1 /*
2 * Copyright (c) 2016, 2021, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2016, 2019 SAP SE. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.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/shared/collectedHeap.hpp"
36 #include "memory/universe.hpp"
37 #include "nativeInst_s390.hpp"
38 #include "oops/objArrayKlass.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/safepointMechanism.inline.hpp"
41 #include "runtime/sharedRuntime.hpp"
42 #include "runtime/stubRoutines.hpp"
43 #include "utilities/powerOfTwo.hpp"
44 #include "vmreg_s390.inline.hpp"
45
46 #define __ _masm->
47
48 #ifndef PRODUCT
49 #undef __
50 #define __ (Verbose ? (_masm->block_comment(FILE_AND_LINE),_masm) : _masm)->
51 #endif
52
53 //------------------------------------------------------------
54
55 bool LIR_Assembler::is_small_constant(LIR_Opr opr) {
56 // Not used on ZARCH_64
57 ShouldNotCallThis();
58 return false;
59 }
60
61 LIR_Opr LIR_Assembler::receiverOpr() {
62 return FrameMap::Z_R2_oop_opr;
63 }
64
65 LIR_Opr LIR_Assembler::osrBufferPointer() {
66 return FrameMap::Z_R2_opr;
67 }
68
69 int LIR_Assembler::initial_frame_size_in_bytes() const {
70 return in_bytes(frame_map()->framesize_in_bytes());
71 }
72
73 // Inline cache check: done before the frame is built.
74 // The inline cached class is in Z_inline_cache(Z_R9).
75 // We fetch the class of the receiver and compare it with the cached class.
76 // If they do not match we jump to the slow case.
77 int LIR_Assembler::check_icache() {
78 Register receiver = receiverOpr()->as_register();
79 int offset = __ offset();
80 __ inline_cache_check(receiver, Z_inline_cache);
81 return offset;
82 }
83
84 void LIR_Assembler::clinit_barrier(ciMethod* method) {
85 assert(!method->holder()->is_not_initialized(), "initialization should have been started");
86
87 Label L_skip_barrier;
88 Register klass = Z_R1_scratch;
89
90 metadata2reg(method->holder()->constant_encoding(), klass);
91 __ clinit_barrier(klass, Z_thread, &L_skip_barrier /*L_fast_path*/);
92
93 __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub());
94 __ z_br(klass);
95
96 __ bind(L_skip_barrier);
97 }
98
99 void LIR_Assembler::osr_entry() {
100 // On-stack-replacement entry sequence (interpreter frame layout described in frame_s390.hpp):
101 //
102 // 1. Create a new compiled activation.
103 // 2. Initialize local variables in the compiled activation. The expression stack must be empty
104 // at the osr_bci; it is not initialized.
105 // 3. Jump to the continuation address in compiled code to resume execution.
106
107 // OSR entry point
108 offsets()->set_value(CodeOffsets::OSR_Entry, code_offset());
109 BlockBegin* osr_entry = compilation()->hir()->osr_entry();
110 ValueStack* entry_state = osr_entry->end()->state();
111 int number_of_locks = entry_state->locks_size();
112
113 // Create a frame for the compiled activation.
114 __ build_frame(initial_frame_size_in_bytes(), bang_size_in_bytes());
115
116 // OSR buffer is
117 //
118 // locals[nlocals-1..0]
119 // monitors[number_of_locks-1..0]
120 //
121 // Locals is a direct copy of the interpreter frame so in the osr buffer
122 // the first slot in the local array is the last local from the interpreter
123 // and the last slot is local[0] (receiver) from the interpreter
124 //
125 // Similarly with locks. The first lock slot in the osr buffer is the nth lock
126 // from the interpreter frame, the nth lock slot in the osr buffer is 0th lock
127 // in the interpreter frame (the method lock if a sync method)
128
129 // Initialize monitors in the compiled activation.
130 // I0: pointer to osr buffer
131 //
132 // All other registers are dead at this point and the locals will be
133 // copied into place by code emitted in the IR.
134
135 Register OSR_buf = osrBufferPointer()->as_register();
136 { assert(frame::interpreter_frame_monitor_size() == BasicObjectLock::size(), "adjust code below");
137 int monitor_offset = BytesPerWord * method()->max_locals() +
138 (2 * BytesPerWord) * (number_of_locks - 1);
139 // SharedRuntime::OSR_migration_begin() packs BasicObjectLocks in
140 // the OSR buffer using 2 word entries: first the lock and then
141 // the oop.
142 for (int i = 0; i < number_of_locks; i++) {
143 int slot_offset = monitor_offset - ((i * 2) * BytesPerWord);
144 // Verify the interpreter's monitor has a non-null object.
145 __ asm_assert_mem8_isnot_zero(slot_offset + 1*BytesPerWord, OSR_buf, "locked object is NULL", __LINE__);
146 // Copy the lock field into the compiled activation.
147 __ z_lg(Z_R1_scratch, slot_offset + 0, OSR_buf);
148 __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_lock(i));
149 __ z_lg(Z_R1_scratch, slot_offset + 1*BytesPerWord, OSR_buf);
150 __ z_stg(Z_R1_scratch, frame_map()->address_for_monitor_object(i));
151 }
152 }
153 }
154
155 // --------------------------------------------------------------------------------------------
156
157 address LIR_Assembler::emit_call_c(address a) {
158 __ align_call_far_patchable(__ pc());
159 address call_addr = __ call_c_opt(a);
160 if (call_addr == NULL) {
161 bailout("const section overflow");
162 }
163 return call_addr;
164 }
165
166 int LIR_Assembler::emit_exception_handler() {
167 // If the last instruction is a call (typically to do a throw which
168 // is coming at the end after block reordering) the return address
169 // must still point into the code area in order to avoid assertion
170 // failures when searching for the corresponding bci. => Add a nop.
171 // (was bug 5/14/1999 - gri)
172 __ nop();
173
174 // Generate code for exception handler.
175 address handler_base = __ start_a_stub(exception_handler_size());
176 if (handler_base == NULL) {
177 // Not enough space left for the handler.
178 bailout("exception handler overflow");
179 return -1;
180 }
181
182 int offset = code_offset();
183
184 address a = Runtime1::entry_for (Runtime1::handle_exception_from_callee_id);
185 address call_addr = emit_call_c(a);
186 CHECK_BAILOUT_(-1);
187 __ should_not_reach_here();
188 guarantee(code_offset() - offset <= exception_handler_size(), "overflow");
189 __ end_a_stub();
190
191 return offset;
192 }
193
194 // Emit the code to remove the frame from the stack in the exception
195 // unwind path.
196 int LIR_Assembler::emit_unwind_handler() {
197 #ifndef PRODUCT
198 if (CommentedAssembly) {
199 _masm->block_comment("Unwind handler");
200 }
201 #endif
202
203 int offset = code_offset();
204 Register exception_oop_callee_saved = Z_R10; // Z_R10 is callee-saved.
205 Register Rtmp1 = Z_R11;
206 Register Rtmp2 = Z_R12;
207
208 // Fetch the exception from TLS and clear out exception related thread state.
209 Address exc_oop_addr = Address(Z_thread, JavaThread::exception_oop_offset());
210 Address exc_pc_addr = Address(Z_thread, JavaThread::exception_pc_offset());
211 __ z_lg(Z_EXC_OOP, exc_oop_addr);
212 __ clear_mem(exc_oop_addr, sizeof(oop));
213 __ clear_mem(exc_pc_addr, sizeof(intptr_t));
214
215 __ bind(_unwind_handler_entry);
216 __ verify_not_null_oop(Z_EXC_OOP);
217 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
218 __ lgr_if_needed(exception_oop_callee_saved, Z_EXC_OOP); // Preserve the exception.
219 }
220
221 // Preform needed unlocking.
222 MonitorExitStub* stub = NULL;
223 if (method()->is_synchronized()) {
224 // Runtime1::monitorexit_id expects lock address in Z_R1_scratch.
225 LIR_Opr lock = FrameMap::as_opr(Z_R1_scratch);
226 monitor_address(0, lock);
227 stub = new MonitorExitStub(lock, true, 0);
228 __ unlock_object(Rtmp1, Rtmp2, lock->as_register(), *stub->entry());
229 __ bind(*stub->continuation());
230 }
231
232 if (compilation()->env()->dtrace_method_probes()) {
233 ShouldNotReachHere(); // Not supported.
234 #if 0
235 __ mov(rdi, r15_thread);
236 __ mov_metadata(rsi, method()->constant_encoding());
237 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit)));
238 #endif
239 }
240
241 if (method()->is_synchronized() || compilation()->env()->dtrace_method_probes()) {
242 __ lgr_if_needed(Z_EXC_OOP, exception_oop_callee_saved); // Restore the exception.
243 }
244
245 // Remove the activation and dispatch to the unwind handler.
246 __ pop_frame();
247 __ z_lg(Z_EXC_PC, _z_abi16(return_pc), Z_SP);
248
249 // Z_EXC_OOP: exception oop
250 // Z_EXC_PC: exception pc
251
252 // Dispatch to the unwind logic.
253 __ load_const_optimized(Z_R5, Runtime1::entry_for (Runtime1::unwind_exception_id));
254 __ z_br(Z_R5);
255
256 // Emit the slow path assembly.
257 if (stub != NULL) {
258 stub->emit_code(this);
259 }
260
261 return offset;
262 }
263
264 int LIR_Assembler::emit_deopt_handler() {
265 // If the last instruction is a call (typically to do a throw which
266 // is coming at the end after block reordering) the return address
267 // must still point into the code area in order to avoid assertion
268 // failures when searching for the corresponding bci. => Add a nop.
269 // (was bug 5/14/1999 - gri)
270 __ nop();
271
272 // Generate code for exception handler.
273 address handler_base = __ start_a_stub(deopt_handler_size());
274 if (handler_base == NULL) {
275 // Not enough space left for the handler.
276 bailout("deopt handler overflow");
277 return -1;
278 } int offset = code_offset();
279 // Size must be constant (see HandlerImpl::emit_deopt_handler).
280 __ load_const(Z_R1_scratch, SharedRuntime::deopt_blob()->unpack());
281 __ call(Z_R1_scratch);
282 guarantee(code_offset() - offset <= deopt_handler_size(), "overflow");
283 __ end_a_stub();
284
285 return offset;
286 }
287
288 void LIR_Assembler::jobject2reg(jobject o, Register reg) {
289 if (o == NULL) {
290 __ clear_reg(reg, true/*64bit*/, false/*set cc*/); // Must not kill cc set by cmove.
291 } else {
292 AddressLiteral a = __ allocate_oop_address(o);
293 bool success = __ load_oop_from_toc(reg, a, reg);
294 if (!success) {
295 bailout("const section overflow");
296 }
297 }
298 }
299
300 void LIR_Assembler::jobject2reg_with_patching(Register reg, CodeEmitInfo *info) {
301 // Allocate a new index in table to hold the object once it's been patched.
302 int oop_index = __ oop_recorder()->allocate_oop_index(NULL);
303 PatchingStub* patch = new PatchingStub(_masm, patching_id(info), oop_index);
304
305 AddressLiteral addrlit((intptr_t)0, oop_Relocation::spec(oop_index));
306 assert(addrlit.rspec().type() == relocInfo::oop_type, "must be an oop reloc");
307 // The NULL will be dynamically patched later so the sequence to
308 // load the address literal must not be optimized.
309 __ load_const(reg, addrlit);
310
311 patching_epilog(patch, lir_patch_normal, reg, info);
312 }
313
314 void LIR_Assembler::metadata2reg(Metadata* md, Register reg) {
315 bool success = __ set_metadata_constant(md, reg);
316 if (!success) {
317 bailout("const section overflow");
318 return;
319 }
320 }
321
322 void LIR_Assembler::klass2reg_with_patching(Register reg, CodeEmitInfo *info) {
323 // Allocate a new index in table to hold the klass once it's been patched.
324 int index = __ oop_recorder()->allocate_metadata_index(NULL);
325 PatchingStub* patch = new PatchingStub(_masm, PatchingStub::load_klass_id, index);
326 AddressLiteral addrlit((intptr_t)0, metadata_Relocation::spec(index));
327 assert(addrlit.rspec().type() == relocInfo::metadata_type, "must be an metadata reloc");
328 // The NULL will be dynamically patched later so the sequence to
329 // load the address literal must not be optimized.
330 __ load_const(reg, addrlit);
331
332 patching_epilog(patch, lir_patch_normal, reg, info);
333 }
334
335 void LIR_Assembler::emit_op3(LIR_Op3* op) {
336 switch (op->code()) {
337 case lir_idiv:
338 case lir_irem:
339 arithmetic_idiv(op->code(),
340 op->in_opr1(),
341 op->in_opr2(),
342 op->in_opr3(),
343 op->result_opr(),
344 op->info());
345 break;
346 case lir_fmad: {
347 const FloatRegister opr1 = op->in_opr1()->as_double_reg(),
348 opr2 = op->in_opr2()->as_double_reg(),
349 opr3 = op->in_opr3()->as_double_reg(),
350 res = op->result_opr()->as_double_reg();
351 __ z_madbr(opr3, opr1, opr2);
352 if (res != opr3) { __ z_ldr(res, opr3); }
353 } break;
354 case lir_fmaf: {
355 const FloatRegister opr1 = op->in_opr1()->as_float_reg(),
356 opr2 = op->in_opr2()->as_float_reg(),
357 opr3 = op->in_opr3()->as_float_reg(),
358 res = op->result_opr()->as_float_reg();
359 __ z_maebr(opr3, opr1, opr2);
360 if (res != opr3) { __ z_ler(res, opr3); }
361 } break;
362 default: ShouldNotReachHere(); break;
363 }
364 }
365
366
367 void LIR_Assembler::emit_opBranch(LIR_OpBranch* op) {
368 #ifdef ASSERT
369 assert(op->block() == NULL || op->block()->label() == op->label(), "wrong label");
370 if (op->block() != NULL) { _branch_target_blocks.append(op->block()); }
371 if (op->ublock() != NULL) { _branch_target_blocks.append(op->ublock()); }
372 #endif
373
374 if (op->cond() == lir_cond_always) {
375 if (op->info() != NULL) { add_debug_info_for_branch(op->info()); }
376 __ branch_optimized(Assembler::bcondAlways, *(op->label()));
377 } else {
378 Assembler::branch_condition acond = Assembler::bcondZero;
379 if (op->code() == lir_cond_float_branch) {
380 assert(op->ublock() != NULL, "must have unordered successor");
381 __ branch_optimized(Assembler::bcondNotOrdered, *(op->ublock()->label()));
382 }
383 switch (op->cond()) {
384 case lir_cond_equal: acond = Assembler::bcondEqual; break;
385 case lir_cond_notEqual: acond = Assembler::bcondNotEqual; break;
386 case lir_cond_less: acond = Assembler::bcondLow; break;
387 case lir_cond_lessEqual: acond = Assembler::bcondNotHigh; break;
388 case lir_cond_greaterEqual: acond = Assembler::bcondNotLow; break;
389 case lir_cond_greater: acond = Assembler::bcondHigh; break;
390 case lir_cond_belowEqual: acond = Assembler::bcondNotHigh; break;
391 case lir_cond_aboveEqual: acond = Assembler::bcondNotLow; break;
392 default: ShouldNotReachHere();
393 }
394 __ branch_optimized(acond,*(op->label()));
395 }
396 }
397
398
399 void LIR_Assembler::emit_opConvert(LIR_OpConvert* op) {
400 LIR_Opr src = op->in_opr();
401 LIR_Opr dest = op->result_opr();
402
403 switch (op->bytecode()) {
404 case Bytecodes::_i2l:
405 __ move_reg_if_needed(dest->as_register_lo(), T_LONG, src->as_register(), T_INT);
406 break;
407
408 case Bytecodes::_l2i:
409 __ move_reg_if_needed(dest->as_register(), T_INT, src->as_register_lo(), T_LONG);
410 break;
411
412 case Bytecodes::_i2b:
413 __ move_reg_if_needed(dest->as_register(), T_BYTE, src->as_register(), T_INT);
414 break;
415
416 case Bytecodes::_i2c:
417 __ move_reg_if_needed(dest->as_register(), T_CHAR, src->as_register(), T_INT);
418 break;
419
420 case Bytecodes::_i2s:
421 __ move_reg_if_needed(dest->as_register(), T_SHORT, src->as_register(), T_INT);
422 break;
423
424 case Bytecodes::_f2d:
425 assert(dest->is_double_fpu(), "check");
426 __ move_freg_if_needed(dest->as_double_reg(), T_DOUBLE, src->as_float_reg(), T_FLOAT);
427 break;
428
429 case Bytecodes::_d2f:
430 assert(dest->is_single_fpu(), "check");
431 __ move_freg_if_needed(dest->as_float_reg(), T_FLOAT, src->as_double_reg(), T_DOUBLE);
432 break;
433
434 case Bytecodes::_i2f:
435 __ z_cefbr(dest->as_float_reg(), src->as_register());
436 break;
437
438 case Bytecodes::_i2d:
439 __ z_cdfbr(dest->as_double_reg(), src->as_register());
440 break;
441
442 case Bytecodes::_l2f:
443 __ z_cegbr(dest->as_float_reg(), src->as_register_lo());
444 break;
445 case Bytecodes::_l2d:
446 __ z_cdgbr(dest->as_double_reg(), src->as_register_lo());
447 break;
448
449 case Bytecodes::_f2i:
450 case Bytecodes::_f2l: {
451 Label done;
452 FloatRegister Rsrc = src->as_float_reg();
453 Register Rdst = (op->bytecode() == Bytecodes::_f2i ? dest->as_register() : dest->as_register_lo());
454 __ clear_reg(Rdst, true, false);
455 __ z_cebr(Rsrc, Rsrc);
456 __ z_brno(done); // NaN -> 0
457 if (op->bytecode() == Bytecodes::_f2i) {
458 __ z_cfebr(Rdst, Rsrc, Assembler::to_zero);
459 } else { // op->bytecode() == Bytecodes::_f2l
460 __ z_cgebr(Rdst, Rsrc, Assembler::to_zero);
461 }
462 __ bind(done);
463 }
464 break;
465
466 case Bytecodes::_d2i:
467 case Bytecodes::_d2l: {
468 Label done;
469 FloatRegister Rsrc = src->as_double_reg();
470 Register Rdst = (op->bytecode() == Bytecodes::_d2i ? dest->as_register() : dest->as_register_lo());
471 __ clear_reg(Rdst, true, false); // Don't set CC.
