1 /*
2 * Copyright (c) 2000, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "c1/c1_Compilation.hpp"
28 #include "c1/c1_LIRAssembler.hpp"
29 #include "c1/c1_MacroAssembler.hpp"
30 #include "c1/c1_Runtime1.hpp"
31 #include "c1/c1_ValueStack.hpp"
32 #include "ci/ciArrayKlass.hpp"
33 #include "ci/ciInstance.hpp"
34 #include "gc_interface/collectedHeap.hpp"
35 #include "memory/barrierSet.hpp"
36 #include "memory/cardTableModRefBS.hpp"
37 #include "nativeInst_x86.hpp"
38 #include "oops/objArrayKlass.hpp"
39 #include "runtime/sharedRuntime.hpp"
40
41
42 // These masks are used to provide 128-bit aligned bitmasks to the XMM
43 // instructions, to allow sign-masking or sign-bit flipping. They allow
44 // fast versions of NegF/NegD and AbsF/AbsD.
45
46 // Note: 'double' and 'long long' have 32-bits alignment on x86.
47 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
48 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
49 // of 128-bits operands for SSE instructions.
50 jlong *operand = (jlong*)(((intptr_t)adr) & ((intptr_t)(~0xF)));
51 // Store the value to a 128-bits operand.
52 operand[0] = lo;
53 operand[1] = hi;
54 return operand;
55 }
56
57 // Buffer for 128-bits masks used by SSE instructions.
58 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
59
60 // Static initialization during VM startup.
61 static jlong *float_signmask_pool = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
62 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
63 static jlong *float_signflip_pool = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
64 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
65
66
67
68 NEEDS_CLEANUP // remove this definitions ?
69 const Register IC_Klass = rax; // where the IC klass is cached
70 const Register SYNC_header = rax; // synchronization header
71 const Register SHIFT_count = rcx; // where count for shift operations must be
72
73 #define __ _masm->
74
75
76 static void select_different_registers(Register preserve,
77 Register extra,
78 Register &tmp1,
79 Register &tmp2) {
80 if (tmp1 == preserve) {
81 assert_different_registers(tmp1, tmp2, extra);
82 tmp1 = extra;
83 } else if (tmp2 == preserve) {
84 assert_different_registers(tmp1, tmp2, extra);
85 tmp2 = extra;
86 }
87 assert_different_registers(preserve, tmp1, tmp2);
88 }
89
90
91
92 static void select_different_registers(Register preserve,
93 Register extra,
94 Register &tmp1,
95 Register &tmp2,
96 Register &tmp3) {
97 if (tmp1 == preserve) {
98 assert_different_registers(tmp1, tmp2, tmp3, extra);
99 tmp1 = extra;
100 } else if (tmp2 == preserve) {
101 assert_different_registers(tmp1, tmp2, tmp3, extra);
102 tmp2 = extra;
103 } else if (tmp3 == preserve) {
104 assert_different_registers(tmp1, tmp2, tmp3, extra);
105 tmp3 = extra;
106 }
107 assert_different_registers(preserve, tmp1, tmp2, tmp3);
108 }
109
110
111
112 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
113 if (opr->is_constant()) {
114 LIR_Const* constant = opr->as_constant_ptr();
115 switch (constant->type()) {
116 case T_INT: {
117 return true;
118 }
119
120 default:
121 return false;
122 }
123 }
124 return false;
125 }
126
127
128 LIR_Opr LIR_Assembler::receiverOpr() {
129 return FrameMap::receiver_opr;
130 }
131
132 LIR_Opr LIR_Assembler::osrBufferPointer() {
133 return FrameMap::as_pointer_opr(receiverOpr()->as_register());
134 }
135
136 //--------------fpu register translations-----------------------
137
138
139 address LIR_Assembler::float_constant(float f) {
140 address const_addr = __ float_constant(f);
141 if (const_addr == NULL) {
142 bailout("const section overflow");
143 return __ code()->consts()->start();
144 } else {
145 return const_addr;
146 }
147 }
148
149
150 address LIR_Assembler::double_constant(double d) {
151 address const_addr = __ double_constant(d);
152 if (const_addr == NULL) {
153 bailout("const section overflow");
154 return __ code()->consts()->start();
155 } else {
156 return const_addr;
157 }
158 }
159
160
161 void LIR_Assembler::set_24bit_FPU() {
162 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
163 }
164
165 void LIR_Assembler::reset_FPU() {
166 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
167 }
168
169 void LIR_Assembler::fpop() {
170 __ fpop();
171 }
172
173 void LIR_Assembler::fxch(int i) {
174 __ fxch(i);
175 }
176
177 void LIR_Assembler::fld(int i) {
178 __ fld_s(i);
179 }
180
181 void LIR_Assembler::ffree(int i) {
182 __ ffree(i);
183 }
184
185 void LIR_Assembler::breakpoint() {
186 __ int3();
187 }
188
189 void LIR_Assembler::push(LIR_Opr opr) {
190 if (opr->is_single_cpu()) {
191 __ push_reg(opr->as_register());
192 } else if (opr->is_double_cpu()) {
193 NOT_LP64(__ push_reg(opr->as_register_hi()));
194 __ push_reg(opr->as_register_lo());
195 } else if (opr->is_stack()) {
196 __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
197 } else if (opr->is_constant()) {
198 LIR_Const* const_opr = opr->as_constant_ptr();
199 if (const_opr->type() == T_OBJECT) {
200 __ push_oop(const_opr->as_jobject());
201 } else if (const_opr->type() == T_INT) {
202 __ push_jint(const_opr->as_jint());
203 } else {
204 ShouldNotReachHere();
205 }
206
207 } else {
208 ShouldNotReachHere();
209 }
210 }
211
212 void LIR_Assembler::pop(LIR_Opr opr) {
213 if (opr->is_single_cpu()) {
214 __ pop_reg(opr->as_register());
215 } else {
216 ShouldNotReachHere();
217 }
218 }
219
220 bool LIR_Assembler::is_literal_address(LIR_Address* addr) {
221 return addr->base()->is_illegal() && addr->index()->is_illegal();
222 }
223
224 //-------------------------------------------
225
226 Address LIR_Assembler::as_Address(LIR_Address* addr) {
227 return as_Address(addr, rscratch1);
228 }
229
230 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
231 if (addr->base()->is_illegal()) {
232 assert(addr->index()->is_illegal(), "must be illegal too");
233 AddressLiteral laddr((address)addr->disp(), relocInfo::none);
234 if (! __ reachable(laddr)) {
235 __ movptr(tmp, laddr.addr());
236 Address res(tmp, 0);
237 return res;
238 } else {
239 return __ as_Address(laddr);
240 }
241 }
242
243 Register base = addr->base()->as_pointer_register();
244
245 if (addr->index()->is_illegal()) {
246 return Address( base, addr->disp());
247 } else if (addr->index()->is_cpu_register()) {
248 Register index = addr->index()->as_pointer_register();
249 return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
250 } else if (addr->index()->is_constant()) {
251 intptr_t addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
252 assert(Assembler::is_simm32(addr_offset), "must be");
253
254 return Address(base, addr_offset);
255 } else {
256 Unimplemented();
257 return Address();
258 }
259 }
260
261
262 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
263 Address base = as_Address(addr);
264 return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
265 }
266
267
268 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
269 return as_Address(addr);
270 }
271
272
273 void LIR_Assembler::osr_entry() {
274 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
275 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
276 ValueStack* entry_state = osr_entry->state();
277 int number_of_locks = entry_state->locks_size();
278
279 // we jump here if osr happens with the interpreter
280 // state set up to continue at the beginning of the
281 // loop that triggered osr - in particular, we have
282 // the following registers setup:
283 //
284 // rcx: osr buffer
285 //
286
287 // build frame
288 ciMethod* m = compilation()->method();
289 __ build_frame(initial_frame_size_in_bytes());
290
291 // OSR buffer is
292 //
293 // locals[nlocals-1..0]
294 // monitors[0..number_of_locks]
295 //
296 // locals is a direct copy of the interpreter frame so in the osr buffer
297 // so first slot in the local array is the last local from the interpreter
298 // and last slot is local[0] (receiver) from the interpreter
299 //
300 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
301 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
302 // in the interpreter frame (the method lock if a sync method)
303
304 // Initialize monitors in the compiled activation.
305 // rcx: pointer to osr buffer
306 //
307 // All other registers are dead at this point and the locals will be
308 // copied into place by code emitted in the IR.
309
310 Register OSR_buf = osrBufferPointer()->as_pointer_register();
311 { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
312 int monitor_offset = BytesPerWord * method()->max_locals() +
313 (2 * BytesPerWord) * (number_of_locks - 1);
314 // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
315 // the OSR buffer using 2 word entries: first the lock and then
316 // the oop.
317 for (int i = 0; i < number_of_locks; i++) {
318 int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
319 #ifdef ASSERT
320 // verify the interpreter's monitor has a non-null object
321 {
322 Label L;
323 __ cmpptr(Address(OSR_buf, slot_offset + 1*BytesPerWord), (int32_t)NULL_WORD);
324 __ jcc(Assembler::notZero, L);
325 __ stop("locked object is NULL");
326 __ bind(L);
327 }
328 #endif
329 __ movptr(rbx, Address(OSR_buf, slot_offset + 0));
330 __ movptr(frame_map()->address_for_monitor_lock(i), rbx);
331 __ movptr(rbx, Address(OSR_buf, slot_offset + 1*BytesPerWord));
332 __ movptr(frame_map()->address_for_monitor_object(i), rbx);
333 }
334 }
335 }
336
337
338 // inline cache check; done before the frame is built.
339 int LIR_Assembler::check_icache() {
340 Register receiver = FrameMap::receiver_opr->as_register();
341 Register ic_klass = IC_Klass;
342 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
343 const bool do_post_padding = VerifyOops || UseCompressedOops;
344 if (!do_post_padding) {
345 // insert some nops so that the verified entry point is aligned on CodeEntryAlignment
346 while ((__ offset() + ic_cmp_size) % CodeEntryAlignment != 0) {
347 __ nop();
348 }
349 }
350 int offset = __ offset();
351 __ inline_cache_check(receiver, IC_Klass);
352 assert(__ offset() % CodeEntryAlignment == 0 || do_post_padding, "alignment must be correct");
353 if (do_post_padding) {
354 // force alignment after the cache check.
355 // It's been verified to be aligned if !VerifyOops
356 __ align(CodeEntryAlignment);
357 }
358 return offset;
359 }
360
361
362 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
363 jobject o = NULL;
364 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
365 __ movoop(reg, o);
366 patching_epilog(patch, lir_patch_normal, reg, info);
367 }
368
369
370 // This specifies the rsp decrement needed to build the frame
371 int LIR_Assembler::initial_frame_size_in_bytes() {
372 // if rounding, must let FrameMap know!
373
374 // The frame_map records size in slots (32bit word)
375
376 // subtract two words to account for return address and link
377 return (frame_map()->framesize() - (2*VMRegImpl::slots_per_word)) * VMRegImpl::stack_slot_size;
378 }
379
380
381 int LIR_Assembler::emit_exception_handler() {
382 // if the last instruction is a call (typically to do a throw which
383 // is coming at the end after block reordering) the return address
384 // must still point into the code area in order to avoid assertion
385 // failures when searching for the corresponding bci => add a nop
386 // (was bug 5/14/1999 - gri)
387 __ nop();
388
389 // generate code for exception handler
390 address handler_base = __ start_a_stub(exception_handler_size);
391 if (handler_base == NULL) {
392 // not enough space left for the handler
393 bailout("exception handler overflow");
394 return -1;
395 }
396
397 int offset = code_offset();
398
399 // the exception oop and pc are in rax, and rdx
400 // no other registers need to be preserved, so invalidate them
401 __ invalidate_registers(false, true, true, false, true, true);
402
403 // check that there is really an exception
404 __ verify_not_null_oop(rax);
405
406 // search an exception handler (rax: exception oop, rdx: throwing pc)
407 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id)));
408 __ should_not_reach_here();
409 guarantee(code_offset() - offset <= exception_handler_size, "overflow");
410 __ end_a_stub();
411
412 return offset;
413 }
414
415
416 // Emit the code to remove the frame from the stack in the exception
417 // unwind path.
418 int LIR_Assembler::emit_unwind_handler() {
419 #ifndef PRODUCT
420 if (CommentedAssembly) {
421 _masm->block_comment("Unwind handler");
422 }
423 #endif
424
425 int offset = code_offset();
426
427 // Fetch the exception from TLS and clear out exception related thread state
428 __ get_thread(rsi);
429 __ movptr(rax, Address(rsi, JavaThread::exception_oop_offset()));
430 __ movptr(Address(rsi, JavaThread::exception_oop_offset()), (intptr_t)NULL_WORD);
431 __ movptr(Address(rsi, JavaThread::exception_pc_offset()), (intptr_t)NULL_WORD);
432
433 __ bind(_unwind_handler_entry);
434 __ verify_not_null_oop(rax);
435 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
436 __ mov(rsi, rax); // Preserve the exception
437 }
438
439 // Preform needed unlocking
440 MonitorExitStub* stub = NULL;
441 if (method()->is_synchronized()) {
442 monitor_address(0, FrameMap::rax_opr);
443 stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
444 __ unlock_object(rdi, rbx, rax, *stub->entry());
445 __ bind(*stub->continuation());
446 }
447
448 if (compilation()->env()->dtrace_method_probes()) {
449 __ get_thread(rax);
450 __ movptr(Address(rsp, 0), rax);
451 __ movoop(Address(rsp, sizeof(void*)), method()->constant_encoding());
452 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
453 }
454
455 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
456 __ mov(rax, rsi); // Restore the exception
457 }
458
459 // remove the activation and dispatch to the unwind handler
460 __ remove_frame(initial_frame_size_in_bytes());
461 __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
462
463 // Emit the slow path assembly
464 if (stub != NULL) {
465 stub->emit_code(this);
466 }
467
468 return offset;
469 }
470
471
472 int LIR_Assembler::emit_deopt_handler() {
473 // if the last instruction is a call (typically to do a throw which
474 // is coming at the end after block reordering) the return address
475 // must still point into the code area in order to avoid assertion
476 // failures when searching for the corresponding bci => add a nop
477 // (was bug 5/14/1999 - gri)
478 __ nop();
479
480 // generate code for exception handler
481 address handler_base = __ start_a_stub(deopt_handler_size);
482 if (handler_base == NULL) {
483 // not enough space left for the handler
484 bailout("deopt handler overflow");
485 return -1;
486 }
487
488 int offset = code_offset();
489 InternalAddress here(__ pc());
490
491 __ pushptr(here.addr());
492 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
493 guarantee(code_offset() - offset <= deopt_handler_size, "overflow");
494 __ end_a_stub();
495
496 return offset;
497 }
498
499
500 // This is the fast version of java.lang.String.compare; it has not
501 // OSR-entry and therefore, we generate a slow version for OSR's
502 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
503 __ movptr (rbx, rcx); // receiver is in rcx
504 __ movptr (rax, arg1->as_register());
505
506 // Get addresses of first characters from both Strings
507 __ load_heap_oop(rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
508 if (java_lang_String::has_offset_field()) {
509 __ movptr (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
510 __ movl (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
511 __ lea (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
512 } else {
513 __ movl (rax, Address(rsi, arrayOopDesc::length_offset_in_bytes()));
514 __ lea (rsi, Address(rsi, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
515 }
516
517 // rbx, may be NULL
518 add_debug_info_for_null_check_here(info);
519 __ load_heap_oop(rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
520 if (java_lang_String::has_offset_field()) {
521 __ movptr (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
522 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
523 __ lea (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
524 } else {
525 __ movl (rbx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
526 __ lea (rdi, Address(rdi, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
527 }
528
529 // compute minimum length (in rax) and difference of lengths (on top of stack)
530 __ mov (rcx, rbx);
531 __ subptr(rbx, rax); // subtract lengths
532 __ push (rbx); // result
533 __ cmov (Assembler::lessEqual, rax, rcx);
534
535 // is minimum length 0?
