Wed, 16 Feb 2011 13:30:31 -0800
7013964: openjdk LICENSE file needs rebranding
Reviewed-by: darcy, katleman, jjg
1 /*
2 * Copyright (c) 2000, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "c1/c1_Compilation.hpp"
27 #include "c1/c1_LIRAssembler.hpp"
28 #include "c1/c1_MacroAssembler.hpp"
29 #include "c1/c1_Runtime1.hpp"
30 #include "c1/c1_ValueStack.hpp"
31 #include "ci/ciArrayKlass.hpp"
32 #include "ci/ciInstance.hpp"
33 #include "gc_interface/collectedHeap.hpp"
34 #include "memory/barrierSet.hpp"
35 #include "memory/cardTableModRefBS.hpp"
36 #include "nativeInst_x86.hpp"
37 #include "oops/objArrayKlass.hpp"
38 #include "runtime/sharedRuntime.hpp"
41 // These masks are used to provide 128-bit aligned bitmasks to the XMM
42 // instructions, to allow sign-masking or sign-bit flipping. They allow
43 // fast versions of NegF/NegD and AbsF/AbsD.
45 // Note: 'double' and 'long long' have 32-bits alignment on x86.
46 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
47 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
48 // of 128-bits operands for SSE instructions.
49 jlong *operand = (jlong*)(((long)adr)&((long)(~0xF)));
50 // Store the value to a 128-bits operand.
51 operand[0] = lo;
52 operand[1] = hi;
53 return operand;
54 }
56 // Buffer for 128-bits masks used by SSE instructions.
57 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
59 // Static initialization during VM startup.
60 static jlong *float_signmask_pool = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
61 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
62 static jlong *float_signflip_pool = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
63 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
67 NEEDS_CLEANUP // remove this definitions ?
68 const Register IC_Klass = rax; // where the IC klass is cached
69 const Register SYNC_header = rax; // synchronization header
70 const Register SHIFT_count = rcx; // where count for shift operations must be
72 #define __ _masm->
75 static void select_different_registers(Register preserve,
76 Register extra,
77 Register &tmp1,
78 Register &tmp2) {
79 if (tmp1 == preserve) {
80 assert_different_registers(tmp1, tmp2, extra);
81 tmp1 = extra;
82 } else if (tmp2 == preserve) {
83 assert_different_registers(tmp1, tmp2, extra);
84 tmp2 = extra;
85 }
86 assert_different_registers(preserve, tmp1, tmp2);
87 }
91 static void select_different_registers(Register preserve,
92 Register extra,
93 Register &tmp1,
94 Register &tmp2,
95 Register &tmp3) {
96 if (tmp1 == preserve) {
97 assert_different_registers(tmp1, tmp2, tmp3, extra);
98 tmp1 = extra;
99 } else if (tmp2 == preserve) {
100 assert_different_registers(tmp1, tmp2, tmp3, extra);
101 tmp2 = extra;
102 } else if (tmp3 == preserve) {
103 assert_different_registers(tmp1, tmp2, tmp3, extra);
104 tmp3 = extra;
105 }
106 assert_different_registers(preserve, tmp1, tmp2, tmp3);
107 }
111 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
112 if (opr->is_constant()) {
113 LIR_Const* constant = opr->as_constant_ptr();
114 switch (constant->type()) {
115 case T_INT: {
116 return true;
117 }
119 default:
120 return false;
121 }
122 }
123 return false;
124 }
127 LIR_Opr LIR_Assembler::receiverOpr() {
128 return FrameMap::receiver_opr;
129 }
131 LIR_Opr LIR_Assembler::incomingReceiverOpr() {
132 return receiverOpr();
133 }
135 LIR_Opr LIR_Assembler::osrBufferPointer() {
136 return FrameMap::as_pointer_opr(receiverOpr()->as_register());
137 }
139 //--------------fpu register translations-----------------------
142 address LIR_Assembler::float_constant(float f) {
143 address const_addr = __ float_constant(f);
144 if (const_addr == NULL) {
145 bailout("const section overflow");
146 return __ code()->consts()->start();
147 } else {
148 return const_addr;
149 }
150 }
153 address LIR_Assembler::double_constant(double d) {
154 address const_addr = __ double_constant(d);
155 if (const_addr == NULL) {
156 bailout("const section overflow");
157 return __ code()->consts()->start();
158 } else {
159 return const_addr;
160 }
161 }
164 void LIR_Assembler::set_24bit_FPU() {
165 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
166 }
168 void LIR_Assembler::reset_FPU() {
169 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
170 }
172 void LIR_Assembler::fpop() {
173 __ fpop();
174 }
176 void LIR_Assembler::fxch(int i) {
177 __ fxch(i);
178 }
180 void LIR_Assembler::fld(int i) {
181 __ fld_s(i);
182 }
184 void LIR_Assembler::ffree(int i) {
185 __ ffree(i);
186 }
188 void LIR_Assembler::breakpoint() {
189 __ int3();
190 }
192 void LIR_Assembler::push(LIR_Opr opr) {
193 if (opr->is_single_cpu()) {
194 __ push_reg(opr->as_register());
195 } else if (opr->is_double_cpu()) {
196 NOT_LP64(__ push_reg(opr->as_register_hi()));
197 __ push_reg(opr->as_register_lo());
198 } else if (opr->is_stack()) {
199 __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
200 } else if (opr->is_constant()) {
201 LIR_Const* const_opr = opr->as_constant_ptr();
202 if (const_opr->type() == T_OBJECT) {
203 __ push_oop(const_opr->as_jobject());
204 } else if (const_opr->type() == T_INT) {
205 __ push_jint(const_opr->as_jint());
206 } else {
207 ShouldNotReachHere();
208 }
210 } else {
211 ShouldNotReachHere();
212 }
213 }
215 void LIR_Assembler::pop(LIR_Opr opr) {
216 if (opr->is_single_cpu()) {
217 __ pop_reg(opr->as_register());
218 } else {
219 ShouldNotReachHere();
220 }
221 }
223 bool LIR_Assembler::is_literal_address(LIR_Address* addr) {
224 return addr->base()->is_illegal() && addr->index()->is_illegal();
225 }
227 //-------------------------------------------
229 Address LIR_Assembler::as_Address(LIR_Address* addr) {
230 return as_Address(addr, rscratch1);
231 }
233 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
234 if (addr->base()->is_illegal()) {
235 assert(addr->index()->is_illegal(), "must be illegal too");
236 AddressLiteral laddr((address)addr->disp(), relocInfo::none);
237 if (! __ reachable(laddr)) {
238 __ movptr(tmp, laddr.addr());
239 Address res(tmp, 0);
240 return res;
241 } else {
242 return __ as_Address(laddr);
243 }
244 }
246 Register base = addr->base()->as_pointer_register();
248 if (addr->index()->is_illegal()) {
249 return Address( base, addr->disp());
250 } else if (addr->index()->is_cpu_register()) {
251 Register index = addr->index()->as_pointer_register();
252 return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
253 } else if (addr->index()->is_constant()) {
254 intptr_t addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
255 assert(Assembler::is_simm32(addr_offset), "must be");
257 return Address(base, addr_offset);
258 } else {
259 Unimplemented();
260 return Address();
261 }
262 }
265 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
266 Address base = as_Address(addr);
267 return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
268 }
271 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
272 return as_Address(addr);
273 }
276 void LIR_Assembler::osr_entry() {
277 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
278 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
279 ValueStack* entry_state = osr_entry->state();
280 int number_of_locks = entry_state->locks_size();
282 // we jump here if osr happens with the interpreter
283 // state set up to continue at the beginning of the
284 // loop that triggered osr - in particular, we have
285 // the following registers setup:
286 //
287 // rcx: osr buffer
288 //
290 // build frame
291 ciMethod* m = compilation()->method();
292 __ build_frame(initial_frame_size_in_bytes());
294 // OSR buffer is
295 //
296 // locals[nlocals-1..0]
297 // monitors[0..number_of_locks]
298 //
299 // locals is a direct copy of the interpreter frame so in the osr buffer
300 // so first slot in the local array is the last local from the interpreter
301 // and last slot is local[0] (receiver) from the interpreter
302 //
303 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
304 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
305 // in the interpreter frame (the method lock if a sync method)
307 // Initialize monitors in the compiled activation.
308 // rcx: pointer to osr buffer
309 //
310 // All other registers are dead at this point and the locals will be
311 // copied into place by code emitted in the IR.
313 Register OSR_buf = osrBufferPointer()->as_pointer_register();
314 { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
315 int monitor_offset = BytesPerWord * method()->max_locals() +
316 (2 * BytesPerWord) * (number_of_locks - 1);
317 // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
318 // the OSR buffer using 2 word entries: first the lock and then
319 // the oop.
320 for (int i = 0; i < number_of_locks; i++) {
321 int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
322 #ifdef ASSERT
323 // verify the interpreter's monitor has a non-null object
324 {
325 Label L;
326 __ cmpptr(Address(OSR_buf, slot_offset + 1*BytesPerWord), (int32_t)NULL_WORD);
327 __ jcc(Assembler::notZero, L);
328 __ stop("locked object is NULL");
329 __ bind(L);
330 }
331 #endif
332 __ movptr(rbx, Address(OSR_buf, slot_offset + 0));
333 __ movptr(frame_map()->address_for_monitor_lock(i), rbx);
334 __ movptr(rbx, Address(OSR_buf, slot_offset + 1*BytesPerWord));
335 __ movptr(frame_map()->address_for_monitor_object(i), rbx);
336 }
337 }
338 }
341 // inline cache check; done before the frame is built.
342 int LIR_Assembler::check_icache() {
343 Register receiver = FrameMap::receiver_opr->as_register();
344 Register ic_klass = IC_Klass;
345 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
346 const bool do_post_padding = VerifyOops || UseCompressedOops;
347 if (!do_post_padding) {
348 // insert some nops so that the verified entry point is aligned on CodeEntryAlignment
349 while ((__ offset() + ic_cmp_size) % CodeEntryAlignment != 0) {
350 __ nop();
351 }
352 }
353 int offset = __ offset();
354 __ inline_cache_check(receiver, IC_Klass);
355 assert(__ offset() % CodeEntryAlignment == 0 || do_post_padding, "alignment must be correct");
356 if (do_post_padding) {
357 // force alignment after the cache check.
358 // It's been verified to be aligned if !VerifyOops
359 __ align(CodeEntryAlignment);
360 }
361 return offset;
362 }
365 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
366 jobject o = NULL;
367 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
368 __ movoop(reg, o);
369 patching_epilog(patch, lir_patch_normal, reg, info);
370 }
373 void LIR_Assembler::monitorexit(LIR_Opr obj_opr, LIR_Opr lock_opr, Register new_hdr, int monitor_no, Register exception) {
374 if (exception->is_valid()) {
375 // preserve exception
376 // note: the monitor_exit runtime call is a leaf routine
377 // and cannot block => no GC can happen
378 // The slow case (MonitorAccessStub) uses the first two stack slots
379 // ([esp+0] and [esp+4]), therefore we store the exception at [esp+8]
380 __ movptr (Address(rsp, 2*wordSize), exception);
381 }
383 Register obj_reg = obj_opr->as_register();
384 Register lock_reg = lock_opr->as_register();
386 // setup registers (lock_reg must be rax, for lock_object)
387 assert(obj_reg != SYNC_header && lock_reg != SYNC_header, "rax, must be available here");
388 Register hdr = lock_reg;
389 assert(new_hdr == SYNC_header, "wrong register");
390 lock_reg = new_hdr;
391 // compute pointer to BasicLock
392 Address lock_addr = frame_map()->address_for_monitor_lock(monitor_no);
393 __ lea(lock_reg, lock_addr);
394 // unlock object
395 MonitorAccessStub* slow_case = new MonitorExitStub(lock_opr, true, monitor_no);
396 // _slow_case_stubs->append(slow_case);
397 // temporary fix: must be created after exceptionhandler, therefore as call stub
398 _slow_case_stubs->append(slow_case);
399 if (UseFastLocking) {
400 // try inlined fast unlocking first, revert to slow locking if it fails
401 // note: lock_reg points to the displaced header since the displaced header offset is 0!
402 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
403 __ unlock_object(hdr, obj_reg, lock_reg, *slow_case->entry());
404 } else {
405 // always do slow unlocking
406 // note: the slow unlocking code could be inlined here, however if we use
407 // slow unlocking, speed doesn't matter anyway and this solution is
408 // simpler and requires less duplicated code - additionally, the
409 // slow unlocking code is the same in either case which simplifies
410 // debugging
411 __ jmp(*slow_case->entry());
412 }
413 // done
414 __ bind(*slow_case->continuation());
416 if (exception->is_valid()) {
417 // restore exception
418 __ movptr (exception, Address(rsp, 2 * wordSize));
419 }
420 }
422 // This specifies the rsp decrement needed to build the frame
423 int LIR_Assembler::initial_frame_size_in_bytes() {
424 // if rounding, must let FrameMap know!
426 // The frame_map records size in slots (32bit word)
428 // subtract two words to account for return address and link
429 return (frame_map()->framesize() - (2*VMRegImpl::slots_per_word)) * VMRegImpl::stack_slot_size;
430 }
433 int LIR_Assembler::emit_exception_handler() {
434 // if the last instruction is a call (typically to do a throw which
435 // is coming at the end after block reordering) the return address
436 // must still point into the code area in order to avoid assertion
437 // failures when searching for the corresponding bci => add a nop
438 // (was bug 5/14/1999 - gri)
439 __ nop();
441 // generate code for exception handler
442 address handler_base = __ start_a_stub(exception_handler_size);
443 if (handler_base == NULL) {
444 // not enough space left for the handler
445 bailout("exception handler overflow");
446 return -1;
447 }
449 int offset = code_offset();
451 // the exception oop and pc are in rax, and rdx
452 // no other registers need to be preserved, so invalidate them
453 __ invalidate_registers(false, true, true, false, true, true);
455 // check that there is really an exception
456 __ verify_not_null_oop(rax);
458 // search an exception handler (rax: exception oop, rdx: throwing pc)
459 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_nofpu_id)));
461 __ stop("should not reach here");
463 assert(code_offset() - offset <= exception_handler_size, "overflow");
464 __ end_a_stub();
466 return offset;
467 }
470 // Emit the code to remove the frame from the stack in the exception
471 // unwind path.
