Sun, 27 Mar 2011 13:17:37 -0700
6964776: c2 should ensure the polling page is reachable on 64 bit
Summary: Materialize the pointer to the polling page in a register instead of using rip-relative addressing when the distance from the code cache is larger than disp32.
Reviewed-by: never, kvn
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_from_callee_id)));
460 __ should_not_reach_here();
461 assert(code_offset() - offset <= exception_handler_size, "overflow");
462 __ end_a_stub();
464 return offset;
465 }
468 // Emit the code to remove the frame from the stack in the exception
469 // unwind path.
470 int LIR_Assembler::emit_unwind_handler() {
471 #ifndef PRODUCT
472 if (CommentedAssembly) {
473 _masm->block_comment("Unwind handler");
474 }
475 #endif
477 int offset = code_offset();
479 // Fetch the exception from TLS and clear out exception related thread state
480 __ get_thread(rsi);
481 __ movptr(rax, Address(rsi, JavaThread::exception_oop_offset()));
482 __ movptr(Address(rsi, JavaThread::exception_oop_offset()), (int32_t)NULL_WORD);
483 __ movptr(Address(rsi, JavaThread::exception_pc_offset()), (int32_t)NULL_WORD);
485 __ bind(_unwind_handler_entry);
486 __ verify_not_null_oop(rax);
487 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
488 __ mov(rsi, rax); // Preserve the exception
489 }
491 // Preform needed unlocking
492 MonitorExitStub* stub = NULL;
493 if (method()->is_synchronized()) {
494 monitor_address(0, FrameMap::rax_opr);
495 stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
496 __ unlock_object(rdi, rbx, rax, *stub->entry());
497 __ bind(*stub->continuation());
498 }
500 if (compilation()->env()->dtrace_method_probes()) {
501 __ get_thread(rax);
502 __ movptr(Address(rsp, 0), rax);
503 __ movoop(Address(rsp, sizeof(void*)), method()->constant_encoding());
504 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
505 }
507 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
508 __ mov(rax, rsi); // Restore the exception
509 }
511 // remove the activation and dispatch to the unwind handler
512 __ remove_frame(initial_frame_size_in_bytes());
513 __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
515 // Emit the slow path assembly
516 if (stub != NULL) {
517 stub->emit_code(this);
518 }
520 return offset;
521 }
524 int LIR_Assembler::emit_deopt_handler() {
525 // if the last instruction is a call (typically to do a throw which
526 // is coming at the end after block reordering) the return address
527 // must still point into the code area in order to avoid assertion
528 // failures when searching for the corresponding bci => add a nop
529 // (was bug 5/14/1999 - gri)
530 __ nop();
532 // generate code for exception handler
533 address handler_base = __ start_a_stub(deopt_handler_size);
534 if (handler_base == NULL) {
535 // not enough space left for the handler
536 bailout("deopt handler overflow");
537 return -1;
538 }
540 int offset = code_offset();
541 InternalAddress here(__ pc());
543 __ pushptr(here.addr());
544 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
546 assert(code_offset() - offset <= deopt_handler_size, "overflow");
547 __ end_a_stub();
549 return offset;
550 }
553 // This is the fast version of java.lang.String.compare; it has not
554 // OSR-entry and therefore, we generate a slow version for OSR's
555 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
556 __ movptr (rbx, rcx); // receiver is in rcx
557 __ movptr (rax, arg1->as_register());
559 // Get addresses of first characters from both Strings
560 __ load_heap_oop(rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
561 __ movptr (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
562 __ lea (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
565 // rbx, may be NULL
566 add_debug_info_for_null_check_here(info);
567 __ load_heap_oop(rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
568 __ movptr (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
569 __ lea (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
571 // compute minimum length (in rax) and difference of lengths (on top of stack)
572 if (VM_Version::supports_cmov()) {
573 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
574 __ movl (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
575 __ mov (rcx, rbx);
576 __ subptr (rbx, rax); // subtract lengths
577 __ push (rbx); // result
578 __ cmov (Assembler::lessEqual, rax, rcx);
579 } else {
580 Label L;
581 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
582 __ movl (rcx, Address(rax, java_lang_String::count_offset_in_bytes()));
583 __ mov (rax, rbx);
584 __ subptr (rbx, rcx);
585 __ push (rbx);
586 __ jcc (Assembler::lessEqual, L);
587 __ mov (rax, rcx);
588 __ bind (L);
589 }
590 // is minimum length 0?
591 Label noLoop, haveResult;
592 __ testptr (rax, rax);
593 __ jcc (Assembler::zero, noLoop);
595 // compare first characters
596 __ load_unsigned_short(rcx, Address(rdi, 0));
597 __ load_unsigned_short(rbx, Address(rsi, 0));
598 __ subl(rcx, rbx);
599 __ jcc(Assembler::notZero, haveResult);
600 // starting loop
601 __ decrement(rax); // we already tested index: skip one
602 __ jcc(Assembler::zero, noLoop);
604 // set rsi.edi to the end of the arrays (arrays have same length)
605 // negate the index
607 __ lea(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
608 __ lea(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
609 __ negptr(rax);
611 // compare the strings in a loop
613 Label loop;
614 __ align(wordSize);
615 __ bind(loop);
616 __ load_unsigned_short(rcx, Address(rdi, rax, Address::times_2, 0));
617 __ load_unsigned_short(rbx, Address(rsi, rax, Address::times_2, 0));
618 __ subl(rcx, rbx);
619 __ jcc(Assembler::notZero, haveResult);
620 __ increment(rax);
621 __ jcc(Assembler::notZero, loop);
623 // strings are equal up to min length
625 __ bind(noLoop);
626 __ pop(rax);
627 return_op(LIR_OprFact::illegalOpr);
629 __ bind(haveResult);
630 // leave instruction is going to discard the TOS value
631 __ mov (rax, rcx); // result of call is in rax,
632 }
635 void LIR_Assembler::return_op(LIR_Opr result) {
636 assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
637 if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
638 assert(result->fpu() == 0, "result must already be on TOS");
639 }
641 // Pop the stack before the safepoint code
642 __ remove_frame(initial_frame_size_in_bytes());
644 bool result_is_oop = result->is_valid() ? result->is_oop() : false;
646 // Note: we do not need to round double result; float result has the right precision
647 // the poll sets the condition code, but no data registers
648 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
649 relocInfo::poll_return_type);
651 if (Assembler::is_polling_page_far()) {
652 __ lea(rscratch1, polling_page);
653 __ relocate(relocInfo::poll_return_type);
654 __ testl(rax, Address(rscratch1, 0));
655 } else {
656 __ testl(rax, polling_page);
657 }
658 __ ret(0);
659 }
662 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
663 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
664 relocInfo::poll_type);
665 guarantee(info != NULL, "Shouldn't be NULL");
666 int offset = __ offset();
667 if (Assembler::is_polling_page_far()) {
668 __ lea(rscratch1, polling_page);
669 offset = __ offset();
670 add_debug_info_for_branch(info);
671 __ testl(rax, Address(rscratch1, 0));
672 } else {
673 add_debug_info_for_branch(info);
674 __ testl(rax, polling_page);
675 }
676 return offset;
677 }
680 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
681 if (from_reg != to_reg) __ mov(to_reg, from_reg);
682 }
684 void LIR_Assembler::swap_reg(Register a, Register b) {
685 __ xchgptr(a, b);
686 }
689 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
690 assert(src->is_constant(), "should not call otherwise");
691 assert(dest->is_register(), "should not call otherwise");
692 LIR_Const* c = src->as_constant_ptr();
694 switch (c->type()) {
695 case T_INT: {
696 assert(patch_code == lir_patch_none, "no patching handled here");
697 __ movl(dest->as_register(), c->as_jint());
698 break;
699 }
701 case T_ADDRESS: {
702 assert(patch_code == lir_patch_none, "no patching handled here");
703 __ movptr(dest->as_register(), c->as_jint());
704 break;
705 }
707 case T_LONG: {
708 assert(patch_code == lir_patch_none, "no patching handled here");
709 #ifdef _LP64
710 __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
711 #else
712 __ movptr(dest->as_register_lo(), c->as_jint_lo());
713 __ movptr(dest->as_register_hi(), c->as_jint_hi());
714 #endif // _LP64
715 break;
716 }
718 case T_OBJECT: {
719 if (patch_code != lir_patch_none) {
720 jobject2reg_with_patching(dest->as_register(), info);
721 } else {
722 __ movoop(dest->as_register(), c->as_jobject());
723 }
724 break;
725 }
727 case T_FLOAT: {
728 if (dest->is_single_xmm()) {
729 if (c->is_zero_float()) {
730 __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
731 } else {
732 __ movflt(dest->as_xmm_float_reg(),
733 InternalAddress(float_constant(c->as_jfloat())));
734 }
735 } else {
736 assert(dest->is_single_fpu(), "must be");
737 assert(dest->fpu_regnr() == 0, "dest must be TOS");
738 if (c->is_zero_float()) {
739 __ fldz();
740 } else if (c->is_one_float()) {
741 __ fld1();
742 } else {
743 __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
744 }
745 }
746 break;
747 }
749 case T_DOUBLE: {
750 if (dest->is_double_xmm()) {
751 if (c->is_zero_double()) {
752 __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
753 } else {
754 __ movdbl(dest->as_xmm_double_reg(),
755 InternalAddress(double_constant(c->as_jdouble())));
756 }
757 } else {
758 assert(dest->is_double_fpu(), "must be");
759 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
760 if (c->is_zero_double()) {
761 __ fldz();
762 } else if (c->is_one_double()) {
763 __ fld1();
764 } else {
765 __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
766 }
767 }
768 break;
769 }
771 default:
772 ShouldNotReachHere();
773 }
774 }
776 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
777 assert(src->is_constant(), "should not call otherwise");
778 assert(dest->is_stack(), "should not call otherwise");
779 LIR_Const* c = src->as_constant_ptr();
781 switch (c->type()) {
782 case T_INT: // fall through
783 case T_FLOAT:
784 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
785 break;
787 case T_ADDRESS:
788 __ movptr(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
789 break;
791 case T_OBJECT:
792 __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
793 break;
795 case T_LONG: // fall through
796 case T_DOUBLE:
797 #ifdef _LP64
798 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
799 lo_word_offset_in_bytes), (intptr_t)c->as_jlong_bits());
800 #else
801 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
802 lo_word_offset_in_bytes), c->as_jint_lo_bits());
803 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
804 hi_word_offset_in_bytes), c->as_jint_hi_bits());
805 #endif // _LP64
806 break;
808 default:
809 ShouldNotReachHere();
810 }
811 }
813 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
814 assert(src->is_constant(), "should not call otherwise");
815 assert(dest->is_address(), "should not call otherwise");
816 LIR_Const* c = src->as_constant_ptr();
817 LIR_Address* addr = dest->as_address_ptr();
819 int null_check_here = code_offset();
820 switch (type) {
821 case T_INT: // fall through
822 case T_FLOAT:
823 __ movl(as_Address(addr), c->as_jint_bits());
824 break;
826 case T_ADDRESS:
827 __ movptr(as_Address(addr), c->as_jint_bits());
828 break;
830 case T_OBJECT: // fall through
831 case T_ARRAY:
832 if (c->as_jobject() == NULL) {
833 if (UseCompressedOops && !wide) {
834 __ movl(as_Address(addr), (int32_t)NULL_WORD);
835 } else {
836 __ movptr(as_Address(addr), NULL_WORD);
837 }
838 } else {
839 if (is_literal_address(addr)) {
840 ShouldNotReachHere();
841 __ movoop(as_Address(addr, noreg), c->as_jobject());
842 } else {
843 #ifdef _LP64
844 __ movoop(rscratch1, c->as_jobject());
845 if (UseCompressedOops && !wide) {
846 __ encode_heap_oop(rscratch1);
847 null_check_here = code_offset();
848 __ movl(as_Address_lo(addr), rscratch1);
849 } else {
850 null_check_here = code_offset();
851 __ movptr(as_Address_lo(addr), rscratch1);
852 }
853 #else
854 __ movoop(as_Address(addr), c->as_jobject());
855 #endif
856 }
857 }
858 break;
860 case T_LONG: // fall through
861 case T_DOUBLE:
862 #ifdef _LP64
863 if (is_literal_address(addr)) {
864 ShouldNotReachHere();
865 __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
866 } else {
867 __ movptr(r10, (intptr_t)c->as_jlong_bits());
868 null_check_here = code_offset();
869 __ movptr(as_Address_lo(addr), r10);
870 }
871 #else
872 // Always reachable in 32bit so this doesn't produce useless move literal
873 __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
874 __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
875 #endif // _LP64
876 break;
878 case T_BOOLEAN: // fall through
879 case T_BYTE:
880 __ movb(as_Address(addr), c->as_jint() & 0xFF);
881 break;
883 case T_CHAR: // fall through
884 case T_SHORT:
885 __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
886 break;
888 default:
889 ShouldNotReachHere();
890 };
892 if (info != NULL) {
893 add_debug_info_for_null_check(null_check_here, info);
894 }
895 }
898 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
899 assert(src->is_register(), "should not call otherwise");
900 assert(dest->is_register(), "should not call otherwise");