472 __ z_cdbr(Rsrc, Rsrc);
473 __ z_brno(done); // NaN -> 0
474 if (op->bytecode() == Bytecodes::_d2i) {
475 __ z_cfdbr(Rdst, Rsrc, Assembler::to_zero);
476 } else { // Bytecodes::_d2l
477 __ z_cgdbr(Rdst, Rsrc, Assembler::to_zero);
478 }
479 __ bind(done);
480 }
481 break;
482
483 default: ShouldNotReachHere();
484 }
485 }
486
487 void LIR_Assembler::align_call(LIR_Code code) {
488 // End of call instruction must be 4 byte aligned.
489 int offset = __ offset();
490 switch (code) {
491 case lir_icvirtual_call:
492 offset += MacroAssembler::load_const_from_toc_size();
493 // no break
494 case lir_static_call:
495 case lir_optvirtual_call:
496 case lir_dynamic_call:
497 offset += NativeCall::call_far_pcrelative_displacement_offset;
498 break;
499 default: ShouldNotReachHere();
500 }
501 if ((offset & (NativeCall::call_far_pcrelative_displacement_alignment-1)) != 0) {
502 __ nop();
503 }
504 }
505
506 void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
507 assert((__ offset() + NativeCall::call_far_pcrelative_displacement_offset) % NativeCall::call_far_pcrelative_displacement_alignment == 0,
508 "must be aligned (offset=%d)", __ offset());
509 assert(rtype == relocInfo::none ||
510 rtype == relocInfo::opt_virtual_call_type ||
511 rtype == relocInfo::static_call_type, "unexpected rtype");
512 // Prepend each BRASL with a nop.
513 __ relocate(rtype);
514 __ z_nop();
515 __ z_brasl(Z_R14, op->addr());
516 add_call_info(code_offset(), op->info());
517 }
518
519 void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
520 address virtual_call_oop_addr = NULL;
521 AddressLiteral empty_ic((address) Universe::non_oop_word());
522 virtual_call_oop_addr = __ pc();
523 bool success = __ load_const_from_toc(Z_inline_cache, empty_ic);
524 if (!success) {
525 bailout("const section overflow");
526 return;
527 }
528
529 // CALL to fixup routine. Fixup routine uses ScopeDesc info
530 // to determine who we intended to call.
531 __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr));
532 call(op, relocInfo::none);
533 }
534
535 void LIR_Assembler::move_regs(Register from_reg, Register to_reg) {
536 if (from_reg != to_reg) __ z_lgr(to_reg, from_reg);
537 }
538
539 void LIR_Assembler::const2stack(LIR_Opr src, LIR_Opr dest) {
540 assert(src->is_constant(), "should not call otherwise");
541 assert(dest->is_stack(), "should not call otherwise");
542 LIR_Const* c = src->as_constant_ptr();
543
544 unsigned int lmem = 0;
545 unsigned int lcon = 0;
546 int64_t cbits = 0;
547 Address dest_addr;
548 switch (c->type()) {
549 case T_INT: // fall through
550 case T_FLOAT:
551 dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
552 lmem = 4; lcon = 4; cbits = c->as_jint_bits();
553 break;
554
555 case T_ADDRESS:
556 dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
557 lmem = 8; lcon = 4; cbits = c->as_jint_bits();
558 break;
559
560 case T_OBJECT:
561 dest_addr = frame_map()->address_for_slot(dest->single_stack_ix());
562 if (c->as_jobject() == NULL) {
563 __ store_const(dest_addr, (int64_t)NULL_WORD, 8, 8);
564 } else {
565 jobject2reg(c->as_jobject(), Z_R1_scratch);
566 __ reg2mem_opt(Z_R1_scratch, dest_addr, true);
567 }
568 return;
569
570 case T_LONG: // fall through
571 case T_DOUBLE:
572 dest_addr = frame_map()->address_for_slot(dest->double_stack_ix());
573 lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
574 break;
575
576 default:
577 ShouldNotReachHere();
578 }
579
580 __ store_const(dest_addr, cbits, lmem, lcon);
581 }
582
583 void LIR_Assembler::const2mem(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info, bool wide) {
584 assert(src->is_constant(), "should not call otherwise");
585 assert(dest->is_address(), "should not call otherwise");
586
587 LIR_Const* c = src->as_constant_ptr();
588 Address addr = as_Address(dest->as_address_ptr());
589
590 int store_offset = -1;
591
592 if (dest->as_address_ptr()->index()->is_valid()) {
593 switch (type) {
594 case T_INT: // fall through
595 case T_FLOAT:
596 __ load_const_optimized(Z_R0_scratch, c->as_jint_bits());
597 store_offset = __ offset();
598 if (Immediate::is_uimm12(addr.disp())) {
599 __ z_st(Z_R0_scratch, addr);
600 } else {
601 __ z_sty(Z_R0_scratch, addr);
602 }
603 break;
604
605 case T_ADDRESS:
606 __ load_const_optimized(Z_R1_scratch, c->as_jint_bits());
607 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
608 break;
609
610 case T_OBJECT: // fall through
611 case T_ARRAY:
612 if (c->as_jobject() == NULL) {
613 if (UseCompressedOops && !wide) {
614 __ clear_reg(Z_R1_scratch, false);
615 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
616 } else {
617 __ clear_reg(Z_R1_scratch, true);
618 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
619 }
620 } else {
621 jobject2reg(c->as_jobject(), Z_R1_scratch);
622 if (UseCompressedOops && !wide) {
623 __ encode_heap_oop(Z_R1_scratch);
624 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
625 } else {
626 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
627 }
628 }
629 assert(store_offset >= 0, "check");
630 break;
631
632 case T_LONG: // fall through
633 case T_DOUBLE:
634 __ load_const_optimized(Z_R1_scratch, (int64_t)(c->as_jlong_bits()));
635 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
636 break;
637
638 case T_BOOLEAN: // fall through
639 case T_BYTE:
640 __ load_const_optimized(Z_R0_scratch, (int8_t)(c->as_jint()));
641 store_offset = __ offset();
642 if (Immediate::is_uimm12(addr.disp())) {
643 __ z_stc(Z_R0_scratch, addr);
644 } else {
645 __ z_stcy(Z_R0_scratch, addr);
646 }
647 break;
648
649 case T_CHAR: // fall through
650 case T_SHORT:
651 __ load_const_optimized(Z_R0_scratch, (int16_t)(c->as_jint()));
652 store_offset = __ offset();
653 if (Immediate::is_uimm12(addr.disp())) {
654 __ z_sth(Z_R0_scratch, addr);
655 } else {
656 __ z_sthy(Z_R0_scratch, addr);
657 }
658 break;
659
660 default:
661 ShouldNotReachHere();
662 }
663
664 } else { // no index
665
666 unsigned int lmem = 0;
667 unsigned int lcon = 0;
668 int64_t cbits = 0;
669
670 switch (type) {
671 case T_INT: // fall through
672 case T_FLOAT:
673 lmem = 4; lcon = 4; cbits = c->as_jint_bits();
674 break;
675
676 case T_ADDRESS:
677 lmem = 8; lcon = 4; cbits = c->as_jint_bits();
678 break;
679
680 case T_OBJECT: // fall through
681 case T_ARRAY:
682 if (c->as_jobject() == NULL) {
683 if (UseCompressedOops && !wide) {
684 store_offset = __ store_const(addr, (int32_t)NULL_WORD, 4, 4);
685 } else {
686 store_offset = __ store_const(addr, (int64_t)NULL_WORD, 8, 8);
687 }
688 } else {
689 jobject2reg(c->as_jobject(), Z_R1_scratch);
690 if (UseCompressedOops && !wide) {
691 __ encode_heap_oop(Z_R1_scratch);
692 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, false);
693 } else {
694 store_offset = __ reg2mem_opt(Z_R1_scratch, addr, true);
695 }
696 }
697 assert(store_offset >= 0, "check");
698 break;
699
700 case T_LONG: // fall through
701 case T_DOUBLE:
702 lmem = 8; lcon = 8; cbits = (int64_t)(c->as_jlong_bits());
703 break;
704
705 case T_BOOLEAN: // fall through
706 case T_BYTE:
707 lmem = 1; lcon = 1; cbits = (int8_t)(c->as_jint());
708 break;
709
710 case T_CHAR: // fall through
711 case T_SHORT:
712 lmem = 2; lcon = 2; cbits = (int16_t)(c->as_jint());
713 break;
714
715 default:
716 ShouldNotReachHere();
717 }
718
719 if (store_offset == -1) {
720 store_offset = __ store_const(addr, cbits, lmem, lcon);
721 assert(store_offset >= 0, "check");
722 }
723 }
724
725 if (info != NULL) {
726 add_debug_info_for_null_check(store_offset, info);
727 }
728 }
729
730 void LIR_Assembler::const2reg(LIR_Opr src, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
731 assert(src->is_constant(), "should not call otherwise");
732 assert(dest->is_register(), "should not call otherwise");
733 LIR_Const* c = src->as_constant_ptr();
734
735 switch (c->type()) {
736 case T_INT: {
737 assert(patch_code == lir_patch_none, "no patching handled here");
738 __ load_const_optimized(dest->as_register(), c->as_jint());
739 break;
740 }
741
742 case T_ADDRESS: {
743 assert(patch_code == lir_patch_none, "no patching handled here");
744 __ load_const_optimized(dest->as_register(), c->as_jint());
745 break;
746 }
747
748 case T_LONG: {
749 assert(patch_code == lir_patch_none, "no patching handled here");
750 __ load_const_optimized(dest->as_register_lo(), (intptr_t)c->as_jlong());
751 break;
752 }
753
754 case T_OBJECT: {
755 if (patch_code != lir_patch_none) {
756 jobject2reg_with_patching(dest->as_register(), info);
757 } else {
758 jobject2reg(c->as_jobject(), dest->as_register());
759 }
760 break;
761 }
762
763 case T_METADATA: {
764 if (patch_code != lir_patch_none) {
765 klass2reg_with_patching(dest->as_register(), info);
766 } else {
767 metadata2reg(c->as_metadata(), dest->as_register());
768 }
769 break;
770 }
771
772 case T_FLOAT: {
773 Register toc_reg = Z_R1_scratch;
774 __ load_toc(toc_reg);
775 address const_addr = __ float_constant(c->as_jfloat());
776 if (const_addr == NULL) {
777 bailout("const section overflow");
778 break;
779 }
780 int displ = const_addr - _masm->code()->consts()->start();
781 if (dest->is_single_fpu()) {
782 __ z_ley(dest->as_float_reg(), displ, toc_reg);
783 } else {
784 assert(dest->is_single_cpu(), "Must be a cpu register.");
785 __ z_ly(dest->as_register(), displ, toc_reg);
786 }
787 }
788 break;
789
790 case T_DOUBLE: {
791 Register toc_reg = Z_R1_scratch;
792 __ load_toc(toc_reg);
793 address const_addr = __ double_constant(c->as_jdouble());
794 if (const_addr == NULL) {
795 bailout("const section overflow");
796 break;
797 }
798 int displ = const_addr - _masm->code()->consts()->start();
799 if (dest->is_double_fpu()) {
800 __ z_ldy(dest->as_double_reg(), displ, toc_reg);
801 } else {
802 assert(dest->is_double_cpu(), "Must be a long register.");
803 __ z_lg(dest->as_register_lo(), displ, toc_reg);
804 }
805 }
806 break;
807
808 default:
809 ShouldNotReachHere();
810 }
811 }
812
813 Address LIR_Assembler::as_Address(LIR_Address* addr) {
814 if (addr->base()->is_illegal()) {
815 Unimplemented();
816 }
817
818 Register base = addr->base()->as_pointer_register();
819
820 if (addr->index()->is_illegal()) {
821 return Address(base, addr->disp());
822 } else if (addr->index()->is_cpu_register()) {
823 Register index = addr->index()->as_pointer_register();
824 return Address(base, index, addr->disp());
825 } else if (addr->index()->is_constant()) {
826 intptr_t addr_offset = addr->index()->as_constant_ptr()->as_jint() + addr->disp();
827 return Address(base, addr_offset);
828 } else {
829 ShouldNotReachHere();
830 return Address();
831 }
832 }
833
834 void LIR_Assembler::stack2stack(LIR_Opr src, LIR_Opr dest, BasicType type) {
835 switch (type) {
836 case T_INT:
837 case T_FLOAT: {
838 Register tmp = Z_R1_scratch;
839 Address from = frame_map()->address_for_slot(src->single_stack_ix());
840 Address to = frame_map()->address_for_slot(dest->single_stack_ix());
841 __ mem2reg_opt(tmp, from, false);
842 __ reg2mem_opt(tmp, to, false);
843 break;
844 }
845 case T_ADDRESS:
846 case T_OBJECT: {
847 Register tmp = Z_R1_scratch;
848 Address from = frame_map()->address_for_slot(src->single_stack_ix());
849 Address to = frame_map()->address_for_slot(dest->single_stack_ix());
850 __ mem2reg_opt(tmp, from, true);
851 __ reg2mem_opt(tmp, to, true);
852 break;
853 }
854 case T_LONG:
855 case T_DOUBLE: {
856 Register tmp = Z_R1_scratch;
857 Address from = frame_map()->address_for_double_slot(src->double_stack_ix());
858 Address to = frame_map()->address_for_double_slot(dest->double_stack_ix());
859 __ mem2reg_opt(tmp, from, true);
860 __ reg2mem_opt(tmp, to, true);
861 break;
862 }
863
864 default:
865 ShouldNotReachHere();
866 }
867 }
868
869 // 4-byte accesses only! Don't use it to access 8 bytes!
870 Address LIR_Assembler::as_Address_hi(LIR_Address* addr) {
871 ShouldNotCallThis();
872 return 0; // unused
873 }
874
875 // 4-byte accesses only! Don't use it to access 8 bytes!
