Fri, 18 Oct 2013 10:41:56 +0200
8026844: Various Math functions needs intrinsification
Reviewed-by: kvn, twisti
1 /*
2 * Copyright (c) 2000, 2013, 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 "asm/macroAssembler.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "c1/c1_Compilation.hpp"
29 #include "c1/c1_LIRAssembler.hpp"
30 #include "c1/c1_MacroAssembler.hpp"
31 #include "c1/c1_Runtime1.hpp"
32 #include "c1/c1_ValueStack.hpp"
33 #include "ci/ciArrayKlass.hpp"
34 #include "ci/ciInstance.hpp"
35 #include "gc_interface/collectedHeap.hpp"
36 #include "memory/barrierSet.hpp"
37 #include "memory/cardTableModRefBS.hpp"
38 #include "nativeInst_x86.hpp"
39 #include "oops/objArrayKlass.hpp"
40 #include "runtime/sharedRuntime.hpp"
43 // These masks are used to provide 128-bit aligned bitmasks to the XMM
44 // instructions, to allow sign-masking or sign-bit flipping. They allow
45 // fast versions of NegF/NegD and AbsF/AbsD.
47 // Note: 'double' and 'long long' have 32-bits alignment on x86.
48 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
49 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
50 // of 128-bits operands for SSE instructions.
51 jlong *operand = (jlong*)(((intptr_t)adr) & ((intptr_t)(~0xF)));
52 // Store the value to a 128-bits operand.
53 operand[0] = lo;
54 operand[1] = hi;
55 return operand;
56 }
58 // Buffer for 128-bits masks used by SSE instructions.
59 static jlong fp_signmask_pool[(4+1)*2]; // 4*128bits(data) + 128bits(alignment)
61 // Static initialization during VM startup.
62 static jlong *float_signmask_pool = double_quadword(&fp_signmask_pool[1*2], CONST64(0x7FFFFFFF7FFFFFFF), CONST64(0x7FFFFFFF7FFFFFFF));
63 static jlong *double_signmask_pool = double_quadword(&fp_signmask_pool[2*2], CONST64(0x7FFFFFFFFFFFFFFF), CONST64(0x7FFFFFFFFFFFFFFF));
64 static jlong *float_signflip_pool = double_quadword(&fp_signmask_pool[3*2], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
65 static jlong *double_signflip_pool = double_quadword(&fp_signmask_pool[4*2], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
69 NEEDS_CLEANUP // remove this definitions ?
70 const Register IC_Klass = rax; // where the IC klass is cached
71 const Register SYNC_header = rax; // synchronization header
72 const Register SHIFT_count = rcx; // where count for shift operations must be
74 #define __ _masm->
77 static void select_different_registers(Register preserve,
78 Register extra,
79 Register &tmp1,
80 Register &tmp2) {
81 if (tmp1 == preserve) {
82 assert_different_registers(tmp1, tmp2, extra);
83 tmp1 = extra;
84 } else if (tmp2 == preserve) {
85 assert_different_registers(tmp1, tmp2, extra);
86 tmp2 = extra;
87 }
88 assert_different_registers(preserve, tmp1, tmp2);
89 }
93 static void select_different_registers(Register preserve,
94 Register extra,
95 Register &tmp1,
96 Register &tmp2,
97 Register &tmp3) {
98 if (tmp1 == preserve) {
99 assert_different_registers(tmp1, tmp2, tmp3, extra);
100 tmp1 = extra;
101 } else if (tmp2 == preserve) {
102 assert_different_registers(tmp1, tmp2, tmp3, extra);
103 tmp2 = extra;
104 } else if (tmp3 == preserve) {
105 assert_different_registers(tmp1, tmp2, tmp3, extra);
106 tmp3 = extra;
107 }
108 assert_different_registers(preserve, tmp1, tmp2, tmp3);
109 }
113 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
114 if (opr->is_constant()) {
115 LIR_Const* constant = opr->as_constant_ptr();
116 switch (constant->type()) {
117 case T_INT: {
118 return true;
119 }
121 default:
122 return false;
123 }
124 }
125 return false;
126 }
129 LIR_Opr LIR_Assembler::receiverOpr() {
130 return FrameMap::receiver_opr;
131 }
133 LIR_Opr LIR_Assembler::osrBufferPointer() {
134 return FrameMap::as_pointer_opr(receiverOpr()->as_register());
135 }
137 //--------------fpu register translations-----------------------
140 address LIR_Assembler::float_constant(float f) {
141 address const_addr = __ float_constant(f);
142 if (const_addr == NULL) {
143 bailout("const section overflow");
144 return __ code()->consts()->start();
145 } else {
146 return const_addr;
147 }
148 }
151 address LIR_Assembler::double_constant(double d) {
152 address const_addr = __ double_constant(d);
153 if (const_addr == NULL) {
154 bailout("const section overflow");
155 return __ code()->consts()->start();
156 } else {
157 return const_addr;
158 }
159 }
162 void LIR_Assembler::set_24bit_FPU() {
163 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_24()));
164 }
166 void LIR_Assembler::reset_FPU() {
167 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
168 }
170 void LIR_Assembler::fpop() {
171 __ fpop();
172 }
174 void LIR_Assembler::fxch(int i) {
175 __ fxch(i);
176 }
178 void LIR_Assembler::fld(int i) {
179 __ fld_s(i);
180 }
182 void LIR_Assembler::ffree(int i) {
183 __ ffree(i);
184 }
186 void LIR_Assembler::breakpoint() {
187 __ int3();
188 }
190 void LIR_Assembler::push(LIR_Opr opr) {
191 if (opr->is_single_cpu()) {
192 __ push_reg(opr->as_register());
193 } else if (opr->is_double_cpu()) {
194 NOT_LP64(__ push_reg(opr->as_register_hi()));
195 __ push_reg(opr->as_register_lo());
196 } else if (opr->is_stack()) {
197 __ push_addr(frame_map()->address_for_slot(opr->single_stack_ix()));
198 } else if (opr->is_constant()) {
199 LIR_Const* const_opr = opr->as_constant_ptr();
200 if (const_opr->type() == T_OBJECT) {
201 __ push_oop(const_opr->as_jobject());
202 } else if (const_opr->type() == T_INT) {
203 __ push_jint(const_opr->as_jint());
204 } else {
205 ShouldNotReachHere();
206 }
208 } else {
209 ShouldNotReachHere();
210 }
211 }
213 void LIR_Assembler::pop(LIR_Opr opr) {
214 if (opr->is_single_cpu()) {
215 __ pop_reg(opr->as_register());
216 } else {
217 ShouldNotReachHere();
218 }
219 }
221 bool LIR_Assembler::is_literal_address(LIR_Address* addr) {
222 return addr->base()->is_illegal() && addr->index()->is_illegal();
223 }
225 //-------------------------------------------
227 Address LIR_Assembler::as_Address(LIR_Address* addr) {
228 return as_Address(addr, rscratch1);
229 }
231 Address LIR_Assembler::as_Address(LIR_Address* addr, Register tmp) {
232 if (addr->base()->is_illegal()) {
233 assert(addr->index()->is_illegal(), "must be illegal too");
234 AddressLiteral laddr((address)addr->disp(), relocInfo::none);
235 if (! __ reachable(laddr)) {
236 __ movptr(tmp, laddr.addr());
237 Address res(tmp, 0);
238 return res;
239 } else {
240 return __ as_Address(laddr);
241 }
242 }
244 Register base = addr->base()->as_pointer_register();
246 if (addr->index()->is_illegal()) {
247 return Address( base, addr->disp());
248 } else if (addr->index()->is_cpu_register()) {
249 Register index = addr->index()->as_pointer_register();
250 return Address(base, index, (Address::ScaleFactor) addr->scale(), addr->disp());
251 } else if (addr->index()->is_constant()) {
252 intptr_t addr_offset = (addr->index()->as_constant_ptr()->as_jint() << addr->scale()) + addr->disp();
253 assert(Assembler::is_simm32(addr_offset), "must be");
255 return Address(base, addr_offset);
256 } else {
257 Unimplemented();
258 return Address();
259 }
260 }
263 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
264 Address base = as_Address(addr);
265 return Address(base._base, base._index, base._scale, base._disp + BytesPerWord);
266 }
269 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
270 return as_Address(addr);
271 }
274 void LIR_Assembler::osr_entry() {
275 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
276 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
277 ValueStack* entry_state = osr_entry->state();
278 int number_of_locks = entry_state->locks_size();
280 // we jump here if osr happens with the interpreter
281 // state set up to continue at the beginning of the
282 // loop that triggered osr - in particular, we have
283 // the following registers setup:
284 //
285 // rcx: osr buffer
286 //
288 // build frame
289 ciMethod* m = compilation()->method();
290 __ build_frame(initial_frame_size_in_bytes());
292 // OSR buffer is
293 //
294 // locals[nlocals-1..0]
295 // monitors[0..number_of_locks]
296 //
297 // locals is a direct copy of the interpreter frame so in the osr buffer
298 // so first slot in the local array is the last local from the interpreter
299 // and last slot is local[0] (receiver) from the interpreter
300 //
301 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
302 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
303 // in the interpreter frame (the method lock if a sync method)
305 // Initialize monitors in the compiled activation.
306 // rcx: pointer to osr buffer
307 //
308 // All other registers are dead at this point and the locals will be
309 // copied into place by code emitted in the IR.
311 Register OSR_buf = osrBufferPointer()->as_pointer_register();
312 { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
313 int monitor_offset = BytesPerWord * method()->max_locals() +
314 (2 * BytesPerWord) * (number_of_locks - 1);
315 // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
316 // the OSR buffer using 2 word entries: first the lock and then
317 // the oop.
318 for (int i = 0; i < number_of_locks; i++) {
319 int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
320 #ifdef ASSERT
321 // verify the interpreter's monitor has a non-null object
322 {
323 Label L;
324 __ cmpptr(Address(OSR_buf, slot_offset + 1*BytesPerWord), (int32_t)NULL_WORD);
325 __ jcc(Assembler::notZero, L);
326 __ stop("locked object is NULL");
327 __ bind(L);
328 }
329 #endif
330 __ movptr(rbx, Address(OSR_buf, slot_offset + 0));
331 __ movptr(frame_map()->address_for_monitor_lock(i), rbx);
332 __ movptr(rbx, Address(OSR_buf, slot_offset + 1*BytesPerWord));
333 __ movptr(frame_map()->address_for_monitor_object(i), rbx);
334 }
335 }
336 }
339 // inline cache check; done before the frame is built.
340 int LIR_Assembler::check_icache() {
341 Register receiver = FrameMap::receiver_opr->as_register();
342 Register ic_klass = IC_Klass;
343 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
344 const bool do_post_padding = VerifyOops || UseCompressedClassPointers;
345 if (!do_post_padding) {
346 // insert some nops so that the verified entry point is aligned on CodeEntryAlignment
347 while ((__ offset() + ic_cmp_size) % CodeEntryAlignment != 0) {
348 __ nop();
349 }
350 }
351 int offset = __ offset();
352 __ inline_cache_check(receiver, IC_Klass);
353 assert(__ offset() % CodeEntryAlignment == 0 || do_post_padding, "alignment must be correct");
354 if (do_post_padding) {
355 // force alignment after the cache check.
356 // It's been verified to be aligned if !VerifyOops
357 __ align(CodeEntryAlignment);
358 }
359 return offset;
360 }
363 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo* info) {
364 jobject o = NULL;
365 PatchingStub* patch = new PatchingStub(_masm, patching_id(info));
366 __ movoop(reg, o);
367 patching_epilog(patch, lir_patch_normal, reg, info);
368 }
370 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo* info) {
371 Metadata* o = NULL;
372 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id);
373 __ mov_metadata(reg, o);
374 patching_epilog(patch, lir_patch_normal, reg, info);
375 }
377 // This specifies the rsp decrement needed to build the frame
378 int LIR_Assembler::initial_frame_size_in_bytes() {
379 // if rounding, must let FrameMap know!
381 // The frame_map records size in slots (32bit word)
383 // subtract two words to account for return address and link
384 return (frame_map()->framesize() - (2*VMRegImpl::slots_per_word)) * VMRegImpl::stack_slot_size;
385 }
388 int LIR_Assembler::emit_exception_handler() {
389 // if the last instruction is a call (typically to do a throw which
390 // is coming at the end after block reordering) the return address
391 // must still point into the code area in order to avoid assertion
392 // failures when searching for the corresponding bci => add a nop
393 // (was bug 5/14/1999 - gri)
394 __ nop();
396 // generate code for exception handler
397 address handler_base = __ start_a_stub(exception_handler_size);
398 if (handler_base == NULL) {
399 // not enough space left for the handler
400 bailout("exception handler overflow");
401 return -1;
402 }
404 int offset = code_offset();
406 // the exception oop and pc are in rax, and rdx
407 // no other registers need to be preserved, so invalidate them
408 __ invalidate_registers(false, true, true, false, true, true);
410 // check that there is really an exception
411 __ verify_not_null_oop(rax);
413 // search an exception handler (rax: exception oop, rdx: throwing pc)
414 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::handle_exception_from_callee_id)));
415 __ should_not_reach_here();
416 guarantee(code_offset() - offset <= exception_handler_size, "overflow");
417 __ end_a_stub();
419 return offset;
420 }
423 // Emit the code to remove the frame from the stack in the exception
424 // unwind path.
425 int LIR_Assembler::emit_unwind_handler() {
426 #ifndef PRODUCT
427 if (CommentedAssembly) {
428 _masm->block_comment("Unwind handler");
429 }
430 #endif
432 int offset = code_offset();
434 // Fetch the exception from TLS and clear out exception related thread state
435 Register thread = NOT_LP64(rsi) LP64_ONLY(r15_thread);
436 NOT_LP64(__ get_thread(rsi));
437 __ movptr(rax, Address(thread, JavaThread::exception_oop_offset()));
438 __ movptr(Address(thread, JavaThread::exception_oop_offset()), (intptr_t)NULL_WORD);
439 __ movptr(Address(thread, JavaThread::exception_pc_offset()), (intptr_t)NULL_WORD);
441 __ bind(_unwind_handler_entry);
442 __ verify_not_null_oop(rax);
443 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
444 __ mov(rbx, rax); // Preserve the exception (rbx is always callee-saved)
445 }
447 // Preform needed unlocking
448 MonitorExitStub* stub = NULL;
449 if (method()->is_synchronized()) {
450 monitor_address(0, FrameMap::rax_opr);
451 stub = new MonitorExitStub(FrameMap::rax_opr, true, 0);
452 __ unlock_object(rdi, rsi, rax, *stub->entry());
453 __ bind(*stub->continuation());
454 }
456 if (compilation()->env()->dtrace_method_probes()) {
457 #ifdef _LP64
458 __ mov(rdi, r15_thread);
459 __ mov_metadata(rsi, method()->constant_encoding());
460 #else
461 __ get_thread(rax);
462 __ movptr(Address(rsp, 0), rax);
463 __ mov_metadata(Address(rsp, sizeof(void*)), method()->constant_encoding());
464 #endif
465 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
466 }
468 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
469 __ mov(rax, rbx); // Restore the exception
470 }
472 // remove the activation and dispatch to the unwind handler
473 __ remove_frame(initial_frame_size_in_bytes());
474 __ jump(RuntimeAddress(Runtime1::entry_for(Runtime1::unwind_exception_id)));
476 // Emit the slow path assembly
477 if (stub != NULL) {
478 stub->emit_code(this);
479 }
481 return offset;
482 }
485 int LIR_Assembler::emit_deopt_handler() {
486 // if the last instruction is a call (typically to do a throw which
487 // is coming at the end after block reordering) the return address
488 // must still point into the code area in order to avoid assertion
489 // failures when searching for the corresponding bci => add a nop
490 // (was bug 5/14/1999 - gri)
491 __ nop();
493 // generate code for exception handler
494 address handler_base = __ start_a_stub(deopt_handler_size);
495 if (handler_base == NULL) {
496 // not enough space left for the handler
497 bailout("deopt handler overflow");
498 return -1;
499 }
501 int offset = code_offset();
502 InternalAddress here(__ pc());
504 __ pushptr(here.addr());
505 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
506 guarantee(code_offset() - offset <= deopt_handler_size, "overflow");
507 __ end_a_stub();
509 return offset;
510 }
513 // This is the fast version of java.lang.String.compare; it has not
514 // OSR-entry and therefore, we generate a slow version for OSR's
515 void LIR_Assembler::emit_string_compare(LIR_Opr arg0, LIR_Opr arg1, LIR_Opr dst, CodeEmitInfo* info) {
516 __ movptr (rbx, rcx); // receiver is in rcx
517 __ movptr (rax, arg1->as_register());
519 // Get addresses of first characters from both Strings
520 __ load_heap_oop(rsi, Address(rax, java_lang_String::value_offset_in_bytes()));
521 if (java_lang_String::has_offset_field()) {
522 __ movptr (rcx, Address(rax, java_lang_String::offset_offset_in_bytes()));
523 __ movl (rax, Address(rax, java_lang_String::count_offset_in_bytes()));
524 __ lea (rsi, Address(rsi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
525 } else {
526 __ movl (rax, Address(rsi, arrayOopDesc::length_offset_in_bytes()));
527 __ lea (rsi, Address(rsi, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
528 }
530 // rbx, may be NULL
531 add_debug_info_for_null_check_here(info);
532 __ load_heap_oop(rdi, Address(rbx, java_lang_String::value_offset_in_bytes()));
533 if (java_lang_String::has_offset_field()) {
534 __ movptr (rcx, Address(rbx, java_lang_String::offset_offset_in_bytes()));
535 __ movl (rbx, Address(rbx, java_lang_String::count_offset_in_bytes()));
536 __ lea (rdi, Address(rdi, rcx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
537 } else {
538 __ movl (rbx, Address(rdi, arrayOopDesc::length_offset_in_bytes()));
539 __ lea (rdi, Address(rdi, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
540 }
542 // compute minimum length (in rax) and difference of lengths (on top of stack)
543 __ mov (rcx, rbx);
544 __ subptr(rbx, rax); // subtract lengths
545 __ push (rbx); // result
546 __ cmov (Assembler::lessEqual, rax, rcx);
548 // is minimum length 0?