902 // move between cpu-registers
903 if (dest->is_single_cpu()) {
904 #ifdef _LP64
905 if (src->type() == T_LONG) {
906 // Can do LONG -> OBJECT
907 move_regs(src->as_register_lo(), dest->as_register());
908 return;
909 }
910 #endif
911 assert(src->is_single_cpu(), "must match");
912 if (src->type() == T_OBJECT) {
913 __ verify_oop(src->as_register());
914 }
915 move_regs(src->as_register(), dest->as_register());
917 } else if (dest->is_double_cpu()) {
918 #ifdef _LP64
919 if (src->type() == T_OBJECT || src->type() == T_ARRAY) {
920 // Surprising to me but we can see move of a long to t_object
921 __ verify_oop(src->as_register());
922 move_regs(src->as_register(), dest->as_register_lo());
923 return;
924 }
925 #endif
926 assert(src->is_double_cpu(), "must match");
927 Register f_lo = src->as_register_lo();
928 Register f_hi = src->as_register_hi();
929 Register t_lo = dest->as_register_lo();
930 Register t_hi = dest->as_register_hi();
931 #ifdef _LP64
932 assert(f_hi == f_lo, "must be same");
933 assert(t_hi == t_lo, "must be same");
934 move_regs(f_lo, t_lo);
935 #else
936 assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
939 if (f_lo == t_hi && f_hi == t_lo) {
940 swap_reg(f_lo, f_hi);
941 } else if (f_hi == t_lo) {
942 assert(f_lo != t_hi, "overwriting register");
943 move_regs(f_hi, t_hi);
944 move_regs(f_lo, t_lo);
945 } else {
946 assert(f_hi != t_lo, "overwriting register");
947 move_regs(f_lo, t_lo);
948 move_regs(f_hi, t_hi);
949 }
950 #endif // LP64
952 // special moves from fpu-register to xmm-register
953 // necessary for method results
954 } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
955 __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
956 __ fld_s(Address(rsp, 0));
957 } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
958 __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
959 __ fld_d(Address(rsp, 0));
960 } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
961 __ fstp_s(Address(rsp, 0));
962 __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
963 } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
964 __ fstp_d(Address(rsp, 0));
965 __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
967 // move between xmm-registers
968 } else if (dest->is_single_xmm()) {
969 assert(src->is_single_xmm(), "must match");
970 __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
971 } else if (dest->is_double_xmm()) {
972 assert(src->is_double_xmm(), "must match");
973 __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
975 // move between fpu-registers (no instruction necessary because of fpu-stack)
976 } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
977 assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
978 assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
979 } else {
980 ShouldNotReachHere();
981 }
982 }
984 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
985 assert(src->is_register(), "should not call otherwise");
986 assert(dest->is_stack(), "should not call otherwise");
988 if (src->is_single_cpu()) {
989 Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
990 if (type == T_OBJECT || type == T_ARRAY) {
991 __ verify_oop(src->as_register());
992 __ movptr (dst, src->as_register());
993 } else {
994 __ movl (dst, src->as_register());
995 }
997 } else if (src->is_double_cpu()) {
998 Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
999 Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
1000 __ movptr (dstLO, src->as_register_lo());
1001 NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
1003 } else if (src->is_single_xmm()) {
1004 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1005 __ movflt(dst_addr, src->as_xmm_float_reg());
1007 } else if (src->is_double_xmm()) {
1008 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1009 __ movdbl(dst_addr, src->as_xmm_double_reg());
1011 } else if (src->is_single_fpu()) {
1012 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1013 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1014 if (pop_fpu_stack) __ fstp_s (dst_addr);
1015 else __ fst_s (dst_addr);
1017 } else if (src->is_double_fpu()) {
1018 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1019 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1020 if (pop_fpu_stack) __ fstp_d (dst_addr);
1021 else __ fst_d (dst_addr);
1023 } else {
1024 ShouldNotReachHere();
1025 }
1026 }
1029 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 */) {
1030 LIR_Address* to_addr = dest->as_address_ptr();
1031 PatchingStub* patch = NULL;
1032 Register compressed_src = rscratch1;
1034 if (type == T_ARRAY || type == T_OBJECT) {
1035 __ verify_oop(src->as_register());
1036 #ifdef _LP64
1037 if (UseCompressedOops && !wide) {
1038 __ movptr(compressed_src, src->as_register());
1039 __ encode_heap_oop(compressed_src);
1040 }
1041 #endif
1042 }
1044 if (patch_code != lir_patch_none) {
1045 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1046 Address toa = as_Address(to_addr);
1047 assert(toa.disp() != 0, "must have");
1048 }
1050 int null_check_here = code_offset();
1051 switch (type) {
1052 case T_FLOAT: {
1053 if (src->is_single_xmm()) {
1054 __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
1055 } else {
1056 assert(src->is_single_fpu(), "must be");
1057 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1058 if (pop_fpu_stack) __ fstp_s(as_Address(to_addr));
1059 else __ fst_s (as_Address(to_addr));
1060 }
1061 break;
1062 }
1064 case T_DOUBLE: {
1065 if (src->is_double_xmm()) {
1066 __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
1067 } else {
1068 assert(src->is_double_fpu(), "must be");
1069 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1070 if (pop_fpu_stack) __ fstp_d(as_Address(to_addr));
1071 else __ fst_d (as_Address(to_addr));
1072 }
1073 break;
1074 }
1076 case T_ARRAY: // fall through
1077 case T_OBJECT: // fall through
1078 if (UseCompressedOops && !wide) {
1079 __ movl(as_Address(to_addr), compressed_src);
1080 } else {
1081 __ movptr(as_Address(to_addr), src->as_register());
1082 }
1083 break;
1084 case T_ADDRESS:
1085 __ movptr(as_Address(to_addr), src->as_register());
1086 break;
1087 case T_INT:
1088 __ movl(as_Address(to_addr), src->as_register());
1089 break;
1091 case T_LONG: {
1092 Register from_lo = src->as_register_lo();
1093 Register from_hi = src->as_register_hi();
1094 #ifdef _LP64
1095 __ movptr(as_Address_lo(to_addr), from_lo);
1096 #else
1097 Register base = to_addr->base()->as_register();
1098 Register index = noreg;
1099 if (to_addr->index()->is_register()) {
1100 index = to_addr->index()->as_register();
1101 }
1102 if (base == from_lo || index == from_lo) {
1103 assert(base != from_hi, "can't be");
1104 assert(index == noreg || (index != base && index != from_hi), "can't handle this");
1105 __ movl(as_Address_hi(to_addr), from_hi);
1106 if (patch != NULL) {
1107 patching_epilog(patch, lir_patch_high, base, info);
1108 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1109 patch_code = lir_patch_low;
1110 }
1111 __ movl(as_Address_lo(to_addr), from_lo);
1112 } else {
1113 assert(index == noreg || (index != base && index != from_lo), "can't handle this");
1114 __ movl(as_Address_lo(to_addr), from_lo);
1115 if (patch != NULL) {
1116 patching_epilog(patch, lir_patch_low, base, info);
1117 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1118 patch_code = lir_patch_high;
1119 }
1120 __ movl(as_Address_hi(to_addr), from_hi);
1121 }
1122 #endif // _LP64
1123 break;
1124 }
1126 case T_BYTE: // fall through
1127 case T_BOOLEAN: {
1128 Register src_reg = src->as_register();
1129 Address dst_addr = as_Address(to_addr);
1130 assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
1131 __ movb(dst_addr, src_reg);
1132 break;
1133 }
1135 case T_CHAR: // fall through
1136 case T_SHORT:
1137 __ movw(as_Address(to_addr), src->as_register());
1138 break;
1140 default:
1141 ShouldNotReachHere();
1142 }
1143 if (info != NULL) {
1144 add_debug_info_for_null_check(null_check_here, info);
1145 }
1147 if (patch_code != lir_patch_none) {
1148 patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
1149 }
1150 }
1153 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1154 assert(src->is_stack(), "should not call otherwise");
1155 assert(dest->is_register(), "should not call otherwise");
1157 if (dest->is_single_cpu()) {
1158 if (type == T_ARRAY || type == T_OBJECT) {
1159 __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1160 __ verify_oop(dest->as_register());
1161 } else {
1162 __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1163 }
1165 } else if (dest->is_double_cpu()) {
1166 Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
1167 Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
1168 __ movptr(dest->as_register_lo(), src_addr_LO);
1169 NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
1171 } else if (dest->is_single_xmm()) {
1172 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1173 __ movflt(dest->as_xmm_float_reg(), src_addr);
1175 } else if (dest->is_double_xmm()) {
1176 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1177 __ movdbl(dest->as_xmm_double_reg(), src_addr);
1179 } else if (dest->is_single_fpu()) {
1180 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1181 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1182 __ fld_s(src_addr);
1184 } else if (dest->is_double_fpu()) {
1185 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1186 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1187 __ fld_d(src_addr);
1189 } else {
1190 ShouldNotReachHere();
1191 }
1192 }
1195 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1196 if (src->is_single_stack()) {
1197 if (type == T_OBJECT || type == T_ARRAY) {
1198 __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
1199 __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
1200 } else {
1201 #ifndef _LP64
1202 __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
1203 __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
1204 #else
1205 //no pushl on 64bits
1206 __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix()));
1207 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1);
1208 #endif
1209 }
1211 } else if (src->is_double_stack()) {
1212 #ifdef _LP64
1213 __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
1214 __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
1215 #else
1216 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
1217 // push and pop the part at src + wordSize, adding wordSize for the previous push
1218 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 2 * wordSize));
1219 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 2 * wordSize));
1220 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
1221 #endif // _LP64
1223 } else {
1224 ShouldNotReachHere();
1225 }
1226 }
1229 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool /* unaligned */) {
1230 assert(src->is_address(), "should not call otherwise");
1231 assert(dest->is_register(), "should not call otherwise");
1233 LIR_Address* addr = src->as_address_ptr();
1234 Address from_addr = as_Address(addr);
1236 switch (type) {
1237 case T_BOOLEAN: // fall through
1238 case T_BYTE: // fall through
1239 case T_CHAR: // fall through
1240 case T_SHORT:
1241 if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
1242 // on pre P6 processors we may get partial register stalls
1243 // so blow away the value of to_rinfo before loading a
1244 // partial word into it. Do it here so that it precedes
1245 // the potential patch point below.
1246 __ xorptr(dest->as_register(), dest->as_register());
1247 }
1248 break;
1249 }
1251 PatchingStub* patch = NULL;
1252 if (patch_code != lir_patch_none) {
1253 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1254 assert(from_addr.disp() != 0, "must have");
1255 }
1256 if (info != NULL) {
1257 add_debug_info_for_null_check_here(info);
1258 }
1260 switch (type) {
1261 case T_FLOAT: {
1262 if (dest->is_single_xmm()) {
1263 __ movflt(dest->as_xmm_float_reg(), from_addr);
1264 } else {
1265 assert(dest->is_single_fpu(), "must be");
1266 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1267 __ fld_s(from_addr);
1268 }
1269 break;
1270 }
1272 case T_DOUBLE: {
1273 if (dest->is_double_xmm()) {
1274 __ movdbl(dest->as_xmm_double_reg(), from_addr);
1275 } else {
1276 assert(dest->is_double_fpu(), "must be");
1277 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1278 __ fld_d(from_addr);
1279 }
1280 break;
1281 }
1283 case T_OBJECT: // fall through
1284 case T_ARRAY: // fall through
1285 if (UseCompressedOops && !wide) {
1286 __ movl(dest->as_register(), from_addr);
1287 } else {
1288 __ movptr(dest->as_register(), from_addr);
1289 }
1290 break;
1292 case T_ADDRESS:
1293 __ movptr(dest->as_register(), from_addr);
1294 break;
1295 case T_INT:
1296 __ movl(dest->as_register(), from_addr);
1297 break;
1299 case T_LONG: {
1300 Register to_lo = dest->as_register_lo();
1301 Register to_hi = dest->as_register_hi();
1302 #ifdef _LP64
1303 __ movptr(to_lo, as_Address_lo(addr));
1304 #else
1305 Register base = addr->base()->as_register();
1306 Register index = noreg;
1307 if (addr->index()->is_register()) {
1308 index = addr->index()->as_register();
1309 }
1310 if ((base == to_lo && index == to_hi) ||
1311 (base == to_hi && index == to_lo)) {
1312 // addresses with 2 registers are only formed as a result of
1313 // array access so this code will never have to deal with
1314 // patches or null checks.
