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