549 Label noLoop, haveResult;
550 __ testptr (rax, rax);
551 __ jcc (Assembler::zero, noLoop);
553 // compare first characters
554 __ load_unsigned_short(rcx, Address(rdi, 0));
555 __ load_unsigned_short(rbx, Address(rsi, 0));
556 __ subl(rcx, rbx);
557 __ jcc(Assembler::notZero, haveResult);
558 // starting loop
559 __ decrement(rax); // we already tested index: skip one
560 __ jcc(Assembler::zero, noLoop);
562 // set rsi.edi to the end of the arrays (arrays have same length)
563 // negate the index
565 __ lea(rsi, Address(rsi, rax, Address::times_2, type2aelembytes(T_CHAR)));
566 __ lea(rdi, Address(rdi, rax, Address::times_2, type2aelembytes(T_CHAR)));
567 __ negptr(rax);
569 // compare the strings in a loop
571 Label loop;
572 __ align(wordSize);
573 __ bind(loop);
574 __ load_unsigned_short(rcx, Address(rdi, rax, Address::times_2, 0));
575 __ load_unsigned_short(rbx, Address(rsi, rax, Address::times_2, 0));
576 __ subl(rcx, rbx);
577 __ jcc(Assembler::notZero, haveResult);
578 __ increment(rax);
579 __ jcc(Assembler::notZero, loop);
581 // strings are equal up to min length
583 __ bind(noLoop);
584 __ pop(rax);
585 return_op(LIR_OprFact::illegalOpr);
587 __ bind(haveResult);
588 // leave instruction is going to discard the TOS value
589 __ mov (rax, rcx); // result of call is in rax,
590 }
593 void LIR_Assembler::return_op(LIR_Opr result) {
594 assert(result->is_illegal() || !result->is_single_cpu() || result->as_register() == rax, "word returns are in rax,");
595 if (!result->is_illegal() && result->is_float_kind() && !result->is_xmm_register()) {
596 assert(result->fpu() == 0, "result must already be on TOS");
597 }
599 // Pop the stack before the safepoint code
600 __ remove_frame(initial_frame_size_in_bytes());
602 bool result_is_oop = result->is_valid() ? result->is_oop() : false;
604 // Note: we do not need to round double result; float result has the right precision
605 // the poll sets the condition code, but no data registers
606 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
607 relocInfo::poll_return_type);
609 if (Assembler::is_polling_page_far()) {
610 __ lea(rscratch1, polling_page);
611 __ relocate(relocInfo::poll_return_type);
612 __ testl(rax, Address(rscratch1, 0));
613 } else {
614 __ testl(rax, polling_page);
615 }
616 __ ret(0);
617 }
620 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
621 AddressLiteral polling_page(os::get_polling_page() + (SafepointPollOffset % os::vm_page_size()),
622 relocInfo::poll_type);
623 guarantee(info != NULL, "Shouldn't be NULL");
624 int offset = __ offset();
625 if (Assembler::is_polling_page_far()) {
626 __ lea(rscratch1, polling_page);
627 offset = __ offset();
628 add_debug_info_for_branch(info);
629 __ testl(rax, Address(rscratch1, 0));
630 } else {
631 add_debug_info_for_branch(info);
632 __ testl(rax, polling_page);
633 }
634 return offset;
635 }
638 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
639 if (from_reg != to_reg) __ mov(to_reg, from_reg);
640 }
642 void LIR_Assembler::swap_reg(Register a, Register b) {
643 __ xchgptr(a, b);
644 }
647 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
648 assert(src->is_constant(), "should not call otherwise");
649 assert(dest->is_register(), "should not call otherwise");
650 LIR_Const* c = src->as_constant_ptr();
652 switch (c->type()) {
653 case T_INT: {
654 assert(patch_code == lir_patch_none, "no patching handled here");
655 __ movl(dest->as_register(), c->as_jint());
656 break;
657 }
659 case T_ADDRESS: {
660 assert(patch_code == lir_patch_none, "no patching handled here");
661 __ movptr(dest->as_register(), c->as_jint());
662 break;
663 }
665 case T_LONG: {
666 assert(patch_code == lir_patch_none, "no patching handled here");
667 #ifdef _LP64
668 __ movptr(dest->as_register_lo(), (intptr_t)c->as_jlong());
669 #else
670 __ movptr(dest->as_register_lo(), c->as_jint_lo());
671 __ movptr(dest->as_register_hi(), c->as_jint_hi());
672 #endif // _LP64
673 break;
674 }
676 case T_OBJECT: {
677 if (patch_code != lir_patch_none) {
678 jobject2reg_with_patching(dest->as_register(), info);
679 } else {
680 __ movoop(dest->as_register(), c->as_jobject());
681 }
682 break;
683 }
685 case T_METADATA: {
686 if (patch_code != lir_patch_none) {
687 klass2reg_with_patching(dest->as_register(), info);
688 } else {
689 __ mov_metadata(dest->as_register(), c->as_metadata());
690 }
691 break;
692 }
694 case T_FLOAT: {
695 if (dest->is_single_xmm()) {
696 if (c->is_zero_float()) {
697 __ xorps(dest->as_xmm_float_reg(), dest->as_xmm_float_reg());
698 } else {
699 __ movflt(dest->as_xmm_float_reg(),
700 InternalAddress(float_constant(c->as_jfloat())));
701 }
702 } else {
703 assert(dest->is_single_fpu(), "must be");
704 assert(dest->fpu_regnr() == 0, "dest must be TOS");
705 if (c->is_zero_float()) {
706 __ fldz();
707 } else if (c->is_one_float()) {
708 __ fld1();
709 } else {
710 __ fld_s (InternalAddress(float_constant(c->as_jfloat())));
711 }
712 }
713 break;
714 }
716 case T_DOUBLE: {
717 if (dest->is_double_xmm()) {
718 if (c->is_zero_double()) {
719 __ xorpd(dest->as_xmm_double_reg(), dest->as_xmm_double_reg());
720 } else {
721 __ movdbl(dest->as_xmm_double_reg(),
722 InternalAddress(double_constant(c->as_jdouble())));
723 }
724 } else {
725 assert(dest->is_double_fpu(), "must be");
726 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
727 if (c->is_zero_double()) {
728 __ fldz();
729 } else if (c->is_one_double()) {
730 __ fld1();
731 } else {
732 __ fld_d (InternalAddress(double_constant(c->as_jdouble())));
733 }
734 }
735 break;
736 }
738 default:
739 ShouldNotReachHere();
740 }
741 }
743 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
744 assert(src->is_constant(), "should not call otherwise");
745 assert(dest->is_stack(), "should not call otherwise");
746 LIR_Const* c = src->as_constant_ptr();
748 switch (c->type()) {
749 case T_INT: // fall through
750 case T_FLOAT:
751 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
752 break;
754 case T_ADDRESS:
755 __ movptr(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jint_bits());
756 break;
758 case T_OBJECT:
759 __ movoop(frame_map()->address_for_slot(dest->single_stack_ix()), c->as_jobject());
760 break;
762 case T_LONG: // fall through
763 case T_DOUBLE:
764 #ifdef _LP64
765 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
766 lo_word_offset_in_bytes), (intptr_t)c->as_jlong_bits());
767 #else
768 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
769 lo_word_offset_in_bytes), c->as_jint_lo_bits());
770 __ movptr(frame_map()->address_for_slot(dest->double_stack_ix(),
771 hi_word_offset_in_bytes), c->as_jint_hi_bits());
772 #endif // _LP64
773 break;
775 default:
776 ShouldNotReachHere();
777 }
778 }
780 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
781 assert(src->is_constant(), "should not call otherwise");
782 assert(dest->is_address(), "should not call otherwise");
783 LIR_Const* c = src->as_constant_ptr();
784 LIR_Address* addr = dest->as_address_ptr();
786 int null_check_here = code_offset();
787 switch (type) {
788 case T_INT: // fall through
789 case T_FLOAT:
790 __ movl(as_Address(addr), c->as_jint_bits());
791 break;
793 case T_ADDRESS:
794 __ movptr(as_Address(addr), c->as_jint_bits());
795 break;
797 case T_OBJECT: // fall through
798 case T_ARRAY:
799 if (c->as_jobject() == NULL) {
800 if (UseCompressedOops && !wide) {
801 __ movl(as_Address(addr), (int32_t)NULL_WORD);
802 } else {
803 __ movptr(as_Address(addr), NULL_WORD);
804 }
805 } else {
806 if (is_literal_address(addr)) {
807 ShouldNotReachHere();
808 __ movoop(as_Address(addr, noreg), c->as_jobject());
809 } else {
810 #ifdef _LP64
811 __ movoop(rscratch1, c->as_jobject());
812 if (UseCompressedOops && !wide) {
813 __ encode_heap_oop(rscratch1);
814 null_check_here = code_offset();
815 __ movl(as_Address_lo(addr), rscratch1);
816 } else {
817 null_check_here = code_offset();
818 __ movptr(as_Address_lo(addr), rscratch1);
819 }
820 #else
821 __ movoop(as_Address(addr), c->as_jobject());
822 #endif
823 }
824 }
825 break;
827 case T_LONG: // fall through
828 case T_DOUBLE:
829 #ifdef _LP64
830 if (is_literal_address(addr)) {
831 ShouldNotReachHere();
832 __ movptr(as_Address(addr, r15_thread), (intptr_t)c->as_jlong_bits());
833 } else {
834 __ movptr(r10, (intptr_t)c->as_jlong_bits());
835 null_check_here = code_offset();
836 __ movptr(as_Address_lo(addr), r10);
837 }
838 #else
839 // Always reachable in 32bit so this doesn't produce useless move literal
840 __ movptr(as_Address_hi(addr), c->as_jint_hi_bits());
841 __ movptr(as_Address_lo(addr), c->as_jint_lo_bits());
842 #endif // _LP64
843 break;
845 case T_BOOLEAN: // fall through
846 case T_BYTE:
847 __ movb(as_Address(addr), c->as_jint() & 0xFF);
848 break;
850 case T_CHAR: // fall through
851 case T_SHORT:
852 __ movw(as_Address(addr), c->as_jint() & 0xFFFF);
853 break;
855 default:
856 ShouldNotReachHere();
857 };
859 if (info != NULL) {
860 add_debug_info_for_null_check(null_check_here, info);
861 }
862 }
865 void LIR_Assembler::reg2reg(LIR_Opr src, LIR_Opr dest) {
866 assert(src->is_register(), "should not call otherwise");
867 assert(dest->is_register(), "should not call otherwise");
869 // move between cpu-registers
870 if (dest->is_single_cpu()) {
871 #ifdef _LP64
872 if (src->type() == T_LONG) {
873 // Can do LONG -> OBJECT
874 move_regs(src->as_register_lo(), dest->as_register());
875 return;
876 }
877 #endif
878 assert(src->is_single_cpu(), "must match");
879 if (src->type() == T_OBJECT) {
880 __ verify_oop(src->as_register());
881 }
882 move_regs(src->as_register(), dest->as_register());
884 } else if (dest->is_double_cpu()) {
885 #ifdef _LP64
886 if (src->type() == T_OBJECT || src->type() == T_ARRAY) {
887 // Surprising to me but we can see move of a long to t_object
888 __ verify_oop(src->as_register());
889 move_regs(src->as_register(), dest->as_register_lo());
890 return;
891 }
892 #endif
893 assert(src->is_double_cpu(), "must match");
894 Register f_lo = src->as_register_lo();
895 Register f_hi = src->as_register_hi();
896 Register t_lo = dest->as_register_lo();
897 Register t_hi = dest->as_register_hi();
898 #ifdef _LP64
899 assert(f_hi == f_lo, "must be same");
900 assert(t_hi == t_lo, "must be same");
901 move_regs(f_lo, t_lo);
902 #else
903 assert(f_lo != f_hi && t_lo != t_hi, "invalid register allocation");
906 if (f_lo == t_hi && f_hi == t_lo) {
907 swap_reg(f_lo, f_hi);
908 } else if (f_hi == t_lo) {
909 assert(f_lo != t_hi, "overwriting register");
910 move_regs(f_hi, t_hi);
911 move_regs(f_lo, t_lo);
912 } else {
913 assert(f_hi != t_lo, "overwriting register");
914 move_regs(f_lo, t_lo);
915 move_regs(f_hi, t_hi);
916 }
917 #endif // LP64
919 // special moves from fpu-register to xmm-register
920 // necessary for method results
921 } else if (src->is_single_xmm() && !dest->is_single_xmm()) {
922 __ movflt(Address(rsp, 0), src->as_xmm_float_reg());
923 __ fld_s(Address(rsp, 0));
924 } else if (src->is_double_xmm() && !dest->is_double_xmm()) {
925 __ movdbl(Address(rsp, 0), src->as_xmm_double_reg());
926 __ fld_d(Address(rsp, 0));
927 } else if (dest->is_single_xmm() && !src->is_single_xmm()) {
928 __ fstp_s(Address(rsp, 0));
929 __ movflt(dest->as_xmm_float_reg(), Address(rsp, 0));
930 } else if (dest->is_double_xmm() && !src->is_double_xmm()) {
931 __ fstp_d(Address(rsp, 0));
932 __ movdbl(dest->as_xmm_double_reg(), Address(rsp, 0));
934 // move between xmm-registers
935 } else if (dest->is_single_xmm()) {
936 assert(src->is_single_xmm(), "must match");
937 __ movflt(dest->as_xmm_float_reg(), src->as_xmm_float_reg());
938 } else if (dest->is_double_xmm()) {
939 assert(src->is_double_xmm(), "must match");
940 __ movdbl(dest->as_xmm_double_reg(), src->as_xmm_double_reg());
942 // move between fpu-registers (no instruction necessary because of fpu-stack)
943 } else if (dest->is_single_fpu() || dest->is_double_fpu()) {
944 assert(src->is_single_fpu() || src->is_double_fpu(), "must match");
945 assert(src->fpu() == dest->fpu(), "currently should be nothing to do");
946 } else {
947 ShouldNotReachHere();
948 }
949 }
951 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
952 assert(src->is_register(), "should not call otherwise");
953 assert(dest->is_stack(), "should not call otherwise");
955 if (src->is_single_cpu()) {
956 Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
957 if (type == T_OBJECT || type == T_ARRAY) {
958 __ verify_oop(src->as_register());
959 __ movptr (dst, src->as_register());
960 } else if (type == T_METADATA) {
961 __ movptr (dst, src->as_register());
962 } else {
963 __ movl (dst, src->as_register());
964 }
966 } else if (src->is_double_cpu()) {
967 Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix(), lo_word_offset_in_bytes);
968 Address dstHI = frame_map()->address_for_slot(dest->double_stack_ix(), hi_word_offset_in_bytes);
969 __ movptr (dstLO, src->as_register_lo());
970 NOT_LP64(__ movptr (dstHI, src->as_register_hi()));
972 } else if (src->is_single_xmm()) {
973 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
974 __ movflt(dst_addr, src->as_xmm_float_reg());
976 } else if (src->is_double_xmm()) {
977 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
978 __ movdbl(dst_addr, src->as_xmm_double_reg());
980 } else if (src->is_single_fpu()) {
981 assert(src->fpu_regnr() == 0, "argument must be on TOS");
982 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
983 if (pop_fpu_stack) __ fstp_s (dst_addr);
984 else __ fst_s (dst_addr);
986 } else if (src->is_double_fpu()) {
987 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
988 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
989 if (pop_fpu_stack) __ fstp_d (dst_addr);
990 else __ fst_d (dst_addr);
992 } else {
993 ShouldNotReachHere();
994 }
995 }
998 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 */) {
999 LIR_Address* to_addr = dest->as_address_ptr();
1000 PatchingStub* patch = NULL;
1001 Register compressed_src = rscratch1;
1003 if (type == T_ARRAY || type == T_OBJECT) {
1004 __ verify_oop(src->as_register());
1005 #ifdef _LP64
1006 if (UseCompressedOops && !wide) {
1007 __ movptr(compressed_src, src->as_register());
1008 __ encode_heap_oop(compressed_src);
1009 }
1010 #endif
1011 }
1013 if (patch_code != lir_patch_none) {
1014 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1015 Address toa = as_Address(to_addr);
1016 assert(toa.disp() != 0, "must have");
1017 }
1019 int null_check_here = code_offset();
1020 switch (type) {
1021 case T_FLOAT: {
1022 if (src->is_single_xmm()) {
1023 __ movflt(as_Address(to_addr), src->as_xmm_float_reg());
1024 } else {
1025 assert(src->is_single_fpu(), "must be");
1026 assert(src->fpu_regnr() == 0, "argument must be on TOS");
1027 if (pop_fpu_stack) __ fstp_s(as_Address(to_addr));
1028 else __ fst_s (as_Address(to_addr));
1029 }
1030 break;
1031 }
1033 case T_DOUBLE: {
1034 if (src->is_double_xmm()) {
1035 __ movdbl(as_Address(to_addr), src->as_xmm_double_reg());
1036 } else {
1037 assert(src->is_double_fpu(), "must be");
1038 assert(src->fpu_regnrLo() == 0, "argument must be on TOS");
1039 if (pop_fpu_stack) __ fstp_d(as_Address(to_addr));
1040 else __ fst_d (as_Address(to_addr));
1041 }
1042 break;
1043 }
1045 case T_ARRAY: // fall through
1046 case T_OBJECT: // fall through
1047 if (UseCompressedOops && !wide) {
1048 __ movl(as_Address(to_addr), compressed_src);
1049 } else {
1050 __ movptr(as_Address(to_addr), src->as_register());
1051 }
1052 break;
1053 case T_METADATA:
1054 // We get here to store a method pointer to the stack to pass to
1055 // a dtrace runtime call. This can't work on 64 bit with
1056 // compressed klass ptrs: T_METADATA can be a compressed klass
1057 // ptr or a 64 bit method pointer.
