)double_signmask_pool));
2462 }
2463 break;
2464
2465 case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
2466 // all other intrinsics are not available in the SSE instruction set, so FPU is used
2467 default : ShouldNotReachHere();
2468 }
2469
2470 } else if (value->is_double_fpu()) {
2471 assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
2472 switch(code) {
2473 case lir_log : __ flog() ; break;
2474 case lir_log10 : __ flog10() ; break;
2475 case lir_abs : __ fabs() ; break;
2476 case lir_sqrt : __ fsqrt(); break;
2477 case lir_sin :
2478 // Should consider not saving rbx, if not necessary
2479 __ trigfunc('s', op->as_Op2()->fpu_stack_size());
2480 break;
2481 case lir_cos :
2482 // Should consider not saving rbx, if not necessary
2483 assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
2484 __ trigfunc('c', op->as_Op2()->fpu_stack_size());
2485 break;
2486 case lir_tan :
2487 // Should consider not saving rbx, if not necessary
2488 __ trigfunc('t', op->as_Op2()->fpu_stack_size());
2489 break;
2490 default : ShouldNotReachHere();
2491 }
2492 } else {
2493 Unimplemented();
2494 }
2495 }
2496
2497 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
2498 // assert(left->destroys_register(), "check");
2499 if (left->is_single_cpu()) {
2500 Register reg = left->as_register();
2501 if (right->is_constant()) {
2502 int val = right->as_constant_ptr()->as_jint();
2503 switch (code) {
2504 case lir_logic_and: __ andl (reg, val); break;
2505 case lir_logic_or: __ orl (reg, val); break;
2506 case lir_logic_xor: __ xorl (reg, val); break;
2507 default: ShouldNotReachHere();
2508 }
2509 } else if (right->is_stack()) {
2510 // added support for stack operands
2511 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2512 switch (code) {
2513 case lir_logic_and: __ andl (reg, raddr); break;
2514 case lir_logic_or: __ orl (reg, raddr); break;
2515 case lir_logic_xor: __ xorl (reg, raddr); break;
2516 default: ShouldNotReachHere();
2517 }
2518 } else {
2519 Register rright = right->as_register();
2520 switch (code) {
2521 case lir_logic_and: __ andptr (reg, rright); break;
2522 case lir_logic_or : __ orptr (reg, rright); break;
2523 case lir_logic_xor: __ xorptr (reg, rright); break;
2524 default: ShouldNotReachHere();
2525 }
2526 }
2527 move_regs(reg, dst->as_register());
2528 } else {
2529 Register l_lo = left->as_register_lo();
2530 Register l_hi = left->as_register_hi();
2531 if (right->is_constant()) {
2532 #ifdef _LP64
2533 __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
2534 switch (code) {
2535 case lir_logic_and:
2536 __ andq(l_lo, rscratch1);
2537 break;
2538 case lir_logic_or:
2539 __ orq(l_lo, rscratch1);
2540 break;
2541 case lir_logic_xor:
2542 __ xorq(l_lo, rscratch1);
2543 break;
2544 default: ShouldNotReachHere();
2545 }
2546 #else
2547 int r_lo = right->as_constant_ptr()->as_jint_lo();
2548 int r_hi = right->as_constant_ptr()->as_jint_hi();
2549 switch (code) {
2550 case lir_logic_and:
2551 __ andl(l_lo, r_lo);
2552 __ andl(l_hi, r_hi);
2553 break;
2554 case lir_logic_or:
2555 __ orl(l_lo, r_lo);
2556 __ orl(l_hi, r_hi);
2557 break;
2558 case lir_logic_xor:
2559 __ xorl(l_lo, r_lo);
2560 __ xorl(l_hi, r_hi);
2561 break;
2562 default: ShouldNotReachHere();
2563 }
2564 #endif // _LP64
2565 } else {
2566 #ifdef _LP64
2567 Register r_lo;
2568 if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
2569 r_lo = right->as_register();
2570 } else {
2571 r_lo = right->as_register_lo();
2572 }
2573 #else
2574 Register r_lo = right->as_register_lo();
2575 Register r_hi = right->as_register_hi();
2576 assert(l_lo != r_hi, "overwriting registers");
2577 #endif
2578 switch (code) {
2579 case lir_logic_and:
2580 __ andptr(l_lo, r_lo);
2581 NOT_LP64(__ andptr(l_hi, r_hi);)
2582 break;
2583 case lir_logic_or:
2584 __ orptr(l_lo, r_lo);
2585 NOT_LP64(__ orptr(l_hi, r_hi);)
2586 break;
2587 case lir_logic_xor:
2588 __ xorptr(l_lo, r_lo);
2589 NOT_LP64(__ xorptr(l_hi, r_hi);)
2590 break;
2591 default: ShouldNotReachHere();
2592 }
2593 }
2594
2595 Register dst_lo = dst->as_register_lo();
2596 Register dst_hi = dst->as_register_hi();
2597
2598 #ifdef _LP64
2599 move_regs(l_lo, dst_lo);
2600 #else
2601 if (dst_lo == l_hi) {
2602 assert(dst_hi != l_lo, "overwriting registers");
2603 move_regs(l_hi, dst_hi);
2604 move_regs(l_lo, dst_lo);
2605 } else {
2606 assert(dst_lo != l_hi, "overwriting registers");
2607 move_regs(l_lo, dst_lo);
2608 move_regs(l_hi, dst_hi);
2609 }
2610 #endif // _LP64
2611 }
2612 }
2613
2614
2615 // we assume that rax, and rdx can be overwritten
2616 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
2617
2618 assert(left->is_single_cpu(), "left must be register");
2619 assert(right->is_single_cpu() || right->is_constant(), "right must be register or constant");
2620 assert(result->is_single_cpu(), "result must be register");
2621
2622 // assert(left->destroys_register(), "check");
2623 // assert(right->destroys_register(), "check");
2624
2625 Register lreg = left->as_register();
2626 Register dreg = result->as_register();
2627
2628 if (right->is_constant()) {
2629 int divisor = right->as_constant_ptr()->as_jint();
2630 assert(divisor > 0 && is_power_of_2(divisor), "must be");
2631 if (code == lir_idiv) {
2632 assert(lreg == rax, "must be rax,");
2633 assert(temp->as_register() == rdx, "tmp register must be rdx");