472 int LIR_Assembler::emit_unwind_handler() {
473 #ifndef PRODUCT
474 if (CommentedAssembly) {
475 _masm->block_comment("Unwind handler");
476 }
477 #endif
479 int offset = code_offset();
481 // Fetch the exception from TLS and clear out exception related thread state
482 __ get_thread(rsi);
483 __ movptr(rax, Address(rsi, JavaThread::exception_oop_offset()));
484 __ movptr(Address(rsi, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD);
485 __ movptr(Address(rsi, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
487 __ bind(_unwind_handler_entry);
488 __ verify_not_null_oop(rax);
489 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
490 __ mov(rsi, rax); // Preserve the exception
491 }
493 // Preform needed unlocking
494 MonitorExitStub* stub = NULL;
495 if (method()->is_synchronized()) {
496 monitor_address(0, FrameMap::rax_opr);
497 stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
498 __ unlock_object(rdi, rbx, rax, *stub->entry());
499 __ bind(*stub->continuation());
500 }
502 if (compilation()->env()->dtrace_method_probes()) {
503 __ get_thread(rax);
504 __ movptr(Address(rsp, 0), rax);
505 __ movoop(Address(rsp, sizeof(void*)), method()->constant_encoding());
506 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
507 }
509 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
510 __ mov(rax, rsi); // Restore the exception
511 }
513 // remove the activation and dispatch to the unwind handler
514 __ remove_frame(initial_frame_size_in_bytes());
515 __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
517 // Emit the slow path assembly
518 if (stub != NULL) {
519 stub->emit_code(this);
520 }
522 return offset;
523 }
526 int LIR_Assembler::emit_deopt_handler() {
527 // if the last instruction is a call (typically to do a throw which
528 // is coming at the end after block reordering) the return address
529 // must still point into the code area in order to avoid assertion
530 // failures when searching for the corresponding bci => add a nop
531 // (was bug 5/14/1999 - gri)
532 __ nop();
534 // generate code for exception handler
535 address handler_base = __ start_a_stub(deopt_handler_size);
536 if (handler_base == NULL) {
537 // not enough space left for the handler
538 bailout("deopt handler overflow");
539 return -1;
540 }
542 int offset = code_offset();
543 InternalAddress here(__ pc());
545 __ pushptr(here.addr());
546 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
548 assert(code_offset() - offset <= deopt_handler_size, "overflow");
549 __ end_a_stub();
551 return offset;
552 }
555 // This is the fast version of java.lang.String.compare; it has not
556 // OSR-entry and therefore, we generate a slow version for OSR's
557 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
558 __ movptr (rbx, rcx); // receiver is in rcx
559 __ movptr (rax, arg1->as_register());
561 // Get addresses of first characters from both Strings
562 __ load_heap_oop(rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
563 __ movptr (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
564 __ lea (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
567 // rbx, may be NULL
568 add_debug_info_for_null_check_here(info);
569 __ load_heap_oop(rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
570 __ movptr (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
571 __ lea (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
573 // compute minimum length (in rax) and difference of lengths (on top of stack)
574 if (VM_Version::supports_cmov()) {
575 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
576 __ movl (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
577 __ mov (rcx, rbx);
578 __ subptr (rbx, rax); // subtract lengths
579 __ push (rbx); // result
580 __ cmov (Assembler::lessEqual, rax, rcx);
581 } else {
582 Label L;
583 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
584 __ movl (rcx, Address(rax, java_lang_String::count_offset_in_bytes()));
585 __ mov (rax, rbx);
586 __ subptr (rbx, rcx);
587 __ push (rbx);
588 __ jcc (Assembler::lessEqual, L);
589 __ mov (rax, rcx);
590 __ bind (L);
591 }
592 // is minimum length 0?
593 Label noLoop, haveResult;
594 __ testptr (rax, rax);
595 __ jcc (Assembler::zero, noLoop);
597 // compare first characters
598 __ load_unsigned_short(rcx, Address(rdi, 0));
599 __ load_unsigned_short(rbx, Address(rsi, 0));
600 __ subl(rcx, rbx);
601 __ jcc(Assembler::notZero, haveResult);
602 // starting loop
603 __ decrement(rax); // we already tested index: skip one
604 __ jcc(Assembler::zero, noLoop);
606 // set rsi.edi to the end of the arrays (arrays have same length)
607 // negate the index
609 __ lea(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
610 __ lea(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
611 __ negptr(rax);
613 // compare the strings in a loop
615 Label loop;
616 __ align(wordSize);
617 __ bind(loop);
618 __ load_unsigned_short(rcx, Address(rdi, rax, Address::times_2, 0));
619 __ load_unsigned_short(rbx, Address(rsi, rax, Address::times_2, 0));
620 __ subl(rcx, rbx);
621 __ jcc(Assembler::notZero, haveResult);
622 __ increment(rax);
623 __ jcc(Assembler::notZero, loop);
625 // strings are equal up to min length
627 __ bind(noLoop);
628 __ pop(rax);
629 return_op(LIR_OprFact::illegalOpr);
631 __ bind(haveResult);
632 // leave instruction is going to discard the TOS value
633 __ mov (rax, rcx); // result of call is in rax,
634 }
637 void LIR_Assembler::return_op(LIR_Opr result) {
638 assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
639 if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
640 assert(result->fpu() == 0, "result must already be on TOS");
641 }
643 // Pop the stack before the safepoint code
644 __ remove_frame(initial_frame_size_in_bytes());
646 bool result_is_oop = result->is_valid() ? result->is_oop() : false;
648 // Note: we do not need to round double result; float result has the right precision
649 // the poll sets the condition code, but no data registers
650 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
651 relocInfo::poll_return_type);
653 // NOTE: the requires that the polling page be reachable else the reloc
654 // goes to the movq that loads the address and not the faulting instruction
655 // which breaks the signal handler code
657 __ test32(rax, polling_page);
659 __ ret(0);
660 }
663 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
664 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
665 relocInfo::poll_type);
667 if (info != NULL) {
668 add_debug_info_for_branch(info);
669 } else {
670 ShouldNotReachHere();
671 }
673 int offset = __ offset();
675 // NOTE: the requires that the polling page be reachable else the reloc
676 // goes to the movq that loads the address and not the faulting instruction
677 // which breaks the signal handler code
679 __ test32(rax, polling_page);
680 return offset;
681 }
684 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
685 if (from_reg != to_reg) __ mov(to_reg, from_reg);
686 }
688 void LIR_Assembler::swap_reg(Register a, Register b) {
689 __ xchgptr(a, b);
690 }
693 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
694 assert(src->is_constant(), "should not call otherwise");
695 assert(dest->is_register(), "should not call otherwise");
696 LIR_Const* c = src->as_constant_ptr();
698 switch (c->type()) {
699 case T_INT: {
700 assert(patch_code == lir_patch_none, "no patching handled here");
701 __ movl(dest->as_register(), c->as_jint());
702 break;
703 }
705 case T_ADDRESS: {
706 assert(patch_code == lir_patch_none, "no patching handled here");
707 __ movptr(dest->as_register(), c->as_jint());
708 break;
709 }
711 case T_LONG: {
712 assert(patch_code == lir_patch_none, "no patching handled here");
713 #ifdef _LP64
714 __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
715 #else
716 __ movptr(dest->as_register_lo(), c->as_jint_lo());
717 __ movptr(dest->as_register_hi(), c->as_jint_hi());
718 #endif // _LP64
719 break;
720 }
722 case T_OBJECT: {
723 if (patch_code != lir_patch_none) {
724 jobject2reg_with_patching(dest->as_register(), info);
725 } else {
726 __ movoop(dest->as_register(), c->as_jobject());
727 }
728 break;
729 }
731 case T_FLOAT: {
732 if (dest->is_single_xmm()) {
733 if (c->is_zero_float()) {
734 __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
735 } else {
736 __ movflt(dest->as_xmm_float_reg(),
737 InternalAddress(float_constant(c->as_jfloat())));
738 }
739 } else {
740 assert(dest->is_single_fpu(), "must be");
741 assert(dest->fpu_regnr() == 0, "dest must be TOS");
742 if (c->is_zero_float()) {
743 __ fldz();
744 } else if (c->is_one_float()) {
745 __ fld1();
746 } else {
747 __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
748 }
749 }
750 break;
751 }
753 case T_DOUBLE: {
754 if (dest->is_double_xmm()) {
755 if (c->is_zero_double()) {
756 __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
757 } else {
758 __ movdbl(dest->as_xmm_double_reg(),
759 InternalAddress(double_constant(c->as_jdouble())));
760 }
761 } else {
762 assert(dest->is_double_fpu(), "must be");
763 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
764 if (c->is_zero_double()) {
765 __ fldz();
766 } else if (c->is_one_double()) {
767 __ fld1();
768 } else {
769 __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
770 }
771 }
772 break;
773 }
775 default:
776 ShouldNotReachHere();
777 }
778 }
780 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
781 assert(src->is_constant(), "should not call otherwise");
782 assert(dest->is_stack(), "should not call otherwise");
783 LIR_Const* c = src->as_constant_ptr();
785 switch (c->type()) {
786 case T_INT: // fall through
787 case T_FLOAT:
788 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
789 break;
791 case T_ADDRESS:
792 __ movptr(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
793 break;
795 case T_OBJECT:
796 __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
797 break;
799 case T_LONG: // fall through
800 case T_DOUBLE:
801 #ifdef _LP64
802 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
803 lo_word_offset_in_bytes), (intptr_t)c->as_jlong_bits());
804 #else
805 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
806 lo_word_offset_in_bytes), c->as_jint_lo_bits());
807 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
808 hi_word_offset_in_bytes), c->as_jint_hi_bits());
809 #endif // _LP64
810 break;
812 default:
813 ShouldNotReachHere();
814 }
815 }
817 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
818 assert(src->is_constant(), "should not call otherwise");
819 assert(dest->is_address(), "should not call otherwise");
820 LIR_Const* c = src->as_constant_ptr();
821 LIR_Address* addr = dest->as_address_ptr();
823 int null_check_here = code_offset();
824 switch (type) {
825 case T_INT: // fall through
826 case T_FLOAT:
827 __ movl(as_Address(addr), c->as_jint_bits());
828 break;
830 case T_ADDRESS:
831 __ movptr(as_Address(addr), c->as_jint_bits());
832 break;
834 case T_OBJECT: // fall through
835 case T_ARRAY:
836 if (c->as_jobject() == NULL) {
837 if (UseCompressedOops && !wide) {
838 __ movl(as_Address(addr), (int32_t)NULL_WORD);
839 } else {
840 __ movptr(as_Address(addr), NULL_WORD);
841 }
842 } else {
843 if (is_literal_address(addr)) {
844 ShouldNotReachHere();
845 __ movoop(as_Address(addr, noreg), c->as_jobject());
846 } else {
847 #ifdef _LP64
848 __ movoop(rscratch1, c->as_jobject());
849 if (UseCompressedOops && !wide) {
850 __ encode_heap_oop(rscratch1);
851 null_check_here = code_offset();
852 __ movl(as_Address_lo(addr), rscratch1);
853 } else {
854 null_check_here = code_offset();
855 __ movptr(as_Address_lo(addr), rscratch1);
856 }
857 #else
858 __ movoop(as_Address(addr), c->as_jobject());
859 #endif
860 }
861 }
862 break;
864 case T_LONG: // fall through
865 case T_DOUBLE:
866 #ifdef _LP64
867 if (is_literal_address(addr)) {
868 ShouldNotReachHere();
869 __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
870 } else {
871 __ movptr(r10, (intptr_t)c->as_jlong_bits());
872 null_check_here = code_offset();
873 __ movptr(as_Address_lo(addr), r10);
874 }
875 #else
876 // Always reachable in 32bit so this doesn't produce useless move literal
877 __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
878 __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
879 #endif // _LP64
880 break;
882 case T_BOOLEAN: // fall through
883 case T_BYTE:
884 __ movb(as_Address(addr), c->as_jint() & 0xFF);
885 break;
887 case T_CHAR: // fall through
888 case T_SHORT:
889 __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
890 break;
892 default:
893 ShouldNotReachHere();
894 };
896 if (info != NULL) {
897 add_debug_info_for_null_check(null_check_here, info);
898 }
899 }
902 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
903 assert(src->is_register(), "should not call otherwise");
904 assert(dest->is_register(), "should not call otherwise");
906 // move between cpu-registers
907 if (dest->is_single_cpu()) {
908 #ifdef _LP64
909 if (src->type() == T_LONG) {
910 // Can do LONG -> OBJECT
911 move_regs(src->as_register_lo(), dest->as_register());
912 return;
913 }
914 #endif
915 assert(src->is_single_cpu(), "must match");
916 if (src->type() == T_OBJECT) {
917 __ verify_oop(src->as_register());
918 }
919 move_regs(src->as_register(), dest->as_register());
921 } else if (dest->is_double_cpu()) {
922 #ifdef _LP64
923 if (src->type() == T_OBJECT || src->type() == T_ARRAY) {
924 // Surprising to me but we can see move of a long to t_object
925 __ verify_oop(src->as_register());
926 move_regs(src->as_register(), dest->as_register_lo());
927 return;
928 }
929 #endif
930 assert(src->is_double_cpu(), "must match");
931 Register f_lo = src->as_register_lo();
932 Register f_hi = src->as_register_hi();
933 Register t_lo = dest->as_register_lo();
934 Register t_hi = dest->as_register_hi();
935 #ifdef _LP64
936 assert(f_hi == f_lo, "must be same");
937 assert(t_hi == t_lo, "must be same");
938 move_regs(f_lo, t_lo);
939 #else
940 assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
943 if (f_lo == t_hi && f_hi == t_lo) {
944 swap_reg(f_lo, f_hi);
945 } else if (f_hi == t_lo) {
946 assert(f_lo != t_hi, "overwriting register");
947 move_regs(f_hi, t_hi);
948 move_regs(f_lo, t_lo);
949 } else {
950 assert(f_hi != t_lo, "overwriting register");
951 move_regs(f_lo, t_lo);
952 move_regs(f_hi, t_hi);
953 }
954 #endif // LP64
956 // special moves from fpu-register to xmm-register
957 // necessary for method results
958 } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
959 __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
960 __ fld_s(Address(rsp, 0));
961 } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
962 __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
963 __ fld_d(Address(rsp, 0));
964 } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
965 __ fstp_s(Address(rsp, 0));