536 Label noLoop, haveResult;
537 __ testptr (rax, rax);
538 __ jcc (Assembler::zero, noLoop);
539
540 // compare first characters
541 __ load_unsigned_short(rcx, Address(rdi, 0));
542 __ load_unsigned_short(rbx, Address(rsi, 0));
543 __ subl(rcx, rbx);
544 __ jcc(Assembler::notZero, haveResult);
545 // starting loop
546 __ decrement(rax); // we already tested index: skip one
547 __ jcc(Assembler::zero, noLoop);
548
549 // set rsi.edi to the end of the arrays (arrays have same length)
550 // negate the index
551
552 __ lea(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
553 __ lea(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
554 __ negptr(rax);
555
556 // compare the strings in a loop
557
558 Label loop;
559 __ align(wordSize);
560 __ bind(loop);
561 __ load_unsigned_short(rcx, Address(rdi, rax, Address::times_2, 0));
562 __ load_unsigned_short(rbx, Address(rsi, rax, Address::times_2, 0));
563 __ subl(rcx, rbx);
564 __ jcc(Assembler::notZero, haveResult);
565 __ increment(rax);
566 __ jcc(Assembler::notZero, loop);
567
568 // strings are equal up to min length
569
570 __ bind(noLoop);
571 __ pop(rax);
572 return_op(LIR_OprFact::illegalOpr);
573
574 __ bind(haveResult);
575 // leave instruction is going to discard the TOS value
576 __ mov (rax, rcx); // result of call is in rax,
577 }
578
579
580 void LIR_Assembler::return_op(LIR_Opr result) {
581 assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
582 if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
583 assert(result->fpu() == 0, "result must already be on TOS");
584 }
585
586 // Pop the stack before the safepoint code
587 __ remove_frame(initial_frame_size_in_bytes());
588
589 bool result_is_oop = result->is_valid() ? result->is_oop() : false;
590
591 // Note: we do not need to round double result; float result has the right precision
592 // the poll sets the condition code, but no data registers
593 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
594 relocInfo::poll_return_type);
595
596 if (Assembler::is_polling_page_far()) {
597 __ lea(rscratch1, polling_page);
598 __ relocate(relocInfo::poll_return_type);
599 __ testl(rax, Address(rscratch1, 0));
600 } else {
601 __ testl(rax, polling_page);
602 }
603 __ ret(0);
604 }
605
606
607 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
608 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
609 relocInfo::poll_type);
610 guarantee(info != NULL, "Shouldn't be NULL");
611 int offset = __ offset();
612 if (Assembler::is_polling_page_far()) {
613 __ lea(rscratch1, polling_page);
614 offset = __ offset();
615 add_debug_info_for_branch(info);
616 __ testl(rax, Address(rscratch1, 0));
617 } else {
618 add_debug_info_for_branch(info);
619 __ testl(rax, polling_page);
620 }
621 return offset;
622 }
623
624
625 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
626 if (from_reg != to_reg) __ mov(to_reg, from_reg);
627 }
628
629 void LIR_Assembler::swap_reg(Register a, Register b) {
630 __ xchgptr(a, b);
631 }
632
633
634 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
635 assert(src->is_constant(), "should not call otherwise");
636 assert(dest->is_register(), "should not call otherwise");
637 LIR_Const* c = src->as_constant_ptr();
638
639 switch (c->type()) {
640 case T_INT: {
641 assert(patch_code == lir_patch_none, "no patching handled here");
642 __ movl(dest->as_register(), c->as_jint());
643 break;
644 }
645
646 case T_ADDRESS: {
647 assert(patch_code == lir_patch_none, "no patching handled here");
648 __ movptr(dest->as_register(), c->as_jint());
649 break;
650 }
651
652 case T_LONG: {
653 assert(patch_code == lir_patch_none, "no patching handled here");
654 #ifdef _LP64
655 __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
656 #else
657 __ movptr(dest->as_register_lo(), c->as_jint_lo());
658 __ movptr(dest->as_register_hi(), c->as_jint_hi());
659 #endif // _LP64
660 break;
661 }
662
663 case T_OBJECT: {
664 if (patch_code != lir_patch_none) {
665 jobject2reg_with_patching(dest->as_register(), info);
666 } else {
667 __ movoop(dest->as_register(), c->as_jobject());
668 }
669 break;
670 }
671
672 case T_FLOAT: {
673 if (dest->is_single_xmm()) {
674 if (c->is_zero_float()) {
675 __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
676 } else {
677 __ movflt(dest->as_xmm_float_reg(),
678 InternalAddress(float_constant(c->as_jfloat())));
679 }
680 } else {
681 assert(dest->is_single_fpu(), "must be");
682 assert(dest->fpu_regnr() == 0, "dest must be TOS");
683 if (c->is_zero_float()) {
684 __ fldz();
685 } else if (c->is_one_float()) {
686 __ fld1();
687 } else {
688 __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
689 }
690 }
691 break;
692 }
693
694 case T_DOUBLE: {
695 if (dest->is_double_xmm()) {
696 if (c->is_zero_double()) {
697 __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
698 } else {
699 __ movdbl(dest->as_xmm_double_reg(),
700 InternalAddress(double_constant(c->as_jdouble())));
701 }
702 } else {
703 assert(dest->is_double_fpu(), "must be");
704 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
705 if (c->is_zero_double()) {
706 __ fldz();
707 } else if (c->is_one_double()) {
708 __ fld1();
709 } else {
710 __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
711 }
712 }
713 break;
714 }
715
716 default:
717 ShouldNotReachHere();
718 }
719 }
720
721 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
722 assert(src->is_constant(), "should not call otherwise");
723 assert(dest->is_stack(), "should not call otherwise");
724 LIR_Const* c = src->as_constant_ptr();
725
726 switch (c->type()) {
727 case T_INT: // fall through
728 case T_FLOAT:
729 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
730 break;
731
732 case T_ADDRESS:
733 __ movptr(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
734 break;
735
736 case T_OBJECT:
737 __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
738 break;
739
740 case T_LONG: // fall through
741 case T_DOUBLE:
742 #ifdef _LP64
743 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
744 lo_word_offset_in_bytes), (intptr_t)c->as_jlong_bits());
745 #else
746 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
747 lo_word_offset_in_bytes), c->as_jint_lo_bits());
748 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
749 hi_word_offset_in_bytes), c->as_jint_hi_bits());
750 #endif // _LP64
751 break;
752
753 default:
754 ShouldNotReachHere();
755 }
756 }
757
758 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
759 assert(src->is_constant(), "should not call otherwise");
760 assert(dest->is_address(), "should not call otherwise");
761 LIR_Const* c = src->as_constant_ptr();
762 LIR_Address* addr = dest->as_address_ptr();
763
764 int null_check_here = code_offset();
765 switch (type) {
766 case T_INT: // fall through
767 case T_FLOAT:
768 __ movl(as_Address(addr), c->as_jint_bits());
769 break;
770
771 case T_ADDRESS:
772 __ movptr(as_Address(addr), c->as_jint_bits());
773 break;
774
775 case T_OBJECT: // fall through
776 case T_ARRAY:
777 if (c->as_jobject() == NULL) {
778 if (UseCompressedOops && !wide) {
779 __ movl(as_Address(addr), (int32_t)NULL_WORD);
780 } else {
781 __ movptr(as_Address(addr), NULL_WORD);
782 }
783 } else {
784 if (is_literal_address(addr)) {
785 ShouldNotReachHere();
786 __ movoop(as_Address(addr, noreg), c->as_jobject());
787 } else {
788 #ifdef _LP64
789 __ movoop(rscratch1, c->as_jobject());
790 if (UseCompressedOops && !wide) {
791 __ encode_heap_oop(rscratch1);
792 null_check_here = code_offset();
793 __ movl(as_Address_lo(addr), rscratch1);
794 } else {
795 null_check_here = code_offset();
796 __ movptr(as_Address_lo(addr), rscratch1);
797 }
798 #else
799 __ movoop(as_Address(addr), c->as_jobject());
800 #endif
801 }
802 }
803 break;
804
805 case T_LONG: // fall through
806 case T_DOUBLE:
807 #ifdef _LP64
808 if (is_literal_address(addr)) {
809 ShouldNotReachHere();
810 __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
811 } else {
812 __ movptr(r10, (intptr_t)c->as_jlong_bits());
813 null_check_here = code_offset();
814 __ movptr(as_Address_lo(addr), r10);
815 }
816 #else
817 // Always reachable in 32bit so this doesn't produce useless move literal
818 __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
819 __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
820 #endif // _LP64
821 break;
822
823 case T_BOOLEAN: // fall through
824 case T_BYTE:
825 __ movb(as_Address(addr), c->as_jint() & 0xFF);
826 break;
827
828 case T_CHAR: // fall through
829 case T_SHORT:
830 __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
831 break;
832
833 default:
834 ShouldNotReachHere();
835 };
836
837 if (info != NULL) {
838 add_debug_info_for_null_check(null_check_here, info);
839 }
840 }
841
842
843 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
844 assert(src->is_register(), "should not call otherwise");
845 assert(dest->is_register(), "should not call otherwise");
846
847 // move between cpu-registers
848 if (dest->is_single_cpu()) {
849 #ifdef _LP64
850 if (src->type() == T_LONG) {
851 // Can do LONG -> OBJECT
852 move_regs(src->as_register_lo(), dest->as_register());
853 return;
854 }
855 #endif
856 assert(src->is_single_cpu(), "must match");
857 if (src->type() == T_OBJECT) {
858 __ verify_oop(src->as_register());
859 }
860 move_regs(src->as_register(), dest->as_register());
861
862 } else if (dest->is_double_cpu()) {
863 #ifdef _LP64
864 if (src->type() == T_OBJECT || src->type() == T_ARRAY) {
865 // Surprising to me but we can see move of a long to t_object
866 __ verify_oop(src->as_register());
867 move_regs(src->as_register(), dest->as_register_lo());
868 return;
869 }
870 #endif
871 assert(src->is_double_cpu(), "must match");
872 Register f_lo = src->as_register_lo();
873 Register f_hi = src->as_register_hi();
874 Register t_lo = dest->as_register_lo();
875 Register t_hi = dest->as_register_hi();
876 #ifdef _LP64
877 assert(f_hi == f_lo, "must be same");
878 assert(t_hi == t_lo, "must be same");
879 move_regs(f_lo, t_lo);
880 #else
881 assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
882
883
884 if (f_lo == t_hi && f_hi == t_lo) {
885 swap_reg(f_lo, f_hi);
886 } else if (f_hi == t_lo) {
887 assert(f_lo != t_hi, "overwriting register");
888 move_regs(f_hi, t_hi);
889 move_regs(f_lo, t_lo);
890 } else {
891 assert(f_hi != t_lo, "overwriting register");
892 move_regs(f_lo, t_lo);
893 move_regs(f_hi, t_hi);
894 }
895 #endif // LP64
896
897 // special moves from fpu-register to xmm-register
898 // necessary for method results
899 } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
900 __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
901 __ fld_s(Address(rsp, 0));
902 } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
903 __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
904 __ fld_d(Address(rsp, 0));
905 } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
906 __ fstp_s(Address(rsp, 0));
907 __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
908 } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
909 __ fstp_d(Address(rsp, 0));
910 __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
911
912 // move between xmm-registers
913 } else if (dest->is_single_xmm()) {
914 assert(src->is_single_xmm(), "must match");
915 __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
916 } else if (dest->is_double_xmm()) {
917 assert(src->is_double_xmm(), "must match");
918 __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
919
920 // move between fpu-registers (no instruction necessary because of fpu-stack)
921 } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
922 assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
923 assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
924 } else {
925 ShouldNotReachHere();
926 }
927 }
928
929 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
930 assert(src->is_register(), "should not call otherwise");
931 assert(dest->is_stack(), "should not call otherwise");
932
933 if (src->is_single_cpu()) {
934 Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
935 if (type == T_OBJECT || type == T_ARRAY) {
936 __ verify_oop(src->as_register());
937 __ movptr (dst, src->as_register());
938 } else {
939 __ movl (dst, src->as_register());
940 }
941
942 } else if (src->is_double_cpu()) {
943 Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
944 Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
945 __ movptr (dstLO, src->as_register_lo());
946 NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
947
948 } else if (src->is_single_xmm()) {
949 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
950 __ movflt(dst_addr, src->as_xmm_float_reg());
951
952 } else if (src->is_double_xmm()) {
953 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
954 __ movdbl(dst_addr, src->as_xmm_double_reg());
955
956 } else if (src->is_single_fpu()) {
957 assert(src->fpu_regnr() == 0, "argument must be on TOS");
958 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
959 if (pop_fpu_stack) __ fstp_s (dst_addr);
960 else __ fst_s (dst_addr);
961
962 } else if (src->is_double_fpu()) {