2634 __ cdql(); // sign extend into rdx:rax
2635 if (divisor == 2) {
2636 __ subl(lreg, rdx);
2637 } else {
2638 __ andl(rdx, divisor - 1);
2639 __ addl(lreg, rdx);
2640 }
2641 __ sarl(lreg, log2_intptr(divisor));
2642 move_regs(lreg, dreg);
2643 } else if (code == lir_irem) {
2644 Label done;
2645 __ mov(dreg, lreg);
2646 __ andl(dreg, 0x80000000 | (divisor - 1));
2647 __ jcc(Assembler::positive, done);
2648 __ decrement(dreg);
2649 __ orl(dreg, ~(divisor - 1));
2650 __ increment(dreg);
2651 __ bind(done);
2652 } else {
2653 ShouldNotReachHere();
2654 }
2655 } else {
2656 Register rreg = right->as_register();
2657 assert(lreg == rax, "left register must be rax,");
2658 assert(rreg != rdx, "right register must not be rdx");
2659 assert(temp->as_register() == rdx, "tmp register must be rdx");
2660
2661 move_regs(lreg, rax);
2662
2663 int idivl_offset = __ corrected_idivl(rreg);
2664 add_debug_info_for_div0(idivl_offset, info);
2665 if (code == lir_irem) {
2666 move_regs(rdx, dreg); // result is in rdx
2667 } else {
2668 move_regs(rax, dreg);
2669 }
2670 }
2671 }
2672
2673
2674 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
2675 if (opr1->is_single_cpu()) {
2676 Register reg1 = opr1->as_register();
2677 if (opr2->is_single_cpu()) {
2678 // cpu register - cpu register
2679 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2680 __ cmpptr(reg1, opr2->as_register());
2681 } else {
2682 assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
2683 __ cmpl(reg1, opr2->as_register());
2684 }
2685 } else if (opr2->is_stack()) {
2686 // cpu register - stack
2687 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2688 __ cmpptr(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2689 } else {
2690 __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2691 }
2692 } else if (opr2->is_constant()) {
2693 // cpu register - constant
2694 LIR_Const* c = opr2->as_constant_ptr();
2695 if (c->type() == T_INT) {
2696 __ cmpl(reg1, c->as_jint());
2697 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2698 // In 64bit oops are single register
2699 jobject o = c->as_jobject();
2700 if (o == NULL) {
2701 __ cmpptr(reg1, (int32_t)NULL_WORD);
2702 } else {
2703 #ifdef _LP64
2704 __ movoop(rscratch1, o);
2705 __ cmpptr(reg1, rscratch1);
2706 #else
2707 __ cmpoop(reg1, c->as_jobject());
2708 #endif // _LP64
2709 }
2710 } else {
2711 ShouldNotReachHere();
2712 }
2713 // cpu register - address
2714 } else if (opr2->is_address()) {
2715 if (op->info() != NULL) {
2716 add_debug_info_for_null_check_here(op->info());
2717 }
2718 __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
2719 } else {
2720 ShouldNotReachHere();
2721 }
2722
2723 } else if(opr1->is_double_cpu()) {
2724 Register xlo = opr1->as_register_lo();
2725 Register xhi = opr1->as_register_hi();
2726 if (opr2->is_double_cpu()) {
2727 #ifdef _LP64
2728 __ cmpptr(xlo, opr2->as_register_lo());
2729 #else
2730 // cpu register - cpu register
2731 Register ylo = opr2->as_register_lo();
2732 Register yhi = opr2->as_register_hi();
2733 __ subl(xlo, ylo);
2734 __ sbbl(xhi, yhi);
2735 if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
2736 __ orl(xhi, xlo);
2737 }
2738 #endif // _LP64
2739 } else if (opr2->is_constant()) {
2740 // cpu register - constant 0
2741 assert(opr2->as_jlong() == (jlong)0, "only handles zero");
2742 #ifdef _LP64
2743 __ cmpptr(xlo, (int32_t)opr2->as_jlong());
2744 #else
2745 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
2746 __ orl(xhi, xlo);
2747 #endif // _LP64
2748 } else {
2749 ShouldNotReachHere();
2750 }
2751
2752 } else if (opr1->is_single_xmm()) {
2753 XMMRegister reg1 = opr1->as_xmm_float_reg();
2754 if (opr2->is_single_xmm()) {
2755 // xmm register - xmm register
2756 __ ucomiss(reg1, opr2->as_xmm_float_reg());
2757 } else if (opr2->is_stack()) {
2758 // xmm register - stack
2759 __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2760 } else if (opr2->is_constant()) {
2761 // xmm register - constant
2762 __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
2763 } else if (opr2->is_address()) {
2764 // xmm register - address
2765 if (op->info() != NULL) {
2766 add_debug_info_for_null_check_here(op->info());
2767 }
2768 __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
2769 } else {
2770 ShouldNotReachHere();
2771 }
2772
2773 } else if (opr1->is_double_xmm()) {
2774 XMMRegister reg1 = opr1->as_xmm_double_reg();
2775 if (opr2->is_double_xmm()) {
2776 // xmm register - xmm register
2777 __ ucomisd(reg1, opr2->as_xmm_double_reg());
2778 } else if (opr2->is_stack()) {
2779 // xmm register - stack
2780 __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
2781 } else if (opr2->is_constant()) {
2782 // xmm register - constant
2783 __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
2784 } else if (opr2->is_address()) {
2785 // xmm register - address
2786 if (op->info() != NULL) {
2787 add_debug_info_for_null_check_here(op->info());
2788 }
2789 __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
2790 } else {
2791 ShouldNotReachHere();
2792 }
2793
2794 } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
2795 assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
2796 assert(opr2->is_fpu_register(), "both must be registers");
2797 __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2798