966 __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
967 } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
968 __ fstp_d(Address(rsp, 0));
969 __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
971 // move between xmm-registers
972 } else if (dest->is_single_xmm()) {
973 assert(src->is_single_xmm(), "must match");
974 __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
975 } else if (dest->is_double_xmm()) {
976 assert(src->is_double_xmm(), "must match");
977 __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
979 // move between fpu-registers (no instruction necessary because of fpu-stack)
980 } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
981 assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
982 assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
983 } else {
984 ShouldNotReachHere();
985 }
986 }
988 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
989 assert(src->is_register(), "should not call otherwise");
990 assert(dest->is_stack(), "should not call otherwise");
992 if (src->is_single_cpu()) {
993 Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
994 if (type == T_OBJECT || type == T_ARRAY) {
995 __ verify_oop(src->as_register());
996 __ movptr (dst, src->as_register());
997 } else {
998 __ movl (dst, src->as_register());
999 }
1001 } else if (src->is_double_cpu()) {
1002 Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
1003 Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
1004 __ movptr (dstLO, src->as_register_lo());
1005 NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
1007 } else if (src->is_single_xmm()) {
1008 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1009 __ movflt(dst_addr, src->as_xmm_float_reg());
1011 } else if (src->is_double_xmm()) {
1012 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1013 __ movdbl(dst_addr, src->as_xmm_double_reg());
1015 } else if (src->is_single_fpu()) {
1016 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1017 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1018 if (pop_fpu_stack) __ fstp_s (dst_addr);
1019 else __ fst_s (dst_addr);
1021 } else if (src->is_double_fpu()) {
1022 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1023 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1024 if (pop_fpu_stack) __ fstp_d (dst_addr);
1025 else __ fst_d (dst_addr);
1027 } else {
1028 ShouldNotReachHere();
1029 }
1030 }
1033 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 */) {
1034 LIR_Address* to_addr = dest->as_address_ptr();
1035 PatchingStub* patch = NULL;
1036 Register compressed_src = rscratch1;
1038 if (type == T_ARRAY || type == T_OBJECT) {
1039 __ verify_oop(src->as_register());
1040 #ifdef _LP64
1041 if (UseCompressedOops && !wide) {
1042 __ movptr(compressed_src, src->as_register());
1043 __ encode_heap_oop(compressed_src);
1044 }
1045 #endif
1046 }
1048 if (patch_code != lir_patch_none) {
1049 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1050 Address toa = as_Address(to_addr);
1051 assert(toa.disp() != 0, "must have");
1052 }
1054 int null_check_here = code_offset();
1055 switch (type) {
1056 case T_FLOAT: {
1057 if (src->is_single_xmm()) {
1058 __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
1059 } else {
1060 assert(src->is_single_fpu(), "must be");
1061 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1062 if (pop_fpu_stack) __ fstp_s(as_Address(to_addr));
1063 else __ fst_s (as_Address(to_addr));
1064 }
1065 break;
1066 }
1068 case T_DOUBLE: {
1069 if (src->is_double_xmm()) {
1070 __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
1071 } else {
1072 assert(src->is_double_fpu(), "must be");
1073 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1074 if (pop_fpu_stack) __ fstp_d(as_Address(to_addr));
1075 else __ fst_d (as_Address(to_addr));
1076 }
1077 break;
1078 }
1080 case T_ARRAY: // fall through
1081 case T_OBJECT: // fall through
1082 if (UseCompressedOops && !wide) {
1083 __ movl(as_Address(to_addr), compressed_src);
1084 } else {
1085 __ movptr(as_Address(to_addr), src->as_register());
1086 }
1087 break;
1088 case T_ADDRESS:
1089 __ movptr(as_Address(to_addr), src->as_register());
1090 break;
1091 case T_INT:
1092 __ movl(as_Address(to_addr), src->as_register());
1093 break;
1095 case T_LONG: {
1096 Register from_lo = src->as_register_lo();
1097 Register from_hi = src->as_register_hi();
1098 #ifdef _LP64
1099 __ movptr(as_Address_lo(to_addr), from_lo);
1100 #else
1101 Register base = to_addr->base()->as_register();
1102 Register index = noreg;
1103 if (to_addr->index()->is_register()) {
1104 index = to_addr->index()->as_register();
1105 }
1106 if (base == from_lo || index == from_lo) {
1107 assert(base != from_hi, "can't be");
1108 assert(index == noreg || (index != base && index != from_hi), "can't handle this");
1109 __ movl(as_Address_hi(to_addr), from_hi);
1110 if (patch != NULL) {
1111 patching_epilog(patch, lir_patch_high, base, info);
1112 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1113 patch_code = lir_patch_low;
1114 }
1115 __ movl(as_Address_lo(to_addr), from_lo);
1116 } else {
1117 assert(index == noreg || (index != base && index != from_lo), "can't handle this");
1118 __ movl(as_Address_lo(to_addr), from_lo);
1119 if (patch != NULL) {
1120 patching_epilog(patch, lir_patch_low, base, info);
1121 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1122 patch_code = lir_patch_high;
1123 }
1124 __ movl(as_Address_hi(to_addr), from_hi);
1125 }
1126 #endif // _LP64
1127 break;
1128 }
1130 case T_BYTE: // fall through
1131 case T_BOOLEAN: {
1132 Register src_reg = src->as_register();
1133 Address dst_addr = as_Address(to_addr);
1134 assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
1135 __ movb(dst_addr, src_reg);
1136 break;
1137 }
1139 case T_CHAR: // fall through
1140 case T_SHORT:
1141 __ movw(as_Address(to_addr), src->as_register());
1142 break;
1144 default:
1145 ShouldNotReachHere();
1146 }
1147 if (info != NULL) {
1148 add_debug_info_for_null_check(null_check_here, info);
1149 }
1151 if (patch_code != lir_patch_none) {
1152 patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
1153 }
1154 }
1157 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1158 assert(src->is_stack(), "should not call otherwise");
1159 assert(dest->is_register(), "should not call otherwise");
1161 if (dest->is_single_cpu()) {
1162 if (type == T_ARRAY || type == T_OBJECT) {
1163 __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1164 __ verify_oop(dest->as_register());
1165 } else {
1166 __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1167 }
1169 } else if (dest->is_double_cpu()) {
1170 Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
1171 Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
1172 __ movptr(dest->as_register_lo(), src_addr_LO);
1173 NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
1175 } else if (dest->is_single_xmm()) {
1176 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1177 __ movflt(dest->as_xmm_float_reg(), src_addr);
1179 } else if (dest->is_double_xmm()) {
1180 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1181 __ movdbl(dest->as_xmm_double_reg(), src_addr);
1183 } else if (dest->is_single_fpu()) {
1184 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1185 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1186 __ fld_s(src_addr);
1188 } else if (dest->is_double_fpu()) {
1189 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1190 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1191 __ fld_d(src_addr);
1193 } else {
1194 ShouldNotReachHere();
1195 }
1196 }
1199 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1200 if (src->is_single_stack()) {
1201 if (type == T_OBJECT || type == T_ARRAY) {
1202 __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
1203 __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
1204 } else {
1205 #ifndef _LP64
1206 __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
1207 __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
1208 #else
1209 //no pushl on 64bits
1210 __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix()));
1211 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1);
1212 #endif
1213 }
1215 } else if (src->is_double_stack()) {
1216 #ifdef _LP64
1217 __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
1218 __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
1219 #else
1220 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
1221 // push and pop the part at src + wordSize, adding wordSize for the previous push
1222 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 2 * wordSize));
1223 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 2 * wordSize));
1224 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
1225 #endif // _LP64
1227 } else {
1228 ShouldNotReachHere();
1229 }
1230 }
1233 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool /* unaligned */) {
1234 assert(src->is_address(), "should not call otherwise");
1235 assert(dest->is_register(), "should not call otherwise");
1237 LIR_Address* addr = src->as_address_ptr();
1238 Address from_addr = as_Address(addr);
1240 switch (type) {
1241 case T_BOOLEAN: // fall through
1242 case T_BYTE: // fall through
1243 case T_CHAR: // fall through
1244 case T_SHORT:
1245 if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
1246 // on pre P6 processors we may get partial register stalls
1247 // so blow away the value of to_rinfo before loading a
1248 // partial word into it. Do it here so that it precedes
1249 // the potential patch point below.
1250 __ xorptr(dest->as_register(), dest->as_register());
1251 }
1252 break;
1253 }
1255 PatchingStub* patch = NULL;
1256 if (patch_code != lir_patch_none) {
1257 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1258 assert(from_addr.disp() != 0, "must have");
1259 }
1260 if (info != NULL) {
1261 add_debug_info_for_null_check_here(info);
1262 }
1264 switch (type) {
1265 case T_FLOAT: {
1266 if (dest->is_single_xmm()) {
1267 __ movflt(dest->as_xmm_float_reg(), from_addr);
1268 } else {
1269 assert(dest->is_single_fpu(), "must be");
1270 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1271 __ fld_s(from_addr);
1272 }
1273 break;
1274 }
1276 case T_DOUBLE: {
1277 if (dest->is_double_xmm()) {
1278 __ movdbl(dest->as_xmm_double_reg(), from_addr);
1279 } else {
1280 assert(dest->is_double_fpu(), "must be");
1281 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1282 __ fld_d(from_addr);
1283 }
1284 break;
1285 }
1287 case T_OBJECT: // fall through
1288 case T_ARRAY: // fall through
1289 if (UseCompressedOops && !wide) {
1290 __ movl(dest->as_register(), from_addr);
1291 } else {
1292 __ movptr(dest->as_register(), from_addr);
1293 }
1294 break;
1296 case T_ADDRESS:
1297 __ movptr(dest->as_register(), from_addr);
1298 break;
1299 case T_INT:
1300 __ movl(dest->as_register(), from_addr);
1301 break;
1303 case T_LONG: {
1304 Register to_lo = dest->as_register_lo();
1305 Register to_hi = dest->as_register_hi();
1306 #ifdef _LP64
1307 __ movptr(to_lo, as_Address_lo(addr));
1308 #else
1309 Register base = addr->base()->as_register();
1310 Register index = noreg;
1311 if (addr->index()->is_register()) {
1312 index = addr->index()->as_register();
1313 }
1314 if ((base == to_lo && index == to_hi) ||
1315 (base == to_hi && index == to_lo)) {
1316 // addresses with 2 registers are only formed as a result of
1317 // array access so this code will never have to deal with
1318 // patches or null checks.
1319 assert(info == NULL && patch == NULL, "must be");
1320 __ lea(to_hi, as_Address(addr));
1321 __ movl(to_lo, Address(to_hi, 0));
1322 __ movl(to_hi, Address(to_hi, BytesPerWord));
1323 } else if (base == to_lo || index == to_lo) {
1324 assert(base != to_hi, "can't be");
1325 assert(index == noreg || (index != base && index != to_hi), "can't handle this");
1326 __ movl(to_hi, as_Address_hi(addr));
1327 if (patch != NULL) {
1328 patching_epilog(patch, lir_patch_high, base, info);
1329 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1330 patch_code = lir_patch_low;
1331 }
1332 __ movl(to_lo, as_Address_lo(addr));
1333 } else {
1334 assert(index == noreg || (index != base && index != to_lo), "can't handle this");
1335 __ movl(to_lo, as_Address_lo(addr));
1336 if (patch != NULL) {
1337 patching_epilog(patch, lir_patch_low, base, info);
1338 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1339 patch_code = lir_patch_high;
1340 }
1341 __ movl(to_hi, as_Address_hi(addr));
1342 }
1343 #endif // _LP64
1344 break;
1345 }
1347 case T_BOOLEAN: // fall through
1348 case T_BYTE: {
1349 Register dest_reg = dest->as_register();
1350 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1351 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1352 __ movsbl(dest_reg, from_addr);
1353 } else {
1354 __ movb(dest_reg, from_addr);
1355 __ shll(dest_reg, 24);
1356 __ sarl(dest_reg, 24);
1357 }
1358 break;
1359 }
1361 case T_CHAR: {
1362 Register dest_reg = dest->as_register();
1363 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1364 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1365 __ movzwl(dest_reg, from_addr);
1366 } else {
1367 __ movw(dest_reg, from_addr);
1368 }
1369 break;
1370 }
1372 case T_SHORT: {
1373 Register dest_reg = dest->as_register();
1374 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1375 __ movswl(dest_reg, from_addr);
1376 } else {
1377 __ movw(dest_reg, from_addr);
1378 __ shll(dest_reg, 16);
1379 __ sarl(dest_reg, 16);
1380 }
1381 break;
1382 }
1384 default:
1385 ShouldNotReachHere();
1386 }
1388 if (patch != NULL) {
1389 patching_epilog(patch, patch_code, addr->base()->as_register(), info);
1390 }
1392 if (type == T_ARRAY || type == T_OBJECT) {
1393 #ifdef _LP64
1394 if (UseCompressedOops && !wide) {
1395 __ decode_heap_oop(dest->as_register());
1396 }
1397 #endif
1398 __ verify_oop(dest->as_register());
1399 }
1400 }
1403 void LIR_Assembler::prefetchr(LIR_Opr src) {
1404 LIR_Address* addr = src->as_address_ptr();
1405 Address from_addr = as_Address(addr);
1407 if (VM_Version::supports_sse()) {
1408 switch (ReadPrefetchInstr) {
1409 case 0:
1410 __ prefetchnta(from_addr); break;
1411 case 1:
1412 __ prefetcht0(from_addr); break;
1413 case 2:
1414 __ prefetcht2(from_addr); break;
1415 default:
1416 ShouldNotReachHere(); break;
1417 }
1418 } else if (VM_Version::supports_3dnow()) {
1419 __ prefetchr(from_addr);
1420 }
1421 }
1424 void LIR_Assembler::prefetchw(LIR_Opr src) {
1425 LIR_Address* addr = src->as_address_ptr();
1426 Address from_addr = as_Address(addr);
1428 if (VM_Version::supports_sse()) {
1429 switch (AllocatePrefetchInstr) {
1430 case 0:
1431 __ prefetchnta(from_addr); break;
1432 case 1:
1433 __ prefetcht0(from_addr); break;
1434 case 2:
1435 __ prefetcht2(from_addr); break;
1436 case 3:
1437 __ prefetchw(from_addr); break;
1438 default:
1439 ShouldNotReachHere(); break;
1440 }
1441 } else if (VM_Version::supports_3dnow()) {
1442 __ prefetchw(from_addr);