1315 assert(info == NULL && patch == NULL, "must be");
1316 __ lea(to_hi, as_Address(addr));
1317 __ movl(to_lo, Address(to_hi, 0));
1318 __ movl(to_hi, Address(to_hi, BytesPerWord));
1319 } else if (base == to_lo || index == to_lo) {
1320 assert(base != to_hi, "can't be");
1321 assert(index == noreg || (index != base && index != to_hi), "can't handle this");
1322 __ movl(to_hi, as_Address_hi(addr));
1323 if (patch != NULL) {
1324 patching_epilog(patch, lir_patch_high, base, info);
1325 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1326 patch_code = lir_patch_low;
1327 }
1328 __ movl(to_lo, as_Address_lo(addr));
1329 } else {
1330 assert(index == noreg || (index != base && index != to_lo), "can't handle this");
1331 __ movl(to_lo, as_Address_lo(addr));
1332 if (patch != NULL) {
1333 patching_epilog(patch, lir_patch_low, base, info);
1334 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1335 patch_code = lir_patch_high;
1336 }
1337 __ movl(to_hi, as_Address_hi(addr));
1338 }
1339 #endif // _LP64
1340 break;
1341 }
1343 case T_BOOLEAN: // fall through
1344 case T_BYTE: {
1345 Register dest_reg = dest->as_register();
1346 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1347 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1348 __ movsbl(dest_reg, from_addr);
1349 } else {
1350 __ movb(dest_reg, from_addr);
1351 __ shll(dest_reg, 24);
1352 __ sarl(dest_reg, 24);
1353 }
1354 break;
1355 }
1357 case T_CHAR: {
1358 Register dest_reg = dest->as_register();
1359 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1360 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1361 __ movzwl(dest_reg, from_addr);
1362 } else {
1363 __ movw(dest_reg, from_addr);
1364 }
1365 break;
1366 }
1368 case T_SHORT: {
1369 Register dest_reg = dest->as_register();
1370 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1371 __ movswl(dest_reg, from_addr);
1372 } else {
1373 __ movw(dest_reg, from_addr);
1374 __ shll(dest_reg, 16);
1375 __ sarl(dest_reg, 16);
1376 }
1377 break;
1378 }
1380 default:
1381 ShouldNotReachHere();
1382 }
1384 if (patch != NULL) {
1385 patching_epilog(patch, patch_code, addr->base()->as_register(), info);
1386 }
1388 if (type == T_ARRAY || type == T_OBJECT) {
1389 #ifdef _LP64
1390 if (UseCompressedOops && !wide) {
1391 __ decode_heap_oop(dest->as_register());
1392 }
1393 #endif
1394 __ verify_oop(dest->as_register());
1395 }
1396 }
1399 void LIR_Assembler::prefetchr(LIR_Opr src) {
1400 LIR_Address* addr = src->as_address_ptr();
1401 Address from_addr = as_Address(addr);
1403 if (VM_Version::supports_sse()) {
1404 switch (ReadPrefetchInstr) {
1405 case 0:
1406 __ prefetchnta(from_addr); break;
1407 case 1:
1408 __ prefetcht0(from_addr); break;
1409 case 2:
1410 __ prefetcht2(from_addr); break;
1411 default:
1412 ShouldNotReachHere(); break;
1413 }
1414 } else if (VM_Version::supports_3dnow()) {
1415 __ prefetchr(from_addr);
1416 }
1417 }
1420 void LIR_Assembler::prefetchw(LIR_Opr src) {
1421 LIR_Address* addr = src->as_address_ptr();
1422 Address from_addr = as_Address(addr);
1424 if (VM_Version::supports_sse()) {
1425 switch (AllocatePrefetchInstr) {
1426 case 0:
1427 __ prefetchnta(from_addr); break;
1428 case 1:
1429 __ prefetcht0(from_addr); break;
1430 case 2:
1431 __ prefetcht2(from_addr); break;
1432 case 3:
1433 __ prefetchw(from_addr); break;
1434 default:
1435 ShouldNotReachHere(); break;
1436 }
1437 } else if (VM_Version::supports_3dnow()) {
1438 __ prefetchw(from_addr);
1439 }
1440 }
1443 NEEDS_CLEANUP; // This could be static?
1444 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
1445 int elem_size = type2aelembytes(type);
1446 switch (elem_size) {
1447 case 1: return Address::times_1;
1448 case 2: return Address::times_2;
1449 case 4: return Address::times_4;
1450 case 8: return Address::times_8;
1451 }
1452 ShouldNotReachHere();
1453 return Address::no_scale;
1454 }
1457 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1458 switch (op->code()) {
1459 case lir_idiv:
1460 case lir_irem:
1461 arithmetic_idiv(op->code(),
1462 op->in_opr1(),
1463 op->in_opr2(),
1464 op->in_opr3(),
1465 op->result_opr(),
1466 op->info());
1467 break;
1468 default: ShouldNotReachHere(); break;
1469 }
1470 }
1472 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1473 #ifdef ASSERT
1474 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
1475 if (op->block() != NULL) _branch_target_blocks.append(op->block());
1476 if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
1477 #endif
1479 if (op->cond() == lir_cond_always) {
1480 if (op->info() != NULL) add_debug_info_for_branch(op->info());
1481 __ jmp (*(op->label()));
1482 } else {
1483 Assembler::Condition acond = Assembler::zero;
1484 if (op->code() == lir_cond_float_branch) {
1485 assert(op->ublock() != NULL, "must have unordered successor");
1486 __ jcc(Assembler::parity, *(op->ublock()->label()));
1487 switch(op->cond()) {
1488 case lir_cond_equal: acond = Assembler::equal; break;
1489 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1490 case lir_cond_less: acond = Assembler::below; break;
1491 case lir_cond_lessEqual: acond = Assembler::belowEqual; break;
1492 case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
1493 case lir_cond_greater: acond = Assembler::above; break;
1494 default: ShouldNotReachHere();
1495 }
1496 } else {
1497 switch (op->cond()) {
1498 case lir_cond_equal: acond = Assembler::equal; break;
1499 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1500 case lir_cond_less: acond = Assembler::less; break;
1501 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
1502 case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
1503 case lir_cond_greater: acond = Assembler::greater; break;
1504 case lir_cond_belowEqual: acond = Assembler::belowEqual; break;
1505 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break;
1506 default: ShouldNotReachHere();
1507 }
1508 }
1509 __ jcc(acond,*(op->label()));
1510 }
1511 }
1513 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1514 LIR_Opr src = op->in_opr();
1515 LIR_Opr dest = op->result_opr();
1517 switch (op->bytecode()) {
1518 case Bytecodes::_i2l:
1519 #ifdef _LP64
1520 __ movl2ptr(dest->as_register_lo(), src->as_register());
1521 #else
1522 move_regs(src->as_register(), dest->as_register_lo());
1523 move_regs(src->as_register(), dest->as_register_hi());
1524 __ sarl(dest->as_register_hi(), 31);
1525 #endif // LP64
1526 break;
1528 case Bytecodes::_l2i:
1529 move_regs(src->as_register_lo(), dest->as_register());
1530 break;
1532 case Bytecodes::_i2b:
1533 move_regs(src->as_register(), dest->as_register());
1534 __ sign_extend_byte(dest->as_register());
1535 break;
1537 case Bytecodes::_i2c:
1538 move_regs(src->as_register(), dest->as_register());
1539 __ andl(dest->as_register(), 0xFFFF);
1540 break;
1542 case Bytecodes::_i2s:
1543 move_regs(src->as_register(), dest->as_register());
1544 __ sign_extend_short(dest->as_register());
1545 break;
1548 case Bytecodes::_f2d:
1549 case Bytecodes::_d2f:
1550 if (dest->is_single_xmm()) {
1551 __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1552 } else if (dest->is_double_xmm()) {
1553 __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1554 } else {
1555 assert(src->fpu() == dest->fpu(), "register must be equal");
1556 // do nothing (float result is rounded later through spilling)
1557 }
1558 break;
1560 case Bytecodes::_i2f:
1561 case Bytecodes::_i2d:
1562 if (dest->is_single_xmm()) {
1563 __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1564 } else if (dest->is_double_xmm()) {
1565 __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1566 } else {
1567 assert(dest->fpu() == 0, "result must be on TOS");
1568 __ movl(Address(rsp, 0), src->as_register());
1569 __ fild_s(Address(rsp, 0));
1570 }
1571 break;
1573 case Bytecodes::_f2i:
1574 case Bytecodes::_d2i:
1575 if (src->is_single_xmm()) {
1576 __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
1577 } else if (src->is_double_xmm()) {
1578 __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
1579 } else {
1580 assert(src->fpu() == 0, "input must be on TOS");
1581 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
1582 __ fist_s(Address(rsp, 0));
1583 __ movl(dest->as_register(), Address(rsp, 0));
1584 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1585 }
1587 // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
1588 assert(op->stub() != NULL, "stub required");
1589 __ cmpl(dest->as_register(), 0x80000000);
1590 __ jcc(Assembler::equal, *op->stub()->entry());
1591 __ bind(*op->stub()->continuation());
1592 break;
1594 case Bytecodes::_l2f:
1595 case Bytecodes::_l2d:
1596 assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
1597 assert(dest->fpu() == 0, "result must be on TOS");
1599 __ movptr(Address(rsp, 0), src->as_register_lo());
1600 NOT_LP64(__ movl(Address(rsp, BytesPerWord), src->as_register_hi()));
1601 __ fild_d(Address(rsp, 0));
1602 // float result is rounded later through spilling
1603 break;
1605 case Bytecodes::_f2l:
1606 case Bytecodes::_d2l:
1607 assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
1608 assert(src->fpu() == 0, "input must be on TOS");
1609 assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
1611 // instruction sequence too long to inline it here
1612 {
1613 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
1614 }
1615 break;
1617 default: ShouldNotReachHere();
1618 }
1619 }
1621 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1622 if (op->init_check()) {
1623 __ cmpl(Address(op->klass()->as_register(),
1624 instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)),
1625 instanceKlass::fully_initialized);
1626 add_debug_info_for_null_check_here(op->stub()->info());
1627 __ jcc(Assembler::notEqual, *op->stub()->entry());
1628 }
1629 __ allocate_object(op->obj()->as_register(),
1630 op->tmp1()->as_register(),
1631 op->tmp2()->as_register(),
1632 op->header_size(),
1633 op->object_size(),
1634 op->klass()->as_register(),
1635 *op->stub()->entry());
1636 __ bind(*op->stub()->continuation());
1637 }
1639 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1640 Register len = op->len()->as_register();
1641 LP64_ONLY( __ movslq(len, len); )
1643 if (UseSlowPath ||
1644 (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
1645 (!UseFastNewTypeArray && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
1646 __ jmp(*op->stub()->entry());
1647 } else {
1648 Register tmp1 = op->tmp1()->as_register();
1649 Register tmp2 = op->tmp2()->as_register();
1650 Register tmp3 = op->tmp3()->as_register();
1651 if (len == tmp1) {
1652 tmp1 = tmp3;
1653 } else if (len == tmp2) {
1654 tmp2 = tmp3;
1655 } else if (len == tmp3) {
1656 // everything is ok
1657 } else {
1658 __ mov(tmp3, len);
1659 }
1660 __ allocate_array(op->obj()->as_register(),
1661 len,
1662 tmp1,
1663 tmp2,
1664 arrayOopDesc::header_size(op->type()),
1665 array_element_size(op->type()),
1666 op->klass()->as_register(),
1667 *op->stub()->entry());
1668 }
1669 __ bind(*op->stub()->continuation());
1670 }
1672 void LIR_Assembler::type_profile_helper(Register mdo,
1673 ciMethodData *md, ciProfileData *data,
1674 Register recv, Label* update_done) {
1675 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1676 Label next_test;
1677 // See if the receiver is receiver[n].
1678 __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
1679 __ jccb(Assembler::notEqual, next_test);
1680 Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
1681 __ addptr(data_addr, DataLayout::counter_increment);
1682 __ jmp(*update_done);
1683 __ bind(next_test);
1684 }
1686 // Didn't find receiver; find next empty slot and fill it in
1687 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1688 Label next_test;
1689 Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
1690 __ cmpptr(recv_addr, (intptr_t)NULL_WORD);
1691 __ jccb(Assembler::notEqual, next_test);
1692 __ movptr(recv_addr, recv);
1693 __ movptr(Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))), DataLayout::counter_increment);
1694 __ jmp(*update_done);
1695 __ bind(next_test);
1696 }
1697 }
1699 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
1700 // we always need a stub for the failure case.