2799 } else if (opr1->is_address() && opr2->is_constant()) {
2800 LIR_Const* c = opr2->as_constant_ptr();
2801 #ifdef _LP64
2802 if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2803 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
2804 __ movoop(rscratch1, c->as_jobject());
2805 }
2806 #endif // LP64
2807 if (op->info() != NULL) {
2808 add_debug_info_for_null_check_here(op->info());
2809 }
2810 // special case: address - constant
2811 LIR_Address* addr = opr1->as_address_ptr();
2812 if (c->type() == T_INT) {
2813 __ cmpl(as_Address(addr), c->as_jint());
2814 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2815 #ifdef _LP64
2816 // %%% Make this explode if addr isn't reachable until we figure out a
2817 // better strategy by giving noreg as the temp for as_Address
2818 __ cmpptr(rscratch1, as_Address(addr, noreg));
2819 #else
2820 __ cmpoop(as_Address(addr), c->as_jobject());
2821 #endif // _LP64
2822 } else {
2823 ShouldNotReachHere();
2824 }
2825
2826 } else {
2827 ShouldNotReachHere();
2828 }
2829 }
2830
2831 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
2832 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
2833 if (left->is_single_xmm()) {
2834 assert(right->is_single_xmm(), "must match");
2835 __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2836 } else if (left->is_double_xmm()) {
2837 assert(right->is_double_xmm(), "must match");
2838 __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2839
2840 } else {
2841 assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
2842 assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
2843
2844 assert(left->fpu() == 0, "left must be on TOS");
2845 __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
2846 op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2847 }
2848 } else {
2849 assert(code == lir_cmp_l2i, "check");
2850 #ifdef _LP64
2851 Label done;
2852 Register dest = dst->as_register();
2853 __ cmpptr(left->as_register_lo(), right->as_register_lo());
2854 __ movl(dest, -1);
2855 __ jccb(Assembler::less, done);
2856 __ set_byte_if_not_zero(dest);
2857 __ movzbl(dest, dest);
2858 __ bind(done);
2859 #else
2860 __ lcmp2int(left->as_register_hi(),
2861 left->as_register_lo(),
2862 right->as_register_hi(),
2863 right->as_register_lo());
2864 move_regs(left->as_register_hi(), dst->as_register());
2865 #endif // _LP64
2866 }
2867 }
2868
2869
2870 void LIR_Assembler::align_call(LIR_Code code) {
2871 if (os::is_MP()) {
2872 // make sure that the displacement word of the call ends up word aligned
2873 int offset = __ offset();
2874 switch (code) {
2875 case lir_static_call:
2876 case lir_optvirtual_call:
2877 case lir_dynamic_call:
2878 offset += NativeCall::displacement_offset;
2879 break;
2880 case lir_icvirtual_call:
2881 offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
2882 break;
2883 case lir_virtual_call: // currently, sparc-specific for niagara
2884 default: ShouldNotReachHere();
2885 }
2886 while (offset++ % BytesPerWord != 0) {
2887 __ nop();
2888 }
2889 }
2890 }
2891
2892
2893 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
2894 assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2895 "must be aligned");
2896 __ call(AddressLiteral(op->addr(), rtype));
2897 add_call_info(code_offset(), op->info());
2898 }
2899
2900
2901 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
2902 RelocationHolder rh = virtual_call_Relocation::spec(pc());
2903 __ movoop(IC_Klass, (jobject)Universe::non_oop_word());
2904 assert(!os::is_MP() ||
2905 (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2906 "must be aligned");
2907 __ call(AddressLiteral(op->addr(), rh));
2908 add_call_info(code_offset(), op->info());
2909 }
2910
2911
2912 /* Currently, vtable-dispatch is only enabled for sparc platforms */
2913 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
2914 ShouldNotReachHere();
2915 }
2916
2917
2918 void LIR_Assembler::emit_static_call_stub() {
2919 address call_pc = __ pc();
2920 address stub = __ start_a_stub(call_stub_size);
2921 if (stub == NULL) {
2922 bailout("static call stub overflow");
2923 return;
2924 }
2925
2926 int start = __ offset();
2927 if (os::is_MP()) {
2928 // make sure that the displacement word of the call ends up word aligned
2929 int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
2930 while (offset++ % BytesPerWord != 0) {
2931 __ nop();
2932 }
2933 }
2934 __ relocate(static_stub_Relocation::spec(call_pc));
2935 __ movoop(rbx, (jobject)NULL);
2936 // must be set to -1 at code generation time
2937 assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
2938 // On 64bit this will die since it will take a movq & jmp, must be only a jmp
2939 __ jump(RuntimeAddress(__ pc()));
2940
2941 assert(__ offset() - start <= call_stub_size, "stub too big");
2942 __ end_a_stub();
2943 }
2944
2945
2946 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2947 assert(exceptionOop->as_register() == rax, "must match");
2948 assert(exceptionPC->as_register() == rdx, "must match");
2949
2950 // exception object is not added to oop map by LinearScan
2951 // (LinearScan assumes that no oops are in fixed registers)
2952 info->add_register_oop(exceptionOop);
2953 Runtime1::StubID unwind_id;
2954
2955 // get current pc information
2956 // pc is only needed if the method has an exception handler, the unwind code does not need it.