1058 LP64_ONLY(ShouldNotReachHere());
1059 __ movptr(as_Address(to_addr), src->as_register());
1060 break;
1061 case T_ADDRESS:
1062 __ movptr(as_Address(to_addr), src->as_register());
1063 break;
1064 case T_INT:
1065 __ movl(as_Address(to_addr), src->as_register());
1066 break;
1068 case T_LONG: {
1069 Register from_lo = src->as_register_lo();
1070 Register from_hi = src->as_register_hi();
1071 #ifdef _LP64
1072 __ movptr(as_Address_lo(to_addr), from_lo);
1073 #else
1074 Register base = to_addr->base()->as_register();
1075 Register index = noreg;
1076 if (to_addr->index()->is_register()) {
1077 index = to_addr->index()->as_register();
1078 }
1079 if (base == from_lo || index == from_lo) {
1080 assert(base != from_hi, "can't be");
1081 assert(index == noreg || (index != base && index != from_hi), "can't handle this");
1082 __ movl(as_Address_hi(to_addr), from_hi);
1083 if (patch != NULL) {
1084 patching_epilog(patch, lir_patch_high, base, info);
1085 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1086 patch_code = lir_patch_low;
1087 }
1088 __ movl(as_Address_lo(to_addr), from_lo);
1089 } else {
1090 assert(index == noreg || (index != base && index != from_lo), "can't handle this");
1091 __ movl(as_Address_lo(to_addr), from_lo);
1092 if (patch != NULL) {
1093 patching_epilog(patch, lir_patch_low, base, info);
1094 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1095 patch_code = lir_patch_high;
1096 }
1097 __ movl(as_Address_hi(to_addr), from_hi);
1098 }
1099 #endif // _LP64
1100 break;
1101 }
1103 case T_BYTE: // fall through
1104 case T_BOOLEAN: {
1105 Register src_reg = src->as_register();
1106 Address dst_addr = as_Address(to_addr);
1107 assert(VM_Version::is_P6() || src_reg->has_byte_register(), "must use byte registers if not P6");
1108 __ movb(dst_addr, src_reg);
1109 break;
1110 }
1112 case T_CHAR: // fall through
1113 case T_SHORT:
1114 __ movw(as_Address(to_addr), src->as_register());
1115 break;
1117 default:
1118 ShouldNotReachHere();
1119 }
1120 if (info != NULL) {
1121 add_debug_info_for_null_check(null_check_here, info);
1122 }
1124 if (patch_code != lir_patch_none) {
1125 patching_epilog(patch, patch_code, to_addr->base()->as_register(), info);
1126 }
1127 }
1130 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
1131 assert(src->is_stack(), "should not call otherwise");
1132 assert(dest->is_register(), "should not call otherwise");
1134 if (dest->is_single_cpu()) {
1135 if (type == T_ARRAY || type == T_OBJECT) {
1136 __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1137 __ verify_oop(dest->as_register());
1138 } else if (type == T_METADATA) {
1139 __ movptr(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1140 } else {
1141 __ movl(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()));
1142 }
1144 } else if (dest->is_double_cpu()) {
1145 Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix(), lo_word_offset_in_bytes);
1146 Address src_addr_HI = frame_map()->address_for_slot(src->double_stack_ix(), hi_word_offset_in_bytes);
1147 __ movptr(dest->as_register_lo(), src_addr_LO);
1148 NOT_LP64(__ movptr(dest->as_register_hi(), src_addr_HI));
1150 } else if (dest->is_single_xmm()) {
1151 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1152 __ movflt(dest->as_xmm_float_reg(), src_addr);
1154 } else if (dest->is_double_xmm()) {
1155 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1156 __ movdbl(dest->as_xmm_double_reg(), src_addr);
1158 } else if (dest->is_single_fpu()) {
1159 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1160 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1161 __ fld_s(src_addr);
1163 } else if (dest->is_double_fpu()) {
1164 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1165 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1166 __ fld_d(src_addr);
1168 } else {
1169 ShouldNotReachHere();
1170 }
1171 }
1174 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
1175 if (src->is_single_stack()) {
1176 if (type == T_OBJECT || type == T_ARRAY) {
1177 __ pushptr(frame_map()->address_for_slot(src ->single_stack_ix()));
1178 __ popptr (frame_map()->address_for_slot(dest->single_stack_ix()));
1179 } else {
1180 #ifndef _LP64
1181 __ pushl(frame_map()->address_for_slot(src ->single_stack_ix()));
1182 __ popl (frame_map()->address_for_slot(dest->single_stack_ix()));
1183 #else
1184 //no pushl on 64bits
1185 __ movl(rscratch1, frame_map()->address_for_slot(src ->single_stack_ix()));
1186 __ movl(frame_map()->address_for_slot(dest->single_stack_ix()), rscratch1);
1187 #endif
1188 }
1190 } else if (src->is_double_stack()) {
1191 #ifdef _LP64
1192 __ pushptr(frame_map()->address_for_slot(src ->double_stack_ix()));
1193 __ popptr (frame_map()->address_for_slot(dest->double_stack_ix()));
1194 #else
1195 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 0));
1196 // push and pop the part at src + wordSize, adding wordSize for the previous push
1197 __ pushl(frame_map()->address_for_slot(src ->double_stack_ix(), 2 * wordSize));
1198 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 2 * wordSize));
1199 __ popl (frame_map()->address_for_slot(dest->double_stack_ix(), 0));
1200 #endif // _LP64
1202 } else {
1203 ShouldNotReachHere();
1204 }
1205 }
1208 void LIR_Assembler::mem2reg(LIR_Opr src, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code, CodeEmitInfo* info, bool wide, bool /* unaligned */) {
1209 assert(src->is_address(), "should not call otherwise");
1210 assert(dest->is_register(), "should not call otherwise");
1212 LIR_Address* addr = src->as_address_ptr();
1213 Address from_addr = as_Address(addr);
1215 if (addr->base()->type() == T_OBJECT) {
1216 __ verify_oop(addr->base()->as_pointer_register());
1217 }
1219 switch (type) {
1220 case T_BOOLEAN: // fall through
1221 case T_BYTE: // fall through
1222 case T_CHAR: // fall through
1223 case T_SHORT:
1224 if (!VM_Version::is_P6() && !from_addr.uses(dest->as_register())) {
1225 // on pre P6 processors we may get partial register stalls
1226 // so blow away the value of to_rinfo before loading a
1227 // partial word into it. Do it here so that it precedes
1228 // the potential patch point below.
1229 __ xorptr(dest->as_register(), dest->as_register());
1230 }
1231 break;
1232 }
1234 PatchingStub* patch = NULL;
1235 if (patch_code != lir_patch_none) {
1236 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1237 assert(from_addr.disp() != 0, "must have");
1238 }
1239 if (info != NULL) {
1240 add_debug_info_for_null_check_here(info);
1241 }
1243 switch (type) {
1244 case T_FLOAT: {
1245 if (dest->is_single_xmm()) {
1246 __ movflt(dest->as_xmm_float_reg(), from_addr);
1247 } else {
1248 assert(dest->is_single_fpu(), "must be");
1249 assert(dest->fpu_regnr() == 0, "dest must be TOS");
1250 __ fld_s(from_addr);
1251 }
1252 break;
1253 }
1255 case T_DOUBLE: {
1256 if (dest->is_double_xmm()) {
1257 __ movdbl(dest->as_xmm_double_reg(), from_addr);
1258 } else {
1259 assert(dest->is_double_fpu(), "must be");
1260 assert(dest->fpu_regnrLo() == 0, "dest must be TOS");
1261 __ fld_d(from_addr);
1262 }
1263 break;
1264 }
1266 case T_OBJECT: // fall through
1267 case T_ARRAY: // fall through
1268 if (UseCompressedOops && !wide) {
1269 __ movl(dest->as_register(), from_addr);
1270 } else {
1271 __ movptr(dest->as_register(), from_addr);
1272 }
1273 break;
1275 case T_ADDRESS:
1276 if (UseCompressedClassPointers && addr->disp() == oopDesc::klass_offset_in_bytes()) {
1277 __ movl(dest->as_register(), from_addr);
1278 } else {
1279 __ movptr(dest->as_register(), from_addr);
1280 }
1281 break;
1282 case T_INT:
1283 __ movl(dest->as_register(), from_addr);
1284 break;
1286 case T_LONG: {
1287 Register to_lo = dest->as_register_lo();
1288 Register to_hi = dest->as_register_hi();
1289 #ifdef _LP64
1290 __ movptr(to_lo, as_Address_lo(addr));
1291 #else
1292 Register base = addr->base()->as_register();
1293 Register index = noreg;
1294 if (addr->index()->is_register()) {
1295 index = addr->index()->as_register();
1296 }
1297 if ((base == to_lo && index == to_hi) ||
1298 (base == to_hi && index == to_lo)) {
1299 // addresses with 2 registers are only formed as a result of
1300 // array access so this code will never have to deal with
1301 // patches or null checks.
1302 assert(info == NULL && patch == NULL, "must be");
1303 __ lea(to_hi, as_Address(addr));
1304 __ movl(to_lo, Address(to_hi, 0));
1305 __ movl(to_hi, Address(to_hi, BytesPerWord));
1306 } else if (base == to_lo || index == to_lo) {
1307 assert(base != to_hi, "can't be");
1308 assert(index == noreg || (index != base && index != to_hi), "can't handle this");
1309 __ movl(to_hi, as_Address_hi(addr));
1310 if (patch != NULL) {
1311 patching_epilog(patch, lir_patch_high, base, info);
1312 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1313 patch_code = lir_patch_low;
1314 }
1315 __ movl(to_lo, as_Address_lo(addr));
1316 } else {
1317 assert(index == noreg || (index != base && index != to_lo), "can't handle this");
1318 __ movl(to_lo, as_Address_lo(addr));
1319 if (patch != NULL) {
1320 patching_epilog(patch, lir_patch_low, base, info);
1321 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1322 patch_code = lir_patch_high;
1323 }
1324 __ movl(to_hi, as_Address_hi(addr));
1325 }
1326 #endif // _LP64
1327 break;
1328 }
1330 case T_BOOLEAN: // fall through
1331 case T_BYTE: {
1332 Register dest_reg = dest->as_register();
1333 assert(VM_Version::is_P6() || dest_reg->has_byte_register(), "must use byte registers if not P6");
1334 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1335 __ movsbl(dest_reg, from_addr);
1336 } else {
1337 __ movb(dest_reg, from_addr);
1338 __ shll(dest_reg, 24);
1339 __ sarl(dest_reg, 24);
1340 }
1341 break;
1342 }
1344 case T_CHAR: {
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 __ movzwl(dest_reg, from_addr);
1349 } else {
1350 __ movw(dest_reg, from_addr);
1351 }
1352 break;
1353 }
1355 case T_SHORT: {
1356 Register dest_reg = dest->as_register();
1357 if (VM_Version::is_P6() || from_addr.uses(dest_reg)) {
1358 __ movswl(dest_reg, from_addr);
1359 } else {
1360 __ movw(dest_reg, from_addr);
1361 __ shll(dest_reg, 16);
1362 __ sarl(dest_reg, 16);
1363 }
1364 break;
1365 }
1367 default:
1368 ShouldNotReachHere();
1369 }
1371 if (patch != NULL) {
1372 patching_epilog(patch, patch_code, addr->base()->as_register(), info);
1373 }
1375 if (type == T_ARRAY || type == T_OBJECT) {
1376 #ifdef _LP64
1377 if (UseCompressedOops && !wide) {
1378 __ decode_heap_oop(dest->as_register());
1379 }
1380 #endif
1381 __ verify_oop(dest->as_register());
1382 } else if (type == T_ADDRESS && addr->disp() == oopDesc::klass_offset_in_bytes()) {
1383 #ifdef _LP64
1384 if (UseCompressedClassPointers) {
1385 __ decode_klass_not_null(dest->as_register());
1386 }
1387 #endif
1388 }
1389 }
1392 void LIR_Assembler::prefetchr(LIR_Opr src) {
1393 LIR_Address* addr = src->as_address_ptr();
1394 Address from_addr = as_Address(addr);
1396 if (VM_Version::supports_sse()) {
1397 switch (ReadPrefetchInstr) {
1398 case 0:
1399 __ prefetchnta(from_addr); break;
1400 case 1:
1401 __ prefetcht0(from_addr); break;
1402 case 2:
1403 __ prefetcht2(from_addr); break;
1404 default:
1405 ShouldNotReachHere(); break;
1406 }
1407 } else if (VM_Version::supports_3dnow_prefetch()) {
1408 __ prefetchr(from_addr);
1409 }
1410 }
1413 void LIR_Assembler::prefetchw(LIR_Opr src) {
1414 LIR_Address* addr = src->as_address_ptr();
1415 Address from_addr = as_Address(addr);
1417 if (VM_Version::supports_sse()) {
1418 switch (AllocatePrefetchInstr) {
1419 case 0:
1420 __ prefetchnta(from_addr); break;
1421 case 1:
1422 __ prefetcht0(from_addr); break;
1423 case 2:
1424 __ prefetcht2(from_addr); break;
1425 case 3:
1426 __ prefetchw(from_addr); break;
1427 default:
1428 ShouldNotReachHere(); break;
1429 }
1430 } else if (VM_Version::supports_3dnow_prefetch()) {
1431 __ prefetchw(from_addr);
1432 }
1433 }
1436 NEEDS_CLEANUP; // This could be static?
1437 Address::ScaleFactor LIR_Assembler::array_element_size(BasicType type) const {
1438 int elem_size = type2aelembytes(type);
1439 switch (elem_size) {
1440 case 1: return Address::times_1;
1441 case 2: return Address::times_2;
1442 case 4: return Address::times_4;
1443 case 8: return Address::times_8;
1444 }
1445 ShouldNotReachHere();
1446 return Address::no_scale;
1447 }
1450 void LIR_Assembler::emit_op3(LIR_Op3* op) {
1451 switch (op->code()) {
1452 case lir_idiv:
1453 case lir_irem:
1454 arithmetic_idiv(op->code(),
1455 op->in_opr1(),
1456 op->in_opr2(),
1457 op->in_opr3(),
1458 op->result_opr(),
1459 op->info());
1460 break;
1461 default: ShouldNotReachHere(); break;
1462 }
1463 }
1465 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
1466 #ifdef ASSERT
1467 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
1468 if (op->block() != NULL) _branch_target_blocks.append(op->block());
1469 if (op->ublock() != NULL) _branch_target_blocks.append(op->ublock());
1470 #endif
1472 if (op->cond() == lir_cond_always) {
1473 if (op->info() != NULL) add_debug_info_for_branch(op->info());
1474 __ jmp (*(op->label()));
1475 } else {
1476 Assembler::Condition acond = Assembler::zero;
1477 if (op->code() == lir_cond_float_branch) {
1478 assert(op->ublock() != NULL, "must have unordered successor");
1479 __ jcc(Assembler::parity, *(op->ublock()->label()));
1480 switch(op->cond()) {
1481 case lir_cond_equal: acond = Assembler::equal; break;
1482 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1483 case lir_cond_less: acond = Assembler::below; break;
1484 case lir_cond_lessEqual: acond = Assembler::belowEqual; break;
1485 case lir_cond_greaterEqual: acond = Assembler::aboveEqual; break;
1486 case lir_cond_greater: acond = Assembler::above; break;
1487 default: ShouldNotReachHere();
1488 }
1489 } else {
1490 switch (op->cond()) {
1491 case lir_cond_equal: acond = Assembler::equal; break;
1492 case lir_cond_notEqual: acond = Assembler::notEqual; break;
1493 case lir_cond_less: acond = Assembler::less; break;
1494 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
1495 case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
1496 case lir_cond_greater: acond = Assembler::greater; break;
1497 case lir_cond_belowEqual: acond = Assembler::belowEqual; break;
1498 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break;
1499 default: ShouldNotReachHere();
1500 }
1501 }
1502 __ jcc(acond,*(op->label()));
1503 }
1504 }
1506 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
1507 LIR_Opr src = op->in_opr();
1508 LIR_Opr dest = op->result_opr();
1510 switch (op->bytecode()) {
1511 case Bytecodes::_i2l:
1512 #ifdef _LP64
1513 __ movl2ptr(dest->as_register_lo(), src->as_register());
1514 #else
1515 move_regs(src->as_register(), dest->as_register_lo());
1516 move_regs(src->as_register(), dest->as_register_hi());
1517 __ sarl(dest->as_register_hi(), 31);
1518 #endif // LP64
1519 break;
1521 case Bytecodes::_l2i:
1522 #ifdef _LP64
1523 __ movl(dest->as_register(), src->as_register_lo());
1524 #else
1525 move_regs(src->as_register_lo(), dest->as_register());
1526 #endif
1527 break;
1529 case Bytecodes::_i2b:
1530 move_regs(src->as_register(), dest->as_register());
1531 __ sign_extend_byte(dest->as_register());
1532 break;
1534 case Bytecodes::_i2c:
1535 move_regs(src->as_register(), dest->as_register());
1536 __ andl(dest->as_register(), 0xFFFF);
1537 break;
1539 case Bytecodes::_i2s:
1540 move_regs(src->as_register(), dest->as_register());
1541 __ sign_extend_short(dest->as_register());
1542 break;
1545 case Bytecodes::_f2d:
1546 case Bytecodes::_d2f:
1547 if (dest->is_single_xmm()) {
1548 __ cvtsd2ss(dest->as_xmm_float_reg(), src->as_xmm_double_reg());
1549 } else if (dest->is_double_xmm()) {
1550 __ cvtss2sd(dest->as_xmm_double_reg(), src->as_xmm_float_reg());
1551 } else {
1552 assert(src->fpu() == dest->fpu(), "register must be equal");
1553 // do nothing (float result is rounded later through spilling)
1554 }
1555 break;
1557 case Bytecodes::_i2f:
1558 case Bytecodes::_i2d:
1559 if (dest->is_single_xmm()) {
1560 __ cvtsi2ssl(dest->as_xmm_float_reg(), src->as_register());
1561 } else if (dest->is_double_xmm()) {
1562 __ cvtsi2sdl(dest->as_xmm_double_reg(), src->as_register());
1563 } else {
1564 assert(dest->fpu() == 0, "result must be on TOS");
1565 __ movl(Address(rsp, 0), src->as_register());
1566 __ fild_s(Address(rsp, 0));
1567 }
1568 break;
1570 case Bytecodes::_f2i:
1571 case Bytecodes::_d2i:
1572 if (src->is_single_xmm()) {
1573 __ cvttss2sil(dest->as_register(), src->as_xmm_float_reg());
1574 } else if (src->is_double_xmm()) {
1575 __ cvttsd2sil(dest->as_register(), src->as_xmm_double_reg());
1576 } else {
1577 assert(src->fpu() == 0, "input must be on TOS");
1578 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_trunc()));
1579 __ fist_s(Address(rsp, 0));
1580 __ movl(dest->as_register(), Address(rsp, 0));
1581 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std()));
1582 }
1584 // IA32 conversion instructions do not match JLS for overflow, underflow and NaN -> fixup in stub
1585 assert(op->stub() != NULL, "stub required");
1586 __ cmpl(dest->as_register(), 0x80000000);
1587 __ jcc(Assembler::equal, *op->stub()->entry());
1588 __ bind(*op->stub()->continuation());
1589 break;
1591 case Bytecodes::_l2f:
1592 case Bytecodes::_l2d:
1593 assert(!dest->is_xmm_register(), "result in xmm register not supported (no SSE instruction present)");
1594 assert(dest->fpu() == 0, "result must be on TOS");
1596 __ movptr(Address(rsp, 0), src->as_register_lo());
1597 NOT_LP64(__ movl(Address(rsp, BytesPerWord), src->as_register_hi()));
1598 __ fild_d(Address(rsp, 0));
1599 // float result is rounded later through spilling
1600 break;
1602 case Bytecodes::_f2l:
1603 case Bytecodes::_d2l:
1604 assert(!src->is_xmm_register(), "input in xmm register not supported (no SSE instruction present)");
1605 assert(src->fpu() == 0, "input must be on TOS");
1606 assert(dest == FrameMap::long0_opr, "runtime stub places result in these registers");
1608 // instruction sequence too long to inline it here
1609 {
1610 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::fpu2long_stub_id)));
1611 }
1612 break;
1614 default: ShouldNotReachHere();
1615 }
1616 }
1618 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
1619 if (op->init_check()) {
1620 __ cmpb(Address(op->klass()->as_register(),
1621 InstanceKlass::init_state_offset()),
1622 InstanceKlass::fully_initialized);
1623 add_debug_info_for_null_check_here(op->stub()->info());
1624 __ jcc(Assembler::notEqual, *op->stub()->entry());
1625 }
1626 __ allocate_object(op->obj()->as_register(),
1627 op->tmp1()->as_register(),
1628 op->tmp2()->as_register(),
1629 op->header_size(),
1630 op->object_size(),
1631 op->klass()->as_register(),
1632 *op->stub()->entry());
1633 __ bind(*op->stub()->continuation());
1634 }
1636 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
1637 Register len = op->len()->as_register();
1638 LP64_ONLY( __ movslq(len, len); )
1640 if (UseSlowPath ||
1641 (!UseFastNewObjectArray && (op->type() == T_OBJECT || op->type() == T_ARRAY)) ||
1642 (!UseFastNewTypeArray && (op->type() != T_OBJECT && op->type() != T_ARRAY))) {
1643 __ jmp(*op->stub()->entry());
1644 } else {
1645 Register tmp1 = op->tmp1()->as_register();
1646 Register tmp2 = op->tmp2()->as_register();
1647 Register tmp3 = op->tmp3()->as_register();
1648 if (len == tmp1) {
1649 tmp1 = tmp3;
1650 } else if (len == tmp2) {
1651 tmp2 = tmp3;
1652 } else if (len == tmp3) {
1653 // everything is ok
1654 } else {
1655 __ mov(tmp3, len);
1656 }
1657 __ allocate_array(op->obj()->as_register(),
1658 len,
1659 tmp1,
1660 tmp2,
1661 arrayOopDesc::header_size(op->type()),
1662 array_element_size(op->type()),
1663 op->klass()->as_register(),
1664 *op->stub()->entry());
1665 }
1666 __ bind(*op->stub()->continuation());
1667 }
1669 void LIR_Assembler::type_profile_helper(Register mdo,
1670 ciMethodData *md, ciProfileData *data,
1671 Register recv, Label* update_done) {
1672 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1673 Label next_test;
1674 // See if the receiver is receiver[n].