1443 }
1444 }
1447 NEEDS_CLEANUP; // This could be static?
1448 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
1449 int elem_size = type2aelembytes(type);
1450 switch (elem_size) {
1451 case 1: return Address::times_1;
1452 case 2: return Address::times_2;
1453 case 4: return Address::times_4;
1454 case 8: return Address::times_8;
1455 }
1456 ShouldNotReachHere();
1457 return Address::no_scale;
1458 }
1461 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1462 switch (op->code()) {
1463 case lir_idiv:
1464 case lir_irem:
1465 arithmetic_idiv(op->code(),
1466 op->in_opr1(),
1467 op->in_opr2(),
1468 op->in_opr3(),
1469 op->result_opr(),
1470 op->info());
1471 break;
1472 default: ShouldNotReachHere(); break;
1473 }
1474 }
1476 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1477 #ifdef ASSERT
1478 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
1479 if (op->block() != NULL) _branch_target_blocks.append(op->block());
1480 if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
1481 #endif
1483 if (op->cond() == lir_cond_always) {
1484 if (op->info() != NULL) add_debug_info_for_branch(op->info());
1485 __ jmp (*(op->label()));
1486 } else {
1487 Assembler::Condition acond = Assembler::zero;
1488 if (op->code() == lir_cond_float_branch) {
1489 assert(op->ublock() != NULL, "must have unordered successor");
1490 __ jcc(Assembler::parity, *(op->ublock()->label()));
1491 switch(op->cond()) {
1492 case lir_cond_equal: acond = Assembler::equal; break;
1493 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1494 case lir_cond_less: acond = Assembler::below; break;
1495 case lir_cond_lessEqual: acond = Assembler::belowEqual; break;
1496 case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
1497 case lir_cond_greater: acond = Assembler::above; break;
1498 default: ShouldNotReachHere();
1499 }
1500 } else {
1501 switch (op->cond()) {
1502 case lir_cond_equal: acond = Assembler::equal; break;
1503 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1504 case lir_cond_less: acond = Assembler::less; break;
1505 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
1506 case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
1507 case lir_cond_greater: acond = Assembler::greater; break;
1508 case lir_cond_belowEqual: acond = Assembler::belowEqual; break;
1509 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break;
1510 default: ShouldNotReachHere();
1511 }
1512 }
1513 __ jcc(acond,*(op->label()));
1514 }
1515 }
1517 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1518 LIR_Opr src = op->in_opr();
1519 LIR_Opr dest = op->result_opr();
1521 switch (op->bytecode()) {
1522 case Bytecodes::_i2l:
1523 #ifdef _LP64
1524 __ movl2ptr(dest->as_register_lo(), src->as_register());
1525 #else
1526 move_regs(src->as_register(), dest->as_register_lo());
1527 move_regs(src->as_register(), dest->as_register_hi());
1528 __ sarl(dest->as_register_hi(), 31);
1529 #endif // LP64
1530 break;
1532 case Bytecodes::_l2i:
1533 move_regs(src->as_register_lo(), dest->as_register());
1534 break;
1536 case Bytecodes::_i2b:
1537 move_regs(src->as_register(), dest->as_register());
1538 __ sign_extend_byte(dest->as_register());
1539 break;
1541 case Bytecodes::_i2c:
1542 move_regs(src->as_register(), dest->as_register());
1543 __ andl(dest->as_register(), 0xFFFF);
1544 break;
1546 case Bytecodes::_i2s:
1547 move_regs(src->as_register(), dest->as_register());
1548 __ sign_extend_short(dest->as_register());
1549 break;
1552 case Bytecodes::_f2d:
1553 case Bytecodes::_d2f:
1554 if (dest->is_single_xmm()) {
1555 __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1556 } else if (dest->is_double_xmm()) {
1557 __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1558 } else {
1559 assert(src->fpu() == dest->fpu(), "register must be equal");
1560 // do nothing (float result is rounded later through spilling)
1561 }
1562 break;
1564 case Bytecodes::_i2f:
1565 case Bytecodes::_i2d:
1566 if (dest->is_single_xmm()) {
1567 __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1568 } else if (dest->is_double_xmm()) {
1569 __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1570 } else {
1571 assert(dest->fpu() == 0, "result must be on TOS");
1572 __ movl(Address(rsp, 0), src->as_register());
1573 __ fild_s(Address(rsp, 0));
1574 }
1575 break;
1577 case Bytecodes::_f2i:
1578 case Bytecodes::_d2i:
1579 if (src->is_single_xmm()) {
1580 __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
1581 } else if (src->is_double_xmm()) {
1582 __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
1583 } else {
1584 assert(src->fpu() == 0, "input must be on TOS");
1585 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
1586 __ fist_s(Address(rsp, 0));
1587 __ movl(dest->as_register(), Address(rsp, 0));
1588 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1589 }
1591 // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
1592 assert(op->stub() != NULL, "stub required");
1593 __ cmpl(dest->as_register(), 0x80000000);
1594 __ jcc(Assembler::equal, *op->stub()->entry());
1595 __ bind(*op->stub()->continuation());
1596 break;
1598 case Bytecodes::_l2f:
1599 case Bytecodes::_l2d:
1600 assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
1601 assert(dest->fpu() == 0, "result must be on TOS");
1603 __ movptr(Address(rsp, 0), src->as_register_lo());
1604 NOT_LP64(__ movl(Address(rsp, BytesPerWord), src->as_register_hi()));
1605 __ fild_d(Address(rsp, 0));
1606 // float result is rounded later through spilling
1607 break;
1609 case Bytecodes::_f2l:
1610 case Bytecodes::_d2l:
1611 assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
1612 assert(src->fpu() == 0, "input must be on TOS");
1613 assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
1615 // instruction sequence too long to inline it here
1616 {
1617 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
1618 }
1619 break;
1621 default: ShouldNotReachHere();
1622 }
1623 }
1625 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1626 if (op->init_check()) {
1627 __ cmpl(Address(op->klass()->as_register(),
1628 instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)),
1629 instanceKlass::fully_initialized);
1630 add_debug_info_for_null_check_here(op->stub()->info());
1631 __ jcc(Assembler::notEqual, *op->stub()->entry());
1632 }
1633 __ allocate_object(op->obj()->as_register(),
1634 op->tmp1()->as_register(),
1635 op->tmp2()->as_register(),
1636 op->header_size(),
1637 op->object_size(),
1638 op->klass()->as_register(),
1639 *op->stub()->entry());
1640 __ bind(*op->stub()->continuation());
1641 }
1643 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1644 Register len = op->len()->as_register();
1645 LP64_ONLY( __ movslq(len, len); )
1647 if (UseSlowPath ||
1648 (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
1649 (!UseFastNewTypeArray && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
1650 __ jmp(*op->stub()->entry());
1651 } else {
1652 Register tmp1 = op->tmp1()->as_register();
1653 Register tmp2 = op->tmp2()->as_register();
1654 Register tmp3 = op->tmp3()->as_register();
1655 if (len == tmp1) {
1656 tmp1 = tmp3;
1657 } else if (len == tmp2) {
1658 tmp2 = tmp3;
1659 } else if (len == tmp3) {
1660 // everything is ok
1661 } else {
1662 __ mov(tmp3, len);
1663 }
1664 __ allocate_array(op->obj()->as_register(),
1665 len,
1666 tmp1,
1667 tmp2,
1668 arrayOopDesc::header_size(op->type()),
1669 array_element_size(op->type()),
1670 op->klass()->as_register(),
1671 *op->stub()->entry());
1672 }
1673 __ bind(*op->stub()->continuation());
1674 }
1676 void LIR_Assembler::type_profile_helper(Register mdo,
1677 ciMethodData *md, ciProfileData *data,
1678 Register recv, Label* update_done) {
1679 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1680 Label next_test;
1681 // See if the receiver is receiver[n].
1682 __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
1683 __ jccb(Assembler::notEqual, next_test);
1684 Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
1685 __ addptr(data_addr, DataLayout::counter_increment);
1686 __ jmp(*update_done);
1687 __ bind(next_test);
1688 }
1690 // Didn't find receiver; find next empty slot and fill it in
1691 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1692 Label next_test;
1693 Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
1694 __ cmpptr(recv_addr, (intptr_t)NULL_WORD);
1695 __ jccb(Assembler::notEqual, next_test);
1696 __ movptr(recv_addr, recv);
1697 __ movptr(Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))), DataLayout::counter_increment);
1698 __ jmp(*update_done);
1699 __ bind(next_test);
1700 }
1701 }
1703 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
1704 // we always need a stub for the failure case.
1705 CodeStub* stub = op->stub();
1706 Register obj = op->object()->as_register();
1707 Register k_RInfo = op->tmp1()->as_register();
1708 Register klass_RInfo = op->tmp2()->as_register();
1709 Register dst = op->result_opr()->as_register();
1710 ciKlass* k = op->klass();
1711 Register Rtmp1 = noreg;
1713 // check if it needs to be profiled
1714 ciMethodData* md;
1715 ciProfileData* data;
1717 if (op->should_profile()) {
1718 ciMethod* method = op->profiled_method();
1719 assert(method != NULL, "Should have method");
1720 int bci = op->profiled_bci();
1721 md = method->method_data_or_null();
1722 assert(md != NULL, "Sanity");
1723 data = md->bci_to_data(bci);
1724 assert(data != NULL, "need data for type check");
1725 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1726 }
1727 Label profile_cast_success, profile_cast_failure;
1728 Label *success_target = op->should_profile() ? &profile_cast_success : success;
1729 Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
1731 if (obj == k_RInfo) {
1732 k_RInfo = dst;
1733 } else if (obj == klass_RInfo) {
1734 klass_RInfo = dst;
1735 }
1736 if (k->is_loaded() && !UseCompressedOops) {
1737 select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1738 } else {
1739 Rtmp1 = op->tmp3()->as_register();
1740 select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1741 }
1743 assert_different_registers(obj, k_RInfo, klass_RInfo);
1744 if (!k->is_loaded()) {
1745 jobject2reg_with_patching(k_RInfo, op->info_for_patch());
1746 } else {
1747 #ifdef _LP64
1748 __ movoop(k_RInfo, k->constant_encoding());
1749 #endif // _LP64
1750 }
1751 assert(obj != k_RInfo, "must be different");
1753 __ cmpptr(obj, (int32_t)NULL_WORD);
1754 if (op->should_profile()) {
1755 Label not_null;
1756 __ jccb(Assembler::notEqual, not_null);
1757 // Object is null; update MDO and exit
1758 Register mdo = klass_RInfo;
1759 __ movoop(mdo, md->constant_encoding());
1760 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1761 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1762 __ orl(data_addr, header_bits);
1763 __ jmp(*obj_is_null);
1764 __ bind(not_null);
1765 } else {
1766 __ jcc(Assembler::equal, *obj_is_null);
1767 }
1768 __ verify_oop(obj);
1770 if (op->fast_check()) {
1771 // get object class
1772 // not a safepoint as obj null check happens earlier
1773 #ifdef _LP64
1774 if (UseCompressedOops) {
1775 __ load_klass(Rtmp1, obj);
1776 __ cmpptr(k_RInfo, Rtmp1);
1777 } else {
1778 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1779 }
1780 #else
1781 if (k->is_loaded()) {
1782 __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->constant_encoding());
1783 } else {
1784 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1785 }
1786 #endif
1787 __ jcc(Assembler::notEqual, *failure_target);
1788 // successful cast, fall through to profile or jump
1789 } else {
1790 // get object class
1791 // not a safepoint as obj null check happens earlier
1792 __ load_klass(klass_RInfo, obj);
1793 if (k->is_loaded()) {
1794 // See if we get an immediate positive hit
1795 #ifdef _LP64
1796 __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
1797 #else
1798 __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->constant_encoding());
1799 #endif // _LP64
1800 if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() != k->super_check_offset()) {
1801 __ jcc(Assembler::notEqual, *failure_target);
1802 // successful cast, fall through to profile or jump
1803 } else {
1804 // See if we get an immediate positive hit
1805 __ jcc(Assembler::equal, *success_target);
1806 // check for self
1807 #ifdef _LP64
1808 __ cmpptr(klass_RInfo, k_RInfo);
1809 #else
1810 __ cmpoop(klass_RInfo, k->constant_encoding());
1811 #endif // _LP64
1812 __ jcc(Assembler::equal, *success_target);
1814 __ push(klass_RInfo);
1815 #ifdef _LP64
1816 __ push(k_RInfo);
1817 #else
1818 __ pushoop(k->constant_encoding());
1819 #endif // _LP64
1820 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1821 __ pop(klass_RInfo);
1822 __ pop(klass_RInfo);
1823 // result is a boolean
1824 __ cmpl(klass_RInfo, 0);
1825 __ jcc(Assembler::equal, *failure_target);
1826 // successful cast, fall through to profile or jump
1827 }
1828 } else {
1829 // perform the fast part of the checking logic
1830 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1831 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1832 __ push(klass_RInfo);
1833 __ push(k_RInfo);
1834 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1835 __ pop(klass_RInfo);
1836 __ pop(k_RInfo);
1837 // result is a boolean
1838 __ cmpl(k_RInfo, 0);
1839 __ jcc(Assembler::equal, *failure_target);
1840 // successful cast, fall through to profile or jump
1841 }
1842 }
1843 if (op->should_profile()) {
1844 Register mdo = klass_RInfo, recv = k_RInfo;
1845 __ bind(profile_cast_success);
1846 __ movoop(mdo, md->constant_encoding());
1847 __ load_klass(recv, obj);
1848 Label update_done;
1849 type_profile_helper(mdo, md, data, recv, success);
1850 __ jmp(*success);
1852 __ bind(profile_cast_failure);
1853 __ movoop(mdo, md->constant_encoding());