2957 int pc_for_athrow_offset = __ offset();
2958 InternalAddress pc_for_athrow(__ pc());
2959 __ lea(exceptionPC->as_register(), pc_for_athrow);
2960 add_call_info(pc_for_athrow_offset, info); // for exception handler
2961
2962 __ verify_not_null_oop(rax);
2963 // search an exception handler (rax: exception oop, rdx: throwing pc)
2964 if (compilation()->has_fpu_code()) {
2965 unwind_id = Runtime1::handle_exception_id;
2966 } else {
2967 unwind_id = Runtime1::handle_exception_nofpu_id;
2968 }
2969 __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2970
2971 // enough room for two byte trap
2972 __ nop();
2973 }
2974
2975
2976 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2977 assert(exceptionOop->as_register() == rax, "must match");
2978
2979 __ jmp(_unwind_handler_entry);
2980 }
2981
2982
2983 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2984
2985 // optimized version for linear scan:
2986 // * count must be already in ECX (guaranteed by LinearScan)
2987 // * left and dest must be equal
2988 // * tmp must be unused
2989 assert(count->as_register() == SHIFT_count, "count must be in ECX");
2990 assert(left == dest, "left and dest must be equal");
2991 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2992
2993 if (left->is_single_cpu()) {
2994 Register value = left->as_register();
2995 assert(value != SHIFT_count, "left cannot be ECX");
2996
2997 switch (code) {
2998 case lir_shl: __ shll(value); break;
2999 case lir_shr: __ sarl(value); break;
3000 case lir_ushr: __ shrl(value); break;
3001 default: ShouldNotReachHere();
3002 }
3003 } else if (left->is_double_cpu()) {
3004 Register lo = left->as_register_lo();
3005 Register hi = left->as_register_hi();
3006 assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
3007 #ifdef _LP64
3008 switch (code) {
3009 case lir_shl: __ shlptr(lo); break;
3010 case lir_shr: __ sarptr(lo); break;
3011 case lir_ushr: __ shrptr(lo); break;
3012 default: ShouldNotReachHere();
3013 }
3014 #else
3015
3016 switch (code) {
3017 case lir_shl: __ lshl(hi, lo); break;
3018 case lir_shr: __ lshr(hi, lo, true); break;
3019 case lir_ushr: __ lshr(hi, lo, false); break;
3020 default: ShouldNotReachHere();
3021 }
3022 #endif // LP64
3023 } else {
3024 ShouldNotReachHere();
3025 }
3026 }
3027
3028
3029 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
3030 if (dest->is_single_cpu()) {
3031 // first move left into dest so that left is not destroyed by the shift
3032 Register value = dest->as_register();
3033 count = count & 0x1F; // Java spec
3034
3035 move_regs(left->as_register(), value);
3036 switch (code) {
3037 case lir_shl: __ shll(value, count); break;
3038 case lir_shr: __ sarl(value, count); break;
3039 case lir_ushr: __ shrl(value, count); break;
3040 default: ShouldNotReachHere();
3041 }
3042 } else if (dest->is_double_cpu()) {
3043 #ifndef _LP64
3044 Unimplemented();
3045 #else
3046 // first move left into dest so that left is not destroyed by the shift
3047 Register value = dest->as_register_lo();
3048 count = count & 0x1F; // Java spec
3049
3050 move_regs(left->as_register_lo(), value);
3051 switch (code) {
3052 case lir_shl: __ shlptr(value, count); break;
3053 case lir_shr: __ sarptr(value, count); break;
3054 case lir_ushr: __ shrptr(value, count); break;
3055 default: ShouldNotReachHere();
3056 }
3057 #endif // _LP64
3058 } else {
3059 ShouldNotReachHere();
3060 }
3061 }
3062
3063
3064 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
3065 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3066 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3067 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3068 __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
3069 }
3070
3071
3072 void LIR_Assembler::store_parameter(jint c, int offset_from_rsp_in_words) {
3073 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3074 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3075 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3076 __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
3077 }
3078
3079
3080 void LIR_Assembler::store_parameter(jobject o, int offset_from_rsp_in_words) {
3081 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3082 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3083 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3084 __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
3085 }
3086
3087
3088 // This code replaces a call to arraycopy; no exception may
3089 // be thrown in this code, they must be thrown in the System.arraycopy
3090 // activation frame; we could save some checks if this would not be the case
3091 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
3092 ciArrayKlass* default_type = op->expected_type();
3093 Register src = op->src()->as_register();
3094 Register dst = op->dst()->as_register();
3095 Register src_pos = op->src_pos()->as_register();
3096 Register dst_pos = op->dst_pos()->as_register();
3097 Register length = op->length()->as_register();
3098 Register tmp = op->tmp()->as_register();
3099
3100 CodeStub* stub = op->stub();
3101 int flags = op->flags();
3102 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
3103 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
3104
3105 // if we don't know anything, just go through the generic arraycopy
3106 if (default_type == NULL) {
3107 Label done;
3108 // save outgoing arguments on stack in case call to System.arraycopy is needed
3109 // HACK ALERT. This code used to push the parameters in a hardwired fashion
3110 // for interpreter calling conventions. Now we have to do it in new style conventions.