876 Address LIR_Assembler::as_Address_lo(LIR_Address* addr) {
877 ShouldNotCallThis();
878 return 0; // unused
879 }
880
881 void LIR_Assembler::mem2reg(LIR_Opr src_opr, LIR_Opr dest, BasicType type, LIR_PatchCode patch_code,
882 CodeEmitInfo* info, bool wide) {
883
884 assert(type != T_METADATA, "load of metadata ptr not supported");
885 LIR_Address* addr = src_opr->as_address_ptr();
886 LIR_Opr to_reg = dest;
887
888 Register src = addr->base()->as_pointer_register();
889 Register disp_reg = Z_R0;
890 int disp_value = addr->disp();
891 bool needs_patching = (patch_code != lir_patch_none);
892
893 if (addr->base()->type() == T_OBJECT) {
894 __ verify_oop(src, FILE_AND_LINE);
895 }
896
897 PatchingStub* patch = NULL;
898 if (needs_patching) {
899 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
900 assert(!to_reg->is_double_cpu() ||
901 patch_code == lir_patch_none ||
902 patch_code == lir_patch_normal, "patching doesn't match register");
903 }
904
905 if (addr->index()->is_illegal()) {
906 if (!Immediate::is_simm20(disp_value)) {
907 if (needs_patching) {
908 __ load_const(Z_R1_scratch, (intptr_t)0);
909 } else {
910 __ load_const_optimized(Z_R1_scratch, disp_value);
911 }
912 disp_reg = Z_R1_scratch;
913 disp_value = 0;
914 }
915 } else {
916 if (!Immediate::is_simm20(disp_value)) {
917 __ load_const_optimized(Z_R1_scratch, disp_value);
918 __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
919 disp_reg = Z_R1_scratch;
920 disp_value = 0;
921 }
922 disp_reg = addr->index()->as_pointer_register();
923 }
924
925 // Remember the offset of the load. The patching_epilog must be done
926 // before the call to add_debug_info, otherwise the PcDescs don't get
927 // entered in increasing order.
928 int offset = code_offset();
929
930 assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
931
932 bool short_disp = Immediate::is_uimm12(disp_value);
933
934 switch (type) {
935 case T_BOOLEAN: // fall through
936 case T_BYTE : __ z_lb(dest->as_register(), disp_value, disp_reg, src); break;
937 case T_CHAR : __ z_llgh(dest->as_register(), disp_value, disp_reg, src); break;
938 case T_SHORT :
939 if (short_disp) {
940 __ z_lh(dest->as_register(), disp_value, disp_reg, src);
941 } else {
942 __ z_lhy(dest->as_register(), disp_value, disp_reg, src);
943 }
944 break;
945 case T_INT :
946 if (short_disp) {
947 __ z_l(dest->as_register(), disp_value, disp_reg, src);
948 } else {
949 __ z_ly(dest->as_register(), disp_value, disp_reg, src);
950 }
951 break;
952 case T_ADDRESS:
953 __ z_lg(dest->as_register(), disp_value, disp_reg, src);
954 break;
955 case T_ARRAY : // fall through
956 case T_OBJECT:
957 {
958 if (UseCompressedOops && !wide) {
959 __ z_llgf(dest->as_register(), disp_value, disp_reg, src);
960 __ oop_decoder(dest->as_register(), dest->as_register(), true);
961 } else {
962 __ z_lg(dest->as_register(), disp_value, disp_reg, src);
963 }
964 __ verify_oop(dest->as_register(), FILE_AND_LINE);
965 break;
966 }
967 case T_FLOAT:
968 if (short_disp) {
969 __ z_le(dest->as_float_reg(), disp_value, disp_reg, src);
970 } else {
971 __ z_ley(dest->as_float_reg(), disp_value, disp_reg, src);
972 }
973 break;
974 case T_DOUBLE:
975 if (short_disp) {
976 __ z_ld(dest->as_double_reg(), disp_value, disp_reg, src);
977 } else {
978 __ z_ldy(dest->as_double_reg(), disp_value, disp_reg, src);
979 }
980 break;
981 case T_LONG : __ z_lg(dest->as_register_lo(), disp_value, disp_reg, src); break;
982 default : ShouldNotReachHere();
983 }
984
985 if (patch != NULL) {
986 patching_epilog(patch, patch_code, src, info);
987 }
988 if (info != NULL) add_debug_info_for_null_check(offset, info);
989 }
990
991 void LIR_Assembler::stack2reg(LIR_Opr src, LIR_Opr dest, BasicType type) {
992 assert(src->is_stack(), "should not call otherwise");
993 assert(dest->is_register(), "should not call otherwise");
994
995 if (dest->is_single_cpu()) {
996 if (is_reference_type(type)) {
997 __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
998 __ verify_oop(dest->as_register(), FILE_AND_LINE);
999 } else if (type == T_METADATA || type == T_ADDRESS) {
1000 __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), true);
1001 } else {
1002 __ mem2reg_opt(dest->as_register(), frame_map()->address_for_slot(src->single_stack_ix()), false);
1003 }
1004 } else if (dest->is_double_cpu()) {
1005 Address src_addr_LO = frame_map()->address_for_slot(src->double_stack_ix());
1006 __ mem2reg_opt(dest->as_register_lo(), src_addr_LO, true);
1007 } else if (dest->is_single_fpu()) {
1008 Address src_addr = frame_map()->address_for_slot(src->single_stack_ix());
1009 __ mem2freg_opt(dest->as_float_reg(), src_addr, false);
1010 } else if (dest->is_double_fpu()) {
1011 Address src_addr = frame_map()->address_for_slot(src->double_stack_ix());
1012 __ mem2freg_opt(dest->as_double_reg(), src_addr, true);
1013 } else {
1014 ShouldNotReachHere();
1015 }
1016 }
1017
1018 void LIR_Assembler::reg2stack(LIR_Opr src, LIR_Opr dest, BasicType type, bool pop_fpu_stack) {
1019 assert(src->is_register(), "should not call otherwise");
1020 assert(dest->is_stack(), "should not call otherwise");
1021
1022 if (src->is_single_cpu()) {
1023 const Address dst = frame_map()->address_for_slot(dest->single_stack_ix());
1024 if (is_reference_type(type)) {
1025 __ verify_oop(src->as_register(), FILE_AND_LINE);
1026 __ reg2mem_opt(src->as_register(), dst, true);
1027 } else if (type == T_METADATA || type == T_ADDRESS) {
1028 __ reg2mem_opt(src->as_register(), dst, true);
1029 } else {
1030 __ reg2mem_opt(src->as_register(), dst, false);
1031 }
1032 } else if (src->is_double_cpu()) {
1033 Address dstLO = frame_map()->address_for_slot(dest->double_stack_ix());
1034 __ reg2mem_opt(src->as_register_lo(), dstLO, true);
1035 } else if (src->is_single_fpu()) {
1036 Address dst_addr = frame_map()->address_for_slot(dest->single_stack_ix());
1037 __ freg2mem_opt(src->as_float_reg(), dst_addr, false);
1038 } else if (src->is_double_fpu()) {
1039 Address dst_addr = frame_map()->address_for_slot(dest->double_stack_ix());
1040 __ freg2mem_opt(src->as_double_reg(), dst_addr, true);
1041 } else {
1042 ShouldNotReachHere();
1043 }
1044 }
1045
1046 void LIR_Assembler::reg2reg(LIR_Opr from_reg, LIR_Opr to_reg) {
1047 if (from_reg->is_float_kind() && to_reg->is_float_kind()) {
1048 if (from_reg->is_double_fpu()) {
1049 // double to double moves
1050 assert(to_reg->is_double_fpu(), "should match");
1051 __ z_ldr(to_reg->as_double_reg(), from_reg->as_double_reg());
1052 } else {
1053 // float to float moves
1054 assert(to_reg->is_single_fpu(), "should match");
1055 __ z_ler(to_reg->as_float_reg(), from_reg->as_float_reg());
1056 }
1057 } else if (!from_reg->is_float_kind() && !to_reg->is_float_kind()) {
1058 if (from_reg->is_double_cpu()) {
1059 __ z_lgr(to_reg->as_pointer_register(), from_reg->as_pointer_register());
1060 } else if (to_reg->is_double_cpu()) {
1061 // int to int moves
1062 __ z_lgr(to_reg->as_register_lo(), from_reg->as_register());
1063 } else {
1064 // int to int moves
1065 __ z_lgr(to_reg->as_register(), from_reg->as_register());
1066 }
1067 } else {
1068 ShouldNotReachHere();
1069 }
1070 if (is_reference_type(to_reg->type())) {
1071 __ verify_oop(to_reg->as_register(), FILE_AND_LINE);
1072 }
1073 }
1074
1075 void LIR_Assembler::reg2mem(LIR_Opr from, LIR_Opr dest_opr, BasicType type,
1076 LIR_PatchCode patch_code, CodeEmitInfo* info, bool pop_fpu_stack,
1077 bool wide) {
1078 assert(type != T_METADATA, "store of metadata ptr not supported");
1079 LIR_Address* addr = dest_opr->as_address_ptr();
1080
1081 Register dest = addr->base()->as_pointer_register();
1082 Register disp_reg = Z_R0;
1083 int disp_value = addr->disp();
1084 bool needs_patching = (patch_code != lir_patch_none);
1085
1086 if (addr->base()->is_oop_register()) {
1087 __ verify_oop(dest, FILE_AND_LINE);
1088 }
1089
1090 PatchingStub* patch = NULL;
1091 if (needs_patching) {
1092 patch = new PatchingStub(_masm, PatchingStub::access_field_id);
1093 assert(!from->is_double_cpu() ||
1094 patch_code == lir_patch_none ||
1095 patch_code == lir_patch_normal, "patching doesn't match register");
1096 }
1097
1098 assert(!needs_patching || (!Immediate::is_simm20(disp_value) && addr->index()->is_illegal()), "assumption");
1099 if (addr->index()->is_illegal()) {
1100 if (!Immediate::is_simm20(disp_value)) {
1101 if (needs_patching) {
1102 __ load_const(Z_R1_scratch, (intptr_t)0);
1103 } else {
1104 __ load_const_optimized(Z_R1_scratch, disp_value);
1105 }
1106 disp_reg = Z_R1_scratch;
1107 disp_value = 0;
1108 }
1109 } else {
1110 if (!Immediate::is_simm20(disp_value)) {
1111 __ load_const_optimized(Z_R1_scratch, disp_value);
1112 __ z_la(Z_R1_scratch, 0, Z_R1_scratch, addr->index()->as_register());
1113 disp_reg = Z_R1_scratch;
1114 disp_value = 0;
1115 }
1116 disp_reg = addr->index()->as_pointer_register();
1117 }
1118
1119 assert(disp_reg != Z_R0 || Immediate::is_simm20(disp_value), "should have set this up");
1120
1121 if (is_reference_type(type)) {
1122 __ verify_oop(from->as_register(), FILE_AND_LINE);
1123 }
1124
1125 bool short_disp = Immediate::is_uimm12(disp_value);
1126
1127 // Remember the offset of the store. The patching_epilog must be done
1128 // before the call to add_debug_info_for_null_check, otherwise the PcDescs don't get
1129 // entered in increasing order.
1130 int offset = code_offset();
1131 switch (type) {
1132 case T_BOOLEAN: // fall through
1133 case T_BYTE :
1134 if (short_disp) {
1135 __ z_stc(from->as_register(), disp_value, disp_reg, dest);
1136 } else {
1137 __ z_stcy(from->as_register(), disp_value, disp_reg, dest);
1138 }
1139 break;
1140 case T_CHAR : // fall through
1141 case T_SHORT :
1142 if (short_disp) {
1143 __ z_sth(from->as_register(), disp_value, disp_reg, dest);
1144 } else {
1145 __ z_sthy(from->as_register(), disp_value, disp_reg, dest);
1146 }
1147 break;
1148 case T_INT :
1149 if (short_disp) {
1150 __ z_st(from->as_register(), disp_value, disp_reg, dest);
1151 } else {
1152 __ z_sty(from->as_register(), disp_value, disp_reg, dest);
1153 }
1154 break;
1155 case T_LONG : __ z_stg(from->as_register_lo(), disp_value, disp_reg, dest); break;
1156 case T_ADDRESS: __ z_stg(from->as_register(), disp_value, disp_reg, dest); break;
1157 break;
1158 case T_ARRAY : // fall through
1159 case T_OBJECT:
1160 {
1161 if (UseCompressedOops && !wide) {
1162 Register compressed_src = Z_R14;
1163 __ oop_encoder(compressed_src, from->as_register(), true, (disp_reg != Z_R1) ? Z_R1 : Z_R0, -1, true);
1164 offset = code_offset();
1165 if (short_disp) {
1166 __ z_st(compressed_src, disp_value, disp_reg, dest);
1167 } else {
1168 __ z_sty(compressed_src, disp_value, disp_reg, dest);
1169 }
1170 } else {
1171 __ z_stg(from->as_register(), disp_value, disp_reg, dest);
1172 }
1173 break;
1174 }
1175 case T_FLOAT :
1176 if (short_disp) {
1177 __ z_ste(from->as_float_reg(), disp_value, disp_reg, dest);
1178 } else {
1179 __ z_stey(from->as_float_reg(), disp_value, disp_reg, dest);
1180 }
1181 break;
1182 case T_DOUBLE:
1183 if (short_disp) {
1184 __ z_std(from->as_double_reg(), disp_value, disp_reg, dest);
1185 } else {
1186 __ z_stdy(from->as_double_reg(), disp_value, disp_reg, dest);
1187 }
1188 break;
1189 default: ShouldNotReachHere();
1190 }
1191
1192 if (patch != NULL) {
1193 patching_epilog(patch, patch_code, dest, info);
1194 }
1195
1196 if (info != NULL) add_debug_info_for_null_check(offset, info);
1197 }
1198
1199
1200 void LIR_Assembler::return_op(LIR_Opr result, C1SafepointPollStub* code_stub) {
1201 assert(result->is_illegal() ||
1202 (result->is_single_cpu() && result->as_register() == Z_R2) ||
1203 (result->is_double_cpu() && result->as_register_lo() == Z_R2) ||
1204 (result->is_single_fpu() && result->as_float_reg() == Z_F0) ||
1205 (result->is_double_fpu() && result->as_double_reg() == Z_F0), "convention");
1206
1207 __ z_lg(Z_R1_scratch, Address(Z_thread, JavaThread::polling_page_offset()));
1208
1209 // Pop the frame before the safepoint code.
1210 __ pop_frame_restore_retPC(initial_frame_size_in_bytes());
1211
1212 if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
1213 __ reserved_stack_check(Z_R14);
1214 }
1215
1216 // We need to mark the code position where the load from the safepoint
1217 // polling page was emitted as relocInfo::poll_return_type here.
1218 __ relocate(relocInfo::poll_return_type);
1219 __ load_from_polling_page(Z_R1_scratch);
1220
1221 __ z_br(Z_R14); // Return to caller.
1222 }
1223
1224 int LIR_Assembler::safepoint_poll(LIR_Opr tmp, CodeEmitInfo* info) {
1225 const Register poll_addr = tmp->as_register_lo();
1226 __ z_lg(poll_addr, Address(Z_thread, JavaThread::polling_page_offset()));
1227 guarantee(info != NULL, "Shouldn't be NULL");
1228 add_debug_info_for_branch(info);
1229 int offset = __ offset();
1230 __ relocate(relocInfo::poll_type);
1231 __ load_from_polling_page(poll_addr);
1232 return offset;
1233 }
1234
1235 void LIR_Assembler::emit_static_call_stub() {
1236
1237 // Stub is fixed up when the corresponding call is converted from calling
1238 // compiled code to calling interpreted code.
1239
1240 address call_pc = __ pc();
1241 address stub = __ start_a_stub(call_stub_size());
1242 if (stub == NULL) {
1243 bailout("static call stub overflow");
1244 return;
1245 }
1246
1247 int start = __ offset();
1248
1249 __ relocate(static_stub_Relocation::spec(call_pc));
1250
1251 // See also Matcher::interpreter_method_reg().