1854 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1855 __ subptr(counter_addr, DataLayout::counter_increment);
1856 __ jmp(*failure);
1857 }
1858 __ jmp(*success);
1859 }
1862 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1863 LIR_Code code = op->code();
1864 if (code == lir_store_check) {
1865 Register value = op->object()->as_register();
1866 Register array = op->array()->as_register();
1867 Register k_RInfo = op->tmp1()->as_register();
1868 Register klass_RInfo = op->tmp2()->as_register();
1869 Register Rtmp1 = op->tmp3()->as_register();
1871 CodeStub* stub = op->stub();
1873 // check if it needs to be profiled
1874 ciMethodData* md;
1875 ciProfileData* data;
1877 if (op->should_profile()) {
1878 ciMethod* method = op->profiled_method();
1879 assert(method != NULL, "Should have method");
1880 int bci = op->profiled_bci();
1881 md = method->method_data_or_null();
1882 assert(md != NULL, "Sanity");
1883 data = md->bci_to_data(bci);
1884 assert(data != NULL, "need data for type check");
1885 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1886 }
1887 Label profile_cast_success, profile_cast_failure, done;
1888 Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1889 Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
1891 __ cmpptr(value, (int32_t)NULL_WORD);
1892 if (op->should_profile()) {
1893 Label not_null;
1894 __ jccb(Assembler::notEqual, not_null);
1895 // Object is null; update MDO and exit
1896 Register mdo = klass_RInfo;
1897 __ movoop(mdo, md->constant_encoding());
1898 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1899 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1900 __ orl(data_addr, header_bits);
1901 __ jmp(done);
1902 __ bind(not_null);
1903 } else {
1904 __ jcc(Assembler::equal, done);
1905 }
1907 add_debug_info_for_null_check_here(op->info_for_exception());
1908 __ load_klass(k_RInfo, array);
1909 __ load_klass(klass_RInfo, value);
1911 // get instance klass (it's already uncompressed)
1912 __ movptr(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
1913 // perform the fast part of the checking logic
1914 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1915 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1916 __ push(klass_RInfo);
1917 __ push(k_RInfo);
1918 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1919 __ pop(klass_RInfo);
1920 __ pop(k_RInfo);
1921 // result is a boolean
1922 __ cmpl(k_RInfo, 0);
1923 __ jcc(Assembler::equal, *failure_target);
1924 // fall through to the success case
1926 if (op->should_profile()) {
1927 Register mdo = klass_RInfo, recv = k_RInfo;
1928 __ bind(profile_cast_success);
1929 __ movoop(mdo, md->constant_encoding());
1930 __ load_klass(recv, value);
1931 Label update_done;
1932 type_profile_helper(mdo, md, data, recv, &done);
1933 __ jmpb(done);
1935 __ bind(profile_cast_failure);
1936 __ movoop(mdo, md->constant_encoding());
1937 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1938 __ subptr(counter_addr, DataLayout::counter_increment);
1939 __ jmp(*stub->entry());
1940 }
1942 __ bind(done);
1943 } else
1944 if (code == lir_checkcast) {
1945 Register obj = op->object()->as_register();
1946 Register dst = op->result_opr()->as_register();
1947 Label success;
1948 emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
1949 __ bind(success);
1950 if (dst != obj) {
1951 __ mov(dst, obj);
1952 }
1953 } else
1954 if (code == lir_instanceof) {
1955 Register obj = op->object()->as_register();
1956 Register dst = op->result_opr()->as_register();
1957 Label success, failure, done;
1958 emit_typecheck_helper(op, &success, &failure, &failure);
1959 __ bind(failure);
1960 __ xorptr(dst, dst);
1961 __ jmpb(done);
1962 __ bind(success);
1963 __ movptr(dst, 1);
1964 __ bind(done);
1965 } else {
1966 ShouldNotReachHere();
1967 }
1969 }
1972 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1973 if (LP64_ONLY(false &&) op->code() == lir_cas_long && VM_Version::supports_cx8()) {
1974 assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
1975 assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
1976 assert(op->new_value()->as_register_lo() == rbx, "wrong register");
1977 assert(op->new_value()->as_register_hi() == rcx, "wrong register");
1978 Register addr = op->addr()->as_register();
1979 if (os::is_MP()) {
1980 __ lock();
1981 }
1982 NOT_LP64(__ cmpxchg8(Address(addr, 0)));
1984 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
1985 NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
1986 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1987 Register newval = op->new_value()->as_register();
1988 Register cmpval = op->cmp_value()->as_register();
1989 assert(cmpval == rax, "wrong register");
1990 assert(newval != NULL, "new val must be register");
1991 assert(cmpval != newval, "cmp and new values must be in different registers");
1992 assert(cmpval != addr, "cmp and addr must be in different registers");
1993 assert(newval != addr, "new value and addr must be in different registers");
1995 if ( op->code() == lir_cas_obj) {
1996 #ifdef _LP64
1997 if (UseCompressedOops) {
1998 __ encode_heap_oop(cmpval);
1999 __ mov(rscratch1, newval);
2000 __ encode_heap_oop(rscratch1);
2001 if (os::is_MP()) {
2002 __ lock();
2003 }
2004 // cmpval (rax) is implicitly used by this instruction
2005 __ cmpxchgl(rscratch1, Address(addr, 0));
2006 } else
2007 #endif
2008 {
2009 if (os::is_MP()) {
2010 __ lock();
2011 }
2012 __ cmpxchgptr(newval, Address(addr, 0));
2013 }
2014 } else {
2015 assert(op->code() == lir_cas_int, "lir_cas_int expected");
2016 if (os::is_MP()) {
2017 __ lock();
2018 }
2019 __ cmpxchgl(newval, Address(addr, 0));
2020 }
2021 #ifdef _LP64
2022 } else if (op->code() == lir_cas_long) {
2023 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
2024 Register newval = op->new_value()->as_register_lo();
2025 Register cmpval = op->cmp_value()->as_register_lo();
2026 assert(cmpval == rax, "wrong register");
2027 assert(newval != NULL, "new val must be register");
2028 assert(cmpval != newval, "cmp and new values must be in different registers");
2029 assert(cmpval != addr, "cmp and addr must be in different registers");
2030 assert(newval != addr, "new value and addr must be in different registers");
2031 if (os::is_MP()) {
2032 __ lock();
2033 }
2034 __ cmpxchgq(newval, Address(addr, 0));
2035 #endif // _LP64
2036 } else {
2037 Unimplemented();
2038 }
2039 }
2041 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
2042 Assembler::Condition acond, ncond;
2043 switch (condition) {
2044 case lir_cond_equal: acond = Assembler::equal; ncond = Assembler::notEqual; break;
2045 case lir_cond_notEqual: acond = Assembler::notEqual; ncond = Assembler::equal; break;
2046 case lir_cond_less: acond = Assembler::less; ncond = Assembler::greaterEqual; break;
2047 case lir_cond_lessEqual: acond = Assembler::lessEqual; ncond = Assembler::greater; break;
2048 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less; break;
2049 case lir_cond_greater: acond = Assembler::greater; ncond = Assembler::lessEqual; break;
2050 case lir_cond_belowEqual: acond = Assembler::belowEqual; ncond = Assembler::above; break;
2051 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; ncond = Assembler::below; break;
2052 default: ShouldNotReachHere();
2053 }
2055 if (opr1->is_cpu_register()) {
2056 reg2reg(opr1, result);
2057 } else if (opr1->is_stack()) {
2058 stack2reg(opr1, result, result->type());
2059 } else if (opr1->is_constant()) {
2060 const2reg(opr1, result, lir_patch_none, NULL);
2061 } else {
2062 ShouldNotReachHere();
2063 }
2065 if (VM_Version::supports_cmov() && !opr2->is_constant()) {
2066 // optimized version that does not require a branch
2067 if (opr2->is_single_cpu()) {
2068 assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
2069 __ cmov(ncond, result->as_register(), opr2->as_register());
2070 } else if (opr2->is_double_cpu()) {
2071 assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2072 assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2073 __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
2074 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
2075 } else if (opr2->is_single_stack()) {
2076 __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
2077 } else if (opr2->is_double_stack()) {
2078 __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
2079 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
2080 } else {
2081 ShouldNotReachHere();
2082 }
2084 } else {
2085 Label skip;
2086 __ jcc (acond, skip);
2087 if (opr2->is_cpu_register()) {
2088 reg2reg(opr2, result);
2089 } else if (opr2->is_stack()) {
2090 stack2reg(opr2, result, result->type());
2091 } else if (opr2->is_constant()) {
2092 const2reg(opr2, result, lir_patch_none, NULL);
2093 } else {
2094 ShouldNotReachHere();
2095 }
2096 __ bind(skip);
2097 }
2098 }
2101 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
2102 assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
2104 if (left->is_single_cpu()) {
2105 assert(left == dest, "left and dest must be equal");
2106 Register lreg = left->as_register();
2108 if (right->is_single_cpu()) {
2109 // cpu register - cpu register
2110 Register rreg = right->as_register();
2111 switch (code) {
2112 case lir_add: __ addl (lreg, rreg); break;
2113 case lir_sub: __ subl (lreg, rreg); break;
2114 case lir_mul: __ imull(lreg, rreg); break;
2115 default: ShouldNotReachHere();
2116 }
2118 } else if (right->is_stack()) {
2119 // cpu register - stack
2120 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2121 switch (code) {
2122 case lir_add: __ addl(lreg, raddr); break;
2123 case lir_sub: __ subl(lreg, raddr); break;
2124 default: ShouldNotReachHere();
2125 }
2127 } else if (right->is_constant()) {
2128 // cpu register - constant
2129 jint c = right->as_constant_ptr()->as_jint();
2130 switch (code) {
2131 case lir_add: {
2132 __ incrementl(lreg, c);
2133 break;
2134 }
2135 case lir_sub: {
2136 __ decrementl(lreg, c);
2137 break;
2138 }
2139 default: ShouldNotReachHere();
2140 }
2142 } else {
2143 ShouldNotReachHere();
2144 }
2146 } else if (left->is_double_cpu()) {
2147 assert(left == dest, "left and dest must be equal");
2148 Register lreg_lo = left->as_register_lo();
2149 Register lreg_hi = left->as_register_hi();
2151 if (right->is_double_cpu()) {
2152 // cpu register - cpu register
2153 Register rreg_lo = right->as_register_lo();
2154 Register rreg_hi = right->as_register_hi();
2155 NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
2156 LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
2157 switch (code) {
2158 case lir_add:
2159 __ addptr(lreg_lo, rreg_lo);
2160 NOT_LP64(__ adcl(lreg_hi, rreg_hi));
2161 break;
2162 case lir_sub:
2163 __ subptr(lreg_lo, rreg_lo);
2164 NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
2165 break;
2166 case lir_mul:
2167 #ifdef _LP64
2168 __ imulq(lreg_lo, rreg_lo);
2169 #else
2170 assert(lreg_lo == rax && lreg_hi == rdx, "must be");
2171 __ imull(lreg_hi, rreg_lo);
2172 __ imull(rreg_hi, lreg_lo);
2173 __ addl (rreg_hi, lreg_hi);
2174 __ mull (rreg_lo);
2175 __ addl (lreg_hi, rreg_hi);
2176 #endif // _LP64
2177 break;
2178 default:
2179 ShouldNotReachHere();
2180 }
2182 } else if (right->is_constant()) {
2183 // cpu register - constant
2184 #ifdef _LP64
2185 jlong c = right->as_constant_ptr()->as_jlong_bits();
2186 __ movptr(r10, (intptr_t) c);
2187 switch (code) {
2188 case lir_add:
2189 __ addptr(lreg_lo, r10);
2190 break;
2191 case lir_sub:
2192 __ subptr(lreg_lo, r10);
2193 break;
2194 default:
2195 ShouldNotReachHere();
2196 }
2197 #else
2198 jint c_lo = right->as_constant_ptr()->as_jint_lo();
2199 jint c_hi = right->as_constant_ptr()->as_jint_hi();
2200 switch (code) {
2201 case lir_add:
2202 __ addptr(lreg_lo, c_lo);
2203 __ adcl(lreg_hi, c_hi);
2204 break;
2205 case lir_sub:
2206 __ subptr(lreg_lo, c_lo);
2207 __ sbbl(lreg_hi, c_hi);
2208 break;
2209 default:
2210 ShouldNotReachHere();
2211 }
2212 #endif // _LP64
2214 } else {
2215 ShouldNotReachHere();
2216 }
2218 } else if (left->is_single_xmm()) {
2219 assert(left == dest, "left and dest must be equal");
2220 XMMRegister lreg = left->as_xmm_float_reg();
2222 if (right->is_single_xmm()) {
2223 XMMRegister rreg = right->as_xmm_float_reg();
2224 switch (code) {
2225 case lir_add: __ addss(lreg, rreg); break;
2226 case lir_sub: __ subss(lreg, rreg); break;
2227 case lir_mul_strictfp: // fall through
2228 case lir_mul: __ mulss(lreg, rreg); break;
2229 case lir_div_strictfp: // fall through
2230 case lir_div: __ divss(lreg, rreg); break;
2231 default: ShouldNotReachHere();
2232 }
2233 } else {
2234 Address raddr;
2235 if (right->is_single_stack()) {
2236 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2237 } else if (right->is_constant()) {
2238 // hack for now
2239 raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
2240 } else {
2241 ShouldNotReachHere();
2242 }
2243 switch (code) {
2244 case lir_add: __ addss(lreg, raddr); break;
2245 case lir_sub: __ subss(lreg, raddr); break;
2246 case lir_mul_strictfp: // fall through
2247 case lir_mul: __ mulss(lreg, raddr); break;
2248 case lir_div_strictfp: // fall through
2249 case lir_div: __ divss(lreg, raddr); break;
2250 default: ShouldNotReachHere();
2251 }
2252 }
2254 } else if (left->is_double_xmm()) {
2255 assert(left == dest, "left and dest must be equal");
2257 XMMRegister lreg = left->as_xmm_double_reg();
2258 if (right->is_double_xmm()) {
2259 XMMRegister rreg = right->as_xmm_double_reg();
2260 switch (code) {
2261 case lir_add: __ addsd(lreg, rreg); break;
2262 case lir_sub: __ subsd(lreg, rreg); break;
2263 case lir_mul_strictfp: // fall through
2264 case lir_mul: __ mulsd(lreg, rreg); break;
2265 case lir_div_strictfp: // fall through
2266 case lir_div: __ divsd(lreg, rreg); break;
2267 default: ShouldNotReachHere();
2268 }
2269 } else {