1675 __ cmpptr(recv, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
1676 __ jccb(Assembler::notEqual, next_test);
1677 Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
1678 __ addptr(data_addr, DataLayout::counter_increment);
1679 __ jmp(*update_done);
1680 __ bind(next_test);
1681 }
1683 // Didn't find receiver; find next empty slot and fill it in
1684 for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
1685 Label next_test;
1686 Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
1687 __ cmpptr(recv_addr, (intptr_t)NULL_WORD);
1688 __ jccb(Assembler::notEqual, next_test);
1689 __ movptr(recv_addr, recv);
1690 __ movptr(Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))), DataLayout::counter_increment);
1691 __ jmp(*update_done);
1692 __ bind(next_test);
1693 }
1694 }
1696 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
1697 // we always need a stub for the failure case.
1698 CodeStub* stub = op->stub();
1699 Register obj = op->object()->as_register();
1700 Register k_RInfo = op->tmp1()->as_register();
1701 Register klass_RInfo = op->tmp2()->as_register();
1702 Register dst = op->result_opr()->as_register();
1703 ciKlass* k = op->klass();
1704 Register Rtmp1 = noreg;
1706 // check if it needs to be profiled
1707 ciMethodData* md;
1708 ciProfileData* data;
1710 if (op->should_profile()) {
1711 ciMethod* method = op->profiled_method();
1712 assert(method != NULL, "Should have method");
1713 int bci = op->profiled_bci();
1714 md = method->method_data_or_null();
1715 assert(md != NULL, "Sanity");
1716 data = md->bci_to_data(bci);
1717 assert(data != NULL, "need data for type check");
1718 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1719 }
1720 Label profile_cast_success, profile_cast_failure;
1721 Label *success_target = op->should_profile() ? &profile_cast_success : success;
1722 Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
1724 if (obj == k_RInfo) {
1725 k_RInfo = dst;
1726 } else if (obj == klass_RInfo) {
1727 klass_RInfo = dst;
1728 }
1729 if (k->is_loaded() && !UseCompressedClassPointers) {
1730 select_different_registers(obj, dst, k_RInfo, klass_RInfo);
1731 } else {
1732 Rtmp1 = op->tmp3()->as_register();
1733 select_different_registers(obj, dst, k_RInfo, klass_RInfo, Rtmp1);
1734 }
1736 assert_different_registers(obj, k_RInfo, klass_RInfo);
1738 __ cmpptr(obj, (int32_t)NULL_WORD);
1739 if (op->should_profile()) {
1740 Label not_null;
1741 __ jccb(Assembler::notEqual, not_null);
1742 // Object is null; update MDO and exit
1743 Register mdo = klass_RInfo;
1744 __ mov_metadata(mdo, md->constant_encoding());
1745 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1746 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1747 __ orl(data_addr, header_bits);
1748 __ jmp(*obj_is_null);
1749 __ bind(not_null);
1750 } else {
1751 __ jcc(Assembler::equal, *obj_is_null);
1752 }
1754 if (!k->is_loaded()) {
1755 klass2reg_with_patching(k_RInfo, op->info_for_patch());
1756 } else {
1757 #ifdef _LP64
1758 __ mov_metadata(k_RInfo, k->constant_encoding());
1759 #endif // _LP64
1760 }
1761 __ verify_oop(obj);
1763 if (op->fast_check()) {
1764 // get object class
1765 // not a safepoint as obj null check happens earlier
1766 #ifdef _LP64
1767 if (UseCompressedClassPointers) {
1768 __ load_klass(Rtmp1, obj);
1769 __ cmpptr(k_RInfo, Rtmp1);
1770 } else {
1771 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1772 }
1773 #else
1774 if (k->is_loaded()) {
1775 __ cmpklass(Address(obj, oopDesc::klass_offset_in_bytes()), k->constant_encoding());
1776 } else {
1777 __ cmpptr(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
1778 }
1779 #endif
1780 __ jcc(Assembler::notEqual, *failure_target);
1781 // successful cast, fall through to profile or jump
1782 } else {
1783 // get object class
1784 // not a safepoint as obj null check happens earlier
1785 __ load_klass(klass_RInfo, obj);
1786 if (k->is_loaded()) {
1787 // See if we get an immediate positive hit
1788 #ifdef _LP64
1789 __ cmpptr(k_RInfo, Address(klass_RInfo, k->super_check_offset()));
1790 #else
1791 __ cmpklass(Address(klass_RInfo, k->super_check_offset()), k->constant_encoding());
1792 #endif // _LP64
1793 if ((juint)in_bytes(Klass::secondary_super_cache_offset()) != k->super_check_offset()) {
1794 __ jcc(Assembler::notEqual, *failure_target);
1795 // successful cast, fall through to profile or jump
1796 } else {
1797 // See if we get an immediate positive hit
1798 __ jcc(Assembler::equal, *success_target);
1799 // check for self
1800 #ifdef _LP64
1801 __ cmpptr(klass_RInfo, k_RInfo);
1802 #else
1803 __ cmpklass(klass_RInfo, k->constant_encoding());
1804 #endif // _LP64
1805 __ jcc(Assembler::equal, *success_target);
1807 __ push(klass_RInfo);
1808 #ifdef _LP64
1809 __ push(k_RInfo);
1810 #else
1811 __ pushklass(k->constant_encoding());
1812 #endif // _LP64
1813 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1814 __ pop(klass_RInfo);
1815 __ pop(klass_RInfo);
1816 // result is a boolean
1817 __ cmpl(klass_RInfo, 0);
1818 __ jcc(Assembler::equal, *failure_target);
1819 // successful cast, fall through to profile or jump
1820 }
1821 } else {
1822 // perform the fast part of the checking logic
1823 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1824 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1825 __ push(klass_RInfo);
1826 __ push(k_RInfo);
1827 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1828 __ pop(klass_RInfo);
1829 __ pop(k_RInfo);
1830 // result is a boolean
1831 __ cmpl(k_RInfo, 0);
1832 __ jcc(Assembler::equal, *failure_target);
1833 // successful cast, fall through to profile or jump
1834 }
1835 }
1836 if (op->should_profile()) {
1837 Register mdo = klass_RInfo, recv = k_RInfo;
1838 __ bind(profile_cast_success);
1839 __ mov_metadata(mdo, md->constant_encoding());
1840 __ load_klass(recv, obj);
1841 Label update_done;
1842 type_profile_helper(mdo, md, data, recv, success);
1843 __ jmp(*success);
1845 __ bind(profile_cast_failure);
1846 __ mov_metadata(mdo, md->constant_encoding());
1847 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1848 __ subptr(counter_addr, DataLayout::counter_increment);
1849 __ jmp(*failure);
1850 }
1851 __ jmp(*success);
1852 }
1855 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
1856 LIR_Code code = op->code();
1857 if (code == lir_store_check) {
1858 Register value = op->object()->as_register();
1859 Register array = op->array()->as_register();
1860 Register k_RInfo = op->tmp1()->as_register();
1861 Register klass_RInfo = op->tmp2()->as_register();
1862 Register Rtmp1 = op->tmp3()->as_register();
1864 CodeStub* stub = op->stub();
1866 // check if it needs to be profiled
1867 ciMethodData* md;
1868 ciProfileData* data;
1870 if (op->should_profile()) {
1871 ciMethod* method = op->profiled_method();
1872 assert(method != NULL, "Should have method");
1873 int bci = op->profiled_bci();
1874 md = method->method_data_or_null();
1875 assert(md != NULL, "Sanity");
1876 data = md->bci_to_data(bci);
1877 assert(data != NULL, "need data for type check");
1878 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
1879 }
1880 Label profile_cast_success, profile_cast_failure, done;
1881 Label *success_target = op->should_profile() ? &profile_cast_success : &done;
1882 Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
1884 __ cmpptr(value, (int32_t)NULL_WORD);
1885 if (op->should_profile()) {
1886 Label not_null;
1887 __ jccb(Assembler::notEqual, not_null);
1888 // Object is null; update MDO and exit
1889 Register mdo = klass_RInfo;
1890 __ mov_metadata(mdo, md->constant_encoding());
1891 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
1892 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
1893 __ orl(data_addr, header_bits);
1894 __ jmp(done);
1895 __ bind(not_null);
1896 } else {
1897 __ jcc(Assembler::equal, done);
1898 }
1900 add_debug_info_for_null_check_here(op->info_for_exception());
1901 __ load_klass(k_RInfo, array);
1902 __ load_klass(klass_RInfo, value);
1904 // get instance klass (it's already uncompressed)
1905 __ movptr(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
1906 // perform the fast part of the checking logic
1907 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
1908 // call out-of-line instance of __ check_klass_subtype_slow_path(...):
1909 __ push(klass_RInfo);
1910 __ push(k_RInfo);
1911 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
1912 __ pop(klass_RInfo);
1913 __ pop(k_RInfo);
1914 // result is a boolean
1915 __ cmpl(k_RInfo, 0);
1916 __ jcc(Assembler::equal, *failure_target);
1917 // fall through to the success case
1919 if (op->should_profile()) {
1920 Register mdo = klass_RInfo, recv = k_RInfo;
1921 __ bind(profile_cast_success);
1922 __ mov_metadata(mdo, md->constant_encoding());
1923 __ load_klass(recv, value);
1924 Label update_done;
1925 type_profile_helper(mdo, md, data, recv, &done);
1926 __ jmpb(done);
1928 __ bind(profile_cast_failure);
1929 __ mov_metadata(mdo, md->constant_encoding());
1930 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
1931 __ subptr(counter_addr, DataLayout::counter_increment);
1932 __ jmp(*stub->entry());
1933 }
1935 __ bind(done);
1936 } else
1937 if (code == lir_checkcast) {
1938 Register obj = op->object()->as_register();
1939 Register dst = op->result_opr()->as_register();
1940 Label success;
1941 emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
1942 __ bind(success);
1943 if (dst != obj) {
1944 __ mov(dst, obj);
1945 }
1946 } else
1947 if (code == lir_instanceof) {
1948 Register obj = op->object()->as_register();
1949 Register dst = op->result_opr()->as_register();
1950 Label success, failure, done;
1951 emit_typecheck_helper(op, &success, &failure, &failure);
1952 __ bind(failure);
1953 __ xorptr(dst, dst);
1954 __ jmpb(done);
1955 __ bind(success);
1956 __ movptr(dst, 1);
1957 __ bind(done);
1958 } else {
1959 ShouldNotReachHere();
1960 }
1962 }
1965 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
1966 if (LP64_ONLY(false &&) op->code() == lir_cas_long && VM_Version::supports_cx8()) {
1967 assert(op->cmp_value()->as_register_lo() == rax, "wrong register");
1968 assert(op->cmp_value()->as_register_hi() == rdx, "wrong register");
1969 assert(op->new_value()->as_register_lo() == rbx, "wrong register");
1970 assert(op->new_value()->as_register_hi() == rcx, "wrong register");
1971 Register addr = op->addr()->as_register();
1972 if (os::is_MP()) {
1973 __ lock();
1974 }
1975 NOT_LP64(__ cmpxchg8(Address(addr, 0)));
1977 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj ) {
1978 NOT_LP64(assert(op->addr()->is_single_cpu(), "must be single");)
1979 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
1980 Register newval = op->new_value()->as_register();
1981 Register cmpval = op->cmp_value()->as_register();
1982 assert(cmpval == rax, "wrong register");
1983 assert(newval != NULL, "new val must be register");
1984 assert(cmpval != newval, "cmp and new values must be in different registers");
1985 assert(cmpval != addr, "cmp and addr must be in different registers");
1986 assert(newval != addr, "new value and addr must be in different registers");
1988 if ( op->code() == lir_cas_obj) {
1989 #ifdef _LP64
1990 if (UseCompressedOops) {
1991 __ encode_heap_oop(cmpval);
1992 __ mov(rscratch1, newval);
1993 __ encode_heap_oop(rscratch1);
1994 if (os::is_MP()) {
1995 __ lock();
1996 }
1997 // cmpval (rax) is implicitly used by this instruction
1998 __ cmpxchgl(rscratch1, Address(addr, 0));
1999 } else
2000 #endif
2001 {
2002 if (os::is_MP()) {
2003 __ lock();
2004 }
2005 __ cmpxchgptr(newval, Address(addr, 0));
2006 }
2007 } else {
2008 assert(op->code() == lir_cas_int, "lir_cas_int expected");
2009 if (os::is_MP()) {
2010 __ lock();
2011 }
2012 __ cmpxchgl(newval, Address(addr, 0));
2013 }
2014 #ifdef _LP64
2015 } else if (op->code() == lir_cas_long) {
2016 Register addr = (op->addr()->is_single_cpu() ? op->addr()->as_register() : op->addr()->as_register_lo());
2017 Register newval = op->new_value()->as_register_lo();
2018 Register cmpval = op->cmp_value()->as_register_lo();
2019 assert(cmpval == rax, "wrong register");
2020 assert(newval != NULL, "new val must be register");
2021 assert(cmpval != newval, "cmp and new values must be in different registers");
2022 assert(cmpval != addr, "cmp and addr must be in different registers");
2023 assert(newval != addr, "new value and addr must be in different registers");
2024 if (os::is_MP()) {
2025 __ lock();
2026 }
2027 __ cmpxchgq(newval, Address(addr, 0));
2028 #endif // _LP64
2029 } else {
2030 Unimplemented();
2031 }
2032 }
2034 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) {
2035 Assembler::Condition acond, ncond;
2036 switch (condition) {
2037 case lir_cond_equal: acond = Assembler::equal; ncond = Assembler::notEqual; break;
2038 case lir_cond_notEqual: acond = Assembler::notEqual; ncond = Assembler::equal; break;
2039 case lir_cond_less: acond = Assembler::less; ncond = Assembler::greaterEqual; break;
2040 case lir_cond_lessEqual: acond = Assembler::lessEqual; ncond = Assembler::greater; break;
2041 case lir_cond_greaterEqual: acond = Assembler::greaterEqual; ncond = Assembler::less; break;
2042 case lir_cond_greater: acond = Assembler::greater; ncond = Assembler::lessEqual; break;
2043 case lir_cond_belowEqual: acond = Assembler::belowEqual; ncond = Assembler::above; break;
2044 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; ncond = Assembler::below; break;
2045 default: ShouldNotReachHere();
2046 }
2048 if (opr1->is_cpu_register()) {
2049 reg2reg(opr1, result);
2050 } else if (opr1->is_stack()) {
2051 stack2reg(opr1, result, result->type());
2052 } else if (opr1->is_constant()) {
2053 const2reg(opr1, result, lir_patch_none, NULL);
2054 } else {
2055 ShouldNotReachHere();
2056 }
2058 if (VM_Version::supports_cmov() && !opr2->is_constant()) {
2059 // optimized version that does not require a branch
2060 if (opr2->is_single_cpu()) {
2061 assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
2062 __ cmov(ncond, result->as_register(), opr2->as_register());
2063 } else if (opr2->is_double_cpu()) {
2064 assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2065 assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
2066 __ cmovptr(ncond, result->as_register_lo(), opr2->as_register_lo());
2067 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), opr2->as_register_hi());)
2068 } else if (opr2->is_single_stack()) {
2069 __ cmovl(ncond, result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()));