963 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
964 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
965 if (pop_fpu_stack) __ fstp_d (dst_addr);
966 else __ fst_d (dst_addr);
967
968 } else {
969 ShouldNotReachHere();
970 }
971 }
972
973
974 void LIR_Assembler::reg2mem(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack, bool wide, bool /* unaligned */) {
975 LIR_Address* to_addr = dest->as_address_ptr();
976 PatchingStub* patch = NULL;
977 Register compressed_src = rscratch1;
978
979 if (type == T_ARRAY || type == T_OBJECT) {
980 __ verify_oop(src->as_register());
981 #ifdef _LP64
982 if (UseCompressedOops && !wide) {
983 __ movptr(compressed_src, src->as_register());
984 __ encode_heap_oop(compressed_src);
985 }
986 #endif
987 }
988
989 if (patch_code != lir_patch_none) {
990 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
991 Address toa = as_Address(to_addr);
992 assert(toa.disp() != 0, "must have");
993 }
994
995 int null_check_here = code_offset();
996 switch (type) {
997 case T_FLOAT: {
998 if (src->is_single_xmm()) {
999 __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
1000 } else {
1001 assert(src->is_single_fpu(), "must be");
1002 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1003 if (pop_fpu_stack) __ fstp_s(as_Address(to_addr));
1004 else __ fst_s (as_Address(to_addr));
1005 }
1006 break;
1007 }
1008
1009 case T_DOUBLE: {
1010 if (src->is_double_xmm()) {
1011 __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
1012 } else {
1013 assert(src->is_double_fpu(), "must be");
1014 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1015 if (pop_fpu_stack) __ fstp_d(as_Address(to_addr));
1016 else __ fst_d (as_Address(to_addr));
1017 }
1018 break;
1019 }
1020
1021 case T_ARRAY: // fall through
1022 case T_OBJECT: // fall through
1023 if (UseCompressedOops && !wide) {
1024 __ movl(as_Address(to_addr), compressed_src);
1025 } else {
1026 __ movptr(as_Address(to_addr), src->as_register());
1027 }
1028 break;
1029 case T_ADDRESS:
1030 __ movptr(as_Address(to_addr), src->as_register());
1031 break;
1032 case T_INT:
1033 __ movl(as_Address(to_addr), src->as_register());
1034 break;
1035
1036 case T_LONG: {
1037 Register from_lo = src->as_register_lo();
1038 Register from_hi = src->as_register_hi();
1039 #ifdef _LP64
1040 __ movptr(as_Address_lo(to_addr), from_lo);
1041 #else
1042 Register base = to_addr->base()->as_register();
1043 Register index = noreg;
1044 if (to_addr->index()->is_register()) {
1045 index = to_addr->index()->as_register();
1046 }
1047 if (base == from_lo || index == from_lo) {
1048 assert(base != from_hi, "can't be");
1049 assert(index == noreg || (index != base && index != from_hi), "can't handle this");
1050 __ movl(as_Address_hi(to_addr), from_hi);
1051 if (patch != NULL) {
1052 patching_epilog(patch, lir_patch_high, base, info);
1053 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1054 patch_code = lir_patch_low;
1055 }
1056 __ movl(as_Address_lo(to_addr), from_lo);
1057 } else {
1058 assert(index == noreg || (index != base && index != from_lo), "can't handle this");
1059 __ movl(as_Address_lo(to_addr), from_lo);
1060 if (patch != NULL) {
1061 patching_epilog(patch, lir_patch_low, base, info);
1062 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1063 patch_code = lir_patch_high;
1064 }
1065 __ movl(as_Address_hi(to_addr), from_hi);
1066 }
1067 #endif // _LP64
1068 break;
1069 }
1070
1071 case T_BYTE: // fall through
1072 case T_BOOLEAN: {
1073 Register src_reg = src->as_register();
1074 Address dst_addr = as_Address(to_addr);
1075 assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
1076 __ movb(dst_addr, src_reg);
1077 break;
1078 }
1079
1080 case T_CHAR: // fall through
1081 case T_SHORT:
1082 __ movw(as_Address(to_addr), src->as_register());
1083 break;
1084
1085 default:
1086 ShouldNotReachHere();
1087 }
1088 if (info != NULL) {
1089 add_debug_info_for_null_check(null_check_here, info);
1090 }
1091
1092 if (patch_code != lir_patch_none) {
1093 patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
1094 }
1095 }
1096
1097
1098 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1099 assert(src->is_stack(), "should not call otherwise");
1100 assert(dest->is_register(), "should not call otherwise");
1101
1102 if (dest->is_single_cpu()) {
1103 if (type == T_ARRAY || type == T_OBJECT) {
1104 __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1105 __ verify_oop(dest->as_register());
1106 } else {
1107 __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1108 }
1109
1110 } else if (dest->is_double_cpu()) {
1111 Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
1112 Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
1113 __ movptr(dest->as_register_lo(), src_addr_LO);
1114 NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
1115
1116 } else if (dest->is_single_xmm()) {
1117 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1118 __ movflt(dest->as_xmm_float_reg(), src_addr);
1119
1120 } else if (dest->is_double_xmm()) {
1121 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1122 __ movdbl(dest->as_xmm_double_reg(), src_addr);
1123
1124 } else if (dest->is_single_fpu()) {
1125 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1126 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1127 __ fld_s(src_addr);
1128
1129 } else if (dest->is_double_fpu()) {
1130 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1131 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1132 __ fld_d(src_addr);
1133
1134 } else {
1135 ShouldNotReachHere();
1136 }
1137 }
1138
1139
1140 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1141 if (src->is_single_stack()) {
1142 if (type == T_OBJECT || type == T_ARRAY) {
1143 __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
1144 __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
1145 } else {
1146 #ifndef _LP64
1147 __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
1148 __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
1149 #else
1150 //no pushl on 64bits
1151 __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix()));
1152 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1);
1153 #endif
1154 }
1155
1156 } else if (src->is_double_stack()) {
1157 #ifdef _LP64
1158 __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
1159 __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
1160 #else
1161 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
1162 // push and pop the part at src + wordSize, adding wordSize for the previous push
1163 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 2 * wordSize));
1164 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 2 * wordSize));
1165 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
1166 #endif // _LP64
1167
1168 } else {
1169 ShouldNotReachHere();
1170 }
1171 }
1172
1173
1174 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool /* unaligned */) {
1175 assert(src->is_address(), "should not call otherwise");
1176 assert(dest->is_register(), "should not call otherwise");
1177
1178 LIR_Address* addr = src->as_address_ptr();
1179 Address from_addr = as_Address(addr);
1180
1181 switch (type) {
1182 case T_BOOLEAN: // fall through
1183 case T_BYTE: // fall through
1184 case T_CHAR: // fall through
1185 case T_SHORT:
1186 if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
1187 // on pre P6 processors we may get partial register stalls
1188 // so blow away the value of to_rinfo before loading a
1189 // partial word into it. Do it here so that it precedes
1190 // the potential patch point below.
1191 __ xorptr(dest->as_register(), dest->as_register());
1192 }
1193 break;
1194 }
1195
1196 PatchingStub* patch = NULL;
1197 if (patch_code != lir_patch_none) {
1198 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1199 assert(from_addr.disp() != 0, "must have");
1200 }
1201 if (info != NULL) {
1202 add_debug_info_for_null_check_here(info);
1203 }
1204
1205 switch (type) {
1206 case T_FLOAT: {
1207 if (dest->is_single_xmm()) {
1208 __ movflt(dest->as_xmm_float_reg(), from_addr);
1209 } else {
1210 assert(dest->is_single_fpu(), "must be");
1211 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1212 __ fld_s(from_addr);
1213 }
1214 break;
1215 }
1216
1217 case T_DOUBLE: {
1218 if (dest->is_double_xmm()) {
1219 __ movdbl(dest->as_xmm_double_reg(), from_addr);
1220 } else {
1221 assert(dest->is_double_fpu(), "must be");
1222 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1223 __ fld_d(from_addr);
1224 }
1225 break;
1226 }
1227
1228 case T_OBJECT: // fall through
1229 case T_ARRAY: // fall through
1230 if (UseCompressedOops && !wide) {
1231 __ movl(dest->as_register(), from_addr);
1232 } else {
1233 __ movptr(dest->as_register(), from_addr);
1234 }
1235 break;
1236
1237 case T_ADDRESS:
1238 __ movptr(dest->as_register(), from_addr);
1239 break;
1240 case T_INT:
1241 __ movl(dest->as_register(), from_addr);
1242 break;
1243
1244 case T_LONG: {
1245 Register to_lo = dest->as_register_lo();
1246 Register to_hi = dest->as_register_hi();
1247 #ifdef _LP64
1248 __ movptr(to_lo, as_Address_lo(addr));
1249 #else
1250 Register base = addr->base()->as_register();
1251 Register index = noreg;
1252 if (addr->index()->is_register()) {
1253 index = addr->index()->as_register();
1254 }
1255 if ((base == to_lo && index == to_hi) ||
1256 (base == to_hi && index == to_lo)) {
1257 // addresses with 2 registers are only formed as a result of
1258 // array access so this code will never have to deal with
1259 // patches or null checks.
1260 assert(info == NULL && patch == NULL, "must be");
1261 __ lea(to_hi, as_Address(addr));
1262 __ movl(to_lo, Address(to_hi, 0));
1263 __ movl(to_hi, Address(to_hi, BytesPerWord));
1264 } else if (base == to_lo || index == to_lo) {
1265 assert(base != to_hi, "can't be");
1266 assert(index == noreg || (index != base && index != to_hi), "can't handle this");
1267 __ movl(to_hi, as_Address_hi(addr));
1268 if (patch != NULL) {
1269 patching_epilog(patch, lir_patch_high, base, info);
1270 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1271 patch_code = lir_patch_low;
1272 }
1273 __ movl(to_lo, as_Address_lo(addr));
1274 } else {
1275 assert(index == noreg || (index != base && index != to_lo), "can't handle this");
1276 __ movl(to_lo, as_Address_lo(addr));
1277 if (patch != NULL) {
1278 patching_epilog(patch, lir_patch_low, base, info);
1279 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1280 patch_code = lir_patch_high;
1281 }
1282 __ movl(to_hi, as_Address_hi(addr));
1283 }
1284 #endif // _LP64
1285 break;
1286 }
1287
1288 case T_BOOLEAN: // fall through
1289 case T_BYTE: {
1290 Register dest_reg = dest->as_register();
1291 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1292 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1293 __ movsbl(dest_reg, from_addr);
1294 } else {
1295 __ movb(dest_reg, from_addr);
1296 __ shll(dest_reg, 24);
1297 __ sarl(dest_reg, 24);
1298 }
1299 break;
1300 }
1301
1302 case T_CHAR: {
1303 Register dest_reg = dest->as_register();
1304 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1305 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1306 __ movzwl(dest_reg, from_addr);
1307 } else {
1308 __ movw(dest_reg, from_addr);
1309 }
1310 break;
1311 }
1312
1313 case T_SHORT: {
1314 Register dest_reg = dest->as_register();
1315 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1316 __ movswl(dest_reg, from_addr);
1317 } else {
1318 __ movw(dest_reg, from_addr);
1319 __ shll(dest_reg, 16);
1320 __ sarl(dest_reg, 16);
1321 }
1322 break;
1323 }
1324
1325 default:
1326 ShouldNotReachHere();
1327 }
1328
1329 if (patch != NULL) {
1330 patching_epilog(patch, patch_code, addr->base()->as_register(), info);
1331 }
1332
1333 if (type == T_ARRAY || type == T_OBJECT) {
1334 #ifdef _LP64
1335 if (UseCompressedOops && !wide) {
1336 __ decode_heap_oop(dest->as_register());
1337 }
1338 #endif
1339 __ verify_oop(dest->as_register());
1340 }
1341 }
1342
1343
1344 void LIR_Assembler::prefetchr(LIR_Opr src) {
1345 LIR_Address* addr = src->as_address_ptr();
1346 Address from_addr = as_Address(addr);
1347
1348 if (VM_Version::supports_sse()) {
1349 switch (ReadPrefetchInstr) {
1350 case 0:
1351 __ prefetchnta(from_addr); break;
1352 case 1:
1353 __ prefetcht0(from_addr); break;
1354 case 2:
1355 __ prefetcht2(from_addr); break;
1356 default:
1357 ShouldNotReachHere(); break;
1358 }
1359 } else if (VM_Version::supports_3dnow_prefetch()) {
1360 __ prefetchr(from_addr);
1361 }
1362 }
1363
1364
1365 void LIR_Assembler::prefetchw(LIR_Opr src) {
1366 LIR_Address* addr = src->as_address_ptr();
1367 Address from_addr = as_Address(addr);
1368
1369 if (VM_Version::supports_sse()) {
1370 switch (AllocatePrefetchInstr) {
1371 case 0:
1372 __ prefetchnta(from_addr); break;
1373 case 1:
1374 __ prefetcht0(from_addr); break;
1375 case 2:
1376 __ prefetcht2(from_addr); break;
1377 case 3:
1378 __ prefetchw(from_addr); break;
1379 default:
1380 ShouldNotReachHere(); break;
1381 }
1382 } else if (VM_Version::supports_3dnow_prefetch()) {
1383 __ prefetchw(from_addr);