2270 Address raddr;
2271 if (right->is_double_stack()) {
2272 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2273 } else if (right->is_constant()) {
2274 // hack for now
2275 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2276 } else {
2277 ShouldNotReachHere();
2278 }
2279 switch (code) {
2280 case lir_add: __ addsd(lreg, raddr); break;
2281 case lir_sub: __ subsd(lreg, raddr); break;
2282 case lir_mul_strictfp: // fall through
2283 case lir_mul: __ mulsd(lreg, raddr); break;
2284 case lir_div_strictfp: // fall through
2285 case lir_div: __ divsd(lreg, raddr); break;
2286 default: ShouldNotReachHere();
2287 }
2288 }
2290 } else if (left->is_single_fpu()) {
2291 assert(dest->is_single_fpu(), "fpu stack allocation required");
2293 if (right->is_single_fpu()) {
2294 arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
2296 } else {
2297 assert(left->fpu_regnr() == 0, "left must be on TOS");
2298 assert(dest->fpu_regnr() == 0, "dest must be on TOS");
2300 Address raddr;
2301 if (right->is_single_stack()) {
2302 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2303 } else if (right->is_constant()) {
2304 address const_addr = float_constant(right->as_jfloat());
2305 assert(const_addr != NULL, "incorrect float/double constant maintainance");
2306 // hack for now
2307 raddr = __ as_Address(InternalAddress(const_addr));
2308 } else {
2309 ShouldNotReachHere();
2310 }
2312 switch (code) {
2313 case lir_add: __ fadd_s(raddr); break;
2314 case lir_sub: __ fsub_s(raddr); break;
2315 case lir_mul_strictfp: // fall through
2316 case lir_mul: __ fmul_s(raddr); break;
2317 case lir_div_strictfp: // fall through
2318 case lir_div: __ fdiv_s(raddr); break;
2319 default: ShouldNotReachHere();
2320 }
2321 }
2323 } else if (left->is_double_fpu()) {
2324 assert(dest->is_double_fpu(), "fpu stack allocation required");
2326 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2327 // Double values require special handling for strictfp mul/div on x86
2328 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
2329 __ fmulp(left->fpu_regnrLo() + 1);
2330 }
2332 if (right->is_double_fpu()) {
2333 arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
2335 } else {
2336 assert(left->fpu_regnrLo() == 0, "left must be on TOS");
2337 assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
2339 Address raddr;
2340 if (right->is_double_stack()) {
2341 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2342 } else if (right->is_constant()) {
2343 // hack for now
2344 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2345 } else {
2346 ShouldNotReachHere();
2347 }
2349 switch (code) {
2350 case lir_add: __ fadd_d(raddr); break;
2351 case lir_sub: __ fsub_d(raddr); break;
2352 case lir_mul_strictfp: // fall through
2353 case lir_mul: __ fmul_d(raddr); break;
2354 case lir_div_strictfp: // fall through
2355 case lir_div: __ fdiv_d(raddr); break;
2356 default: ShouldNotReachHere();
2357 }
2358 }
2360 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2361 // Double values require special handling for strictfp mul/div on x86
2362 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
2363 __ fmulp(dest->fpu_regnrLo() + 1);
2364 }
2366 } else if (left->is_single_stack() || left->is_address()) {
2367 assert(left == dest, "left and dest must be equal");
2369 Address laddr;
2370 if (left->is_single_stack()) {
2371 laddr = frame_map()->address_for_slot(left->single_stack_ix());
2372 } else if (left->is_address()) {
2373 laddr = as_Address(left->as_address_ptr());
2374 } else {
2375 ShouldNotReachHere();
2376 }
2378 if (right->is_single_cpu()) {
2379 Register rreg = right->as_register();
2380 switch (code) {
2381 case lir_add: __ addl(laddr, rreg); break;
2382 case lir_sub: __ subl(laddr, rreg); break;
2383 default: ShouldNotReachHere();
2384 }
2385 } else if (right->is_constant()) {
2386 jint c = right->as_constant_ptr()->as_jint();
2387 switch (code) {
2388 case lir_add: {
2389 __ incrementl(laddr, c);
2390 break;
2391 }
2392 case lir_sub: {
2393 __ decrementl(laddr, c);
2394 break;
2395 }
2396 default: ShouldNotReachHere();
2397 }
2398 } else {
2399 ShouldNotReachHere();
2400 }
2402 } else {
2403 ShouldNotReachHere();
2404 }
2405 }
2407 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
2408 assert(pop_fpu_stack || (left_index == dest_index || right_index == dest_index), "invalid LIR");
2409 assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
2410 assert(left_index == 0 || right_index == 0, "either must be on top of stack");
2412 bool left_is_tos = (left_index == 0);
2413 bool dest_is_tos = (dest_index == 0);
2414 int non_tos_index = (left_is_tos ? right_index : left_index);
2416 switch (code) {
2417 case lir_add:
2418 if (pop_fpu_stack) __ faddp(non_tos_index);
2419 else if (dest_is_tos) __ fadd (non_tos_index);
2420 else __ fadda(non_tos_index);
2421 break;
2423 case lir_sub:
2424 if (left_is_tos) {
2425 if (pop_fpu_stack) __ fsubrp(non_tos_index);
2426 else if (dest_is_tos) __ fsub (non_tos_index);
2427 else __ fsubra(non_tos_index);
2428 } else {
2429 if (pop_fpu_stack) __ fsubp (non_tos_index);
2430 else if (dest_is_tos) __ fsubr (non_tos_index);
2431 else __ fsuba (non_tos_index);
2432 }
2433 break;
2435 case lir_mul_strictfp: // fall through
2436 case lir_mul:
2437 if (pop_fpu_stack) __ fmulp(non_tos_index);
2438 else if (dest_is_tos) __ fmul (non_tos_index);
2439 else __ fmula(non_tos_index);
2440 break;
2442 case lir_div_strictfp: // fall through
2443 case lir_div:
2444 if (left_is_tos) {
2445 if (pop_fpu_stack) __ fdivrp(non_tos_index);
2446 else if (dest_is_tos) __ fdiv (non_tos_index);
2447 else __ fdivra(non_tos_index);
2448 } else {
2449 if (pop_fpu_stack) __ fdivp (non_tos_index);
2450 else if (dest_is_tos) __ fdivr (non_tos_index);
2451 else __ fdiva (non_tos_index);
2452 }
2453 break;
2455 case lir_rem:
2456 assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
2457 __ fremr(noreg);
2458 break;
2460 default:
2461 ShouldNotReachHere();
2462 }
2463 }
2466 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
2467 if (value->is_double_xmm()) {
2468 switch(code) {
2469 case lir_abs :
2470 {
2471 if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
2472 __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
2473 }
2474 __ andpd(dest->as_xmm_double_reg(),
2475 ExternalAddress((address)double_signmask_pool));
2476 }
2477 break;
2479 case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
2480 // all other intrinsics are not available in the SSE instruction set, so FPU is used
2481 default : ShouldNotReachHere();
2482 }
2484 } else if (value->is_double_fpu()) {
2485 assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
2486 switch(code) {
2487 case lir_log : __ flog() ; break;
2488 case lir_log10 : __ flog10() ; break;
2489 case lir_abs : __ fabs() ; break;
2490 case lir_sqrt : __ fsqrt(); break;
2491 case lir_sin :
2492 // Should consider not saving rbx, if not necessary
2493 __ trigfunc('s', op->as_Op2()->fpu_stack_size());
2494 break;
2495 case lir_cos :
2496 // Should consider not saving rbx, if not necessary
2497 assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
2498 __ trigfunc('c', op->as_Op2()->fpu_stack_size());
2499 break;
2500 case lir_tan :
2501 // Should consider not saving rbx, if not necessary
2502 __ trigfunc('t', op->as_Op2()->fpu_stack_size());
2503 break;
2504 default : ShouldNotReachHere();
2505 }
2506 } else {
2507 Unimplemented();
2508 }
2509 }
2511 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
2512 // assert(left->destroys_register(), "check");
2513 if (left->is_single_cpu()) {
2514 Register reg = left->as_register();
2515 if (right->is_constant()) {
2516 int val = right->as_constant_ptr()->as_jint();
2517 switch (code) {
2518 case lir_logic_and: __ andl (reg, val); break;
2519 case lir_logic_or: __ orl (reg, val); break;
2520 case lir_logic_xor: __ xorl (reg, val); break;
2521 default: ShouldNotReachHere();
2522 }
2523 } else if (right->is_stack()) {
2524 // added support for stack operands
2525 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2526 switch (code) {
2527 case lir_logic_and: __ andl (reg, raddr); break;
2528 case lir_logic_or: __ orl (reg, raddr); break;
2529 case lir_logic_xor: __ xorl (reg, raddr); break;
2530 default: ShouldNotReachHere();
2531 }
2532 } else {
2533 Register rright = right->as_register();
2534 switch (code) {
2535 case lir_logic_and: __ andptr (reg, rright); break;
2536 case lir_logic_or : __ orptr (reg, rright); break;
2537 case lir_logic_xor: __ xorptr (reg, rright); break;
2538 default: ShouldNotReachHere();
2539 }
2540 }
2541 move_regs(reg, dst->as_register());
2542 } else {
2543 Register l_lo = left->as_register_lo();
2544 Register l_hi = left->as_register_hi();
2545 if (right->is_constant()) {
2546 #ifdef _LP64
2547 __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
2548 switch (code) {
2549 case lir_logic_and:
2550 __ andq(l_lo, rscratch1);
2551 break;
2552 case lir_logic_or:
2553 __ orq(l_lo, rscratch1);
2554 break;
2555 case lir_logic_xor:
2556 __ xorq(l_lo, rscratch1);
2557 break;
2558 default: ShouldNotReachHere();
2559 }
2560 #else
2561 int r_lo = right->as_constant_ptr()->as_jint_lo();
2562 int r_hi = right->as_constant_ptr()->as_jint_hi();
2563 switch (code) {
2564 case lir_logic_and:
2565 __ andl(l_lo, r_lo);
2566 __ andl(l_hi, r_hi);
2567 break;
2568 case lir_logic_or:
2569 __ orl(l_lo, r_lo);
2570 __ orl(l_hi, r_hi);
2571 break;
2572 case lir_logic_xor:
2573 __ xorl(l_lo, r_lo);
2574 __ xorl(l_hi, r_hi);
2575 break;
2576 default: ShouldNotReachHere();
2577 }
2578 #endif // _LP64
2579 } else {
2580 #ifdef _LP64
2581 Register r_lo;
2582 if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
2583 r_lo = right->as_register();
2584 } else {
2585 r_lo = right->as_register_lo();
2586 }
2587 #else
2588 Register r_lo = right->as_register_lo();
2589 Register r_hi = right->as_register_hi();
2590 assert(l_lo != r_hi, "overwriting registers");
2591 #endif
2592 switch (code) {
2593 case lir_logic_and:
2594 __ andptr(l_lo, r_lo);
2595 NOT_LP64(__ andptr(l_hi, r_hi);)
2596 break;
2597 case lir_logic_or:
2598 __ orptr(l_lo, r_lo);
2599 NOT_LP64(__ orptr(l_hi, r_hi);)
2600 break;
2601 case lir_logic_xor:
2602 __ xorptr(l_lo, r_lo);
2603 NOT_LP64(__ xorptr(l_hi, r_hi);)
2604 break;
2605 default: ShouldNotReachHere();
2606 }
2607 }
2609 Register dst_lo = dst->as_register_lo();
2610 Register dst_hi = dst->as_register_hi();
2612 #ifdef _LP64
2613 move_regs(l_lo, dst_lo);
2614 #else
2615 if (dst_lo == l_hi) {
2616 assert(dst_hi != l_lo, "overwriting registers");
2617 move_regs(l_hi, dst_hi);
2618 move_regs(l_lo, dst_lo);
2619 } else {
2620 assert(dst_lo != l_hi, "overwriting registers");
2621 move_regs(l_lo, dst_lo);
2622 move_regs(l_hi, dst_hi);
2623 }
2624 #endif // _LP64
2625 }
2626 }
2629 // we assume that rax, and rdx can be overwritten
2630 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
2632 assert(left->is_single_cpu(), "left must be register");
2633 assert(right->is_single_cpu() || right->is_constant(), "right must be register or constant");
2634 assert(result->is_single_cpu(), "result must be register");
2636 // assert(left->destroys_register(), "check");
2637 // assert(right->destroys_register(), "check");
2639 Register lreg = left->as_register();
2640 Register dreg = result->as_register();
2642 if (right->is_constant()) {
2643 int divisor = right->as_constant_ptr()->as_jint();
2644 assert(divisor > 0 && is_power_of_2(divisor), "must be");
2645 if (code == lir_idiv) {
2646 assert(lreg == rax, "must be rax,");
2647 assert(temp->as_register() == rdx, "tmp register must be rdx");
2648 __ cdql(); // sign extend into rdx:rax
2649 if (divisor == 2) {
2650 __ subl(lreg, rdx);
2651 } else {
2652 __ andl(rdx, divisor - 1);
2653 __ addl(lreg, rdx);
2654 }
2655 __ sarl(lreg, log2_intptr(divisor));
2656 move_regs(lreg, dreg);
2657 } else if (code == lir_irem) {
2658 Label done;
2659 __ mov(dreg, lreg);
2660 __ andl(dreg, 0x80000000 | (divisor - 1));
2661 __ jcc(Assembler::positive, done);
2662 __ decrement(dreg);
2663 __ orl(dreg, ~(divisor - 1));
2664 __ increment(dreg);
2665 __ bind(done);
2666 } else {
2667 ShouldNotReachHere();
2668 }
2669 } else {
2670 Register rreg = right->as_register();
2671 assert(lreg == rax, "left register must be rax,");
2672 assert(rreg != rdx, "right register must not be rdx");
2673 assert(temp->as_register() == rdx, "tmp register must be rdx");
2675 move_regs(lreg, rax);
2677 int idivl_offset = __ corrected_idivl(rreg);
2678 add_debug_info_for_div0(idivl_offset, info);
2679 if (code == lir_irem) {
2680 move_regs(rdx, dreg); // result is in rdx
2681 } else {
2682 move_regs(rax, dreg);
2683 }
2684 }
2685 }
2688 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
2689 if (opr1->is_single_cpu()) {
2690 Register reg1 = opr1->as_register();
2691 if (opr2->is_single_cpu()) {
2692 // cpu register - cpu register
2693 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2694 __ cmpptr(reg1, opr2->as_register());
2695 } else {
2696 assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
2697 __ cmpl(reg1, opr2->as_register());
2698 }
2699 } else if (opr2->is_stack()) {
2700 // cpu register - stack
2701 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2702 __ cmpptr(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2703 } else {
2704 __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2705 }
2706 } else if (opr2->is_constant()) {
2707 // cpu register - constant
2708 LIR_Const* c = opr2->as_constant_ptr();
2709 if (c->type() == T_INT) {
2710 __ cmpl(reg1, c->as_jint());
2711 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2712 // In 64bit oops are single register
2713 jobject o = c->as_jobject();
2714 if (o == NULL) {
2715 __ cmpptr(reg1, (int32_t)NULL_WORD);
2716 } else {
2717 #ifdef _LP64