1252 AddressLiteral meta = __ allocate_metadata_address(NULL);
1253 bool success = __ load_const_from_toc(Z_method, meta);
1254
1255 __ set_inst_mark();
1256 AddressLiteral a((address)-1);
1257 success = success && __ load_const_from_toc(Z_R1, a);
1258 if (!success) {
1259 bailout("const section overflow");
1260 return;
1261 }
1262
1263 __ z_br(Z_R1);
1264 assert(__ offset() - start <= call_stub_size(), "stub too big");
1265 __ end_a_stub(); // Update current stubs pointer and restore insts_end.
1266 }
1267
1268 void LIR_Assembler::comp_op(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Op2* op) {
1269 bool unsigned_comp = condition == lir_cond_belowEqual || condition == lir_cond_aboveEqual;
1270 if (opr1->is_single_cpu()) {
1271 Register reg1 = opr1->as_register();
1272 if (opr2->is_single_cpu()) {
1273 // cpu register - cpu register
1274 if (is_reference_type(opr1->type())) {
1275 __ z_clgr(reg1, opr2->as_register());
1276 } else {
1277 assert(!is_reference_type(opr2->type()), "cmp int, oop?");
1278 if (unsigned_comp) {
1279 __ z_clr(reg1, opr2->as_register());
1280 } else {
1281 __ z_cr(reg1, opr2->as_register());
1282 }
1283 }
1284 } else if (opr2->is_stack()) {
1285 // cpu register - stack
1286 if (is_reference_type(opr1->type())) {
1287 __ z_cg(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1288 } else {
1289 if (unsigned_comp) {
1290 __ z_cly(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1291 } else {
1292 __ z_cy(reg1, frame_map()->address_for_slot(opr2->single_stack_ix()));
1293 }
1294 }
1295 } else if (opr2->is_constant()) {
1296 // cpu register - constant
1297 LIR_Const* c = opr2->as_constant_ptr();
1298 if (c->type() == T_INT) {
1299 if (unsigned_comp) {
1300 __ z_clfi(reg1, c->as_jint());
1301 } else {
1302 __ z_cfi(reg1, c->as_jint());
1303 }
1304 } else if (c->type() == T_METADATA) {
1305 // We only need, for now, comparison with NULL for metadata.
1306 assert(condition == lir_cond_equal || condition == lir_cond_notEqual, "oops");
1307 Metadata* m = c->as_metadata();
1308 if (m == NULL) {
1309 __ z_cghi(reg1, 0);
1310 } else {
1311 ShouldNotReachHere();
1312 }
1313 } else if (is_reference_type(c->type())) {
1314 // In 64bit oops are single register.
1315 jobject o = c->as_jobject();
1316 if (o == NULL) {
1317 __ z_ltgr(reg1, reg1);
1318 } else {
1319 jobject2reg(o, Z_R1_scratch);
1320 __ z_cgr(reg1, Z_R1_scratch);
1321 }
1322 } else {
1323 fatal("unexpected type: %s", basictype_to_str(c->type()));
1324 }
1325 // cpu register - address
1326 } else if (opr2->is_address()) {
1327 if (op->info() != NULL) {
1328 add_debug_info_for_null_check_here(op->info());
1329 }
1330 if (unsigned_comp) {
1331 __ z_cly(reg1, as_Address(opr2->as_address_ptr()));
1332 } else {
1333 __ z_cy(reg1, as_Address(opr2->as_address_ptr()));
1334 }
1335 } else {
1336 ShouldNotReachHere();
1337 }
1338
1339 } else if (opr1->is_double_cpu()) {
1340 assert(!unsigned_comp, "unexpected");
1341 Register xlo = opr1->as_register_lo();
1342 Register xhi = opr1->as_register_hi();
1343 if (opr2->is_double_cpu()) {
1344 __ z_cgr(xlo, opr2->as_register_lo());
1345 } else if (opr2->is_constant()) {
1346 // cpu register - constant 0
1347 assert(opr2->as_jlong() == (jlong)0, "only handles zero");
1348 __ z_ltgr(xlo, xlo);
1349 } else {
1350 ShouldNotReachHere();
1351 }
1352
1353 } else if (opr1->is_single_fpu()) {
1354 if (opr2->is_single_fpu()) {
1355 __ z_cebr(opr1->as_float_reg(), opr2->as_float_reg());
1356 } else {
1357 // stack slot
1358 Address addr = frame_map()->address_for_slot(opr2->single_stack_ix());
1359 if (Immediate::is_uimm12(addr.disp())) {
1360 __ z_ceb(opr1->as_float_reg(), addr);
1361 } else {
1362 __ z_ley(Z_fscratch_1, addr);
1363 __ z_cebr(opr1->as_float_reg(), Z_fscratch_1);
1364 }
1365 }
1366 } else if (opr1->is_double_fpu()) {
1367 if (opr2->is_double_fpu()) {
1368 __ z_cdbr(opr1->as_double_reg(), opr2->as_double_reg());
1369 } else {
1370 // stack slot
1371 Address addr = frame_map()->address_for_slot(opr2->double_stack_ix());
1372 if (Immediate::is_uimm12(addr.disp())) {
1373 __ z_cdb(opr1->as_double_reg(), addr);
1374 } else {
1375 __ z_ldy(Z_fscratch_1, addr);
1376 __ z_cdbr(opr1->as_double_reg(), Z_fscratch_1);
1377 }
1378 }
1379 } else {
1380 ShouldNotReachHere();
1381 }
1382 }
1383
1384 void LIR_Assembler::comp_fl2i(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst, LIR_Op2* op) {
1385 Label done;
1386 Register dreg = dst->as_register();
1387
1388 if (code == lir_cmp_fd2i || code == lir_ucmp_fd2i) {
1389 assert((left->is_single_fpu() && right->is_single_fpu()) ||
1390 (left->is_double_fpu() && right->is_double_fpu()), "unexpected operand types");
1391 bool is_single = left->is_single_fpu();
1392 bool is_unordered_less = (code == lir_ucmp_fd2i);
1393 FloatRegister lreg = is_single ? left->as_float_reg() : left->as_double_reg();
1394 FloatRegister rreg = is_single ? right->as_float_reg() : right->as_double_reg();
1395 if (is_single) {
1396 __ z_cebr(lreg, rreg);
1397 } else {
1398 __ z_cdbr(lreg, rreg);
1399 }
1400 if (VM_Version::has_LoadStoreConditional()) {
1401 Register one = Z_R0_scratch;
1402 Register minus_one = Z_R1_scratch;
1403 __ z_lghi(minus_one, -1);
1404 __ z_lghi(one, 1);
1405 __ z_lghi(dreg, 0);
1406 __ z_locgr(dreg, one, is_unordered_less ? Assembler::bcondHigh : Assembler::bcondHighOrNotOrdered);
1407 __ z_locgr(dreg, minus_one, is_unordered_less ? Assembler::bcondLowOrNotOrdered : Assembler::bcondLow);
1408 } else {
1409 __ clear_reg(dreg, true, false);
1410 __ z_bre(done); // if (left == right) dst = 0
1411
1412 // if (left > right || ((code ~= cmpg) && (left <> right)) dst := 1
1413 __ z_lhi(dreg, 1);
1414 __ z_brc(is_unordered_less ? Assembler::bcondHigh : Assembler::bcondHighOrNotOrdered, done);
1415
1416 // if (left < right || ((code ~= cmpl) && (left <> right)) dst := -1
1417 __ z_lhi(dreg, -1);
1418 }
1419 } else {
1420 assert(code == lir_cmp_l2i, "check");
1421 if (VM_Version::has_LoadStoreConditional()) {
1422 Register one = Z_R0_scratch;
1423 Register minus_one = Z_R1_scratch;
1424 __ z_cgr(left->as_register_lo(), right->as_register_lo());
1425 __ z_lghi(minus_one, -1);
1426 __ z_lghi(one, 1);
1427 __ z_lghi(dreg, 0);
1428 __ z_locgr(dreg, one, Assembler::bcondHigh);
1429 __ z_locgr(dreg, minus_one, Assembler::bcondLow);
1430 } else {
1431 __ z_cgr(left->as_register_lo(), right->as_register_lo());
1432 __ z_lghi(dreg, 0); // eq value
1433 __ z_bre(done);
1434 __ z_lghi(dreg, 1); // gt value
1435 __ z_brh(done);
1436 __ z_lghi(dreg, -1); // lt value
1437 }
1438 }
1439 __ bind(done);
1440 }
1441
1442 // result = condition ? opr1 : opr2
1443 void LIR_Assembler::cmove(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type,
1444 LIR_Opr cmp_opr1, LIR_Opr cmp_opr2) {
1445 assert(cmp_opr1 == LIR_OprFact::illegalOpr && cmp_opr2 == LIR_OprFact::illegalOpr, "unnecessary cmp oprs on s390");
1446
1447 Assembler::branch_condition acond = Assembler::bcondEqual, ncond = Assembler::bcondNotEqual;
1448 switch (condition) {
1449 case lir_cond_equal: acond = Assembler::bcondEqual; ncond = Assembler::bcondNotEqual; break;
1450 case lir_cond_notEqual: acond = Assembler::bcondNotEqual; ncond = Assembler::bcondEqual; break;
1451 case lir_cond_less: acond = Assembler::bcondLow; ncond = Assembler::bcondNotLow; break;
1452 case lir_cond_lessEqual: acond = Assembler::bcondNotHigh; ncond = Assembler::bcondHigh; break;
1453 case lir_cond_greaterEqual: acond = Assembler::bcondNotLow; ncond = Assembler::bcondLow; break;
1454 case lir_cond_greater: acond = Assembler::bcondHigh; ncond = Assembler::bcondNotHigh; break;
1455 case lir_cond_belowEqual: acond = Assembler::bcondNotHigh; ncond = Assembler::bcondHigh; break;
1456 case lir_cond_aboveEqual: acond = Assembler::bcondNotLow; ncond = Assembler::bcondLow; break;
1457 default: ShouldNotReachHere();
1458 }
1459
1460 if (opr1->is_cpu_register()) {
1461 reg2reg(opr1, result);
1462 } else if (opr1->is_stack()) {
1463 stack2reg(opr1, result, result->type());
1464 } else if (opr1->is_constant()) {
1465 const2reg(opr1, result, lir_patch_none, NULL);
1466 } else {
1467 ShouldNotReachHere();
1468 }
1469
1470 if (VM_Version::has_LoadStoreConditional() && !opr2->is_constant()) {
1471 // Optimized version that does not require a branch.
1472 if (opr2->is_single_cpu()) {
1473 assert(opr2->cpu_regnr() != result->cpu_regnr(), "opr2 already overwritten by previous move");
1474 __ z_locgr(result->as_register(), opr2->as_register(), ncond);
1475 } else if (opr2->is_double_cpu()) {
1476 assert(opr2->cpu_regnrLo() != result->cpu_regnrLo() && opr2->cpu_regnrLo() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1477 assert(opr2->cpu_regnrHi() != result->cpu_regnrLo() && opr2->cpu_regnrHi() != result->cpu_regnrHi(), "opr2 already overwritten by previous move");
1478 __ z_locgr(result->as_register_lo(), opr2->as_register_lo(), ncond);
1479 } else if (opr2->is_single_stack()) {
1480 __ z_loc(result->as_register(), frame_map()->address_for_slot(opr2->single_stack_ix()), ncond);
1481 } else if (opr2->is_double_stack()) {
1482 __ z_locg(result->as_register_lo(), frame_map()->address_for_slot(opr2->double_stack_ix()), ncond);
1483 } else {
1484 ShouldNotReachHere();
1485 }
1486 } else {
1487 Label skip;
1488 __ z_brc(acond, skip);
1489 if (opr2->is_cpu_register()) {
1490 reg2reg(opr2, result);
1491 } else if (opr2->is_stack()) {
1492 stack2reg(opr2, result, result->type());
1493 } else if (opr2->is_constant()) {
1494 const2reg(opr2, result, lir_patch_none, NULL);
1495 } else {
1496 ShouldNotReachHere();
1497 }
1498 __ bind(skip);
1499 }
1500 }
1501
1502 void LIR_Assembler::arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest,
1503 CodeEmitInfo* info, bool pop_fpu_stack) {
1504 assert(info == NULL, "should never be used, idiv/irem and ldiv/lrem not handled by this method");
1505
1506 if (left->is_single_cpu()) {
1507 assert(left == dest, "left and dest must be equal");
1508 Register lreg = left->as_register();
1509
1510 if (right->is_single_cpu()) {
1511 // cpu register - cpu register
1512 Register rreg = right->as_register();
1513 switch (code) {
1514 case lir_add: __ z_ar (lreg, rreg); break;
1515 case lir_sub: __ z_sr (lreg, rreg); break;
1516 case lir_mul: __ z_msr(lreg, rreg); break;
1517 default: ShouldNotReachHere();
1518 }
1519
1520 } else if (right->is_stack()) {
1521 // cpu register - stack
1522 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1523 switch (code) {
1524 case lir_add: __ z_ay(lreg, raddr); break;
1525 case lir_sub: __ z_sy(lreg, raddr); break;
1526 default: ShouldNotReachHere();
1527 }
1528
1529 } else if (right->is_constant()) {
1530 // cpu register - constant
1531 jint c = right->as_constant_ptr()->as_jint();
1532 switch (code) {
1533 case lir_add: __ z_agfi(lreg, c); break;
1534 case lir_sub: __ z_agfi(lreg, -c); break; // note: -min_jint == min_jint
1535 case lir_mul: __ z_msfi(lreg, c); break;
1536 default: ShouldNotReachHere();
1537 }
1538
1539 } else {
1540 ShouldNotReachHere();
1541 }
1542
1543 } else if (left->is_double_cpu()) {
1544 assert(left == dest, "left and dest must be equal");
1545 Register lreg_lo = left->as_register_lo();
1546 Register lreg_hi = left->as_register_hi();
1547
1548 if (right->is_double_cpu()) {
1549 // cpu register - cpu register
1550 Register rreg_lo = right->as_register_lo();
1551 Register rreg_hi = right->as_register_hi();
1552 assert_different_registers(lreg_lo, rreg_lo);
1553 switch (code) {
1554 case lir_add:
1555 __ z_agr(lreg_lo, rreg_lo);
1556 break;
1557 case lir_sub:
1558 __ z_sgr(lreg_lo, rreg_lo);
1559 break;
1560 case lir_mul:
1561 __ z_msgr(lreg_lo, rreg_lo);
1562 break;
1563 default:
1564 ShouldNotReachHere();
1565 }
1566
1567 } else if (right->is_constant()) {
1568 // cpu register - constant
1569 jlong c = right->as_constant_ptr()->as_jlong_bits();
1570 switch (code) {
1571 case lir_add: __ z_agfi(lreg_lo, c); break;
1572 case lir_sub:
1573 if (c != min_jint) {
1574 __ z_agfi(lreg_lo, -c);
1575 } else {
1576 // -min_jint cannot be represented as simm32 in z_agfi
1577 // min_jint sign extended: 0xffffffff80000000
1578 // -min_jint as 64 bit integer: 0x0000000080000000
1579 // 0x80000000 can be represented as uimm32 in z_algfi
1580 // lreg_lo := lreg_lo + -min_jint == lreg_lo + 0x80000000
1581 __ z_algfi(lreg_lo, UCONST64(0x80000000));
1582 }
1583 break;
1584 case lir_mul: __ z_msgfi(lreg_lo, c); break;
1585 default:
1586 ShouldNotReachHere();
1587 }
1588
1589 } else {
1590 ShouldNotReachHere();
1591 }
1592
1593 } else if (left->is_single_fpu()) {
1594 assert(left == dest, "left and dest must be equal");
1595 FloatRegister lreg = left->as_float_reg();
1596 FloatRegister rreg = right->is_single_fpu() ? right->as_float_reg() : fnoreg;
1597 Address raddr;
1598
1599 if (rreg == fnoreg) {
1600 assert(right->is_single_stack(), "constants should be loaded into register");
1601 raddr = frame_map()->address_for_slot(right->single_stack_ix());
1602 if (!Immediate::is_uimm12(raddr.disp())) {
1603 __ mem2freg_opt(rreg = Z_fscratch_1, raddr, false);
1604 }
1605 }
1606
1607 if (rreg != fnoreg) {
1608 switch (code) {
1609 case lir_add: __ z_aebr(lreg, rreg); break;
1610 case lir_sub: __ z_sebr(lreg, rreg); break;
1611 case lir_mul: __ z_meebr(lreg, rreg); break;
1612 case lir_div: __ z_debr(lreg, rreg); break;
1613 default: ShouldNotReachHere();
1614 }
1615 } else {