1384 }
1385 }
1386
1387
1388 NEEDS_CLEANUP; // This could be static?
1389 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
1390 int elem_size = type2aelembytes(type);
1391 switch (elem_size) {
1392 case 1: return Address::times_1;
1393 case 2: return Address::times_2;
1394 case 4: return Address::times_4;
1395 case 8: return Address::times_8;
1396 }
1397 ShouldNotReachHere();
1398 return Address::no_scale;
1399 }
1400
1401
1402 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1403 switch (op->code()) {
1404 case lir_idiv:
1405 case lir_irem:
1406 arithmetic_idiv(op->code(),
1407 op->in_opr1(),
1408 op->in_opr2(),
1409 op->in_opr3(),
1410 op->result_opr(),
1411 op->info());
1412 break;
1413 default: ShouldNotReachHere(); break;
1414 }
1415 }
1416
1417 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1418 #ifdef ASSERT
1419 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
1420 if (op->block() != NULL) _branch_target_blocks.append(op->block());
1421 if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
1422 #endif
1423
1424 if (op->cond() == lir_cond_always) {
1425 if (op->info() != NULL) add_debug_info_for_branch(op->info());
1426 __ jmp (*(op->label()));
1427 } else {
1428 Assembler::Condition acond = Assembler::zero;
1429 if (op->code() == lir_cond_float_branch) {
1430 assert(op->ublock() != NULL, "must have unordered successor");
1431 __ jcc(Assembler::parity, *(op->ublock()->label()));
1432 switch(op->cond()) {
1433 case lir_cond_equal: acond = Assembler::equal; break;
1434 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1435 case lir_cond_less: acond = Assembler::below; break;
1436 case lir_cond_lessEqual: acond = Assembler::belowEqual; break;
1437 case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
1438 case lir_cond_greater: acond = Assembler::above; break;
1439 default: ShouldNotReachHere();
1440 }
1441 } else {
1442 switch (op->cond()) {
1443 case lir_cond_equal: acond = Assembler::equal; break;
1444 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1445 case lir_cond_less: acond = Assembler::less; break;
1446 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
1447 case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
1448 case lir_cond_greater: acond = Assembler::greater; break;
1449 case lir_cond_belowEqual: acond = Assembler::belowEqual; break;
1450 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break;
1451 default: ShouldNotReachHere();
1452 }
1453 }
1454 __ jcc(acond,*(op->label()));
1455 }
1456 }
1457
1458 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1459 LIR_Opr src = op->in_opr();
1460 LIR_Opr dest = op->result_opr();
1461
1462 switch (op->bytecode()) {
1463 case Bytecodes::_i2l:
1464 #ifdef _LP64
1465 __ movl2ptr(dest->as_register_lo(), src->as_register());
1466 #else
1467 move_regs(src->as_register(), dest->as_register_lo());
1468 move_regs(src->as_register(), dest->as_register_hi());
1469 __ sarl(dest->as_register_hi(), 31);
1470 #endif // LP64
1471 break;
1472
1473 case Bytecodes::_l2i:
1474 #ifdef _LP64
1475 __ movl(dest->as_register(), src->as_register_lo());
1476 #else
1477 move_regs(src->as_register_lo(), dest->as_register());
1478 #endif
1479 break;
1480
1481 case Bytecodes::_i2b:
1482 move_regs(src->as_register(), dest->as_register());
1483 __ sign_extend_byte(dest->as_register());
1484 break;
1485
1486 case Bytecodes::_i2c:
1487 move_regs(src->as_register(), dest->as_register());
1488 __ andl(dest->as_register(), 0xFFFF);
1489 break;
1490
1491 case Bytecodes::_i2s:
1492 move_regs(src->as_register(), dest->as_register());
1493 __ sign_extend_short(dest->as_register());
1494 break;
1495
1496
1497 case Bytecodes::_f2d:
1498 case Bytecodes::_d2f:
1499 if (dest->is_single_xmm()) {
1500 __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1501 } else if (dest->is_double_xmm()) {
1502 __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1503 } else {
1504 assert(src->fpu() == dest->fpu(), "register must be equal");
1505 // do nothing (float result is rounded later through spilling)
1506 }
1507 break;
1508
1509 case Bytecodes::_i2f:
1510 case Bytecodes::_i2d:
1511 if (dest->is_single_xmm()) {
1512 __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1513 } else if (dest->is_double_xmm()) {
1514 __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1515 } else {
1516 assert(dest->fpu() == 0, "result must be on TOS");
1517 __ movl(Address(rsp, 0), src->as_register());
1518 __ fild_s(Address(rsp, 0));
1519 }
1520 break;
1521
1522 case Bytecodes::_f2i:
1523 case Bytecodes::_d2i:
1524 if (src->is_single_xmm()) {
1525 __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
1526 } else if (src->is_double_xmm()) {
1527 __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
1528 } else {
1529 assert(src->fpu() == 0, "input must be on TOS");
1530 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
1531 __ fist_s(Address(rsp, 0));
1532 __ movl(dest->as_register(), Address(rsp, 0));
1533 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1534 }
1535
1536 // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
1537 assert(op->stub() != NULL, "stub required");
1538 __ cmpl(dest->as_register(), 0x80000000);
1539 __ jcc(Assembler::equal, *op->stub()->entry());
1540 __ bind(*op->stub()->continuation());
1541 break;
1542
1543 case Bytecodes::_l2f:
1544 case Bytecodes::_l2d:
1545 assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
1546 assert(dest->fpu() == 0, "result must be on TOS");
1547
1548 __ movptr(Address(rsp, 0), src->as_register_lo());
1549 NOT_LP64(__ movl(Address(rsp, BytesPerWord), src->as_register_hi()));
1550 __ fild_d(Address(rsp, 0));
1551 // float result is rounded later through spilling
1552 break;
1553
1554 case Bytecodes::_f2l:
1555 case Bytecodes::_d2l:
1556 assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
1557 assert(src->fpu() == 0, "input must be on TOS");
1558 assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
1559
1560 // instruction sequence too long to inline it here
1561 {
1562 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
1563 }
1564 break;
1565
1566 default: ShouldNotReachHere();
1567 }
1568 }
1569
1570 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1571 if (op->init_check()) {
1572 __ cmpb(Address(op->klass()->as_register(),
1573 instanceKlass::init_state_offset()),
1574 instanceKlass::fully_initialized);
1575 add_debug_info_for_null_check_here(op->stub()->info());
1576 __ jcc(Assembler::notEqual, *op->stub()->entry());
1577 }
1578 __ allocate_object(op->obj()->as_register(),
1579 op->tmp1()->as_register(),
1580 op->tmp2()->as_register(),
1581 op->header_size(),
1582 op->object_size(),
1583 op->klass()->as_register(),
1584 *op->stub()->entry());
1585 __ bind(*op->stub()->continuation());
1586 }
1587
1588 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1589 Register len = op->len()->as_register();
1590 LP64_ONLY( __ movslq(len, len); )
1591
1592 if (UseSlowPath ||
1593 (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
1594 (!UseFastNewTypeArray && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
1595 __ jmp(*op->stub()->entry());
1596 } else {
1597 Register tmp1 = op->tmp1()->as_register();
1598 Register tmp2 = op->tmp2()->as_register();
1599 Register tmp3 = op->tmp3()->as_register();
1600 if (len == tmp1) {
1601 tmp1 = tmp3;
1602 } else if (len == tmp2) {
1603 tmp2 = tmp3;
1604 } else if (len == tmp3) {
1605 // everything is ok
1606 } else {
1607 __ mov(tmp3, len);
1608 }
1609 __ allocate_array(op->obj()->as_register(),
1610 len,
1611 tmp1,
1612 tmp2,
1613 arrayOopDesc::header_size(op->type()),
1614 array_element_size(op->type()),
1615 op->klass()->as_register(),
1616 *op->stub()->entry());
1617 }
1618 __ bind(*op->stub()->continuation());
1619 }
1620
1621 void LIR_Assembler::type_profile_helper(Register mdo,
1622 ciMethodData *md, ciProfileData *data,
1623 Register recv, Label* update_done) {
1624 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1625 Label next_test;
1626 // See if the receiver is receiver[n].
1627 __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
1628 __ jccb(Assembler::notEqual, next_test);
1629 Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
1630 __ addptr(data_addr, DataLayout::counter_increment);
1631 __ jmp(*update_done);
1632 __ bind(next_test);
1633 }
1634
1635 // Didn't find receiver; find next empty slot and fill it in
1636 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1637 Label next_test;
1638 Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
1639 __ cmpptr(recv_addr, (intptr_t)NULL_WORD);
1640 __ jccb(Assembler::notEqual, next_test);
1641 __ movptr(recv_addr, recv);
1642 __ movptr(Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))), DataLayout::counter_increment);
1643 __ jmp(*update_done);
1644 __ bind(next_test);
1645 }
1646 }
1647
1648 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
1649 // we always need a stub for the failure case.
1650 CodeStub* stub = op->stub();
1651 Register obj = op->object()->as_register();
1652 Register k_RInfo = op->tmp1()->as_register();
1653 Register klass_RInfo = op->tmp2()->as_register();
1654 Register dst = op->result_opr()->as_register();
1655 ciKlass* k = op->klass();
1656 Register Rtmp1 = noreg;
1657
1658 // check if it needs to be profiled
1659 ciMethodData* md;
1660 ciProfileData* data;
1661
1662 if (op->should_profile()) {
1663 ciMethod* method = op->profiled_method();
1664 assert(method != NULL, "Should have method");
1665 int bci = op->profiled_bci();
1666 md = method->method_data_or_null();
1667 assert(md != NULL, "Sanity");
1668 data = md->bci_to_data(bci);
1669 assert(data != NULL, "need data for type check");
1670 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1671 }
1672 Label profile_cast_success, profile_cast_failure;
1673 Label *success_target = op->should_profile() ? &profile_cast_success : success;
1674 Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
1675
1676 if (obj == k_RInfo) {
1677 k_RInfo = dst;
1678 } else if (obj == klass_RInfo) {
1679 klass_RInfo = dst;
1680 }
1681 if (k->is_loaded() && !UseCompressedOops) {
1682 select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1683 } else {
1684 Rtmp1 = op->tmp3()->as_register();
1685 select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1686 }
1687
1688 assert_different_registers(obj, k_RInfo, klass_RInfo);
1689 if (!k->is_loaded()) {
1690 jobject2reg_with_patching(k_RInfo, op->info_for_patch());
1691 } else {
1692 #ifdef _LP64
1693 __ movoop(k_RInfo, k->constant_encoding());
1694 #endif // _LP64
1695 }
1696 assert(obj != k_RInfo, "must be different");
1697
1698 __ cmpptr(obj, (int32_t)NULL_WORD);
1699 if (op->should_profile()) {
1700 Label not_null;
1701 __ jccb(Assembler::notEqual, not_null);
1702 // Object is null; update MDO and exit
1703 Register mdo = klass_RInfo;
1704 __ movoop(mdo, md->constant_encoding());
1705 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1706 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1707 __ orl(data_addr, header_bits);
1708 __ jmp(*obj_is_null);
1709 __ bind(not_null);
1710 } else {
1711 __ jcc(Assembler::equal, *obj_is_null);
1712 }
1713 __ verify_oop(obj);
1714
1715 if (op->fast_check()) {
1716 // get object class
1717 // not a safepoint as obj null check happens earlier
1718 #ifdef _LP64
1719 if (UseCompressedOops) {
1720 __ load_klass(Rtmp1, obj);
1721 __ cmpptr(k_RInfo, Rtmp1);
1722 } else {
1723 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1724 }
1725 #else
1726 if (k->is_loaded()) {
1727 __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->constant_encoding());
1728 } else {
1729 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1730 }
1731 #endif
1732 __ jcc(Assembler::notEqual, *failure_target);
1733 // successful cast, fall through to profile or jump
1734 } else {
1735 // get object class
1736 // not a safepoint as obj null check happens earlier
1737 __ load_klass(klass_RInfo, obj);
1738 if (k->is_loaded()) {
1739 // See if we get an immediate positive hit
1740 #ifdef _LP64
1741 __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
1742 #else
1743 __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->constant_encoding());
1744 #endif // _LP64
1745 if ((juint)in_bytes(Klass::secondary_super_cache_offset()) != k->super_check_offset()) {
1746 __ jcc(Assembler::notEqual, *failure_target);
1747 // successful cast, fall through to profile or jump
1748 } else {
1749 // See if we get an immediate positive hit
1750 __ jcc(Assembler::equal, *success_target);
1751 // check for self
1752 #ifdef _LP64
1753 __ cmpptr(klass_RInfo, k_RInfo);
1754 #else
1755 __ cmpoop(klass_RInfo, k->constant_encoding());
1756 #endif // _LP64
1757 __ jcc(Assembler::equal, *success_target);
1758
1759 __ push(klass_RInfo);
1760 #ifdef _LP64
1761 __ push(k_RInfo);
1762 #else
1763 __ pushoop(k->constant_encoding());
1764 #endif // _LP64
1765 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1766 __ pop(klass_RInfo);
1767 __ pop(klass_RInfo);
1768 // result is a boolean
1769 __ cmpl(klass_RInfo, 0);
1770 __ jcc(Assembler::equal, *failure_target);
1771 // successful cast, fall through to profile or jump
1772 }
1773 } else {
1774 // perform the fast part of the checking logic
1775 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1776 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1777 __ push(klass_RInfo);
1778 __ push(k_RInfo);
1779 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1780 __ pop(klass_RInfo);
1781 __ pop(k_RInfo);
1782 // result is a boolean
1783 __ cmpl(k_RInfo, 0);
1784 __ jcc(Assembler::equal, *failure_target);
1785 // successful cast, fall through to profile or jump
1786 }
1787 }
1788 if (op->should_profile()) {
1789 Register mdo = klass_RInfo, recv = k_RInfo;
1790 __ bind(profile_cast_success);
1791 __ movoop(mdo, md->constant_encoding());
1792 __ load_klass(recv, obj);
1793 Label update_done;
1794 type_profile_helper(mdo, md, data, recv, success);
1795 __ jmp(*success);
1796
1797 __ bind(profile_cast_failure);
1798 __ movoop(mdo, md->constant_encoding());
1799 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1800 __ subptr(counter_addr, DataLayout::counter_increment);
1801 __ jmp(*failure);
1802 }
1803 __ jmp(*success);
1804 }
1805
1806
1807 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1808 LIR_Code code = op->code();
1809 if (code == lir_store_check) {
1810 Register value = op->object()->as_register();
1811 Register array = op->array()->as_register();
1812 Register k_RInfo = op->tmp1()->as_register();
1813 Register klass_RInfo = op->tmp2()->as_register();
1814 Register Rtmp1 = op->tmp3()->as_register();
1815
1816 CodeStub* stub = op->stub();
1817
1818 // check if it needs to be profiled
1819 ciMethodData* md;
1820 ciProfileData* data;
1821
1822 if (op->should_profile()) {
1823 ciMethod* method = op->profiled_method();
1824 assert(method != NULL, "Should have method");
1825 int bci = op->profiled_bci();
1826 md = method->method_data_or_null();
1827 assert(md != NULL, "Sanity");
1828 data = md->bci_to_data(bci);
1829 assert(data != NULL, "need data for type check");
1830 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1831 }
1832 Label profile_cast_success, profile_cast_failure, done;
1833 Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1834 Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
1835
1836 __ cmpptr(value, (int32_t)NULL_WORD);
1837 if (op->should_profile()) {
1838 Label not_null;
1839 __ jccb(Assembler::notEqual, not_null);
1840 // Object is null; update MDO and exit
1841 Register mdo = klass_RInfo;
1842 __ movoop(mdo, md->constant_encoding());
1843 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1844 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1845 __ orl(data_addr, header_bits);
1846 __ jmp(done);
1847 __ bind(not_null);
1848 } else {
1849 __ jcc(Assembler::equal, done);
1850 }
1851
1852 add_debug_info_for_null_check_here(op->info_for_exception());
1853 __ load_klass(k_RInfo, array);
1854 __ load_klass(klass_RInfo, value);
1855
1856 // get instance klass (it's already uncompressed)
1857 __ movptr(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset()));
1858 // perform the fast part of the checking logic
1859 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1860 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1861 __ push(klass_RInfo);
1862 __ push(k_RInfo);
1863 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1864 __ pop(klass_RInfo);
1865 __ pop(k_RInfo);
1866 // result is a boolean
1867 __ cmpl(k_RInfo, 0);
1868 __ jcc(Assembler::equal, *failure_target);
1869 // fall through to the success case
1870
1871 if (op->should_profile()) {
1872 Register mdo = klass_RInfo, recv = k_RInfo;
1873 __ bind(profile_cast_success);
1874 __ movoop(mdo, md->constant_encoding());
1875 __ load_klass(recv, value);
1876 Label update_done;