2070 } else if (opr2->is_double_stack()) {
2071 __ cmovptr(ncond, result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix(), lo_word_offset_in_bytes));
2072 NOT_LP64(__ cmovptr(ncond, result->as_register_hi(), frame_map()->address_for_slot(opr2->double_stack_ix(), hi_word_offset_in_bytes));)
2073 } else {
2074 ShouldNotReachHere();
2075 }
2077 } else {
2078 Label skip;
2079 __ jcc (acond, skip);
2080 if (opr2->is_cpu_register()) {
2081 reg2reg(opr2, result);
2082 } else if (opr2->is_stack()) {
2083 stack2reg(opr2, result, result->type());
2084 } else if (opr2->is_constant()) {
2085 const2reg(opr2, result, lir_patch_none, NULL);
2086 } else {
2087 ShouldNotReachHere();
2088 }
2089 __ bind(skip);
2090 }
2091 }
2094 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack) {
2095 assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
2097 if (left->is_single_cpu()) {
2098 assert(left == dest, "left and dest must be equal");
2099 Register lreg = left->as_register();
2101 if (right->is_single_cpu()) {
2102 // cpu register - cpu register
2103 Register rreg = right->as_register();
2104 switch (code) {
2105 case lir_add: __ addl (lreg, rreg); break;
2106 case lir_sub: __ subl (lreg, rreg); break;
2107 case lir_mul: __ imull(lreg, rreg); break;
2108 default: ShouldNotReachHere();
2109 }
2111 } else if (right->is_stack()) {
2112 // cpu register - stack
2113 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2114 switch (code) {
2115 case lir_add: __ addl(lreg, raddr); break;
2116 case lir_sub: __ subl(lreg, raddr); break;
2117 default: ShouldNotReachHere();
2118 }
2120 } else if (right->is_constant()) {
2121 // cpu register - constant
2122 jint c = right->as_constant_ptr()->as_jint();
2123 switch (code) {
2124 case lir_add: {
2125 __ incrementl(lreg, c);
2126 break;
2127 }
2128 case lir_sub: {
2129 __ decrementl(lreg, c);
2130 break;
2131 }
2132 default: ShouldNotReachHere();
2133 }
2135 } else {
2136 ShouldNotReachHere();
2137 }
2139 } else if (left->is_double_cpu()) {
2140 assert(left == dest, "left and dest must be equal");
2141 Register lreg_lo = left->as_register_lo();
2142 Register lreg_hi = left->as_register_hi();
2144 if (right->is_double_cpu()) {
2145 // cpu register - cpu register
2146 Register rreg_lo = right->as_register_lo();
2147 Register rreg_hi = right->as_register_hi();
2148 NOT_LP64(assert_different_registers(lreg_lo, lreg_hi, rreg_lo, rreg_hi));
2149 LP64_ONLY(assert_different_registers(lreg_lo, rreg_lo));
2150 switch (code) {
2151 case lir_add:
2152 __ addptr(lreg_lo, rreg_lo);
2153 NOT_LP64(__ adcl(lreg_hi, rreg_hi));
2154 break;
2155 case lir_sub:
2156 __ subptr(lreg_lo, rreg_lo);
2157 NOT_LP64(__ sbbl(lreg_hi, rreg_hi));
2158 break;
2159 case lir_mul:
2160 #ifdef _LP64
2161 __ imulq(lreg_lo, rreg_lo);
2162 #else
2163 assert(lreg_lo == rax && lreg_hi == rdx, "must be");
2164 __ imull(lreg_hi, rreg_lo);
2165 __ imull(rreg_hi, lreg_lo);
2166 __ addl (rreg_hi, lreg_hi);
2167 __ mull (rreg_lo);
2168 __ addl (lreg_hi, rreg_hi);
2169 #endif // _LP64
2170 break;
2171 default:
2172 ShouldNotReachHere();
2173 }
2175 } else if (right->is_constant()) {
2176 // cpu register - constant
2177 #ifdef _LP64
2178 jlong c = right->as_constant_ptr()->as_jlong_bits();
2179 __ movptr(r10, (intptr_t) c);
2180 switch (code) {
2181 case lir_add:
2182 __ addptr(lreg_lo, r10);
2183 break;
2184 case lir_sub:
2185 __ subptr(lreg_lo, r10);
2186 break;
2187 default:
2188 ShouldNotReachHere();
2189 }
2190 #else
2191 jint c_lo = right->as_constant_ptr()->as_jint_lo();
2192 jint c_hi = right->as_constant_ptr()->as_jint_hi();
2193 switch (code) {
2194 case lir_add:
2195 __ addptr(lreg_lo, c_lo);
2196 __ adcl(lreg_hi, c_hi);
2197 break;
2198 case lir_sub:
2199 __ subptr(lreg_lo, c_lo);
2200 __ sbbl(lreg_hi, c_hi);
2201 break;
2202 default:
2203 ShouldNotReachHere();
2204 }
2205 #endif // _LP64
2207 } else {
2208 ShouldNotReachHere();
2209 }
2211 } else if (left->is_single_xmm()) {
2212 assert(left == dest, "left and dest must be equal");
2213 XMMRegister lreg = left->as_xmm_float_reg();
2215 if (right->is_single_xmm()) {
2216 XMMRegister rreg = right->as_xmm_float_reg();
2217 switch (code) {
2218 case lir_add: __ addss(lreg, rreg); break;
2219 case lir_sub: __ subss(lreg, rreg); break;
2220 case lir_mul_strictfp: // fall through
2221 case lir_mul: __ mulss(lreg, rreg); break;
2222 case lir_div_strictfp: // fall through
2223 case lir_div: __ divss(lreg, rreg); break;
2224 default: ShouldNotReachHere();
2225 }
2226 } else {
2227 Address raddr;
2228 if (right->is_single_stack()) {
2229 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2230 } else if (right->is_constant()) {
2231 // hack for now
2232 raddr = __ as_Address(InternalAddress(float_constant(right->as_jfloat())));
2233 } else {
2234 ShouldNotReachHere();
2235 }
2236 switch (code) {
2237 case lir_add: __ addss(lreg, raddr); break;
2238 case lir_sub: __ subss(lreg, raddr); break;
2239 case lir_mul_strictfp: // fall through
2240 case lir_mul: __ mulss(lreg, raddr); break;
2241 case lir_div_strictfp: // fall through
2242 case lir_div: __ divss(lreg, raddr); break;
2243 default: ShouldNotReachHere();
2244 }
2245 }
2247 } else if (left->is_double_xmm()) {
2248 assert(left == dest, "left and dest must be equal");
2250 XMMRegister lreg = left->as_xmm_double_reg();
2251 if (right->is_double_xmm()) {
2252 XMMRegister rreg = right->as_xmm_double_reg();
2253 switch (code) {
2254 case lir_add: __ addsd(lreg, rreg); break;
2255 case lir_sub: __ subsd(lreg, rreg); break;
2256 case lir_mul_strictfp: // fall through
2257 case lir_mul: __ mulsd(lreg, rreg); break;
2258 case lir_div_strictfp: // fall through
2259 case lir_div: __ divsd(lreg, rreg); break;
2260 default: ShouldNotReachHere();
2261 }
2262 } else {
2263 Address raddr;
2264 if (right->is_double_stack()) {
2265 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2266 } else if (right->is_constant()) {
2267 // hack for now
2268 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2269 } else {
2270 ShouldNotReachHere();
2271 }
2272 switch (code) {
2273 case lir_add: __ addsd(lreg, raddr); break;
2274 case lir_sub: __ subsd(lreg, raddr); break;
2275 case lir_mul_strictfp: // fall through
2276 case lir_mul: __ mulsd(lreg, raddr); break;
2277 case lir_div_strictfp: // fall through
2278 case lir_div: __ divsd(lreg, raddr); break;
2279 default: ShouldNotReachHere();
2280 }
2281 }
2283 } else if (left->is_single_fpu()) {
2284 assert(dest->is_single_fpu(), "fpu stack allocation required");
2286 if (right->is_single_fpu()) {
2287 arith_fpu_implementation(code, left->fpu_regnr(), right->fpu_regnr(), dest->fpu_regnr(), pop_fpu_stack);
2289 } else {
2290 assert(left->fpu_regnr() == 0, "left must be on TOS");
2291 assert(dest->fpu_regnr() == 0, "dest must be on TOS");
2293 Address raddr;
2294 if (right->is_single_stack()) {
2295 raddr = frame_map()->address_for_slot(right->single_stack_ix());
2296 } else if (right->is_constant()) {
2297 address const_addr = float_constant(right->as_jfloat());
2298 assert(const_addr != NULL, "incorrect float/double constant maintainance");
2299 // hack for now
2300 raddr = __ as_Address(InternalAddress(const_addr));
2301 } else {
2302 ShouldNotReachHere();
2303 }
2305 switch (code) {
2306 case lir_add: __ fadd_s(raddr); break;
2307 case lir_sub: __ fsub_s(raddr); break;
2308 case lir_mul_strictfp: // fall through
2309 case lir_mul: __ fmul_s(raddr); break;
2310 case lir_div_strictfp: // fall through
2311 case lir_div: __ fdiv_s(raddr); break;
2312 default: ShouldNotReachHere();
2313 }
2314 }
2316 } else if (left->is_double_fpu()) {
2317 assert(dest->is_double_fpu(), "fpu stack allocation required");
2319 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2320 // Double values require special handling for strictfp mul/div on x86
2321 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
2322 __ fmulp(left->fpu_regnrLo() + 1);
2323 }
2325 if (right->is_double_fpu()) {
2326 arith_fpu_implementation(code, left->fpu_regnrLo(), right->fpu_regnrLo(), dest->fpu_regnrLo(), pop_fpu_stack);
2328 } else {
2329 assert(left->fpu_regnrLo() == 0, "left must be on TOS");
2330 assert(dest->fpu_regnrLo() == 0, "dest must be on TOS");
2332 Address raddr;
2333 if (right->is_double_stack()) {
2334 raddr = frame_map()->address_for_slot(right->double_stack_ix());
2335 } else if (right->is_constant()) {
2336 // hack for now
2337 raddr = __ as_Address(InternalAddress(double_constant(right->as_jdouble())));
2338 } else {
2339 ShouldNotReachHere();
2340 }
2342 switch (code) {
2343 case lir_add: __ fadd_d(raddr); break;
2344 case lir_sub: __ fsub_d(raddr); break;
2345 case lir_mul_strictfp: // fall through
2346 case lir_mul: __ fmul_d(raddr); break;
2347 case lir_div_strictfp: // fall through
2348 case lir_div: __ fdiv_d(raddr); break;
2349 default: ShouldNotReachHere();
2350 }
2351 }
2353 if (code == lir_mul_strictfp || code == lir_div_strictfp) {
2354 // Double values require special handling for strictfp mul/div on x86
2355 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
2356 __ fmulp(dest->fpu_regnrLo() + 1);
2357 }
2359 } else if (left->is_single_stack() || left->is_address()) {
2360 assert(left == dest, "left and dest must be equal");
2362 Address laddr;
2363 if (left->is_single_stack()) {
2364 laddr = frame_map()->address_for_slot(left->single_stack_ix());
2365 } else if (left->is_address()) {
2366 laddr = as_Address(left->as_address_ptr());
2367 } else {
2368 ShouldNotReachHere();
2369 }
2371 if (right->is_single_cpu()) {
2372 Register rreg = right->as_register();
2373 switch (code) {
2374 case lir_add: __ addl(laddr, rreg); break;
2375 case lir_sub: __ subl(laddr, rreg); break;
2376 default: ShouldNotReachHere();
2377 }
2378 } else if (right->is_constant()) {
2379 jint c = right->as_constant_ptr()->as_jint();
2380 switch (code) {
2381 case lir_add: {
2382 __ incrementl(laddr, c);
2383 break;
2384 }
2385 case lir_sub: {
2386 __ decrementl(laddr, c);
2387 break;
2388 }
2389 default: ShouldNotReachHere();
2390 }
2391 } else {
2392 ShouldNotReachHere();
2393 }
2395 } else {
2396 ShouldNotReachHere();
2397 }
2398 }
2400 void LIR_Assembler::arith_fpu_implementation(LIR_Code code, int left_index, int right_index, int dest_index, bool pop_fpu_stack) {
2401 assert(pop_fpu_stack || (left_index == dest_index || right_index == dest_index), "invalid LIR");
2402 assert(!pop_fpu_stack || (left_index - 1 == dest_index || right_index - 1 == dest_index), "invalid LIR");
2403 assert(left_index == 0 || right_index == 0, "either must be on top of stack");
2405 bool left_is_tos = (left_index == 0);
2406 bool dest_is_tos = (dest_index == 0);
2407 int non_tos_index = (left_is_tos ? right_index : left_index);
2409 switch (code) {
2410 case lir_add:
2411 if (pop_fpu_stack) __ faddp(non_tos_index);
2412 else if (dest_is_tos) __ fadd (non_tos_index);
2413 else __ fadda(non_tos_index);
2414 break;
2416 case lir_sub:
2417 if (left_is_tos) {
2418 if (pop_fpu_stack) __ fsubrp(non_tos_index);
2419 else if (dest_is_tos) __ fsub (non_tos_index);
2420 else __ fsubra(non_tos_index);
2421 } else {
2422 if (pop_fpu_stack) __ fsubp (non_tos_index);
2423 else if (dest_is_tos) __ fsubr (non_tos_index);
2424 else __ fsuba (non_tos_index);
2425 }
2426 break;
2428 case lir_mul_strictfp: // fall through
2429 case lir_mul:
2430 if (pop_fpu_stack) __ fmulp(non_tos_index);
2431 else if (dest_is_tos) __ fmul (non_tos_index);
2432 else __ fmula(non_tos_index);
2433 break;
2435 case lir_div_strictfp: // fall through
2436 case lir_div:
2437 if (left_is_tos) {
2438 if (pop_fpu_stack) __ fdivrp(non_tos_index);
2439 else if (dest_is_tos) __ fdiv (non_tos_index);
2440 else __ fdivra(non_tos_index);
2441 } else {
2442 if (pop_fpu_stack) __ fdivp (non_tos_index);
2443 else if (dest_is_tos) __ fdivr (non_tos_index);
2444 else __ fdiva (non_tos_index);
2445 }
2446 break;
2448 case lir_rem:
2449 assert(left_is_tos && dest_is_tos && right_index == 1, "must be guaranteed by FPU stack allocation");
2450 __ fremr(noreg);
2451 break;
2453 default:
2454 ShouldNotReachHere();
2455 }
2456 }
2459 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op) {
2460 if (value->is_double_xmm()) {
2461 switch(code) {
2462 case lir_abs :
2463 {
2464 if (dest->as_xmm_double_reg() != value->as_xmm_double_reg()) {
2465 __ movdbl(dest->as_xmm_double_reg(), value->as_xmm_double_reg());
2466 }
2467 __ andpd(dest->as_xmm_double_reg(),
2468 ExternalAddress((address)double_signmask_pool));
2469 }
2470 break;
2472 case lir_sqrt: __ sqrtsd(dest->as_xmm_double_reg(), value->as_xmm_double_reg()); break;
2473 // all other intrinsics are not available in the SSE instruction set, so FPU is used
2474 default : ShouldNotReachHere();
2475 }
2477 } else if (value->is_double_fpu()) {
2478 assert(value->fpu_regnrLo() == 0 && dest->fpu_regnrLo() == 0, "both must be on TOS");
2479 switch(code) {
2480 case lir_log : __ flog() ; break;
2481 case lir_log10 : __ flog10() ; break;
2482 case lir_abs : __ fabs() ; break;
2483 case lir_sqrt : __ fsqrt(); break;
2484 case lir_sin :
2485 // Should consider not saving rbx, if not necessary
2486 __ trigfunc('s', op->as_Op2()->fpu_stack_size());
2487 break;
2488 case lir_cos :
2489 // Should consider not saving rbx, if not necessary
2490 assert(op->as_Op2()->fpu_stack_size() <= 6, "sin and cos need two free stack slots");
2491 __ trigfunc('c', op->as_Op2()->fpu_stack_size());
2492 break;
2493 case lir_tan :
2494 // Should consider not saving rbx, if not necessary
2495 __ trigfunc('t', op->as_Op2()->fpu_stack_size());
2496 break;
2497 case lir_exp :
2498 __ exp_with_fallback(op->as_Op2()->fpu_stack_size());
2499 break;
2500 case lir_pow :
2501 __ pow_with_fallback(op->as_Op2()->fpu_stack_size());
2502 break;
2503 default : ShouldNotReachHere();
2504 }
2505 } else {
2506 Unimplemented();
2507 }
2508 }
2510 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
2511 // assert(left->destroys_register(), "check");
2512 if (left->is_single_cpu()) {
2513 Register reg = left->as_register();
2514 if (right->is_constant()) {
2515 int val = right->as_constant_ptr()->as_jint();
2516 switch (code) {
2517 case lir_logic_and: __ andl (reg, val); break;
2518 case lir_logic_or: __ orl (reg, val); break;
2519 case lir_logic_xor: __ xorl (reg, val); break;
2520 default: ShouldNotReachHere();
2521 }
2522 } else if (right->is_stack()) {
2523 // added support for stack operands
2524 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
2525 switch (code) {
2526 case lir_logic_and: __ andl (reg, raddr); break;