1616 switch (code) {
1617 case lir_add: __ z_aeb(lreg, raddr); break;
1618 case lir_sub: __ z_seb(lreg, raddr); break;
1619 case lir_mul: __ z_meeb(lreg, raddr); break;
1620 case lir_div: __ z_deb(lreg, raddr); break;
1621 default: ShouldNotReachHere();
1622 }
1623 }
1624 } else if (left->is_double_fpu()) {
1625 assert(left == dest, "left and dest must be equal");
1626 FloatRegister lreg = left->as_double_reg();
1627 FloatRegister rreg = right->is_double_fpu() ? right->as_double_reg() : fnoreg;
1628 Address raddr;
1629
1630 if (rreg == fnoreg) {
1631 assert(right->is_double_stack(), "constants should be loaded into register");
1632 raddr = frame_map()->address_for_slot(right->double_stack_ix());
1633 if (!Immediate::is_uimm12(raddr.disp())) {
1634 __ mem2freg_opt(rreg = Z_fscratch_1, raddr, true);
1635 }
1636 }
1637
1638 if (rreg != fnoreg) {
1639 switch (code) {
1640 case lir_add: __ z_adbr(lreg, rreg); break;
1641 case lir_sub: __ z_sdbr(lreg, rreg); break;
1642 case lir_mul: __ z_mdbr(lreg, rreg); break;
1643 case lir_div: __ z_ddbr(lreg, rreg); break;
1644 default: ShouldNotReachHere();
1645 }
1646 } else {
1647 switch (code) {
1648 case lir_add: __ z_adb(lreg, raddr); break;
1649 case lir_sub: __ z_sdb(lreg, raddr); break;
1650 case lir_mul: __ z_mdb(lreg, raddr); break;
1651 case lir_div: __ z_ddb(lreg, raddr); break;
1652 default: ShouldNotReachHere();
1653 }
1654 }
1655 } else if (left->is_address()) {
1656 assert(left == dest, "left and dest must be equal");
1657 assert(code == lir_add, "unsupported operation");
1658 assert(right->is_constant(), "unsupported operand");
1659 jint c = right->as_constant_ptr()->as_jint();
1660 LIR_Address* lir_addr = left->as_address_ptr();
1661 Address addr = as_Address(lir_addr);
1662 switch (lir_addr->type()) {
1663 case T_INT:
1664 __ add2mem_32(addr, c, Z_R1_scratch);
1665 break;
1666 case T_LONG:
1667 __ add2mem_64(addr, c, Z_R1_scratch);
1668 break;
1669 default:
1670 ShouldNotReachHere();
1671 }
1672 } else {
1673 ShouldNotReachHere();
1674 }
1675 }
1676
1677 void LIR_Assembler::intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr thread, LIR_Opr dest, LIR_Op* op) {
1678 switch (code) {
1679 case lir_sqrt: {
1680 assert(!thread->is_valid(), "there is no need for a thread_reg for dsqrt");
1681 FloatRegister src_reg = value->as_double_reg();
1682 FloatRegister dst_reg = dest->as_double_reg();
1683 __ z_sqdbr(dst_reg, src_reg);
1684 break;
1685 }
1686 case lir_abs: {
1687 assert(!thread->is_valid(), "there is no need for a thread_reg for fabs");
1688 FloatRegister src_reg = value->as_double_reg();
1689 FloatRegister dst_reg = dest->as_double_reg();
1690 __ z_lpdbr(dst_reg, src_reg);
1691 break;
1692 }
1693 default: {
1694 ShouldNotReachHere();
1695 break;
1696 }
1697 }
1698 }
1699
1700 void LIR_Assembler::logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dst) {
1701 if (left->is_single_cpu()) {
1702 Register reg = left->as_register();
1703 if (right->is_constant()) {
1704 int val = right->as_constant_ptr()->as_jint();
1705 switch (code) {
1706 case lir_logic_and: __ z_nilf(reg, val); break;
1707 case lir_logic_or: __ z_oilf(reg, val); break;
1708 case lir_logic_xor: __ z_xilf(reg, val); break;
1709 default: ShouldNotReachHere();
1710 }
1711 } else if (right->is_stack()) {
1712 Address raddr = frame_map()->address_for_slot(right->single_stack_ix());
1713 switch (code) {
1714 case lir_logic_and: __ z_ny(reg, raddr); break;
1715 case lir_logic_or: __ z_oy(reg, raddr); break;
1716 case lir_logic_xor: __ z_xy(reg, raddr); break;
1717 default: ShouldNotReachHere();
1718 }
1719 } else {
1720 Register rright = right->as_register();
1721 switch (code) {
1722 case lir_logic_and: __ z_nr(reg, rright); break;
1723 case lir_logic_or : __ z_or(reg, rright); break;
1724 case lir_logic_xor: __ z_xr(reg, rright); break;
1725 default: ShouldNotReachHere();
1726 }
1727 }
1728 move_regs(reg, dst->as_register());
1729 } else {
1730 Register l_lo = left->as_register_lo();
1731 if (right->is_constant()) {
1732 __ load_const_optimized(Z_R1_scratch, right->as_constant_ptr()->as_jlong());
1733 switch (code) {
1734 case lir_logic_and:
1735 __ z_ngr(l_lo, Z_R1_scratch);
1736 break;
1737 case lir_logic_or:
1738 __ z_ogr(l_lo, Z_R1_scratch);
1739 break;
1740 case lir_logic_xor:
1741 __ z_xgr(l_lo, Z_R1_scratch);
1742 break;
1743 default: ShouldNotReachHere();
1744 }
1745 } else {
1746 Register r_lo;
1747 if (is_reference_type(right->type())) {
1748 r_lo = right->as_register();
1749 } else {
1750 r_lo = right->as_register_lo();
1751 }
1752 switch (code) {
1753 case lir_logic_and:
1754 __ z_ngr(l_lo, r_lo);
1755 break;
1756 case lir_logic_or:
1757 __ z_ogr(l_lo, r_lo);
1758 break;
1759 case lir_logic_xor:
1760 __ z_xgr(l_lo, r_lo);
1761 break;
1762 default: ShouldNotReachHere();
1763 }
1764 }
1765
1766 Register dst_lo = dst->as_register_lo();
1767
1768 move_regs(l_lo, dst_lo);
1769 }
1770 }
1771
1772 // See operand selection in LIRGenerator::do_ArithmeticOp_Int().
1773 void LIR_Assembler::arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info) {
1774 if (left->is_double_cpu()) {
1775 // 64 bit integer case
1776 assert(left->is_double_cpu(), "left must be register");
1777 assert(right->is_double_cpu() || is_power_of_2(right->as_jlong()),
1778 "right must be register or power of 2 constant");
1779 assert(result->is_double_cpu(), "result must be register");
1780
1781 Register lreg = left->as_register_lo();
1782 Register dreg = result->as_register_lo();
1783
1784 if (right->is_constant()) {
1785 // Convert division by a power of two into some shifts and logical operations.
1786 Register treg1 = Z_R0_scratch;
1787 Register treg2 = Z_R1_scratch;
1788 jlong divisor = right->as_jlong();
1789 jlong log_divisor = log2i_exact(right->as_jlong());
1790
1791 if (divisor == min_jlong) {
1792 // Min_jlong is special. Result is '0' except for min_jlong/min_jlong = 1.
1793 if (dreg == lreg) {
1794 NearLabel done;
1795 __ load_const_optimized(treg2, min_jlong);
1796 __ z_cgr(lreg, treg2);
1797 __ z_lghi(dreg, 0); // Preserves condition code.
1798 __ z_brne(done);
1799 __ z_lghi(dreg, 1); // min_jlong / min_jlong = 1
1800 __ bind(done);
1801 } else {
1802 assert_different_registers(dreg, lreg);
1803 NearLabel done;
1804 __ z_lghi(dreg, 0);
1805 __ compare64_and_branch(lreg, min_jlong, Assembler::bcondNotEqual, done);
1806 __ z_lghi(dreg, 1);
1807 __ bind(done);
1808 }
1809 return;
1810 }
1811 __ move_reg_if_needed(dreg, T_LONG, lreg, T_LONG);
1812 if (divisor == 2) {
1813 __ z_srlg(treg2, dreg, 63); // dividend < 0 ? 1 : 0
1814 } else {
1815 __ z_srag(treg2, dreg, 63); // dividend < 0 ? -1 : 0
1816 __ and_imm(treg2, divisor - 1, treg1, true);
1817 }
1818 if (code == lir_idiv) {
1819 __ z_agr(dreg, treg2);
1820 __ z_srag(dreg, dreg, log_divisor);
1821 } else {
1822 assert(code == lir_irem, "check");
1823 __ z_agr(treg2, dreg);
1824 __ and_imm(treg2, ~(divisor - 1), treg1, true);
1825 __ z_sgr(dreg, treg2);
1826 }
1827 return;
1828 }
1829
1830 // Divisor is not a power of 2 constant.
1831 Register rreg = right->as_register_lo();
1832 Register treg = temp->as_register_lo();
1833 assert(right->is_double_cpu(), "right must be register");
1834 assert(lreg == Z_R11, "see ldivInOpr()");
1835 assert(rreg != lreg, "right register must not be same as left register");
1836 assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10) ||
1837 (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see ldivInOpr(), ldivOutOpr(), lremOutOpr()");
1838
1839 Register R1 = lreg->predecessor();
1840 Register R2 = rreg;
1841 assert(code != lir_idiv || lreg==dreg, "see code below");
1842 if (code == lir_idiv) {
1843 __ z_lcgr(lreg, lreg);
1844 } else {
1845 __ clear_reg(dreg, true, false);
1846 }
1847 NearLabel done;
1848 __ compare64_and_branch(R2, -1, Assembler::bcondEqual, done);
1849 if (code == lir_idiv) {
1850 __ z_lcgr(lreg, lreg); // Revert lcgr above.
1851 }
1852 if (ImplicitDiv0Checks) {
1853 // No debug info because the idiv won't trap.
1854 // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1855 // which is unnecessary, too.
1856 add_debug_info_for_div0(__ offset(), info);
1857 }
1858 __ z_dsgr(R1, R2);
1859 __ bind(done);
1860 return;
1861 }
1862
1863 // 32 bit integer case
1864
1865 assert(left->is_single_cpu(), "left must be register");
1866 assert(right->is_single_cpu() || is_power_of_2(right->as_jint()), "right must be register or power of 2 constant");
1867 assert(result->is_single_cpu(), "result must be register");
1868
1869 Register lreg = left->as_register();
1870 Register dreg = result->as_register();
1871
1872 if (right->is_constant()) {
1873 // Convert division by a power of two into some shifts and logical operations.
1874 Register treg1 = Z_R0_scratch;
1875 Register treg2 = Z_R1_scratch;
1876 jlong divisor = right->as_jint();
1877 jlong log_divisor = log2i_exact(right->as_jint());
1878 __ move_reg_if_needed(dreg, T_LONG, lreg, T_INT); // sign extend
1879 if (divisor == 2) {
1880 __ z_srlg(treg2, dreg, 63); // dividend < 0 ? 1 : 0
1881 } else {
1882 __ z_srag(treg2, dreg, 63); // dividend < 0 ? -1 : 0
1883 __ and_imm(treg2, divisor - 1, treg1, true);
1884 }
1885 if (code == lir_idiv) {
1886 __ z_agr(dreg, treg2);
1887 __ z_srag(dreg, dreg, log_divisor);
1888 } else {
1889 assert(code == lir_irem, "check");
1890 __ z_agr(treg2, dreg);
1891 __ and_imm(treg2, ~(divisor - 1), treg1, true);
1892 __ z_sgr(dreg, treg2);
1893 }
1894 return;
1895 }
1896
1897 // Divisor is not a power of 2 constant.
1898 Register rreg = right->as_register();
1899 Register treg = temp->as_register();
1900 assert(right->is_single_cpu(), "right must be register");
1901 assert(lreg == Z_R11, "left register must be rax,");
1902 assert(rreg != lreg, "right register must not be same as left register");
1903 assert((code == lir_idiv && dreg == Z_R11 && treg == Z_R10)
1904 || (code == lir_irem && dreg == Z_R10 && treg == Z_R11), "see divInOpr(), divOutOpr(), remOutOpr()");
1905
1906 Register R1 = lreg->predecessor();
1907 Register R2 = rreg;
1908 __ move_reg_if_needed(lreg, T_LONG, lreg, T_INT); // sign extend
1909 if (ImplicitDiv0Checks) {
1910 // No debug info because the idiv won't trap.
1911 // Add_debug_info_for_div0 would instantiate another DivByZeroStub,
1912 // which is unnecessary, too.
1913 add_debug_info_for_div0(__ offset(), info);
1914 }
1915 __ z_dsgfr(R1, R2);
1916 }
1917
1918 void LIR_Assembler::throw_op(LIR_Opr exceptionPC, LIR_Opr exceptionOop, CodeEmitInfo* info) {
1919 assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1920 assert(exceptionPC->as_register() == Z_EXC_PC, "should match");
1921
1922 // Exception object is not added to oop map by LinearScan
1923 // (LinearScan assumes that no oops are in fixed registers).
1924 info->add_register_oop(exceptionOop);
1925
1926 // Reuse the debug info from the safepoint poll for the throw op itself.
1927 __ get_PC(Z_EXC_PC);
1928 add_call_info(__ offset(), info); // for exception handler
1929 address stub = Runtime1::entry_for (compilation()->has_fpu_code() ? Runtime1::handle_exception_id
1930 : Runtime1::handle_exception_nofpu_id);
1931 emit_call_c(stub);
1932 }
1933
1934 void LIR_Assembler::unwind_op(LIR_Opr exceptionOop) {
1935 assert(exceptionOop->as_register() == Z_EXC_OOP, "should match");
1936
1937 __ branch_optimized(Assembler::bcondAlways, _unwind_handler_entry);
1938 }
1939
1940 void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
1941 ciArrayKlass* default_type = op->expected_type();
1942 Register src = op->src()->as_register();
1943 Register dst = op->dst()->as_register();
1944 Register src_pos = op->src_pos()->as_register();
1945 Register dst_pos = op->dst_pos()->as_register();
1946 Register length = op->length()->as_register();
1947 Register tmp = op->tmp()->as_register();
1948
1949 CodeStub* stub = op->stub();
1950 int flags = op->flags();
1951 BasicType basic_type = default_type != NULL ? default_type->element_type()->basic_type() : T_ILLEGAL;
1952 if (basic_type == T_ARRAY) basic_type = T_OBJECT;
1953
1954 // If we don't know anything, just go through the generic arraycopy.
1955 if (default_type == NULL) {
1956 address copyfunc_addr = StubRoutines::generic_arraycopy();
1957
1958 if (copyfunc_addr == NULL) {
1959 // Take a slow path for generic arraycopy.
1960 __ branch_optimized(Assembler::bcondAlways, *stub->entry());
1961 __ bind(*stub->continuation());
1962 return;
1963 }
1964
1965 // Save outgoing arguments in callee saved registers (C convention) in case
1966 // a call to System.arraycopy is needed.
1967 Register callee_saved_src = Z_R10;
1968 Register callee_saved_src_pos = Z_R11;
1969 Register callee_saved_dst = Z_R12;
1970 Register callee_saved_dst_pos = Z_R13;
1971 Register callee_saved_length = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
1972
1973 __ lgr_if_needed(callee_saved_src, src);
1974 __ lgr_if_needed(callee_saved_src_pos, src_pos);
1975 __ lgr_if_needed(callee_saved_dst, dst);
1976 __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
1977 __ lgr_if_needed(callee_saved_length, length);
1978
1979 // C function requires 64 bit values.
1980 __ z_lgfr(src_pos, src_pos);
1981 __ z_lgfr(dst_pos, dst_pos);
1982 __ z_lgfr(length, length);
1983
1984 // Pass arguments: may push as this is not a safepoint; SP must be fix at each safepoint.
1985
1986 // The arguments are in the corresponding registers.