3111 // For the moment until C1 gets the new register allocator I just force all the
3112 // args to the right place (except the register args) and then on the back side
3113 // reload the register args properly if we go slow path. Yuck
3114
3115 // These are proper for the calling convention
3116 store_parameter(length, 2);
3117 store_parameter(dst_pos, 1);
3118 store_parameter(dst, 0);
3119
3120 // these are just temporary placements until we need to reload
3121 store_parameter(src_pos, 3);
3122 store_parameter(src, 4);
3123 NOT_LP64(assert(src == rcx && src_pos == rdx, "mismatch in calling convention");)
3124
3125 address C_entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
3126
3127 address copyfunc_addr = StubRoutines::generic_arraycopy();
3128
3129 // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
3130 #ifdef _LP64
3131 // The arguments are in java calling convention so we can trivially shift them to C
3132 // convention
3133 assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
3134 __ mov(c_rarg0, j_rarg0);
3135 assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
3136 __ mov(c_rarg1, j_rarg1);
3137 assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
3138 __ mov(c_rarg2, j_rarg2);
3139 assert_different_registers(c_rarg3, j_rarg4);
3140 __ mov(c_rarg3, j_rarg3);
3141 #ifdef _WIN64
3142 // Allocate abi space for args but be sure to keep stack aligned
3143 __ subptr(rsp, 6*wordSize);
3144 store_parameter(j_rarg4, 4);
3145 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3146 __ call(RuntimeAddress(C_entry));
3147 } else {
3148 #ifndef PRODUCT
3149 if (PrintC1Statistics) {
3150 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3151 }
3152 #endif
3153 __ call(RuntimeAddress(copyfunc_addr));
3154 }
3155 __ addptr(rsp, 6*wordSize);
3156 #else
3157 __ mov(c_rarg4, j_rarg4);
3158 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3159 __ call(RuntimeAddress(C_entry));
3160 } else {
3161 #ifndef PRODUCT
3162 if (PrintC1Statistics) {
3163 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3164 }
3165 #endif
3166 __ call(RuntimeAddress(copyfunc_addr));
3167 }
3168 #endif // _WIN64
3169 #else
3170 __ push(length);
3171 __ push(dst_pos);
3172 __ push(dst);
3173 __ push(src_pos);
3174 __ push(src);
3175
3176 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3177 __ call_VM_leaf(C_entry, 5); // removes pushed parameter from the stack
3178 } else {
3179 #ifndef PRODUCT
3180 if (PrintC1Statistics) {
3181 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3182 }
3183 #endif
3184 __ call_VM_leaf(copyfunc_addr, 5); // removes pushed parameter from the stack
3185 }
3186
3187 #endif // _LP64
3188
3189 __ cmpl(rax, 0);
3190 __ jcc(Assembler::equal, *stub->continuation());
3191
3192 if (copyfunc_addr != NULL) {
3193 __ mov(tmp, rax);
3194 __ xorl(tmp, -1);
3195 }
3196
3197 // Reload values from the stack so they are where the stub
3198 // expects them.
3199 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3200 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3201 __ movptr (length, Address(rsp, 2*BytesPerWord));
3202 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3203 __ movptr (src, Address(rsp, 4*BytesPerWord));
3204
3205 if (copyfunc_addr != NULL) {
3206 __ subl(length, tmp);
3207 __ addl(src_pos, tmp);
3208 __ addl(dst_pos, tmp);
3209 }
3210 __ jmp(*stub->entry());
3211
3212 __ bind(*stub->continuation());
3213 return;
3214 }
3215
3216 assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
3217
3218 int elem_size = type2aelembytes(basic_type);
3219 int shift_amount;
3220 Address::ScaleFactor scale;
3221
3222 switch (elem_size) {
3223 case 1 :
3224 shift_amount = 0;
3225 scale = Address::times_1;
3226 break;
3227 case 2 :
3228 shift_amount = 1;
3229 scale = Address::times_2;
3230 break;
3231 case 4 :
3232 shift_amount = 2;
3233 scale = Address::times_4;
3234 break;
3235 case 8 :
3236 shift_amount = 3;
3237 scale = Address::times_8;
3238 break;
3239 default:
3240 ShouldNotReachHere();
3241 }
3242
3243 Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
3244 Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
3245 Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
3246 Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
3247
3248 // length and pos's are all sign extended at this point on 64bit
3249
3250 // test for NULL
3251 if (flags & LIR_OpArrayCopy::src_null_check) {
3252 __ testptr(src, src);
3253 __ jcc(Assembler::zero, *stub->entry());
3254 }
3255 if (flags & LIR_OpArrayCopy::dst_null_check) {
3256 __ testptr(dst, dst);
3257 __ jcc(Assembler::zero, *stub->entry());
3258 }
3259
3260 // check if negative
3261 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
3262 __ testl(src_pos, src_pos);
3263 __ jcc(Assembler::less, *stub->entry());
3264 }
3265 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
3266 __ testl(dst_pos, dst_pos);
3267 __ jcc(Assembler::less, *stub->entry());
3268 }
3269
3270 if (flags & LIR_OpArrayCopy::src_range_check) {
3271 __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
3272 __ cmpl(tmp, src_length_addr);
3273 __ jcc(Assembler::above, *stub->entry());
3274 }
3275 if (flags & LIR_OpArrayCopy::dst_range_check) {
3276 __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
3277 __ cmpl(tmp, dst_length_addr);
3278 __ jcc(Assembler::above, *stub->entry());
3279 }
3280
3281 if (flags & LIR_OpArrayCopy::length_positive_check) {
3282 __ testl(length, length);
3283 __ jcc(Assembler::less, *stub->entry());
3284 __ jcc(Assembler::zero, *stub->continuation());
3285 }
3286
3287 #ifdef _LP64
3288 __ movl2ptr(src_pos, src_pos); //higher 32bits must be null
3289 __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null
3290 #endif
3291
3292 if (flags & LIR_OpArrayCopy::type_check) {
3293 // We don't know the array types are compatible
3294 if (basic_type != T_OBJECT) {
3295 // Simple test for basic type arrays
3296 if (UseCompressedOops) {
3297 __ movl(tmp, src_klass_addr);
3298 __ cmpl(tmp, dst_klass_addr);
3299 } else {
3300 __ movptr(tmp, src_klass_addr);
3301 __ cmpptr(tmp, dst_klass_addr);
3302 }
3303 __ jcc(Assembler::notEqual, *stub->entry());
3304 } else {
3305 // For object arrays, if src is a sub class of dst then we can
3306 // safely do the copy.