2718 __ movoop(rscratch1, o);
2719 __ cmpptr(reg1, rscratch1);
2720 #else
2721 __ cmpoop(reg1, c->as_jobject());
2722 #endif // _LP64
2723 }
2724 } else {
2725 ShouldNotReachHere();
2726 }
2727 // cpu register - address
2728 } else if (opr2->is_address()) {
2729 if (op->info() != NULL) {
2730 add_debug_info_for_null_check_here(op->info());
2731 }
2732 __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
2733 } else {
2734 ShouldNotReachHere();
2735 }
2737 } else if(opr1->is_double_cpu()) {
2738 Register xlo = opr1->as_register_lo();
2739 Register xhi = opr1->as_register_hi();
2740 if (opr2->is_double_cpu()) {
2741 #ifdef _LP64
2742 __ cmpptr(xlo, opr2->as_register_lo());
2743 #else
2744 // cpu register - cpu register
2745 Register ylo = opr2->as_register_lo();
2746 Register yhi = opr2->as_register_hi();
2747 __ subl(xlo, ylo);
2748 __ sbbl(xhi, yhi);
2749 if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
2750 __ orl(xhi, xlo);
2751 }
2752 #endif // _LP64
2753 } else if (opr2->is_constant()) {
2754 // cpu register - constant 0
2755 assert(opr2->as_jlong() == (jlong)0, "only handles zero");
2756 #ifdef _LP64
2757 __ cmpptr(xlo, (int32_t)opr2->as_jlong());
2758 #else
2759 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
2760 __ orl(xhi, xlo);
2761 #endif // _LP64
2762 } else {
2763 ShouldNotReachHere();
2764 }
2766 } else if (opr1->is_single_xmm()) {
2767 XMMRegister reg1 = opr1->as_xmm_float_reg();
2768 if (opr2->is_single_xmm()) {
2769 // xmm register - xmm register
2770 __ ucomiss(reg1, opr2->as_xmm_float_reg());
2771 } else if (opr2->is_stack()) {
2772 // xmm register - stack
2773 __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2774 } else if (opr2->is_constant()) {
2775 // xmm register - constant
2776 __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
2777 } else if (opr2->is_address()) {
2778 // xmm register - address
2779 if (op->info() != NULL) {
2780 add_debug_info_for_null_check_here(op->info());
2781 }
2782 __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
2783 } else {
2784 ShouldNotReachHere();
2785 }
2787 } else if (opr1->is_double_xmm()) {
2788 XMMRegister reg1 = opr1->as_xmm_double_reg();
2789 if (opr2->is_double_xmm()) {
2790 // xmm register - xmm register
2791 __ ucomisd(reg1, opr2->as_xmm_double_reg());
2792 } else if (opr2->is_stack()) {
2793 // xmm register - stack
2794 __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
2795 } else if (opr2->is_constant()) {
2796 // xmm register - constant
2797 __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
2798 } else if (opr2->is_address()) {
2799 // xmm register - address
2800 if (op->info() != NULL) {
2801 add_debug_info_for_null_check_here(op->info());
2802 }
2803 __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
2804 } else {
2805 ShouldNotReachHere();
2806 }
2808 } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
2809 assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
2810 assert(opr2->is_fpu_register(), "both must be registers");
2811 __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2813 } else if (opr1->is_address() && opr2->is_constant()) {
2814 LIR_Const* c = opr2->as_constant_ptr();
2815 #ifdef _LP64
2816 if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2817 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
2818 __ movoop(rscratch1, c->as_jobject());
2819 }
2820 #endif // LP64
2821 if (op->info() != NULL) {
2822 add_debug_info_for_null_check_here(op->info());
2823 }
2824 // special case: address - constant
2825 LIR_Address* addr = opr1->as_address_ptr();
2826 if (c->type() == T_INT) {
2827 __ cmpl(as_Address(addr), c->as_jint());
2828 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2829 #ifdef _LP64
2830 // %%% Make this explode if addr isn't reachable until we figure out a
2831 // better strategy by giving noreg as the temp for as_Address
2832 __ cmpptr(rscratch1, as_Address(addr, noreg));
2833 #else
2834 __ cmpoop(as_Address(addr), c->as_jobject());
2835 #endif // _LP64
2836 } else {
2837 ShouldNotReachHere();
2838 }
2840 } else {
2841 ShouldNotReachHere();
2842 }
2843 }
2845 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
2846 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
2847 if (left->is_single_xmm()) {
2848 assert(right->is_single_xmm(), "must match");
2849 __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2850 } else if (left->is_double_xmm()) {
2851 assert(right->is_double_xmm(), "must match");
2852 __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2854 } else {
2855 assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
2856 assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
2858 assert(left->fpu() == 0, "left must be on TOS");
2859 __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
2860 op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2861 }
2862 } else {
2863 assert(code == lir_cmp_l2i, "check");
2864 #ifdef _LP64
2865 Label done;
2866 Register dest = dst->as_register();
2867 __ cmpptr(left->as_register_lo(), right->as_register_lo());
2868 __ movl(dest, -1);
2869 __ jccb(Assembler::less, done);
2870 __ set_byte_if_not_zero(dest);
2871 __ movzbl(dest, dest);
2872 __ bind(done);
2873 #else
2874 __ lcmp2int(left->as_register_hi(),
2875 left->as_register_lo(),
2876 right->as_register_hi(),
2877 right->as_register_lo());
2878 move_regs(left->as_register_hi(), dst->as_register());
2879 #endif // _LP64
2880 }
2881 }
2884 void LIR_Assembler::align_call(LIR_Code code) {
2885 if (os::is_MP()) {
2886 // make sure that the displacement word of the call ends up word aligned
2887 int offset = __ offset();
2888 switch (code) {
2889 case lir_static_call:
2890 case lir_optvirtual_call:
2891 case lir_dynamic_call:
2892 offset += NativeCall::displacement_offset;
2893 break;
2894 case lir_icvirtual_call:
2895 offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
2896 break;
2897 case lir_virtual_call: // currently, sparc-specific for niagara
2898 default: ShouldNotReachHere();
2899 }
2900 while (offset++ % BytesPerWord != 0) {
2901 __ nop();
2902 }
2903 }
2904 }
2907 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
2908 assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2909 "must be aligned");
2910 __ call(AddressLiteral(op->addr(), rtype));
2911 add_call_info(code_offset(), op->info());
2912 }
2915 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
2916 RelocationHolder rh = virtual_call_Relocation::spec(pc());
2917 __ movoop(IC_Klass, (jobject)Universe::non_oop_word());
2918 assert(!os::is_MP() ||
2919 (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2920 "must be aligned");
2921 __ call(AddressLiteral(op->addr(), rh));
2922 add_call_info(code_offset(), op->info());
2923 }
2926 /* Currently, vtable-dispatch is only enabled for sparc platforms */
2927 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
2928 ShouldNotReachHere();
2929 }
2932 void LIR_Assembler::emit_static_call_stub() {
2933 address call_pc = __ pc();
2934 address stub = __ start_a_stub(call_stub_size);
2935 if (stub == NULL) {
2936 bailout("static call stub overflow");
2937 return;
2938 }
2940 int start = __ offset();
2941 if (os::is_MP()) {
2942 // make sure that the displacement word of the call ends up word aligned
2943 int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
2944 while (offset++ % BytesPerWord != 0) {
2945 __ nop();
2946 }
2947 }
2948 __ relocate(static_stub_Relocation::spec(call_pc));
2949 __ movoop(rbx, (jobject)NULL);
2950 // must be set to -1 at code generation time
2951 assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
2952 // On 64bit this will die since it will take a movq & jmp, must be only a jmp
2953 __ jump(RuntimeAddress(__ pc()));
2955 assert(__ offset() - start <= call_stub_size, "stub too big");
2956 __ end_a_stub();
2957 }
2960 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2961 assert(exceptionOop->as_register() == rax, "must match");
2962 assert(exceptionPC->as_register() == rdx, "must match");
2964 // exception object is not added to oop map by LinearScan
2965 // (LinearScan assumes that no oops are in fixed registers)
2966 info->add_register_oop(exceptionOop);
2967 Runtime1::StubID unwind_id;
2969 // get current pc information
2970 // pc is only needed if the method has an exception handler, the unwind code does not need it.
2971 int pc_for_athrow_offset = __ offset();
2972 InternalAddress pc_for_athrow(__ pc());
2973 __ lea(exceptionPC->as_register(), pc_for_athrow);
2974 add_call_info(pc_for_athrow_offset, info); // for exception handler
2976 __ verify_not_null_oop(rax);
2977 // search an exception handler (rax: exception oop, rdx: throwing pc)
2978 if (compilation()->has_fpu_code()) {
2979 unwind_id = Runtime1::handle_exception_id;
2980 } else {
2981 unwind_id = Runtime1::handle_exception_nofpu_id;
2982 }
2983 __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2985 // enough room for two byte trap
2986 __ nop();
2987 }
2990 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2991 assert(exceptionOop->as_register() == rax, "must match");
2993 __ jmp(_unwind_handler_entry);
2994 }
2997 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2999 // optimized version for linear scan:
3000 // * count must be already in ECX (guaranteed by LinearScan)
3001 // * left and dest must be equal
3002 // * tmp must be unused
3003 assert(count->as_register() == SHIFT_count, "count must be in ECX");
3004 assert(left == dest, "left and dest must be equal");
3005 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
3007 if (left->is_single_cpu()) {
3008 Register value = left->as_register();
3009 assert(value != SHIFT_count, "left cannot be ECX");
3011 switch (code) {
3012 case lir_shl: __ shll(value); break;
3013 case lir_shr: __ sarl(value); break;
3014 case lir_ushr: __ shrl(value); break;
3015 default: ShouldNotReachHere();
3016 }
3017 } else if (left->is_double_cpu()) {
3018 Register lo = left->as_register_lo();
3019 Register hi = left->as_register_hi();
3020 assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
3021 #ifdef _LP64
3022 switch (code) {
3023 case lir_shl: __ shlptr(lo); break;
3024 case lir_shr: __ sarptr(lo); break;
3025 case lir_ushr: __ shrptr(lo); break;
3026 default: ShouldNotReachHere();
3027 }
3028 #else
3030 switch (code) {
3031 case lir_shl: __ lshl(hi, lo); break;
3032 case lir_shr: __ lshr(hi, lo, true); break;
3033 case lir_ushr: __ lshr(hi, lo, false); break;
3034 default: ShouldNotReachHere();
3035 }
3036 #endif // LP64
3037 } else {
3038 ShouldNotReachHere();
3039 }
3040 }
3043 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
3044 if (dest->is_single_cpu()) {
3045 // first move left into dest so that left is not destroyed by the shift
3046 Register value = dest->as_register();
3047 count = count & 0x1F; // Java spec
3049 move_regs(left->as_register(), value);
3050 switch (code) {
3051 case lir_shl: __ shll(value, count); break;
3052 case lir_shr: __ sarl(value, count); break;
3053 case lir_ushr: __ shrl(value, count); break;
3054 default: ShouldNotReachHere();
3055 }
3056 } else if (dest->is_double_cpu()) {
3057 #ifndef _LP64
3058 Unimplemented();
3059 #else
3060 // first move left into dest so that left is not destroyed by the shift
3061 Register value = dest->as_register_lo();
3062 count = count & 0x1F; // Java spec
3064 move_regs(left->as_register_lo(), value);
3065 switch (code) {
3066 case lir_shl: __ shlptr(value, count); break;
3067 case lir_shr: __ sarptr(value, count); break;
3068 case lir_ushr: __ shrptr(value, count); break;
3069 default: ShouldNotReachHere();
3070 }
3071 #endif // _LP64
3072 } else {
3073 ShouldNotReachHere();
3074 }
3075 }
3078 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
3079 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3080 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3081 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3082 __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
3083 }
3086 void LIR_Assembler::store_parameter(jint c, int offset_from_rsp_in_words) {
3087 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3088 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3089 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3090 __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
3091 }
3094 void LIR_Assembler::store_parameter(jobject o, int offset_from_rsp_in_words) {
3095 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3096 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3097 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3098 __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
3099 }
3102 // This code replaces a call to arraycopy; no exception may
3103 // be thrown in this code, they must be thrown in the System.arraycopy
3104 // activation frame; we could save some checks if this would not be the case
3105 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
3106 ciArrayKlass* default_type = op->expected_type();
3107 Register src = op->src()->as_register();
3108 Register dst = op->dst()->as_register();
3109 Register src_pos = op->src_pos()->as_register();
3110 Register dst_pos = op->dst_pos()->as_register();
3111 Register length = op->length()->as_register();
3112 Register tmp = op->tmp()->as_register();
3114 CodeStub* stub = op->stub();
3115 int flags = op->flags();
3116 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
3117 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
3119 // if we don't know anything or it's an object array, just go through the generic arraycopy
3120 if (default_type == NULL) {
3121 Label done;
3122 // save outgoing arguments on stack in case call to System.arraycopy is needed
3123 // HACK ALERT. This code used to push the parameters in a hardwired fashion
3124 // for interpreter calling conventions. Now we have to do it in new style conventions.