2527 case lir_logic_or: __ orl (reg, raddr); break;
2528 case lir_logic_xor: __ xorl (reg, raddr); break;
2529 default: ShouldNotReachHere();
2530 }
2531 } else {
2532 Register rright = right->as_register();
2533 switch (code) {
2534 case lir_logic_and: __ andptr (reg, rright); break;
2535 case lir_logic_or : __ orptr (reg, rright); break;
2536 case lir_logic_xor: __ xorptr (reg, rright); break;
2537 default: ShouldNotReachHere();
2538 }
2539 }
2540 move_regs(reg, dst->as_register());
2541 } else {
2542 Register l_lo = left->as_register_lo();
2543 Register l_hi = left->as_register_hi();
2544 if (right->is_constant()) {
2545 #ifdef _LP64
2546 __ mov64(rscratch1, right->as_constant_ptr()->as_jlong());
2547 switch (code) {
2548 case lir_logic_and:
2549 __ andq(l_lo, rscratch1);
2550 break;
2551 case lir_logic_or:
2552 __ orq(l_lo, rscratch1);
2553 break;
2554 case lir_logic_xor:
2555 __ xorq(l_lo, rscratch1);
2556 break;
2557 default: ShouldNotReachHere();
2558 }
2559 #else
2560 int r_lo = right->as_constant_ptr()->as_jint_lo();
2561 int r_hi = right->as_constant_ptr()->as_jint_hi();
2562 switch (code) {
2563 case lir_logic_and:
2564 __ andl(l_lo, r_lo);
2565 __ andl(l_hi, r_hi);
2566 break;
2567 case lir_logic_or:
2568 __ orl(l_lo, r_lo);
2569 __ orl(l_hi, r_hi);
2570 break;
2571 case lir_logic_xor:
2572 __ xorl(l_lo, r_lo);
2573 __ xorl(l_hi, r_hi);
2574 break;
2575 default: ShouldNotReachHere();
2576 }
2577 #endif // _LP64
2578 } else {
2579 #ifdef _LP64
2580 Register r_lo;
2581 if (right->type() == T_OBJECT || right->type() == T_ARRAY) {
2582 r_lo = right->as_register();
2583 } else {
2584 r_lo = right->as_register_lo();
2585 }
2586 #else
2587 Register r_lo = right->as_register_lo();
2588 Register r_hi = right->as_register_hi();
2589 assert(l_lo != r_hi, "overwriting registers");
2590 #endif
2591 switch (code) {
2592 case lir_logic_and:
2593 __ andptr(l_lo, r_lo);
2594 NOT_LP64(__ andptr(l_hi, r_hi);)
2595 break;
2596 case lir_logic_or:
2597 __ orptr(l_lo, r_lo);
2598 NOT_LP64(__ orptr(l_hi, r_hi);)
2599 break;
2600 case lir_logic_xor:
2601 __ xorptr(l_lo, r_lo);
2602 NOT_LP64(__ xorptr(l_hi, r_hi);)
2603 break;
2604 default: ShouldNotReachHere();
2605 }
2606 }
2608 Register dst_lo = dst->as_register_lo();
2609 Register dst_hi = dst->as_register_hi();
2611 #ifdef _LP64
2612 move_regs(l_lo, dst_lo);
2613 #else
2614 if (dst_lo == l_hi) {
2615 assert(dst_hi != l_lo, "overwriting registers");
2616 move_regs(l_hi, dst_hi);
2617 move_regs(l_lo, dst_lo);
2618 } else {
2619 assert(dst_lo != l_hi, "overwriting registers");
2620 move_regs(l_lo, dst_lo);
2621 move_regs(l_hi, dst_hi);
2622 }
2623 #endif // _LP64
2624 }
2625 }
2628 // we assume that rax, and rdx can be overwritten
2629 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
2631 assert(left->is_single_cpu(), "left must be register");
2632 assert(right->is_single_cpu() || right->is_constant(), "right must be register or constant");
2633 assert(result->is_single_cpu(), "result must be register");
2635 // assert(left->destroys_register(), "check");
2636 // assert(right->destroys_register(), "check");
2638 Register lreg = left->as_register();
2639 Register dreg = result->as_register();
2641 if (right->is_constant()) {
2642 int divisor = right->as_constant_ptr()->as_jint();
2643 assert(divisor > 0 && is_power_of_2(divisor), "must be");
2644 if (code == lir_idiv) {
2645 assert(lreg == rax, "must be rax,");
2646 assert(temp->as_register() == rdx, "tmp register must be rdx");
2647 __ cdql(); // sign extend into rdx:rax
2648 if (divisor == 2) {
2649 __ subl(lreg, rdx);
2650 } else {
2651 __ andl(rdx, divisor - 1);
2652 __ addl(lreg, rdx);
2653 }
2654 __ sarl(lreg, log2_intptr(divisor));
2655 move_regs(lreg, dreg);
2656 } else if (code == lir_irem) {
2657 Label done;
2658 __ mov(dreg, lreg);
2659 __ andl(dreg, 0x80000000 | (divisor - 1));
2660 __ jcc(Assembler::positive, done);
2661 __ decrement(dreg);
2662 __ orl(dreg, ~(divisor - 1));
2663 __ increment(dreg);
2664 __ bind(done);
2665 } else {
2666 ShouldNotReachHere();
2667 }
2668 } else {
2669 Register rreg = right->as_register();
2670 assert(lreg == rax, "left register must be rax,");
2671 assert(rreg != rdx, "right register must not be rdx");
2672 assert(temp->as_register() == rdx, "tmp register must be rdx");
2674 move_regs(lreg, rax);
2676 int idivl_offset = __ corrected_idivl(rreg);
2677 add_debug_info_for_div0(idivl_offset, info);
2678 if (code == lir_irem) {
2679 move_regs(rdx, dreg); // result is in rdx
2680 } else {
2681 move_regs(rax, dreg);
2682 }
2683 }
2684 }
2687 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
2688 if (opr1->is_single_cpu()) {
2689 Register reg1 = opr1->as_register();
2690 if (opr2->is_single_cpu()) {
2691 // cpu register - cpu register
2692 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2693 __ cmpptr(reg1, opr2->as_register());
2694 } else {
2695 assert(opr2->type() != T_OBJECT && opr2->type() != T_ARRAY, "cmp int, oop?");
2696 __ cmpl(reg1, opr2->as_register());
2697 }
2698 } else if (opr2->is_stack()) {
2699 // cpu register - stack
2700 if (opr1->type() == T_OBJECT || opr1->type() == T_ARRAY) {
2701 __ cmpptr(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2702 } else {
2703 __ cmpl(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2704 }
2705 } else if (opr2->is_constant()) {
2706 // cpu register - constant
2707 LIR_Const* c = opr2->as_constant_ptr();
2708 if (c->type() == T_INT) {
2709 __ cmpl(reg1, c->as_jint());
2710 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2711 // In 64bit oops are single register
2712 jobject o = c->as_jobject();
2713 if (o == NULL) {
2714 __ cmpptr(reg1, (int32_t)NULL_WORD);
2715 } else {
2716 #ifdef _LP64
2717 __ movoop(rscratch1, o);
2718 __ cmpptr(reg1, rscratch1);
2719 #else
2720 __ cmpoop(reg1, c->as_jobject());
2721 #endif // _LP64
2722 }
2723 } else {
2724 fatal(err_msg("unexpected type: %s", basictype_to_str(c->type())));
2725 }
2726 // cpu register - address
2727 } else if (opr2->is_address()) {
2728 if (op->info() != NULL) {
2729 add_debug_info_for_null_check_here(op->info());
2730 }
2731 __ cmpl(reg1, as_Address(opr2->as_address_ptr()));
2732 } else {
2733 ShouldNotReachHere();
2734 }
2736 } else if(opr1->is_double_cpu()) {
2737 Register xlo = opr1->as_register_lo();
2738 Register xhi = opr1->as_register_hi();
2739 if (opr2->is_double_cpu()) {
2740 #ifdef _LP64
2741 __ cmpptr(xlo, opr2->as_register_lo());
2742 #else
2743 // cpu register - cpu register
2744 Register ylo = opr2->as_register_lo();
2745 Register yhi = opr2->as_register_hi();
2746 __ subl(xlo, ylo);
2747 __ sbbl(xhi, yhi);
2748 if (condition == lir_cond_equal || condition == lir_cond_notEqual) {
2749 __ orl(xhi, xlo);
2750 }
2751 #endif // _LP64
2752 } else if (opr2->is_constant()) {
2753 // cpu register - constant 0
2754 assert(opr2->as_jlong() == (jlong)0, "only handles zero");
2755 #ifdef _LP64
2756 __ cmpptr(xlo, (int32_t)opr2->as_jlong());
2757 #else
2758 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "only handles equals case");
2759 __ orl(xhi, xlo);
2760 #endif // _LP64
2761 } else {
2762 ShouldNotReachHere();
2763 }
2765 } else if (opr1->is_single_xmm()) {
2766 XMMRegister reg1 = opr1->as_xmm_float_reg();
2767 if (opr2->is_single_xmm()) {
2768 // xmm register - xmm register
2769 __ ucomiss(reg1, opr2->as_xmm_float_reg());
2770 } else if (opr2->is_stack()) {
2771 // xmm register - stack
2772 __ ucomiss(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
2773 } else if (opr2->is_constant()) {
2774 // xmm register - constant
2775 __ ucomiss(reg1, InternalAddress(float_constant(opr2->as_jfloat())));
2776 } else if (opr2->is_address()) {
2777 // xmm register - address
2778 if (op->info() != NULL) {
2779 add_debug_info_for_null_check_here(op->info());
2780 }
2781 __ ucomiss(reg1, as_Address(opr2->as_address_ptr()));
2782 } else {
2783 ShouldNotReachHere();
2784 }
2786 } else if (opr1->is_double_xmm()) {
2787 XMMRegister reg1 = opr1->as_xmm_double_reg();
2788 if (opr2->is_double_xmm()) {
2789 // xmm register - xmm register
2790 __ ucomisd(reg1, opr2->as_xmm_double_reg());
2791 } else if (opr2->is_stack()) {
2792 // xmm register - stack
2793 __ ucomisd(reg1, frame_map()->address_for_slot(opr2->double_stack_ix()));
2794 } else if (opr2->is_constant()) {
2795 // xmm register - constant
2796 __ ucomisd(reg1, InternalAddress(double_constant(opr2->as_jdouble())));
2797 } else if (opr2->is_address()) {
2798 // xmm register - address
2799 if (op->info() != NULL) {
2800 add_debug_info_for_null_check_here(op->info());
2801 }
2802 __ ucomisd(reg1, as_Address(opr2->pointer()->as_address()));
2803 } else {
2804 ShouldNotReachHere();
2805 }
2807 } else if(opr1->is_single_fpu() || opr1->is_double_fpu()) {
2808 assert(opr1->is_fpu_register() && opr1->fpu() == 0, "currently left-hand side must be on TOS (relax this restriction)");
2809 assert(opr2->is_fpu_register(), "both must be registers");
2810 __ fcmp(noreg, opr2->fpu(), op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2812 } else if (opr1->is_address() && opr2->is_constant()) {
2813 LIR_Const* c = opr2->as_constant_ptr();
2814 #ifdef _LP64
2815 if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2816 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "need to reverse");
2817 __ movoop(rscratch1, c->as_jobject());
2818 }
2819 #endif // LP64
2820 if (op->info() != NULL) {
2821 add_debug_info_for_null_check_here(op->info());
2822 }
2823 // special case: address - constant
2824 LIR_Address* addr = opr1->as_address_ptr();
2825 if (c->type() == T_INT) {
2826 __ cmpl(as_Address(addr), c->as_jint());
2827 } else if (c->type() == T_OBJECT || c->type() == T_ARRAY) {
2828 #ifdef _LP64
2829 // %%% Make this explode if addr isn't reachable until we figure out a
2830 // better strategy by giving noreg as the temp for as_Address
2831 __ cmpptr(rscratch1, as_Address(addr, noreg));
2832 #else
2833 __ cmpoop(as_Address(addr), c->as_jobject());
2834 #endif // _LP64
2835 } else {
2836 ShouldNotReachHere();
2837 }
2839 } else {
2840 ShouldNotReachHere();
2841 }
2842 }
2844 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
2845 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
2846 if (left->is_single_xmm()) {
2847 assert(right->is_single_xmm(), "must match");
2848 __ cmpss2int(left->as_xmm_float_reg(), right->as_xmm_float_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2849 } else if (left->is_double_xmm()) {
2850 assert(right->is_double_xmm(), "must match");
2851 __ cmpsd2int(left->as_xmm_double_reg(), right->as_xmm_double_reg(), dst->as_register(), code == lir_ucmp_fd2i);
2853 } else {
2854 assert(left->is_single_fpu() || left->is_double_fpu(), "must be");
2855 assert(right->is_single_fpu() || right->is_double_fpu(), "must match");
2857 assert(left->fpu() == 0, "left must be on TOS");
2858 __ fcmp2int(dst->as_register(), code == lir_ucmp_fd2i, right->fpu(),
2859 op->fpu_pop_count() > 0, op->fpu_pop_count() > 1);
2860 }
2861 } else {
2862 assert(code == lir_cmp_l2i, "check");
2863 #ifdef _LP64
2864 Label done;
2865 Register dest = dst->as_register();
2866 __ cmpptr(left->as_register_lo(), right->as_register_lo());
2867 __ movl(dest, -1);
2868 __ jccb(Assembler::less, done);
2869 __ set_byte_if_not_zero(dest);
2870 __ movzbl(dest, dest);
2871 __ bind(done);
2872 #else
2873 __ lcmp2int(left->as_register_hi(),
2874 left->as_register_lo(),
2875 right->as_register_hi(),
2876 right->as_register_lo());
2877 move_regs(left->as_register_hi(), dst->as_register());
2878 #endif // _LP64
2879 }
2880 }
2883 void LIR_Assembler::align_call(LIR_Code code) {
2884 if (os::is_MP()) {
2885 // make sure that the displacement word of the call ends up word aligned
2886 int offset = __ offset();
2887 switch (code) {
2888 case lir_static_call:
2889 case lir_optvirtual_call:
2890 case lir_dynamic_call:
2891 offset += NativeCall::displacement_offset;
2892 break;
2893 case lir_icvirtual_call:
2894 offset += NativeCall::displacement_offset + NativeMovConstReg::instruction_size;
2895 break;
2896 case lir_virtual_call: // currently, sparc-specific for niagara
2897 default: ShouldNotReachHere();
2898 }
2899 while (offset++ % BytesPerWord != 0) {
2900 __ nop();
2901 }
2902 }
2903 }
2906 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
2907 assert(!os::is_MP() || (__ offset() + NativeCall::displacement_offset) % BytesPerWord == 0,
2908 "must be aligned");
2909 __ call(AddressLiteral(op->addr(), rtype));
2910 add_call_info(code_offset(), op->info());
2911 }
2914 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
2915 __ ic_call(op->addr());
2916 add_call_info(code_offset(), op->info());
2917 assert(!os::is_MP() ||
2918 (__ offset() - NativeCall::instruction_size + NativeCall::displacement_offset) % BytesPerWord == 0,
2919 "must be aligned");
2920 }
2923 /* Currently, vtable-dispatch is only enabled for sparc platforms */
2924 void LIR_Assembler::vtable_call(LIR_OpJavaCall* op) {
2925 ShouldNotReachHere();
2926 }
2929 void LIR_Assembler::emit_static_call_stub() {
2930 address call_pc = __ pc();
2931 address stub = __ start_a_stub(call_stub_size);
2932 if (stub == NULL) {
2933 bailout("static call stub overflow");
2934 return;
2935 }
2937 int start = __ offset();
2938 if (os::is_MP()) {
2939 // make sure that the displacement word of the call ends up word aligned
2940 int offset = __ offset() + NativeMovConstReg::instruction_size + NativeCall::displacement_offset;
2941 while (offset++ % BytesPerWord != 0) {
2942 __ nop();
2943 }
2944 }
2945 __ relocate(static_stub_Relocation::spec(call_pc));
2946 __ mov_metadata(rbx, (Metadata*)NULL);
2947 // must be set to -1 at code generation time
2948 assert(!os::is_MP() || ((__ offset() + 1) % BytesPerWord) == 0, "must be aligned on MP");
2949 // On 64bit this will die since it will take a movq & jmp, must be only a jmp
2950 __ jump(RuntimeAddress(__ pc()));
2952 assert(__ offset() - start <= call_stub_size, "stub too big");
2953 __ end_a_stub();
2954 }
2957 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
2958 assert(exceptionOop->as_register() == rax, "must match");
2959 assert(exceptionPC->as_register() == rdx, "must match");
2961 // exception object is not added to oop map by LinearScan
2962 // (LinearScan assumes that no oops are in fixed registers)
2963 info->add_register_oop(exceptionOop);
2964 Runtime1::StubID unwind_id;
2966 // get current pc information
2967 // pc is only needed if the method has an exception handler, the unwind code does not need it.