1877 type_profile_helper(mdo, md, data, recv, &done);
1878 __ jmpb(done);
1879
1880 __ bind(profile_cast_failure);
1881 __ movoop(mdo, md->constant_encoding());
1882 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1883 __ subptr(counter_addr, DataLayout::counter_increment);
1884 __ jmp(*stub->entry());
1885 }
1886
1887 __ bind(done);
1888 } else
1889 if (code == lir_checkcast) {
1890 Register obj = op->object()->as_register();
1891 Register dst = op->result_opr()->as_register();
1892 Label success;
1893 emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
1894 __ bind(success);
1895 if (dst != obj) {
1896 __ mov(dst, obj);
1897 }
1898 } else
1899 if (code == lir_instanceof) {
1900 Register obj = op->object()->as_register();
1901 Register dst = op->result_opr()->as_register();
1902 Label success, failure, done;
1903 emit_typecheck_helper(op, &success, &failure, &failure);
1904 __ bind(failure);
1905 __ xorptr(dst, dst);
1906 __ jmpb(done);
1907 __ bind(success);
1908 __ movptr(dst, 1);
1909 __ bind(done);
1910 } else {
1911 ShouldNotReachHere();
1912 }
1913
1914 }
1915
1916
1917 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1918 if (LP64_ONLY(false &&) op->code() == lir_cas_long && VM_Version::supports_cx8()) {
1919 assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
1920 assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
1921 assert(op->new_value()->as_register_lo() == rbx, "wrong register");
1922 assert(op->new_value()->as_register_hi() == rcx, "wrong register");
1923 Register addr = op->addr()->as_register();
1924 if (os::is_MP()) {
1925 __ lock();
1926 }
1927 NOT_LP64(__ cmpxchg8(Address(addr, 0)));
1928
1929 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
1930 NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
1931 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1932 Register newval = op->new_value()->as_register();
1933 Register cmpval = op->cmp_value()->as_register();
1934 assert(cmpval == rax, "wrong register");
1935 assert(newval != NULL, "new val must be register");
1936 assert(cmpval != newval, "cmp and new values must be in different registers");
1937 assert(cmpval != addr, "cmp and addr must be in different registers");
1938 assert(newval != addr, "new value and addr must be in different registers");
1939
1940 if ( op->code() == lir_cas_obj) {
1941 #ifdef _LP64
1942 if (UseCompressedOops) {
1943 __ encode_heap_oop(cmpval);
1944 __ mov(rscratch1, newval);
1945 __ encode_heap_oop(rscratch1);
1946 if (os::is_MP()) {
1947 __ lock();
1948 }
1949 // cmpval (rax) is implicitly used by this instruction
1950 __ cmpxchgl(rscratch1, Address(addr, 0));
1951 } else
1952 #endif
1953 {
1954 if (os::is_MP()) {
1955 __ lock();
1956 }
1957 __ cmpxchgptr(newval, Address(addr, 0));
1958 }
1959 } else {
1960 assert(op->code() == lir_cas_int, "lir_cas_int expected");
1961 if (os::is_MP()) {
1962 __ lock();
1963 }
1964 __ cmpxchgl(newval, Address(addr, 0));
1965 }
1966 #ifdef _LP64
1967 } else if (op->code() == lir_cas_long) {
1968 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1969 Register newval = op->new_value()->as_register_lo();
1970 Register cmpval = op->cmp_value()->as_register_lo();
1971 assert(cmpval == rax, "wrong register");
1972 assert(newval != NULL, "new val must be register");
1973 assert(cmpval != newval, "cmp and new values must be in different registers");
1974 assert(cmpval != addr, "cmp and addr must be in different registers");
1975 assert(newval != addr, "new value and addr must be in different registers");
1976 if (os::is_MP()) {
1977 __ lock();
1978 }
1979 __ cmpxchgq(newval, Address(addr, 0));
1980 #endif // _LP64
1981 } else {
1982 Unimplemented();
1983 }
1984 }
1985
1986 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
1987 Assembler::Condition acond, ncond;
1988 switch (condition) {
1989 case lir_cond_equal: acond = Assembler::equal; ncond = Assembler::notEqual; break;
1990 case lir_cond_notEqual: acond = Assembler::notEqual; ncond = Assembler::equal; break;
1991 case lir_cond_less: acond = Assembler::less; ncond = Assembler::greaterEqual; break;
1992 case lir_cond_lessEqual: acond = Assembler::lessEqual; ncond = Assembler::greater; break;
1993 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less; break;
1994 case lir_cond_greater: acond = Assembler::greater; ncond = Assembler::lessEqual; break;
1995 case lir_cond_belowEqual: acond = Assembler::belowEqual; ncond = Assembler::above; break;
1996 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; ncond = Assembler::below; break;
1997 default: ShouldNotReachHere();
1998 }
1999
2000 if (opr1->is_cpu_register()) {
2001 reg2reg(opr1, result);
2002 } else if (opr1->is_stack()) {
2003 stack2reg(opr1, result, result->type());
2004 } else if (opr1->is_constant()) {
2005 const2reg(opr1, result, lir_patch_none, NULL);
2006 } else {
2007 ShouldNotReachHere();
2008 }
2009
2010 if (VM_Version::supports_cmov() && !opr2->is_constant()) {
2011 // optimized version that does not require a branch
2012 if (opr2->is_single_cpu()) {
2013 assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
2014 __ cmov(ncond, result->as_register(), opr2->as_register());
2015 } else if (opr2->is_double_cpu()) {
2016 assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2017 assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2018 __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
2019 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
2020 } else if (opr2->is_single_stack()) {
2021 __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
2022 } else if (opr2->is_double_stack()) {
2023 __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
2024 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
2025 } else {
2026 ShouldNotReachHere();
2027 }
2028
2029 } else {
2030 Label skip;
2031 __ jcc (acond, skip);
2032 if (opr2->is_cpu_register()) {
2033 reg2reg(opr2, result);
2034 } else if (opr2->is_stack()) {
2035 stack2reg(opr2, result, result->type());
2036 } else if (opr2->is_constant()) {
2037 const2reg(opr2, result, lir_patch_none, NULL);
2038 } else {
2039 ShouldNotReachHere();
2040 }
2041 __ bind(skip);
2042 }
2043 }
2044
2045
2046 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
2047 assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
2048
2049 if (left->is_single_cpu()) {
2050 assert(left == dest, "left and dest must be equal");
2051 Register lreg = left->as_register();
2052
2053 if (right->is_single_cpu()) {
2054 // cpu register - cpu register
2055 Register rreg = right->as_register();
2056 switch (code) {
2057 case lir_add: __ addl (lreg, rreg); break;
2058 case lir_sub: __ subl (lreg, rreg); break;
2059 case lir_mul: __ imull(lreg, rreg); break;
2060 default: ShouldNotReachHere();
2061 }
2062
2063 } else if (right->is_stack()) {
2064 // cpu register - stack
2065 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2066 switch (code) {
2067 case lir_add: __ addl(lreg, raddr); break;
2068 case lir_sub: __ subl(lreg, raddr); break;
2069 default: ShouldNotReachHere();
2070 }
2071
2072 } else if (right->is_constant()) {
2073 // cpu register - constant
2074 jint c = right->as_constant_ptr()->as_jint();
2075 switch (code) {
2076 case lir_add: {
2077 __ incrementl(lreg, c);
2078 break;
2079 }
2080 case lir_sub: {
2081 __ decrementl(lreg, c);
2082 break;
2083 }
2084 default: ShouldNotReachHere();
2085 }
2086
2087 } else {
2088 ShouldNotReachHere();
2089 }
2090
2091 } else if (left->is_double_cpu()) {
2092 assert(left == dest, "left and dest must be equal");
2093 Register lreg_lo = left->as_register_lo();
2094 Register lreg_hi = left->as_register_hi();
2095
2096 if (right->is_double_cpu()) {
2097 // cpu register - cpu register
2098 Register rreg_lo = right->as_register_lo();
2099 Register rreg_hi = right->as_register_hi();
2100 NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
2101 LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
2102 switch (code) {
2103 case lir_add:
2104 __ addptr(lreg_lo, rreg_lo);
2105 NOT_LP64(__ adcl(lreg_hi, rreg_hi));
2106 break;
2107 case lir_sub:
2108 __ subptr(lreg_lo, rreg_lo);
2109 NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
2110 break;
2111 case lir_mul:
2112 #ifdef _LP64
2113 __ imulq(lreg_lo, rreg_lo);
2114 #else
2115 assert(lreg_lo == rax && lreg_hi == rdx, "must be");
2116 __ imull(lreg_hi, rreg_lo);
2117 __ imull(rreg_hi, lreg_lo);
2118 __ addl (rreg_hi, lreg_hi);
2119 __ mull (rreg_lo);
2120 __ addl (lreg_hi, rreg_hi);
2121 #endif // _LP64
2122 break;
2123 default:
2124 ShouldNotReachHere();
2125 }
2126
2127 } else if (right->is_constant()) {
2128 // cpu register - constant
2129 #ifdef _LP64
2130 jlong c = right->as_constant_ptr()->as_jlong_bits();
2131 __ movptr(r10, (intptr_t) c);
2132 switch (code) {
2133 case lir_add:
2134 __ addptr(lreg_lo, r10);
2135 break;
2136 case lir_sub:
2137 __ subptr(lreg_lo, r10);
2138 break;
2139 default:
2140 ShouldNotReachHere();
2141 }
2142 #else
2143 jint c_lo = right->as_constant_ptr()->as_jint_lo();
2144 jint c_hi = right->as_constant_ptr()->as_jint_hi();
2145 switch (code) {
2146 case lir_add:
2147 __ addptr(lreg_lo, c_lo);
2148 __ adcl(lreg_hi, c_hi);
2149 break;
2150 case lir_sub:
2151 __ subptr(lreg_lo, c_lo);
2152 __ sbbl(lreg_hi, c_hi);
2153 break;
2154 default:
2155 ShouldNotReachHere();
2156 }
2157 #endif // _LP64
2158
2159 } else {
2160 ShouldNotReachHere();
2161 }
2162
2163 } else if (left->is_single_xmm()) {
2164 assert(left == dest, "left and dest must be equal");
2165 XMMRegister lreg = left->as_xmm_float_reg();
2166
2167 if (right->is_single_xmm()) {
2168 XMMRegister rreg = right->as_xmm_float_reg();
2169 switch (code) {
2170 case lir_add: __ addss(lreg, rreg); break;
2171 case lir_sub: __ subss(lreg, rreg); break;
2172 case lir_mul_strictfp: // fall through
2173 case lir_mul: __ mulss(lreg, rreg); break;
2174 case lir_div_strictfp: // fall through
2175 case lir_div: __ divss(lreg, rreg); break;
2176 default: ShouldNotReachHere();
2177 }
2178 } else {
2179 Address raddr;
2180 if (right->is_single_stack()) {
2181 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2182 } else if (right->is_constant()) {
2183 // hack for now
2184 raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
2185 } else {
2186 ShouldNotReachHere();
2187 }
2188 switch (code) {
2189 case lir_add: __ addss(lreg, raddr); break;
2190 case lir_sub: __ subss(lreg, raddr); break;
2191 case lir_mul_strictfp: // fall through
2192 case lir_mul: __ mulss(lreg, raddr); break;
2193 case lir_div_strictfp: // fall through
2194 case lir_div: __ divss(lreg, raddr); break;
2195 default: ShouldNotReachHere();
2196 }
2197 }
2198
2199 } else if (left->is_double_xmm()) {
2200 assert(left == dest, "left and dest must be equal");
2201
2202 XMMRegister lreg = left->as_xmm_double_reg();
2203 if (right->is_double_xmm()) {
2204 XMMRegister rreg = right->as_xmm_double_reg();
2205 switch (code) {
2206 case lir_add: __ addsd(lreg, rreg); break;
2207 case lir_sub: __ subsd(lreg, rreg); break;
2208 case lir_mul_strictfp: // fall through
2209 case lir_mul: __ mulsd(lreg, rreg); break;
2210 case lir_div_strictfp: // fall through
2211 case lir_div: __ divsd(lreg, rreg); break;
2212 default: ShouldNotReachHere();
2213 }
2214 } else {
2215 Address raddr;
2216 if (right->is_double_stack()) {
2217 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2218 } else if (right->is_constant()) {
2219 // hack for now
2220 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2221 } else {
2222 ShouldNotReachHere();
2223 }
2224 switch (code) {
2225 case lir_add: __ addsd(lreg, raddr); break;
2226 case lir_sub: __ subsd(lreg, raddr); break;
2227 case lir_mul_strictfp: // fall through
2228 case lir_mul: __ mulsd(lreg, raddr); break;
2229 case lir_div_strictfp: // fall through
2230 case lir_div: __ divsd(lreg, raddr); break;
2231 default: ShouldNotReachHere();
2232 }
2233 }
2234
2235 } else if (left->is_single_fpu()) {
2236 assert(dest->is_single_fpu(), "fpu stack allocation required");
2237
2238 if (right->is_single_fpu()) {
2239 arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
2240
2241 } else {
2242 assert(left->fpu_regnr() == 0, "left must be on TOS");
2243 assert(dest->fpu_regnr() == 0, "dest must be on TOS");
2244
2245 Address raddr;
2246 if (right->is_single_stack()) {
2247 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2248 } else if (right->is_constant()) {
2249 address const_addr = float_constant(right->as_jfloat());
2250 assert(const_addr != NULL, "incorrect float/double constant maintainance");
2251 // hack for now
2252 raddr = __ as_Address(InternalAddress(const_addr));
2253 } else {
2254 ShouldNotReachHere();
2255 }
2256
2257 switch (code) {
2258 case lir_add: __ fadd_s(raddr); break;
2259 case lir_sub: __ fsub_s(raddr); break;
2260 case lir_mul_strictfp: // fall through
2261 case lir_mul: __ fmul_s(raddr); break;
2262 case lir_div_strictfp: // fall through
2263 case lir_div: __ fdiv_s(raddr); break;
2264 default: ShouldNotReachHere();
2265 }
2266 }
2267
2268 } else if (left->is_double_fpu()) {
2269 assert(dest->is_double_fpu(), "fpu stack allocation required");
2270
2271 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2272 // Double values require special handling for strictfp mul/div on x86
2273 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
2274 __ fmulp(left->fpu_regnrLo() + 1);
2275 }
2276
2277 if (right->is_double_fpu()) {
2278 arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
2279
2280 } else {
2281 assert(left->fpu_regnrLo() == 0, "left must be on TOS");
2282 assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
2283
2284 Address raddr;
2285 if (right->is_double_stack()) {
2286 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2287 } else if (right->is_constant()) {
2288 // hack for now
2289 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2290 } else {
2291 ShouldNotReachHere();
2292 }
2293
2294 switch (code) {
2295 case lir_add: __ fadd_d(raddr); break;
2296 case lir_sub: __ fsub_d(raddr); break;
2297 case lir_mul_strictfp: // fall through
2298 case lir_mul: __ fmul_d(raddr); break;
2299 case lir_div_strictfp: // fall through
2300 case lir_div: __ fdiv_d(raddr); break;
2301 default: ShouldNotReachHere();
2302 }
2303 }
2304
2305 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2306 // Double values require special handling for strictfp mul/div on x86
2307 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
2308 __ fmulp(dest->fpu_regnrLo() + 1);
2309 }
2310
2311 } else if (left->is_single_stack() || left->is_address()) {
2312 assert(left == dest, "left and dest must be equal");
2313
2314 Address laddr;
2315 if (left->is_single_stack()) {
2316 laddr = frame_map()->address_for_slot(left->single_stack_ix());
2317 } else if (left->is_address()) {
2318 laddr = as_Address(left->as_address_ptr());
2319 } else {
2320 ShouldNotReachHere();
2321 }
2322
2323 if (right->is_single_cpu()) {
2324 Register rreg = right->as_register();
2325 switch (code) {
2326 case lir_add: __ addl(laddr, rreg); break;
2327 case lir_sub: __ subl(laddr, rreg); break;
2328 default: ShouldNotReachHere();
2329 }
2330 } else if (right->is_constant()) {
2331 jint c = right->as_constant_ptr()->as_jint();
2332 switch (code) {
2333 case lir_add: {
2334 __ incrementl(laddr, c);
2335 break;
2336 }
2337 case lir_sub: {
2338 __ decrementl(laddr, c);
2339 break;
2340 }
2341 default: ShouldNotReachHere();
2342 }
2343 } else {
2344 ShouldNotReachHere();
2345 }
2346
2347 } else {
2348 ShouldNotReachHere();
2349 }
2350 }
2351
2352 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
2353 assert(pop_fpu_stack || (left_index == dest_index || right_index == dest_index), "invalid LIR");
2354 assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
2355 assert(left_index == 0 || right_index == 0, "either must be on top of stack");
2356
2357 bool left_is_tos = (left_index == 0);
2358 bool dest_is_tos = (dest_index == 0);
2359 int non_tos_index = (left_is_tos ? right_index : left_index);
2360
2361 switch (code) {
2362 case lir_add:
2363 if (pop_fpu_stack) __ faddp(non_tos_index);
2364 else if (dest_is_tos) __ fadd (non_tos_index);
2365 else __ fadda(non_tos_index);
2366 break;
2367
2368 case lir_sub:
2369 if (left_is_tos) {
2370 if (pop_fpu_stack) __ fsubrp(non_tos_index);
2371 else if (dest_is_tos) __ fsub (non_tos_index);
2372 else __ fsubra(non_tos_index);
2373 } else {
2374 if (pop_fpu_stack) __ fsubp (non_tos_index);
2375 else if (dest_is_tos) __ fsubr (non_tos_index);
2376 else __ fsuba (non_tos_index);
2377 }
2378 break;
2379
2380 case lir_mul_strictfp: // fall through
2381 case lir_mul:
2382 if (pop_fpu_stack) __ fmulp(non_tos_index);
2383 else if (dest_is_tos) __ fmul (non_tos_index);
2384 else __ fmula(non_tos_index);
2385 break;
2386
2387 case lir_div_strictfp: // fall through
2388 case lir_div:
2389 if (left_is_tos) {
2390 if (pop_fpu_stack) __ fdivrp(non_tos_index);