3307 Label cont, slow;
3308
3309 __ push(src);
3310 __ push(dst);
3311
3312 __ load_klass(src, src);
3313 __ load_klass(dst, dst);
3314
3315 __ check_klass_subtype_fast_path(src, dst, tmp, &cont, &slow, NULL);
3316
3317 __ push(src);
3318 __ push(dst);
3319 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
3320 __ pop(dst);
3321 __ pop(src);
3322
3323 __ cmpl(src, 0);
3324 __ jcc(Assembler::notEqual, cont);
3325
3326 __ bind(slow);
3327 __ pop(dst);
3328 __ pop(src);
3329
3330 address copyfunc_addr = StubRoutines::checkcast_arraycopy();
3331 if (copyfunc_addr != NULL) { // use stub if available
3332 // src is not a sub class of dst so we have to do a
3333 // per-element check.
3334
3335 int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
3336 if ((flags & mask) != mask) {
3337 // Check that at least both of them object arrays.
3338 assert(flags & mask, "one of the two should be known to be an object array");
3339
3340 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
3341 __ load_klass(tmp, src);
3342 } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
3343 __ load_klass(tmp, dst);
3344 }
3345 int lh_offset = klassOopDesc::header_size() * HeapWordSize +
3346 Klass::layout_helper_offset_in_bytes();
3347 Address klass_lh_addr(tmp, lh_offset);
3348 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
3349 __ cmpl(klass_lh_addr, objArray_lh);
3350 __ jcc(Assembler::notEqual, *stub->entry());
3351 }
3352
3353 // Spill because stubs can use any register they like and it's
3354 // easier to restore just those that we care about.
3355 store_parameter(dst, 0);
3356 store_parameter(dst_pos, 1);
3357 store_parameter(length, 2);
3358 store_parameter(src_pos, 3);
3359 store_parameter(src, 4);
3360
3361 #ifndef _LP64
3362 __ movptr(tmp, dst_klass_addr);
3363 __ movptr(tmp, Address(tmp, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
3364 __ push(tmp);
3365 __ movl(tmp, Address(tmp, Klass::super_check_offset_offset_in_bytes() + sizeof(oopDesc)));
3366 __ push(tmp);
3367 __ push(length);
3368 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3369 __ push(tmp);
3370 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3371 __ push(tmp);
3372
3373 __ call_VM_leaf(copyfunc_addr, 5);
3374 #else
3375 __ movl2ptr(length, length); //higher 32bits must be null
3376
3377 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3378 assert_different_registers(c_rarg0, dst, dst_pos, length);
3379 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3380 assert_different_registers(c_rarg1, dst, length);
3381
3382 __ mov(c_rarg2, length);
3383 assert_different_registers(c_rarg2, dst);
3384
3385 #ifdef _WIN64
3386 // Allocate abi space for args but be sure to keep stack aligned
3387 __ subptr(rsp, 6*wordSize);
3388 __ load_klass(c_rarg3, dst);
3389 __ movptr(c_rarg3, Address(c_rarg3, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
3390 store_parameter(c_rarg3, 4);
3391 __ movl(c_rarg3, Address(c_rarg3, Klass::super_check_offset_offset_in_bytes() + sizeof(oopDesc)));
3392 __ call(RuntimeAddress(copyfunc_addr));
3393 __ addptr(rsp, 6*wordSize);
3394 #else
3395 __ load_klass(c_rarg4, dst);
3396 __ movptr(c_rarg4, Address(c_rarg4, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
3397 __ movl(c_rarg3, Address(c_rarg4, Klass::super_check_offset_offset_in_bytes() + sizeof(oopDesc)));
3398 __ call(RuntimeAddress(copyfunc_addr));
3399 #endif
3400
3401 #endif
3402
3403 #ifndef PRODUCT
3404 if (PrintC1Statistics) {
3405 Label failed;
3406 __ testl(rax, rax);
3407 __ jcc(Assembler::notZero, failed);
3408 __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_cnt));
3409 __ bind(failed);
3410 }
3411 #endif
3412
3413 __ testl(rax, rax);
3414 __ jcc(Assembler::zero, *stub->continuation());
3415
3416 #ifndef PRODUCT
3417 if (PrintC1Statistics) {
3418 __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_attempt_cnt));
3419 }
3420 #endif
3421
3422 __ mov(tmp, rax);
3423
3424 __ xorl(tmp, -1);
3425
3426 // Restore previously spilled arguments
3427 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3428 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3429 __ movptr (length, Address(rsp, 2*BytesPerWord));
3430 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3431 __ movptr (src, Address(rsp, 4*BytesPerWord));
3432
3433
3434 __ subl(length, tmp);
3435 __ addl(src_pos, tmp);
3436 __ addl(dst_pos, tmp);
3437 }
3438
3439 __ jmp(*stub->entry());
3440
3441 __ bind(cont);
3442 __ pop(dst);
3443 __ pop(src);
3444 }
3445 }
3446
3447 #ifdef ASSERT
3448 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
3449 // Sanity check the known type with the incoming class. For the
3450 // primitive case the types must match exactly with src.klass and
3451 // dst.klass each exactly matching the default type. For the
3452 // object array case, if no type check is needed then either the
3453 // dst type is exactly the expected type and the src type is a
3454 // subtype which we can't check or src is the same array as dst
3455 // but not necessarily exactly of type default_type.