2968 int pc_for_athrow_offset = __ offset();
2969 InternalAddress pc_for_athrow(__ pc());
2970 __ lea(exceptionPC->as_register(), pc_for_athrow);
2971 add_call_info(pc_for_athrow_offset, info); // for exception handler
2973 __ verify_not_null_oop(rax);
2974 // search an exception handler (rax: exception oop, rdx: throwing pc)
2975 if (compilation()->has_fpu_code()) {
2976 unwind_id = Runtime1::handle_exception_id;
2977 } else {
2978 unwind_id = Runtime1::handle_exception_nofpu_id;
2979 }
2980 __ call(RuntimeAddress(Runtime1::entry_for(unwind_id)));
2982 // enough room for two byte trap
2983 __ nop();
2984 }
2987 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
2988 assert(exceptionOop->as_register() == rax, "must match");
2990 __ jmp(_unwind_handler_entry);
2991 }
2994 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2996 // optimized version for linear scan:
2997 // * count must be already in ECX (guaranteed by LinearScan)
2998 // * left and dest must be equal
2999 // * tmp must be unused
3000 assert(count->as_register() == SHIFT_count, "count must be in ECX");
3001 assert(left == dest, "left and dest must be equal");
3002 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
3004 if (left->is_single_cpu()) {
3005 Register value = left->as_register();
3006 assert(value != SHIFT_count, "left cannot be ECX");
3008 switch (code) {
3009 case lir_shl: __ shll(value); break;
3010 case lir_shr: __ sarl(value); break;
3011 case lir_ushr: __ shrl(value); break;
3012 default: ShouldNotReachHere();
3013 }
3014 } else if (left->is_double_cpu()) {
3015 Register lo = left->as_register_lo();
3016 Register hi = left->as_register_hi();
3017 assert(lo != SHIFT_count && hi != SHIFT_count, "left cannot be ECX");
3018 #ifdef _LP64
3019 switch (code) {
3020 case lir_shl: __ shlptr(lo); break;
3021 case lir_shr: __ sarptr(lo); break;
3022 case lir_ushr: __ shrptr(lo); break;
3023 default: ShouldNotReachHere();
3024 }
3025 #else
3027 switch (code) {
3028 case lir_shl: __ lshl(hi, lo); break;
3029 case lir_shr: __ lshr(hi, lo, true); break;
3030 case lir_ushr: __ lshr(hi, lo, false); break;
3031 default: ShouldNotReachHere();
3032 }
3033 #endif // LP64
3034 } else {
3035 ShouldNotReachHere();
3036 }
3037 }
3040 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
3041 if (dest->is_single_cpu()) {
3042 // first move left into dest so that left is not destroyed by the shift
3043 Register value = dest->as_register();
3044 count = count & 0x1F; // Java spec
3046 move_regs(left->as_register(), value);
3047 switch (code) {
3048 case lir_shl: __ shll(value, count); break;
3049 case lir_shr: __ sarl(value, count); break;
3050 case lir_ushr: __ shrl(value, count); break;
3051 default: ShouldNotReachHere();
3052 }
3053 } else if (dest->is_double_cpu()) {
3054 #ifndef _LP64
3055 Unimplemented();
3056 #else
3057 // first move left into dest so that left is not destroyed by the shift
3058 Register value = dest->as_register_lo();
3059 count = count & 0x1F; // Java spec
3061 move_regs(left->as_register_lo(), value);
3062 switch (code) {
3063 case lir_shl: __ shlptr(value, count); break;
3064 case lir_shr: __ sarptr(value, count); break;
3065 case lir_ushr: __ shrptr(value, count); break;
3066 default: ShouldNotReachHere();
3067 }
3068 #endif // _LP64
3069 } else {
3070 ShouldNotReachHere();
3071 }
3072 }
3075 void LIR_Assembler::store_parameter(Register r, int offset_from_rsp_in_words) {
3076 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3077 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3078 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3079 __ movptr (Address(rsp, offset_from_rsp_in_bytes), r);
3080 }
3083 void LIR_Assembler::store_parameter(jint c, int offset_from_rsp_in_words) {
3084 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3085 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3086 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3087 __ movptr (Address(rsp, offset_from_rsp_in_bytes), c);
3088 }
3091 void LIR_Assembler::store_parameter(jobject o, int offset_from_rsp_in_words) {
3092 assert(offset_from_rsp_in_words >= 0, "invalid offset from rsp");
3093 int offset_from_rsp_in_bytes = offset_from_rsp_in_words * BytesPerWord;
3094 assert(offset_from_rsp_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
3095 __ movoop (Address(rsp, offset_from_rsp_in_bytes), o);
3096 }
3099 // This code replaces a call to arraycopy; no exception may
3100 // be thrown in this code, they must be thrown in the System.arraycopy
3101 // activation frame; we could save some checks if this would not be the case
3102 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
3103 ciArrayKlass* default_type = op->expected_type();
3104 Register src = op->src()->as_register();
3105 Register dst = op->dst()->as_register();
3106 Register src_pos = op->src_pos()->as_register();
3107 Register dst_pos = op->dst_pos()->as_register();
3108 Register length = op->length()->as_register();
3109 Register tmp = op->tmp()->as_register();
3111 CodeStub* stub = op->stub();
3112 int flags = op->flags();
3113 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
3114 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
3116 // if we don't know anything, just go through the generic arraycopy
3117 if (default_type == NULL) {
3118 Label done;
3119 // save outgoing arguments on stack in case call to System.arraycopy is needed
3120 // HACK ALERT. This code used to push the parameters in a hardwired fashion
3121 // for interpreter calling conventions. Now we have to do it in new style conventions.
3122 // For the moment until C1 gets the new register allocator I just force all the
3123 // args to the right place (except the register args) and then on the back side
3124 // reload the register args properly if we go slow path. Yuck
3126 // 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 C_entry = CAST_FROM_FN_PTR(address, Runtime1::arraycopy);
3138 address copyfunc_addr = StubRoutines::generic_arraycopy();
3140 // pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint
3141 #ifdef _LP64
3142 // The arguments are in java calling convention so we can trivially shift them to C
3143 // convention
3144 assert_different_registers(c_rarg0, j_rarg1, j_rarg2, j_rarg3, j_rarg4);
3145 __ mov(c_rarg0, j_rarg0);
3146 assert_different_registers(c_rarg1, j_rarg2, j_rarg3, j_rarg4);
3147 __ mov(c_rarg1, j_rarg1);
3148 assert_different_registers(c_rarg2, j_rarg3, j_rarg4);
3149 __ mov(c_rarg2, j_rarg2);
3150 assert_different_registers(c_rarg3, j_rarg4);
3151 __ mov(c_rarg3, j_rarg3);
3152 #ifdef _WIN64
3153 // Allocate abi space for args but be sure to keep stack aligned
3154 __ subptr(rsp, 6*wordSize);
3155 store_parameter(j_rarg4, 4);
3156 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3157 __ call(RuntimeAddress(C_entry));
3158 } else {
3159 #ifndef PRODUCT
3160 if (PrintC1Statistics) {
3161 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3162 }
3163 #endif
3164 __ call(RuntimeAddress(copyfunc_addr));
3165 }
3166 __ addptr(rsp, 6*wordSize);
3167 #else
3168 __ mov(c_rarg4, j_rarg4);
3169 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3170 __ call(RuntimeAddress(C_entry));
3171 } else {
3172 #ifndef PRODUCT
3173 if (PrintC1Statistics) {
3174 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3175 }
3176 #endif
3177 __ call(RuntimeAddress(copyfunc_addr));
3178 }
3179 #endif // _WIN64
3180 #else
3181 __ push(length);
3182 __ push(dst_pos);
3183 __ push(dst);
3184 __ push(src_pos);
3185 __ push(src);
3187 if (copyfunc_addr == NULL) { // Use C version if stub was not generated
3188 __ call_VM_leaf(C_entry, 5); // removes pushed parameter from the stack
3189 } else {
3190 #ifndef PRODUCT
3191 if (PrintC1Statistics) {
3192 __ incrementl(ExternalAddress((address)&Runtime1::_generic_arraycopystub_cnt));
3193 }
3194 #endif
3195 __ call_VM_leaf(copyfunc_addr, 5); // removes pushed parameter from the stack
3196 }
3198 #endif // _LP64
3200 __ cmpl(rax, 0);
3201 __ jcc(Assembler::equal, *stub->continuation());
3203 if (copyfunc_addr != NULL) {
3204 __ mov(tmp, rax);
3205 __ xorl(tmp, -1);
3206 }
3208 // Reload values from the stack so they are where the stub
3209 // expects them.
3210 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3211 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3212 __ movptr (length, Address(rsp, 2*BytesPerWord));
3213 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3214 __ movptr (src, Address(rsp, 4*BytesPerWord));
3216 if (copyfunc_addr != NULL) {
3217 __ subl(length, tmp);
3218 __ addl(src_pos, tmp);
3219 __ addl(dst_pos, tmp);
3220 }
3221 __ jmp(*stub->entry());
3223 __ bind(*stub->continuation());
3224 return;
3225 }
3227 assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
3229 int elem_size = type2aelembytes(basic_type);
3230 int shift_amount;
3231 Address::ScaleFactor scale;
3233 switch (elem_size) {
3234 case 1 :
3235 shift_amount = 0;
3236 scale = Address::times_1;
3237 break;
3238 case 2 :
3239 shift_amount = 1;
3240 scale = Address::times_2;
3241 break;
3242 case 4 :
3243 shift_amount = 2;
3244 scale = Address::times_4;
3245 break;
3246 case 8 :
3247 shift_amount = 3;
3248 scale = Address::times_8;
3249 break;
3250 default:
3251 ShouldNotReachHere();
3252 }
3254 Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
3255 Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
3256 Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
3257 Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
3259 // length and pos's are all sign extended at this point on 64bit
3261 // test for NULL
3262 if (flags & LIR_OpArrayCopy::src_null_check) {
3263 __ testptr(src, src);
3264 __ jcc(Assembler::zero, *stub->entry());
3265 }
3266 if (flags & LIR_OpArrayCopy::dst_null_check) {
3267 __ testptr(dst, dst);
3268 __ jcc(Assembler::zero, *stub->entry());
3269 }
3271 // check if negative
3272 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
3273 __ testl(src_pos, src_pos);
3274 __ jcc(Assembler::less, *stub->entry());
3275 }
3276 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
3277 __ testl(dst_pos, dst_pos);
3278 __ jcc(Assembler::less, *stub->entry());
3279 }
3281 if (flags & LIR_OpArrayCopy::src_range_check) {
3282 __ lea(tmp, Address(src_pos, length, Address::times_1, 0));
3283 __ cmpl(tmp, src_length_addr);
3284 __ jcc(Assembler::above, *stub->entry());
3285 }
3286 if (flags & LIR_OpArrayCopy::dst_range_check) {
3287 __ lea(tmp, Address(dst_pos, length, Address::times_1, 0));
3288 __ cmpl(tmp, dst_length_addr);
3289 __ jcc(Assembler::above, *stub->entry());
3290 }
3292 if (flags & LIR_OpArrayCopy::length_positive_check) {
3293 __ testl(length, length);
3294 __ jcc(Assembler::less, *stub->entry());
3295 __ jcc(Assembler::zero, *stub->continuation());
3296 }
3298 #ifdef _LP64
3299 __ movl2ptr(src_pos, src_pos); //higher 32bits must be null
3300 __ movl2ptr(dst_pos, dst_pos); //higher 32bits must be null
3301 #endif
3303 if (flags & LIR_OpArrayCopy::type_check) {
3304 // We don't know the array types are compatible
3305 if (basic_type != T_OBJECT) {
3306 // Simple test for basic type arrays
3307 if (UseCompressedClassPointers) {
3308 __ movl(tmp, src_klass_addr);
3309 __ cmpl(tmp, dst_klass_addr);
3310 } else {
3311 __ movptr(tmp, src_klass_addr);
3312 __ cmpptr(tmp, dst_klass_addr);
3313 }
3314 __ jcc(Assembler::notEqual, *stub->entry());
3315 } else {
3316 // For object arrays, if src is a sub class of dst then we can
3317 // safely do the copy.
3318 Label cont, slow;
3320 __ push(src);
3321 __ push(dst);
3323 __ load_klass(src, src);
3324 __ load_klass(dst, dst);
3326 __ check_klass_subtype_fast_path(src, dst, tmp, &cont, &slow, NULL);
3328 __ push(src);
3329 __ push(dst);
3330 __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::slow_subtype_check_id)));
3331 __ pop(dst);
3332 __ pop(src);
3334 __ cmpl(src, 0);
3335 __ jcc(Assembler::notEqual, cont);
3337 __ bind(slow);
3338 __ pop(dst);
3339 __ pop(src);
3341 address copyfunc_addr = StubRoutines::checkcast_arraycopy();
3342 if (copyfunc_addr != NULL) { // use stub if available
3343 // src is not a sub class of dst so we have to do a
3344 // per-element check.
3346 int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
3347 if ((flags & mask) != mask) {
3348 // Check that at least both of them object arrays.
3349 assert(flags & mask, "one of the two should be known to be an object array");
3351 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
3352 __ load_klass(tmp, src);
3353 } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
3354 __ load_klass(tmp, dst);
3355 }
3356 int lh_offset = in_bytes(Klass::layout_helper_offset());
3357 Address klass_lh_addr(tmp, lh_offset);
3358 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
3359 __ cmpl(klass_lh_addr, objArray_lh);
3360 __ jcc(Assembler::notEqual, *stub->entry());
3361 }
3363 // Spill because stubs can use any register they like and it's
3364 // easier to restore just those that we care about.
3365 store_parameter(dst, 0);
3366 store_parameter(dst_pos, 1);
3367 store_parameter(length, 2);
3368 store_parameter(src_pos, 3);
3369 store_parameter(src, 4);
3371 #ifndef _LP64
3372 __ movptr(tmp, dst_klass_addr);
3373 __ movptr(tmp, Address(tmp, ObjArrayKlass::element_klass_offset()));
3374 __ push(tmp);
3375 __ movl(tmp, Address(tmp, Klass::super_check_offset_offset()));
3376 __ push(tmp);
3377 __ push(length);
3378 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3379 __ push(tmp);
3380 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3381 __ push(tmp);
3383 __ call_VM_leaf(copyfunc_addr, 5);
3384 #else
3385 __ movl2ptr(length, length); //higher 32bits must be null
3387 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3388 assert_different_registers(c_rarg0, dst, dst_pos, length);
3389 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3390 assert_different_registers(c_rarg1, dst, length);
3392 __ mov(c_rarg2, length);
3393 assert_different_registers(c_rarg2, dst);
3395 #ifdef _WIN64
3396 // Allocate abi space for args but be sure to keep stack aligned
3397 __ subptr(rsp, 6*wordSize);
3398 __ load_klass(c_rarg3, dst);
3399 __ movptr(c_rarg3, Address(c_rarg3, ObjArrayKlass::element_klass_offset()));
3400 store_parameter(c_rarg3, 4);
3401 __ movl(c_rarg3, Address(c_rarg3, Klass::super_check_offset_offset()));
3402 __ call(RuntimeAddress(copyfunc_addr));
3403 __ addptr(rsp, 6*wordSize);
3404 #else
3405 __ load_klass(c_rarg4, dst);
3406 __ movptr(c_rarg4, Address(c_rarg4, ObjArrayKlass::element_klass_offset()));
3407 __ movl(c_rarg3, Address(c_rarg4, Klass::super_check_offset_offset()));
3408 __ call(RuntimeAddress(copyfunc_addr));
3409 #endif
3411 #endif
3413 #ifndef PRODUCT
3414 if (PrintC1Statistics) {
3415 Label failed;
3416 __ testl(rax, rax);
3417 __ jcc(Assembler::notZero, failed);
3418 __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_cnt));
3419 __ bind(failed);
3420 }
3421 #endif
3423 __ testl(rax, rax);
3424 __ jcc(Assembler::zero, *stub->continuation());
3426 #ifndef PRODUCT
3427 if (PrintC1Statistics) {
3428 __ incrementl(ExternalAddress((address)&Runtime1::_arraycopy_checkcast_attempt_cnt));
3429 }
3430 #endif
3432 __ mov(tmp, rax);
3434 __ xorl(tmp, -1);
3436 // Restore previously spilled arguments
3437 __ movptr (dst, Address(rsp, 0*BytesPerWord));
3438 __ movptr (dst_pos, Address(rsp, 1*BytesPerWord));
3439 __ movptr (length, Address(rsp, 2*BytesPerWord));
3440 __ movptr (src_pos, Address(rsp, 3*BytesPerWord));
3441 __ movptr (src, Address(rsp, 4*BytesPerWord));
3444 __ subl(length, tmp);
3445 __ addl(src_pos, tmp);
3446 __ addl(dst_pos, tmp);
3447 }
3449 __ jmp(*stub->entry());
3451 __ bind(cont);
3452 __ pop(dst);
3453 __ pop(src);
3454 }
3455 }
3457 #ifdef ASSERT
3458 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
3459 // Sanity check the known type with the incoming class. For the
3460 // primitive case the types must match exactly with src.klass and
3461 // dst.klass each exactly matching the default type. For the
3462 // object array case, if no type check is needed then either the
3463 // dst type is exactly the expected type and the src type is a
3464 // subtype which we can't check or src is the same array as dst
3465 // but not necessarily exactly of type default_type.
3466 Label known_ok, halt;
3467 __ mov_metadata(tmp, default_type->constant_encoding());
3468 #ifdef _LP64
3469 if (UseCompressedClassPointers) {
3470 __ encode_klass_not_null(tmp);
3471 }
3472 #endif
3474 if (basic_type != T_OBJECT) {
3476 if (UseCompressedClassPointers) __ cmpl(tmp, dst_klass_addr);
3477 else __ cmpptr(tmp, dst_klass_addr);
3478 __ jcc(Assembler::notEqual, halt);
3479 if (UseCompressedClassPointers) __ cmpl(tmp, src_klass_addr);
3480 else __ cmpptr(tmp, src_klass_addr);
3481 __ jcc(Assembler::equal, known_ok);
3482 } else {
3483 if (UseCompressedClassPointers) __ cmpl(tmp, dst_klass_addr);
3484 else __ cmpptr(tmp, dst_klass_addr);
3485 __ jcc(Assembler::equal, known_ok);
3486 __ cmpptr(src, dst);
3487 __ jcc(Assembler::equal, known_ok);
3488 }
3489 __ bind(halt);
3490 __ stop("incorrect type information in arraycopy");
3491 __ bind(known_ok);
3492 }
3493 #endif
3495 #ifndef PRODUCT
3496 if (PrintC1Statistics) {
3497 __ incrementl(ExternalAddress(Runtime1::arraycopy_count_address(basic_type)));
3498 }
3499 #endif
3501 #ifdef _LP64
3502 assert_different_registers(c_rarg0, dst, dst_pos, length);
3503 __ lea(c_rarg0, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3504 assert_different_registers(c_rarg1, length);
3505 __ lea(c_rarg1, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3506 __ mov(c_rarg2, length);
3508 #else
3509 __ lea(tmp, Address(src, src_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3510 store_parameter(tmp, 0);
3511 __ lea(tmp, Address(dst, dst_pos, scale, arrayOopDesc::base_offset_in_bytes(basic_type)));
3512 store_parameter(tmp, 1);
3513 store_parameter(length, 2);
3514 #endif // _LP64
3516 bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
3517 bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
3518 const char *name;
3519 address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
3520 __ call_VM_leaf(entry, 0);
3522 __ bind(*stub->continuation());
3523 }
3525 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3526 assert(op->crc()->is_single_cpu(), "crc must be register");
3527 assert(op->val()->is_single_cpu(), "byte value must be register");
3528 assert(op->result_opr()->is_single_cpu(), "result must be register");
3529 Register crc = op->crc()->as_register();
3530 Register val = op->val()->as_register();
3531 Register res = op->result_opr()->as_register();
3533 assert_different_registers(val, crc, res);
3535 __ lea(res, ExternalAddress(StubRoutines::crc_table_addr()));
3536 __ notl(crc); // ~crc
3537 __ update_byte_crc32(crc, val, res);
3538 __ notl(crc); // ~crc
3539 __ mov(res, crc);
3540 }
3542 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
3543 Register obj = op->obj_opr()->as_register(); // may not be an oop
3544 Register hdr = op->hdr_opr()->as_register();
3545 Register lock = op->lock_opr()->as_register();
3546 if (!UseFastLocking) {
3547 __ jmp(*op->stub()->entry());
3548 } else if (op->code() == lir_lock) {
3549 Register scratch = noreg;
3550 if (UseBiasedLocking) {
3551 scratch = op->scratch_opr()->as_register();
3552 }
3553 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3554 // add debug info for NullPointerException only if one is possible
3555 int null_check_offset = __ lock_object(hdr, obj, lock, scratch, *op->stub()->entry());
3556 if (op->info() != NULL) {
3557 add_debug_info_for_null_check(null_check_offset, op->info());
3558 }
3559 // done
3560 } else if (op->code() == lir_unlock) {
3561 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
3562 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
3563 } else {
3564 Unimplemented();
3565 }
3566 __ bind(*op->stub()->continuation());
3567 }
3570 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
3571 ciMethod* method = op->profiled_method();
3572 int bci = op->profiled_bci();
3573 ciMethod* callee = op->profiled_callee();
3575 // Update counter for all call types
3576 ciMethodData* md = method->method_data_or_null();
3577 assert(md != NULL, "Sanity");
3578 ciProfileData* data = md->bci_to_data(bci);
3579 assert(data->is_CounterData(), "need CounterData for calls");
3580 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
3581 Register mdo = op->mdo()->as_register();
3582 __ mov_metadata(mdo, md->constant_encoding());
3583 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
3584 Bytecodes::Code bc = method->java_code_at_bci(bci);
3585 const bool callee_is_static = callee->is_loaded() && callee->is_static();
3586 // Perform additional virtual call profiling for invokevirtual and
3587 // invokeinterface bytecodes
3588 if ((bc == Bytecodes::_invokevirtual || bc == Bytecodes::_invokeinterface) &&
3589 !callee_is_static && // required for optimized MH invokes
3590 C1ProfileVirtualCalls) {
3591 assert(op->recv()->is_single_cpu(), "recv must be allocated");
3592 Register recv = op->recv()->as_register();
3593 assert_different_registers(mdo, recv);
3594 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
3595 ciKlass* known_klass = op->known_holder();
3596 if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
3597 // We know the type that will be seen at this call site; we can
3598 // statically update the MethodData* rather than needing to do
3599 // dynamic tests on the receiver type
3601 // NOTE: we should probably put a lock around this search to
3602 // avoid collisions by concurrent compilations
3603 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
3604 uint i;
3605 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3606 ciKlass* receiver = vc_data->receiver(i);
3607 if (known_klass->equals(receiver)) {
3608 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3609 __ addptr(data_addr, DataLayout::counter_increment);
3610 return;
3611 }
3612 }
3614 // Receiver type not found in profile data; select an empty slot
3616 // Note that this is less efficient than it should be because it
3617 // always does a write to the receiver part of the
3618 // VirtualCallData rather than just the first time
3619 for (i = 0; i < VirtualCallData::row_limit(); i++) {
3620 ciKlass* receiver = vc_data->receiver(i);
3621 if (receiver == NULL) {
3622 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
3623 __ mov_metadata(recv_addr, known_klass->constant_encoding());
3624 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
3625 __ addptr(data_addr, DataLayout::counter_increment);
3626 return;
3627 }
3628 }
3629 } else {
3630 __ load_klass(recv, recv);
3631 Label update_done;
3632 type_profile_helper(mdo, md, data, recv, &update_done);
3633 // Receiver did not match any saved receiver and there is no empty row for it.