2391 else if (dest_is_tos) __ fdiv (non_tos_index);
2392 else __ fdivra(non_tos_index);
2393 } else {
2394 if (pop_fpu_stack) __ fdivp (non_tos_index);
2395 else if (dest_is_tos) __ fdivr (non_tos_index);
2396 else __ fdiva (non_tos_index);
2397 }
2398 break;
2399
2400 case lir_rem:
2401 assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
2402 __ fremr(noreg);
2403 break;
2404
2405 default:
2406 ShouldNotReachHere();
2407 }
2408 }
2409
2410
2411 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
2412 if (value->is_double_xmm()) {
2413 switch(code) {
2414 case lir_abs :
2415 {
2416 if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
2417 __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
2418 }
2419 __ andpd(dest->as_xmm_double_reg(),
2420 ExternalAddress((address)double_signmask_pool));
2421 }
2422 break;
2423
2424 case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
2425 // all other intrinsics are not available in the SSE instruction set, so FPU is used
2426 default : ShouldNotReachHere();
2427 }
2428
2429 } else if (value->is_double_fpu()) {
2430 assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
2431 switch(code) {
2432 case lir_log : __ flog() ; break;
2433 case lir_log10 : __ flog10() ; break;
2434 case lir_abs : __ fabs() ; break;
2435 case lir_sqrt : __ fsqrt(); break;
2436 case lir_sin :
2437 // Should consider not saving rbx, if not necessary
2438 __ trigfunc('s', op->as_Op2()->fpu_stack_size());
2439 break;
2440 case lir_cos :
2441 // Should consider not saving rbx, if not necessary
2442 assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
2443 __ trigfunc('c', op->as_Op2()->fpu_stack_size());
2444 break;
2445 case lir_tan :
2446 // Should consider not saving rbx, if not necessary
2447 __ trigfunc('t', op->as_Op2()->fpu_stack_size());
2448 break;
2449 default : ShouldNotReachHere();
2450 }
2451 } else {
2452 Unimplemented();
2453 }
2454 }
2455
2456 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
2457 // assert(left->destroys_register(), "check");
2458 if (left->is_single_cpu()) {
2459 Register reg = left->as_register();
2460 if (right->is_constant()) {
2461 int val = right->as_constant_ptr()->as_jint();
2462 switch (code) {
2463 case lir_logic_and: __ andl (reg, val); break;
2464 case lir_logic_or: __ orl (reg, val); break;
2465 case lir_logic_xor: __ xorl (reg, val); break;
2466 default: ShouldNotReachHere();
2467 }
2468 } else if (right->is_stack()) {
2469 // added support for stack operands
2470 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2471 switch (code) {
2472 case lir_logic_and: __ andl (reg, raddr); break;
2473 case lir_logic_or: __ orl (reg, raddr); break;
2474 case lir_logic_xor: __ xorl (reg, raddr); break;
2475 default: ShouldNotReachHere();
2476 }
2477 } else {
2478 Register rright = right->as_register();
2479 switch (code) {
2480 case lir_logic_and: __ andptr (reg, rright); break;
2481 case lir_logic_or : __ orptr (reg, rright); break;
2482 case lir_logic_xor: __ xorptr (reg, rright); break;
2483 default: ShouldNotReachHere();
2484 }
2485 }
2486 move_regs(reg, dst->as_register());
2487 } else {
2488 Register l_lo = left->as_register_lo();
2489 Register l_hi = left->as_register_hi();
2490 if (right->is_constant()) {
2491 #ifdef _LP64
2492 __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
2493 switch (code) {
2494 case lir_logic_and:
2495 __ andq(l_lo, rscratch1);
2496 break;
2497 case lir_logic_or:
2498 __ orq(l_lo, rscratch1);
2499 break;
2500 case lir_logic_xor:
2501 __ xorq(l_lo, rscratch1);
2502 break;
2503 default: ShouldNotReachHere();
2504 }
2505 #else
2506 int r_lo = right->as_constant_ptr()->as_jint_lo();
2507 int r_hi = right->as_constant_ptr()->as_jint_hi();
2508 switch (code) {
2509 case lir_logic_and:
2510 __ andl(l_lo, r_lo);
2511 __ andl(l_hi, r_hi);
2512 break;
2513 case lir_logic_or:
2514 __ orl(l_lo, r_lo);
2515 __ orl(l_hi, r_hi);
2516 break;
2517 case lir_logic_xor:
2518 __ xorl(l_lo, r_lo);
2519 __ xorl(l_hi, r_hi);
2520 break;
2521 default: ShouldNotReachHere();
2522 }
2523 #endif // _LP64
2524 } else {
2525 #ifdef _LP64
2526 Register r_lo;
2527 if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
2528 r_lo = right->as_register();
2529 } else {
2530 r_lo = right->as_register_lo();
2531 }
2532 #else
2533 Register r_lo = right->as_register_lo();
2534 Register r_hi = right->as_register_hi();
2535 assert(l_lo != r_hi, "overwriting registers");
2536 #endif
2537 switch (code) {
2538 case lir_logic_and:
2539 __ andptr(l_lo, r_lo);
2540 NOT_LP64(__ andptr(l_hi, r_hi);)
2541 break;
2542 case lir_logic_or:
2543 __ orptr(l_lo, r_lo);
2544 NOT_LP64(__ orptr(l_hi, r_hi);)
2545 break;
2546 case lir_logic_xor:
2547 __ xorptr(l_lo, r_lo);
2548 NOT_LP64(__ xorptr(l_hi, r_hi);)
2549 break;
2550 default: ShouldNotReachHere();
2551 }
2552 }
2553
2554 Register dst_lo = dst->as_register_lo();
2555 Register dst_hi = dst->as_register_hi();
2556
2557 #ifdef _LP64
2558 move_regs(l_lo, dst_lo);
2559 #else
2560 if (dst_lo == l_hi) {
2561 assert(dst_hi != l_lo, "overwriting registers");
2562 move_regs(l_hi, dst_hi);
2563 move_regs(l_lo, dst_lo);
2564 } else {
2565 assert(dst_lo != l_hi, "overwriting registers");
2566 move_regs(l_lo, dst_lo);
2567 move_regs(l_hi, dst_hi);
2568 }
2569 #endif // _LP64
2570 }
2571 }
2572
2573
2574 // we assume that rax, and rdx can be overwritten
2575 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
2576
2577 assert(left->is_single_cpu(), "left must be register");
2578 assert(right->is_single_cpu() || right->is_constant(), "right must be register or constant");
2579 assert(result->is_single_cpu(), "result must be register");
2580
2581 // assert(left->destroys_register(), "check");
2582 // assert(right->destroys_register(), "check");
2583
2584 Register lreg = left->as_register();
2585 Register dreg = result->as_register();
2586
2587 if (right->is_constant()) {
2588 int divisor = right->as_constant_ptr()->as_jint();
2589 assert(divisor > 0 && is_power_of_2(divisor), "must be");
2590 if (code == lir_idiv) {
2591 assert(lreg == rax, "must be rax,");
2592 assert(temp->as_register() == rdx, "tmp register must be rdx");
2593 __ cdql(); // sign extend into rdx:rax
2594 if (divisor == 2) {
2595 __ subl(lreg, rdx);
2596 } else {
2597 __ andl(rdx, divisor - 1);
2598 __ addl(lreg, rdx);
2599 }
2600 __ sarl(lreg, log2_intptr(divisor));
2601 move_regs(lreg, dreg);
2602 } else if (code == lir_irem) {
2603 Label done;
2604 __ mov(dreg, lreg);
2605 __ andl(dreg, 0x80000000 | (divisor - 1));
2606 __ jcc(Assembler::positive, done);
2607 __ decrement(dreg);
2608 __ orl(dreg, ~(divisor - 1));
2609 __ increment(dreg);
2610 __ bind(done);
2611 } else {
2612 ShouldNotReachHere();
2613 }
2614 } else {
2615 Register rreg = right->as_register();
2616 assert(lreg == rax, "left register must be rax,");
2617 assert(rreg != rdx, "right register must not be rdx");
2618 assert(temp->as_register() == rdx, "tmp register must be rdx");
2619
2620 move_regs(lreg, rax);
2621
2622 int idivl_offset = __ corrected_idivl(rreg);
2623 add_debug_info_for_div0(idivl_offset, info);
2624 if (code == lir_irem) {
2625 move_regs(rdx, dreg); // result is in rdx
2626 } else {
2627 move_regs(rax, dreg);
2628 }
2629 }
2630 }
2631
2632
2633 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
2634 if (opr1->is_single_cpu()) {
2635 Register reg1 = opr1->as_register();
2636 if (opr2->is_single_cpu()) {
2637 // cpu register - cpu register
2638 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2639 __ cmpptr(reg1, opr2->as_register());
2640 } else {
2641 assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
2642 __ cmpl(reg1, opr2->as_register());
2643 }
2644 } else if (opr2->is_stack()) {
2645 // cpu register - stack
2646 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2647 __ cmpptr(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2648 } else {
2649 __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2650 }
2651 } else if (opr2->is_constant()) {
2652 // cpu register - constant
2653 LIR_Const* c = opr2->as_constant_ptr();
2654 if (c->type() == T_INT) {
2655 __ cmpl(reg1, c->as_jint());
2656 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2657 // In 64bit oops are single register
2658 jobject o = c->as_jobject();
2659 if (o == NULL) {
2660 __ cmpptr(reg1, (int32_t)NULL_WORD);
2661 } else {
2662 #ifdef _LP64
2663 __ movoop(rscratch1, o);
2664 __ cmpptr(reg1, rscratch1);
2665 #else
2666 __ cmpoop(reg1, c->as_jobject());
2667 #endif // _LP64
2668 }
2669 } else {
2670 fatal(err_msg("unexpected type: %s", basictype_to_str(c->type())));
2671 }
2672 // cpu register - address
2673 } else if (opr2->is_address()) {
2674 if (op->info() != NULL) {
2675 add_debug_info_for_null_check_here(op->info());
2676 }
2677 __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
2678 } else {
2679 ShouldNotReachHere();
2680 }
2681
2682 } else if(opr1->is_double_cpu()) {
2683 Register xlo = opr1->as_register_lo();
2684 Register xhi = opr1->as_register_hi();
2685 if (opr2->is_double_cpu()) {
2686 #ifdef _LP64
2687 __ cmpptr(xlo, opr2->as_register_lo());
2688 #else
2689 // cpu register - cpu register
2690 Register ylo = opr2->as_register_lo();
2691 Register yhi = opr2->as_register_hi();
2692 __ subl(xlo, ylo);
2693 __ sbbl(xhi, yhi);
2694 if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
2695 __ orl(xhi, xlo);
2696 }
2697 #endif // _LP64
2698 } else if (opr2->is_constant()) {
2699 // cpu register - constant 0
2700 assert(opr2->as_jlong() == (jlong)0, "only handles zero");
2701 #ifdef _LP64
2702 __ cmpptr(xlo, (int32_t)opr2->as_jlong());
2703 #else
2704 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
2705 __ orl(xhi, xlo);
2706 #endif // _LP64
2707 } else {
2708 ShouldNotReachHere();
2709 }
2710
2711 } else if (opr1->is_single_xmm()) {
2712 XMMRegister reg1 = opr1->as_xmm_float_reg();
2713 if (opr2->is_single_xmm()) {
2714 // xmm register - xmm register
2715 __ ucomiss(reg1, opr2->as_xmm_float_reg());
2716 } else if (opr2->is_stack()) {
2717 // xmm register - stack
2718 __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2719 } else if (opr2->is_constant()) {
2720 // xmm register - constant
2721 __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
2722 } else if (opr2->is_address()) {
2723 // xmm register - address
2724 if (op->info() != NULL) {
2725 add_debug_info_for_null_check_here(op->info());
2726 }
2727 __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
2728 } else {
2729 ShouldNotReachHere();
2730 }
2731
2732 } else if (opr1->is_double_xmm()) {
2733 XMMRegister reg1 = opr1->as_xmm_double_reg();
2734 if (opr2->is_double_xmm()) {
2735 // xmm register - xmm register
2736 __ ucomisd(reg1, opr2->as_xmm_double_reg());
2737 } else if (opr2->is_stack()) {
2738 // xmm register - stack
2739 __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
2740 } else if (opr2->is_constant()) {
2741 // xmm register - constant
2742 __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
2743 } else if (opr2->is_address()) {
2744 // xmm register - address
2745 if (op->info() != NULL) {
2746 add_debug_info_for_null_check_here(op->info());
2747 }
2748 __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
2749 } else {
2750 ShouldNotReachHere();
2751 }
2752
2753 } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
2754 assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
2755 assert(opr2->is_fpu_register(), "both must be registers");
2756 __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2757
2758 } else if (opr1->is_address() && opr2->is_constant()) {
2759 LIR_Const* c = opr2->as_constant_ptr();
2760 #ifdef _LP64
2761 if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2762 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
2763 __ movoop(rscratch1, c->as_jobject());
2764 }
2765 #endif // LP64
2766 if (op->info() != NULL) {
2767 add_debug_info_for_null_check_here(op->info());
2768 }
2769 // special case: address - constant
2770 LIR_Address* addr = opr1->as_address_ptr();
2771 if (c->type() == T_INT) {
2772 __ cmpl(as_Address(addr), c->as_jint());
2773 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2774 #ifdef _LP64
2775 // %%% Make this explode if addr isn't reachable until we figure out a
2776 // better strategy by giving noreg as the temp for as_Address
2777 __ cmpptr(rscratch1, as_Address(addr, noreg));
2778 #else
2779 __ cmpoop(as_Address(addr), c->as_jobject());
2780 #endif // _LP64
2781 } else {
2782 ShouldNotReachHere();
2783 }
2784
2785 } else {
2786 ShouldNotReachHere();
2787 }
2788 }
2789
2790 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
2791 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
2792 if (left->is_single_xmm()) {
2793 assert(right->is_single_xmm(), "must match");
2794 __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2795 } else if (left->is_double_xmm()) {
2796 assert(right->is_double_xmm(), "must match");
2797 __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2798
2799 } else {
2800 assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
2801 assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
2802
2803 assert(left->fpu() == 0, "left must be on TOS");
2804 __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
2805 op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2806 }
2807 } else {
2808 assert(code == lir_cmp_l2i, "check");
2809 #ifdef _LP64
2810 Label done;
2811 Register dest = dst->as_register();
2812 __ cmpptr(left->as_register_lo(), right->as_register_lo());
2813 __ movl(dest, -1);
2814 __ jccb(Assembler::less, done);
2815 __ set_byte_if_not_zero(dest);
2816 __ movzbl(dest, dest);
2817 __ bind(done);
2818 #else
2819 __ lcmp2int(left->as_register_hi(),
2820 left->as_register_lo(),
2821 right->as_register_hi(),
2822 right->as_register_lo());
2823 move_regs(left->as_register_hi(), dst->as_register());
2824 #endif // _LP64
2825 }
2826 }
2827
2828
2829 void LIR_Assembler::align_call(LIR_Code code) {
2830 if (os::is_MP()) {
2831 // make sure that the displacement word of the call ends up word aligned
2832 int offset = __ offset();
2833 switch (code) {
2834 case lir_static_call:
2835 case lir_optvirtual_call:
2836 case lir_dynamic_call:
2837 offset += NativeCall::displacement_offset;
2838 break;
2839 case lir_icvirtual_call:
2840 offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
2841 break;
2842 case lir_virtual_call: // currently, sparc-specific for niagara
2843 default: ShouldNotReachHere();
2844 }
2845 while (offset++ % BytesPerWord != 0) {
2846 __ nop();
2847 }
2848 }
2849 }
2850
2851
2852 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
2853 assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2854 "must be aligned");
2855 __ call(AddressLiteral(op->addr(), rtype));
2856 add_call_info(code_offset(), op->info());
2857 }
2858
2859
2860 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
2861 RelocationHolder rh = virtual_call_Relocation::spec(pc());
2862 __ movoop(IC_Klass, (jobject)Universe::non_oop_word());
2863 assert(!os::is_MP() ||
2864 (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2865 "must be aligned");
2866 __ call(AddressLiteral(op->addr(), rh));
2867 add_call_info(code_offset(), op->info());
2868 }
2869
2870
2871 /* Currently, vtable-dispatch is only enabled for sparc platforms */
2872 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
2873 ShouldNotReachHere();
2874 }
2875
2876
2877 void LIR_Assembler::emit_static_call_stub() {
2878 address call_pc = __ pc();
2879 address stub = __ start_a_stub(call_stub_size);
2880 if (stub == NULL) {
2881 bailout("static call stub overflow");
2882 return;
2883 }
2884
2885 int start = __ offset();
2886 if (os::is_MP()) {
2887 // make sure that the displacement word of the call ends up word aligned
2888 int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
2889 while (offset++ % BytesPerWord != 0) {
2890 __ nop();
2891 }
2892 }
2893 __ relocate(static_stub_Relocation::spec(call_pc));
2894 __ movoop(rbx, (jobject)NULL);
2895 // must be set to -1 at code generation time
2896 assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
2897 // On 64bit this will die since it will take a movq & jmp, must be only a jmp
2898 __ jump(RuntimeAddress(__ pc()));
2899
2900 assert(__ offset() - start <= call_stub_size, "stub too big");
2901 __ end_a_stub();
2902 }
2903
2904
2905 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2906 assert(exceptionOop->as_register() == rax, "must match");
2907 assert(exceptionPC->as_register() == rdx, "must match");
2908
2909 // exception object is not added to oop map by LinearScan
2910 // (LinearScan assumes that no oops are in fixed registers)
2911 info->add_register_oop(exceptionOop);
2912 Runtime1::StubID unwind_id;
2913
2914 // get current pc information
2915 // pc is only needed if the method has an exception handler, the unwind code does not need it.