3125 // For the moment until C1 gets the new register allocator I just force all the
3126 // args to the right place (except the register args) and then on the back side
3127 // reload the register args properly if we go slow path. Yuck
3129 // These are proper for the calling convention
3131 store_parameter(length, 2);
3132 store_parameter(dst_pos, 1);
3133 store_parameter(dst, 0);
3135 // these are just temporary placements until we need to reload
3136 store_parameter(src_pos, 3);
3137 store_parameter(src, 4);
3138 NOT_LP64(assert(src == rcx && src_pos == rdx, "mismatch in calling convention");)
3140 address entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
3142 // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
3143 #ifdef _LP64
3144 // The arguments are in java calling convention so we can trivially shift them to C
3145 // convention
3146 assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
3147 __ mov(c_rarg0, j_rarg0);
3148 assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
3149 __ mov(c_rarg1, j_rarg1);
3150 assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
3151 __ mov(c_rarg2, j_rarg2);
3152 assert_different_registers(c_rarg3, j_rarg4);
3153 __ mov(c_rarg3, j_rarg3);
3154 #ifdef _WIN64
3155 // Allocate abi space for args but be sure to keep stack aligned
3156 __ subptr(rsp, 6*wordSize);
3157 store_parameter(j_rarg4, 4);
3158 __ call(RuntimeAddress(entry));
3159 __ addptr(rsp, 6*wordSize);
3160 #else
3161 __ mov(c_rarg4, j_rarg4);
3162 __ call(RuntimeAddress(entry));
3163 #endif // _WIN64
3164 #else
3165 __ push(length);
3166 __ push(dst_pos);
3167 __ push(dst);
3168 __ push(src_pos);
3169 __ push(src);
3170 __ call_VM_leaf(entry, 5); // removes pushed parameter from the stack
3172 #endif // _LP64
3174 __ cmpl(rax, 0);
3175 __ jcc(Assembler::equal, *stub->continuation());
3177 // Reload values from the stack so they are where the stub
3178 // expects them.
3179 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3180 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3181 __ movptr (length, Address(rsp, 2*BytesPerWord));
3182 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3183 __ movptr (src, Address(rsp, 4*BytesPerWord));
3184 __ jmp(*stub->entry());
3186 __ bind(*stub->continuation());
3187 return;
3188 }
3190 assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
3192 int elem_size = type2aelembytes(basic_type);
3193 int shift_amount;
3194 Address::ScaleFactor scale;
3196 switch (elem_size) {
3197 case 1 :
3198 shift_amount = 0;
3199 scale = Address::times_1;
3200 break;
3201 case 2 :
3202 shift_amount = 1;
3203 scale = Address::times_2;
3204 break;
3205 case 4 :
3206 shift_amount = 2;
3207 scale = Address::times_4;
3208 break;
3209 case 8 :
3210 shift_amount = 3;
3211 scale = Address::times_8;
3212 break;
3213 default:
3214 ShouldNotReachHere();
3215 }
3217 Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
3218 Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
3219 Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
3220 Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
3222 // length and pos's are all sign extended at this point on 64bit
3224 // test for NULL
3225 if (flags & LIR_OpArrayCopy::src_null_check) {
3226 __ testptr(src, src);
3227 __ jcc(Assembler::zero, *stub->entry());
3228 }
3229 if (flags & LIR_OpArrayCopy::dst_null_check) {
3230 __ testptr(dst, dst);
3231 __ jcc(Assembler::zero, *stub->entry());
3232 }
3234 // check if negative
3235 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
3236 __ testl(src_pos, src_pos);
3237 __ jcc(Assembler::less, *stub->entry());
3238 }
3239 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
3240 __ testl(dst_pos, dst_pos);
3241 __ jcc(Assembler::less, *stub->entry());
3242 }
3243 if (flags & LIR_OpArrayCopy::length_positive_check) {
3244 __ testl(length, length);
3245 __ jcc(Assembler::less, *stub->entry());
3246 }
3248 if (flags & LIR_OpArrayCopy::src_range_check) {
3249 __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
3250 __ cmpl(tmp, src_length_addr);
3251 __ jcc(Assembler::above, *stub->entry());
3252 }
3253 if (flags & LIR_OpArrayCopy::dst_range_check) {
3254 __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
3255 __ cmpl(tmp, dst_length_addr);
3256 __ jcc(Assembler::above, *stub->entry());
3257 }
3259 if (flags & LIR_OpArrayCopy::type_check) {
3260 if (UseCompressedOops) {
3261 __ movl(tmp, src_klass_addr);
3262 __ cmpl(tmp, dst_klass_addr);
3263 } else {
3264 __ movptr(tmp, src_klass_addr);
3265 __ cmpptr(tmp, dst_klass_addr);
3266 }
3267 __ jcc(Assembler::notEqual, *stub->entry());
3268 }
3270 #ifdef ASSERT
3271 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
3272 // Sanity check the known type with the incoming class. For the
3273 // primitive case the types must match exactly with src.klass and
3274 // dst.klass each exactly matching the default type. For the
3275 // object array case, if no type check is needed then either the
3276 // dst type is exactly the expected type and the src type is a
3277 // subtype which we can't check or src is the same array as dst
3278 // but not necessarily exactly of type default_type.
3279 Label known_ok, halt;
3280 __ movoop(tmp, default_type->constant_encoding());
3281 #ifdef _LP64
3282 if (UseCompressedOops) {
3283 __ encode_heap_oop(tmp);
3284 }
3285 #endif
3287 if (basic_type != T_OBJECT) {
3289 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3290 else __ cmpptr(tmp, dst_klass_addr);
3291 __ jcc(Assembler::notEqual, halt);
3292 if (UseCompressedOops) __ cmpl(tmp, src_klass_addr);
3293 else __ cmpptr(tmp, src_klass_addr);
3294 __ jcc(Assembler::equal, known_ok);
3295 } else {
3296 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3297 else __ cmpptr(tmp, dst_klass_addr);
3298 __ jcc(Assembler::equal, known_ok);
3299 __ cmpptr(src, dst);
3300 __ jcc(Assembler::equal, known_ok);
3301 }
3302 __ bind(halt);
3303 __ stop("incorrect type information in arraycopy");
3304 __ bind(known_ok);
3305 }
3306 #endif
3308 if (shift_amount > 0 && basic_type != T_OBJECT) {
3309 __ shlptr(length, shift_amount);
3310 }
3312 #ifdef _LP64
3313 assert_different_registers(c_rarg0, dst, dst_pos, length);
3314 __ movl2ptr(src_pos, src_pos); //higher 32bits must be null
3315 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3316 assert_different_registers(c_rarg1, length);
3317 __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null
3318 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3319 __ mov(c_rarg2, length);
3321 #else
3322 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3323 store_parameter(tmp, 0);
3324 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3325 store_parameter(tmp, 1);
3326 store_parameter(length, 2);
3327 #endif // _LP64
3328 if (basic_type == T_OBJECT) {
3329 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::oop_arraycopy), 0);
3330 } else {
3331 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::primitive_arraycopy), 0);
3332 }
3334 __ bind(*stub->continuation());
3335 }
3338 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
3339 Register obj = op->obj_opr()->as_register(); // may not be an oop
3340 Register hdr = op->hdr_opr()->as_register();
3341 Register lock = op->lock_opr()->as_register();
3342 if (!UseFastLocking) {
3343 __ jmp(*op->stub()->entry());
3344 } else if (op->code() == lir_lock) {
3345 Register scratch = noreg;
3346 if (UseBiasedLocking) {
3347 scratch = op->scratch_opr()->as_register();
3348 }
3349 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3350 // add debug info for NullPointerException only if one is possible
3351 int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
3352 if (op->info() != NULL) {
3353 add_debug_info_for_null_check(null_check_offset, op->info());
3354 }
3355 // done
3356 } else if (op->code() == lir_unlock) {
3357 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3358 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
3359 } else {
3360 Unimplemented();
3361 }
3362 __ bind(*op->stub()->continuation());
3363 }
3366 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
3367 ciMethod* method = op->profiled_method();
3368 int bci = op->profiled_bci();
3370 // Update counter for all call types
3371 ciMethodData* md = method->method_data_or_null();
3372 assert(md != NULL, "Sanity");
3373 ciProfileData* data = md->bci_to_data(bci);
3374 assert(data->is_CounterData(), "need CounterData for calls");
3375 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
3376 Register mdo = op->mdo()->as_register();
3377 __ movoop(mdo, md->constant_encoding());
3378 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
3379 Bytecodes::Code bc = method->java_code_at_bci(bci);
3380 // Perform additional virtual call profiling for invokevirtual and
3381 // invokeinterface bytecodes
3382 if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3383 C1ProfileVirtualCalls) {
3384 assert(op->recv()->is_single_cpu(), "recv must be allocated");
3385 Register recv = op->recv()->as_register();
3386 assert_different_registers(mdo, recv);
3387 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3388 ciKlass* known_klass = op->known_holder();
3389 if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
3390 // We know the type that will be seen at this call site; we can
3391 // statically update the methodDataOop rather than needing to do
3392 // dynamic tests on the receiver type
3394 // NOTE: we should probably put a lock around this search to
3395 // avoid collisions by concurrent compilations
3396 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3397 uint i;
3398 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3399 ciKlass* receiver = vc_data->receiver(i);
3400 if (known_klass->equals(receiver)) {
3401 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3402 __ addptr(data_addr, DataLayout::counter_increment);
3403 return;
3404 }
3405 }
3407 // Receiver type not found in profile data; select an empty slot
3409 // Note that this is less efficient than it should be because it
3410 // always does a write to the receiver part of the
3411 // VirtualCallData rather than just the first time
3412 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3413 ciKlass* receiver = vc_data->receiver(i);
3414 if (receiver == NULL) {
3415 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3416 __ movoop(recv_addr, known_klass->constant_encoding());
3417 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3418 __ addptr(data_addr, DataLayout::counter_increment);
3419 return;
3420 }
3421 }
3422 } else {
3423 __ load_klass(recv, recv);
3424 Label update_done;
3425 type_profile_helper(mdo, md, data, recv, &update_done);
3426 // Receiver did not match any saved receiver and there is no empty row for it.
3427 // Increment total counter to indicate polymorphic case.
3428 __ addptr(counter_addr, DataLayout::counter_increment);
3430 __ bind(update_done);
3431 }
3432 } else {
3433 // Static call
3434 __ addptr(counter_addr, DataLayout::counter_increment);
3435 }
3436 }
3438 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
3439 Unimplemented();
3440 }
3443 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
3444 __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
3445 }
3448 void LIR_Assembler::align_backward_branch_target() {
3449 __ align(BytesPerWord);
3450 }
3453 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3454 if (left->is_single_cpu()) {
3455 __ negl(left->as_register());
3456 move_regs(left->as_register(), dest->as_register());
3458 } else if (left->is_double_cpu()) {
3459 Register lo = left->as_register_lo();
3460 #ifdef _LP64
3461 Register dst = dest->as_register_lo();
3462 __ movptr(dst, lo);
3463 __ negptr(dst);
3464 #else
3465 Register hi = left->as_register_hi();
3466 __ lneg(hi, lo);
3467 if (dest->as_register_lo() == hi) {
3468 assert(dest->as_register_hi() != lo, "destroying register");
3469 move_regs(hi, dest->as_register_hi());
3470 move_regs(lo, dest->as_register_lo());
3471 } else {
3472 move_regs(lo, dest->as_register_lo());
3473 move_regs(hi, dest->as_register_hi());
3474 }
3475 #endif // _LP64
3477 } else if (dest->is_single_xmm()) {
3478 if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
3479 __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
3480 }
3481 __ xorps(dest->as_xmm_float_reg(),
3482 ExternalAddress((address)float_signflip_pool));
3484 } else if (dest->is_double_xmm()) {
3485 if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
3486 __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
3487 }
3488 __ xorpd(dest->as_xmm_double_reg(),
3489 ExternalAddress((address)double_signflip_pool));
3491 } else if (left->is_single_fpu() || left->is_double_fpu()) {
3492 assert(left->fpu() == 0, "arg must be on TOS");
3493 assert(dest->fpu() == 0, "dest must be TOS");
3494 __ fchs();
3496 } else {
3497 ShouldNotReachHere();
3498 }
3499 }
3502 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
3503 assert(addr->is_address() && dest->is_register(), "check");
3504 Register reg;
3505 reg = dest->as_pointer_register();
3506 __ lea(reg, as_Address(addr->as_address_ptr()));
3507 }
3511 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3512 assert(!tmp->is_valid(), "don't need temporary");
3513 __ call(RuntimeAddress(dest));
3514 if (info != NULL) {
3515 add_call_info_here(info);
3516 }
3517 }
3520 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3521 assert(type == T_LONG, "only for volatile long fields");
3523 if (info != NULL) {
3524 add_debug_info_for_null_check_here(info);
3525 }
3527 if (src->is_double_xmm()) {
3528 if (dest->is_double_cpu()) {
3529 #ifdef _LP64
3530 __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
3531 #else
3532 __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
3533 __ psrlq(src->as_xmm_double_reg(), 32);
3534 __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
3535 #endif // _LP64
3536 } else if (dest->is_double_stack()) {
3537 __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
3538 } else if (dest->is_address()) {
3539 __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
3540 } else {
3541 ShouldNotReachHere();
3542 }
3544 } else if (dest->is_double_xmm()) {
3545 if (src->is_double_stack()) {
3546 __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
3547 } else if (src->is_address()) {
3548 __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
3549 } else {
3550 ShouldNotReachHere();
3551 }
3553 } else if (src->is_double_fpu()) {
3554 assert(src->fpu_regnrLo() == 0, "must be TOS");
3555 if (dest->is_double_stack()) {
3556 __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
3557 } else if (dest->is_address()) {
3558 __ fistp_d(as_Address(dest->as_address_ptr()));
3559 } else {
3560 ShouldNotReachHere();
3561 }
3563 } else if (dest->is_double_fpu()) {
3564 assert(dest->fpu_regnrLo() == 0, "must be TOS");
3565 if (src->is_double_stack()) {
3566 __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
3567 } else if (src->is_address()) {
3568 __ fild_d(as_Address(src->as_address_ptr()));
3569 } else {
3570 ShouldNotReachHere();
3571 }
3572 } else {
3573 ShouldNotReachHere();
3574 }
3575 }
3578 void LIR_Assembler::membar() {
3579 // QQQ sparc TSO uses this,
3580 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
3581 }
3583 void LIR_Assembler::membar_acquire() {
3584 // No x86 machines currently require load fences
3585 // __ load_fence();
3586 }
3588 void LIR_Assembler::membar_release() {
3589 // No x86 machines currently require store fences
3590 // __ store_fence();
3591 }
3593 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3594 assert(result_reg->is_register(), "check");
3595 #ifdef _LP64
3596 // __ get_thread(result_reg->as_register_lo());
3597 __ mov(result_reg->as_register(), r15_thread);
3598 #else
3599 __ get_thread(result_reg->as_register());
3600 #endif // _LP64
3601 }
3604 void LIR_Assembler::peephole(LIR_List*) {
3605 // do nothing for now
3606 }
3609 #undef __