1701 CodeStub* stub = op->stub();
1702 Register obj = op->object()->as_register();
1703 Register k_RInfo = op->tmp1()->as_register();
1704 Register klass_RInfo = op->tmp2()->as_register();
1705 Register dst = op->result_opr()->as_register();
1706 ciKlass* k = op->klass();
1707 Register Rtmp1 = noreg;
1709 // check if it needs to be profiled
1710 ciMethodData* md;
1711 ciProfileData* data;
1713 if (op->should_profile()) {
1714 ciMethod* method = op->profiled_method();
1715 assert(method != NULL, "Should have method");
1716 int bci = op->profiled_bci();
1717 md = method->method_data_or_null();
1718 assert(md != NULL, "Sanity");
1719 data = md->bci_to_data(bci);
1720 assert(data != NULL, "need data for type check");
1721 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1722 }
1723 Label profile_cast_success, profile_cast_failure;
1724 Label *success_target = op->should_profile() ? &profile_cast_success : success;
1725 Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
1727 if (obj == k_RInfo) {
1728 k_RInfo = dst;
1729 } else if (obj == klass_RInfo) {
1730 klass_RInfo = dst;
1731 }
1732 if (k->is_loaded() && !UseCompressedOops) {
1733 select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1734 } else {
1735 Rtmp1 = op->tmp3()->as_register();
1736 select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1737 }
1739 assert_different_registers(obj, k_RInfo, klass_RInfo);
1740 if (!k->is_loaded()) {
1741 jobject2reg_with_patching(k_RInfo, op->info_for_patch());
1742 } else {
1743 #ifdef _LP64
1744 __ movoop(k_RInfo, k->constant_encoding());
1745 #endif // _LP64
1746 }
1747 assert(obj != k_RInfo, "must be different");
1749 __ cmpptr(obj, (int32_t)NULL_WORD);
1750 if (op->should_profile()) {
1751 Label not_null;
1752 __ jccb(Assembler::notEqual, not_null);
1753 // Object is null; update MDO and exit
1754 Register mdo = klass_RInfo;
1755 __ movoop(mdo, md->constant_encoding());
1756 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1757 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1758 __ orl(data_addr, header_bits);
1759 __ jmp(*obj_is_null);
1760 __ bind(not_null);
1761 } else {
1762 __ jcc(Assembler::equal, *obj_is_null);
1763 }
1764 __ verify_oop(obj);
1766 if (op->fast_check()) {
1767 // get object class
1768 // not a safepoint as obj null check happens earlier
1769 #ifdef _LP64
1770 if (UseCompressedOops) {
1771 __ load_klass(Rtmp1, obj);
1772 __ cmpptr(k_RInfo, Rtmp1);
1773 } else {
1774 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1775 }
1776 #else
1777 if (k->is_loaded()) {
1778 __ cmpoop(Address(obj, oopDesc::klass_offset_in_bytes()), k->constant_encoding());
1779 } else {
1780 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1781 }
1782 #endif
1783 __ jcc(Assembler::notEqual, *failure_target);
1784 // successful cast, fall through to profile or jump
1785 } else {
1786 // get object class
1787 // not a safepoint as obj null check happens earlier
1788 __ load_klass(klass_RInfo, obj);
1789 if (k->is_loaded()) {
1790 // See if we get an immediate positive hit
1791 #ifdef _LP64
1792 __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
1793 #else
1794 __ cmpoop(Address(klass_RInfo, k->super_check_offset()), k->constant_encoding());
1795 #endif // _LP64
1796 if (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes() != k->super_check_offset()) {
1797 __ jcc(Assembler::notEqual, *failure_target);
1798 // successful cast, fall through to profile or jump
1799 } else {
1800 // See if we get an immediate positive hit
1801 __ jcc(Assembler::equal, *success_target);
1802 // check for self
1803 #ifdef _LP64
1804 __ cmpptr(klass_RInfo, k_RInfo);
1805 #else
1806 __ cmpoop(klass_RInfo, k->constant_encoding());
1807 #endif // _LP64
1808 __ jcc(Assembler::equal, *success_target);
1810 __ push(klass_RInfo);
1811 #ifdef _LP64
1812 __ push(k_RInfo);
1813 #else
1814 __ pushoop(k->constant_encoding());
1815 #endif // _LP64
1816 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1817 __ pop(klass_RInfo);
1818 __ pop(klass_RInfo);
1819 // result is a boolean
1820 __ cmpl(klass_RInfo, 0);
1821 __ jcc(Assembler::equal, *failure_target);
1822 // successful cast, fall through to profile or jump
1823 }
1824 } else {
1825 // perform the fast part of the checking logic
1826 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1827 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1828 __ push(klass_RInfo);
1829 __ push(k_RInfo);
1830 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1831 __ pop(klass_RInfo);
1832 __ pop(k_RInfo);
1833 // result is a boolean
1834 __ cmpl(k_RInfo, 0);
1835 __ jcc(Assembler::equal, *failure_target);
1836 // successful cast, fall through to profile or jump
1837 }
1838 }
1839 if (op->should_profile()) {
1840 Register mdo = klass_RInfo, recv = k_RInfo;
1841 __ bind(profile_cast_success);
1842 __ movoop(mdo, md->constant_encoding());
1843 __ load_klass(recv, obj);
1844 Label update_done;
1845 type_profile_helper(mdo, md, data, recv, success);
1846 __ jmp(*success);
1848 __ bind(profile_cast_failure);
1849 __ movoop(mdo, md->constant_encoding());
1850 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1851 __ subptr(counter_addr, DataLayout::counter_increment);
1852 __ jmp(*failure);
1853 }
1854 __ jmp(*success);
1855 }
1858 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1859 LIR_Code code = op->code();
1860 if (code == lir_store_check) {
1861 Register value = op->object()->as_register();
1862 Register array = op->array()->as_register();
1863 Register k_RInfo = op->tmp1()->as_register();
1864 Register klass_RInfo = op->tmp2()->as_register();
1865 Register Rtmp1 = op->tmp3()->as_register();
1867 CodeStub* stub = op->stub();
1869 // check if it needs to be profiled
1870 ciMethodData* md;
1871 ciProfileData* data;
1873 if (op->should_profile()) {
1874 ciMethod* method = op->profiled_method();
1875 assert(method != NULL, "Should have method");
1876 int bci = op->profiled_bci();
1877 md = method->method_data_or_null();
1878 assert(md != NULL, "Sanity");
1879 data = md->bci_to_data(bci);
1880 assert(data != NULL, "need data for type check");
1881 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1882 }
1883 Label profile_cast_success, profile_cast_failure, done;
1884 Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1885 Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
1887 __ cmpptr(value, (int32_t)NULL_WORD);
1888 if (op->should_profile()) {
1889 Label not_null;
1890 __ jccb(Assembler::notEqual, not_null);
1891 // Object is null; update MDO and exit
1892 Register mdo = klass_RInfo;
1893 __ movoop(mdo, md->constant_encoding());
1894 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1895 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1896 __ orl(data_addr, header_bits);
1897 __ jmp(done);
1898 __ bind(not_null);
1899 } else {
1900 __ jcc(Assembler::equal, done);
1901 }
1903 add_debug_info_for_null_check_here(op->info_for_exception());
1904 __ load_klass(k_RInfo, array);
1905 __ load_klass(klass_RInfo, value);
1907 // get instance klass (it's already uncompressed)
1908 __ movptr(k_RInfo, Address(k_RInfo, objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)));
1909 // perform the fast part of the checking logic
1910 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1911 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1912 __ push(klass_RInfo);
1913 __ push(k_RInfo);
1914 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1915 __ pop(klass_RInfo);
1916 __ pop(k_RInfo);
1917 // result is a boolean
1918 __ cmpl(k_RInfo, 0);
1919 __ jcc(Assembler::equal, *failure_target);
1920 // fall through to the success case
1922 if (op->should_profile()) {
1923 Register mdo = klass_RInfo, recv = k_RInfo;
1924 __ bind(profile_cast_success);
1925 __ movoop(mdo, md->constant_encoding());
1926 __ load_klass(recv, value);
1927 Label update_done;
1928 type_profile_helper(mdo, md, data, recv, &done);
1929 __ jmpb(done);
1931 __ bind(profile_cast_failure);
1932 __ movoop(mdo, md->constant_encoding());
1933 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1934 __ subptr(counter_addr, DataLayout::counter_increment);
1935 __ jmp(*stub->entry());
1936 }
1938 __ bind(done);
1939 } else
1940 if (code == lir_checkcast) {
1941 Register obj = op->object()->as_register();
1942 Register dst = op->result_opr()->as_register();
1943 Label success;
1944 emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
1945 __ bind(success);
1946 if (dst != obj) {
1947 __ mov(dst, obj);
1948 }
1949 } else
1950 if (code == lir_instanceof) {
1951 Register obj = op->object()->as_register();
1952 Register dst = op->result_opr()->as_register();
1953 Label success, failure, done;
1954 emit_typecheck_helper(op, &success, &failure, &failure);
1955 __ bind(failure);
1956 __ xorptr(dst, dst);
1957 __ jmpb(done);
1958 __ bind(success);
1959 __ movptr(dst, 1);
1960 __ bind(done);
1961 } else {
1962 ShouldNotReachHere();
1963 }
1965 }
1968 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1969 if (LP64_ONLY(false &&) op->code() == lir_cas_long && VM_Version::supports_cx8()) {
1970 assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
1971 assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
1972 assert(op->new_value()->as_register_lo() == rbx, "wrong register");
1973 assert(op->new_value()->as_register_hi() == rcx, "wrong register");
1974 Register addr = op->addr()->as_register();
1975 if (os::is_MP()) {
1976 __ lock();
1977 }
1978 NOT_LP64(__ cmpxchg8(Address(addr, 0)));
1980 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
1981 NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
1982 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1983 Register newval = op->new_value()->as_register();
1984 Register cmpval = op->cmp_value()->as_register();
1985 assert(cmpval == rax, "wrong register");
1986 assert(newval != NULL, "new val must be register");
1987 assert(cmpval != newval, "cmp and new values must be in different registers");
1988 assert(cmpval != addr, "cmp and addr must be in different registers");
1989 assert(newval != addr, "new value and addr must be in different registers");
1991 if ( op->code() == lir_cas_obj) {
1992 #ifdef _LP64
1993 if (UseCompressedOops) {
1994 __ encode_heap_oop(cmpval);
1995 __ mov(rscratch1, newval);
1996 __ encode_heap_oop(rscratch1);
1997 if (os::is_MP()) {
1998 __ lock();
1999 }
2000 // cmpval (rax) is implicitly used by this instruction
2001 __ cmpxchgl(rscratch1, Address(addr, 0));
2002 } else
2003 #endif
2004 {
2005 if (os::is_MP()) {
2006 __ lock();
2007 }
2008 __ cmpxchgptr(newval, Address(addr, 0));
2009 }
2010 } else {
2011 assert(op->code() == lir_cas_int, "lir_cas_int expected");
2012 if (os::is_MP()) {
2013 __ lock();
2014 }
2015 __ cmpxchgl(newval, Address(addr, 0));
2016 }
2017 #ifdef _LP64
2018 } else if (op->code() == lir_cas_long) {
2019 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
2020 Register newval = op->new_value()->as_register_lo();
2021 Register cmpval = op->cmp_value()->as_register_lo();
2022 assert(cmpval == rax, "wrong register");
2023 assert(newval != NULL, "new val must be register");
2024 assert(cmpval != newval, "cmp and new values must be in different registers");
2025 assert(cmpval != addr, "cmp and addr must be in different registers");
2026 assert(newval != addr, "new value and addr must be in different registers");
2027 if (os::is_MP()) {
2028 __ lock();
2029 }
2030 __ cmpxchgq(newval, Address(addr, 0));
2031 #endif // _LP64
2032 } else {
2033 Unimplemented();
2034 }
2035 }
2037 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
2038 Assembler::Condition acond, ncond;
2039 switch (condition) {
2040 case lir_cond_equal: acond = Assembler::equal; ncond = Assembler::notEqual; break;
2041 case lir_cond_notEqual: acond = Assembler::notEqual; ncond = Assembler::equal; break;
2042 case lir_cond_less: acond = Assembler::less; ncond = Assembler::greaterEqual; break;
2043 case lir_cond_lessEqual: acond = Assembler::lessEqual; ncond = Assembler::greater; break;
2044 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less; break;