3456 Label known_ok, halt;
3457 __ movoop(tmp, default_type->constant_encoding());
3458 #ifdef _LP64
3459 if (UseCompressedOops) {
3460 __ encode_heap_oop(tmp);
3461 }
3462 #endif
3463
3464 if (basic_type != T_OBJECT) {
3465
3466 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3467 else __ cmpptr(tmp, dst_klass_addr);
3468 __ jcc(Assembler::notEqual, halt);
3469 if (UseCompressedOops) __ cmpl(tmp, src_klass_addr);
3470 else __ cmpptr(tmp, src_klass_addr);
3471 __ jcc(Assembler::equal, known_ok);
3472 } else {
3473 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3474 else __ cmpptr(tmp, dst_klass_addr);
3475 __ jcc(Assembler::equal, known_ok);
3476 __ cmpptr(src, dst);
3477 __ jcc(Assembler::equal, known_ok);
3478 }
3479 __ bind(halt);
3480 __ stop("incorrect type information in arraycopy");
3481 __ bind(known_ok);
3482 }
3483 #endif
3484
3485 #ifndef PRODUCT
3486 if (PrintC1Statistics) {
3487 __ incrementl(ExternalAddress(Runtime1::arraycopy_count_address(basic_type)));
3488 }
3489 #endif
3490
3491 #ifdef _LP64
3492 assert_different_registers(c_rarg0, dst, dst_pos, length);
3493 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3494 assert_different_registers(c_rarg1, length);
3495 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3496 __ mov(c_rarg2, length);
3497
3498 #else
3499 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3500 store_parameter(tmp, 0);
3501 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3502 store_parameter(tmp, 1);
3503 store_parameter(length, 2);
3504 #endif // _LP64
3505
3506 bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
3507 bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
3508 const char *name;
3509 address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
3510 __ call_VM_leaf(entry, 0);
3511
3512 __ bind(*stub->continuation());
3513 }
3514
3515
3516 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
3517 Register obj = op->obj_opr()->as_register(); // may not be an oop
3518 Register hdr = op->hdr_opr()->as_register();
3519 Register lock = op->lock_opr()->as_register();
3520 if (!UseFastLocking) {
3521 __ jmp(*op->stub()->entry());
3522 } else if (op->code() == lir_lock) {
3523 Register scratch = noreg;
3524 if (UseBiasedLocking) {
3525 scratch = op->scratch_opr()->as_register();
3526 }
3527 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3528 // add debug info for NullPointerException only if one is possible
3529 int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
3530 if (op->info() != NULL) {
3531 add_debug_info_for_null_check(null_check_offset, op->info());
3532 }
3533 // done
3534 } else if (op->code() == lir_unlock) {
3535 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3536 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
3537 } else {
3538 Unimplemented();
3539 }
3540 __ bind(*op->stub()->continuation());
3541 }
3542
3543
3544 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
3545 ciMethod* method = op->profiled_method();
3546 int bci = op->profiled_bci();
3547
3548 // Update counter for all call types
3549 ciMethodData* md = method->method_data_or_null();
3550 assert(md != NULL, "Sanity");
3551 ciProfileData* data = md->bci_to_data(bci);
3552 assert(data->is_CounterData(), "need CounterData for calls");
3553 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
3554 Register mdo = op->mdo()->as_register();
3555 __ movoop(mdo, md->constant_encoding());
3556 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
3557 Bytecodes::Code bc = method->java_code_at_bci(bci);
3558 // Perform additional virtual call profiling for invokevirtual and
3559 // invokeinterface bytecodes
3560 if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3561 C1ProfileVirtualCalls) {
3562 assert(op->recv()->is_single_cpu(), "recv must be allocated");
3563 Register recv = op->recv()->as_register();
3564 assert_different_registers(mdo, recv);
3565 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3566 ciKlass* known_klass = op->known_holder();
3567 if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
3568 // We know the type that will be seen at this call site; we can
3569 // statically update the methodDataOop rather than needing to do
3570 // dynamic tests on the receiver type
3571
3572 // NOTE: we should probably put a lock around this search to
3573 // avoid collisions by concurrent compilations
3574 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3575 uint i;
3576 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3577 ciKlass* receiver = vc_data->receiver(i);
3578 if (known_klass->equals(receiver)) {
3579 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3580 __ addptr(data_addr, DataLayout::counter_increment);
3581 return;
3582 }
3583 }
3584
3585 // Receiver type not found in profile data; select an empty slot
3586
3587 // Note that this is less efficient than it should be because it
3588 // always does a write to the receiver part of the
3589 // VirtualCallData rather than just the first time
3590 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3591 ciKlass* receiver = vc_data->receiver(i);
3592 if (receiver == NULL) {
3593 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3594 __ movoop(recv_addr, known_klass->constant_encoding());
3595 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3596 __ addptr(data_addr, DataLayout::counter_increment);
3597 return;
3598 }
3599 }
3600 } else {
3601 __ load_klass(recv, recv);
3602 Label update_done;
3603 type_profile_helper(mdo, md, data, recv, &update_done);
3604 // Receiver did not match any saved receiver and there is no empty row for it.