2916 int pc_for_athrow_offset = __ offset();
2917 InternalAddress pc_for_athrow(__ pc());
2918 __ lea(exceptionPC->as_register(), pc_for_athrow);
2919 add_call_info(pc_for_athrow_offset, info); // for exception handler
2920
2921 __ verify_not_null_oop(rax);
2922 // search an exception handler (rax: exception oop, rdx: throwing pc)
2923 if (compilation()->has_fpu_code()) {
2924 unwind_id = Runtime1::handle_exception_id;
2925 } else {
2926 unwind_id = Runtime1::handle_exception_nofpu_id;
2927 }
2928 __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2929
2930 // enough room for two byte trap
2931 __ nop();
2932 }
2933
2934
2935 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2936 assert(exceptionOop->as_register() == rax, "must match");
2937
2938 __ jmp(_unwind_handler_entry);
2939 }
2940
2941
2942 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2943
2944 // optimized version for linear scan:
2945 // * count must be already in ECX (guaranteed by LinearScan)
2946 // * left and dest must be equal
2947 // * tmp must be unused
2948 assert(count->as_register() == SHIFT_count, "count must be in ECX");
2949 assert(left == dest, "left and dest must be equal");
2950 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2951
2952 if (left->is_single_cpu()) {
2953 Register value = left->as_register();
2954 assert(value != SHIFT_count, "left cannot be ECX");
2955
2956 switch (code) {
2957 case lir_shl: __ shll(value); break;
2958 case lir_shr: __ sarl(value); break;
2959 case lir_ushr: __ shrl(value); break;
2960 default: ShouldNotReachHere();
2961 }
2962 } else if (left->is_double_cpu()) {
2963 Register lo = left->as_register_lo();
2964 Register hi = left->as_register_hi();
2965 assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
2966 #ifdef _LP64
2967 switch (code) {
2968 case lir_shl: __ shlptr(lo); break;
2969 case lir_shr: __ sarptr(lo); break;
2970 case lir_ushr: __ shrptr(lo); break;
2971 default: ShouldNotReachHere();
2972 }
2973 #else
2974
2975 switch (code) {
2976 case lir_shl: __ lshl(hi, lo); break;
2977 case lir_shr: __ lshr(hi, lo, true); break;
2978 case lir_ushr: __ lshr(hi, lo, false); break;
2979 default: ShouldNotReachHere();
2980 }
2981 #endif // LP64
2982 } else {
2983 ShouldNotReachHere();
2984 }
2985 }
2986
2987
2988 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2989 if (dest->is_single_cpu()) {
2990 // first move left into dest so that left is not destroyed by the shift
2991 Register value = dest->as_register();
2992 count = count & 0x1F; // Java spec
2993
2994 move_regs(left->as_register(), value);
2995 switch (code) {
2996 case lir_shl: __ shll(value, count); break;
2997 case lir_shr: __ sarl(value, count); break;
2998 case lir_ushr: __ shrl(value, count); break;
2999 default: ShouldNotReachHere();
3000 }
3001 } else if (dest->is_double_cpu()) {
3002 #ifndef _LP64
3003 Unimplemented();
3004 #else
3005 // first move left into dest so that left is not destroyed by the shift
3006 Register value = dest->as_register_lo();
3007 count = count & 0x1F; // Java spec
3008
3009 move_regs(left->as_register_lo(), value);
3010 switch (code) {
3011 case lir_shl: __ shlptr(value, count); break;
3012 case lir_shr: __ sarptr(value, count); break;
3013 case lir_ushr: __ shrptr(value, count); break;
3014 default: ShouldNotReachHere();
3015 }
3016 #endif // _LP64
3017 } else {
3018 ShouldNotReachHere();
3019 }
3020 }
3021
3022
3023 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
3024 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3025 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3026 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3027 __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
3028 }
3029
3030
3031 void LIR_Assembler::store_parameter(jint c, int offset_from_rsp_in_words) {
3032 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3033 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3034 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3035 __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
3036 }
3037
3038
3039 void LIR_Assembler::store_parameter(jobject o, int offset_from_rsp_in_words) {
3040 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3041 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3042 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3043 __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
3044 }
3045
3046
3047 // This code replaces a call to arraycopy; no exception may
3048 // be thrown in this code, they must be thrown in the System.arraycopy
3049 // activation frame; we could save some checks if this would not be the case
3050 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
3051 ciArrayKlass* default_type = op->expected_type();
3052 Register src = op->src()->as_register();
3053 Register dst = op->dst()->as_register();
3054 Register src_pos = op->src_pos()->as_register();
3055 Register dst_pos = op->dst_pos()->as_register();
3056 Register length = op->length()->as_register();
3057 Register tmp = op->tmp()->as_register();
3058
3059 CodeStub* stub = op->stub();
3060 int flags = op->flags();
3061 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
3062 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
3063
3064 // if we don't know anything, just go through the generic arraycopy
3065 if (default_type == NULL) {
3066 Label done;
3067 // save outgoing arguments on stack in case call to System.arraycopy is needed
3068 // HACK ALERT. This code used to push the parameters in a hardwired fashion
3069 // for interpreter calling conventions. Now we have to do it in new style conventions.
3070 // For the moment until C1 gets the new register allocator I just force all the
3071 // args to the right place (except the register args) and then on the back side
3072 // reload the register args properly if we go slow path. Yuck
3073
3074 // These are proper for the calling convention
3075 store_parameter(length, 2);
3076 store_parameter(dst_pos, 1);
3077 store_parameter(dst, 0);
3078
3079 // these are just temporary placements until we need to reload
3080 store_parameter(src_pos, 3);
3081 store_parameter(src, 4);
3082 NOT_LP64(assert(src == rcx && src_pos == rdx, "mismatch in calling convention");)
3083
3084 address C_entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
3085
3086 address copyfunc_addr = StubRoutines::generic_arraycopy();
3087
3088 // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
3089 #ifdef _LP64
3090 // The arguments are in java calling convention so we can trivially shift them to C
3091 // convention
3092 assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
3093 __ mov(c_rarg0, j_rarg0);
3094 assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
3095 __ mov(c_rarg1, j_rarg1);
3096 assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
3097 __ mov(c_rarg2, j_rarg2);
3098 assert_different_registers(c_rarg3, j_rarg4);
3099 __ mov(c_rarg3, j_rarg3);
3100 #ifdef _WIN64
3101 // Allocate abi space for args but be sure to keep stack aligned
3102 __ subptr(rsp, 6*wordSize);
3103 store_parameter(j_rarg4, 4);
3104 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3105 __ call(RuntimeAddress(C_entry));
3106 } else {
3107 #ifndef PRODUCT
3108 if (PrintC1Statistics) {
3109 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3110 }
3111 #endif
3112 __ call(RuntimeAddress(copyfunc_addr));
3113 }
3114 __ addptr(rsp, 6*wordSize);
3115 #else
3116 __ mov(c_rarg4, j_rarg4);
3117 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3118 __ call(RuntimeAddress(C_entry));
3119 } else {
3120 #ifndef PRODUCT
3121 if (PrintC1Statistics) {
3122 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3123 }
3124 #endif
3125 __ call(RuntimeAddress(copyfunc_addr));
3126 }
3127 #endif // _WIN64
3128 #else
3129 __ push(length);
3130 __ push(dst_pos);
3131 __ push(dst);
3132 __ push(src_pos);
3133 __ push(src);
3134
3135 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3136 __ call_VM_leaf(C_entry, 5); // removes pushed parameter from the stack
3137 } else {
3138 #ifndef PRODUCT
3139 if (PrintC1Statistics) {
3140 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3141 }
3142 #endif
3143 __ call_VM_leaf(copyfunc_addr, 5); // removes pushed parameter from the stack
3144 }
3145
3146 #endif // _LP64
3147
3148 __ cmpl(rax, 0);
3149 __ jcc(Assembler::equal, *stub->continuation());
3150
3151 if (copyfunc_addr != NULL) {
3152 __ mov(tmp, rax);
3153 __ xorl(tmp, -1);
3154 }
3155
3156 // Reload values from the stack so they are where the stub
3157 // expects them.
3158 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3159 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3160 __ movptr (length, Address(rsp, 2*BytesPerWord));
3161 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3162 __ movptr (src, Address(rsp, 4*BytesPerWord));
3163
3164 if (copyfunc_addr != NULL) {
3165 __ subl(length, tmp);
3166 __ addl(src_pos, tmp);
3167 __ addl(dst_pos, tmp);
3168 }
3169 __ jmp(*stub->entry());
3170
3171 __ bind(*stub->continuation());
3172 return;
3173 }
3174
3175 assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
3176
3177 int elem_size = type2aelembytes(basic_type);
3178 int shift_amount;
3179 Address::ScaleFactor scale;
3180
3181 switch (elem_size) {
3182 case 1 :
3183 shift_amount = 0;
3184 scale = Address::times_1;
3185 break;
3186 case 2 :
3187 shift_amount = 1;
3188 scale = Address::times_2;
3189 break;
3190 case 4 :
3191 shift_amount = 2;
3192 scale = Address::times_4;
3193 break;
3194 case 8 :
3195 shift_amount = 3;
3196 scale = Address::times_8;
3197 break;
3198 default:
3199 ShouldNotReachHere();
3200 }
3201
3202 Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
3203 Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
3204 Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
3205 Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
3206
3207 // length and pos's are all sign extended at this point on 64bit
3208
3209 // test for NULL
3210 if (flags & LIR_OpArrayCopy::src_null_check) {
3211 __ testptr(src, src);
3212 __ jcc(Assembler::zero, *stub->entry());
3213 }
3214 if (flags & LIR_OpArrayCopy::dst_null_check) {
3215 __ testptr(dst, dst);
3216 __ jcc(Assembler::zero, *stub->entry());
3217 }
3218
3219 // check if negative
3220 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
3221 __ testl(src_pos, src_pos);
3222 __ jcc(Assembler::less, *stub->entry());
3223 }
3224 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
3225 __ testl(dst_pos, dst_pos);
3226 __ jcc(Assembler::less, *stub->entry());
3227 }
3228
3229 if (flags & LIR_OpArrayCopy::src_range_check) {
3230 __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
3231 __ cmpl(tmp, src_length_addr);
3232 __ jcc(Assembler::above, *stub->entry());
3233 }
3234 if (flags & LIR_OpArrayCopy::dst_range_check) {
3235 __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
3236 __ cmpl(tmp, dst_length_addr);
3237 __ jcc(Assembler::above, *stub->entry());
3238 }
3239
3240 if (flags & LIR_OpArrayCopy::length_positive_check) {
3241 __ testl(length, length);
3242 __ jcc(Assembler::less, *stub->entry());
3243 __ jcc(Assembler::zero, *stub->continuation());
3244 }
3245
3246 #ifdef _LP64
3247 __ movl2ptr(src_pos, src_pos); //higher 32bits must be null
3248 __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null
3249 #endif
3250
3251 if (flags & LIR_OpArrayCopy::type_check) {
3252 // We don't know the array types are compatible
3253 if (basic_type != T_OBJECT) {
3254 // Simple test for basic type arrays
3255 if (UseCompressedOops) {
3256 __ movl(tmp, src_klass_addr);
3257 __ cmpl(tmp, dst_klass_addr);
3258 } else {
3259 __ movptr(tmp, src_klass_addr);
3260 __ cmpptr(tmp, dst_klass_addr);
3261 }
3262 __ jcc(Assembler::notEqual, *stub->entry());
3263 } else {
3264 // For object arrays, if src is a sub class of dst then we can
3265 // safely do the copy.
3266 Label cont, slow;
3267
3268 __ push(src);
3269 __ push(dst);
3270
3271 __ load_klass(src, src);
3272 __ load_klass(dst, dst);
3273
3274 __ check_klass_subtype_fast_path(src, dst, tmp, &cont, &slow, NULL);
3275
3276 __ push(src);
3277 __ push(dst);
3278 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
3279 __ pop(dst);
3280 __ pop(src);
3281
3282 __ cmpl(src, 0);
3283 __ jcc(Assembler::notEqual, cont);
3284
3285 __ bind(slow);
3286 __ pop(dst);
3287 __ pop(src);
3288
3289 address copyfunc_addr = StubRoutines::checkcast_arraycopy();
3290 if (copyfunc_addr != NULL) { // use stub if available
3291 // src is not a sub class of dst so we have to do a
3292 // per-element check.