2045 case lir_cond_greater: acond = Assembler::greater; ncond = Assembler::lessEqual; break;
2046 case lir_cond_belowEqual: acond = Assembler::belowEqual; ncond = Assembler::above; break;
2047 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; ncond = Assembler::below; break;
2048 default: ShouldNotReachHere();
2049 }
2051 if (opr1->is_cpu_register()) {
2052 reg2reg(opr1, result);
2053 } else if (opr1->is_stack()) {
2054 stack2reg(opr1, result, result->type());
2055 } else if (opr1->is_constant()) {
2056 const2reg(opr1, result, lir_patch_none, NULL);
2057 } else {
2058 ShouldNotReachHere();
2059 }
2061 if (VM_Version::supports_cmov() && !opr2->is_constant()) {
2062 // optimized version that does not require a branch
2063 if (opr2->is_single_cpu()) {
2064 assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
2065 __ cmov(ncond, result->as_register(), opr2->as_register());
2066 } else if (opr2->is_double_cpu()) {
2067 assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2068 assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2069 __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
2070 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
2071 } else if (opr2->is_single_stack()) {
2072 __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
2073 } else if (opr2->is_double_stack()) {
2074 __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
2075 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
2076 } else {
2077 ShouldNotReachHere();
2078 }
2080 } else {
2081 Label skip;
2082 __ jcc (acond, skip);
2083 if (opr2->is_cpu_register()) {
2084 reg2reg(opr2, result);
2085 } else if (opr2->is_stack()) {
2086 stack2reg(opr2, result, result->type());
2087 } else if (opr2->is_constant()) {
2088 const2reg(opr2, result, lir_patch_none, NULL);
2089 } else {
2090 ShouldNotReachHere();
2091 }
2092 __ bind(skip);
2093 }
2094 }
2097 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
2098 assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
2100 if (left->is_single_cpu()) {
2101 assert(left == dest, "left and dest must be equal");
2102 Register lreg = left->as_register();
2104 if (right->is_single_cpu()) {
2105 // cpu register - cpu register
2106 Register rreg = right->as_register();
2107 switch (code) {
2108 case lir_add: __ addl (lreg, rreg); break;
2109 case lir_sub: __ subl (lreg, rreg); break;
2110 case lir_mul: __ imull(lreg, rreg); break;
2111 default: ShouldNotReachHere();
2112 }
2114 } else if (right->is_stack()) {
2115 // cpu register - stack
2116 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2117 switch (code) {
2118 case lir_add: __ addl(lreg, raddr); break;
2119 case lir_sub: __ subl(lreg, raddr); break;
2120 default: ShouldNotReachHere();
2121 }
2123 } else if (right->is_constant()) {
2124 // cpu register - constant
2125 jint c = right->as_constant_ptr()->as_jint();
2126 switch (code) {
2127 case lir_add: {
2128 __ incrementl(lreg, c);
2129 break;
2130 }
2131 case lir_sub: {
2132 __ decrementl(lreg, c);
2133 break;
2134 }
2135 default: ShouldNotReachHere();
2136 }
2138 } else {
2139 ShouldNotReachHere();
2140 }
2142 } else if (left->is_double_cpu()) {
2143 assert(left == dest, "left and dest must be equal");
2144 Register lreg_lo = left->as_register_lo();
2145 Register lreg_hi = left->as_register_hi();
2147 if (right->is_double_cpu()) {
2148 // cpu register - cpu register
2149 Register rreg_lo = right->as_register_lo();
2150 Register rreg_hi = right->as_register_hi();
2151 NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
2152 LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
2153 switch (code) {
2154 case lir_add:
2155 __ addptr(lreg_lo, rreg_lo);
2156 NOT_LP64(__ adcl(lreg_hi, rreg_hi));
2157 break;
2158 case lir_sub:
2159 __ subptr(lreg_lo, rreg_lo);
2160 NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
2161 break;
2162 case lir_mul:
2163 #ifdef _LP64
2164 __ imulq(lreg_lo, rreg_lo);
2165 #else
2166 assert(lreg_lo == rax && lreg_hi == rdx, "must be");
2167 __ imull(lreg_hi, rreg_lo);
2168 __ imull(rreg_hi, lreg_lo);
2169 __ addl (rreg_hi, lreg_hi);
2170 __ mull (rreg_lo);
2171 __ addl (lreg_hi, rreg_hi);
2172 #endif // _LP64
2173 break;
2174 default:
2175 ShouldNotReachHere();
2176 }
2178 } else if (right->is_constant()) {
2179 // cpu register - constant
2180 #ifdef _LP64
2181 jlong c = right->as_constant_ptr()->as_jlong_bits();
2182 __ movptr(r10, (intptr_t) c);
2183 switch (code) {
2184 case lir_add:
2185 __ addptr(lreg_lo, r10);
2186 break;
2187 case lir_sub:
2188 __ subptr(lreg_lo, r10);
2189 break;
2190 default:
2191 ShouldNotReachHere();
2192 }
2193 #else
2194 jint c_lo = right->as_constant_ptr()->as_jint_lo();
2195 jint c_hi = right->as_constant_ptr()->as_jint_hi();
2196 switch (code) {
2197 case lir_add:
2198 __ addptr(lreg_lo, c_lo);
2199 __ adcl(lreg_hi, c_hi);
2200 break;
2201 case lir_sub:
2202 __ subptr(lreg_lo, c_lo);
2203 __ sbbl(lreg_hi, c_hi);
2204 break;
2205 default:
2206 ShouldNotReachHere();
2207 }
2208 #endif // _LP64
2210 } else {
2211 ShouldNotReachHere();
2212 }
2214 } else if (left->is_single_xmm()) {
2215 assert(left == dest, "left and dest must be equal");
2216 XMMRegister lreg = left->as_xmm_float_reg();
2218 if (right->is_single_xmm()) {
2219 XMMRegister rreg = right->as_xmm_float_reg();
2220 switch (code) {
2221 case lir_add: __ addss(lreg, rreg); break;
2222 case lir_sub: __ subss(lreg, rreg); break;
2223 case lir_mul_strictfp: // fall through
2224 case lir_mul: __ mulss(lreg, rreg); break;
2225 case lir_div_strictfp: // fall through
2226 case lir_div: __ divss(lreg, rreg); break;
2227 default: ShouldNotReachHere();
2228 }
2229 } else {
2230 Address raddr;
2231 if (right->is_single_stack()) {
2232 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2233 } else if (right->is_constant()) {
2234 // hack for now
2235 raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
2236 } else {
2237 ShouldNotReachHere();
2238 }
2239 switch (code) {
2240 case lir_add: __ addss(lreg, raddr); break;
2241 case lir_sub: __ subss(lreg, raddr); break;
2242 case lir_mul_strictfp: // fall through
2243 case lir_mul: __ mulss(lreg, raddr); break;
2244 case lir_div_strictfp: // fall through
2245 case lir_div: __ divss(lreg, raddr); break;
2246 default: ShouldNotReachHere();
2247 }
2248 }
2250 } else if (left->is_double_xmm()) {
2251 assert(left == dest, "left and dest must be equal");
2253 XMMRegister lreg = left->as_xmm_double_reg();
2254 if (right->is_double_xmm()) {
2255 XMMRegister rreg = right->as_xmm_double_reg();
2256 switch (code) {
2257 case lir_add: __ addsd(lreg, rreg); break;
2258 case lir_sub: __ subsd(lreg, rreg); break;
2259 case lir_mul_strictfp: // fall through
2260 case lir_mul: __ mulsd(lreg, rreg); break;
2261 case lir_div_strictfp: // fall through
2262 case lir_div: __ divsd(lreg, rreg); break;
2263 default: ShouldNotReachHere();
2264 }
2265 } else {
2266 Address raddr;
2267 if (right->is_double_stack()) {
2268 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2269 } else if (right->is_constant()) {
2270 // hack for now
2271 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2272 } else {
2273 ShouldNotReachHere();
2274 }
2275 switch (code) {
2276 case lir_add: __ addsd(lreg, raddr); break;
2277 case lir_sub: __ subsd(lreg, raddr); break;
2278 case lir_mul_strictfp: // fall through
2279 case lir_mul: __ mulsd(lreg, raddr); break;
2280 case lir_div_strictfp: // fall through
2281 case lir_div: __ divsd(lreg, raddr); break;
2282 default: ShouldNotReachHere();
2283 }
2284 }
2286 } else if (left->is_single_fpu()) {
2287 assert(dest->is_single_fpu(), "fpu stack allocation required");
2289 if (right->is_single_fpu()) {
2290 arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
2292 } else {
2293 assert(left->fpu_regnr() == 0, "left must be on TOS");
2294 assert(dest->fpu_regnr() == 0, "dest must be on TOS");
2296 Address raddr;
2297 if (right->is_single_stack()) {
2298 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2299 } else if (right->is_constant()) {
2300 address const_addr = float_constant(right->as_jfloat());
2301 assert(const_addr != NULL, "incorrect float/double constant maintainance");
2302 // hack for now
2303 raddr = __ as_Address(InternalAddress(const_addr));
2304 } else {
2305 ShouldNotReachHere();
2306 }
2308 switch (code) {
2309 case lir_add: __ fadd_s(raddr); break;
2310 case lir_sub: __ fsub_s(raddr); break;
2311 case lir_mul_strictfp: // fall through
2312 case lir_mul: __ fmul_s(raddr); break;
2313 case lir_div_strictfp: // fall through
2314 case lir_div: __ fdiv_s(raddr); break;
2315 default: ShouldNotReachHere();
2316 }
2317 }
2319 } else if (left->is_double_fpu()) {
2320 assert(dest->is_double_fpu(), "fpu stack allocation required");
2322 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2323 // Double values require special handling for strictfp mul/div on x86
2324 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
2325 __ fmulp(left->fpu_regnrLo() + 1);
2326 }
2328 if (right->is_double_fpu()) {
2329 arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
2331 } else {
2332 assert(left->fpu_regnrLo() == 0, "left must be on TOS");
2333 assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
2335 Address raddr;
2336 if (right->is_double_stack()) {
2337 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2338 } else if (right->is_constant()) {
2339 // hack for now
2340 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2341 } else {
2342 ShouldNotReachHere();
2343 }
2345 switch (code) {
2346 case lir_add: __ fadd_d(raddr); break;
2347 case lir_sub: __ fsub_d(raddr); break;
2348 case lir_mul_strictfp: // fall through
2349 case lir_mul: __ fmul_d(raddr); break;
2350 case lir_div_strictfp: // fall through
2351 case lir_div: __ fdiv_d(raddr); break;
2352 default: ShouldNotReachHere();
2353 }
2354 }
2356 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2357 // Double values require special handling for strictfp mul/div on x86
2358 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
2359 __ fmulp(dest->fpu_regnrLo() + 1);
2360 }
2362 } else if (left->is_single_stack() || left->is_address()) {
2363 assert(left == dest, "left and dest must be equal");
2365 Address laddr;
2366 if (left->is_single_stack()) {
2367 laddr = frame_map()->address_for_slot(left->single_stack_ix());
2368 } else if (left->is_address()) {
2369 laddr = as_Address(left->as_address_ptr());
2370 } else {
2371 ShouldNotReachHere();
2372 }
2374 if (right->is_single_cpu()) {
2375 Register rreg = right->as_register();
2376 switch (code) {
2377 case lir_add: __ addl(laddr, rreg); break;
2378 case lir_sub: __ subl(laddr, rreg); break;
2379 default: ShouldNotReachHere();
2380 }
2381 } else if (right->is_constant()) {
2382 jint c = right->as_constant_ptr()->as_jint();
2383 switch (code) {
2384 case lir_add: {
2385 __ incrementl(laddr, c);
2386 break;
2387 }
2388 case lir_sub: {
2389 __ decrementl(laddr, c);
2390 break;
2391 }
2392 default: ShouldNotReachHere();
2393 }
2394 } else {
2395 ShouldNotReachHere();
2396 }
2398 } else {
2399 ShouldNotReachHere();
2400 }
2401 }
2403 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
2404 assert(pop_fpu_stack || (left_index == dest_index || right_index == dest_index), "invalid LIR");
2405 assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
2406 assert(left_index == 0 || right_index == 0, "either must be on top of stack");
2408 bool left_is_tos = (left_index == 0);
2409 bool dest_is_tos = (dest_index == 0);
2410 int non_tos_index = (left_is_tos ? right_index : left_index);
2412 switch (code) {
2413 case lir_add:
2414 if (pop_fpu_stack) __ faddp(non_tos_index);
2415 else if (dest_is_tos) __ fadd (non_tos_index);
2416 else __ fadda(non_tos_index);
2417 break;
2419 case lir_sub:
2420 if (left_is_tos) {
2421 if (pop_fpu_stack) __ fsubrp(non_tos_index);
2422 else if (dest_is_tos) __ fsub (non_tos_index);
2423 else __ fsubra(non_tos_index);
2424 } else {
2425 if (pop_fpu_stack) __ fsubp (non_tos_index);
2426 else if (dest_is_tos) __ fsubr (non_tos_index);
2427 else __ fsuba (non_tos_index);
2428 }
2429 break;
2431 case lir_mul_strictfp: // fall through
2432 case lir_mul:
2433 if (pop_fpu_stack) __ fmulp(non_tos_index);