3605 // Increment total counter to indicate polymorphic case.
3606 __ addptr(counter_addr, DataLayout::counter_increment);
3607
3608 __ bind(update_done);
3609 }
3610 } else {
3611 // Static call
3612 __ addptr(counter_addr, DataLayout::counter_increment);
3613 }
3614 }
3615
3616 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
3617 Unimplemented();
3618 }
3619
3620
3621 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
3622 __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
3623 }
3624
3625
3626 void LIR_Assembler::align_backward_branch_target() {
3627 __ align(BytesPerWord);
3628 }
3629
3630
3631 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3632 if (left->is_single_cpu()) {
3633 __ negl(left->as_register());
3634 move_regs(left->as_register(), dest->as_register());
3635
3636 } else if (left->is_double_cpu()) {
3637 Register lo = left->as_register_lo();
3638 #ifdef _LP64
3639 Register dst = dest->as_register_lo();
3640 __ movptr(dst, lo);
3641 __ negptr(dst);
3642 #else
3643 Register hi = left->as_register_hi();
3644 __ lneg(hi, lo);
3645 if (dest->as_register_lo() == hi) {
3646 assert(dest->as_register_hi() != lo, "destroying register");
3647 move_regs(hi, dest->as_register_hi());
3648 move_regs(lo, dest->as_register_lo());
3649 } else {
3650 move_regs(lo, dest->as_register_lo());
3651 move_regs(hi, dest->as_register_hi());
3652 }
3653 #endif // _LP64
3654
3655 } else if (dest->is_single_xmm()) {
3656 if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
3657 __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
3658 }
3659 __ xorps(dest->as_xmm_float_reg(),
3660 ExternalAddress((address)float_signflip_pool));
3661
3662 } else if (dest->is_double_xmm()) {
3663 if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
3664 __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
3665 }
3666 __ xorpd(dest->as_xmm_double_reg(),
3667 ExternalAddress((address)double_signflip_pool));
3668
3669 } else if (left->is_single_fpu() || left->is_double_fpu()) {
3670 assert(left->fpu() == 0, "arg must be on TOS");
3671 assert(dest->fpu() == 0, "dest must be TOS");
3672 __ fchs();
3673
3674 } else {
3675 ShouldNotReachHere();
3676 }
3677 }
3678
3679
3680 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
3681 assert(addr->is_address() && dest->is_register(), "check");
3682 Register reg;
3683 reg = dest->as_pointer_register();
3684 __ lea(reg, as_Address(addr->as_address_ptr()));
3685 }
3686
3687
3688
3689 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3690 assert(!tmp->is_valid(), "don't need temporary");
3691 __ call(RuntimeAddress(dest));
3692 if (info != NULL) {
3693 add_call_info_here(info);
3694 }
3695 }
3696
3697
3698 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3699 assert(type == T_LONG, "only for volatile long fields");
3700
3701 if (info != NULL) {
3702 add_debug_info_for_null_check_here(info);
3703 }
3704
3705 if (src->is_double_xmm()) {
3706 if (dest->is_double_cpu()) {
3707 #ifdef _LP64
3708 __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
3709 #else
3710 __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
3711 __ psrlq(src->as_xmm_double_reg(), 32);
3712 __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
3713 #endif // _LP64
3714 } else if (dest->is_double_stack()) {
3715 __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
3716 } else if (dest->is_address()) {
3717 __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
3718 } else {
3719 ShouldNotReachHere();
3720 }
3721
3722 } else if (dest->is_double_xmm()) {
3723 if (src->is_double_stack()) {
3724 __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
3725 } else if (src->is_address()) {
3726 __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
3727 } else {
3728 ShouldNotReachHere();
3729 }
3730
3731 } else if (src->is_double_fpu()) {
3732 assert(src->fpu_regnrLo() == 0, "must be TOS");
3733 if (dest->is_double_stack()) {
3734 __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
3735 } else if (dest->is_address()) {
3736 __ fistp_d(as_Address(dest->as_address_ptr()));
3737 } else {
3738 ShouldNotReachHere();
3739 }
3740
3741 } else if (dest->is_double_fpu()) {
3742 assert(dest->fpu_regnrLo() == 0, "must be TOS");
3743 if (src->is_double_stack()) {
3744 __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
3745 } else if (src->is_address()) {
3746 __ fild_d(as_Address(src->as_address_ptr()));
3747 } else {
3748 ShouldNotReachHere();
3749 }
3750 } else {
3751 ShouldNotReachHere();
3752 }
3753 }
3754
3755
3756 void LIR_Assembler::membar() {
3757 // QQQ sparc TSO uses this,
3758 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
3759 }
3760
3761 void LIR_Assembler::membar_acquire() {
3762 // No x86 machines currently require load fences
3763 // __ load_fence();
3764 }
3765
3766 void LIR_Assembler::membar_release() {
3767 // No x86 machines currently require store fences
3768 // __ store_fence();
3769 }
3770
3771 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3772 assert(result_reg->is_register(), "check");
3773 #ifdef _LP64
3774 // __ get_thread(result_reg->as_register_lo());
3775 __ mov(result_reg->as_register(), r15_thread);
3776 #else
3777 __ get_thread(result_reg->as_register());
3778 #endif // _LP64
3779 }
3780
3781
3782 void LIR_Assembler::peephole(LIR_List*) {
3783 // do nothing for now
3784 }
3785
3786
3787 #undef __