3634 // Increment total counter to indicate polymorphic case.
3635 __ addptr(counter_addr, DataLayout::counter_increment);
3637 __ bind(update_done);
3638 }
3639 } else {
3640 // Static call
3641 __ addptr(counter_addr, DataLayout::counter_increment);
3642 }
3643 }
3645 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
3646 Register obj = op->obj()->as_register();
3647 Register tmp = op->tmp()->as_pointer_register();
3648 Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
3649 ciKlass* exact_klass = op->exact_klass();
3650 intptr_t current_klass = op->current_klass();
3651 bool not_null = op->not_null();
3652 bool no_conflict = op->no_conflict();
3654 Label update, next, none;
3656 bool do_null = !not_null;
3657 bool exact_klass_set = exact_klass != NULL && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
3658 bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
3660 assert(do_null || do_update, "why are we here?");
3661 assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
3663 __ verify_oop(obj);
3665 if (tmp != obj) {
3666 __ mov(tmp, obj);
3667 }
3668 if (do_null) {
3669 __ testptr(tmp, tmp);
3670 __ jccb(Assembler::notZero, update);
3671 if (!TypeEntries::was_null_seen(current_klass)) {
3672 __ orptr(mdo_addr, TypeEntries::null_seen);
3673 }
3674 if (do_update) {
3675 #ifndef ASSERT
3676 __ jmpb(next);
3677 }
3678 #else
3679 __ jmp(next);
3680 }
3681 } else {
3682 __ testptr(tmp, tmp);
3683 __ jccb(Assembler::notZero, update);
3684 __ stop("unexpect null obj");
3685 #endif
3686 }
3688 __ bind(update);
3690 if (do_update) {
3691 #ifdef ASSERT
3692 if (exact_klass != NULL) {
3693 Label ok;
3694 __ load_klass(tmp, tmp);
3695 __ push(tmp);
3696 __ mov_metadata(tmp, exact_klass->constant_encoding());
3697 __ cmpptr(tmp, Address(rsp, 0));
3698 __ jccb(Assembler::equal, ok);
3699 __ stop("exact klass and actual klass differ");
3700 __ bind(ok);
3701 __ pop(tmp);
3702 }
3703 #endif
3704 if (!no_conflict) {
3705 if (exact_klass == NULL || TypeEntries::is_type_none(current_klass)) {
3706 if (exact_klass != NULL) {
3707 __ mov_metadata(tmp, exact_klass->constant_encoding());
3708 } else {
3709 __ load_klass(tmp, tmp);
3710 }
3712 __ xorptr(tmp, mdo_addr);
3713 __ testptr(tmp, TypeEntries::type_klass_mask);
3714 // klass seen before, nothing to do. The unknown bit may have been
3715 // set already but no need to check.
3716 __ jccb(Assembler::zero, next);
3718 __ testptr(tmp, TypeEntries::type_unknown);
3719 __ jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
3721 if (TypeEntries::is_type_none(current_klass)) {
3722 __ cmpptr(mdo_addr, 0);
3723 __ jccb(Assembler::equal, none);
3724 __ cmpptr(mdo_addr, TypeEntries::null_seen);
3725 __ jccb(Assembler::equal, none);
3726 // There is a chance that the checks above (re-reading profiling
3727 // data from memory) fail if another thread has just set the
3728 // profiling to this obj's klass
3729 __ xorptr(tmp, mdo_addr);
3730 __ testptr(tmp, TypeEntries::type_klass_mask);
3731 __ jccb(Assembler::zero, next);
3732 }
3733 } else {
3734 assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3735 ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
3737 __ movptr(tmp, mdo_addr);
3738 __ testptr(tmp, TypeEntries::type_unknown);
3739 __ jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
3740 }
3742 // different than before. Cannot keep accurate profile.
3743 __ orptr(mdo_addr, TypeEntries::type_unknown);
3745 if (TypeEntries::is_type_none(current_klass)) {
3746 __ jmpb(next);
3748 __ bind(none);
3749 // first time here. Set profile type.
3750 __ movptr(mdo_addr, tmp);
3751 }
3752 } else {
3753 // There's a single possible klass at this profile point
3754 assert(exact_klass != NULL, "should be");
3755 if (TypeEntries::is_type_none(current_klass)) {
3756 __ mov_metadata(tmp, exact_klass->constant_encoding());
3757 __ xorptr(tmp, mdo_addr);
3758 __ testptr(tmp, TypeEntries::type_klass_mask);
3759 #ifdef ASSERT
3760 __ jcc(Assembler::zero, next);
3762 {
3763 Label ok;
3764 __ push(tmp);
3765 __ cmpptr(mdo_addr, 0);
3766 __ jcc(Assembler::equal, ok);
3767 __ cmpptr(mdo_addr, TypeEntries::null_seen);
3768 __ jcc(Assembler::equal, ok);
3769 // may have been set by another thread
3770 __ mov_metadata(tmp, exact_klass->constant_encoding());
3771 __ xorptr(tmp, mdo_addr);
3772 __ testptr(tmp, TypeEntries::type_mask);
3773 __ jcc(Assembler::zero, ok);
3775 __ stop("unexpected profiling mismatch");
3776 __ bind(ok);
3777 __ pop(tmp);
3778 }
3779 #else
3780 __ jccb(Assembler::zero, next);
3781 #endif
3782 // first time here. Set profile type.
3783 __ movptr(mdo_addr, tmp);
3784 } else {
3785 assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3786 ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3788 __ movptr(tmp, mdo_addr);
3789 __ testptr(tmp, TypeEntries::type_unknown);
3790 __ jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
3792 __ orptr(mdo_addr, TypeEntries::type_unknown);
3793 }
3794 }
3796 __ bind(next);
3797 }
3798 }
3800 void LIR_Assembler::emit_delay(LIR_OpDelay*) {
3801 Unimplemented();
3802 }
3805 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst) {
3806 __ lea(dst->as_register(), frame_map()->address_for_monitor_lock(monitor_no));
3807 }
3810 void LIR_Assembler::align_backward_branch_target() {
3811 __ align(BytesPerWord);
3812 }
3815 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest) {
3816 if (left->is_single_cpu()) {
3817 __ negl(left->as_register());
3818 move_regs(left->as_register(), dest->as_register());
3820 } else if (left->is_double_cpu()) {
3821 Register lo = left->as_register_lo();
3822 #ifdef _LP64
3823 Register dst = dest->as_register_lo();
3824 __ movptr(dst, lo);
3825 __ negptr(dst);
3826 #else
3827 Register hi = left->as_register_hi();
3828 __ lneg(hi, lo);
3829 if (dest->as_register_lo() == hi) {
3830 assert(dest->as_register_hi() != lo, "destroying register");
3831 move_regs(hi, dest->as_register_hi());
3832 move_regs(lo, dest->as_register_lo());
3833 } else {
3834 move_regs(lo, dest->as_register_lo());
3835 move_regs(hi, dest->as_register_hi());
3836 }
3837 #endif // _LP64
3839 } else if (dest->is_single_xmm()) {
3840 if (left->as_xmm_float_reg() != dest->as_xmm_float_reg()) {
3841 __ movflt(dest->as_xmm_float_reg(), left->as_xmm_float_reg());
3842 }
3843 __ xorps(dest->as_xmm_float_reg(),
3844 ExternalAddress((address)float_signflip_pool));
3846 } else if (dest->is_double_xmm()) {
3847 if (left->as_xmm_double_reg() != dest->as_xmm_double_reg()) {
3848 __ movdbl(dest->as_xmm_double_reg(), left->as_xmm_double_reg());
3849 }
3850 __ xorpd(dest->as_xmm_double_reg(),
3851 ExternalAddress((address)double_signflip_pool));
3853 } else if (left->is_single_fpu() || left->is_double_fpu()) {
3854 assert(left->fpu() == 0, "arg must be on TOS");
3855 assert(dest->fpu() == 0, "dest must be TOS");
3856 __ fchs();
3858 } else {
3859 ShouldNotReachHere();
3860 }
3861 }
3864 void LIR_Assembler::leal(LIR_Opr addr, LIR_Opr dest) {
3865 assert(addr->is_address() && dest->is_register(), "check");
3866 Register reg;
3867 reg = dest->as_pointer_register();
3868 __ lea(reg, as_Address(addr->as_address_ptr()));
3869 }
3873 void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
3874 assert(!tmp->is_valid(), "don't need temporary");
3875 __ call(RuntimeAddress(dest));
3876 if (info != NULL) {
3877 add_call_info_here(info);
3878 }
3879 }
3882 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
3883 assert(type == T_LONG, "only for volatile long fields");
3885 if (info != NULL) {
3886 add_debug_info_for_null_check_here(info);
3887 }
3889 if (src->is_double_xmm()) {
3890 if (dest->is_double_cpu()) {
3891 #ifdef _LP64
3892 __ movdq(dest->as_register_lo(), src->as_xmm_double_reg());
3893 #else
3894 __ movdl(dest->as_register_lo(), src->as_xmm_double_reg());
3895 __ psrlq(src->as_xmm_double_reg(), 32);
3896 __ movdl(dest->as_register_hi(), src->as_xmm_double_reg());
3897 #endif // _LP64
3898 } else if (dest->is_double_stack()) {
3899 __ movdbl(frame_map()->address_for_slot(dest->double_stack_ix()), src->as_xmm_double_reg());
3900 } else if (dest->is_address()) {
3901 __ movdbl(as_Address(dest->as_address_ptr()), src->as_xmm_double_reg());
3902 } else {
3903 ShouldNotReachHere();
3904 }
3906 } else if (dest->is_double_xmm()) {
3907 if (src->is_double_stack()) {
3908 __ movdbl(dest->as_xmm_double_reg(), frame_map()->address_for_slot(src->double_stack_ix()));
3909 } else if (src->is_address()) {
3910 __ movdbl(dest->as_xmm_double_reg(), as_Address(src->as_address_ptr()));
3911 } else {
3912 ShouldNotReachHere();
3913 }
3915 } else if (src->is_double_fpu()) {
3916 assert(src->fpu_regnrLo() == 0, "must be TOS");
3917 if (dest->is_double_stack()) {
3918 __ fistp_d(frame_map()->address_for_slot(dest->double_stack_ix()));
3919 } else if (dest->is_address()) {
3920 __ fistp_d(as_Address(dest->as_address_ptr()));
3921 } else {
3922 ShouldNotReachHere();
3923 }
3925 } else if (dest->is_double_fpu()) {
3926 assert(dest->fpu_regnrLo() == 0, "must be TOS");
3927 if (src->is_double_stack()) {
3928 __ fild_d(frame_map()->address_for_slot(src->double_stack_ix()));
3929 } else if (src->is_address()) {
3930 __ fild_d(as_Address(src->as_address_ptr()));
3931 } else {
3932 ShouldNotReachHere();
3933 }
3934 } else {
3935 ShouldNotReachHere();
3936 }
3937 }
3939 #ifdef ASSERT
3940 // emit run-time assertion
3941 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
3942 assert(op->code() == lir_assert, "must be");
3944 if (op->in_opr1()->is_valid()) {
3945 assert(op->in_opr2()->is_valid(), "both operands must be valid");
3946 comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op);
3947 } else {
3948 assert(op->in_opr2()->is_illegal(), "both operands must be illegal");
3949 assert(op->condition() == lir_cond_always, "no other conditions allowed");
3950 }
3952 Label ok;
3953 if (op->condition() != lir_cond_always) {
3954 Assembler::Condition acond = Assembler::zero;
3955 switch (op->condition()) {
3956 case lir_cond_equal: acond = Assembler::equal; break;
3957 case lir_cond_notEqual: acond = Assembler::notEqual; break;
3958 case lir_cond_less: acond = Assembler::less; break;
3959 case lir_cond_lessEqual: acond = Assembler::lessEqual; break;
3960 case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break;
3961 case lir_cond_greater: acond = Assembler::greater; break;
3962 case lir_cond_belowEqual: acond = Assembler::belowEqual; break;
3963 case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break;
3964 default: ShouldNotReachHere();
3965 }
3966 __ jcc(acond, ok);
3967 }
3968 if (op->halt()) {
3969 const char* str = __ code_string(op->msg());
3970 __ stop(str);
3971 } else {
3972 breakpoint();
3973 }
3974 __ bind(ok);
3975 }
3976 #endif
3978 void LIR_Assembler::membar() {
3979 // QQQ sparc TSO uses this,
3980 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad));
3981 }
3983 void LIR_Assembler::membar_acquire() {
3984 // No x86 machines currently require load fences
3985 // __ load_fence();
3986 }
3988 void LIR_Assembler::membar_release() {
3989 // No x86 machines currently require store fences
3990 // __ store_fence();
3991 }
3993 void LIR_Assembler::membar_loadload() {
3994 // no-op
3995 //__ membar(Assembler::Membar_mask_bits(Assembler::loadload));
3996 }
3998 void LIR_Assembler::membar_storestore() {
3999 // no-op
4000 //__ membar(Assembler::Membar_mask_bits(Assembler::storestore));
4001 }
4003 void LIR_Assembler::membar_loadstore() {
4004 // no-op
4005 //__ membar(Assembler::Membar_mask_bits(Assembler::loadstore));
4006 }
4008 void LIR_Assembler::membar_storeload() {
4009 __ membar(Assembler::Membar_mask_bits(Assembler::StoreLoad));
4010 }
4012 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
4013 assert(result_reg->is_register(), "check");
4014 #ifdef _LP64
4015 // __ get_thread(result_reg->as_register_lo());
4016 __ mov(result_reg->as_register(), r15_thread);
4017 #else
4018 __ get_thread(result_reg->as_register());
4019 #endif // _LP64
4020 }
4023 void LIR_Assembler::peephole(LIR_List*) {
4024 // do nothing for now
4025 }
4027 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
4028 assert(data == dest, "xchg/xadd uses only 2 operands");
4030 if (data->type() == T_INT) {
4031 if (code == lir_xadd) {
4032 if (os::is_MP()) {
4033 __ lock();
4034 }
4035 __ xaddl(as_Address(src->as_address_ptr()), data->as_register());
4036 } else {
4037 __ xchgl(data->as_register(), as_Address(src->as_address_ptr()));
4038 }
4039 } else if (data->is_oop()) {
4040 assert (code == lir_xchg, "xadd for oops");
4041 Register obj = data->as_register();
4042 #ifdef _LP64
4043 if (UseCompressedOops) {
4044 __ encode_heap_oop(obj);
4045 __ xchgl(obj, as_Address(src->as_address_ptr()));
4046 __ decode_heap_oop(obj);
4047 } else {
4048 __ xchgptr(obj, as_Address(src->as_address_ptr()));
4049 }
4050 #else
4051 __ xchgl(obj, as_Address(src->as_address_ptr()));
4052 #endif
4053 } else if (data->type() == T_LONG) {
4054 #ifdef _LP64
4055 assert(data->as_register_lo() == data->as_register_hi(), "should be a single register");
4056 if (code == lir_xadd) {
4057 if (os::is_MP()) {
4058 __ lock();
4059 }
4060 __ xaddq(as_Address(src->as_address_ptr()), data->as_register_lo());
4061 } else {
4062 __ xchgq(data->as_register_lo(), as_Address(src->as_address_ptr()));
4063 }
4064 #else
4065 ShouldNotReachHere();
4066 #endif
4067 } else {
4068 ShouldNotReachHere();
4069 }
4070 }
4072 #undef __