2434 else if (dest_is_tos) __ fmul (non_tos_index);
2435 else __ fmula(non_tos_index);
2436 break;
2438 case lir_div_strictfp: // fall through
2439 case lir_div:
2440 if (left_is_tos) {
2441 if (pop_fpu_stack) __ fdivrp(non_tos_index);
2442 else if (dest_is_tos) __ fdiv (non_tos_index);
2443 else __ fdivra(non_tos_index);
2444 } else {
2445 if (pop_fpu_stack) __ fdivp (non_tos_index);
2446 else if (dest_is_tos) __ fdivr (non_tos_index);
2447 else __ fdiva (non_tos_index);
2448 }
2449 break;
2451 case lir_rem:
2452 assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
2453 __ fremr(noreg);
2454 break;
2456 default:
2457 ShouldNotReachHere();
2458 }
2459 }
2462 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
2463 if (value->is_double_xmm()) {
2464 switch(code) {
2465 case lir_abs :
2466 {
2467 if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
2468 __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
2469 }
2470 __ andpd(dest->as_xmm_double_reg(),
2471 ExternalAddress((address)double_signmask_pool));
2472 }
2473 break;
2475 case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
2476 // all other intrinsics are not available in the SSE instruction set, so FPU is used
2477 default : ShouldNotReachHere();
2478 }
2480 } else if (value->is_double_fpu()) {
2481 assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
2482 switch(code) {
2483 case lir_log : __ flog() ; break;
2484 case lir_log10 : __ flog10() ; break;
2485 case lir_abs : __ fabs() ; break;
2486 case lir_sqrt : __ fsqrt(); break;
2487 case lir_sin :
2488 // Should consider not saving rbx, if not necessary
2489 __ trigfunc('s', op->as_Op2()->fpu_stack_size());
2490 break;
2491 case lir_cos :
2492 // Should consider not saving rbx, if not necessary
2493 assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
2494 __ trigfunc('c', op->as_Op2()->fpu_stack_size());
2495 break;
2496 case lir_tan :
2497 // Should consider not saving rbx, if not necessary
2498 __ trigfunc('t', op->as_Op2()->fpu_stack_size());
2499 break;
2500 default : ShouldNotReachHere();
2501 }
2502 } else {
2503 Unimplemented();
2504 }
2505 }
2507 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
2508 // assert(left->destroys_register(), "check");
2509 if (left->is_single_cpu()) {
2510 Register reg = left->as_register();
2511 if (right->is_constant()) {
2512 int val = right->as_constant_ptr()->as_jint();
2513 switch (code) {
2514 case lir_logic_and: __ andl (reg, val); break;
2515 case lir_logic_or: __ orl (reg, val); break;
2516 case lir_logic_xor: __ xorl (reg, val); break;
2517 default: ShouldNotReachHere();
2518 }
2519 } else if (right->is_stack()) {
2520 // added support for stack operands
2521 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2522 switch (code) {
2523 case lir_logic_and: __ andl (reg, raddr); break;
2524 case lir_logic_or: __ orl (reg, raddr); break;
2525 case lir_logic_xor: __ xorl (reg, raddr); break;
2526 default: ShouldNotReachHere();
2527 }
2528 } else {
2529 Register rright = right->as_register();
2530 switch (code) {
2531 case lir_logic_and: __ andptr (reg, rright); break;
2532 case lir_logic_or : __ orptr (reg, rright); break;
2533 case lir_logic_xor: __ xorptr (reg, rright); break;
2534 default: ShouldNotReachHere();
2535 }
2536 }
2537 move_regs(reg, dst->as_register());
2538 } else {
2539 Register l_lo = left->as_register_lo();
2540 Register l_hi = left->as_register_hi();
2541 if (right->is_constant()) {
2542 #ifdef _LP64
2543 __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
2544 switch (code) {
2545 case lir_logic_and:
2546 __ andq(l_lo, rscratch1);
2547 break;
2548 case lir_logic_or:
2549 __ orq(l_lo, rscratch1);
2550 break;
2551 case lir_logic_xor:
2552 __ xorq(l_lo, rscratch1);
2553 break;
2554 default: ShouldNotReachHere();
2555 }
2556 #else
2557 int r_lo = right->as_constant_ptr()->as_jint_lo();
2558 int r_hi = right->as_constant_ptr()->as_jint_hi();
2559 switch (code) {
2560 case lir_logic_and:
2561 __ andl(l_lo, r_lo);
2562 __ andl(l_hi, r_hi);
2563 break;
2564 case lir_logic_or:
2565 __ orl(l_lo, r_lo);
2566 __ orl(l_hi, r_hi);
2567 break;
2568 case lir_logic_xor:
2569 __ xorl(l_lo, r_lo);
2570 __ xorl(l_hi, r_hi);
2571 break;
2572 default: ShouldNotReachHere();
2573 }
2574 #endif // _LP64
2575 } else {
2576 #ifdef _LP64
2577 Register r_lo;
2578 if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
2579 r_lo = right->as_register();
2580 } else {
2581 r_lo = right->as_register_lo();
2582 }
2583 #else
2584 Register r_lo = right->as_register_lo();
2585 Register r_hi = right->as_register_hi();
2586 assert(l_lo != r_hi, "overwriting registers");
2587 #endif
2588 switch (code) {
2589 case lir_logic_and:
2590 __ andptr(l_lo, r_lo);
2591 NOT_LP64(__ andptr(l_hi, r_hi);)
2592 break;
2593 case lir_logic_or:
2594 __ orptr(l_lo, r_lo);
2595 NOT_LP64(__ orptr(l_hi, r_hi);)
2596 break;
2597 case lir_logic_xor:
2598 __ xorptr(l_lo, r_lo);
2599 NOT_LP64(__ xorptr(l_hi, r_hi);)
2600 break;
2601 default: ShouldNotReachHere();
2602 }
2603 }
2605 Register dst_lo = dst->as_register_lo();
2606 Register dst_hi = dst->as_register_hi();
2608 #ifdef _LP64
2609 move_regs(l_lo, dst_lo);
2610 #else
2611 if (dst_lo == l_hi) {
2612 assert(dst_hi != l_lo, "overwriting registers");
2613 move_regs(l_hi, dst_hi);
2614 move_regs(l_lo, dst_lo);
2615 } else {
2616 assert(dst_lo != l_hi, "overwriting registers");
2617 move_regs(l_lo, dst_lo);
2618 move_regs(l_hi, dst_hi);
2619 }
2620 #endif // _LP64
2621 }
2622 }
2625 // we assume that rax, and rdx can be overwritten
2626 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
2628 assert(left->is_single_cpu(), "left must be register");
2629 assert(right->is_single_cpu() || right->is_constant(), "right must be register or constant");
2630 assert(result->is_single_cpu(), "result must be register");
2632 // assert(left->destroys_register(), "check");
2633 // assert(right->destroys_register(), "check");
2635 Register lreg = left->as_register();
2636 Register dreg = result->as_register();
2638 if (right->is_constant()) {
2639 int divisor = right->as_constant_ptr()->as_jint();
2640 assert(divisor > 0 && is_power_of_2(divisor), "must be");
2641 if (code == lir_idiv) {
2642 assert(lreg == rax, "must be rax,");
2643 assert(temp->as_register() == rdx, "tmp register must be rdx");
2644 __ cdql(); // sign extend into rdx:rax
2645 if (divisor == 2) {
2646 __ subl(lreg, rdx);
2647 } else {
2648 __ andl(rdx, divisor - 1);
2649 __ addl(lreg, rdx);
2650 }
2651 __ sarl(lreg, log2_intptr(divisor));
2652 move_regs(lreg, dreg);
2653 } else if (code == lir_irem) {
2654 Label done;
2655 __ mov(dreg, lreg);
2656 __ andl(dreg, 0x80000000 | (divisor - 1));
2657 __ jcc(Assembler::positive, done);
2658 __ decrement(dreg);
2659 __ orl(dreg, ~(divisor - 1));
2660 __ increment(dreg);
2661 __ bind(done);
2662 } else {
2663 ShouldNotReachHere();
2664 }
2665 } else {
2666 Register rreg = right->as_register();
2667 assert(lreg == rax, "left register must be rax,");
2668 assert(rreg != rdx, "right register must not be rdx");
2669 assert(temp->as_register() == rdx, "tmp register must be rdx");
2671 move_regs(lreg, rax);
2673 int idivl_offset = __ corrected_idivl(rreg);
2674 add_debug_info_for_div0(idivl_offset, info);
2675 if (code == lir_irem) {
2676 move_regs(rdx, dreg); // result is in rdx
2677 } else {
2678 move_regs(rax, dreg);
2679 }
2680 }
2681 }
2684 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
2685 if (opr1->is_single_cpu()) {
2686 Register reg1 = opr1->as_register();
2687 if (opr2->is_single_cpu()) {
2688 // cpu register - cpu register
2689 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2690 __ cmpptr(reg1, opr2->as_register());
2691 } else {
2692 assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
2693 __ cmpl(reg1, opr2->as_register());
2694 }
2695 } else if (opr2->is_stack()) {
2696 // cpu register - stack
2697 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2698 __ cmpptr(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2699 } else {
2700 __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2701 }
2702 } else if (opr2->is_constant()) {
2703 // cpu register - constant
2704 LIR_Const* c = opr2->as_constant_ptr();
2705 if (c->type() == T_INT) {
2706 __ cmpl(reg1, c->as_jint());
2707 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2708 // In 64bit oops are single register
2709 jobject o = c->as_jobject();
2710 if (o == NULL) {
2711 __ cmpptr(reg1, (int32_t)NULL_WORD);
2712 } else {
2713 #ifdef _LP64
2714 __ movoop(rscratch1, o);
2715 __ cmpptr(reg1, rscratch1);
2716 #else
2717 __ cmpoop(reg1, c->as_jobject());
2718 #endif // _LP64
2719 }
2720 } else {
2721 ShouldNotReachHere();
2722 }
2723 // cpu register - address
2724 } else if (opr2->is_address()) {
2725 if (op->info() != NULL) {
2726 add_debug_info_for_null_check_here(op->info());
2727 }
2728 __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
2729 } else {
2730 ShouldNotReachHere();
2731 }
2733 } else if(opr1->is_double_cpu()) {
2734 Register xlo = opr1->as_register_lo();
2735 Register xhi = opr1->as_register_hi();
2736 if (opr2->is_double_cpu()) {
2737 #ifdef _LP64
2738 __ cmpptr(xlo, opr2->as_register_lo());
2739 #else
2740 // cpu register - cpu register
2741 Register ylo = opr2->as_register_lo();
2742 Register yhi = opr2->as_register_hi();
2743 __ subl(xlo, ylo);
2744 __ sbbl(xhi, yhi);
2745 if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
2746 __ orl(xhi, xlo);
2747 }
2748 #endif // _LP64
2749 } else if (opr2->is_constant()) {
2750 // cpu register - constant 0
2751 assert(opr2->as_jlong() == (jlong)0, "only handles zero");
2752 #ifdef _LP64
2753 __ cmpptr(xlo, (int32_t)opr2->as_jlong());
2754 #else
2755 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
2756 __ orl(xhi, xlo);
2757 #endif // _LP64
2758 } else {
2759 ShouldNotReachHere();
2760 }
2762 } else if (opr1->is_single_xmm()) {
2763 XMMRegister reg1 = opr1->as_xmm_float_reg();
2764 if (opr2->is_single_xmm()) {
2765 // xmm register - xmm register
2766 __ ucomiss(reg1, opr2->as_xmm_float_reg());
2767 } else if (opr2->is_stack()) {
2768 // xmm register - stack
2769 __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2770 } else if (opr2->is_constant()) {
2771 // xmm register - constant
2772 __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
2773 } else if (opr2->is_address()) {
2774 // xmm register - address
2775 if (op->info() != NULL) {
2776 add_debug_info_for_null_check_here(op->info());
2777 }
2778 __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
2779 } else {
2780 ShouldNotReachHere();
2781 }
2783 } else if (opr1->is_double_xmm()) {
2784 XMMRegister reg1 = opr1->as_xmm_double_reg();
2785 if (opr2->is_double_xmm()) {
2786 // xmm register - xmm register
2787 __ ucomisd(reg1, opr2->as_xmm_double_reg());
2788 } else if (opr2->is_stack()) {
2789 // xmm register - stack
2790 __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
2791 } else if (opr2->is_constant()) {
2792 // xmm register - constant
2793 __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
2794 } else if (opr2->is_address()) {
2795 // xmm register - address
2796 if (op->info() != NULL) {
2797 add_debug_info_for_null_check_here(op->info());
2798 }
2799 __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
2800 } else {
2801 ShouldNotReachHere();
2802 }
2804 } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
2805 assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
2806 assert(opr2->is_fpu_register(), "both must be registers");
2807 __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2809 } else if (opr1->is_address() && opr2->is_constant()) {
2810 LIR_Const* c = opr2->as_constant_ptr();
2811 #ifdef _LP64
2812 if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2813 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
2814 __ movoop(rscratch1, c->as_jobject());
2815 }
2816 #endif // LP64
2817 if (op->info() != NULL) {
2818 add_debug_info_for_null_check_here(op->info());
2819 }
2820 // special case: address - constant
2821 LIR_Address* addr = opr1->as_address_ptr();
2822 if (c->type() == T_INT) {