1987 assert(Z_ARG1 == src, "assumption");
1988 assert(Z_ARG2 == src_pos, "assumption");
1989 assert(Z_ARG3 == dst, "assumption");
1990 assert(Z_ARG4 == dst_pos, "assumption");
1991 assert(Z_ARG5 == length, "assumption");
1992 #ifndef PRODUCT
1993 if (PrintC1Statistics) {
1994 __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_generic_arraycopystub_cnt);
1995 __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
1996 }
1997 #endif
1998 emit_call_c(copyfunc_addr);
1999 CHECK_BAILOUT();
2000
2001 __ compare32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
2002
2003 __ z_lgr(tmp, Z_RET);
2004 __ z_xilf(tmp, -1);
2005
2006 // Restore values from callee saved registers so they are where the stub
2007 // expects them.
2008 __ lgr_if_needed(src, callee_saved_src);
2009 __ lgr_if_needed(src_pos, callee_saved_src_pos);
2010 __ lgr_if_needed(dst, callee_saved_dst);
2011 __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
2012 __ lgr_if_needed(length, callee_saved_length);
2013
2014 __ z_sr(length, tmp);
2015 __ z_ar(src_pos, tmp);
2016 __ z_ar(dst_pos, tmp);
2017 __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2018
2019 __ bind(*stub->continuation());
2020 return;
2021 }
2022
2023 assert(default_type != NULL && default_type->is_array_klass() && default_type->is_loaded(), "must be true at this point");
2024
2025 int elem_size = type2aelembytes(basic_type);
2026 int shift_amount;
2027
2028 switch (elem_size) {
2029 case 1 :
2030 shift_amount = 0;
2031 break;
2032 case 2 :
2033 shift_amount = 1;
2034 break;
2035 case 4 :
2036 shift_amount = 2;
2037 break;
2038 case 8 :
2039 shift_amount = 3;
2040 break;
2041 default:
2042 shift_amount = -1;
2043 ShouldNotReachHere();
2044 }
2045
2046 Address src_length_addr = Address(src, arrayOopDesc::length_offset_in_bytes());
2047 Address dst_length_addr = Address(dst, arrayOopDesc::length_offset_in_bytes());
2048 Address src_klass_addr = Address(src, oopDesc::klass_offset_in_bytes());
2049 Address dst_klass_addr = Address(dst, oopDesc::klass_offset_in_bytes());
2050
2051 // Length and pos's are all sign extended at this point on 64bit.
2052
2053 // test for NULL
2054 if (flags & LIR_OpArrayCopy::src_null_check) {
2055 __ compareU64_and_branch(src, (intptr_t)0, Assembler::bcondZero, *stub->entry());
2056 }
2057 if (flags & LIR_OpArrayCopy::dst_null_check) {
2058 __ compareU64_and_branch(dst, (intptr_t)0, Assembler::bcondZero, *stub->entry());
2059 }
2060
2061 // Check if negative.
2062 if (flags & LIR_OpArrayCopy::src_pos_positive_check) {
2063 __ compare32_and_branch(src_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2064 }
2065 if (flags & LIR_OpArrayCopy::dst_pos_positive_check) {
2066 __ compare32_and_branch(dst_pos, (intptr_t)0, Assembler::bcondLow, *stub->entry());
2067 }
2068
2069 // If the compiler was not able to prove that exact type of the source or the destination
2070 // of the arraycopy is an array type, check at runtime if the source or the destination is
2071 // an instance type.
2072 if (flags & LIR_OpArrayCopy::type_check) {
2073 assert(Klass::_lh_neutral_value == 0, "or replace z_lt instructions");
2074
2075 if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2076 __ load_klass(tmp, dst);
2077 __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2078 __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2079 }
2080
2081 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2082 __ load_klass(tmp, src);
2083 __ z_lt(tmp, Address(tmp, in_bytes(Klass::layout_helper_offset())));
2084 __ branch_optimized(Assembler::bcondNotLow, *stub->entry());
2085 }
2086 }
2087
2088 if (flags & LIR_OpArrayCopy::src_range_check) {
2089 __ z_la(tmp, Address(src_pos, length));
2090 __ z_cl(tmp, src_length_addr);
2091 __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2092 }
2093 if (flags & LIR_OpArrayCopy::dst_range_check) {
2094 __ z_la(tmp, Address(dst_pos, length));
2095 __ z_cl(tmp, dst_length_addr);
2096 __ branch_optimized(Assembler::bcondHigh, *stub->entry());
2097 }
2098
2099 if (flags & LIR_OpArrayCopy::length_positive_check) {
2100 __ z_ltr(length, length);
2101 __ branch_optimized(Assembler::bcondNegative, *stub->entry());
2102 }
2103
2104 // Stubs require 64 bit values.
2105 __ z_lgfr(src_pos, src_pos); // int -> long
2106 __ z_lgfr(dst_pos, dst_pos); // int -> long
2107 __ z_lgfr(length, length); // int -> long
2108
2109 if (flags & LIR_OpArrayCopy::type_check) {
2110 // We don't know the array types are compatible.
2111 if (basic_type != T_OBJECT) {
2112 // Simple test for basic type arrays.
2113 if (UseCompressedClassPointers) {
2114 __ z_l(tmp, src_klass_addr);
2115 __ z_c(tmp, dst_klass_addr);
2116 } else {
2117 __ z_lg(tmp, src_klass_addr);
2118 __ z_cg(tmp, dst_klass_addr);
2119 }
2120 __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2121 } else {
2122 // For object arrays, if src is a sub class of dst then we can
2123 // safely do the copy.
2124 NearLabel cont, slow;
2125 Register src_klass = Z_R1_scratch;
2126 Register dst_klass = Z_R10;
2127
2128 __ load_klass(src_klass, src);
2129 __ load_klass(dst_klass, dst);
2130
2131 __ check_klass_subtype_fast_path(src_klass, dst_klass, tmp, &cont, &slow, NULL);
2132
2133 store_parameter(src_klass, 0); // sub
2134 store_parameter(dst_klass, 1); // super
2135 emit_call_c(Runtime1::entry_for (Runtime1::slow_subtype_check_id));
2136 CHECK_BAILOUT2(cont, slow);
2137 // Sets condition code 0 for match (2 otherwise).
2138 __ branch_optimized(Assembler::bcondEqual, cont);
2139
2140 __ bind(slow);
2141
2142 address copyfunc_addr = StubRoutines::checkcast_arraycopy();
2143 if (copyfunc_addr != NULL) { // use stub if available
2144 // Src is not a sub class of dst so we have to do a
2145 // per-element check.
2146
2147 int mask = LIR_OpArrayCopy::src_objarray|LIR_OpArrayCopy::dst_objarray;
2148 if ((flags & mask) != mask) {
2149 // Check that at least both of them object arrays.
2150 assert(flags & mask, "one of the two should be known to be an object array");
2151
2152 if (!(flags & LIR_OpArrayCopy::src_objarray)) {
2153 __ load_klass(tmp, src);
2154 } else if (!(flags & LIR_OpArrayCopy::dst_objarray)) {
2155 __ load_klass(tmp, dst);
2156 }
2157 Address klass_lh_addr(tmp, Klass::layout_helper_offset());
2158 jint objArray_lh = Klass::array_layout_helper(T_OBJECT);
2159 __ load_const_optimized(Z_R1_scratch, objArray_lh);
2160 __ z_c(Z_R1_scratch, klass_lh_addr);
2161 __ branch_optimized(Assembler::bcondNotEqual, *stub->entry());
2162 }
2163
2164 // Save outgoing arguments in callee saved registers (C convention) in case
2165 // a call to System.arraycopy is needed.
2166 Register callee_saved_src = Z_R10;
2167 Register callee_saved_src_pos = Z_R11;
2168 Register callee_saved_dst = Z_R12;
2169 Register callee_saved_dst_pos = Z_R13;
2170 Register callee_saved_length = Z_ARG5; // Z_ARG5 == Z_R6 is callee saved.
2171
2172 __ lgr_if_needed(callee_saved_src, src);
2173 __ lgr_if_needed(callee_saved_src_pos, src_pos);
2174 __ lgr_if_needed(callee_saved_dst, dst);
2175 __ lgr_if_needed(callee_saved_dst_pos, dst_pos);
2176 __ lgr_if_needed(callee_saved_length, length);
2177
2178 __ z_llgfr(length, length); // Higher 32bits must be null.
2179
2180 __ z_sllg(Z_ARG1, src_pos, shift_amount); // index -> byte offset
2181 __ z_sllg(Z_ARG2, dst_pos, shift_amount); // index -> byte offset
2182
2183 __ z_la(Z_ARG1, Address(src, Z_ARG1, arrayOopDesc::base_offset_in_bytes(basic_type)));
2184 assert_different_registers(Z_ARG1, dst, dst_pos, length);
2185 __ z_la(Z_ARG2, Address(dst, Z_ARG2, arrayOopDesc::base_offset_in_bytes(basic_type)));
2186 assert_different_registers(Z_ARG2, dst, length);
2187
2188 __ z_lgr(Z_ARG3, length);
2189 assert_different_registers(Z_ARG3, dst);
2190
2191 __ load_klass(Z_ARG5, dst);
2192 __ z_lg(Z_ARG5, Address(Z_ARG5, ObjArrayKlass::element_klass_offset()));
2193 __ z_lg(Z_ARG4, Address(Z_ARG5, Klass::super_check_offset_offset()));
2194 emit_call_c(copyfunc_addr);
2195 CHECK_BAILOUT2(cont, slow);
2196
2197 #ifndef PRODUCT
2198 if (PrintC1Statistics) {
2199 NearLabel failed;
2200 __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondNotEqual, failed);
2201 __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_cnt);
2202 __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2203 __ bind(failed);
2204 }
2205 #endif
2206
2207 __ compareU32_and_branch(Z_RET, (intptr_t)0, Assembler::bcondEqual, *stub->continuation());
2208
2209 #ifndef PRODUCT
2210 if (PrintC1Statistics) {
2211 __ load_const_optimized(Z_R1_scratch, (address)&Runtime1::_arraycopy_checkcast_attempt_cnt);
2212 __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2213 }
2214 #endif
2215
2216 __ z_lgr(tmp, Z_RET);
2217 __ z_xilf(tmp, -1);
2218
2219 // Restore previously spilled arguments
2220 __ lgr_if_needed(src, callee_saved_src);
2221 __ lgr_if_needed(src_pos, callee_saved_src_pos);
2222 __ lgr_if_needed(dst, callee_saved_dst);
2223 __ lgr_if_needed(dst_pos, callee_saved_dst_pos);
2224 __ lgr_if_needed(length, callee_saved_length);
2225
2226 __ z_sr(length, tmp);
2227 __ z_ar(src_pos, tmp);
2228 __ z_ar(dst_pos, tmp);
2229 }
2230
2231 __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2232
2233 __ bind(cont);
2234 }
2235 }
2236
2237 #ifdef ASSERT
2238 if (basic_type != T_OBJECT || !(flags & LIR_OpArrayCopy::type_check)) {
2239 // Sanity check the known type with the incoming class. For the
2240 // primitive case the types must match exactly with src.klass and
2241 // dst.klass each exactly matching the default type. For the
2242 // object array case, if no type check is needed then either the
2243 // dst type is exactly the expected type and the src type is a
2244 // subtype which we can't check or src is the same array as dst
2245 // but not necessarily exactly of type default_type.
2246 NearLabel known_ok, halt;
2247 metadata2reg(default_type->constant_encoding(), tmp);
2248 if (UseCompressedClassPointers) {
2249 __ encode_klass_not_null(tmp);
2250 }
2251
2252 if (basic_type != T_OBJECT) {
2253 if (UseCompressedClassPointers) { __ z_c (tmp, dst_klass_addr); }
2254 else { __ z_cg(tmp, dst_klass_addr); }
2255 __ branch_optimized(Assembler::bcondNotEqual, halt);
2256 if (UseCompressedClassPointers) { __ z_c (tmp, src_klass_addr); }
2257 else { __ z_cg(tmp, src_klass_addr); }
2258 __ branch_optimized(Assembler::bcondEqual, known_ok);
2259 } else {
2260 if (UseCompressedClassPointers) { __ z_c (tmp, dst_klass_addr); }
2261 else { __ z_cg(tmp, dst_klass_addr); }
2262 __ branch_optimized(Assembler::bcondEqual, known_ok);
2263 __ compareU64_and_branch(src, dst, Assembler::bcondEqual, known_ok);
2264 }
2265 __ bind(halt);
2266 __ stop("incorrect type information in arraycopy");
2267 __ bind(known_ok);
2268 }
2269 #endif
2270
2271 #ifndef PRODUCT
2272 if (PrintC1Statistics) {
2273 __ load_const_optimized(Z_R1_scratch, Runtime1::arraycopy_count_address(basic_type));
2274 __ add2mem_32(Address(Z_R1_scratch), 1, Z_R0_scratch);
2275 }
2276 #endif
2277
2278 __ z_sllg(tmp, src_pos, shift_amount); // index -> byte offset
2279 __ z_sllg(Z_R1_scratch, dst_pos, shift_amount); // index -> byte offset
2280
2281 assert_different_registers(Z_ARG1, dst, dst_pos, length);
2282 __ z_la(Z_ARG1, Address(src, tmp, arrayOopDesc::base_offset_in_bytes(basic_type)));
2283 assert_different_registers(Z_ARG2, length);
2284 __ z_la(Z_ARG2, Address(dst, Z_R1_scratch, arrayOopDesc::base_offset_in_bytes(basic_type)));
2285 __ lgr_if_needed(Z_ARG3, length);
2286
2287 bool disjoint = (flags & LIR_OpArrayCopy::overlapping) == 0;
2288 bool aligned = (flags & LIR_OpArrayCopy::unaligned) == 0;
2289 const char *name;
2290 address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
2291 __ call_VM_leaf(entry);
2292
2293 __ bind(*stub->continuation());
2294 }
2295
2296 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
2297 if (dest->is_single_cpu()) {
2298 if (left->type() == T_OBJECT) {
2299 switch (code) {
2300 case lir_shl: __ z_sllg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2301 case lir_shr: __ z_srag (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2302 case lir_ushr: __ z_srlg (dest->as_register(), left->as_register(), 0, count->as_register()); break;
2303 default: ShouldNotReachHere();
2304 }
2305 } else {
2306 assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2307 Register masked_count = Z_R1_scratch;
2308 __ z_lr(masked_count, count->as_register());
2309 __ z_nill(masked_count, 31);
2310 switch (code) {
2311 case lir_shl: __ z_sllg (dest->as_register(), left->as_register(), 0, masked_count); break;
2312 case lir_shr: __ z_sra (dest->as_register(), 0, masked_count); break;
2313 case lir_ushr: __ z_srl (dest->as_register(), 0, masked_count); break;
2314 default: ShouldNotReachHere();
2315 }
2316 }
2317 } else {
2318 switch (code) {
2319 case lir_shl: __ z_sllg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2320 case lir_shr: __ z_srag (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2321 case lir_ushr: __ z_srlg (dest->as_register_lo(), left->as_register_lo(), 0, count->as_register()); break;
2322 default: ShouldNotReachHere();
2323 }
2324 }
2325 }
2326
2327 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, jint count, LIR_Opr dest) {
2328 if (left->type() == T_OBJECT) {
2329 count = count & 63; // Shouldn't shift by more than sizeof(intptr_t).
2330 Register l = left->as_register();
2331 Register d = dest->as_register_lo();
2332 switch (code) {
2333 case lir_shl: __ z_sllg (d, l, count); break;
2334 case lir_shr: __ z_srag (d, l, count); break;
2335 case lir_ushr: __ z_srlg (d, l, count); break;
2336 default: ShouldNotReachHere();
2337 }
2338 return;
2339 }
2340 if (dest->is_single_cpu()) {
2341 assert(code == lir_shl || left == dest, "left and dest must be equal for 2 operand form right shifts");
2342 count = count & 0x1F; // Java spec
2343 switch (code) {
2344 case lir_shl: __ z_sllg (dest->as_register(), left->as_register(), count); break;
2345 case lir_shr: __ z_sra (dest->as_register(), count); break;
2346 case lir_ushr: __ z_srl (dest->as_register(), count); break;
2347 default: ShouldNotReachHere();
2348 }
2349 } else if (dest->is_double_cpu()) {
2350 count = count & 63; // Java spec
2351 Register l = left->as_pointer_register();
2352 Register d = dest->as_pointer_register();
2353 switch (code) {
2354 case lir_shl: __ z_sllg (d, l, count); break;
2355 case lir_shr: __ z_srag (d, l, count); break;
2356 case lir_ushr: __ z_srlg (d, l, count); break;
2357 default: ShouldNotReachHere();
2358 }
2359 } else {
2360 ShouldNotReachHere();
2361 }
2362 }
2363
2364 void LIR_Assembler::emit_alloc_obj(LIR_OpAllocObj* op) {
2365 if (op->init_check()) {
2366 // Make sure klass is initialized & doesn't have finalizer.