3293
3294 int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
3295 if ((flags & mask) != mask) {
3296 // Check that at least both of them object arrays.
3297 assert(flags & mask, "one of the two should be known to be an object array");
3298
3299 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
3300 __ load_klass(tmp, src);
3301 } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
3302 __ load_klass(tmp, dst);
3303 }
3304 int lh_offset = in_bytes(Klass::layout_helper_offset());
3305 Address klass_lh_addr(tmp, lh_offset);
3306 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
3307 __ cmpl(klass_lh_addr, objArray_lh);
3308 __ jcc(Assembler::notEqual, *stub->entry());
3309 }
3310
3311 // Spill because stubs can use any register they like and it's
3312 // easier to restore just those that we care about.
3313 store_parameter(dst, 0);
3314 store_parameter(dst_pos, 1);
3315 store_parameter(length, 2);
3316 store_parameter(src_pos, 3);
3317 store_parameter(src, 4);
3318
3319 #ifndef _LP64
3320 __ movptr(tmp, dst_klass_addr);
3321 __ movptr(tmp, Address(tmp, objArrayKlass::element_klass_offset()));
3322 __ push(tmp);
3323 __ movl(tmp, Address(tmp, Klass::super_check_offset_offset()));
3324 __ push(tmp);
3325 __ push(length);
3326 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3327 __ push(tmp);
3328 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3329 __ push(tmp);
3330
3331 __ call_VM_leaf(copyfunc_addr, 5);
3332 #else
3333 __ movl2ptr(length, length); //higher 32bits must be null
3334
3335 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3336 assert_different_registers(c_rarg0, dst, dst_pos, length);
3337 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3338 assert_different_registers(c_rarg1, dst, length);
3339
3340 __ mov(c_rarg2, length);
3341 assert_different_registers(c_rarg2, dst);
3342
3343 #ifdef _WIN64
3344 // Allocate abi space for args but be sure to keep stack aligned
3345 __ subptr(rsp, 6*wordSize);
3346 __ load_klass(c_rarg3, dst);
3347 __ movptr(c_rarg3, Address(c_rarg3, objArrayKlass::element_klass_offset()));
3348 store_parameter(c_rarg3, 4);
3349 __ movl(c_rarg3, Address(c_rarg3, Klass::super_check_offset_offset()));
3350 __ call(RuntimeAddress(copyfunc_addr));
3351 __ addptr(rsp, 6*wordSize);
3352 #else
3353 __ load_klass(c_rarg4, dst);
3354 __ movptr(c_rarg4, Address(c_rarg4, objArrayKlass::element_klass_offset()));
3355 __ movl(c_rarg3, Address(c_rarg4, Klass::super_check_offset_offset()));
3356 __ call(RuntimeAddress(copyfunc_addr));
3357 #endif
3358
3359 #endif
3360
3361 #ifndef PRODUCT
3362 if (PrintC1Statistics) {
3363 Label failed;
3364 __ testl(rax, rax);
3365 __ jcc(Assembler::notZero, failed);
3366 __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_cnt));
3367 __ bind(failed);
3368 }
3369 #endif
3370
3371 __ testl(rax, rax);
3372 __ jcc(Assembler::zero, *stub->continuation());
3373
3374 #ifndef PRODUCT
3375 if (PrintC1Statistics) {
3376 __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_attempt_cnt));
3377 }
3378 #endif
3379
3380 __ mov(tmp, rax);
3381
3382 __ xorl(tmp, -1);
3383
3384 // Restore previously spilled arguments
3385 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3386 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3387 __ movptr (length, Address(rsp, 2*BytesPerWord));
3388 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3389 __ movptr (src, Address(rsp, 4*BytesPerWord));
3390
3391
3392 __ subl(length, tmp);
3393 __ addl(src_pos, tmp);
3394 __ addl(dst_pos, tmp);
3395 }
3396
3397 __ jmp(*stub->entry());
3398
3399 __ bind(cont);
3400 __ pop(dst);
3401 __ pop(src);
3402 }
3403 }
3404
3405 #ifdef ASSERT
3406 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
3407 // Sanity check the known type with the incoming class. For the
3408 // primitive case the types must match exactly with src.klass and
3409 // dst.klass each exactly matching the default type. For the
3410 // object array case, if no type check is needed then either the
3411 // dst type is exactly the expected type and the src type is a
3412 // subtype which we can't check or src is the same array as dst
3413 // but not necessarily exactly of type default_type.
3414 Label known_ok, halt;
3415 __ movoop(tmp, default_type->constant_encoding());
3416 #ifdef _LP64
3417 if (UseCompressedOops) {
3418 __ encode_heap_oop(tmp);
3419 }
3420 #endif
3421
3422 if (basic_type != T_OBJECT) {
3423
3424 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3425 else __ cmpptr(tmp, dst_klass_addr);
3426 __ jcc(Assembler::notEqual, halt);
3427 if (UseCompressedOops) __ cmpl(tmp, src_klass_addr);
3428 else __ cmpptr(tmp, src_klass_addr);
3429 __ jcc(Assembler::equal, known_ok);
3430 } else {
3431 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3432 else __ cmpptr(tmp, dst_klass_addr);
3433 __ jcc(Assembler::equal, known_ok);
3434 __ cmpptr(src, dst);
3435 __ jcc(Assembler::equal, known_ok);
3436 }
3437 __ bind(halt);
3438 __ stop("incorrect type information in arraycopy");
3439 __ bind(known_ok);
3440 }
3441 #endif
3442
3443 #ifndef PRODUCT
3444 if (PrintC1Statistics) {
3445 __ incrementl(ExternalAddress(Runtime1::arraycopy_count_address(basic_type)));
3446 }
3447 #endif
3448
3449 #ifdef _LP64
3450 assert_different_registers(c_rarg0, dst, dst_pos, length);
3451 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3452 assert_different_registers(c_rarg1, length);
3453 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3454 __ mov(c_rarg2, length);
3455
3456 #else
3457 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3458 store_parameter(tmp, 0);
3459 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3460 store_parameter(tmp, 1);
3461 store_parameter(length, 2);
3462 #endif // _LP64
3463
3464 bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
3465 bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
3466 const char *name;
3467 address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
3468 __ call_VM_leaf(entry, 0);
3469
3470 __ bind(*stub->continuation());
3471 }
3472
3473
3474 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
3475 Register obj = op->obj_opr()->as_register(); // may not be an oop
3476 Register hdr = op->hdr_opr()->as_register();
3477 Register lock = op->lock_opr()->as_register();
3478 if (!UseFastLocking) {
3479 __ jmp(*op->stub()->entry());
3480 } else if (op->code() == lir_lock) {
3481 Register scratch = noreg;
3482 if (UseBiasedLocking) {
3483 scratch = op->scratch_opr()->as_register();
3484 }
3485 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3486 // add debug info for NullPointerException only if one is possible
3487 int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
3488 if (op->info() != NULL) {
3489 add_debug_info_for_null_check(null_check_offset, op->info());
3490 }
3491 // done
3492 } else if (op->code() == lir_unlock) {
3493 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3494 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
3495 } else {
3496 Unimplemented();
3497 }
3498 __ bind(*op->stub()->continuation());
3499 }
3500
3501
3502 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
3503 ciMethod* method = op->profiled_method();
3504 int bci = op->profiled_bci();
3505
3506 // Update counter for all call types
3507 ciMethodData* md = method->method_data_or_null();
3508 assert(md != NULL, "Sanity");
3509 ciProfileData* data = md->bci_to_data(bci);
3510 assert(data->is_CounterData(), "need CounterData for calls");
3511 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
3512 Register mdo = op->mdo()->as_register();
3513 __ movoop(mdo, md->constant_encoding());
3514 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
3515 Bytecodes::Code bc = method->java_code_at_bci(bci);
3516 // Perform additional virtual call profiling for invokevirtual and
3517 // invokeinterface bytecodes
3518 if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3519 C1ProfileVirtualCalls) {
3520 assert(op->recv()->is_single_cpu(), "recv must be allocated");
3521 Register recv = op->recv()->as_register();
3522 assert_different_registers(mdo, recv);
3523 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3524 ciKlass* known_klass = op->known_holder();
3525 if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
3526 // We know the type that will be seen at this call site; we can
3527 // statically update the methodDataOop rather than needing to do
3528 // dynamic tests on the receiver type
3529
3530 // NOTE: we should probably put a lock around this search to
3531 // avoid collisions by concurrent compilations
3532 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3533 uint i;
3534 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3535 ciKlass* receiver = vc_data->receiver(i);
3536 if (known_klass->equals(receiver)) {
3537 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3538 __ addptr(data_addr, DataLayout::counter_increment);
3539 return;
3540 }
3541 }
3542
3543 // Receiver type not found in profile data; select an empty slot
3544
3545 // Note that this is less efficient than it should be because it
3546 // always does a write to the receiver part of the
3547 // VirtualCallData rather than just the first time
3548 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3549 ciKlass* receiver = vc_data->receiver(i);
3550 if (receiver == NULL) {
3551 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3552 __ movoop(recv_addr, known_klass->constant_encoding());
3553 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3554 __ addptr(data_addr, DataLayout::counter_increment);
3555 return;
3556 }
3557 }
3558 } else {
3559 __ load_klass(recv, recv);
3560 Label update_done;
3561 type_profile_helper(mdo, md, data, recv, &update_done);
3562 // Receiver did not match any saved receiver and there is no empty row for it.
3563 // Increment total counter to indicate polymorphic case.
3564 __ addptr(counter_addr, DataLayout::counter_increment);
3565
3566 __ bind(update_done);
3567 }
3568 } else {
3569 // Static call
3570 __ addptr(counter_addr, DataLayout::counter_increment);
3571 }
3572 }
3573
3574 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
3575 Unimplemented();
3576 }
3577
3578
3579 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
3580 __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
3581 }
3582
3583
3584 void LIR_Assembler::align_backward_branch_target() {
3585 __ align(BytesPerWord);
3586 }
3587
3588
3589 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3590 if (left->is_single_cpu()) {
3591 __ negl(left->as_register());
3592 move_regs(left->as_register(), dest->as_register());
3593
3594 } else if (left->is_double_cpu()) {
3595 Register lo = left->as_register_lo();
3596 #ifdef _LP64
3597 Register dst = dest->as_register_lo();
3598 __ movptr(dst, lo);
3599 __ negptr(dst);
3600 #else
3601 Register hi = left->as_register_hi();
3602 __ lneg(hi, lo);
3603 if (dest->as_register_lo() == hi) {
3604 assert(dest->as_register_hi() != lo, "destroying register");
3605 move_regs(hi, dest->as_register_hi());
3606 move_regs(lo, dest->as_register_lo());
3607 } else {
3608 move_regs(lo, dest->as_register_lo());
3609 move_regs(hi, dest->as_register_hi());
3610 }
3611 #endif // _LP64
3612
3613 } else if (dest->is_single_xmm()) {
3614 if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
3615 __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
3616 }
3617 __ xorps(dest->as_xmm_float_reg(),
3618 ExternalAddress((address)float_signflip_pool));
3619
3620 } else if (dest->is_double_xmm()) {
3621 if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
3622 __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
3623 }
3624 __ xorpd(dest->as_xmm_double_reg(),
3625 ExternalAddress((address)double_signflip_pool));
3626
3627 } else if (left->is_single_fpu() || left->is_double_fpu()) {
3628 assert(left->fpu() == 0, "arg must be on TOS");
3629 assert(dest->fpu() == 0, "dest must be TOS");
3630 __ fchs();
3631
3632 } else {
3633 ShouldNotReachHere();
3634 }
3635 }
3636
3637
3638 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
3639 assert(addr->is_address() && dest->is_register(), "check");
3640 Register reg;
3641 reg = dest->as_pointer_register();
3642 __ lea(reg, as_Address(addr->as_address_ptr()));
3643 }
3644
3645
3646
3647 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3648 assert(!tmp->is_valid(), "don't need temporary");
3649 __ call(RuntimeAddress(dest));
3650 if (info != NULL) {
3651 add_call_info_here(info);
3652 }
3653 }
3654
3655
3656 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3657 assert(type == T_LONG, "only for volatile long fields");
3658
3659 if (info != NULL) {
3660 add_debug_info_for_null_check_here(info);
3661 }
3662
3663 if (src->is_double_xmm()) {
3664 if (dest->is_double_cpu()) {
3665 #ifdef _LP64
3666 __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
3667 #else
3668 __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
3669 __ psrlq(src->as_xmm_double_reg(), 32);
3670 __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
3671 #endif // _LP64
3672 } else if (dest->is_double_stack()) {
3673 __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
3674 } else if (dest->is_address()) {
3675 __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
3676 } else {
3677 ShouldNotReachHere();
3678 }
3679
3680 } else if (dest->is_double_xmm()) {
3681 if (src->is_double_stack()) {
3682 __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
3683 } else if (src->is_address()) {
3684 __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
3685 } else {
3686 ShouldNotReachHere();
3687 }
3688
3689 } else if (src->is_double_fpu()) {
3690 assert(src->fpu_regnrLo() == 0, "must be TOS");
3691 if (dest->is_double_stack()) {
3692 __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
3693 } else if (dest->is_address()) {
3694 __ fistp_d(as_Address(dest->as_address_ptr()));
3695 } else {
3696 ShouldNotReachHere();
3697 }
3698
3699 } else if (dest->is_double_fpu()) {
3700 assert(dest->fpu_regnrLo() == 0, "must be TOS");
3701 if (src->is_double_stack()) {
3702 __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
3703 } else if (src->is_address()) {
3704 __ fild_d(as_Address(src->as_address_ptr()));
3705 } else {
3706 ShouldNotReachHere();
3707 }
3708 } else {
3709 ShouldNotReachHere();
3710 }
3711 }
3712
3713
3714 void LIR_Assembler::membar() {
3715 // QQQ sparc TSO uses this,
3716 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
3717 }
3718
3719 void LIR_Assembler::membar_acquire() {
3720 // No x86 machines currently require load fences
3721 // __ load_fence();
3722 }
3723
3724 void LIR_Assembler::membar_release() {
3725 // No x86 machines currently require store fences
3726 // __ store_fence();
3727 }
3728
3729 void LIR_Assembler::membar_loadload() {
3730 // no-op
3731 //__ membar(Assembler::Membar_mask_bits(Assembler::loadload));
3732 }
3733
3734 void LIR_Assembler::membar_storestore() {
3735 // no-op
3736 //__ membar(Assembler::Membar_mask_bits(Assembler::storestore));
3737 }
3738
3739 void LIR_Assembler::membar_loadstore() {
3740 // no-op
3741 //__ membar(Assembler::Membar_mask_bits(Assembler::loadstore));
3742 }
3743
3744 void LIR_Assembler::membar_storeload() {
3745 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
3746 }
3747
3748 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3749 assert(result_reg->is_register(), "check");
3750 #ifdef _LP64
3751 // __ get_thread(result_reg->as_register_lo());
3752 __ mov(result_reg->as_register(), r15_thread);
3753 #else
3754 __ get_thread(result_reg->as_register());
3755 #endif // _LP64
3756 }
3757
3758
3759 void LIR_Assembler::peephole(LIR_List*) {
3760 // do nothing for now
3761 }
3762
3763
3764 #undef __