2823 __ cmpl(as_Address(addr), c->as_jint());
2824 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2825 #ifdef _LP64
2826 // %%% Make this explode if addr isn't reachable until we figure out a
2827 // better strategy by giving noreg as the temp for as_Address
2828 __ cmpptr(rscratch1, as_Address(addr, noreg));
2829 #else
2830 __ cmpoop(as_Address(addr), c->as_jobject());
2831 #endif // _LP64
2832 } else {
2833 ShouldNotReachHere();
2834 }
2836 } else {
2837 ShouldNotReachHere();
2838 }
2839 }
2841 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
2842 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
2843 if (left->is_single_xmm()) {
2844 assert(right->is_single_xmm(), "must match");
2845 __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2846 } else if (left->is_double_xmm()) {
2847 assert(right->is_double_xmm(), "must match");
2848 __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2850 } else {
2851 assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
2852 assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
2854 assert(left->fpu() == 0, "left must be on TOS");
2855 __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
2856 op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2857 }
2858 } else {
2859 assert(code == lir_cmp_l2i, "check");
2860 #ifdef _LP64
2861 Label done;
2862 Register dest = dst->as_register();
2863 __ cmpptr(left->as_register_lo(), right->as_register_lo());
2864 __ movl(dest, -1);
2865 __ jccb(Assembler::less, done);
2866 __ set_byte_if_not_zero(dest);
2867 __ movzbl(dest, dest);
2868 __ bind(done);
2869 #else
2870 __ lcmp2int(left->as_register_hi(),
2871 left->as_register_lo(),
2872 right->as_register_hi(),
2873 right->as_register_lo());
2874 move_regs(left->as_register_hi(), dst->as_register());
2875 #endif // _LP64
2876 }
2877 }
2880 void LIR_Assembler::align_call(LIR_Code code) {
2881 if (os::is_MP()) {
2882 // make sure that the displacement word of the call ends up word aligned
2883 int offset = __ offset();
2884 switch (code) {
2885 case lir_static_call:
2886 case lir_optvirtual_call:
2887 case lir_dynamic_call:
2888 offset += NativeCall::displacement_offset;
2889 break;
2890 case lir_icvirtual_call:
2891 offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
2892 break;
2893 case lir_virtual_call: // currently, sparc-specific for niagara
2894 default: ShouldNotReachHere();
2895 }
2896 while (offset++ % BytesPerWord != 0) {
2897 __ nop();
2898 }
2899 }
2900 }
2903 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
2904 assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2905 "must be aligned");
2906 __ call(AddressLiteral(op->addr(), rtype));
2907 add_call_info(code_offset(), op->info());
2908 }
2911 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
2912 RelocationHolder rh = virtual_call_Relocation::spec(pc());
2913 __ movoop(IC_Klass, (jobject)Universe::non_oop_word());
2914 assert(!os::is_MP() ||
2915 (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2916 "must be aligned");
2917 __ call(AddressLiteral(op->addr(), rh));
2918 add_call_info(code_offset(), op->info());
2919 }
2922 /* Currently, vtable-dispatch is only enabled for sparc platforms */
2923 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
2924 ShouldNotReachHere();
2925 }
2928 void LIR_Assembler::emit_static_call_stub() {
2929 address call_pc = __ pc();
2930 address stub = __ start_a_stub(call_stub_size);
2931 if (stub == NULL) {
2932 bailout("static call stub overflow");
2933 return;
2934 }
2936 int start = __ offset();
2937 if (os::is_MP()) {
2938 // make sure that the displacement word of the call ends up word aligned
2939 int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
2940 while (offset++ % BytesPerWord != 0) {
2941 __ nop();
2942 }
2943 }
2944 __ relocate(static_stub_Relocation::spec(call_pc));
2945 __ movoop(rbx, (jobject)NULL);
2946 // must be set to -1 at code generation time
2947 assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
2948 // On 64bit this will die since it will take a movq & jmp, must be only a jmp
2949 __ jump(RuntimeAddress(__ pc()));
2951 assert(__ offset() - start <= call_stub_size, "stub too big");
2952 __ end_a_stub();
2953 }
2956 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2957 assert(exceptionOop->as_register() == rax, "must match");
2958 assert(exceptionPC->as_register() == rdx, "must match");
2960 // exception object is not added to oop map by LinearScan
2961 // (LinearScan assumes that no oops are in fixed registers)
2962 info->add_register_oop(exceptionOop);
2963 Runtime1::StubID unwind_id;
2965 // get current pc information
2966 // pc is only needed if the method has an exception handler, the unwind code does not need it.
2967 int pc_for_athrow_offset = __ offset();
2968 InternalAddress pc_for_athrow(__ pc());
2969 __ lea(exceptionPC->as_register(), pc_for_athrow);
2970 add_call_info(pc_for_athrow_offset, info); // for exception handler
2972 __ verify_not_null_oop(rax);
2973 // search an exception handler (rax: exception oop, rdx: throwing pc)
2974 if (compilation()->has_fpu_code()) {
2975 unwind_id = Runtime1::handle_exception_id;
2976 } else {
2977 unwind_id = Runtime1::handle_exception_nofpu_id;
2978 }
2979 __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2981 // enough room for two byte trap
2982 __ nop();
2983 }
2986 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2987 assert(exceptionOop->as_register() == rax, "must match");
2989 __ jmp(_unwind_handler_entry);
2990 }
2993 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2995 // optimized version for linear scan:
2996 // * count must be already in ECX (guaranteed by LinearScan)
2997 // * left and dest must be equal
2998 // * tmp must be unused
2999 assert(count->as_register() == SHIFT_count, "count must be in ECX");
3000 assert(left == dest, "left and dest must be equal");
3001 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
3003 if (left->is_single_cpu()) {
3004 Register value = left->as_register();
3005 assert(value != SHIFT_count, "left cannot be ECX");
3007 switch (code) {
3008 case lir_shl: __ shll(value); break;
3009 case lir_shr: __ sarl(value); break;
3010 case lir_ushr: __ shrl(value); break;
3011 default: ShouldNotReachHere();
3012 }
3013 } else if (left->is_double_cpu()) {
3014 Register lo = left->as_register_lo();
3015 Register hi = left->as_register_hi();
3016 assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
3017 #ifdef _LP64
3018 switch (code) {
3019 case lir_shl: __ shlptr(lo); break;
3020 case lir_shr: __ sarptr(lo); break;
3021 case lir_ushr: __ shrptr(lo); break;
3022 default: ShouldNotReachHere();
3023 }
3024 #else
3026 switch (code) {
3027 case lir_shl: __ lshl(hi, lo); break;
3028 case lir_shr: __ lshr(hi, lo, true); break;
3029 case lir_ushr: __ lshr(hi, lo, false); break;
3030 default: ShouldNotReachHere();
3031 }
3032 #endif // LP64
3033 } else {
3034 ShouldNotReachHere();
3035 }
3036 }
3039 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
3040 if (dest->is_single_cpu()) {
3041 // first move left into dest so that left is not destroyed by the shift
3042 Register value = dest->as_register();
3043 count = count & 0x1F; // Java spec
3045 move_regs(left->as_register(), value);
3046 switch (code) {
3047 case lir_shl: __ shll(value, count); break;
3048 case lir_shr: __ sarl(value, count); break;
3049 case lir_ushr: __ shrl(value, count); break;
3050 default: ShouldNotReachHere();
3051 }
3052 } else if (dest->is_double_cpu()) {
3053 #ifndef _LP64
3054 Unimplemented();
3055 #else
3056 // first move left into dest so that left is not destroyed by the shift
3057 Register value = dest->as_register_lo();
3058 count = count & 0x1F; // Java spec
3060 move_regs(left->as_register_lo(), value);
3061 switch (code) {
3062 case lir_shl: __ shlptr(value, count); break;
3063 case lir_shr: __ sarptr(value, count); break;
3064 case lir_ushr: __ shrptr(value, count); break;
3065 default: ShouldNotReachHere();
3066 }
3067 #endif // _LP64
3068 } else {
3069 ShouldNotReachHere();
3070 }
3071 }
3074 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
3075 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3076 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3077 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3078 __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
3079 }
3082 void LIR_Assembler::store_parameter(jint c, int offset_from_rsp_in_words) {
3083 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3084 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3085 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3086 __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
3087 }
3090 void LIR_Assembler::store_parameter(jobject o, int offset_from_rsp_in_words) {
3091 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3092 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3093 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3094 __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
3095 }
3098 // This code replaces a call to arraycopy; no exception may
3099 // be thrown in this code, they must be thrown in the System.arraycopy
3100 // activation frame; we could save some checks if this would not be the case
3101 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
3102 ciArrayKlass* default_type = op->expected_type();
3103 Register src = op->src()->as_register();
3104 Register dst = op->dst()->as_register();
3105 Register src_pos = op->src_pos()->as_register();
3106 Register dst_pos = op->dst_pos()->as_register();
3107 Register length = op->length()->as_register();
3108 Register tmp = op->tmp()->as_register();
3110 CodeStub* stub = op->stub();
3111 int flags = op->flags();
3112 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
3113 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
3115 // if we don't know anything or it's an object array, just go through the generic arraycopy
3116 if (default_type == NULL) {
3117 Label done;
3118 // save outgoing arguments on stack in case call to System.arraycopy is needed
3119 // HACK ALERT. This code used to push the parameters in a hardwired fashion
3120 // for interpreter calling conventions. Now we have to do it in new style conventions.
3121 // For the moment until C1 gets the new register allocator I just force all the
3122 // args to the right place (except the register args) and then on the back side
3123 // reload the register args properly if we go slow path. Yuck
3125 // These are proper for the calling convention
3127 store_parameter(length, 2);
3128 store_parameter(dst_pos, 1);
3129 store_parameter(dst, 0);
3131 // these are just temporary placements until we need to reload
3132 store_parameter(src_pos, 3);
3133 store_parameter(src, 4);
3134 NOT_LP64(assert(src == rcx && src_pos == rdx, "mismatch in calling convention");)
3136 address entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
3138 // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
3139 #ifdef _LP64
3140 // The arguments are in java calling convention so we can trivially shift them to C
3141 // convention
3142 assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
3143 __ mov(c_rarg0, j_rarg0);
3144 assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
3145 __ mov(c_rarg1, j_rarg1);
3146 assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
3147 __ mov(c_rarg2, j_rarg2);
3148 assert_different_registers(c_rarg3, j_rarg4);
3149 __ mov(c_rarg3, j_rarg3);
3150 #ifdef _WIN64
3151 // Allocate abi space for args but be sure to keep stack aligned
3152 __ subptr(rsp, 6*wordSize);
3153 store_parameter(j_rarg4, 4);
3154 __ call(RuntimeAddress(entry));
3155 __ addptr(rsp, 6*wordSize);
3156 #else
3157 __ mov(c_rarg4, j_rarg4);
3158 __ call(RuntimeAddress(entry));
3159 #endif // _WIN64
3160 #else
3161 __ push(length);
3162 __ push(dst_pos);
3163 __ push(dst);
3164 __ push(src_pos);
3165 __ push(src);
3166 __ call_VM_leaf(entry, 5); // removes pushed parameter from the stack
3168 #endif // _LP64
3170 __ cmpl(rax, 0);
3171 __ jcc(Assembler::equal, *stub->continuation());
3173 // Reload values from the stack so they are where the stub
3174 // expects them.