2367 const int state_offset = in_bytes(InstanceKlass::init_state_offset());
2368 Register iklass = op->klass()->as_register();
2369 add_debug_info_for_null_check_here(op->stub()->info());
2370 if (Immediate::is_uimm12(state_offset)) {
2371 __ z_cli(state_offset, iklass, InstanceKlass::fully_initialized);
2372 } else {
2373 __ z_cliy(state_offset, iklass, InstanceKlass::fully_initialized);
2374 }
2375 __ branch_optimized(Assembler::bcondNotEqual, *op->stub()->entry()); // Use long branch, because slow_case might be far.
2376 }
2377 __ allocate_object(op->obj()->as_register(),
2378 op->tmp1()->as_register(),
2379 op->tmp2()->as_register(),
2380 op->header_size(),
2381 op->object_size(),
2382 op->klass()->as_register(),
2383 *op->stub()->entry());
2384 __ bind(*op->stub()->continuation());
2385 __ verify_oop(op->obj()->as_register(), FILE_AND_LINE);
2386 }
2387
2388 void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
2389 Register len = op->len()->as_register();
2390 __ move_reg_if_needed(len, T_LONG, len, T_INT); // sign extend
2391
2392 if (UseSlowPath ||
2393 (!UseFastNewObjectArray && (is_reference_type(op->type()))) ||
2394 (!UseFastNewTypeArray && (!is_reference_type(op->type())))) {
2395 __ z_brul(*op->stub()->entry());
2396 } else {
2397 __ allocate_array(op->obj()->as_register(),
2398 op->len()->as_register(),
2399 op->tmp1()->as_register(),
2400 op->tmp2()->as_register(),
2401 arrayOopDesc::header_size(op->type()),
2402 type2aelembytes(op->type()),
2403 op->klass()->as_register(),
2404 *op->stub()->entry());
2405 }
2406 __ bind(*op->stub()->continuation());
2407 }
2408
2409 void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md, ciProfileData *data,
2410 Register recv, Register tmp1, Label* update_done) {
2411 uint i;
2412 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2413 Label next_test;
2414 // See if the receiver is receiver[n].
2415 Address receiver_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2416 __ z_cg(recv, receiver_addr);
2417 __ z_brne(next_test);
2418 Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
2419 __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2420 __ branch_optimized(Assembler::bcondAlways, *update_done);
2421 __ bind(next_test);
2422 }
2423
2424 // Didn't find receiver; find next empty slot and fill it in.
2425 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2426 Label next_test;
2427 Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
2428 __ z_ltg(Z_R0_scratch, recv_addr);
2429 __ z_brne(next_test);
2430 __ z_stg(recv, recv_addr);
2431 __ load_const_optimized(tmp1, DataLayout::counter_increment);
2432 __ z_stg(tmp1, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)), mdo);
2433 __ branch_optimized(Assembler::bcondAlways, *update_done);
2434 __ bind(next_test);
2435 }
2436 }
2437
2438 void LIR_Assembler::setup_md_access(ciMethod* method, int bci,
2439 ciMethodData*& md, ciProfileData*& data, int& mdo_offset_bias) {
2440 Unimplemented();
2441 }
2442
2443 void LIR_Assembler::store_parameter(Register r, int param_num) {
2444 assert(param_num >= 0, "invalid num");
2445 int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2446 assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2447 __ z_stg(r, offset_in_bytes, Z_SP);
2448 }
2449
2450 void LIR_Assembler::store_parameter(jint c, int param_num) {
2451 assert(param_num >= 0, "invalid num");
2452 int offset_in_bytes = param_num * BytesPerWord + FrameMap::first_available_sp_in_frame;
2453 assert(offset_in_bytes < frame_map()->reserved_argument_area_size(), "invalid offset");
2454 __ store_const(Address(Z_SP, offset_in_bytes), c, Z_R1_scratch, true);
2455 }
2456
2457 void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
2458 // We always need a stub for the failure case.
2459 CodeStub* stub = op->stub();
2460 Register obj = op->object()->as_register();
2461 Register k_RInfo = op->tmp1()->as_register();
2462 Register klass_RInfo = op->tmp2()->as_register();
2463 Register dst = op->result_opr()->as_register();
2464 Register Rtmp1 = Z_R1_scratch;
2465 ciKlass* k = op->klass();
2466
2467 assert(!op->tmp3()->is_valid(), "tmp3's not needed");
2468
2469 // Check if it needs to be profiled.
2470 ciMethodData* md = NULL;
2471 ciProfileData* data = NULL;
2472
2473 if (op->should_profile()) {
2474 ciMethod* method = op->profiled_method();
2475 assert(method != NULL, "Should have method");
2476 int bci = op->profiled_bci();
2477 md = method->method_data_or_null();
2478 assert(md != NULL, "Sanity");
2479 data = md->bci_to_data(bci);
2480 assert(data != NULL, "need data for type check");
2481 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2482 }
2483
2484 // Temp operands do not overlap with inputs, if this is their last
2485 // use (end of range is exclusive), so a register conflict is possible.
2486 if (obj == k_RInfo) {
2487 k_RInfo = dst;
2488 } else if (obj == klass_RInfo) {
2489 klass_RInfo = dst;
2490 }
2491 assert_different_registers(obj, k_RInfo, klass_RInfo);
2492
2493 if (op->should_profile()) {
2494 NearLabel not_null;
2495 __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2496 // Object is null; update MDO and exit.
2497 Register mdo = klass_RInfo;
2498 metadata2reg(md->constant_encoding(), mdo);
2499 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2500 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2501 __ or2mem_8(data_addr, header_bits);
2502 __ branch_optimized(Assembler::bcondAlways, *obj_is_null);
2503 __ bind(not_null);
2504 } else {
2505 __ compareU64_and_branch(obj, (intptr_t) 0, Assembler::bcondEqual, *obj_is_null);
2506 }
2507
2508 NearLabel profile_cast_failure, profile_cast_success;
2509 Label *failure_target = op->should_profile() ? &profile_cast_failure : failure;
2510 Label *success_target = op->should_profile() ? &profile_cast_success : success;
2511
2512 // Patching may screw with our temporaries,
2513 // so let's do it before loading the class.
2514 if (k->is_loaded()) {
2515 metadata2reg(k->constant_encoding(), k_RInfo);
2516 } else {
2517 klass2reg_with_patching(k_RInfo, op->info_for_patch());
2518 }
2519 assert(obj != k_RInfo, "must be different");
2520
2521 __ verify_oop(obj, FILE_AND_LINE);
2522
2523 // Get object class.
2524 // Not a safepoint as obj null check happens earlier.
2525 if (op->fast_check()) {
2526 if (UseCompressedClassPointers) {
2527 __ load_klass(klass_RInfo, obj);
2528 __ compareU64_and_branch(k_RInfo, klass_RInfo, Assembler::bcondNotEqual, *failure_target);
2529 } else {
2530 __ z_cg(k_RInfo, Address(obj, oopDesc::klass_offset_in_bytes()));
2531 __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2532 }
2533 // Successful cast, fall through to profile or jump.
2534 } else {
2535 bool need_slow_path = !k->is_loaded() ||
2536 ((int) k->super_check_offset() == in_bytes(Klass::secondary_super_cache_offset()));
2537 intptr_t super_check_offset = k->is_loaded() ? k->super_check_offset() : -1L;
2538 __ load_klass(klass_RInfo, obj);
2539 // Perform the fast part of the checking logic.
2540 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1,
2541 (need_slow_path ? success_target : NULL),
2542 failure_target, NULL,
2543 RegisterOrConstant(super_check_offset));
2544 if (need_slow_path) {
2545 // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2546 address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2547 store_parameter(klass_RInfo, 0); // sub
2548 store_parameter(k_RInfo, 1); // super
2549 emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2550 CHECK_BAILOUT2(profile_cast_failure, profile_cast_success);
2551 __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2552 // Fall through to success case.
2553 }
2554 }
2555
2556 if (op->should_profile()) {
2557 Register mdo = klass_RInfo, recv = k_RInfo;
2558 assert_different_registers(obj, mdo, recv);
2559 __ bind(profile_cast_success);
2560 metadata2reg(md->constant_encoding(), mdo);
2561 __ load_klass(recv, obj);
2562 type_profile_helper(mdo, md, data, recv, Rtmp1, success);
2563 __ branch_optimized(Assembler::bcondAlways, *success);
2564
2565 __ bind(profile_cast_failure);
2566 metadata2reg(md->constant_encoding(), mdo);
2567 __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2568 __ branch_optimized(Assembler::bcondAlways, *failure);
2569 } else {
2570 __ branch_optimized(Assembler::bcondAlways, *success);
2571 }
2572 }
2573
2574 void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
2575 LIR_Code code = op->code();
2576 if (code == lir_store_check) {
2577 Register value = op->object()->as_register();
2578 Register array = op->array()->as_register();
2579 Register k_RInfo = op->tmp1()->as_register();
2580 Register klass_RInfo = op->tmp2()->as_register();
2581 Register Rtmp1 = Z_R1_scratch;
2582
2583 CodeStub* stub = op->stub();
2584
2585 // Check if it needs to be profiled.
2586 ciMethodData* md = NULL;
2587 ciProfileData* data = NULL;
2588
2589 assert_different_registers(value, k_RInfo, klass_RInfo);
2590
2591 if (op->should_profile()) {
2592 ciMethod* method = op->profiled_method();
2593 assert(method != NULL, "Should have method");
2594 int bci = op->profiled_bci();
2595 md = method->method_data_or_null();
2596 assert(md != NULL, "Sanity");
2597 data = md->bci_to_data(bci);
2598 assert(data != NULL, "need data for type check");
2599 assert(data->is_ReceiverTypeData(), "need ReceiverTypeData for type check");
2600 }
2601 NearLabel profile_cast_success, profile_cast_failure, done;
2602 Label *success_target = op->should_profile() ? &profile_cast_success : &done;
2603 Label *failure_target = op->should_profile() ? &profile_cast_failure : stub->entry();
2604
2605 if (op->should_profile()) {
2606 NearLabel not_null;
2607 __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondNotEqual, not_null);
2608 // Object is null; update MDO and exit.
2609 Register mdo = klass_RInfo;
2610 metadata2reg(md->constant_encoding(), mdo);
2611 Address data_addr(mdo, md->byte_offset_of_slot(data, DataLayout::header_offset()));
2612 int header_bits = DataLayout::flag_mask_to_header_mask(BitData::null_seen_byte_constant());
2613 __ or2mem_8(data_addr, header_bits);
2614 __ branch_optimized(Assembler::bcondAlways, done);
2615 __ bind(not_null);
2616 } else {
2617 __ compareU64_and_branch(value, (intptr_t) 0, Assembler::bcondEqual, done);
2618 }
2619
2620 add_debug_info_for_null_check_here(op->info_for_exception());
2621 __ load_klass(k_RInfo, array);
2622 __ load_klass(klass_RInfo, value);
2623
2624 // Get instance klass (it's already uncompressed).
2625 __ z_lg(k_RInfo, Address(k_RInfo, ObjArrayKlass::element_klass_offset()));
2626 // Perform the fast part of the checking logic.
2627 __ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, success_target, failure_target, NULL);
2628 // Call out-of-line instance of __ check_klass_subtype_slow_path(...):
2629 address a = Runtime1::entry_for (Runtime1::slow_subtype_check_id);
2630 store_parameter(klass_RInfo, 0); // sub
2631 store_parameter(k_RInfo, 1); // super
2632 emit_call_c(a); // Sets condition code 0 for match (2 otherwise).
2633 CHECK_BAILOUT3(profile_cast_success, profile_cast_failure, done);
2634 __ branch_optimized(Assembler::bcondNotEqual, *failure_target);
2635 // Fall through to success case.
2636
2637 if (op->should_profile()) {
2638 Register mdo = klass_RInfo, recv = k_RInfo;
2639 assert_different_registers(value, mdo, recv);
2640 __ bind(profile_cast_success);
2641 metadata2reg(md->constant_encoding(), mdo);
2642 __ load_klass(recv, value);
2643 type_profile_helper(mdo, md, data, recv, Rtmp1, &done);
2644 __ branch_optimized(Assembler::bcondAlways, done);
2645
2646 __ bind(profile_cast_failure);
2647 metadata2reg(md->constant_encoding(), mdo);
2648 __ add2mem_64(Address(mdo, md->byte_offset_of_slot(data, CounterData::count_offset())), -(int)DataLayout::counter_increment, Rtmp1);
2649 __ branch_optimized(Assembler::bcondAlways, *stub->entry());
2650 }
2651
2652 __ bind(done);
2653 } else {
2654 if (code == lir_checkcast) {
2655 Register obj = op->object()->as_register();
2656 Register dst = op->result_opr()->as_register();
2657 NearLabel success;
2658 emit_typecheck_helper(op, &success, op->stub()->entry(), &success);
2659 __ bind(success);
2660 __ lgr_if_needed(dst, obj);
2661 } else {
2662 if (code == lir_instanceof) {
2663 Register obj = op->object()->as_register();
2664 Register dst = op->result_opr()->as_register();
2665 NearLabel success, failure, done;
2666 emit_typecheck_helper(op, &success, &failure, &failure);
2667 __ bind(failure);
2668 __ clear_reg(dst);
2669 __ branch_optimized(Assembler::bcondAlways, done);
2670 __ bind(success);
2671 __ load_const_optimized(dst, 1);
2672 __ bind(done);
2673 } else {
2674 ShouldNotReachHere();
2675 }
2676 }
2677 }
2678 }
2679
2680 void LIR_Assembler::emit_compare_and_swap(LIR_OpCompareAndSwap* op) {
2681 Register addr = op->addr()->as_pointer_register();
2682 Register t1_cmp = Z_R1_scratch;
2683 if (op->code() == lir_cas_long) {
2684 assert(VM_Version::supports_cx8(), "wrong machine");
2685 Register cmp_value_lo = op->cmp_value()->as_register_lo();
2686 Register new_value_lo = op->new_value()->as_register_lo();
2687 __ z_lgr(t1_cmp, cmp_value_lo);
2688 // Perform the compare and swap operation.
2689 __ z_csg(t1_cmp, new_value_lo, 0, addr);
2690 } else if (op->code() == lir_cas_int || op->code() == lir_cas_obj) {
2691 Register cmp_value = op->cmp_value()->as_register();
2692 Register new_value = op->new_value()->as_register();
2693 if (op->code() == lir_cas_obj) {
2694 if (UseCompressedOops) {
2695 t1_cmp = op->tmp1()->as_register();
2696 Register t2_new = op->tmp2()->as_register();
2697 assert_different_registers(cmp_value, new_value, addr, t1_cmp, t2_new);
2698 __ oop_encoder(t1_cmp, cmp_value, true /*maybe null*/);
2699 __ oop_encoder(t2_new, new_value, true /*maybe null*/);
2700 __ z_cs(t1_cmp, t2_new, 0, addr);
2701 } else {
2702 __ z_lgr(t1_cmp, cmp_value);
2703 __ z_csg(t1_cmp, new_value, 0, addr);
2704 }
2705 } else {
2706 __ z_lr(t1_cmp, cmp_value);
2707 __ z_cs(t1_cmp, new_value, 0, addr);
2708 }
2709 } else {
2710 ShouldNotReachHere(); // new lir_cas_??
2711 }
2712 }
2713
2714 void LIR_Assembler::breakpoint() {
2715 Unimplemented();
2716 // __ breakpoint_trap();
2717 }
2718
2719 void LIR_Assembler::push(LIR_Opr opr) {
2720 ShouldNotCallThis(); // unused
2721 }
2722
2723 void LIR_Assembler::pop(LIR_Opr opr) {
2724 ShouldNotCallThis(); // unused
2725 }
2726
2727 void LIR_Assembler::monitor_address(int monitor_no, LIR_Opr dst_opr) {
2728 Address addr = frame_map()->address_for_monitor_lock(monitor_no);
2729 __ add2reg(dst_opr->as_register(), addr.disp(), addr.base());
2730 }
2731
2732 void LIR_Assembler::emit_lock(LIR_OpLock* op) {
2733 Register obj = op->obj_opr()->as_register(); // May not be an oop.
2734 Register hdr = op->hdr_opr()->as_register();
2735 Register lock = op->lock_opr()->as_register();
2736 if (UseHeavyMonitors) {
2737 __ branch_optimized(Assembler::bcondAlways, *op->stub()->entry());
2738 } else if (op->code() == lir_lock) {
2739 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2740 // Add debug info for NullPointerException only if one is possible.
2741 if (op->info() != NULL) {
2742 add_debug_info_for_null_check_here(op->info());
2743 }
2744 __ lock_object(hdr, obj, lock, *op->stub()->entry());
2745 // done
2746 } else if (op->code() == lir_unlock) {
2747 assert(BasicLock::displaced_header_offset_in_bytes() == 0, "lock_reg must point to the displaced header");
2748 __ unlock_object(hdr, obj, lock, *op->stub()->entry());
2749 } else {
2750 ShouldNotReachHere();
2751 }
2752 __ bind(*op->stub()->continuation());
2753 }
2754
2755 void LIR_Assembler::emit_load_klass(LIR_OpLoadKlass* op) {
2756 Register obj = op->obj()->as_pointer_register();
2757 Register result = op->result_opr()->as_pointer_register();
2758
2759 CodeEmitInfo* info = op->info();
2760 if (info != NULL) {
2761 add_debug_info_for_null_check_here(info);
2762 }
2763
2764 if (UseCompressedClassPointers) {
2765 __ z_llgf(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2766 __ decode_klass_not_null(result);
2767 } else {
2768 __ z_lg(result, Address(obj, oopDesc::klass_offset_in_bytes()));
2769 }
2770 }
2771 void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
2772 ciMethod* method = op->profiled_method();
2773 int bci = op->profiled_bci();
2774 ciMethod* callee = op->profiled_callee();
2775
2776 // Update counter for all call types.
2777 ciMethodData* md = method->method_data_or_null();
2778 assert(md != NULL, "Sanity");
2779 ciProfileData* data = md->bci_to_data(bci);
2780 assert(data != NULL && data->is_CounterData(), "need CounterData for calls");
2781 assert(op->mdo()->is_single_cpu(), "mdo must be allocated");
2782 Register mdo = op->mdo()->as_register();
2783 assert(op->tmp1()->is_double_cpu(), "tmp1 must be allocated");
2784 Register tmp1 = op->tmp1()->as_register_lo();
2785 metadata2reg(md->constant_encoding(), mdo);
2786
2787 Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
2788 // Perform additional virtual call profiling for invokevirtual and
2789 // invokeinterface bytecodes
2790 if (op->should_profile_receiver_type()) {
2791 assert(op->recv()->is_single_cpu(), "recv must be allocated");
2792 Register recv = op->recv()->as_register();
2793 assert_different_registers(mdo, tmp1, recv);
2794 assert(data->is_VirtualCallData(), "need VirtualCallData for virtual calls");
2795 ciKlass* known_klass = op->known_holder();
2796 if (C1OptimizeVirtualCallProfiling && known_klass != NULL) {
2797 // We know the type that will be seen at this call site; we can
2798 // statically update the MethodData* rather than needing to do
2799 // dynamic tests on the receiver type.
2800
2801 // NOTE: we should probably put a lock around this search to
2802 // avoid collisions by concurrent compilations.
2803 ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
2804 uint i;
2805 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2806 ciKlass* receiver = vc_data->receiver(i);
2807 if (known_klass->equals(receiver)) {
2808 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2809 __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2810 return;
2811 }
2812 }
2813
2814 // Receiver type not found in profile data. Select an empty slot.
2815
2816 // Note that this is less efficient than it should be because it
2817 // always does a write to the receiver part of the
2818 // VirtualCallData rather than just the first time.
2819 for (i = 0; i < VirtualCallData::row_limit(); i++) {
2820 ciKlass* receiver = vc_data->receiver(i);
2821 if (receiver == NULL) {
2822 Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
2823 metadata2reg(known_klass->constant_encoding(), tmp1);
2824 __ z_stg(tmp1, recv_addr);
2825 Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
2826 __ add2mem_64(data_addr, DataLayout::counter_increment, tmp1);
2827 return;
2828 }
2829 }
2830 } else {
2831 __ load_klass(recv, recv);
2832 NearLabel update_done;
2833 type_profile_helper(mdo, md, data, recv, tmp1, &update_done);
2834 // Receiver did not match any saved receiver and there is no empty row for it.
2835 // Increment total counter to indicate polymorphic case.
2836 __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2837 __ bind(update_done);
2838 }
2839 } else {
2840 // static call
2841 __ add2mem_64(counter_addr, DataLayout::counter_increment, tmp1);
2842 }
2843 }
2844
2845 void LIR_Assembler::align_backward_branch_target() {
2846 __ align(OptoLoopAlignment);
2847 }
2848
2849 void LIR_Assembler::emit_delay(LIR_OpDelay* op) {
2850 ShouldNotCallThis(); // There are no delay slots on ZARCH_64.
2851 }
2852
2853 void LIR_Assembler::negate(LIR_Opr left, LIR_Opr dest, LIR_Opr tmp) {
2854 // tmp must be unused
2855 assert(tmp->is_illegal(), "wasting a register if tmp is allocated");
2856 assert(left->is_register(), "can only handle registers");
2857
2858 if (left->is_single_cpu()) {
2859 __ z_lcr(dest->as_register(), left->as_register());
2860 } else if (left->is_single_fpu()) {
2861 __ z_lcebr(dest->as_float_reg(), left->as_float_reg());
2862 } else if (left->is_double_fpu()) {
2863 __ z_lcdbr(dest->as_double_reg(), left->as_double_reg());
2864 } else {
2865 assert(left->is_double_cpu(), "Must be a long");
2866 __ z_lcgr(dest->as_register_lo(), left->as_register_lo());
2867 }
2868 }
2869
2870 void LIR_Assembler::rt_call(LIR_Opr result, address dest,
2871 const LIR_OprList* args, LIR_Opr tmp, CodeEmitInfo* info) {
2872 assert(!tmp->is_valid(), "don't need temporary");
2873 emit_call_c(dest);
2874 CHECK_BAILOUT();
2875 if (info != NULL) {
2876 add_call_info_here(info);
2877 }
2878 }
2879
2880 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {
2881 ShouldNotCallThis(); // not needed on ZARCH_64
2882 }
2883
2884 void LIR_Assembler::membar() {
2885 __ z_fence();
2886 }
2887
2888 void LIR_Assembler::membar_acquire() {
2889 __ z_acquire();
2890 }
2891
2892 void LIR_Assembler::membar_release() {
2893 __ z_release();
2894 }
2895
2896 void LIR_Assembler::membar_loadload() {
2897 __ z_acquire();
2898 }
2899
2900 void LIR_Assembler::membar_storestore() {
2901 __ z_release();
2902 }
2903
2904 void LIR_Assembler::membar_loadstore() {
2905 __ z_acquire();
2906 }
2907
2908 void LIR_Assembler::membar_storeload() {
2909 __ z_fence();
2910 }
2911
2912 void LIR_Assembler::on_spin_wait() {
2913 Unimplemented();
2914 }
2915
2916 void LIR_Assembler::leal(LIR_Opr addr_opr, LIR_Opr dest, LIR_PatchCode patch_code, CodeEmitInfo* info) {
2917 assert(patch_code == lir_patch_none, "Patch code not supported");
2918 LIR_Address* addr = addr_opr->as_address_ptr();
2919 assert(addr->scale() == LIR_Address::times_1, "scaling unsupported");
2920 __ load_address(dest->as_pointer_register(), as_Address(addr));
2921 }
2922
2923 void LIR_Assembler::get_thread(LIR_Opr result_reg) {
2924 ShouldNotCallThis(); // unused
2925 }
2926
2927 #ifdef ASSERT
2928 // Emit run-time assertion.
2929 void LIR_Assembler::emit_assert(LIR_OpAssert* op) {
2930 Unimplemented();
2931 }
2932 #endif
2933
2934 void LIR_Assembler::peephole(LIR_List*) {
2935 // Do nothing for now.
2936 }
2937
2938 void LIR_Assembler::atomic_op(LIR_Code code, LIR_Opr src, LIR_Opr data, LIR_Opr dest, LIR_Opr tmp) {
2939 assert(code == lir_xadd, "lir_xchg not supported");
2940 Address src_addr = as_Address(src->as_address_ptr());
2941 Register base = src_addr.base();
2942 intptr_t disp = src_addr.disp();
2943 if (src_addr.index()->is_valid()) {
2944 // LAA and LAAG do not support index register.
2945 __ load_address(Z_R1_scratch, src_addr);
2946 base = Z_R1_scratch;
2947 disp = 0;
2948 }
2949 if (data->type() == T_INT) {
2950 __ z_laa(dest->as_register(), data->as_register(), disp, base);
2951 } else if (data->type() == T_LONG) {
2952 assert(data->as_register_lo() == data->as_register_hi(), "should be a single register");
2953 __ z_laag(dest->as_register_lo(), data->as_register_lo(), disp, base);
2954 } else {
2955 ShouldNotReachHere();
2956 }
2957 }
2958
2959 void LIR_Assembler::emit_profile_type(LIR_OpProfileType* op) {
2960 Register obj = op->obj()->as_register();
2961 Register tmp1 = op->tmp()->as_pointer_register();
2962 Register tmp2 = Z_R1_scratch;
2963 Address mdo_addr = as_Address(op->mdp()->as_address_ptr());
2964 ciKlass* exact_klass = op->exact_klass();
2965 intptr_t current_klass = op->current_klass();
2966 bool not_null = op->not_null();
2967 bool no_conflict = op->no_conflict();
2968
2969 Label update, next, none, null_seen, init_klass;
2970
2971 bool do_null = !not_null;
2972 bool exact_klass_set = exact_klass != NULL && ciTypeEntries::valid_ciklass(current_klass) == exact_klass;
2973 bool do_update = !TypeEntries::is_type_unknown(current_klass) && !exact_klass_set;
2974
2975 assert(do_null || do_update, "why are we here?");
2976 assert(!TypeEntries::was_null_seen(current_klass) || do_update, "why are we here?");
2977
2978 __ verify_oop(obj, FILE_AND_LINE);
2979
2980 if (do_null || tmp1 != obj DEBUG_ONLY(|| true)) {
2981 __ z_ltgr(tmp1, obj);
2982 }
2983 if (do_null) {
2984 __ z_brnz(update);
2985 if (!TypeEntries::was_null_seen(current_klass)) {
2986 __ z_lg(tmp1, mdo_addr);
2987 __ z_oill(tmp1, TypeEntries::null_seen);
2988 __ z_stg(tmp1, mdo_addr);
2989 }
2990 if (do_update) {
2991 __ z_bru(next);
2992 }
2993 } else {
2994 __ asm_assert_ne("unexpect null obj", __LINE__);
2995 }
2996
2997 __ bind(update);
2998
2999 if (do_update) {
3000 #ifdef ASSERT
3001 if (exact_klass != NULL) {
3002 __ load_klass(tmp1, tmp1);
3003 metadata2reg(exact_klass->constant_encoding(), tmp2);
3004 __ z_cgr(tmp1, tmp2);
3005 __ asm_assert_eq("exact klass and actual klass differ", __LINE__);
3006 }
3007 #endif
3008
3009 Label do_update;
3010 __ z_lg(tmp2, mdo_addr);
3011
3012 if (!no_conflict) {
3013 if (exact_klass == NULL || TypeEntries::is_type_none(current_klass)) {
3014 if (exact_klass != NULL) {
3015 metadata2reg(exact_klass->constant_encoding(), tmp1);
3016 } else {
3017 __ load_klass(tmp1, tmp1);
3018 }
3019
3020 // Klass seen before: nothing to do (regardless of unknown bit).
3021 __ z_lgr(Z_R0_scratch, tmp2);
3022 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
3023 __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
3024 __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
3025
3026 // Already unknown: Nothing to do anymore.
3027 __ z_tmll(tmp2, TypeEntries::type_unknown);
3028 __ z_brc(Assembler::bcondAllOne, next);
3029
3030 if (TypeEntries::is_type_none(current_klass)) {
3031 __ z_lgr(Z_R0_scratch, tmp2);
3032 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
3033 __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
3034 __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, init_klass);
3035 }
3036 } else {
3037 assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3038 ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "conflict only");
3039
3040 // Already unknown: Nothing to do anymore.
3041 __ z_tmll(tmp2, TypeEntries::type_unknown);
3042 __ z_brc(Assembler::bcondAllOne, next);
3043 }
3044
3045 // Different than before. Cannot keep accurate profile.
3046 __ z_oill(tmp2, TypeEntries::type_unknown);
3047 __ z_bru(do_update);
3048 } else {
3049 // There's a single possible klass at this profile point.
3050 assert(exact_klass != NULL, "should be");
3051 if (TypeEntries::is_type_none(current_klass)) {
3052 metadata2reg(exact_klass->constant_encoding(), tmp1);
3053 __ z_lgr(Z_R0_scratch, tmp2);
3054 assert(Immediate::is_uimm(~TypeEntries::type_klass_mask, 16), "or change following instruction");
3055 __ z_nill(Z_R0_scratch, TypeEntries::type_klass_mask & 0xFFFF);
3056 __ compareU64_and_branch(Z_R0_scratch, tmp1, Assembler::bcondEqual, next);
3057 #ifdef ASSERT
3058 {
3059 Label ok;
3060 __ z_lgr(Z_R0_scratch, tmp2);
3061 assert(Immediate::is_uimm(~TypeEntries::type_mask, 16), "or change following instruction");
3062 __ z_nill(Z_R0_scratch, TypeEntries::type_mask & 0xFFFF);
3063 __ compareU64_and_branch(Z_R0_scratch, (intptr_t)0, Assembler::bcondEqual, ok);
3064 __ stop("unexpected profiling mismatch");
3065 __ bind(ok);
3066 }
3067 #endif
3068
3069 } else {
3070 assert(ciTypeEntries::valid_ciklass(current_klass) != NULL &&
3071 ciTypeEntries::valid_ciklass(current_klass) != exact_klass, "inconsistent");
3072
3073 // Already unknown: Nothing to do anymore.
3074 __ z_tmll(tmp2, TypeEntries::type_unknown);
3075 __ z_brc(Assembler::bcondAllOne, next);
3076 __ z_oill(tmp2, TypeEntries::type_unknown);
3077 __ z_bru(do_update);
3078 }
3079 }
3080
3081 __ bind(init_klass);
3082 // Combine klass and null_seen bit (only used if (tmp & type_mask)==0).
3083 __ z_ogr(tmp2, tmp1);
3084
3085 __ bind(do_update);
3086 __ z_stg(tmp2, mdo_addr);
3087
3088 __ bind(next);
3089 }
3090 }
3091
3092 void LIR_Assembler::emit_updatecrc32(LIR_OpUpdateCRC32* op) {
3093 assert(op->crc()->is_single_cpu(), "crc must be register");
3094 assert(op->val()->is_single_cpu(), "byte value must be register");
3095 assert(op->result_opr()->is_single_cpu(), "result must be register");
3096 Register crc = op->crc()->as_register();
3097 Register val = op->val()->as_register();
3098 Register res = op->result_opr()->as_register();
3099
3100 assert_different_registers(val, crc, res);
3101
3102 __ load_const_optimized(res, StubRoutines::crc_table_addr());
3103 __ kernel_crc32_singleByteReg(crc, val, res, true);
3104 __ z_lgfr(res, crc);
3105 }
3106
3107 #undef __