3175 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3176 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3177 __ movptr (length, Address(rsp, 2*BytesPerWord));
3178 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3179 __ movptr (src, Address(rsp, 4*BytesPerWord));
3180 __ jmp(*stub->entry());
3182 __ bind(*stub->continuation());
3183 return;
3184 }
3186 assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
3188 int elem_size = type2aelembytes(basic_type);
3189 int shift_amount;
3190 Address::ScaleFactor scale;
3192 switch (elem_size) {
3193 case 1 :
3194 shift_amount = 0;
3195 scale = Address::times_1;
3196 break;
3197 case 2 :
3198 shift_amount = 1;
3199 scale = Address::times_2;
3200 break;
3201 case 4 :
3202 shift_amount = 2;
3203 scale = Address::times_4;
3204 break;
3205 case 8 :
3206 shift_amount = 3;
3207 scale = Address::times_8;
3208 break;
3209 default:
3210 ShouldNotReachHere();
3211 }
3213 Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
3214 Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
3215 Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
3216 Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
3218 // length and pos's are all sign extended at this point on 64bit
3220 // test for NULL
3221 if (flags & LIR_OpArrayCopy::src_null_check) {
3222 __ testptr(src, src);
3223 __ jcc(Assembler::zero, *stub->entry());
3224 }
3225 if (flags & LIR_OpArrayCopy::dst_null_check) {
3226 __ testptr(dst, dst);
3227 __ jcc(Assembler::zero, *stub->entry());
3228 }
3230 // check if negative
3231 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
3232 __ testl(src_pos, src_pos);
3233 __ jcc(Assembler::less, *stub->entry());
3234 }
3235 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
3236 __ testl(dst_pos, dst_pos);
3237 __ jcc(Assembler::less, *stub->entry());
3238 }
3239 if (flags & LIR_OpArrayCopy::length_positive_check) {
3240 __ testl(length, length);
3241 __ jcc(Assembler::less, *stub->entry());
3242 }
3244 if (flags & LIR_OpArrayCopy::src_range_check) {
3245 __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
3246 __ cmpl(tmp, src_length_addr);
3247 __ jcc(Assembler::above, *stub->entry());
3248 }
3249 if (flags & LIR_OpArrayCopy::dst_range_check) {
3250 __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
3251 __ cmpl(tmp, dst_length_addr);
3252 __ jcc(Assembler::above, *stub->entry());
3253 }
3255 if (flags & LIR_OpArrayCopy::type_check) {
3256 if (UseCompressedOops) {
3257 __ movl(tmp, src_klass_addr);
3258 __ cmpl(tmp, dst_klass_addr);
3259 } else {
3260 __ movptr(tmp, src_klass_addr);
3261 __ cmpptr(tmp, dst_klass_addr);
3262 }
3263 __ jcc(Assembler::notEqual, *stub->entry());
3264 }
3266 #ifdef ASSERT
3267 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
3268 // Sanity check the known type with the incoming class. For the
3269 // primitive case the types must match exactly with src.klass and
3270 // dst.klass each exactly matching the default type. For the
3271 // object array case, if no type check is needed then either the
3272 // dst type is exactly the expected type and the src type is a
3273 // subtype which we can't check or src is the same array as dst
3274 // but not necessarily exactly of type default_type.
3275 Label known_ok, halt;
3276 __ movoop(tmp, default_type->constant_encoding());
3277 #ifdef _LP64
3278 if (UseCompressedOops) {
3279 __ encode_heap_oop(tmp);
3280 }
3281 #endif
3283 if (basic_type != T_OBJECT) {
3285 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3286 else __ cmpptr(tmp, dst_klass_addr);
3287 __ jcc(Assembler::notEqual, halt);
3288 if (UseCompressedOops) __ cmpl(tmp, src_klass_addr);
3289 else __ cmpptr(tmp, src_klass_addr);
3290 __ jcc(Assembler::equal, known_ok);
3291 } else {
3292 if (UseCompressedOops) __ cmpl(tmp, dst_klass_addr);
3293 else __ cmpptr(tmp, dst_klass_addr);
3294 __ jcc(Assembler::equal, known_ok);
3295 __ cmpptr(src, dst);
3296 __ jcc(Assembler::equal, known_ok);
3297 }
3298 __ bind(halt);
3299 __ stop("incorrect type information in arraycopy");
3300 __ bind(known_ok);
3301 }
3302 #endif
3304 if (shift_amount > 0 && basic_type != T_OBJECT) {
3305 __ shlptr(length, shift_amount);
3306 }
3308 #ifdef _LP64
3309 assert_different_registers(c_rarg0, dst, dst_pos, length);
3310 __ movl2ptr(src_pos, src_pos); //higher 32bits must be null
3311 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3312 assert_different_registers(c_rarg1, length);
3313 __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null
3314 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3315 __ mov(c_rarg2, length);
3317 #else
3318 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3319 store_parameter(tmp, 0);
3320 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3321 store_parameter(tmp, 1);
3322 store_parameter(length, 2);
3323 #endif // _LP64
3324 if (basic_type == T_OBJECT) {
3325 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::oop_arraycopy), 0);
3326 } else {
3327 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Runtime1::primitive_arraycopy), 0);
3328 }
3330 __ bind(*stub->continuation());
3331 }
3334 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
3335 Register obj = op->obj_opr()->as_register(); // may not be an oop
3336 Register hdr = op->hdr_opr()->as_register();
3337 Register lock = op->lock_opr()->as_register();
3338 if (!UseFastLocking) {
3339 __ jmp(*op->stub()->entry());
3340 } else if (op->code() == lir_lock) {
3341 Register scratch = noreg;
3342 if (UseBiasedLocking) {
3343 scratch = op->scratch_opr()->as_register();
3344 }
3345 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3346 // add debug info for NullPointerException only if one is possible
3347 int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
3348 if (op->info() != NULL) {
3349 add_debug_info_for_null_check(null_check_offset, op->info());
3350 }
3351 // done
3352 } else if (op->code() == lir_unlock) {
3353 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3354 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
3355 } else {
3356 Unimplemented();
3357 }
3358 __ bind(*op->stub()->continuation());
3359 }
3362 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
3363 ciMethod* method = op->profiled_method();
3364 int bci = op->profiled_bci();
3366 // Update counter for all call types
3367 ciMethodData* md = method->method_data_or_null();
3368 assert(md != NULL, "Sanity");
3369 ciProfileData* data = md->bci_to_data(bci);
3370 assert(data->is_CounterData(), "need CounterData for calls");
3371 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
3372 Register mdo = op->mdo()->as_register();
3373 __ movoop(mdo, md->constant_encoding());
3374 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
3375 Bytecodes::Code bc = method->java_code_at_bci(bci);
3376 // Perform additional virtual call profiling for invokevirtual and
3377 // invokeinterface bytecodes
3378 if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3379 C1ProfileVirtualCalls) {
3380 assert(op->recv()->is_single_cpu(), "recv must be allocated");
3381 Register recv = op->recv()->as_register();
3382 assert_different_registers(mdo, recv);
3383 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3384 ciKlass* known_klass = op->known_holder();
3385 if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
3386 // We know the type that will be seen at this call site; we can
3387 // statically update the methodDataOop rather than needing to do
3388 // dynamic tests on the receiver type
3390 // NOTE: we should probably put a lock around this search to
3391 // avoid collisions by concurrent compilations
3392 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3393 uint i;
3394 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3395 ciKlass* receiver = vc_data->receiver(i);
3396 if (known_klass->equals(receiver)) {
3397 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3398 __ addptr(data_addr, DataLayout::counter_increment);
3399 return;
3400 }
3401 }
3403 // Receiver type not found in profile data; select an empty slot
3405 // Note that this is less efficient than it should be because it
3406 // always does a write to the receiver part of the
3407 // VirtualCallData rather than just the first time
3408 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3409 ciKlass* receiver = vc_data->receiver(i);
3410 if (receiver == NULL) {
3411 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3412 __ movoop(recv_addr, known_klass->constant_encoding());
3413 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3414 __ addptr(data_addr, DataLayout::counter_increment);
3415 return;
3416 }
3417 }
3418 } else {
3419 __ load_klass(recv, recv);
3420 Label update_done;
3421 type_profile_helper(mdo, md, data, recv, &update_done);
3422 // Receiver did not match any saved receiver and there is no empty row for it.
3423 // Increment total counter to indicate polymorphic case.
3424 __ addptr(counter_addr, DataLayout::counter_increment);
3426 __ bind(update_done);
3427 }
3428 } else {
3429 // Static call
3430 __ addptr(counter_addr, DataLayout::counter_increment);
3431 }
3432 }
3434 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
3435 Unimplemented();
3436 }
3439 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
3440 __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
3441 }
3444 void LIR_Assembler::align_backward_branch_target() {
3445 __ align(BytesPerWord);
3446 }
3449 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3450 if (left->is_single_cpu()) {
3451 __ negl(left->as_register());
3452 move_regs(left->as_register(), dest->as_register());
3454 } else if (left->is_double_cpu()) {
3455 Register lo = left->as_register_lo();
3456 #ifdef _LP64
3457 Register dst = dest->as_register_lo();
3458 __ movptr(dst, lo);
3459 __ negptr(dst);
3460 #else
3461 Register hi = left->as_register_hi();
3462 __ lneg(hi, lo);
3463 if (dest->as_register_lo() == hi) {
3464 assert(dest->as_register_hi() != lo, "destroying register");
3465 move_regs(hi, dest->as_register_hi());
3466 move_regs(lo, dest->as_register_lo());
3467 } else {
3468 move_regs(lo, dest->as_register_lo());
3469 move_regs(hi, dest->as_register_hi());
3470 }
3471 #endif // _LP64
3473 } else if (dest->is_single_xmm()) {
3474 if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
3475 __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
3476 }
3477 __ xorps(dest->as_xmm_float_reg(),
3478 ExternalAddress((address)float_signflip_pool));
3480 } else if (dest->is_double_xmm()) {
3481 if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
3482 __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
3483 }
3484 __ xorpd(dest->as_xmm_double_reg(),
3485 ExternalAddress((address)double_signflip_pool));
3487 } else if (left->is_single_fpu() || left->is_double_fpu()) {
3488 assert(left->fpu() == 0, "arg must be on TOS");
3489 assert(dest->fpu() == 0, "dest must be TOS");
3490 __ fchs();
3492 } else {
3493 ShouldNotReachHere();
3494 }
3495 }
3498 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
3499 assert(addr->is_address() && dest->is_register(), "check");
3500 Register reg;
3501 reg = dest->as_pointer_register();
3502 __ lea(reg, as_Address(addr->as_address_ptr()));
3503 }
3507 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3508 assert(!tmp->is_valid(), "don't need temporary");
3509 __ call(RuntimeAddress(dest));
3510 if (info != NULL) {
3511 add_call_info_here(info);
3512 }
3513 }
3516 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3517 assert(type == T_LONG, "only for volatile long fields");
3519 if (info != NULL) {
3520 add_debug_info_for_null_check_here(info);
3521 }
3523 if (src->is_double_xmm()) {
3524 if (dest->is_double_cpu()) {
3525 #ifdef _LP64
3526 __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
3527 #else
3528 __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
3529 __ psrlq(src->as_xmm_double_reg(), 32);
3530 __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
3531 #endif // _LP64
3532 } else if (dest->is_double_stack()) {
3533 __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
3534 } else if (dest->is_address()) {
3535 __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
3536 } else {
3537 ShouldNotReachHere();
3538 }
3540 } else if (dest->is_double_xmm()) {
3541 if (src->is_double_stack()) {
3542 __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
3543 } else if (src->is_address()) {
3544 __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
3545 } else {
3546 ShouldNotReachHere();
3547 }
3549 } else if (src->is_double_fpu()) {
3550 assert(src->fpu_regnrLo() == 0, "must be TOS");
3551 if (dest->is_double_stack()) {
3552 __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
3553 } else if (dest->is_address()) {
3554 __ fistp_d(as_Address(dest->as_address_ptr()));
3555 } else {
3556 ShouldNotReachHere();
3557 }
3559 } else if (dest->is_double_fpu()) {
3560 assert(dest->fpu_regnrLo() == 0, "must be TOS");
3561 if (src->is_double_stack()) {
3562 __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
3563 } else if (src->is_address()) {
3564 __ fild_d(as_Address(src->as_address_ptr()));
3565 } else {
3566 ShouldNotReachHere();
3567 }
3568 } else {
3569 ShouldNotReachHere();
3570 }
3571 }
3574 void LIR_Assembler::membar() {
3575 // QQQ sparc TSO uses this,
3576 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
3577 }
3579 void LIR_Assembler::membar_acquire() {
3580 // No x86 machines currently require load fences
3581 // __ load_fence();
3582 }
3584 void LIR_Assembler::membar_release() {
3585 // No x86 machines currently require store fences
3586 // __ store_fence();
3587 }
3589 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
3590 assert(result_reg->is_register(), "check");
3591 #ifdef _LP64
3592 // __ get_thread(result_reg->as_register_lo());
3593 __ mov(result_reg->as_register(), r15_thread);
3594 #else
3595 __ get_thread(result_reg->as_register());
3596 #endif // _LP64
3597 }
3600 void LIR_Assembler::peephole(LIR_List*) {
3601 // do nothing for now
3602 }
3605 #undef __